Datasets:

chemNLP

/

ncbi-text

like 0

Malaria is caused by species of , a protozoan parasite belonging to the phylum Apicomplexa. The parasite grows within erythrocytes, which it invades in a rapid, multistep process. Initial attachment of the merozoite is followed by reorientation, formation of an electron-dense junction between its apical prominence and the erythrocyte surface, and entry through this “moving junction” into a parasitophorous vacuole. Invasion is facilitated by the discharge of apical secretory organelles called micronemes. The various steps in invasion involve different receptor–ligand interactions. Primary binding is low affinity and is probably mediated by a glycosyl phosphatidylinositol–anchored protein complex mostly composed of fragments of merozoite surface protein 1 (MSP1), which is expressed around the merozoite circumference (), or by other surface-resident glycosyl phosphatidylinositol–anchored proteins (). Reorientation requires a microneme protein called apical membrane antigen 1 (AMA1), which is released onto the merozoite surface before attachment (). Junction formation constitutes an essentially irreversible, high-affinity binding step and involves members of a family of adhesive type I integral membrane microneme proteins known as the Duffy binding ligand-erythrocyte binding proteins (DBL-EBPs). The DBL-EBPs share a similar overall structure, with one or two erythrocyte binding DBL domains situated near their N terminus, a juxtamembrane cysteine-rich domain of unknown function called region VI, a transmembrane domain (TMD), and short cytoplasmic domain (). Despite this overall similarity, the DBL-EBP exhibit extensive sequence diversity and interact with a range of receptors. The best-characterized DBL-EBP is erythrocyte binding antigen 175 (EBA-175), which binds to a sialic acid–containing structure on glycophorin A (; ; ). Although some strains use predominantly this interaction for invasion, others (including many field isolates) do not, invading through alternative pathways (; ) that use other members of the DBL-EBP family (; ). There is extensive redundancy in invasion pathways, and the variety of DBL-EBP paralogues expressed by together with polymorphism within them may allow the parasite to evade immune responses and to invade a range of host cells despite widespread polymorphism in human erythrocyte surface molecules (; ). The structural and antigenic diversity across the DBL-EBP family poses hurdles for the development of vaccines or drugs that target these important ligands. Invasion by all Apicomplexa is accompanied by extensive proteolysis of surface and microneme proteins. During invasion by merozoites, both the MSP1 complex and AMA1 are quantitatively shed by cleavage at juxtamembrane sites. The enzyme responsible is a subtilisin-like “sheddase” called PfSUB2 that translocates across the parasite surface (; ). A distinct mode of shedding has been observed in the related apicomplexan . Here, several transmembrane microneme proteins that translocate across the parasite through interactions with a subplasmalemmal actinomyosin motor are shed in the final stages of invasion by cleavage within their TMD (; ). This “capping proteolysis” is mediated by an enzyme, initially dubbed microneme processing protease 1 (MPP1; ), that is now thought to be a rhomboid, a class of polytopic membrane serine proteases that cleave within the TMD of their substrates (; ; ; ; ). All the proteins shed by PfSUB2 and MPP1 have demonstrated or putative binding activity, and the fact that interventions that prevent shedding also inhibit invasion suggests that shedding is required to disengage binding interactions between the parasite and host cell surface, enabling rapid entry into the parasitophorous vacuole (). It has not been established whether intramembrane proteolysis plays a role in the release of adhesins. EBA-175 and all other known DBL-EBPs accumulate in soluble form in culture supernatants (; ; ), suggesting that their role at invasion may culminate in their being shed in a manner similar to the adhesins described above. The significance of DBL-EBP shedding has not been investigated. Here, we address this issue. We show that EBA-175 is secreted onto the surface of merozoites and is shed at or around the point of invasion. Shedding occurs by cleavage within the TMD at a site that is conserved in all DBL-EBPs. Cleavage is mediated by a rhomboid called PfROM4 that is expressed at the merozoite plasma membrane but not by other rhomboids tested. Importantly, parasite lines carrying mutations that prevent PfROM4-mediated cleavage could not be established, suggesting that shedding of EBA-175 is important for successful invasion. Our results show that intramembrane proteolysis by PfROM4 is critical to maintenance of the parasite life cycle and suggest that targeting PfROM4 activity may provide a means to interfere with the function of the entire DBL-EBP family. Antibodies raised against a folded, recombinant form of EBA- 175 region VI () were specific for schizont and merozoite-derived EBA-175 on Western blots and also reacted with the shed form in culture supernatants (). The shed protein therefore retains some or all of region VI, as previously suggested (; ). The antibodies produced a strong punctate immunofluorescence assay (IFA) pattern in acetone-fixed schizonts and free merozoites, consistent with the micronemal location of EBA-175 (). They also reacted specifically with the apical surface of many nonpermeabilized free merozoites, suggesting that EBA-175 is secreted onto the merozoite surface just before invasion (). In contrast, the anti–region VI antibodies did not react at all with most newly invaded rings of either the W2mef or 3D7 clones, labeling only a small fraction with a weak, punctate pattern ( and , second column). To examine this in more detail, we took advantage of the fact that parasite lines vary in their degree of dependence on EBA-175 as an invasion ligand. W2mef relies predominantly on sialic acid–dependent pathways for invasion and uses EBA-175 as its dominant invasion ligand, whereas 3D7 is capable of invading by both sialic acid–dependent and –independent routes (; ). Pretreatment of host erythrocytes with neuraminidase removes the sialic acids required for EBA-175 binding to glycophorin A (Fig. S1, available at ), precluding EBA-175–mediated invasion and forcing invasion to proceed by alternative pathways. As shown in , invasion into neuraminidase-treated cells did not alter the proportion of 3D7 rings reactive with the anti–region VI antibodies. Collectively, these results imply that at around the time of invasion, EBA-175 is predominantly shed from the merozoite and that this occurs to the same degree regardless of whether EBA-175 is used as a primary ligand mediating invasion. As EBA-175 plays its role in invasion at the parasite–host interface, shedding presumably takes place from the merozoite surface. The truncated form of EBA-175 that appears in culture supernatants is the product of this. To study the mechanism of EBA-175 shedding, we used an assay developed for characterizing the sheddase PfSUB2 (; ). Merozoites were incubated at 37°C with a range of test additives, and supernatants were analyzed for the presence of shed proteins. shows that shedding of EBA-175 was readily detectable, as was shedding of the expected MSP1 and AMA1 fragments. Release of EBA-175 was partially sensitive to PMSF and dichloroisocoumarin, indicating the involvement of a serine protease. However, in contrast to MSP1 and AMA1, EBA-175 shedding was not inhibited by either the calcium chelating agent EGTA or recombinant PfSUB2 prodomain (PfSUB2PD), a selective inhibitor of PfSUB2 (). These data show that EBA-175 is not shed by PfSUB2 but by a distinct, calcium-independent, serine protease. Shed EBA-175 can be recovered from culture medium by being bound to erythrocytes and then eluted with high salt (). We adapted this approach to purify sufficient EBA-175 for structural analysis. SDS-PAGE of the partially purified fractions revealed a major protein species at ∼175 kD (). Matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) analysis of in-gel tryptic digests confirmed this as EBA-175, with a MASCOT probability score of 462 (Table S1, available at ). The most C-terminal peptide identified in these digests corresponds to EFDDPSYTCFRK, which lies just upstream of the TMD (). No anomalous peptides were identified that might represent the N- or C-terminal end of the protein, so further samples were digested with Asp-N in the presence of 50% (vol/vol) O water. Under these conditions, products of digestion incorporate both O and O at their C termini as a result of proteolytic hydrolysis, whereas any peptide derived from the extreme C terminus of the protein substrate should retain a normal O isotope spectrum (, ). Examination of these digests by MALDI- TOF again identified only peptides derived from the EBA-175 ectodomain (MASCOT probability score 199), including a region VI–derived ion at 1740.716, matching DSEEYYNCTKREF (calculated 1740.717). As expected, all peptides identified with an N-terminal Asp and ending at a residue adjacent to an Asp-N cleavage site exhibited an O-containing spectrum. Within the anomalous peptides observed were only two unlabeled ions that could be assigned as being products of EBA-175 digestion: one at 2435.242, corresponding closely to the EBA-175–derived peptide DDPSYTCFRKEAFSSMPYYA (calculated 2435.032; ), and a second, less intense ion at 2451.262, matching its expected Met oxidized form (calculated 2451.032; unpublished data). The absence of an Asp-N cleavage site at the C terminus of this peptide, together with the absence of O label, implicated it as the C terminus of the shed EBA-175. Collision-induced fragmentation in a nanospray mass spectrometer () confirmed its identity and clearly detected the C-terminal sequence Y-A. These data map the C terminus of shed EBA-175 to an Ala residue that lies three residues into the TMD. Importantly, this Ala is conserved across the DBL-EBP family (; ). Of the few proteases that can hydrolyze intramembrane substrates, only rhomboids cleave close to the luminal side of the TMD; also, cleavage after an Ala is a hallmark of rhomboids (; ; ). Collectively, our data show that EBA-175 is shed by a protease with the characteristics of a rhomboid. Apicomplexan parasites possess several rhomboid-like genes (). has six, of which three (encoding TgROM1, -4, and -5) are expressed in tachyzoites in locations where they might play a role in invasion; TgROM1 is micronemal, whereas the other two larger rhomboids are expressed at the parasite plasma membrane (; ). Based on these expression patterns, the best candidates for the protease mediating shedding of EBA-175 were considered to be PfROM1, the orthologue of TgROM1, and PfROM4, the single orthologue of TgROM4 and -5. Microarray studies have shown that both (PlasmoDB ID PF11_0150) and (PFE0340c) are expressed in blood stages (; ) The predominantly intramembrane nature of rhomboids can make them poor antigens; therefore, to localize these gene products, we attempted to epitope tag the genes using single-crossover homologous recombination. For PfROM1, we transfected parasites with a plasmid containing a targeting sequence that included coding sequence fused to three copies of the HA epitope (HA3; Fig. S2 A, available at ). Drug-selected parasites (called 3D7ROM1HA) were cloned for further characterization. Southern blot (Fig. S2 B) and PCR analysis (unpublished data) confirmed that the plasmid had integrated through the expected recombination event, placing the HA3 tag at the 3′ end of the gene. Western blot of the clones (Fig. S3 A, available at ) revealed an anti-HA–reactive protein matching the predicted mass (26.7 kD) of HA-tagged PfROM1. Schizonts probed with mAb 3F10 showed a punctate IFA pattern that colocalized with the microneme protein AMA1 (Fig. S3 B), indicating that, like its homologue, PfROM1 is a microneme protein. IFA of free merozoites and ring stages showed that, unlike AMA1, which redistributes onto the merozoite surface just before invasion (), PfROM1 remains exclusively in an apical location; it could not be detected at the merozoite surface (Fig. S3, C and D) and did not colocalize with the plasma membrane marker MSP1 in rings (Fig. S3 E). Our results show that PfROM1 is restricted to micronemes, where it is unlikely to play a part in shedding of surface-located EBA-175. Attempts to epitope tag the 3′ end of the gene using a similar strategy failed (unpublished data), perhaps indicating that modification of the gene at this site is deleterious to the parasite. In an alternative approach, we transfected parasites with a construct for episomal expression of N-terminal HA3-tagged PfROM4, under the control of the promoter. Western blot of the resulting parasite line (called 3D7HAROM4) revealed expression of a novel, anti-HA–reactive protein of 84 kD, close to the expected mass of HA-tagged PfROM4 (). IFA showed this colocalized with the plasma membrane marker MSP1 in schizonts (). Identical results were obtained using the promoter to express the gene (unpublished data). These findings indicate that PfROM4 is expressed at the merozoite plasma membrane, an appropriate location to play a role in shedding of merozoite surface EBA-175. To explore whether EBA-175 could be a substrate for PfROM4, we used a heterologous expression system that has been used to characterize several rhomboids, including those from and . A synthetic “minigene,” encoding HA-tagged EBA-175 region VI plus its cognate TMD and cytoplasmic domain, was cloned into a vector for constitutive expression in mammalian cells (). The gene product was targeted to the secretory pathway by the murine Igκ signal sequence. COS-7 cells transfected with this construct showed strong surface fluorescence when probed without prior fixation with mAb 3F10, indicating expression of the protein in membrane bound form at the cell surface (). Western blot with the same mAb detected a major, ∼32-kD, and minor, 25-kD, protein in transfected cells. Region VI contains two potential -glycosylation sites at N1333 and N1401. -glycosylation is rare or absent in (), so one or both of these residues was substituted with Asp, resulting in expressed products EBAregVIgl1 and EBAregVI, respectively. The latter migrated as a single 25-kD species, close to its predicted nonglycosylated mass of 24 kD; it was also equally well expressed at the surface of transfected COS-7 cells (). Cells were then cotransfected with constructs designed for expression of either wild-type PfROM4 or a mutant form with the predicted active-site Ser residue replaced with Ala, and medium from the cells was examined by Western blot for the appearance of shed forms of the HA-tagged region VI. Coexpression with wild-type PfROM4 resulted in the appearance in cell supernatants of anti-HA–reactive proteins that migrated on SDS-PAGE more rapidly than the forms in cell extracts (). Importantly, levels of these shed forms were significantly lower, and not above background levels (unpublished data), when coexpressed with the mutant PfROM4. These results suggested that the shed protein is a result of PfROM4 proteolytic activity. All DBL-EBP possess a GA or GG motif just proximal to the extracellular side of their TMD (), reminiscent of the helix-destabilizing structures that determine sensitivity to rhomboids (). To explore the requirements for recognition of the EBA-175 TMD by PfROM4, three further forms of EBAregVI with mutations in the TMD were evaluated in the COS-7 cell system (). Substitution of A1427, the residue at which cleavage occurs to release EBA-175 from the merozoite surface (mutant EBAregVI-mutA), prevented PfROM4-mediated shedding. Similarly, substitution of the GA motif (called GA) closest to the extracellular end of the TMD (EBAregVI-mutGA), predicted as the site required for rhomboid recognition, abolished specific cleavage. In contrast, substitution of another GA motif (GA) lying near the cytoplasmic end of the TMD (EBAregVI-mutGA) had no effect on cleavage. These data show that the EBA-175 TMD is recognized and cleaved by PfROM4 and suggest that the requirements for recognition are similar to those characterized for other rhomboids. These experiments were extended, cotransfecting with constructs expressing PfROM1 or the rhomboid Rho-1 instead of PfROM4. In repeated experiments, and under conditions where Rho-1 efficiently cleaved AMA1 as shown previously (), we were unable to detect any activity of these rhomboids against the EBAregVI constructs (unpublished data). These findings imply a specific interaction between PfROM4 and EBA-175. To directly address the importance of EBA-175 shedding in the erythrocytic life cycle of , we attempted to generate transgenic parasites expressing mutant forms of EBA-175 that lacked one or other of the two TMD GA motifs examined in the previous section. To do this, we constructed plasmids (pHH1-175-GA/FF and pHH1-175-GA/FF) designed to introduce the desired substitutions by recombination into the 3′ end of the gene (). Each plasmid was first transfected into W2mef, and the resulting parasite lines (called EBA175GA/FF and EBA175GA/FF) were drug cycled to select for integrants. Southern blot analysis of EBA175GA/FF revealed a hybridization profile consistent with integration into the locus as expected, in the bulk of the parasite population (). Confirmation of the expected GA/FF substitution within the gene was obtained by PCR amplification () and sequencing (not depicted) of the modified locus. Expression and correct localization of the mutant EBA-175 protein was confirmed by IFA using antibodies against EBA-175 and -181, another micronemal EBL-DBP (). In contrast to the ease with which the EBA175GA/FF transgenic line was derived, establishment of a viable parasite line carrying a substitution of the GA motif could not be achieved. In two independent transfection experiments in W2mef and one in 3D7, the input plasmid remained episomal for up to six drug cycles (unpublished data), a typical finding where integration is lethal or severely detrimental to the parasite. Because the two integration constructs used here were identical aside from the few base-pairs within the TMD encoding the mutant codons, it is unlikely that the failure to obtain integration of pHH1-175-GA/FF was a technical problem associated with construct design. Our results support a critical role for the GA motif and suggest that modifications that interfere with PfROM4-mediated cleavage of EBA-175 cannot be tolerated by the parasite. Because shedding of EBA-175 occurs at or around the point of erythrocyte invasion, this is likely to be the stage in the life cycle at which arrest occurs after substitution of the GA motif. Many microneme proteins are secreted onto the parasite surface to play a role in host cell entry and then ultimately shed. This study demonstrates that EBA-175, and, by extrapolation, all other DBL-EBPs, are subject to a similar fate. Given their role in invasion and their capacity to bind erythrocyte surface receptors with high affinity, these ligands presumably function in membrane bound form at the merozoite surface. Our results show that the truncated form of EBA-175 released into supernatants is a result of a physiologically important, precise cleavage event that takes place at the merozoite surface and is mediated via intramembrane cleavage by a rhomboid-like malarial protease. IFA of newly invaded rings showed that, irrespective of whether EBA-175 was used as the dominant invasion ligand, invasion is associated with shedding of EBA-175. Western blot showed that the shed protein retains much or all of region VI, and mass spectrometric analysis allowed us to map its C terminus to an Ala residue that lies three residues into the predicted TMD, a specificity identical to that of the rhomboid-like cleavage sites previously mapped in apicomplexan microneme proteins (; ; ). We then showed that the homologues of three rhomboids previously localized to the secretory pathway in are expressed in merozoites. Whereas PfROM1 was found to be exclusively micronemal, PfROM4 localizes to the merozoite plasma membrane. This is consistent with the findings of and who showed that TgROM4 and -5, the two homologues of PfROM4, are both expressed at the tachyzoite surface. Of these, only TgROM5 showed protease activity against microneme proteins; thus, it was proposed that this is MPP1. also found that TgROM5 expression was concentrated at the posterior surface of the tachyzoite, where it may be optimally placed to release transmembrane microneme proteins as they translocate rearwards at invasion. We observed no such bias in the distribution of PfROM4, but our findings do indicate that this rhomboid is expressed at an appropriate time and place to mediate shedding of EBA-175. Early studies on the protease activity of rhomboids indicated that substrate recognition requires only a suitably disordered TMD resulting from the presence of a helix-disrupting GA or GG motif disposed toward the luminal end of the sequence (). Subsequent studies suggest that this may not be universally true, and there are now indications that structures on the extracellular (; ; ) or intracellular side () of the TMD are also important for efficient substrate recognition. The minigene constructs used here to explore cleavage of the EBA-175 TMD incorporated the entire region VI as well as the cytoplasmic domain in anticipation that elements critical for recognition might reside within these juxtamembrane sequences. Of the three rhomboids tested, only PfROM4 cleaved these proteins. Substitution of either the Ala mapped as the site of EBA-175 cleavage, or the GA motif, abolished cleavage. The lack of cleavage by either PfROM1 or Rho-1 shows that these motifs alone are not sufficient to render the EBA-175 TMD susceptible to all rhomboids and implies a degree of specificity in the interaction between PfROM4 and EBA-175. Eukaryotic rhomboids cluster into two major groups referred to as the RHO (typified by Rho-1) and PARL (typified by the mitochondrial presenilins-associated rhomboid-like protein) subfamilies (). Both the malarial rhomboids investigated here are more closely related to Rho-1 than to PARL, but PfROM4 differs noticeably from Rho-1 and PfROM1 in possessing unusually large predicted loops between its first and second, and sixth and seventh TMD, as well as extended N- and C-terminal domains (), structures that may be involved in EBA-175 recognition. Similarly, although region VI is known to be important for the trafficking and structural integrity of EBA-175 (), it might also play a role in recognition by PfROM4. EBA-175 is the dominant DBL-EBP mediating invasion in the W2mef line, so this isolate was first chosen to examine the effects of TMD mutations that prevent PfROM4-mediated cleavage. Although transgenic parasites lacking the GA motif were readily generated, parasites lacking the GA motif critical for PfROM4-mediated cleavage could not be. Further work showed that the GA mutation could not be introduced into the 3D7 line either. This was initially surprising in view of previous work showing that functional inactivation of EBA-175 by removal of the cytoplasmic tail () or deletion of the TMD () can be readily achieved in 3D7 and is easily compensated for in W2mef by a switch to an EBA-175–independent invasion pathway. We accept that caution should be exercised in interpretation of our “negative” result. However, together with our observation that shedding of EBA-175 takes place to the same degree in both parasite lines irrespective of whether the ligand is used for invasion, our results can be reconciled by a model that postulates that if EBA-175 is present in a functional, membrane bound form, it must be capable of being shed for invasion to go to completion. At present we can only speculate on how a defect in shedding could stall invasion. EBA-175 functions as a dimer (), and if shedding-defective mutants can form heterodimers with other DBL-EBPs, this may exert a dominant-negative effect, interfering with the release of these ligands too from the merozoite surface. EBA-175 is the most abundant DBL-EBP in most isolates examined; thus, an alternative possibility is that a defect in EBA-175 shedding leads to an insurmountable physical barrier at invasion, with an accumulation of unshed EBA-175 preventing the rapid recruitment of other DBL-EBPs to the confined space of the moving junction. It is worth noting that, somewhat analogous to our results, mutations that prevent MPP1-mediated shedding of the microneme protein TgMIC6 are deleterious to that parasite, despite the fact that TgMIC6 is not essential for host cell invasion (). Could PfROM4 shed merozoite proteins other than the DBL-EBPs? Recent studies have implicated a second family of apical membrane proteins in erythrocyte invasion. Called the reticulocyte binding protein homologues, several lines of evidence indicate an important role, including gene disruption experiments that result in a switch in invasion phenotype (). The precise function of these proteins is unclear, but they may have adhesive properties (), and proteolytic shedding may be important for their function (). Some members of this family possess helix-disrupting motifs in their TMD, and it will be interesting to determine whether they too can be substrates for PfROM4 or -1. We have been unable to disrupt the gene by targeted homologous recombination (unpublished data), suggesting that it is indispensable in the blood-stage life cycle of the parasite. Our findings show that erythrocyte invasion involves the activity of at least two serine proteases, PfSUB2 and PfROM4, which have distinct roles and which belong to different protease families. This is consistent with protease inhibitor studies performed over two decades ago indicating that invasion requires the activity of more than one serine protease (). Our evidence implicating PfROM4 in invasion by the malaria parasite supports the emerging view that rhomboid-mediated intramembrane proteolysis likely plays multiple roles in biological systems across evolution (). Our findings may have important practical implications for antimalarial drug development. has evolved a complex strategy to overcome receptor heterogeneity in its human host, exploiting a diverse family of ligands able to recognize a range of erythrocyte types. The conserved nature of the EBA-175 cleavage site across the DBL-EBP family raises the provocative notion that all these ligands may share a common requirement to be released at or around the point of invasion by PfROM4-mediated cleavage. Targeting PfROM4 with suitably selective drugs may transcend this ligand diversity, providing a more global approach to blocking invasion than attempting to target the receptor binding function of the ligands themselves. clones 3D7 and W2mef were cultivated and synchronized as described previously (). For invasion assays or production of newly invaded ring stages, schizonts enriched by centrifugation over cushions of 63% Percoll (GE Healthcare) were cultured with fresh erythrocytes for up to 3 h to allow rupture and reinvasion before removal of residual schizonts as described previously (). For transfection, ring-stage parasites (2–10% parasitaemia) were electroporated with 70–100 μg of plasmid DNA using standard procedures (). After initial selection for ∼4 wk in 2.5 or 10 nM of the antifolate WR99210 (Jacobus Pharmaceutical Co.), parasites were subjected to repeated cycles of selection for 3 wk followed by removal of the drug for 3 wk. Clonal populations were then obtained by limiting dilution. Recombinant region VI was expressed in the yeast . , encoding the C-terminal 170 residues (N1333-K1502) of 3D7 EBA-175 (PlasmoDB gene ID PF07_0128) was obtained from GENEART. An -glycosylation site at N1401 was substituted with Asp using QuikChange site-directed mutagenesis (Stratagene) with primers EBAmut12for and -rev (see Table S2, available at , for details of all primers used). Sequence encoding N1333 to M1423 was then PCR amplified with primers PPeba-A and -B to remove a second -glycosylation site at N1333 and add a 3′ hexa-histidine sequence. The product was digested with EcoRI and NotI and cloned into similarly digested pPIK9K (Invitrogen). This was electroporated into strain GS115, and integrant clones were selected as described previously (). Secreted recombinant protein was purified by nickel chelate chromatography and gel filtration and used to raise antisera in mice by standard protocols. Circular dichroism and SDS-PAGE revealed that the protein was folded. The mAbs 3F10 (anti-HA; Roche), X509 (anti-MSP1), 1E1 (anti-PfMSP1), 61.3 (anti-PfRhopH2), and 4G2 (anti-AMA1), used unmodified or conjugated to Alexa Fluor 594 (Invitrogen), as well as a mouse antiserum specific for AMA1 have been described (; ), as have rabbit and mouse polyclonal antibodies against EBA-175 () and EBA-181 (). Extracts of parasites, COS-7 cells, medium, and partially purified or purified proteins were subjected to SDS-PAGE under reducing or nonreducing conditions and transferred to Hybond-C extra (GE Healthcare), and membranes were probed with antibodies as described previously (; ). Binding was revealed by ECL detection (Pierce Chemical Co.) Shedding from purified merozoites and the effects of various inhibitors on it was assayed as described previously (; ). In brief, 3D7 merozoites suspended in 50 mM Tris-HCl and 5 mM CaCl, pH 7.6, were divided into equal aliquots of ∼4 × 10 merozoites on ice, supplemented with test inhibitors or control buffers, and transferred to 37°C for 1 h. Merozoites were pelleted, and shed proteins were detected in supernatants by Western blot using mAb 4G2, mAb X509, or anti–region VI antibodies. Thin films of were fixed with acetone or methanol and processed for IFA with mAbs as described previously (). For surface labeling of unfixed free merozoites, purified schizonts were cultured for 2 h in medium supplemented with mAb 3F10 (1:500) to allow merozoite release. Parasites were then pelleted and washed, thin films were prepared and fixed in acetone, and labeling was continued at the secondary antibody step. For probing with polyclonal antibodies, slides were incubated for 1 h with anti–region VI antisera (1:2,000) or a mixture of rabbit anti–EBA-175 (1:1,000) and mouse anti–EBA-181 (1:1,250) and then washed and incubated for 1 h with FITC-conjugated anti-mouse IgG (1:200; Sigma-Aldrich) or a mixture of Alexa Fluor 594 goat anti–rabbit IgG antibodies (1:2,000) and Alexa Fluor 488 anti-mouse IgG antibodies (1:2,000; Invitrogen) and DAPI (1:1,000, Roche). For IFA of unfixed COS-7 cells, cells adherent to glass coverslips were transfected and cultured for 48 h and then probed without prior fixation or permeabilization with mAb 3F10 (1:500) followed by biotinylated goat anti–rat IgG (1:500; Chemicon) and FITC-conjugated streptavidin (1:500; Vector Laboratories). Samples were mounted in Citifluor (Citifluor Ltd.). Images were captured at 21°C using AxioVision 3.1 software on an imaging system (Axioplan 2; Carl Zeiss MicroImaging, Inc.) using a Plan-APOCHROMAT 100×/1.4 (or Plan-APOCHROMAT 63×/1.4 for COS-7 cells) oil-immersion objective, and annotated using Photoshop (Adobe). Medium harvested from cultures of 3D7 schizonts that had undergone rupture and erythrocyte reinvasion overnight was concentrated 10-fold by ultrafiltration using a 100,000-D molecular mass cutoff membrane (PBHK; Millipore). Human erythrocytes were added to a 5% haematocrit, and the suspension was mixed for 30 min at room temperature to allow binding of EBA-175. The erythrocytes were recovered and resuspended for 10 min in 2 volumes of protein-free RPMI 1640 supplemented with 1.5 M NaCl, 10 mM EDTA, and 1 mM PMSF to elute bound protein. Several batches of protein produced in this manner were pooled and fractionated on a HiLoad 26/60 Superdex 200 pg gel filtration column equilibrated in 20 mM Tris-HCl, pH 8.0, and 150 mM NaCl. Fractions containing EBA-175 were identified by Western blot. In-gel proteolytic digestion of reduced, alkylated protein and peptide analysis by MALDI-TOF and nanospray mass spectrometry was done as described previously (, ; ). Erythrocytes were neuraminidase treated as described by . In brief, 150 μl of packed erythrocytes were washed and resuspended in 300 μl of protein-free RPMI 1640 25 mM Hepes, pH 6.7, containing 30 mU (30 μl) α2-3,6,8-neuraminidase (Calbiochem), and incubated at 37°C for 2.5 h. Mock-treated cells were incubated similarly without neuraminidase. The cells were washed in culture medium before use in invasion assays. EBA-175 overlay assays were performed as described by . In brief, untreated, mock-treated, or neuraminidase-treated cells were lysed in ice-cold 5 mM sodium phosphate, pH 7.0, and the resulting ghosts were washed extensively in the same buffer before being fractionated by SDS-PAGE on 7.5% gels and transferred to Hybond C. Membranes were blocked in 5% (wt/vol) milk powder in PBS 0.5% (vol/vol) Tween 20, incubated for 1 h at room temperature with gel filtration fractions containing partially purified 3D7 EBA-175 prepared as described above, and probed sequentially with anti–region VI antibodies (diluted 1:1,000) and horseradish peroxidase–conjugated goat anti–mouse IgG (1:8,000; Chemicon). Binding was revealed by ECL. Plasmid pPfROM1HA for epitope tagging of the gene by homologous recombination was constructed by the insertion of sequence encoding the HA3 epitope, downstream of the coding sequence, into plasmid pHH1 (). The HA3 tag was amplified from pREP(HA3)42 () using primers HAROMtagF and -R and inserted into the EspI and XhoI sites of pHH1 to produce pHH1HA. A target fragment comprising the entire 609 bp of the gene and the immediate 340-bp upstream sequence was amplified from 3D7 genomic DNA (gDNA) using primers PfROM1F and -R and inserted into pHH1HA using AflII–AvrII sites within the tag. Primers were designed to remove the stop codon from the end of the gene and form a continuous reading frame into the tag, at the end of which was a new stop codon. Synthetic and genes, each with an HA3 epitope tag introduced immediately after the N-terminal initiator Met residue and recodoned for mammalian expression, were gifts of M. Freeman (Medical Research Council Laboratory of Molecular Biology, Cambridge, UK). The gene, called , was excised from the provided plasmid and blunted with Klenow. For episomal expression in under control of the promoter, the 5′ flanking region was amplified from 3D7 gDNA using primers PfROM4upF and -R and cloned into the BglII–XhoI sites of pHH1. This vector was digested with XhoI and blunt ended, and the fragment was inserted. A 1.4-kb NotI Rep20 fragment was included in the final vector, called pHAROM4. The 3′ ∼1 kb of was amplified from cDNA using primers EBA175-S and EBA175-AS. Mutagenesis of this fragment was achieved in a two-step primer-directed PCR method with Vent polymerase (Stratagene). Substitution of codons encoding 1428GA1429 (GA) and 1440GA1441 (GA) with phenylalanine codons was achieved using primer pairs EBA175GA1/FF-S and EBA175GA1/FF-AS or EBA175GA2/FF-S and EBA175GA2/FF-AS, respectively. PCR products were sequenced to confirm the absence of undesired mutations, digested with BamHI and XhoI, and cloned into BglII–XhoI–digested pHH1 to produce constructs pHH1-175-GA/FF and pHH1-175-GA/FF. For Southern blot analysis, gDNA was extracted from parasites as described previously (), restricted and electrophoresed on 0.6% agarose gel, and transferred to Hybond N nylon membrane (GE Healthcare) by standard procedures. Probe preparation and Southern blotting was performed using the Prime-It II random primer labeling kit (Stratagene). Sequence encoding HA3 was PCR amplified from pREP(HA3)42 using primers pSecHAfor and -rev. The product was digested with EcoRI and ligated into pSecTag2a (Invitrogen) that had been digested with SfiI, blunted with T4 polymerase, and further digested with EcoRI. to produce expression vector pSecEBAregVIgl2. Mutagenesis to remove -glycosylation sites at N1333 and N1401 and to substitute codons for TMD residues A1427, 1428GA1429, and 1440GA1441 was performed by QuikChange mutagenesis (Stratagene) using primer pairs EBAmut13for and -rev, EBAmut12for and -rev, EBAactivefor1 and -rev1, EBATMmutfor1 and -rev1, and EBATMmutfor2 and -rev2, respectively. Constructs based on pcDNA3.1 (Invitrogen) for transient expression of HA-tagged Rho-1 in COS-7 cells (; ) were gifts of M. Freeman. Similar constructs for expression of HA-tagged PfROM4 and -1, containing the synthetic genes described above, were also provided by M. Freeman. Mutagenesis of the PfROM4 construct to convert the predicted active site Ser residue to Ala was done by QuikChange, using primers ROM4mutfor and -rev. Transfection of COS-7 cells in 6-well plates was done as described previously (; ) using 200 ng of each construct. 24 h after transfection, the medium was replaced with 1 ml per well of serum-free medium containing 25 μM Ilomastat (Calbiochem). After a further 24 h of incubation, the medium was harvested and concentrated to a final volume of 25 μl using Vivaspin 5000 concentrators (Sartorius). Cells from each well were also harvested and resuspended in 50 μl PBS. All samples were analyzed by Western blot with mAb 3F10. Fig. S1 shows that neuraminidase treatment of erythrocytes abolishes EBA-175 binding. Fig. S2 shows generation of a transgenic line expressing HA-tagged PfROM1. Fig. S3 demonstrates that PfROM1 is exclusively micronemal. Table S1 lists peptides identified by MALDI-TOF analysis of EBA-175 tryptic digests. Table S2 gives oligonucleotide primers used in this study. Online supplemental material is available at .

Cilia are microtubule (MT)-based nanomachines that perform diverse roles in motility, sensory perception, and signaling, and their dysfunction contributes to ciliary diseases (; ; ; ; ). Cilia are built and maintained by intraflagellar transport (IFT) motors that deliver ciliary precursors bound to protein complexes called IFT particles (consisting of subcomplexes IFT-A and -B; ) from the basal bodies to their sites of incorporation into cilia (). Understanding the mechanism of IFT and how defects in this process contribute to ciliary diseases is currently a topic of great interest. One ciliopathy that may reflect defects in IFT is Bardet-Biedl syndrome (BBS), a genetically heterogeneous disorder characterized by a pleiotropic phenotype that encompasses truncal obesity, pigmentary retinopathy, polydactyly, renal malformations, learning disabilities, hypogenitalism, and anomisa (). Mutations in 11 BBS genes are thought to cause defects in basal bodies or cilia, which may be a significant factor underlying this disease (; ; ). In , BBS proteins control the IFT motors that build cilia on sensory neurons, and, consequently, sensory cilia represent an appealing model to address mechanisms of IFT and the roles of BBS proteins in cilium biogenesis and disease (). amphid channel ciliary axonemes are made up of two domains: an initial segment (called the middle segment) containing 4-μm–long MT doublets extending from the 1-μm–long transition zone (a modified basal body that is also called the proximal segment) that together form the cilium foundation and a distal segment comprising 2.5-μm–long MT singlets (). We have previously shown that the IFT particles assembling these sensory cilia are moved by the coordinate action of two anterograde IFT motors called kinesin-II and OSM-3, which are both members of the kinesin-2 family (; ; ; ). These motors function redundantly to move the same IFT particles along the initial segment and build the cilium foundation, with either motor but not both being dispensable for this function (; ). Then, OSM-3 alone extends singlet MTs on the distal ends of the cilium core in a process involving OSM-3 movement along these distal singlets (; ; ). In this study, we focus on the question of how kinesin-II and OSM-3 are functionally coordinated to move the same IFT particle along the initial segment of amphid channel cilia. The rates of IFT seen in , and mutants suggest that kinesin-II alone moves along MTs at 0.5 μm/s and OSM-3 alone moves at 1.3 μm/s. This also suggests that the intermediate rate of transport seen in the initial segment of wild-type cilia (0.7 μm/s) results from the action of both motors (; ), but the rates of MT motility predicted for the purified motors have not been tested using in vitro motility assays. A related question is how BBS proteins contribute to the functional coordination of kinesin-2 motors. In , BBS-1, -2, -3, -5, -7, and -8 were shown to be ciliary proteins (), and, of these, the loss of BBS-7 and -8 function in mutant animals leads to the loss of ciliary distal segments and sensory defects (). Using in vivo transport assays, we observed that in wild-type animals, IFT particle subcomplexes IFT-A and -B move together along the initial segment with kinesin-II and OSM-3 at a single rate of 0.7 μm/s. However, in -7/-8 mutants, kinesin-II and IFT-A move together at 0.5 μm/s, but OSM-3–kinesin and IFT-B move as a distinct complex at 1.3 μm/s. This suggested that BBS-7/-8 proteins coordinate IFT by holding subcomplexes IFT-A and -B together and stabilizing the integrity of the IFT particles (). This offers a unique system for probing the mechanism by which BBS proteins contribute to kinesin-II and OSM-3 motor coordination. The power of as a system for addressing these questions would be enhanced if in vivo time-lapse microscopy assays of IFT () could be complemented by in vitro motility assays (), but this has not been performed because of the low abundance of native kinesin-2 motors (). Here, we have initiated such in vitro assays using purified recombinant kinesin-II and OSM-3. In this study, we combine both in vivo and in vitro motility assays of kinesin-II and OSM-3 to determine whether the cooperative motility observed in cilia is an intrinsic property of the motors alone or whether it depends on additional ciliary factors; what the mechanism is by which the two IFT kinesins cooperate to move IFT particles to redundantly assemble the cilium foundation and whether motor cooperation contributes to the dissociation of the IFT particles in the mutants; and whether we can develop quantitative models for motor coordination that account for the in vivo and in vitro velocities of the motors. The results illuminate the mechanism by which the two same-polarity IFT motors cooperate to move an IFT particle along a cilium. The question of whether the motor coordination that gives rise to an intermediate rate of transport is an intrinsic property of the motors or whether it requires additional ciliary cofactors can be addressed using competitive in vitro motility assays with mixtures of purified kinesin-II and OSM-3. Because it is difficult to purify native holoenzymes of kinesin-II and OSM-3 from (), we used baculovirus and systems to overexpress and purify recombinant kinesin-II (see next paragraph) and OSM-3 (; see Imanishi et al. on p. of this issue), and we sought conditions under which these preparations drive motility at the rates predicted from in vivo experiments (). Purified kinesin-II behaved as a monodisperse, heterotrimeric complex on sucrose gradients () and gel filtration columns () consisting of 1 mol each of its subunits KLP-11, KLP-20, and KAP-1 with a native molecular mass of 287 kD (supplemental material and Fig. S1, available at ). These hydrodynamic properties are very similar to those of the native kinesin-II holoenzyme in extracts (). In MT gliding assays, kinesin-II–driven motility conformed to Michaelis-Menten kinetics with a nucleotide profile very similar to that of kinesin-1 (), indicating that Mg-ATP is the preferred substrate for kinesin-2 motors (., supplemental material, and Fig. S2). In MT gliding assays performed under standard conditions (in BRB80, which contains 80 mM Pipes), we observed that purified kinesin-II and OSM-3 both used Mg-ATP according to Michaelis-Menten kinetics (). However, kinesin-II alone moved MTs at a maximal rate of 0.3 μm/s, and OSM-3 alone moved them at 0.7 μm/s. Although the relative rates of motility driven by the two motors were consistent with in vivo transport assays, the actual rates observed in vitro were approximately twofold lower than those seen in vivo. Therefore, we sought in vitro assay conditions that supported the rates of transport observed in vivo and found that by lowering the K-Pipes buffer concentration, rates very similar to those predicted from in vivo assays were observed (; also see next section). Recombinant OSM-3 is purified in an active homodimeric state, but it displays autoinhibition and drives a low MT gliding velocity (0.3 μm/s) when assayed on antibody-coated surfaces, and it can be activated by the mutagenesis of glycine residue 444 to glutamate (; ). To control for this potential complication, we compared MT gliding driven by wild-type OSM-3 on antibody-coated surfaces or directly coated onto the coverslip with that of the OSM-3–G444E mutant protein in different Pipes concentrations () and observed that directly adsorbed OSM-3 and OSM-3–G444E both supported MT gliding at similar rates of 1.1 and 0.97 μm/s, respectively. We conclude that direct adsorption onto the coverslip, as in our standard gliding assays, activates autoinhibited OSM-3 to the same extent as mutagenesis, and, consequently, the assays of both the wild-type OSM-3 and OSM-3–G444E discussed in the following section refer to the active state. Based on the aforementioned results, we analyzed the rate of MT-based motility driven by mixtures of varying molar ratios of pure kinesin-II and OSM-3 in the presence of 45 mM Pipes for wild-type OSM-3 and 25 mM Pipes for OSM-3–G444E ( and ). Under these optimized conditions, kinesin-II alone moved at ∼0.5 μm/s, whereas OSM-3 alone (wild type or the G444E mutant) moved at ∼1.1 μm/s (, , , and Video 1; available at ), which is very close to the rates of transport driven by kinesin-II (0.5 μm/s) and OSM-3 (1.3 μm/s) in vivo (; ). Moreover, mixtures of the two motors displayed intermediate rates of motility, with the rate varying in a nonlinear fashion with the molar ratio and replicating the in vivo rate of 0.7 μm/s at a mole fraction of OSM-3 between 0.6 and 0.8 (, , and Video 1). This suggests that intermediate velocities of motility, like those observed along the initial segments of cilia, can be generated by simple functional interactions between kinesin-II and OSM-3 without any requirement for additional ciliary factors such as regulators of motility, the presence of IFT particles bound to the motors as cargo, an overlying ciliary membrane, or cilia-specific axonemal MT doublets to serve as tracks for the motors. The observation that kinesin-II and OSM-3 cooperate to drive MT motility at an intermediate rate that depends on their molar ratio can be explained using two types of models, as described in detail in the supplemental material. In the first model, the alternating action model, we propose that the motors act sequentially so that a fast step or a run of several fast steps driven by OSM-3 alternates with a slow step or a run of slow steps taken by kinesin-II (Fig. S3 A, available at ). The periodic switching of the two motors would average out to produce the intermediate velocity of motility observed in vitro and along the cilium. In the second model, the mechanical competition model, the motors act simultaneously in a concerted fashion, with the slower moving kinesin-II exerting drag on the faster OSM-3 and the faster moving OSM-3 pulling kinesin-II along and speeding it up (Fig. S3 B). We developed quantitative alternating action and mechanical competition models from which we derived equations that relate the speed of motility to the mole fraction of the two motors (supplemental material). We observed that by using reasonable parameter adjustments, the equations derived from both models displayed excellent fit to the data points in plots of MT gliding velocity versus the mole fraction of OSM-3 (), and, thus, the two models can account for the gliding assay data and the rates of IFT particle transport observed along the cilia of wild-type, , and animals (see Discussion). Thus, a significant conclusion from our quantitative modeling is that both the alternating action and the mechanical competition models are highly plausible, but this analysis by itself did not allow us to decide whether alternating action or mechanical competition is the more likely mechanism of motor coordination. However, the alternating action and mechanical competition models do make distinct predictions concerning the transport of IFT particles in double mutants lacking BBS proteins and either kinesin-II or OSM-3 (, , or mutants). Let us consider why IFT particle subcomplexes A and B apparently move together along the initial segment of the cilium at 0.7 μm/s in wild types, whereas in mutants, IFT-A is apparently moved by kinesin-II at 0.5 μm/s, and IFT-B is moved at ∼1.1–1.3 μm/s by OSM-3 (). We assume that the BBS proteins stabilize intact IFT particles, which therefore dissociate into IFT-A and -B in the loss of function mutants. Although this dissociation could be passive, we reasoned that it might instead be an active process caused by stresses imposed on the IFT particles by the concerted action of kinesin-II and OSM-3 motors moving the IFT particles together if the slow motor exerts drag on the fast motor and vice versa, as in the mechanical competition model (supplemental material). Such stresses could not be developed by motors acting sequentially, as in the alternating action model, because only one type of motor will be moving the particle at any one time (supplemental material). Furthermore, if the stresses that dissociate IFT particles in single mutants require mechanical competition between kinesin-II and OSM-3, the loss of either motor together with BBS protein function, as in double mutants, should prevent IFT particle dissociation. Based on these arguments, the alternating action model predicts that in double mutants, IFT particles will dissociate passively into IFT-A and -B, only one of which is moved along the cilium (). On the other hand, the mechanical competition model predicts that the IFT particles should remain intact in the absence of the tension exerted on IFT particles by the concerted action of the competing motors (). We tested the aforementioned model predictions by assaying IFT and examining the ciliary phenotypes in and double mutants. Specifically, we made double mutant strains using CHE-11∷GFP to mark IFT-A and CHE-2∷GFP or OSM-6∷GFP to track IFT-B. In wild-type animals, as we demonstrated before, CHE-11∷GFP and CHE-2∷GFP move identically along both initial and distal segments of sensory cilia (), but in single mutants (), IFT particles A and B dissociate and move separately (). Significantly, however, in double mutants (; , and Videos 2 and 4; available at ), CHE-11∷GFP enters the distal segments and moves at the same velocity as CHE-2∷GFP or OSM-6∷GFP, which is characteristic of OSM-3–kinesin's fast speed along both the initial and distal segments. In double mutants (; , and Videos 3 and 5), both CHE-2∷GFP and OSM-6∷GFP enter the remaining initial segments and are moved by kinesin-II at its characteristic slow velocity. Thus, we observed the following: The and double mutants have phenotypes identical to the and mutants, respectively. In particular, the double mutants have almost full-length cilia, which is similar to the mutants, and the - double mutants have truncated cilia similar to the mutants. Unlike in the single mutants, in the and double mutants, the IFT particles remain intact, and the IFT-A and -B subcomplexes move together at an identical rate, which is characteristic of OSM-3 or kinesin-II, respectively ( and ). These results support the predictions of the mechanical competition model (). Accordingly, we propose that in wild-type , IFT particles are moved along sensory cilia by kinesin-II and OSM-3–kinesin acting together, with the slower moving kinesin-II exerting drag on the faster moving OSM-3, whereas the faster moving OSM-3 tends to pull the slower moving kinesin-II along (). This produces a mechanical competition that translates into tension across the IFT particles, leading to their dissociation in the absence of the BBS proteins (, mutant). Thus, the BBS-7 and -8 proteins antagonize this tension force and maintain the integrity of the IFT particles by stabilizing the association of IFT-A with IFT-B (, wild type [WT]). Conversely, in or single mutants, this stabilization is lost, and the motor-dependent stresses dissociate subcomplexes A and B, which are moved separately by kinesin-II or OSM-3 (). In the double mutants, however, the remaining kinesin-2 motor lacks its antagonistic partner, so the stresses required to dissociate IFT-A from IFT-B are absent. Consequently, the IFT particles are moved intact along the cilium by kinesin-II or OSM-3–kinesin alone (, and ). This study used competitive motility assays, quantitative modeling, and in vivo transport assays in ciliary mutants to investigate how kinesin-II and OSM-3 cooperate to move IFT particles along the initial segment of the cilium. This required purified kinesin-II (this study) and OSM-3 kinesin (). Native kinesin-II had been purified from sea urchin embryos (, ; ; ) and subsequently from other systems (; ; ; ; ; ; ) in amounts that allowed the basic motility properties of several members of the kinesin-2 family to be analyzed, but the low abundance of native kinesin-II precluded motility assays (). This problem was circumvented here using baculovirus expression. The availability of purified kinesin-II and OSM-3 allowed us to assess their role in IFT by comparing in vivo and in vitro motility assays. The rates of MT gliding driven by kinesin-II and OSM-3 (0.4–0.5 and 1.1 μm/s) are similar to the rates of anterograde movement of GFP∷kinesin-II and GFP∷OSM-3 alone along sensory cilia (0.5 and 1.3 μm/s), and the intermediate rate of 0.7 μm/s seen along the initial segment of the cilium can be recapitulated in gliding assays using mixtures of kinesin-II and OSM-3 (; ; ). This is striking given the different conditions under which the motors drive MT gliding over glass coverslips versus driving IFT particle transport along cilia, where different ATP concentrations, different MT tracks, and the presence of IFT particles or other regulatory cofactors could influence motility. Both quantitative modeling and in vivo IFT assays using double mutants are concordant with the intermediate rate being caused by simple mechanical competition between kinesin-II and OSM-3, which further suggests that motor coordination does not require sophisticated regulatory mechanisms to turn the motors on and off. In this model, BBS proteins coordinate the motors simply by maintaining the association of kinesin-II–IFT-A with OSM-3–IFT-B, which otherwise dissociate because of tension exerted by the two motors (e.g., in mutants). In gliding assays, BBS proteins are not required because the coverslip forms a physical connection between adjacent kinesin-II and OSM-3 proteins. Our model assumes that IFT subcomplexes A and B normally interact to form a single IFT particle complex that is moved along the wild-type cilium by both kinesin-II and OSM-3 (which may move along the MT A and B subfibers, respectively; ). However, although IFT particles isolated from flagella (which probably lack BBS proteins) sediment as a 16-S complex, they can also be separated by varying the solution conditions, raising the possibility that IFT-A and -B usually exist as separate complexes in vivo (; ). For example, our previous data could be explained if in wild types, separate complexes of kinesin-II–OSM-3–IFT-A and kinesin-II–OSM-3–IFT-B move at the same rate along the initial segment of the cilium, with BBS proteins being required to dock OSM-3 onto the kinesin-II–IFT-B complex and kinesin-II onto the OSM-3–kinesin-II complex (). However, this predicts that in and double mutants, IFT-A and -B, respectively, would not move along the cilium at all, which is inconsistent with our IFT assays. More work is required to firmly establish that IFT-A and -B normally interact to form a single transport complex in vivo. Although BBS proteins appear to be required for the stabilization of IFT particles in systems that use both kinesin-II and OSM-3 for ciliogenesis, in organisms where kinesin-II acts alone (for example, ), the two-motor–dependent mechanical competition is lacking, so IFT particles should remain intact in mutants, and BBS proteins may not be needed. Indeed, BBS-7/-8 proteins are absent in the flagellome (), although it is possible that they, along with OSM-3, enter the flagellum and elongate distal singlets during mating (). In vertebrates, the OSM-3 homologue KIF17 is required to target cyclic nucleotide-gated channels to the cilium (). Moreover, prominent distal singlets of the type found in sensory cilia occur in various organisms (e.g., human [] and frog olfactory cilia []), and cyclic nucleotide-gated channels cluster over a region of the distal segments in the latter cilia (). Interestingly, BBS knockout mice (BBS-1, -2, and -4) specifically lose the distal segments of their olfactory cilia, which results in anosmia (; ; ). Thus, BBS proteins may contribute to IFT specifically in cilia containing distal segments, where two anterograde IFT motors are used. The BBS protein–dependent mechanical competition between IFT motors, which was uncovered in sensory neurons, may be relevant to the assembly of similar two-domain cilia in some vertebrates in which defects in this process may underlie BBS. The antagonistic competition between opposite polarity mitotic MT motors has been discussed extensively (), and precedents also exist for functional interaction between same-polarity motors. In muscles, for example, slower cycling cross-bridges can exert drag on faster cycling bridges and, thereby, slow down the maximal velocity of the shortening of the muscle (). It has also been shown that a motor can be accelerated by a force applied in the direction of MT gliding driven by that motor (). Moreover, the biophysical plausibility of our proposed mechanism is supported by experiments in which plus end–directed kinesin motors with different speeds were shown to interact to produce intermediate speeds in motility assays, including kinesin-1 and -5 () and mixed Kif3A and Kif3B homodimers (). The striking studies of illuminate how the distinct Kif3A and Kif3B motor domains can cooperate within the processive kinesin-II holoenzyme, but the competition between kinesin-1 versus kinesin-5 and Kif3A versus Kif3B homodimers are unlikely to reflect true in vivo interactions. To our knowledge, our study is the first to uncover competitive motility between distinct, same-polarity MT-based intracellular transport motor holoenzymes that are known to cooperate in vivo. This system differs from the controlled coordination that exists between kinesin-II and cytoplasmic dynein on melanosomes () and other cargoes (), in which the simultaneous activation and inhibition of the anterograde and retrograde motors (and vice versa) facilitates alternating runs in the anterograde and retrograde direction (). The distinct mechanical competition proposed here for same-polarity IFT motors may represent yet another general way in which motors are coordinated and controlled to produce coherent networks of intracellular transport within eukaryotic cells. The quantitative mechanical competition model makes testable predictions. For example, the unloaded velocities of kinesin-II and OSM-3, and, are known, but the ratio of their stall forces, γ, is an unknown free parameter (supplemental material). By adjusting γ and comparing the resulting curve to the gliding assay data, we find the best fit when the ratio of OSM-3 to kinesin-II stall forces issuggesting that kinesin-II and OSM-3 stall at similar forces, which can be tested in future experiments. We can also estimate the molar ratio of OSM-3 and kinesin-II on IFT particles by combining the stall force ratio (γ = 0.98), the unloaded velocities of the two motors ( and ), and the velocity of IFT particle transport driven by the concerted action of the two motors ( = 0.72 μm/s, yielding C = 1.25). This leads to the prediction that in vivo, the mole fraction of OSM-3 is α ≈ 0.45, reflecting an approximately equimolar ratio of the two motors on the IFT particles within the initial segment of the cilium. The analysis of isolated motor–IFT particle complexes may allow us to test this prediction. Overall, this work illuminates how two anterograde IFT motors cooperate to move IFT particles along the initial segment of the axoneme at a rate that is intermediate between the free-sliding rate of each motor alone to build the cilium foundation on dendritic endings of sensory neurons. Sf9 cells infected with baculovirus containing the three genes , , and that encode kinesin-II expressed sufficient quantities of the corresponding subunits to permit the purification of the heterotrimeric complex in a monodisperse, active state in high yields (). To accomplish this, PCR amplifications containing the cDNA sequences for KAP-1, KLP-11, and KLP-20 (GenBank/EMBL/DDBJ accession no. , , and , respectively) were inserted into Gateway vector pDONR221 (Invitrogen) and cloned into target vector pDEST8 (Invitrogen), and recombinant baculoviruses were generated according to the manufacturer's instructions. The infected cells were then incubated for 3 d at 27°C before being harvested. Cell pellets from 400 ml of culture were suspended in 80 ml of ice-cold lysis buffer containing 50 mM Pipes, pH 7.0, 300 mM NaCl, 1 mM MgCl, 1 mM 2-mercaptoethanol, and EDTA-free protease inhibitor mixture (Roche) and passed twice through a French press at 1,000 pounds per square inch. The cell lysate was centrifuged for 30 min at 1,5000 and 4°C. The supernatant was purified using Talon affinity beads (BD Biosciences) according to the manufacturer's instructions. The Talon-purified protein was dialyzed against gel filtration column buffer containing 80 mM Pipes, pH 6.9, 200 mM NaCl, 1 mM MgCl, 1 mM EGTA, 1 mM DTT, and 0.1 mM ATP. The kinesin-II complex was further purified on a column (Sephacryl S-300HR; GE Healthcare) in gel filtration buffer and concentrated by ultrafiltration with Centriprep 30K (Amicon). This simple procedure of Talon column affinity followed by Sephacryl S-300 gel filtration chromatography routinely yielded 2 mg of highly purified kinesin-II per 400 ml of starting material ( and Fig. S1 A). We estimate that we would need to start with 4,000 liters of mixed stage worm culture (i.e., 10,000 times more starting material) to purify kinesin-II in comparable amounts (). Sucrose density gradient centrifugation and gel filtration chromatography were performed as described previously (). The molecular weight of the kinesin-II complex was calculated using the Siegel and Monty equation (; ; ). MT motility assays were performed as described previously at 21°C (). The velocities of 10–60 MTs were measured for each data point. IFT was assayed as described previously (; ). The GFP transgenic worms were anesthetized with 10 mM levamisole, mounted on agar pads, and maintained at 21°C. We collected images with a microscope (IX70; Olympus) equipped with a 100× NA 1.35 objective and a spinning disc confocal head (UltraVIEW; PerkinElmer) with excitation by 488-argon ion lasers at 0.3 s/frame for 2–3 min. All images were acquired using cooled charge-coupled device cameras (ORCA-ER; Hamamatsu), and kymographs and videos were created using MetaMorph software (Universal Imaging Corp.). Transgenic animals expressing , and were crossed with , and , or to create double mutants, and their genotypes were confirmed by their dye-filling phenotype and/or PCR. Supplemental material provides data supporting purified kinesin-II as a monodisperse heterotrimeric complex whose motility conforms to Michaelis-Menten kinetics and describing quantitative models for the functional coordination of the two anterograde IFT kinesins. Fig. S1 shows the purification and hydrodynamic analysis of the heterotrimeric kinesin-II from Sf9 cell extracts. Fig. S2 shows Michaelis-Menten analysis of the motility activity of purified kinesin-II in the presence and absence of nucleotide substrates and inhibitors. Fig. S3 shows the alternating action and mechanical competition models that explain the comovement of kinesin-II and OSM-3–kinesin along sensory cilia. Fig. S4 shows motility assay data obtained using kinesin-II or OSM-3 alone and mixtures of the two motors together with the gliding velocity versus mole fraction relationship. Video 1 shows in vitro assays of MT gliding induced by purified kinesin-II, purified OSM-3, and a mixture of the two motors. Videos 2–5 show in vivo transport assays of the movement of IFT particle proteins along sensory cilia of double mutant strains ; and ;. Online supplemental material is available at .

The development of new blood vessels from existing vasculature termed angiogenesis is an essential physiological process and is critical for certain pathological disorders including rheumatoid arthritis, diabetic retinopathy, atherosclerosis, psoriasis, and tumor growth/metastasis (; ; ). Multiple endogenous factors have been implicated in promoting and suppressing angiogenesis, and a balance between pro- and antiangiogenic activities determines the angiogenic response. Both intact proteins and proteolytic fragments can be endogenous angiogenesis inhibitors. Although plasminogen, type XVIII collagen (Col XVIII), Col XV, Col IVα1, Col IVα2, Col IVα3, and fibronectin lack antiangiogenic activity, proteolytic cleavage of these proteins yields angiostatin, endostatin, endostatin-like fragment from type XV collagen, arresten, canstatin, tumstatin, and anastelin, respectively, which are antiangiogenic (). Thrombospondin 1 exemplifies an antiangiogenic protein that functions as an intact protein (; ). A protein precursor and proteolytic product can both be antiangiogenic as illustrated by calreticulin and its N-terminal fragment vasostatin (, ). Endogenous angiogenesis inhibitors target endothelium via suppressing cell proliferation and migration, inducing apoptosis, down-regulating proangiogenic factors and signaling pathways, and inducing antiangiogenic factors (; ; ). Endogenous angiogenesis inhibitors can antagonize cell surface integrins (; ). Novel antiangiogenic mechanisms include impaired tubulin polymerization (), inhibition of ATP synthase (), and induction of VEGF receptor (VEGFR) proteolysis (). Our studies to dissect genetic networks initiated by the transcription factor GATA-1 in erythroid cells revealed that GATA-1 activates transcription of the murine preprotachykinin-B gene (; ), which encodes a neurokinin-B (NK-B) precursor protein (). The human orthologue is highly induced upon ex vivo differentiation of peripheral blood hematopoietic precursors (). is also expressed by neurons (), the uterus (), and syncytiotrophoblasts of the placenta (). Biological functions of tachykinins include smooth muscle contraction (), vasodilation (), neurotransmission (), neurogenic inflammation (), and immune system activation (). NK-B activates NK1, NK2, and NK3 G protein−coupled receptors (Shigemoto et al., 1989; ; ). As quantitative RT-PCR analysis did not reveal NK receptor expression in erythroid cells, erythroid cell-derived NK-B might act on neighboring cells within the hematopoietic and/or vascular microenvironments. NK receptors are expressed on mouse yolk sac endothelial cells (YSECs) and mouse aortic endothelial cells, and NK-B induces cAMP accumulation in these cells (). NK receptors are also expressed on endothelial cells of rat postcapillary venules (), human umbilical vein endothelial cells (HUVECs; ; ), and bovine corpus luteal endothelial cells (). These findings led us to hypothesize that NK-B targets endothelial cells, although the ensemble of physiological targets and the molecular and physiological consequences were unclear. Here, we demonstrate that NK-B blocks endothelial cell motility, represses VEGFR expression, and induces an antiangiogenic protein, all of which collectively oppose vascular network assembly. NK-B activity is greatly potentiated by thromboxane A2 (TXA2) signaling or phosphodiesterase inhibition, providing evidence for an antiangiogenic NK-B/TXA2 axis. Because endothelial cells express NK receptors (), we tested whether NK-B regulates endothelial cell proliferation, migration, and vascular network assembly. Endothelial cell growth supplement containing 2% fetal calf serum, FGF2, EGF, hydrocortisone, and heparin (see online supplemental material, available at ) stimulates YSEC and HUVEC proliferation. NK-B did not inhibit the proliferation response (). YSECs and HUVECs assemble extensive vascular networks on Matrigel containing the supplement. Real-time video microscopy revealed that NK-B treatment before culturing cells on Matrigel reduced the number of migrating cells (; Video 1, Vehicle; and Video 2, NK-B) and strongly reduced the motility of cells that remained competent to migrate (). Importantly, the rate and extent of vascular network assembly were suppressed by NK-B (). We tested another tachykinin, substance P (SP; ), which was reported to promote HUVEC vascular network assembly and angiogenesis (; ; ). SP did not affect supplement-induced YSEC and HUVEC network assembly (unpublished data). A consequence of NK receptor activation is elevation of intracellular cAMP (). We demonstrated that NK-B increases cAMP in YSECs in the presence of the phosphodiesterase inhibitor 3-isobutyl-1-methylxanthine (IBMX; ). IBMX increased the magnitude of the cAMP induction ∼1.7-fold and converted the transient cAMP induction into a sustained response (). As elevated cAMP impairs endothelial cell survival, motility, and angiogenesis (, ), we tested whether IBMX potentiates NK-B activity. Although IBMX or NK-B alone did not affect YSEC proliferation (), NK-B/IBMX suppressed proliferation (). NK-B/IBMX nearly ablated YSEC motility (; Video 3, IBMX; and Video 4, NK-B/IBMX, available at ). Because the motility blockade was prevented by a combination of NK1- and NK3-selective small molecule inhibitors ( and Video 5, NK-B/IBMX/NK inhibitors), NK receptor interactions appear to mediate the blockade. Given the importance of proliferation and motility for vascular network organization, we predicted that NK-B/IBMX would severely compromise network assembly. NK-B/IBMX abrogated YSEC network assembly on Matrigel (). The NK-B/IBMX−mediated induction of cAMP would be expected to activate PKA and downstream effectors. The PKA inhibitor H-89 prevented the NK-B/IBMX−mediated abrogation of vascular network assembly (), indicating that PKA is required for NK-B/IBMX actions. Replacement of NK-B/IBMX−containing medium with medium lacking NK-B/IBMX after 20 h restored network assembly by 36 h (Fig. S1, available at ), indicating the lack of cytotoxicity. Analogous to the IBMX potentiation of NK-B activity, we reasoned that physiological factors such as TXA2 that activate adenylyl cyclase functionally interact with NK-B. NK-B () and TXA2 are overexpressed in preeclampsia (). TXA2 mediates FGF-COX-2−stimulated angiogenesis () and inhibits VEGF- and FGF2-induced endothelial cell migration and vascular network formation (; ). The stable prostaglandin U46619 recapitulates endogenous TXA2 signaling (). Because IBMX potentiation is associated with sustained and increased magnitude of the cAMP induction, we asked whether U46619 similarly affects cAMP. YSECs and HUVECs express TXA2 receptors (Fig. S2, available at ). U46619 increased the magnitude of the cAMP induction, without sustaining the response (). NK-B and U46619 together (NK-B/U46619; Video 7) abrogated network assembly and motility, whereas U46619 alone (Video 6) had no effect (). The U46619 activity required TXA2 receptor activation, as the selective TXA2 receptor antagonist SQ29548 strongly reduced the NK-B/U46619 activity to block vascular network assembly (). Analogous to NK-B/IBMX, H-89 significantly reduced the NK-B/U46619 activity to abrogate network assembly (). Thus, TXA2 potentiates NK-B activity to oppose vascular remodeling. Endothelial cell subtypes have unique gene expression profiles, reflecting their distinct functional roles (; ). We tested if NK-B/IBMX regulates the capacity of human endothelial cells from the umbilical vein (HUVEC), aorta (human aortic endothelial cell [HAEC]), and neonatal dermis microvasculature (human microvascular endothelial cell [HMVEC]) to assemble vascular networks. NK-B/IBMX abrogated HUVEC and HAEC network assembly, whereas HMVEC vascular network assembly was unaffected (). The differential responsiveness correlates with NK receptor expression. HUVECs and HAECs, but not HMVECs, express NK receptors (). NK-B/IBMX induced cAMP accumulation in HUVECs and HAECs, but not in HMVECs (). Direct activation of adenylyl cyclase with forskolin induced cAMP similarly in the three cell types. In contrast to NK-B/IBMX, 8-Br cAMP and forskolin did not affect HUVEC vascular network assembly, whereas NK-B/forskolin and NK-B/ 8-Br cAMP abrogated network assembly (Fig. S3, A and B, available at ). These findings indicate that cAMP induction is necessary but insufficient for abrogating network assembly. Vascular network assembly in vitro reflects the capacity of endothelial cells to form blood vessels, but the milieu that regulates angiogenesis in vivo is considerably more complex due in part to the three dimensionality of the microenvironment (; ; ; ). An alternative assay uses collagen gels to recapitulate a three-dimensional microenvironment. NK-B/IBMX, but not IBMX, abrogated HUVEC vascular network assembly in collagen gels (Fig. S3 C). To test whether NK-B alone or NK-B/IBMX inhibits angiogenesis in vivo, we asked whether NK-B and NK-B/IBMX oppose FGF2-dependent angiogenesis in the chicken chorioallantoic membrane (CAM) assay. NK-B inhibited FGF2-mediated expansion of the microvasculature by ∼40%, and NK-B/IBMX almost completely blocked angiogenesis (). A mutant NK-B (mNK-B) inactive in the Matrigel assay (not depicted), in which the methionine residue at position 2 was substituted with the β amino acid analogue β-homomethionine, did not influence FGF2-induced angiogenesis with or without IBMX. Thus, the structural integrity of NK-B is crucial for antiangiogenic activity. If NK-B functions physiologically to oppose proangiogenic factors, blocking endogenous NK-B signaling with NK receptor inhibitors should promote angiogenesis. A combination of NK1- and NK3-selective inhibitors induced a strong angiogenic response equal to FGF2 (). Based on the inhibitor specificities (; ), these data provide evidence that endogenous neurokinin signaling suppresses endogenous angiogenic pathways. Furthermore, NK-B might function physiologically with other factors such as TXA2. Whereas U46619 alone lacked activity in the CAM assay, NK-B/U46619 suppressed angiogenesis, and this activity was antagonized by the TXA2 receptor antagonist SQ29548 (); SQ29548 was not active in the CAM assay (not depicted). Quantitative RT-PCR confirmed NK1 and NK3 expression in the CAM (). The NK receptor inhibitor sensitivity of the NK-B/IBMX−induced motility blockade suggested that the blockade requires NK receptor signaling. To genetically assess the role of NK1 and NK3 in vascular network assembly, NK1 and NK3 were knocked down individually via stable expression of short hairpin RNAs. The knockdowns almost completely ablated the respective mRNA (). Under conditions in which NK1 or NK3 were knocked down, the ability of NK-B/IBMX to block network assembly was reduced (). Both NK1 and NK3 are therefore required for NK-B to inhibit network assembly. The role of NK receptors in opposing vascular remodeling was further demonstrated with NK1- and NK3-selective inhibitors, which prevented NK-B/IBMX from abrogating network assembly (). Tachykinins induce phosphoinositide hydrolysis yielding metabolites that mobilize intracellular Ca (; ). Because dynamic changes in intracellular Ca have an important role in regulating motility () and differentiation (), we investigated whether NK-B alters intracellular Ca levels. The supplement induced Ca oscillations in YSECs, which were blocked by NK-B or NK-B/IBMX, but not mNK-B (). Forskolin did not affect the percentage of cells exhibiting oscillations (). NK-B–mediated ablation of Ca oscillations is therefore independent of the cAMP response. If the NK-B−dependent suppression of Ca oscillations is functionally important, we reasoned that an independent strategy to ablate Ca oscillations should mimic the NK-B activity to oppose vascular network assembly. The Ca chelator BAPTA ablated Ca oscillations () and significantly reduced network assembly (), consistent with the notion that suppression of Ca oscillations represents a functionally important component of the NK-B mechanism. FGF2 induces VEGF synthesis in certain contexts (), and both YSECs and HUVECs produce more VEGF when cultured in FGF2-containing medium (). We tested whether NK-B opposes FGF2-dependent angiogenesis by acting upstream or downstream of FGF2-induced VEGF production. NK-B alone, or in the presence of IBMX or U46619, prevented supplement-induced VEGF production (), indicating that NK-B acts upstream of FGF2-induced VEGF production. Furthermore, blocking Ca oscillations with BAPTA also inhibited VEGF production (, left), indicating that NK-B−mediated abrogation of Ca oscillations is important for reduced VEGF output. Because VEGF and FGF2 promote vascular network assembly (), we tested whether NK-B affects expression of Type I (VEGFR1; ) and type II (VEGFR2; ) VEGFRs and FGF receptor 1 (FGFR1; ). When starved HUVECs and YSECs were treated with supplement, and mRNA increased five- and twofold, respectively, by 6 h (). NK-B, NK-B/IBMX, and NK-B/U46619 inhibited mRNA induction. NK-B/IBMX and NK-B/U46619, but not NK-B alone, inhibited expression. Analysis of VEGFR1 and VEGFR2 protein expression validated the mRNA analysis (). and expression were also measured in HUVECs and YSECs plated on Matrigel. and mRNA increased considerably, and NK-B/IBMX strongly suppressed the induction (). induction in YSECs was not affected by NK-B/IBMX. The supplement did not induce expression in HUVECs. Because VEGF stimulates endothelial cell proliferation and motility (; ; ), VEGFR down-regulation should diminish proliferation, migration, and vascular network assembly. As VEGF induces VEGFR1 and VEGFR2 expression (; ) and NK-B suppresses VEGF production (), we tested whether NK-B/IBMX−mediated down-regulation of VEGFR1 and VEGFR2 is counteracted by elevating VEGF. VEGF induced and mRNA to 30–60% of the control level after 5 or 20 h in YSECs; FGF2 had no effect (Fig. S4 A, available at ). These results support an NK-B action upstream of FGF2-dependent VEGF production. FGF2 was active, as FGF2 induced FGFR1 expression (Fig. S4 A, right). Concomitant with reactivated VEGFR expression, network assembly was partially rescued (Fig. S4 B). These results indicate that a balance between NK-B/IBMX and VEGF signaling dictates the extent of network assembly. A proteomics screen was conducted to test if NK-B/IBMX regulates the levels of proteins linked to angiogenesis. Two-dimensional SDS-PAGE analysis identified four regulated proteins (). MALDI-TOF (matrix-assisted, laser-desorption-ionization/time of flight) mass spectroscopy identified the proteins as β-tubulin (1 and 2), -lactate dehydrogenase, and calreticulin, respectively. The Ca binding protein calreticulin, () which functions as a chaperone and extracellularly as a signaling molecule with thrombospondin () and integrins (), was verified by Western blotting (). Whereas NK-B alone and U46619 alone did not induce calreticulin, NK-B/U46619 increased calreticulin approximately twofold. The N-terminal 180 amino acids of calreticulin, termed vasostatin, were purified previously as an antiangiogenic factor (). Both calreticulin and vasostatin are antiangiogenic (; ). To assess whether YSEC and HUVEC vascular network assembly is responsive to calreticulin, calreticulin and vasostatin were overexpressed in as GST fusion proteins, and GST was liberated via site-specific proteolysis (). Purified calreticulin and vasostatin impaired network assembly (). These results support a model in which NK-B/IBMX−mediated induction of calreticulin, VEGF, VEGFR1, and VEGFR2 down-regulation, reduced cell proliferation, and reduced cell motility constitute a multicomponent antiangiogenic mechanism. Our results demonstrate NK-B targeting of endothelium via a multicomponent mechanism to oppose vascular remodeling. Not only is NK-B antiangiogenic, but disruption of endogenous neurokinin signaling stimulated angiogenesis comparable to FGF2. These results suggest a new mode of vascular regulation in which NK-B functions as an endogenous angiogenesis inhibitor. The TXA2 potentiation of NK-B activity exemplifies how NK-B can interact with other physiological factors to regulate endothelial cell function. Although a link between NK-B function and vascular regulation had not been established previously, SP can regulate endothelial cell function. SP promotes angiogenesis during acute neurogenic inflammation (), and expression of the NK receptor preferred by SP, NK1, is up-regulated in endothelial cells upon angiogenesis during chronic airway inflammation in rats (). These results indicate that NK receptor signaling by endothelial cells mediates vascular regulation in certain contexts, but the underlying mechanisms were not defined. In addition to regulating angiogenesis, NK receptor modulation alters vascular tone (). Synthetic NK receptor agonists induced hypertension in guinea pigs (), continuous infusion of NK-B in rats increased blood pressure (), and NK-B induced vasodilation of TXA2-constricted perfused placental cotyledon (). However, mechanisms underlying these activities are unknown. NK-B resembles certain known endogenous angiogenesis inhibitors () that suppress endothelial cell proliferation and migration. NK-B down-regulated VEGFR1 and VEGFR2 expression, and endostatin down-regulates VEGFR2 transcription approximately twofold (). Both endostatin and TIMP2 inhibit VEGFR2 phosphorylation (). We are unaware of reports in which endogenous angiogenesis inhibitors down-regulate VEGFR1 expression. Pigment epithelium-derived factor inhibits growth factor−induced angiogenesis in microvascular endothelial cells through a mechanism involving cleavage and intracellular translocation of the VEGFR1 transmembrane domain (). Another potentially important component of the NK-B mechanism is calreticulin induction. Other examples of an endogenous angiogenesis inhibitor inducing a distinct angiogenesis inhibitor include endostatin and fibulin 5 up-regulation of thrombospondin 1 and pigment epithelium-derived factor up-regulation of plasminogen kringle 5 (; ; ). Besides NK-B, angiogenesis inhibitors have not been reported to induce calreticulin expression. The molecular steps of the multicomponent mechanism segregate into extracellular (), intracellular signaling (), and effector () modules. Multiple lines of evidence indicate that NK-B−mediated cAMP induction is required for a subset of the NK-B activities. Inclusion of IBMX with NK-B further reduced YSEC motility () and vascular network assembly () and inhibited mRNA induction (). Forskolin resembled NK-B in inhibiting and mRNA induction (Fig. S4 C). H-89 blocked NK-B/IBMX−mediated abrogation of vascular network assembly (). Elevated cAMP can impair endothelial cell survival, motility, and angiogenesis (). VEGFR1 and VEGFR2 down-regulation would further suppress proliferation and motility, given the VEGF activity to stimulate proliferation and motility (; ; ). Although tachykinins can mobilize intracellular Ca, our results show that NK-B blocks growth factor−dependent Ca oscillations (). As intracellular Ca stimulates motility via calpain-dependent and -independent mechanisms (), the suppressive activity of NK-B on Ca oscillations should amplify the motility blockade established via cAMP induction. The finding that BAPTA mimics NK-B in ablating Ca oscillations and opposing vascular network assembly () reinforces the functional significance of the NK-B−mediated abrogation of Ca oscillations. Recombinant calreticulin and vasostatin, which are antiangiogenic in vivo (; ), suppressed vascular network assembly but only modestly inhibited motility (unpublished data). Vasostatin inhibits endothelial cell attachment to laminin (). It is attractive to propose that elevated calreticulin/vasostatin synergizes with the NK-B−instigated motility blockade to oppose vascular remodeling. Forskolin did not induce calreticulin (unpublished data), indicating that a distinct NK-B activity mediates calreticulin induction. The Ca ionophore ionomycin, which increases intracellular Ca, elevated calreticulin expression by approximately twofold (Fig. S4 D), suggesting that reduced intracellular Ca does not explain NK-B−mediated calreticulin induction. Because IBMX increased the magnitude and sustained the cAMP induction, whereas U46619 only increased the magnitude of the response, a sustained response is not required for potentiating NK-B activity. Considering that both NK-B and TXA2 are elevated in preeclampsia (; ), it is attractive to propose that NK-B targeting of vascular endothelium and TXA2 potentiation of NK-B activity have important implications for vascular deregulation in preeclampsia. As TXA2 is not likely the sole physiological factor that interacts with NK-B to oppose vascular remodeling in vivo, it will be important to identify additional factors that function combinatorially with NK-B and TXA2. Nevertheless, our demonstration that NK-B directly targets endothelium, that perturbation of endogenous neurokinin signaling is angiogenic, and that TXA2 potentiates NK-B activity and our elucidation of the NK-B signaling circuitry establish an NK-B/TXA2 axis as a new mode of vascular regulation. NK-B, NK1 receptor−selective antagonist L733060 ((2S, 3S)-3-(93,5-bis (trifluoromethyl) phenyl) methoxy)-2-phenylpiperidine hydrochloride), NK3 receptor−selective antagonist SB222200 ((S)-3 methyl-2-phenyl-N- (1-phenylpropyl)-4-quinilinecarboxamide), 8-BrcAMP, forskolin, and H-89 were purchased from Sigma-Aldrich. U46619 and IBMX were purchased from Calbiochem and were solubilized in DMSO. U46619 and H-89 were diluted in ethanol and 50% ethanol, respectively. SQ 29548 (Cayman Chemical) was dissolved in ethanol. The final concentrations of NK-B, IBMX, 8-Br cAMP, forskolin, H-89, U46619, and SQ 29548 in cell culture medium were 100, 100, 100, 10, 10, 1, and 2 μM, respectively. NK-B was maximally effective at 75 μM. The final concentrations of DMSO and ethanol did not exceed 1%. Recombinant bovine FGF2 (provided by A. Rapraeger, University of Wisconsin School of Medicine, Madison, WI) and mouse VEGF164 (R&D Systems) were reconstituted in PBS containing 1% BSA. An NK-B mutant peptide was produced via microwave-assisted solid- phase peptide synthesis (). Fmoc amino acids (Calbiochem-Novabiochem) were activated with HBTU/HOBt in dimethylformamide and coupled using microwave irradiation (600 W maximum power, 70°C, ramp 2 min, hold 2 min; MARS multimode microwave [CEM Corporation]). Removal of the Fmoc protecting group was accomplished by treatment with 20% piperidine in dimethylformamide with microwave irradiation (600 W maximum power, 80°C, ramp 2 min, hold 2 min). After cleavage from the solid support (NovaSyn TGR resin; Novabiochem) with trifluoroacetic acid, the crude peptide mixture was purified by reverse phase HPLC and structurally validated by MALDI-TOF mass spectrometry. Mouse YSECs were derived from a hypervascular transgenic mouse expressing the / protooncogene (). YSECs exhibit a normal endothelial phenotype and are not tumorigenic. HUVECs, HAECs, and HMVECs were purchased from Cascade Biologics, Inc. Cells were maintained as described in the online supplemental material (available at ). YSECs and HUVECs were grown in 48-well plates (1.5 × 10 cells/well) with supplement-free M200 medium to allow synchronization of cells in G/G. After 24 h, supplemented M200 medium containing either vehicle, NK-B, IBMX, or NK-B/IBMX was added and incubated at 37°C for another 24 h. The cell proliferation was quantified using CellTiter 96 AQueous One Solution reagent (Promega) according to the manufacturer's directions. The results from triplicate determinations (mean ± SD) are presented as the percent increase of proliferation compared with nonstimulated growth in supplement-free medium. Vascular network assembly was assessed by measuring the formation of capillary-like structures by endothelial cells on Matrigel (BD Biosciences). Matrigel was diluted 1:1 with supplement-free M200 medium, poured in 24-well plates, and allowed to solidify at 37°C. Subconfluent endothelial cells were harvested and preincubated under different experimental conditions (YSECs, 1 h; HUVECs, HAECs, and HMVECs, 2 h) in growth supplement−free M200 medium in microfuge tubes. An equal volume of supplemented medium containing the indicated reagents was added. Cells were plated on Matrigel (1.5 × 10 cells/well) and incubated at 37°C. Vascular network assembly was measured as a function of time, and digital pictures were captured with an inverted microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.; 10× objective, NA 0.25), using the Axiovision 3.1 software (Carl Zeiss MicroImaging, Inc.). To quantitate the vascular network assembly, digital pictures of several adjacent frames were taken, and the lengths of the tubelike structures were measured. Subconfluent YSECs and HUVECs were pretreated with vehicle, NK-B, IBMX, or IBMX/NK-B for 1 or 2 h, respectively, in supplement-free M200 medium, 5–10 × 10 cells were plated per well in a 12-well Matrigel-coated plate, and an equal volume of supplemented medium with or without various reagents was added. NK receptor inhibitors were added 30 min before adding NK-B. Cells were allowed to adhere for 30 min, and time-lapse images were acquired at 37°C with an inverted microscope (TE300; Nikon; 20× objective, NA 0.45), equipped with a CoolSnap fx charge-coupled device camera (Photometrics) and a cube temperature controller (Life Imaging Services). Images were captured with E-See Inovision Software (Inovision) and with Slidebook 4 software (Intelligent Imaging Innovations; one image per minute for 120 min). Cells migrating in a field were quantitated by putting a digital mark in the center of the cell body at time 0 and determining the migration of the cell body away from the digital mark after 2 h. Migrating cells were further validated by Z-stack projection in Image J. The migrating cells were expressed as a percentage of the total cells in each field. The rate of cell migration was quantitated using the Slidebook 4 software. The center of each migrating cell was marked digitally at time 0, and the distance migrated from the digital mark after 2 h was calculated. For each field, migration of at least 20 cells was analyzed and the mean distance migrated per cell was calculated. Real-time RT-PCR was conducted as described in the online supplemental material. Real time RT-PCR primers were designed using PRIMER EXPRESS 1.0 (Applied Biosystems) to amplify 50–150-bp amplicons and were based on GenBank or Ensembl sequences. Sequences are provided in Table S1 (available at ). Cells were cultured in 6-well plates (2 × 10 cells/well), washed with supplement-free M200 medium, and treated with the indicated reagents at 37°C. Reactions were terminated by aspirating medium. Cells were washed with PBS and lysed in 500 μl of 0.1 N HCl containing 0.5% Triton X-100. cAMP was assayed in cell lysates (100 μl) using a direct cAMP enzyme-linked immunosorbent assay (Assay Designs, Inc.). Fertilized chicken eggs were incubated at 39°C with 90% humidity. On embryonic day 7, a small hole was drilled at the end of the egg that contains the air sac to lower the embryo. A second hole was made above the CAM, and 200 μl of a 1.5% solution of methylcellulose (StemCell Technologies Inc.) containing vehicle, FGF2 (30 pmol)/IBMX (10 nmol), FGF2/NK-B (33 nmol) with or without IBMX, FGF2/mNK-B (33 nmol)/IBMX (mNK-B, inactive mutant of NK-B), a combination of NK1- (L733060, 10 nmol) and NK3- (SB222200, 4 nmol) selective inhibitors, FGF2/U46619 (10 nmol) with or without NK-B, or FGF2/NK- B/U46619 with SQ29548 (40 nmol) was applied to the CAM of day 7 chicken embryos. Vasculature was analyzed after 48 h, and digital images were captured at room temperature with a Wild M3Z dissecting microscope (40×; Leica) attached to a SPOT RT KE charge-coupled device camera, using the software SPOT 3.5.9 (Diagnostic Instruments). Cells were incubated for 30 min in 4 μM fluo-4/AM (Invitrogen) in the presence or absence of 4 μM BAPTA/AM (Invitrogen) in supplement-free M200 medium at 37°C, trypsinized, plated on Matrigel with indicated reagents, and allowed to adhere for 10 min. Time-lapse video microscopy with a FluoView 500 laser scanning confocal microscope (Olympus; 40×, NA 0.8 water-immersion objective) was used to measure intracellular Ca as described previously (). An argon laser at 488 nm was used to excite Fluo-4, and the emitted light was detected at wavelengths >510 nm. Images were acquired at 15-s intervals and analyzed using Image J. When indicated, the fluorescence intensity in the region of interest (F) was normalized by dividing fluorescence intensity of the resting cell at time 0 (F). To generate stably expressing siRNAs, oligonucleotides were designed according to the criteria specified by Dharmacon and Oligoengine and cloned into the BglII−HindIII sites of pSUPER Puro (Oligoengine). The NK1 target sequence 5′-CAACAGGACTTACGAGAAA-3′ corresponded to nucleotides 1480–1497 of the mouse NK1 cDNA; the NK3 target sequence 5′-AGATTTCGTGCAGGCTTCA-3′ corresponded to nucleotides 1044–1061 of the mouse NK3 cDNA. The empty vector was used as a negative control. YSECs were transfected with Lipofectin (Invitrogen), and positive clones were selected with 2 μg/ml puromycin. YSECs and HUVECs were cultured in 6-well plates (3 × 10 cells/well), incubated in supplement-free M200 medium overnight, and treated with vehicle, NK-B, IBMX, NK-B/IBMX, U46619, NK-B/U46619, or BAPTA/AM in supplement-containing medium for 24 h. Secreted VEGF was quantitated with mouse or human VEGF Quantikine ELISA kits (R&D Systems) according to the manufacturer's instructions. YSECs were treated with IBMX or NK-B/IBMX for 1 h in supplement-free M200 medium, followed by 10 min in supplemented medium. Cells were processed and proteins were analyzed as described in the online supplemental material. The antibodies used and methodology are described in the online supplemental material. For construction of the GST-calreticulin and -vasostatin fusion constructs, the coding regions for calreticulin and vasostatin were cloned as C-terminal translational fusions with the GST gene for expression in Purification of GST-calreticulin was achieved by lysis of the bacteria, followed by sonication, centrifugation, adjusting the pH of supernatants to pH 7.0, and mixing with preequilibrated Glutathione Sepharose 4B (GE Healthcare) in PBS containing 1.0% Triton X-100. After a 30-min incubation, beads were washed, bound protein was cleaved with factor Xa (Novagen), and liberated protein was separated from immobilized GST via centrifugation. Table S1 shows the forward and reverse primers for RT-PCR analysis. Videos 1–7 document YSEC migration on Matrigel after treatment with vehicle, NK-B, IBMX, NK-B/IBMX, NK-B/IBMX + NK1- and NK3-selective inhibitors, U46619, and NK-B/U46619. Fig. S1 shows the reversibility of the NK-B−mediated vascular network assembly blockade. Fig. S2 shows an analysis of TXA2 receptor expression in YSECs and HUVECs. Fig. S3 (A and B) shows that cAMP induction is insufficient to inhibit vascular network assembly, but cAMP potentiates NK-B−mediated inhibition of vascular network assembly. Fig. S3 C shows NK-B/IBMX−mediated abrogation of vascular network assembly in three-dimensional collagen gels. Fig. S4 (A and B) shows VEGF rescue of the NK-B−mediated suppression of VEGFR expression and vascular network assembly. Fig. S4 (C and D) shows that cAMP induction, but not the antiangiogenic proteins calreticulin and vasostatin, down-regulates VEGFR1 and VEGFR2 in YSECs. Online supplemental material is available at .

Embryonic stem (ES) cells have pluripotent capacity for unlimited growth and self-renewal and the ability to differentiate into mature cells, including vascular cell lineages. Recent studies have revealed the effects of ES cell–derived endothelial cells (ECs) on the formation of new vessels and improvements in cardiac function, implicating the potential of these cells to be used for progenitor cell–associated applications (). Differentiation of ES cells toward the endothelial lineage is an important step in vasculogenesis, in which signals initiated by vascular endothelial growth factor (VEGF), FGF, and other cytokines have been implicated. Most of our knowledge concerning endothelial differentiation from stem cells is derived from studies using growth factors and hypoxia environments, but less is known about the effect of mechanical forces on endothelial differentiation (; ; ). ECs are critical cellular components of blood vessels, functioning as selectively permeable barriers between blood and tissues. The denudation or dysfunction of the intact endothelial monolayer causes lipid accumulation, monocyte adhension, and inflammatory reactions that initiate atherosclerotic lesion development (; ). Recently, accumulating evidences indicate that endothelial progenitor cells (EPCs), possibly derived from adult stem cells, play an important role in endothelial repair (, ; ; ; ; ). During this process, incorporated EPCs are exposed to shear stress, which is a mechanical force generated by blood flow. Several studies have shown that EPCs can differentiate into endothelial phenotypes when shear stress is applied (; ; ; ). However, the underlying mechanism remains to be elucidated. The homeostasis between histone acetyltransferases and histone deacetylases (HDACs) regulates the structure and function of chromatin and some transcription factors, leading to gene transcription regulation. The HDAC family consists of 18 members, which are categorized into three classes (). HDAC3 belongs to class I HDACs, which contains four members, HDAC1, 2, 3, and 8, and is sensitive to trichostatin A (TSA). HDAC8 is reported to link with smooth muscle cell function (). As HDAC1–3 are found expressed in most cell types in vitro, their role in cell differentiation is neglected. The aim of the present study was to investigate how laminar shear-initiated signal pathways were involved in stem cell differentiation into ECs. We demonstrated that shear stress activated HDAC3 through the VEGF receptor 2 (Flk-1)–PI3K–Akt signal pathway, and that HDAC3-mediated p53 deacetylation and p21 activation was crucial for shear stress and VEGF-induced EC differentiation. ES- or stem cell antigen-1–positive (Sca1) cells (vascular progenitors) were cultured on collagen IV–coated plates in the absence of LIF, and then subjected to 12 dyne/cm laminar shear stress. As shown in , laminar flow increased the proliferation by ∼70% for both cells, as compared with static control. RT-PCR analysis revealed that withdrawal of LIF and culture on collagen IV–coated slides for 4 d increased mRNA expression of PECAM-1 (CD31), prominin 1 (CD133), VE-cadherin (CD144), VEGF receptor 1 (Flt-1), and VEGF receptor 2 (Flk-1) in both ES and Sca1 progenitor cells, which was further enhanced by shear stress (unpublished data). Data shown in indicate the increase of Flk-1 and endothelial nitric oxide synthase (eNOS) proteins by shear in ES cells. FACS analysis revealed that CD31 or CD133 cells increased slightly, whereas VCAM-1–positive (CD106) cells increased significantly, in response to shear stress (). These results suggest that laminar flow increases the ES cell differentiation toward ECs. To further explore whether the shear stress–induced, ES cell–derived ECs were functional, in vitro tube formation and in vivo neovascularization assays were performed. As shown in , sheared ES cells formed a tubelike structure on Matrigel, and no such structure was observed in the static control. After subcutaneous injection of cells mixed with Matrigel, the sheared cells formed capillary-like structures in all inoculations (6/6) at 1 and 2 wk (, shear), whereas only 1/6 of the cells from static group formed capillary-like structures at 2 wk, but 3/6 at 1 wk and 1/6 at 2 wk formed tumorlike structures (, static). To distinguish the exogenous ES cells from host cells, the ES cells were infected with the Ad-LacZ virus 12 h before injection. The contribution of shear-induced ECs to the formation of new blood vessels was shown in from X-gal staining. Because the p53–p21 pathway was reported in shear-induced cell cycle arrest in mature ECs (; ; ), we wondered whether such a pathway was involved in shear-induced EC differentiation. When laminar flow was applied, the p53 and p21 proteins increased gradually as shear proceeded (). Double immunostaining showed a colocalization of p53 or p21 with EC markers, e.g., CD106 () and CD144 (not depicted). Overexpression of p53 increased CD144- and eNOS-Luc reporter gene expression, but slightly decreased Flk-1–Luc reporter gene expression in ES cells, whereas overexpression of p21 increased expression of all three reporter genes (). To further confirm the impact of p53 and p21 in EC differentiation, we overexpressed p53 in ES cells via adenoviral gene transfer. FACS analysis showed that overexpression of p53 increased CD31, CD106, and CD144 cell numbers, but caused no change in CD133 cells (). Western blot analysis indicated that p53 overexpression led to the increase of CD31, CD144, eNOS, and p21 proteins in a dose-dependent manner (unpublished data). When cells were infected with adenovirus carrying p53 and injected into mice with Matrigel, all inoculates (6/6) showed vascular structures, and only 1/6 of controls displayed such a structure. Immunohistological staining revealed that all vascular structures were positive for p53, p21, and CD144 (). These results indicate that p53 up-regulation and the concomitant p21 activation can promote EC differentiation in vitro and in vivo. To explore whether p53 and p21 were essential for shear-induced EC differentiation, ES cells were transfected with p53 or p21 siRNA for 2 d, followed by 12 dyne/cm laminar flow treatment for 24 h. Western blot analysis showed that p53 siRNA treatment decreased shear-induced p53 and p21 protein production, and that shear-induced eNOS and Flt-1 protein levels were also decreased (). Meanwhile, p21 siRNA treatment decreased p21 proteins as expected, and shear-induced eNOS, Flk-1, and Flt-1 induction was ablated (). Shear-induced CD144 reporter gene expression was significantly decreased by treatment with either p53 or p21 siRNAs, although both siRNA treatments did not completely block CD144 reporter expression (unpublished data). As shown in , p53 acetylation at Lys317 and Lys370 sites was decreased, although p53 proteins were increased under shear, suggesting that laminar flow increases p53 deacetylation. To explore the role of HDACs in p53 deacetylation by shear, HDAC activity was detected in cell lysates from static and sheared samples. Refreshment of growth medium slightly increased HDAC activity in static cells, which may be caused by growth factor stimulation present in the serum, whereas application of laminar flow on ES cells resulted in a significant elevation in HDAC activity, with a peak at ∼6 h (). Western blot analysis revealed that HDAC1 and HDAC3 proteins were up-regulated by shear in a pattern similar to HDAC activity (), whereas other types of HDACs were undetectable or slightly decreased by shear (not depicted). As further studies revealed that HDAC1 was less relevant in EC differentiation, we focused our studies on HDAC3. Cotransfection assay showed that overexpression of HDAC3 could enhance p53-mediated p21-Luc reporter gene expression (). To further confirm the role of HDAC3 in p53 deacetylation and p21 activation and its potential role in shear-induced EC differentiation, the HDAC inhibitor TSA was included in shear experiments. TSA completely abolished shear-induced HDAC activation (). As expected, shear-induced HDAC3 activation and p53 deacetylation were ablated, as revealed by an increase in p53 acetylation. TSA treatment led to an increase in p21 and eNOS proteins in static cells; shear stress could not increase these proteins any more in the presence of TSA, indicating that shear stress does not cooperate with TSA in the enhancement of these gene expressions (). When ES cells were transfected with HDAC3 siRNA, shear-induced p53 deacetylation, p21 activation, and Flt-1 expression were significantly reduced, as demonstrated by Western blot analysis (). Luciferase reporter assay revealed that HDAC3 siRNA treatment slightly inhibited CD144 reporter gene expression under static conditions, but significantly reduced shear-induced CD144 reporter gene expression (unpublished data). Overexpression of p21 partially rescued HDAC3 siRNA-mediated suppression of Flt-1 reporter gene expression (). These results indicate that HDAC3-mediated p53 deacetylation and p21 activation are essential for shear-induced EC differentiation from ES cells. Western blot analysis demonstrated that both RNA and the protein synthesis inhibitors actinomycin D and cycloheximide decreased HDAC3 protein levels in static cells. When laminar flow was applied, HDAC3 increased by a similar fold (∼1.5-fold) in the absence or presence of the inhibitors (), implicating that shear stress up-regulates HDAC3 by posttranslation stabilization. To clarify the upstream signal pathways, we studied VEGF and related signal transducers. In the presence of VEGF antagonist SU1498, shear stress could not increase HDAC3 proteins further, and the HDAC activity was abolished (). Consequently, shear-induced p53 deacetylation, p21 activation, and eNOS expression were ablated (). These results indicate that VEGF receptor Flk-1 is crucial for shear-induced HDAC3 stabilization and EC differentiation. It has been reported that shear stress activated Flk-1 in a ligand-independent manner (; ). To explore whether ligand-dependent activation of Flk-1 could also contribute to HDAC3 stabilization, the effect of VEGF-165 on HDAC activity and HDAC3 proteins were detected in ES cells. VEGF-165 treatment transiently increased HDAC3 proteins () with concomitant increase of HDAC activity (not depicted). Furthermore, both laminar flow and VEGF-165 treatment resulted in Akt phosphorylation, but different effects were observed on ERK phosphorylation (). Flk-1 inhibitor SU1498 abolished both shear- and VEGF-induced HDAC3 stabilization and Akt phosphorylation. ERK Kinase (MAKK) inhibitor PD98059 inhibited VEGF-induced ERK phosphorylation, but no effect on shear and VEGF-induced HDAC3 stabilization and Akt phosphorylation, whereas PI3K inhibitor LY294002 inhibited both shear- and VEGF-induced HDAC3 stabilization and Akt phosphorylation. Similarly, HDAC3 was also demonstrated to be crucial for VEGF-induced EC differentiation, as HDAC3 siRNA ablated VEGF-induced EC marker gene reporter (eNOS-Luc, Flt-1-Luc, and vWF-Luc) expression in ES cells (), and high levels of HDAC3, p53, and p21 were detected in VEGF-induced blood vessels (). As described above, the HDAC3–p53–p21 pathway was essential for shear- and VEGF-induced ES cell differentiation toward endothelial lineage in vitro, and might function in the in vivo blood vessel formation derived from these cells. To explore whether such a pathway was also involved in angiogenesis during embryonic development, their expression was checked in blood vessels of 13.5 d –transgenic mice embryos. As shown in , blood vessels were positively stained for β-gal and CD31 in endothelium, whereas HDAC3 and p21 labeled the whole vessel wall. These results suggest that HDAC3 and p21 may be involved in vasculogenesis during embryo development. To explore whether the in vitro–produced EPCs could incorporate into the injured vessel and increase the repair in vivo, a femoral artery injury model was conducted in ApoE mice. Considering that in vitro–differentiated ECs directly from ES cells take up only a small portion of the cell population, and that other types of cells might interfere with the experiment, Sca1 progenitor cells were alternatively used. As shown in , β-gal–labeled exogenous VEGF-treated Sca1 cells incorporated into the injured vessel wall when checked on day 3, which also showed CD31-positive staining. The injured vessel was completely occluded by neointimal lesions 2 wk postoperatively (). However, local transfer of VEGF-treated Sca1 cells significantly reduced neointima lesions (). Statistically, the progenitor cell–treated group had neointimal lesions decreased by ∼30–40% (). To further explore whether shear stress–induced EPCs had similar function, Sca1 cells were selected from 24-h sheared ES cells. FACS analysis showed that Sca1 cells took up ∼15% of the population in sheared ES cells before sorting, and >95% after sorting. After expansion, the shear stress–induced Sca1 cells were used in similar femoral artery injury repair experiments. As expected, shear stress–induced Sca1 cells could also incorporate into the injured vessel () and reduced the neointima formation () to a lesser extent, as compared with VEGF-treated Sca1 cells (). These results demonstrate that the in vitro–differentiated EPCs and/or ECs can incorporate into injured artery and reduce neointimal lesions by increased repairing. Under specific stimulation, e.g., growth factors (), extracellular matrix (), mechanical forces (; ), and coculture with other cell types (; ), progenitor cells undergo specific lineage differentiation. In this study, we demonstrate that laminar flow enhances the differentiation of stem cells into functional ECs. We show for the first time that HDAC3 deacetylates p53, leading to p21 activation and resulting in differentiation of stem/progenitor cells into mature ECs. We also provide evidence that shear activates HDAC3 by posttranslation stabilization through Flk-1–PI3K–Akt signal pathways, and that HDAC3 and p21 may be involved in blood vessel formation in vivo. Our findings of mechanistic aspects for shear-induced stem cell differentiation toward ECs could significantly enhance our knowledge on stem cells involved in vasculogenesis, angiogenesis, and vascular repair. Cultured on collagen IV–coated plates in the absence of LIF for 3–4 d, ES cells could spontaneously differentiate into EPCs, some even into mature ECs, as identified by the up-regulation of EPC markers, e.g., Flk-1, CD31, CD133, eNOS, and the EC marker CD144. We propose that shear stress significantly enhances both processes, i.e., vascular progenitor differentiation into EPCs, and EPCs into ECs. Shear-produced EPCs could be derived from the differentiation of bipotential vascular progenitor cells. Supporting this notion is the fact that vascular progenitor cells can differentiate into both vascular smooth muscle cells and ECs in response to PDGF and VEGF treatment, respectively (). Previous studies showed that shear stress induced EC differentiation, while suppressing smooth muscle differentiation (). In our experiments, we also observed that smooth muscle differentiation was down-regulated by shear stress (unpublished data). It seems that shear-induced EC differentiation is at the cost of reduced smooth muscle differentiation. If this is also true for adult vessels in vivo, a better explanation could be offered as to why high-shear areas in the arterial wall have an integrative endothelium that is resistant to atherosclerosis. Thus, vascular progenitor differentiation toward ECs could be influenced by shear stress in vivo. VEGF receptor 2 (Flk-1) is a major progenitor cell marker for hematopoietic and endothelial lineage, expressed from hemangioblast to mature ECs (). Flk-1 mediates not only VEGF-induced EC differentiation but also VEGF-induced mature endothelial function, such as proliferation, migration, permeability, and cell survival (; ; ; ). It was reported that shear stress could activate Flk-1 in a ligand-independent manner (; ).We demonstrated that shear stress–induced EC differentiation was mediated through Flk-1, as the Flk-1–specific inhibitor SU1498 abolishes shear-induced eNOS expression. Activation of Flk-1 can trigger several downstream signal pathways, among which the PI3K–Akt pathway is defined to cell survival, whereas the PLC–ERK pathway is defined to proliferation (; ; ). In ES-derived progenitor cells, shear stress activated Akt phosphorylation but suppressed ERK phosphorylation, whereas VEGF activated both Akt and ERK phosphorylation. Moreover, we demonstrated for the first time that the PI3K–Akt pathway led to HDAC3 activation. Akt phosphorylation status correlated with HDAC3 protein levels, and its inhibition by SU1498 and LY294002 abolished shear-induced HDAC3 protein increase. Further investigation of how Akt phosphorylation modulates HDAC3 is necessary. Previous studies reported that inhibition of HDAC activity by TSA blocked tumor angiogenesis, especially hypoxia-induced angiogenesis (; ). It is well known that hypoxia induces tumor angiogenesis and cell survival through the up-regulation of VEGF expression in tumor cells (; ). These studies provide indirect evidence for the potential role of HDACs in EC differentiation. Although reported that shear-activated HDAC1 and HDAC3 were involved in shear-induced adult EPC differentiation through up-regulation of HoxA9, we demonstrated that HDAC3 activation led to p53 deacetylation and, in turn, to p21 activation. In this study, HDAC3 was found to be activated by shear stress and VEGF, whereas inhibition of HDAC activity by TSA or siRNA-mediated knockdown of HDAC3 abolished shear- and VEGF-induced EC marker gene expression. Collectively, these results suggest an essential role of HDAC3 in shear- and VEGF-induced stem cell differentiation in vitro, and a potential role in blood vessel formation in vivo. Evidence indicates that lack of p21 expression increases EPC sensitivity to apoptosis, with retarded EC maturation (; ). In this study, we demonstrated that shear stress activated p21 through p53 deacetylation mediated by HDAC3, and that p21 activation was essential for shear-induced EC differentiation. p53 knockdown ablated shear-induced p21 activation, indicating that p21 activation by shear is p53-dependent. Both inhibition of HDAC activity by TSA and HDAC3 siRNA decreased shear-induced p53 deacetylation and p21 activation, implicating that p21 activation is also HDAC3 dependent. Other evidence came from the observations that overexpression of HDAC3 could enhance p53-mediated p21 expression, and high levels of HDAC3 and p21 were detected in blood vessels formed from in vitro–differentiated ECs and during embryogenesis. Other studies showed that p53 acetylation was linked to p21 activation (; ). The discrepancy may be caused by different cell types. In response to different stimuli, native p53 and acetylated p53 recruit different coactivators to form a complex binding to p21 promoter. We show that HDAC3, p53, and p21 are directly linked to EC differentiation in vitro. These findings suggest that p21 is downstream of HDAC3 and plays a central role in shear-induced EC differentiation. In this study, we also observed that in vitro–differentiated EPCs/ECs could form blood vessels in vivo, and appear to play a crucial role in endothelial restoration and vascular repair. Treatment with these vascular progenitors led to a significant reduction of neointimal lesions in a mouse model of femoral artery injury, through replacement of the denuded endothelium. Newly isolated Sca1 progenitor cells were less effective than VEGF-treated ones. The discrepancy was derived from the observation that vascular progenitor cells were able to differentiate toward both smooth muscle cell and EC, whereas VEGF induced specific differentiation toward endothelial lineage (). Indeed, VEGF-treated cells showed much more EC marker expression. The functionally intact endothelial monolayer suppresses the recruitment of inflammatory cells and, thus, modulates the process of vascular remodeling and down-regulates intima hyperplasia. These observations provide further support to the concept that differentiation of stem cells toward ECs is beneficial to the blood vessel. In summary, this study demonstrates that shear stress activates Flk-1 and its downstream PI3K–Akt cascade, resulting in the activation of HDAC3, which in turn deacetylates p53 and activates p21. p21 activation results in stem cell differentiation into ECs and an increase in their survival. HDAC3 may also be involved in epigenetic modification of the chromatin to regulate gene transcription, which is essential for modulation of EC marker gene expression, together, leading to stem cell differentiation. Thus, HDAC3 is essential for shear- and VEGF-mediated stem cell differentiation, which can be a new target for therapeutic use. Antibodies against CD31 (rat), Flk-1, and Flt-1 were purchased from Abcam; antibodies against CD106 (rat) and CD144 (rat) were obtained from BD Biosciences; antibody against CD133 (rat) was purchased from eBioscience; antibodies against β-gal, CD31 (goat), CD144 (mouse), HDAC1, p21, p53, p-AktThr308, Akt1/2 (goat), pERK42/44, and ERK2 were obtained from Santa Cruz Biotechnology, Inc.; antibodies against eNOS, HDAC3, and α-tubulin (mouse) were purchased from Sigma-Aldrich; and antibodies against acetylated p53Lys320 and p53Lys373 were obtained from Millipore. All antibodies were raised in rabbit, except those indicated. All secondary antibodies were purchased from DakoCytomation. The inhibitors actinomycin D (1 μg/ml), cycloheximide (30 μg/ml), LY294002 (5 μM), PD98059 (5 μM), SU1498 (1 μM), and TSA (50 nM) were purchased from Sigma-Aldrich, dissolved in DMSO, and used at the indicated concentrations. The cells were pretreated with a specific inhibitor for 1 h before further treatment in the presence of the inhibitor. Other chemicals were also purchased from Sigma-Aldrich. Mouse ES cells (ES-D3 cell line, CRL-1934; American Type Culture Collection [ATCC]) were maintained as previously described (). In brief, the ES cells were cultured in gelatin-coated flasks in DME (ATCC) supplemented with 10% FBS (ATCC), 10 ng/ml LIF (CHEMICON International, Inc.), 0.1 mM 2-mercaptoethanol, 100 U/ml penicillin, and 100 μg/ml streptomycin in a humidified incubator supplemented with 5% CO, and they were split at a 1:6 ratio every other day. Sca1 cells were isolated as previously described (), and maintained in the same condition as ES cells. Both ES and Sca1 cells were used for <20 passages in this study. For differentiation, ES or Sca1 cells were predifferentiated by culture on mouse collagen IV (5 μg/ml)–coated glass slides or flasks or plates in differentiation medium (DM; α-MEM medium [Invitrogen] supplemented with 10% FBS [Invitrogen], 0.05 mM 2-mecaptoethanol, 100 U/ml penicillin, and 100 μg/ml streptomycin) for 3 or 4 d before further treatment. The medium was refreshed every other day. Shear experiments were conducted as previously described (). In brief, the predifferentiated cells were exposed to a laminar flow generated by the pressure difference between the upper and lower reservoirs, with the effluent DM medium circulated back to the upper reservoir through a peristaltic pump. The shear stress, which is determined by the flow rate and the channel dimensions, was 12 dyne/cm, which is comparable with the physiological range in human major arteries. The entire flow system was placed in an incubator at 37°C and supplemented with 5% CO to maintain pH. Static controls were cells cultured on slides not exposed to flow. To prevent contamination, 5 μg/ml gentamycin and 0.1 μg/ml doxycycline were included in shear medium. For VEGF-induced HDAC activation assay, the predifferentiated ES cells were pretreated with serum-free α-MEM supplemented with 0.05 mM 2-mercaptoethanol and antibiotics for 2 h, followed by treatment with 10 ng/ml mouse VEGF-165 (Bender Medsystems) for the time indicated in . For HDAC3 siRNA assay, the predifferentiated ES cells were transfected with HDAC3 siRNA or control siRNA for 2 d, followed by 10 ng/ml VEGF-165 treatment in serum-free α-MEM supplemented with 0.05 mM 2-mercaptoethanol, 1% BSA, 10 ng/ml insulin, and antibiotics for 24 h. For VEGF-induced differentiation, Sca1 cells were cultured in DM medium in the presence of 10 ng/ml VEGF-165 for 1 wk, followed by either an in vivo angiogenesis assay or a femoral injury repair experiment. A portion of the 24-h sheared or static cells was seeded in 12-well plates in the presence of MTT reagents (Promega) for 1 h, and the cell proliferation was detected according to the protocol provided. The procedure used for flow cytometry was similar to that previously described (). Rat anti-CD31, CD106, CD133, and CD144 antibodies were used for EC markers. The procedure used was similar to that described previously (). Cell suspension containing 4 × 10 static or sheared ES cells was placed on top of the 50 μl/well Matrigel (10 mg/ml; Becton Dickinson) in 8-well chamber slides (Nunc). Rearrangement of cells and the formation of capillary-like structures were observed at 18–24 h. Cells were fixed with 4% paraformaldehyde in PBS at 4°C overnight, and then HE staining was performed. Images were assessed with Axioplan 2 imaging microscope with Plan-NEOFLUAR 10×, NA 0.3, objective lenses, AxioCam camera, and Axiovision software (all Carl Zeiss MicroImaging, Inc.) at room temperature, and were processed with Photoshop software (Adobe). All cells were labeled with the Ad-LacZ virus (MOI = 20) before in vivo angiogenesis assay to distinguish the in vitro–differentiated cells from the host cells. 10 cells were pelleted and mixed with 50 μl Matrigel, and then injected subcutaneously into the back or flank of C57BL/6 mice. Six injections were conducted for each group. The mice were killed on days 7 and 14, and the plaques were harvested and frozen immediately in liquid nitrogen, followed by embedding with OCT and sectioning. 13.5-d embryos were collected from /ApoE mice () and cryosectioned. For artery injury model, both the left and right femoral arteries of ApoE mice were injured by wire. Ad-LacZ virus–labeled Sca1 cells in 30 μl PBS were injected into one artery, 30 μl PBS was injected into the other side as a control, and both were incubated for 30 min, followed by restoring the blood flow. Arteries were harvested at day 3 for cryosections or day 14 for formalin-fixed sections. The sections were stained with HE or X-gal (), or immunohistologically stained with specific antibodies, and observed under the microscope. Images were assessed and processed as described above except 20×, NA 0.5, and 40×, NA 0.75, objective lenses were used instead. All animal experiments were performed according to protocols approved by the Institutional Committee for Use and Care of Laboratory Animals. Cells were harvested and lysed by sonication. 50 μg of whole cell lysate was subjected to standard immunoblotting procedure and probed with primary antibodies as indicated. The bound primary antibody was detected by use of HRP-conjugated secondary antibody and the ECL detection system (GE Healthcare). The procedure used for immunofluorescence staining was similar to that described previously (). Rabbit anti–β-gal, p53, and p21, and rat anti-CD31, CD106, and CD144 antibodies were used. Swine anti–rabbit IgG-FITC and swine anti–rat IgG–TRITC antibodies were used to label the bound primary antibodies, which were illuminated by blue and green light, respectively. Images were assessed and processed as described in the previous paragraph, except that 40×, NA 0.75, objective lenses were used instead. HDAC activity was determined by using the Colorimetric HDAC Activity Assay kit (BioVision) with the protocol provided. The cells were lysed by sonication, and 50 μg of whole cell lysate was used to detect HDAC activity; relative HDAC activity was defined as A405 nm/μg protein, with that of control set as 1.0. The expression vectors for HDAC3, p21, and p53, and the reporter systems for CD144, eNOS, Flk-1, Flt-1, and vWF were cloned with vectors and primers in . For transient transfection, ES cells were cultured in a collagen IV–coated 12-well plate for 3 d, and then transfected with reporter gene (0.33 μg/well) together with expression plasmids (0.33 μg/well) and/or siRNA (0.033 nmol/well) using Fugene 6 reagent (Invitrogen) according to the protocol provided. pShuttle2-LacZ plasmid (0.15 μg/well) was included in all transfection assays as internal control, and pShuttle2 vector and control siRNA were used as a mock control. Luciferase activity assay was performed at 36 or 72 h (siRNA) after transfection. Luciferase and β-galactosidase activities were detected with standard protocol. Relative luciferase unit was defined as the ration of luciferase activity to β-galactosidase activity, with that of controls set as 1.0. The siRNAs for p53(sc-29436) and p21(sc-29428) were purchased from Santa Cruz Biotechnology, Inc. The control siRNA and the siRNA for HDAC3 (5′-ccucaucgccuggcauugatt-3′ and 5′-ucaaugccaggcgaugaggtt-3′) were purchased or synthesized from Ambion. For the siRNA knockdown experiments, ES cells were cultured on collagen IV–coated slides (38 × 76 mm) for 3 d, and the medium was refreshed at 24 h and 1 h before transfection. 10 μl of 10 mM siRNA per slide was introduced into the cells with siIMPORTER transfection reagents (Millipore), according to the manufacture's protocol. The untreated and nonrelated siRNA-transfected cells were included as controls. For specific target genes, siRNA transfection was performed in duplicate. The transfected cells were cultured for an additional 48 h and subjected to 12 dyne/cm laminar flow for 24 h, followed by Western blot analysis. Data expressed as the mean ± the SEM were analyzed with a two-tailed test, for two groups, or pair-wise comparisons. A value of P < 0.05 was considered to be significant.

At the ultrastructural level, the mature postsynaptic density (PSD) represents a weblike structure composed of filamentous components that hold together the particulate components (; ). The particulate components represent membrane-associated guanylate kinase proteins and signaling molecules, such as the Ca/calmodulin-dependent protein kinase II (CaMKII), whereas the filamentous components are formed by actin and spectrin filaments. The assembly of these components appears to be important for the formation and maintenance of the highly organized PSD. In search of the molecular mechanisms underlying this activity-dependent organization, we show that it is influenced by the neural cell adhesion molecule (NCAM) that associates with the membrane–cytoskeleton linker protein spectrin, which is highly enriched in PSDs (; ). The spectrin scaffold is thought to accumulate functionally interactive proteins in many cell types (). In neurons, spectrin binds to the cytoplasmic domains of -methyl--aspartate (NMDA) receptor subunits () and associates with other cytoskeletal structures, protein kinases, and phosphatases enriched in synapses (; ; ). Spectrin may, thus, serve to accumulate the postsynaptic machinery in PSDs. The mechanisms regulating this accumulation have, however, remained largely unknown. Spectrin interacts with the intracellular domains of the two major transmembrane isoforms of NCAM, with molecular masses of 140 (NCAM140) and 180 kD (NCAM180; , ; ; ), and may be indirectly linked to the 120-kD GPI-anchored NCAM isoform (NCAM120) via lipid rafts (). NCAM180, which is the most potent spectrin-binding isoform (), is enriched in PSDs (; ). NCAM is recruited to axodendritic contacts within minutes after initial contact formation and promotes stabilization of the newly formed contacts (, ). Reintroduction of NCAM into neurons of NCAM-deficient (NCAM−/−) mice stimulates synapse formation with NCAM-expressing neurons in an NMDA receptor–dependent manner (). Moreover, the NMDA receptor–dependent form of long-term potentiation (LTP) is impaired in NCAM−/− mice (; ) or after application of NCAM antibodies (). In humans, genetic variations in the NCAM gene have been reported to confer risk factors associated with bipolar affective disorders (), and abnormal levels of NCAM have been found in the brains of patients with schizophrenia and bipolar disorders (, ). Again, the molecular mechanisms underlying NCAM-mediated synaptogenesis and synaptic plasticity are not known. We show that NCAM promotes accumulation of spectrin in PSDs, and thus assembles the spectrin-based postsynaptic scaffold, recruiting NMDA receptors and CaMKIIα. Assembly of these components is impaired in NCAM−/− mice and reduced in NCAM+/+ synapses by a dominant-negative spectrin fragment containing the NCAM-binding site or when βI spectrin expression is reduced by using siRNA technology. Reduced accumulation of spectrin and NMDA receptors in NCAM−/− PSDs results in an inability to recruit CaMKIIα to synapses in an activity-dependent manner that is required for longer lasting changes in synaptic strength. The combined observations indicate that NCAM promotes assembly and maintenance of the spectrin-based postsynaptic-signaling complex, and thus reveal a hitherto unknown mechanism that contributes to the organization of the PSD in excitatory synapses. Because NCAMs accumulate at interneuronal contacts before their transformation into synapses, we first analyzed whether NCAM recruits spectrin to these contacts, thus transforming them into nascent synapses (). The distribution of spectrin was analyzed in cultured hippocampal neurons along thick, tapering MAP2-positive neurites, identified as dendrites, in the vicinity of contacts formed on the dendrites by thin neurites of a uniform diameter with multiple varicosities, identified as axons. In a separate set of experiments, we found that these thin neurites were positive for the established axonal markers, such as tau, L1, and synaptophysin (). In 4-d-old cultures, synaptophysin had not yet started to accumulate at contact sites (unpublished data), indicating that the immature axodendritic contacts were analyzed before elaboration of the presynaptic vesicle recycling machinery. In NCAM+/+ neurons, contacts accumulated spectrin. The average distribution of spectrin at these contacts was bell-shaped, with a peak in the center of the contact. In contrast, in NCAM−/− neurons, accumulation of spectrin at contacts was reduced. MAP2 was not accumulated at contacts of neurons from NCAM+/+ or NCAM−/− mice and was uniformly distributed in their vicinity. The expression of spectrin has been shown to be decreased in adult NCAM−/− brains (), a phenomenon that may at least partially explain the reduced accumulation of spectrin in contact sites between NCAM−/− neurons. To investigate whether NCAM recruits spectrin to contact sites, we analyzed the ratio (designated hereafter as ) of spectrin-labeling intensities at the center of a contact and 2.5 μm away from the contact center, a parameter that describes the extent of spectrin accumulation at the contact site and that is independent of spectrin expression levels. was reduced at contacts between NCAM−/− neurons (). Moreover, the percentage of contacts accumulating spectrin and defined as contacts with > 1.5 was also reduced in NCAM−/− neurons (63.5% in NCAM+/+ neurons and 34.3% in NCAM−/− neurons), indicating that NCAM is required to recruit spectrin to axodendritic contacts. To confirm this and analyze whether homophilic transinteractions of NCAM are important for spectrin accumulation at axodendritic contacts, we analyzed contacts formed between NCAM+/+ and NCAM−/− neurons in coculture (; ). The distribution of spectrin was then analyzed along dendrites of NCAM+/+ neurons in the vicinity of contacts formed by NCAM−/− axons or along dendrites of NCAM−/− neurons in the vicinity of contacts formed by NCAM+/+ axons. For this type of contact, which we hereafter call heterogenotypic contacts, only transinteractions of NCAM, but not spectrin protein levels, were affected in NCAM+/+ neurons. . Only 32% of the heterogenotypic contacts formed by NCAM+/+ axons contacting NCAM−/− dendrites and 25.4% of the heterogenotypic contacts formed by NCAM−/− axons contacting NCAM+/+ dendrites showed > 1.5, indicating that NCAM homophilic interactions are required for recruitment of spectrin to axodendritic contacts. Because spectrin is a potent scaffolding protein (), and because NCAM promotes accumulation of spectrin at axodendritic contacts, we analyzed whether clustering of NCAM at the cell surface also redistributes other postsynaptic components into the NCAM-associated molecular scaffold (). Thus, we clustered NCAM with antibodies against its extracellular domain at the surface of hippocampal neurons maintained in culture for 4 d, i.e., before synapse formation. To analyze whether proteins of interest were indeed integrated in a scaffold with NCAM, we treated neurons with 1% Triton X-100, thereby removing detergent-soluble molecules, but sparing cytoskeletal scaffolds (). Spectrin redistributed to NCAM clusters and formed detergent-insoluble complexes with NCAM, as indicated by an increase in the correlation between distributions of the two proteins when neurons treated with NCAM antibodies were compared with neurons treated with nonspecific immunoglobulins (). Interestingly, other postsynaptic scaffolding proteins, such as PSD95 and α-actinin, which are located proximal to the plasma membrane, and Shank, which is located more on the cytoplasmic face of the PSD (), also redistributed to detergent-insoluble clusters of NCAM, suggesting that NCAM not only clusters PSD components, but does so in an organized fashion, generating a structure that resembles the actual PSD. Strikingly, CaMKIIα, which is an enzyme enriched in PSDs, was also redistributed. Clustering of NCAM not only induced redistribution but also increased levels of detergent-insoluble spectrin, CaMKIIα, PSD95, Shank, and α-actinin, indicating that these proteins are recruited from the soluble pool and integrated in the NCAM-assembled scaffold (). In contrast, neither distribution nor levels of detergent-insoluble actin, tubulin, and MAP2 were affected by clustering of NCAM (). Redistribution to NCAM clusters does not imply that proteins form a tight molecular complex with NCAM, but may reflect that proteins coredistribute to an NCAM-assembled scaffold in a more indirect manner. To analyze which postsynaptic proteins form a firmer complex with NCAM, we immunoprecipitated NCAM from detergent-solubilized brain homogenates and analyzed the immunoprecipitates with antibodies against postsynaptic components. Spectrin and CaMKIIα, but not PSD95 and α-actinin, coimmunoprecipitated with NCAM (). The ability of NCAM to redistribute PSD95 and α-actinin in cultured neurons, thus, suggests that although these proteins may not be in as tight a complex with NCAM as spectrin and CaMKIIα, PSD95 and α-actinin may relate to NCAM-associated proteins. PDZ domains have been shown to bind to spectrinlike motifs in α-actinin-2 (), and indeed, when we immunoprecipitated spectrin from brain homogenates, PSD95 was coimmunoprecipitated (). We also analyzed whether postsynaptic glutamate receptors associate with NCAM. The NMDA receptor subunits NR1 and NR2B, but not the AMPA receptor subunits GluR1 and GluR2/3, coimmunoprecipitated with NCAM, and NCAM coimmunoprecipitated with NR1 (). NR1 and NR2B also coclustered with NCAM at the surface of cultured hippocampal neurons (Fig. S1, available at ). Because PSD95 and α-actinin did not coimmunoprecipitate with NCAM, these results also indicate that NMDA receptors and CaMKIIα form a firmer complex with NCAM, most likely independent of PSD95 and α-actinin. Association of NCAM with postsynaptic components was developmentally regulated; spectrin and NMDA receptors were already associated with NCAM in detergent-solubilized homogenates of brains from 1-d-old mice (the earliest age tested), whereas only low levels of CaMKIIα were found in NCAM immunoprecipitates in brains of mice of this age (). The levels of CaMKIIα in NCAM immunoprecipitates were significantly increased in adult brains reflecting the increased expression levels of CaMKIIα (unpublished data). In the adult brain, 27 ± 9.5% of total NR1 protein and 23 ± 13% of total CaMKIIα protein were associated with NCAM. These values may represent an underestimation of the actual levels because only ∼80% of NCAM, 60% of NR1, and 70% of CaMKIIα were solubilized under the detergent lysis conditions used (Fig. S2, available at ). In adult brains, the majority of NMDA receptors was associated with NCAM180, which is the predominant spectrin-binding isoform localized postsynaptically (). In brains of 1- and 7-d-old mice, NCAM140 and the predominantly glial NCAM120 isoform were also detected in the NR1 immunoprecipitates. The presence of NCAM120 in the immunoprecipitates may be caused by the presence of NMDA receptors in glia (). The highest number of NCAM–NR1 complexes was found at postnatal day 7, the time of most active overall synaptogenesis. To analyze the role of NCAM in the accumulation of PSD components in synapses in intact brains, we compared levels of spectrin, NR1, and CaMKIIα in synapses in the CA1 region of hippocampus of NCAM+/+ and NCAM−/− mice by postembedding immunogold labeling (). In accordance with previous reports (; ), spectrin was accumulated in PSDs. Spectrin labeling was often associated with filamentous structures extending from the PSD (). In NCAM−/− mice, fewer synapses were labeled for spectrin, NR1, or CaMKIIα. Moreover, spectrin, NR1, and CaMKIIα-positive synapses in NCAM−/− mice contained fewer immunogold particles when compared with NCAM+/+ synapses. In contrast, synaptic accumulation of the GluR1 subunit of the AMPA receptor was not influenced in NCAM−/− brains. To study whether reduced accumulation of spectrin, NMDA receptors, and CaMKIIα in NCAM−/− synapses may correlate with structural abnormalities of PSDs, we analyzed some ultrastructural features of PSDs in the CA1 region of the hippocampus; the length and thickness of PSDs in NCAM−/− brains were reduced by ∼10 and 15%, respectively, when compared with NCAM+/+ mice (215.9 ± 3.1 nm and 24.0 ± 0.2 nm in NCAM+/+ vs. 193.9 ± 3.3 nm and 20.9 ± 0.2 nm in NCAM−/− mice; ). Assuming that the PSD is a disk, these apparently small changes would, however, yield a 20 and 30% decrease in the surface and volume of PSDs, respectively. To extend this analysis, we isolated membrane fractions from adult NCAM+/+ and NCAM−/− mouse brains, which were then used to isolate PSDs (), and compared levels of NCAM-associated postsynaptic proteins in these fractions (). Spectrin levels were reduced in membrane and PSD fractions from NCAM−/− brains, confirming the role of NCAM in recruiting spectrin to membrane subdomains. Levels of NR1 and NR2B were reduced in the PSD fraction and increased in the membrane fraction isolated from NCAM−/− brains, suggesting that these transmembrane proteins are not efficiently recruited to PSDs in the absence of NCAM as a synapse-targeting cue and redistributed to other membrane domains. Interestingly, NR1 levels were increased in NCAM−/− brain homogenates (Fig. S3, available at ), indicating that the overall expression of NMDA receptors in NCAM−/− brains is increased, possibly as a compensatory reaction to its inefficient synaptic targeting. Levels of CaMKIIα, PSD95, and α-actinin were also reduced in PSDs from NCAM−/− mice, further confirming the relationship between NCAM and these postsynaptic components. In contrast, levels of GluR1, GluR2/3, actin, tubulin, and the postsynaptic scaffolding protein of inhibitory synapses, gephyrin, were similar in PSDs and membrane fractions from NCAM+/+ and NCAM−/− mice (). As a scaffolding protein, spectrin has been shown to cross- link NCAM with transmembrane and cytosolic proteins (; ). To characterize the role of spectrin in complex formation between NCAM and the NMDA receptor and CaMKIIα, we analyzed the formation of these complexes in the presence of a βI spectrin fragment consisting of the 2 and 3 spectrin repeats (βI). This fragment contains the NCAM-binding site and disrupts the association of NCAM with endogenous spectrin by a dominant-negative mechanism (). CHO cells or cultured cortical neurons were therefore cotransfected with βI, NR1 carrying a HA tag, and NR2B carrying a Flag tag. CHO cells, which do not express NCAM, were additionally cotransfected with NCAM180. In cotransfected CHO cells, NMDA receptors were detected at the cell surface, where they colocalized with NCAM (Fig. S4, available at ). For control, we used GFP or fragments containing the N-terminal domain of βI spectrin (βI), the N-terminal and 1–2 spectrin repeats (βI), or the 3–5 spectrin repeats (βI) for transfection. Whereas all transfected proteins were expressed in approximately equal amounts, as measured by Western blot analysis in cell lysates, coimmunoprecipitation of NMDA receptors and CaMKIIα with NCAM was reduced by βI but was not affected by βI or βI in CHO cells or in cortical neurons (). Similar results were obtained when CHO cells were cotransfected with NCAM140 (unpublished data). However, NCAM180 was more potent in precipitating the NMDA receptor subunits and CaMKIIα when compared with NCAM140, correlating with their ability to bind spectrin (). Interestingly, the βI spectrin fragment also blocked the association between NCAM and NR1, but did not affect binding of NCAM to spectrin (), CaMKIIα, or NR2B, suggesting that this fragment contains the binding site for NR1. To verify this, NR1 was immunoprecipitated from CHO cells cotransfected with different βI spectrin fragments and assayed for its ability to bind βI spectrin fragments. Only βI coimmunoprecipitated with NR1, indicating that this spectrin fragment contains the binding site for NR1 (). Because in these experiments the NR1 splice variant that was transfected into CHO cells without its heterodimeric partner NR2B was largely retained in the ER (), our data suggest that the interaction between NMDA receptors and spectrin already occurs in this intracellular compartment. Coimmunoprecipitation of NR1 with spectrin was not affected in NCAM−/− brains, indicating that NCAM is not required for this interaction (). To further confirm that spectrin is necessary for the formation of the NCAM–NMDA receptor complex, CHO cells or cultured hippocampal neurons were cotransfected with βI spectrin siRNA. This reduction of spectrin expression abolished the interaction of NMDA receptors and CaMKIIα with NCAM180 in cotransfected CHO cells, and disrupted the complex of endogenous NCAM and NMDA receptors in hippocampal neurons (). In hippocampal neurons maintained in culture for 14 d, NCAM180 accumulated in PSDs (), as previously observed in brain (; ), indicating that cultured neurons are relevant for further analysis of the role of spectrin in NCAM-mediated recruitment of postsynaptic proteins to PSDs. To this aim, we transfected hippocampal neurons with the βI spectrin fragment and analyzed accumulation of postsynaptic components in synapses (). To distinguish transfected neurons, they were cotransfected with GFP. In neurons transfected with βI, postsynaptic accumulation of NR1 and PSD95 was reduced when compared with GFP-only–transfected neurons, whereas presynaptic synaptophysin accumulations were not altered. We also found a small reduction in synaptic targeting of GluR1 in βI-transfected neurons. Similar results were obtained when MAP2-GFP was used to identify dendrites of transfected neurons (unpublished data). Transfection with the βI spectrin fragment did not affect synaptic targeting of the NMDA receptors (unpublished data). Interestingly, the number of synaptophysin puncta along dendrites of βI-transfected neurons was reduced when compared with GFP-only–transfected neurons (). To investigate whether this phenomenon reflects a reduction in the number of functional synapses, presynaptic boutons capable of active exo- and endocytosis were loaded with the lipophilic dye FM4-64 applied to neurons in the presence of 47 mM K+ (). The number of FM4-64–loaded boutons along βI-transfected dendrites was reduced by ∼40% when compared with βI-transfected cells, indicating that NCAM-mediated postsynaptic accumulation of spectrin and spectrin-associated proteins is required for synapse formation. In contrast, uptake of FM4-64 into individual synaptic boutons along βI- transfected dendrites was not affected, indicating that disruption of the postsynaptic NCAM–spectrin complex does not affect the presynaptic machinery. Similar to βI-transfected neurons, postsynaptic accumulations of NR1, NR2B, and PSD95 were reduced in NCAM−/− neurons, whereas accumulation of synaptophysin in presynaptic boutons was not changed ( and ). In contrast, although postsynaptic accumulation of GluR1 was slightly reduced in NCAM−/− neurons, synaptic targeting of GluR2/3 subunits was not changed (). It is noteworthy, in this context, that GluR1 interacts with protein 4.1, which is a prominent binding partner of spectrin that anchors AMPA receptors to the cytoskeleton (). Reduced association of GluR1 with the synaptic scaffold because of spectrin deficiency may cause reduced synaptic targeting of GluR1 in cultured NCAM−/− and βI-transfected neurons. However, PSDs from NCAM−/− brains contain normal levels of GluR1, suggesting more potent synapse-targeting cues than NCAM. Among those could be secreted molecules, such as neuronal activity–regulated pentraxin (Narp), which is involved in the clustering of AMPA receptors (). Alternatively, NCAM-independent accumulation of PSD95 may suffice for proper synaptic targeting of AMPA receptors. The current model for enhancement of synaptic strength suggests that Ca entering through NMDA receptors activates CaMKIIα, which in turn leads to an increase in the number and conductivity of AMPA receptors within PSDs (; ). Interestingly, this type of synaptic plasticity is abnormal in NCAM−/− mice showing impaired NMDA receptor–dependent LTP (; ). Thus, we hypothesized that abnormal PSD organization in NCAM−/− mice may affect signaling events in PSDs required for LTP. Indeed, whereas CaMKIIα protein levels were increased, levels of constitutively active CaMKIIα autophosphorylated at threonine 286 were reduced in NCAM−/− brains (), correlating with reduced levels of NMDA receptors in PSDs (, , and ). NMDA receptor–dependent activation of CaMKIIα is accompanied by its translocation to postsynaptic sites () and is regulated by the number of docking sites for CaMKIIα in PSDs (). Threonine 286–autophosphorylated CaMKIIα coimmunoprecipitated with NCAM (), indicating that the NCAM-associated scaffold may be a synapse-targeting cue for activated CaMKIIα. In accordance with this notion, clustering of NCAM induces redistribution and recruitment of CaMKIIα to the NCAM-assembled scaffold (). Accordingly, levels of CaMKIIα were reduced in NCAM−/− PSDs ( and ). To extend this analysis and verify whether activity-dependent translocation of CaMKIIα to synapses is affected in NCAM−/− neurons, we compared levels of CaMKIIα in synapses of cultured NCAM+/+ and NCAM−/− neurons at resting conditions and after incubation with 50 μM glutamate plus 5 μM glycine (), a protocol that induces NMDA receptor–dependent translocation of CaMKIIα to synapses (). Control and activated neurons were colabeled with antibodies against CaMKIIα and PSD95 to label PSDs and to determine CaMKIIα immunofluorescence intensity within PSD95 clusters. In nonstimulated neurons, CaMKIIα levels were reduced in PSDs of NCAM−/− neurons when compared with NCAM+/+ neurons, in accordance with data for brain tissue ( and ). Stimulation with glutamate enhanced levels of CaMKIIα in PSDs of NCAM+/+ neurons, in accordance with previous reports (). However, redistribution of CaMKIIα to PSDs of NCAM−/− neurons was strongly inhibited. Because in these experiments glutamate was applied to neurons in the culture medium, not only synaptic but also extrasynaptic NMDA receptors were activated. Because the levels of extrasynaptic NMDA receptors were increased in NCAM−/− versus NCAM+/+ neurons (), reduced redistribution of CaMKIIα to NCAM−/− PSDs was not likely caused by reduced activation of NMDA receptors by glutamate in NCAM−/− neurons. It is noteworthy, in this respect, that overall CaMKIIα protein expression was increased in NCAM−/− when compared with NCAM+/+ brains (). Redistribution of CaMKIIα to synapses was also inhibited in cultured NCAM+/+ neurons transfected with βI spectrin siRNA (). The combined observations indicate that the spectrin-based NCAM-associated postsynaptic signaling complex is required for efficient activity-dependent translocation of CaMKIIα to PSDs. It is now well-documented that NCAM plays an important role in memory formation and synaptic plasticity (), but the mechanisms by which NCAM exerts its functions have remained poorly understood. Using a variety of methods, we show that NCAM is an important PSD-targeting cue for spectrin, and that NCAM initiates and maintains a targeting platform for functionally crucial components of the synaptic machinery on dendrites. The fact that NCAM already promotes spectrin accumulation at immature axodendritic contacts before synapse formation suggests that NCAM clusters spectrin at contact sites, thereby promoting recruitment of the crucial binding partners, predominantly NMDA receptors and CaMKIIα, and thus transforming the initial contact into a functional synapse. Clustering of NCAM with antibodies in cultured neurons induces the formation of detergent- insoluble complexes enriched not only in NCAM, spectrin, and CaMKIIα but also in PSD95, Shank, and α-actinin, which are all constituents of PSDs. We call these complexes spectrin-based postsynaptic specializations, thus, highlighting spectrin as a cross-linking platform between NCAM and other components of the PSD. Whether Ca influx and intracellular signaling cascades activated by NCAM clustering (; ) induce posttranslational modifications of postsynaptic proteins, thereby priming them for recruitment to PSDs, is an important topic for further investigation. NCAM is required not only for spectrin accumulation at immature axodendritic contacts but also for the maintenance of spectrin and the accumulation of the associated proteins in PSDs in differentiated neurons. In culture, disruption of the NCAM–spectrin complex in differentiated NCAM-deficient neurons or by a functionally blocking spectrin fragment in differentiated wild-type neurons reduces the accumulation of associated proteins in PSDs. This was also seen by immunohistology at the electron microscopic level in brain tissue and in biochemically isolated PSDs from adult brain tissue. The NCAM-associated spectrin scaffold is interconnected with other scaffolding proteins in PSDs, such as PSD95 and α-actinin. However, PSD95 and α-actinin are not tightly associated with NCAM because we could not detect them in NCAM immunoprecipitates. Nevertheless, disruption of the NCAM–spectrin complex also reduces synaptic accumulation of PSD95 and α-actinin. PSD95 and α-actinin are also recruited to detergent-insoluble complexes of NCAM after the NCAM antibody induced clustering at the cell surface, probably by interaction with NCAM-associated proteins. An additional linkage between NCAM and postsynaptic scaffolds is provided by the interaction between spectrin and actin. At the cell surface, NCAM forms complexes with other cell adhesion molecules, such as N-cadherin (), L1 (), and the cellular prion protein (). The interconnectivity of postsynaptic scaffolds implies that disruption of a single scaffolding system in PSDs would not necessarily lead to a complete dissociation of the PSD complex because other scaffolding proteins may compensate, at least partially, for the missing protein. Accordingly, synaptic targeting of NMDA receptors and synapse density of the receptor subunits are normal in PSD95-deficient mice or when PSD95 family proteins are dispersed by an interfering peptide (; ). Similarly, disruption of N-cadherin adhesion by a dominant-negative construct only delays, but does not abolish, synapse formation, resulting in smaller synapses (). We also show that disruption of the NCAM–spectrin complex does not abolish synapse formation, but produces smaller PSDs containing reduced levels of spectrin, NMDA receptors, and CaMKIIα. Although synapse formation per se is not disrupted in NCAM−/− mice, the ability of synapses to produce NMDA receptor–dependent LTP is severely impaired in this mutant (; ). Our observations yield at least a partial explanation to this phenomenon; one of the major enzymes required to produce long-lasting changes in the synaptic strength is CaMKIIα (; ). Upon NMDA receptor activation and Ca influx, CaMKIIα redistributes to PSDs, where it phosphorylates several substrates, including AMPA receptors. We show that this pathway is impaired in NCAM−/− neurons, which accumulate reduced levels of CaMKIIα in PSDs already under nonstimulated conditions and are not able to increase levels of CaMKIIα in response to synapse activation. A possible explanation for this could be twofold; lower numbers of NMDA receptors in NCAM−/− PSDs result in reduced activation of CaMKIIα, whereas reduced levels of spectrin in NCAM−/− PSDs result in impaired recruitment of CaMKIIα to synapses. NCAM deficiency affects not only the structural organization of PSDs, resulting in reduction in their size, but disruption of the NCAM–spectrin complex also reduces the efficacy of excitatory synapse formation, thus, providing support for NCAM as a signal that recruits postsynaptic components to nascent synapses. Genetic variations in the NCAM gene have been reported a risk factor in bipolar affective disorders in humans (). Patients with schizophrenia and bipolar disorder accumulate secreted extracellular cleavage products of NCAM in hippocampus and prefrontal cortex (, ). These cleavage products may interfere with homophilic or heterophilic interactions of NCAM. Strikingly, transgenic mice ectopically expressing secreted NCAM display disturbed synaptic connectivity and abnormal behavior that may be relevant to schizophrenia (). Because neurological dysfunctions are, in many instances, likely to relate to synaptic abnormalities, our study lends support to the notion that NCAM is an important player in the assembly and function of a unique invention: the synapse. Rabbit polyclonal antibodies and rat monoclonal antibody H28 against mouse NCAM recognizing extracellular epitopes of the protein were as previously described (). NCAM180 was detected with monoclonal antibody D3 (). Rabbit polyclonal antibodies against NR2B and goat polyclonal antibodies against synaptophysin were obtained from Santa Cruz Biotechnology, Inc.; rabbit polyclonal antibodies against erythrocyte spectrin, CaMKIIα, and HA tag, mouse monoclonal antibodies against Flag tag, α-actinin, and MAP2, and nonspecific rabbit and rat immunoglobulins were purchased from Sigma-Aldrich; mouse monoclonal antibodies against NR1 (clone 54.1), NR2B, and gephyrin were obtained from BD Biosciences; mouse monoclonal antibodies against PSD95 were purchased from Upstate Biotechnology; mouse monoclonal antibodies against CaMKII were obtained from Stressgen Biotechnologies; rabbit polyclonal antibodies against phospho-CaMKII (Thr286) were purchased from Cell Signaling Technology; mouse monoclonal antibodies against actin (clone JLA20) and β-tubulin (clone E7) were obtained from the Developmental Studies Hybridoma Bank; and mouse monoclonal antibodies against HA tag (clone 12CA5) were purchased from Roche. Rabbit polyclonal antibodies against GluR1, GluR2/3, NR1, and rabbit polyclonal antibodies reactive with mouse Shank were gifts from R.J. Wenthold (National Institute on Deafness and Other Communication Disorders, Bethesda, MA) and H.-J. Kreienkamp (Universitätskrankenhaus Hamburg-Eppendorf, Hamburg, Germany), respectively. Secondary antibodies against rabbit, goat, rat, and mouse Ig coupled to HRP, Cy2, Cy3, or Cy5 were obtained from Dianova. DL-AP5 was purchased from Tocris. The vector containing the HA-NR1-1-GFP subunit of the NMDA receptor with the HA tag in the extracellular part of the protein and GFP tag in the intracellular part of the protein was a gift from G.A. Dekaban (The John P. Robarts Research Institute, London, Ontario, Canada). NR2B-Flag with the Flag tag in the extracellular part of the protein was a gift from L. Hawkins (National Institute on Deafness and Other Communication Disorders, Bethesda, Maryland). Constructs encoding βI spectrin fragments and NCAM isoforms have been described previously (). The eGFP plasmid was purchased from CLONTECH Laboratories, Inc. Spectrin βI siRNA was obtained from Santa Cruz Biotechnology, Inc. Control (nonsilencing) siRNA was purchased from QIAGEN. NCAM−/− mice were provided by H. Cremer () and were inbred for at least nine generations onto the C57BL/6J background. Animals for biochemical experiments (ages as indicated in the text) and electron microscopy (3-mo-old) were NCAM+/+ and NCAM−/− littermates obtained from heterozygous breeding. To prepare cultures of hippocampal neurons, 1- to 3-day-old C57BL/6J and NCAM−/− mice from homozygous breeding pairs were used. Cultures of hippocampal and cortical neurons were maintained on glass coverslips (for immunocytochemistry) or in 6-well plates (for biochemistry) in hormonally supplemented culture medium containing 5% horse serum (Sigma-Aldrich; ). Coverslips and plates were coated overnight with 100 μg/ml poly--lysine in conjunction with 20 μg/ml laminin. For analysis of synaptic accumulation of NMDA receptors, neurons were chronically treated with 100 μM DL-AP5 to increase synaptic targeting of NMDA receptors (). CHO cells were maintained in Glasgow's modified Eagle's medium containing 10% of fetal calf serum. Neurons were transfected 4 d after plating with DNA constructs or 12 d after plating with siRNA using Lipofectamine 2000 (Invitrogen), according to the manufacturer's instructions, and analyzed 14 d after plating. CHO cells were transfected using Lipofectamine with Plus reagent (Invitrogen) following the manufacturer's instructions. In experiments analyzing activity-dependent synaptic translocation of CaMKIIα, neurons were treated for 20 s with 50 μM glutamate plus 5 μM glycine (). Immunolabeling was performed essentially as previously described (). For labeling of GluR1 and GluR2/3, cultures were fixed in 4% formaldehyde/4% sucrose in PBS, pH 7.3, for 10 min at 37°C, washed, and permeabilized with 0.25% Triton X-100 in PBS for 5 min. This type of mild fixation was also used for the labeling of GFP-transfected neurons to preserve GFP fluorescence. For labeling of NR1, NR2B, PSD95, CaMKIIα, and NCAM180 in nontransfected neurons, fixation with methanol (−20°C for 7 min) was used. Neurons were blocked in 3% BSA in PBS. The antibodies were applied in 3% BSA in PBS for 2 h at 37°C and detected with corresponding secondary antibodies after incubation for 45 min at room temperature. To label NCAM, Flag, or HA tag at the cell surface of live neurons, antibodies against NCAM, HA, or Flag tag were applied in culture medium to live cultures for 10 min and detected with fluorochrome-coupled secondary antibodies applied for 5 min, all in a CO incubator (). Detergent extraction was performed as previously described (). Clustering of NCAM was induced by incubating live 4-d-old neurons for 15 min (5% CO at 37°C) with NCAM antibodies, and was visualized with secondary antibodies applied for 5 min (). Detergent-extracted neurons were fixed in 4% formaldehyde in PBS and labeled with antibodies. Neurons were briefly washed in modified Tyrode solution containing 150 mM NaCl, 4 mM KCl, 2 mM MgCl, 10 mM glucose, 10 mM Hepes, and 2 mM CaCl, pH 7.4, (∼310 mosM) and incubated for 90 s in 47 mM K+ solution (modified Tyrode solution containing equimolar substitution of KCl for NaCl) containing 15 μM FM4-64 (Invitrogen) to induce synaptic vesicle exo- and endocytosis (). Neurons were then extensively washed with modified Tyrode solution. All staining protocols were performed with 10 μM CNQX and 50 μM AP-5 to prevent recurrent activity. Images were acquired from live neurons maintained in modified Tyrode solution at room temperature using a confocal laser scanning microscope (LSM510; Carl Zeiss MicroImaging, Inc.). Coverslips were embedded in Aqua-Poly/Mount (Polysciences, Inc.). Images were acquired at room temperature using a LSM510, LSM510 software (version 3), and an oil Plan-Neofluar 40× objective, NA 1.3 (all Carl Zeiss MicroImaging, Inc.) at 3× digital zoom. Contrast and brightness of the images were further adjusted in Corel Photo-Paint 9 (Corel Corporation). Immunofluorescence quantification was performed essentially as previously described (). In brief, to define synaptic NR1, NR2B, GluR1, GluR2/3, or PSD95 clusters, the thresholds for each individual experiment and corresponding channels were chosen manually and corresponded approximately to two times the average intensity of fluorescence in the dendritic shafts. The same threshold was used for all neurons in one experiment. Binary images of clusters of NR1, NR2B, GluR1, GluR2/3, or PSD95 were compared with binary images of synaptophysin clusters. Any postsynaptic cluster that had at least one pixel of overlap with a presynaptic cluster was defined as synaptic. Cluster size and mean intensity of the labeling within a cluster were measured using Scion Image for Windows (Scion Corporation). Synaptic accumulation of the protein was then defined as a product of cluster size and mean labeling intensity of the protein and expressed in arbitrary units. CaMKIIα intensity was measured in PSD95 clusters. Values indicate the mean ± the SEM. Group comparisons were made by test. Each experiment was reproduced at least two times. Colocalization profiles were analyzed using ImageJ software (National Institutes of Health, Bethesda, MD). Correlation coefficients were calculated using Excel software (Microsoft). Proteins were separated by 6–16% SDS-PAGE and electroblotted to Nitrocellulose Transfer Membrane (PROTRAN; Schleicher & Schuell) for 3 h at 250 mA. Immunoblots were incubated with the appropriate primary antibodies, followed by incubation with HRP-labeled secondary antibodies, and visualized using Super Signal West Pico or West Dura reagents (Pierce Chemical Co.) on BIOMAX film (Sigma-Aldrich). Molecular mass markers were prestained protein standards obtained from Bio-Rad laboratories. For quantitative comparisons of chemiluminescence between the lanes, the same amounts of total protein or equal amounts of immunoprecipitates were loaded in each lane. All preparations were performed three times, and at least two Western blots were performed with an individual sample ( ≥ 6). In experiments, when NCAM+/+ and NCAM−/− total membrane and PSD fractions were compared, NCAM+/+ total membrane values were set to 1 and other intensities were normalized to these values. Values of all experiments were used to calculate mean values and SEMs. The chemiluminescence quantification was performed using TINA 2.09 software (University of Manchester, Manchester, England) or Scion Image for Windows (Microsoft). Group comparisons were made by paired test. Brain homogenates were prepared in 5 mM Tris-HCl buffer, pH 7.5, containing 1 mM of CaCl, 1 mM MgCl, and 1 mM NaHCO. Samples containing 1 mg of total protein were lysed for 40 min at +4°C with lysis buffer, pH 7.5, containing 50 mM Tris-HCl, 150 mM NaCl, 1% NP-40, 1 mM NaPO, 1 mM NaF, 1 mM EDTA, 2 mM NaVO, 0.1 mM PMSF, and complete protease inhibitor cocktail (Roche), and centrifuged for 15 min at 20,000 at 4°C. Supernatants were cleared with protein A/G–agarose beads (Santa Cruz Biotechnology, Inc.) for 3 h at 4°C and incubated with corresponding antibodies or control Ig overnight at 4°C, followed by precipitation with protein A/G–agarose beads for 1 h at 4°C. The beads were washed three times with lysis buffer and two times with PBS, and then analyzed by immunoblotting. Membrane fractions and PSDs were isolated as previously described (). Total protein concentration was measured using the BC kit (Interchim). To measure PSD size, tissue was processed as previously described (). Postembedding immunocytochemistry was also performed essentially as previously described (). In brief, mice were deeply anesthetized and transcardially perfused with 2% ice-cold dextran (70,000 kD) in PBS, and then with 4% formaldehyde/0.3% glutaraldehyde in PBS at room temperature. Brains were removed, incubated in the fixative overnight at 4°C, washed, and cut in 100-μm-thick coronal sections with a VT1000S vibratome (Leica) all in 4% sucrose in PBS. Sections were cryoprotected using a series of 10, 20, and 30% sucrose in PBS, plunge-frozen in 2-methyl-butane precooled with liquid nitrogen, and subjected to freeze-substitution and low-temperature embedding in Monostep Lowicryl HM20 resin. Frozen tissue was immersed in methanol at −90°C in an AFS instrument (Leica), infiltrated in Lowicryl HM 20 resin at −45°C, and polymerized with UV light (−45–0°C). 90-nm sections were cut on an Ultracut S ultramicrotome (Leica) and collected on parlodion-coated nickel grids. Grids were incubated in 0.1% sodium borohydride and 50 mM glycine in Tris-buffered saline, pH 7.4, containing 0.1% Triton X-100 (TBST) for 10 min, and incubated for 10 min in 10% normal donkey serum (NDS) in TBST, followed by primary antibodies in 1% NDS in TBST for 2 h. Primary antibody concentrations were selected to produce no background immunogold labeling. Sections from both genotypes were labeled in parallel. Grids were washed in TBST, blocked in 10% NDS/TBST, and incubated with 12-nm immunogold particles (GE Healthcare; 1:20) in NDS/TBST plus 0.5% polyethylene glycol (20,000 kD) for 1 h. Sections were counterstained with uranyl acetate and lead citrate. Fig. S1 shows that NMDA receptors cocluster with NCAM at the surface of hippocampal neurons. Fig. S2 shows efficiency of protein solubilization with 1% NP-40. Fig. S3 shows that expression of NMDA receptors is increased in NCAM−/− brains. Fig. S4 shows that NMDA receptors colocalize with NCAM at the surface of transfected CHO cells. Online supplemental material is available at .

E-cadherin is the main epithelial cell–cell adhesion molecule, and either it is functionally disrupted or its expression is lost altogether during tumor progression (; ). The effect is likely to be direct, because reestablishing E-cadherin function in cadherin-deficient cell lines can reverse the invasive phenotype (). Moreover, experiments in transgenic mice strongly suggest that loss of E-cadherin directly promotes the transition of a benign adenoma into a carcinoma (). The mechanism by which E-cadherin suppresses invasiveness is still unclear. The intracellular domain of E-cadherin interacts directly with β-catenin and p120 catenin (p120) via separate conserved interaction domains. β-Catenin binding was recently shown to be important for the anti-invasive properties of E-cadherin (), although neither increased cell adhesion nor reduced nuclear β-catenin signaling was required for this effect. Unlike β-catenin, p120 has not been implicated in E-cadherin–mediated suppression of invasiveness, although it mislocalizes to the cytoplasm of E-cadherin–deficient cells. This altered localization of p120 in breast or colon carcinomas is prognostic for aggressive disease (; ). Epithelial-to-mesenchymal transition is a process associated with normal development and wound healing, but its aberrant regulation contributes to cancer progression and metastasis (). Epithelial-to-mesenchymal transition is associated with loss of E-cadherin expression and increased expression of mesenchymal cadherins. Indeed, overexpression studies have suggested that increased expression of mesenchymal cadherins (N-cadherin, R-cadherin, and cadherin 11) increases the motility and invasiveness of epithelial cells (; ; ; ). It is currently unclear whether endogenous mesenchymal cadherins are required for the increased motility/invasiveness of E-cadherin–deficient cells. The Rho family of GTPases (e.g., RhoA, Rac1, and Cdc42) mediate cytoskeletal dynamics () and are crucial regulators of both cell motility () and cadherin-dependent cell adhesion (). As such, Rho GTPases are thought either to promote intercellular adhesion or to induce cell migration, depending on signals received from the microenvironment. Signaling from the cadherin complexes to Rho GTPases is thought to depend on p120 (). Recent data indicate that p120 binding promotes the stabilization of cadherin complexes on the plasma membrane and thus strengthens cell–cell adhesion (; ). In some cases, p120 can also negatively affect cell adhesion, although the mechanism of this effect remains unclear. p120 overexpression induces dramatic changes in cell morphology and increases cell motility (for review see ). These effects are apparently mediated by the ability of p120 to suppress RhoA activity (; ) and induce the activities of the related Rho GTPases Rac1 and Cdc42 (; ). E-cadherin overexpression blocks the effects of p120 on cell morphology, suggesting that the recruitment of p120 to E-cadherin complexes reduces its effects toward Rho GTPases and possibly affects the balance between sessile and motile states. Using E-cadherin–deficient cells, we show that endogenous p120 mediates both the invasion-promoting effects of mesenchymal cadherins and the invasion-suppressing action of ectopically expressed E-cadherin. Endogenously expressed mesenchymal cadherins are essential for the invasiveness of E-cadherin–deficient cells, and their levels depend on p120 association. Furthermore, p120-induced Rac activation requires binding of p120 to mesenchymal cadherins and promotes invasiveness. p120 also promotes invasiveness by inhibiting RhoA in a cadherin-independent manner. The data indicate that endogenous p120 is an important contributor to both the invasive phenotype of E-cadherin–deficient carcinomas and the sessile phenotype of E-cadherin–expressing epithelial cells. To test whether endogenous p120 promotes the invasiveness of E-cadherin–deficient tumor cells, we measured the invasiveness of cells with diminished p120 expression. MDA-MB-231 (MDA-231) and UMRC3 cells were infected with a retroviral vector (pRS) expressing short hairpin RNA (shRNA) targeted against human p120, as described previously (). UMRC3 cells are highly metastatic renal carcinoma cells that lack E-cadherin expression. Polyclonal cell populations with reduced p120 expression were tested for invasiveness in Matrigel-coated transwells toward a gradient of hepatocyte growth factor (HGF; MDA-231 cells) or 5% FBS (UMRC3 cells). HGF was used as a chemoattractant because blocking HGF signaling prevents MDA-231 metastasis in nude mice (). As shown in , reduction of endogenous p120 levels resulted in significantly reduced invasiveness in both cell lines. Next, we selected individual clones of shRNA-expressing MDA-231 cells with varying expression of endogenous p120 and tested their invasiveness toward HGF. demonstrates that the invasiveness of MDA-231 cells in vitro is proportional to the levels of endogenous p120. Control experiments verified that cell line growth was not significantly affected by p120 depletion under our experimental conditions. Identical results were obtained using anti-p120–specific RNAi compared with control RNAi (unpublished data). Combined, the data strongly suggest that endogenous p120 is required for the invasiveness of E-cadherin–deficient cells. To further validate the in vitro invasion data, we measured cell migration using a scratch-wound assay. Within 12 h, control pRS/Neo-expressing cells moved to fill in the wound much more efficiently than cells expressing human-specific p120 shRNA (p120KD; ). Expression of full-length murine p120 (mp120) in p120KD cells rescued cell migration in response to HGF. The data argue that endogenous p120 mediates HGF-induced cell migration. To address the possibility that p120 promotes motility by decreasing cell–cell adhesion, we performed cell aggregation–disaggregation assays. Cells were allowed to form multicellular spheroid aggregates after overnight incubation in a hanging drop of media. Cell aggregates were then pipetted 10 times with a small bore tip, and the extent of cell disaggregation was assessed as a measure of adhesion strength. As can be seen in , ectopic expression of mp120 in p120KD cells moderately increased cell–cell adhesion, suggesting that increased motility and invasiveness in response to p120 is not due to reduced adhesion. Finally, we examined the invasiveness of p120KD cells expressing murine p120 or a p120 mutant that lacks the first Armadillo domain (A1) and is unable to associate with E-cadherin (). As shown in , expression of full-length p120 reversed the p120KD effect and increased cell invasiveness. In contrast, ectopic expression of the p120 A1 mutant (mp120-A1) failed to rescue invasiveness. The data indicate that the effects of p120 depletion on cell invasion and motility are indeed p120 dependent. They also suggest that cadherin binding is required for the effects of p120 on invasion. The inability of a cadherin-uncoupled p120 mutant to promote invasiveness suggested the possibility that in these E-cadherin–deficient cells, p120 induces invasiveness via its association with mesenchymal cadherins. MDA-231 cells lack E-cadherin expression but express mesenchymal cadherin 11, whereas UMRC3 cells lack E-cadherin expression but express mesenchymal N-cadherin (unpublished data). Both of these cadherins have been shown to promote motility and invasiveness when overexpressed in tumor cells (; ). To test the possibility that endogenous p120 acts via mesenchymal cadherins to promote invasiveness, we initially expressed a short myc-tagged fragment of the E-cadherin cytoplasmic tail in MDA-231 cells. This small E-cadherin fragment (ΔCB) binds avidly to p120, but not β-catenin, and recruits it to the cytosol away from any endogenous cadherins (). As can be seen in , expression of ΔCB significantly reduced MDA-231 cell invasiveness. Furthermore, ΔCB expression caused a marked reduction in endogenous cadherin 11 levels, consistent with the hypothesis that p120 binding regulates the levels of endogenous cadherins (; ; ). To directly test whether endogenous p120 regulates the levels of mesenchymal cadherins in E-cadherin–deficient cells, we determined the expression of mesenchymal cadherins after p120 depletion. shows that depletion of endogenous p120 reduced mesenchymal cadherin expression in both MDA-231 and UMRC3 cells. Finally, to determine whether endogenously expressed mesenchymal cadherins promote invasion, we treated MDA-231 and UMRC3 cells with siRNAs against cadherin 11 or N-cadherin, respectively, and measured cell invasiveness. In both cases, reduction of the endogenous mesenchymal cadherin levels resulted in significant inhibition of cell invasiveness, indicating that endogenous mesenchymal cadherins mediate motility and invasiveness in E-cadherin–deficient cells (). The data in suggested that one way by which p120 promotes invasiveness is by binding to and regulating the levels of mesenchymal cadherins. Consistent with this hypothesis, the E-cadherin–uncoupled p120 A1 mutant was unable to promote invasiveness when expressed in p120KD cells (). However, it is not known whether this p120 mutant (p120-A1) is also unable to associate with mesenchymal cadherins and whether its expression affects mesenchymal cadherin levels. To answer the first question, we performed coimmunoprecipitation assays using full-length p120 as a control. As shown in , ectopically expressed murine p120 coprecipitated with endogenous cadherin 11 from MDA-231 cells, whereas the p120-A1 mutant did not. Similar results were also obtained for N-cadherin (unpublished data), indicating that the p120-A1 mutant is uncoupled from both epithelial and mesenchymal cadherins. As shown earlier, p120 depletion caused a reduction in mesenchymal cadherin levels (). shows that p120 reexpression increased the levels of both cadherin 11 and N-cadherin in MDA-231 and UMRC3 cells, respectively. In contrast, expression of the cadherin-uncoupled p120-A1 mutant had no effect on mesenchymal cadherin expression, in agreement with a requirement for p120 association. We next tested the ability of a p120-uncoupled N-cadherin mutant to promote invasiveness. shows that ectopic expression of murine N-cadherin in UMRC3 cells depleted of endogenous N-cadherin rescued cell invasiveness. However, expression of a p120-uncoupled N-cadherin mutant (N-cad-AAA; ) failed to promote invasiveness. The data argue that p120 promotes motility and invasiveness by binding to mesenchymal cadherins. The association promotes cadherin stability and possibly signaling events that induce cell migration. Initially, we examined the activity of Rac1 in control cells (pRS-neo), p120-depleted cells (p120KD-neo), and cells reexpressing p120 (p120KD-mp120), under basal, serum-starved conditions. As shown in , under basal conditions, p120 depletion resulted in significant reduction of Rac1 activity, which was reversed by expression of murine p120. p120 depletion also affected Rac1 activity of HGF-treated cells. shows that Rac1 activity was increased in control pRS-neo cells treated with HGF but not in p120-depleted cells (p120KD-neo). Expression of murine p120 in p120-depleted cells (p120KD-mp120) restored HGF-mediated Rac1 activation. In contrast, expression of the cadherin-uncoupled p120-A1 mutant failed to restore Rac1 activation, suggesting that p120-mediated activation of Rac1 requires mesenchymal cadherin binding. Finally, if the ability of p120 to induce Rac1 activation requires cadherin association, depletion of endogenous cadherins should block p120-mediated Rac1 activation. To test this, we used either control p120-depleted cells (p120KD-neo) or p120-depleted cells reexpressing p120 (p120KD-mp120). Rac1 activation in response to HGF () is p120 dependent under these conditions. shows that the ability of p120 to mediate HGF-induced Rac1 activation depends on the levels of cadherin 11, as a cadherin 11–specific siRNA blocks p120-mediated Rac1 activation, whereas a control siRNA has no effect. As expected, cadherin 11 depletion also inhibited HGF-induced Rac1 activation in parental MDA-231 cells (unpublished data). Collectively, these data strongly argue that the p120-mediated activation of Rac1 requires association of p120 with mesenchymal cadherins. Previous studies have shown that p120 overexpression decreases RhoA activity (; ; ). Thus, we examined the effect of p120 on RhoA activity under either basal (serum-starved) conditions or after HGF treatment. shows that under basal conditions, RhoA activity was significantly increased in p120-depleted cells (p120KD-neo) when compared with control cells (pRS-neo). Consistent with the increased RhoA activity, p120-depleted cells were flatter and contained more stress fibers than control cells (unpublished data). Expression of murine p120 reduced basal RhoA activity levels and restored normal cell morphology, indicating that p120 regulates RhoA activity. Similar results were obtained when cells were treated for 20 min with HGF; control cells and p120-reexpressing cells exhibited reduced Rho activity compared with p120-depleted cells (). We then asked whether the p120-mediated reduction of RhoA activity requires binding to endogenous cadherins. Again, we used p120-depleted (p120KD-neo) and p120-reexpressing cells (p120KD-mp120) and examined RhoA activity in response to HGF treatment in the presence of cadherin 11 or control siRNA. shows that p120 can suppress RhoA activity even in cadherin 11–depleted cells, suggesting that cadherin 11 is not required for p120-induced RhoA inhibition. This conclusion was corroborated by the observation that the cadherin-uncoupled p120 A1 mutant, like wild-type (wt) p120, inhibits RhoA activity in MDA-231 cells (). Finally, like wt p120, which when overexpressed induces a branching morphology by inhibiting RhoA activity (), mp120-A1 was able to induce a branching morphology in NIH3T3 cells (). The data suggest that, unlike p120-mediated Rac1 activation, the p120-mediated inhibition of RhoA is independent of mesenchymal cadherin expression in these cells. We next sought to establish that the p120-mediated changes in Rho GTPase activities were related to the increased cell migration and invasiveness of E-cadherin–deficient cells. First, we overexpressed constitutively active or dominant-negative Rac1 in MDA-231 cells and measured the in vitro invasiveness of serum-starved cells toward a gradient of HGF. Expression of dominant-negative Rac1 (DN-Rac1) significantly reduced cell invasiveness (), suggesting that Rac1 activation is required for HGF-induced invasiveness. Ectopic expression of constitutively active Rac1 did not increase cell invasiveness over cells expressing a vehicle control (pcDNA), suggesting that parental MDA-231 cells maximally activate Rac1 in response to HGF (unpublished data). To test the potential involvement of Rho activation in the invasiveness of MDA-231 cells, we expressed constitutively active RhoA and measured cell invasiveness in response to HGF. As shown in , increased levels of active RhoA significantly reduced the invasiveness of MDA-231 cells. To directly correlate changes in Rho GTPase activities with p120-induced invasiveness, we tested whether activation of Rac1 and inhibition of RhoA signaling can restore the invasiveness of p120-depleted cells. As shown in , stable expression of constitutively active Rac1 (CA-Rac1) in p120-depleted MDA-231 cells increased invasiveness toward HGF in vitro. Inhibition of Rho signaling by incubating p120-depleted cells expressing CA-Rac1 with a Rho kinase inhibitor (H1152; 1.6 nM; ) further promoted cell invasiveness. Together, the data argue that changes in Rac and Rho signaling are causally involved in p120-mediated effects on cell motility and invasiveness. Having established that endogenous p120 promotes the invasive behavior of E-cadherin–deficient cells, we tested the hypothesis that recruitment of p120 to E-cadherin can suppress invasiveness. Ectopic expression of wt human E-cadherin in MDA-231 or UMRC3 cells was accomplished by retroviral infection, followed by G418 selection. More than 95% of infected cells expressed E-cadherin under these conditions. As shown in , MDA-231 cells infected with a control neo virus express no E-cadherin, have primarily cytoplasmic p120 staining, and exhibit scattered cell morphology. In contrast, cells expressing wt E-cadherin form epithelial colonies in which p120 is recruited to the E-cadherin–mediated cell–cell junctions. Identical results were also obtained with UMRC3 cells. The ectopic expression of E-cadherin significantly inhibited cell migration () and blocked the ability of either MDA-231 or UMRC3 cells to invade in vitro (). Together, these experiments establish that both cell lines represent excellent model systems for studying the mechanism by which E-cadherin suppresses invasiveness and the possible involvement of p120. Next, we compared the invasiveness of MDA-231 cells expressing a p120-uncoupled E-cadherin mutant (764-AAA; ) to that of control cells or cells expressing wt E-cadherin. As shown in , cells expressing the p120-uncoupled 764-AAA E-cadherin mutant were significantly more invasive than cells expressing wt E-cadherin but not as invasive as the neo controls. Control experiments verified that both cadherins were expressed at comparable levels and that cells expressing the p120-uncoupled E-cadherin mutant do form colonies and recruit β-catenin to the membrane (unpublished data; ). To directly implicate endogenous p120 in E-cadherin–mediated suppression of invasiveness, we used two complimentary approaches. In the first, we overexpressed p120 in E-cadherin–expressing cells. In the second, we inhibited p120 function in cells expressing the p120-uncoupled E-cadherin mutant. shows that p120 overexpression increased the invasiveness of wt E-cadherin–expressing cells in vitro, suggesting that recruitment of endogenous p120 by E-cadherin suppresses invasion. Furthermore, the invasiveness of cells expressing the 764-AAA E-cadherin mutant was potently blocked by coexpression of ΔCB, the small cytoplasmic fragment of E-cadherin that binds selectively to p120 but not β-catenin. The data suggest that the increased invasiveness of cells expressing p120-uncoupled E-cadherin is due to its inability to recruit endogenous p120 to the E-cadherin complex. To confirm this hypothesis, we examined the effect of p120 depletion on the invasiveness of cells expressing the p120-uncoupled E-cadherin mutant. Retroviral expression of p120-specific shRNA, but not control shRNA, significantly reduced the invasiveness of these cells (). Examination of individual p120 shRNA clonal lines with varied endogenous p120 expression indicated that cell invasion is directly correlated with endogenous p120 levels (). Collectively, the data indicate that p120 binding is required for E-cadherin–mediated suppression of invasiveness. Finally, we also examined the effects of E-cadherin reexpression on the levels of endogenous mesenchymal cadherins. As can be seen in , E-cadherin expression in MDA-231 cells resulted in a marked reduction of cadherin 11 levels. The data are consistent with the hypothesis that p120 recruitment to E-cadherin causes the loss of p120 binding to mesenchymal cadherins and subsequent reduction of their protein levels. Indeed, expression of p120-uncoupled E-cadherin largely reverses the loss of mesenchymal cadherin expression observed in cells expressing wt E-cadherin and, as shown earlier (), restores their invasiveness. Similar results were observed in UMRC3 cells (which express N-cadherin), demonstrating that this effect is not restricted to MDA-231 cells (unpublished data). Several previous studies have suggested that overexpression of p120 catenin (; ) or overexpression of mesenchymal cadherins (; ; ) promotes cell migration. We have used E-cadherin–deficient cancer cells to test the hypothesis that p120 is required for the increased migration and invasiveness of these cells in vitro. The results reveal an essential role for p120 in both the migration and invasiveness of these cells and an unexpected role in mediating the proinvasive function of endogenous mesenchymal cadherins. Remarkably, association of endogenous p120 with E-cadherin is required for E-cadherin–mediated suppression of invasiveness and is accompanied by a concomitant reduction in mesenchymal cadherin levels. Mechanistically, p120 seems to regulate migration and invasiveness via three seemingly independent pathways: the p120 association–dependent regulation of mesenchymal cadherin levels, the induction of Rac1 activity after mesenchymal cadherin binding, and the cadherin-independent inhibition of RhoA. Our data demonstrate that mesenchymal cadherins are essential for the migration and invasiveness of E-cadherin–deficient tumor cells and that both the physical and the functional interaction with p120 are required for the proinvasive function of mesenchymal cadherins. The data are also consistent with a recent study suggesting that the p120-binding juxtamembrane domain of cadherin 11 is responsible for cadherin 11–mediated cell motility (). It has been proposed that p120 mediates its effects on cell migration through regulation of Rho GTPases (for review see ). Interestingly, we show that both the basal and HGF-induced Rac1 activities are inhibited by p120 depletion. Furthermore, experiments using cadherin 11–depleted cells or cadherin-uncoupled p120 mutants indicate that the p120-induced Rac1 activation requires mesenchymal cadherin association and are in agreement with recent data suggesting that the p120-binding juxtamembrane domain is required for E-cadherin– or N-cadherin–induced Rac1 activation upon cell–cell adhesion (; ). p120 depletion increased RhoA activity and decreased the activity of Rac1, providing a potential mechanistic explanation for the ability of dominant-active Rac mutants and dominant-negative RhoA mutants to rescue the defects induced by p120 depletion on gastrulation (). However, unlike Rac activation, the data suggest that p120 inhibits RhoA in a cadherin-independent manner in these cells, in agreement with several previous investigations (; ; ; ; ). To address the involvement of Rho GTPases in the invasiveness of our E-cadherin–deficient cells, we initially demonstrated that either reduced Rac1 or increased RhoA activities result in decreased cell invasiveness in vitro Reduced Rac1 and increased RhoA activities mimic the effects of p120 depletion in these cells. Interestingly, activation of Rac1 and inhibition of RhoA signaling cooperatively restored the invasiveness of p120-depleted cells, arguing that changes in Rho GTPase signaling are causally involved in p120-mediated effects on cell motility and invasiveness. The observation that endogenous p120 promotes the expression of mesenchymal cadherins and increases the invasiveness of E-cadherin–deficient cells suggested the possibility that p120 is also involved in the invasion-suppressive function of E-cadherin. In the simplest scenario, E-cadherin association would reduce the amount of p120 available to bind mesenchymal cadherins and promote invasiveness. Indeed, we show that E-cadherin suppresses invasion, at least in part, by binding endogenous p120. Furthermore, as predicted, mesenchymal cadherin levels were significantly reduced upon the expression of wt E-cadherin but not the p120-uncoupled E-cadherin mutant (764-AAA). The data reveal an important role for p120 binding in E-cadherin–mediated suppression of invasiveness and regulation of the motile or sessile phenotype of epithelial cells. It should be noted that the invasiveness of 764-AAA E-cadherin–expressing cells was lower than that of neo controls (), suggesting that a portion of the invasion-suppressive function of E-cadherin may not be related to p120 binding. It is likely that other factors, including recruitment of β-catenin (), play important roles in this process. Our data indicate that p120 binds to and cooperates with mesenchymal cadherins to activate Rac1 and promote motility and invasiveness. However, it is unclear why E-cadherin suppresses migration, despite its ability to activate Rac1 in a p120-dependent manner after its homophilic interaction. One possibility is that the activation of Rac1 in the context of cadherin ligation is effectively different from activation of Rac1 in response to certain growth factors. In agreement with this, lamellipodium extension in response to cadherin activation is reportedly dependent on a PI-3-kinase–Rac1 pathway, whereas cadherin-mediated adhesion proceeds via a PI-3-kinase independent, Rac1-dependent pathway; both responses require the membrane association of p120 with the cadherin complex (). In addition, E-cadherin may be less effective than mesenchymal cadherins in promoting Rac1 activation in response to promigratory signals, or it may be more capable of suppressing growth factor signaling by sequestering and preventing the ligand-dependent activation of their receptors (). Another possibility is that the differential ability of cadherins to recruit p120 to cell junctions may result in differential regulation of Rho activities. In support of these possibilities, the increased migration of R-cadherin overexpressing BT-20 cells, which normally express E-cadherin, correlates with increased Rac1 and reduced RhoA activities (), suggesting that E-cadherin and mesenchymal cadherins differentially affect Rho GTPases. Finally, it is important to note that collagen-mediated integrin signaling can switch the effect of increased Rac1 activation from promoting E-cadherin–mediated adhesion to promoting cell migration (). The data indicate that contextual signals can misdirect Rac signaling to promote cell migration, even in the presence of E-cadherin. Clearly, understanding the functional differences between E-cadherin and mesenchymal cadherins in regulating cell adhesion versus migration will be critical for understanding tumor progression to metastasis and events involved in tissue morphogenesis. The data presented here imply that E-cadherin competes p120 away from mesenchymal cadherins, which then become destabilized. Further studies will be needed to address the relative affinities of p120 for different cadherins and how these affinities are affected by posttranslational modifications (e.g., p120 phosphorylation). It is possible that the functional disruption of the cadherin–catenin complex, which is often the result of Ras mutations or constitutive receptor tyrosine kinase signaling, promotes a more invasive phenotype by reducing the affinity of p120 for E-cadherin. As invasiveness was tested here using cell culture models, future studies are needed to show whether these results reflect invasive behavior in vivo In any case, our data indicate that endogenous p120 acts as a rheostat, promoting a sessile cellular phenotype when associated with E-cadherin or a motile phenotype when associated with mesenchymal cadherins. Culture conditions have been described previously (). For shRNA expression, cells were infected with pRS and selected with 5 μg/ml puromycin. As indicated, some cells were infected again with LZRS-neo or -zeo and selected with 1 mg/ml G418 or 350 μg/ml zeocin. pRS and LZRS amphotropic retroviruses were produced as described previously (). Clonal MDA-MB-231 cell lines were generated by limiting dilution. Amaxa electroporations were performed according to the company's protocol. In brief, 10 cells were resuspended in 100 μl solution T (Amaxa) containing 2 μg of plasmid DNA. Electroporation was performed using program A-23. Electroporated cells were plated in 60-mm dishes and incubated for 24 h in normal culture media. Cells were then washed in PBS and incubated for another 12 h in serum-free Dulbecco's minimal essential medium. Immunofluorescence localization procedures have been described in detail (). The following primary antibodies were used: 0.5 μg/ml F1αSH p120 polyclonal antibody () and 1 μg/ml HECD-1 (Zymed Laboratories). The secondary antibodies used were goat anti-mouse Alexa 488 (Invitrogen) and goat anti-rabbit Alexa 596 at 1:600. Cells were visualized under a fluorescent microscope (DM5000B; Leica) using a 63×/1.4 HCX planApo oil objective (Leica). Photos were acquired with the FX4000 program (Leica) using a charge-coupled device camera (DFC350FX; Leica) and compiled in Photoshop (Adobe) and PowerPoint (Microsoft). LZRS-mp120 isoform 1A-neo, LZRS-mp120 A1-neo, LZRS-wt-E-cadherin-neo, and LZRS-764-AAA-neo were described previously (). The pRS vector was a gift from R. Agami (The Netherlands Cancer Institute, Amsterdam, Netherlands). pRS human p120 shRNA was also described previously (). LZRS-MS-zeocin was provided by A. Reynolds (Vanderbilt University, Nashville, TN) and encodes for zeocin instead of neomycin resistance. Initially, LZRS-ΔCB-GFP was generated by subcloning an EcoRI fragment of pCAN-ΔCB () into the EcoRI site of LZRS-MS-GFP (LZRS-ΔCB-GFP). LZRS-ΔCB-zeocin was generated by subcloning a SgfI–SfiI fragment of LZRS-ΔCB-GFP (containing ΔCB) into the respective sites of the LZRS-MS-zeocin vector. The RhoA-V14-myc (CA-RhoA), Rac1-V12-myc (CA-Rac1), and Rac1-N17-myc (DN-Rac1) pcDNA3 constructs were all described previously (). Murine N-cadherin-YFP and N-cadherin-AAA-YFP were provided by K.J. Green (Northwestern University, Chicago, IL; ). All constructs were verified by sequencing. Smartpool siRNAs against human cadherin 11 and N-cadherin were obtained from Dharmacon. Silencing specificity was confirmed using ON-Targetplus nontargeting siRNAs (Dharmacon). Western blotting procedures were conducted as described previously (). Primary antibodies were used as follows: 0.25 μg/ml anti-p120 mAbs pp120 and 1 μg/ml 8D11 (does not recognize human p120), anti–E-cadherin mAbs (1/2,500; C-20820; BD Biosciences) and 1 μg/ml HECD-1 (Zymed Laboratories), 1 μg/ml anti–c-Met (C-28; Santa Cruz Biotechnology, Inc.), anti–β-catenin polyclonal antibody (1/1000; C2206; Sigma-Aldrich), 5 μg/ml anti-Flag tag mAb (M2; Sigma-Aldrich), 1 μg/ml anti-myc tag (9E10; Sigma-Aldrich), anti–cadherin 11 antibodies (WTID1 [polyclonal antibody] and 5B2H5 [mAb]; Zymed Laboratories), and 0.6 μg/ml anti-actin goat polyclonal antibody (I-19; Santa Cruz Biotechnology, Inc.). Secondary antibodies were peroxidase-conjugated donkey anti–mouse IgG and mouse anti–rabbit IgG (Jackson ImmunoResearch Laboratories) and donkey anti-goat IgG (Santa Cruz Biotechnology, Inc.) used at 1:10,000. Cell invasion was measured in vitro using BioCoat Matrigel-coated invasion chambers (8 μm pore size; Becton Dickinson). Culture medium was changed to Dulbecco's minimal essential medium supplemented with 250 μg/ml BSA, and cells were incubated overnight at 37°C. Cells were then harvested using Cell Stripper (Mediatech, Inc.), to prevent the proteolytic degradation of cadherins, and resuspended in Dulbecco's minimal essential medium/BSA at a density of 5 × 10 cells/ml. 100 μl (5 × 10 cells) of cell suspension was added to the top chamber, whereas Dulbecco's minimal essential medium/BSA containing either 20 ng/ml HGF (Reprotech, Inc.) or 5% FBS was added to the lower chamber as a chemoattractant. Cells were allowed to invade the Matrigel and migrate to the underside of the invasion chamber for 20 h at 37°C in 5% CO. Cells on the top surface of the chamber were removed by gentle scrubbing with a cotton swab, and cells on the underside were stained with crystal violet and counted. Control experiments established that no growth differences existed between all cell lines tested under the conditions of this assay. Data from several experiments were expressed as percentage of control and represent the mean ± SEM of at least three independent determinations performed in duplicate. One, two, and three asterisks represent P < 0.05, P < 0.01, and P < 0.001, respectively ( test, or one way ANOVA followed by post-hoc comparisons using the Newman-Keuls test). The H-1152 Rho kinase inhibitor (Calbiochem) was used in some experiments at 1.6 nM, which is the reported Ki for this compound. Cells were harvested using Cell Stripper, washed twice in PBS, and resuspended at 1 × 10 cells/ml in Dulbecco's minimal essential medium. 3 × 10 cells in 300 μl of media were then cultured in 4-well chamber slides (Nunc). 24 h later, cells were washed again with PBS and supplemented with serum-free media for 12 h. Confluent cell monolayers were scratched using a 200-μl Finnpipette tip, and serum-free medium containing 20 ng/ml HGF was added to the cells. Migration of cells into the wound was monitored in multiple wells using a live cell imaging workstation (AS-MDW; Leica) with a 20×/04 N Plan objective. Images were captured every 60 min, and images shown represent 0 and 12 h after HGF addition. Cells were tested for their ability to aggregate in hanging drop suspension cultures, as previously described (). In brief, cells were suspended using Cell Stripper, washed in PBS, and resuspended in Dulbecco's minimal essential medium. 1.5 × 10 cells in 30 μl of media were suspended as hanging drops from the lid of a 24-well culture dish and allowed to aggregate overnight in a humid 5% CO incubator at 37°C. Aggregation was assessed 18 h after plating. To assay for tightness of cell–cell adhesion, cells were subjected to shear force by passing them 10 times through a standard 200-μl Finnpipette tip. Cells were photographed within 10 min through the AS-MDW live cell imaging workstation using a 10× phase-contrast objective. MDA-231 cells were plated in 100-mm dishes. 18 h later, the cells were washed and incubated for an additional 12 h in serum-free media, and RhoA or Rac1 activities were determined as described previously (). In some cases, serum-deprived cells were treated with 20 ng/ml HGF for the indicated times before cell lysis. Cells were lysed for 5 min at 0°C in 500 μl of lysis buffer (20 mM Hepes, pH 7.5, 0.5% NP-40, 100 mM NaCl, 0.2% deoxycholic acid, 10% glycerol, and 10 mM MgCl) supplemented with protease and phosphatase inhibitors. Lysates were clarified with a 5-min microcentrifugation, and supernatants were transferred to new tubes containing 30 μg of either Rhotekin RBD or PAK-1 PBD (Upstate Biotechnology) bound to glutathione beads. A 20-μl aliquot of supernatant was also saved for the determination of total RhoA/Rac1 and p120 levels in each sample. After a 45-min incubation at 4°C, beads were washed in wash buffer (20 mM Hepes, pH 7.5, 0.5% NP-40, 100 mM NaCl, 10% glycerol, and 10 mM MgCl), and bound RhoA- or Rac1-GTP, as well as total RhoA/Rac1, were visualized after SDS-PAGE and Western blotting using either a RhoA-specific mAb (26C4; Santa Cruz Biotechnology, Inc.) or a Rac-1–specific mAb (BD Biosciences). GTPγS- or GDP-labeled cell lysates were used as positive and negative controls, respectively.

Cell–cell interactions of selectins with glycosylated ligands mediate leukocyte tethering to and rolling on vascular surfaces during inflammation and immune responses (). L-selectin is expressed on leukocytes, whereas P- and E-selectin are expressed on activated platelets and/or endothelial cells. The hydrodynamic environment of the circulation imposes kinetic and mechanical constraints on selectin–ligand interactions. For a flowing cell to tether, a selectin must encounter its ligand and interact with it before flow again separates the molecules. For a cell to roll, new interactions must form at the leading edge to replace those that dissociate at the trailing edge. Force applied to these interactions affects their dissociation rates and, hence, their lifetimes. L-selectin requires a counterintuitive threshold shear to mediate both tethering and rolling (; ). As flow increases to an optimal level, more cells tether and the cells roll more slowly. Above the flow optimum, fewer cells tether, and the cells roll more rapidly. Distinct physical mechanisms regulate flow-enhanced tethering and rolling. Transport augments tethering through the following three mechanisms: sliding of the cell bottom on the surface, Brownian motions of the cell, and rotational diffusion of L-selectin and its ligand (). As flow increases, these mechanisms increase the collision frequency between L-selectin and its ligands, which favors productive interactions because the intrinsic docking rate is very rapid. Above the flow optimum, the tethering rate declines as the encounter times become shorter than the time scale for docking, and thus become limiting. It is not known whether, and if so how, changes in the structure of a selectin might affect its molecular diffusivity. Force augments rolling by decreasing the dissociation of L-selectin from its ligands. Normally, forces shorten the lifetimes of receptor–ligand interactions (slip bonds; ). However, at low levels force paradoxically prolongs the lifetimes of selectin–ligand interactions (catch bonds) before they convert to slip bonds at higher forces (; ). Catch bonds are particularly evident for L-selectin (). As flow increases from the threshold to an optimal value, rolling becomes slower and more regular as force lengthens the lifetimes of L-selectin catch bonds (). Above the flow optimum, rolling becomes faster and less regular as higher forces shorten the lifetimes of slip bonds. Several models explaining transitions from catch to slip bonds have been proposed (). Studies of bacterial variants support a model in which force applied to a linker region in the adhesin FimH allosterically regulates ligand binding (), perhaps by generating catch bonds (). However, the structural basis for catch bonds remains poorly understood. Each selectin has an N-terminal C-type lectin domain, followed by an EGF-like module, a series of short consensus repeats, a transmembrane domain, and a cytoplasmic tail (). Crystal structures of the lectin and EGF domains of P- and E-selectin have been published (; ). The ligand-binding region is a broad shallow surface at the top of the lectin domain opposite to where the EGF domain is attached (; ). This region includes a Ca-coordination site that is shared with the fucose in sialyl Lewis x (sLe; NeuAcα2-3Galβ1-4[Fucα1-3]GlcNAcβ1-R), which is a capping structure on glycans of selectin ligands. The lectin domain forms other contacts with sialic acid and galactose, as well as with the sulfated components of some glycoproteins. P- and L-selectin bind to the N-terminal region of the leukocyte mucin P-selectin glycoprotein ligand-1 (PSGL-1) through cooperative interactions, with sLe capping a core 2 O-glycan, and with adjacent sulfated tyrosines and other amino acids (, ; ). L-selectin also binds to the peripheral node addressin, which is a group of mucins on high endothelial venules of lymph nodes. The major binding determinant on the O-glycans of these mucins is 6-sulfo- sLe, a form of sLe with a sulfate ester attached to the C-6 position of GlcNAc (). There are only a few noncovalent interactions between the lectin and EGF domains, most of which are conserved among the three selectins. Two P-selectin structures with different angles between the lectin and EGF domains have been described (; ). The structures suggest that the P-selectin lectin domain can pivot on a hinge over the EGF domain. We postulated that this conformational change is common to all three selectins and plays an important role in regulating the kinetic on/off-rate of selectin–ligand interactions. We show that eliminating a hydrogen bond to increase the flexibility of the interdomain hinge in L-selectin reduced the shear threshold for adhesion by increasing tethering through greater rotational diffusion and by strengthening rolling through augmented catch bonds with longer lifetimes at smaller forces. Thus, allosteric changes remote from the ligand-binding interface regulate both bond formation and dissociation. At the putative hinge region of P-selectin, Tyr37 of the lectin domain is located close to Gly138 of the EGF domain (). As previously proposed (), the lack of a side chain in Gly138 should favor flexibility between the domains. In contrast, E- and L-selectin have an Asn at residue 138, and published structures of E-selectin reveal a hydrogen bond between Tyr37 and the side chain of Asn138 in a closed-angle conformation (; ). We have solved a crystal structure of the lectin and EGF domains of L-selectin (unpublished data), which also has a hydrogen bond between Tyr37 and Asn138 in a closed-angle conformation (, left). We used molecular dynamics (MD) simulations to examine whether L-selectin can assume an open-angle conformation (Videos 1 and 2, available at ). The closed-angle structure of L-selectin aligned well with the closed-angle structure of P-selectin (), with only a 1.4-Å root mean square distance (RMSD) between the corresponding backbone atoms from residues 1–156. Five of seven L-selectin free-dynamics simulations (each for 6 ns) exhibited only small conformational fluctuations around the crystal structure, suggesting that the closed-angle conformation is stable (Video 1). However, transitions between the closed-angle and open-angle conformations were observed in two simulations (Video 2). The dynamics of conformational transitions were quantified by the time courses of RMSD between the corresponding backbone atoms of the simulated L-selectin structure and the closed-angle L-selectin crystal structure or the open-angle P-selectin crystal structure (). A RMSD value of ∼2 Å indicated good alignment, and a RMSD value of ∼15 Å indicated poor alignment. After 2.5 ns of free dynamics, the simulated L-selectin structure changed from the original closed-angle conformation to an open-angle conformation (, right), which lasted for 1.5 ns and aligned well with the open-angle crystal structure of P-selectin (). These results suggest that L-selectin primarily resides in the closed-angle conformation, but occasionally makes spontaneous transitions to the open-angle conformation. MD simulations also indicated that force applied to unbind PSGL-1 from P-selectin promoted transitions from the closed-angle to the open-angle conformation (unpublished data). Thus, the closed-angle and open-angle conformations are in dynamic equilibrium; force can shift this equilibrium to a higher fraction of time in the open-angle conformation and a lower fraction of time in the closed-angle conformation. The force required to set the same conformational equilibrium between the closed and open interdomain angle should be higher for L-selectin than for P-selectin, because it must disrupt the Tyr37–Asn138 hydrogen bond and overcome steric interference in the hinge to enable transition from the closed-angle to open-angle conformation. How might the hinge flexibility of a selectin affect its interactions with ligand at an interface several nanometers from the hinge? A flexible hinge will increase the frequency of transitions between the closed- and open-angle conformations. This will facilitate rotational diffusion of the lectin domain, thereby contributing to the on-rate for ligand binding. Substituting Gly for Asn138 in L-selectin (L-selectinN138G) will eliminate the hydrogen bond between Try37 and Asn138. This might increase rotational diffusivity and enhance cellular on-rate. To test this hypothesis, we compared the tether rate, which is a metric of cellular on-rate, of microspheres bearing recombinant L-selectin or L-selectinN138G as they flowed over two distinct immobilized ligands. Microspheres with a 3-μm radius () bearing L-selectin or L-selectinN138G at a site density ( ) of 750 μm in media of 1 cP viscosity (μ) were flowed at different wall shear rates () through a PSGL-1–coated chamber at a site density ( ) of 120 μm or coated with 6-sulfo-sLe at a density that supported rolling, but could not be precisely measured with IgM mAb. Tether rates plotted against the product exhibited a biphasic shape characteristic of the shear threshold phenomenon, increasing initially, reaching a maximum, and then decreasing with further increase in (). As predicted by our hypothesis, the tether rates were higher for L-selectinN138G than for L-selectin at all values tested for both ligands, whereas the values where tether rates reached maximum were not significantly different. /(6πμ), where is the Boltzmann constant and is the absolute temperature), and molecular diffusion (proportional to /(6πμ), where is a characteristic length in the molecular scale; ). = −ln (1 − )/ ≈ /, where (= 224 μm) is the length of the microscopic field of view. to remove the mass action effects caused by different numbers of interacting molecules in the contact area. /( ) is determined by , /(6πμ), and /(6πμ). /(6πμ) and /(6πμ) are used as respective metrics for microsphere and molecular diffusivity, based on the Stokes-Einstein relationship. In the previous study, we independently varied these values by using microspheres of different radii ( = 1, 2.25, and 3 μm) and media of different viscosities (μ = 1, 1.8, 2.6, and 4.2 cP). /( ) increased with until it reached maximum, when microsphere Brownian motions became the limiting transport mechanism. Increasing microsphere Brownian motions enabled sliding to further enhance tether rate (reaching maximum at higher ). /( ) curve reached maximum (optimal ) increased linearly with /(6πμ) (). /( ) versus curve increased linearly with /(6πμ) (). These previously defined linear relationships were used as calibrations to estimate the difference in molecular diffusivities of L-selectin and L-selectinN138G in the present study. As expected, microspheres bearing L-selectin or L-selectinN138G had comparable optimal values that were similar to the previous data in the calibration curve because they had the same diffusivities calculated from the same radius (3 μm) and media viscosity (1 cP; ). The characteristic length for molecular diffusion was taken as = 100 nm in the data in , which serves as an order-of-magnitude estimate. /(6πμ) = 2.18 μm/s could be calculated from the same value for L-selectin in this study. /( ) (= 1.56 × 10 μm) as a y-axis coordinate, this L-selectin datum agreed with the previous data measured in media with 1-cP viscosity, regardless of the microsphere radius (). We hypothesized that the higher tether rate of microspheres bearing L-selectinN138G rather than L-selectin was caused by increased relative molecular diffusivity for L-selectinN138G. /( ) (= 3.08 × 10 μm), a relative molecular diffusivity of 4.02 μm/s for L-selectinN138G was extrapolated from the calibration curve, which was increased by 85% over that of L-selectin (). To further test our hypothesis, we designed a set of conditions to counter the predicted increase in the L-selectinN138G diffusivity. The medium viscosity was increased from 1 to 1.8 cP to reduce the diffusivity of L-selectinN138G by 80%. The radius of the microspheres was decreased from 3 to 1.5 μm to keep the product μ (and hence the microsphere diffusivity) approximately constant. constant. Because the microsphere diffusivities, molecular diffusivities, and normalizing factors were matched, we predicted that the tether rate versus curve for L-selectinN138G measured under the designed conditions would match the tether rate versus curve for L-selectin measured in 1-cP viscosity media with 3-μm radius microspheres. This was, indeed, the case (). The tether rate curves for 6-sulfo-sLe () were qualitatively similar to those for PSGL-1 (), but the former curves had smaller maximum tether rates that occurred at higher . Such quantitative differences reveal the impact of molecular docking rates specific to the ligands. /( ) value significantly below the calibration curve in , because the calibration curves were based on interactions of L-selectin with PSGL-1 rather than 6-sulfo-sLe. /( ) value from the L-selectin–6-sulfo-sLe tether rate curve to the calibration curve. Nevertheless, increasing rotational diffusivity of L-selectinN138G should augment tethering to molecularly distinct ligands because this mechanism does not require alterations of the ligand-binding surface. In the preceding paragraph, we estimated an 85% increase in diffusivity of L-selectinN138G over that of L-selectin by comparing the tether rate curves for interactions with PSGL-1 (). If this value represents the true difference in rotational diffusivities between the two selectins, instead of a fortuitous value that enabled curve fitting, it should account for the higher tether rate curve of L-selectinN138G over L-selectin for 6-sulfo-sLe (), as well as for PSGL-1 (). This hypothesis predicts that designing conditions to counter the increase in L-selectinN138G diffusivity would also align the designed L-selectinN138G curve with the original L-selectin curve for 6-sulfo-sLe. To test this prediction, we increased the medium viscosity from 1.0 to 1.8 cP and reduced the radius of the microspheres from 3.0 to 1.5 μm. Because the density of 6-sulfo-sLe could not be precisely measured, we increased the L-selectinN138G density from 750 to 1,500 μm. As predicted, the tether rate curve for these designed conditions aligned with the tether rate versus curve for L-selectin measured in 1-cP viscosity media with 3-μm radius microspheres (). Collectively, these data quantitatively demonstrate that eliminating the Tyr37–Asn138 hydrogen bond enhances cell tethering to different ligands by increasing the rotational diffusion of L-selectin. A flexible hinge might also affect force-dependent dissociation of L-selectin from its ligands. At low applied forces, the interdomain angle should remain mostly closed, as suggested by our MD simulations (; and Videos 1 and 2). Noncovalent interactions with the ligand may dissociate as it detaches from the lectin domain at the interface that is perpendicular to the direction of force (). As applied force increases, the equilibrium between the interdomain angles should shift toward a higher probability in the open conformation, which tilts the interface to align with the force direction, as suggested by MD simulations of the unbinding of P-selectin from PSGL-1 (unpublished data). Consequently, the ligand may slide across the lectin domain as preexisting interactions dissociate (). As observed in MD simulations of the unbinding of P-selectin from PSGL-1 (unpublished data), the sliding motion provides an opportunity for new interactions to replace those that are disrupted, or for the original interactions to reform before the ligand fully dissociates, thereby slowing dissociation. This force-dependent deceleration of dissociation is a hallmark of catch bonds (; ). Thus, a flexible hinge between the lectin and EGF domains may allow force to allosterically elicit catch bonds with ligand by sliding and rebinding. Once the interdomain angle is fully open, further increases in force can no longer increase rebinding, resulting in the transition from catch bonds to slip bonds. Force-dependent sliding of ligand over a pivoting lectin domain might at least partially explain why P-selectin, whose interdomain hinge is predicted to be more flexible than that of L-selectin, forms augmented catch bonds with longer lifetimes that convert to slip bonds at lower force than L-selectin (; ). We have formulated this sliding–rebinding mechanism into a mathematical model (see Materials and methods) whose solution exhibits catch–slip transitional bonds that fit the respective lifetime versus force relationships observed for P- and L-selectin interacting with PSGL-1 (; ; unpublished data). The sliding–rebinding model also predicts that substituting Gly for Asn138 in L-selectin (L-selectinN138G) will reduce the force required to elicit catch bonds, prolong their lifetimes, and lower the force where catch bonds convert to slip bonds, even with molecularly distinct ligands. To test these predictions, we measured how force affected the lifetimes of interactions of L-selectin and L-selectinN138G with PSGL-1 and 6-sulfo-sLe. Both biomembrane force probe (BFP) and flow chamber experiments were used to obtain complementary data. For BFP experiments, interactions of L-selectin or L-selectinN138G coated on a target bead with PSGL-1 or 6-sulfo-sLe coated on a probe bead, were stressed through a red blood cell to allow lifetime measurements at various levels of constant force. For flow chamber experiments, microspheres displaying each selectin were perfused at various wall shear stresses over low densities of PSGL-1 or 6-sulfo-sLe. The lifetimes of transient tethers were measured by high-speed video microscopy. A large number of lifetime measurements were used to derive the mean lifetime (which equals the reciprocal off-rate 1/ for first-order dissociation of single bonds) for each interaction at each tensile force (for BFP) or each wall shear stress (for the flow chamber). Both methods yielded similar results. As previously observed (; ), the lifetimes of L-selectin interactions with both ligands demonstrated a biphasic pattern characteristic of transitions from catch to slip bonds (). Initial increases in force prolonged mean lifetimes until an optimal value was reached; further increases in force shortened lifetimes. Although L-selectinN138G interactions with both ligands also exhibited transitions between catch and slip bonds (), the lifetimes in the catch bond regime were significantly longer and the transitions to slip bonds occurred at lower forces. In contrast, there was little difference in the lifetimes of L-selectin and L-selectinN138G interactions in the slip bond regime. The lifetime versus force curves were sensitive to the selectin and ligand used, although both selectins were captured by the same antibody and both biotinylated ligands were captured by streptavidin (see Materials and methods). The present L-selectin–PSGL-1 data also agree quantitatively with previous measurements using different capturing methods () and using neutrophils expressing membrane-bound L-selectin (). This strongly suggests that the data in are dominated by lifetimes of selectin–ligand bonds rather than of antibody–antigen or biotin–streptavidin bonds. Because we demonstrated that L-selectin interacts as a monomer in the previous measurements (), the quantitative agreement with the current data suggests that they also represent monomeric interactions. However, the sliding–rebinding model of can be translated to dimeric bonds (). Although the crystal structure of the P-selectin–PSGL-1 complex reveals many noncovalent interactions at the atomic level (), they were simplified into two pseudoatomic interactions in this study to model the sliding–rebinding mechanism of forced-dissociation of a selectin–ligand complex (). and unbinds at a dissociation rate that obeys the Bell equation (), with a stress-free dissociation rate of and an energy well of width . < < . (see Materials and methods for details). Curve fitting by the model yielded excellent agreement with the data (). ) of the six fitting parameters could be the same for the curves of both selectins interacting with the same ligand. ) to fully open the interdomain angle and exhibited a higher rebinding rate (larger ). Thus, both experimental data and their theoretical fits strongly support the model for transitions from catch to slip bonds through force-dependent sliding of the ligand over a pivoting lectin domain that promotes rebinding. Although alterations in the hinge region might propagate conformational changes across the lectin domain to the ligand-binding interface, the model does not require such changes. To determine whether the augmented catch bonds lowered the shear threshold for L-selectin-dependent rolling, we perfused microspheres bearing each selectin over higher densities of PSGL-1 or 6-sulfo-sLe. Rolling motions were visualized by high-speed video microscopy. As wall shear stress increased, the mean rolling velocities of microspheres displaying either selectin first decreased and then reached a minimum (), a characteristic of the shear threshold phenomenon that is mediated by catch bonds (). At higher shear stresses, the rolling velocities again increased as catch bonds converted to slip bonds. Remarkably, the descending phases of the rolling velocity curves for L-selectinN138G shifted downward and to the left, with significantly slower mean velocities and with minimal velocities at much lower shear stresses than for L-selectin. Rolling motions at these suboptimal flow rates were more regular for L-selectinN138G than for L-selectin, with longer mean stop times () and higher fractions of time in the stop phase (). In contrast, the ascending phases of the rolling velocity curves were similar for both selectins. These data confirm that substituting Gly for Asn138 lowers the shear threshold for L-selectin-dependent rolling. An unexplained property of circulating leukocytes is that they do not aggregate, even though they express both L-selectin and its ligand PSGL-1. We hypothesized that specific kinetic and mechanical properties of L-selectin–PSGL-1 interactions prevent leukocyte aggregation. Because of these properties, randomly colliding leukocytes might form bonds between L-selectin and PSGL-1, but the number of bonds would be too small and/or their lifetimes would be too short at small forces for stable adhesion. This hypothesis predicts that increasing interdomain flexibility will cause spontaneous aggregation of leukocytes because both bond formation and bond lifetime are increased. To test this hypothesis, we flowed neutrophils or mixtures of neutrophils and microspheres bearing L-selectin or L-selectinN138G in a shear field to promote collisions. Cells or microspheres were perfused in a flow chamber coated with the nonadherent protein human serum albumin (HSA) at a wall shear stress of 1 dyn/cm. This shear stress level applies sufficient force to bonds at the trailing edge of a rolling cell, thereby promoting optimal L-selectin–dependent rolling on PSGL-1 when it is coated in the flow chamber (), but should exert much smaller forces on bonds bridging cells or microspheres in flowing aggregates (). High-speed video microscopy revealed that some free-flowing neutrophils formed doublets after they collided. However, doublet lifetimes were very brief, and the interacting cells dissociated within 0.02 to 0.03 s ( and Video 3, available at ). Doublets of neutrophils and L-selectin microspheres were similarly short lived ( and Video 4). In contrast, neutrophils formed stable doublets with L-selectinN138G microspheres that persisted until they flowed out of the field of view ( and Video 5). To quantify this phenomenon, we perfused mixtures of neutrophils and microspheres that were labeled with different fluorescent dyes and fixed the mixtures after they exited the flow chamber. Flow cytometry revealed very few particles containing fluorescence markers for both neutrophils and L-selectin microspheres (), but many particles labeled for both neutrophils and L-selectinN138G microspheres (). Addition of the anti–L-selectin mAb DREG-56 blocked aggregate formation, confirming that aggregates developed through engagement of L-selectinN138G (). Similarly, fluorescence microscopy revealed few aggregates between neutrophils and L-selectin microspheres (), whereas neutrophils and L-selectinN138G microspheres formed many doublets and larger aggregates (). Such aggregates did not develop in the presence of DREG-56 (). These data suggest that a subtle change in the interdomain hinge of L-selectin is sufficient to cause flowing leukocytes to aggregate. Leukocyte adhesion to vascular surfaces requires that the kinetics of formation and dissociation of selectin–ligand bonds be regulated by the mechanics of blood flow. The shear threshold for tethering and rolling through L-selectin provides a striking illustration of this regulation. Transport mechanisms, including molecular diffusion, govern flow-enhanced tethering, whereas force governs flow-enhanced rolling through catch bonds. How an atomic-level structure dictates such mechanisms was not understood. In this study, we used crystal structures and MD simulations to formulate a model for these mechanisms. MD simulations revealed that the hinge between the lectin and EGF domains of L-selectin can assume dual conformations that are in dynamic equilibrium but allow rapid transitions. Applied force or changes in hinge flexibility can alter the stability of the conformations, tilt the dynamic equilibrium, and induce conformational transitions. Structural analysis identified a single residue that contributes to the rigidity of the interdomain hinge by forming a hydrogen bond. Eliminating this hydrogen bond demonstrated that hinge flexibility regulates both the kinetic and mechanical properties of L-selectin. Thus, the fine tuning of hinge flexibility modulates flow-enhanced cell adhesion and likely prevents inappropriate leukocyte aggregation. Cells expressing a chimeric L-selectin, in which the native EGF domain was replaced with that of P-selectin, were shown to have a lowered shear threshold for adhesion because of an enhanced cellular on-rate that increased tethering to surface-bound ligand (, ). This property may have resulted, at least in part, from increased rotational diffusivity caused by substituting Gly for Asn138 in the chimeric molecule. Our results suggest that augmented catch bonds between the chimeric L-selectin and its ligands also may have contributed to the lowered shear threshold for adhesion. That L-selectinN138G promotes aggregation of microspheres with free-flowing neutrophils suggests that the kinetic and mechanical properties of L-selectin–PSGL-1 interactions must be finely regulated to allow rolling on vascular surfaces, but not aggregation of flowing leukocytes. This regulation is essential because both L-selectin and PSGL-1 are constitutively expressed on the surfaces of leukocytes. In contrast, platelets mobilize P-selectin on their surfaces only after they are activated, which normally does not occur until they adhere to damaged vessel walls (). Our model suggests that leukocytes will form aggregates with circulating activated platelets because the lifetimes of PSGL-1 bonds with P-selectin are longer than those with L-selectin at all force levels (; ), perhaps in part because of the increased interdomain flexibility of P-selectin conferred by Gly138. Indeed, circulating platelet-leukocyte aggregates are observed in pathological disorders that increase platelet activation (). Our findings exemplify how force-induced conformational changes allosterically regulate protein function, which may be applicable to other proteins. The sliding–rebinding model provides a structural explanation for catch bonds and may be applicable to interactions of other proteins with an interdomain hinge. For example, the adhesin FimH mediates shear stress-enhanced adhesion of bacteria to epithelial cells (), which may result from catch bonds between FimH and its mannosylated ligands (). MD simulations revealed that force applied at the ligand-binding site of the lectin domain extends a linker chain that connects the lectin domain to the adjacent pilin domain. Mutations in this segment decrease the shear threshold for bacterial adhesion (). Such mutations might make the linker region more flexible and reduce the force required to slide ligand across the binding interface. The increased flexibility could also favor bond formation through greater rotational diffusion. Interactions of glycoprotein Ibα with von Willebrand factor mediate flow-enhanced adhesion of platelets to damaged vascular surfaces (; ). Catch bonds might contribute to flow-dependent platelet adhesion, and a sliding–rebinding mechanism might explain why mutations in glycoprotein Ibα or von Willebrand factor that are remote from the interface can alter binding (; ). Force-induced allosteric changes may also regulate catch–slip transitional bonds between actin and myosin () and the conversion of integrins to their active conformations (; ). L-selectin–Ig containing the lectin domain, EGF domain, and both consensus repeats of human L-selectin fused to the Fc portion of human IgG1 was expressed as previously described (). The cDNA encoding L-selectin was used as a template to alter the codon for Asn138 to Gly, using the QuikChange Mutagenesis kit (Stratagene). The mutations in the construct, termed L-selectinN138G, were confirmed by DNA sequencing. L-selectinN138G-Ig was expressed as described for L-selectin–Ig. The aminopropyl glycoside of 6-sulfo-sLe was biotinylated as previously described (). Soluble recombinant monomeric PSGL-1, anti–PSGL-1 mAb PL1, and anti–L-selectin mAb DREG-56 have been previously described (). Anti–human IgG Fc polyclonal antibody was obtained from CHEMICON International, Inc. A crystal structure of the lectin and EGF domains of human L-selectin served as the starting coordinate for MD simulations, except that the glycan attached to residue Asn66 was deleted to reduce the system size and to avoid the use of less reliable glycan force fields. Because this glycan extends out from Asn66 on the protein surface, its deletion is unlikely to affect the structure of the lectin and EGF domains. The molecule was solvated in a 90 × 60 × 60 Å TIP3 water box, together with 8 Ca and 18 Cl ions to neutralize the system, which included 29,969 atoms. MD simulations were performed using NAMD () with a CHARMM22 all-atom force field (). The system was first subjected to a two-step energy minimization. Each step consisted of 10,000 conjugate gradient iterations. In the first step, the heavy atoms of the protein were fixed and the rest of the atoms were allowed to move. In the second step, all atoms were allowed to move. After energy minimization, the system was heated gradually from 0 to 300 K in 100 ps. Then the system was equilibrated for 1 ns with pressure and temperature control. The temperature was held at 300 K using Langevin dynamics and the pressure was held at 1 atm by the Langevin piston method. Free dynamics were then simulated with the equilibrated system for 5 ns. A periodic boundary condition was adopted. Particle Mesh Ewald was used for electrostatic interactions, and a cutoff of 12 Å was used for van der Waals interactions. A total of seven independent simulations were performed, following the same procedure. The trajectory of each simulation was compared with the initial L-selectin crystal structure and with the P-selectin open-angle crystal structure (PDB 1G1S; ). The sliding–rebinding model can be formulated with increasing complexity, depending on the number of pseudoatomic interactions assumed to describe the noncovalent interactions distributed across the selectin–ligand interface. A minimal model was solved by Monte Carlo simulations. The simulation starts with two pseudoatomic interactions () and advances in 1-μs time steps (). To determine what would happen in each time step, a random number (uniformly distributed between 0 and 1) is compared with one of the probabilities given by , , , and , depending on the stage of the simulation. In the first stage, the dissociation from the initial bound state is simulated. The fate of each of the two pseudoatomic interactions in the current time step is determined using one of two probabilities, depending on whether that pseudoatomic interaction survived in the previous time step. If so, it would survive if the random number were smaller than the probability,or it would dissociate. If not, i.e., if in the previous time step the pseudoatomic interaction in question dissociated, but the other pseudoatomic interaction survived, it would associate if the random number were smaller than the probability,or it would remain dissociated. is a force-dependent dissociation rate that obeys the Bell equation (), as follows:where and are model parameters, is applied force, and = 1 or 2, depending on whether one or both pseudoatomic interactions were intact in the previous time step. After both preexisting pseudoatomic interactions dissociate, the simulation advances to the second stage, which simulates sliding and the formation of a new interaction. In the next simulation step, a new pseudoatomic interaction would form if the random number were smaller than the probability, as follows:or if the ligand would dissociate. relates the probability of forming a new interaction to force through the interdomain angle. , the interdomain angle would be confined to the closed conformation (as depicted in the top left portion of ), so no new interaction would be allowed. , i.e., compressive force. < < , force shifts the equilibrium between the two hinge conformations toward what is more likely to be in the open-angle conformation, allows the binding interface to slide more readily, and increases the probability of forming a new pseudoatomic interaction (as depicted in the top right portion of ). , the interdomain angle reaches maximum and can increase no further (i.e., = 1). , i.e., tensile force. Formation of a new interaction advances the simulation to the third stage, where two tests are performed in the next time step. One compares a random number to the probability given by to determine whether the new interaction would survive. The other test determines whether the original double-pseudoatomic interactions would reform (as depicted in ). is a constant rebinding rate, which returns the simulation back to the first stage. The simulation continues until the ligand dissociates from the selectin when no new interaction forms after both old interactions dissociate or the new interaction dissociates before rebinding occurs. The accumulated time steps are the lifetime of the selectin–ligand complex in that simulation run, which is repeated 1,000,000 times for a given force to obtain an ensemble of exponentially distributed lifetimes and their average at that force. The six model parameters ( , , , , , ) were adjusted to obtain a mean lifetime versus force curve that fit the experimental data for a given selectin–ligand pair. For each ligand interacting with two L-selectin molecules, the first three parameters ( , , and ) describing the rates of association and dissociation of the pseudoatomic interactions were kept the same for L-selectin and L-selectinN138G. Of the two parameters describing the probability ( ) of forming a new pseudoatomic interaction after sliding, one ( ) was also kept invariant. The other parameter ( ) had to be smaller for L-selectinN138G than for L-selectin, to reflect the more flexible interdomain hinge of L-selectinN138G. For the same reason, the rebinding rate ( ) had to be larger for L-selectinN138G than for L-selectin. . (compared with ) results from the greatly reduced number of interactions assumed and the same kinetic rates assumed for the new interaction formed after sliding as the preexisting interactions. values. Our in-house–built BFP apparatus is a duplicate of that developed in the laboratory of E. Evans (University of British Columbia and Boston University; ). L-selectin-Ig or L-selectinN138G-Ig was captured by goat anti–human Fc antibody covalently precoupled to the target bead (3–4 μm in diameter), as previously described (). The same protocol, but without the step of linking proteins to polyethyleneglycol polymers, was used to couple streptavidin-maleimide (Sigma-Aldrich) to the probe bead (2 μm in diameter), which captured either biotinylated PSGL-1 or 6-sulfo-sLe. The streptavidin also attached the probe bead to a biotinylated red blood cell, which, when pressurized by micropipette aspiration, served as an ultrasensitive force transducer. Low densities of selectins and ligands ensured infrequent adhesion (30%), which was specific, as EDTA and anti–L-selectin mAb DREG-56 abolished adhesion. The in-house online image analysis software tracked the red blood cell deflection with a 0.6-ms temporal resolution and 5-nm spatial resolution, which, for a spring constant of 0.3 pN/nm, translates to 1.5-pN force resolution. Driven by a computer-controlled piezoelectric translator, the force-clamp test cycle consisted of an approach of the target bead (1,500 nm/s) to touch the probe bead, a gentle (15 pN) and brief (0.1 s) contact period, a retraction of the target bead to load the selectin–ligand bond (if a bond was formed) at 1,000 pN/s to the desired level of force, and a waiting period during which the bond (if the bond survived ramping) was subject to a constant force until dissociation, which was then repeated thousands of times. The bond lifetime was measured from the moment when the bond force reached the desired level to the moment when the bond dissociated. A total of ∼800 lifetimes were measured at forces ranging from 3–90 pN for each interaction of L-selectin or L-selectinN138G with PSGL-1 or 6-sulfo-sLe, which were segregated into 7–9 force bins. For each bin, the natural log of the number of measurements with a lifetime greater than was plotted versus , which exhibited linear decay as predicted by first-order dissociation kinetics, except for longer lifetimes of probable multiple bonds that represented <8% of the total interactions, which were excluded as outliers. The reciprocal of the negative slope was found equal to the mean and SD of lifetime, as predicted by first-order dissociation kinetics. The mean ± the SEM of lifetimes in each bin were plotted against force for each interaction, as shown in . Each selectin-Ig was captured on polystyrene microspheres (6- or 3-μm diam; Polysciences, Inc.) coated with anti–human Fc polyclonal antibody (). Matched densities of each selectin were confirmed by flow cytometry (). Biotinylated PSGL-1 or 6-sulfo-sLe was captured on streptavidin (Pierce Chemical Co.) adsorbed to flow chamber floors (). Microspheres (2 × 10/ml in HBSS containing 0.5% HSA) were perfused in media without or with 3% (wt/vol) Ficoll (molecular weight 400,000; Sigma-Aldrich) at various flow rates over PSGL-1 or 6-sulfo-sLe in a parallel-plate flow chamber (). The viscosities of the media without and with Ficoll were 1.0 and 1.8 cp, respectively, at room temperature as previously described (). Images were captured with a digital video camera (Fastcam Super 10 K; Photron) at 250 frames/s. Tether rates were measured by a previously described method (; ). The tether rate, , was calculated by normalizing the number of observed tethering events in 1 min by the total number of microspheres flowing through the field of view in the same focal plane in the same period of time (). Mean rolling velocities were measured by tracking individual microspheres frame by frame. Rolling step analysis was performed with custom-designed macros prepared in Excel (Microsoft; ). In some experiments, microspheres were perfused in media containing 20 μg/ml DREG-56 or PL1, or 10 mM EDTA. All tethering and rolling events were specific because they were eliminated by inclusion of mAb or EDTA in the media. Transient tether lifetimes were measured on low densities of PSGL-1 or 6-sulfo-sLe that did not support rolling or skipping (). Images captured at 250 fps were replayed in slow motion, and durations of transient tethers were measured using frame-by-frame analysis. For each mean lifetime curve, five sets of lifetimes at each wall shear stress (∼100 tethering events in each set) were measured. At each wall shear stress, the exponentially distributed transient tether lifetimes in each set were averaged. The data are presented as mean ± SD of the five sets of average lifetimes. Microspheres were labeled with FITC-conjugated anti–human Fc antibody (Sigma-Aldrich). In some experiments, microspheres directly conjugated with FITC (Polysciences) were coated with unlabeled anti-human Fc antibody. After blocking with HBSS containing 1% HSA, L-selectin-Ig, or L-selectinN138-Ig was captured on the labeled microspheres. Isolated human neutrophils () were labeled with the fluorochrome PKH26 (Sigma-Aldrich). Neutrophils were mixed with microspheres bearing L-selectin or L-selectinN138G (final concentration of 2 × 10/ml for both cells and microspheres) in HBSS without Ca and Mg containing 0.5% HSA. Immediately before perfusion, 1 M CaCl and 1 M MgCl were added to the suspensions to achieve final concentrations of 2 mM Ca and 2 mM Mg. The cell/microsphere suspensions were perfused through a flow chamber coated with 1% HSA at a wall shear stress of 1 dyn/cm. In some experiments, 20 μg/ml DREG-56 was added to the suspensions. After exiting the flow chamber, samples were collected and fixed with 1% paraformaldehyde. Some samples were analyzed by flow cytometry without gate selection on a FACScalibur instrument (Becton Dickinson). Other samples were visualized in a fluorescence microscope (ECLIPSE E800; Nikon) connected to a CCD digital camera (Dxm1200; Nikon). Digital images were stored with Nikon software. FITC and PKH26 images were merged using Photoshop (Adobe). In some experiments, unlabeled neutrophils or mixtures of neutrophils with microspheres bearing L-selectin or L-selectinN138G were perfused at 1 dyn/cm. Video microscopy images of interactions among flowing neutrophils or microspheres were captured at 250 frames/s. The online material consists of five videos. Videos 1 and 2 are MD simulations of movements around the hinge between the lectin and EGF domains of L-selectin. Videos 3, 4, and 5 depict collisions of free-flowing neutrophils with other neutrophils or with microspheres bearing L-selectin or L-selectinN138G. Online supplemental material is available at .

I was actually a math major at first, but after taking some psychology courses, I got really interested in science. I switched my major to chemistry with the idea of going on to graduate school, but I first wanted to get a little bit of lab experience. So I went to work as a technician in a lab that was differentiating muscle cells from myoblasts into myofibers. I thought that was really exciting, that you could get cells to change their behavior in a culture setting. I wanted to know how cells understood where they were in the body and how they would know what to do based on their environment. Yes, in graduate school (at the University of Utah), I worked with breast tumor cells. We would culture them on matrigel, where they could form these little acini structures. At that point, I decided that I was really interested in the mechanisms that regulated the changes in cell shape and cell movement that would allow the cells to form those sorts of structures. Then I went to the University of North Carolina to work with Keith Burridge on Rho signaling and the regulation of the cytoskeleton and cell migration. A lot of their work was with fibroblasts and epithelial cells at the time. I decided to look at the migration of leukocytes because leukocytes are normally highly migratory cells, even in adults, whereas fibroblasts aren't as motile. In particular, I was interested in their migration during inflammation, when they interact with the lining of the blood vessel wall, penetrate, and invade across the endothelium. I am. We're also expanding on it to understand how the same sort of mechanisms regulate tumor cell migration during metastasis. This would be during the late stages of metastasis, when the tumor cell is circulating in the bloodstream. There are a lot of molecular parallels between tumor cell and leukocyte extravasation. Some of the adhesion molecules that mediate the interaction with the endothelial cells are the same. A few years ago, it also became clear that some tumor cells overexpress chemokine receptors that are normally found on leukocytes. Less than a year before, yes. My husband is also a faculty member here in the biochemistry department. We were here for about nine months, and then the storm hit. We had evacuated beforehand. We were in Houston when it hit, and then from what we saw on TV, we knew we weren't going to be going back anytime soon. My husband and I drove across the country to Oregon because that's home to both of us. We just got in the car and kept going. I would say most scientists who were displaced found labs to go and stay in. Many times, people went back to places where you already knew people, like the lab they had been in before, or a university where you had a collaborator. And there were a lot of offers of help within the scientific community to help house people temporarily. I was contacted by several people asking if I wanted to come and have space. But we actually declined for personal reasons. So my husband and I stayed with his parents for three months before we could go back to New Orleans. We both wrote grants, sitting at my in-laws' dining room table. Also, our house had flooded, so there were really a lot of issues to deal with outside of work. At that point I only had one research associate, and she stayed with family in Baton Rouge. Then she started working at a lab in Baton Rouge, where she had worked as an undergraduate. For instance, Andy Catling had previously worked in Michael Weber's lab at the University of Virginia. He went back there, and they made space for him. He took his whole lab and continued on with his own research there. Then there was Suresh Alahari. He did some collaborative work at different labs, where he was able to learn some techniques that he didn't know before. So for him it worked out almost like a sabbatical. The main thing was having to replace all your reagent stocks. We remade some, and then called up collaborators to get some, so we didn't have to make them all from scratch. We also had to move to new lab space. When we first came back to New Orleans, LSU was not open. So we were put up at Children's Hospital uptown in New Orleans, which is an area that did not flood. They opened space for several LSU researchers there. In June last year, the LSU Medical School reopened, and then I moved in here. That was initially also a temporary space, and since then we've moved again. So we've actually had our lab in three different places since the storm—all in New Orleans, but still it's disruptive. There was support in the form of supplements to grants so that you could replace all of the materials and time that you had lost. I had an American Heart Association grant, so I got a supplement from them. People who had NIH grants got supplements from NIH. These were basically special funding mechanisms to help you restore lost reagents. I would say it took until the beginning of this year to feel like we were hitting stride again. It took that long to feel like experiments were going as smoothly as they should be. We got tired of replacing everything, and then we started just replacing the things that we needed to do certain key experiments. We built up slowly. The other thing that slowed us down was personnel. I had hired a post-doc, but then that was canceled because of the storm. He went someplace else. Before I even felt comfortable to try to recruit after the storm, I needed to feel sure that the situation in New Orleans was stable. I finally hired a post-doc last October, but that was essentially a year delay. There are four of us right now. Andy Catling is working on MAP kinase signaling and scaffolding molecules and how that's involved in adhesion and migration. Suresh Alahari works on PAK kinase and on a protein that he discovered called Nischarin, which is down-regulated in advanced breast cancers. He's working on how those signaling proteins regulate adhesion and migration. My husband David uses x-ray crystallography to study the structures of proteins involved in Rho family signaling. We all get together every other week and have a group lab meeting. We borrow reagents, and we talk about experiments in the hallway all the time. We have some collaborative studies, but right now it's more that we're sharing ideas, as we've each got our own individual take on things. For me that makes the whole environment here good. All the recovery plans for the city call for developing health care industry and biomedical research as a driver of the economy in New Orleans. It adds to the traditional tourism industry. There's actually been a lot of support from the state to improve the facilities here. We've had the opportunity to purchase state-of-the-art equipment recently for imaging and proteomics facilities, to increase our technical capabilities. There's also a large recruiting effort, which we hope will add to our group. We've had several very talented people come through for interviews. I think there was some natural selection. The people who have remained here are tough and optimistic about the future. In general the attitudes that people have here are, I guess you could say, cautiously optimistic. Yes. I'm hoping that we're going to continue to recruit more people, because we did lose some faculty. I'm hopeful that we're going to keep building up on our cell adhesion and migration group as well, because it's a really great environment. I think we managed to convince our families that we weren't completely insane by staying in New Orleans!

53BP1, identified as a p53-interacting protein, is involved in DNA damage–induced checkpoint arrest. After ionizing radiation (IR) treatment, 53BP1 becomes progressively, yet transiently, immobilized around the double-strand break–flanking chromatin (). 53BP1 contains a 381-amino-acid region (1235–1616), called the kinetochore binding domain, which is essential for the accumulation of IR-induced 53BP1 foci (). Methylated lysine residues in the kinetochore binding domain modulate the accessibility of 53BP1 to the chromatin (; ). 53BP1 foci have also been detected in response to UV radiation, hydroxyurea (HU), camptothecin, and etoposide treatment (). We have recently shown that 53BP1 is involved in the recruitment of two important caretaker tumor suppressors, BLM and p53, to the sites of HU-induced stalled replication forks in S-phase (). Incidentally, like p53 and BLM, 53BP1 also displays a “hyper-rec” phenotype (). Hence, 53BP1 may lead to aberrant recognition of the stalled replication forks, resulting in chromosomal abnormalities, as observed in p5353BP1 mice (; ). BLM helicase has been proposed to function at the interface of replication and recombination and facilitate the repair of damaged DNA (). The major characteristic of BLM patients is elevated recombination events. BLM may regulate homologous recombination (HR) by modulating the functions of other proteins involved in the process. BLM resides in a nuclear matrix bound complex with prorecombinogenic protein RAD51 (). The direct interaction between BLM and RAD51 is evolutionarily conserved and has been proposed to have a role during recombinational repair. The focal expression of BLM and RAD51 has an inverse correlation in human cells (). Exposure of cells to replication stress and IR causes BLM to colocalize and physically interact with RAD51 at the sites of DNA damage (; ). Inactivation of p53 in Bloom syndrome (BS) cells causes a further increase of sister chromatid exchange (SCE), thereby demonstrating that p53 and BLM cooperatively affect HR (). Because BLM and 53BP1 have the potential to be functionally involved during replication stress, we sought to investigate whether and, if so, how these two proteins can affect HR by modulating the functions of RAD51. To determine whether 53BP1 interacts with BLM and RAD51 during replication stress, we performed reciprocal immunoprecipitation (IP; ) experiments on human telomerase reverse transcriptase (hTERT)–immortalized normal human fibroblasts (NHFs), left asynchronous (−HU), arrested in S-phase by HU (+HU), or allowed to proceed after washing away HU (postwash; Fig. S1 A, available at ). 53BP1 and RAD51 protein levels in the nuclear extract (whose integrity was verified; Fig. S1 B) remained unchanged in −HU, +HU, or postwash conditions. In contrast, BLM accumulated during +HU treatment, and the protein levels remained high during the postwash stage (). Importantly, both 53BP1 and RAD51 were present in the BLM IPs only during HU treatment (). IPs with 53BP1 () and RAD51 () antibodies confirmed the formation of a complex where BLM, 53BP1, and RAD51 were present. However, these results do not rule out the concurrent presence of 53BP1–RAD51 and BLM–RAD51 complexes during +HU condition. Low levels of 53BP1–RAD51 interactions also occurred during −HU and postwash conditions (). Because both 53BP1 and BLM interact with RAD51, there is a possibility that both of these proteins play a role in controlling HR and subsequent cell survival. To determine whether loss of 53BP1 alone could indeed cause a change in cell survival after stalling of the replication forks, embryonic stem cells from 53BP1 and 53BP1 mice were used (; , inset). A substantial increase in sensitivity to HU was observed in 53BP1 cells during clonogenic survival assays (), which may be due to increased chromosomal instability. To confirm this hypothesis, metaphase spreads were analyzed from asynchronous or HU-treated 53BP1 and 53BP1 cells (). A statistically significant increase in the number of chromosomal aberrations was observed in HU-treated 53BP1 cells, consistent with the possibility that 53BP1 is required in the regulation of chromosomal instability and cell survival during replication arrest. A role for 53BP1 in controlling genomic integrity and cell survival during S-phase may be due to its effect on HR. A role for 53BP1 in suppressing HR has been recently indicated (). It has also been reported that although spontaneous HR by gene conversion did not depend on 53BP1, class switch recombination was severely impaired by lack of 53BP1 (). However, the effect of 53BP1 on stress-induced HR has never been addressed. The effect of 53BP1 alone on HR was measured by SCE assays in 53BP1 and 53BP1 cells. Loss of 53BP1 resulted in a statistically significant increase in SCE level during replication stress (), indicating that 53BP1 has an antirecombinogenic effect on HR during replication stress. The antirecombinogenic effect of 53BP1 was subsequently verified in human cells by host cell reactivation assays in NHF, BS, and A-15 (corrected BS fibroblasts) cells. Overexpression of RAD51 in NHF cells led to the expected increase in HR (). Conversely, overexpression of wild-type 53BP1 led to a four- to fivefold decrease in the rate of HR, confirming its antirecombinogenic role in human cells. Interestingly, overexpression of wild-type 53BP1 in A-15 and BS cells cause a fourfold and twofold decrease, respectively. Hence, the interaction of BLM with 53BP1 was essential for the latter to attain its full antirecombinogenic potential. To determine whether and, if so, how BLM and 53BP1 together negatively regulate HR, we transfected BLM siRNA oligos into 53BP1 and 53BP1 cells (Fig. S2 A, available at ) and 53BP1 siRNA oligos into BS/A-15 cells (). Western analysis revealed that both BLM and 53BP1 siRNAs were able to acutely down-regulate the expression of their cognate proteins ( and Fig. S2 A). Cells with both BLM and 53BP1 expression (A-15 and 53BP1) exhibited low rates of HR, which increased 2–2.5-fold after replication arrest. BS cells or loss of BLM expression in 53BP1 cells led to an increase in the rate of HR. Loss of 53BP1 in either the mouse or human cells led to a HU-dependent increase in HR. In cells lacking the expression of both 53BP1 and BLM, the rate of HR increased synergistically in the presence of HU, thereby indicating that both BLM and 53BP1 together negatively regulated HR during replication arrest ( and Fig. S2 B). To further demonstrate that 53BP1 and BLM together regulate HR under physiological conditions, we needed to measure SCE in stable cell lines. 53BP1 and 53BP1 cells could not be used, as stable lines expressing mouse BLM siRNA sequences could not be generated. Hence, the stable lines were derived from BS and A-15 cells by stable expression of human 53BP1 siRNA (). BS cells exhibited a pronounced increase in both spontaneous and stress-induced SCE compared with A-15 cells (). Loss of 53BP1 alone (A-15 53BP1 cells) resulted in a three- to fourfold increase in SCE compared with A-15 only during HU treatment. However, BS 53BP1 cells exhibit statistically significant synergistic increase in the stalled replication–induced SCE rate, indicating that 53BP1 negatively regulates HR during stalled replication both alone and in cooperation with BLM. To determine how BLM and 53BP1 regulate HR, we hypothesized that the two proteins may be controlling RAD51 protein levels or its accumulation during HR. Although there was an increase in RAD51 protein level because of loss of BLM, there was no further change as a result of the subsequent loss of 53BP1 (Fig. S3 A, available at ), indicating different modes of action of BLM and 53BP1 on RAD51 function. RAD51 foci that are formed after stalled replication mark a subset of cells that have entered the HR pathway and contain functional recombination complexes (). It is possible that BLM and/or 53BP1 affected the formation of RAD51 foci. To investigate this possibility, BLM and/or 53BP1 were transiently transfected into BS 53BP1 cells and the rate of formation of endogenous RAD51 foci was determined. Transfection of either BLM or 53BP1 in HU-treated BS 53BP1 cells resulted in a decrease in the number of RAD51 foci (Fig. S3, B and C). These results indicate that both BLM and 53BP1 may affect the formation of RAD51 nucleoprotein filaments and thereby negatively regulate HR. Overexpression of RAD51 results in the formation of RAD51 nucleoprotein filaments reminiscent of presynaptic RAD51 complexes (). Adenoviral overexpression of RAD51 in HU-treated BS 53BP1 cells resulted in a two- to threefold increase in the level of RAD51 protein (Fig. S3 D) and formation of high-order filamentous nuclear structures up to 5 μm in length throughout the nuclei ( and Video 1, available at ). Around 80% of the cells infected with adenoviral RAD51 (Ad-RAD51) showed the appearance of these structures (). Overexpression of BLM or 53BP1 decreased the number of cells with complete RAD51 structures, instead showing either disrupted structures or diffused RAD51 staining (, A and B; and Videos 2 and 3). Both BLM and 53BP1 colocalized with the disrupted RAD51 structures (). Very few cells had RAD51 filaments when both 53BP1 and BLM were overexpressed (). Hence, apart from the BLM-independent role, 53BP1 also cooperates with BLM to negatively regulate HR. It will be interesting to know the biochemical parameters that govern the interactions among BLM, 53BP1, and RAD51. BLM is known to exert its antirecombinogenic functions in conjunction with other cellular proteins, like topoisomerase IIIα (). Additional interacting partners of (the yeast orthologue of BLM) have been identified in large-scale genetic screens (). Deletion of (the yeast orthologue of 53BP1) alone was able to partially suppress the cell cycle arrest observed in cells. Moreover, viability of cells was reduced compared with their counterparts, consistent with the hypothesis that lack of creates DNA damage that is recognized by a -dependent checkpoint (). More recently it was observed that is critical for suppressing spontaneous translocations between highly diverged genes in cells mutated for (). Hence, it is possible that in higher eukaryotes, BLM and 53BP1 can together maintain chromosomal integrity by modulating HR. We hypothesize that BLM and 53BP1 may recognize different recombination substrates and thereby disrupt the RAD51 filaments by different mechanisms. The disappearance of the filaments when BLM and 53BP1 are both expressed () lends credence to this hypothesis. In vitro experiments have unmasked a translocase activity for the yeast DNA helicase , which is involved in disrupting the RAD51 nucleofilament (). It is possible that a similar mechanism may exist for BLM. Our data also provides a logical explanation for the reported physical interaction of BLM with RAD51 (). This mechanism possibly occurs parallel to the reported ability of BLM and topoisomerase IIIα to resolve recombination intermediates like double Holliday junctions into noncrossover products (). The presence of multiple mechanisms may be required for effecting exact control of RAD51 function. Recent studies have demonstrated that signal transducer proteins can have other functions in addition to their known roles in the transmission of DNA damage signal. Like 53BP1, mediator of DNA damage checkpoint protein (MDC1) and breast cancer type 1 susceptibility gene 1 (BRCA1) are also known to participate in DNA damage sensing and signaling pathways, particularly in S-phase (; ). Although 53BP1 has an antirecombinogenic function on HR (; this work), MDC1 interacts with RAD51 and facilitates HR in response to IR (). BRCA1 decreases spontaneous HR and gene conversion and deletion events without affecting recombination intermediates, but it also promotes HR repair of chromosomal double-strand breaks and activates RAD51 function (). This effect on HR by BRCT (C-terminal domain of BRCA1 protein BRCT)–containing adaptor proteins may be modulated by the presence of specific domains (like Tudor domain for 53BP1 or forkhead-associated domain of MDC1) and their interacting partners. The observations that transducer proteins, like BLM and 53BP1, function in HR can have important implications in our understanding the pathway of DNA damage response. Because lack of BLM leads to almost all kinds of cancer in human and mouse models (), BLM is probably at the center of such a DNA damage signal transmission highway. BLM may obtain the signal upstream kinases (; ) and, depending on the type and extent of damage, transmits the signal to repair pathways like HR and simultaneously synergizes with its interacting partners (like p53 and 53BP1; ; this work) to ensure that the extent of recombination is exact. BLM takes into account the antirecombinogenic role of 53BP1, enhancing its interaction with RAD51. Subsequently, the two antirecombinogenic proteins modulate RAD51 function, with 53BP1 probably acting as a backup for BLM. The fact that 53BP1 is required for efficient accumulation of BLM to the sites of stalled replication (), and BLM in turn enhances the 53BP1–RAD51 interaction, suggests the presence of a feed-forward system involved in efficient recognition, transmission, and resolution of the DNA damage. pCMH6K53BP1 was a gift from K. Iwabuchi (Kanazawa Medical University, Ishikawa, Japan). pSilencer2.1-U6 hygro-53BP1 was obtained by cloning the 19-nucleotide 53BP1 siRNA sequence GAACGAGGAGACGGTAATA into the BamHI–HindIII sites of pSilencer2.1-U6 hygro (Ambion). The following antibodies were used: anti-BLM goat polyclonal C- 18 (Santa Cruz Biotechnology, Inc.; Western blots and immunofluorescence), anti-BLM rabbit polyclonal (Novus; IPs), anti-53BP1 mAb (BD Biosciences; immunofluorescence), anti-53BP1 rabbit polyclonal (a gift from Y. Adachi, University of Edinburgh, Edinburgh, UK; Western blots and IPs), anti-TBP mAb l 58C9 (Santa Cruz Biotechnology, Inc.), anti-RAD51 rabbit polyclonal Ab-1 (Calbiochem), and anti-actin mAb C-2 (Santa Cruz Biotechnology, Inc.). Secondary antibodies were purchased from Jackson ImmunoResearch Laboratories and Southern Biotechnology Associates, Inc. hTERT-immortalized BS fibroblasts, BS cells, and the corrected BS fibroblasts A-15 were a gift from J. Shay (University of Texas Southwestern Medical Center, Dallas, TX). Wild-type and knockout primary 53BP1 mouse embryonic fibroblasts were a gift from J. Chen (Yale University School of Medicine, New Haven, CT). All of the above cell lines, along with the hTERT-immortalized NHF strain GM07532 (NHF), were maintained as described previously (, ). 1 mM HU (Sigma-Aldrich) treatments were done as described previously (). Cells were either left untreated (–HU) or treated (+HU) for 16 h. Parallel HU-treated plates were washed, and incubation was continued for a further 6 h (postwash). To generate BS 53BP1 and A-15 53BP1 stable cell lines, the parental BS and A-15 cells were transfected with pSilencer2.1-U6 hygro-53BP1 plasmid. Standard procedures were followed to select stably transfected cells expressing 53BP1 siRNA with 50 μg/ml hygromycin B. Immunofluorescence was performed as described previously (). Cells were fixed, and the stained cells were visualized in a motorized epifluorescence microscope (Upright Axioimager M1; Carl Zeiss MicroImaging, Inc.) equipped with a high-resolution camera (AxioCam MRm Rev. 2; Carl Zeiss MicroImaging, Inc.). Images were analyzed by AxioVision Software Rel. 4.4 (Carl Zeiss MicroImaging, Inc.). The images were taken with Plan Apochromat objective 100×/1.40 oil immersion using FITC or Texas red fluorophore. At least 100 cells were analyzed for a colocalization experiment. The images were subsequently imported into Canvas graphics software, where adjustments were made for the whole image for brightness, contrast, and color balance. Based on two published sequences (; ), siRNAs for 53BP1 were synthesized by Dharmacon. Both the siRNA sequences were alternately used and gave the same results. Commercially available mouse BLM siRNA was purchased from Santa Cruz Biotechnology, Inc. Scrambled duplex RNA and luciferase siRNA (Dharmacon) were used as controls. Cells were fixed for flow cytometry according to standard protocols. Cells were analyzed using a flow cytometer (BD LSR; BD Biosciences) for both cell cycle analysis and host cell reactivation assay. The host cell reactivation assay to determine HR rate was performed as described previously (). Transfections were performed with the substrate (pBHRF), encoding an intact, emission-shifted, “blue” variant of GFP (BFP), with a 300-nucleotide stretch of homology to a nonfunctional copy of GFP. In the absence of HR, only BFP is present, whereas HR can also create a functional GFP. 53BP1 and 53BP1 cells were plated out at low dilutions and treated with different concentrations of HU, and the treatment was continued for 48 h. After removal of the drug-treated medium, the cells were washed and subsequently allowed to grow in complete growth medium undisturbed for 14 d. The colonies formed were fixed and subsequently stained with Giemsa. The whole experiment was done twice in triplicates, and the data shown are cumulative of both experiments. SCE analysis for −HU cultures was performed according to standard protocols. For +HU cultures, BrdU-treated cells were cotreated with HU for 16 h during the second round of DNA duplication. The cells arrested in S-phase were subsequently washed, and incubation was continued in the presence of BrDU for a further 8 h before colcemid treatment. 25 metaphase spreads were imaged at 100× in a microscope (Axioimager M1) for each cell line. The whole experiment was performed twice, and for each cell line, SCE was scored blind. The p-values were obtained by test. The conditions are two-tailed, unpaired data with unequal variance. BS 53BP1 cells were transfected with either BLM or 53BP1. 6 h after transfection, the cells were infected with replication-defective adenoviral genomes generated as described previously (). 34 h after infection, the cells were treated with HU, and the incubation was continued for a further 16 h. The cells were washed and fixed for immunofluorescence. The experiments were repeated thrice, and a total of 150 cells showing RAD51 expression were counted. Fig. S1 shows cell cycle analysis and nucleocytoplasmic separation of NHFs grown under different conditions. Fig. S2 show that 53BP1 and BLM together modulate HR in mouse cells. Fig. S3 shows that 53BP1 and BLM regulate RAD51 foci formation. Video 1 shows visualization of RAD51 filaments. Video 2 shows visualization of the disruption of RAD51 filaments as a result of BLM overexpression. Video 3 shows visualization of the disruption of RAD51 filaments as a result of 53BP1 overexpression. Online supplemental material is available at .

PKD is a family of serine/threonine-specific protein kinases comprising three structurally related members: PKD1/PKCμ, PKD2, and PKD3/PKCν. PKD contains two zinc finger–like cysteine-rich motifs that bind DAG, a pleckstrin homology (PH), and a kinase domain. PKD localizes to the cytosol, nucleus, Golgi complex, and plasma membrane, where it regulates diverse cellular processes, including vesicle trafficking (; ). Thus far, only a few physiological PKD substrates are known (e.g., the neuronal protein Kidins220, the Ras effector RIN1, HDAC5, and PI4KIIIβ; ; ; ; ). At the TGN, PKD is critically involved in the fission of transport carriers en route to the cell surface (; ). PKD is recruited to the TGN by its cysteine-rich regions (; ; ), where it is activated by PKCη-mediated phosphorylation (). PKD-mediated phosphorylation of PI4KIIIβ stimulates its lipid kinase activity, resulting in enhanced phosphatidylinositol 4-phosphate (PIP) production and cargo transport to the plasma membrane (). In this study, we demonstrate that PKD also phosphorylates and regulates the activity of the Golgi-localized ceramide transfer protein (CERT; also known as Goodpasture antigen-binding protein), a cytosolic protein essential for the nonvesicular delivery of ceramide from its site of production at the ER to Golgi membranes, where conversion to sphingomyelin (SM) takes place (). Two CERT isoforms exist: the more abundantly expressed, alternatively spliced form missing a 26–amino acid serine-rich region and the full-length 624–amino acid protein, which is designated CERT (). Both CERT isoforms possess a steroidogenic acute regulatory lipid transfer (START) domain that is necessary and sufficient for ceramide binding and transport (). START domains are ∼210 amino acids in length and form a hydrophobic tunnel that accommodates a monomeric lipid (; ). They are found in 15 mammalian proteins, with CERT being most closely related to Pctp, which binds and shuttles phosphatidylcholine (PC) between membranes, and StarD10, a lipid transfer protein specific for PC and phosphatidylethanolamine (; ; ). CERT proteins further contain an N-terminal PH domain with specificity for PIP that contributes to Golgi localization (; ) and an FFAT motif (two phenylalanines in an acidic tract) that targets the protein to the ER via interaction with the ER resident transmembrane proteins VAP-A and VAP-B (vesicle-associated membrane protein–associated protein; ; ). Nonvesicular lipid transfer is thought to occur at membrane contact sites, at which the ER comes into close apposition with other organelles (). CERT may thus shuttle a very short distance between ER and Golgi membranes or perhaps contact both compartments simultaneously. When overexpressed, the START domain of CERT is sufficient for ceramide transfer to the Golgi complex (). However, under physiological conditions, both Golgi and ER targeting motifs are essential for CERT function. In the CHO cell line LY-A, CERT was identified to contain a mutation within its PH domain (G67E), rendering the protein defective in PIP binding, which resulted in reduced cellular SM levels (). The PIP requirement for CERT function is further supported by a recent study showing that PI4KIIIβ activity is necessary for efficient ceramide trafficking to the Golgi (). We now provide evidence that PKD phosphorylates CERT on serine 132 adjacent to the PH domain, whereby PIP binding, Golgi targeting, and ceramide transfer activity are negatively regulated. Furthermore, by transferring ceramide that is required for DAG production to Golgi membranes, CERT stimulates PKD activity and ensures the maintenance of constitutive secretory transport. PKD is a key regulator at the Golgi complex, with PI4KIIIβ being the only local substrate identified thus far (). To test whether the Golgi complex–localized CERT protein may serve as a substrate for PKD, we made use of a phosphospecific substrate antibody, termed pMOTIF, that was raised against consensus motifs phosphorylated by PKD (). HEK293T cells were transfected with expression vectors encoding Flag-tagged CERT and CERT. Immunoprecipitated CERT isoforms were analyzed by Western blotting with the pMOTIF antibody (). A pMOTIF signal corresponding to the molecular weight of CERT and, more weakly, to that of CERT was detected (). The weaker detection of the CERT isoform by ∼25% compared with CERT may be related to its known behavior to form aggregates, which may impact phosphosite accessibility to kinases (). To investigate whether recognition of CERT by the pMOTIF antibody was dependent on PKD, we expressed CERT together with a kinase-dead (KD) dominant-negative PKD1 variant (PKD1-KD) in HEK293T cells. Coexpression of inactive PKD1 abolished CERT detection by the pMOTIF antibody, suggesting that the signal was indeed the result of PKD-mediated CERT phosphorylation (). To address the question of which PKD isoform was responsible for CERT phosphorylation, we used an RNAi approach to down-regulate PKD. Silencing of only one isoform did not influence the level of CERT phosphorylation as judged by immunoblotting with the pMOTIF antibody (unpublished data). However, simultaneous knockdown of PKD1 and PKD2 greatly reduced CERT phosphorylation (), suggesting that these two isoforms were primarily responsible for phosphorylating CERT, whereas PKD3 appeared to play a minor role. This is in accordance with previously reported overlapping substrate specificities of PKD1 and PKD2, which both phosphorylate PI4KIIIβ, whereas PKD3 fails to do so (). The phosphorylation status of CERT was strongly reduced in serum-deprived cells and could be restored by the readdition of serum (), indicating that CERT phosphorylation is dependent on extracellular stimuli. It was recently reported that OSBP (oxysterol-binding protein) promotes CERT translocation to the Golgi complex in response to stimulation with its ligand, 25-hydroxycholesterol, thereby integrating sterol signaling and SM synthesis (). In line with these studies, 25-hydroxycholesterol treatment was found to augment CERT phosphorylation (), possibly by bringing CERT to the Golgi in the vicinity of PKD. CERT has been demonstrated to colocalize with the cis/medial-Golgi marker GS28 (). Immunofluorescence analysis of GFP-tagged CERT expressed in COS7 cells showed that the protein localized to GS28-positive Golgi regions (Fig. S1, available at ). However, lipid transfer proteins are thought to act at membrane contact sites, which are formed between the ER and TGN (), where PKD is localized. Immunofluorescence staining of Flag-tagged CERT coexpressed with GFP-tagged PKD in COS7 cells revealed that the two proteins colocalize at the Golgi complex. Furthermore, staining of the TGN-specific marker protein TGN46 verified that CERT partially localizes to this compartment (). To identify pMOTIF recognition sites in CERT, we searched for potential PKD consensus motifs characterized by a leucine, isoleucine, or valine residue in the −5 position and arginine in the −3 position relative to a serine or threonine. Two serines at positions 132 and 272 matching the PKD consensus motif () were exchanged for alanines by site-directed mutagenesis. Mutants were expressed in HEK293T cells and tested for recognition by the pMOTIF antibody. Interestingly, mutation of serine 132 to alanine abrogated the detection of CERT with the pMOTIF antibody and caused an increase in electrophoretic mobility, which is indicative of the loss of phosphorylation, whereas the S272A mutation did not affect the pMOTIF signal (). On low percentage gels, the wild-type (WT) protein migrated as two distinct bands, indicating the presence of a phosphorylated and a nonphosphorylated CERT pool (unpublished data). To confirm that PKD was capable of directly phosphorylating serine 132, we performed in vitro kinase assays with purified PKD1 and recombinant CERT GST fusion proteins comprising the first 138 amino acids of the protein. WT CERT was efficiently phosphorylated by PKD1, whereas the CERT-S132A protein showed a strongly reduced incorporation of radioactivity in this assay (). Furthermore, in vitro PKD phosphorylation of WT but not CERT-S132A generated a recognition site for the pMOTIF antibody (). Collectively, these results prove that CERT is a genuine PKD substrate in vitro and in vivo and identify serine 132 as a specific PKD phosphorylation site in CERT that can be monitored with the pMOTIF antibody. Serine 132 is in close proximity to the CERT PH domain (aa 23–117), making it possible that phosphorylation on this site affects PIP binding by increasing the local negative charge. Therefore, we quantified PIP binding of CERT-WT and -S132A by performing protein–lipid overlay assays. Cytosol from cells transiently expressing the CERT variants was incubated with membranes spotted with a concentration gradient of the different phosphoinositides, and bound CERT proteins were detected via their GFP tag. As reported previously, the WT protein demonstrated weak binding to several phospholipid species but displayed strong interaction with PIP (; ). CERT-S132A binding to PIP was detectable at two- to fourfold lower concentrations as compared with that of the WT protein (). To corroborate these results, the association of CERT with multilamellar vesicles (MLVs) consisting of PC alone or PC plus 5% PIP was measured. Although the addition of PIP to PC vesicles increased the membrane binding of CERT-WT 1.5-fold, the binding of CERT-S132A was enhanced 1.9-fold, suggesting an increased affinity of the CERT-S132A mutant to PIP (). To investigate whether this affected the association with Golgi membranes in intact cells, we performed fractionation studies with cells expressing CERT-WT and -S132A. To estimate the level of ER binding, we included a CERT mutant (G67E) defective in PIP binding. Only a small proportion of the WT and G67E protein were recovered in the pellet fraction, suggesting that under the experimental conditions used, ER binding was negligible, and Golgi association of the WT protein was not maintained (). The CERT-S132A mutant protein was highly enriched in the pellet fraction, confirming that the enhanced affinity for PIP stabilizes membrane association in vivo. Together, these data imply that CERT, once bound to the Golgi complex, is phosphorylated by PKD. This then decreases the affinity of CERT to PIP and, thereby, regulates the interaction of CERT with the Golgi complex. Because PIP is also present at the plasma membrane, additional factors must specify CERT targeting to the Golgi complex. One candidate is Arf1, which has been shown to interact with the structurally related proteins OSBP and FAPP1 (). Whether CERT phosphorylation influences binding to such additional factors remains to be tested in the future. The CERT protein has been shown to function as a lipid transfer protein (). Thus, we investigated whether CERT phosphorylation on serine 132 influenced its ability to bind and transfer ceramide between membranes. To this end, GFP-tagged versions of CERT-WT and -S132A were transiently expressed in HEK239T cells, and the cytosol fraction was analyzed for ceramide-specific lipid transfer activity using a fluorescence resonance energy transfer–based assay. In this assay, vesicles containing pyrene-labeled ceramide as a fluorescent donor and quenching amounts of 2,4,6-trinitrophenyl-phosphatidylethanolamine (TNP-PE) were used (; ). The lipid preparation used was total extract from porcine brain, which is likely to contain PIP. Upon the addition of cytosol-containing CERT-WT, a steady increase in fluorescence was noted, which was not observed when control cytosol of vector-transfected cells was used (). Compared with the WT protein, CERT-S132A displayed a higher rate of lipid transfer, which was evident from a more rapid increase in pyrene fluorescence (). Similar results were obtained when 0.5% PIP was added to donor liposomes (unpublished data). This suggests that CERT phosphorylation on serine 132 down-regulates ceramide transfer activity, most likely by decreasing association of the protein with membranes. Previous data have already shown that PKD regulates the level of PIP at the Golgi complex by the phosphorylation-mediated activation of PI4KIIIβ (). Interestingly, PI4KIIIβ is critical for the transport of ceramide between the ER and the Golgi complex (). Accordingly, together with the data presented in this study, a dual role for PKD in maintaining lipid homeostasis of Golgi membranes becomes apparent by controlling the on rate (via PIP levels) and off rate (via direct phosphorylation) of CERT. The transfer of ceramide from the ER to the TGN is essential for SM synthesis at this compartment (). Golgi-localized SM synthase 1 utilizes ceramide and PC to generate SM and DAG (), the latter being a prerequisite for PKD recruitment and activation. Compounds that block DAG production at the TGN inhibit the binding of PKD to TGN membranes and interfere with secretory transport (). Therefore, increased ceramide transfer from the ER to the TGN by the overexpression of CERT should result in an elevated local DAG pool and may consequently stimulate PKD activity and secretory transport. To test this hypothesis, we transiently expressed CERT-WT and -S132A in HEK293T cells and analyzed the autophosphorylation of endogenous PKD. Compared with the control, the expression of both CERT-WT and -S132A increased PKD activity, as revealed by analyses with a phosphospecific PKD antibody (). CERT has been reported to possess kinase activity (), making it possible that it activates PKD by direct phosphorylation. However, kinase assays clearly demonstrated that PKD is not phosphorylated by CERT. Moreover, a kinase activity was associated with the CERT protein only under mild detergent conditions (Fig. S1). Thus, our results show that PKD activation is regulated by CERT proteins, most likely as a result of increased ceramide delivery and enforced SM/DAG synthesis. A similar function has recently been described for the lipid transfer protein Nir2 in the maintenance of DAG levels at the Golgi apparatus via regulation of the cytidine-5′-diphosphate–choline pathway (). RNAi-mediated knockdown of Nir2 decreased DAG and PKD levels at the Golgi complex and blocked secretory transport. Interestingly, this effect could be rescued by the addition of exogenous C-ceramide (), indicating a critical role for ceramide in DAG synthesis and PKD recruitment to the Golgi complex. To address the question of whether CERT-mediated PKD activation indeed translated into enhanced secretory transport, we made use of a plasmid encoding HRP fused to a signal sequence (ss). The fusion protein ssHRP can be used as a reporter for constitutive protein secretion (). In control cells, secretion of ssHRP could be detected within 1 h and increased over time (). Coexpression of PKD1-KD, which inhibits the secretory transport of cargo protein (; ), almost entirely abrogated ssHRP secretion. This confirmed that HRP was secreted in a PKD-dependent manner in our assay. Coexpression of CERT-WT and -S132A strongly augmented the amount of secreted HRP (). Conversely, knockdown of CERT by RNAi in COS7 cells inhibited the secretion of HRP (), confirming the essential role for CERT in the constitutive exocytosis of cargo proteins. We could only detect a slight increase in secretion with the S132A mutant compared with the one observed with the WT protein. This is in accordance with the comparable activation of PKD by CERT-WT and -S132A () but was unexpected in light of the substantially enhanced in vitro lipid transfer activity of the CERT mutant (). However, increased levels of ceramide may not necessarily translate into equivalent increases in DAG because DAG synthesis might be limited by the availability of PC and the activity of SM synthase. The accumulation of ceramide is known to affect Golgi membrane stability and induces vesicle fission (; ). Therefore, we investigated whether overexpression of the CERT-S132A mutant affected its localization and/or caused morphological changes of the Golgi apparatus. In addition to concentrating in GS28-positive regions of the Golgi complex, the CERT-S132A mutant displayed a dispersed punctate staining (). However, the distribution of GS28 itself and that of TGN46 was not affected by the expression of CERT-S132A, nor were these proteins present in the vesicular structures observed with the mutant CERT protein (). This rules out fragmentation of the Golgi apparatus as a consequence of CERT-S132A overexpression. Some of the vesicular structures were found to contain the cargo protein ssHRP, providing evidence that these structures represent Golgi-derived transport carriers (). It thus appears that the increased membrane affinity of CERT-S132A prevents its dissociation from budding vesicles. Interestingly, when coexpressed with CERT-S132A, the PH domain of OSBP also localized to these vesicles, indicating that these structures are PIP positive (). The CERT-S132A mutant may therefore inhibit PIP turnover, thus stabilizing the lipid on transport carriers. Of note, a CERT-S132E protein was indistinguishable from the alanine mutant in terms of cellular localization and, thus, could not be used to mimic the phosphorylated state (unpublished data). Collectively, our data support the following working model: PKD is recruited to the TGN by a local DAG pool that can be generated via different metabolic pathways. PKD then activates PI4KIIIβ, increasing PIP levels at the TGN. This, in turn, recruits the CERT protein to the Golgi complex, where it contributes to PKD activation and vesicular transport processes by providing ceramide as a precursor for further DAG production. The system is tightly regulated by a negative feedback loop: active PKD phosphorylates CERT at serine 132, thus decreasing the affinity of CERT toward its lipid target PIP to ensure continuous rounds of lipid transfer from the ER to the Golgi compartment. In conclusion, we have identified CERT as a PKD substrate and provide evidence for a novel relationship between membrane lipid biogenesis and protein secretion. Cells were fixed in 4% PFA for 10 min, washed, and incubated with PBS containing 0.1 M glycine for 15 min. Cells were permeabilized with PBS containing 0.1% Triton X-100 for 5 min and blocked with 5% goat serum in PBS containing 0.1% Tween 20 for 30 min. Cells were then incubated with primary antibody diluted in blocking buffer for 2 h followed by incubation with secondary antibodies diluted in blocking buffer for 1 h. Coverslips were mounted in Fluoromount G (Southern Biotechnology Associates, Inc.) and analyzed on a confocal laser-scanning microscope (TCS SL; Leica) using 488- and 543-nm excitation and a 40.0/1.25 HCX PL APO objective lens. Images were processed with Photoshop (Adobe). All images shown are stacks of several confocal sections. Whole cell extracts were obtained by solubilizing cells in NP-40 extraction buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 1% NP-40, 1 mM sodium orthovanadate, 10 mM sodium fluoride, and 20 mM β-glycerophosphate plus Complete protease inhibitors [Roche]). Lysates were clarified by centrifugation at 16,000 for 10 min. For immunoprecipitations, equal amounts of protein were incubated with specific antibodies for 2 h on ice. Immune complexes were collected with protein G–Sepharose beads (GE Healthcare) and washed three times with NP-40 extraction buffer. Whole cell extracts or immunoprecipitated proteins were subjected to SDS-PAGE, and proteins were blotted onto polyvinylidene difluoride membranes (Roth). After blocking with 0.5% blocking reagent (Roche) in PBS containing 0.1% Tween 20, filters were probed with specific antibodies. Proteins were visualized with HRP-coupled secondary antibody using the ECL system (Pierce Chemical Co.). Stripping of membranes was performed in 62.5 mM Tris, pH 6.8, 2% SDS, and 100 mM β-mercaptoethanol for 30 min at 60°C. Membranes were then reprobed with the indicated antibodies. BL21 bacteria were transformed with pGEX6P-Flag-CERT-WT(1–138) and -S132A(1–138) vectors. Expression was induced with 0.5 mM IPTG for 4 h at 30°C. Bacteria were harvested and resuspended in PBS containing 50 μg/ml lysozyme, Complete protease inhibitors (Roche), 10 mM sodium fluoride, and 20 mM β-glycerophosphate. Triton X-100 was added to a final concentration of 1% before sonication. GST-CERT fusions were purified from clarified lysate with glutathione resin (GE Healthcare). Recombinant proteins were incubated with purified PKD1 from insect cells in kinase buffer (50 mM Tris, pH 7.5, 10 mM MgCl, and 1 mM DTT) in the presence of either 2 μCi γ-[P]ATP or 75 μM of cold ATP for 30 min. Samples were resolved by SDS-PAGE, blotted onto membrane, analyzed on a phosphorimager (Storm 860; Molecular Dynamics), and detected with the indicated antibodies. Cells were harvested in hypotonic buffer (50 mM Tris, pH 7.4, containing Complete protease inhibitors, 1 mM PMSF, 5 mM β-glycerophosphate, and 5 mM sodium fluoride) and sheared by passage through a 25-G/16-mm needle. Nuclei were removed by centrifugation at 500 , and cytosol and membrane fractions were obtained by centrifugation at 100,000 . The amount of expressed CERT protein in the cytosolic fraction was quantified by GFP peak emission at 480–550 nm (excitation of 466 nm). Phosphatidylinositol phosphate arrays (Echelon) were blocked in TBS-T (10 mM Tris, pH 8, 150 mM NaCl, and 0.1% Tween 20) containing 3% fatty acid–free BSA (Roth) followed by incubation with 500 μg cytosol containing equal amounts of GFP proteins in 5 ml of blocking buffer for 1 h. Bound proteins were detected with anti-GFP antibody followed by HRP-conjugated secondary antibody. Flotation assays were performed by incubating 50 μl cytosol containing equal amounts of GFP-tagged CERT proteins with 100 μl MLVs in 50 mM Tris, pH 7.5, and 50 mM NaCl buffer for 10 min at RT. The suspension was adjusted to 30% sucrose by the addition of 100 μl of 75% sucrose and overlayed with 200 μl of 25% sucrose in buffer and 50 μl sucrose-free buffer. Samples were centrifuged at 240,000 for 1 h. The bottom (250 μl) and top (100 μl) fractions were collected and analyzed by fluorescence spectrometry. Protein-mediated transfer of ceramide between small unilamellar vesicles was measured as described previously (). The transfer assay mixture contained donor vesicles (2 nmol of lipid/ml) composed of brain lipids, pyrene-labeled C-ceramide, TNP-PE (provided by P. Somerharju, University of Helsinki, Helsinki, Finland; 88.6:0.4:11 mol percent), and a 10-fold excess of acceptor vesicles composed of brain lipids. Fluorescence intensity was recorded at 395 nm (excitation of 345 nm and slit widths of 4 nm) before and after the addition of 75 μg cytosol from HEK293T cells transiently expressing GFP-tagged CERT-WT and -S132A. Fluorescence intensities were normalized to the maximum intensity obtained after the addition of 0.5% Triton X-100 and the maximum GFP fluorescence to account for different protein expression levels. HEK293T cells were cotransfected with ssHRP-Flag plasmid together with empty vector, pEGFPN1-PKD1-KD, pcDNA3-Flag-CERT-WT, and -S132A at a ratio of 1:6.5, respectively. For CERT RNAi, COS7 cells were transfected with ssHRP-Flag plasmid, harvested after 8 h, replated, and transfected with siRNAs. HEK293T and COS7 cells were washed with serum-free medium 24 and 48 h after transfection, respectively, and HRP secretion was quantified by incubation of clarified cell supernatant with ECL reagent. Measurements were performed with a luminometer (Lucy2; Anthos) at 450 nm. Fig. S1 shows that CERT does not phosphorylate PKD directly. Fig. S2 shows the colocalization of CERT-WT and GS28. Supplemental materials and methods provides information about the antibodies and reagents used, DNA constructs, and cell culture and transfection. Online supplemental material is available at .

A critical event during tumorigenesis is the conversion from a static primary tumor to an invasive, disseminating metastasis. Moreover, tumor cells show an increased capacity to migrate. Numerous intracellular signaling molecules have been implicated in migratory processes. Among them, the Rho GTPase family plays a pivotal role in regulating the biochemical and cytoskeletal pathways relevant to cell migration. The Rho GTPases Rac, Cdc42, and Rho control cell protrusions during migration. Aberrant regulation of Rho proteins is believed to associate with metastasis by promoting cell motility (; ; ; ). The bona fide environment for migrating cells is the extracellular matrix, which permits movement in a 3D scaffold that is biochemically complex and shows dynamic flexibility. 3D tissue culture models reconstitute an environment that resembles the in vivo situation in regard to cell shape and movement. This model provides important insights into the mechanisms of cell motion during carcinogenesis. Recent advances have identified two modes of cell motility in 3D matrices. The elongated mode of migration is a consequence of Rac activity and generates membrane protrusions, the lamellipodia that drive motility. In contrast, a novel rounded mode of motility depends on RhoA and its main effector, Rho-associated coil-containing protein kinase (ROCK), and resembles the amoeboid movement. This involves a rounded bleb-associated movement that generates propulsive motion through the matrix independently of proteolysis (). Curiously, genes encoding Rho GTPases are rarely found mutated in human cancers, in which only their functional activities seem to be deregulated (; ). This suggests that alterations in others genes account for functional modifications of Rho GTPases accompanying actin cytoskeleton remodeling during metastasis. We hypothesize that this set of genes controls the cell cycle. Their mutation during the initiation phase in cancer would modify the behavior of proteins involved in actin cytoskeleton dynamics, such as Rho GTPases, leading to a migratory and invasive phenotype. Beyond these genes is the tumor suppressor p53, whose mutations occurs in >50% of human tumors (). p53 protects cells from malignant transformation by regulating cell cycle arrest or by promoting apoptosis (; ). We and others have recently shown that p53 modulates cell migration: p53 negatively modulates Rho GTPases and regulates cell polarization and migration (, ; ). However, little is known about the role of p53 in cells moving in a 3D matrix that mimics the in vivo microenvironment of tumor cells. Identifying the mechanisms by which p53 modulates cell migration is important to understand how invasive cells arise. In this study, we show that the elongated spindle-shaped fibroblastoid mode of motility can be converted to a rounded blebbing movement by blocking p53 function. This amoeboid mode of motility requires RhoA–ROCK signaling and confers higher velocity and invasive properties to p53-deficient cells. Thus, the range of p53 tumor suppressor activity extends to the control of the mode of migration of invasive cells. Initially, we compared the movement of p53-deficient primary mouse embryonic fibroblasts (MEFs; p53 MEFs) with that of wild-type (wt) MEFs cultured in 3D Matrigel matrix using time-lapse video microscopy. The migration of the two cell types is strikingly different ( and Videos 1 and 2, available at ). wt MEF maintained a spindle-shaped elongated morphology, generated extensions, and moved slowly at 2 ± 1 μm/h, whereas p53 MEF moved as rounded cells using a translatory motion in the direction of the serum growth factor gradient. They show many dynamic bleblike structures, alternating rapid cycles of squeezing and expansion, and considerable deformability. p53 cells moved at 12 ± 5 μm/h, maintaining their rounded morphology. Expression of the GFP-tagged mutant p53 H273, a dominant-negative form of p53, converts the elongated morphology of MEF to a marked spherical and flexible one associated with highly dynamic membrane blebs similar to those observed in p53 MEFs ( and Video 3). The expression of GFP-tagged wt p53 in p53 MEFs inhibits dynamic blebbing and leads to an elongated morphology ( and Video 4), whereas that of GFP-tagged p53 H273 has no effect. Thus, the loss of wt p53 activity is sufficient to drive the switch from an elongated fibroblast-like to a rounded type of migration. In Matrigel, cells moving with a rounded or elongated morphology show a distinct subcellular distribution of integrin β1 and ezrin that helps in establishing the direction of movement (). In Matrigel, rounded blebbing MEFs that arise from p53 inactivation show integrin β1 and ezrin relocalization to the moving front of the cell. A similar clustering was also observed in rounded p53 MEFs. In contrast, conversion of p53 MEFs to elongated morphology by the expression of p53 wt led to their relocalization throughout the cell, notably in the newly formed long extensions (). Thus, p53 controls the clustering of integrin β1 and ezrin in rounded cells. Culture of fibroblastic cells in monolayers imposes an artificial polarity between the lower and upper surface of these apolar cells. Their morphology and migration change in a 3D matrix, suggesting that the spatial constraints and surrounding biochemical microenvironment are a better approximation of the in vivo situation (; ). This is particularly striking for p53-deficient MEFs that become blebbing spherical cells once suspended in 3D matrices, losing their initial fibroblastoid morphology observed in 2D culture. This rounded blebbing movement observed in p53 MEFs is strikingly similar to amoeboid-like motility, which is dependent on Rho–ROCK signaling in a 3D matrix (). We compared the level of GTP-bound RhoA in p53 and wt MEFs. GTP-RhoA was increased 3.4-fold in p53 MEFs, whereas GTP-Rac and Cdc42 were barely affected (). This change was functional because protein expression levels were unaltered by p53 deletion. Reintroduction of wt p53 in p53 MEFs lowered the level of GTP-RhoA to that of wt MEFs (). Conversely, p53 H273 did not affect the overactivation of RhoA in p53 MEFs. Finally, blocking p53 in MEFs using p53 H273 or siRNA-mediated knockdown of p53 was sufficient to increase GTP-RhoA (). These results show that wt p53 inhibits RhoA activation and has a role in regulating RhoA signaling pathways. Translocation from the cytoplasm to the membrane area is essential for both the activation and function of RhoA (). We compared the localization of RhoA and filamentous actin (F-actin) in p53 and wt MEFs. In wt MEFs, RhoA exhibits a punctuate distribution throughout the cytoplasm with a marked concentration in the perinuclear region. F-actin stress fibers cross the cytoplasm and accumulate slightly around the edge of the cell (, top). In p53 MEFs, RhoA shows its punctuate cytoplasmic localization but is excluded from the perinuclear region (, bottom). Instead, RhoA colocalizes with large patches of polymerized actin in bleblike globular structures that are distributed over and protruding from the cell surface (, arrows). In addition, MEFs lacking p53 activity show more peripheral microspikes, which were previously characterized as filopodia (). They also show an accumulation of peripheral polymerized actin bundles (i.e., a redistribution of F-actin stress fibers; ). The expression of wt p53 led to loss of the bleblike protrusions and recovery of the punctuate RhoA localization in the perinuclear region. This was not seen with p53 H273, indicating that wt p53 activity is required for RhoA localization (, compare D with E). Treatment of p53 MEFs with TAT-C3, which inactivates RhoA, RhoB, and RhoC (), abolished bleblike structures and led to RhoA redistribution in the perinuclear region (). Thus, RhoA activation is required for bleb formation in p53 cells. RhoA-driven actin reorganization largely depends on the serine–threonine protein kinase ROCK (; ). To determine whether ROCK mediates membrane blebbing in p53 MEFs cultured in monolayer, we used Y27632 and H1152, which are two distinct, structurally unrelated pharmacological inhibitors of ROCK. p53 MEFs promptly lost their extensive dynamic bleblike activities and their rounded morphology within 40 min, adopting a flattened, spread shape when treated with Y27632 ( and Video 5, available at ) or H1152 (not depicted). Consistent with this, the expression of ROCK-Δ1, an active mutant, in wt MEF promoted a rounded morphology accompanied by numerous dynamic bleblike extensions similar to p53 MEF ( and Video 6). GFP alone had no effect (). The spreading of p53 MEFs induced by Y27632 or H1152 was accompanied by the reorganization of F-actin from cortical bundles into scattered cytoplasmic dotlike structures concomitant with the loss of filopodia (). Similar to other studies (; ), ROCK inhibitors strongly reduced the number of stress fibers in both wt and p53 MEFs. In addition, ROCK inhibition in p53 MEFs led to RhoA relocalization to the perinuclear region (compare with ). Thus, ROCK inhibition causes RhoA to lose its membrane localization. We tested whether ROCK activity is required for the rounded blebbing movement of p53 MEFs cultured in Matrigel. Transfection of ROCK-Δ1 converts MEFs from an elongated to a rounded blebbing morphology; GFP alone had no effect ( and Video 7, available at ). In contrast, treatment of p53 MEFs with Y27632 or H1152 inhibits dynamic blebbing and leads to an elongated morphology even upon the expression of p53 H273 ( and Videos 8 and 9). This transition is accompanied by the redistribution of integrin β1 and ezrin to the moving front of the cell (Fig. S1). Thus, ROCK activity is required for the amoeboid-like rounded morphology driven by p53 deficiency. Collectively, our results indicate that characteristics of the fibroblast type of motility (i.e., that using elongation and traction) are lost when p53 function is abrogated using either p53 H273 or siRNA. Cells adopt a rounded bleb morphology similar to that observed in p53 MEFs cultured in 3D matrices. A p53 defect leads to the aberrant overactivation of RhoA, which consequently translocates to specific membrane blebbing structures. Importantly, we show that the RhoA–ROCK signaling pathway is involved in p53-dependent morphological changes by modulating round blebbing. It also appears to promote blebbing in p53 cells. Notably, RhoA signaling through ROCK was previously shown to be important in the rounded blebbing movement but not for the elongated protrusive movement of cancer cells cultured in 3D matrices (). RhoA–ROCK signaling also plays a key role in invasion by blebbing cells moving in 3D matrix (). Similarly, p53 MEFs, which have an overactivation of this pathway, showed substantially higher invasiveness than wt MEFs in Matrigel (). Inhibition of Rho or ROCK by treatment with TAT-C3, Y27632, or H1152 diminished the ability of p53 MEFs to invade Matrigel by around 60%. Expression of wt p53 but not p53 H273 blocked this behavior. The ability of MEFs to invade Matrigel was increased when p53 activity was blocked either by expression of the H273 p53 mutant or using siRNA to p53 in MEFs. Similarly, invasiveness was enhanced by an active mutant of RhoA (RhoA-V14) and ROCK-Δ1. Invasiveness of blebbing cells is independent of protease activities (); accordingly, the invasive activity of p53 MEFs was unaffected by the inhibition of proteases. Thus, the Rho–ROCK signaling module is necessary for the invasive behavior of p53 cells in Matrigel. RhoE, a member of the Rho family that blocks actin stress fibers (; ), is a transcriptional target of p53 in response to genotoxic stress; RhoE promotes cell survival through the inhibition of ROCK1-mediated apoptosis (). The overexpression of RhoE clearly prevented actin polymerization and abolished the bleblike protrusions in p53 MEFs, converting them to an elongated morphology (; and Video 10, available at ). RhoE also considerably reduced the invasiveness of p53 MEFs (), suggesting that RhoE helps mediate the control of cell morphology and invasiveness by p53. We sought to extend these observations to cancer cells. We chose A375P melanoma cells, which have a wt form of p53 and show an elongated mode of migration in 3D matrix (). Cell morphology was visualized using fluorescent wheat germ agglutinin, which provides highly selective labeling of the external plasma membrane. The expression of GFP-tagged p53 H273 confers a rounded blebbing morphology to A375P cells, which associates with membrane blebs similar to those observed in p53 MEFs (). In contrast, the elongated morphology of A375P was unchanged when p53 accumulation was induced by etoposide, which was verified using the induction of p21WAF, a transcriptional target of p53 (). These morphologic changes correlated with the level of GTP-RhoA, which was 4.6-fold higher upon the expression of p53 H273 (). Thus, the amoeboid-like motility of A375P melanoma cells induced by the loss of p53 activity is dependent on RhoA activation. Amoeboid cancer cell motility is particularly efficient in promoting the invasiveness of metastatic cells in 3D environments. Therefore, we performed invasion assays in Matrigel using this model. The expression of p53 H273 strongly increased the invasiveness of A375P, whereas wt p53 inhibited it (). TAT-C3, Y27632, or H1152 treatment largely reduced this response in p53 H273–expressing A375P cells, arguing that mutated p53 drives this behavior through RhoA–ROCK signaling. In contrast, protease inhibition did not alter p53 H273–driven invasiveness (). This stresses the role of the amoeboid-like rounded morphology of p53-deficient A375P cells in their invasive potential. The tumor suppressor p53 is mutated in >50% of human cancers () and, thus, constitutes an appealing target for anticancer therapy. Diverse therapeutic strategies already exist that attempt to restore wt p53 function in cancer cells, including the rescue of mutant p53 function and reactivation of wt p53 (). Given that there is no obvious selection for a metastatic phenotype, proposed that classic oncogenes and tumor suppressor genes implicated in the early stages of tumorigenesis may also play a role in metastasis. Our studies indicate that p53 is an ideal candidate, as it is involved both in the control of tumor cell apoptosis and tumor cell invasion. This extends the applications for anticancer agents aimed at restoring p53 wt functions: such agents may combine antiproliferative and antiinvasive activities. wt and p53 MEFs were generated, cultured, and transfected as previously described (). A375P cells were cultured in DME with 10% FCS and transfected using electroporation (Nucleofector; Amaxa). In brief, 10 cells were suspended in 100 μl of solution R and mixed with 3 μg DNA for electroporation (program T-016). The primary antibodies used were anti-p21 (C-19; Santa Cruz Biotechnology, Inc.), anti-p53 (DO-1; Novocastra), and anti–α-tubulin (DM1A; Sigma-Aldrich). The inhibitors used were 10 μM Y27632 (Calbiochem) and 5 μM H1152 (Calbiochem). Etoposide (Sigma-Aldrich) was used at 25 μg/ml. GFP-tagged p53 wt and p53 H273 were previously described (). pSuper-siRNA p53 and plasmid EGFPC1-RhoE were gifts from J.-C. Bourdon (University of Dundee, Dundee, Scotland, UK) and Ph. Fort (Centre National de la Recherche Scientifique, Montpellier, France), respectively. The Rho GTPase activities were processed as previously described (). The quantification of cell invasion was performed in transwell cell culture chambers containing fluorescence-blocking polycarbonate porous membrane inserts (pore size of 8 μm; Fluoroblock; BD Biosciences). 100 μl Matrigel with reduced growth factors (a commercially prepared reconstituted basement membrane from Englebreth-Holm-Swarm tumors; BD Biosciences) were prepared in a transwell. Cells were transfected and treated with TAT-C3 for 16 h and Y27632 or H1152 for 2 h as monolayer before trypsinization and plating (5 × 10) in serum-free media on top of a thick layer (around 500 μm) of Matrigel contained within the upper chamber of a transwell. Controls were left untreated. The upper and lower chambers were then filled with serum-free DME and DME with 10% FCS, respectively, thus establishing a soluble gradient of chemoattractant that permits cell invasion throughout the Matrigel. Inhibitors were added immediately after cell plating at the aforementioned concentrations. Cells were allowed to invade at 37°C and 5% CO through the gel for 36 h (for MEF and p53 MEF) or for 24 h (for A375P) before fixing for 15 min in 3.7% formaldehyde. Cells that had invaded through the Matrigel were detected on the lower side of the filter by GFP fluorescence and counted for cell number. Six fields per filter were counted, and each assay was performed twice in triplicate for each cell line. Cells were transfected and/or treated with TAT-C3, Y27632, or H1152 on coverslips at a confluence of ∼30% before fixation in 3.7% formaldehyde in PBS followed by a 5-min permeabilization in 0.1% Triton X-100 in PBS and were incubated in PBS containing 0.1% BSA before staining with primary antibodies as follows: RhoA (Santa Cruz Biotechnology, Inc.), antiezrin polyclonal antibody (provided by P. Mangeat, Centre de Recherche en Biochimie Macromoléculaire, Montpellier, France), and integrin β1 (Santa Cruz Biotechnology, Inc.). All of the antibodies were revealed with either an AlexaFluor546- or -488–conjugated goat anti–mouse or anti–rabbit antibody (Invitrogen and Interchim). Cells were stained for F-actin using TRITC-conjugated phalloidin (Sigma-Aldrich) or for membrane morphology using cell-impermeant AlexaFluor594 wheat germ agglutinin, which binds selectively to -acetylglucosamin and -acetylneuraminic (sialic) acid residues and provides highly selective labeling of the plasma membrane (Invitrogen). Cells were prepared as described previously (). Stacks of 16-bit fluorescent images (z step of 0.1 μm) were captured with a MetaMorph-driven microscope (DMRB; Leica) using a 63× NA 1.4 Apochromat oil immersion objective (Leica), a piezo stepper (E662; Physik Instruments), and a camera (CoolSNAP HQ2; Photometrics). Epifluorescence of all images (in 2D and 3D) were restored and deconvolved with Huygens (Scientific Volume Imaging) using the maximum likelihood estimation algorithm. Restored stacks were processed with Imaris (Bitplane) for visualization. The restored images were saved as tiff files that were mounted using Photoshop and Illustrator (Adobe). For immunofluorescence of cells in 3D Matrigel network, cultures were transfected and treated with TAT-C3, Y27632, or H1152 as monolayer followed by trypsinization and embedding into Matrigel in transwell cell culture chambers (as previously described in the Invasion assay section except that cells were allowed to invade through the gel for 6 h). Controls were left untreated. 50-μm cryosections were then cut at −20°C before fixation in 4% PFA in PBS followed by a 2-min permeabilization in 0.5% Triton X-100. Immunolabeling with various antibodies was performed as in the previous section. Cells were analyzed using a microscope (LSM510 Meta; Carl Zeiss MicroImaging, Inc.) with a 63× NA 1.32 Apochromat water immersion objective (Carl Zeiss MicroImaging, Inc.) and a photomultiplicator. Stacks of images were captured with LSM510 expert mode and were restored as described in the previous section. Time-lapse microscopy was performed on an inverted microscope (DL IRBE; Leica) equipped with differential interference contrast (DIC) and GFP optics using a 63× NA 1.3 oil-immersion objective (Leica), sample heater (37°C), and a CO incubation chamber (Leica). Images were captured with a CCD camera (MicroMax 1300; Princeton Instruments) using MetaMorph 6 software (Molecular Devices), converted to tiff files that were edited with ImageJ (National Institutes of Health), and compiled into QuickTime videos (Macintosh). The exposure time was set to 50 ms. All videos were recorded at the frequency of one image every 5 s. For 3D videos, we devised home-made bicompartment chambers comprised of an 8-mm metal ring placed in the center of a 30-mm Petri dish. The cells were mixed with serum-free Matrigel, and the resultant suspension was poured into the well formed by the metal ring. The assembly was placed in an incubator for 30 min while the Matrigel solidified, after which the metal ring was removed. The resultant disc of Matrigel was surrounded with 10% serum-complemented medium, taking care not to allow any medium to flow on top, thus exposing the cells to a lateral serum gradient. Cells close to the periphery of the Matrigel disc were time lapsed to monitor their ability to migrate along the serum gradient. Z sections recorded over a series of time points were combined into QuickTime videos. Cell velocity () was determined by measuring the speed of 10–12 cells, and the values are the means of three independent experiments. Video 1 shows MEFs move using an elongated mode of motility in 3D matrix. Video 2 shows p53 MEFs move using a rounded mode of motility in 3D matrix. Video 3 shows that rounded blebbing movement is provoked by p53 deficiency in 3D matrix. Video 4 shows that elongated movement is driven by p53 activity in 3D matrix. Video 5 shows the spreading of a p53 MEF treated with Y27632. Video 6 shows that ROCK-Δ1 promotes a rounded morphology. Video 7 shows that rounded movement is promoted by ROCK-Δ1 in 3D matrix. Videos 8 and 9 show that p53 deficiency–driven rounded morphology depends on ROCK activity in 3D matrix. Video 10 shows that p53 deficiency–driven rounded morphology is prevented by RhoE in 3D matrix. Fig. S1 shows that ROCK mediates p53-dependent regulation of the subcellular distribution of integrin β1 and ezrin in rounded migrating cells cultured in 3D Matrigel. Online supplemental material is available at .

The mammalian centrosome contains two barrel-shaped centrioles made of nine microtubule triplets, surrounded by a proteinaceous pericentriolar matrix. In 1898, Henneguy and Lenhossék independently observed that the centrioles of the centrosome and basal bodies that anchor ciliary and flagellar axonemes are identical structures (); we will use centriole to refer to the free structure and basal body to refer to the structure at the base of cilia. In addition to flagellated sperm cells, many other animal cells generate cilia (). The majority of cells produce a single immotile cilium, the primary cilium, that transduces mechanical and chemical signals from the extracellular environment (). Sensory cilia on specialized retinal, olfactory, and auditory cells are also essential for communicating sensory stimuli to the nervous system. Multiple motile cilia are made by ciliated protists, flagellated sperm of lower plants, and certain animal epithelial cell types. In mammals, multiciliated epithelium is found in the airways, the oviduct, and the ventricular system of the brain. Each of hundreds of basal bodies in multiciliated epithelial cells anchors a motile cilium; the concerted beating of cilia propels substances over the epithelial surface. Receptor proteins have been found in the ciliary membrane of motile cilia as well, suggesting that both types of cilia might function in signaling (; ). Intraflagellar transport (IFT), which involves the bidirectional trafficking of molecules along the axonemal microtubules, is common to all types of cilia and is required for axoneme formation and ciliary signal transduction (). In mammals, a hypomorphic mutation in polaris (also known as IFT88/Tg737), a core component of the IFT machinery, results in shorter or absent primary cilia in kidney epithelial cells and leads to polycystic kidney disease (). Polaris mutation also results in sparser, shorter motile cilia in ventricular epithelial cells (). However, because of the embryonic lethality of the polaris-null mutation in mouse (), the function of polaris and IFT in general has not been fully characterized in ciliated epithelial cells. In contrast to cycling cells, multiciliated cells have the ability to assemble hundreds of centrioles. EM shows that these centrioles arise through two parallel pathways initiated in the vicinity of the cell's existing centrosome (; ). In the centriolar pathway, multiple new centrioles form around an existing mother centriole, similar to the process in cycling cells, with the exception that only a single centriole is generated there. In the acentriolar pathway, by which the majority of centrioles in multiciliated cells are generated, new centrioles form around the deuterosome, a non–microtubule-based structure. In both cases, protein-rich fibrous granules are found surrounding the elongating centrioles. Centrioles assemble in the cytoplasm and then move to the apical cell surface, where they align at the plasma membrane and begin forming the ciliary axoneme. Centriole formation in ciliating cells differs from centrosome duplication in normal cycling cells in four key ways: more than two daughter centrioles are generated in the presence of the existing centrosome, a mother centriole simultaneously nucleates more than one daughter centriole, noncentriolar structures (deuterosomes) nucleate multiple centrioles, and centrioles are generated in nondividing cells. Despite these differences, ciliogenesis and centrosome duplication produce seemingly identical structures, raising the possibility of a common regulatory mechanism. A potential common regulator is SAS-6, a conserved centrosomal protein that is required for the initial steps of centriole formation in (; ; ). In human cells, HsSAS-6 depletion by RNAi blocks centriole assembly, and overexpression leads to the formation of excess centrosomal foci (). The creation of hundreds of centrioles is likely to require a dramatic increase in the expression and transport of constituent proteins. In cycling cells, some centrosomal proteins rely on dynein-mediated transport for localization. Ninein, centrin, pericentrin, and other centrosomal proteins are found in centriolar satellites (), dynein-containing protein complexes that traffic toward microtubule minus ends (; ). Two centriolar satellite proteins, PCM-1 and BBS4, are thought to tether cargo proteins to dynein, and in their absence, assembly of cargo proteins at the centrosome is decreased (; ). PCM-1 also localizes to the abundant fibrous granules found in the ciliating cell cytoplasm () and, therefore, might be particularly important for ciliogenesis. Comparative genomic (; ), proteomic (; ), and gene expression () studies have identified conserved ciliary components and potential regulators; however, mechanistic understanding of centriole formation in ciliated epithelial cells remains limited. The forkhead family transcription factor, Foxj1, is uniquely expressed in cells with flagella or motile cilia (). mice make centrioles but lack motile multicilia due to failure to anchor centrioles at the plasma membrane (). Several proteins important for centrosome structure and function, including γ-tubulin, centrin, and pericentrin, localize to the basal body region of ciliated epithelial cells (; ; ), but their role in ciliogenesis has not been tested. Many genes required for basal body and axoneme formation have been characterized in the biflagellate alga (; ; ) and in multiciliated cells, such as and (; ). Despite many conserved elements, ciliated epithelial cells possess features that cannot be modeled by lower organisms. For example, has only two basal bodies, and in multiciliated protists, basal bodies assemble on the cell cortex rather than in the cytoplasm. Here, we report the development of a model system using in vitro differentiated, multiciliated epithelial cells to study the pathway of centriole assembly during ciliogenesis. We define the localization of centrosomal proteins during the process and examine the role of the IFT component polaris, the mouse orthologue of the centrosomal protein SAS-6, and PCM-1–containing fibrous granules in centriole formation. We adapted the mouse tracheal epithelial cell (MTEC) culture system developed by to study centriole formation during ciliogenesis. The culture is started by seeding freshly isolated tracheal cells onto a porous filter suspended in medium. Cells proliferate into a confluent, polarized epithelium while submerged in medium. Although ciliated cells are present in the isolated tracheal cell population at the time of plating, most are unable to attach to the filter; therefore, the ciliated cells that appear later in the culture are due to in vitro differentiation (). Ciliogenesis is initiated by altering the medium and creating an air–liquid interface (ALI) by supplying medium only from below the filter. We defined three phases of the culture based on landmark events observed by light and electron microscopy (). No ciliogenesis occurs during the pre-ALI phase comprising the first 5 d of culture. Ciliogenesis begins in the second phase, a period of 2–3 d after ALI creation when centriole formation begins, but cilia are not yet detected at the surface. The third phase consists of a period of active ciliogenesis leading to a maximally ciliated epithelium at ∼14 d after ALI creation. The timing of experiments in this paper is reported in days relative to ALI creation (noted as “day ALI ± of culture”). The mature culture contains many fully ciliated cells () with occasional cells at earlier steps of ciliogenesis. In day ALI + 14 cultures, typically 40–60% of cells are ciliated, consistent with a previous report (). Transmission EM (TEM) of day ALI + 2 cultures revealed deuterosomes and fibrous granules in the apical cytoplasm of ciliating cells (), similar to structures seen in vivo (). These results indicate that in vitro ciliogenesis proceeds through the same steps as in vivo. Finally, cultured MTECs acquire cilia over the course of several days (), similar to the timing of ciliogenesis during airway development and tracheal epithelium reformation in vivo after damage (). Centriole formation during in vitro ciliogenesis was characterized by immunofluorescence localization of the marker proteins γ-tubulin and Cep135 (centrosomes and centrioles), acetylated α-tubulin (centrioles, cytoplasmic, and axonemal microtubules), and ZO-1 (epithelial cell boundaries). Before ciliogenesis, MTECs formed a confluent, polarized epithelium that appeared to consist of multiple cell layers (), resembling the pseudostratified tracheal epithelium in vivo. Because of the multilayered nature of the in vitro cultures, a maximum projection of deconvolved image planes through the entire epithelium showed a variable number of γ-tubulin–labeling centrosomes within a cell boundary (, image layers 1–39). However, a yz projection showed that the centrosomes are actually found at different depths, as expected for multiple cell layers. Most subsequent depictions include only the apical portion of the image stack (, image layers 1–10). Although cell boundaries are not always shown, all images are of fully confluent epithelia, except where noted. We identified four stages (stages I–IV) of centriole formation in MTECs during ciliogenesis (). At the time of ALI creation, the culture consisted of nonciliated cells that were no longer proliferating. Most cells had two separated centrosomes; however, each had only a single centrin-labeling centriole, consistent with G1 cells in which the centriole pair had separated (not depicted). Most cells had a primary cilium extending from one of the two centrioles (, nonciliated). The first detectable sign of centriole formation was at stage I, when foci of centrosomal proteins appeared near the centrosome in the apical cytoplasm (, stage I). The foci formed at approximately day ALI + 2 and preceded the formation of centrioles, based on the absence of acetylated α-tubulin labeling at the foci. The appearance of cytoplasmic foci coincided with an increase in the amount of centrosomal proteins at the existing centrosome (, stage I). Primary cilia in these cells were approximately fivefold longer than in nonciliated cells (nonciliated = 0.98 ± 0.09 μm, stage I = 5.39 ± 0.04 μm; see the supplemental text, available at ). Ciliating cells also had more cytoplasmic microtubules during stage I (Fig. S1 A), and these microtubules were more resistant to depolymerization than those of neighboring nonciliating cells (Fig. S1 B). During stage II, centrosomal proteins began to localize to a single dense cluster per cell (, stage II). Acetylated α-tubulin labeling indicated that centrioles were present in these clusters. In most ciliating cells, this nascent centriole cluster was closer to one side of the cell, as judged by cell boundary labeling, but the orientation of this eccentric localization in neighboring ciliating cells appeared to be random. Primary cilia were no longer present on ciliating cells beginning in stage II, although they remained on adjacent nonciliating cells (unpublished data). In stage III, centrioles dispersed from the cytoplasmic cluster toward the plasma membrane (, stage III). Axoneme formation began during stage IV, shortly after the centrioles reached the plasma membrane but before all centrioles were distributed evenly at the surface. Basal bodies in mature ciliated cells were evenly distributed at the apical membrane, and each anchored a cilium (, stage IV) of 2.89 ± 0.09 μm mean length. Cells in the stages defined above appeared sequentially during the culture period (), suggesting that these morphological states represent stages in the pathway of ciliogenesis. The above results suggest that the accumulations of material in ciliating cells previously observed by EM likely represent the accumulation of centrosomal material before assembly into centrioles. In addition to γ-tubulin and Cep135, we examined the localization of many centrosomal proteins during ciliogenesis (Fig. S1 C, centriolin). All tested proteins localized to centrosomes in nonciliated cells and to cytoplasmic foci and centrioles during ciliogenesis (Table S1, available at ). Ciliogenesis is accompanied by an increase in centrin expression (), and we tested whether other centrosomal proteins are also up-regulated. The MTEC culture contains both ciliated and nonciliated cells, and to analyze ciliated cells specifically, we cultured tracheal epithelial cells from a transgenic FOXJ1/EGFP mouse strain that expresses EGFP under the control of the ciliated cell–specific promoter (; ). We found that in mature cultures of these cells, all ciliated cells were EGFP+, and all EGFP+ cells were ciliated (unpublished data). FOXJ1/EGFP expression began at about day ALI + 2 of culture (), coinciding with the first signs of centriole formation (, stages I and II). Thus, FOXJ1/EGFP is a useful marker for both early ciliating and mature ciliated cells. To assay centrosomal protein abundance in ciliating cells, EGFP+ cells from a ciliating FOXJ1/EGFP culture were obtained by FACS at day ALI + 4 of culture (). Relative protein levels were compared by Western blotting for the centrosomal proteins indicated in from cell lysate prepared from equal numbers of cells. The examined proteins were 2- to 86-fold more abundant in ciliating (EGFP+) than nonciliated (EGFP−) cell types (). In sum, these results indicate that many centrosomal proteins are up-regulated during ciliogenesis, appear in cytoplasmic foci at the site of centriole assembly, and are recruited to the centrioles. To determine the role of individual proteins in ciliogenesis, we used the in vitro culture system to interfere with their function. For this purpose, we developed a means of efficiently introducing RNAi constructs into MTECs using lentiviral infection (see Materials and methods). We chose to focus on three proteins, representing different functional classes: the IFT component polaris; the centrosome component SAS-6; and PCM-1, a component of fibrous granules. The localization of the IFT component polaris in MTECs was determined by immunofluorescence (). Polaris localized along both primary cilia and motile cilia (, nonciliated, stages I and IV) in a punctate pattern with enrichment at the base and tip of the axoneme, consistent with previous work (). During ciliogenesis, polaris was present on the enlarged primary cilium in stage I but was not in pericentrosomal cytoplasmic foci, like most other centriolar components (, stage I). Polaris colocalized with nascent centrioles during stages II and III and then with axonemes in stage IV (, stages II–IV). Polaris was fivefold more abundant in ciliating cells (FOXJ1/EGFP+) than in nonciliated cells (FOXJ1/EGFP−) at day ALI + 4 and 14-fold more abundant at day ALI + 10 (Fig. S2 A, available at ). A lentivirally expressed short hairpin RNA (shRNA) construct targeting polaris was used to address its role in motile multicilia formation. The lentivirus effectively depleted polaris from NIH/3T3 cells (Fig. S2 B) and disrupted the formation of primary cilia in NIH/3T3 cells and MTECs before ciliogenesis (unpublished data). To examine polaris function during ciliogenesis, MTECs derived from FOXJ1/EGFP mice were infected on day ALI − 2 and were assayed on day ALI + 10 of culture. Depletion by RNAi was demonstrated by decrease in polaris labeling in cells, with most having no detectable polaris signal (Fig. S2 C). Control and shRNA-treated cultures had similar numbers of FOXJ1/EGFP+ cells (unpublished data), indicating that polaris depletion did not affect the adoption of the ciliated cell fate. Complete polaris depletion blocked axoneme formation in FOXJ1/EGFP+ cells (156/156 cells; ). FOXJ1/EGFP+ cells with a partial depletion of polaris had normal (60/127), short and sparse (54/127), or absent (13/127) axonemes. At the same culture stage, virtually all FOXJ1/EGFP+ cells in control infected cultures had fully formed axonemes (182/189), indicating that ciliogenesis was nearly complete. In contrast to the effect on ciliary axoneme assembly, basal bodies were still present in FOXJ1/EGFP cells without detectable polaris signal (), although they were fewer in number (128 per cell; = 10) when compared with control MTECs (325 per cell; = 10) and less evenly distributed on the cell surface (). Extensive similarities between basal bodies and centrioles suggest that proteins involved in centrosome duplication may also be required for centriole assembly during ciliogenesis. We chose to focus on SAS-6, as it is a highly conserved centriolar protein that appears to specifically regulate centriole formation (; ; ), and our initial assessment showed that it localizes to basal bodies (Table S1) and is up-regulated during ciliogenesis (). Immunofluorescence and lentiviral expression of a SAS-6–GFP construct in MTECs showed that SAS-6 distribution was similar to that of other centrosomal proteins throughout ciliogenesis ( and Fig. S3 A, available at ). SAS-6 localized to pericentrosomal cytoplasmic foci during stage I and to nascent centrioles from stage II and on ( and Fig. S3 A). SAS-6 did not colocalize precisely with either centrin or γ-tubulin on centrioles, suggesting that these proteins reside in different structural domains. Surprisingly, in fully ciliated cells from mature MTEC cultures, SAS-6 localized both to basal bodies and to the proximal region of axonemes (); this localization was confirmed with a second antibody to SAS-6 (not depicted). An xz projection through a mature ciliated cell from day ALI + 10 of culture revealed two distinct regions of SAS-6 labeling (, right, xz projection), with overlap between the top domain and the acetylated α-tubulin signal marking the axonemal microtubules. In a newly formed ciliated cell from day ALI + 5, the xz projection showed only the characteristic basal body localization, a single SAS-6–labeling region distinct from the axoneme (, left, xz projection). This axonemal localization in mature cells was unique to SAS-6 among the analyzed basal body components and suggested that it might be involved in both basal body and axoneme formation. A lentivirally expressed shRNA targeting SAS-6 was used to determine its role during ciliogenesis. The lentivirus effectively depleted SAS-6 from NIH/3T3 cells (Fig. S3 B), and consistent with published results (), SAS-6 depletion blocked centriole formation during mitotic cycles in NIH/3T3 cells (Fig. S3 C). To examine SAS-6 function during ciliogenesis, MTECs were infected on day ALI − 2 and were assayed on day ALI + 10 of culture. RNAi resulted in a substantial, but not complete, depletion of SAS-6 in most infected cells, as judged by decrease in the intensity of SAS-6 antibody signal (Fig. S3 D). Cep135 antibody labeling showed that most of the ciliated cells (147/150) from the control population had a full complement of mature basal bodies (). In contrast, most of the SAS-6 shRNA–depleted ciliated cells (112/150) had only a small number of Cep135-labeling dots at the cell surface (). These dots also contained acetylated α-tubulin, confirming that they were centrioles and not foci of centrosomal material (). Control cells contained a mean of 317 centrioles/cell ( = 10), whereas SAS-6 shRNA–treated cells had a mean of 33 centrioles/cell ( = 10). Interestingly, ciliary axonemes were also absent in depleted cells (), raising the possibility that SAS-6 is also involved in axoneme formation, although it is also possible that the few basal bodies that form under SAS-6 depletion are abnormal and are not capable of initiating axoneme formation. A hallmark of centriole formation in ciliated epithelial cells is the presence of PCM-1–containing fibrous granules in close proximity to nascent centrioles (). These might be transporting centriolar proteins by analogy with the purported role of centriolar satellites in dividing cells (). In nonciliated cells, PCM-1 localized to an apical layer of disperse cytoplasmic granules similar to centriolar satellites (, nonciliated). Unlike in cycling cells, these granules were not clustered around the centrioles marked by lentivirally expressed centrin2-GFP. In the transition to stage I, PCM-1 formed several large aggregates around the existing centrioles (, stage I). These PCM-1 aggregates appeared at the same time as, and partially colocalized with, the cytoplasmic foci of centrosomal proteins in stage I. In stage II cells, PCM-1 aggregates were smaller and more dispersed, in close association with the nascent centrioles (, stage II). These aggregates tracked with the centrioles to the plasma membrane through stage III, while continuing to decrease in size and abundance (, stage III). In mature ciliated cells, the little remaining PCM-1 was associated with basal bodies at the apical surface (, stage IV). PCM-1 distribution was supported by exact colocalization of the endogenous protein with lentivirally expressed, GFP-tagged PCM-1 (Fig. S4 A, available at ; nonciliated). PCM-1 also localized to fibrous granules in ciliating MTECs by immuno-EM, confirming previous results for PCM-1 localization (; Fig. S4 B). Furthermore, similar to cycling cells (), PCM-1 also colocalized with cytoplasmic granules of ninein and centrin in both nonciliated cells and during stage I of ciliogenesis (Fig. S4 C [ninein] and not depicted [centrin]). These results are consistent with the observed PCM-1–labeling structures in ciliating cells being analogous to centriolar satellites. A lentivirally expressed shRNA construct targeting PCM-1 was used to assess its role in centriole generation. The lentivirus effectively depleted PCM-1 from NIH/3T3 cells (Fig. S4 D), and consistent with published results (), PCM-1 depletion led to decreased centrosomal accumulation of ninein and centrin in shRNA-treated NIH/3T3 cells (unpublished data). To determine the function of PCM-1 during ciliogenesis, MTECs derived from FOXJ1/EGFP mice were infected on day ALI − 2 and were assayed on day ALI + 10 of culture. RNAi resulted in a range of depletion as judged by PCM-1 antibody labeling, with most cells having no or very few PCM-1 granules. Similar to NIH/3T3 cells, PCM-1–depleted MTECs had decreased amounts of centrosomal ninein and γ-tubulin (Fig. S4 E [ninein] and not depicted [γ-tubulin]). Remarkably, PCM-1–depleted FOXJ1/EGFP+ cells showed no observable defects in either centriole or axoneme formation (). All examined centrosomal proteins were present on mature basal bodies, although often at reduced levels (, γ-tubulin), suggesting that PCM-1 depletion did indeed affect protein targeting in fully ciliated cells but that this did not prevent centriole formation. As substantial depletion of PCM-1 had no effect on ciliogenesis, we attempted to interfere with PCM-1 by lentivirus-mediated overexpression of BBS4, which results in the formation of large aggregates trapping PCM-1 and associated cargo proteins in cycling cells (). In some infected cells, BBS4-myc perfectly colocalized with PCM-1 granules (, left), but in the majority of cells, infection resulted in aggregates containing BBS4-myc, PCM-1, and some centrin and ninein (, right [PCM-1], , image layer 22 of 31 [centrin], and not depicted [ninein]). Although the aggregates effectively trapped all visible PCM-1, cells with BBS4-myc–induced aggregates still assembled centrioles (). Individual z slices of fully ciliated cells revealed that the induced aggregates still contained centrosomal proteins and occasionally trapped centrioles in the cytoplasm (, image layer 22 of 31). The combination of the RNAi results and BBS4-induced aggregation suggests that normal levels of PCM-1 are dispensable for centriole formation and ciliogenesis. We have investigated the process of ciliogenesis in MTECs, as a model for centriole formation. These cells are unique in that they generate hundreds of centrioles during differentiation, whereas dividing cells produce only two centrioles per cell cycle. We adapted an established in vitro culture system for these cells and examined centriole formation with molecular markers. We found that proteins defined as centrosomal in dividing cells also localized to the basal bodies of ciliated cells and that these proteins were up-regulated during ciliogenesis. Based on the localization of these proteins, we identified four stages of centriole assembly, which are consistent with previous EM studies of the process. We used the culture system to investigate the role of three proteins in ciliogenesis—polaris, SAS-6, and PCM-1—and found that polaris and SAS-6 are required for distinct stages in ciliogenesis. Although PCM-1, a component of fibrous granules, colocalized with particles containing centrosomal proteins during ciliogenesis, normal levels of PCM-1 were not required for ciliogenesis. Here, we consider the implications of these findings. We characterized the pathway of centriole and axoneme formation during ciliogenesis by observing the localization of centrosomal proteins; this process had been previously studied only by EM (; ). We have shown that most centrosomal proteins behave as expected for being components of the protein-rich particles observed by EM during ciliogenesis, appearing first as small foci in the apical domain of cells, before the appearance of centrioles. Similar concentrations of centrosomal proteins have been observed during de novo centriole formation in , in which a protein complex containing γ-tubulin, pericentrin, and myosin II forms before centrioles appear (). Also, cytoplasmic foci of centrin were seen before the de novo generation of centrioles in HeLa cells in which the original centrioles were destroyed (). Given that centrosomal protein levels increased greatly during the early phase of ciliogenesis, this suggests that pools of precursor material are amassed in particulate form and deposited at the site of assembly before incorporation into centrioles. It will be important to determine the full extent of gene expression changes that take place specifically in ciliating cells, as it is possible that the transcriptional program is ultimately responsible not just for increased levels of structural components of centrioles, but also of the regulators that allow the assembly of hundreds of centrioles. We found that centrioles formed in ciliating cells and centrosomes in dividing cells have similar protein constituents, including both centriolar and pericentriolar matrix proteins. This suggests that centrioles, even when acting as basal bodies, are associated with proteins normally thought of as being limited to the cycling cell centrosome. We noted a key difference in centriole maturation during ciliogenesis: proteins that are specific to the mature mother centriole in cycling cells, such as ninein and -tubulin, localized to new centrioles in ciliating cells with timing similar to that of other components. Thus, the normal maturation cycle by which a centriole acquires these proteins is bypassed in ciliating cells, perhaps because the relative abundance of these proteins in ciliating cells drives their association with centrioles. Similar to published results from mouse mutants (; ), we found that the IFT component polaris is required for ciliary axoneme formation in MTECs. In addition, polaris depletion caused a modest decrease in centriole number in MTECs. It is possible that polaris directly affects centriole assembly, as it localizes to nascent centrioles during ciliogenesis, and it has recently been described as a functionally important component of the centrosome (). Another possibility is that polaris is involved in generating a regulatory signal for induction of the ciliogenesis transcriptional program; this could, for example, be transduced by the elongated primary cilium found during stage I of ciliogenesis. Finally, decreased numbers of centrioles could result from degeneration of basal bodies that failed to form axonemes. Future experiments are required to determine whether the polaris depletion phenotype is unique to polaris or is common to disruption of IFT in general. Our results show that polaris, and possibly IFT, is required for ciliogenesis in the multiciliated epithelium. Because several axonemal components have been implicated in human disease, ciliopathies should be examined to investigate the possible phenotypic contribution of motile cilium defects, in addition to the established defects in primary cilium function. We found that SAS-6 is a direct effector of centriole assembly during ciliogenesis. This is consistent with the role of SAS-6 in centrosome duplication in human cells and in embryos (; ). In worms, SAS-6 is required for the early steps of procentriole formation, and SAS-6 depletion results in failure to form complete centrioles (). Our results show that SAS-6–depleted MTECs are able to assemble only a few centrioles, presumably because of residual SAS-6 protein in depleted cells. Interestingly, SAS-6 was found at the basal bodies in newly formed ciliated cells but localized to both basal bodies and axonemes in more mature ciliated cells. During this period, the ciliary axoneme is thought to become fully functional by acquiring additional length, motility, and perhaps other components associated with ciliary signaling. This redistribution was unique to SAS-6 among basal body proteins, and it is possible that SAS-6, in addition to regulating centriole formation, might also function later in axoneme maturation. Fibrous granules, defined by EM, have been observed in ciliating epithelial cells in many tissues and organisms. They are known to contain PCM-1 and are likely to be identical to the granules in cycling cells containing PCM-1, dynein, and centrosomal components that traffic on cytoplasmic microtubules and are enriched around the centrosome (). However, the formation of centrioles and axonemes proceeded normally in PCM-1–depleted cells. This is consistent with the presence of respiratory cilia in mice deficient for BBS4, another component of PCM-1 granules (). Although we interfered with PCM-1 function in two ways, it is possible that residual PCM-1 activity was sufficient to support the formation of centrioles or that we were not able to detect subtle kinetic differences in ciliogenesis. Conclusive results regarding the role of PCM-1 in ciliogenesis will likely require examination of the process in the absence of the protein in PCM-1–null cells. Our results from SAS-6 depletion suggest that common mechanisms control centriole assembly during centrosome duplication and ciliogenesis. However, there are some differences in the processes that will ultimately have to be resolved. For example, recent results () suggest that in cycling cells, separase activity is required to disengage centrioles at the end of mitosis to allow duplication in the subsequent cell cycle, thus limiting centriole assembly to two new centrioles per cell cycle. In contrast, in ciliating epithelial cells, many centrioles grow orthogonally to existing centrioles or deuterosomes and then dissociate from these structures as ciliogenesis progresses, without passage through mitosis. It is unknown whether this dissociation is the same as anaphase disengagement of centrioles and whether separase activity is required. If the processes are analogous, then disengagement might control the availability of sites on the organizing structures and thus play a role in centriole number control during ciliogenesis. Ultimately, some mechanism must limit the number of centrioles formed in ciliating cells. Interestingly, we have noticed that although most ciliated cells in the MTEC culture have ∼300 basal bodies, there are occasional cells with a large apical surface area with >1,000 basal bodies, but distributed with a similar density (unpublished data). This suggests that cell size or apical surface area might control centriole number, but how this is communicated to the centriole assembly machinery is unknown. We have developed an in vitro culture system that undergoes ciliogenesis, can be manipulated by infection with viral vectors, and from which ciliated cells can be sorted on the basis of FOXJ1/EGFP expression. We believe that this is an ideal model system for addressing many of the outstanding questions in centriole and ciliary biology. Much recent attention has been focused on centriole generation, including number control and pathways of assembly (; ), and on centriole function in microtubule organization and cilium formation (; ). These processes have been studied individually in cycling mammalian cells, often under nonphysiological conditions, whereas in ciliated epithelial cells, they can all be observed as part of the naturally occurring ciliogenesis pathway. MTECs were derived from wild-type C3H × C57Bl/6J F1 hybrid or FOXJ1/EGFP transgenic mice (a gift from L. Ostrowski, University of North Carolina at Chapel Hill, Chapel Hill, NC) generated on C3H × C57Bl/6J F1 hybrid background (). FOXJ1/EGFP mice were bred by mating transgenic heterozygous males to wild-type C3H × C57Bl/6J F1 hybrid females (Taconic). Offspring were genotyped using PCR with EGFP-specific PCR primers. All procedures involving animals were approved by the Institutional Animal Care and Use Committee in accordance with established guidelines for animal care. NIH/3T3 and 293T cells were grown in DME with 10% FBS (Invitrogen). MTEC culture was based on . Mice were killed at 4–6 mo of age, and trachea were excised, trimmed of excess tissue, opened longitudinally to expose the lumen, and placed in 1.5 mg/ml pronase E in F-12K nutrient mixture (Invitrogen) at 4°C overnight. Tracheal epithelial cells were dislodged by gentle agitation and collected in F-12K with 10% FBS. Cells were treated with 0.5 mg/ml DNase I for 5 min on ice and centrifuged at 4°C for 10 min at 400 . Cells were resuspended in DME/F-12 (Invitrogen) with 10% FBS and plated in a tissue culture dish for 3 h at 37°C and 5% CO to adhere contaminating fibroblasts. Nonadhered cells were resuspended in an appropriate volume of MTEC Plus medium () and seeded onto Transwell-Clear (Corning) permeable filter supports at 10 cells/cm. The ALI was created ∼2 d after cells reached confluence, by feeding MTEC Serum-free or MTEC NuSerum medium () only from below the filter. Cells were cultured at 37°C and 5% CO and fed fresh medium every 2 d. Beating cilia were observed by phase microscopy 2–3 d after ALI creation. All chemicals were obtained from Sigma-Aldrich unless otherwise indicated. All media were supplemented with 100 U/ml penicillin, 100 mg/ml streptomycin, and 0.25 mg/ml Fungizone (all obtained from Invitrogen). HIV-derived recombinant lentivirus expressing GFP-tagged constructs were made from the lentiviral transfer vector pRRL.sin-18.PPT.PGK.GFP.pre () by inserting the ORF in frame with the GFP cassette at the AgeI site using PCR. Additional tagged cDNA constructs were made by inserting a PCR fragment of the tagged cDNA into the AgeI site of the lentiviral transfer vector pRRL.sin-18.PPT.PGK.IRES.GFP.pre () after removing the IRES and GFP sequences by digestion with NheI and BsrGI, blunting, and religation. Lentivirus encoding the mouse polaris shRNA (targeting nt 2164–2182; available from GenBank/EMBL/DDBJ under accession no. ) was made using the pSicoR PGK puro () lentiviral vector. Lentiviruses encoding the mouse SAS-6 (targeting nt 1273–1291; accession no. ) and PCM-1 shRNAs (targeting nt 1213–1231; accession no. ) were made using the pLentiLox3.7 () lentiviral transfer vector that also expresses GFP from a separate CMV promoter to mark infected cells. For infecting FOXJ1/EGFP MTECs, the modified lentiviral vector, pLentiRFP3.7, was generated by replacing the GFP cassette in pLentiLox3.7 with monomeric RFP using the NheI and EcoRI sites. shRNA constructs were verified by sequencing. Lentiviral vectors were propagated in XL2-Blue cells (Stratagene) and isolated from bacteria using the QIAfilter Maxi Plasmid Purification kit (QIAGEN). Recombinant lentivirus was produced by transient cotransfection of 293T cells with the appropriate transfer and lentiviral helper plasmids (pCMVDR8.74 packaging vector and pMD2.VSVG envelope vector; a gift from P. Kowalski, Stanford University, Stanford, CA; ) using the FuGENE6 transfection reagent (Roche Applied Science) or the calcium phosphate coprecipitation method. 18 h after transfection, cells were given fresh medium. The lentiviral supernatant was harvested 48–72 h after transfection and filtered though a 0.45-μm PES filter (Nalgene). Some lentiviral supernatants were concentrated 100- to 500-fold by ultracentrifugation at 20°C for 180 min at 50,000 . Lentiviruses were titered on NIH/3T3 cells by flow cytometry or immunofluorescence. Titers for preparations used were 10–10 infectious units/ml. To infect MTECs, medium was removed and cells were rinsed twice with PBS. Efficient lentiviral transduction of polarized airway epithelial cells only occurs at the basolateral surface (). To allow access to the basolateral surface, epithelial tight junctions were disrupted by treating cells with 12 mM EGTA in 10 mM Hepes, pH 7.4, at 37°C for 20 min. Cells were rinsed twice with PBS. Fresh medium was added to the bottom of the dish, and a mix of lentivirus, 5 μg/ml hexadimethrine bromide, and medium was placed on top of the cells. The plate was sealed with parafilm and centrifuged at 32°C for 80 min at 1,500 . After centrifugation, the plate was unsealed and placed at 37°C. Centrifugation greatly enhanced transduction efficiency in MTECs and had no adverse effects on cell morphology or viability. Epithelial junctions were completely reformed by 24 h after infection as monitored by ZO-1 antibody signal. Virus was removed 24 h after infection. Cells were assayed at least 48 h after infection; based on cytoplasmic GFP or monomeric RFP expression from the lentivirus, 20–50% of cells at the surface of the epithelium were transduced. Control infections were performed using virus made from transfer vectors without the transgene or short hairpin construct of interest. For indirect immunofluorescence, MTECs were rinsed twice with PBS and fixed in either methanol at −20°C for 7 min or 4% paraformaldehyde in PBS at room temperature for 10 min, depending on antigen. To preserve both cytoplasmic GFP signal and epitopes sensitive to aldehyde cross-linking, cells were fixed in 0.5% paraformaldehyde in PBS at room temperature for 5 min, followed by methanol at −20°C for 7 min. After fixation, cells were rinsed twice with PBS and filters were excised from plastic supports. Filters were cut in quarters to provide multiple equivalent samples for consistent observation. Cells were incubated two times for 5 min each in 0.2% Triton X-100 in PBS and blocked for 1 h at room temperature in 5% normal goat serum (Invitrogen) and 3% BSA (Sigma-Aldrich) in PBS. Primary antibodies were applied to filters at 37°C for 1 h or 4°C overnight. Alexa dye–conjugated goat secondary antibodies (Invitrogen) were applied to filters at room temperature for 30 min. Cells were incubated two times for 5 min each in 0.2% Triton X-100 in PBS between changes of antibody. Filters were mounted with 12-mm coverslips (1.5; Erie Scientific) using Mowiol mounting medium containing N-propyl gallate (Sigma-Aldrich). Cells were observed using Openlab 4.0.4 (Improvision) controlling a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.). Image stacks were collected with a z-step size of 0.2 μm and were then deconvolved and processed with AutoDeblur 9.3 and AutoVisualize 9.3 (AutoQuant Imaging). For a list of antibodies and appropriate fixation conditions used, see the supplemental text. To prepare a single cell suspension of MTECs and FOXJ1/EGFP MTECs for FACS, filters were incubated with a 1:1 mix of 0.5% Trypsin/EDTA (Invitrogen) and Cell Dissociation Solution (Sigma-Aldrich) at 37°C for 1 h. Cells were gently pipetted up and down every 15 min. Cells were washed in 1× PBS and resuspended in ice-cold PBS + 10% FBS at 10 cells/ml. EGFP+ and EGFP− cells were sorted with a FACStar (Becton Dickinson) sorter using a 488-nm Argon ion laser. For Western blotting on total cell lysates, sorted cells were rinsed in PBS and resuspended directly in SDS sample buffer at 10 cells/μl; 10 cells were loaded per well. Blots were blocked for 1 h in 5% milk in TBS + 0.05% Tween-20 and incubated overnight with primary antibody. Primary antibodies were detected with Alexa 635– conjugated goat secondary antibody (Invitrogen) and scanning with a Typhoon 9210 Variable Mode Imager using a 633-nm HeNe laser and an emission filter (670 BP 30; GE Healthcare). Images were quantitated with ImageQuant 5.2 (GE Healthcare). Fig. S1 shows microtubule and centrosomal protein distribution in ciliating cells. Fig. S2 shows polaris expression during ciliogenesis and polaris depletion by lentiviral RNAi. Fig. S3 shows SAS-6 localization during ciliogenesis and SAS-6 depletion by lentiviral RNAi. Fig. S4 shows PCM-1– containing fibrous granules during ciliogenesis and PCM-1 depletion by lentiviral RNAi. Table S1 shows localization of centrosomal proteins in MTECs. Table S2 lists antibodies used for immunofluorescence and Western blotting. Online supplemental material is available at .

The ER is a single contiguous compartment (; ; ) that is differentiated into at least three functionally distinct domains: rough ER, smooth ER, and nuclear envelope (; ; ). Cumulatively, these ER domains partition the nuclear contents from the cytoplasm and direct the synthesis of lipids, as well as membrane and secretory proteins (; ; ; ; ). The ER is also a signaling organelle that serves as a storage site for intracellular calcium and regulates its uptake and release into the cytoplasm (). Structurally, the ER network consists of membrane tubules, flattened sheets, and cisternae. The thickness of ER sheets is similar to the diameter of ER tubules, typically 60–100 nm, suggesting that common structural elements underlie these morphologically distinct forms (). The differing morphologies exhibited by ER domains likely contribute to their distinct functions. Rough ER, specialized for protein synthesis and folding, is often found in ribosome-studded sheets. In contrast, smooth ER, a site for lipid synthesis, contact with other organelles, and vesicle budding and fusion, lacks ribosomes and is often tubular (). The nuclear envelope, perhaps the most highly differentiated region of the ER, is a polarized sheet that regulates the movement of macromolecules between the nuclear space and the cytoplasm (; ). The membrane on one side of the sheet, the outer nuclear membrane (ONM), faces the cytoplasm, and on the opposite side of the lumen, the inner nuclear membrane (INM) faces the chromatin. Nuclear pores, gated channels between the cytoplasm and the nuclear interior, pass through both membrane bilayers and are sites where the INMs and ONMs are fused to each other (; ; ). Resident INM proteins pass from the ONM to the INM by diffusion or active transport through the nuclear pores and concentrate in the INM as a result of interactions with the underlying chromatin and the nuclear lamina (; ; ; ; ; ; ). Visualization in living cells has revealed the dynamic nature of the ER network. ER tubules in the periphery of mammalian cells continuously form and fuse, generating a meshwork characterized by the presence of “three-way” junctions between tubules that can move relative to one another (; ; ). The ER is also structurally reorganized during cell cycle progression. One prominent example is in animal cells, where the nuclear envelope disassembles during mitotic entry to promote spindle assembly. After the chromosomes separate in anaphase, nuclear envelopes reform around each of the separated chromatin masses (; ; ). The peripheral ER also undergoes cell cycle–dependent changes. In eggs from a variety of vertebrate and invertebrate species, there is a dramatic clustering of the peripheral ER network during mitosis (; ; ). This has been particularly well characterized in oocytes, where electron microscopy revealed the formation of “mitotic ER clusters” between 1 and 5 μm in diameter composed of packed smooth ER tubules and cisternae (). Relatively little is known about the factors that shape ER tubules and sheets, how the domains within the contiguous ER network maintain their distinct morphologies, or how transitions in the organization of the ER network during cell cycle progression are orchestrated. However, the development of systems for assembly of ER tubules from vesicles in vitro has led to some molecular insight. The reticulon family member Rtn4a was identified based on its modification by sulfhydryl reagents that inhibit the assembly of ER tubules (). Inhibition of Rtn4a by antibody addition to the in vitro reaction did not block vesicle fusion but prevented the fused vesicles from adopting an elongated tubule-like morphology. These results suggest that ER tubule formation requires both homotypic vesicle fusion and factors, including Rtn4a, that confer a tubular, rather than spherical, geometry. Rtn4a interacts with a second integral membrane protein called DP1/NogoA, and in , Ret1p and Yop1p, homologues of Rtn4a and DP1/NogoA, localize to the ER and play a functionally redundant role in the maintenance of peripheral ER tubules (). The mechanisms that drive homotypic fusion during the assembly of ER tubules remain unclear. In vitro, fusion requires GTP and is inhibited by a nonhydrolyzable GTP analogue (). In addition, a previous study demonstrated that the homotypic fusion of mammalian ER microsomes could be blocked by Rab GDP-dissociation inhibitor (GDI), implicating a Rab-type GTPase in the fusion reaction (). Here, we use the embryo as a model system to explore the molecular requirements for ER structure and dynamics. We show that simultaneous inhibition of YOP-1 and RET-1, the homologues of DP1 and Rtn4a, results in a defect in ER morphology that is most pronounced during mitosis, when the ER is coalesced. Using this phenotype as a guide, we examined embryos depleted of each of the 29 Rab family GTPases. Surprisingly, we found that specific depletion of the endosomal Rab-type GTPase, RAB-5, resulted in a defect in ER morphology that closely resembled depletion of YOP-1/RET-1. Both RAB-5 and YOP-1/RET-1 are also required for timely disassembly of the nuclear envelope during mitosis. Cumulatively, these results demonstrate a role for Rab5 in ER structure that is independent of its previously characterized functions during endocytosis and suggest that the morphology of the peripheral ER is important for nuclear envelope disassembly during mitotic entry. To explore the molecular requirements for ER structure, we used the first division of the embryo as a model system. In (), as in eggs from other vertebrate and invertebrate species (; ; ), the ER coalesces during mitosis into a reticular network of thick tubules and “mitotic ER clusters.” Because the interphase network of fine ER tubules is difficult to resolve at the light microscopic level, examination of the ER in its coalesced mitotic state provides a convenient means to identify defects in ER structure. We monitored ER dynamics by imaging embryos expressing a GFP fusion with the lumenal signal peptidase SP-12 (). After fertilization, the oocyte- and sperm-derived pronuclei migrate toward each other. The two pronuclei meet, and the nuclear/centrosome complex moves to the embryo center and rotates onto the long axis of the cell. Subsequently, the nuclear envelopes become permeable, and the mitotic spindle assembles (). During interphase, before pronuclear migration, GFP:SP-12 was present in the pronuclear envelopes as well as in a network of fine tubules and punctate structures dispersed throughout the cytoplasm. As the embryos entered mitosis, the ER coalesced to form a reticular network of thicker tubules and mitotic clusters (; and Video 1, available at ; ). The clusters began to form ∼80 s before the onset of nuclear envelope disassembly and were enriched around the centrosomes and mitotic spindle (, 0-s time point). The first embryonic division in is polarized, and mitotic ER clusters were concentrated near the cortex in the embryo anterior (, 0-s time point, cortical plane), similar to the cortical actomyosin cytoskeleton. After chromosome segregation in anaphase, the thick tubules and clusters abruptly dispersed, and nuclear envelopes re-formed around the separated chromosome masses, returning the ER to its interphase state (Video 1). Rtn4a and its associated protein DP1/NogoA have been implicated in ER tubule assembly in vitro (). We therefore characterized the localization and depletion phenotypes for their homologues, RET-1 and YOP-1. Both endogenous RET-1 () and a GFP fusion with YOP-1 (Fig. S1, A and B, available at ) localized to the ER. Like their vertebrate and yeast homologues (), RET-1 and YOP-1 concentrate in the peripheral ER and are largely excluded from the nuclear envelope, where they are de-enriched relative to the luminal ER marker SP-12 ( and Fig. S1 A). Consistent with their redundant functions in budding yeast, embryos depleted of YOP-1 or RET-1 alone exhibited no appreciable defects in embryo viability or ER structure (unpublished data). However, simultaneous depletion of YOP-1 and RET-1 dramatically altered ER morphology and reduced embryonic viability to ∼40% (473 of 1191 embryos survived to hatching). The ER morphology defect was particularly evident during mitosis, when the network is normally coalesced. In YOP-1/RET-1–depleted embryos, there were fewer thick tubules that appeared shorter and more poorly organized than in control embryos, and no mitotic ER clusters were formed (; and Video 1). We conclude that YOP-1/RET-1 play a critical role in ER structure. In addition, their simultaneous inhibition results in a phenotype that is easily detected by monitoring ER organization in mitotic embryos using light microscopy. A visual assay for formation of a tubular ER network from salt-washed ER-enriched membranes isolated from oocytes revealed that tubule assembly requires GTP, in addition to Rtn4a and DP1/NogoA (; ). To investigate the origin of this GTP requirement, we compared the ability of ER-enriched membranes to form tubules after incubation with either Rab or Rho GDI, which inhibit Rab or Rho family GTPases, respectively. Although Rho GDI failed to have any effect, addition of Rab GDI potently blocked ER tubule formation in vitro (). This result suggested that, in addition to YOP-1 and RET-1, ER structure is controlled by a Rab family GTPase. To investigate whether a Rab GTPase contributes to ER structural dynamics in vivo, we systematically depleted each of the 29 Rabs (Table S1, available at ). This analysis identified six Rab activities necessary for embryo production and/or early embryogenesis (). Of these, only reduction of RAB-5 levels resulted in an effect on ER structure similar to depletion of YOP-1/RET-1 (compare with ; Video 1). Depletion of RAB-5, which resulted in 100% embryonic lethality, inhibited both the formation of a reticular network of thick tubules and the appearance of mitotic ER clusters (; and Videos 1 and 2). In contrast to the dramatic effect on ER structure, staining of control and RAB-5–depleted fixed embryos with antibodies against a Golgi marker, the glucuronyltransferase SQV-8, revealed that Golgi size and distribution were not altered by RAB-5 depletion (). Rab5 has an established role in endocytosis and early endosome fusion (; ; ; ). Previous work in has shown that these functions are mediated by two RAB-5 guanine nucleotide exchange factors (GEFs), RME-6 and RABX-5, which have overlapping functions (). Examination of the ER in embryos mutant for in which RME-6 was depleted by RNAi revealed a structural defect essentially identical to that in embryos (), indicating that the role of RAB-5 in ER structure is also mediated by RME-6 and RABX-5. To determine whether the effect of RAB-5 depletion on ER structure is an indirect consequence of its effect on endocytosis, we inhibited endocytosis by depleting other proteins that function in the early endocytic pathway, including clathrin heavy chain (CHC-1; ), dynamin, and α-adaptin (not depicted). Depletion of CHC-1 redistributed clathrin light chain into the cytoplasm (not depicted) and resulted in a dramatic decline in cytoplasmic yolk granule density (), a hallmark of a pronounced defect in endocytosis during oocyte development (). However, despite a reduction in yolk density similar to that in RAB-5–depleted embryos, ER morphology was not perturbed by CHC-1 depletion (). RAB-5 still localized to punctate endosomal structures in CHC-1–depleted embryos (), although the RAB-5 structures were enlarged and relatively more concentrated in the embryo anterior compared with controls. Examination of ER structure in embryos individually depleted of candidate RAB-5 effector proteins, including the homologues of EEA-1, Rabenosyn-5, Rabaptin-5, and the catalytic subunit of a type 3 PI 3-kinase, hVps34 (), also failed to reveal any detectable alteration in ER morphology (). Depletion of RAB-5 also did not substantially alter the appearance of the microtubule cytoskeleton (Fig. S1 C), which has been shown to play a role in the distribution of the ER within cells (; ). In addition, mitotic ER tubules and clusters still formed in nocodazole-treated embryos that lacked detectable microtubule polymer (Fig. S1 D). As previous work suggested that the actomyosin cytoskeleton also plays a role in ER structure (), we examined the distributions of GFP fusions with myosin II (GFP:NMY-2; ) and the actin binding domain of moesin (GFP:Moe), an established probe for filamentous actin in the embryo (). In the period leading up to metaphase, the actin cytoskeleton appeared similar in control and depleted embryos (Fig. S1 E). A comparison of the distribution of GFP:NMY-2 in control, , and embryos revealed that during the establishment of polarity, the organization of cortical myosin II was also similar under all three conditions (Fig. S1 F). During mitosis, the organization of GFP:NMY-2 in embryos, although distinct from that in controls, was similar to that in the embryos (Fig. S1 G). These results suggest that although compromised endocytosis, which is common to both and embryos, does alter the organization of cortical NMY-2, this mild perturbation is not responsible for the impact of RAB-5 depletion on ER structure. Small GTPases cycle between an active GTP-bound form and an inactive GDP-bound form. To determine whether ER structure is sensitive to activation as well as inhibition of RAB-5, we constructed transgenic strains that stably expressed RFP () fused to either wild-type RAB-5 or a mutant form of RAB-5 that dramatically slows GTP hydrolysis (Q78L). Expression of the RFP fusion with wild-type RAB-5 did not affect either endosome distribution or ER morphology (Fig. S2 A, available at ). Expression of the RFP fusion with RAB-5, at ∼25% of the level of endogenous RAB-5 (), reduced the normal anterior bias of endosomes in the polarized one-cell embryo but did not dramatically alter endosome morphology as assessed by examination of the localization of endogenous RAB-5 (), or a GFP fusion with the endosomal marker EEA-1 (Fig. S2 F). However, expression of RFP:RAB-5 had a striking effect on ER structure, monitored by coexpression with GFP:SP-12. More mitotic clusters that were also considerably larger than those in control embryos were observed ( and Video 3). Additionally, the clusters failed to fully disperse after anaphase and often persisted into the following cell division (Video 3). Depletion of YOP-1 and RET-1 in embryos expressing RFP:RAB-5 revealed that YOP-1 and RET-1 are required for the dramatic changes in ER structure caused by expression of the dominant-active RAB-5 mutant (). These results demonstrate that ER structure is sensitive to either depletion or activation of RAB-5, exhibiting opposing responses to the two perturbations. Expression of mammalian Rab5A also potentiated the formation of mitotic ER clusters in human HeLa cells. Compared with early embryos, control HeLa cells exhibited relatively few mitotic ER clusters, as did cells expressing Rab5A, a dominant-negative version of Rab5A. In contrast, expression of wild-type Rab5A resulted in a small increase, and expression of Rab5A resulted in a dramatic and consistent increase in the number of mitotic ER clusters >0.5 μm in diameter (Fig. S3, A and B). Cumulatively, these results suggest a conserved role for RAB-5 in controlling ER structure. Depletion of RAB-5 or YOP-1/RET-1 has a similar effect on ER structure. In addition, YOP-1/RET-1 are required for RFP:RAB-5 to exert a dominant effect. These results raised the possibility that YOP-1 and RET-1 are downstream effectors of RAB-5. To determine whether YOP-1 and RET-1 are likely to be direct effectors of RAB-5, we tested whether recombinant FLAG-tagged YOP-1 or RET-1 could bind to GST:RAB-5 in vitro. However, no substantial binding of either protein to GST:RAB-5 prebound to either GDP or GTPγS was observed (Fig. S3 C). A role for YOP-1/RET-1 as downstream effectors of RAB-5 would also predict that simultaneous depletion of YOP-1/RET-1 would not enhance the effect of RAB-5 depletion on ER morphology. However, simultaneous depletion of YOP- 1/RET-1 with RAB-5 resulted in a dramatic synergistic defect evident throughout the cell cycle (). In the triple-depleted embryos, ER morphology was aberrant in interphase, when GFP:SP-12 was present in numerous loop-shaped structures. These ER loops accumulated around the spindle poles as the triple-depleted embryos entered mitosis, leaving the embryo periphery nearly devoid of ER (). In addition, as in the individual depletions, the ER failed to coalesce into a network of thick tubules and clusters ( and Video 4, available at ). Although it is possible that RAB-5 and YOP-1/RET-1 are on a single pathway and the synergistic defect in ER structure is due to a greater effectiveness in inhibiting this pathway when two of its constituents are depleted by RNAi, we favor the idea that the strong synergistic phenotype indicates that RAB-5 and YOP-1/RET-1 make distinct essential contributions to ER structure. It is worth pointing out, however, that the idea that RAB-5 and YOP-1/RET-1 each retain some function when the other is absent does not rule out the possibility that RAB-5 exerts its effect in part by altering the function of YOP-1/RET-1. The ER is a single contiguous compartment that includes the nuclear envelope. We were therefore interested in whether, in addition to the peripheral ER, inhibition of RAB-5 or YOP-1/RET-1 altered nuclear envelope dynamics. One clue that this might be the case came from the observation that depletion of either RAB-5 or YOP-1/RET-1 resulted in a characteristic “four-eyes” phenotype at the end of the first division, in which two nuclei instead of one re-formed in each daughter cell. In the zygote, after the two pronuclei meet, the nuclear envelopes are permeabilized and cleared from the region near the centrosomes, allowing spindle microtubules to interact with and align chromosomes before their segregation (; ). After anaphase chromosome segregation, a single nuclear envelope reforms around each of the separated chromatin masses, thereby mixing the haploid sperm and oocyte genomes. In embryos in which either RAB-5 or YOP-1/RET-1 was depleted, separate nuclear envelopes formed around the sperm- and oocyte-derived chromosomes after their segregation, resulting in two nuclei in each daughter cell (). To understand the basis for this phenotype, we filmed embryos coexpressing a GFP fusion with the INM protein LEM-2 and an RFP fusion with histone H2B. This analysis revealed that ∼35 s before anaphase onset, the juxtaposed oocyte and sperm pronuclear envelopes in control embryos undergo a scission event in close proximity to the aligned chromosomes and are cleared from the region between the chromosomes, allowing the chromosomes from the two pronuclei to mix and form a single nucleus after segregation ( and Video 5, left, available at ). In embryos in which either RAB-5 or YOP-1/RET-1 were depleted, this scission event never occurred, and the oocyte- and sperm-derived chromosomes remained separate during their segregation on the spindle (; and Video 5, right). Consequently, two nuclei were formed in each daughter cell, and the mixing of the genomes derived from the two gametes failed. The four-eyes phenotype observed in YOP-1/RET-1– or RAB-5– depleted embryos is highly unusual. Functional genomic analysis of cell division in identified only one other protein, the nuclear pore component NPP-12, whose depletion reproducibly results in this phenotype (). NPP-12 is the homologue of gp210, one of two integral membrane proteins associated with the nuclear pore (; ). Imaging of embryos coexpressing GFP:LEM-2 and RFP:histone confirmed that depletion of NPP-12 results in a defect in nuclear envelope disassembly similar to that resulting from depletion of RAB-5 or YOP-1/RET-1 (). However, analysis of GFP:SP-12 indicated that, in contrast to depletion of YOP-1/RET-1 or RAB-5, depletion of NPP-12 does not noticeably perturb peripheral ER structure (). These results indicate that the defect in nuclear envelope disassembly in the YOP-1/RET-1– and RAB-5–depleted embryos is likely to be a consequence rather than a cause of the defect in ER structure. They also point to an interesting functional connection between the role of peripheral ER structure in nuclear envelope disassembly and events occurring at nuclear pores. Diffusion of membrane proteins between the peripheral ER and INM is proposed to require energy-dependent restructuring that creates transient channels through the nuclear pore membrane (). Antibodies to gp210 inhibit diffusion between the peripheral ER and INM, suggesting that gp210 plays a role in this restructuring (). A role in promoting the diffusion of integral membrane proteins between the peripheral ER and INM would provide a plausible explanation for why perturbing gp210 or peripheral ER structure has a similar effect on nuclear envelope disassembly. To explore the phenotypic similarity of these two very different perturbations further, we used a series of assays to compare nuclear envelope disassembly in NPP-12– and RAB-5–depleted embryos. Nuclear envelope disassembly is accompanied by several distinct events: loss of the peripheral pore components, which renders the nuclear envelope permeable to macromolecules of progressively larger diameter (; ); removal of pores from the membrane; disassembly of the nuclear lamina; and release of resident INM proteins back into the peripheral ER (; ). Because the dependencies between these different events are not clear, we compared the effects of RAB-5 and NPP-12 depletion on all of them. To determine if nuclear envelope permeabilization was affected, we filmed embryos expressing GFP:histone H2B and measured the timing of nuclear envelope permeabilization with respect to the kinetics of chromosome condensation. Permeabilization was followed by monitoring the diffusion of free GFP:histone out of the nucleus, which occurs within 60 s of the entry of cytoplasmic 70-kD dextran into the nucleus (). We refer to the time point when the free nuclear GFP:histone fluorescence has equilibrated with the cytoplasm as nuclear envelope permeabilization. To analyze chromosome condensation, we used a recently developed image analysis method that can be used to quantitatively compare condensation kinetics between control and specifically perturbed embryos (). This analysis revealed that the kinetics of chromosome condensation were not altered by depletion of either RAB-5 or NPP-12. However, both perturbations resulted in a 50–60-s delay between the completion of chromosome condensation and nuclear envelope permeabilization (). Timing the release of a NLS-containing GFP fusion protein (GFP-LacI) from the nucleus relative to anaphase onset provided independent support that permeabilization is delayed by ∼50 s in RAB-5–depleted embryos (Fig. S4, A–C, available at ). To analyze the removal of nuclear pores from the envelope, we filmed embryos expressing a GFP fusion with NUP-155, a stable component of the pore wall (). Redistribution of NUP-155 from the nuclear rim to the cytoplasm was delayed by ∼50 s in RAB-5– and NPP-12–depleted embryos ( and not depicted). Disassembly of the nuclear lamina was also similarly delayed by both perturbations. Although a YFP fusion with the B-type lamin LMN-1 is nearly completely removed from the nuclear envelope by anaphase onset in control embryos, substantial amounts of LMN-1 remained at anaphase onset in RAB-5– or NPP-12–depleted embryos (). Fluorescence intensity measurements revealed that until anaphase onset, lamin disassembly is quantitatively similar between RAB-5– and NPP-12–depleted embryos (). We next investigated the effect of depletion of RAB-5 and NPP-12 on the release of resident INM proteins into the peripheral ER. In the embryo, the INM protein LEM-2 is present in the peripheral ER throughout the cell cycle. As the nuclear envelope forms in telophase, LEM-2 becomes selectively enriched in this ER domain (Fig. S4 E). Because the presence of LEM-2 in ER concentrated around mitotic spindle poles interfered with visualization of LEM-2 associated with the nuclear envelope, we examined the effect of RAB-5 depletion in (RNAi) embryos, which lack functional centrosomes and fail to build mitotic spindles (). In embryos depleted of SPD-5 alone, the peripheral ER exhibits normal dynamics and is coalesced during mitosis to form a network of thick tubules and clusters (). As expected, when both SPD-5 and RAB-5 were depleted, there was a profound defect in the formation of thick tubules and clusters during mitosis (). In embryos depleted of SPD-5 alone, LEM-2 is completely dispersed into the surrounding ER and then reaccumulates around the chromatin as it begins to decondense. In contrast, in embryos codepleted of either RAB-5 or NPP-12 with SPD-5, the redistribution of LEM-2 failed to occur (). The defects in ER morphology in RAB-5– and YOP-1/ RET-1–depleted embryos are most prominent during mitosis and could therefore reflect a specific role for RAB-5 and YOP-1/RET-1 in restructuring the interphase peripheral ER network during mitotic entry. Alternatively, YOP-1/RET-1 and RAB-5 may be required to maintain a normal ER network during interphase, which is then the substrate for mitotic coalescence. We favor the later possibility for three reasons. First, in embryos that are simultaneously depleted of YOP-1/RET-1 and RAB-5, obvious defects in ER structure are apparent in interphase (). Second, reticulon/Rab-dependent assembly of ER tubules in vitro from oocyte membranes is inhibited by mitotic but not interphase cytosol (). Third, defects in ER structure are apparent when Yop1p and the reticulons are deleted in budding yeast, regardless of cell cycle state (; ). Previous work has identified >20 different effectors that link Rab5 to a variety of cellular activities (). However, technical difficulties have prevented the isolation of integral membrane proteins that respond to Rab5 signaling. The fact that depletion of YOP-1/RET-1 has an effect on ER structure that is similar to that of depletion of RAB-5, and the requirement for YOP-1/RET-1 for RFP:RAB-5 to exert a dominant effect, make YOP-1 and RET-1 attractive candidates for integral membrane effectors that link RAB-5 to the ER. Interestingly, the budding yeast homologue of YOP-1, Yop1p, was identified based on the fact that a short amino acid stretch in its N terminus interacts with Yip1, a protein implicated in the dissociation of Rab GTPases from GDI. In addition, Yop1p was found to interact with multiple Rab-type GTPases in yeast, potentially via their prenylation motifs (). Although a conventional in vitro approach using GST:RAB-5 isolated from bacteria failed to uncover evidence for a direct interaction between RAB-5 and either YOP-1 or RET-1 (Fig. S3 C), it is possible that this interaction requires prenylation of RAB-5 and/or membrane association of YOP-1/RET-1, and hence a different approach may be needed to uncover an interaction. In addition to YOP-1 and RET-1, our data suggest the existence of additional effectors, as simultaneous depletion of RAB-5 exacerbates the defect in ER morphology present in embryos depleted of YOP-1/RET-1 alone (). A combination of genetic analysis in with in vitro biochemical reconstitution of ER tubules using membranes will likely be useful to uncover these important factors. RAB-5 was the only Rab GTPase whose depletion phenocopied the effect of YOP-1/RET-1 depletion on ER structure. This finding was surprising because RAB-5 localizes to endosomes and has well-studied functions in endocytosis and early endosome fusion (for reviews see ; ; ; ). Inhibition of endocytosis by other means, such as CHC-1 depletion, does not affect ER structure but also does not abolish localization of RAB-5 (), indicating that endosomes, although abnormal, are still present. In contrast to CHC-1 depletion, simultaneous inhibition of the two characterized RAB-5 GEFs, RME-6 and RABX-5, redistributes RAB-5 to the cytoplasm () and phenocopies the effect of RAB-5 depletion (). These results suggest that endosomal RAB-5 acts in trans to control ER structure. Consistent with this idea, ER tubule assembly in vitro does not require ongoing endocytosis but does require the function of a Rab-type GTPase. We propose that active RAB-5 on endosomes transiently interacts with effectors on ER membranes to promote their homotypic fusion (). Consistent with the idea that interactions with endosomes might alter ER structure, the ER has also been shown to contact several other organelles, including the Golgi, plasma membrane, lysosomes, and late endosomes (). In the one-cell embryo, the asymmetric clustering of the ER also mirrors the asymmetric distribution of endosomes (Fig. S2 A), consistent with a trans-acting mechanism. The hypothesis that RAB-5 acts in trans to control ER structure is similar to the proposal that the small GTPase Ran, together with chromatin, promotes the homotypic fusion of ER membranes to form the nuclear envelope (). In this case, a chromatin-activated small GTPase signals in trans to promote ER membrane fusion. Depletion of RAB-5 does not prevent nuclear envelope assembly, and depletion of RAN-1 does not disrupt ER structure (unpublished data), indicating that these two reactions, which both require fusion of ER membranes, depend on signaling from distinct small GTPases. Another instance in which ER structural changes might be triggered by signals presented in trans is during autophagy, when a double-membrane layer that may be derived from ER membranes forms around a portion of the cytoplasm to create an autophagosome (; ). Interestingly, recent work suggests a role for the Rab24 GTPase during starvation-induced autophagy, although its precise function in this process remains unknown (). Why would ER structure be controlled by trans-acting GTPases? One possibility is that trans-action restricts homotypic ER fusion to specific times and places within the cell to facilitate the formation of a network or envelope as opposed to large membrane aggregates. We show that depletion of YOP-1/RET-1 or RAB-5 leads to a defect in nuclear envelope disassembly in addition to disrupting the morphology of the peripheral ER. Further characterization of this defect revealed a striking phenotypic similarity to depletion of NPP-12, the homologue of the nuclear pore-associated integral membrane protein, gp210. Although both perturbations delay nuclear envelope permeabilization, pore removal, and lamina disassembly, the most pronounced consequence of either depletion is an essentially complete block in the release of INM components into the peripheral ER. RAB-5 and YOP-1/RET-1 may directly restructure the nuclear envelope in addition to the peripheral ER. However, we suspect that the contribution of RAB-5 and YOP-1/RET-1 to nuclear envelope disassembly occurs indirectly via their function in structuring the peripheral ER. Consistent with this idea, YOP-1/RET-1 and their homologues in other organisms are selectively enriched in the peripheral ER relative to the nuclear envelope (; Fig. S1 A; ). Previous work has shown that INM components disperse into the peripheral ER during mitosis after their release from chromatin and the nuclear lamina (; ). We therefore propose that disruption of peripheral ER morphology in RAB-5– and YOP-1/RET-1–depleted embryos prevents the diffusion of INM proteins, trapping them in the spindle region (). In this context, the fact that depletion of NPP-12/gp210 phenocopies the effect of disruption of the peripheral ER on nuclear envelope disassembly would support the previous suggestion that NPP-12/gp210 acts as a gatekeeper that positively regulates the diffusion of membrane proteins between the INM and peripheral ER (). We would like to emphasize that although depletion of RAB-5 or YOP-1/RET-1 results in defects in both mitotic clustering of the peripheral ER and in nuclear envelope breakdown, our data do not demonstrate that these phenotypes are linked (i.e., cluster formation stimulates nuclear envelope breakdown). It remains equally likely that these two phenotypes are independent consequences of disruption of peripheral ER structure. The generation of strains expressing fluorescent fusions with histone H2B (), LEM-2 (), LMN-1 (), SP-12 (), and NUP-155 () has been described previously. Other fluorescent fusions with YOP-1 (Y71F9B.3), RAB-5 (F26H9.6), and SP-12 (C34B2.10) were generated by cloning the unspliced genomic loci into the SpeI sites of pIC26 (GFP; ) or pAA64 (RFP; ). The activated form of RAB-5 (Q78L) was generated by subjecting the spliced genomic locus of F26H9.6 to site-directed mutagenesis before cloning into pAA64. Constructs were integrated into DP38 ( ()) as described previously () using a particle delivery system (Biolistic PDS-1000/He; Bio-Rad Laboratories, Inc.). All strains used in this study are listed in Table S2 (available at ). The strains expressing GFP:2xFYVE and GFP:NUP-155 were provided by B. Grant (Rutgers University, Piscataway, NJ) and P. Askjaer (Parc Científic de Barcelona, Barcelona, Spain), respectively. Double-stranded RNA (dsRNA) was prepared as described previously () from templates prepared by using the primers listed in Table S3 (available at ) to amplify N2 genomic DNA. For complete and partial depletions, L4 hermaphrodites were injected with dsRNA and incubated at 20°C for 45 or 22–24 h, respectively, before analysis. Antibodies against RAB-5 and RET-1 were generated by cloning nucleotides 124–612 of F26H9.6 (RAB-5) and the entire coding sequence of W06A7.3b (RET-1), amplified from a cDNA library, into pGEX6P-1 (GE Healthcare). Purified GST fusion proteins were outsourced for injection into rabbits (Covance). Both antibodies were affinity purified from serum as described previously () by binding to columns of the same antigen after removal of the GST tag by cleavage with PreScission protease. Antibodies directed against SQV-8 and GFP have been described previously (; ). For analysis of fixed embryos, images were acquired using a 100×, 1.35 NA U-Planapo oil-objective lens (Olympus) mounted on a DeltaVision microscope system (Applied Precision) equipped with an Olympus IX70 base and a charge-coupled device camera (CoolSnap; Roper Scientific). Immunofluorescence of fixed embryos was performed as described previously (), using the following rabbit antibodies at a concentration of 1 μg/ml: α-RAB-5 (Cy3 labeled), the mouse monoclonal antibody DM1α (Oregon green 488 labeled; Sigma-Aldrich), α-SQV-8 (Cy-5 labeled), and the goat polyclonal GFP antibody (Oregon green 488 labeled). For live analysis, embryos were mounted as described previously () and imaged at 20°C on a spinning-disc confocal microscope (Eclipse TE2000-E; Nikon) equipped with a Nikon 60×, 1.4 NA Planapo oil-objective lens and a charge-coupled device camera (Orca-ER; Hamamatsu). To depolymerize microtubules, meiotic embryos were dissected directly into meiosis media (25 mM Hepes, pH 7.4, 60% Leibowitz L-15 Media, 20% fetal bovine serum, and 500 μg/ml inulin) containing 10 μg/ml nocodazole and imaged in a depression slide sealed with petroleum jelly. Quantification of ER cluster formation and lamin and LacI fluorescence intensity measurements were performed using MetaMorph software. Analysis of chromosome condensation was performed as described previously (). The mean value of the condensation parameter was calculated after aligning the image sequences with respect to nuclear envelope permeabilization. To simplify presentation, the plots of the condensation parameters are displayed aligned with respect to the onset of condensation. A light membrane fraction from egg extracts was prepared as described previously (), stained with octadecyl rhodamine, and mounted on glass slides to visualize tubule formation. To test the effect of GDI addition, membranes were preincubated for 10 min with either 10 μM Rab GDI or 30 μM Rho GDI, followed by the addition of 1 mM ATP and 0.5 mM GTP. Reactions were allowed to proceed for 60 min before staining and pipette transfer onto slides. Purified Rho GDI was provided by G. Bokoch (The Scripps Research Institute, La Jolla, CA), and the construct to express Rab GDI was a gift from W. Balch (The Scripps Research Institute). Fig. S1 shows that YOP-1 localizes to the ER and nuclear envelope, reorganization of the ER during mitosis does not require microtubules, and the defect in ER organization observed after RAB-5 depletion is not caused by a perturbation in the actomyosin cytoskeleton. Fig. S2 shows that RAB-5–containing early endosomes contact the ER more frequently than RAB-7–containing late endosomes, and EEA-1 localization is lost after RAB-5 depletion but only subtly perturbed upon RAB-5 expression. Fig. S3 shows that Rab5A overexpression potentiates ER cluster formation in HeLa cells, and neither YOP-1 nor RET-1 binds to GTP-loaded GST:RAB-5. Fig. S4 shows that nuclear envelope permeabilization is delayed in embryos depleted of RAB-5, GFP:2xFYVE does not localize to the nuclear envelope, and LEM-2 is present in the peripheral ER throughout the cell cycle. Table S1 lists the putative Rab-type GTPases in . Table S2 lists the strains used in this study. Table S3 lists the dsRNAs used in this study. Video 1 shows the ER morphology in embryos depleted of RAB-5 or YOP-1/RET-1. Video 2 shows that RAB-5 is required for ER structure and dynamics. Video 3 shows that expression of activated RAB-5 potentiates ER clustering. Video 4 shows that codepletion of RAB-5 and YOP-1/RET-1 leads to a synergistic defect in ER morphology. Video 5 shows that RAB-5 depletion blocks nuclear envelope disassembly. Online supplemental material is available at .

A key function of activated macrophages is to secrete cytokines, releasing these mediators in a tightly orchestrated manner to regulate the progression of an inflammatory response. The rapid production and secretion of TNFα by macrophages in response to microbial stimuli is central to induction of the early stages of inflammation, especially the recruitment of neutrophils. In addition to the protective role of TNFα during acute inflammatory responses, the overproduction or prolonged secretion of TNFα underlies the pathology of several chronic diseases, such as rheumatoid arthritis, graft-versus-host disease, and Crohn's disease (). For this reason, TNFα is the target of drugs and therapeutic antibodies, which are now increasingly used to treat these chronic inflammatory conditions (). Knowledge of the intracellular pathways responsible for TNFα trafficking and secretion could further aid drug development in this field. To this end, we recently defined the route for biosynthetic trafficking of membrane-bound TNFα and its release from activated macrophages (; ,; ). Newly synthesized TNFα rapidly accumulates in the Golgi complex upon the stimulation of macrophages with lipopolysaccharide (LPS) and is then transported to an intermediate compartment, the recycling endosome, before delivery to the plasma membrane for secretion. A second proinflammatory cytokine produced by activated macrophages is interleukin 6 (IL-6), which, lacking a transmembrane domain, is trafficked and secreted directly as a soluble protein. IL-6 is produced somewhat later in the inflammatory response and is important in the early resolution phase of innate responses and in the induction of acquired immunity (; ; ; for review see ). The inappropriate or chronic production of IL-6 has also been implicated in several inflammatory diseases, in which it is often targeted, alone or in conjunction with TNFα, by drugs and antibodies (). In activated macrophages, the expression and secretion profiles for TNFα and IL-6 generally overlap, although in a sequential fashion (; ). Nothing is currently known about the intracellular trafficking of IL-6. Therefore, the question begs as to whether soluble IL-6 is secreted via the same pathway as the membrane-bound TNFα and, furthermore, whether both cytokines rely on the same intracellular carriers and trafficking machinery. In recent years, the combination of high resolution microscopy and live cell imaging of fluorescently tagged proteins has largely redefined the post-Golgi trafficking and sorting of biosynthetic cargo in polarized and nonpolarized cells. Importantly, these studies have demonstrated that biosynthetic cargo exits the TGN in pleomorphic, often tubular carriers that fuse with intermediate compartments in transit to the cell surface (; ; ; ). All of the aforementioned studies implicate the recycling endosome as a way station for biosynthetic cargo and one wherein the cargo is likely sorted for transport to apical or basolateral membranes in polarized cells. The recycling endosome also, and more traditionally, handles cargo recycling from endosomal compartments to the cell surface, such as transferrin (Tfn) and Tfn receptor (TfnR; ). The recycling endosome in epithelial cells is increasingly recognized as having roles in trafficking through endocytic, recycling, and exocytic pathways (for reviews see ; ). The recycling endosome in macrophages also subserves both endocytic and exocytic processes, resulting in the delivery of biosynthetic TNFα to the cell surface at sites of phagocytic cup formation (), along with the recycling endosome membrane required for extension and expansion of the phagocytic cup around captured microbes (; ; ). Although this represents a particularly efficient pathway for the release of TNFα, whether other cytokines are secreted via recycling endosomes and, indeed, via phagocytic cups is not known. To address this, we directly compared the trafficking and secretion of endogenous and fluorescently tagged TNFα and IL-6 in activated macrophages. We show that IL-6 is loaded into tubulovesicular structures that bud off the TGN and fuse with recycling endosomes in transit to the plasma membrane. Surprisingly, in macrophages simultaneously producing TNFα and IL-6, it became clear that the trafficking of these two cytokines use overlapping but distinct trafficking pathways. Importantly, both TNFα and IL-6 are delivered to recycling endosomes but still maintain the ability to exit independently from this endosomal compartment. We have previously described the rapid production and secretion of TNFα from LPS-stimulated RAW264.7 macrophages (). Accumulation of TNFα precursors can be seen in the Golgi complex by 20 min, and secretion of TNFα can be detected as early as 40 min after LPS stimulation. A high level of secretion is maintained for 4–6 h before declining (). In response to LPS, low levels of the soluble cytokine IL-6 can also be detected in the supernatant of activated macrophages from 4 h (); however, its accumulation in the Golgi complex is not visible until 6 h after stimulation. Priming and activation of macrophages with IFNγ and LPS led to substantially higher levels of IL-6 secretion () and high levels of intracellular IL-6 by 6 h after stimulation. Thus, there is a considerable temporal overlap in the secretion of these two proinflammatory cytokines from activated macrophages. Consistent with this, dual-color immunofluorescence microscopy revealed the simultaneous localization of both IL-6 and TNFα in the Golgi complex in macrophages stimulated with IFN and LPS for 6 h (). This localization of IL-6 and TNFα to the Golgi stacks was also confirmed by immuno-EM (). A signal sequence in the translated IL-6 protein drives its delivery into the endoplasmic reticulum, but, to confirm its passage through the Golgi complex and the constitutive secretory pathway, macrophages were stimulated in the presence of brefeldin A, a drug known to disrupt the structure and secretory function of the Golgi complex. This treatment led to the substantial accumulation of intracellular IL-6 within activated macrophages () and completely blocked IL-6 secretion into the medium (not depicted). In fixed cells, endogenous IL-6 can be immunostained and clearly observed in the Golgi region (), but no other organelles or cytoplasmic vesicles containing this soluble cytokine were readily visible, probably because of the insufficient density of the cytokine during trafficking. Therefore, to track the secretory pathway of IL-6, we generated fluorescently tagged constructs containing the full-length sequence of IL-6 (GFP–IL-6 and mCherry–IL-6) and induced the expression of these constructs by transient transfection of macrophages. Within 4 h after transfection, high levels of fluorescently tagged IL-6 are visible within the Golgi complex (), where immuno-EM shows the soluble cargo accumulating in the dilated ends of Golgi cisternae (). IL-6–GFP was now also present in large vesicular structures throughout the cytoplasm. The fluorescently tagged IL-6 is actively secreted from macrophages even in the absence of exogenous stimulation, but higher levels are secreted from LPS-stimulated cells (). All IL-6 detected in this ELISA is presumably GFP tagged, as no endogenous IL-6 is secreted from mock-transfected cells, and, after LPS stimulation, only very low levels were detected at the 4-h time point (unpublished data). Thus, the expression of fluorescently tagged IL-6 is a suitable model with which to study IL-6 biosynthetic trafficking. Live imaging of IL-6–GFP-transfected cells was performed to visualize its exit from the TGN and subsequent post-Golgi movement. As highlighted in and Video 1 (available at ), vesicular structures were readily observed leaving the brightly stained Golgi pool of IL-6–GFP. In addition, large tubular structures containing IL-6–GFP were observed in some cells as they exited the Golgi region. These structures were often maintained for >30 s, and vesicular structures budding off the Golgi complex were also observed in these same cells ( and Video 2). These carriers are similar in form and kinetics to those described previously for TNFα transport (), leading us to compare the trafficking of these cargoes. Live imaging of cells coexpressing TNFα-mCherry and IL-6–GFP was performed. Both TNFα-mCherry and IL-6–GFP were visualized exiting the TGN in tubular and vesicular carriers. However, the budding of carriers containing IL-6–GFP was more frequent than for those with TNFα-mCherry, and, most of the time, IL-6–GFP appeared as the sole cargo in budding carriers (). In contrast, carriers containing TNFα that budded off the TGN almost always contained IL-6 (). A single video depicting both of these trafficking events is provided as Video 3 (available at ). A biochemical approach was also used to investigate the localization of TNFα and IL-6 in post-Golgi carriers. Golgi membranes were isolated from homogenates of macrophages stimulated with IFN/LPS for 4 h. Both pro-TNFα and IL-6 were readily detected in these Golgi fractions, which were then incubated in the presence of cytosol and GTPγs to induce budding of Golgi-derived membrane carriers. The resulting budded vesicles were then segregated into different populations by passage over a sucrose density gradient (; ), and eluted fractions were assayed by Western blotting. IL-6 and TNFα were both present in subsets of the Golgi-derived vesicles that overlapped with the distribution of another vesicle-associated protein, γ-adaptin. IL-6 and TNFα appeared in an overlapping distribution but notably peaked in different vesicle fractions, with IL-6 vesicle carriers in heavier fractions than TNFα-containing carriers (). This result is consistent with the live imaging in showing that these two cytokines exit the Golgi complex in carriers that contain cytokines as individual or mixed cargo. In live cells, vesicles budding off the TGN containing IL-6–GFP were observed fusing with large, preexisting stable compartments within the cytoplasm ( and Video 4, available at ). In fixed macrophages, IL-6–GFP was also visible in these large peripheral structures, where, in many cases, it colocalized with either TNFα-mCherry or endogenous TNFα (). Quantification of these structures containing IL-6–GFP demonstrated that ∼70% colocalize with endogenous TNFα or with TNFα-mCherry (), identifying this as a common destination for both newly synthesized cytokines. In light of our previous work demonstrating that TNFα is delivered from the TGN to recycling endosomes (), we hypothesized that these large structures containing both TNFα and IL-6, as shown in , may represent recycling endosomes. Immuno-EM also confirmed the colocalization of TNFα and IL-6 to endosomes and large vesicular structures (). Among these structures were some containing IL-6 and TNFα together and others with only one cytokine. Of interest, IL-6 appears to consistently localize to the luminal face of the membrane, which is suggestive of membrane association. To formally investigate the recycling endosome as a site for TNFα and IL-6 delivery, high resolution confocal microscopy was performed on cotransfected fixed cells. We have previously shown that vesicle-associated membrane protein 3 (VAMP3) is a reliable marker of recycling endosomes in macrophages (), with its expression leading to the localization of this SNARE protein to ringlike structures that represent recycling endosomes. Coexpressed IL-6–mCherry strongly localized to these VAMP3-positive ringlike recycling endosomes, as highlighted by the inset in . In other cases, the known recycling endosome cargo of internalized Tfn was used in conjunction with the overexpression of IL-6–GFP and TNFα-mCherry. clearly shows several structures that contain both GFP-TNFα and IL-6–mCherry (indicated by arrows) and also colocalize with internalized Tfn, which is consistent with their identity as recycling endosomes. The localization of IL-6 and TNFα in recycling endosomes was also confirmed by immuno-EM, in which the cytokines colocalized with TfnR () in both peri-Golgi and more peripheral recycling endosomes. Quantification showed that 51% of labeled structures contained IL-6 alone, whereas 29% of TfnR-labeled endosomes were also colabeled for IL-6; this included structures in which the TfnR and IL-6 gold particles were at opposite ends of tubules. Similarly, 72% of labeled endosomes had TNFα alone, whereas 14% were colabeled with TfnR. Thus, a substantial proportion of recycling endosomes, or parts thereof, contain each cytokine as a single cargo. Based on a fluorescence intensity line scan through several recycling endosomes (), we noted that Tfn did not always peak precisely with TNFα and IL-6, but, instead, all three proteins appeared in very close proximity. In the structures containing IL-6, the peak fluorescence overlaps tightly with the peak of TNFα (, structures 2 and 3), but Tfn fluorescence is more dispersed in these structures compared with those containing TNFα and Tfn alone (, structures 1 and 4). This raises two questions: are all three of these cargoes truly localizing to the same structure, and, if so, are they segregated within that structure? To address these questions, two different approaches were undertaken. First, 3D reconstructions of recycling endosomes containing IL-6, TNFα, and Tfn were performed to access the distribution of cargo within these structures. Specified recycling endosomes were excised and subjected to surface rendering, allowing their visualization in 3D (). In structure A, IL-6 overlaps with TNFα, and Tfn appears to occupy a separate domain within this structure. In contrast, structure B contains substantial amounts of all three cargoes, but each occupies a separate domain, with very little direct overlap. These images clearly suggest the existence of the dynamic compartmentalization of cargoes within the recycling endosome. A second approach was used to confirm that cargoes are in a single recycling endosome. Cells expressing IL-6–GFP were sequentially incubated in the presence of the DilC18-DS lipophilic dye followed by a chase period devoid of dye in which the cells were incubated with Texas red–conjugated Tfn. The lipophilic dye initially labels the plasma membrane, but, under this protocol, it is internalized into all surface-derived endosomes. In , a structure is highlighted where IL-6 and Tfn are found side by side but are linked within the same endosomal structure by the overlap of both cargoes with lipophilic dye. Thus, dye labeling helps to demonstrate that IL-6 and Tfn can exist in the same recycling endosome but occupy separate regions of the compartment. Collectively, our results show that IL-6 and TNFα are delivered to the same recycling endosome but can be segregated as cargo within the recycling endosome, with each also being able to be segregated from the internalized cargo Tfn. Thus, the recycling endosome has the potential to compartmentalize cargo. We have previously described a Q-SNARE complex consisting of Vti1b, syntaxin 7 (Stx7), and Stx6 on post-Golgi carriers that mediates the trafficking of TNFα from the TGN to the recycling endosome in activated macrophages (,). In the present study, we sought to examine whether some of these SNARE proteins also function in IL-6 trafficking. Transient overexpression of the TGN-localized Q-SNARE proteins Stx6 and Vti1b led to significant increases (P < 0.05) in the level of IL-6 secretion compared with controls or with cells transfected with an irrelevant SNARE, Stx2, or an unrelated TGN protein, mannose-6-phosphate receptor (M6PR; ). In addition, siRNA knockdown of Stx6 and particularly of Vti1b significantly decreased (P < 0.05) the level of IL-6 secretion in cultures in which 50–70% of cells demonstrated knockdown of the relevant gene (). Thus, it is likely that the same Q-SNARE complex, perhaps in addition to others, regulates the post-Golgi traffic of IL-6. VAMP3 on the recycling endosome is the complementary R-SNARE for this transport step, and its overexpression or siRNA knockdown both affect TNFα trafficking (). Transient overexpression of VAMP3 here resulted in a significantly increased (P < 0.05) secretion of IL-6 (). Similarly, siRNA knockdown of VAMP3 decreased IL-6 secretion (). This is functional evidence that endogenous IL-6 is trafficked via the recycling endosome for secretion and that these SNAREs can regulate both TNFα and IL-6 trafficking. To test whether the recycling endosome is an essential intermediary for IL-6 secretion, an endosomal inactivation assay was performed as previously described (; ; ). Inactivation of recycling endosome function by sequestration and activation of HRP-Tfn blocks the secretion of TNFα, (), and, in this study, we show that IL-6 secretion is similarly and completely blocked under these same conditions (). This block occurs at a post-Golgi step at recycling endosomes because Golgi staining of IL-6 and TNFα is still visible in endosome-inactivated cells. Quantification shows a big increase in cells showing Golgi-accumulated TNFα and a small decrease in cells with Golgi-accumulated IL-6, perhaps revealing a difference in the accumulation of membrane-bound and soluble cargo under conditions in which subsequent trafficking and secretion are blocked (). Alternatively, this decrease in the number of cells producing IL-6 may result from the blockage of TNF secretion, which is known to have an autocrine effect on macrophages and enhance subsequent cytokine secretion. These results confirm that the recycling endosome has a pivotal functional role as an intermediate compartment during the secretion of cytokines from activated macrophages. We have recently shown that TNFα in the recycling endosome is trafficked to the phagocytic cup along with membranes recruited during phagocytosis (). Therefore, a central question becomes whether IL-6 is also delivered to the cell surface via the phagocytic cups. As soluble cargo, IL-6 is released immediately from cells, and it is not normally stained at the cell surface. Therefore, the movement of IL-6–GFP in the vicinity of phagocytic cups was examined in live cells. When videos of cells coexpressing VAMP3-GFP and mCherry–IL-6 were analyzed by particle tracking software, compilations of the tracks taken by these two proteins over time reveal different targeting (). Although VAMP3-GFP tracks show concerted movement toward and directly around a phagocytic cup, there are no mCherry–IL-6 tracks leading toward or even in the vicinity of the phagocytic cup in this and other examples. In no experiments using live or fixed cells did we detect IL-6 around or near phagocytic cups, suggesting that it and TNFα are segregated and are targeted distinctly upon exit from the recycling endosome. TNFα is delivered to the phagocytic cups, but IL-6 is not. The selective targeting of recycling endosome cargo to the phagocytic cup was also examined in the case of TfnR. Macrophages phagocytosing yeast could be seen with VAMP3-GFP and TNF-GFP at the actin-rich phagocytic cups, but, in contrast, TfnR was not concentrated at the cups (). This confirms that there is selective delivery of cargo such as TNFα to the phagocytic cup, but other cargo, such as IL-6 and TfnR, are excluded from movement to this site. This is a functional demonstration of cargo within the recycling endosomes being compartmentalized. We also demonstrate that different cargo can exit the recycling endosome separately. In live cells coexpressing IL-6–GFP and TNF-mCherry, we see both cytokines together in recycling endosomes, and (also see Video 5, available at ) shows the exit of IL-6–GFP from this structure in a carrier but not of TNF-mCherry. In addition, in a macrophage undergoing phagocytosis of an IgG-coated bead, VAMP3 can be seen exiting a recycling endosome independently of IL-6, which is retained while VAMP3 moves in a carrier toward a phagocytic cup ( and Video 6). Thus, cargo delivered to the recycling endosomes is compartmentalized, presumably undergoing sorting, and can then exit in specific carriers that are targeted to different destinations. The recycling endosome in the macrophage has final and selective control over the fate and destination of its proinflammatory cytokines. By examining endogenous IL-6 and fluorescently tagged IL-6 in live and fixed macrophages, we describe here, for the first time, the secretory pathway for the soluble cytokine IL-6. Fluorescent IL-6 was observed exiting the Golgi complex in tubulovesicular carriers, where it appeared as labeled cargo alone or in conjunction with TNFα. Overall, our results are in agreement with the limited observations of intracellular IL-6 in the literature, including an early study showing the costaining of TNFα and IL-6 in the Golgi complex of activated monocytes () and electron microscopic labeling showing that, as a constitutive secretory product in mast cells, IL-6 was excluded from entry into secretory granules and was instead found clustered in small constitutive vesicles after leaving the Golgi complex (). Now, a major revelation in this study is that IL-6, upon leaving the Golgi complex, is trafficked to the recycling endosome before it is delivered to the cell surface. Moreover, we show that the recycling endosome represents a critical point of divergence for the cytokines IL-6 and TNFα, with TNFα but not IL-6 delivered to phagocytic cups. Compartmentalization of cargo within the recycling endosome underpins the individual exit and release of these cytokines, revealing new capacities and an important role for this organelle in orchestrating the macrophage immune response. Fluorescent IL-6 is loaded into tubulovesicular structures budding from the TGN in live macrophages. The size, appearance, and kinetics of these carriers are consistent with carriers seen previously in macrophages labeled with TNFα as cargo, and they are also similar to post-Golgi carriers visualized in other cell types (, ; ; ). By immunofluorescence, EM, and biochemical analyses, we show that IL-6 is the lone labeled cargo in some carriers, whereas in other cases, it is in carriers that also contain TNFα. Although TGN-derived carriers containing fluorescent IL-6 were abundant (for example, see Video 3), those containing TNFα alone or IL-6 were seen less frequently. Thus, there is little apparent sorting of soluble cargo like IL-6 at the TGN, whereas membrane-bound TNFα must be actively sorted for more selective loading into carriers. This is also borne out by previous data from HeLa cells, in which we showed that TNFα is selectively loaded into only a subset of golgin-labeled carriers budding off the TGN (). In general, there is little understanding of how soluble cargo in constitutive secretory pathways is handled at the TGN. Other soluble cargo such as the lysosomal enzyme is sorted in the TGN by binding to M6PR for trafficking to endosomes (), and, in cells with regulated secretion, secretory products are clustered by binding to chromogranins in the TGN for loading into secretory granules (; ). An earlier study suggests that in epithelial cells, soluble proteins are sorted in a pH-dependent compartment for polarized delivery to the cell surface (). Our current findings now highlight the fact that the TGN may not be the only place for the sorting of biosynthetic cargo such as cytokines. The recycling endosome, which is appropriately slightly acidic (; ), must now also be considered as a possible sorting site for membrane-bound () and soluble cargoes. Several lines of evidence show that the post-Golgi carriers containing IL-6 are indeed delivered to recycling endosomes. First, overexpression or knockdown of the SNARE proteins Vti1b, Stx6, and VAMP3 affected the secretion of IL-6. We have previously shown that these proteins are part of the Q-SNARE complex (Stx6–Stx7–Vti1b) on the TGN with the cognate R-SNARE VAMP3 on recycling endosome membranes and that this complex regulates TNFα trafficking in macrophages (). Implicating these SNAREs also in IL-6 trafficking is consistent with both cytokines sharing some carriers and/or being trafficked in separate carriers regulated by the same SNAREs, although it is possible that additional SNARE complexes could be involved. VAMP3 function in IL-6 secretion confirms the recycling endosome as a post-Golgi destination for this cytokine. Second, by live cell imaging, fluorescently tagged IL-6 in tubular and vesicular carriers budding off the TGN was frequently observed fusing with preexisting recycling endosomes in the cell periphery. In these same structures, IL-6 was colocalized with recycling endosome markers, including VAMP3 and endocytosed Tfn. Finally, inactivation of the recycling endosome completely ablated the secretion of IL-6, showing that this is a requisite compartment en route to the cell surface for newly synthesized IL-6. The direct delivery of all or most of IL-6 to the recycling endosome is in agreement with the transport of other biosynthetic cargo from the TGN to this compartment, including TNFα in macrophages and endothelial cadherin and vesicular stomatitis virus-G in epithelial cells (; ). IL-6 is the first example of a constitutively transported soluble cargo trafficking to the cell surface via the recycling endosome, and, as such, it underscores the diverse roles played by the recycling endosome in exocytosis. The recycling endosome handles cargo from both endocytic/recycling routes as well as exocytic cargo (for review see ). In the present study, we present evidence that IL-6 and TNFα converge with recycling Tfn and with the resident SNARE VAMP3 within recycling endosomes. Importantly, however, all three of these cargo proteins appear to be partially segregated within this compartment based on colocalization and image reconstruction in fixed and live cells. This compartmentalization differentiates not only recycling and exocytic cargo but even segregates different exocytic cargo within the same structure. Live imaging and image reconstructions demonstrate that such a compartmentalization of cargo is an extremely dynamic process, with cargo-rich domains that continuously merge and then segregate (Video 3). Mechanisms for sorting and compartmentalizing the membrane-bound cargoes such as TNFα and Tfn/TfnR can be envisioned, but very little is known about how soluble proteins are or could be sequestered (; for review see ). It is possible that surface receptors for IL-6, such as IL6R or gp130 (), could be involved, but this awaits further investigation. In addition, it is notable that IL-GFP switched from accumulating in the lumens of Golgi cisternae to being on the luminal face of recycling endosome membranes in our immuno-EM images. This is suggestive of membrane association that could provide a mechanism for IL-6 in sorting at the level of the recycling endosome. The concept of having compartmentalized recycling endosomes is already soundly established within the literature, particularly with reference to the membrane-associated trafficking machinery. Light, fluorescence, and EM studies show this to be a highly reticulated and tubular compartment (; ; ; ). It has a complex and segmented molecular landscape with multiple resident GTPases such as Arf6 and the Rab11 subfamily proteins (; ; ). Members of the Rab11 family demark different compartments of the recycling endosome in MDCK cells with the locations of Rab11a and Rab25 distinct from that of Rab11b (). Similarly, the large Rab11-FIP family of Rab11a effectors offers many opportunities for further spatial and functional complexity, as does another Rab11 effector, myosinVb, which has also been shown to differentially regulate different cargo trafficking through the recycling endosome (; ; ). Resident recycling endosome markers such as Rab11 and the SNARE VAMP3 only partially colocalize with Tfn as cargo (; ). A targeted fluorescence method used by showed only a 20–30% overlap between Tfn and VAMP3 on pericentriolar recycling endosomes. There is also a dynamic overlap or continuity of recycling endosomes with early endosomes, as demonstrated by the sequential but overlapping distributions of Rabs 11, 4, and 5, portending the possibility of recycling endosomes being part of a continuous tubular network (; ; ; ; ). Thus, the complex handling and segregation of cargoes within the recycling endosome that we show here now suggests a functional consequence for this array of machinery found associated with these membranes. TNFα and IL-6 are both proinflammatory cytokines that are secreted temporally in overlapping profiles from macrophages, but, because both cytokines have different immune functions, it is important for their secretion to be tightly coordinated throughout the inflammatory response. In this study, we show for the first time that the secretion of these cytokines can be differentially regulated at the level of intracellular trafficking in addition to their known regulation at the transcriptional and translational levels (). Similarly, recently identified the existence of two separate exocytic pathways for the targeted or polarized trafficking of cytokines to the immunological synapse or to the entire plasma membrane in activated T cells. This bears analogy to the targeted delivery of TNFα directly to the phagocytic cup and the more generalized secretion of IL-6, which is shown here in macrophages. It is not known whether the recycling endosome is involved in the differential trafficking of cytokines in T cells, as we show here for macrophages. Our results may also be comparable with the differential release of IL-4 and -12 from eosinophil crystalloid granules (; ). In these granules, IL-4 is selectively mobilized to move from the granules into secretory vesicles by IL-4Rα after the stimulation of eosinophils (). Such examples highlight the importance for immune cells to independently regulate and release cytokines. Indeed, by analogy, we propose that this is a function assigned to recycling endosomes in macrophages, which, unlike granulocytes, do not have granules for packaging and selective release of cytokines. Unlike TNFα, the secretion of IL-6 was not linked to the formation of phagocytic cups. This possibly reflects the temporal nature of cytokine secretion, with an essential role for TNFα in very early inflammation, whereas IL-6 is important during the transition from innate to adaptive immunity (for review see ). Our studies to date do not help define the location or nature of the cell surface sites to which IL-6 is delivered for release. Tracking fluorescently tagged IL-6 during its delivery to the plasma membrane has not yet revealed any particular feature or special sites for plasma membrane fusion. There may well be organization of IL-6 secretion sites at the level of membrane lipid microdomains, as there is for TNFα (), or exocyst-demarked delivery sites, as seen on the apico-lateral membranes of epithelial cells (; ). Interestingly, we also found that TfnR in recycling endosomes is not delivered to the actin-rich phagocytic cups. The cognate delivery of VAMP3 and TNFα to the cups and the concomitant exclusion of IL-6 and TfnR is evidence once again that sorting occurs in the recycling endosomes. Importantly, it shows that the recruitment of recycling endosome membrane for the formation of phagocytic cups at the cell surface occurs in a highly selective fashion, perhaps being designed to ensure that other trafficking and other cell functions can proceed unfettered by the onset of phagocytosis. It also highlights the temporal nature of phagosome maturation, a process that requires the sequential recruitment of different organelles in a SNARE-mediated fashion (; ; ). Although TfnR is excluded from early stage phagocytic cups, it can appear in more mature phagosomes (), suggesting that the recycling endosome plays a dynamic and evolving role throughout the phagocytic process. In conclusion, the results shown in this study demonstrate the differential trafficking and secretion of cytokines from activated macrophages. Aspects of the post-Golgi trafficking that are shared or not shared by IL-6 and TNFα may suggest strategies for the joint or separate therapeutic targeting of these cytokines in inflammatory disease. A new route is demonstrated for IL-6 trafficking, one including recycling endosomes as a way station and possible sorting site. Our studies suggest that mechanistically, sorting and compartmentalization of cargo within the macrophage recycling endosome enables the differential secretion of these proinflammatory cytokines and helps orchestrate the immune response. The following primary antibodies were used: rabbit polyclonal antibody to mouse IL-6 (Serotec), rabbit polyclonal and rat monoclonal anti–mouse TNFα antibodies (Calbiochem and Auspep, respectively), monoclonal antibodies to Stx6 and Vti1b (Translabs; BD Biosciences), monoclonal antibody to TfnR (Zymed Laboratories and Invitrogen), and a polyclonal rabbit antibody to VAMP3 (Abcam). Secondary antibodies used for immunofluorescence microscopy include Cy3-conjugated sheep anti–mouse IgG, Cy3-conjugated goat anti–rat IgG, and Cy3-conjugated goat anti–rabbit IgG, which were all purchased from Jackson ImmunoResearch Laboratories, as well as AlexaFluor488-conjugated goat anti–rat IgG and AlexaFluor647-conjugated goat anti–mouse IgG, which were both purchased from Invitrogen. Tetramethylrhodamine- and AlexaFluor647-conjugated Tfn and the DilC-DS lipophilic dye were purchased from Invitrogen. LPS from serotype 0111:B4 was purchased from Sigma-Aldrich. IFNγ was purchased from R&D Systems. RAW264.7 murine macrophages were cultured in RPMI 1640 medium (BioWhittaker) containing 10% heat-inactivated Serum Supreme (BioWhittaker) and 1% -glutamine (Invitrogen) as previously described (). Macrophages were activated by the addition of 100–1,000 ng/ml LPS or 500 pg/ml IFNγ/LPS. Full-length constructs of M6PR, TNFα, and the SNAREs Stx2, Stx6, VAMP3, and Vti1b were cloned into the pEGF-C1 vector (CLONTECH Laboratories, Inc., and BD Biosciences) to produce an N-terminal GFP-tagged protein as previously described (). TNFα and VAMP3 were also cloned into the pCherry-C1 vector (provided by R.Y. Tsien, University of California, San Diego, La Jolla, CA). IL-6 was cloned into the pGFP-N1 vector and the pCherry-N1 vector (adapted from the pCherry-C1 vector) to produce proteins that were C terminal tagged with either GFP or mCherry, respectively. Macrophages were transfected for the transient expression of cDNAs by either electroporation or LipofectAMINE 2000 (Invitrogen). For electroporation, using an electroporation system (Gene Pulser II; Bio-Rad Laboratories), 2.5 × 10 cells were mixed with 10 μg DNA with a high capacitance setting (280 mV and 950 μF). Cells were then washed and typically cultured for 2–24 h before stimulation and/or imaging. For transfection with LipofectAMINE 2000, 15 μl of the reagent was mixed with 4 μg DNA in Opti-MEM (Invitrogen) per P6 plate containing 2 × 10 adherent cells. After 2 h, the cells were washed and further cultured for at least 1 h in complete medium before stimulation/imaging. To label the recycling endosome, RAW264.7 cells were incubated with 10 μg/ml AlexaFluor647-conjugated Tfn in media for 15–60 min at 37°C before fixation in 4% PFA. Alternatively, cells were washed twice in serum-free media before the addition of 1 μM of the DilC-DS lipophilic dye, diluted in Opti-MEM (Invitrogen), and incubated for 5 min at 37°C followed by 15 min on ice. Unbound dye was removed by washing the cells three times in complete media before the addition of 10 μg/ml tetramethylrhodamine-conjugated Tfn in media and incubation for 20 min at 37°C. Finally, cells were washed before fixation in 4% PFA. The trafficking of TNFα and IL-6 from the Golgi complex to the cell surface was measured using an immunofluorescence-based assay. In brief, macrophages were incubated in the presence of 100 ng/ml LPS or 500 pg/ml IFN/100 ng/ml LPS for 4–6 h, fixed in 4% PFA, permeabilized with 0.1% Triton X-100, and stained by immunofluorescence for cell surface and intracellular TNFα (as previously described; ) and IL-6. In some experiments, 5 μg/ml brefeldin A was added to the macrophages for the final 2 h of incubation. To determine the levels of secreted TNFα and IL-6, commercial ELISA kits (BD OptEIA; BD Biosciences) were used according to the manufacturer's instructions. In addition, to determine the levels of total IL-6 and TNFα within activated macrophages, cell lysates were prepared at the end of each time point and analyzed by Western blotting as described in Preparation of cell lysates, SDS-PAGE, and Western blotting. In some experiments, macrophages were primed for 18 h in the presence of 500 pg/ml IFNγ before their incubation with live (at a ratio of 10 yeast per macrophage). Plates were briefly spun at 50 for 2 min to encourage yeast–macrophage interactions, and a chase period of 5–20 min was performed at 37°C to allow phagocytosis to proceed. For live cell imaging, IgG-opsonized beads were added directly to the media immersing the imaging macrophages, and the videos were collected for up to a further 30 min. RAW264.7 cells were washed three times with PBS and lysed in buffer A (10 mM Tris, pH 7.4, containing 1 mM EDTA, 0.5% Triton X-100, and Complete protease inhibitors [Roche]). Cells were harvested by scraping followed by disruption via passaging through a series of successively smaller needles and centrifuged for 10 min at 17,000 . The supernatant was assayed for protein content (Bio-Rad Laboratories protein assay), and 20–50 μg of total protein from each sample was subjected to SDS-PAGE separation and analyzed by immunoblotting. For live cell experiments, cells were cultured on 25-mm round glass coverslips or in glass-bottom 35-mm dishes (MatTek Corporation). Live cell epifluorescence imaging was performed using inverted microscopes (IX71 or IX81; Olympus). Coverslips were mounted in a temperature-controlled heating block, warmed to 37°C, and maintained to within ±0.2°C using a heated water bath or with cells in MatTek dishes maintained at 37°C by use of a microscope incubator (Solvent Scientific). In both cases, the cells were incubated in CO-independent medium containing 10% heat-inactivated FCS (). Fluorophore excitation was achieved with a Xenon lamp-based monochromatic light generator (Polychrome IV; TILL Photonics) controlled by a digital signal processor board (TILL Photonics) on the IX71 microscope and an illumination system (MT20; Olympus) on the IX81 microscope. Images were captured through either a UPlanSApo 60×/NA 1.35 or 100×/NA 1.40 oil objective (Olympus). A 12-bit 1,280 × 1,024-pixel camera (IMAGO Super VGA; TILL Photonics) was used to capture the images on the IX71 microscope, and a 12-bit 1,376 × 1,032-pixel high resolution camera (F-VIEW; Olympus) was used on the IX81 inverted microscope. For single-color videos, frame capture rates were between 200 and 2,000 ms, with total capture periods ranging from 2 to 45 min. For dual-color videos, individual channels were captured sequentially, with typical capture rates of 100–1,000 ms for GFP and capture rates of 80–400 ms for mCherry. Importantly, the GFP signal was captured first, followed by mCherry. Image control and postcapture image analysis were performed using either TILLVISION (TILL Photonics) or CelR v2.5 software (Olympus). Videos were analyzed, cropped, and constructed using ImageJ v1.37p (National Institutes of Health), Volocity v3.7 (Improvision), Photoshop CS2 (Adobe), and TILLVISION software and were exported as QuickTime videos (Apple) with a playback speed of 10 frames per second. An additional ImageJ plugin, ParticleTracker, was used for automated detection and tracking of particles recorded by video imaging (). 200 frames were analyzed for each channel, the particle radius was set at two pixels, and the maximum pixel displacement allowed between frames for particles was two. The resulting detected particles and their tracks were filtered so only those that were continuous for five or more frames were displayed on an overlay of the channel. Each detected track is shown at its full length as a continuous color, with red sections being interpolated by the plugin to handle occlusions (). A stacked Golgi membrane fraction and subsequent in vitro budded Golgi-derived vesicles were prepared from RAW264.7 cells stimulated with 100 ng/ml LPS for 6 h by density gradient centrifugation based on previously published methods (; ). The resulting fractions were analyzed by SDS-PAGE and immunoblotting. RAW264.7 cells plated on glass coverslips were subjected to two rounds of transfection of siRNA constructs using LipofectAMINE 2000 (Invitrogen) according to the manufacturer's instructions, each 24 h apart, and were finally cultured for a further 24 h before LPS stimulation. The following siRNAs were used in these experiments: scramble (siRNA ID# 4635), glyceraldehyde-3-phosphate dehydrogenase (ID# 4631), mouse VAMP3 (ID# 186990), Vti1b (ID# 184555), and Stx6 (ID#7 5897). All constructs were purchased from Ambion. Western blotting and immunofluorescence were used to confirm the knockdown of the target proteins. An HRP inactivation assay was modified from the protocol of and as previously described (; ). In brief, RAW264.7 cells were incubated with 10 μg/ml Tfn-HRP in media for 30 min in the dark at 37°C. Cells were washed twice in ice-cold PBS, and surface-bound Tfn-HRP was removed by two 5-min washes with 0.15 M NaCl and 20 mM citric acid, pH 5. Cells were then washed twice with ice-cold PBS and incubated in the dark for 1 h with PBS containing 0.1 mg/ml DAB and 0.025% HO to the inactivation sample (the control contained DAB but no HO). Cells were finally washed twice in PBS containing 1% BSA to stop the reaction and were incubated in prewarmed media containing 500 pg/ml IFN and 100 ng/ml LPS for 6 h at 37°C. Supernatant was collected after 2 h and 6 h to determine the level of TNFα and IL-6 secretion, respectively. Coverslips were fixed in 4% PFA and immunostained for internal TNFα and IL-6. Immuno-EM of ultrathin cryosections was performed as previously described (). In brief, after fixation in 4% PFA (EM grade; ProSciTech), RAW264.7 cells were embedded in warm gelatin and frozen onto cryostubs. Ultra-thin cryosections were collected onto copper grids and immunolabeled according to . Antibodies were detected with either different-sized protein A–gold conjugates (provided by J. Slot, University of Utrecht, Utrecht, Netherlands) or species-specific gold probes (British BioCell, Australian Laboratory Services). Sections were viewed on an electron microscope (model 1011; JEOL), whereas images were captured using the iTEM analysis program (Soft Imaging System). Video 1 shows IL-6 exiting the Golgi in a budded vesicle. Video 2 shows IL-6 exiting the Golgi in vesicular and budded vesicles. Video 3 shows that IL-6 exits the TGN in carriers both with and without TNF. Video 4 shows IL-6 exiting the Golgi in a budded vesicle and fusing with the recycling endosome. Video 5 shows that IL-6 exits the recycling endosome without TNF. Video 6 shows that IL-6 does not traffic to the forming phagocytic cup. Online supplemental material is available at .

In cells mitochondria fuse and divide continuously with balanced rates, resulting in the formation of highly dynamic reticular networks (). Dynamic rearrangements of the mitochondrial network contribute to diverse cellular processes (; ; ). For example, the proper regulation of mitochondrial fission and fusion rates appears to be vital for Ca buffering () and maintenance of synaptic plasticity (). Mitochondrial fragmentation and fusion inhibition are also closely associated with outer mitochondrial membrane (OMM) permeabilization and subsequent release of certain proteins from the mitochondria to the cytosol, early during apoptosis (; ; ). The contribution of a balanced mitochondrial network for cellular fitness was further illuminated by reports showing that mutations of fusion proteins Opa1 and Mfn2 are associated with the neurodegenerative diseases dominant optic atrophy and Charcot-Marie-Tooth disease, respectively (for review see ; ). Mitofusins (Fzo1p in yeast), Opa1 (Mgm1p in yeast), and Drp1 (Dnm1p in yeast) are large dynamin-related GTPases essential for mitochondrial fusion and fission (; ; ; ; ; ; ; ). In addition to Dnm1p, mitochondrial fission in yeast depends on at least three additional factors. Fis1p is an approximate 18-kD, C-tail–anchored integral protein of the OMM () with a tetratricopeptide repeat domain serving as a receptor for Mdv1p (; ; ; ) or its structural homologue Caf4p (). On mitochondria Dnm1p (as well as Drp1) localizes to relatively large punctiform structures, particularly abundant at prospective scission sites as well as at mitochondrial tips (; ; ). As detailed recently (; ), mitochondrial Dnm1p complexes are dynamic structures formed and maintained as a result of transient interactions of different components of the fission machinery. The N-terminal extension and the C-terminal WD-40 repeat of Mdv1p or Caf4p are thought to act as adaptors linking Fis1p and Dnm1p into the mitochondrial division process (; ). In contrast to yeast, mitochondrial division in other organisms, including mammalian cells, is less well understood. As in budding yeast, mammalian cells require Drp1 () and Fis1 (; ) for mitochondrial fission, suggesting that mitochondrial division mechanisms are to some extent conserved. However, the mechanisms of mitochondrial translocation of Drp1 as well as scission foci assembly in mammalian cells remain elusive. The conjugation of the 76-amino acid protein ubiquitin to substrate proteins is involved in the regulation of a variety of cellular processes (; ) ranging from selective protein degradation to DNA repair () and membrane protein trafficking (). Recent findings indicate that ubiquitylation plays a direct role in mitochondrial membrane remodeling (; ; ; ). For example, it has been shown that treatment with the mating pheromone α-factor induces a 26S proteasome-dependent degradation of the mitofusin Fzo1p, resulting in the fragmentation of mitochondria (; ). In this study we demonstrate that the mitochondrial E3 ubiquitin ligase MARCH5 (MarchV) (; ), also known as MITOL (), participates in the regulation of mitochondrial division. RING finger–containing proteins comprise a large family of potential ubiquitin ligases involved in regulation of 26S proteasome-dependent degradation of misfolded and superfluous proteins () and in signal transduction in various cellular locations (). Using bioinformatics combined with molecular biology techniques, we cloned several uncharacterized RING finger proteins associated with mitochondria. One of the mitochondrial proteins identified in our screen was MARCH5 (RNF153/MarchV), which has been previously described to associate with the ER (), and more recently shown to localize to mitochondria (; ). MARCH5 consists of four predicted transmembrane domains and an N-terminal RINGv-type RING finger domain critical for ubiquitin transfer activity of several E3 ubiquitin ligases (). MARCH5 was recently shown to mediate ubiquitylation of some mitochondrial proteins, including mitochondrial morphogenesis regulators Fis1 and Drp1 (; ). However, the role of MARCH5 in the regulation of mitochondrial network dynamics remains to be clarified. When fused with YFP and expressed in HeLa cells, both YFP-MARCH5 (unpublished data) and MARCH5-YFP localized to mitochondria (), consistent with two recent studies that showed mitochondrial localization of endogenous MARCH5 (; ). We failed to detect any association of MARCH5 with the ER (unpublished data). To analyze the function of MARCH5 we mutated a conserved Zn-binding histidine residue inside the RING domain to tryptophan (MARCH5) (). Based on the available RING finger protein structure (; ), the substitution of His-43 with tryptophan is predicted to inhibit Zn coordination and therefore the overall function of the RING domain. Mitochondrial morphology in cells transiently transfected with MARCH5, or MARCH5 expression vectors was accessed by confocal microscopy with immunostaining of cytochrome to visualize mitochondria (). As described previously (), overexpression of wild-type MARCH5 did not affect the overall organization of the mitochondrial network in various cell types as determined by cytochrome (; unpublished data) and Tom20 staining patterns (unpublished data). This indicates that MARCH5 is well tolerated even at high expression levels and appears not to be a rate-limiting factor in mitochondrial morphogenesis. However, contrary to previous observations reporting fragmentation of the mitochondrial network as a consequence of MARCH5 inhibition (; ), the majority of cells (52 ± 8.7%) expressing MARCH5 displayed highly interconnected mitochondria (). The aberrant mitochondria range in morphology from thin and elongated, often forming tubule nets (), to highly interconnected thickened tubular and blebbed structures (). In cells expressing higher amounts of MARCH5 a further increase in the diameter and rounding of mitochondria was associated with the perinuclear accumulation of these organelles (30.5 ± 10%; ; Fig. S1 A, available at ). In 17.5 ± 10.4% of MARCH5-expressing cells, a control-like organization of the mitochondrial network was observed (). We also constructed another MARCH5 RING mutant by replacing cysteines 65 and 68 with serines (MARCH5), as this mutation has been already shown to inhibit MARCH5 activity (). Consistent with the effect of MARCH5, expression of MARCH5 led to the formation of interconnected mitochondria in ∼63% of cells (). The remaining ∼31% of cells displayed a perinuclear accumulation of mitochondria ( and Fig. S1 B), or normal mitochondrial networks (6 ± 2%; ). The range of mitochondrial phenotypes also resembled those induced by MARCH5-YFP. The expression of a MARCH5 construct missing the entire RING domain (MARCH5) also led to the formation of abnormally interconnected mitochondria, although in a smaller percentage of cells (29.7 ± 5.9% cells with abnormally elongated mitochondria and 16.6 ± 8.3% with perinuclear clustering of mitochondria; ; Fig. S1, C and D). These data indicate that inactivation of the MARCH5 RING domain induces an abnormal interconnection of the mitochondrial network. To further test the role of MARCH5 in the regulation of mitochondrial dynamics, we applied short hairpin RNA interference (shRNAi)–mediated down-regulation of MARCH5 protein expression. HeLa cells were transfected with control RNAi vector targeting GFP (control RNAi cells; ), or a construct targeting MARCH5 (MARCH5 RNAi cells). As shown by Western blot, the MARCH5 RNAi led to a substantial down-regulation of MARCH5 expression compared with control RNAi cells (). We also tested the expression levels of Mfn2, Drp1, and Fis1, mitochondrial dynamics proteins reported to interact with MARCH5 (; ). Whole cell lysates from control RNAi and MARCH5 RNAi cells analyzed by Western blot show that a reduction of MARCH5 expression was not associated with apparent changes in the protein levels of Mfn2, Drp1, and Fis1 (). To analyze mitochondrial morphology, control RNAi () and MARCH5 RNAi () cells were immunostained for Tom20. In ∼40% of MARCH5 RNAi cells, mitochondria formed elongated and interconnected tubules (). In contrast to the previous reports, we did not detect a significant fragmentation of mitochondria in MARCH5 RNAi cells (5.67 ± 2.31% of MARCH5 RNAi cells, compared with 6.67 ± 4.72% of control RNAi cells; ). To further characterize the effect of MARCH5 activity, we analyzed mitochondrial network unit size and contiguity using a mitochondrial matrix–targeted photoactivable variant of GFP (mito-PAGFP) () in cells expressing different MARCH5 constructs. To this end in double-transfected cells, 4-μm wide regions of interests (ROIs) (; black squares) were briefly irradiated with a 413-nm laser followed by imaging with a 488-nm laser immediately after 413-nm photoactivation (; middle panels). This results in a photoactivation of mito-PAGFP within the ROIs. The 488-nm confocal imaging, performed immediately after ROI photoactivation, reveals the area of mitochondrial network units located within the ROIs and those with matrix compartments contiguous with the ROIs (). We found that, under the conditions applied here, in control (unpublished data) and wild-type MARCH5-expressing cells () the activated mitochondria were usually more restricted to the photoactivated ROIs, whereas in MARCH5 () and MARCH5 (unpublished data) expressing cells the mito-PAGFP signal redistributes further from the ROIs. This indicates an increased size of mitochondrial network units upon inhibition of MARCH5 activity. To quantify the relative area covered by ROI-activated mitochondria we also photoactivated the whole imaging fields. This results in depiction of all the mitochondria within the same cells (; right panels). The quantification of the ratio of area covered by ROI-activated mitochondria (r) to that covered by the whole cell mitochondria (w) revealed a significant increase in mitochondrial unit size in cells expressing MARCH5 (r/w = 0.595 ± 0.179) and MARCH5 (r/w = 0.626 ± 0.199), compared with control (r/w = 0.243 ± 0.086) and wild-type MARCH5 (r/w = 0.240 ± 0.074) expressing cells (). To analyze mitochondrial network complexity, we also applied FRAP, as described previously (; ), in cells expressing MARCH5-CFP, MARCH5-CFP, or MARCH5-CFP together with a mitochondrial matrix–targeted yellow fluorescent protein (mito-YFP). To measure FRAP, 3-μm ROIs (; red circles) in double-transfected cells were photobleached to ∼50% of the initial mito-YFP fluorescence by irradiating with a high power laser. This was followed by the analysis of the YFP fluorescence recovery every 400 msec for ∼14 s. Under conditions in which mito-YFP fluorescence in control cells recovered to 65.35 ± 7.6% of prebleach fluorescence, as accessed at the last time points of each measurement, fluorescence recoveries in MARCH5 and MARCH5-expressing cells were significantly higher (; 76.3 ± 9.74 and 76.1 ± 7.02, respectively). Confirming the immunofluorescence results (), mito-PAGFP analysis (), and published data (), there was no statistically significant difference in FRAP between control and wild-type MARCH5-expressing cells (; 68.22 ± 9.22, P > 0.2). This demonstrates that ectopic expression of wild-type MARCH5 does not affect the overall volume of mitochondria, whereas the increased FRAP in the MARCH5 RING mutant–expressing cells reflects increased mitochondrial connectivity. We also performed control FRAP experiments in cells expressing Drp1, a dominant-negative mutant of mitochondrial fission protein Drp1, or in cells transfected with the viral mitochondria-associated inhibitor of apoptosis (vMIA) (). It has been shown that Drp1 induces aberrant mitochondrial elongation (), whereas vMIA causes mitochondrial fragmentation (; ). There was a significant increase of FRAP in Drp1-expressing cells (; 80.2 ± 9.24), and a decrease in vMIA-expressing cells (; 54.03 ± 6.89). There was no significant difference between mitochondrial FRAP values in cells expressing MARCH5 RING mutants and Drp1 (MARCH5/ Drp1, P > 0.1; and MARCH5/Drp1, P > 0.1), leading to the conclusion that MARCH5 RING domain mutation-induced mitochondrial defects are qualitatively and quantitatively comparable to those induced by Drp1, an established inhibitor of mitochondrial division (see also Supplemental text and Fig. S2 and Fig. S3, available at ). Our observations appear to contradict recently published interpretations of mitochondrial phenotypes in MARCH5 RING domain mutant–expressing cells (; ). It has been reported that mitochondrial fragmentation, but not elongation, is a predominant response to inhibition of MARCH5 activity. We found that in all cases analyzed here, after expression of: N- and C-terminal YFP fusion of MARCH5; and C-terminal fusions of MARCH5 and MARCH5 (Fig. S2 A), as well as Myc-tagged MARCH5 (Fig. S2 B), normal morphology or an abnormal elongation and connection of mitochondria prevail in cells expressing low to moderate levels of the mutant proteins. One explanation is that the enlarged and rounded perinuclear mitochondria that are more apparent in cells producing high levels of MARCH, MARCH5, and MARCH5 could be interpreted as “fragmented” at low magnification (Fig. S1, A–C). Yet, when compared with small vesicular mitochondria formed as a result of unbalanced division when mitochondrial fusion is impaired in Opa1 RNAi cells (Fig. S3 A) or Mfn1KO and Mfn2KO mouse embryonic fibroblasts (MEFs) (see ), the round perinuclear organelles in MARCH5 mutant–expressing cells are much larger (Fig. S3 B). Furthermore, expression of the MARCH5 RING mutants does not cause an obvious increase in mitochondrial number (Fig. S3, A and B; unpublished data), expected upon activation of mitochondrial division, suggesting that mitochondrial fission does not cause the MARCH5 inhibition-induced mitochondrial aberrations. In fact, it has been demonstrated, both in mammalian cells () and in yeast (), that strong fission defects can lead to the perinuclear collapse and a rounding of mitochondria, whereas lesser inhibition of this process induces the formation of spread, net-like mitochondria. Taking into consideration that only relatively high levels of MARCH5 mutant expression can stimulate perinuclear collapse and rounding of mitochondria (unpublished data), one can picture that the degree of fission in MARCH5 or MARCH5 transfected cells is inversely proportional to the expression levels of the mutant proteins, as expected to occur in the case of dominant-negative mutants. As a consequence of this relation, in the most extreme cases mitochondrial alterations comparable to those induced by “strong” dominant-negative mutants of Drp1 may take place (). We therefore attempted to test whether round collapsed mitochondria in MARCH5 mutant–expressing cells are separate units or whether they form aberrant but interconnected mitochondrial networks. To attain this we used another FRAP-based approach (Fig. S4, available at ). Multiple cycles of bleach (marked with “V” in Fig. S4, B–D) and fluorescence recovery were applied to seemingly separate round mitochondria (ROI 1 in Fig. S4 A; bleached area diameter ∼2 μm), while avoiding direct photobleaching of the adjacent mitochondria. Changes of fluorescence intensities of the photobleached ROI 1 (Fig. S4 B), a directly adjacent mitochondrion (ROI 2; Fig. S4 C), and a mitochondrion located relatively far from the bleached area (ROI 3; Fig. S4 D) were plotted as a function of time. We found that efficient fluorescence recoveries in ROI 1 occur after each bleach that were associated with a gradual depletion, but also slightly delayed fluorescence recovery in ROI 2 and little change in ROI 3. From these data we can conclude that an exchange of mito-YFP occurs between ROI 1 and ROI 2, and that the mitochondrion in ROI 3 is separate from ROIs 1 and 2. Moreover, because the mitochondrion located in ROI 2 also recovers mito-YFP fluorescence, it most likely is also connected with another mitochondria. Because mitochondrial matrix–targeted GFP can diffuse freely with the rate of 15–19 msec () in normal tubular mitochondria in HeLa cells, the delayed response of ROI 2 indicates that it is likely connected with a mitochondrion located within ROI 1 by abnormally thin mitochondrial tubules. This is consistent with the observations that seemingly separate, round mitochondria visible in Dnm1p mutant are able to exchange matrix-targeted GFP (). In sum, based on the data discussed above, we conclude that in cells expressing MARCH5 and MARCH5 an abnormal interconnection of mitochondria is the main morphological manifestation and that inhibition of MARCH5 activity increases the volume and decreases the number of mitochondrial units within the network. This points to a possible inhibition of mitochondrial fission or activation of fusion by the loss of MARCH5 activity. During the analysis of mitochondrial morphology in wild-type MARCH5-YFP, MARCH5-YFP, and MARCH5-YFP expressing cells we noticed that while wild-type MARCH5 decorated mitochondria uniformly (), MARCH5 (; ) and MARCH5 () localized mainly to submitochondrial aggregates with some diffusely circumscribing the OMM. To quantify the degree of MARCH5 association with the OMM, cells were transfected with wild-type MARCH5-YFP, MARCH5-YFP, or MARCH5-YFP followed by immunostaining for Tom20 and confocal image acquisition (). The quantification of the colocalization of green channel pixels (MARCH5 and MARCH5 RING mutants) with red channel pixels (Tom20) was performed using the “colocalization” function of the image analysis software Volocity. The association degree of different variants of MARCH5 with the OMM is depicted as a Pearson's Correlation coefficient (r) () where the r-values between −0.3 ± 0.3 indicate little or no association, +0.3 ± 0.7 weak positive association, and +0.7 ± 1.0 strong positive association. Using this analysis we found that MARCH5-YFP, which localizes to the OMM in a diffuse manner with only occasional submitochondrial coalescence and extramitochondrial localization, shows a high degree of association with the OMM (r = 0.77 ± 0.03; ). MARCH5-YFP (r = 0.46 ± 0.11; ) and MARCH5-YFP (r = 0.36 ± 0.09; B) display a distinctly lower degree of association with the OMM, indicating that mutations perturbing the activity of the MARCH5 RING domain lead to a redistribution of this protein in the OMM. Next, we studied whether inhibition of the MARCH5 RING domain may affect oligomerization of this protein that may contribute to the focal appearance on mitochondria. To achieve this, mitochondria-enriched heavy membrane fractions (HM) were treated with the chemical cross-linker Bis- Maleimidohexane (BMH) and analyzed by Western blot for Myc-tagged MARCH5 (). We found, in contrast to wild-type MARCH5, which was mostly detected as a 33-kD monomer and an ∼65-kD form, that MARCH5 also formed higher molecular weight complexes. The ∼65-kD form of MARCH5 most likely represents homodimers of this protein (Fig. S4). The molecular components of the high molecular weight complexes containing the MARCH5 RING mutant are currently unknown. However, most likely these structures contribute to the formation of MARCH5 RING mutant submitochondrial foci. This difference in submitochondrial localization and oligomerization of the MARCH5 RING mutants compared with the wild-type protein suggests that protein complexes containing inactive MARCH5 and its substrates are stabilized and/or abnormally enlarged, likely due to perturbed ubiquitylation. We investigated whether certain OMM-associated proteins, including those participating in mitochondrial morphogenesis, also accumulate in MARCH5- and MARCH5- enriched mitochondrial aggregates. Cells transfected with MARCH5-YFP and MARCH5-YFP, in the presence or absence of cotransfection with mammalian expression vectors encoding Myc-tagged integral membrane proteins of the OMM, OMP25 and Fis1, were immunostained with anti-Myc antibodies or antibodies detecting endogenous Tom20, Tom22, endophilin B1, and Drp1. Although changes in the distribution of Tom20 (), Tom22, OMP25 and endophilin B1, as well as YFP-tagged mitochondrial fusion proteins Mfn1 and Mfn2 were not noticeable (unpublished data), an abnormal localization pattern of the fission protein Drp1 was apparent (′). In the most extreme cases, occurring in cells with a high degree of MARCH5-YFP expression, formation of extremely enlarged perinuclear clusters of Drp1 was observed (Fig. S5 A, available at ). Consistent with MARCH5, expression of MARCH5-YFP also induced abnormal mitochondrial clustering of Drp1 (). Furthermore, confocal analysis () and quantification of Pearson's Correlation values () of cells expressing MARCH5-YFP or MARCH5-YFP and stained for Drp1 revealed clear colocalization of these proteins (), suggesting that MARCH5 participates in the regulation of Drp1. However, the fission protein Fis1 (; ) did not localize to these MARCH5/Drp1-positive structures (Fig. S5 B). Because yeast Fis1p is required to recruit Dnm1p early during mitochondrial scission (), MARCH5 could be involved in mitochondrial scission downstream of human Fis1. Membrane translocation of the Drp1 homologue, dynamin, and dynamin-mediated endocytic vesicle scission are regulated by binding and hydrolysis of GTP (; ). Therefore, we tested whether Drp1 colocalization with MARCH5 depends on Drp1 GTPase activity. Cells were cotransfected with MARCH5-YFP and GTPase- inactive Drp1, immunostained for Drp1 and analyzed by confocal microscopy. As shown before (), unlike overexpressed wild-type Drp1 that correctly localizes to small submitochondrial foci (; ), Drp1 forms enlarged aggregates that are only partly associated with mitochondria. We found that these Drp1 containing structures do not colocalize with MARCH5-YFP (Fig. S5 C), whereas overexpressed wild-type Drp1 does (unpublished data). This suggests that the GTPase activity of Drp1 is required for its translocation to MARCH5-YFP–enriched complexes on the OMM. Interestingly, there was no detectable effect of Drp1 on the submitochondrial localization of wild-type MARCH5-YFP and MARCH5-YFP. In contrast, knockdown of Drp1 expression resulted in the redistribution of MARCH5-YFP from foci () into a more diffuse mitochondrial localization () that colocalized with Tom20 to a significantly greater extent than when expressed in control RNAi cells (). This demonstrates that MARCH5-YFP clusters are to some extent Drp1 dependent. However, MARCH5 RNAi did not induce abnormal distribution of Drp1, as revealed by immunofluorescence and Western blot (unpublished data). Collectively, these data suggest a functional interdependence between MARCH5 and Drp1. The data showing that expression of MARCH5 or MARCH5 has a strong impact on the localization of Drp1, and that Drp1 expression impacts MARCH5 distribution, suggest that MARCH5 is involved in the regulation of the mitochondrial fission complexes. The abnormal organization of Drp1 on mitochondria in MARCH5 and MARCH5, but not wild-type MARCH5-expressing cells (), suggests that the transition between certain mitochondrial steps of the Drp1-dependent fission cycle is blocked by the inhibition of MARCH5 activity. In mammalian cells as well as in yeast, Drp1/Dnm1p cycles between the cytosol and mitochondria (; ). It is possible that the formation of abnormal fission complexes in response to the expression of MARCH5 mutants is caused by changes in the subcellular dynamics of Drp1. To test this hypothesis, the mobility of YFP-Drp1 in cells expressing MARCH5-CFP, MARCH5-CFP, and MARCH5-CFP (MARCH5/Drp1 plasmid ratio 5:1) or in control cells was analyzed by FRAP (). We confirmed that, like endogenous Drp1 (), the subcellular localization of YFP-Drp1 is altered by the expression of MARCH5 RING mutants and that the YFP tag does not influence colocalization of these proteins (unpublished data). To measure FRAP, a 4-μm strip through YFP-Drp1–expressing cells was photobleached by irradiation with a high power laser and subsequent fluorescence recovery was monitored over ∼20 s. The recovery curves were normalized, with the last time point values of the fastest recovering wild-type MARCH5 taken as 100%, and immediate postbleach values of each sample (1.06 s postbleach) as 0%. As shown in , a distinct decline in the mobility of assembled complexes of YFP-Drp1 in cells expressing MARCH5 or MARCH5 compared with YFP-Drp1, or Drp1 and wild-type MARCH5 cotransfected cells was evident. We also tested the effects of MARCH5 activity on the expression level, mitochondrial association (), and the oligomerization () of endogenous Drp1. Whole cell lysates (WCL) and heavy membrane (HM) fractions obtained from HeLa cells expressing Myc-tagged, wild-type MARCH5 or MARCH5 were analyzed by Western blot (). The data show that, while the total expression level of Drp1 is not influenced by MARCH5 RING domain activity, the quantity of Drp1 associating with mitochondria is slightly increased in MARCH5-expressing cells and, to a lesser degree, in cells expressing wild-type MARCH5 (). The WCL and HM fractions used in these analyses were obtained. Thus, one can conclude that the MARCH5 RING mutants affect the cellular dynamics of Drp1, but not the degradation of this protein (see also ). Next, to reveal the oligomeric state of mitochondria-associated Drp1, aliquots of the HM fractions were also treated with BMH (). Using this approach, we did not detect any clear alterations in the Drp1 oligomerization. These data suggest that the organization of the subunits forming the light microscopy-detectable Drp1 complexes is not influenced by MARCH5 activity. It has been reported that in addition to Drp1 and Fis1 (; ), proteins required for mitochondrial division, MARCH5 also coimmunoprecipitates with Mfn2 (), a mitochondrial fusion protein. Therefore, we analyzed the effect of wild-type MARCH5 and MARCH5 expression on the proteasome-dependent degradation of Mfn2 (). To inhibit a proteasome-dependent degradation of Mfn2, cells were treated with the proteasome inhibitor MG132. Mitochondria-enriched HM fractions were obtained from MG132-treated or untreated HeLa cells expressing Myc-tagged, wild-type MARCH5 or MARCH5. Although proteasomal activity is required for Mfn2 degradation, MARCH5 does not influence Mfn2 levels in either MG132-treated or untreated cells (). The relative contribution of MARCH5 to mitochondrial division and fusion is not known. By analyzing whether wild-type MARCH5 and MARCH5 RING mutants can complement mitochondrial defects in Mfn1 (Mfn1KO) and Mfn2 (Mfn2KO) knock-out MEFs, we tested the degree to which MARCH5 activity influences mitochondrial fission versus fusion. It has been shown that the abnormally fragmented mitochondria in Mfn1KO and Mfn2KO MEFs are formed as a result of reduced fusion and unrestrained fission of these organelles (). The ectopic expression of Drp1, a strong inhibitor of mitochondrial division, reverses the mitochondrial defects in Mfn1KO and Mfn2KO cells (). This indicates that the residual fusion of mitochondria in these cells is sufficient to rebuild tubular mitochondria. We found that, in contrast to wild-type MARCH5 (), when expressed in Mfn1KO or Mfn2KO cells, both MARCH5 () and MARCH5 () distinctly reversed mitochondrial fragmentation, inducing the formation of normal reticular mitochondria () or, in the most extreme cases, induced abnormally interconnected organelles (). The MARCH5 RING mutant–induced reconstitution of reticular mitochondria in Mfn1KO and Mfn2KO cells indicates that the activity of the MARCH5 RING domain is a strong determinant of mitochondrial membrane dynamics. Because MARCH5 coimmunoprecipitates with the mitochondrial fusion protein Mfn2 (), it is possible that, due to a partial functional redundancy between Mfn1 and Mfn2 (), MARCH5 RING mutants may stimulate the activity of Mfn2 in Mfn1KO cells, and Mfn1 activity in Mfn2KO cells. Therefore, we tested the relative contribution of mitochondrial fusion to the MARCH5 RING mutant– induced elongation of Mfn2KO cell mitochondria (). We measured mitochondrial fusion in control Mfn2KO cells (), and in Mfn2KO cells expressing MARCH5 () using a mito-PAGFP-dilution–based assay (). As a positive control we used Mfn2KO cells ectopically expressing Mfn2 (). The relative rates of mitochondrial fusion were quantified at different time points after ROI photoactivation, as indicated (). At 45 min after ROI photoactivation, mito-PAGFP fluorescence dilution in Mfn2KO MEFs was dramatically increased by ectopic expression of Mfn2 (; Δ ∼28.5% in Mfn2KO MEFs, compared with Δ ∼76% in Mfn2KO MEFs expressing Mfn2; P = 1.176 × 10). These data indicate that the mitochondrial fusion in Mfn2KO cells was reconstituted by the reexpression of Mfn2. In contrast, mitochondrial fusion in Mfn2KO MEFs was only slightly improved by the expression of MARCH5 (; at 45 min after photoactivation Δ ∼39.7% in Mfn2KO MEFs expressing MARCH5; P = 0.026). These MARCH5-induced improvements of mitochondrial fusion in Mfn2KO MEFs suggest that to some degree mitochondrial fusion could be regulated by MARCH5 activity. However, we found that there is no additional affect of MARCH5 RING mutants on mitochondrial morphology in Drp1 RNAi cells (), whereas overexpression of Mfn2 dramatically elongates mitochondria in Drp1 activity-deficient cells (). Thus, one can conclude that MARCH5 RING mutants reconstitute tubular mitochondria in Mfn1KO MEFs and Mfn2KO MEFs mainly through their inhibition of mitochondrial fission. Altogether, the data presented above suggest that the abnormal mitochondrial interconnection in MARCH5 or MARCH5 expressing cells is due to an inhibition of Drp1-dependent mitochondrial fission. Therefore, we tested whether the ectopic expression of Drp1 could reverse this mitochondrial phenotype. Cells were cotransfected with MARCH5 () or MARCH5 RING mutants (; green on overlay images) as well as Drp1 (MARCH5/Drp1 molar ratio 1:5), followed by immunostaining for Drp1 (red on overlay images) and Tom20 (blue on overlay images), to reveal mitochondrial morphology. Cells expressing elevated levels of Drp1 together with different variants of MARCH5 (, labeled with asterisk) were scored for mitochondrial morphology (). We found a substantial reversal of the aberrant mitochondrial phenotypes in cells cotransfected with MARCH5 mutants and Drp1. In 64.25% of the cells expressing MARCH5 and 65.5% of cells expressing MARCH5 together with elevated levels of Drp1, mitochondrial network organization resembled that of control cells. The overexpression of Drp1 alone only slightly affected mitochondrial structure (), inducing mitochondrial fragmentation in <10% of the cells. This indicates that ectopic expression of Drp1 specifically compensates for the MARCH5 RING mutant–induced alterations of mitochondria and suggests further that MARCH5 RING mutants may compete with Drp1 for certain factors involved in mitochondrial fission. Because expression of the mitochondrial fission protein Fis1 (; ) does not reverse mitochondrial alterations induced by expression of MARCH5 RING mutants (unpublished data; ) we conclude that MARCH5 regulates mitochondrial fission downstream of Fis1. This conclusion is also supported by the fact that expression of MARCH5 RING mutants had no detectable impact on the submitochondrial distribution of Fis1 (Fig. S5 B). The data described in this study indicate that MARCH5, an integral ubiquitin E3 ligase of the OMM, is required for mitochondrial division in mammalian cells. This conclusion is based on the following evidence: Expression of MARCH5 RING mutants induces abnormal interconnection of mitochondria, as well as an increase in size, and decrease in the number of mitochondrial network units. MARCH5 RNAi induces mitochondrial elongation. The subcellular distribution and mitochondrial association of Drp1, a protein essential for mitochondrial division, is significantly influenced by MARCH5 RING domain activity and the localization of MARCH5 RING mutants on mitochondria is altered by knockdown of Drp1 expression. Moreover, MARCH5 RING mutants colocalize with Drp1-enriched mitochondrial foci. Ectopic expression of Drp1 reverses the MARCH5 RING mutant–induced mitochondrial abnormalities, suggesting strong functional interdependence of MARCH5 and Drp1. MARCH5 RING mutants reverse mitochondria defects in Mfn1KO and Mfn2KO MEFs, most likely through the inhibition of mitochondrial division. The data showing the formation of abnormally enlarged clusters of Drp1, mitochondrial accumulation of Drp1, as well as an inhibition of Drp1 mobility in MARCH5 RING mutant–expressing cells support the hypothesis that MARCH5 is a positive regulator of mitochondrial division. Since it has been reported before that MARCH5 can be coimmunoprecipitated only with the ubiquitylated form of Drp1 (), the mitochondrial scission complexes are likely regulated by MARCH5-mediated conjugation of ubiquitin. Consistent with this, a reduction of MARCH5 ubiquitin ligase activity apparent after expression of MARCH5 RING mutants or by RNAi leads to major defects in mitochondrial division. The opposite phenotype, mitochondrial fragmentation, could be expected to occur if MARCH5 mediated degradation of Drp1, Fis1, or any positive regulator of mitochondrial division. The data showing no clear effect of the MARCH5 activity on the degradation of Drp1 or Mfn2, as well as the rescue of the fragmented mitochondria in Mfn1KO and Mfn2KO MEFs by MARCH5 RING mutants, support the model of a “molecular switch”. In this scenario, MARCH5 could be essential for the activity of mitochondrial fission complexes, for example by affecting Drp1 GTPase activity or influencing protein interactions within the mitochondrial fission machinery. Increased colocalization of MARCH5 RING mutants and Drp1, as well as abnormal mitochondrial association of Drp1 that could occur due to inhibition or delay of a specific step of the fission reaction, perhaps via perturbed ubiquitylation of an essential fission protein, further support this hypothesis. Yet, it cannot be excluded that MARCH5 regulates the stability of a currently unknown protein repressor of mitochondrial fission. The abnormal accumulation of such a factor in MARCH5 RING mutant–expressing cells could lead to the inhibition of fission complexes and elongation of mitochondria. It has been proposed that conjugation of the ubiquitin-like molecule SUMO-1 stabilizes Drp1 in mammalian cells (), especially in the context of apoptosis induction (). This implies that conjugation of SUMO-1 could be the main determinant of Drp1trafficking onto or off mitochondria, however participation of a ubiquitin-dependent pathway, as well as yet-uncharacterized E3 ubiquitin ligases, cannot be excluded. Indeed, we have identified several novel RING finger proteins that localize to the mitochondria (unpublished data), some of which could mediate ubiquitin-dependent degradation of proteins from the OMM, including mitochondrial morphogenesis factors. However, as specific Lys residues of the target proteins may be modified by ubiquitin, as well as SUMO, resulting in different functional consequences, it is also possible that both sumoylation and ubiquitylation participate in regulation of Drp-1–dependent mitochondrial fission. HeLa cells (American Type Culture Collection) were cultured in complete Dulbecco's minimum essential medium supplemented with 10% heat- inactivated fetal calf serum, 100 U/ml penicillin, and 100 μg/ml streptomycin in 5% CO at 37°C. Wild-type, Mfn1KO, and Mfn2 KO cells were cultured as described previously (, ). Transfections were performed using FuGENE 6 (Roche) according to the manufacturer's instructions. Human MARCH5 (GenBank/EMBL/DDBJ accession no. ) was amplified using the proof reading polymerase Pfx (Invitrogen) from IMAGE clone #3905766 (Invitrogen) with the following primers: 5′-ATCGCCTCGAGCCATGCCGGACCAAGCCCTACA-3′ and 5′-ACTAGGGATCCCGTGCTTCTTCTTGTTCTGGATAATT-3′ and cloned into the XhoI and BamHI site of YFP-N1 and YFP-C1 (BD Biosciences) to generate YFP-MARCH5 and MARCH5-YFP. MARCH5-YFP and MARCH5-YFP were generated using the mutagenic primers: 5′-GAGGATCTACAAAATGGGTTTGGCAAGCTTGTCTACAACGCTGGGTG-3′ and 5′-CACCCAGCGTTGTAGACAAGCTTGCCAAACCCATTTTGTAGATCCTC-3′ for MARCH5-YFP and 5′-GTACAGCCAGAGTGGCATCTCCATGGTCCAATGCTGAATACCTAATAG-3′ and 5′-CTATTAGGTATTCAGCATTGGACCATGGAGATGCCACTCTGGCTGTAC-3′ for and MARCH5-YFP and Pfu polymerase (Stratagene) with MARCH5-YFP as template. MARCH5 was obtained using the following primers: 5′-ATCGCCTCGAGCCATG GCTGAATACCTAATAGTTTTTCC-3′ and 5′-ACTAGGGATCCCGTGCTTCTTCTTGTTCTGGATAATT-3′ and cloned XhoI/BamHI into YFP-N1 to generate MARCH5-YFP. All constructs were verified by sequencing. The mouse Mfn2 expression vector () was provided by D. Chan (Caltech, Pasadena, CA). The short hairpin RNAi was performed as described previously (). The oligonucleotides used for MARCH5 shRNAi generation were: 5′-CTGGAATTCGAGGAACAGGACAACCTATGAAGCTTGATAGGTTGTCTTGTTCTTCGAATTCTAGTTGTTTTTTG-3′, and 5′-GATCCAAAAAACAACTAGAATTCGAAGAACAAGACAACCTATCAAGCTTCATAGGTTGTCCTGTTCCTCGAATTCCAGCG-3′. Drp1 shRNAi- and control GFP shRNAi-generating vectors were used as described previously (). For confocal microscopy analyses cells were cultured in 2-well chambered Borosilicate coverglasses (Nunc). Images were captured using an LSM510 Meta imaging station (Carl Zeiss MicroImaging, Inc.). The images of immunostained, fixed cells were acquired with 100×/1.4 color corrected Plan-Apochromat (Carl Zeiss MicroImaging, Inc.) objective lens. For live cell imaging, cells were maintained at 35°C for no longer than 30 min and imaged using 100×/1.45 α-Plan-FLUAR objective lens (Carl Zeiss MicroImaging, Inc.). FRAP analysis of mitochondrial volume was performed as described previously (; ), unless stated otherwise. Subcellular mobility of Drp1 was determined using FRAP analysis of YFP-Drp1. Cells were cotransfected with YFP-Drp1 and CFP fused MARCH5 constructs in the molar ratio 1:5. The 4-μm strip was photobleached in YFP-Drp1 expressing cells using a 30.0-mV argon laser set to 514 nm with 50% of the laser power output and 100% of transmission. This was followed by the time-lapse 12-bit image acquisition of YFP-Drp1 fluorescence recovery with the same laser set to 50% of the laser power output and 1% of transmission. Time-lapse imaging was performed for 19.66 s after photobleaching with the interval of 400 ms. The detector gain was maintained at the levels below saturation of the brightest structures, resulting in FRAP of mostly focal Drp1. The FRAP curves were obtained using LSM510 software (Carl Zeiss MicroImaging, Inc.). The photoactivation studies of mitochondrial matrix targeted photoactivable GFP (mito-PAGFP) were performed as described previously (). For the mito-PAGFP and FRAP analysis in cotransfected cells, cells expressing WT MARCH5 and MARCH5 RING mutants were identified by CFP fluorescence. Cells were fixed with 4% paraformaldehyde in PBS for 30 min, permeabilized with 0.15% Triton X-100 for 20 min, and then blocked with 10% BSA for 45 min at room temperature. The cells were then immunostained for 90 min at RT, or overnight at 4°C with the following primary antibodies: anti-cytochrome mAbs (BD Biosciences; 1:250), anti-Tom20 mAbs (BD Biosciences; 1:250), anti-Tom20 polyclonal antibodies (Santa Cruz Biotechnology, Inc.; 1:500), anti-Dlp1 mAbs (Drp1; clone 8; BD Biosciences; 1:100), and anti-Myc mAbs (Santa Cruz Biotechnology, Inc., 1:1,000). Cells were washed and stained with AlexaFluor (488-, 546- and 633)- conjugated secondary antibodies (BD Biosciences; 1:250) for 1 h at RT. For the colocalization assay the projections of 12-bit confocal z-sections covering the entire depth of the cell, obtained with an interval of 0.25 μm, were used. Images in all experimental groups were obtained with the same settings of the confocal microscope, except for detector gain adjustments in the green channel that were performed to normalize saturations levels. Images were analyzed using the colocalization menu of the Volocity software (Improvision). The ROIs covering the entire area of the cell were created, followed then by the adjustment of the red-channel lower threshold used in order to exclude background staining from further analysis. The colocalization statistics were determined from the points included within ROIs and presented as Pearson's Correlation values that can vary between −1 and 1. Pearson's Correlation reflects the linear relationship between two variables. When applied to image analysis it can be used to establish the statistical significance of the association of pixels from different channels of confocal images, and therefore to determine the colocalization degree of different proteins in the cell. A correlation of zero means that there is no linear relationship between the variables whereas a correlation of 1 indicates perfect positive linear relationship. For antibody production, full-length MARCH5 was amplified using 5′-AACTAGCATATGCCGGACCAAGCCCTACA-3′ and 5′-ACTAGGGATCCTTATGCTTCTTCTTGTTCTGGATA-3′ and cloned into pET15b (CLONTECH Laboratories, Inc.) using NdeI/BamHI to obtain pET15b-MARCH5. were transformed with pET15b-MARCH5. The inclusion bodies containing full-length MARCH5 were isolated and solubilized in denaturizing bind buffer (20 mM Tris, pH 7.4, 500 mM NaCl, 40 mM imidazole, and 6M Guanindium HCl) and purified using HISTrap HP (GE Healthcare) according to the manufacturer's suggestions. MARCH5 was refolded by dialysis against refolding buffer (20 mM Tris, pH 8.0, 500 mM NaCl, 500 mM Arginin/HCl, and 0.2% Triton X-100). MARCH5-specific antibodies were raised using recombinant full-length MARCH5 injected into a New Zealand White rabbit. The antibodies were affinity purified from rabbit serum using MARCH5 bound to HiTrap NHS-activated HP (GE Healthcare) according to the manufacturer's recommendations. Anti-Mfn2 polyclonal antibodies were obtained using the same protocol, with the 1–405 Mfn2 fragment used to immunize rabbits. For Western blot analysis HeLa cells were transfected with MARCH5, MARCH5, or control vector (YFP-N1; CLONTECH Laboratories, Inc.) using Effectene (QIAGEN) according to the manufacturer's recommendation. Cells were harvested 18 h after transfection and then suspended in cell lysis buffer (250 mM Sucrose, 20 mM Hepes/KOH, pH 7.5, 10 mM KCl, 1.5 mM MgCl, 1 mM EDTA, 1 mM EGTA, and complete protease inhibitor [Roche]). Cells were disrupted by 15 passages through a 25-gauge needle (with 1-ml syringe). The heavy membrane fraction (HM) was obtained by centrifugation at 10,000 for 10 min. For cross-linking experiments the HM fraction was resuspended in lysis buffer and incubated on ice for 30 min either with 1 mM BMH (Pierce Chemical Co.) or with DMSO as control. The reaction was stopped by addition of 100 mM glycine, and samples were boiled in SDS PAGE sample buffer and analyzed by Western blot. The commercial antibodies used in Western blot analysis were: anti-Drp1 (Dlp1) mAbs (clone 8; BD Biosciences), anti-Hsp60 mAbs (StressGen Biotechnologies), anti-α-tubulin mAbs (Sigma-Aldrich), and anti-Fis1 polyclonal antibodies (Biovision). Supplemental text describes additional analyses of mitochondrial phenotypes in MARCH5-expressing cells. Figs. S1 and S2 show additional examples of mitochondrial phenotypes in MARCH5 RING mutant transfected cells. Fig. S3 shows FRAP analysis in cells expressing high amounts of MARCH5. Fig. S4 shows analyses of the oligomeric state of MARCH5. Fig. S5 demonstrates the degree of colocalization of Drp1, Drp1, and Fis1 with MARCH5. Online supplemental material is available at .

Competence has long been recognized as a critical mechanism for restricting the developmental potential of cells to specific fates (). Competence is progressively regulated such that cells achieve a series of competent states, each of which sets up the subsequent developmental response during the life cycle. In this manner, widespread signals are refined to direct spatially and temporally restricted biological responses. Although often invoked, relatively little is known about how competence is achieved at a molecular level. We are studying the programmed cell death of larval tissues during metamorphosis as a model system for understanding the molecular basis of competence and its role in determining the appropriate temporal progression of steroid-triggered biological responses during development. Programmed cell death plays a central role in animal development, eliminating unwanted tissues, controlling cell numbers, and sculpting complex structures. The decision to live or die is determined by each cell based on a critical balance between evolutionarily conserved death activators and death inhibitors (; ). Several signals can affect this balance, regulating the patterns of programmed cell death in a precise temporal and spatial manner. In frogs, thyroid hormone signals the destruction of the tadpole tail and remodeling of the intestine as the animal progresses from a juvenile to adult form (). Similarly, steroid hormones regulate mammalian apoptotic pathways, including the glucocorticoid-induced apoptosis of immature thymocytes and mature T cells (). A dramatic manifestation of steroid-triggered cell death can be seen in the fruit fly , where the steroid hormone ecdysone directs the massive and rapid destruction of larval tissues during metamorphosis. Although relatively little is known about the mechanisms of hormone-regulated programmed cell death in vertebrates, considerable insights into this pathway have been gained in . A high titer pulse of ecdysone at the end of larval development acts through the ecdysone receptor (EcR)/Ultraspiracle nuclear receptor heterodimer to signal puparium formation and the destruction of several larval tissues, including the midgut (; ). A second ecdysone pulse, ∼10 h after puparium formation (APF), triggers adult head eversion, marking the prepupal–pupal transition and signaling the rapid destruction of the larval salivary glands (; ). Destruction of the larval midguts and salivary glands is accompanied by classic hallmarks of apoptosis, including acridine orange staining, TUNEL staining, and caspase activation (). These larval tissues, however, undergo a distinct form of programmed cell death referred to as autophagy, characterized by the formation of intracellular autophagic vesicles (; ). Three related death activator genes play a central role in the control of programmed cell death: (), (), and (; ; ). Elimination of these genes completely blocks programmed cell death, whereas ectopic expression of any is sufficient to trigger a cell death response. The Rpr, Hid, and Grim proteins interact with the cell death inhibitor inhibitor of apoptosis 1 (DIAP1), disrupting its interaction with caspases and targeting DIAP1 for degradation, allowing caspase activation and cell death (for review see ). A similar regulatory pathway is used in the destruction of larval salivary glands during metamorphosis, where the prepupal pulse of ecdysone triggers a transcriptional cascade that culminates in and induction, overcoming the inhibitory effects of DIAP1 and initiating tissue destruction (, ; ; ; ). Here, we show that during most of larval development, DIAP1 cannot be overcome by death activator expression in larval salivary glands. This is due to high levels of DIAP1 in this tissue at early stages, substantially higher than are present at the onset of metamorphosis. This switch in DIAP1 levels occurs in the mid-third instar and depends on EcR and the CREB binding protein (CBP), defining a new transcriptional hierarchy that regulates salivary gland cell death. CBP is both necessary and sufficient to down-regulate DIAP1, indicating a central role for this factor in mediating the switch in DIAP1 levels. The resulting threshold level of DIAP1 is sufficient to hold back cell death, but low enough to be sensitive to subsequent ecdysone-induced and expression in prepupae. Larval salivary gland cell death is thus a two-step temporally regulated response that involves sequential effects on DIAP1 levels. The first step, mediated by CBP, reduces DIAP1 levels, providing the competence to die. The second step, mediated by and , eliminates residual DIAP1 and triggers tissue destruction. This work provides a new context for understanding how steroid hormones regulate programmed cell death during development. It also provides a molecular basis for understanding how competence is achieved to allow the appropriate temporal progression of hormone-regulated biological responses during development. To further our understanding of the regulation of steroid-triggered cell death, we screened for mutations that block the destruction of the larval salivary glands in living animals using a salivary gland–specific GFP reporter (). Starting with a collection of -element–induced lethal mutations (), we identified several complementation groups that resulted in persistent larval salivary glands (unpublished data). One complementation group of three alleles mapped to an ∼400-bp region near the 5′ end of the gene that encodes the homologue of CBP/Nejire (Nej), adjacent to a previously uncharacterized exon of (, linked to by RT-PCR; unpublished data). All three mutations (, , and ) fail to complement a known mutation, . The -null allele, like the three -element–induced mutations, leads to developmental arrest soon after the prepupal–pupal transition, with persistent larval salivary glands (). This block in cell death can be phenocopied by salivary gland–specific inactivation of using RNAi, demonstrating that the effects of on cell death are tissue autonomous (). mutant pupae are virtually indistinguishable from their wild-type counterparts at 12 h APF. The absence of defects in morphogenesis of adult structures and the lack of developmental delay between puparium formation and the prepupal–pupal transition suggest that the global ecdysone-regulated control of metamorphosis is normal in mutant animals. Similarly, mutant salivary glands synthesize glue proteins at the appropriate time during third-instar larval (L3) development and secrete their contents in response to ecdysone at puparium formation, as normal (unpublished data). In contrast to wild-type glands, however, which show clear signs of cell death at ∼13.5 h APF (), mutant salivary glands survive for at least 72 h, with 70% of mutant pupae displaying persistent glands ( = 60). The mutant glands fail to display morphological signs of tissue breakdown, showing tight cellular association at 13.5 and 16 h APF (). Moreover, although wild-type glands show staining for active caspase-3 at 13.5 h APF (), mutant salivary glands do not display this response, consistent with their block in programmed cell death (). This phenotype is specific to the locus, as it can be rescued by the expression of a wild-type transgene in combination with a driver, which is induced in mid-L3 salivary glands and expressed through puparium formation (). Of 51 mutant pupae examined from a control cross of /, ; [], [] and , flies, 47% had persistent salivary glands. In contrast, of 12 mutants examined from the rescue cross of /, ; [], [] and , ; [] flies, 17% had persistent salivary glands, a threefold reduction from the control. CBP is a key transcriptional coactivator, bridging transcription factors with the basal transcriptional machinery and directing histone acetylation. CBP can physically interact with a large number of transcription factors, including nuclear receptors, and has been implicated in regulating cell death (). Given its general role in transcriptional control, we predicted that mutations would act at the top of the ecdysone-triggered death cascade, disrupting induction of the key death activators and at the prepupal–pupal transition. Surprisingly, however, Northern blot analysis of RNA isolated from control and mutant salivary glands revealed that and are induced normally in mutant glands, as are the apical caspase gene and the Apaf-1 adaptor encoded by (). Expression of any one of these four death regulators is sufficient to trigger cell death, an observation that appears inconsistent with the healthy morphology of mutant salivary glands (; ; ; ). This apparent paradox, however, can be explained by up-regulation of , the only known death inhibitor that acts parallel to or downstream from these death activators (; ; ; ; ). In addition, salivary gland–specific overexpression of phenocopies the persistent salivary gland phenotype of mutants, indicating that selective up-regulation of is sufficient to block salivary gland cell death (). The observation that levels are up-regulated in mutant salivary glands () suggested that levels may be high early in development and that may down-regulate at a specific time, achieving the final level seen at the onset of metamorphosis. We thus shifted our attention to examining DIAP1 protein levels and looking at earlier stages of wild-type development. Consistent with this model, DIAP1 is abundant in the salivary glands of control first (L1), second (L2), and early third instar larvae (see ), but drops to low levels by the end of L3 (see ). Staining of salivary glands from staged L3 with DIAP1 antibodies revealed that this switch in DIAP1 expression levels occurs between 20–24 and 34–38 h after the L2–L3 molt () and fails to occur in mutant salivary glands (). Moreover, ectopic expression of in early L3 is sufficient to direct premature DIAP1 down-regulation, defining CBP as a critical regulator of DIAP1 levels and suggesting that stage-specific expression of CBP in mid-L3 directs this switch in expression (). A critical developmental transition occurs about 1 d after the L2–L3 molt, in preparation for puparium formation 1 d later, defining what has been called the mid-third instar transition (; ; ). This transition is manifested by a global switch in gene expression as well as behavioral and metabolic changes that culminate in larval wandering and puparium formation. The mid-L3 transition occurs in synchrony with a low titer pulse of ecdysone, and at least some aspects are dependent on (). Northern blot analysis of RNA isolated from staged larval salivary glands revealed that both and are expressed at 20–24 h after the L2–L3 molt and down-regulated by 34–38 h, in synchrony with the switch from to glue gene expression, a hallmark of the mid-third instar transition (; ; ; ). This change in levels, combined with our genetic studies, indicates that although is required for DIAP1 down-regulation in mid-L3, it is not needed to maintain expression in salivary glands. The observation that transcript levels and protein levels drop at the same time is consistent with the short half-life of DIAP1 protein (). A more detailed understanding, however, of the precise timing and mechanism of transcriptional repression remains to be addressed. The down-regulation of fails to occur when function is disrupted by RNAi () or upon expression of a dominant-negative form of (; unpublished data). In addition, low levels of mRNA can be induced by ecdysone in cultured larval organs, and this response is enhanced upon addition of the protein synthesis inhibitor cycloheximide, defining it as a primary response to the hormone (). Collectively, these observations suggest that ecdysone-induced expression of in larval salivary glands leads to down-regulation as part of the mid-L3 transition. The high level of DIAP1 in early larval salivary glands raises the interesting possibility that this tissue may be resistant to cell death, even in the presence of ectopic death activator expression. Consistent with this hypothesis, control L1, L2, and early L3 salivary glands are resistant to ectopic overexpression, surviving normally and showing no staining for active caspase-3 (; ; and not depicted). In contrast, salivary glands from animals after the mid-L3 transition are susceptible to -induced cell death ( and ). This sensitization for -mediated cell death fails to occur in mutant salivary glands (). Thus, although wild-type salivary glands do not display caspase activation immediately after the mid-L3 transition (), 77% of salivary glands are positive for this assay at the same stage in development in the presence of ectopic expression (heat-treated animals; = 30; ). This susceptibility to -mediated cell death drops to 8% in mutant salivary glands (heat-treated ; animals; = 26; ). These observations demonstrate that expression at the mid-L3 transition is a prerequisite for destruction of this tissue by death activators and suggest that this sensitization is conferred through specific down-regulation of DIAP1. #text Further information on the alleles described in this paper can be found on FlyBase using the Bloomington stock numbers: , , and (12265; Fig. S1, available at ). The allele was provided by S. Smolik (Oregon Health & Science University, Portland, OR). Embryos were collected in 6-h intervals on molasses/agar plates supplemented with yeast paste, transferred to fresh egg caps, and allowed to age as necessary. For analysis of the mid-third instar transition, larvae were staged relative to the second-to-third instar molt. For analysis of the prepupal–pupal transition, animals were staged relative to the white prepupal stage (0 h APF). For heat shock–driven expression of transgenes, staged animals were transferred to a 1.5-ml microcentrifuge tube plugged with cotton, subjected to a 30-min heat treatment in a 37.5°C water bath, and allowed to recover at 25°C for at least 6 h before dissection and fixation. Larval salivary glands were dissected from the appropriate stage and fixed with 4% formaldehyde, 1× PBST (PBS and 0.1% Tween-20), and 3 vol of heptane for 30 min at room temperature. Samples were washed four times in 1× PBST and blocked in 1× PBST and 4% normal goat serum (NGS) for 2 h at room temperature. The samples were stained with antibodies directed against either DIAP1 (a gift from B. Hay, California Institute of Technology, Pasadena, CA) at 1:500 dilution or against the cleaved/active form of caspase-3 (Cell Signaling Technologies) at 1:1,000 in 1× PBST and 4% NGS overnight at 4°C. Samples were washed four times in 1× PBST and stained with either Cy3 donkey anti–mouse secondary antibody (Jackson ImmunoResearch Laboratories) for DIAP1 or Cy3 donkey anti–rabbit secondary antibody (Jackson ImmunoResearch Laboratories) for active caspase-3, at 1:400 in 1× PBST and 4% NGS for 2 h at room temperature. Samples were mounted in Vectashield (Vector Laboratories). Images in were obtained using a 2.5×/0.075 Plan-neofluar objective (Carl Zeiss MicroImaging, Inc.) on a microscope (Axiophot; Carl Zeiss MicroImaging, Inc.) and captured with a digital charge-coupled device camera (SensiCamQE; Cooke Corp.) and Slidebook 3.0 software (Intelligent Imaging Innovations, Inc.). Images in were obtained using a 10×/0.30 Plan-neofluar objective (Carl Zeiss MicroImaging, Inc.) on a microscope (Axioskop2; Carl Zeiss MicroImaging, Inc.) and captured with a charge-coupled device camera (CoolSNAP-Pro; Media Cybernetics) and Image-Pro Plus software (Media Cybernetics). The remaining images were captured as a z series with a confocal laser-scanning microscope (MRC1024; Bio-Rad Laboratories) using 10×/0.30 Plan-neofluar () or 20×/0.50 Plan-neofluar (; ; and ) objectives. All images were acquired at room temperature. Each experiment was performed using identical laser intensity and acquisition parameters to allow direct comparison of control and mutant salivary glands. All figures were processed in parallel with Photoshop CS (Adobe) and assembled in Illustrator CS (Adobe). Wild-type Canton-S late third-instar larvae, 18 h before puparium formation, were dissected and cultured in Schneider's medium (Invitrogen) in the presence of 5 μM 20-hydroxyecdysone (Sigma-Aldrich), 85 μM cycloheximide (Sigma-Aldrich), or both compounds, for 2 or 6 h. 20-hydroxyecdysone is the physiologically active form of ecdysone. Total RNA was extracted from whole animals and analyzed by Northern blot analysis, as described previously (). Larval salivary glands were dissected from animals staged at 20–24 or 34–38 h from the second-to-third instar larval molt or relative to puparium formation in 2-h intervals. Equal amounts of total RNA, isolated using Tripure (Roche), were fractionated on 1% formaldehyde gels and transferred to nylon membranes for Northern blot hybridization. Probes were prepared as described previously (). A pair of oligonucleotides, 5′-CTCTGTCAACGTCGGTGGC-3′ and 5′-CTGTTGCTGCTGTCCT-3′, was used to amplify a 942-bp fragment spanning the coding region for the conserved KIX domain (nt 2778–3720; available from GenBank/EMBL/DDBJ under accession no. ) from adult genomic DNA with EcoRI and XbaI sites at the ends. Amplification with a second pair of oligonucleotides, 5′-CTCTGTCAACGTCGGTGGC-3′ and 5′-GCCTGCTGCCGTCTGTTGGC-3′, generated the same fragment with KpnI and EcoRV sites at the ends. Each fragment was sequentially inserted into the appropriate restriction sites of the vector pUAST, generating a construct with an inverted repeat of the 942-bp fragments in a tail-to-tail orientation under the control of the Gal4-dependent UAS promoter. The -element carrying the inverted repeats was introduced into the germline using standard protocols. Multiple independent lines of were isolated and tested. Expression from these transgenes using a larval salivary gland–specific GAL4 () showed a low penetrance of persistent larval salivary glands (27%; = 33), probably a result of inefficient RNAi processing of CBP (persistent larval salivary glands at 24 h APF is never seen in wild-type animals). The allele was used in all mutant stocks. SG-GAL4 was provided by A. Andres (University of Nevada, Las Vegas, Las Vegas, NV). For broad assessment of death competence, animals carrying a heat-shock inducible transgene (; provided by H. Steller, The Rockefeller University, New York, NY; ) were staged from egg lay and subjected to a single 30-min heat treatment during first, second, early third, or late third larval instars to induce expression and allowed to recover for 6 h before dissecting and staining salivary glands with antibodies to detect either DIAP1 or the cleaved active form of caspase-3. Salivary glands dissected from control larvae () did not exhibit signs of necrosis or apoptosis. For a more precise assessment of the change in death competence during the third larval instar, control and mutant animals ( or or ) were staged from the second-to-third instar larval molt in 4-h intervals, subjected to a single 30-min heat treatment, and allowed to recover for 6 h before dissecting salivary glands and staining. Transgenic expression of heat-inducible double-stranded RNA common to all mRNA isoforms () was used to disrupt function, as described previously (). Animals expressing -RNAi that were allowed to continue in their development displayed a fully penetrant block in puparium formation, demonstrating efficient inactivation of (; unpublished data). Fig. S1 shows the genomic location of the -element–induced mutations described in the text. Online supplemental material is available at .

The differences between cancer and normal cells in consumption and metabolism of basic nutrients have been considered a promising cancer therapy target for several decades (for review see ; ). The metabolism of glucose and glutamine has attracted a particular interest, as these molecules are the main nutrients consumed by cancer cells. Glucose metabolism has drawn much of the attention (for review see ) from the time when Otto Warburg found that some cancer cells consume more glucose than normal cells and proposed that a failure to make ATP by oxidative phosphorylation is the cause of cancer (). Many cancer cells indeed accumulate glucose derivatives faster than normal tissues, the ability that has been used successfully for cancer diagnosis and monitoring (for review see ). The more recent findings that glucose metabolism is linked to signaling pathways of cell growth and survival, and apoptosis in particular (for review see ; ) only reinforced the view of glucose metabolism as the Achilles heel of cancer cells. This view was also supported by the observation that depletion of glucose induced apoptosis in rodent cells that were transduced with , an oncogene that is abnormally expressed in many human cancers (), but not in the parental cell line (). This finding was remarkable as it suggested that the very processes that make cells cancerous can make cells dependent on glucose. However, the idea that restricting supply of glucose or imitating this restriction can kill cancer cells selectively by depleting ATP is not without its critics (for review see ). Indeed, in contrast to Warburg's proposal, both normal and cancer cells can produce ATP by oxidative phosphorylation, and the contribution of glycolysis to ATP production varies greatly among both normal and cancer cells. Some cancer cells can even survive in vitro without glucose at all if supplied with glutamine and nucleosides (; ; ). Metabolism of glutamine, the major nutrient consumed by cancer cells besides glucose, has also been considered as a target of cancer therapy, even though with a much lower intensity than metabolism of glucose. The rate of glutamine transport is higher in cancerous than in normal cells (), the activity of glutaminase, the first enzyme in glutamine metabolism, correlates with the growth rates of tumors (; ; ), and some cancer cell lines require glutamine or its metabolites for survival (for review see ; ). The ability to deplete glutamine is a therapeutically relevant component of asparaginase that has been used to treat acute lymphoblastic leukemia (). Antibiotics DON (6-diazo-5-oxo--norleucine) and acivicin, which are glutamine analogues, were very effective in animal models of cancer, perhaps due to the ability to inhibit nucleotide synthesis, but proved to be unacceptably toxic in clinical trials (for review see ). Surprisingly, why glutamine is required for cell viability, or, in other words, why depletion of glutamine would kill cells, is not entirely clear. On the one hand, glutamine is the most abundant amino acid in the body and is used by the cells in a variety of ways, including oxidation by the Krebs cycle to produce ATP (for review see ), providing nitrogen required for nucleotide synthesis, and being a precursor of glutathione, the major nonenzymatic cellular anti-oxidant (for review see ). However, glutamine is made by cells, which makes the requirement for exogenous glutamine appear unnecessary. Moreover, only a few percent of the consumed glutamine is used for macromolecular biosynthesis, while the rest is metabolized into molecules, such as lactate, that are released by the cells (for review see ). These observations raise the possibility that metabolism of glutamine not merely supports accelerated proliferation or an increased requirement for ATP, but is used for yet unrecognized pathways that are required for survival of cells and cancer cells in particular. This study was designed to find nutrients that are required for cells that express abnormally high amounts of MYC but not for their normal counterparts. Considering that metabolism can vary among species, and that human rather than animal tumors are of primary interest, we used human cells. To our surprise, we found that in contrast to rodent cells () glucose depletion killed normal human cells irrespective of the activity of ectopic MYC and did so not by inducing apoptosis. On the other hand, depletion of glutamine induced apoptosis that depended on activity of ectopic MYC, which led us to investigate the mechanisms of this death. Overall, our study raises again the concern that the requirement of normal cells for glucose and intraspecies differences in glucose metabolism need to be considered in evaluating depletion of this nutrient for therapeutic needs. This study also suggests that learning why cells require glutamine to maintain their viability may provide new approaches for killing cancer cells selectively. To test how an abnormally high expression of affects the requirement of human cells for nutrients, we used a well-characterized system in which MYC activity can be induced acutely. This system uses a fusion of MYC with a mutated hormone- binding domain of the estrogen receptor, ER-MYC (; ). ER-MYC is excluded from the nucleus, which makes the MYC moiety inactive transcriptionally, but translocates there upon binding a synthetic estrogen analogue 4-hydroxytamoxifen (OHT). We will refer to ER-MYC activated by OHT as active ER-MYC, and indicate active ER-MYC in the figures as MYC and inactive as MYC. This system allows studying immediate effects of MYC by minimizing the artifacts associated with clonal selection of cells that because of additional genetic changes avoided apoptosis induced by ectopic expression of . To further decrease the chances of clonal selection, we used retroviral transduction rather than transient transfection to generate populations of normal human fibroblasts IMR90 that expressed (I) or (I); the latter served to exclude effects of OHT unrelated to MYC activity. Adding OHT to the tissue culture medium caused translocation of ER-MYC to the nucleus (Fig. S1 A, available at ) and increased sensitivity of the cells to DNA damage (Fig. S1 B), which confirmed that the system functioned as expected (; ). The first part of our study reiterated the previously raised concern () that the knowledge about the relative requirement of normal and transformed cells for glucose is still insufficient to make a reliable judgment as to whether agents that mimic glucose deprivation can be effective in cancer therapy. In particular, our results demonstrated that dependence of cells on glucose and the effects of its depletion vary among cell types and can depend on the combination of oncogenes expressed in these cells. Our finding that a deficiency in glutamine killed most of the human cells that we tested depending on the activity of MYC raised the possibility that the requirement for this nutrient can be explored for learning how to kill cancer cells in which MYC activity is abnormally high. Testing this possibility will require knowing why and how glutamine depletion kills, and how MYC activity synergizes with the depletion of glutamine to induce apoptosis. Normal diploid human lung fibroblasts IMR-90 (American Type Culture Collection), normal diploid human foreskin fibroblasts (HFFs) (obtained from Donald Ganem, University of California at San Francisco, San Francisco, CA), and cell lines derived from them were maintained in DME (Invitrogen) without phenol red supplemented with 4.5 g/L -glucose, 2 mM -glutamine, 1 mM sodium pyruvate, 10% fetal bovine serum, and mixture of penicillin and streptomycin (100 g/ml each). Cells were used between passages 12 and 23. Human retinal epithelial RPE cells, immortalized with hTERT and transduced either with pMig-neo vector expressing c- (RPE-) or with the vector alone, were obtained from Dun Yang (University of California, San Francisco) and maintained in the DMEM media described above. Genetically defined immortalized kidney epithelial cells (HA1E) were obtained from Robert Weinberg (Whitehead Institute, Cambridge, MA). HA1E and HA1E-derived cells were grown in MEM-α medium (Invitrogen) supplemented with 10% fetal bovine serum and antibiotics as described above. Retroviral vectors pBABE-hygroER, expressing the estrogen receptor binding domain, and pBABE-hygroER-MYC expressing c- fused to the estrogen receptor, were a gift from William Tansey (Cold Spring Harbor Laboratory, Cold Spring Harbor, NY). pBABE-hygroER-MYC and pBABE-puro -ER plasmids were described previously (). cDNAs encoding caspase-9 Cys to Ser mutant (caspase 9 dominant negative, C9DN) (), crmA, Bcl-2, and human c- were cloned into a MarX-IV-puro retroviral gene transfer vector (). Retroviruses were produced and used to infect primary cells as described previously (). In the case when two vectors were transduced, cells were infected with supernatants containing two different viruses simultaneously. Cells were selected with appropriate selection agents and obtained cell clones were pooled. Cells were seeded at 1.2 × 10 cells per well into 6-well plates or 5 × 10 cells per well into 12-well plates and were allowed to attach for 7–12 h. ER, or the fusion of ER with , were activated by adding 4-hydroxytamoxifen (OHT) (Sigma-Aldrich) to the media to a final concentration 200 nM. Ethanol was used as a vehicle. Cells were washed at 24 h once with PBS, supplemented with fresh medium, and used for the experiments. Unless indicated, all media used for the experiments contained ethanol or OHT. To deplete glutamine or glucose, cells were cultured in glutamine or glucose-free DMEM that contained neither phenol red nor pyruvate (Invitrogen) unless indicated. When these nutrients had to be present, glucose or glutamine was added into the media to the final concentrations, 10 mM and 2 mM, respectively. To prepare media deficient in one of the essential amino acids, Earle's balanced salt solution was supplemented with 4.5 g/L glucose (Sigma-Aldrich), MEM vitamin solution (Invitrogen), 0.37 mM sodium bicarbonate (Invitrogen), 24.8 μM ferric nitrate (Sigma-Aldrich), 2 mM glutamine, and all essential amino acids, except one, at the concentrations of complete DMEM. Media for all experiments was supplemented with 10% dialyzed FBS prepared by dialyzing 50 ml of FBS against 2 liters of PBS using a membrane with a molecular cut-off of 3,500 D. PBS was changed every 12 h for 72 h. Cells were harvested by combining floating cells in the media with adherent cells detached with 0.05% trypsin/0.53 mM EDTA solution (Invitrogen). Cells were washed with cold PBS, resuspended in 10 mM Hepes, 140 mM NaCl, and 2.5 mM CaCl, pH 7.4, and incubated with 2 μg/ml of recombinant annexin V conjugated to FITC for 15 min and with 4 μg/ml of PI for 1 min and analyzed by Becton Dickinson LSR-II cell analyzer with FACSVantage DiVa software. For analyzing nuclear morphology, cells were resuspended in staining solution containing 4% parafomaldehyde, 50% glycerol, 0.1% Triton, and 2 μg/ml Hoechst 33342. To identify cells that failed to exclude Trypan Blue, the cells were collected and resuspended in 50 μl of PBS. The suspension was complemented with equal volume of 0.08% Trypan Blue solution and incubated on ice for 5 min. The Trypan Blue–containing cells were scored by light microscopy. To analyze DNA content, 5 × 10 cells were collected by trypsinization, washed with PBS, fixed with 70% ethanol (−20°C) and kept at −20°C for at least 4 h. After washing with PBS, cells were resuspended in PBS containing 80 μg/ml RNase and 30 μg/ml PI. After incubating for 30 min at room temperature, samples were analyzed by LSR-II cell analyzer with ModFit software. Cells were grown on coverslips, fixed with 4% paraformadehyde, stained with appropriate primary antibody and a secondary antibody labeled with AlexaFluor 488, and mounted using Prolong antifade medium (Invitrogen). The images were taken with an Axiophot microscope (Carl Zeiss MicroImaging, Inc.) with Plan-Apochromat 63×/1.40 oil objective lens, using a CCD camera (DFC420C; Leica) and Firecam acquisition software (Leica). The images were formatted using Adobe Photoshop to fit the size of the figures. All changes in brightness or contrast, if any done, were applied to the entire image. Cells grown in 12-well plates were collected and washed with cold PBS. ATP was extracted by adding 1 ml of boiling water to the cell pellet as described previously (). The extract was vortexed and cleared for 5 min at 12,000 at 4°C. 700 μl of the cleared supernatant were taken to measure ATP by Luciferin-Luciferase kit (Sigma-Aldrich) following the manufacturer's instructions. The pellet was resuspended in the remaining 300 μl of the supernatant and used to determine protein concentration. The concentration of protein was also determined in the supernatant used to determine ATP concentration to correct for any protein left after centrifugation of the extract. The amount of protein in both fractions was combined to obtain the total protein amount in the extract. The amount of reduced glutathione in cells was measured by flow cytometry as described previously (). In brief, cells grown in 12-well plates were collected, resuspended in 0.5 ml of PBS containing 1% FBS, and incubated with 40 μM monobromobimane (Biochemika) for 10 min at room temperature. After incubation, cells were moved on ice and the fluorescence at 485 nm was measured by flow cytometry (DiVa; Becton Dickinson). The intracellular concentrations of NADH and NAD were measured by enzymatic cycling assay as described previously (). In brief, cells grown in 6-well plates were collected and washed with cold PBS. NADH was extracted by adding 100 μl of 0.05 N NaOH and 0.5 mM cysteine and incubating cell suspension at 60°C for 10 min. NAD was extracted by adding 100 μl of 0.04 N HCl and 4 mM cysteine and incubating the suspension at room temperature for 15 min. NADH and NAD extracts were sonicated for 30 s and used in a cycling system consisting of lactic dehydrogenase and glutamic dehydrogenase to produce pyruvate in 2,000-fold yield in 30 min. Reaction was stopped by heating for 2 min at 100°C and the concentration of pyruvate was measured using lactic dehydrogenase reaction fluorimetrically. Protein was measured in the extracts by Bradford assay (Bio-Rad Laboratories). Cells were grown in 12-well plates with 1 ml of media in each well. At the indicated time points the media was collected and pelleted at 2,000 for 10 min at 4°C. The supernatant was snap-frozen in liquid nitrogen and stored at −80°C until the assay was performed. Lactate concentration was measured using lactate dehydrogenase reaction (). Metabolites were extracted from adherent cells by a procedure adopted from . In brief, cells were collected by trypsinization, pelleted, and resuspended in 1 ml of cold PBS. 50 μl of cell suspension was taken to measure protein concentration by BCA assay (Pierce Chemical Co.). The rest of the cells were pelleted and the pellets were frozen in liquid nitrogen. The mixture of methanol and chloroform (−20°C) in a 2:1 ratio (vol/vol; 360 μl/cell pellet) was added to the frozen cell pellet. The pellet–solvent mixture was allowed to thaw on ice and was then sonicated. The sonicated extracts were supplemented with internal standards, methionine sulfone and Pipes (pH 7.4), at 2 nmol each. To separate hydrophilic and lipophylic fractions, 120 μl of chloroform and 90 μl of water were added. Suspension was centrifuged at 14,000 for 20 min at 4°C. Equal volumes of the upper aqueous fraction were dried in a SpeedVac concentrator. The extracted metabolites were resuspended in 50 μl of water and stored at −80°C. Extracts were analyzed by capillary-electrophoresis mass spectrometry (CE-MS), according to the method devised by for the quantitation of anions in cellular extracts. Agilent CE system was interfaced to an Applied Biosystems QSTAR Pulsar I using Applied Biosystems' CE-MS interface. Electrophoresis was performed in positively coated 110 cm × 50 μm i.d. SMILE (+) capillary (Nacalay Tesque), using 50 mM ammonium acetate (pH 8.5) as electrolyte. Injection was done with 50 mbar overpressure for 30 s; separation was carried with a constant potential of −30 kV. The voltage of the electrosprayer was −4.5 kV (thus reducing the effective electrophoresis voltage to 25.5 kV), and sheat liquid 5 mM ammonium acetate in 50% methanol was delivered at a flow rate of 2 μl/min. The mass spectrometer was run in full MS mode, scanning 50–600 at 3 Hz. Identification was based on retention time and exact mass measurements after alignment of both axes with reference compounds, and confirmed by spiked standards. For semi-quantitation, peak areas where normalized to that of the internal standard Pipes and to the amount of protein measured by the BCA assay. All processing steps were performed with custom routines written in the Matlab environment (The Mathworks). Rabbit polyclonal antibodies to S6 and pS6 were obtained from Cell Signaling Technology and were used in 1:1,000 and 1:4,000 dilutions, respectively. Monoclonal antibody to p53, clone DO-1, was obtained from Calbiochem and used in 1:1,000 dilution. Anti-Fas activating antibody, clone CH11, was obtained from Upstate Biotechnology. Anti-pH2AX (Ser139) antibodies were from Upstate Biotechnology (1:1,000 dilution). The 9E10 antibody to Myc was obtained from the Cold Spring Harbor Laboratory Monoclonal Antibody Facility. Anti-actin antibodies were purchased from Sigma-Aldrich. Rapamycin was obtained from Biosource International. All other reagents were obtained from Sigma-Aldrich. Fig. S1, A and B document that ER-MYC is activated as expected; Fig. S1 C demonstrates that the MYC moiety of ER-MYC makes cells sensitive to glutamine depletion. Fig. S2 visualizes localization of phosphorylated histone H2AX in cells deprived of glutamine or treated with etoposide. Fig. S3 documents that lactate production remains largely unchanged after depletion of glutamine and is independent of ER-MYC activity. Fig. S4 compares the concentrations of glutathione and ATP measured biochemically or by mass spectrometry. Table S1 shows how concentrations of the measured metabolites change after glutamine depletion. Online supplemental material is available at .

Wiring of the intricate nervous system requires guided axonal growth to form specific neuronal connections. Motile growth cones at the tip of elongating axons sense spatiotemporally distributed guidance cues to steer through the complex environment to reach the correct targets. A variety of attractive and repulsive guidance molecules and their corresponding receptors have now been identified (; ). Recent studies also show that several classic morphogens, including the Hedgehog, Wingless/Wnt, and Bone Morphogenic Protein (BMP)/TGFβ families, play a key role in axon guidance (). During neural tube development, several members of the BMP/TGFβ family, such as BMP7, GDF7, and BMP6, are expressed in the dorsal roof plate, and it is believed that their main function is to control the induction and differentiation of dorsal interneurons (; ). Interestingly, BMP7 at the roof plate was recently shown to exhibit a chemorepellent function in the initial trajectory of commissural axons in the developing spinal cord (; ). However, a generalized role for BMP molecules in growth cone motility and guidance has not been established. Little is also known about the signaling mechanisms underlying BMP guidance effects. The canonical BMP–TGFβ pathway controls neuronal cell fate and involves activation of the Smad transcription factors (), but this long-term transcription-dependent signaling is unlikely to be involved in rapid growth cone responses to BMP7 (). Rather, the cytoskeletal dynamics controlling growth cone motility is expected to be targeted by BMP7 to generate specific guidance responses (; ). In this study, we used cultured embryonic spinal neurons and a well-established turning assay to investigate the cellular mechanisms underlying BMP7 guidance. We report that a BMP7 gradient elicits bidirectional turning responses of growth cones in culture. growth cones were initially attracted to a BMP7 gradient 4–8 h after plating but became repelled by BMP7 after overnight culture (20–24 h). We further show that LIM kinase (LIMK) and Slingshot (SSH) phosphatase mediate attraction and repulsion, respectively, through the phosphorylation regulation of their common target, actin-depolymerizing factor (ADF)/cofilin. Finally, we demonstrate that the switching of attraction to repulsion results from the emergence of Ca signals from a transient receptor potential (TRP) channel (TRPC1) in overnight cultures. The elevated amount of TRPC1 on the growth cone surface of overnight neurons allows Ca signaling that activates calcineurin (CaN) phosphatase and SSH, which is dominant over the LIMK activation, to induce repulsion. Thus, these data indicate that BMP7 acts through distinct LIMK and Ca–CaN–SSH pathways that converge on ADF/cofilin to locally regulate the actin cytoskeleton and control the direction of growth cone steering. To study the guidance effects of BMP molecules, we performed the turning assay on growth cones in culture (). We first examined the growth cone responses to the BMP7 gradient in 4–8-h neuronal cultures. A gradient created by pulsatile ejection of BMP7 at 5 μM (in pipette) induced marked attractive turning of the growth cone toward the pipette (), whereas the control solution (without BMP7) had no influence on the direction of growth cone extension (). We quantified the turning responses by measuring the turning angle and net extension of each growth cone over a 30-min assay period and summarized the overall responses in means (). BMP7 gradients induced attraction at the concentrations of 0.5 or 5 μM in pipette, whereas 0.05 μM BMP7 (in pipette) or the control exerted no directional influence. Based on the previous estimate and fluorescent imaging (; ), the effective concentrations of BMP7 at the growth cone for eliciting attraction are estimated to be ∼0.5 nM or 5 nM, respectively, which are within the physiological range of BMP molecules reported previously (; ). Although 5 μM BMP7 in pipette appears to be more effective in inducing attraction, no statistical significance was found between these two groups (P > 0.5; Mann-Whitney test). Identical gradients of BMP2, a member of the BMP2/4 subfamily, did not induce any turning responses (). Although the mean lengths of growth cone extension varied slightly among different groups (), no statistical difference was detected (P > 0.5; one-way analysis of variance [ANOVA]). Therefore, BMP7 gradients mainly influenced the direction, not the rate, of growth cone extension. neurons exhibit different responses to certain guidance cues when cultured for different times (; ). Therefore, we examined the turning response to BMP7 in overnight cultures (20–24 h). The same BMP7 gradient (5 μM in pipette) was found to induce marked repulsive turning in overnight neurons (), whereas the control exhibited no effect (), which was further confirmed by the mean turning angles (). Consistently, BMP2 gradients had no influence on overnight growth cones (), supporting the notion that BMP7 guidance effects unlikely result from nonspecific actions of BMP molecules. Finally, the net extension of these overnight growth cones was not significantly affected by BMP7 or BMP2 (P > 0.5; one-way ANOVA; ). To eliminate the possibility that different populations of neurons were responsible for distinct turning responses, we performed a double turning assay in which the same growth cone was examined at 4–8 h and again at 20–24 h after plating. To track the same growth cone, a special coverslip with etched grids and labels was used (). We found that the same growth cone responded to the BMP7 gradient (5 μM in pipette) with marked attraction at 5 h after plating but was repelled by the same BMP7 gradient at 22 h (). Of 20 growth cones subjected to the double turning assay, a great majority switched their turning responses from attraction to repulsion overnight, as indicated by the cumulative distribution and the means of turning angles (). No significant change in neurite extension was observed (P > 0.5; one-way ANOVA; ). These data demonstrate that the same growth cone exhibits bidirectional responses to BMP7 after different times in culture. We also examined the BMP7-induced turning responses in the presence of follistatin, an antagonist of BMP7, and noggin, an antagonist of the BMP2/4 subclass (; ). Bath presence of follistatin, not noggin, eliminated BMP7-induced attraction in 4–8-h neurons () and repulsion in overnight cultures (), supporting the notion that BMP7, not BMP2, selectively induces growth cone turning. BMP molecules typically act through BMP receptors I and II (BMPRI and BMPRII), and their kinase activity is responsible for Smad activation and transcription regulation (). neurons express a long form of BMPRII (), and we have verified it by RT-PCR analysis (Fig. S1 A, available at ). To evaluate the involvement of BMP receptors in BMP7 guidance, we overexpressed in spinal neurons a dominant-negative (DN) BMPRII, which lacks the cytoplasmic region (). Blastomere injection of mRNA encoding DN-BMPRII together with FITC-dextran as a cell tracer was used to express DN-BMPRII in neurons (). We found that both attractive and repulsive turning responses to BMP7 were abolished by the expression of DN-BMPRII but not the wild-type (WT) BMPRII (). Unlike other TGFβ type II receptors, BMPRII has a long carboxy tail after the kinase domain, which is not required for BMP activation of the Smad pathway (; ) but interacts with other signaling pathways (; ; ; ). When we overexpressed a short form of BMPRII lacking the carboxy tail in neurons, we found that short form BMPRII completely abolished the bidirectional turning responses to BMP7 (). These data indicate that the BMP7 effects on growth cone turning do not depend on the Smad pathway. This conclusion is consistent with the rapid turning responses to BMP7 gradients (in minutes) and is further supported by the finding that the general transcription inhibitor actinomycin D could not block BMP7-induced attraction in 4–8-h neurons () and repulsion in overnight cultures () when used at a concentration (2.5 μg/ml) that is shown to inhibit [H]uridine incorporation into RNA in cultures (; ). Together, these findings show that BMP7 guidance is mediated by BMP receptors, involves signaling independent of the canonical Smad transcription pathway, and requires the carboxy-terminal regions of BMPRII. The carboxy tail of BMPRII interacts with LIMK1 in mammals and (; ; ). LIMK1 is a key regulator of the actin cytoskeleton through its phosphorylation of ADF/cofilin at serine-3 for inactivation (). A peptide containing the serine-3 sequence of ADF/cofilin (32 amino acids; referred to as the S3 peptide) has been widely used as an effective competitive inhibitor of LIMK1 (; ; ). We synthesized the S3 peptide containing the specific ADF/cofilin (XAC) sequence of the LIMK phosphorylation site as a chimera with a penetratin sequence for uptake into cultured neurons (). Bath application of 10 μg/ml of the S3 peptide completely blocked growth cone attraction induced by the BMP7 gradient (), whereas 10 μg/ml of the control peptide (the reversed sequence of the XAC S3 peptide with the normal penetratin internalization sequence; RV-S3 peptide) did not affect BMP7-induced attraction (). The inhibition of BMP7-induced attraction by S3 peptides is clearly depicted by the cumulative distribution of turning angles of all growth cones examined (). A majority of growth cones exposed to BMP7 gradients alone or with RV-S3 in bath exhibited positive turning angles, resulting in a marked shift of the angle distribution to the positive territory. However, growth cones treated with S3 peptides did not show a directional preference in response to the BMP7 gradient, resulting in an angle distribution that overlaps with that of the control and centers at zero. The mean turning angles further highlight the effective inhibition of BMP7-induced attractive responses of growth cones by S3 treatment (). All of these treatments did not significantly affect the growth cone extension (P > 0.5; one-way ANOVA). We further tested the role of LIMK1 by expressing DN-LIMK1 (Asp 460 replaced with Asn), which is a catalytically dead mutant of LIMK1. Consistent with the S3 results, BMP7-induced growth cone attraction was completely abolished by DN-LIMK1 (), whereas WT-LIMK1 had no effect (). The overall turning responses of DN-LIMK1–expressing growth cones are similar to those treated with the S3 peptide, both exhibiting a mean turning angle of about zero (). On the other hand, blocking Ca influx by Ca-free solution or buffering intracellular Ca concentration ([Ca]) by the intracellular loading of BAPTA, a strong Ca chelator, did not generate any effect on BMP7-induced attraction. Similarly, inhibition of either Ca–calmodulin-dependent kinase II by 5 μM of the specific inhibitor KN93 or inhibition of PKA by 50 μM of a membrane-permeable cAMP antagonist, Rp-cAMPS, did not affect BMP7-induced attraction. Therefore, BMP7-induced attraction is mediated by LIMK and is independent of the Ca and cAMP signaling pathways. We first tested the involvement of LIMK1 in BMP7-induced repulsion in overnight cultures and found that bath application of S3 or RV-S3 peptides did not affect the repulsion (). Similarly, DN-LIMK1 expression did not affect BMP7-induced repulsion (). However, the overexpression of WT-LIMK1 was found to abolish the repulsion. Recent studies show that a family of phosphatases, named SSH, dephosphorylate ADF/ cofilin at serine-3 to oppose LIMK1 effects on ADF/cofilin activity (; ; ). If SSH mediated the repulsion, its action could have been attenuated by overexpressing LIMK1. To test this possibility, we overexpressed DN-SSH (a phosphatase-defective mutant of SSH in which the conserved Cys residue in the catalytic pocket is replaced by Ser; ) in neurons. Interestingly, growth cones of overnight neurons expressing DN-SSH responded to the BMP7 gradient with marked attraction instead of repulsion, whereas WT-SSH expression did not influence the repulsion (). Furthermore, the attraction in overnight neurons overexpressing DN-SSH was blocked by S3 peptides but not RV-S3 (). These findings suggest that BMP7 activates both LIMK1 and SSH pathways in overnight neurons, but the SSH pathway dominates to result in repulsion; however, when SSH is inhibited, localized LIMK1 activation takes in charge for attraction. Overall, these results show that BMP7 signals through the LIMK1 and SSH pathways to regulate opposite turning responses and that a balancing act of LIMK1 and SSH controls specific growth cone responses to the BMP7 gradient. LIMK1 and SSH oppositely control the activity of ADF/cofilin to regulate the actin cytoskeleton dynamics in neurite outgrowth, cell migration, and polarity (; ). Phosphorylation of serine-3 of ADF/cofilin inhibits its ability to sever filaments and enhance their subunit turnover. Therefore, BMP7-induced growth cone attraction and repulsion may involve localized ADF/cofilin-mediated stabilization and destabilization of filamentous actin, respectively. Our high resolution live imaging of the actin cytoskeleton in the growth cone appears to support this notion. In 4–8-h neurons, the BMP7 gradient rapidly induced local actin-based protrusions (both lamellipodia and filopodia) on the near side of the growth cone facing BMP7 before the turning of the growth cone shaft (Fig. S2, A and B; and Video 1, available at ). In overnight cultures, however, a BMP7 gradient elicited a rapid onset of asymmetric protrusion of the actin-based lamellipodia and filopodia on the distal side of the growth cone (with respect to the BMP7 pipette) followed by repulsion (Fig. S2, C and D). At the same time, limited protrusive activity on the BMP7 pipette side of the growth cone was observed, suggesting an inhibition of actin polymerization (Fig. S2 C). The inhibition of protrusive activity was most apparent in the middle of this time-lapse sequence (Video 2). We next performed quantitative immunofluorescence (IF) microscopy to examine the phosphorylation level of ADF/cofilin in growth cones exposed to bath application of BMP7 in 4–8-h and overnight cultures. neurons of 4–8 h exhibited a low level of phosphorylated XAC (p-XAC) in the growth cone (), which was markedly increased by 10 min of exposure to 5 nM BMP7. However, the level of p-XAC in the growth cone of overnight cultures was much higher than that of the untreated 4–8-h neurons, which was largely reduced by 5 nM BMP7 (). We quantified the intensities of p-XAC and total XAC of all of the growth cones examined and normalized the intensity values against that of the corresponding control group of 4–8-h cultures (). The total level of endogenous XAC was not changed by BMP7 and other treatments, but BMP7 induced a substantial increase in the level of p-XAC in 4–8-h neurons and a decrease in overnight neurons. Importantly, the BMP7-induced increase in p-XAC was abolished by the bath application of S3 peptides but not RV-S3, supporting the notion that ADF/cofilin is phosphorylated by LIMK1 in response to BMP7 in 4–8-h neurons. Expression of DN-SSH in the overnight neurons abolished the decrease in the level of p-XAC upon BMP7 exposure and, interestingly, further increased the overall p-XAC in the growth cone (). It should be noted that the lack of further elevation of the p-XAC level upon BMP7 exposure in the overnight neurons expressing DN SSH may reflect the limited sensitivity of quantitative IF in detecting small changes. Finally, the expression of short form BMPRII completely abolished BMP7-induced changes in p-XAC levels in both culture conditions (). These data support a model in which BMP7 signaling through BMPRII acts upon LIMK1 to phosphorylate ADF/cofilin in 4–8-h neurons but activates SSH to dephosphorylate ADF/cofilin in overnight neurons. BMP7 gradients likely regulate ADF/cofilin in a spatially restricted manner to induce distinct turning responses. Using 5-(4,6-dichlorotriazinyl)aminofluorescein (DTAF) as the volume label (), we examined the spatial distribution of p-XAC in response to a BMP7 gradient using ratiometric imaging. We found that a BMP7 gradient induced a preferential increase of p-XAC on the near side of the growth cone facing the BMP7 pipette in 6-h cultures, resulting in an asymmetry of p-XAC (). However, the same BMP7 gradient appears to induce a reversed asymmetry of p-XAC in overnight neurons (). Similar ratiometric imaging of the total XAC against the volume (DTAF) showed a relatively even distribution upon the BMP7 gradient in both 6-h and overnight cultures. Quantification of the asymmetry using the five-box analysis method () confirms that a BMP7 gradient induces opposite asymmetries of p-XAC in the 6-h and overnight growth cones, the former with a higher p-XAC level on the near side and the latter with a lower p-XAC level on the near side (). Consistently, the distribution of total XAC is uniform with and without BMP7 gradients. Together with the bath application data (), these findings support the notion that a BMP7 gradient induces a local elevation of ADF/cofilin phosphorylation at the near side for attraction in 4–8-h cultures but a local decrease of ADF/cofilin phosphorylation for repulsion in overnight neurons. To directly assess the functional role of ADF/cofilin phosphorylation in BMP7-induced growth cone guidance, we expressed various forms of XAC in neurons, including WT-XAC and a constitutively active (XAC3A; serine-3 replaced with alanine) or inactive (XAC3E; serine-3 replaced with glutamate) form. Expression of WT-XAC did not affect BMP7-induced attraction in 4–8-h neurons () and repulsion in overnight cultures (). However, the overexpression of either XAC3E or XAC3A totally abolished the turning responses to BMP7 both in 4–8-h () and overnight () cultures. No significant effects were observed on neurite extension among these groups of growth cones (P > 0.5; one-way ANOVA; ). Together with the imaging data, these findings show that BMP7 gradients elicit the asymmetric activation of LIMK1 and SSH that converge on ADF/cofilin phosphorylation to control the bidirectional responses of the growth cone. What caused the switching from LIMK1 activation (for attraction) to SSH activation (for repulsion) in overnight cultures? SSH could be activated by the Ca–calmodulin-dependent phosphatase CaN (), which is known to play a key role in Ca-dependent growth cone repulsion (). CaN inhibition by 10 nM of the specific inhibitors cyclosporin or deltamethrin not only abolished repulsion but converted it to LIMK-dependent attraction (). However, inhibition of phosphatase-1 by 4 nM tautomycin did not affect BMP7-induced repulsion (). CaN is activated by small Ca signals (; ); thus, we tested the involvement of Ca signals by either removing extracellular Ca using a Ca-free solution or intracellular loading of neurons with BAPTA to buffer the changes in [Ca]. We found that the Ca-free or BAPTA treatment caused the growth cones in overnight cultures to respond to BMP7 gradients with marked attraction (). This result confirms that Ca signaling through CaN and SHH mediates BMP7-induced repulsion. We next performed ratiometric Ca imaging and found that growth cones in overnight cultures responded to the BMP7 gradient with a small but substantial increase in [Ca], whereas the growth cones of 6-h neurons did not show any increase in [Ca] (). The Ca involvement in BMP7 effects on overnight neurons was further confirmed by the findings that the BMP7-induced decrease of p-XAC in overnight growth cones was completely abolished by CaN inhibition or by Ca-free or BAPTA treatment (). Thus, it is conceivable that the BMP7 gradient elicits small asymmetric Ca increases to activate CaN, which, in turn, acts on SSH to induce repulsion. Indeed, a small asymmetric elevation of [Ca] was observed in some overnight growth cones upon onset of the BMP7 gradient (Fig. S3, available at ). Recent studies show that TRP channels play an important role in growth cone guidance (; ; ). TRPC has been shown to interact with the carboxy-terminal tail domain of BMPRII (), and BMP7-induced small Ca elevation in overnight growth cones was eliminated by 2 μM of the TRP channel inhibitor SKF96365 (). To further examine the role of TRPC, we expressed either DN-TRPC1 (phenylalanine 561 replaced with alanine) or WT-TRPC1. We found that DN- but not WT-TRPC1 abolished the BMP7-induced reduction of p-XAC, indicating that TRPC1 is involved in the BMP7-induced dephosphorylation of XAC (). Therefore, Ca signals from TRPC1 are likely responsible for the activation of CaN and SSH to induce XAC dephosphorylation for repulsion. Because Ca–CaN–SSH signaling was only involved in overnight neurons for repulsive guidance by BMP7, the mechanism for generating Ca signals likely developed between 8 and 20 h in culture. Using quantitative IF of TRPC1, we found that TRPC1 was expressed at a much higher level on the surface of growth cones in overnight cultures compared with 4–8-h cultures (). This difference in the level of TRPC1 on the growth cone provides an attractive model in which TRPC expression may enable BMP7 to elicit Ca signaling for CaN–SSH activation to initiate growth cone repulsion in overnight neurons, whereas 4–8-h cultures responded to BMP7 with Ca-independent LIMK1-mediated attraction. In support of this model, the expression of DN-TRPC1 abrogated BMP7-induced repulsion and, importantly, switched it to attraction (). On the other hand, WT-TRPC1 expression did not affect the repulsion induced by BMP7 in overnight cultures. Furthermore, the overexpression of WT- but not DN-TRPC1 in 4–8-h neurons enabled consistent repulsion in the BMP7 gradient, which is a switch from the original attraction (). Therefore, these findings strongly support our hypothesis that the late emergence of TRPC1 on growth cones of overnight cultures couples BMP7 signaling to the Ca–CaN–SSH pathway for repulsion and demonstrate that TRPCs play a crucial role in regulating the balance of phosphorylation and dephosphorylation of ADF/cofilin to control distinct turning responses by the guidance signal BMP7 at different developmental stages. BMPs represent one of the three families of classic morphogens that have recently been shown to guide developing axons (). In this study, we used cultured spinal neurons and the turning assay to dissect the signaling cascades involved in BMP guidance. We find that BMP7 can act as a bidirectional guidance molecule to induce attractive and repulsive turning responses of growth cones. Importantly, our study has elucidated the downstream mechanisms that control the bidirectional responses of the growth cone (): BMP7 attracts growth cones through the LIMK pathway (4–8-h cultures) but repels growth cones through the CaN–SSH phosphatases pathway (overnight cultures). Our data further identify ADF/cofilin as the common downstream target of these two signaling pathways to locally regulate the actin cytoskeleton for distinct turning responses: local phosphorylation (inactivation) of ADF/cofilin by LIMK leads to attraction, whereas dephosphorylation (activation) of ADF/coflin by SSH results in repulsion. Finally, we find that the key trigger for the switching of BMP7-induced attraction to repulsion is the emerging TRP channels on the growth cone membrane of overnight neurons, which enable BMP7-induced Ca signals to activate CaN and, subsequently, SSH. Although BMP7 repulsion has been shown for commissural axons in vivo (; ), it remains to be examined whether BMP7-induced attraction plays a role in axon guidance in vivo. Our real-time PCR analysis of the expression of TRPC1 and BMPRII shows that TRPC1 emerges later than BMPRII, and it correlates with the ventral projection of commissural axons (Fig. S4, available at ). Therefore, the observed delayed emergence of TRPC1 in cultured neurons is unlikely to be an artifact. Because ADF/cofilin regulation by LIMK1 is important for neurite outgrowth (; ), BMP7 signaling through LIMK1 may play a role in commissural axon initiation and outgrowth in the early stage of development. The subsequent appearance of TRPC1 would then enable the commissural repulsion by BMP7 from the roof plate and, at the same time, the attraction by netrin-1 from the floor plate. Indeed, TRPC involvement in netrin-1–induced attraction was observed in overnight spinal neurons (; ), coinciding with the increased TRPC1 observed in this study. However, netrin-1–induced attraction appears to involve both TRPC and L-type Ca channels to generate a relatively large Ca signal (; ). Therefore, although TRPC is involved in both cases, BMP7 and netrin-1 could signal through distinct Ca pathways to generate different responses. Finally, BMP7-induced attraction might also play a potential role in the initial guidance of commissural axons because BMP7 expression was also observed in a second location ventral to the roof plate (). Nevertheless, our findings from the cultured cells have illustrated an intriguing and complex pattern of signal transduction elicited by BMP molecules that involves signal divergence and convergence. The remarkable ability of TRPC to tilt the balance to favor SSH-mediated repulsion indicates a novel role for TRP channels in influencing specific growth cone responses and further substantiates the importance of TRPC in guidance signaling. BMP signaling through BMPRII and BMPRI is linked to the Smad activation for transcription regulation, leading to neuronal differentiation (; ). We present evidence that the guidance effects are independent of the Smad pathway and selectively associated with BMPRII. In particular, BMP7 guidance was abolished by the short form BMPRII lacking the carboxy tail, which is not required for BMP activation of the Smad pathway (; ). It has been shown that LIMK1 interacts with this tail domain of BMPRII directly, and this interaction is required for its activation induced by BMPs (; ; ). LIMK1 regulates the actin cytoskeleton through its phosphorylation and inactivation of ADF/cofilin and has been implicated in neuronal development, including dendritogenesis, synaptic stability, and growth cone motility (). Our data indicate that LIMK1 mediates growth cone attraction induced by the BMP7 gradient in the 4–8-h neurons. This conclusion is based on the results from an extensive set of experiments involving pharmacological and molecular manipulations of LIMK in conjunction with quantitative fluorescent imaging of ADF/cofilin phosphorylation and turning assays. Our live imaging of the actin cytoskeleton in 4–8-h growth cones also indicates a locally enhanced protrusion of actin-based lamellipodia and filopodia upon the onset of the BMP7 gradient. Together, these findings support a model in which the BMP7 gradient induces growth cone attraction by eliciting asymmetric actin polymerization/stabilization through the spatial regulation of ADF/cofilin activity via LIMK1. Exactly how BMP7 activates LIMK1 is still unclear. Although interaction with BMPRII is required for LIMK1 activation induced by BMP7, LIMK1 can be activated by the Rho GTPase effectors Rho kinase (Rho-associated coil-containing protein kinase [ROCK]) and p21-activated kinases (). Considering the crucial role for Rho GTPases in regulating the actin cytoskeleton, it is plausible that Rho GTPases may participate in and contribute to BMP7 signaling during growth cone attraction. Furthermore, their linkage to virtually all receptors that mediate growth cone turning () suggests that attractive guidance cues might all have a final common target in ADF/cofilin, which serves to integrate multiple environmental signals into a single growth cone response. Our data indicate that BMP7-induced repulsion in overnight neurons is mediated by the SSH phosphatase, which is activated by Ca signaling through CaN. This conclusion is again supported by a large set of experiments involving Ca imaging and pharmacological and molecular manipulations of Ca, CaN, and SSH. Importantly, the inhibition of CaN or SSH by pharmacological and molecular manipulations not only blocked BMP7-induced repulsion but consistently switched it to attraction in the overnight neurons. Furthermore, blockade of Ca signaling also converted the repulsion to attraction in overnight cultures (). These results not only indicate a key role for Ca–CaN–SSH signaling in BMP7-induced repulsion but also demonstrate that LIMK1 can be activated in overnight neurons, although activated SSH predominates. SSH dephosphorylates ADF/cofilin at serine-3 to reactivate its actin-severing activity, and our quantitative IF shows that BMP7 induces ADF/cofilin dephosphorylation in overnight neurons, supporting the notion that BMP7 induces repulsion through the local SSH-dependent dephosphorylation of ADF/cofilin. It is notable that neurons from the overnight cultures exhibit a higher level of phospho-ADF/cofilin than those in 4–8-h cultures. Although the underlying mechanisms are unknown, such a difference in the phospho-ADF/cofilin level may reflect a difference in the actin dynamics and/or architecture in these neurons. For example, dynamic actin cytoskeleton in motile growth cones of 4–8-h neurons may undergo rapid turnover, which may involve the ADF/cofilin-severing activity (). Alternatively, there are factors other than phosphorylation that affect ADF/cofilin activity, and these may also change during development to maintain actin dynamics. Among these factors are levels of phosphatidylinositol-4-phosphate and phosphatidylinositol-4,5-bis phosphate, pH, and tropomyosin (). The key molecule triggering the switching of attraction to repulsion appears to be TRPC1 channels. TRPC is involved in Ca-dependent guidance by netrin-1, brain-derived neurotrophic factor, and myelin-associated glycoprotein in vitro and in vivo (; ; ). How these guidance molecules activate TRPCs remains unclear. A recent study found that TRPC can interact with the carboxy-terminal tail domain of BMPRII, suggesting a potential role of TRPC in the BMP7 signaling pathway (). The elimination of BMP-induced repulsion and XAC dephosphorylation by the short form BMPRII suggests the importance of the carboxy tail in repulsive signaling. Future experiments are clearly required to dissect the molecular mechanisms of how BMPRII interacts with and activates TRPCs. Nonetheless, our findings indicate that the switching of BMP7-induced attraction to repulsion is a result of an elevated level of TRPC1 on the growth cone surface of overnight cultures. Our model for BMP-induced bidirectional turning responses is as follows (): at the early stage of culture (4–8 h), BMP7 acts through BMPRII to mainly activate LIMK to asymmetrically phosphorylate ADF/cofilin for attraction. By 20–24 h in culture, a substantial amount of TRPC1 emerges on the growth cone surface, and BMPRII-TRPC coupling allows BMP7 to elicit Ca signaling to act on CaN and SSH for repulsion. Whereas LIMK1 can still be activated by BMP7 in overnight cultures, the Ca–CaN–SSH activation predominates for repulsion. This dominant effect could be caused by the ability of active SSH to dephosphorylate and inactivate LIMK1 () in addition to activating ADF/cofilin. In support of this possibility, the overexpression of WT-SSH was found to attenuate BMP7-induced LIMK-dependent attraction in 4–8-h cultures (Fig. S5, available at ). Thus, this model indicates that a balancing act of LIMK1 and CaN–SSH on ADF/cofilin phosphorylation controls specific turning responses. Two crucial sets of evidence support the aforementioned model. First, in the overnight cultures, inhibition of TRPC1 by expressing DN-TRPC1 not only blocked the repulsion but also switched it back to LIMK1-dependent attraction, similar to the switching from the inhibition of Ca signals, CaN, or SSH. Second, in the 4–8-h neurons that exhibit little TRPC1, the overexpression of WT-TRPC1 enabled the growth cones to respond to BMP7 with repulsion, confirming that TRPC1 expression is the key for BMP7-induced repulsion. ADF/cofilin family proteins are key regulators of the actin cytoskeleton and can increase the rate of dissociation of ADP-actin from the pointed end of actin filaments and generate new filament ends through severing actin filaments (; ). Furthermore, the severing and enhanced depolymerization are counteracted by an enhanced stabilization of actin at a high ADF/cofilin to actin ratio (), providing for exquisite spatial and temporal control of actin dynamics. Previous studies have shown that ADF/cofilin is involved in regulating growth cone motility, morphology, and neurite extension in response to extracellular signaling molecules, including brain-derived neurotrophic factor, semaphorin 3A, and myelin-associated inhibitors (; ; ; ; ). However, our findings here indicate a detailed molecular cascade that links extracellular guidance signals to intracellular cytoskeletal dynamics through phosphoregulation of ADF/cofilin activity. Furthermore, the inhibition of both attraction and repulsion by mutating the serine-3 of XAC to either alanine (XAC3A) or glutamate (XAC3E) indicates that spatial phosphoregulation of ADF/cofilin activity, not just the activity itself, is key for directional responses of growth cones to BMP7 gradients. The global presence of constitutive active (3A) or DN (3E) ADF/cofilin is likely to override any local regulation of ADF/cofilin activity elicited by BMP7 gradients, thus abolishing both turning responses. At this moment, we do not know whether local synthesis of ADF/cofilin plays a role in BMP7 guidance. Cofilin is one of the molecules whose mRNA has been detected in growth cones (), and local synthesis of cofilin appears to be involved in Slit-2–induced growth cone collapse. BMP molecules could signal through TGFβ1-activated tyrosine kinase 1 (a MAPK kinase kinase) and MAPK (; ), which could regulate local protein synthesis. Nevertheless, even locally synthesized ADF/cofilin will be subjected to the same phosphoregulation for its effects on actin dynamics. In summary, our study has provided important mechanistic insights into the BMP7 guidance of nerve growth cones. Specifically, our data present evidence that two distinct signaling pathways, LIMK1 and Ca–CaN–SSH, elicited by the BMP7 gradients control attractive and repulsive growth cone responses, respectively, through a balancing act on the phosphorylation and dephosphorylation of ADF/cofilin to regulate the actin motility. A similar balancing act of kinase and phosphatase in controlling growth cone steering has also been observed for Ca-dependent bidirectional turning (). It is conceivable that spatial regulation of the balance of phosphorylation and dephosphorylation of molecules involved in motility may represent a common scheme for directional responses of growth cones (). Loss of LIMK1 has been linked to William's syndrome, a disorder characterized by visuospatial defects and mild mental retardation (). Because the loss of LIMK1 has been linked to synaptic deficits (), it would be interesting to see whether defects in LIMK1, SSH, or ADF/cofilin may contribute to the wiring abnormality and mental retardation in William's syndrome and other neurological diseases. Blastomere injections of mRNA molecules encoding different proteins into embryo were performed as described previously (; ). The DNA constructs used for mRNA preparation were provided by the following laboratories: WT and DN BMPRII (provided by C. Wright, Vanderbilt University, Nashville, TN), human short form of BMPRII (provided by J. Massague, Memorial Sloan-Kettering Cancer Center, New York, NY), human WT LIMK1 and its mutant forms (provided by G. Bokoch, The Scripps Research Institute, La Jolla, CA), SSH and its mutant forms (provided by H. Abe, Chiba University, Chiba, Japan), XAC or its mutant forms (), and human TRPC1 or its mutant forms (). In brief, mRNA of each construct was prepared using the mMESSAGE mMACHINE kit (Ambion), and 2–3 ng was microinjected into one blastomere of embryos at the one- or two-cell stage together with 10 mg/ml of fixable FITC-dextran as the marker. Injected embryos were screened for their fluorescence and used later for biochemical assays or cell cultures. Embryonic spinal neurons isolated from stage 20–22 embryos were cultured on glass coverslips coated with poly--lysine and laminin as described previously (). The cultures were kept at 20–22°C in a serum-free medium (SFM) containing 50% (vol/vol) Leibovitz L-15 medium (Invitrogen), 50% (vol/vol) Ringer's solution (115 mM KCl, 2 mM CaCl, 2.6 mM KCl, and 10 mM Hepes, pH 7.4), and 1% (wt/vol) BSA (Sigma-Aldrich). Growth cone turning induced by BMP7 or BMP2 gradients was performed at room temperature according to the method described previously (), except assays were performed in a modified Ringer's solution (). Microscopic gradients of chemicals were produced by the pipette application method described previously (; ). A standard pressure pulse of 3 pounds per square inch was applied to a glass pipette (1-μm opening) at a frequency of 2 Hz with a pulse duration of 20 ms. The 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. With these settings, the growth cone is estimated to receive ∼1/1,000th of the concentration of the tested molecule in the pipette and an ∼10% relative concentration difference across the surface (; ). We typically used a 20× NA 0.45 dry objective for microscopy and imaging during the turning assay. The digital images of the growth cone at the onset and end of the 30-min period were acquired by a CCD camera (C2400; Hamamatsu) and saved onto the hard drive. The images were then 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 experiment onset and at the end of 30-min exposure to the gradient. Neurite extension was quantified by measuring the entire trajectory of net neurite growth over the 30-min period. Only growth cones extending 5 μm or more were scored for turning responses. We used the nonparametric Mann-Whitney test to analyze turning angles because they do not follow a normal distribution. Recombinant human BMP7 and BMP2 proteins and their antagonists follistatin (2 μg/ml) and noggin (500 ng/ml) were purchased from R&D Systems. Most of the chemical agonists and antagonists were purchased from Calbiochem. Typically, different agonists and antagonists were added to the bath medium 20 min before the onset of turning assays. The S3 and RV-S3 peptides were synthesized by GenScript containing the amino-terminal unique phosphorylation site of XAC (MASGVMVSDDVVKVFN) and were synthesized in reverse order for the RV-S3 peptide as well as a penetratin sequence (RQIKIWFQNRRMKWKK) that allowed the peptide to be internalized from the cell culture medium (). BAPTA was used to buffer any changes in intracellular Ca concentrations. In brief, cultures were incubated with 1 μM BAPTA-acetoxymethyl ester (Sigma-Aldrich) for 30 min, rinsed three times, and incubated with fresh SFM for 90 min before turning assays. To prevent Ca influx, a Ca-free solution based on Ringer's saline (115 mM KCl, 2.6 mM KCl, 2 mM MgCl, 1 mM EGTA, and 10 mM Hepes, pH 7.4) was used to replace the culture medium. neurons were plated on poly--lysine– and laminin-coated 18 × 18-mm square coverslips with etched grids and labels (Bellco Biotechnology). The cells were cultured in SFM, and the first round of turning assays was performed in modified Ringer's solution at 4–8 h after plating as described in the previous section. The location of the cell was noted by the etched grids and labels. Afterward, the Ringer's solution was replaced by fresh SFM containing 100 U/ml penicillin and 0.1 mg/ml streptomycin, and the cells were cultured overnight for the second round of turning assays. The same neuron and its growth cone were identified by its location and verified by its morphology using the image acquired before. Quantification of the turning angles and lengths of extension was performed as described in the previous section. cultures with different treatments were exposed to 5 nM of bath BMP7 or to a control solution for 10 min. The cells were then rapidly fixed with 4% PFA and 0.25% glutaraldehyde in a cacodylate buffer (0.1 M sodium cacodylate and 0.1 M sucrose, pH 7.4) for 30 min, washed three times in 100% Ringer's saline, and permeabilized with 0.1% Triton X-100 for 10 min. The cells were first incubated with 1% donkey serum to block nonspecific binding sites for 1 h at room temperature. The cells were then incubated with a polyclonal antibody against phospho-ADF/cofilin (Santa Cruz Biotechnology, Inc.) or total XAC (). Both primary antibodies have been verified by Western blots for their reactivity to neuronal tissues (Fig. S1 B). Afterward, the cells were labeled by Cy3-conjugated goat anti–rabbit secondary antibodies (The Jackson Laboratory). For mRNA-injected embryos, cells identified by their FITC-dextran fluorescence were used for quantification. Fluorescent imaging was performed on an inverted microscope (TE2000; Nikon) using a 60× NA 1.4 plan Apo objective with identical settings between the control and exposed groups. Digital images were acquired with a CCD camera (SensiCam QE; Cooke Scientific) through the use of IPLab software (BD Biosciences). Background-subtracted images were analyzed by creating a region of interest that circumscribed the growth cone using ImageJ software (National Institutes of Health [NIH]). For each growth cone, the region of interest intensity was normalized to the mean from the parallel control. Data for each condition were obtained from at least two separate batches of cultures on different days. For quantitative imaging of TRPC1, cultures of different ages were fixed for 30 min and labeled (without permeabilization) with an antibody that recognizes the extracellular domain of TRPC1 () followed by incubation with AlexaFluor488 donkey anti–rabbit IgG secondary antibodies (Invitrogen). Fluorescent imaging was performed as described in the previous paragraph. To analyze the asymmetry of p-XAC or XAC in the growth cone, we placed five equal-size boxes (typically 10 × 10 but adjusted to fit different growth cones) across the growth cone according to a previously described method (). All of the quantitative measurements were performed on the original 16-bit images using ImageJ software (NIH). neurons were loaded with both fluo-4 and fura red using their acetoxymethyl forms (4 μM and 2 μM, respectively) for 30 min followed by three washes and 30-min recovery in the culture medium. Before the imaging, the bath medium was replaced with the modified Ringer's solution. The cells were imaged on a confocal microscope (C1; Nikon) using the 488-nm excitation of an argon laser and both photon multiplier tubes to simultaneously collect both the green (fluo-4) and red (fura red) fluorescence. The confocal microscope was equipped with an inverted microscope (TE300; Nikon), and a 60× NA 1.4 oil immersion objective (Nikon) was used. Time-lapse recording was performed on the growth cone at a rate of one pair every 10 s. Typically, 10 pairs of images were acquired before the onset of the BMP7 gradient followed by 30 pairs (5 min) of image acquisition. Background-subtracted ratiometric calculation was performed between fluo-4 and fura red images using ImageJ software (NIH). Measurements of the ratio were performed in a region of interest placed at the center of the growth cone (). Fig. S1 shows PCR analysis of BMPRII expression and Western blotting of ADF/cofilin in neural tube tissues. Fig. S2 shows fluorescence live imaging of GFP–γ-actin and mRFP-EB3 in growth cones during attraction and repulsion. Fig. S3 shows asymmetric Ca elevation in a growth cone subjected to the BMP7 gradient in overnight neurons. Fig. S4 shows real-time PCR analysis of the expression profile of TRPC1 and BMPRII in the neural tube. Fig. S5 shows the attenuation of BMP7-induced attraction by overexpression of WT SSH in 4–8-h cultures. Videos 1 and 2 show the dynamic distributions of the actin and microtubule cytoskeleton in a growth cone during attraction (Video 1) and during repulsion (Video 2) to a BMP7 gradient. Supplemental text provides the methods used for the data in supplemental figures. Online supplemental material is available at .

Skeletogenesis occurs through two mechanisms: intramembranous ossification and endochondral ossification (). During intramembranous ossification, which occurs in some cranial bones and the clavicle, mesenchymal cells differentiate directly into osteoblasts without a chondrocyte intermediary step. In endochondral bone formation, mesenchymal cells first condense to shape the future skeletal elements. The mesenchymal cells within these condensations differentiate into chondrocytes, which express α1(II) collagen, whereas the α1(I) collagen–expressing undifferentiated mesenchymal cells at the periphery of the condensations form the perichondrium. After the formation of the cartilaginous templates, chondrocytes further differentiate into prehypertrophic chondrocytes and then into hypertrophic chondrocytes, which express α1(X) collagen. Terminally differentiated hypertrophic chondrocytes also express matrix metalloproteinase 13 (Mmp13; also called collagenase 3). This hypertrophic chondrocyte differentiation step does not occur in skeletal elements that are destined to be permanent cartilage. Finally, the osteoblasts in the perichondrium invade the hypertrophic chondrocyte area and deposit bone matrix. Skeletogenesis is tightly controlled by transcription factors. Among them, Runx2, a transcription factor containing a DNA binding domain, is a key gene for osteoblast differentiation (). In addition, Runx2 plays multiple and opposite roles during chondrogenesis (; ; ). On one hand, Runx2 is transiently expressed in prehypertrophic chondrocytes, where it, in conjunction with Runx3 for some skeletal elements, is necessary to initiate chondrocyte hypertrophy (), the ultimate event of chondrogenesis. Accordingly, Runx2 constitutive expression in nonhypertrophic chondrocytes leads to premature and ectopic chondrocyte hypertrophy and, thereby, bone formation (). On the other hand, through its perichondrial expression, Runx2 prevents chondrocyte maturation (). To date, histone deacetylase 4 (HDAC4) is the only protein known to regulate Runx2 function during chondrogenesis (). This contrasts sharply with the larger number of molecules regulating Runx2 function during osteogenesis (; ; ) and suggests that additional regulators of Runx2's ability to favor chondrocyte hypertrophy may exist. Filamins are a family of large cytoplasmic actin binding proteins that cross-link filamentous actin into a three-dimensional network (). The family consists of three members in mammals, Filamin A, B, and C (Flna, Flnb, and Flnc). All three members share well-conserved actin binding domains and a rodlike domain of 24 repeats, which is interrupted by two poorly conserved hinge domains (). All three members are widely expressed, although Flnc is more restricted to skeletal and cardiac muscle (). Studies in different model organisms indicate that Filamins may interact with >30 proteins, thus implicating them in many cellular functions ranging from mechanical stability, to cell–cell and cell–matrix interactions, to integrators of signal transductions (; ; ). Genetic evidence indicates that Filamins are essential for the development of multiple organs in human, including skeleton (; ). For instance, mutations in are found in five human skeletal disorders: spondylocarpotarsal synostosis syndrome (SSS), Boomerang dysplasia, atelosteogenesis I, atelosteogenesis III, and Larsen syndrome (). The recessive SSS, characterized by short stature and fusions of vertebral, carpal, and tarsal bones, is caused by stop codon mutations within rod domain repeats, whereas the other dominant disorders are associated with missense mutations or small in-frame deletions. Patients with Boomerang dysplasia () have underossification of the limb bones and vertebrae. Atelosteogenesis I and III () patients have vertebral abnormalities, disharmonious skeletal maturation, hypoplastic long bones, and joint dislocations. Patients with Larsen syndrome () have multiple joint dislocations, craniofacial abnormalities, and accessory carpal bones. The molecular mechanisms whereby mutations in result in skeletal abnormalities remain unknown. To study how mutations could lead to skeletal disorders, we generated mice deficient in full-length Flnb protein and found that the mutant mice display a phenotype similar to the recessive SSS. Cellular and histological analysis showed that these defects are due to premature and/or ectopic chondrocyte hypertrophy in skeletal elements that should be permanent cartilage. Molecular studies showed that these abnormalities are due to an increase in Runx2 activity. Here, we also show that Flnb does not interact with either Runx2 or HDAC4 but can bind to Smad3, a protein expressed in prehypertrophic chondrocytes and able to interact with Runx2 (; ). These results demonstrate for the first time that Flnb can influence the function of a transcription factor, identify a new mechanism whereby Runx2's function during chondrogenesis is regulated, and provide a molecular basis for SSS. The gene trap embryonic stem (ES) cell line contains a β-galactosidase neomycin insertion within the 20th intron of (; see Materials and methods). The 20th exon of splices into the insertion to generate a fusion transcript, but it does not produce any wild-type transcript ( and Fig. S1, available at ). The mutant transcript is predicted to encode a truncated Flnb protein of 1,624 of the full-length 2,602 amino acids and to be fused with (). Intercrosses of heterozygotes in the 129/Ola-C57BL/6 mixed genetic background generate homozygous mice at Mendelian ratios at embryonic day (E) 18.5 (). However, at postnatal day (P) 5 or P21, the number of homozygous mutants is 50 and 90% decreased, respectively. The few surviving homozygous mutants are runted and have abnormal postures, with x-ray analysis showing severe kyphosis (). All these features are reminiscent of the clinical manifestation of SSS (). Staining of skeletal preparations with Alizarin red for mineralized ECM and Alcian blue for unmineralized cartilaginous ECM reveals multiple skeletal defects in the mutants. For instance, similar to what is seen in SSS patients, homozygous mice have both cervical vertebrae and carpal bone fusions. Indeed, although in wild-type P30 mice, unmineralized cartilaginous ECM separates the seven cervical vertebrae, the intervertebral ECM of the mutant mice was stained by Alizarin red, indicating that this ECM was mineralized, causing vertebral fusions (). In P0 mutant mice, mineralization of the cartilage ECM, although detectable, is less extensive, suggesting that the fusion of cervical vertebrae develops progressively. The mineralization of intervertebral ECM was first detectable at E17.5, whereas no difference in cervical vertebrae staining was seen between wild type and mutant at E16.5, suggesting that the cartilage develops properly in the homozygous mutants but becomes mineralized (). In addition, the cartilage that would normally separate carpal bones in the wild-type mice is ectopically mineralized in the mutant mice. At P0, blue staining cartilaginous ECM separates the carpal bones, whereas by P15, this ECM is becoming mineralized, leading to fusions of carpal bones (). Homozygous mutant mice also display abnormal ossification in the chondrocostal cartilage of the rib cage and sterna. In wild-type mice at all stages examined, chondrocostal cartilage and cartilage separating the sternebrae never ossify and are stained by Alcian blue. In contrast, in homozygous mutant mice, although the chondrocostal cartilage is stained with Alcian blue at P0, it stains progressively with Alizarin red at P15 and P30, indicating that this ECM is becoming prematurely mineralized (). Similarly, cartilaginous areas between sternebrae become mineralized in the mutant mice. Collectively, these ectopic ECM mineralization processes observed in cervical, carpal, and rib cage suggest that one of the Flnb functions is to regulate chondrocyte differentiation. To determine if this ectopic mineralization was caused by ectopic chondrocyte hypertrophy, we analyzed the expression of molecular markers of proliferating, prehypertrophic, and hypertrophic chondrocytes in ribs by in situ hybridization. We used probes to , a gene expressed in proliferating and prehypertrophic chondrocytes; () and (), two prehypertrophic chondrocyte markers; , a marker of hypertrophic chondrocytes; , a terminally differentiated hypertrophic marker; and , which is expressed in osteoblasts and mesenchymal cells in the bone collar. At P0, although is expressed in the cartilage area between sternebrae 3/4 of wild-type mice, its expression cannot be detected in mutant mice (). cells express and (), indicating that the chondrocytes have differentiated into hypertrophic chondrocytes. Similarly, is expressed in the cartilaginous area between sternebrae 2/3 of P0 in wild-type mice, but its expression at this location is diminished in mutant mice. Instead, these cells express , , and . These results indicate that the chondrocytes within that area are undergoing hypertrophy in the homozygous mutant mice at P0 (). At P5, is no longer expressed in the cartilage area between sternebrae 2/3 and 3/4, but osteoblasts are present, as demonstrated by the expression of (). This observation explains the presence of ectopic bone formation in the sterna. chondrocostal cartilage express but not , indicating that chondrocytes in the chondrocostal cartilage are not hypertrophic (). However, 5 d later, expression can be detected in the chondrocostal cartilage of the homozygous mutant mice, suggesting that hypertrophic chondrocyte differentiation has occurred (). chondrocostal cartilage, a cellular abnormality reminiscent of what was seen in mice overexpressing Runx2 in chondrocytes ( transgenic mice; ). The similarity between mutant and phenotypes suggests that Flnb and Runx2 may interact genetically, with Flnb acting as a negative regulator of Runx2 activity. To test this hypothesis, we generated mutant mice on a haploinsufficiency background and examined them for ectopic mineralization. ; mice have completely normal cervical vertebrae (), and the sterna phenotype observed in mice is partially rescued (). ; mice (). The rescue, complete or partial, is always fully penetrant (). These results provide genetic evidence that Flnb and Runx2 interact and suggest that Flnb regulates chondrocyte hypertrophy by inhibiting Runx2 activity. How does a cytoplasmic molecule like Flnb inhibit the function of a nuclear transcription factor like Runx2? We were unable to detect mislocalization of Runx2 and HDAC4 proteins in the mutants or to identify biochemical interactions between Flnb and Runx2 or HDAC4 (; and Fig. S2, available at ). Although negative and therefore to be interpreted cautiously, these observations suggest that Flnb must interact with another protein. Filamin A, another member of the Filamin protein family, has been shown to bind CBFβ, a nuclear factor that can interact with Runx proteins. Through this interaction, Filamin A prevents CBFβ from entering the nucleus to activate the Runx1 transcription factor (). Therefore, we examined whether Flnb also binds CBFβ. We failed, however, to detect any interaction between Flnb and CBFβ (), nor could we see abnormal intracellular localization of CBFβ in chondrocytes (Fig. S2). Thus, the Runx2 interaction with Flnb appears to be CBFβ independent under the conditions of these assays. Deletion of a phosphorylation site on Smad3 causes ectopic chondrocyte hypertrophy, suggesting that TGF-β–Smad3 signals repress chondrocyte differentiation (; ). Because phosphorylated Smad3 has been shown to recruit HDAC4 to inhibit Runx2 activity in osteoblasts by forming a HDAC4–Smad3–Runx2 complex (), we tested whether Flnb could repress Runx2 activity by interacting with Smad3. Indeed, three lines of evidence support the notion that a direct interaction between Flnb and Smad3 exists. First, Smad3 can be immunoprecipitated with Flnb in an in vitro overexpression assay (). Notably, compared with Smad3, Smad2 and -5 show no detectable interaction with Flnb (), although an interaction with Smad4 and -6 is also detected (Fig. S2). Second, endogenous Smad3 protein can be immunoprecipitated by tagged Flnb protein (). Third, Flnb directly interacts with Smad3 in a pull-down assay (). Because the phosphorylation of Smad3 is important in chondrocyte differentiation, we further examined the amount of active Smad3 in mutant versus wild type. The amount of phosphorylated Smad3 is significantly increased in the rib chondrocytes of mutants by Western blot analysis (), although this could not be readily detected by immunostaining (Fig. S2). Meanwhile, the total amount of Smad3 is not changed in the mutants (). Thus, Flnb may regulate Smad3 to ensure the accumulation of an appropriate amount of activated Smad3 in the nucleus (). An excessive amount of phosphorylated Smad3 may titrate out HDAC4, leading to a failure to form the repressive HDAC4–Smad3–Runx2 complex. In this model, amounts of HDAC4 need not to be changed (Fig. S2). rib chondrocytes contain more p-Smad3–HDAC4 complex than in wild-type cells because more HDAC4 can be coimmunoprecipitated by anti–p-Smad3 antibody (). The titration of HDAC4 proposed in this model may disrupt the repressive HDAC4–Smad3–Runx2 complex, resulting in the improper activation of Runx2 (). Together, our data show that disruption of leads to abnormal differentiation of chondrocytes into hypertrophic chondrocytes, causing ectopic bone formation in cartilaginous elements of the cervical vertebrae, carpal bone, and rib cage. That this phenotype can be rescued by inactivating one allele identifies an / genetic cascade regulating chondrogenesis. Thus, the regulation of Runx2 function during chondrogenesis is more complicated than previously thought. Our data indicate that the action of Flnb on Runx2 is mediated through the TGF-β–Smad3 pathway, likely by controlling phosphorylated Smad3. Interestingly, similar to the effect of the accumulation of phosphorylated Smad3, loss of TGF-β–Smad3 signals also causes excessive chondrocyte hypertrophy (). These seemingly contradictory results led us to propose that the homeostasis of activated Smad3 protein is essential for suppressing Runx2 activity because the readout of TGF-β depends on the amount of HDAC4–Smad3–Runx2 triple protein complex in the nucleus (). Consequently, either excessive Smad3 or lack of Smad3 will result in Runx2 activation. This is consistent with both the role of Smad3 in preventing chondrocyte differentiation (; ; ) and with reports suggesting that TGF-β can up-regulate Mmp13 production in human osteoarthritic chondrocytes (; ). We remain aware that further experiments are needed to test all aspects of this model. These results do not exclude the possibility that other mechanisms also contribute to the phenotype observed in the mutant mice. The interaction between Flnb and Smad3 in chondrocyte differentiation revealed in this study indicates that Filamins are involved in TGF-β–Smad signaling. Supporting this notion, Filamin A physically interacts with Smad2 and -5 and functionally promotes Smad2 phosphorylation, which is suggested by the lack of phosphorylation in –defective human melanoma cells (). Although our results indicate that Flnb normally prevents excessive Smad3 phosphorylation, we hypothesize that the action of Filamins on Smad proteins may vary in different cell types. From a biomedical point of view, this study provides a mouse model for SSS, a disease associated with nonsense mutations in ; more important, it uncovers a molecular mechanism explaining the vertebral and carpal bone fusions seen in these patients. In contrast to the ectopic ossification caused by nonsense mutations in , missense mutations in cause the absence or underossification of limb bones and vertebrae in Boomerang dysplasia () and atelosteogenesis I and III (). In these dominant disorders, we speculate that the mutant Flnb causes delayed chondrocyte hypertrophy or affects osteoblast differentiation. In fact, Flnb may also regulate Runx2 function in osteoblast differentiation, as indicated by the less developed occipital bone, which is formed through intramembranous ossification, in Runx2; Flnb than in Runx2; Flnb mice (Fig. S3, available at ). Examining more alleles of in the mouse will provide further insights into its additional roles in skeletal development. In summary, our study not only reveals a new mechanism for regulating chondrocyte differentiation but also provides a molecular basis for the skeletal dysplasia caused by nonsense mutations. The gene trap mice were generated with the 129/Ola ES cell line XD076 from BayGenomics (). The ES cell line was expanded and injected into C57BL/6-Albino blastocysts in the Darwin Transgenic Mouse Core facility at Baylor College of Medicine (). Chimeras were bred to C57BL/6 mice, and germline transmission was achieved. Mice were genotyped using Southern blotting after BglII digestion with a probe to the 5′ end of the insertion, which is amplified with the primer pair: filb-5southF (AACATGGCTTGCTGTGACTG) and filb-5southR (GGAGAGGGAAATCCGAAGTC). The knockout was used previously (). The primers for genotyping the mice are as follows: Cbfa1DB (CACGGAGCACAGGAAGTTGGG), Cbfa1DF (TGAGCGACGTGAGCCCGGTGG), and Neo3F (AAGATGGATTGCACGCAGGTTCTC). The Cbfa1DB/Neo3F primer pair amplifies the mutant allele, whereas the Cbfa1DB/Cbfa1DF primer pair amplifies the wild-type allele. For timed matings, the morning of the vaginal plug was designated E0.5. All animal protocols were approved by the Institutional Animal Care and Use Committee of Baylor College of Medicine. Total RNA was prepared from liver or kidney dissected from P21 homozygous, heterozygous, and wild-type mice ( = 3 for each genotype) using RNAqueous-4PCR kit (Ambion). Transcripts were amplified using Superscript One-Step RT-PCR Systems (Invitrogen). The primers to amplify the wild-type transcript spanning the β-geo insertion are as follows: rtf1 (TCTTCCCACATACGATGCAA) and rtr (TCCACTACAAAGCCCACCTC). The primers to amplify the mutant transcript (fusion of and insertion) are as follows: rtf2 (CCTATATCCCTGATAAGACCGGACGC) and gal (GACAGTATCGGCCTCAGGAAGATCG). Mice were dissected, fixed in 95% ethanol overnight, and stained in 0.015% alcian blue (Sigma-Aldrich) dye overnight. The carcasses were treated with 2% KOH for 1 to several days until most soft tissue disappeared. After removing the remaining soft tissues carefully with forceps, the mice were stained in 0.005% alizarin red (Sigma-Aldrich) solution for 30 min to several hours depending on the age of the mice. Finally, skeletons were cleared in a 20% glycerol/1% KOH solution. At least four mice of each genotype were analyzed for each stage. Mice at P0 were dissected so that most tissues other than skeleton are removed. The skeletons were fixed in 0.2% glutaraldehyde for 1 h at RT followed by washing in PBS twice. Fixed specimens were then put in fresh staining solution overnight at RT. The staining solution was made by mixing 1 ml 10× stock solution, 250 μl 0.2 M EGTA, and 250 μl 40 mg/ml X-gal into 8.5 ml PBS. The 10× stock solution was made as follows: for 100 ml solution, add 1.646 g KFe(CN) Ferro, 2.112 g KFe(CN) Ferri, 2 ml 1 M MgCl, 2 ml 1% NP40, and 0.01 g sodium deoxycholate. X-gal stock solution was made by dissolving 100 mg X-gal in 2.5 ml dimethylformamide. All microscopy for skeletal preparation and X-gal staining of the skeleton are performed with Stemi SV11 microscopy (Carl Zeiss MicroImaging, Inc.) with 0.5× (Plan) and 1.0× (Planapo) objectives at RT. Pictures were taken with a color video camera (DXC 300 3CCD; Sony), captured through the Spot software (Diagnostic Instruments), and further processed with Photoshop (Adobe). Tissues were fixed in 4% paraformaldehyde/PBS overnight at 4°C. Mice obtained after P0 were decalcified in TBD-2 solution (Thermo Shandon) overnight at 4°C. Specimens were dehydrated and embedded in paraffin and sectioned at 5 μm. For histological analysis, sections were stained with hematoxylin and eosin. In situ hybridization was performed using digoxigenin-labeled riboprobes (Roche). The α1(I), α1(II), and α1(X) probes were used previously (). Hybridizations were performed overnight at 60°C, and washes were performed at 65°C. Sections were then incubated for 2 h with alkaline phosphatase–linked anti-digoxigenin antibody (1:2,000; Roche) at RT followed by color reactions with BM purple substrate (Roche). Sections were finally counterstained with nuclear fast red (Vector Laboratories) and mounted with Permount. Three mice of each genotype were analyzed for each stage. The microscopy for all histology and in situ sections were performed using a microscope (AxioPlan2; Carl Zeiss MicroImaging, Inc.) with 5× (Plan-Neofluar), 10× (Plan-Apochromat), or 20× (Plan-Apochromat) objectives at RT. All images were taken with AxioCam (Carl Zeiss MicroImaging, Inc.), captured through the AxioVision software (Carl Zeiss MicroImaging, Inc.), and processed with Photoshop software (Adobe). Smad-expressing mammalian plasmids were provided by K. Watanabe (National Institute for Longevity Sciences, Aichi, Japan; pF:Smad1, -2, -4, -5, and -6) and R. Derynck (University of California, San Francisco, San Francisco, CA; flag-Smad3; ; ). The CBFβ- expressing plasmid was obtained from T. Watanabe (Tohoku University, Sendai, Japan; ). To generate myc-tagged N terminus–truncated mammalian-expressing Filamin B, the C-terminal 231-amino-acid coding region was generated by PCR and inserted into the StuI–XbaI sites of the CS2+MT vector. To produce his-tagged Smad3 protein, PCR-amplified Smad3 gene was subcloned into NdeI and BamHI sites of a bacterial expression vector, pET-15b (Novagen). To produce GST fusion protein, the C terminus of Filamin B was PCR amplified and subcloned into NdeI and HindIII sites of a pGEX2TL. BL21-pLysS was transformed with His-tagged Smad3 and GST-fused Filamin B bacterial expression vectors, respectively. After induction with IPTG, the bacterial pellets were lysed by sonication in the presence of Ni-column binding buffer (20 mM TrisCl, pH 7.9, 500 mM NaCl, and 5 mM Imidazole) and GST binding buffer (20 mM TrisCl, pH 7.9, 1 M NaCl, 0.2 mM EDTA, 2 mM DTT, and 10% Glycerol), respectively, and centrifugation was followed. From each supernatant, His-tagged Smad3 and GST-fused Filamin B proteins were isolated by Ni-NTA (GE Healthcare) chromatography and glutathione-coupled Sepharose (GE Healthcare) chromatography. 5 mg of purified His-tagged Smad3 protein was incubated with either 10 mg immobilized GST or GST-fused Filamin B on glutathione-coupled Sepharose in BC300 (20 mM Tris, pH 7.9, 0.2 mM EDTA, 300 mM KCl, 20% glycerol, 0.1% NP-40, and 1 mM PMSF) solution containing 2 mg/ml BSA for 6 h at 4°C. The immobilized GST or GST-fused Filamin B beads were washed extensively with BC300 and analyzed by immunoblot after SDS-PAGE. Fig. S1 shows the expression of different portions of the transcript in the mutant compared with the wild type. Fig. S2 shows the expression of CBFβ, HDAC4, p-Smad3, and Runx2 in the chondrocytes and interaction between Filamin B and Smad4 or -6. Fig. S3 shows the aggravated occipital bone phenotype by deleting in the background of haploinsufficiency. Online supplemental material is available at .

Cell cycle genes and specifically those genes that regulate the G1/S transition have been shown to play an important role in regulating the neural precursor population. Members of the cyclin-dependent kinase inhibitor (CDKI) family have received much of the attention. CDKIs, p21, and p27 negatively regulate embryonic and adult neural precursor proliferation (; ). Bmi-1 promotes self-renewing cell division in both hematopoeitic and neural precursors through the transcriptional repression of CDKIs, p16, and p19 (, ). However, cell cycle regulators impacting the neural precursor population are not only restricted to CDKIs (). We have recently shown that the Retinoblastoma (Rb) family member p107, an inhibitor of the cell cycle G1/S transition, negatively regulates the neural precursor pool in the developing and adult brain by regulating self-renewal (). p107 has been shown to function by interacting with E2F transcription factors (preferentially E2F4) to repress the transcription of genes required for cell cycle progression (). Distinct from other Rb family members, p107 is only expressed in cycling neural precursor cells in the ventricular zone (VZ; ). The Notch–Hes pathway is necessary for self-renewing cell division and, thus, maintenance of the neural precursor population (; ; ; ). Whereas the deletion of either Notch1, Hes1, or Hes1 and Hes5 causes premature differentiation of embryonic neural precursors, resulting in their depletion (; ; ), the overexpression of activated Notch1 or Hes1 results in an expansion of neural precursor numbers (). Hes1 and Hes5 inhibit differentiation by repressing the expression of the proneural genes , , and (; ). Because the Notch–Hes signaling pathway is crucial for neural precursor self-renewal and inhibition of premature differentiation, we asked whether the cell cycle protein p107 may be regulating the neural precursor population and progenitor differentiation by the repression of Hes1. In this study, we demonstrate that the p107-mediated regulation of neural precursor number occurs through the repression of transcription. Hes1 is elevated in p107-deficient brains. Loss of a single allele restores the neural precursor population to wild-type levels both in vitro and in vivo. Despite the expanded progenitor population, p107- deficient brains exhibit a reduction in the number of cortical neurons that cannot be accounted for by apoptosis. Short- and long-term BrdU labeling studies revealed a striking defect in the rate at which p107-null progenitors commit to a neuronal fate. Loss of a single Hes1 allele on a p107-null background rescues the number of neurons born during cortical development. Together, these results identify that the mechanism by which p107 regulates both neural precursor self-renewal and differentiation is through regulation of the Notch–Hes1 signaling pathway. In summary, we identify a novel function for p107, a cell cycle regulatory protein, in controlling the onset of differentiation. To determine the temporal requirement for p107 in regulating the neural precursor population, we counted the number of proliferating precursors in the brains of mice at three different ages: in adults and in embryos at embryonic days (E) 10.5 and 13.5. Using antibodies to label cells in the cell cycle (proliferating cell nuclear antigen [PCNA], which labels cells in all phases of the cell cycle; phosphohistone H3 [PH3], which labels cells in M phase; and BrdU, which gets incorporated during S phase), we demonstrate an increase in the proliferating precursor population in p107-null mice. Adult p107-null mice have a 50% increase in the number of precursors, as demonstrated by both cumulative BrdU labeling of proliferating progenitor cells and PCNA immunostaining (). Similarly, at both embryonic time points E10.5 and 13.5, p107-null embryos had more precursor cells (). The difference is most pronounced in adult mice when proliferation rates are slower, with a cell cycle time of ∼12.7 h (). These studies demonstrate that p107 mutants have an expanded precursor population. As the increased number of neural precursor cells in p107-deficient mice could be the result of an increased total population or an enhanced rate of cell division, we assessed the proliferative index in embryonic and adult wild-type and p107 mice. Double immunohistochemistry for BrdU and PCNA was performed on adult brains after 10.5 h of cumulative BrdU labeling and on embryonic brains after a 2-h pulse of BrdU. The number of cells that entered S phase (BrdU cells) was compared with the total proliferating population (PCNA- expressing cells) in wild-type and p107 mice. A more rapid cell cycle time in p107 knockouts would result in a greater percentage of cells in S phase (). In adult wild-type mice, 78.2 ± 1.4% of the proliferating population incorporated BrdU, whereas in p107 mice, 78.0 ± 1.9% were BrdU positive (). After a 2-h BrdU pulse at E13.5, 53.6 ± 1.8% of wild-type and 51.3 ± 3.2% of p107 cells incorporated BrdU (). Thus, comparable proliferative indices in adult and embryonic ages indicate similar cell cycle times. Therefore, the increased number of total progenitor cells at each stage of development represents an overall expansion of the precursor population in p107 mice. This increase in total precursor number is consistent with our previous studies revealing that p107-deficient animals exhibit elevated levels of the Notch1–Hes1 pathway and increased stem cell self-renewal (). As the Notch1 pathway has been shown to play an essential role in stem cell self-renewal (), our expanded precursor population in p107 deficiency is consistent with an increase in the Notch–Hes1 pathway. Therefore, we asked whether the expanded precursor population in p107-deficient mice is the result of the deregulation of Hes1. Because we have previously demonstrated the deregulation of Notch signaling in p107-deficient neural precursors (), we examined the expression of Notch targets Hes1 and Hes5 in p107 deficiency. Hes1 and Hes5 are basic helix-loop-helix transcription factors that act downstream of Notch to regulate neural precursor self-renewal (). In situ hybridization revealed elevated levels of transcript in cells of the VZ in p107 mice at E14.5 (). Quantitative real-time RT-PCR further demonstrated increased mRNA in embryonic p107-deficient cortices (). In contrast to expression, no difference in mRNA was detected by in situ hybridization or real-time RT-PCR (). An examination of Hes1 protein by Western blotting also revealed enhanced expression in neurospheres derived from p107 embryos (). Although both Hes1 and Hes5 are Notch1 targets, only is deregulated in p107 mice. The selective deregulation of expression leads us to question whether p107 could regulate transcription. Because in situ hybridization and real-time RT-PCR reveal that is deregulated in p107 neural precursors, we questioned whether p107 could regulate transcription. To determine whether the promoter is responsive to p107, we used a luciferase promotor assay. The promoter (∼1,500 bp) was inserted into a pGL3-Basic reporter vector containing the luciferase gene (B-Hes1; ). This construct was transfected into HEK 293A cells along with 3 μg of a p107 expression vector or Rb expression vector as a control. Cotransfection of 3 μg p107 resulted in a 2.5-fold reduction in promoter activity (). This repression was dose dependent because a further increase in p107 (10 μg) resulted in a >10-fold repression. In contrast, protein Rb (pRb) did not repress promoter activity. These results demonstrate that p107 represses promotor activity. Because p107 regulates transcription by interacting with E2F transcription factors, we asked whether E2Fs were required for p107-mediated repression. The promotor was analyzed, and three putative E2F-binding sites (BSs [E2F-BS]) were found at positions −560, 161, and 398 bp relative to the transcription start site (). To test whether E2Fs were required for p107-mediated repression, we deleted all three BSs from the promotor (Hes1-3xBS). Cotransfection of the Hes1-3xBS construct with the p107 expression vector resulted in pronounced repression of luciferase activity (). These results demonstrate that p107-mediated repression of the promotor occurs indirectly, independent of these E2F sites. Nevertheless, our results showing decreased levels of transcript and protein combined with p107-mediated repression of the promoter demonstrate that p107 is required in the neural precursor population for the repression of . To ask whether p107 regulates the neural precursor population by the repression of , we interbred p107-deficient mice with animals carrying a null mutation for . We hypothesized that if p107 regulated neural precursor cells by controlling the levels of Hes1, the loss of one or more alleles of in p107-null mice should partially or completely restore the expanded precursor population to wild-type levels. Conversely, if p107 acted through an independent pathway, the number of neural precursors would not be affected by the loss of Hes1, as seen in wild-type cells. mice are embryonic lethal after E12.5, embryos were taken at E10.5 (). The neuroepithelia from each embryo was dissociated into a single-cell suspension, cells were plated at clonal density, and, after 7 d, neurospheres were counted. Cultures from p107-null mice produced substantially more neurospheres than cultures from all other genotypes (). Consistent with previous findings (), the loss of Hes1 alone had no effect on neural precursor numbers from wild-type animals, whereas the absence of one or both alleles of in p107-deficient precursor cells restored neurosphere numbers to wild-type levels. Together with the demonstration that p107 represses gene expression, these results demonstrate that the expanded neural precursor population in p107 embryos results from the deregulation of . We next questioned whether p107 controlled the size of the neural precursor population in vivo through the repression of . As a result of the embryonic lethality of double mutant mice, we asked whether the loss of a single allele could restore the number of neural precursor cells in p107-null mice to wild-type levels. In adult mice, the precursor population was labeled with BrdU. Counts of BrdU-positive cells in adult brains showed that a reduction in Hes1 could restore the number of proliferating precursors in p107-null mice (). PH3 immunohistochemistry of the rapidly dividing E10.5 neural precursors demonstrated that loss of a single allele in p107-null embryos (Hes1:p107) reduced the number of cells in M phase to levels comparable with the wild type (). Similarly, a 2-h BrdU pulse in E13.5 embryos also revealed a reduction in the number of cells in S phase in p107 embryos lacking a single allele (). Loss of a single allele in p107 mice restored the number of proliferating cells to wild-type levels at all three developmental time points. In contrast, loss of a single allele on a wild-type background (Hes1) did not result in a reduction in the number of proliferating cells (). These results demonstrate that p107 regulates the neural precursor population by controlling the levels of Hes1 in progenitor cells. The expanded neural precursor numbers in p107 mice lead us to question whether the number of cortical neurons was also affected. Specifically, we asked whether the expanded precursor population leads to increased neurogenesis. Immunohistochemistry with NeuN, a marker for mature central nervous system neurons, revealed a significant reduction (P < 0.05) in the number of NeuN-expressing cells in p107 cortices (). The decreased number of cortical neurons is further demonstrated by a reduction in the overall size/thickness of the cortex (), whereas measurements of the VZ did not reveal any differences between p107 mice and wild type at E18.5 (). Because p107-deficient mice exhibit a reduction in the total number of cortical neurons despite an expanded progenitor pool, we questioned whether p107 was required for the survival of cortical neurons. To determine whether an increase in cell death was responsible for the reduction in cortical neuron numbers in p107-deficient brains and at what stage in development cells were dying, we counted the number of apoptotic cells in the VZ/sub-VZ (SVZ), the intermediate zone (IZ), and in the cortical plate and marginal zone. Apoptotic cells were identified by immunohistochemistry for active caspase-3 and Hoechst nuclear staining in E13.5 embryonic brains (). The number of apoptotic cells in the brains of wild-type and p107 mice was low, with less than one cell per 1,000-μm area (). As a result of the low frequency of apoptotic cells, counts were performed in 12 sections throughout the telencephalon in both left and right hemispheres, and the total number of apoptotic cells was compared between genotypes (). A twofold increase in the number of apoptotic cells was observed only in the VZ/SVZ of p107-null brains, whereas no differences were observed in the IZ or cortical plate/marginal zone. An increase in cell death in the p107 mutant VZ/SVZ suggests that newly committed neurons may be dying in the VZ before they initiate migration. To identify at which stage in progenitor cell development cells were dying, double immunolabeling was performed with antibodies to active caspase-3 and Nestin (progenitor cell marker), doublecortin (migratory neuroblast), and βIII-tubulin (Tuj1, an early neuronal marker). Because the increase in apoptotic cells is only in the VZ/SVZ, we counted the double-labeled cells in this region. At E13.5, most cells were double labeled with active caspase-3/Nestin (), suggesting that p107 affects the survival of uncommitted progenitor cells. No double labeling was detected with doublecortin or Tuj1 in any of our samples, ruling out the possibility that p107 deficiency results in enhanced cell death in newly committed neuroblasts. As apoptosis progresses, dying cells will lose their cell type–specific markers; thus, some pyknotic cells did not colabel with Nestin or any other marker examined. The colabeling of apoptotic cells with caspase-3 and Nestin is consistent with the interpretation that cell death is occurring in uncommitted progenitor cells in the VZ/SVZ. The increase in apoptosis in the VZ/SVZ did not reveal any death of newly committed neurons. In conclusion, these studies show that the absence of p107 results in the increased cell death of Nestin- expressing progenitor cells. As the increased apoptosis in the VZs of p107-null mice is caused by the death of uncommitted Nestin-expressing progenitor cells and the rate of cell death was still very low, we questioned whether the absence of p107 may result in a defect in the rate of neuronal commitment. To address this question, we performed neuronal birthdating assays using a BrdU protocol to label cells undergoing terminal mitosis at the time of BrdU injection. Neuronal birthdating is based on the demonstration that neurogenesis occurs between E12 and 17, during which cohorts of neural precursors are born (neuronal commitment) at distinct time points and migrate out of the VZ to form the layers of the cortex (; ; ). A single BrdU injection labels all cells in S phase, but only neural precursors undergoing terminal mitosis retain the BrdU label. Therefore, BrdU birthdating provides a quantitative analysis of cells that commit to a neuronal fate, undergo terminal mitosis, and migrate to their ultimate destination in the cerebral cortex. Accordingly, pregnant dams were injected with BrdU at E13.5 (), the time at which deep layer cortical neurons are generated. Embryos were collected 5 d after injection at E18.5. BrdU cell counts revealed that p107 mice (36 ± 5; = 5) had a dramatic twofold reduction in the number of neurons that were born at injection time (E13.5) and reached the cortical plate by E18.5 relative to wild-type littermates (68 ± 9; = 4; ). These results show that there is a decrease in the number of neurons reaching the cortical plate in p107-deficient brains. Because our examination of cell death did not reveal a loss of newly generated neurons in the cortical plate, we asked whether the reduced neuronal numbers in p107-deficient mice was caused by fewer neurons born at E13.5 (the time of BrdU injection). To address this question, we performed a 24-h BrdU incorporation to measure the rate of neuronal commitment. Pregnant dams were injected at E13.5, embryos were collected 24 h later at E14.5, and the number of strongly labeled BrdU-positive cells were counted (i.e., cells that underwent terminal mitosis at the time of injection). In addition, sections were double stained with PCNA to show that these cells are no longer cycling. Double labeling with BrdU and PCNA revealed that most BrdU-positive cells within the SVZ and IZ were no longer expressing PCNA, indicating that they were newly postmitotic (). Cell counts of this newly postmitotic population revealed a two-fold reduction in p107-deficient brains compared with wild-type controls (). These results demonstrate that there is a decrease in the number of newly postmitotic cells leaving the VZ in p107-deficient animals. We hypothesized that a decrease in the number of newly postmitotic cells migrating out of the VZ/SVZ in p107-deficient mice indicates that p107 may be required for the regulation of neuronal commitment. To test this possibility, we performed the aforementioned 24-h BrdU commitment assay followed by double labeling with BrdU and doublecortin to identify migrating neuroblasts or BrdU and Tuj1, an early panneuronal marker induced just after terminal mitosis. Double labeling revealed that p107-deficient mice exhibited a striking decrease in the number of BrdU-positive cells expressing doublecortin (wild type, 76.7 ± 0.7; and p107, 56.1 ± 4.1) and Tuj1 (wild type, 78.3 ± 3.2; and p107, 49.5 ± 5.4; ). These findings support a model whereby fewer p107-deficient progenitor cells commit to a neuronal fate, resulting in a twofold reduction in cortical neurons in the brain at E18. In summary, our results reveal that p107 is required for neuronal commitment and promotes the decision to exit the progenitor pool and commit to a neuronal fate. Because our results show that Hes1 is involved in p107-mediated neural precursor self-renewal and Hes1 functions to maintain the neural precursor population by repressing the expression of proneural genes (; ), we asked whether deregulated Hes1 could account for the defect in neurogenesis in p107-null mice. The loss of a single allele could partially restore the rate of neurogenesis by increasing the number of neurons that commit to a neuronal fate at E13.5. Specifically, BrdU birthdating revealed that a reduction in Hes1 could restore the number of neurons born at E13.5 from 37 ± 5 ( = 4) in p107 to 59 ± 3 ( = 6) in Hes1:p107 comparable with wild-type levels of 68 ± 9 ( = 4; ). These results highlight that p107 regulates the neural precursor pool by regulating self-renewal and controlling the decision to exit the progenitor pool and commit to a neuronal fate. Furthermore, we show that the mechanisms underlying p107-mediated regulation of the neural precursor population is through repression of the Notch–Hes1 pathway. t h i s s t u d y , w e i d e n t i f y a n o v e l r o l e f o r p 1 0 7 i n r e g u l a t i n g t h e t r a n s i t i o n f r o m p r o l i f e r a t i n g n e u r a l p r o g e n i t o r c e l l t o c o m m i t t e d n e u r o b l a s t . D u r i n g n e u r a l d e v e l o p m e n t , p 1 0 7 r e g u l a t e s t h e s i z e o f t h e n e u r a l p r e c u r s o r p o o l b y l i m i t i n g s e l f - r e n e w a l c a p a c i t y a n d p r o m o t i n g t h e t r a n s i t i o n f r o m p r o g e n i t o r c e l l t o c o m m i t t e d n e u r o b l a s t . F u r t h e r m o r e , w e s h o w t h a t t h e m e c h a n i s m b y w h i c h p 1 0 7 r e g u l a t e s t h e n e u r a l p r e c u r s o r p o o l i s t h r o u g h r e p r e s s i o n o f N o t c h – H e s 1 s i g n a l i n g . Germline p107-null mice were generated previously by and maintained on a mixed SV-129 and C57BL/6 background. Germline Hes1-null mice on an ICR background were originally generated by R. Kageyama (). Hes1/p107-deficient embryos were generated by interbreeding heterozygous (Hes1) mice with p107 mice. Animals were genotyped according to standard protocols with previously published primers for () and (). For embryonic time points, the time of plug identification was counted as E0.5. All experiments were approved by the University of Ottawa's Animal Care Ethics Committee, adhering to the guidelines of the Canadian Council on Animal Care. Pregnant dams and adult mice were killed with a lethal injection of sodium pentobarbitol. Embryos were dissected and submersion fixed overnight in 4% PFA in 1× PBS, pH 7.4. Adult mice were perfused with 1× PBS followed by cold 4% PFA, and brains were removed. Brains were postfixed overnight in 4% PFA, cryoprotected in 22% sucrose in 1× PBS, and frozen, and 14-μm coronal sections through the forebrain were collected on Superfrost Plus slides (Fisher Scientific). Nonradioactive in situ hybridization and digoxigenin probe labeling was performed according to previously described protocols (). Antisense riboprobes for and were generated according to previously published sequences (). Total RNA was isolated from cortices from wild-type and p107-deficient embryos using TRIzol reagent according to the manufacturer's instructions (Invitrogen). RNA was reverse transcribed using the GeneAmp RNA PCR Core kit (Applied Biosystems). Real-time PCR was performed on cDNA using the TaqMan Universal PCR Master Mix (Applied Biosystems) with commercially provided PCR primers for , , and from TaqMan Gene Expression Assays (Applied Biosystems). Immunohistochemistry was performed on coronal cryostat sections from embryonic and adult brains with primary antibodies to mouse anti-NeuN (1:100; Chemicon), rabbit antiactive caspase-3 (1:500; BD Biosciences), rabbit anti-PH3 (1:400; Upstate Biotechnology), rat anti-BrdU (1:100; Accurate Chemicals), mouse anti-BrdU (1:100; Becton Dickinson), mouse anti-PCNA (1:300; Vector Laboratories), goat antidoublecortin (1:100; Santa Cruz Biotechnology, Inc.), mouse anti–βIII-tubulin (mouse monoclonal hybridoma supernatant; 1:100; ), and mouse anti-Nestin (1:400; Research Diagnostics). For BrdU birthdating experiments (E13.5–18.5), pregnant dams received a single injection of BrdU at 20 μg/g of body weight; for short-term BrdU incorporation experiments, pregnant dams and adult mice received intraperitoneal injections of BrdU at 100 μg/g of body weight. For BrdU detection, sections were denatured in 2 N HCl at 37°C for 15 min followed by neutralization in 0.1 M Na borate, pH 8.5, for 10 min at room temperature before incubation with the primary antibody. For PCNA detection, sodium citrate antigen retrieval, pH 6.0, was performed on sections before incubation with PCNA antibodies. In all double labeling with BrdU antibodies, denaturation and BrdU immunohistochemistry were performed after the first primary and secondary antibody incubation. Protein was isolated from cultured neurospheres in lysis buffer, run on a 15% SDS-polyacrylamide gel, and transferred to a nitrocellulose membrane as described previously (). Immunoblotting was performed with antibodies directed against Hes1 (provided by H. Kitamura, Yokohama City University School of Medicine, Yokohama, Japan; ), Hes5 (Chemicon), and actin (Santa Cruz Biotechnology, Inc.). Blots were developed by chemiluminescence according to the manufacturer's instructions (NEL100; PerkinElmer). In the adult brain, BrdU cells were counted along the entire length of the ventricular surface (dorsal and ventral) in every 10th section from the rostral crossing of the corpus callosum to the start of the third ventricle and crossing of the anterior commissure with an equal number of sections counted per brain as previously described (; ). PH3 cells were counted along a 1,000-μm length of the ventricle in three representative sections through the forebrains of E10.5 embryos. In E13.5, 14.5, and 18.5 embryos, BrdU cells and/or NeuN cells were counted along a 750-μm length of the ventricle up through to the pial surface in four representative regions through the forebrain (). Cortical plate, VZ, and cortical mantle measurements were performed on cresyl violet sections of E18 brains from wild type. Triplicate measurements were performed on three representative sections through the forebrain. The promotor (1,270 bp) sequence, including a 224-bp 3′ sequence of the transcription start site, was amplified from mouse genomic DNA by PCR and inserted into a pGL3-Basic luciferase reporter construct (forward primer 5′-CGCGGCGGCAATAAAACATC-3′; reverse primer 5′-GATGAGTGCACAGGGGGAGAAAAGAGGTC-3′; Promega; ; ). To assess whether promotor activity was affected by p107, the pGL3B-Hes1 construct or the E2F-BS mutant (Hes1-3xBS) was cotransfected with expression vectors for either (pCMV-p107) or (pGK-RB) into HEK 293A cells by standard calcium phosphate precipitation (). 2 μg pMLV-LacZ was cotransfected with each sample to control for transfection efficiency. 4-methylumbelliferyl--galactoside assay was performed to standardize the transfection reaction, and luciferase activity was assessed according to standard procedures (). Statistical analysis was performed on the means of three different experiments. Putative E2F consensus sites were identified by MatInspector software (Genomatix; ). A mutant promotor minus all three E2F-binding sequences (Hes1-3xBS) was constructed by linking PCR fragments on each side of the E2F-BSs and inserting them into the pGL3B-luciferase construct (). The neurosphere assay was performed on neuroepithelia from E10.5 embryos as previously described (; ; ). Statistical comparisons were performed on the mean number of neurospheres per embryo per genotype. Sections treated for immunohistochemistry or in situ hybridization were examined by a microscope (Axioskop 2; Carl Zeiss MicroImaging, Inc.) with standard fluorescence and brightfield/darkfield settings at 5× NA 0.25 or 20× NA 0.50 objectives. Images were captured using a color video camera (Power HAD 3CCD; Sony) with Northern Eclipse software (Empix Imaging). For confocal microscopy, images were captured using a microscope (LSM 510 META; Carl Zeiss MicroImaging, Inc.) on an inverted microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with the manufacturer's integrated digital imaging software. Figures were compiled using Photoshop 6.0 (Adobe). Manipulations of brightness and intensity were made equally to all treatment groups.

Eukaryotic cells can detect and move up concentration gradients of chemoattractants, a process known as chemotaxis (; ; ; ). This behavior plays an important role in a number of processes, including metastasis, angiogenesis, immune responses, and inflammation (; ; ). Furthermore, chemotaxis is essential for cell aggregation in the life cycle of the social amoebae, (; ; ; ). Chemotaxis is a coordinated phenomenon of three fundamental cell processes: gradient sensing, cell polarization, and cell motility. Chemotactic cells, such as neutrophils and , display polarized morphology, involving asymmetric distributions of many signaling molecules (; ; ; ), and heightened responsiveness to the attractant at their leading edge (; ; ). These crawling cells extend their leading edges by assembling a force-generating network of actin filaments beneath the plasma membrane (; ). Elsewhere in the cell, actin collaborates with myosin to retract the rear of the advancing cell and to prevent errant pseudopod extension (). Consequently, the actin-depolymerizing agent Latrunculin can be used to eliminate polarization and motility of cells, and thus facilitate quantitative spatiotemporal analyses of the mechanisms underlying gradient sensing (; ; ). Gradient sensing is mediated by G protein–coupled receptors (GPCRs) and associated signaling components that detect the spatiotemporal changes of chemoattractants and translate shallow gradients of chemoattractants into steep intracellular gradients of signaling components (; ; ; ). Binding of cAMP to the GPCR cAR1 induces the dissociation of heterotrimeric G proteins into Gα2 and Gβγ subunits (; ; ). Free Gβγ activates Ras, thereby leading to the activation of PI3K, which converts PIP (PIP) to PIP (PIP) in the plasma membrane (; ; ; ; ). The phosphatase PTEN acts as an antagonist of PI3K, dephosphorylating PIP to regenerate PIP (; ; ). PIP mediates cellular processes by recruiting proteins with pleckstrin homology (PH) domains, such as cytosolic regulator of adenylyl cyclase (CRAC) and Akt/PKB, to the plasma membrane (; ). Both CRAC and Akt/PKB play roles in the regulation of actin polymerization during chemotaxis (; ). Recent progress in fluorescence microscopy has permitted measurements of the spatiotemporal changes of many signaling events in living cells with high spatiotemporal resolution required to test models of gradient sensing (; ; ). There are several key features of gradient sensing. First, cells have the ability to spontaneously terminate responses under a sustained cAMP stimulation in a process called “adaptation” (; ). Second, if cAMP is removed from adapted cells, the cells will enter a de-adaptation phase—a refractory period lasting several minutes during which the cells progressively regain their ability to respond to another cAMP stimulation (,). Third, cells have the capability of translating shallow cAMP gradients across the cell diameter into highly polarized intracellular responses, a process called “amplification” (; ; ). To explain these features, it has been proposed that an increase in receptor occupancy activates two antagonistic signaling processes: a rapid “excitation” that triggers cell responses, such as the membrane accumulation of PIP, and a slower “inhibition” that turns off those responses (). Although many of the molecular mechanisms of the excitatory process have been identified, those of the inhibitory process have remained elusive. The dynamic relationship between excitation and inhibition that leads to activation, adaptation, and amplification has been studied by direct visualization and quantitative analysis of the spatiotemporal changes in receptor occupancy, G protein dissociation, PI3K and PTEN distribution, and PIP level along the membrane (; ). Over the years, models have been proposed to explain how the excitatory and the inhibitory processes interact in cells responding to chemoattractants to achieve adaptation or amplification (; ; ; ; ; ; ; ). Although inhibitors are essential components of all gradient sensing models, the spatial-temporal presence of inhibitors has not been examined experimentally. In this study, we designed sequential stimulation protocols to detect temporal and spatial aspects of the inhibition process in single living cells. We found that repeated transient activations of cAR1 receptor trigger repetitive PH-GFP membrane translocations without detectable refractory periods, demonstrating that a short pulse of cAR1 activation elevates excitation but little long-lasting inhibition. This result provides evidence that cAR1 activation induces an immediate excitation and a delayed recruitment of long-term inhibition leading to PIP accumulation. More significantly, we have revealed spatial distribution of the inhibition process induced by a cAMP gradient. Exposing a cell to a sustained cAMP gradient leads to a stable PH-GFP accumulation in the front of the cell. We found that a sudden withdrawal of the cAMP gradient from this biochemically polarized cell leads to a rapid return of G protein activation, PTEN, and PIP distributions to basal levels around the cell membrane. However, there was a short time period during which reactivation of receptors and G proteins around the membrane induced a clear PIP response in the back but not the front of the cell. This inverted PIP response indicates that a cAMP gradient induces a stronger inhibition of PI3K in the front of a cell. Previous studies suggest that activation of cAR1 triggers a fast increase of the excitation level and a slower elevation of the inhibition, allowing a cell to respond transiently and then adapt (; ; ; ). When a sustained cAMP stimulus is removed, cells that had adapted enter a refractory period during which the cells progressively regain their ability to respond to cAMP, suggesting that the inhibition returns to its prestimulus level more slowly than does excitation (; ,). To test whether there is a temporal difference between the cAR1-induced excitatory and inhibitory processes controlling PIP production, we measured the kinetics of PIP levels around the membrane of cells that were stimulated by multiple transient cAMP stimuli (). We simultaneously visualized and quantitatively measured transiently applied cAMP stimulations and PIP production, reported by the membrane translocation of PH-GFP, a PIP reporter (). Chemotaxis-competent PH-GFP-expressing cells (“PH cells”) were treated with Latrunculin B, which eliminates morphological polarity and motility by disrupting the actin cytoskeleton. A micropipette filled with cAMP (1 μM) was placed within 30 μm of PH cells and used in conjunction with a microinjector to deliver a series of brief cAMP stimulations. Alexa 594 was included in the micropipette as a measure of the applied cAMP concentration (). Each cAMP stimulation induced a transient response of PH-GFP membrane translocation (). Temporal changes in cAMP concentration around the cell were determined as the average intensity change of the dye in the R1 and R2 regions (). The kinetics of PH-GFP membrane translocation were measured as intensity changes in cytosolic PH-GFP pool (), which is inversely related to the amount of membrane-associated PH-GFP (). Quantitative analyses showed that PH-GFP translocation reached its maximal level in 4 s after the cAMP concentration reached its peak, which reflects the temporal delay that is expected for PIP production upon the activation of cAR1 (). In our experimental setup, the shortest interval between two transient cAMP stimuli was ∼24 s, a minimal time required for the cAMP concentration to return to its basal level between stimuli (). Sequential transient cAMP stimuli with as short as 24-s intervals generated repetitive transient responses of PH-GFP translocation, and transient responses displayed kinetics without a refractory period (). Our data indicate that a transient receptor activation quickly activates excitatory pathways leading to an increase in PIP and upon cAMP removal, these pathways quickly return to prestimulated levels and can be activated again by another cAMP stimulation. Transient cAR1 activations do not signal long enough to substantially elevate the slower inhibition process from its basal level. Therefore, this result supports the idea that cAR1-mediated excitation and inhibition process increases and decreases by following a fast and a slow temporal mode, respectively. The spatial distribution of inhibition in a cell exposed to a cAMP gradient has never been tested. Several models propose that cAR1-induced inhibitors diffuse quickly in a cell, and thus are assumed to be evenly distributed in the inner membrane (; ; ). However, our dynamic analyses and computer simulations suggest that inhibition mechanisms act locally and predict that inhibition affecting PI3K activity is strongest in the front of a cell at steady state in a cAMP gradient (; ). Because the molecular nature of the inhibitors in chemosensing of is still unknown, we developed an approach to indirectly measure the relative extent of inhibition in the front and back of a biochemically polarized cell. We reasoned that after a removal of the cAMP gradient, the signaling network would rapidly return from the polarized to the resting steady state. During this transition, the time required to regain responsiveness to cAMP (the refractory period) may differ between the front and back depending on local concentrations of inhibitors induced by the prior gradient. In our experiments (), PH cells were first exposed to a cAMP gradient until they achieved a stably polarized state, in which PH-GFP accumulated in the front. After quickly withdrawing the gradient at time 0, we observed that PH-GFP rapidly returned to the cytosol. Before the cells could fully return to the resting state, which may take ∼6 min (; ), they were challenged with a uniformly applied cAMP stimulus. Interestingly, this induced an “inverted” response in which PH-GFP transiently translocated to the back of the cells, demonstrating that the original front sides of the cells were less responsive to cAMP than were the back sides. Moreover, cells that had been exposed to gradients of various cAMP concentrations for the initial stimulus also exhibited inverted PH-GFP responses upon a uniform cAMP stimulation (; and Fig. S1, available at ). The observed inverted PIP response is likely caused by a slow return of the intracellular components to their “resting” states because the average time period for cAMP binding to the receptor is in the range of seconds (). Several components in the pathways may be responsible for the differential refractory behaviors of the front and back of the cells. For example, receptors may remain asymmetrically desensitized or G proteins may not be completely reassociated upon the second cAMP stimulation. We addressed this issue by directly measuring the kinetics of G protein reassociation and reactivation in living cells using FRET analysis (). Cells expressing Gα2-CFP and YFP-Gβ (“G cells”) were suddenly exposed to 10 μM cAMP (), a saturating dose for cAR1, or 1 μM cAMP (Fig. S2, available at ). Addition of cAMP induced a rapid FRET loss, which reached a steady state in <20 s, indicating G protein dissociation (; Fig. S2) (). After the removal of cAMP, FRET returned to the prestimulus level in ∼60 s, indicating that the G proteins were completely reassociated. A second sudden exposure to the same concentration of cAMP triggered an instant FRET loss that displayed kinetics very similar to those in response to the first stimulation (; Fig. S2), demonstrating that cAR1 receptor and G proteins rapidly returned to their prestimulation states and could be fully reactivated within 60 s after the removal of cAMP. We then simultaneously monitored temporal changes in the receptor/G protein activation and PH-GFP translocation in the front and back of the cells previously exposed to a cAMP gradient (). We measured FRET changes in one G cell and PH-GFP translocation in another PH cell. The cells were located within 20 μm and thus both were exposed to very similar cAMP stimuli (). After a rapid withdrawal of a cAMP gradient at time 0, the G protein reassociated around the G cell membrane and PH-GFP returned to the cytosol (). A uniformly applied stimulation at 134.7 s triggered a similar degree of G protein dissociation, which was measured as a CFP fluorescence increase (FRET loss), in both the front and back of the G cell (), while a distinctive accumulation of PH-GFP only occurred in the back of the PH cell (). Because the subsequent cAMP stimulation induced G protein activation in both the front and back, we concluded that the possible asymmetrical desensitization of cAR1 receptors, such as cAR1 phosphorylation induced by the first cAMP gradient, could not explain the subsequent PH-GFP translocation only to the back of the cell (). Thus, the inverted PIP response is likely caused by inhibitory mechanisms acting on the signaling components downstream of cAR1 and the heterotrimeric G proteins. cAR1 activates an excitatory signaling branch that induces PTEN to translocate from the membrane to the cytosol and also elevates an inhibitory mechanism that allows cytosolic PTEN to return to the membrane (; ). cells and compared the kinetics to those in wild-type (WT) cells (; ), in response to uniformly applied cAMP and a cAMP gradient (). cells expressing PH-GFP were stimulated uniformly with cAMP (1 μM) at 0 s (). The cAMP-triggered PH-GFP membrane translocation is fast and transient in WT cells. cells was clearly slower, peaking in ∼12 s and returning to prestimulus levels in more than 40 s (). cells. cells. There was no clear decrease in PH-GFP amount at the front for more than 150 s, which differs from the biphasic response in WT cells (). cells (). cells. Upon a removal of the gradient at 0 s, PH-GFP gradually returned to the cytosol. cells than in the WT cells, whose t were ∼22 s and 14 s, respectively (). PTEN is involved in regulating spatiotemporal dynamics of PIP levels around the membrane of a cell in response to cAMP stimulation. Therefore, the dynamic distribution of PTEN affects the local PIP levels. When a cell reaches the “polarized” steady state in a stable cAMP gradient, PTEN is enriched at the back side of a cell (; ; ). After withdrawal of the cAMP gradient, PTEN starts to redistribute itself from the polarized to the resting steady state. During this transition, a transient accumulation of PTEN in the front could potentially explain the inverted PIP response. To address this possibility, we measured the spatiotemporal dynamics of PTEN under these conditions (). After a rapid withdrawal of the gradient at time 0 (), PTEN redistributed from the back and became uniformly distributed around the membrane in ∼80 s without over-accumulating in the front (). Furthermore, reapplying a uniform stimulus () or gradient () of cAMP induces PTEN translocation with kinetics () similar to those observed in the cells that had not previously been stimulated (), indicating that the cAR1-controlled regulatory components of PTEN returned to their “resting” states and PTEN molecules in both the front and back were responsive to a second cAMP stimulation when the inverted PH-GPF response occurred. Therefore, the excitatory and the inhibitory mechanisms that control PTEN membrane distribution are not the likely explanation for this inverted response. Other mechanisms may also be involved in inhibition in the signaling network of cAMP gradient sensing. For example, Gα9 and Gα1-mediated PLC pathways in have been shown to function as negative regulators in the cAR1-mediated signaling (; ). To test whether either pathway is essential for the gradient sensing, we examined PH-GFP responses in Gα9 and Gα1 null cells (). We measured PH-GFP membrane translocation by monitoring intensity changes of GFP fluorescence in the cell membrane (). In response to a uniform stimulation, the spatiotemporal kinetics of PH-GFP membrane translocation in either gα9 or gα1 were similar to those in the WT cells (). When the gα9 or gα1 cells were suddenly exposed to stable cAMP gradients, PH-GFP translocation, as in WT cells, consisted of two phases, an initial transient translocation around the cell membrane followed by a second phase producing a highly polarized distribution (). Because our observed dynamics in both mutant cells are similar to those displayed in WT cells (; ), we suggest that Gα1 or Gα9 controlled signaling are not essential inhibitory mechanisms for cAR1-mediated gradient sensing. We speculated that the inverted PH-GFP translocation may be induced by a reapplied cAMP gradient. shows this experiment. A WT cell expressing PH-GFP was first equilibrated in a cAMP gradient to achieve the polarized state. After a withdrawal of the cAMP gradient at time 0 s, PH-GFP returned to cytosol. At 81 s, the identical gradient was reapplied (). In response to this second gradient, the cells exhibited a much stronger transient accumulation of PH-GFP in the back than in the front (). shows the comparison of normalized increase in cAMP concentration delivered by the second gradient in the front and back regions of the cells; and shows the maximal increase in PH-GFP in the front and back regions during the inverted responses ( = 8). Our results show that, despite the cAMP stimulus being higher in the front than in the back, the cells responded only in the back regions. Thus, after a withdrawal of the gradient, the cells displayed an asymmetrical refractory period. During this period, PIP initially accumulated in an intracellular gradient that had the opposite direction of the external cAMP gradient. The asymmetrical refractory period could be detected in some cells for more than four minutes after the withdrawal of the previous gradient (Fig. S3, available at ). It was previously reported that cAMP receptor-mediated PI3K activation consists of two layers in chemotaxing cells. First, free Gβγ activates Ras that stimulates a small amount of preexisting, membrane-associated PI3K. The resulting actin polymerization leads to recruitment of additional PI3K from cytosol to the membrane, thereby increasing the amount of active PI3K (). In Latrunculin-treated cells, PI3Ks were uniformly distributed around the membrane of the cells even when they were exposed to the cAMP gradient (; unpublished data). Therefore, under our experimental condition, we monitored the spatiotemporal regulations of PI3K activity without complications from the second layer of actin-dependent PI3K recruitment. In addition to the signaling pathway leading to PI3K activation, the cAMP receptor also regulates another pathway mediating the redistribution of membrane-bound PTEN, which is important for the proper directional response of PIP. In cells, PH-GFP was still able to accumulate in the front when the cells were exposed to a cAMP gradient (; ). cells were broader than those formed in WT cells (, E and F; at time 0), as previously described (; ). Furthermore, the directions of the crescents, unlike those in WT cells, did not always perfectly point to the direction of the gradient (; at time 0). These results indicate, as expected, that a cAMP gradient-induced PTEN redistribution ensures the PIP response in the restricted front region, and this directional response was not precise without PTEN. However, after a withdrawal of the cAMP gradient, PH-GFP returned to cytosol. cells (), indicating that a cAMP gradient-induced asymmetrical inhibition occurred in the absent of PTEN. Collectively, our results suggest that the previous gradient induced an asymmetrically distributed and locally controlled inhibition and this localized inhibition acts on the signaling pathway between free Gβγ to PI3K. We examined the temporal appearance and disappearance of the gradient-induced asymmetrical inhibition (; Fig. S4, B and C, available at ). We found that a brief gradient stimulation of ∼50 s was not sufficient to induce an inverted PH-GFP response (, A and B; Fig. S4 B). Thus, exposure to a stable gradient for ∼2 min is needed to establish an asymmetrical inhibition. Furthermore, cells that were removed from a gradient for 6 min and rechallenged with either uniform cAMP stimulation or a cAMP gradient displayed a noninverted PH-GFP translocation response as in naive cells (, C and D; Fig. S4 C), indicating that asymmetrical inhibition disappears within 6 min after the gradient is removed. The existence of inhibitory components in GPCR-mediated chemosensing has been proposed for more than thirty years, but the molecular mechanisms are still unknown and, thus, cannot yet be visualized directly. Here, we report insights into the temporal and spatial aspects of the inhibition based on measurements of the spatiotemporal dynamics of known components of the gradient sensing machinery aided by our computational modeling study. We have constructed a quantitative model for cAR1-mediated signaling network (). The model, which includes receptor-mediated and locally controlled inhibitory mechanisms that regulate PI3K and PTEN (B), simulates experimentally determined dynamics of receptor activation, G protein dissociation, PTEN membrane localization, and PIP accumulation (). For example, in response to a uniform cAMP stimulus, the model generates a transient PIP response that quickly returns to the resting stage (adaptation). When exposed to a cAMP gradient, a cell generates a steeper PIP gradient by initially inducing a PIP increase followed by a PIP decrease around the membrane, and then producing a highly polarized distribution of PIP in 120 s (amplification) (Fig. S5, available at ). During the amplification process, the membrane-bound PTEN gradually translocates from the front to the back, while the amount of the membrane-associated PI3K remains the same around the membrane (). Temporal changes in PI3K activity in the front and back, which cannot be directly visualized, have been simulated by the model based on dynamics of PIP and membrane-bound PTEN (; Fig. S5). Previous study and our measurements indicated that when a cell reaches the “polarized” steady state, the amount of PI3K in the front is almost equal to that in the back (). PIP is at a higher steady-state level in the front. However, this level does not continue to increase (ΔPIP/Δt = 0) in spite of a lower level of PTEN. At the same time, in the back, a higher level of PTEN does not result in a continuous decrease in the PIP level (ΔPIP/Δt = 0). Two possible models may explain different steady states of PIP in a polarized cell. First, PI3K activity is stronger than that of PTEN in the front, and PIPs are continually produced. The PIP level remains steady in the front because it diffuses fast enough to be degraded by PTEN that is enriched in the back, which is expected from models containing only globe inhibition mechanisms (; ; ). Second, balances between the activities of PI3K and PTEN have been reached in both the front and back, and the balances are achieved by a stronger inhibition of PI3K activity in the front, which have been proposed in our model that includes local inhibition mechanisms (; ). Because different proposed mechanisms could lead to high chemotactic sensitivity in theory (; ; ; ; ; ), we designed experiments to determine which inhibitory mechanisms are likely used in GPCR-mediated chemosensing. In this study, we revealed spatiotemporal features of an inhibitory process that acts locally on the activation pathway between Gβγ and PI3K. There is a general agreement that inhibition increases and decrease slowly in response to the changes of cAMP receptor occupancy (; ; ; ). Several lines of evidence are consistence with this notion. We showed that a series of cAMP short pulses induce multiple transient PIP responses without detectable refractory periods, suggesting that a quick increase and decrease of cAR1 receptor occupancy immediately turns on and off the excitation process leading to PIP production but does not significantly elevate the slower inhibition process (). We previously reported that a sustained cAMP stimulation induces persistent G protein dissociation while the PIP increases transiently, returning to basal levels within a minute (). Thus, the inhibition that is responsible for the adaptation is most likely caused by an increase in the level of negative regulators controlling the signaling components other than cAR1 and G proteins around the membrane. When a cAMP stimulation was rapidly removed from an adapted cell, the G proteins reassociated and PTEN returned to its prestimulus, membrane-associated state in about one minute and could be fully reactivated by another cAMP stimulation (, and ). When a cAMP gradient was removed from a polarized cell, there was a short period of time during which another cAMP stimulation triggers G protein dissociation and PTEN translocation in both the front and back but induces PIP responses only in the back of the cell (). This suggests that inhibitors that are more abundant in the front block transmission of activating signals from Gβγ to PI3K (). The relatively slower recovery of the responsiveness in PIP production in the front of the cell revealed that the inhibitory effect diminished slowly. cells indicated that the recruitment of inhibitors does not depend on PTEN (). reported that cells that were stimulated with a sustained uniform cAMP field did not result in a clear decrease in PIP production to another cAMP stimulation, suggesting that the recovery period is very short in a cell that has adapted to a uniform cAMP concentration. It is possible that a high level of the inhibitor is only induced by a cAMP gradient at the front of a cell where the PIP level is high but not by a uniform cAMP around the cell membrane where the PIP level remains low. We can only speculate on this point before the putative inhibitors in GPCR-mediated chemosensing network are identified. The inhibition has been assumed to be “global” or uniformly distributed throughout the plasma membrane even when a cell is exposed to a cAMP gradient (; ; ; ; ). Our findings demonstrate that the concept of a purely global inhibition cannot be reconciled with the observed spatial distributions of some inhibitory mechanisms. The inverted PIP response upon restimulation indicates that a sustained cAMP gradient induces an asymmetrically distributed inhibition that acts on the signaling pathway between G protein and PI3K (). This inhibition is stronger in the front of the cell. The spatiotemporal features of the inhibition can shed light on unknown molecular mechanisms. Based on the fast-diffusive-inhibition models, small molecules, such as Ca or cGMP, were suggested to be candidate inhibitors, which have not been verified by experiments. The “local excitation and global inhibition” model assumes the presence of a negative regulator, and suggests that it is likely to be PTEN. Based on our detailed spatiotemporal dynamics of PTEN and PIP, our computational model showed that PTEN alone cannot fully explain the experimentally determined dynamics (). We proposed, in addition to PTEN, other inhibitory mechanisms that may involve reversible modifications of components in the pathway from free Gβγ to Ras and then to PI3K. Previous studies in mammalian GPCR signaling indicated several inhibitory components. After GPCR activation, free Gβγ dimers interact with the receptor-associated kinase GRK2, blocking Gβγ signaling (). GPCR activation can also induce a translocation of a RasGAP, which binds to PIP, to inner membrane deactivating Ras thereby inhibiting PI3K (). In , it has been shown that a sustained cAR1 activation, which triggers a persistent G protein dissociation, induces a transient activation of RasG, which activates PI3K (). The transient nature of RasG activation is consistent with the idea that the cAR1 activation also recruits inhibitors to the membrane to shut down signals from free Gβγ to Ras activation. Our computational model is able to simulate the observed spatiotemporal dynamics of known components in adaptation and in gradient sensing by including these putative inhibitor(s) (). Therefore, we propose that the inhibition process is performed by these negative regulators acting locally on the PI3K signaling branch and those on PTEN branch, which act in concert to control the spatiotemporal dynamic of PIP around the cell membrane. Future studies are needed to identify inhibitors involved in the GPCR-mediated chemosensing network. , gα1 and gα9 cells expressing PH-GFP were developed to the chemotactic stage. Cells were plated on a 1-well chamber for the microinjector delivered cAMP stimulation (Nalge Nunc International), allowed to adhere to the cover glass for 10 min, and then covered with additional DB buffer. Live cells were imaged using a Zeiss Laser Scanning Microscope, LSM 510 META, with a 40× NA 1.3 DIC Plan-Neofluar objective. To monitor cAMP and PH-GFP, PTEN-GFP, PI3K1-GFP cells were excited with two laser lines, 488 nm for GFP and 543 nm for Alexa 594, a water-soluble fluorescence dye. Images were simultaneously recorded in three channels. Channel one: fluorescent emissions from 505–530 nm for GFP (green); channel two: emissions from 580–650 nm for Alexa 594 (red). The temporal-spatial intensity changes of Alexa 594 and cells expressing PH-GFP, PTEN-GFP, or PI3K1-GFP were directly imaged using a confocal microscope with Z-axis resolution of ∼2 μm. Fluorescence intensities of Alexa 594 and GFP within the focal plane were simultaneously recorded in two different channels. To establish a steady gradient, we set an external supply pressure to 70 hPa (Femtojet and micromanipulator 5171; Eppendorf) to ensure the injection of a constant and small volume of cAMP and Alxea 594 into a one well chamber. Under this condition, a stable gradient was established within 100 μm around the tip of the micropipette. To suddenly expose a cell to a stable gradient, a micropipette filled with a mixture of cAMP and 0.1 g/μl Alexa 594 linked to a FemtoJet was positioned 1,000 μm away from the cells, and then was quickly moved to a position within 100 μm to the cells. During the experiments, we only changed the distance between the micropipette and the cells. The speed of the movement determines how fast a stable gradient can form around a cell (). To withdrawal a gradient, the micropipette was quickly moved away from a cell. Using a spectral confocal fluorescence microscope (LSM510 META), we measured intensity decrease of acceptor (YFP) and increase of donor (CFP) in response to stimuli. We monitored intensity changes of CFP (donor) and (YFP) acceptor following a stimulation using a time-lapse acquisition of Lambda Stacks. The cells were excited with a 454-nm laser line and the spectral emissions in each pixel of the fluorescence images were simultaneously recorded in 8 channels, each with a 10-nm width, from 464 to 544 nm. To separate multi-fluorescence signals, each of the fluorescence images was collected using Lambda Stack acquisition. The spectral emissions of fluorescence images were simultaneously recorded in a CHS-1 from 464 nm to 544 nm. The spectra of the cells expressing CFP, YFP or GFP only were obtained and used as the references for the Linear Unmixing Function. The digitally separated images of CFP and YFP of the G cells, and GFP of the PH cells were obtained. The intensities of each fluorophore in the regions of interest in the time-lapse experiments were measured, normalized, and expressed as a function of time in responses to cAMP stimulations, using the software of LSM510 META (). Images were processed and analyzed by the LSM 510 META software, and converted to TIFF files by the Adobe Photoshop software. All frames of any given series were processed identically. Selected frames of the series were assembled as montages using Photoshop 7.0. Quantification of fluorescence intensities of Alexa 594, GFP, CFP, and YFP in the regions of interest was performed using the LSM 510 META software. Fig. S1 shows inverted PIP responses. Fig. S2 shows FRET measurement of G protein dissociation and association and redissociation. Fig. S3 shows reapplied a cAMP gradient induced an inverted PH-GFP membrane translocation. Fig. S4 A shows Gα9 null cells detect cAMP gradient normally. Fig. S4 B and C show kinetics of the formation the asymmetrically distributed inhibition. Fig. S5 shows PI3K activity, membrane-bound PTEN and the resulting dynamics of PIP in a cell when it is exposed to a cAMP gradient in a computer simulation and a schematic representation of the signaling network that describes spatiotemporal changes. Videos 1 and 2 show uniformly applied cAMP stimulation triggered inverted PH-GFP translocation. Video 3 shows simultaneously measurement of G protein activation in the front and back of a cell and the inverted PH-GFP response. Video 4 shows dynamics of PTEN in a cell upon a withdrawal of a cAMP gradient and then reapplied the gradient. Videos 5 and 6 show a cAMP gradient induced the inverted PH-GFP membrane translocation. Online supplemental material is available at .

To colonize new areas and establish cell masses sufficient for further development, both normal progenitor cells and malignant tumor cells must be able to migrate and proliferate. There is evidence to suggest that cells may not be able to engage in both of these activities simultaneously. In the rat brain, invading glioma cells are observed to halt their migration along blood vessels during periods of mitosis (), mimicking the saltatory pattern of normal glial progenitor migration (). Correspondingly, the most highly invasive glioma cells in human tumors are often found to exhibit the lowest rates of proliferation and vice versa (). We have investigated the role of the NG2 proteoglycan in regulating the choice between glioma cell proliferation and motility. NG2 is expressed by a variety of immature progenitor cell types and also by several types of tumors (). Several studies suggest a role for NG2 in promoting the proliferation and motility that are characteristic of both normal progenitor cells and malignant tumor cells (; ; ; ). As a membrane-spanning molecule, NG2 affects proliferation and migration via interactions with both extracellular and intracellular binding partners. The NG2 ectodomain has the ability to sequester growth factors and bind to growth factor receptors (; ; ), influence the processing of kringle domain proteins (; ), interact with extracellular matrix ligands such as collagen VI (; , ), and form signaling complexes with α3β1-integrin and galectin-3 (; ). The cytoplasmic domain of NG2 is involved in activation of the Rho family GTPases Rac and Cdc42 (; ; ) as well as in anchorage via the PDZ-containing scaffolding proteins MUPP1 and GRIP1 (; ). Although NG2 exhibits some signal-transducing capabilities of its own (; ; ), its ability to enhance signaling by growth factor receptors (; ) and β1-integrins (; ; ; ) greatly expands the proteoglycan's scope of action. Posttranslational modifications of NG2 provide an important means for regulating its interaction with extracellular and cytoplasmic binding partners. We have reported that PKCα-mediated phosphorylation of Thr in the NG2 cytoplasmic domain triggers the redistribution of NG2 from apical microprocesses to lamellipodia on the leading edge of the cell. This molecular redistribution is accompanied by enhanced cell motility (). This has led to our current exploration of the extracellular signal–regulated kinase (ERK)–mediated phosphorylation of NG2 at Thr and the contrasting consequences of the two respective phosphorylation events on proliferation and motility. The Thr residue phosphorylated by PKCα is one of several Thr residues in the NG2 cytoplasmic domain (; ). Analysis of the cytoplasmic domain of rat NG2 for sequences involved in protein–protein interaction (Scansite 2.0) identified the amino acid residues 2,278–2,290 as a D domain, which is postulated to be a docking site for ERK (; ). shows the alignment of the NG2 D-domain sequence with that of other proteins with known ERK-docking motifs. Notably, NG2 also possesses a potential proline-dependent phosphorylation site (RTPNP; ) that conforms to the minimal recognition sequence (S/T)P required for phosphorylation by ERK (). The residues surrounding Thr are conserved in the human (), chimpanzee (GenBank/EMBL/DDBJ accession no. ), mouse (), and dog (GenBank/EMBL/DDBJ accession no. ) homologues of NG2, which is suggestive of the functional importance of this region. To detect the association between NG2 and ERK, the proteoglycan was immunoprecipitated from NG2-transfected U251 cells (U251/NG2). Immunoblotting for phosphorylated ERK (P-ERK) reveals that serum-starved U251/NG2 cells contain low levels of P-ERK associated with NG2 (). When serum-starved cells are treated with 10% serum, increased coimmunoprecipitation of P-ERK with NG2 is observed. Pretreatment of cells with the MAPK kinase (MEK) 1/2 inhibitor U0126 largely blocks the P-ERK–NG2 interaction, suggesting that the activation of ERK is required for its association with NG2. To directly test the ability of ERK to phosphorylate NG2, we used active recombinant ERK in conjunction with γ-[P]ATP to carry out in vitro phosphorylation of GST fusion proteins containing the cytoplasmic regions of wild-type NG2 (GST-NG2c) and the Thr point mutants GST-NG2c-T2256E, -NG2c-T2265E, -NG2c-T2278E (), and -NG2c-T2314E. Results of the phosphorylation reactions were analyzed by two-dimensional mapping of tryptic phosphopeptides. The phosphopeptide maps of the GST-NG2c-T2265E and -NG2c-T2314E fusion proteins differ from those of the other species (). GST-NG2c-T2265E lacks the more highly charged of two major phosphopeptides derived from the wild-type fusion protein (spot 1), whereas the more neutral phosphopeptide (spot 2) is missing from GST-NG2c-T2314E. Simultaneous mutation of both of these Thr sites leads to the disappearance of all phosphopeptides from the GST-NG2c-T2265E/T2314E map, confirming the ability of recombinant ERK to phosphorylate the NG2 cytoplasmic domain at positions Thr and Thr in vitro. In light of artifacts that may occur in vitro, caution must be used when interpreting the phosphorylation data in (; ). Therefore, we examined ERK phosphorylation of full-length NG2 in the context of living cells using two different motif-specific antibodies to detect Thr phosphorylation. The first antibody recognizes the sequence xTpx(K/R) found at Thr, whereas the second recognizes the xTpP motif found at Thr. The specificity of these antibodies in the context of the NG2 cytoplasmic domain was confirmed via immunoblotting of the GST-NG2c fusion protein that had been phosphorylated in vitro by nonradioactive ERK or PKCα-catalyzed reactions (unpublished data). To stimulate the ERK-mediated phosphorylation of full-length NG2 in living cells, we treated U251/NG2 cells with PDGF-BB. This ligand activates both the α and β isoforms of the PDGF receptor, triggering the downstream activation of numerous intracellular signaling cascades, including both the PKC and MAPK pathways. After PDGF-BB stimulation, total cell extracts were examined to confirm the activation of PKCα and ERK. Levels of P-ERK and PKCα were found to increase within 15 to 30 min (). In parallel, NG2 was immunoprecipitated and immunoblotted with the respective phospho-Thr antibodies, revealing large increases in the quantity of phosphorylated NG2 detected by both the anti-xTpP and -xTpx(K/R) antibodies (). Similar patterns of NG2 phosphorylation were detected by these same phosphospecific antibodies in A375 human melanoma cells (). Therefore, phosphorylation of the proteoglycan is not an artifact of its overexpression but a normal physiological consequence of PDGF-BB treatment. It was initially unclear whether NG2 phosphorylation detected by the xTpx(K/R) antibody was caused by the ERK-mediated phosphorylation of Thr observed in vitro or by the reported PKCα-mediated phosphorylation of Thr (). To resolve this issue, U251/NG2 cells were pretreated either with the MEK1/2 inhibitor U0126 or the PKCα inhibitor Gö6976 before the addition of PDGF-BB. U0126 inhibition of ERK activation did not alter the level of xTpx(K/R) phosphorylation (). In contrast, this phosphorylation was blocked by treatment with Gö6976. These data suggest that the PDGF-BB–induced xTpx(K/R) phosphorylation of NG2 is not caused by ERK phosphorylation at Thr but instead by PKCα-mediated phosphorylation at Thr. Thus, Thr is unlikely to be an ERK phosphorylation site in NG2 expressed in living cells. The substantial increase in PDGF-BB–induced xTpP motif phosphorylation implicates Thr as an in vivo NG2 phosphorylation site (). This phosphorylation is not observed in the presence of U0126 (), establishing ERK as the likely mediator of Thr phosphorylation. Interestingly, PDGF-BB–dependent ERK-mediated Thr phosphorylation is not observed when PKCα activation is blocked by Gö6976 (), suggesting that PKCα is a key upstream effector of ERK activation in these cells. To further establish that Thr is the sole site of ERK-mediated NG2 phosphorylation in living cells, cells transfected with full-length NG2 constructs containing individual T→V substitutions (NG2-T2256V, -T2265V, and -T2314V) were stimulated with PDGF-BB in the absence or presence of the U0126 and Gö6976 inhibitors followed by immunoprecipitation of the mutant NG2 proteins. With the exception of the NG2-T2314V mutant, phosphorylation was detected in all NG2 species by blotting with the anti-xTpP antibody (), further establishing Thr as the site of ERK phosphorylation. In addition, PDGF-BB treatment induced heavy xTpx(R/K) phosphorylation of the NG2-T2265 and -T2314 mutants but failed to phosphorylate an xTpx(K/R) site in the NG2-T2256V mutant. This identifies Thr as the sole PDGF-BB–dependent xTpx(K/R) phosphorylation site in NG2. Thus, by activating both PKCα and ERK downstream of PDGF receptors, PDGF-BB stimulation results in the phosphorylation of NG2 at both Thr and Thr. As a final demonstration that ERK phosphorylates only Thr in cells, we transfected U251/NG2 cells with a constitutively active MEK mutant (MEK-DD) (; ). As expected, an increase in ERK phosphorylation was observed in U251/NG2/MEK-DD cells compared with U251/NG2 cells, whereas PKCα phosphorylation was unchanged (). Correspondingly, when NG2 was immunoprecipitated from the two sets of cells and immunoblotted with the respective phospho-Thr antibodies, the phosphorylation of NG2 at Thr was increased in U251/NG2/MEK-DD cells relative to that seen in U251/NG2 cells, whereas Thr phosphorylation was unaffected. In U251/NG2 cells, the proteoglycan is present on microprotrusions from the apical cell surface (). It is noteworthy that these microprotrusions are infrequently seen on parental U251 cells (not depicted) and on NG2-negative cells in the U251/NG2 population (; arrows). In contrast, the mutant NG2-T2256E species that mimics Thr phosphorylation () is localized to broad lamellipodia (). As suggested by its coimmunoprecipitation with NG2 (), phospho-PKC is also localized with NG2-T2256E in lamellipodia (). Furthermore, in both microprotrusions and lamellipodia, NG2 labeling is colocalized with β1-integrin labeling (). The α3-integrin subunit has an identical localization in these cell types (unpublished data), which is consistent with our observation that α3β1 is a predominant integrin heterodimer in U251 cells (). Unlike NG2-T2256E, an NG2-T2314E species designed to mimic Thr phosphorylation is largely present in β1-integrin–positive arrays of apical microprotrusions reminiscent of those seen on U251/NG2 cells. However, in U251/NG2-T2314E transfectants, these structures are more highly clustered or bundled than in U251/NG2 cells (). Cells expressing the NG2-T2314V variant resemble parental U251 cells in that they rarely display apical protrusions (). Collectively, these data suggest that Thr phosphorylation not only directs NG2 localization to apical microspikes but may actually stimulate the formation of these protrusions. We propose that the level of Thr phosphorylation found in U251/NG2 cells under basal conditions () is sufficient to induce detectable microspike formation. Phosphorylation of additional NG2 molecules at Thr, as mimicked by the NG2-T2314E mutant, appears to result in aggregation of the microprotrusions. We were unable to detect a consistent signal with phospho-ERK antibody in any of these cell types and, therefore, were not able to determine whether phospho-ERK localizes with NG2 or NG2-T2314E in apical microspikes. NG2 expression has been shown to have a stimulatory effect on cell motility (; ; ). We used transwell migration assays to study the effect of PDGF-BB on the motility of U251 cells expressing a variety of NG2 species (). Under basal conditions (i.e., no PDGF-BB), no reproducible differences in cell migration are observed between parental U251 cells and any of the NG2 transfectants except for the NG2-T2256E transfectant, which exhibits strongly enhanced migration. This supports our previous finding that actual or mimicked phosphorylation at Thr promotes cell motility (). Notably, glutamic acid substitution at Thr does not have such an effect on basal cell motility. Although PDGF-BB treatment has a small stimulatory effect on the migration of parental U251 cells, the magnitude of this stimulation is enhanced in U251/NG2 cells, which migrate at a level comparable with the high basal rate seen with the NG2-T2256E transfectants (). Thus, although PDGF-BB treatment is capable of stimulating the motility of U251 cells via NG2-independent mechanisms, the expression of NG2 permits an additional increase in PDGF-BB–induced motility via PKCα-mediated phosphorylation at Thr. This NG2-dependent enhancement of PDGF-BB stimulation is also seen in the other mutant transfectants, with the exception of NG2-T2256V and -T2314E, which respond to PDGF-BB at the same modest level as parental U251 cells. For the T2256V transfectant, this loss of responsiveness is explained by the inability of PKCα to phosphorylate the mutated site. Regarding the T2314E transfectant, it is critical to note that the T2314V mutation does not exhibit this same depressed response to PDGF-BB. Therefore, the inhibitory effect of the T→E substitution is not the result of a block of Thr phosphorylation. Instead, the ability of the mutation to mimic ERK-mediated phosphorylation at Thr must counteract, to some degree, the stimulatory effect on motility caused by NG2 phosphorylation at Thr. There are several reports of interactions between NG2 and β1-integrins, including α4β1 and α3β1 (; ; ; ). We have documented the ability of exogenous NG2 to activate α3β1-integrin signaling on the endothelial cell surface (). Because membrane-bound NG2 might also be able to activate β1-integrin signaling in a cis manner in U251/NG2 cells, we repeated the cell motility studies in the presence of the β1-integrin–blocking antibody P4C10 (). This antibody substantially inhibits both the PDGF-BB–enhanced motility of U251/NG2 cells and the spontaneous motility of NG2-T2256E transfectants. An antibody against the α1-integrin subunit, which is present at very low levels on these cells, provides a negative control for possible nonspecific antibody effects. These results suggest that the enhanced motility of U251/NG2 cells in response to PDGF-BB is a β1-integrin–dependent process whose activation is at least partially dependent on the phosphorylation of NG2 at Thr. Because previous in vitro and in vivo studies have demonstrated the involvement of NG2 in cell proliferation (; ; ), we investigated the involvement of ERK-mediated NG2 phosphorylation in this process. Under basal conditions in the absence of exogenous PDGF-BB over the course of 48 h, the number of NG2-expressing U251 cells is ∼50% greater than that of parental U251 cells (). This is true for both conventional U251/NG2 transfectants and U251 cells with adenovirally expressed NG2 (). To determine whether the increased cell number is caused by increased proliferation or improved survival in the NG2-expressing cells, apoptosis levels were assessed via flow cytometry (see Materials and methods). The low apoptosis levels we observed (1.2% for U251 cells and 2.3% for U251/NG2 cells) cannot explain the difference in cell expansion seen in , indicating that NG2 expression enhances cell proliferation rather than cell survival under low serum conditions. Interestingly, the constitutively active MEK-DD construct further enhances the proliferation of U251/NG2 cells (), suggesting the involvement of the ERK-dependent phosphorylation of NG2 at Thr (). To further evaluate this possibility, we studied the effect of PDGF-BB treatment on cell proliferation (). PDGF-BB has little effect on NG2-negative U251 cells and actually decreases the proliferation of U251/NG2 cells. To determine the respective effects on proliferation of NG2 phosphorylation at Thr versus Thr, we performed proliferation studies on our panel of mutant NG2 transfectants. Under basal conditions, the T2256V and T2265V transfectants exhibit enhanced proliferation (relative to U251 cells) similar to that of the wild-type NG2 transfectant. In contrast, T2314V transfectants have the nonenhanced proliferation characteristics of NG2-negative U251 cells. Together with the enhancement produced by MEK-DD, these findings suggest that basal levels of ERK-mediated phosphorylation at Thr are responsible for NG2-dependent proliferation under nonstimulatory conditions. After treatment with PDGF-BB, decreased cell proliferation is observed for all valine-substituted NG2 transfectants except for the T2256V variant, in which PDGF-BB enhances proliferation. This behavior may be explained by an inhibitory effect of phosphorylation at Thr by PKCα, which is relieved by the NG2-T2256V mutation. These possibilities were further assessed by examination of the T→E substitution variants. By mimicking phosphorylation at Thr, the NG2-T2256E transfectant reduces both basal and PDGF-BB–stimulated cell proliferation to the level of parental cells, which is in agreement with the hypothesis that the Thr phosphorylation event is inhibitory to cell proliferation. In contrast, by mimicking phosphorylation at Thr, the NG2-T2314E transfectant exhibits increased proliferation beyond that seen with any other species, confirming the important role of Thr phosphorylation in promoting cell proliferation. Treatment of U251/NG2-T2314E cells with PDGF-BB partially overcomes this stimulatory effect, most likely as a result of the negative effect of Thr phosphorylation. In light of the effect of β1-integrin–blocking antibody on U251/NG2 cell motility, we repeated the proliferation studies in the presence of this same antibody (). This antibody completely inhibits the increased proliferation seen in U251/NG2 cells under basal conditions. The effect of the β1 antibody is somewhat less striking in the case of NG2-T2314E transfectants but is still significant (P < 0.05) compared with the lack of effect of the control α1 antibody. These results indicate that the NG2-dependent increase in proliferation seen in U251/NG2 cells is caused by enhanced β1-integrin signaling in response to NG2 phosphorylation at Thr. We sought additional evidence for NG2-dependent β1-integrin signaling through use of the HUTS-21 monoclonal antibody, which binds to the activated form of the β1 subunit (; ). Because we were unable to detect HUTS-21 labeling on fixed cells, we used living cells for these experiments. Although this resulted in the loss of much of the morphological detail seen in , several interesting patterns of β1-integrin activation were nevertheless revealed. Whereas HUTS-21 binding is detectable only at very low levels in U251/NG2 cells (), labeling is greatly increased along lamellipodial edges in NG2-T2256E transfectants (), corresponding to the site of NG2 and β1-integrin localization in these cells (). HUTS-21 labeling of NG2-T2314E transfectants is also increased () but is more evenly distributed across the cell surface, which is consistent with the localization of NG2 and β1-integrin to apical microprotrusions (). In contrast, HUTS-21 binds to the NG2-T2314V () and -T2256V (not depicted) transfectants at the basal level seen in U251/NG2 cells, supporting the idea that phosphorylations at Thr and Thr trigger NG2-dependent β1-integrin activation. As shown by labeling with the β1 antibody TS2/16 (), each of the cell types examined express comparable levels of total β1-integrin, although integrin distribution is skewed toward lamellipodia in the case of U251/NG2-T2256E cells. Integrin activation is known to trigger the assembly of signaling complexes containing nonreceptor protein tyrosine kinases such as focal adhesion kinase (FAK; for review see ). Accordingly, we found high levels of tyrosine phosphorylation colocalized with NG2 in the lamellipodia of U251/NG2-T2256E cells () and in the microprotrusions of U251/NG2-T2314E cells (). Consistent with the scarcity of microprotrusions in the absence of NG2 (), this type of phosphotyrosine labeling is not seen in cells in the U251/NG2-T2314E culture that have lost NG2 expression (; arrows). In these cells, phosphotyrosine labeling is faintly visible in focal adhesion plaques, which are completely obscured by the labeling of microprotrusions in NG2-positive cells. Phosphorylation and dephosphorylation have proved to be among the most versatile and functionally important types of posttranslational modifications (; ). In the case of NG2-dependent mechanisms, PKCα-mediated phosphorylation of NG2 at Thr triggers increased cell motility via a mechanism that involves the translocation of NG2 to sites in leading edge lamellipodia of motile cells (). This profound effect of phosphorylation on NG2 localization/function prompted us to investigate the existence of additional NG2 phosphorylation sites capable of influencing other aspects of cell biology. Sequence analysis allows us to identify two previously unrecognized features of the NG2 cytoplasmic domain: a putative D domain–docking site for ERK (residues 2,278–2,290) and a potential site for ERK-mediated phosphorylation at Thr. Immunoprecipitation/immunoblotting results confirm the existence of a physical interaction between NG2 and activated ERK. In addition, purified recombinant ERK is able to phosphorylate the isolated NG2 cytoplasmic domain at Thr and Thr. In extending these studies to full-length NG2 expressed endogenously by A375 melanoma cells or by transfection in U251 glioma cells, we used PDGF-BB stimulation to activate both PKCα and ERK. This provides a more physiologically relevant means of stimulating phosphorylation than the PMA treatment we previously used to identify PKCα-mediated modification at Thr (). We evaluated PDGF-BB–induced phosphorylation of NG2 by immunoblotting with phosphospecific antibodies against the xTpx(K/R) motif present at Thr and Thr and against the xTpP motif present at Thr. The use of specific inhibitors of PKCα and ERK coupled with analysis of NG2 species with valine substitutions at the putative phosphorylation sites reveals that PKCα-mediated phosphorylation occurs at Thr, whereas ERK-mediated phosphorylation occurs only at Thr. Contrary to the results obtained in vitro with recombinant ERK treatment of the isolated cytoplasmic domain, Thr is not used as a site for ERK phosphorylation in full-length NG2 expressed by living cells. This is more in keeping with predictions based on NG2 sequence analysis because the residues in the immediate vicinity of Thr do not represent a canonical ERK phosphorylation motif. An important mechanistic observation from the experiments with NG2-transfected U251 cells is that ERK activation in response to PDGF-BB is highly dependent on the activity of PKCα. This seems somewhat surprising because conventional wisdom would suggest that PDGF receptors should also be able to activate ERK via a Son of sevenless–Ras–Raf–MEK-dependent pathway (for review see ). However, the ability of the PKCα inhibitor Gö6976 to inhibit both the PKCα-catalyzed phosphorylation of Thr and the ERK-catalyzed phosphorylation of Thr does not provide evidence for an important contribution of the growth factor–activated Son of sevenless–Ras pathway to ERK activation in U251 cells. A similar dependence of ERK on PKC activity has been observed in PDGF stimulation of smooth muscle cells (; ). The consequences of these findings in U251 cells are twofold: PDGF-BB treatment inevitably activates both PKCα and ERK, thus leading to NG2 phosphorylation at both Thr and Thr, and we cannot inhibit PKCα without also affecting ERK activity. In this regard, the constitutively active MEK-DD construct has provided an effective means for activating ERK independent of PKCα activation, enabling us to restrict NG2 phosphorylation to Thr. Having established the existence of two distinct types of Thr phosphorylation sites in the NG2 cytoplasmic domain, we addressed the effects of these two phosphorylation events on cell behavior. Use of the transwell cell motility assay allows us to confirm our previous report () that both the actual and mimicked phosphorylation of NG2 at Thr leads to enhanced cell migration. Not surprisingly, PDGF-BB increases the motility of parental U251 cells via mechanisms that are independent of NG2. Nevertheless, the presence of NG2 further enhances the response to PDGF-BB. The fact that this is caused by NG2 phosphorylation at Thr is demonstrated by the loss of enhanced PDGF-BB responsiveness in NG2-T2256V transfectants and by the increased motility of NG2-T2256E variants even in the absence of growth factor. Notably, the other NG2 transfectant that fails to exhibit enhanced NG2-dependent motility in response to PDGF-BB is the NG2-T2314E variant. Because the NG2-T2314V variant does not exhibit this loss of function, the depressed motility of NG2-T2314E transfectants cannot be caused by the blockage of Thr phosphorylation. Instead, the defect must be the result of mimicked phosphorylation at this site, indicating that phosphorylation at Thr counteracts the stimulation of motility produced by the phosphorylation of Thr. We were also able to identify effects of NG2 phosphorylation on cell proliferation. In the absence of growth factor stimulation, U251 cells expressing NG2 exhibit a greater rate of cell proliferation than parental U251 cells. This phenomenon is observed in both conventionally transfected and adenovirally transformed cells, demonstrating that the behavior is independent of the means of NG2 expression. Elevated basal rates of proliferation are also observed in each of the valine-substituted NG2 variants with the exception of NG2-T2314V, suggesting that basal levels of phosphorylation at Thr in nonstimulated cells () may be responsible for the increased proliferation of NG2-expressing U251 cells relative to parental cells. Support for this idea is provided by the behavior of U251/NG2/MEK-DD transfectants and NG2-T2314E transfectants, both of which proliferate faster under basal conditions than any other species examined because of phosphorylation and mimicked phosphorylation at Thr, respectively. Interestingly, although the addition of PDGF-BB has no effect on the proliferation of parental U251 cells, it negatively affects the proliferation of U251/NG2 transfectants. This negative effect of PDGF is also seen in the valine-substituted NG2 variants with the exception of NG2-T2256V, in which proliferation is enhanced. The possibility that PDGF-induced phosphorylation at Thr is capable of reversing the stimulatory effect on proliferation caused by Thr phosphorylation is supported by the low rates of proliferation observed in NG2-T2256E transfectants, which mimic Thr phosphorylation. Our findings suggest the existence of an intriguing balance between cell proliferation and migration that can be regulated, in part, by phosphorylation of the NG2 cytoplasmic domain at two different sites. PKCα-mediated phosphorylation at Thr appears to stimulate cell motility while inhibiting cell proliferation. Conversely, ERK-mediated phosphorylation at Thr tends to block cell motility while promoting cell proliferation. In unstimulated cells, it appears that low levels of Thr phosphorylation are responsible for the increased rate of proliferation of U251/NG2 cells compared with parental U251 cells. When cells are treated with PDGF-BB, increased phosphorylation occurs at both Thr and Thr. Under these conditions, our data suggest that signals generated via Thr phosphorylation may be dominant over those resulting from Thr phosphorylation. In this context, and considering that both PDGF-induced phosphorylation events seem to depend on PKCα activation, it is logical to ask under what circumstances Thr phosphorylation would lead to increased cell proliferation. An answer may lie in the ability of non–growth factor–driven mechanisms to activate ERK independently of PKCα activation. An artificial example of such a mechanism is provided by our use of the MEK-DD construct to drive ERK activation independently of PKCα. Under these circumstances, Thr phosphorylation is achieved in the absence of Thr phosphorylation, resulting in the enhancement of cell proliferation. A more biologically relevant means of achieving Thr phosphorylation independent of Thr phosphorylation might involve the activation of ERK via a G protein–coupled receptor-dependent pathway (). Alternatively, integrin-mediated activation of the FAK–src–p130cas pathway could serve to stimulate ERK independently of PKCα (). With respect to integrins, it is noteworthy that we and others have reported the ability of NG2 to activate β1-integrin signaling (; ; ). NG2 is able to form a signaling complex with α3β1-integrin when the proteoglycan is present in soluble exogenous form or when it is expressed in cis fashion on the same cells as the integrin (). The relationship between NG2 and β1-integrins may be especially relevant to the behavior of U251/NG2 cells. Labeling with the activation-dependent HUTS-21 antibody reveals β1-integrin activation in U251 cells that express the NG2-T2256E and NG2-T2314E variants that mimic phosphorylation at Thr and Thr, respectively. In addition, a β1-blocking antibody has inhibitory effects on both the proliferation and motility induced by NG2 phosphorylation at Thr and Thr, respectively. These results are initially paradoxical because it is not immediately clear how integrin activation by NG2 might be able to stimulate proliferation in one case (response to Thr phosphorylation) and motility in the other (response to Thr phosphorylation). We suggest two possible resolutions to this paradox. First, the two phosphorylated NG2 species might differentially influence integrin signaling by interacting with different β1-integrin heterodimers or by recruiting additional distinct cytoplasmic binding partners to the NG2–β1-integrin complex. Second, differential localization of the NG2–integrin complexes may be determined by the NG2 phosphorylation pattern with the result that integrin signaling occurs in distinctly different microdomains of the cell. We have presented evidence consistent with this second alternative. Specifically, the NG2-T2256E species is localized along with β1-integrin in broad lamellipodia so that integrin activation and the resulting tyrosine phosphorylation of downstream signaling intermediates are localized to a microdomain that is critical to cell motility. In contrast, the NG2-T2314E species is colocalized with β1-integrin and elevated tyrosine phosphorylation on apical microprotrusions that appear to be dependent on NG2 phosphorylation at Thr for their formation. These observations are consistent with the concept that integrin-mediated signal transduction can activate both motility and proliferation via intermediates such as FAK and Crk-associated substrate, whose specific localization patterns determine the outcome of signaling (for review see ). These proliferation and motility results obtained with NG2-transfected U251 cells contradict, to some extent, our previous results with aortic smooth muscle cells from wild-type and NG2-null mice (). Wild-type smooth muscle cells proliferated and migrated in response to both PDGF-AA and -BB. NG2-null cells failed to respond well to PDGF-AA but had normal responses to PDGF-BB. Thus, unlike our results with U251 glioma cells, the presence or absence of NG2 did not affect smooth muscle cell responses to PDGF-BB. It seems likely to us that these apparent discrepancies reflect the primitive status of our understanding of differences in the details of signaling mechanisms that occur from one cell type to another. Several examples will serve to illustrate the complexity of the situation. We have no information about the phosphorylation of NG2 in smooth muscle cells. If PKCα and ERK are not activated by PDGF-BB in these cells in the same way we have seen in U251 and A375 cells, the phosphorylation of NG2 at Thr and Thr will not occur, and NG2 will have no influence on motility or proliferation. As a result of the proposed link between integrin signaling and PDGF receptor activation (), the NG2-dependent integrin activation described in this study could have an effect on PDGF receptor signaling. Because we have not determined how the spectrum of integrins in smooth muscle cells compares with that of U251 cells, we cannot say whether integrin–PDGF receptor or NG2–integrin interactions would be comparable between the two cell types. PDGF-BB activates signaling through both α and β receptors. We have presented evidence for an interaction between NG2 and PDGFα receptor that potentiates signaling via this receptor (; ). The relative abundance of α and β receptors has been found to differ between smooth muscle and glioma cells (; ), and we have not determined the α/β ratio in the specific cases of our two cell types. Therefore, we cannot predict the extent to which NG2 itself affects PDGF receptor signaling in either cell type. These uncertainties make it clear that our understanding of the functional role of NG2 is at an early stage and that much additional work will be required to elucidate details of the mechanisms by which NG2–integrin–PDGF receptor interactions regulate cell proliferation and motility. Gö6976 and U0126 were purchased from Calbiochem. PDGF-BB was purchased from Chemicon International, and type I collagen was obtained from Cohesion. Affinity-purified rabbit antibody against rat NG2 has been described previously (). The 9.2.27 monoclonal anti-NG2 antibody was a gift from R. Reisfeld (The Scripps Research Institute, La Jolla, CA). Rabbit or mouse antibodies specific for the phospho-Thr motifs xTpx(K/R) and xTpP were purchased from Cell Signaling Technology, as were polyclonal anti-ERK1/2 antibody and phosphospecific ERK1/2 and PKCα antibodies. Rabbit antibody against phosphotyrosine was provided by E. Pasquale (Burnham Institute for Medical Research, La Jolla, CA). Monoclonal antibody against PKCα and monoclonal HUTS-21 antibody against activated β1-integrin subunit were obtained from BD Biosciences. Monoclonal antibodies against human β1-, α3-, and α1-integrin subunits were purchased from Chemicon. Fluorescein and rhodamine-coupled second antibodies were obtained from Biosource International. NG2-negative U251MG human astrocytoma cells were maintained in DME containing 10% FCS, 2 mM glutamine, 100 IU/ml penicillin, and 100 μg/ml streptomycin sulfate. U251 cells transfected with cDNA for rat NG2 (U251/NG2) have been described previously (). NG2-positive A375 human melanoma cells and the B5 (anti-NG2) hybridoma were obtained from American Type Culture Collection. An adenoviral vector containing rat NG2 cDNA (provided by A. Nishiyama, University of Connecticut, Storrs, CT) was also used to express NG2 in U251 cells (). For this purpose, viral stocks of 10 plaque-forming units/ml were added to semiconfluent U251 cultures. Stable transfection with cDNA coding for mutant NG2 species was performed as described previously (; ). Mutant NG2 cDNA species were made using the QuikChange mutagenesis kit (Stratagene). For some experiments, U251/NG2 cells were further transfected with a pUSE vector containing the constitutively active MEK-DD mutant (). Immunoprecipitation and immunoblotting were performed as previously described (). In the case of human NG2 from A375 cells, immunoprecipitation and immunoblotting were performed with the 9.2.27 and B5 monoclonal antibodies, respectively. Cells in culture were fixed and labeled as previously described (; ). In the case of HUTS-21 labeling, living cells were stained at 37°C for 15 min with both first and second antibodies. For phospho-PKC, phospho-ERK, and phosphotyrosine labeling, 0.1% Triton X-100 was included in the incubation mix. In double-labeling studies, controls were included to rule out the cross-reactivity of second antibodies with inappropriate Ig species. Specimens were examined at room temperature using a microscope (Optiphot; Nikon) equipped for epifluorescence. Images were captured on color slide film (Fujichrome 400 ASA; Fuji) using a plan Apo 40× 1.0 NA oil immersion objective (Nikon) and scanned at 300 dots per inch resolution using a scanner (UY-S77; Sony). Images were saved as Photoshop tif files (3,072 × 2,048 pixels) and subsequently processed in Photoshop 5.0 (Adobe). GST fusion proteins containing the wild-type or mutant NG2 cytoplasmic domain were prepared as described previously (). 10 μg of fusion protein bound to glutathione–Sepharose beads were used for phosphorylation reactions. Protein-laden beads were suspended in 30 μl of kinase buffer (50 mM MES, pH 6.5, 10 mM MgCl, 1 mM EGTA, 2 mM DTT, 100 μM ATP, 10 μCi γ-[P]ATP; PerkinElmer) and incubated with 100 ng of recombinant ERK2 (Calbiochem) for 30 min at 30°C. In vitro phosphorylation catalyzed by recombinant PKCα was performed as described previously (). Samples were fractionated on 4–20% SDS-PAGE gels and transferred to Immobilon P. Labeled components were detected by autoradiography. In some cases, in vitro phosphorylation reactions were performed using unlabeled ATP. In these cases, phosphorylated bands were recognized by immunoblotting with specific phospho-Thr antibodies. After SDS-PAGE fractionation, P-radiolabeled fusion proteins were cut from dried gels and digested with TPCK trypsin (; ). Phosphopeptide mapping was performed in two dimensions on cellulose TLC plates (). After overnight incubation in DME containing 0.5% FCS, cells were harvested using enzyme-free cell dissociation buffer and seeded in 0.5% FCS medium in 96-well plates (7 × 10 cells/well) in the presence or absence of 20 ng/ml PDGF-BB. After a 48-h growth period at 37°C, the viable cell mass was determined by the addition of MTS (3-[4,5-dimethylthiazol-2yl]-5-[3-carboxymethoxyphenyl]-2-[4-sulfophenyl]-2H tetrazolium) phenozine methosulfate solution (Promega) to each well. Incubation was continued for 2 h at 37°C in the dark. Absorbance (490 nm) was then measured on a microplate reader (model 550; Bio-Rad Laboratories). 5 × 10 cells were seeded in six-well plates and maintained in DME/0.5 FCS% for 2 d. Both adherent and floating cells were collected and stained with annexin V and propidium iodide using the annexin V–FITC apoptosis detection kit (Invitrogen). 10 cells of each type were analyzed using a flow cytometer (FACSort; Becton Dickinson). Parental U251 and U251/NG2 cell populations were found to contain 1.2% and 2.3% apoptotic cells, respectively. Cell migration was examined in transwell cell culture chambers (Costar). Polycarbonate membrane inserts (6.5-mm diameter and 8-μM pores) were coated overnight by immersion in 30 μg/ml of type I collagen at 4°C. Cells grown overnight in DME containing 0.5% FCS were harvested using enzyme-free cell dissociation buffer and were resuspended in medium containing 0.5% FCS. Cells were added to the upper chamber of each well (5 × 10 cells/well) and incubated at 37°C for 16 h. Where indicated, medium in the lower chamber of the well contained 20 ng/ml PDGF-BB. After incubation, cells on the upper side of the membrane were removed with a cotton swab. The membrane was fixed, stained with DAPI, and coverslipped with Vectashield (Vector Laboratories). DAPI-positive nuclei were then counted under 200× magnification in four contiguous fields of the membrane, excluding the edges. Statistical analyses were performed using the two-tailed test.

Fibronectin (FN) is a ubiquitous and abundant ECM protein, which is secreted as a disulfide-bonded dimer consisting primarily of three types of repeating modules (I, II, and III; ; ). Distinct FN modules bind cell surface receptors such as integrins and syndecans () and ECM proteins such as collagens, fibrin, and FN itself (). FN plays an essential role during development, in physiology (tissue repair), and in disease (cancer progression and invasion, inflammation, atherosclerosis, etc.). The functions of FN are critically dependent on the assembly of the secreted FN dimers into a fibrillar network. FN assembly is a cell-driven process that occurs in a stepwise manner: the dimeric FN protein is secreted in a compact or inactive conformation; integrin binding converts FN into an active and extended dimer; the dimeric state of FN induces integrin clustering and accumulation of integrin-bound FN, which in turn allows FN–FN interactions at different sites along the FN molecule and the formation of fibrils (). The activation of FN is most efficiently induced by α5β1 integrin binding to the RGD motif in the type III-10 module along with the synergy sequence located in the adjacent type III-9 module (; ; ; ). Additional RGD binding integrins, including αvβ3 and αIIbβ3, can substitute for α5β1 in FN matrix assembly but are less able to form a dense and delicate fibrillar network in vitro (, ; ). Furthermore, it has also been reported that FN can be assembled in an RGD-independent manner through binding of α4β1 integrin to the connecting segment-1 (CS1) in the alternatively spliced V region near the C terminus of FN (). It is not clear, however, whether the CS1-dependent FN assembly occurs and is of significance in vivo. Gene ablation studies in mice confirmed that the two RGD-dependent FN assembly mechanisms induced by either α5β1 integrin or αv integrins exist in vivo and that each of them can compensate for the absence of the other. Mice lacking the α5 integrin gene, as well as mice lacking the αv integrin gene, exhibit normal assembly of FN matrix (; ). In contrast, mice carrying deletions of both the α5 and the αv integrin genes display severely compromised FN matrix formation (). Antibody and peptide inhibition studies in amphibians and chicks (; ), chemical mutagenesis in zebrafish (; ), and gene targeting in mice () identified a crucial role for the FN matrix in the development of mesoderm and mesoderm-derived structures. The mesoderm is one of the three germ layers that forms during gastrulation and is essential for morphogenesis and organ development. FN-deficient mice commence gastrulation and die at embryonic day (E) 8–8.5 with several mesodermal defects, including shortened anterior–posterior axis, disorganized notochord, absence of heart and somites, and abnormal vasculogenesis (). Mice lacking the expression of both the α5 and the αv integrins (and hence all FN-RGD binding integrins) also display arrested development at the beginning of gastrulation (). α5/αv double-null mice are arrested at an earlier stage than the FN-deficient mice. The α5/αv double mutants completely lack the mesodermal germ layer and contain only a few mesodermal cells in the developing head fold region and allantois, indicating that the two integrins must perform functions in addition to FN matrix assembly. Interestingly, mice lacking only the α5 integrin die ∼E9.5–10, and despite their ability to develop a normal FN matrix, they also have mesodermal defects. These are restricted to the posterior region of the embryo and include an arrest in posterior axis elongation, the absence of posterior somites, and vascular defects (). These observations indicate that the α5β1–FN interaction conveys essential signals in the posterior trunk mesoderm that are not compensated by αv integrins or other FN binding proteins. It is also possible, however, that the α5β1 integrin requires interaction with ECM proteins other than FN to extend the posterior trunk and develop posterior somites. In contrast, mice without αv integrin display no mesodermal abnormalities and die either of placenta defects or perinatally as a result of severe intracerebral hemorrhage, indicating that the role of the αv integrins for mesoderm formation is fully compensated by α5β1 (). To test how FN binding receptors compensate for each other during development and FN fibril formation, it would be necessary to generate double, triple, or even higher order combinations of integrin and syndecan knockout mice. However, the necessary intercross combinations cannot be performed because most integrin-null mice are recessive lethal (α4, α5, α8, and αv). Moreover, integrins and syndecans interact with multiple ECM proteins, and ECM proteins interact with multiple integrins, which would further complicate the interpretation of the phenotypes. Therefore, we decided to specifically test the role of FN-RGD binding integrins in vivo by generating mice in which the aspartic acid (D) residue of the RGD motif was replaced with a glutamate (E). Many previous studies have shown that this replacement completely inactivates this site for cell adhesion. We found that the FN mice died early in embryogenesis from multiple defects, but surprisingly they assembled an apparently normal FN matrix. Furthermore, our studies revealed a novel integrin binding site in the fifth N-terminal type I module of FN, which was essential for the assembly of the FN-RGE matrix. To test the role of the RGD motif in FN-III in vivo, we cloned part of the mouse FN gene, changed the GAC codon in exon 30 to GAG by site-directed mutagenesis, and inserted a flanked neo cassette into intron 30 (Fig. S1, A and B, available at ). The C>G nucleotide change resulted in a replacement of the aspartate (D) residue of the RGD motif with a glutamate (E) and created a novel HinfI restriction site in the mutant FN gene (Fig. S1, A and B). The DNA vector was used to transfer the mutation via embryonic stem cells into the germline of mice. The neo cassette was removed by breeding to a deleter-Cre strain. Mice were genotyped by PCR analysis of tail or yolk sac biopsies, and the knockin mutation was confirmed by sequencing of the PCR amplicon (Fig. S1, B and C). Mice heterozygous for the mutant FN allele (FN) were normal. Heterozygous mice were intercrossed to obtain homozygosity (FN). Out of 157 viable offspring, we obtained 65% heterozygous and 35% wild type but no homozygous mutant mice (). This indicates that the RGE mutation in FN represents a recessive embryonic lethal trait. To determine the exact time point of lethality, we dissected embryos derived from heterozygous intercrosses and determined their genotypes by genomic PCR using DNA extracted from either the yolk sac or embryos. We found that genotypes showed a normal Mendelian distribution until E10.5, whereas beyond E10.5, no homozygous mutant embryos with beating hearts were recovered (). At ∼E11.5, FN embryos started to disintegrate or were already partially resorbed. To search for defects and identify the cause for the lethality, we isolated whole-mount embryos from heterozygous intercrosses at different time points of development. At E8.5, the FN embryos had slightly shorter anterior–posterior axes but were otherwise indistinguishable from their wild-type or heterozygous littermates (). Both the FN embryos and wild-type littermates contained three to four somite pairs and had a well-developed head fold, neural tube, and beating heart. At E9.5, wild-type embryos were turned into the fetal position and their posterior trunks extended into a C-like shape (). In contrast, FN embryos were only partially turned and displayed a kinked neural tube and severely shortened posterior trunk without somites (). E9.5 FN embryos contained never more than 12 or 13 somite pairs, whereas wild-type littermates had ∼21 somite pairs (). The neural tubes were noticeably kinked, the limb buds were absent, and the embryonic and yolk sac vasculature was frequently dilated and sometimes ruptured, leading to blood cell leakage into the amniotic cavity (unpublished data). At E10.5, the head structures, heart, and fore limb buds of FN embryos were further developed, whereas the number of somites had not increased further (). To test whether proliferation or survival was affected by the FN-RGE mutation, we analyzed Ki67, TUNEL, and activated caspase-3 signals. Proliferation as tested by Ki67 staining was similar in control and FN embryos (unpublished data). In sharp contrast, apoptotic cell death of tail bud–derived mesodermal cells was dramatically elevated, as shown in a double staining for TUNEL () and activated caspase-3 signals (). It has become generally accepted that the RGD motif in FN is essential for FN matrix assembly (; ; ). Consequently, we expected mutation of the RGD motif to abolish fibril assembly and hamper fibril stability or function. To evaluate FN-RGE matrix assembly, we performed immunostaining and Western blot experiments. Immunostaining of parasagittal, paraffin-embedded sections from PFA-fixed E9.5 embryos revealed that FN-RGE was widely expressed. Closer inspection of different areas within the FN embryos revealed strong FN-RGE matrix surrounding the nervous tissue, somites, and large vessels and a fine, network-like distribution in the heart, developing facial, and trunk mesenchyme, within the somites and the tail bud mesoderm (). Neither the distribution nor the intensity of the FN immunosignal differed between wild-type and FN littermates (). Western blot assays of whole embryo lysates confirmed the normal expression levels of FN-RGE in FN embryos (). The apparently normal FN-RGE fibrils in FN embryos was unexpected and could be due to a cross-linking artifact induced by the PFA fixation or, alternatively, to α4β1 integrin–dependent assembly. To test whether either was the case, we immunostained nonfixed cryosections derived from E9.5 embryos for FN and α4 integrin expression. The FN staining revealed a similar distribution and amount of FN fibrils in control and FN embryos (). The α4 integrin signals were scarce and only observed in the cranial region (likely on migratory cranial neural crest cells) and in the epicardium and the proepicardial serosa of control and FN sections ( and not depicted). Importantly, FN signals did not colocalize with α4 integrin–positive cells in the cranial region or in the heart ( and not depicted). Furthermore, FACS analysis of single-cell suspensions derived from collagenase-treated control and FN E9.5 embryos identified a small α4 integrin–positive cell population of indistinguishable size and comparable α4 integrin levels (Fig. S2, available at ), indicating that expression of α4 integrin is not altered in FN mice. To examine the mechanism underlying the assembly of FN-RGE, we immortalized fibroblast-like cells from wild-type and FN embryos and established several clonal cell lines with comparable adhesion properties to FN, collagen I, laminin-111 (LM111), and vitronectin (VN; unpublished data). All clones chosen for the subsequent assays expressed similar levels of β1, β3, αv, and α5 integrins on their cell surface (Fig. S3 A, available at ). Importantly, to rule out α4 integrin–dependent FN fibril assembly, we used only clones that lacked α4 expression (Fig. S3 A). When labeled wild-type plasma FN (pFN) was added to the culture, FN cell clones assembled it into a normal fibril network, with α5 integrin colocalizing into typical fibrillar adhesions (Fig. S3 B). Next, we tested whether FN cells are able to assemble their endogenously produced mutant FN-RGE. Assembly assays with self-produced FN critically depend on the absence of exogenous serum-derived FN, which is a widely used supplement of cell culture media. To avoid serum-derived FN, we grew control and FN cells in serum replacement medium and, because they lacked the serum-derived ECM components, seeded them on defined substrates. Control and FN cells plated on LM111 adhered via their laminin binding α6β1 integrin, leaving the α5β1 integrin, as well as the αv integrins, free to contribute to the assembly of the secreted FN. In contrast to the previous notion that the RGD sequence of FN is essential for the initiation of matrix assembly, FN cells initiated and completed the assembly of FN-RGE into fibrils on their cell surface (). Similar to mouse embryos, FN cells were able to assemble a dense fibrillar network (). Although the kinetic of fibril formation was similar between control and FN cells, the size and the distribution of fibrils differed between the two cell types. Control cells produced an elaborate network consisting of thin and long FN fibrils (). In contrast, FN fibrils formed by FN cells appeared short and thick. To determine which integrin, αv and/or α5, mediates assembly of FN-RGE, we plated cells on VN. On VN, control and FN cells adhered via their αv integrins. A previous study showed that this depletes αvβ3 integrin from the cell surface but leaves α5β1 integrin diffusely distributed and presumably free to participate in matrix assembly (). Control cells developed a regular FN network when plated on VN, whereas FN cells were unable to form FN-RGE fibrils (). These data suggest that αv, but not α5β1, integrins mediate the assembly of FN-RGE into matrix fibrils. To further investigate whether αv and not α5β1 integrins executed FN-RGE fibril formation, we immunostained FN cells growing on LM111 with α5 integrin and FN antibodies. As expected, control cells showed coalignment of the α5 integrin with the thin and elaborate FN network (). In sharp contrast, LM111-growing FN cells failed to align α5 integrin with FN-RGE–containing fibrils. Instead, the α5 integrin signal was diffusely distributed over the entire cell surface, further indicating that it is not engaged in FN-RGE binding (). Unfortunately, because of the poor quality of the αv integrin antibodies that were available to us, we could not directly demonstrate a colocalization of αv integrins with FN-RGE fibrils. To confirm that αv integrins were indeed assembling FN-RGE fibrils, we depleted αv integrins in FN cells by viral transduction of αv-specific siRNAs and assayed for FN assembly. FN cells expressed high levels of αv integrins (Fig. S3 A and Fig. S4, available at ), which were reduced by ∼80% after retroviral expression of siRNAs (Fig. S4). The αv integrin–depleted FN cells were cultured in serum replacement medium and seeded on LM111. Despite the availability of free α5β1 integrins, the αv-depleted FN cells were unable to form FN-RGE fibrils (). Occasionally, we could observe a few thick fibrils after longer culture periods, which likely developed as a result of the remaining low levels of αv integrins on FN cells (). To address the question of whether assembly of FN-RGE is accomplished via the RGD binding pocket of αv integrins, we tested FN-RGE fibrillogenesis in the presence of linear Gly-Arg-Gly-Asp-Ser-Pro peptides (linRGD) or cyclic (-Arg-Gly-Asp-D-Phe-Val-) peptides (RGD). The more constrained RGD was reported to preferentially bind αvβ3 integrin (). Incubation of control and FN cells with the linRGD peptide considerably inhibited assembly of wild-type as well as FN-RGE fibrils (). Treatment with the RGD peptide inhibited assembly of FN-RGE much more strongly but showed only partial inhibition of assembly of wild-type FN (). Treatment of control and FN cells with peptides, in which the RGD binding motif is mutated (linear Gly-Arg-Gly-Glu-Ser-Pro peptide [linRGE] or cyclic [-Arg-Ala-Asp-D-Phe-Val-] peptides [RAD]), had no effect on assembly of wild-type FN or FN-RGE (). Similarly, heparin treatment, which blocks the binding of FN to cell surface proteoglycans such as syndecans, partially inhibited assembly of wild-type FN but had almost no effect on FN-RGE fibrils. These data demonstrate that the RGD binding pocket of αv integrins is essential for the assembly of the FN-RGE matrix. The data so far indicate that αv integrins bind and assemble FN-RGE into fibrils via a novel integrin binding site on FN. To localize the novel αv binding/assembly sites, we cultured FN cells on LM111 and treated them with recombinant FN fragments spanning almost the entire FN molecule (). Although FN-I, FN-III, and FN-III strongly inhibited the formation of both wild-type and FN-RGE fibrils, FN-III, FN-III, FN-IIIRGE, and FN-III had no effect on either ( and Fig. S5, available at ). FN-I is generally recognized as a potent inhibitor of matrix assembly and is thought to participate in binding to receptors on the cell surface or to specific FN-III domains of another FN molecule (; ; ; ). Because assembly of the FN-RGE matrix depended on αvβ3 integrin, we wondered if this integrin might bind directly to FN-I. We therefore examined whether αvβ3 integrin is capable of binding to the recombinant FN-I fragment. To this end, we performed direct and competitive solid-phase binding assays, in which the recombinant FN fragments were adsorbed to microtiter plates and then incubated with serially diluted biotinylated αvβ3 integrin. Direct binding assays showed comparable binding of αvβ3 integrins to FN-I, FN-III, and FN-III but no binding to the other fragments tested ( and Fig. S5). Interestingly, α5β1 was also able to bind FN-I, although α5β1 was incapable of assembling FN-RGE fibrils (). Competitive binding assays showed that incubation with the linRGD peptide inhibited binding of both αvβ3 and α5β1 to FN-I (). These results indicate that FN-I contains novel and functional αvβ3 binding/assembly sites, FN-I contains a binding site for α5β1 that is not functional for matrix assembly, the binding sites on both integrins involve their RGD binding pocket, and the binding affinity of FN-I to αvβ3 integrin is of somewhat lower affinity than the RGD binding motif of wild-type FN. Primary and tertiary structure analyses showed that murine FN contains several GNGRG loops within the FN-I that are well conserved in human, bovine, rat, amphibian, and fish (). While we were preparing this manuscript, reported that a single GNGRG loop in FN-I can convert to GDGRG (where D is isoasparte) and that the GDGRG provided a novel binding site for αvβ3 integrin. To explore this observation, we synthesized the (*Cys-Asn-Gly-Arg-Cys*) peptide (NGR-2C) as well as the (*Cys-Asp-Gly-Arg-Cys*) peptide (DGR-2C), which adopts a conformation similar to the deamidated, rearranged GNGRG loop of FN-I () and used them in our competitive solid-phase binding and FN in vitro assembly assays. As shown in , DGR-2C strongly inhibited αvβ3, but not α5β1, binding to FN-I, indicating that αvβ3 but not α5β1 can recognize the DGR motif. Incubation of FN cells grown on LM111 with the NGR-2C or the DGR-2C peptides effectively inhibited assembly of the FN-RGE matrix, whereas assembly of wild-type FN by control cells was unaffected (). have shown that the GNGRG, as well as the GDGRG, motif exhibits a lower affinity to αvβ3 integrins compared with the GDGRG motif. The rearrangement of GNGRG to GDGRG and GDGRG is a well-known reaction occurring spontaneously under physiological conditions (). Because NGR-2C inhibited FN-RGE assembly equally as well as DGR-2C (), we tested the kinetics of the rearrangement of the NGR-2C into DGR-2C. HPLC assays revealed that ∼10% of NGR-2C was converted into DGR-2C after an incubation of 16 h at 37°C at pH 7.2 (). The experiment also shows that before incubation at 37°C, the peptide was nearly 100% NGR-2C. Therefore, when testing for inhibition of FN matrix assembly, only a small amount of DGR-2C would have accumulated in the later time of the 14-h assay. The inhibition observed is likely due to unmodified NGR-2C interacting with αvβ3. Altogether, these data indicate that the GNGRG motif in FN-I represents a novel and selective αvβ3 binding that can function for FN fibril assembly in vitro and in vivo. xref #text FN mice were mated with a deleter-Cre strain to eliminate the loxP-flanked neomycin cassette. Heterozygous FN mice were intercrossed to obtain homozygous FN mutants. The following primers were used to amplify the RGE mutation: pRGEGenof (5′-CAAAGAAGACCCCAAGAGCA-3′) and pRGEGenor (5-ACAAGCCCTGGCCTTTAGTT-3′). The following primers were used to amplify FN mRNA by RT-PCR from total RNA extracted from embryonic tissue: pmRNAf3 (5′-TATCACCGCCAACTCATTCA-3′) and pmRNAr3 (5′-GGGAGTGGTGGTCACTCTGT-3′). The PRC amplicon was then digested with HinfI to obtain two bands of 155 and 315 bp. To test FN expression, E9.5 embryos were dissected and either immediately embedded in optimal cutting temperature compound (OCT; Thermo Savant) to obtain unfixed sections or fixed in 4% PFA and embedded in paraffin (Paraplast X-tra; Sigma-Aldrich) to obtain paraffin sections. Cryosections of 10-μm thickness and paraffin sections of 6-μm thickness were cut and used for histochemistry and immunofluorescence as described previously (). TUNEL staining was performed according to the manufacturer's instructions (Roche). Antibodies against the following proteins were used: anti-FN (Chemicon), anti–α4 integrin (BioLegend), and anti–cleaved caspase-3 (New England Biolabs, Inc.). Cy3- or FITC-conjugated secondary antibodies were purchased from Sigma-Aldrich or Jackson ImmunoResearch Laboratories. E8.5 embryos were solubilized in RIPA buffer, briefly sonicated, and boiled for 2 min in nonreducing sample buffer. The samples were separated on 6% SDS polyacrylamide gels, blotted onto Immobilon-P polyvinylidene difluoride membranes (Millipore), and hybridized with polyclonal antibodies specific for FN (Chemicon) and tubulin (Sigma-Aldrich). Bound antibodies were hybridized with horseradish peroxidase–conjugated secondary antibodies and detected using the ECL kit (GE Healthcare). To establish cell lines, FN mice were intercrossed and embryos harvested at E9. To release cells, embryos were trypsinized for 5 min at 37°C and disrupted by pipetting up and down. Cells were cultured in DME supplemented with 10% FBS. Cells were genotyped by PCR, immortalized by retroviral transduction of the SV-40 large T antigen, and cloned. Control (FN) and FN cell lines were adapted to grow in serum replacement medium (47.5:47.5:5:1 ratio of DME/Aim-V Medium [Invitrogen]/RPM1640/nonessential amino acids) and then used for FN assembly. Cell adhesion assays were performed as described previously (). Flow cytometry was performed as described by . The following antibodies were used: anti–β1, –β3, –α2, –α4, –α5, –α6, and –αv integrins (all obtained from BD Biosciences), which were FITC conjugated (anti–β1, –α2, and –α6 integrins), biotinylated (anti–β3, –α5, and –αv integrins), or PE conjugated (anti–α4 integrin). Biotinylated antibodies were detected with streptavidin-Cy5 (Jackson ImmunoResearch Laboratories). To assess autofluorescence and nonspecific staining, cells were stained with IgM isotype-FITC (β1 and α2), IgG2a isotype-FITC (α6), IgG2a isotype-PE (for α4 integrin), or streptavidin-Cy5 (β3, α5, and αv; all obtained from BD Biosciences). The sequence of the siRNA constructs was chosen according to the protocol of and . Full-length mRNA sequences of the target gene were checked using the online software siDirect for 19-bp stem-loop structure. The sequence obtained was synthesized as a 5′-terminally phosphorylated primer with a 5′-HindIII and 3′-BglII recognition site (5′-GATCCCCCGTTAGGGCAATTAGGATTTTCAAGAGAAATCCTAATTGCCCTAACGTTTTTA-3′). Subsequently, primers were annealed and inserted into HindIII–BamHI linearized pSuper.retro.puro (OligoEngine). The primer 5′-GATCCCCAGCAGTGCATGTATGCTTCTTCAAGAGAGAAGCATACATGCACTGTTTTTTA-3′ was used as a scrambled control. Stable expression of the siRNAs in FN cell was achieved by retroviral gene transfer (). In brief, siRNA inserts were cloned into a retroviral vector (pclMFG). After harvesting, the recombinant retrovirus FN cells were infected, selected in 6 μg/ml puromycin, and cloned. The human FN-III fragment was subcloned into pET15b plasmid (Invitrogen) and mutated into the pET15bFN-IIIRGE and pET15bFN-IIIδRGD by using the QuikChange II XL Site-Directed Mutagenesis kit. The following primers were used to introduce the mutation: RGD>RGE forward, 5′-CACTGGCCGTGGAGAGAGCCCCGCAAGCAG-3′; RGD>RGE reverse, 5′-CTGCTTGCGGGGCTCTCTCCACGGCCAGTGA-3′; RGD>δRGD forward, 5′-GTGTATGCTGTCACTGGCAGCCCCGCAAGCAGCAAGC-3′; and RGD>δRGD reverse, 5′-GCTTGCTGCTTGCGGGGCTGCCAGTGACAGCATACAC-3′. The plasmids were transformed into BL21 (DE3)pLysS cells. Protein expression and purification of these and other fragments were performed as described previously (). All experiments were performed in serum replacement medium. Control and FN cells were seeded on 10 μg/ml LM111–coated 8-well Lab-Tek Chamber Slides (80,000 cells/well), allowed to spread for 2 h, and incubated with inhibitory peptides dissolved in PBS for another 14 h at 37°C and 5% CO. The assay was stopped by fixation with 4% PFA in PBS and immunostained. The final concentrations of the used peptides were as follows: linRGD and linRGE, 0.5 mg/ml (900 μM; BIOMOL Research Laboratories, Inc.); RGD and RAD, 0.25 mg/ml (410 μM; BIOMOL Research Laboratories, Inc.); DGR-2C and NGR-2C, 800 μM (for synthesis, see Peptide synthesis); FN-I, 2.5 μM; FN-III, FN-IIIRGE, and FN-IIIΔRGD, 10 μM; and FN-III, FN-III, FN-III, and FN-III, 25 μM. The FN polypeptide synthesis was previously described (). Heparin was used at a concentration of 0.1 mg/ml. Direct solid-phase binding assay for determining the relative binding of biotinylated αvβ3 integrins to various FN fragments was performed as described previously (). In brief, 0.25 μM solutions of recombinant FN fragments in TBS were used to coat 96-well polyvinylchloride microtiter plates (Maxisorp; Nunc) by an overnight incubation at 4°C. Coating with 0.25 μM BSA was used to determine the background values of unspecific binding. After a 1-h blocking step (1% BSA in TBS), serially diluted integrins were added to the plates and allowed to bind the absorbed FN fragments for 5 h at room temperature in the presence of 1 mM MnCl. The plates were washed with TBS containing 1 mM MnCl, and the bound integrins were quantified by the addition of a 1:1,000 dilution of streptavidin peroxidase conjugate (Vector Laboratories) in TBS/1mM MnCl to the plate for 15 min. After several washing steps with TBS/1mM MnCl, bound peroxidase was detected by a chromogenic reaction at 405 nm (ABTS; Vector Laboratories). Soluble recombinant αvβ3 and α5β1 integrins were purified from the culture supernatants from CHO-lec 3.2.8.1 cells stably transfected with corresponding subunits (). They were then biotinylated with Sulfo-LC-NHS-biotin (Pierce Chemical Co.) according to the manufacturer's recommendation. The linear peptides Cys-Asn-Gly-Arg-Cys (NGR-2C) and Cys-Asp-Gly-Arg-Cys (DGR-2C) were synthesized according to standard solid-phase peptide synthesis on TCP resin using the Fmoc strategy. Fmoc-Asp-OBu was used as Asp building block. Cys was used trityl protected, and Arg was used Pbf protected. To avoid positive charge at the N terminus and to facilitate HPLC purification by enhancing the UV absorption, the terminal amine was benzoylated with 3 eq. benzoyl chloride and 5 eq. diisopropyl amine in -methylpyrrolidinone for 20 min. The linear peptide was cleaved and simultaneously deprotected using a mixture of 95% trifluoroacetic acid, 2.5% water, and 2.5% triisopropyl silane for 1 h at ambient temperature and precipitated in diethyl ether. Cys-Cys cyclization was performed by treatment of the highly diluted (10–10 mol/L) solution in water/ acetonitrile with 2 eq. HO at pH 8–8.5 (NaHCO). The reaction mixture was concentrated, and the DGR-2C peptide was purified by reverse-phase HPLC (purity >95%). For the measurement of the kinetics of the deamidation of asparagine, NGR-2C was dissolved in PSB buffer, pH 7.2, at a concentration of 0.5 mg/ml and put in a heated shaker at 37°C. In the following hours, 50-μl samples were injected on an analytical HPLC, and the composition of the solution was determined by integration of the peaks. Both products of the deamidation rearrangement were identified by comparison of their retention times with the pure compounds. Fig. S1 shows the gene targeting strategy and the analysis of the FN allele. Fig. S2 shows a FACS histogram demonstrating similar α4 integrin expression on cells derived from E9.0 embryos. Fig. S3 shows the expression profile of different integrins on control and FN cells assessed by FACS and the assembly of normal, Cy5-conjugated, plasma-derived FN by control and FN cells. Fig. S4 displays an FCS histogram demonstrating the αv integrin levels in FN cells after siRNA-mediated depletion of αv integrin mRNA. Fig. S5 shows the effect of recombinant FN fragments on FN-RGE assembly and a solid-phase binding assay of recombinant, soluble biotinylated αvβ3 to different recombinant FN fragments. Online supplemental material is available at .

Phosphoinositide 3-kinase (PI3K) is conserved across eukaryotic organisms and regulates many facets of pathways involving cellular growth, survival, metabolism, vehicle trafficking, and chemotaxis. PI3Ks are classified into Class I, II, and III based on their structures and substrate preferences. Class I PI3K is primarily responsible for the production of PIP via phosphorylation of PIP in response to extracellular stimulation (; ; ). All mammalian Class I PI3Ks contain a Ras binding domain (RBD) and can be activated by interacting with GTP-Ras (). The Class IB PI3K, PI3Kγ, is expressed most highly in neutrophils and activated by binding to Gβγ subunits and Ras upon G protein–coupled receptor activation (, ; ; ). Overexpression of Gβγ subunits in HEK293 cells leads to PI3Kγ activation via its interaction with the p101 catalytic subunit (; ). Recently, p101 knockout mice and RBD-mutated PI3Kγ knock-in mice have been generated. In the neutrophils from these mutant mice, chemoattractant-induced PI3Kγ activation is significantly decreased (). These in vitro and in vivo studies demonstrate a linkage between chemoattractant stimulation and PI3Kγ activation. Interestingly, lipid PI3K assays demonstrated that cells have a basal level of PI3K activity in the absence of chemoattractants or serum (; ). It is unclear whether this basal activity of PI3K is actively controlled or merely a passive property of this enzyme. Class I PI3Ks are considered to be the functional counterpart of PI3Kγ on the basis of biochemical and structural characteristics (; ). As implicated in PI3Kγ regulation in mammalian cells, cells lacking Gβ cannot activate PI3K in response to chemoattractant or GTPγS stimulation (; ). cells, however, manage to divide and undergo random motility in the absence of functional heterotrimeric G proteins (). It is known that these complex cell shape changes involve the polymerization of F-actin, but the upstream signals regulating these pathways have yet to be determined. Cells have the intrinsic ability to produce pseudopodia and move in the absence of chemoattractants or nutrients (; ). This random cell migration allows cells to explore their environment and is associated with metastasis of tumor cells. Although much progress has been made in elucidating the molecular mechanisms of chemoattractant-mediated migration (chemotaxis), random cell migration has barely been investigated. In this study, we genetically separate the GPCR heterotrimeric G protein–dependent cellular events that regulate chemotaxis from the basal regulatory signaling loops that control random cell movement and cytokinesis. We demonstrate that both PI3K and Ras activation occur actively, at the same sites of new pseudopod formation in the absence of extracellular stimuli and without heterotrimeric G protein input. PI3K and Ras activation also occur at the poles of dividing cells in wild-type strains and in cells lacking functional heterotrimeric G proteins, implying that cell shape changes during cytokinesis are controlled by a similar mechanism. We suggest that this “unprompted” Ras activity and PIP accumulation, which are independent of external stimuli, constitute a core regulatory pathway involved in a variety of physiological responses and provide the basis for many ligand- or substrate-mediated processes, such as chemotaxis and phagocytosis. In this study, we used vegetative cells rather than starved, developmentally competent cells, which produce and respond to cAMP and are normally used to investigate chemoattractant-mediated cell movement in , as Gβγ signaling is required for cells to become developmentally competent. We expressed the GFP-tagged PH domain of CRAC (PH), a reporter for PIP and PIP, in the KAx-3 wild-type strain and a null strain of Gβ, the sole Gβ subunit in (; ). Consistent with previous studies (; ), β null (β) cells show no detectable folic acid–mediated translocation of PH to the cell cortex (). However, we noticed that, in the buffer containing only Na/KPO, the PH domains spontaneously accumulated at both pseudopodial extensions and to macropinosomes (the structures responsible for fluid-phase uptake) of randomly moving KAx-3 cells (). The PH accumulation in randomly moving cells is a consequence of PI3K activation because its membrane localization was abolished ∼1 min after the addition of LY294002, a PI3K inhibitor (). Importantly, β cells also exhibited spontaneous PH accumulation, which was sensitive to the PI3K inhibitor. This finding indicates that PI3K is spontaneously activated without heterotrimeric G protein signaling. To validate whether β cells activate PI3K signaling, we measured the kinase activity of Akt, a downstream effector of PI3K. illustrates that β cells have a robust Akt activity in the absence of a chemoattractant or nutrients. The treatment of cells with LY294002 completely suppresses this spontaneous Akt activity in both strains to background levels ( cells; unpublished data). Notably, basal Akt activity in β cells is higher than that in the wild-type cells, suggesting a possibility that Gβ attenuates on the basal PI3K/Akt signaling in the absence of stimuli. Although the importance of the Gβ subunit in receptor-mediated PI3K activation has been demonstrated (; ), these data reveal that PI3K and its downstream signaling are activated without a chemoattractant or nutrients and Gβ, whereas Gβ is essential for ligand-induced PI3K activation. As spontaneous PIP accumulation appears to occur strictly at sites of F-actin protrusion, including both the sites of macropinosomes and pseudopodial extensions, we examined whether spontaneous PI3K activation requires F-actin synthesis. (In many instances, we could not resolve whether protrusions began as macropinosomes or pseudopodia, but consider them pseudopodia if they ultimately protruded from the cell and gave rise to a net movement.) After treatment with Latrunculin B (LatB), an inhibitor of F-actin polymerization, spontaneous PIP accumulation (as visualized through PH recruitment) was lost and cells rounded up. After washout of LatB, cells regained the spontaneous PIP accumulation at the sites of new pseudopodial projections (). This finding suggests that Gβ-independent, spontaneous PI3K activation requires and occurs at sites of F-actin polymerization. Furthermore, we found that GFP-PI3K2 labeled the sites of new F-actin projections in both wild-type and β cells ( and Video 1, which is available at ), whereas PTEN-GFP detached from the membrane in pseudopodial extensions in both wild-type cells and β cells (unpublished data). The β cells did not exhibit PI3K translocalization in response to a chemoattractant, whereas wild-type cells did (unpublished data). Thus, although PI3K and PTEN localization is regulated by extracellular stimuli in the chemotaxing cells, the reciprocal localization of PI3K and PTEN occurs during random cell movement, even in the absence of chemoattractant, nutrients, and heterotrimeric G proteins. As Ras is required for PI3K activation in response to extracellular stimuli in neutrophils and , we monitored the localization of activated Ras during random movement using a GFP fusion of the human Raf1 RBD (GFP-RBD; ). cells. Surprisingly, the Ras activation was abrogated with LY294002 treatment and recovered after removal of the drug ( and Video 2, which is available at ). Biochemical assays confirmed a basal Ras activity in both wild-type and β vegetative cells in the absence of exogenous stimulation or nutrients (). The basal Ras and PI3K activation do not require cellular attachment to the substratum, as cells in suspension display the LY294002- sensitive spontaneous Ras and Akt activations (Fig. S1 A, available at ). cells (), supporting our notion of the requirement of PI3K. Furthermore, like PI3K activation, this unprompted Ras activation requires F-actin ( and Video 3), although it is possible that a very low level of Ras and PI3K activity exists in these drug-treated cells. These drug-treated wild-type cells can activate Ras in response to chemoattractant stimulation (; ). Next, we examined whether Gβ was required for chemoattractant-induced Ras activation. illustrates that chemoattractant-induced Ras activation occurs in wild-type cells but not in β cells. There was no detectable translocation of GFP-RBD in response to chemoattractant stimulation in cells lacking the Gβ subunit (). These results uncover two different pathways that lead to detectable levels of activated Ras: a receptor-mediated, Gβ-dependent pathway and a Gβ-independent, PI3K-dependent pathway. It is worth noting that we previously observed spontaneous PI3K and Ras activation in developed PTEN-deficient cells (). cells likely increase the level of the chemoattractant (cAMP) secretion (), which would evoke a GPCR/heterotrimeric G protein–mediated autocrine Ras activation (). To compare the sites and timing of Ras activation to those of PI3K in randomly moving cells, we simultaneously imaged both GFP-RBD and RFP-PH by using a dual-wavelength beam splitter with high time resolution (∼200 msec). The RBD and PH domain appeared concurrently (within the limits of our resolution) at the same sites where pseudopodia start to form (, arrow, and Video 4, which is available at ). The kinetics of the disappearance of the RBD and PH probes were also closely correlated (). cells. RBD and PH also appear at micropinosome cups (clathrin-dependent small invaginating pits) with the same timing; however, PH is retained considerably longer at these sites than RBD, suggesting differential regulation of Ras at later stages of micropinocytosis (Fig. S1 B). These data demonstrate that the timing of PI3K and Ras activation occurs in parallel and at the same sites during random movement. In experiments in which cells were stimulated with a chemoattractant, the RBD translocated ∼0.6 s faster than PH to the plasma membrane (, cAMP, and Fig. S2 C, folic acid, which is available at ). Thus, Ras activation is not synchronized with PI3K activity in the initial phases (∼2 s) of chemoattractant signaling. cells stimulated with either folic acid or cAMP, PH was retained at the plasma membrane for >1 min, whereas RBD returned to the cytosol in ∼20 s (). These results suggest that a chemoattractant/Gβ-dependent pathway induces the signaling response that overrides or disrupts the intrinsic feedback activation of Ras/PI3K. In response to a chemoattractant gradient, cells activate a localized response at the site on the plasma membrane closest to the chemoattractant source while inhibiting the spontaneous activation of Ras/PI3K at the lateral sides of cells, thereby repressing random cell movement. We further investigated whether Gβ-independent PI3K activation is involved in random motility. illustrates that random movement of two strains of wild-type cells (KAx-3 and NC-4) is rapidly blocked by 50 μM LY294002 (Video 5, available at ; similar results are observed with 30 μM LY294002, unpublished data). As NC-4 cells do not have macropinosomes, the cellular movement is driven by LY294002-sensitive pseudopodial extensions. This LY294002-sensitive random cell movement does not require heterotrimeric G proteins because the β cells exhibit random cell movement (). cells (a strain lacking the three major PI3Ks responsible for PIP production in []). cells exhibit defects in random cell motility ( and Video 6). cells still produce small pseudopodial projections, suggesting that a low level of PI3K-independent F-actin synthesis pathway is active. On the other hand, wild-type cells expressing membrane-bound PI3K2 (Myr-PI3K2; ), thus bypassing F-actin–induced PI3K translocalization, periodically exhibit sudden robust protrusions that are associated with rapid cell movement (Video 7). This finding suggests that F-actin–mediated PI3K translocation plays an important role in amplifying F-actin projection during random movement. Overall, these data demonstrate that PI3K regulates random cell movement and the extent of F-actin projections in the absence of extracellular stimuli. In randomly moving wild-type cells, RFP-PH and GFP-RBD colocalize on F-actin projection sites ( and Video 8). strain is a PI3K hypomorphic cell line and displays reduced random cell movement and abolished PH domain localization at the plasma membrane. cells complements this random cell movement defect and spontaneous PIP accumulation at sites where Ras is activated (). Furthermore, an isogenic strain lacking Aimless, one of the RasGEFs responsible for chemoattractant-induced Ras activation (; ), displays reduced random movement (). Collectively, these results show that Gβ-independent activation of PI3K requires interaction with Ras-GTP, and that Ras activation is required for basal cell motility. This suggests that cells use a Gβ-independent Ras/PI3K feedback amplification pathway to form pseudopodia. As Dd-target of rapamycin (Dd-TOR) may be another LY294002-sensitive kinase, we examined the role of TOR in random cell movement. cells, which lack the ortholog of mammalian Rictor/mAVO3 (), did not result in random movement defects (unpublished data and Video 9, which is available at ). Akt is under the regulation of TORC2. cells, which have a defect in macropinocytosis, display normal RFP-PH and GFP-RBD colocalization on F-actin projections linked to cellular movement, consistent with TORC2 not playing a major role in random cell movement (; Video 10). We previously demonstrated that the regulation of PIP plays a central role in cell shape changes during cytokinesis (). We show here that Ras activation as examined by RBD cortical localization, like PI3K activation, was uniformly suppressed at the onset of cytokinesis as cells rounded up. As cells progressed through mitosis, the Ras activity reporter gradually localized to the polar ruffles during spindle assembly, cell elongation, and cytokinesis (). When the daughter cells separate from one another, cortically localized RBD and PIP activity increase dramatically, resulting in high levels of random pseudopod extension. The β cells exhibit activation of Ras indistinguishable from that of wild-type cells during cytokinesis, suggesting that the Gβ-independent Ras/PI3K circuit plays a fundamental role in cytokinesis. The pathways controlling random movement parallel those of the amplification step of chemotaxis that is controlled through a regulatory loop containing Ras, PI3K, PTEN, and F-actin (). We suggest that, in chemotaxis, the directed activation of the pathway by the chemoattractant/G proteins biases the localized activation of the intrinsic Ras/PI3K circuit and locally restricts the positive feedback loop that is the basis for random cell movement (Fig. S2). The output from the sensing mechanism activates Ras and generates PI3K binding sites and simultaneously results in the loss of PTEN binding sites. This reciprocal regulation, along with the activation of Ras, leads to the local production of PIP and pseudopod extension. cells is decreased (; ). There are likely other Ras isoforms that are integrated into the PI3K feedback loop, as null cells show modest defects. The challenge for future investigation will be to elucidate the molecular mechanisms by which Ras, RasGEF, and RasGAP regulate basic cell motility in a Gβ-independent fashion in the absence of extracellular stimuli. We obtained folic acid, LatB, and LY294002 from Sigma-Aldrich, and monoclonal anti-Ras (Ab-3) antibody from Oncogene Research Products. We used 50 μM LY294002 for the experiments because this concentration has become the standard and at this concentration Akt/protein kinase B (PKB) activity and GFP-PH translocation are blocked, but PKBR1 (which is not PI3K dependent but is a TORC2-dependent, Akt/PKB-related kinase in ) is not blocked. We observed that 30 μM LY294002 similarly suppressed random movement of wild-type cells. GST-RBD, GFP-RBD, GFP-PH, N-PI3K-GFP, and PTEN-GFP were described previously (; ; ). RFP-PH was cloned into a hygromycin-resistant CV5 vector (pYu34). We created an null strain in the KAx-3 background using a targeting vector similar to that used to generate null in a KAx2 background (). The β, , and null strains were described previously (; ; ). All cell lines except for NC-4 were cultured axenically in HL5 medium at 22°C. NC-4 cells were grown on bacterial lawns. Transformants were maintained in 40 μg/ml G418, 50 μg/ml hygromycin, or both as required. PKB activation and Ras activation were measured as described previously (; ). Cells were treated or not treated with 50 μM LY294002 for 5 min. Endogenous Akt was immunoprecipitated and subjected to an in vitro kinase assay using H2B as a substrate. To monitor the Akt and Ras activation in vegetative cells, cells were harvested at 2–4 × 10 cells/ml, washed twice with 12 mM sodium/potassium phosphate buffer, pH 6.1, buffer A, and then resuspended in buffer A at 10 cells/ml. After 30 min of starvation, cells were subjected to the assay. For comparing Ras activation of wild-type cells to that of the mutant cells, washed cells were placed on the plate for 1 h and stimulated with 50 μM folic acid for the indicated duration, and then lysed on the plate. In a random movement assay, vegetative cells growing on plates were harvested and seeded onto a chambered coverglass in starvation buffer. Cells were rinsed three times with an excess amount of buffer A at 10 min after seeding, and then sat for 1 h. Images were collected on a microscope (model TE300; Nikon) with DIC and 40x/0.60 objectives. Initial images were captured using Metamorph software and analyzed with the DIAS program (). Speed refers to the speed of the cell's centroid movement along the total path. The cell movement during the 1 min between measured frames was measured to calculate speed so that genuine movement was measured rather than cytoplasmic rearrangement. Parallel experiments were performed with cells in HL5 axenic growth medium or cells starved for 2 h. No differences in the results were observed under these three conditions. Cytokinesis was measured as described previously (). Confocal images were obtained by using a CSU10 scanner unit (Yokogawa) on a Leica inverted DMIRE2 microscope with a 63x/1.4 objective using an ORCA-ER camera (Hamamatsu) or a Dual-View OI-11-EM–equipped EM-CCD camera (Hamamatsu) for simultaneous imaging. Imaging was described previously (; ). Fig. S1 A shows spontaneous Ras and PI3K activation in low density suspended cells, and Fig. S1 B shows the differences of RBD and PH accumulation kinetics in micropinosome formation from those in pseudopodial formation. Fig. S1 C shows translocation of GFP-RBD and PH in vegetative wild-type cells by folic acid. Fig. S2 illustrates a model for the Ras/PI3K circuit during random movement and chemotaxis. Video 1 shows GFP-N-PI3K1 localization in β null cells without extracellular stimuli. In Video 2, GFP-RBD localization in a β null cell corresponds to A in the text. In Video 3, GFP-RBD localization in a β null cell corresponds to D in the text. In Video 4, simultaneous imaging of GFP-RBD and RFP-PH in the wild-type vegetative cells corresponds to A in the text. In Video 5, random movement analysis of nonaxenic NC4 cells corresponds to A in the text. In Video 6, random movement analysis of cells corresponds to A in the text. Video 7 shows random movement analysis of myristoylated PI3K2-overexpressing wild-type cells. In Video 8, GFP-RBD and RFP-PH in the wild-type cells correspond to B in the text. Video 9 shows random movement analysis of null cells. Video 10 shows GFP-RBD and RFP-PH in the null cells. The online supplemental material can be found at .

Regulation of the actin cytoskeleton is a key mechanism for the control of cell shape and motility involved in numerous complex and dynamic processes such as cell polarization, adhesion, endocytosis, and phagocytosis. These events require the localization and activation of specific actin nucleation factors to generate de novo actin filament networks of different sizes and shapes. The intensively studied actin-related protein 2/3 (Arp2/3) complex generates branched actin filaments, whereas the family of formin proteins produces unbranched actin filaments (for reviews see ; ). The Diaphanous-related formins (DRFs) stimulate barbed end actin filament elongation through the dimeric formin homology (FH) 2 domain preceded by a proline-rich FH1 domain (; ; ). Dia1 is characterized by regulatory domains in which the N terminus encompasses a RhoA-binding domain (RBD) followed by a four armadillo repeat–containing Diaphanous inhibitory domain (DID) that binds the C-terminal Diaphanous autoregulatory domain (DAD), thereby maintaining the protein in a dormant conformation (; ). Dia1 autoinhibition between DAD and DID was shown to be released through the binding of RhoA-GTP to the RBD (). The armadillo repeat–containing DID and the adjacent dimerization domain have also been called the FH3 region and are thought to be involved in subcellular DRF location through interaction with unknown factors (). Thus, targeting of Dia1 to actin dynamic regions is not yet understood but likely involves activated RhoA (). The proposed biological functions of mammalian Dia1 so far are rather diverse and include cell adhesion and migration, microtubule stabilization, serum response factor (SRF) transcriptional activity, and endocytosis and phagocytosis (). Phagocytosis is the activity performed by professional phagocytes to engulf large particles (>0.5 μm). This activity is crucial for tissue homeostasis and clearance of pathogenic microorganisms. Dia1 has been shown to be essential for CR3- mediated phagocytosis in macrophages, which is a Rho-mediated process (). The DRF FRLα (formin- related gene in leukocytes α) is required for Fcγ receptor–mediated phagocytosis in macrophages, which is a Cdc42/Rac-mediated process (; ). Both DRFs are recruited at the phagocytic cup, where de novo actin polymerization occurs during pseudopod extension around the particle. It has been suggested that N-WASP (neural Wiskott Aldrich syndrome protein) and the Arp2/3 complex activate actin nucleation at the nascent phagosome in both CR3 and RFcγ-mediated phagocytosis. The precise role of DRFs during phagocytic cup formation is still unknown but likely involves the regulation of actin dynamics or the de novo assembly of actin filaments together with the Arp2/3 complex. In this study, we identify IQGAP1 as a new binding partner of Dia1. IQGAP1 regulates Dia1 localization at the leading edge of migrating cells as well as at the phagocytic cup in macrophages. Furthermore, we show that the deregulation of IQGAP1 activity fully inhibited phagocytosis in mouse macrophages, thus suggesting a crucial role for this protein in the immune response. To identify factors responsible for the localization and regulation of DRFs, we used GST affinity columns containing an N-terminal region of Dia1 spanning amino acids 256–567 (formerly FH3). Using that approach, we eluted a specific band migrating at 190 kD from HeLa cell extracts (). Mass spectrometric analysis identified this band as the cytoskeletal scaffold protein IQGAP1 (Fig. S1, available at ), and these data were confirmed by immunoblotting using an IQGAP1-specific antiserum (). IQGAP1 is a widely expressed Rac and Cdc42-binding protein that is involved in polarized cell migration by regulating microtubule capture through CLIP170 and adenomatosis polyposis coli (; ). Furthermore, IQGAP1 is enriched at cell–cell contact sites as well as at the leading edge of migrating cells, where it influences the actin cytoskeletal morphology through mechanisms that are less well understood (). However, it has recently been shown that IQGAP1 binds N-WASP and, thereby, activates Arp2/3-mediated actin nucleation (). The fact that IQGAP1 and Dia1 both influence the actin filament and the microtubule system in migrating cells led us to further characterize a possible relationship between these two cytoskeletal regulators. First, we tested whether endogenous Dia1 and IQGAP1 would associate using immunoprecipitations from serum-starved and -stimulated cells. Interestingly, detectable coprecipitated IQGAP1 increased by 3.4 ± 0.9-fold within 10 min in a serum-dependent manner (), suggesting that extracellular signals dynamically regulate the association of Dia1 with IQGAP1. Second, immunofluorescence analysis on wounded migrating fibroblasts confirmed that endogenous IQGAP1 accumulates at the leading edge, where it colocalizes with microinjected Dia1-FH3, whereas this construct did not inhibit cell migration (). These data suggest that this N-terminal region mediates the interaction of Dia1 with IQGAP1 in polarized cell movement. Interestingly, recent data demonstrated that active RhoA and Dia1 colocalize at the leading front of migrating cells, and this Dia1 localization depends on activated RhoA (), indicating that RhoA regulates Dia1 activity in this specific location. Indeed, when cells were treated with C3 to inactivate RhoA, Dia1 no longer localized to the wound edge of migrating cells as expected, whereas IQGAP1 localization to the leading edge was unaffected (unpublished data). Thus, Dia1 and IQGAP1 associate dynamically during cell migration, and localization of Dia1 but not IQGAP1 depends on RhoA activation at the front of protruding cells. In an attempt to characterize the domains responsible for association with Dia1, we generated a series of IQGAP1 mutants according to known domain boundaries () and performed coimmunoprecipitation experiments in HEK293 cells. The FH3 region was reciprocally immunoprecipitated with IQGAP1 and interacted preferentially with the C terminus of IQGAP1 (). These data also revealed that IQGAP1 interacted specifically with Dia1 or its N-terminal region but not with Dia2 or Dia3 (), whereas both IQGAP1 and IQGAP2 coimmunoprecipitated with the FH3 region (Fig. S2 A, available at ). The minimal IQGAP1-binding region of Dia1 spans amino acids 256–346, representing the armadillo repeats 3 and 4 of the N-terminal region (), and it is both sufficient and required for IQGAP1 binding (). The Dia1-binding region (DBR) of IQGAP1 lies within the IQGAP1 C terminus (Fig. S2, B and C) and was narrowed down to a recombinantly purifiable fragment of amino acids 1,503–1,657. In vitro binding studies with purified N-terminus Dia1 (Dia1-nt) and IQGAP1 DBR revealed a single class of affinity binding sites with a K of ∼60 nM (). This IQGAP1 C-terminal region was previously reported to bind to the armadillo repeat domain of adenomatosis polyposis coli (), suggesting a similar mode of protein–protein interaction as observed for Dia1. Interestingly, the DBR directly interacted with Dia1-nt that was dependent on the release of C-terminus Dia1 (Dia1-ct)–mediated autoinhibition by the addition of active RhoA, whereas it did not interact with autoinhibited Dia1-nt/Dia1-ct in the absence of RhoA (), demonstrating that only activated Dia1 associates with IQGAP. Because the Dia1–IQGAP1-binding region partially overlaps with regulatory domains such as the DAD–DID autoinhibitory binding interface, we tested whether the DBR of IQGAP1 directly affects the actin polymerizing activity of Dia1 in vitro. For this, we functionally reconstituted Dia1 autoinhibition using N- and C-terminal constructs. The N-terminal construct contains the RBD, DID, dimerization domain, and the coiled-coil regions, and the C-terminal construct spans the FH2 and DAD domains (). As reported by , we observed that the C terminus stimulates actin filament assembly, which can be inhibited by the N terminus (). The addition of purified V14RhoA fully restored FH2-mediated actin polymerization in a dose-dependent manner, whereas the addition of IQGAP1-DBR displayed no detectable effects on Dia1-mediated actin filament assembly either alone or in combination with RhoA (). Consistent with this, IQGAP1 or its C terminus neither stimulated SRF nor interfered with Dia1- induced SRF activity (Fig. S2 D), which is known to depend on actin dynamics (). Thus, although only RhoA-activated Dia1 binds IQGAP1, we observed that IQGAP1 neither activated nor inhibited actin polymerization by Dia1. Of course, this does not rule out that full-length IQGAP1 can modulate Dia1 activity under more physiological conditions (for example, to stabilize its active conformation). Because the FH3 of Dia1 colocalized with IQGAP1 at the front of migrating cells, we speculated that IQGAP1 is involved in the subcellular location of Dia1. To test this, NIH3T3 cells were treated with control siRNA or with siRNA specific for Dia1 or IQGAP1, and monolayers were scratch wounded to induce polarized cell migration. Previous RNAi studies demonstrated a critical role for both proteins in wounding-induced fibroblast cell migration (; ). Using this assay, we observed that localization of Dia1 to the wound edge of migrating cells depends on the presence of IQGAP1, but, when Dia1 was knocked down, IQGAP1 localization was normal (). In agreement with this, we found that cells expressing the DBR failed to localize Dia1 to the wound edge (Fig. S2 E). This argues that IQGAP1 is required to localize Dia1 at the front of the migrating cell. The front of migrating cells is a cortical region of high local actin filament assembly and disassembly (for review see ). Using RNAi, it has recently been suggested that Dia1 is involved in phagocytosis by a yet unidentified mechanism (). Therefore, we decided to investigate the potential role of IQGAP1 in localizing Dia1-mediated actin filament assembly by studying phagocytic cup formation, a process in which localized actin polymerization provides the driving force by which cells internalize particles (). First, we examined IQGAP1 and Dia1 localization during the internalization of 3-μm avidin-coated latex beads in RAW macrophages. Avidin coating stimulates internalization via unknown receptors. Immunofluorescence analysis showed that endogenous IQGAP1 localizes to the phagocytic cup of avidin-coated beads along with filamentous actin (F-actin; ) as well as to CR3-coated beads (not depicted). Likewise, we also observed a distinct signal for endogenous Dia1 around the phagocytic cup (), indicating that Dia1 functions locally at these structures. Consistent with this, using a previously described in situ probe for Rho-GTP (), we could detect active Rho at the phagocytic cup, suggesting that it may promote local Dia1 activation during phagocytosis. Further supporting these findings are the observations that IQGAP1 and Dia1 colocalize at the phagocytic cup (). These data suggest that Dia1 and IQGAP1 interact during phagocytic cup formation, as observed at the leading edge of migrating cells, and that RhoA is also locally activated to promote internalization. Thus, IQGAP1 localization at the phagocytic cup may be essential for positioning RhoA-induced Dia1 activity. We examined the localization of IQGAP1 during phagocytosis in a more dynamic way by expressing GFP-IQGAP1 in macrophages. We confirmed that GFP-IQGAP1 but not GFP is recruited to the phagocytic cup (). After phagocytic closure, IQGAP1 was not detected around the internalized phagosome (, asterisk). However, on a later stage of maturation, we observed that some phagosomes exhibited rocketing movement throughout the cell cytoplasm, which was characterized by the formation of an actin tail ( and Video 1, available at ). This phenomenon has been observed for endosomes and more recently for phagosomes in (). We could monitor such phenomenon in RAW macrophages stably expressing GFP-actin (Video 1). To our surprise, we observed that GFP-IQGAP1 is enriched in the comet tails during phagosome rocketing ( and Video 2). This implies that IQGAP1 is involved in different phagocytosis stages in which actin filament assembly is required, suggesting a critical function of IQGAP1 for activating local de novo actin nucleation. Rocketing movement of phagosomes likely does not involve microtubules, suggesting that IQGAP1 could act in the cell as a regulator of actin assembly independently of its function as a microtubule plus end–capturing protein. To investigate the functional role of the IQGAP1–Dia1 interaction in phagocytosis, we overexpressed the IQGAP1- DBR in RAW macrophages. In contrast to wild-type IQGAP1, which localized to the plasma membrane and to filopodia, this domain showed a diffuse distribution throughout the cytoplasm (). We observed that cells expressing the IQGAP1-DBR were unable to initiate phagocytic cup formation and to subsequently take up opsonized particles (). However, binding of the beads to the surface of the cell was not impaired (, arrows). Furthermore, in DBR-expressing cells, endogenous IQGAP1 still localized to the cell cortex (, top). In contrast, Dia1 showed a diffuse distribution throughout the cytoplasm and was not detected at the bead-binding site (, bottom). These data suggest that the DBR can function as an interfering version of IQGAP1, probably by inhibiting its interaction with endogenous Dia1, and show that this interaction is crucial to initiate phagocytic cup formation. To confirm the functional role of IQGAP1 in phagocytosis, we inhibited IQGAP1 production using two different siRNAs. Western blot analysis showed that IQGAP1 expression was decreased by 50% in cells treated with siRNA #1 and for 90% in cells treated with siRNA #2 (). The IQGAP1 knocked down cells showed an almost complete inhibition of phagocytic activity (), demonstrating the crucial role of IQGAP in the phagocytic process. In this study, we identify IQGAP1 as a binding partner for Dia1 and as a novel regulator for phagocytosis. The interaction of IQGAP1 and Dia1 appears to be critical for phagocytic cup formation, implicating IQGAP proteins as factors that regulate Dia1 localization in addition to or in cooperation with RhoA, thereby adding another level of complexity in how cells control formin function. Thus, it is tempting to speculate that local actin polymerization by Dia1 acts as a driving force for phagocytosis. Interestingly, a recent report has demonstrated that the Cdc42-regulated DRF FRLα is important for phagocytosis and has suggested that, in addition to GTPase activity, a factor X may bind at the DRF N terminus to promote membrane localization (). In the case of Dia1, we would like to propose that IQGAP1 represents such a factor. Because IQGAP specifically binds Dia1 but not Dia2 or Dia3, it is likely that different scaffold proteins regulate other DRFs in a specific manner for various cellular functions involving cytoskeletal dynamics. All cell culture reagents were obtained from Invitrogen. All chemicals were obtained from Sigma-Aldrich. 3 μm of latex beads were purchased from Polyscience. HRP-conjugated α-Flag, α-myc antibodies, and α-Flag agarose were purchased from Sigma-Aldrich, monoclonal α-Dia1 antibodies were obtained from BD Biosciences, and α-His antibodies were purchased from QIAGEN. α-Dia1 antibodies were raised in rabbits against synthetic peptides representing residues 727–765 and were purified using affinity chromatography on immobilized peptides. Generation of IQGAP1 antiserum was described previously (). All secondary antibodies were obtained from Dianova. IQGAP2 cDNA was a gift from A. Bernards (Massachusetts General Hospital, Boston, MA), and pGEX-RhoAV14 was a gift from J. Faix (Medizinische Hochschule Hannover, Hannover, Germany). Desired IQGAP and Dia plasmids were amplified by PCR using specific primers (Table S1, available at ). To obtain Dia1 plasmids deleted for residues 256–346 or 750–770, the corresponding codons were replaced with three alanines, introducing a Not1 site. Proteins were purified in the strain DE3 using standard protocols (). Proteins were dialyzed against 25 mM Tris, pH 7.4, 50 mM NaCl, 2 mM EDTA, 2 mM DTT, and 2% glycerol. His-tagged Dia1-ct (748–1,203) and -nt (1–548) were prepared according to . 9 × 10 Hela cell extracts (Cil) were lysed with a Dounce homogenizer, and lysates were cleared by centrifugation at 100,000 and passed through glutathione beads to remove endogenous GST. 25 mg/ml of cell extracts was incubated with 0.5 mg GST fusion proteins coupled to glutathione–Sepharose beads and eluted by salt gradient before isopropanol precipitation. Eluates were analyzed by SDS-PAGE (4–12% Bis-Tris gels; Invitrogen) and colloidal Coomassie. Protein bands of interest were excised and analyzed by mass spectrometry (matrix-assisted laser desorption/ionization-time of flight; Biochemiezentrum Heidelberg). RAW 264.1 mouse macrophages, NIH3T3, and HEK293 cells were maintained in DME supplemented with 10% FBS, 2 mM glutamine, 100 IU/ml penicillin, and 100 mg/ml streptomycin at 37°C in a CO atmosphere. Cells were transfected using LipofectAMINE 2000 (Invitrogen). IQGAP1-specific siRNA (#2, 5′-UUAUCGCCCAGAAACAUCUUGUUGG-3′; and #1, 5′-UUCUUCAUGAGACAAGGCUUGUUCA-3′) was obtained from Invitrogen, and Dia1 siRNA (; ; ) was obtained from IBA. Wound-healing assays and stainings on NIH3T3 cells were performed as described previously (). RAW cells grown on 12-mm glass coverslips were incubated with 3 μm of latex beads coated with avidin (1:100 final dilution) in internalization medium (MEM, 10 mM Hepes, and 5 mM -glucose, pH 7.4). Cells were fixed and permeabilized before PFA was quenched in 50 mM PBS/NHCl for 15 min. Staining of F-actin was performed by incubating the cells with 0.5 μM phalloidin-TRITC for 30 min at RT. Detection of IQGAP1 and Dia1 was performed by incubating cells for 1 h at RT with polyclonal α-IQGAP1 antibodies (dilution of 1:100) or monoclonal α-p140mDia antibodies (1:100). For RhoA-GTP detection, purified GFP-RBD protein was added at a final concentration of 0.02 mg/ml for 2 h at 4°C as described previously (). Slides were mounted in MOWIOL. Immunofluorescence images were collected with a microscope (DMIRE2; Leica) equipped a camera (DC350F; Leica) and IM50 imaging software (Leica) or with an FCS confocal laser-scanning microscope (LSM510; Carl Zeiss MicroImaging, Inc.) equipped with an oil immersion UV PLAPO 63× NA 1.32 lens (Carl Zeiss MicroImaging, Inc.). Images were processed using Photoshop CS (Adobe). Cell extracts were subjected to immunoprecipitation and immunoblotting as described previously (). For in vitro interactions, increasing concentrations of purified His–Dia1-nt were mixed with immobilized GST-DBR in binding buffer (20 mM Tris-HCl, pH 8, 100 mM NaCl, 0.1% CHAPS, 0.1% Triton X-100, and 1 mM DTT), and bound proteins were analyzed by Coomassie brilliant blue staining. The amount of His–Dia1-nt was detected in a linear range using serial dilutions of standards (BSA) by Coomassie staining and was estimated by densitometric analysis using Photoshop (Adobe). Nonlinear regressions were determined using Prism software (GraphPad). To reconstitute Dia1 autoinhibition, 0.03 μM His–Dia1-nt was incubated in the absence or presence of 0.06 μM His–Dia1-ct or His–Dia1-ct together with 3 μM V14RhoA for 10 min on ice. GST-DBR–coupled glutathione–Sepharose beads were added, and reactions were incubated for 30 min at 4°C. Precipitates were washed four times, and proteins were analyzed by immunoblotting. Protein solutions were mixed with 40 μl of G buffer (20 μM CaCl, 20 μM ATP, and 0.5 mM Tris-HCl, pH 8) and 10 μl of polymerization buffer (12.5 mM KCl, 0.5 mM MgCl, 2and 5 μM ATP) and added to 30 μl of G actin (5% pyrene labeled; final concentration of 4 μM; Cytoskeleton, Inc.). Pyrene fluorescence was excited at 365 nm and recorded at 407 nm every 12.5 s for a period of 500 s for every experiment. 48 h after transfection, RAW cells were replated onto 35-mm glass-bottom dishes and allowed to adhere for 8 h. For imaging, cells were placed at 37°C on a heated stage connected to a humidifier module. To start phagocytosis, cells were incubated with 1 ml of imaging medium (10 mM Hepes and 5 mM -glucose, pH 7.4) containing latex beads coupled to avidin. Imaging runs were started after bead internalization. For each time point, a 10-plan z stack of GFP images spaced 0.5 μm apart was acquired at an exposure time of 500 ms for GFP-actin–expressing cells and at 800 ms for GFP-IQGAP1 cells. A total of 10 images per time point were collected, which resulted in one time point per 5 s for actin-GFP and 8 s for IQGAP-GFP. Beads were visualized in the GFP channel as a result of their autofluorescence. 48 h after transfection, RAW cells grown on 12-mm glass coverslips were incubated at 37°C with 3-μm latex beads coated with avidin (1:100 final dilution) in internalization medium (MEM, 10 mM Hepes, and 5 mM -glucose, pH 7.4). After 1 h, cells were washed and fixed with 4% PFA. Noninternalized beads were stained using either biotin-rhodamine or monoclonal α-avidin antibody (1:100) detected by α-mouse-Cy5. Internalized beads were counted using phase contrast. Three independent experiments were performed. For each, 30 cells positive for GFP-protein transfection were examined for their phagocytic activity. Fig. S1 represents the peptide sequences of IQGAP1 identified by mass spectrometric analysis. Fig. S2 shows Dia1-IQGAP interaction and its effects on SRF activity and Dia1 localization in migrating cells. Table S1 provides detailed information about the plasmids generated in this study. Video 1 shows phagosome rocketing in a GFP-actin–expressing RAW cell. Video 2 reveals the presence of GFP-IQGAP1 during phagosome rocketing. Online supplemental material is available at .

TGF-β superfamily members are multifunctional cytokines that regulate a broad range of cellular functions, including cell proliferation, differentiation, and apoptosis (; ). TGF-β signals through a heteromeric complex of two types of transmembrane Ser/Thr kinases: TGF-β type I receptor and TGF-β type II receptor (TβRII). TGF-β binding to TβRII induces the recruitment and phosphorylation of TGF-β type I receptor, which, in turn, phosphorylates the receptor-regulated Smads (R-Smads) Smad2 and Smad3. Once phosphorylated, Smad2 and Smad3 associate with the common partner Smad, Smad4, and translocate to the nucleus, where they regulate the expression of TGF-β target genes. In contrast to R-Smads and Smad4, the inhibitory Smad, Smad7, appears to block signal transduction by preventing access of R-Smads to the TGF-β receptor or by recruiting distinct E3 ubiquitin ligases that target the receptor–Smad7 complex for degradation (; ; ). Upon TGF-β stimulation, Smad2 is recruited to the receptor complex by an adaptor molecule called Smad anchor for receptor activation (SARA). At steady state, SARA-bound Smad2 is localized in early endosomes to which the receptor is internalized via clathrin-coated pits (; ). The importance of the clathrin-mediated endocytic pathway in TGF-β signaling is also manifested by the recent finding that cPML (cytoplasmic form of the promyelocytic leukemia protein) mediates TGF-β signaling by facilitating recruitment of the SARA–Smad2 complex and TGF-β receptors to early endosomes (). In addition to clathrin, TGF-β receptors can also associate with caveolin (), which leads to their internalization into caveolin1-positive vesicles with subsequent degradation through the proteasome pathway. Consistent with this notion, the caveolin1-positive vesicles were found to associate with Smad7 (), which is known to mediate the association of the E3 ligases Smurf1 and Smurf2 to receptors, leading to their degradation. To gain more insight into the regulation of TGF-β signaling, we have performed yeast two-hybrid screens using TβRII as bait. ADAM12 (a disintegrin and metalloproteinase 12) was one of the TβRII interactors that exhibited specific and strong binding to TβRII. ADAM12 belongs to the ADAMs family, which are glycoproteins characterized by a multidomain structure comprised of pro-, metalloproteinase, disintegrin, cysteine-rich, transmembrane, and cytoplasmic domains (; ). ADAMs exhibit proteolytic, cell adhesion, and signaling properties, and perturbations of ADAM expression are associated with several human diseases, including cancers (). In the present study, we provide the first evidence that ADAM12 interacts with TβRII and enhances TGF-β signaling by controlling the localization of TGF-β receptors to early endosomes. These results reveal a new role for ADAM12 in the regulation of TGF-β receptor trafficking. Using the extracellular domain of human TβRII as bait, we performed a yeast two-hybrid screen of a human placental cDNA library. Eight different fragments of ADAM12 were found to interact with TβRII (). Two variants were previously described for ADAM12: a transmembrane glycoprotein () and a shorter secreted form (). The common sequences shared by the overlapping fragments of the prey span the metalloproteinase and disintegrin domains common to the two variants (). To confirm the association of ADAM12 with TβRII, a fragment of ADAM12 isolated in the yeast two-hybrid screen (amino acids 142–739 that include the metalloproteinase and cysteine-rich domains; ) was tagged with Flag and cotransfected into 293 cells alone or in combination with HA-TβRII. Immunoprecipitation with anti-Flag followed by immunoblotting with anti-HA revealed that TβRII can interact with ADAM12, and this interaction was not affected by TGF-β (). To provide further evidence that ADAM12 interacts with TβRII, we examined their colocalization by immunofluorescence. As expected, TβRII is localized predominantly in patched areas near the cell surface. Interestingly, we found that ADAM12 extensively colocalized with TβRII, confirming their interaction (). To examine whether the association of ADAM12 with TβRII can occur under physiological conditions, we used hepatic stellate cells (HSCs), Rhabdomyosarcoma (RD), and C2C12 cells, three cell lines that were previously described to express detectable ADAM12 (; ; ). In immunoprecipitates prepared with preimmune antisera, no TβRII was coprecipitated. However, in the anti-ADAM12 immunoprecipitates, we could clearly detect TβRII coprecipitating with ADAM12 (). Formation of the endogenous ADAM12–TβRII complex was also demonstrated by anti-ADAM12 immunoblotting of anti-TβRII immunoprecipitates (). The interaction of ADAM12 with TβRII is specific because we were unable to detect an interaction between TβRII and ADAM10 or ADAM17 (), which share the structure organization with ADAM12. Similarly, we were unable to see an interaction between ADAM12 and the bone morphogenetic protein type II receptor (). To explore the functional significance of the interaction between ADAM12 and TβRII, we investigated whether the expression of ADAM12 may influence TGF-β–mediated transcriptional esponses. For this, we made use of the TGF-β/Smad2-responsive reporter ARE-Lux () and found that the expression of ADAM12 resulted in an approximately fivefold increase in TGF-β–induced transcription (). A similar effect of ADAM12 was observed with the TGF-β/Smad3-responsive reporter CAGA-Lux (approximately threefold in and sixfold in ; ). Next, we attempted to confirm the role of ADAM12 in enhancing TGF-β signaling by investigating its effect on the expression of endogenous plasminogen activator inhibitor-1 (PAI-1), which contains CAGA boxes in the promoter. The results showed that the TGF-β–dependent expression of PAI-1 was increased by the expression of ADAM12 (Fig. S1 A, available at ). During the course of these analyses, we also investigated the role of endogenous ADAM12 in enhancing the transcriptional activation of collagen I (COL1A2) by TGF-β. For this, HSC cells were treated by ADAM12 antisense oligonucleotides before TGF-β stimulation, and the expression of ADAM12 or COL1A2 was analyzed. As we recently reported (), TGF-β treatment induces an accumulation of ADAM12 mRNA and protein, and this increase was reduced to the background level by ADAM12 antisense. Similarly, treatment of cells with antisense to ADAM12 attenuated the TGF-β–dependent induction of COL1A2 mRNA (Fig. S1 B). To confirm these results, we depleted HSC, RD, and C2C12 cells from ADAM12 by RNAi. When ADAM12 was targeted in these cells using a specific short hairpin RNA (shRNA), both the steady-state levels and the TGF-β–dependent accumulation of ADAM12 were reduced. Interestingly, the knockdown of ADAM12 resulted in a decrease in the TGF-β–induced expression of PAI-1 (). A similar result was obtained with JunB (), the expression of which is up-regulated by TGF-β through a mechanism similar to that of PAI-1. To investigate the mechanism underlying the effects of ADAM12 on TGF-β signaling, we investigated whether the expression of ADAM12 may regulate the TGF-β–dependent phosphorylation of Smad2. We observed that exposure of cells to TGF-β resulted in increased Smad2 phosphorylation, and this effect was further enhanced by the expression of ADAM12 (). Consistent with this, the expression of ADAM12 enhanced the ability of TGF-β to induce assembly of the Smad2–Smad4 complex (). In addition, the depletion of endogenous ADAM12 by RNAi suppressed Smad2 phosphorylation (). Collectively, these data suggest that ADAM12 may function to enhance TGF-β signaling by facilitating Smad2 phosphorylation and its subsequent heterodimerization with Smad4. At least six members of the ADAM family have been demonstrated to have proteolytic activity, including ADAM12 (; ). In initial experiments, we found that a truncated form of ADAM12 (ADAM12-tail), which lacks the cytoplasmic domain, retains its ability to enhance TGF-β signaling (). Therefore, we sought to investigate whether the increase in TGF-β transcriptional activity mediated by ADAM12 may involve its catalytic activity. To approach this question, we investigated the effect of phenanthroline, a specific metalloproteinase inhibitor, on the ability of ADAM12 to enhance TGF-β transcriptional responses. Surprisingly, exposure of cells to phenanthroline failed to suppress the effect of ADAM12 on TGF-β–induced CAGA-Lux (). In another approach, we used ADAM12-E351Q, a protease inactive mutant. As shown in , the expression of ADAM12-E351Q enhanced TGF-β–induced transcription with an activity similar to that of wild-type ADAM12. Together, these results indicate that ADAM12 enhances TGF-β signaling through a protease-independent mechanism. During our immunofluorescence analyses, we observed that ADAM12 and TβRII are colocalized predominantly in patched areas near the cell surface in C2C12 cells, but a substantial fraction of both proteins can also colocalize in endosome vesicle-like structures (). This pattern of colocalization of ADAM12 and TβRII in the two compartments was also evident in Mv1Lu cells (), but their distribution is more pronounced in endosomal vesicles when compared with C2C12 cells (). Based on the findings that TβRII colocalizes with early endosomal antigen 1 (EEA1), a marker of early endosomes (), we sought to investigate whether ADAM12 colocalizes with TβRII in the EEA1-enriched compartment using Mv1Lu cells that exhibit extensive staining of these proteins in early endosomes (; ). As for TβRII, there is some colocalization of ADAM12 with EEA1 in Mv1Lu cells ( and Fig. S2, available at ), suggesting that ADAM12 may accumulate in early endosomes to which TβRII is internalized via clathrin-coated pits. To examine whether the localization of ADAM12 in early endosomes plays a role in TGF-β signaling, we examined the effect of inhibition of clathrin-mediated endocytosis by potassium depletion, which was reported to prevent endosome-dependent TGF-β signaling (). As shown in , potassium depletion decreased the ability of ADAM12 to enhance TGF-β–induced transcription. Potassium depletion also decreased TGF-β signaling in the absence of transfected ADAM12, but this effect seems to depend on ADAM12 because it was lost in cells depleted from endogenous ADAM12 by RNAi. In a control experiment, we found that potassium depletion can further decrease TGF-β–induced transcription in cells depleted from Smad3 (), supporting the hypothesis that potassium depletion may inhibit TGF-β signaling by specifically interfering with ADAM12 function. To provide further evidence that ADAM12 functions in TGF-β signaling by facilitating the trafficking of TβRII to early endosomes, we examined the localization of SARA, which has been shown to interact with TβRII at the plasma membrane and in EEA1-positive early endosomes (; ). We observed that the expression of ADAM12 caused the redistribution of the TβRII–SARA complexes from the plasma membrane into early endosomes (). This effect is likely to be direct because the expression of ADAM12 had no effect on the association of TβRII with several transmembrane proteins that could potentially prevent or enhance its trafficking (Fig. S3, available at ). To provide further evidence that ADAM12 facilitates the localization of TβRII in early endosomes, we tested the effect of nystatin, a sterol-binding antibiotic that is known to induce the redistribution of TGF-β receptors into EEA1-positive endosomes by affecting the raft structures (). We reasoned that if we induce the majority of TβRII to accumulate in early endosomes by an alternative approach, such as the treatment of cells with nystatin, ADAM12 should have no further effect on TGF-β–mediated transcription. As shown in , exposure of cells to nystatin caused a considerable increase in the TGF-β–mediated activation of CAGA-Lux, and this increase was not affected by the expression of ADAM12. Under these experimental conditions, the expression of Smad3 can synergize with nystatin to enhance TGF-β–induced transcription, arguing against the possibility that the lack of ADAM12 effect is caused by the ability of nystatin to elicit the maximum threshold level of TGF-β signaling in this cell system. Collectively, these results suggest that ADAM12 may function as an important component in TGF-β signaling by modulating the trafficking of the TGF-β receptor. The clathrin-dependent internalization into early endosomes promotes TGF-β signaling, whereas the lipid raft–caveolar internalization pathway is required for receptor turnover. To obtain direct evidence that the accumulation of ADAM12 in early endosomes plays a role in the up-regulation of TGF-β signaling, we examine whether the expression of ADAM12 interferes with TβRII degradation. To approach this question, we first investigated the effect of ADAM12 on TβRII ubiquitination. We observed that the coexpression of ADAM12 resulted in a substantial decrease in the ubiquitination of TβRII (). In support of this result, the expression of ADAM12 increased the steady-state levels of TβRII (). Furthermore, in pulse-chase experiments, the expression of ADAM12 resulted in a marked decrease in the turnover of TβRII (). A similar result was obtained with the cytoplasmic truncated form ADAM12-tail, which, like the wild-type counterpart, can enhance TGF-β signaling (). As a control, we found that expression of the extracellular soluble form of ADAM12 failed to stabilize TβRII (), providing support to the theory that ADAM12 may stabilize TβRII by facilitating its intracellular redistribution from the plasma membrane to early endosomes. In contrast to clathrin-enriched vesicles, TβRII enriched in caveolin1-positive vesicles was found to associate with Smad7, which is known to mediate the association of Smurf1/2 to receptors, leading to their degradation. To confirm that ADAM12 can interfere with the ubiquitin-dependent degradation of TβRII, we examined its effect on the association of TβRII with Smad7. We observed that the expression of ADAM12 induced a reduced assembly of the TβRII–Smad7 complex (). Further evidence that ADAM12 can modulate the interaction of Smad7 with TβRII was obtained by experiments showing a considerable increase in accumulation of the endogenous Smad7–TβRII complex in cells depleted from endogenous ADAM12 (). As Smad7 can restrict the access of Smad2 to TGF-β receptor, we also investigated whether endogenous ADAM12 regulates the association of endogenous Smad2 with endogenous TβRII. In fact, we found that the depletion of ADAM12 can interfere with the association of Smad2 with TβRII (). These results suggest that ADAM12 may counteract the internalization of TβRII into caveolin1-positive vesicles and may counteract its subsequent degradation. A fragment corresponding to the extracellular domain (20–160 amino acids) of human TβRII was cloned into pBTM116. The human cDNA libraries from placenta were constructed in pGADGH. A total of 10 × 10 independent colonies were screened as previously described (). The prey fragments of the positive clones were PCR amplified and sequenced. The human embryonic kidney cell line 293T, HSCs, human RD cells, mouse C2C12 cells, monkey kidney COS7 cells, and mink lung MvLu1 cells were transfected using LipofectAMINE-Plus reagent (Invitrogen) according to the manufacturer's instructions. For experiments with ADAM12 antisense, cells were incubated with 2 μM of antisense oligonucleotides to ADAM12 (CTCTCTTTTATGCCTTCT and CCCCATTCCTTTCTCC) or random control oligonucleotides (ACTACTACACTAGACTAC and GCTCTATGACTCCCAG) as previously described (). For RNAi experiments, cells were transfected with 0.5 μg of expression vector encoding the indicated shRNA. ARE-Lux, GAGA-Lux, FAST1, HA-Smad4, myc-Smad2, and myc-Smad7 were previously described (; ). The expression vector for HA-TβRII was provided by J. Wrana (Samuel Lunenfeld Research Institute, Mount Sinai Hospital, Toronto, Ontario, Canada). Expression constructs for wild type or mutants of ADAM12 and ADAM12 fused to EGFP were prepared as previously described (). The expression vector encoding ADAM12 shRNA or scrambled shRNA was constructed using the BLOCK-IT U6 RNA System (Invitrogen) according to the manufacturer's instructions. The expression vector for Flag-ADAM12 was obtained by fusing the Flag epitope to the N terminus of the ADAM12 fragment (amino acids 142–739) isolated in the yeast two-hybrid screen. HepG2, C2C12, or 293T cells were transfected by LipofectAMINE, and, 30 h later, they were treated for 18 h with 2 ng/ml human TGF-β1 (Sigma-Aldrich). Cell extracts were assayed for luciferase activity using the Dual Luciferase Reporter Assay System (Promega), and luciferase activities were normalized on the basis of Renilla luciferase expression from the pRL-TK control vector. For potassium depletion experiments, transfected cells were incubated in medium and water (1:1) for 5 min at 37°C followed by incubation in medium depleted or not depleted in KCl for 1 h at 37°C before stimulation with TGF-β. After transfection, cells were lysed in lysis buffer (), and cell lysates were subjected to immunoprecipitation with the appropriate antibody for 2 h followed by adsorption to Sepharose bead–coupled protein G for 1 h. Immunoprecipitates were washed five times with lysis buffer containing 0.5% NP-40. For the association of endogenous TβRII with endogenous ADAMs, immunoprecipitates were washed three times with lysis buffer containing 0.5% NP-40 and two times with lysis buffer containing 1% NP-40. Then, samples were separated by SDS-PAGE and analyzed by immunoblotting with the indicated antibodies. The following antibodies were used: anti-ADAM12 Rb 122 (), anti-Flag M2 (Sigma-Aldrich), anti-HA and anti–myc-9E10 (Boehringer Manheim), antiphospho-Smad2 (Cell Signaling Technologies), anti-Smad2 (Zymed Laboratories), anti-ADAM10 (ProSci), anti-ADAM17 (Chemicon), and antiactin, anti-TβRII, anti–bone morphogenetic protein RII, anti-Smad7, anti–PAI-1, and anti-JunB (Santa Cruz Biotechnology, Inc.). Cells were fixed in 3% PFA, permeabilized with 0.1% Triton X-100, and incubated for 60 min at room temperature with the appropriate primary antibody followed by the appropriate secondary antibody. The coverslips were washed, mounted in PBS containing 50% glycerol and 1 mg/ml 1,4-diazabicyclo[2.2.2]octane, and viewed on an automated microscope (DMRXA2; Leica) equipped with a camera (CoolSNAP ES N&B; Roper Scientific) and a 63× Hcx Pl Apo NA 1.32 oil objective (Leica). Z steps were submitted to deconvolution (nearest neighbor method) by using MetaMorph software (Universal Imaging Corp.). Total RNA were extracted by the guanidinium thiocianate/cesium chloride method, and real-time quantitative PCR was performed by the fluorescent dye SYBR green methodology as previously described (). Primer pairs for target genes were as follows: PAI-1, sense (5′-GTCTTTCCGACCAAGAGCAG-3′) and antisense (5′-CGATCCTGACCTTTTGCAGT-3′); ADAM12, sense (5′-GTTTGGCTTTGGAGGAAGCACAG-3′) and antisense (5′-TGCAGGCAGAGGCTTCTGAGG-3′); COL1A2, sense (5′-GGTGGTGGTTATGACTTTG-3′) and antisense (5′-ATACAGGTTTCGCCGGTAG-3′); and 18S, sense (5′-CGCCGCTAGAGGTGAAATTC-3′) and antisense (5′-TTGGCAAATGCTTTCGCTC-3′). Fig. S1 A shows the effect of increasing amounts of ADAM12 on expression of the TGF-β–responsive gene PAI-1. Fig. S1 B shows the TGF-β–dependent expression of endogenous ADAM12 or COL1A2 in cells treated with ADAM12 antisense or control oligonucleotides. Fig. S2 shows the colocalization of ADAM12 with EEA1 or TβRII in Mv1Lu cells. Fig. S3 shows the association of TβRII with several transmembrane proteins as indicated by labeling with a membrane-impermeable biotinylation reagent. Online supplemental material is available at .

DNA damage such as double-strand breaks (DSBs) causes rapid alterations of chromatin structure, including the posttranslational modification of histones through the activated PI3K-like kinases ATM (ataxia telangiectasia mutated) and ATR (ataxia telangiectasia and Rad3 related; ; ; ; ). The best studied of these modifications is the phosphorylation of histone H2AX, an isotype of histone H2A. Phosphorylated H2AX (γ-H2AX) is detected in mammalian cells within several minutes after ionizing radiation; this modification spreads along a megabase of chromatin surrounding a DSB (, ). The precise roles of this modification are still under investigation, but γ-H2AX formation appears to be required for and/or maintain the association of proteins involved in DNA repair and damage signaling, including Nbs1, Mdc1, and 53BP1 (; ; ). The absence of γ-H2AX impairs DNA repair, most notably the repair of sister chromatids (; ; ). In , the major H2A histone contains a phosphorylatable SQE site at its C terminus, and, for simplicity, we will refer to the yeast's histone H2A as H2AX. In budding yeast, a single HO nuclease-induced DSB also promotes the modification of chromatin around the break over a domain of ∼100 kb (). The presence of γ-H2AX in yeast leads to the recruitment of both cohesins (; ) and the Smc5/6 complex (; ) and, consequently, promotes sister chromatid repair of ionizing radiation. Moreover, there is a γ-H2AX–dependent association of chromatin remodeling complexes such as Ino80, SWR1, and NuA4 at the damage site (; ; ). The alteration of chromatin through γ-H2AX has been suggested to provide the platform to recruit or maintain activities needed for the efficient repair of DSB damage (; ; ; ). In addition, the presence of γ-H2AX acts as a signal to retard cells from reentering the cell cycle after DNA is repaired (); this signaling may be important to maintain genomic integrity. Although there have been extensive studies to understand the role of γ-H2AX in the DSB-responsive pathway, the mechanism by which γ-H2AX spreads along a global chromatin in response to a DSB is still not clear. Whether there are boundaries to γ-H2AX spreading is not known. We have investigated the factors responsible for the extent of spreading of γ-H2AX around a site-specific DSB. To understand the alteration of chromatin structure after DSB formation and its involvement in the DSB-responsive pathway, we decided to study the extent and the regulation of γ-H2AX spreading in budding yeast. Our finding that γ-H2AX cannot form in regions of silenced heterochromatin led us to explore similar questions in mammalian cells. In this study, we report that γ-H2AX is largely excluded from regions of heterochromatin both in yeast and in mammalian cells but that heterochromatin is not a barrier to spreading γ-H2AX beyond that region. We report the surprising finding that in yeast, subtelomeric regions in budding yeast are constitutively modified by γ-H2AX, suggesting that telomeres are at least transiently recognized as a form of DSB damage. To examine how far γ-H2AX spreads along the chromosome in response to a DSB, we performed chromatin immunoprecipitation (ChIP) with an antibody specific to budding yeast γ-H2AX (). A DSB was generated at by expressing a galactose-inducible HO endonuclease. Because the homologous sequences and were deleted in this strain, the DSB at could not be repaired by homologous recombination (), and, thus, both of the DSB-responsive kinases Mec1 and Tel1 could be activated by the persistence of DSB. 1 h after HO induction, a DSB had formed in > 90% of the cells. The extent of γ-H2AX surrounding the DSB was examined by quantitative PCR using primer pairs for sites on either side of the DSB (Table S1, available at ) compared with results before HO cleavage. Consistent with the previous studies (; ; ), γ-H2AX was enriched over a ∼50-kb region on either side of the DSB except for the regions very close to the DSB (). The highest increase of γ-H2AX was seen 10–30 kb from the either side of the DSB, but increased signal could be seen as far as 40–50 kb from the DSB. Previous studies have shown that either Tel1 or Mec1 could carry out γ-H2AX modification in asynchronous cells and that in G1-arrested cells, in which Mec1 is inactive and there is very little resection of DSB ends (), Tel1 was necessary and sufficient to modify the entire region (). Now, we report that in G2-arrested Δ cells, in which ends are continually resected, Tel1 itself can nevertheless fully modify the region in G2-arrested cells (). One possible reason for modifying histones over such a large domain would be to eliminate transcription that might compete with DNA repair proteins in binding to DNA. We report that the modification of histones over this domain does not appear to affect the state of transcription in this domain. Cells were arrested in nocodazole so that we would not monitor changes in cell cycle–regulated genes when the HO-induced population arrested at G2/M. About 150 genes were either turned off or turned on in response to DNA damage (); the complete data can be found at . Here, we focus on genes surrounding the DSB (the genes between α and are indicated; ). 1 h after HO-induced creation of an unrepaired break at , there were no notable changes in gene expression even though γ-H2AX modification can be seen in 15 min (). Over time, gene expression progressively stopped for genes further from the DSB; these changes correlate with the time it takes for these sequences to be rendered single stranded by 5′ to 3′ resection, moving at 4 kb/h (). The lack of additional spreading beyond 50 kb at 1 h after HO induction could be caused by the presence of barrier sequences. To test whether the spreading of γ-H2AX was limited by a barrier sequence to the right of the DSB, we deleted the normal HO cleavage site and inserted a site 17 kb to the right. In this circumstance, the entire region of γ-H2AX was displaced to the right (). Because γ-H2AX spread further to the right and did not extend as far as it had to the left, the spreading of γ-H2AX does not seem to be limited by boundaries. A similar result was found when we deleted the original HO cleavage site and inserted an HO cut site ∼600 bp from the centromere of chromosome III, (). Again, spreading of γ-H2AX covered ∼50 kb on both sides of the DSB. These results also demonstrated that the budding yeast's small centromere, which lacks pericentric heterochromatin (), is not a barrier to γ-H2AX spreading. As a further test, we introduced an HO recognition site into the left arm of chromosome VI in the strain lacking the HO recognition sequence at . γ-H2AX spreading on chromosome VI was nearly identical to that on chromosome III (). By creating a strain with DSBs on both chromosomes III and VI, we showed that the extent of γ-H2AX spreading was largely not affected by the presence of a second DSB in a different chromosome, although there seems to be a slight increase of the signal at 50 kb distal from the DSB on chromosome III (). We also examined γ-H2AX spreading when two DSBs were created in the same chromosome, chromosome III, with one DSB at and the other 600 bp to the left of . The distance between the two DSBs is ∼86 kb. With two DSBs, the overall extent was similar to the results expected if one added the results of strains with only a DSB at or at (; also see B). γ-H2AX covered the 86 kb of chromatin between the two breaks and spread ∼50 kb to the left of the DSB near , as it did with the single DSB next to There does seem to be a modest increase in γ-H2AX spreading around the DSB at . Thus, the presence of a second DSB in the same chromosome did not significantly affect the extent of γ-H2AX spreading at each DSB. γ-H2AX formed after DSB formation does not seem to be turned over or replaced around the DSB before loss by resection. The Mec1 and Tel1 kinases, which are responsible for creating γ-H2AX, can be inactivated by the PI3KK inhibitor caffeine (; ). If γ-H2AX was rapidly turned over and replaced by unphosphorylated H2AX and/or by the alternative histone H2A.Z (), we would expect to lose the γ-H2AX ChIP signal after inhibiting the Mec1 and Tel1 kinase activities by caffeine. Caffeine treatment mimics the deletion of both Mec1 and Tel1 in that it prevents γ-H2AX formation after methylmethanesulfonate (MMS) treatment (). 30 min after HO induction, we treated cells with 10 mg/ml of caffeine and examined γ-H2AX on the chromatin located 20 kb from the DSB 30 min and 1 h later, long before 5′ to 3′ resection of the DNA end would reach this region. The amount of γ-H2AX remaining at this site was not affected by caffeine treatment (), showing that γ-H2AX is not removed from DNA by rapid turnover. Our previous study showed that γ-H2AX is lost when a region becomes single stranded by DNA end resection, presumably because histones, or at least H2A–H2B dimers, are lost (). Here, we confirm and extend this conclusion, showing that there is a progressive loss of γ-H2AX from the region beginning near the DSB (, top). To establish that the loss of γ-H2AX depends on DNA resection, we looked at γ-H2AX when DNA end resection was inactive either in G1-arrested cells or in the cells overexpressing the CDK1/Clb inhibitor Sic1 (). The formation of HO-induced γ-H2AX was not impaired by arrest. In both conditions, a substantial amount of γ-H2AX modification persisted as long as 8 h after HO induction, when resection was severely impaired (, middle and bottom). In fact, there was a slight increase in the level of γ-H2AX 10 kb away from the DSB 4 h after HO induction when resection was blocked. These results show that the loss of γ-H2AX does not occur without DNA resection. The decrease of γ-H2AX ChIP signals 2 or 10 kb from the DBS at 8 h after HO induction is likely caused by residual DNA resection (Fig. S1, available at ). The results we have shown so far demonstrate that once γ-H2AX is formed in the first 30–60 min, there is little additional spreading of the modification to adjacent regions in the next few hours. But, if one looks at later times, when γ-H2AX is lost from single-stranded DNA at sites as far as 10 kb from the DSB, there is a large increase of γ-H2AX in the intact chromatin 70 kb away from the DSB (). We also note that there was a slight general increase (approximately two- to threefold) of γ-H2AX at sites on other chromosomes that did not suffer a DSB 8 h after DSB formation (Fig. S2, available at ). Similarly, Western blot analysis reveals a striking increase in γ-H2AX 4 and 8 h after HO induction compared with the level 1 h after HO induction (). However, when resection was blocked, the amount of γ-H2AX 70 kb away from the DSB did not significantly increase (). We suggest that single-strand DNA generated by resection results in the displacement of γ-H2AX but, at the same time, mediates the repositioning of kinases responsible for γ-H2AX so that more distant regions can now be modified. To examine whether Mec1 and Tel1 kinase are responsible for the late spreading of γ-H2AX, we looked at γ-H2AX by the Western blot analysis either in Δ or Δ cells. As we showed previously (), there is a substantial increase in total γ-H2AX 4–8 h after HO induction; however, this increase is absent in Δ cells but clearly seen in Δ cells (). In addition, γ-H2AX ChIP signal in the chromatin 70 kb away from the DSB did not increase in Δ cells 4 and 8 h as resection proceeded compared with that 1 h after HO induction (). The Mec1-interacting ATRIP protein Ddc2 is known to associate with replication protein A–coated single-strand DNA (). Therefore, we suggest that Mec1 is recruited to the newly generated single-strand DNA by DNA end resection and then can phosphorylate γ-H2AX in adjacent chromatin domains that were not initially modified when Mec1 and Tel1 were bound near the DSB end. Previously, we showed that the histone phosphatase complex containing Pph3 is responsible for dephosphorylating γ-H2AX, but only after it had been released from chromatin (). However, the cellular pool of γ-H2AX continues to increase despite the loss of γ-H2AX from single-stranded DNA (, A [top] and C). One source of the increase can be the late spreading of γ-H2AX to more distant regions, but there may also be a decrease in the rate of dephosphorylation of released γ-H2AX. Thus, in checkpoint-arrested cells, with an unrepaired DSB, γ-H2AX may not be dephosphorylated immediately after it is released from chromosomes. Alternatively but not exclusively, some of the released γ-H2AX may be reincorporated randomly into the chromosomes and/or H2AX in the random chromosomes may be moderately phosphorylated in a Mec1-dependent manner. The distribution of γ-H2AX across from a nearby DSB both in strains with a second DSB at () and with a single DSB near () showed that the centromere-specific chromatin structure did not inhibit γ-H2AX spreading. However, in contrast to the centromeres of fission yeast and of most other organisms, centromeres of budding yeast are very small and not especially heterochromatic (). To examine whether the presence of heterochromatin affects γ-H2AX modification, we examined silent heterochromatic and loci, which have highly positioned, largely deacetylated nucleosomes (; ). An HO cut site was introduced 7 kb to the right of . After DSB formation, γ-H2AX spread ∼50 kb from the right side of the DSB, as seen with DSBs at other sites; however, spreading of γ-H2AX in the left side of the DSB stopped at (, left). The lack of γ-H2AX increase at relative to a site on another chromosome is also seen in the 10 kb between and the telomere; however, the small fold increase in this region after HO-induced damage turns out to be attributable to an unexpectedly high level of γ-H2AX in subtelomeric regions in the absence of DNA damage (, right), as we discuss in detail below. To separate the possible overlapping effects of and the telomere on γ-H2AX spreading, we inserted the heterochromatic locus ∼27 kb to the right of the HO cut site and ∼40 kb from the telomere in a strain in which the normal sequence was replaced by marker (). After DSB induction, γ-H2AX spread ∼50 kb to the right of the DSB, but there was much less enrichment of γ-H2AX within the chromatin (). Although γ-H2AX increased 19-fold to the right of the inserted HMR sequences (at ∼38 kb from the DSB), there was only a fourfold increase within chromatin, which was 10 kb more proximal to the DSB. It is possible that the estimate of modification within is too high because it is possible that there is some ChIP of γ-H2AX that is actually in the immediately adjacent regions, as not all sonicated fragments will be of the average 500-bp size. However, in any case, it is clear that sequences are refractory to modification compared with sequences on either side. In the absence of the ectopic domain, there was 15-fold γ-H2AX induction ∼30 kb away from the DSB, which is approximately the same distance from the DSB as the inserted . These results suggest that the heterochromatic and loci do not act as barriers to γ-H2AX spreading but are themselves refractory to γ-H2AX formation. These observations suggest that the spreading of γ-H2AX from a DSB does not occur by a processive mechanism in which the kinase modification of one histone provides a platform for the modification of an adjacent or nearby histone octamer, or at least the kinase must be able to reach over a 3-kb segment of heterochromatin. As a further demonstration that heterochromatin and not simply chromatin covering specific sequences are unable to be modified, we desilenced the normal locus by deletion of the gene. Here, the α locus carries a single base pair mutation that prevents HO cleavage. After DSB induction, γ-H2AX was enriched over the unsilenced sequence in Δ cells four or nine times more compared with modification in cells (). We conclude that the inhibition of γ-H2AX formation depends on the silent chromatin status of . However, we note that the γ-H2AX epitope could be occluded in silent chromatin, preventing the detection of γ-H2AX by ChIP. We next examined γ-H2AX distribution in mammalian cells. The introduction of DNA breaks is known to induce chromatin remodeling to a more open state, including when the DNA breaks occur within heterochromatin (). We hypothesized that if neighboring condensed regions of chromatin not containing DSBs are refractory to the spreading of γ-H2AX, treatment with a histone deacetylase inhibitor that has the potential to open chromatin would result in the increased spreading of phosphorylated H2AX in response to DNA damage. Therefore, we examined the distribution, mean fluorescence intensity, and size (volume) of γ-H2AX containing foci in primary mouse embryo fibroblasts treated with 1 μM trichostatin A (TSA) for 8 h before exposure with 10 ng/ml of the radiomimetic neocarzinostatin (NCS) for 1 h. We found that the mean number of γ-H2AX foci per nucleus increased slightly from 17.4 ± 5.9 in NCS-treated cells to 28.4 ± 7.1 in TSA plus NCS–treated cells (). TSA treatment led to a greater accumulation of acetylated histones but, in the absence of NCS treatment, did not influence the formation of γ-H2AX foci (Fig. S3, available at ). More strikingly, a subset of foci per nucleus (on average 5/28) had a volume >1.0 μm compared with ∼1/18 in non-TSA–treated cells. Altogether, there were 256 large foci out of 1,305 foci ( = 46) in TSA-pretreated cells compared with only 35/784 ( = 45) in the untreated cells (P < 0.0001). Interestingly, several of the larger γ-H2AX foci localized adjacent to pericentric heterochromatin (), but the γ-H2AX signal was restricted to the periphery of the heterochromatin and did not extend into the interior of the heterochromatin. Although the mean fluorescence intensity measured for the individual foci did not change dramatically in the NCS-treated cells compared with the TSA plus NCS–treated cells, the volume of the foci did increase, leading to a shift in the foci distribution to a greater total or integrated fluorescence intensity in the TSA plus NCS–treated cells (Fig. S4). This indicates that more total H2AX is phosphorylated in the larger foci compared with the smaller foci. Thus, chromatin remodeling to a more open state is accompanied by an increase in the total amount of H2AX phosphorylated within the damaged region. We also examined the degree of colocalization of γ-H2AX with histone H3 acetylated at lysines 9 and 14 (acH3K9K14), a marker for highly acetylated histone H3 and a more open accessible chromatin state. The amount of acH3K9K14 increased in the TSA-treated cells (Fig. S5, available at ), and there was an increase in the colocalization coefficient of γ-H2AX colocalized with acH3K9K14 from 0.67 ± 0.13 ( = 20) in NCS-treated cells to 0.91 ± 0.09 ( = 20) in the TSA plus NCS–treated cells. Acetylated H3 (acH3K9K14) colocalized with γ-H2AX foci in cells not treated with TSA, but blocking histone deacetylase activity with TSA increased the amount of acetylated H3 and bolstered the association of γ-H2AX with acH3K9K14. Although more γ-H2AX colocalized with acH3K9K14 in the TSA plus NCS–treated cell, γ-H2AX and acH3K9K14 were found at the periphery of pericentric heterochromatin but not throughout the entire heterochromatin domain (Fig. S5). The accessibility of heterochromatin to immunostaining was substantiated by the localized labeling of methylated histone H3-K9, a marker for heterochromatin (unpublished data). Similar pretreatment with 5 mM sodium butyrate increased the volume of γ-H2AX foci in NCS-treated fibroblasts (unpublished data). Thus, global increases in histone acetylation leads to the further spreading of γ-H2AX in a subset of foci, and those foci are absent from heterochromatin. In budding yeast, we find that γ-H2AX is constitutively located at telomeres (, right); consequently, when one examines the fold increase after creating DSB damage, there is only a small fold increase over the already high base line (, left). The high level of γ-H2AX near telomeres is seen when one compares γ-H2AX ChIP signals at telomeres with those at other locations, such as (, right). We confirmed the high level of telomere-adjacent γ-H2AX at three chromosome ends by probing unique sequences near the left and right ends of chromosome III () as well as near the left telomere of chromosome V (not depicted). The extent of γ-H2AX distribution at telomere-adjacent regions is ∼10 kb, which was smaller than that in response to an HO-induced DSB. As expected, there was no γ-H2AX signal at telomeres or elsewhere in cells in which the normal H2A gene is replaced by H2A-S129A (unpublished data). These results suggest that γ-H2AX is constitutively distributed on the telomere-adjacent chromatin, leading to a small increase fold of γ-H2AX after a DSB formation. Telomeric 3′ single-stranded DNA is associated with various proteins, including Cdc13, that prevent telomeres from activating DSB repair or cell cycle arrest (; ). We asked whether more γ-H2AX would be generated in telomere-adjacent chromatin in response to decapping the telomere. In a mutant strain at the nonpermissive temperature, the telomeric DNA is uncapped and degraded by 5′ to 3′ end resection, leading to the cell cycle arrest in G2 (; ). We grew the cells at 25°C and shifted the temperature to 37°C. At the same tine, HO endonuclease was induced to make a DBS at . In response to the HO-induced DSB, we detected an ∼20-fold increase of γ-H2AX 20 kb away from the HO-induced DSB. Because of the constitutive presence of the modification near telomeres, there was only a twofold increase of γ-H2AX at the telomere-adjacent chromatin after the temperature shift (). Nonetheless, the absolute amounts of γ-H2AX normalized to the signal at were almost the same at both locations (). Moreover, γ-H2AX spread further in response to uncapping telomeres, past , in which little γ-H2AX was induced, for example. Indeed, γ-H2AX near the telomeres was lost at 4 h after temperature shift, coinciding with the degradation of telomeric DNA (). Thus, when telomeres are deprotected, they elicit the same modification as an HO-induced DSB. Telomeres are localized at the nuclear periphery by either the yeast's Ku proteins or by Esc1, both of which require Sir4 for the telomere tethering. We asked whether the displacement of telomeres from the periphery would influence the level of constitutive γ-H2AX near the telomere by examining cells lacking either Yku80 or Sir4 (), but almost the same amount of γ-H2AX was detected near the telomere in both mutants (). In Δ or Δ mutants, gene silencing by telomere position effect (TPE) is defective; the fact that the level of γ-H2AX is the same in wild-type or TPE-defective mutants suggests that checkpoint-mediated chromatin modification is independent of TPE. To be replicated, telomeres likely transiently dissociate from their capping proteins; indeed, the cell cycle checkpoint kinases Mec1p and Tel1p are recruited to telomere ends, and both kinases influence telomere length (; ). We hypothesized that γ-H2AX near the telomere could be generated when the telomere was associated with the kinases during DNA replication and, therefore, might be higher in G2 cells than in G1. We examined this possibility by comparing γ-H2AX at the telomere-proximal chromatin in G1-arrested cells with that in G2-arrested cells. Interestingly, a similar amount of γ-H2AX was detected in G2-arrested cells as well as in G1-arrested cells, in which the telomeres were fully capped and protected (). Indeed, γ-H2AX appeared even when Δ cells were arrested in G1. This result was surprising because Tel1 has been shown to be the sole kinase responsible for γ-H2AX when G1-arrested cells suffer a DSB. These results lead us to suggest that γ-H2AX formed near the telomere in S and G2 persists on the chromatin until the next cell cycle. Consistent with this idea, we show in the previous sections that there is little turnover of γ-H2AX when kinases are inactivated; consequently, γ-H2AX at telomeres may persist for some time. On the other hand, showed that Mec1 associated with the short telomeres in G1-arrested Δ cells. Therefore, it is also possible that H2AX at telomeres in G1 cells can be phosphorylated by Mec1 when Tel1 is not functional. We have shown that the extent of γ-H2AX spreading is not dependent on the location of DSB. There is a rather constant 50 kb of rapid modification (within 1 h) on either side of a DSB. This constraint is not apparently imposed by barrier sequences but may reflect a fundamental aspect of chromosome architecture that confines kinases associating with DSB ends from modifying more distant regions. A second important finding is that histone H2AX in heterochromatin is not efficiently modified in response to DSBs either in budding yeast or in mammalian cells. This does not imply that the DNA damage response is necessarily less efficient if the lesion occurs within heterochromatin. When DSBs are generated within heterochromatin itself, the chromatin surrounding the lesion remodels to a more decondensed configuration that appears to permit the recruitment of repair factors and γ-H2AX formation (). When a DSB is generated outside of heterochromatin, it is possible that the hypoacetylated state of yeast histones at lysine residues in their N termini precludes the spreading of the C-terminal phosphorylation to heterochromatin; thus, eliminating histone deacetylation in yeast by deleting restores γ-H2AX modification. A very important conclusion from our study is that the presence of a heterochromatic region does not preclude the modification of histones further away from the DSB. This finding argues that the process of modification is not strictly a processive hand-off process in which a newly phosphorylated γ-H2AX serves as a platform for Mec1 or Tel1 kinase to modify an adjacent histone. It is possible that the ends of the heterochromatic silent region form a loop that would allow a processive modification process to step over the silent region instead of traversing it nucleosome by nucleosome (Bystricky, K., personal communication). In addition, we have documented that γ-H2AX is lost when DNA is rendered single stranded by resection. The loss of γ-H2AX is progressive, beginning from the DSB. On the contrary, there is no significant loss of γ-H2AX when DNA end resection is inhibited either in G1-arrested cells or in the cells overexpressing the CDK1/Clb inhibitor SIC1. An important new result is that chromatin that is far distal from the DSB and, thus, not initially modified eventually does become modified as resection proceeds. This late, distant modification is dependent on but not on . Mec1 and its interacting partner Ddc2 are known to associate with replication protein A–coated single-strand DNA (). As resection proceeds, Mec1–Ddc2 can associate with newly generated single-stranded DNA and create γ-H2AX on still more distant regions. Thus, although 5′ to 3′ resection displaces the initially modified γ-H2AX/H2B dimer, it also provides new recognition sites for Mec1, leading to an extension of the modified chromatin domain. Finally, we have found that chromatin near telomeres have constitutively high levels of γ-H2AX. Previously, the DSB-responsive kinases Mec1 and Tel1 were shown to be recruited to the telomeres at specific times in the cell cycle, playing roles in telomerase recruitment (; ). Our current observation suggests that telomeres are at least transiently recognized as DSBs, although without triggering cell cycle arrest. Unlike what occurs with a single DSB, in which γ-H2AX is lost either because the locus is repaired or by extensive 5′ to 3′ resection (), γ-H2AX at undamaged telomeres seems to be persistent. Telomeres do not normally undergo extensive resection that would remove γ-H2AX nor do telomeres engage frequently in recombinational repair, so the modification is not removed. It is not possible to determine what proportion of any given telomere is modified by γ-H2AX in a population of cells; it is quite possible that there is a changing subset of ends that are detected as DSBs. Strain JKM179, in which a galactose-induced DSB at created by HO endonuclease cannot be repaired by homologous recombination, has been previously described (). Strains lacking the HO cleavage site were selected as rare, imperfect, nonhomologous end-joining events from cells in which the HO endonuclease is continually expressed. The ectopic HO cut site cassette was constructed by inserting a 117-bp cleavage site derived from a adjacent to a HPH-MX marker, and the consequent 1.8-kb HO cassette containing the HPH marker was introduced at 217,940 bp on chromosome III to create the HO cut site 17 kb to the right of (YJK1) by lithium acetate–mediated transformation. The HO cassette was also integrated on chromosome III either at 113,500 bp on chromosome III, 600 bp to the left of (YJK2), or at 18,719 bp, which is 7 kb to the right of (YJK21). The HO cut site on chromosome VI was generated by integrating the HO cassette at 195,680 bp on chromosome VI (YFD032). To construct the strains containing two HO-inducible break sites, either JKM179 or YFD032 was transformed with the HO cassette for YJK9 (the second break near ) or YJK15 (the second break on chromosome VI), respectively. To delete the endogenous domain (YJK52), YJK21 was transformed with BamHI-digested pJH455 (Δ∷). The strain containing the ectopic α at 41 kb away from the left end of chromosome III (YJK79) was constructed by crosses between a derivative of YJK52 containing pa and a derivative of XW426 (α Δ∷ 41kb∷∷α Δ∷). Deletions of (YJK53) and (YJK75) were created by PCR amplification of KAN-MX–marked gene deletions from a collection of yeast deletion mutants (Research Genetics) transformed into YJK52. To delete , a derivative of YJK21 containing α-inc (YJK27) was transformed with Δ∷KAN-MX, which was PCR amplified from a Research Genetics yeast deletion mutant collection. To overexpress the CDK1 inhibitor SIC1, the cells containing the gal promoter–fused allele was used (). YSL187 (a Δ∷), a derivative of JKM179, was used for the α-factor arrest experiment. The allele (YMV021) was introduced into JKM179 by pop-in/pop-out of a -containing plasmid, pVL451, provided by V. Lundblad (Salk Institute, La Jolla, CA). ChIP of γ-H2AX was performed as previously described () using an antibody against yeast γ-H2AX, which was provided by C. Redon and W. Bonner (National Institutes of Health [NIH], Bethesda, MD). Cells were grown to a density between 5 ×10 cells and 1 × 10 cells/ml in yeast extract/lactate medium, and HO endonuclease was induced by adding 2% (wt/vol final concentration) galactose. DNA and proteins in the cells were cross-linked by the addition of 1.4% formaldehyde to 45 ml of cultures for 10 min. Cells were lysed with glass beads, and the extracts were sonicated to shear DNA to an average size of 0.5 kb. Immunoprecipitation samples were incubated with 50 ng/ml anti–γ-H2AX serum for 1 h at 4°C and bound to protein G–agarose beads for 1 h at 4°C. After a series of washing, samples were eluted from the beads followed by the reversal of cross-linking for 6 h at 65°C. Finally, proteins were removed from the sample by proteinase K and phenol extraction, and DNA was ethanol precipitated. The γ-H2AX ChIP signal was quantified by quantitative PCR with multiple primer pairs specific to chromatin regions surrounding the HO-induced DSB (Table S1). PCR was performed with a real-time PCR machine (Chromo 4; MJ Research) except for . With each primer pair, the number of amplification cycles that were required for the sample's response curve to reach a particular threshold fluorescence signal level was measured. The amount of chromatin immunoprecipitated template DNA for the reaction was then estimated from a standard curve based on serial dilution of a standard PCR product (). The ChIP signals in were measured by quantifying the band intensities on 1.5% agarose gels with Quantity One (Bio-Rad Laboratories). The band intensities were also converted to the initial amount of DNA template by being estimated from a standard curve. The ChIP signal at each locus was normalized to that at in chromosome VIII, in which DSB was not induced. The fold increase of γ-H2AX was calculated by dividing the ChIP signal at 1 h after HO induction (T1) by that without HO induction (T0). Cell extracts were prepared by trichloroacetic acid and were subjected to Western blot analysis. γ-H2AX was detected by polyclonal anti-yeast γ-H2AX antibody (1:10,000; a gift from C. Redon and W. Bonner). As a loading control, carboxy peptidase Y (CPY) was visualized with the monoclonal anti-CPY antibody (1:10,000; Invitrogen). Wild-type primary mouse embryo fibroblasts were grown on glass coverslips and treated with 10 ng/ml NCS for 1 h or with 1 μM TSA for 8 h and were treated with 10 ng/ml NCS for an additional hour before being fixed in 2% freshly prepared PFA in PBS for 5 min at room temperature. Cells were immunolabeled using the following antibodies, either individually or in combination: monoclonal antiphospho-H2AX antibody (clone JBW301; 1:1,000; Upstate Biotechnology), rabbit polyclonal antiacetylated histone H3 (K9 and K14; 1:1,000; Upstate Biotechnology), rabbit polyclonal lysine-methylated histone H3 (H3meK9; 1:200; Upstate Biotechnology), and goat anti–rabbit AlexaFluor488 (1:500) or goat anti–mouse AlexaFluor546 secondary antibody (Invitrogen). The cells were labeled for 30 min in PBS containing 100 nM DAPI, rinsed with PBS, and mounted on glass microscope slides in glycerol-based mounting media containing -propyl gallate antifade (Sigma-Aldrich). For volume measurements, image stacks were background subtracted and imported into Imaris version 4.2 (Bitplane) image processing and analysis software using the Surpass volume rendering module. Maximum intensity projections were calculated for all fluorescence channels. Isosurface volume renderings of the γ-H2AX fluorescence channel were calculated without resampling or smoothing the dataset. The consistent threshold value, which is representative of signal above background from the negative control, was used in the isosurface calculation. Based on the visual inspection of all image stacks collected, the individual isosurface renderings were split into individual objects within a group using a value of 100 as the maximum number of objects per group. Statistics generated from the objects within the group were exported to a spreadsheet program, and the mean volume (micrometers) was calculated. A snapshot of the γ-H2AX isosurface rendering superimposed onto the DAPI maximum intensity projection was saved and organized into figures using Photoshop version 8.0 (Adobe). The mean volume of 1.0 μm was empirically chosen as a cut-off value for significantly large γ-H2AX foci based on the fact that in the NCS-only treated cells, the vast majority (95%) of foci fell below 1.0 μm in size. Colocalization analysis was performed on 3D reconstructions from image stacks that were background subtracted and set at a consistent threshold value (the same γ-H2AX threshold value used above for individual foci volume measurements). The colocalization module in Imaris software (Bitplane) was used, and coefficients were measured for γ-H2AX colocalizing with acH3K9K14 within a range of 0 to 1.0. Fig. S1 documents the inhibition of end resection at a DSB in Cdk1-inhbitied or G1-arrested cells. Fig. S2 shows the extent of γ-H2AX modification when a DSB was created on chromosome VI. Fig. S3 presents γ-H2AX foci and acetylated histone H3 in NCS-treated or TSA plus NCS–treated wild-type mouse embryonic fibroblasts. Fig. S4 summarizes the distributions of γ-H2AX foci in NCS- and TSA plus NCS–treated mouse embryonic fibroblasts in terms of mean fluorescence intensity, foci volume, or integrated fluorescence intensity. Fig. S5 presents the colocalization of γ-H2AX and acetylated histone H3 in mouse embryo fibroblasts containing hyperacetylated histones. Table S1 lists all primer pairs used in ChIP experiments. Online supplemental material is available at .

Polycomb group (PcG) proteins are conserved transcriptional regulators with roles in cell identity, lineage specification, cell cycle control, and X inactivation (; ; ; ). Their function in regulating homeotic genes has been established in many organisms, including flies and mammals. Several PcG genes are essential for development. PcG proteins exert their function, in part, via histone-modifying activities. Two biochemically distinct complexes have been isolated and possess catalytic activity. Polycomb repressive complex 1 (PRC1) contains the RING finger domain proteins Ring1A and Ring1B, which mediate the monoubiquitination of histone H2A lysine 119 (H2AK119ub1) via an E3 ubiquitin ligase activity. PRC2 consists of the PcG proteins Eed, Suz12, and Ezh2 and catalyzes histone H3 lysine 27 di- and trimethylation (H3K27me3; ; ; ; ) as well as the methylation of histone H1 lysine 26 (). PcG complex–mediated histone modifications have been associated with silent chromatin. H3K27me3 has been shown to increase the affinity for binding of chromodomain-containing Polycomb proteins such as Cbx7, which are also components of PRC1 (; ). Based on this, PRC2-mediated H3K27me3 has been proposed to act as a recruitment signal for PRC1, which, in turn, would catalyze H2AK119ub1. Consistently, Ring1B binding is compromised in embryonic stem (ES) cells carrying a mutation in the PRC2 gene , which causes a loss of H3K27me3 (). Furthermore, loss of PRC1 components results in the disruption of PRC1 binding at genes (, ). Mammals achieve dosage compensation between XX females and XY males by the inactivation of one of the two X chromosomes in female cells. X inactivation is initiated by RNA, which associates with the inactive X chromosome (Xi) and initiates chromosome-wide silencing. is crucial for the initiation of X inactivation but is dispensable for maintaining the Xi at later stages of differentiation, when other epigenetic mechanisms, including DNA methylation, ensure stable silencing (; ; ). PcG proteins are recruited by and contribute to the establishment of histone modifications along the Xi (). The initiation of X inactivation is characterized by chromosome-wide histone modifications, including H3K27me3, H2AK119ub1, and monomethylation of histone H4 lysine 20 (H4K20me1; ; ; ; ). A mutant RNA, which lacks the repeat A sequence and, thus, cannot cause transcriptional repression, is still able to recruit PcG proteins and establish chromosome-wide histone modifications. This indicates that PcG recruitment occurs independently of the initiation of silencing (; ; ) and that PcG recruitment is not sufficient for the initiation of chromosome-wide silencing. An involvement of PcG proteins in the maintenance of X inactivation has been proposed based on their function in maintaining the repression of homeotic genes. However, is required for the recruitment of PcG proteins and histone modifications throughout ES cell differentiation and in differentiated cell types. This suggests that in X chromosome inactivation, PcG complexes have a function in the establishment of the maintenance of stable silencing rather than being silencing factors themselves. Thus, recruitment of PcG complexes in X inactivation might differ from recruitment to developmental control genes. Consistent with an involvement in the maintenance of X inactivation, the PRC2 gene is required for maintenance of the Xi in differentiating trophoblast stem cells (). In contrast, PRC2 function is dispensable for X inactivation in embryonic cells (; ), and and H2AK119ub1 can be recruited to the -expressing chromosome in cells lacking PRC2 function caused by disruption of the gene (). This suggests a PRC2-independent mode of Ring1B recruitment in X inactivation. The ability of -deficient ES cells to initiate chromosome-wide silencing could either be explained by a potential redundancy of PRC1 and PRC2 or, alternatively, Ring1B could be of primary functional importance for X inactivation in embryonic cells. Previously, it has been shown that both Ring1A and Ring1B mediate H2AK119ub1 on the Xi in mouse embryonic fibroblasts (). is an essential gene in the mouse, and its mutation leads to gastrulation arrest and cell cycle inhibition (). An involvement in embryonic axis specification and regulation of homeotic genes has also been demonstrated (). Ring1B appears to be associated with several distinct complexes. Apart from its function as a catalytic E3 ubiquitin ligase in the PRC1 complex, recruitment of Ring1 proteins by the transcriptional repressor E2F6 () and the spliceosomal component Sf3b1 () has been observed. It is conceivable that histones are not the only targets to be modified by PcG proteins. Recent results indicate a function for Ring1B in ubiquitination of the PcG-associated protein Ring1 and YY1-binding protein (RYBP; ). In the present study, we address the function of Ring1B in the regulation of developmental control genes, PRC1 protein levels, and the initiation of X inactivation in mouse ES cells. To investigate the function of in clone 36 ES cells (), we generated a targeting vector that replaced the start codon and the catalytically active RING finger domain with a floxed hygromycin selection marker (). A splice acceptor site and an SV40 polyA sequence flanking the selection marker were inserted to avoid production of truncated protein products. Targeting of the first allele was efficient with a frequency of 15% and was confirmed by Southern analysis (). The second allele could only be targeted with an efficiency of 0.3%, and clones could not be isolated as a result of a strong tendency to differentiate. Following a conditional targeting strategy (), ES cells were obtained with a frequency of 5% (). Cre-mediated recombination established 36 clones with a frequency of 43% as confirmed by Southern analysis (). About half of these clones were lost as a result of spontaneous differentiation, but the other half could be recovered and cultured for >20 passages. However, 36 ES cells appeared to have a strong propensity to differentiate, were extremely sensitive to stress, especially upon freezing and thawing, and could only be maintained under pristine culture conditions. The absence of Ring1B protein was confirmed by Western analysis in two independently derived 36 ES clones (). In 36 ES cells, Ring1B protein levels were reduced, indicating that the conditional targeting vector yielded a hypomorphic allele before Cre-mediated recombination (), which is similar to a hypomorphic allele reported previously (). Notably, the abundance of the PRC1 proteins Mph1, Mel18, and Rybp was reduced to undetectable levels in -deficient 36 ES cells (). The levels of Mph2 and Mpc2 were strongly reduced (). All PRC1 proteins were abundantly detected in control clone 36 ES cells. We conclude that disruption of leads to the reduction of several PRC1 proteins in -deficient ES cells. PcG proteins have been implicated in the repression of developmental control genes in ES cells (). To investigate whether the derepression of such genes occurs in -deficient ES cells and could contribute to the instability of stem cell identity, we performed an expression analysis of lineage-specific genes, including the trophoblast stem cell markers and and the markers for extraembryonic endoderm , , and , which are normally not expressed during ES cell differentiation. All trophoblast stem cell and extraembryonic endoderm markers were repressed in control clone 36 ES cells but were up-regulated in 36 ES cells (). In 36 ES cells, which are deficient for PRC2 function as a result of a null mutation in (), a substantial up-regulation of , , and but only a weak derepression of and was observed (). The pattern of derepression of lineage-specific genes in - and -deficient ES cells is largely consistent with the previously reported binding of and to the respective chromosomal loci in mouse ES cells (). has not been reported as a PcG target, and derepression could be an indirect effect of the loss of . Deregulation of developmental control genes is not limited to markers for extraembryonic development, as , a marker for neuronal differentiation, is slightly up-regulated in - and -deficient ES cells. Expression of the pluripotency-associated gene was observed in -deficient, -deficient, and control ES cell lines at comparable levels ( and Fig. S1 A, available at ). We conclude that -deficient ES cells can be isolated and maintained but show the derepression of lineage genes, which contributes to a predisposition to differentiation and compromises stem cell maintenance. To analyze the differentiation potential of -deficient ES cells, we investigated their ability to form embryoid bodies (EBs; ). After 7 d in suspension culture, a portion of 36 EBs formed large, hollow spheres. In contrast, EBs derived from control clone 36 ES cells formed compact aggregates (). When these EBs were plated on gelatine-coated dishes, they attached and formed beating structures indicative of the development of contractile cardiomyocytes. EBs derived from 36 ES cells neither attached nor formed contractile cardiomyocytes after 7 d in suspension culture but continued to grow in suspension as hollow spheres, reaching a diameter of up to 5 mm after 3 wk (). 36 EBs, which have reduced Ring1B protein levels, did not attach efficiently but formed contractile structures in suspension culture after 1 wk (Video 1, available at ). These peculiar beating spheres were not observed in control clone 36 EBs and could indicate cardiomyocyte development at reduced Ring1B protein levels. A deregulation of lineage gene expression was observed in 36 EBs after 2 wk of differentiation, which is consistent with the aberrant differentiation potential (Fig. S1 D). When 36 EBs were plated on gelatine after 48 h, some attached, and, after 3 wk, cells with a morphology reminiscent of trophoblast giant cells developed (unpublished data). Consistent with this, we observed the expression of , which is normally exclusively expressed in trophoblast giant cells (). To investigate whether Ring1B controls the expression of other PcG genes, we performed expression analysis of the PRC1 genes , , , , , , , , and the PRC2 member (). As expected, we observed a loss of expression in 36 ES cells (). Transcription of the PcG genes , , , , and was unaffected by the loss of either or (). However, the levels of and transcript were up-regulated in 36 and 36 ES cells (), which is consistent with the reported binding of Ring1B and Eed to the and promoters in mouse ES cells (). Transcription of was found to be slightly up-regulated in - but not -deficient ES cells. We conclude that in general, PcG genes are not regulated by Ring1B at the transcriptional level, but we find that , , and transcription is negatively regulated by . This showed that the loss of PRC1 proteins in -deficient ES cells was not mediated by transcriptional repression but occurred at the level of protein stability or translation. Compared with clone 36 ES cells, Bmi1 protein levels were reduced to undetectable levels in 36 ES cells but were more abundant in 36 ES cells (). Thus, the up-regulation of transcription in 36 and 36 ES cells resulted in an accumulation of Bmi1 protein in -deficient but not -deficient ES cells. This could be explained by a critical role of Ring1B in stabilization of the PRC1 complex. Consistent with this, several PRC1 proteins could not be detected in -deficient ES cells () despite unaffected transcription (). The PRC2 protein Suz12 was unaffected by the loss of in 36 ES cells (). We conclude that Ring1B is critical for PRC1 but not PRC2 protein levels in ES cells, possibly by the regulation of translation or protein stabilization. To characterize the effect of PRC1 disruption on histone modifications associated with X inactivation, we performed Western analysis of ES cells lacking . H2AK119ub1 was absent in 36 ES cells compared with clone 36 and 36 ES cells (), which is consistent with a previous report of a crucial function of Ring1B in the ubiquitination of histone H2A (). H3K27me3 was unaffected in 36 ES cells but was absent in 36 ES cells, which lack PRC2 (). Global levels of H4K20me1 as well as macroH2A were unchanged in - and -deficient ES cells compared with control clone 36 ES cells (). To analyze the recruitment of PcG proteins by and the establishment of histone marks, we performed immunofluorescence analysis combined with RNA FISH. In clone 36 and 36 ES cells, expression can be induced from a transgene inserted into chromosome 11 by the addition of doxycycline (). Upon the addition of doxycycline for 3 d, was induced efficiently in 36 ES cells, and a focal cluster was observed in 57 ± 5% of the nuclei compared with 62 ± 5% in control clone 36 ES cells. In 36 ES cells, colocalization of focal H2AK119ub1 staining with was reduced and observed in 7 ± 4% of the nuclei compared with 90 ± 6% in control clone 36 ES cells after 3 d of induction with doxycycline (). Colocalization of H3K27me3 with was unaffected by the loss of with 92 ± 5% and 95 ± 3% of the nuclei showing focal staining in wild-type and 36 ES cells, respectively ( and Fig. S2 A, available at ). Similarly, the establishment of H4K20me1 on the -expressing chromosome was not impaired by the loss of , and focal staining was observed in 51 ± 5% and 46 ± 6% of wild-type and 36 ES cells, respectively (). We next characterized the recruitment of PcG proteins by (). Mph1 was recruited in 30% of control 36 but not in 36 ES cells. In addition, the immunofluorescence signal for Mph1 was weaker in 36 ES cells compared with wild-type ES cells (not depicted), which is consistent with our observation that the levels of several PRC1 proteins were strongly reduced in -deficient ES cells (). In contrast, recruitment of the PRC2 members Ezh2 and Suz12 was not affected by the loss of in ES cells (). Colocalization of Ezh2 with was observed in 96 ± 1% and 91 ± 2% in wild-type and -deficient ES cells, respectively. Similarly, Suz12 colocalized with in 89 ± 4% of wild-type clone 36 and 90 ± 6% of 36 ES cells. To demonstrate the specificity of the effect of the deletion on H2AK119ub1 and PcG recruitment in 36 ES cells, a knockin strategy was used to rescue the disruption after attempts to transiently or stably express transgenes were unsuccessful. For this, we used the conditional vector to establish 36 ES cells. In 36 ES cells, H2AK119ub1 is observed on the -expressing chromosome in 75% of analyzed nuclei (). Furthermore, Mph1 protein levels and recruitment by in 36 ES cells were comparable to clone 36 ES cells (unpublished data). This demonstrated that the loss of specifically disrupts PRC1 function and H2AK119ub1 in ES cells. However, PRC2 function as well as H4K20me1 is recruited by independent of PRC1 in 36 ES cells. After 3 d of retinoic acid–induced differentiation in the presence of doxycycline, the colocalization of H2AK119ub1 with became evident in 36 ES cells, and, after 8 d, 72 ± 6% of the cells showed the colocalization of focal H2AK119ub1 staining with compared with 90 ± 2% of control clone 36 cells. We found that Ring1A protein levels were strongly up-regulated upon the differentiation of 36 and control ES cells (Fig. S1, A and B), and we observed Ring1A colocalization with on day 8 of differentiation (Fig. S2 C). This suggested that Ring1A could possibly contribute to H2AK119ub1 in differentiated -deficient cells, which is consistent with a previous report that can compensate for the disruption of in embryonic fibroblasts (). Furthermore, the establishment of H2AK119ub1 early in the differentiation of 36 cells could explain the small proportion of -deficient ES cells showing the colocalization of H2AK119ub1 and . Nonetheless, Western analysis demonstrated that global H2AK119ub1 levels were not restored upon differentiation in 36 cells (). In addition, was unable to recruit Mph2 efficiently in -deficient cells despite the recovery of H2AK119ub1. On day 8 of differentiation, 30% of control clone 36 but only 2 ± 2% of 36 cells showed the colocalization of Mph2 with (). This could be explained by reduced Mph2, Bmi1, and Mel18 protein levels in differentiated 36 cells compared with controls (Fig. S1 E). H3K27me3 colocalization with Xist was unaffected and was observed in 63 ± 16% of differentiated 36 cells comparable with 61 ± 1% in controls (). Furthermore, macroH2A recruitment by Xist was not affected in Ring1B-deficient cells after 8 d of differentiation, and 78 ± 5% of H3K27me3-positive cells showed colocalizing macroH2A signals compared with 76% of control clone 36 cells (Fig. S3 A, available at ). Control 36 cells showed a 34 ± 6% colocalization of Ezh2 and a 44 ± 2% colocalization of Suz12 with Xist. In 36 cells, the percentages decreased to 12 ± 4% and 16 ± 7% for Ezh2 and Suz12, respectively, after 8 d of differentiation (). This is consistent with Western analysis showing a reduction of the PRC2 protein levels of Suz12 and Ezh2 (), possibly as a result of the heterogeneous expression of PRC2 proteins in a subset of cells (not depicted). However, the reduction in the abundance of PRC2 proteins in 36 was not as severe as in Eed-deficient cells () and did not lead to a measurable difference in H3K27me3; thus, this might not be of functional relevance. We conclude that despite a recovery of H2AK119ub1 colocalization with Xist upon the differentiation of Ring1B-deficient ES cells, the stability of the PcG system critically depends on the presence of Ring1B. A redundant E3 ligase activity can remedy defects in ubiquitination in X inactivation but not in global histone H2A ubiquitination. We next assessed the ability of to initiate gene silencing in the absence of Ring1B and PRC1. The induction of expression in clone 36 ES cells causes repression of a puromycin marker gene (puro), which is cointegrated with the transgene. Thus, -mediated silencing can be analyzed by Northern analysis of puro expression. After the induction of for 3 d, repression of the puro marker in 36 ES cells was comparable with control 36 ES cells (). We further confirmed this result by analysis of cell growth under puromycin selection. -deficient as well as control 36 ES cells became puromycin sensitive upon the addition of doxycycline to the medium (Fig. S3 B). A control heterozygous 36 ES cell clone that had lost the ability to express remained puromycin resistant upon exposure to doxycycline. We conclude that initiation of silencing by is independent of and H2AK119ub1. To investigate whether is essential for the maintenance of silencing, ES cell differentiation was induced with all-trans–retinoic acid. was either turned on from the beginning of differentiation, for 4 d followed by 4 d without induction or cells were differentiated without doxycycline for 8 d in parallel cultures (). Expression of the puro marker gene was quantified on day 8 of differentiation by Northern analysis. Repression of the puro marker was observed in -deficient 36 cells comparable with control 36 ES cells after 8 d of differentiation in the presence of doxycycline (). Furthermore, silencing was efficiently maintained independent of expression in -deficient cells, which were differentiated in the presence of doxycycline for 4 d followed by 4 d without. To confirm that the maintenance of -mediated silencing is not limited to the cointegrated puro marker, we performed Northern analysis of the imprinted gene that is expressed from the maternal chromosome 11, into which the transgene was integrated (; ). We found that is repressed by expression in clone 36 control and 36 cells after day 8 of differentiation in the presence of doxycycline. Repression was further stably maintained if was turned off after 4 d of differentiation (). We further confirmed these results by real-time PCR analysis of expression, a nonimprinted gene on chromosome 11 (Fig. S3 C). This demonstrated that is dispensable for the chromosome-wide maintenance of silencing in differentiated cells. We next assessed the ability of -deficient cells to establish a chromosomal memory that is set up by the expression of early in differentiation and allows for the efficient recruitment of H3K27me3 by in differentiated cells (). We found that the establishment of memory is independent of (Fig. S3 D). After 15 d of doxycycline treatment, 58% of clone 36 and 71% of 36 cells with a focus also showed colocalizing H3K27me3. The delayed induction of after 4 d of differentiation without doxycycline resulted in reduced H3K27me3 recruitment, with 37% of clone 36 and 35% of 36 cells showing focal H3K27me3 colocalizing with . When was turned on for the first 4 d of differentiation followed by 4 d without doxycycline and reinduction for 7 d more, H3K27me3 recruitment was observed in 58% of control clone 36 and 65% of 36 cells comparable with differentiation in the continuous presence of doxycycline. This shows that a chromosomal memory regulating H3K27me3 in differentiated cells can be established by independent of . We find that a null mutation in leads to a reduction of PRC1 proteins, including Mph1, Bmi1, and Mel18, and a loss of H2AK119ub1 in ES cells. Consequently, the loss of PRC1 causes the derepression of lineage-restricted genes in ES cells and leads to aberrant differentiation. The genes , , , and are derepressed in -deficient ES cells, which is consistent with a previous report of Ring1B binding to their promoters (). Moreover, , , and , which are bound by Ring1B and PRC2, are derepressed in either -deficient or -deficient ES cells. This demonstrates that both and PRC2 are essential for the repression of developmental genes, which is consistent with reports that PRC2 is required for PRC1 recruitment to the locus in cells (; ). Notably, , a target of PRC1 but not PRC2, is derepressed strongly in -deficient but only weakly in -deficient ES cells. This indicates that PRC2-dependent and independent modes of PRC1 recruitment to developmental control genes exist, similar to our previous observation in X inactivation (). Loss of the repression of lineage-specific genes in -deficient ES cells contributes to a marked predisposition to differentiation. Nonetheless, if -deficient ES cells are cultured under optimal conditions, they proliferate normally and express the pluripotency-associated marker comparable with wild-type ES cells. Differentiation of -deficient ES cells leads to abnormal EB formation, which is possibly the result of a failure to generate the normal spectrum of cell types. This results in the inability of the EB to form contractile cardiomyocytes but does not impair the proliferation of differentiating cells. Aberrant differentiation is consistent with the observation that disruption of the gene in mice results in gastrulation arrest (). Notably, we find the expression of , a gene that is specific for terminally differentiated trophoblast cells, upon the differentiation of -deficient ES cells. This could indicate an aberrant differentiation potential toward extraembryonic lineages, which is not observed in normal mouse ES cells. The effect of on lineage specification is dosage sensitive, as we observe a partial phenotype in 36 ES cells, which show reduced levels of Ring1B protein as a result of a hypomorphic allele. These cells can form contractile cell types but attach to culture plates only inefficiently, resulting in the formation of peculiar contracting spherical structures. Several PcG proteins were present in reduced amounts in -deficient cells. By Western and immunofluorescence analyses, we found that Rybp, Mel18, Mpc2, and Mph1 are virtually absent in -deficient ES cells. The finding that these PRC1 transcripts were detected in 36 ES cells suggests regulation at the protein level. The promoter has been reported as a target of both PRC1 and PRC2 (). Consistent with this, we found elevated transcript levels in - and -deficient cells. However, Bmi1 protein accumulates in -deficient but is virtually absent in -deficient ES cells despite elevated mRNA levels. This suggests that Ring1B is needed for Bmi1 protein translation or stabilization, possibly by complex formation. This is in line with a recent report that Ring1B and Bmi1 are required for mutual stabilization (). Notably, Ring1B and PRC2 regulate expression at the transcriptional and protein levels. The requirement of Ring1B for the regulation of protein levels of other PRC1 members is somewhat reminiscent of the situation in PRC2, in which Eed controls the abundance of Ezh2 protein but transcription is unaltered in - deficient cells (). This suggests that PcG proteins in general might be regulated at the protein level to achieve proper complex composition. We conclude that has a dual function in the regulation of PRC1 protein levels and in the maintenance of transcriptional repression of developmental control genes in ES cells. expression cannot establish chromosome-wide H2AK119ub1 in -deficient ES cells. This is in contrast to the situation in mouse embryonic fibroblasts, in which the disruption of has no effect on H2AK119ub1 on the Xi, but only the double deficiency of and leads to a loss of H2AK119ub1 (). Likewise, we find that H2AK119ub1 colocalization with is restored upon the differentiation of -deficient ES cells. This indicates the presence of a redundantly acting E3 ligase activity similar to that of Ring1A in embryonic fibroblasts. Consistent with this, we observe Ring1A colocalization with in differentiated ES cells. We conclude that in ES cells, the establishment of H2AK119ub1 on the -expressing chromosome as well as on developmental control genes requires the specific recruitment of Ring1B. In differentiated 36 ES cells, H2AK119ub1 is observed on the -expressing chromosome despite the absence of Ring1B and several PRC1 proteins. H2A ubiquitination activity is specifically recruited by , but global levels of H2AK119ub1 are not restored upon the differentiation of 36 cells. Similar results were reported in mouse embryonic fibroblasts, in which global H2AK119ub1 was lost, but H2AK119ub1 on the Xi was unaffected upon the deletion of (). Ring1A E3 ligase activity in the absence of Mph2 has been shown in vitro (; ). Additionally, our previous observation that Ring1B can catalyze H2AK119ub1 without Mph1 recruitment in -deficient ES cells () supports the idea of the PRC1-independent recruitment of Ring1A to the -expressing chromosome in differentiating 36 cells. Bmi1 is sufficient for the H2A ubiquitination activity of Ring1A when a Ring1A–Bmi1 complex is reconstituted in vitro (). In contrast, our data suggest that Bmi1 and Mel18 are not essential for the recruitment of E3 ligase activity by . Our findings indicate that H2A ubiquitination in X inactivation depends on a special mode of PcG recruitment by , and Ring1B appears to be critical for global H2AK119ub1 in ES cells and differentiated cells. We have previously shown that H2AK119ub1 can be recruited by a mutant RNA, which lacks the 5′ repeat A and does not initiate gene silencing in ES cells (). Thus, H2AK119ub1 is not sufficient for gene silencing in X inactivation. However, it remained conceivable that H2AK119ub1 could be a prerequisite for silencing. In this study, we find that initiates silencing in the absence of H2AK119ub1 in -deficient ES cells. From this and from our previous data (), we conclude that neither H2AK119ub1 nor H3K27me3 are essential for silencing in X inactivation. This is in contrast to the finding that developmentally regulated genes are derepressed in -deficient ES cells. Thus, we conclude that the requirement for PcG recruitment differs between the silencing of developmental genes and X inactivation. The reason for this discrepancy could be that PRC1 and PRC2 are recruited in parallel by RNA and, thus, could compensate for each other's loss of function. Consistent with this notion, the other initiation marks of X inactivation, namely H3K27me3 and H4K20me1, are efficiently recruited by in -deficient cells. expression in ES cells initiates reversible chromosome-wide gene repression. Therefore, a potential repressive activity of might be masked by active repression by . Upon differentiation, loses its ability to initiate silencing, and repression is maintained independently of . cells, as the abundance of several PRC1 and PRC2 proteins is strongly reduced. However, we observe that chromosome-wide histone modifications characteristic of the Xi are not affected by the absence of in differentiated cells. Moreover, chromosomal silencing is stably maintained independently of in differentiated -deficient cells. This is in stark contrast to the regulation of developmental control genes, which are derepressed in ES cells carrying mutations in either or . We note that the chromosome-wide silencing of X inactivation is more robust in the face of a loss of PcG proteins than the repression of developmental regulators. This might suggest that in X inactivation, several levels of control act synergistically, and the loss of Ring1B causes only a minor destabilization, which we could not detect by our assays. In the future, it will be imperative to study the simultaneous loss of PRC1 and PRC2 function and examine whether such a mutant background is compatible with stem cell maintenance. Thus, X inactivation can provide a sophisticated model system for studying aspects of PcG protein recruitment and to dissect their effect on chromatin and gene expression. ES cell culture was described previously (). expression was induced with 1 μg/ml doxycycline. Differentiation medium contained 100 nM of all-trans–retinoic acid and no Leukemia inhibitory factor (LIF). EBs were generated by the hanging drop method in medium without LIF for 2 d. Then, aggregates were cultured in suspension and subsequently plated on gelatin-coated dishes for up to 3 wk. Cells were counted with a Casy 1 cell counter (Schaerfe System GmbH). For targeting, a 10-kb HindIII–BamHI genomic fragment was isolated from a bacterial artificial chromosome clone (RP22-287N19) from the RPCI22 129 mouse bacterial artificial chromosome library (Children's Hospital Oakland Research Institute). For the minus targeting vector, a 3-kb AvrII–SphI fragment containing three exons, including the start codon and RING domain, was replaced by a stop cassette containing the adenoviral splice acceptor, a -flanked hygromycin-thymidine kinase cassette, and a polyadenylation signal. For counter selection, a diphtheria toxin A chain cassette was added (). Clone 36 ES cells () were electroporated with 50 μg of linearized targeting vector. After selection with 130 μg/ml hygromycin B, targeted clones were identified by Southern analysis of BamHI-digested DNA by a 5-kb band (wild type at 12 kb). The targeting frequency was 15%. The selection cassette was removed by electroporation of 30 μg Cre recombinase expression vector followed by 2 μM gancyclovir selection. For the conditional targeting vector, a -flanked hygromycin-thymidine kinase cassette was integrated into the SphI restriction site in intron 4. A and a BamHI site were inserted into an AvrII site in intron 1 (). 36 ES cells were obtained with a frequency of 5%, and, after Cre-mediated recombination, 36 ES cells were established with a frequency of 43%. ES cells were preplated twice for 30 min to remove feeder cells and were spun onto poly--lysine–coated slides (Sigma-Aldrich) using a centrifuge (Cytospin 3; Thermo Shandon). Differentiated cells were grown on Roboz slides (CellPoint Scientific). Immunostaining was performed as described previously (). In brief, cells were fixed for 10 min in 4% PFA in PBS, permeabilized for 5 min in 0.1% Na citrate/0.5% Triton X-100, and blocked for 30 min in PBS containing 5% BSA and 0.1% Tween 20. For H2AK119ub1 immunostaining, cells were preextracted in 100 mM NaCl, 300 mM sucrose, 3 mM MgCl, 10 mM Pipes, pH 6.8, and 0.5% Triton X-100 for 2 min before fixation, and washes after incubation with primary and secondary antibody were performed in KCM buffer (120 mM KCl, 20 mM NaCl, 10 mM Tris, pH 8.0, and 0.5 mM EDTA)/0.1% Tween 20. RNA FISH probes were generated by random priming (Stratagene) using Cy3-dCTP (GE Healthcare). After immunostaining, cells were fixed in 4% PFA in PBS for 10 min, dehydrated, hybridized, and washed as described previously (). Vectashield (Vector Laboratories) was used as imaging medium. Images were obtained at room temperature with a fluorescence microscope (Axioplan; Carl Zeiss MicroImaging, Inc.) at a magnification of 100× using a plan Neofluar NA 1.3 objective, a CCD camera (CoolSNAP ; Photometrics), and MetaMorph image analysis software (Universal Imaging Corp.). Color levels were adjusted in Photoshop 7.0 (Adobe). ES cell lines were analyzed, and the means and SDs of at least two experiments were calculated and normalized to the number of -expressing cells unless stated differently. Northern analysis was performed using 15 μg RNA (TRIzol; Invitrogen) as described previously (). Quantification was performed using a scanner (STORM 860; Molecular Dynamics) and ImageQuant TL software v2003.03 (GE Healthcare). Mean and SD was calculated from at least two 36 cell lines and from at least two independent experiments. Histones were acid extracted in 0.2 N HCl. Nuclear proteins were extracted in 10 mM Hepes, pH 7.9, 1.5 mM MgCl, 0.1 mM EDTA, 25% glycerol, and 0.4 M NaCl after the cytoplasm had been separated. Protein concentration was measured by the Bradford assay. Loading was controlled by Ponceau S staining and lamin B1. The following antibodies were used for immunofluorescence/Western analysis (Antisera dilutions are given in immunofluorescence/Western blot pairs. “−/…” identifies that the antisera was not used for immunofluorescence; “…/−” was not used in Western blot): α-Ring1B (1:100/1:100; ), α-Ring1A (1:100/1:100; ), α-MPc2 (−/1:300; Santa Cruz Biotechnology, Inc.), α-Bmi1 (−/1:500; Abcam), α-Mph1 (1:5/1:2; ), α-Mph2 (1:100/1:50; ), α-Mel18 (1:300/1:500; Santa Cruz Biotechnology, Inc.), α-Suz12 (1:1,000/1:1,000; Upstate Biotechnology), α-Ezh2 (1:500/1:500; ), α-H3K27me3 (1:1,000/1:1,000; ), α-H4K20me1 (1:1,000/1:1,000; ), α-H2AK119ub1 (1:50/1:500; Upstate Biotechnology), α-RYBP (−/1:1,000; Chemicon), α-histone macroH2A–containing antiserum (1:500/−), α-histone macroH2A (−/1:500; Upstate Biotechnology), and α-lamin B1 (−/1:5,000; Abcam). Secondary antibodies used are as follows: AlexaFluor488 goat anti–rabbit IgG (1:500/−), AlexaFluor488 goat anti–mouse IgG (1:500/−), and AlexaFluor568 rabbit anti–goat IgG (1:500/−); and HRP-conjugated Affinipure goat α-rabbit IgG (−/1:10,000), HRP-conjugated Affinipure goat α-mouse IgG (−/1:5,000), HRP-conjugated donkey α-goat IgG (−/1:2,000), and HRP-conjugated donkey α-human IgG (−/1:2,000) from Jackson ImmunoResearch Laboratories. cDNA was generated from 400 ng of total RNA from clone 36, 36, 36, 36 ES cells, and female trophoblast stem cells using the Superscript II Reverse transcription kit (Invitrogen) and dT primers. Expression of the genes , , , , , , , , , , , , , , , , , and β was analyzed by PCR (for primer sequences and conditions, see Table S1, available at ). Real-time PCR analysis was performed as described previously (). Fig. S1 describes the expression analysis of differentiated -deficient ES cells. Fig. S2 presents immunofluorescence analysis of H3K27me3 and Ring1A recruitment in clone 36 and 36 cells. In Fig. S3, we present the analysis of chromosome-wide silencing in clone 36 and 36 cells. Video 1 shows contractile spheres formed by differentiating Ring1B ES cells. Table S1 provides PCR primer sequences for semiquantitative expression analysis. Online supplemental material is available at .

Nuclear factor κB (NF-κB) is a widely expressed primary transcription factor composed of a heterodimeric complex (p65–p50). A myriad of unrelated exogenous and endogenous stimuli are capable of inducing NF-κB activity. In turn, NF-κB regulates the expression of an equally diverse array of cellular genes that are important in immunity, inflammation, and development (; ). Aberrance of its function has been linked to such pathological processes as cancer and abnormal development (for review see ). Determining molecular mechanisms that regulate the activation of NF-κB is crucial to understand how multiple intracellular signaling pathways converge to activate a single transcription factor. NF-κB is normally localized in the cytoplasm as an inactive complex through physically associating with its inhibitory molecule IκBα. Extensive studies have been performed to address how various stimuli trigger its translocation from the cytoplasm into the nucleus (). Seminal works from several laboratories have determined a sequence of biochemical events that result in the ubiquitin-dependent degradation of IκB proteins (; ; ). Consequently, this releases NF-κB to move into the nucleus and switch on the expression of target genes (; ). NF-κB belongs to the Rel homology domain (RHD) family of transcription factors that exploit similar strategies to achieve initial activation. Recently, an alternative pathway was identified to regulate another member (p100) of this family (). Compared with the fruitful know-how about the molecular events of NF-κB activation in cytoplasm, much less is understood concerning its active regulation and functional interaction with other proteins inside the nucleus. Recent progress has shed light on the importance of nuclear events in shaping the strength and duration of the NF-κB transcriptional response, which is achieved partly by posttranslational modification of the NF-κB transcription factor complex or the histones that surround various NF-κB target genes (). For example, I κ-B kinase α (IKKα) was demonstrated to accelerate both the turnover of NF-κB and its removal from proinflammatory gene promoters (). This kinase could also phosphorylate histone H3 and was critical for NF-κB–responsive gene expression (). Additionally, the acetyltransferase activity of p300/CREB-binding protein (CBP) was required for the activation of NF-κB–dependent transcriptions. p300/CBP proteins were also found to directly associate with NF-κB, forming a bridge to the basal transcriptional machinery (; ). Although a couple of cofactors were recently shown to reside in the NF-κB enhanceosome, much remains to be done to understand their specific functions and regulatory mechanisms. This indicates that NF-κB assembles a much higher order transcription complex than once expected and that there are additional important layers of regulation for the NF-κB transactivation process. Ubiquitously expressed transcript (UXT) is ∼18 kD and predominantly localizes in the nucleus (). It was demonstrated to be widely expressed in human and mouse tissues, and its expression was markedly elevated in some human tumors (; ). Computer modeling predicted that UXT was an α-class prefoldin (PFD) family protein (). Most members of this family are small molecular mass proteins (14–23 kD) and are composed of coiled-coil structures. Yeast and human PFDs 1–6 were previously found to assemble into a hexameric complex that functioned as a new type of molecular chaperone (; ). Until now, the functional characterization of UXT was scarce. One recent study indicated that UXT bound to the N terminus of the androgen receptor and regulated androgen receptor–responsive genes that are important in prostate growth suppression and differentiation (; ). Another investigation suggested UXT to be a component of the centrosome (). However, much remains to be done as to the in vivo function of UXT and its regulatory roles in cellular processes. During a systematic screening for proteins that interacted with components of the NF-κB enhanceosome, we identified UXT as a novel p65-interacting protein. This interaction is confirmed both in vitro and in vivo. In this study, we show that RNAi knockdown of UXT leads to impaired NF-κB activity and dramatically attenuates the expression of NF-κB– dependent genes. This interference also sensitizes cells to apoptosis by TNF-α. Furthermore, UXT forms a signal- dependent complex with NF-κB and is recruited to the NF-κB enhanceosome upon stimulation. Interestingly, the UXT protein level correlates with constitutive NF-κB activity in human prostate cancer cell lines. The presence of NF-κB within the nucleus of stimulated or constitutively active cells is considerably diminished with decreased endogenous UXT protein levels. Collectively, our investigation reveals that UXT is an integral component of the NF-κB enhanceosome and is essential for its nuclear function, which uncovers a new mechanism of NF-κB regulation. To identify new components of the NF-κB enhanceosome, we performed a systematic yeast two-hybrid screening in which the cDNA fragment harboring the RHD of p65 (amino acids 1–312) was used as bait. Several positive clones were identified to encode full-length UXT (). In addition, previously confirmed p65-interacting proteins (e.g., IκBα and PIAS3) were screened out. UXT was previously reported to be expressed almost exclusively inside the nucleus of most cells (). This was confirmed in our investigation for either endogenous or overexpressed UXT (). To further substantiate its interaction with p65, an in vitro coimmunoprecipitation assay was applied in which full-length HA-p65 and FLAG-UXT proteins were generated and labeled, respectively, with [S]methionine by in vitro translation. The products were mixed and immunoprecipitated with either control IgG or anti-HA antibody. As shown in , UXT could be coprecipitated by antibody against the HA epitope but not by control IgG, which suggests that UXT indeed interacts directly with full-length p65. To address the physiological relevance of this interaction in mammalian cells, we expressed HA-UXT in 293T cells and then stimulated cells with or without TNF-α for the indicated times. The fractionated cytoplasmic or nuclear extracts were immunoprecipitated with either anti-p65 antibody or IgG as a control, respectively. There was no detectable UXT that interacted with cytoplasmic p65 in the presence or absence of TNF-α (), which was consistent with the unique subcellular location of UXT. In addition, there was only a marginal amount of endogenous p65 in the nucleus devoid of TNF-α treatment. Consequently, no UXT was coimmunoprecipitated from this nuclear extract even though there existed a large amount of UXT. In contrast, there exhibited a strong interaction between nuclear p65 and UXT upon TNF-α stimulation. Furthermore, we tested whether endogenous UXT and p65 could interact in response to TNF-α. As shown in , endogenous UXT was coimmunoprecipitated by p65 antibody from cells treated with TNF-α. In contrast, UXT was barely detected in the immunoprecipitates without TNF-α treatment. One possible explanation for this phenomenon is that only after p65 translocation into the nucleus could UXT have access to p65. However, we could not formally rule out the possibility that posttranslational modifications of either protein were prerequisites for this interaction in vivo. Collectively, these results indicate that UXT interacts in vivo with p65 upon TNF-α stimulation. The p65 subunit of NF-κB harbors an N-terminal conserved region (∼300 amino acid residues) known as RHD and a C-terminal transactivation domain. To explore the UXT-binding region within p65, we constructed a series of p65 deletion mutants (). It was found that the loss of amino acids 1–285 at the N terminus of p65 resulted in its complete inability to interact with UXT (, top). In contrast, p65 fragments spanning amino acids 1–286, 1–312, or 1–372 fully retained their binding capability and interacted with UXT as well as the wild type (, middle). Because the RHD of p65 consisted of two Ig-like domains (), we made two additional deletion mutants of p65 (amino acids 1–190 and 191–551), each of which contained only one Ig-like domain. Immunoprecipitation assays revealed that neither of them was able to interact with UXT (, bottom). In addition, we generated several point mutations of UXT and did not observe a considerable change on UXT and p65 interaction. We also had attempted in vain to express truncation mutants of UXT in mammalian cells, which prevented us from further dissecting UXT. Collectively, these data suggest that the intact RHD of p65 is both essential and sufficient to mediate interaction with UXT. The RHD structure defined a highly conserved family of transcription factors (Rel family) that are important in immunity and inflammation (). This led us to wonder whether this interaction was also applicable to other proteins of this family. To explore this possibility, we transfected UXT into 293T cells along with p50 or c-Rel. Interestingly, UXT was also capable of binding to p50 or c-Rel specifically and strongly (). In contrast, lymphoid enhancer binding factor 1, a transcription factor unrelated to the Rel family, did not have any affinity to UXT. This phenomenon suggested that UXT recognized a consensus structure fold and that there was probably a unified theme for its interaction with p65 and other members of the Rel family. Because UXT interacted specifically inside the nucleus with p65 in a signal-dependent manner, we went on to address whether this interaction was important for regulating NF-κB activity and genes responsive to it. We transfected 293T cells with UXT or other control plasmids, prepared nuclear extracts, and performed electrophoretic mobility shift assay (EMSA) as indicated. Consistently, there was no NF-κB binding to its cognate probe without TNF-α treatment. Notably, the overexpression of UXT alone did not induce any detectable basal NF-κB binding activity. In contrast, robust NF-κB binding activity was induced upon TNF-α stimulation. Expectedly, this activity was severely impaired by A20, a potent inhibitor of NF-κB signaling (). However, in response to TNF-α, we observed neither inhibitory nor synergically stimulatory effects on NF-κB binding affinity (). Alternatively, we explored whether overexpressing UXT had any measurable effects on inducible genes such as A20 or interleukin-8 (IL-8) that were regulated by NF-κB. On transfection of UXT alone, we did not observe any change on the basal expression of these two genes (unpublished data). Thus, we transfected 293T cells with UXT or IκBα super repressor (SR) and analyzed the amount of A20 or IL-8 mRNA induced by TNF-α, respectively, via quantitative real-time RT-PCR. IκBαSR was a well-established potent inhibitor of NF-κB activation (). Consistently, A20 or IL-8 inductions by TNF-α were severely attenuated in the presence of IκBαSR. Interestingly, we observed marginal synergic inductions of both A20 and IL-8 by TNF-α in the presence of UXT (), which suggested that UXT might modulate NF-κB function. Because we failed via overexpressing UXT to convincingly demonstrate the involvement of UXT in NF-κB regulation, we turned to investigate the effect if endogenous UXT expression was reduced via RNAi. Two UXT siRNAs were fished out, which are designated as 242 and 428 hereafter. We confirmed the effectiveness of them against UXT by monitoring the mRNA level (), protein level (), and cellular immunofluorescence of UXT (). 428 was reproducibly better than 242 and was used more frequently in later experiments. In addition, two negative control siRNAs were used. One was a nonspecific siRNA and was named as control; the other is a mutant form of siRNA 428 (428m) that could partially bind to UXT mRNA but lost the interfering ability. Initially, we used a κB-Luc reporter gene to evaluate the effect of UXT knockdown on NF-κB activation status. Excitedly, luciferase assay revealed that the decrease of endogenous UXT considerably inhibited NF-κB transcriptional activity induced by TNF-α, IL-1β (), or lipopolysaccharide (). Notably, this phenomenon also held true for basal luciferase expression. Likewise, similar effects were observed in cells with reduced UXT expression when the cells were stimulated by overexpressing MyD88 or TRAF6, which are well known to induce NF-κB activity (; ; ). In addition, gene activation induced by p65 alone was also markedly attenuated in cells with a decreased expression of endogenous UXT (). Our data have shown that UXT could interact with other Rel family proteins like c-Rel (). c-Rel was previously found to regulate the expression of the IL-12 p40 subunit by specifically interacting with its promoter (). Thus, a reporter gene was used that harbored firefly luciferase under the control of the mouse p40 promoter. Consistently, c-Rel activation was severely impaired in cells expressing reduced amounts of UXT. This attenuation was in direct proportion to the intensity of RNAi, as evidenced in oligonucleotides 428 versus 242 (). To make it more physiologically relevant, we also investigated how the induction of NF-κB–dependent genes (IL-8, A20, and IκBα) was influenced by knocking down endogenous UXT expression. For cells transfected with control siRNA or UXT siRNA, quantitative real-time RT-PCR was used to measure endogenous mRNA levels of IL-8, A20, and IκBα induced by TNF-α during a time course. Consistently, these inductions were sharply attenuated when endogenous UXT expression was suppressed (). Collectively, our data strongly indicate that UXT is required for inducing genes tightly regulated by NF-κB and that it plays an essential role in NF-κB function. As was stated in the Introduction, NF-κB activation involved a series of molecular events both in the cytoplasm and in the nucleus. To rule out the possibility that UXT might act on processes other than directly on NF-κB itself, we examined the effects of UXT knockdown on the phosphorylation and degradation of IκBα. We did not find any differences in this regard between UXT-deficient and normal cells during TNF-α stimulation (). This was further supported by the observation that IKK kinase activity was not affected at all in terms of UXT knockdown (unpublished data). Given that UXT was expressed almost exclusively inside the nucleus and interacted directly with NF-κB, these data strongly indicate that UXT performed its function via targeting NF-κB itself. Via EMSA, we then analyzed endogenous NF-κB DNA binding activity in cells with reduced UXT expression. Expectedly, TNF-α alone induced endogenous NF-κB to bind to its cognate probe strongly and specifically. Considerably, this interaction was markedly diminished in nuclear extracts from UXT-specific knockdown cells. In addition, this reduction was correlated with the effectiveness of the siRNA administered (). To substantiate this finding, we took advantage of the chromatin immunoprecipitation (ChIP) assay to examine whether binding of NF-κB to its endogenous promoter was influenced in vivo when the expression of UXT was diminished. Consistently, deficiency of the endogenous UXT level resulted in considerable decreases in the amount of p65 associated with its endogenous cognate promoters upon TNF-α stimulation. In addition, this was also true for other components of the NF-κB transcriptional enhanceosome such as p50 and RNA polymerase II. Understandably, deficiency of the endogenous UXT level had no effect on the glyceraldehyde-3-phosphate dehydrogenase (GAPDH) transcriptional complex (). Collectively, these finding indicate that UXT plays an important role in the NF-κB enhanceosome. A previous study has predicted UXT as an α-class PFD family protein via computer modeling (). Some members of this family were previously found to assemble into a hexameric complex that functioned as a molecular chaperone in protein folding and stability (). Thus, we hypothesized that UXT performed its function by serving as a specific molecular chaperone for NF-κB inside the nucleus. To explore this hypothesis, we investigated whether the nuclear presence of p65 was influenced by UXT. Therefore, 293T cells were transfected with or without the indicated siRNAs and stimulated with or without TNF-α as shown. Interestingly, the amount of nuclear p65 was markedly diminished in cells that were treated with siRNA against UXT even though these cells were stimulated with TNF-α. Notably, siRNA 428 was better than siRNA 242 in causing the loss of nuclear p65. In contrast, control siRNA displayed no such effect on the nuclear presence of p65. The transcription factor sp1 as a negative control was expressed constitutively inside the nucleus and remained intact in response to UXT knockdown. The cytoplasmic reservoir of p65 did not seem to be affected by siRNA against UXT (). This suggested that UXT positively regulated p65 inside the nucleus, and the loss of it affected the sustained action of p65 in the nucleus, which led to the attenuated EMSA results as observed in . Alternatively, we performed immunofluorescence analysis to support this speculation. Interestingly, when transfecting cells with siRNA against UXT and stimulating them with TNF-α, there were an apparently decreased percentage of cells that displayed focused nuclear p65. Approximately 90% of control cells displayed p65 inside the nucleus upon stimulation, whereas only 35–50% displayed p65 in the case of the UXT knockdown specimen (). Collectively, these data indicate that UXT is a stabilizing factor for the NF-κB transcriptional enhanceosome. One possible mechanism for the aforementioned phenomenon is that the loss of UXT releases NF-κB from the enhanceosome, which consequently causes NF-κB to be exported out into the cytoplasm. To address this hypothesis, leptomycin B (LMB), a specific inhibitor of nuclear export (; ), was used. Cells were transfected with siRNA against UXT and stimulated with TNF-α in the presence or absence of LMB. Immunofluorescence analysis indicated that LMB considerably increased the nuclear accumulation of p65 even though the endogenous UXT was knocked down, which suggests that UXT contributed to the sustained presence of p65 inside the nucleus. We have also tested other possibilities, but no obvious correlation could be drawn at present (see Discussion). Our data formerly indicated that RHD of p65 was essential and sufficient to mediate the interaction between p65 and UXT (). We wondered whether this stabilizing effect of UXT was applicable to any proteins containing this domain. To address this speculation, we constructed two fusing proteins, p65 (1–312)-VP16 and Gal4 BD-p65 (285–551): the former one harbors the RHD, whereas the latter one spans all of p65 except RHD. Luciferase assays were performed to monitor the effect of UXT knockdown on activations of indicated reporters induced by these chimeric proteins. Interestingly, the knockdown of UXT inhibited the activity induced by p65 (1–312)-VP16 () but not that induced by Gal4 BD-p65 (285–551) (). Collectively, these data indicate that UXT directly modulates the nuclear function of NF-κB. To rule out possible off-target effects of the siRNA against UXT, a siRNA-resistant UXT (UXTr) was generated in which silent mutations were introduced into the sequence targeted by siRNA 428 and without changing the amino acid sequence of the proteins expressed. The usefulness of UXTr was confirmed so that siRNA 428 no longer had any effects on UXTr but siRNA 242 could still interfere with UXTr expression (). The exogenous wild-type UXT or UXTr was transfected together with siRNA 428 and stimulated with TNF-α as indicated, respectively. Nuclear and cytoplasmic extracts were probed with indicated antibodies. As shown in , the expression of UXTr rescued the loss of nuclear p65 caused by endogenous UXT impairment, whereas the expression of wild-type UXT failed to do so. This rescuing effect was also confirmed with EMSA assays (). Collectively, these data indicate that the loss of nuclear p65 was directly caused by impairment of UXT function and that there was a direct functional connection between p65 and UXT. Given that UXT interacted with p65 and was essential to maintain the presence of NF-κB inside the nucleus, we wondered whether UXT was an integral component of the NF-κB transcriptional enhanceosome in vivo. To address this possibility, we transfected HA-UXT into 293T cells and performed systematic ChIP assays on the promoters of A20 and IκBα as described in Materials and methods. It turned out that UXT was indeed present within the NF-κB transcriptional enhanceosome. Notably, its presence became much more prominent upon stimulation, which suggested that UXT was dynamically recruited onto the enhanceosome. In addition, UXT had nothing to do with the transcription complex on the GAPDH promoter, indicating the selectivity of UXT action (). We also stimulated 293T cells and performed similar ChIP assays to confirm again that endogenous UXT was recruited onto the NF-κB enhanceosome in response to stimulation (). Alternatively, we performed supershift assay to further substantiate this observation. Stimulation of 293T cells with TNF-α led to a strong NF-κB DNA-binding band composed of κB probe, NF-κB, and its cofactors in EMSA. Interestingly, this EMSA band was markedly diminished when antibody against UXT was introduced into the reaction mixture (), which strongly suggests that UXT is an integral component in this band and that it is of importance to foster an NF-κB conformation amenable to its binding. Collectively, these data indicate that UXT forms a dynamic complex with p65 in vivo and is recruited to the NF-κB enhanceosome after stimulation. The NF-κB enhanceosome consisted of a growing list of transcriptional cofactors. Our current finding that UXT was one of them and that it was able to maintain the stability of the NF-κB enhanceosome led us to ask whether other factors could also accomplish this function. Coactivator-associated arginine methyltransferase 1 (CARM1) is known to be a transcriptional cofactor in the NF-κB enhanceosome. Cells with a reduced expression of CARM1 showed an impaired expression of NF-κB–dependent genes upon stimulation (; ). However, immunofluorescence and EMSA experiments indicated that the knockdown of CARM1 did not affect the nuclear presence of p65 (), which suggested that UXT played a unique role toward NF-κB function. It was well established that NF-κB promotes the survival of most cells through the transcriptional induction of antiapoptotic genes (; ). Normally, 293T cells would not display apoptotic phenomenon in the presence of TNF-α (<3%). However, loss of NF-κB function would make cells prone to apoptosis. This was confirmed by knocking down UXT upon TNF-α stimulation (∼15%; unpublished data). Cycloheximide (CHX) is a potent protein synthesis inhibitor. However, at the concentration of ≤10 μg/ml, CHX only reduces but does not completely block de novo protein synthesis (). Interestingly, a low concentration of CHX could dramatically augment the apoptotic effects (; ). Taking advantage of this cell model, we explored whether UXT was important for cell survival in response to TNF-α treatment. As was shown in , TNF plus 5 μg/ml CHX resulted in ∼10% cell death after 18 h of treatment. More importantly, when siRNA of UXT, which blocked NF-κB activity, was used in combination with TNF and CHX, we detected drastically morphological changes under light microscopy and ∼50% of cells undergoing apoptosis. Alternatively, we used annexin V and TUNEL methods to quantitatively measure the percentages of cells undergoing apoptosis. Consistently, a considerable increase of apoptotic cells was observed in samples with reduced UXT (). These indicated that the knockdown of UXT sensitizes 293T cells to apoptosis induced by TNF-α. Prostate cancer began as an androgen-dependent tumor and progressed into an androgen-independent tumor. In this progress, NF-κB activity started to behave out of control (; ). In androgen-independent PC-3 cells, more NF-κB localized inside the nucleus and constitutively displayed DNA binding activity, whereas it exhibited barely detectable activity in androgen-sensitive LNCaP cells (). We confirmed these observations as shown in , which indicate that these two cells are ideal models for studying NF-κB regulation within the nucleus. In addition, a former study indicated that the UXT mRNA level was considerably elevated in PC-3 as compared with LNCaP cells (). We confirmed this observation in terms of UXT mRNA (not depicted) and also found that this was true for the protein level of endogenous UXT (). Given that UXT was recruited onto the NF-κB enhanceosome upon stimulation and that it is essential to maintain the presence of NF-κB inside the nucleus, these results suggest a possible correlation between UXT levels and NF-κB activation. To further explore this possibility, we determined whether the inhibition of endogenous UXT expression would affect the constitutive amount of nuclear NF-κB in PC-3 cells. As shown in , this interference indeed reduced the amount of nuclear NF-κB in this cell, which is consistent with that observed in 293T cells stimulated by TNF-α. In addition, luciferase assay revealed that the decrease of UXT expression suppressed the constitutive transcriptional activity of NF-κB in PC-3 cells (). Furthermore, EMSA confirmed that the loss of UXT also suppressed the constitutive DNA binding activity of NF-κB but not the control sp1 (). These results indicate that the elevated expression of UXT strongly correlates with constitutive NF-κB activity in prostate cancer cell lines and again substantiates the notion that UXT is essential for NF-κB function in the nucleus. The NF-κB family of transcription factors is crucial for many key cellular processes. Recently, the regulation of NF-κB activity has been cast in the limelight (; ). A handful of transcriptional cofactors were implicated in this process. For example, p300/CBP played a major role in the acetylation of p65 in vivo (). Conversely, acetylated p65 was subjected to deacetylation by HDAC3 (). Recent work also revealed that SIRT1 physically interacted with p65 and promoted p65 deacetylation (). In addition, NF-κB was reported to be phosphorylated at multiple sites during activation. For example, IKKα was demonstrated to accelerate both the turnover of NF-κB and its removal from proinflammatory gene promoters (). An emerging theme is that there are additional important layers of regulation for nuclear NF-κB. Regulating the duration of the nuclear presence of p65 is another potential mechanism to modulate the NF-κB transcriptional response. In this study, we characterized UXT as a novel and essential cofactor for NF-κB function in its enhanceosome. Several lines of findings support this argument. UXT was shown to interact directly with p65 both in vitro and in vivo. Importantly, this interaction was dependent on external stimuli. Because UXT was almost exclusively present inside the nucleus, we reasoned that only after NF-κB translocation into the nucleus could this interaction take place, which was also substantiated by the observation that NF-κB was constitutively inside the nucleus in PC-3 cells and regulated by UXT as a result. ChIP and EMSA assays demonstrated that UXT localized inside the NF-κB enhanceosome in vivo and was recruited to it upon stimulation. The knockdown of UXT did not influence the molecular events of NF-κB activation outside the nucleus. Instead, it decreased the amount of nuclear p65 and severely impaired NF-κB activation. This decrease could be rescued by a siRNA-resistant exogenous UXT. Overexpression of UXT caused marginal synergic inductions of both A20 and IL-8 in response to TNF-α. We reasoned that there was a sufficient amount of endogenous UXT within the nucleus in 293T cells so that it would be difficult to demonstrate the synergic effect more dramatically. Convincingly, RNAi of UXT resulted in apparent attenuations of NF-κB–responsive reporter expression by various stimuli and, ultimately, the inducible expression of genes tightly regulated by NF-κB. The reduction of endogenous UXT tamed cells prone to apoptosis that was induced by TNF-α, which is a known index for the impairment of NF-κB function. PC-3 cells displayed constitutive NF-κB activity inside the nucleus. This was nicely correlated with the elevated presence of both NF-κB and UXT within the nucleus, which was also consistently substantiated by the loss of NF-κB and its activity when endogenous UXT was diminished in PC-3 cells. Thus, UXT might function to extend the duration of NF-κB or its enhanceosome inside the nucleus. This function was probably achieved by fostering a favorable conformation for NF-κB in its enhanceosome. UXT was recently predicted as a new member of the α-class PFD family protein. Yeast and human PFDs 1–6 assembled into a hexameric complex, which functioned as a molecular chaperone in protein folding. Our preliminary data also suggested that UXT forms oligomers in vivo (unpublished data). Some members of the PFD family protein were implicated to participate in transcriptional regulation, such as PFDN5 (MM1) in c-myc transcription () and URI in the rapamycin-sensitive transcription response (). However, the specific mechanisms of their action remain unknown. We have generated several point mutations of UXT (C75A, L32P, L50P, L59P, and L32P/L50P/L59P) and did not reveal any substantial correlation between UXT and p65 interaction. Our speculation is that a three-dimensional juxtaposed motif may be involved in this interaction. Structural analysis of UXT and NF-κB interaction is under way in our laboratory, and hopefully this will shed light on how UXT performs this regulatory role. Recently, studies suggested that the ubiquitin–proteasomal degradation pathway was also involved in stringent control of the promoter-bound p65 (; ). An NF-κB coactivator, Pin1, was found to bind p65 and prevent SOCS-1–mediated ubiquitination (). We have tried to address whether the ubiquitin–proteasome pathway is critical for p65 stability in the context of UXT, but no conclusion can be made as of now. The problem lies in the fact that most proteins are instantly degraded after ubiquitination, which is also the case for p65. To prevent this obstacle, proteasome inhibitors were used to stabilize ubiquitinated proteins. Unfortunately, a critical step during NF-κB activation involved proteasome function (i.e., IκBα degradation), which makes it unpractical to probe endogenous p65 degradation inside the nucleus. The few papers that reported nuclear p65 ubiquitination often used the overexpression of p65 to answer this question, which was a controversial approach. After this method, we did observe traces of p65 ubiquitination when UXT was knocked down (unpublished data), which was comparable with published reports (). However, we believe it is too early to correlate the loss of UXT with p65 ubiquitination. Substantially, we found that LMB, a specific inhibitor of nuclear export, increased the nuclear accumulation of p65 even though the endogenous UXT was knocked down. More likely, UXT influenced p65 nucleocytoplasmic shuttling. Given the findings from this study, we favor the notion that UXT is a nuclear chaperone that promotes formation of the NF-κB enhanceosome. Nuclear chaperones usually mediate nucleosome assembly and remodeling (; ). Furthermore, they were implicated to directly regulate transcription factors. For example, a nuclear chaperone termed FACT (facilitates chromatin transcription) could facilitate transcript elongation through nucleosomes (). Another human nuclear chaperone (bZIP-enhancing factor) promoted the transcriptional activity of bZIP (basic region–leucine zipper DNA-binding domain) proteins (). JDP2 was recently demonstrated to be a chaperone for AP1 (). In addition, nuclear chaperones can also negatively regulate transcription. For example, nuclear chaperones p23 and Hsp90 bound to glucocorticoid-induced regulatory complexes and consequently disassembled these complexes (). In conclusion, nuclear chaperones have regulatory functions targeted to their specific client proteins other than the classic functions involved in protein synthesis and maturation. More work needs to be done before the function and mechanism of UXT action are fully understood. Numerous reports have documented considerable correlations between NF-κB activation and specific types of cancer (for review see ). By promoting proliferation and inhibiting apoptosis, NF-κB could tip the balance between proliferation and apoptosis toward malignant behavior in tumor cells (for review see ). Prostate cancer begins as an androgen-dependent tumor and progresses into an androgen-independent tumor. In this progress, NF-κB activity is up-regulated (; ). In this study, we provide a new thread of explanation for this correlation and demonstrate that UXT is essential for the constitutive activity of NF-κB in prostate cancer cells. Collectively, our study reveals that UXT is an integral component of the NF-κB enhanceosome and is essential for its function in the nucleus. UXT may function as a specific molecular chaperone for NF-κB in this process. Future investigations will focus on how UXT interacts with NF-κB and regulates the dynamic processes of NF-κB within the nucleus. Monoclonal UXT antibodies 6D3 and 105.128 were provided by M.I. Greene (University of Pennsylvania, Philadelphia, PA) and W. Krek (Eidgenössische Technische Hochschule Honggerberg, Zurich, Switzerland), respectively. UXT siRNA duplexes were chemically synthesized by GenePharma. The UXT siRNA sequences were as follows: 242, AGCACUCGGAGUUAUAUAUdTdT; 428, CCAAGGACUCCAUGAAUAUdTdT; and 428M1, CCAACCCCUCCAUGAAUAUdTdT. The control siRNA sequence was UUCUCCGAACGUGUCACGUdTdT, and the CARM1 siRNA sequence was GCAGUCCUUCAUCAUCACCdTdT. Other commercially available reagents and antibodies used were as follows: HA, p65, p65, p50, IκBα, and RNA polymerase II antibodies were purchased from Santa Cruz Biotechnology, Inc. Flag, sp1, and β-actin antibodies were obtained from Sigma-Aldrich. Anti-myc was purchased from Wolwobiotech. rhTNFα and RhIL-1β was purchased from R&D Systems, and lipopolysaccharide was obtained from Sigma-Aldrich. UXT and its deletion mutants were constructed by PCR from the human thymus library and subsequently cloned into mammalian expression vectors as indicated. The UXT siRNA-resistant form was generated by introducing four silent mutations (373 ACAAAAGATAGC 384) in the siRNA 428 target sequence. Flag-IκBαSR, Flag-TRAF6, and the reporter genes (3×κB-luc and pRL-SV40) have been described previously (). pcDNA3-HA-p65 was a gift from G. Pei (Shanghai Institute of Biochemistry and Cell Biology [SIBCB], Shanghai, China), and its deletion mutants were constructed by PCR. MyD88 was amplified from the human thymus library and subcloned into pcDNA3.1-Flag vector. HA–lymphoid enhancer binding factor 1 was provided by L. Li (SIBCB, Shanghai, China), HA-p50 and mp40-luc were provided by B. Sun (SIBCB, Shanghai, China), myc-cRel was provided by B. Huang (Northeast Normal University, Changchun, China), and A20 was provided by L. Guo (SIBCB, Shanghai, China). A cDNA fragment encoding residues 1–312 of human p65 was inserted in frame into the Gal4 DNA-binding domain vector pGBKT7. A human thymus cDNA library (CLONTECH Laboratories, Inc.) was screened according to protocols recommended by the manufacturer. 293T and RAW264.7 cells were cultured in DME (Invitrogen) supplemented with 10% FBS (Hyclone). PC-3 cells were cultured in RPMI 1640 medium (Invitrogen) supplemented with 10% FBS, and LNCaP cells were cultured in Ham's F12 medium (Invitrogen) supplemented with 10% FBS. All transfections were performed using LipofectAMINE 2000 (Invitrogen) according to the manufacturer's instructions. Cells were seeded in 24-well plates and transfected with 40 pmol siRNA combined with reporters and other constructs as indicated. The total amount of DNA was kept constant by supplementing with pcDNA3. pRL-SV40 (Promega) was cotransfected to normalize transfection efficiency. 48 h after transfection, cells were treated with the indicated reagents or left untreated. Luciferase activity was analyzed with the Dual Luciferase Reporter Assay System (Promega). UXT and p65 were in vitro translated and labeled with [S]methionine using the TNT Quick Coupled Transcription/Translation System (Promega) according to the manufacturer's instructions. Total RNA was isolated with TRIzol (Invitrogen) according to the manufacturer's instructions. Reverse transcription of purified RNA was performed using oligonucleotide dT primer. The quantification of gene transcripts was performed by real-time PCR using SYBR green I dye (Invitrogen). Expression values were normalized with control β-actin. The primers used are listed as follows: IL-8, sense (AGGTGCAGTTTTGCCAAGGA) and antisense (TTTCTGTGTTGGCGCAGTGT); IκBα, sense (CTGAGCTCCGAGACTTTCGAGG) and antisense (CACGTGTGGCCATTGTAGTTGG); A20, sense (GCGTTCAGGACACAGACTTG) and antisense (GCAAAGCCCCGTTTCAACAA); UXT, sense (TTTGGGCTGTAACTTCTTCGT) and antisense (ATATTCATGGAGTCCTTGGTG); CARM1, sense (TGCCGACCGCCTATGACT) and antisense (CCCGTGTTGGCTAAAGGAA); β-actin, sense (AAAGACCTGTACGCCAACAC) and antisense (GTCATACTCCTGCTTGCTGAT). EMSAs were performed as described previously (). The probes used are as follows: wild-type κB probe (AGTTGAGGGGACTTTCCCAGGC), mutant κB probe (AGTTGAGGCGACTTTCCCAGGC), and sp1 probe (ATTCGATCGGGGCGGGGCGAGC). The ChIP assay kit (Upstate Biotechnology) was used according to the manufacturer's instructions with some variations. Formaldehyde cross-linking was performed at room temperature for 10 min before glycine was added to a final concentration of 125 mM for 5 min. The cells were rapidly collected and lysed in SDS lysis buffer. Suspended chromatin was sheared by sonication to a mean size of 200–1,000 bp, centrifuged to pellet debris, and diluted 10 times with dilution buffer. Extracts were precleared for 2 h with salmon sperm DNA and BSA-saturated protein A/G beads. Immunoprecipitations were performed at 4°C overnight using antibodies as indicated with IgG as a negative control. Immune complexes were collected and washed sequentially with Tris-SDS-EDTA buffer I, II, and III followed by two washes with Tris-EDTA buffer. Immune complexes were then extracted with elution buffer and DNA: protein complexes were disrupted by heating at 65°C overnight. After proteinase K digestion for 1 h, DNA was extracted with phenolchloroform and precipitated in ethanol. About one twentieth of precipitated DNA was used as template in each PCR reaction. The following promoter-specific primers were used: human A20, sense (CAGCCCGACCCAGAGAGTCAC) and antisense (CGGGCTCCAAGCTCGCTT); human GAPDH, sense (AGCTCAGGCCTCAAGACCTT) and antisense (AAGAAGATGCGGCTGACTGT); and human IκBα, sense (TAGTGGCTCATCGCAGGGAG) and antisense (TCAGGCTCGGGGAATTTCC). Cells grown on coverslips were fixed with 4% PFA, permeabilized in 0.1% Triton X-100, blocked by 1% BSA, and stained with the indicated primary antibodies followed by FITC-conjugated anti–mouse IgG (Jackson ImmunoResearch Laboratories). Nuclei were counterstained with DAPI (Sigma-Aldrich). Slides were mounted by Aqua-Poly/Mount (Polysciences). Images were captured at room temperature using a confocal microscope (TCS SP2 ACBS; Leica) with a 63× NA 1.4 oil objective (Leica) except those in , which were captured using a camera (DP70; Olympus) on a microscope (BX51; Olympus) with a 10× NA 0.3 objective. The acquiring software was TCS (Leica) or DPcontrol (Olympus). 293T cells were transfected with the indicated siRNA. 48 h after transfection, cells were treated with 50 ng/ml TNF-α and 5 μg/ml CHX or left untreated for 18 h. Floating and adherent cells were collected and analyzed using the annexin V–phycoerythrin Apoptosis Detection kit I (BD Biosciences) and In Situ Cell Death Detection kit (TUNEL; Roche) according to the manufacturer's instructions. The flow cytometer used was a FACSCalibur (BD Biosciences).

Besides their established role at the plasma membrane, heterotrimeric G proteins and their regulators including guanine nucleotide exchange factors (GEFs), guanine nucleotide dissociation inhibitors (GDIs), and regulator of G protein signaling (RGS) proteins play a critical role in regulating microtubule (MT) pulling force during asymmetric cell division in and (). G-class GDIs, such as GPR1/2 and Pins, inhibit the release of nucleotide from G-GDP via their GoLoco domain. A GEF, Ric-8, likely stimulates nucleotide exchange of GoLoco protein–G–GDP complex, producing free G-GTP and signals force generation (). RGS, a G GTPase-activating protein (GAP), may also act as an effector by positively regulating the pulling force (). Altered expression of G proteins or their regulators in results in symmetric cell division, which causes inappropriate cell lineage determination and, ultimately, embryonic lethality. Emerging evidence suggests that mammalian heterotrimeric G proteins and their regulators also localize in the intracellular organelles and regulate MT pulling force (; ). However, the consequence of altered expression or function of these mammalian proteins on cell division has not yet been described. Unique among RGS and GDI proteins, RGS14 and RGS12 contain both an RGS domain for GAP activity and a GoLoco domain for GDI activity (). Both domains of RGS14 target members of the G subclass (). RGS14 also possesses two Raf-like Ras-binding domains, which overlap with the small GTPase, Rap-interacting domain (). RGS14 associates with centrosomes and MTs, and loss of expression in mice is catastrophic, resulting in the failure of zygotes to progress to the two-cell stage (; ). Very little is known about which activity of RGS14 is involved in centrosome/MT-related function and how the different activities of RGS14 are regulated in vivo. We show that the G proteins, targets for RGS14 regulation, localize in the centrosomes and midbody. We also demonstrate a direct interaction of RGS14 with G in the centrosomes and the necessity for normal G and RGS14 function for proper cell division. These results implicate heterotrimeric G protein–mediated signal transduction in centrosome biology and in cytokinesis. Based on RGS14 expression in centrosome and its G selectivity, we examined whether G, G, or G localized in the centrosomes (). A YFP fusion protein of G localized at the plasma membrane and cytoplasm, but it also colocalized with CFP-tagged RGS14 in centrosomes (). YFP expressed from the vector control evenly localized throughout the cell, except in the areas that appeared to be nucleoli. G-YFP also colocalized with endogenous centrosome proteins, including RGS14, γ-tubulin, and ninein (). Expression of G-YFP did not displace the endogenous centrosome proteins examined, suggesting that G-YFP expression did not interfere with centrosome recruitment of these proteins. G- and G-YFP also targeted to the centrosomes, colocalizing with another centrosome marker, pericentrin, as did the G-YFP (). Coexpression of RGS14-CFP was not necessary for targeting of YFP fusions of G, G, or G to the centrosomes. The YFP tag in the G-YFP constructs was shown not to interfere with G function (). The Glu-Glu (EE)–tagged G proteins also localized to the centrosomes, excluding the possibility of altered targeting caused by the YFP tagging (). The Alexa Fluor–conjugated secondary antibodies used in this study yielded no substantial staining of cells when used without primary antibodies (). Imaging of live cells transfected with the G-YFP constructs demonstrated that the fusion proteins were predominantly localized at the plasma membrane in most cells, although not in all cells (). The fixation of cells with 50% acetone/50% methanol (used for centrosome staining) and subsequent immunostaining resulted in a considerable loss of fusion proteins localized at the plasma membrane (). The YFP fusions of all three G proteins also showed strong expression at the junction between two daughter cells during cytokinesis in live cell imaging, unlike the YFP expressed from the vector control (). This may represent expression at the cleavage furrow, and/or possibly at the midbody, although these structures were difficult to discern in our low-resolution epifluorescence imaging of live cells. Next, we verified that endogenous G proteins localized in centrosomes of HeLa and NIH3T3 cells, which are known to express G, G, and G using G-, G-, or G-specific antibodies (; ). To verify the specificity of each antibody, lysates of HeLa cells expressing EE-tagged human G, G, or G were immunoblotted with the anti-G and -EE antibodies. The results showed minimal cross-reactivity (Fig. S1, available at ). To determine cell cycle position, we stained cells with anti- pericentrin or anti– γ-tubulin antibody and with Hoechst 33342 (). G and G localized in the centrosomes of interphase and mitotic cells, but G did not, contrary to what we observed with G-YFP () and to the recently published finding (). This is likely caused by centrosome expression of G at a level below detection by the antibody or epitope masking in the centrosomes. G expression is observed at the midbody; however, it is no longer detected in centrosomes during cytokinesis (). The expression pattern of G during cytokinesis mirrors that of pericentrin with modest midbody and strong centrosome expression (). We also detected midbody G and RGS14 staining (). Although RGS14 was reported to colocalize with MTs (), neither N- or C-terminally tagged RGS14 did so nor did we observe significant MT staining with three RGS14 antibodies raised independently (unpublished data). Finally, we reconfirmed the intracellular staining pattern of G and G using additional anti-G and -G antibodies raised independently and by demonstrating the absence of centrosomal staining of G in the cells isolated from G knockout mice (Fig. S2; ). We observed some inconsistency in the plasma membrane staining with various anti-G antibodies. The varying staining patterns by the antibodies raised against the same G proteins are likely caused by the difference in epitope recognition and affinity. Furthermore, the fixation with 50% acetone/50% methanol also contributed to the inconsistency in plasma membrane staining of endogenous G proteins. The fixation may weaken the integrity of plasma membrane and/or alter the antigenicity of G proteins. This inconsistency has also been observed in previous studies. In the study by , anti-G antibody stained only Golgi and cytoplasm, whereas reported strong plasma membrane and Golgi staining by an anti-G antibody in certain cells. However, the same anti-G antibody stained only Golgi (not the plasma membrane) in another type of cell (). To investigate whether G and RGS14 interacted in the centrosomes, we performed two independent fluorescence resonance energy transfer (FRET) analyses on HeLa cells transfected with RGS14-CFP and G-YFP. First, the acceptor photobleaching method was used on fixed cells expressing the two fusion proteins. Cells containing centrosomes expressing both fusion proteins at similar levels were found on the basis of their specific fluorescence intensities. A representative FRET image of acceptor photobleaching assays performed is shown in . Unlike the nonbleached centrosome, the CFP fluorescence intensity of the bleached centrosome increased considerably after YFP bleaching. The presence of FRET was also confirmed by using the sensitized emission FRET method on live cells. Initial images of live cells expressing a CFP/YFP fusion, RGS14-CFP, or G-YFP were use to adjust confocal microscope settings. The subsequent acquisition of FRET efficiency image clearly demonstrated a robust FRET signal in the centrosomes of live cells expressing wild-type RGS14-CFP and G-YFP (). To figure out which domain of RGS14 was involved in interaction with G in the centrosomes, we generated CFP fusions of various deletion and point mutants () and tested the ability of the various fusion proteins to bind G by acceptor photobleaching (). The average FRET efficiency of unbleached centrosomes (1%) served as a negative control. Bleached centrosomes expressing wild-type RGS14-CFP (HC30) and G-YFP showed robust FRET signals, with an average FRET efficiency of 10.7%. Neither the HC31 lacking the N-terminal 184 amino acids nor HC32 containing centrosome-targeted, Rap-interacting domain () yielded a true FRET. Various GoLoco domain deletion mutants, including HC33, could not be tested because they localized only in the nucleus (unpublished data). True FRET signals were observed from the HC34 and HC35 with an RGS and a GoLoco domain defective in G binding, respectively. A likely explanation for why the RGS domain deletion mutant behaves differently from the RGS domain point mutant is that the N-terminal deletion may have globally affected RGS14 protein conformation, interfering with the G–RGS14 GoLoco interaction. The HC36 containing both defective RGS and GoLoco domains showed no true FRET signal. Together, our data indicate that both the RGS and GoLoco domains are involved in G binding in the centrosome. A constitutively active Q204L mutant of G-YFP also produced a true FRET with wild-type RGS14 in the centrosomes. In contrast to the centrosome interaction, no or low-efficiency FRET signals were observed between RGS14 and wild-type or the mutant G in the cytoplasm, suggesting a different mechanism of interaction in the centrosomes. RGS14 was recently shown to bind both wild-type and GTPase-deficient forms of G and G at the plasma membrane (). Next, we examined the effect of altered G expression or function on cell division by time-lapse videomicroscopy. To assess progression of cell division and measure the duration of mitosis or cytokinesis, we used visual landmarks. Rounding up of attached cells was used as the initiation of mitosis, emergence of a daughter cell as initiation of cytokinesis, and the loss of roundness in conjunction with inability to detect intercellular bridge (reattachment) as the end time point for cytokinesis. These measurable durations may not truly reflect the actual durations, although they can be used for comparison. For example, absolving of midbody after mitosis, which cannot be visualized in our live imaging, may take up to several hours (; ; ). Of almost all the cells expressing YFP, only control, wild-type G, or G QL mutant underwent relatively normal cell division ( and Videos 1 and 2, available at ). The measurable durations of mitosis and cytokinesis of these cells were ∼30 min for these cells. However, a considerable number of dividing cells expressing wild-type or QL mutant of G (∼35 and 20% of cells examined, respectively) showed prolonged mitosis with an average duration of 176 min (from the metaphase to the initiation of cytokinesis), but underwent relatively normal cytokinesis ( and Video 3). In an extreme case, a cell stayed in the mitotic phase for >8 h before initiation of cytokinesis. In contrast, cells expressing the G QL mutant exhibited relatively normal mitosis, but ∼40% had prolonged or unresolved cytokinesis ( and Video 4). More than 7 h lapsed between the initiation of cytokinesis and the appearance of midbody in the cell shown in . Two daughter cells remained unattached (rounded up). Mitosis was also slightly delayed in ∼15% of cells expressing wild-type G (unpublished data). We treated cells with pertussis toxin (PTX) and monitored cell cycle progression using videomicroscopy ( and Videos 5 and 6, available at ). PTX interferes G GDP/GTP exchange (except for G) triggered by G protein–coupled receptors (GPCRs). Differential interference contrast (DIC) images of almost all dividing control NIH3T3 cells showed short or no visible intercellular bridges during cytokinesis with two dividing cells being aligned at nearly 180° angles, whereas PTX treatment resulted in formation of abnormally extended intercellular bridges and misalignment of MTs (). Some cells interconnected via an intercellular bridge appeared to coalesce into single cells (Video 6). No apparent defects in mitotic spindle formation were observed with the PTX-treated cells (based on α-tubulin staining; unpublished data). An average of 12% of PTX-treated cells showed multinucleation, compared with an average of 3% of nontreated cells, which is likely the result of failed cytokinesis seen in videomicroscopy (). We took advantage of a recent report demonstrating efficient silencing of G, G, and G in HeLa cells using siRNAs and closely followed the previously described method (). A significant compensatory increase in G expression was observed after silencing of G or G in HeLa cells, demonstrating the adaptability of G protein signaling networks. Different G isoforms may also be able to functionally compensate for the lack of others. Therefore, we used two siRNAs reported to simultaneously silence all three G isoforms (G siRNA; ). Cells were harvested for immunoblotting and immunocytochemistry or subjected to videomicroscopy 6–7 d after the first transfection. Transfection efficiency was monitored with DY547-tagged siRNA yielding an average of ∼90% efficiency (unpublished data). Immunoblotting demonstrated a significant reduction in expression of all three G isoforms in G siRNA–transfected cells compared with cells transfected with siRNA control (). The centrosome staining by anti-G or -G antibody was also significantly reduced (an average of ∼40% reduction for both G and G) in HeLa cells transfected with G siRNA (). Because the localization of G in the centrosomes was recently reported () and our anti-G antibody did not recognize the centrosomal G, we did not perform immunostaining with this antibody. HeLa cells transfected with siRNAs were then monitored for cell cycle progression by videomicroscopy ( and Videos 7 and 8, available at ). DIC images of dividing cells transfected with control siRNA showed normal cell division with short or no visible intercellular bridges during cytokinesis. Cells with reduced expression of all three G isoforms exhibited mainly cytokinesis defects. Most of these cells appeared to progress normally through mitosis, except for ∼7% of cells with moderately prolonged mitosis. The average duration from metaphase to the initiation of cytokinesis of those cells with prolonged mitosis was 127 min compared with ∼30 min in control cells. More strikingly, ∼50% of the cells examined exhibited cytokinesis defects. A majority of the defective cells separated into two daughter cells after prolonged periods of time in cytokinesis (an average duration of 322 min compared with 33 min in control cells). A few interconnected cells coalesced to form single cells (, si-Gb). Abnormally extended intercellular bridges and misalignment of MTs were also observed with HeLa cells transfected with G siRNA (). No apparent defects in mitotic spindle formation were observed (unpublished data). DNA staining using Hoechst 33342 revealed a threefold increase in multinucleation with G siRNA–transfected cells compared with control siRNA–transfected cells (). Flow cytometric analysis of live cells confirmed the same threefold increase in multinucleation (). The higher percentage of multinucleation observed with the immunocytochemistry experiment is likely caused by the inclusion of cells with micronuclei. RGS14 localized in the centrosomes and at the midbody, and silencing of RGS14 caused multinucleation in HeLa (). Using videomicroscopy, we tested whether the multinucleation caused by RGS14 silencing might also be the result of defective cytokinesis. The aforementioned method was used for siRNA silencing of RGS14 in HeLa. Cells were harvested for RT-PCR or subjected to videomicroscopy 6–7 d after the first transfection. To demonstrate reduced expression of RGS14, RT-PCR was used instead of immunoblotting because our antibody recognized RGS14 protein only in immune cell lysate, where RGS14 expression was relatively high, but not in HeLa cell lysate with a lower level of RGS14 expression (). Compared with control cells, 12% of cells transfected with RGS14 siRNA showed slightly longer duration in mitosis, with an average duration of 106 min (unpublished data). Approximately 30% of the RGS14 siRNA-transfected cells exhibited cytokinesis defects ( and Videos 9 and 10, available at ). A majority of the defective cells remained in cytokinesis for prolonged periods of time (average duration of 381 min) before separating into two daughter cells. A few interconnected cells with reduced RGS14 expression also coalesced to form single cells. MT staining by anti–α-tubulin antibody revealed a striking defect in HeLa cells transfected with RGS14 siRNA (). The thickness of MTs in the intercellular bridge extending from one daughter cell was much thinner than that of the other daughter cell in many interconnected cells. Emerging evidence suggests that heterotrimeric G proteins and their regulators localize in diverse intracellular compartments and may function independently of GPCRs. A recent genetic study has demonstrated a new mechanism for G signaling in yeast (). The yeast G, Gpa1, localizes to endosomes and directly binds PI3K instead of pairing with G. Intriguingly, the catalytic subunit of PI3K binds preferentially to the activated form of G, whereas the regulatory subunit of PI3K prefers the inactive GDP-bound form, suggesting cycling between active and inactive forms of G in the endosome. In , both G and G localize on asters and are implicated in regulation of centrosome movement and spindle positioning (Gotta and Ahringer, 2001). Mammalian proteins that regulate G protein activity, such as LGN and RGS14, are also reported to localize in the centrosomes and at the midbody (; ; ; ). In this study, we show that three mammalian G isoforms, G, G, and G, localize in the centrosomes and at the midbody. FRET assays demonstrate that RGS14 can bind G protein in the centrosomes via both RGS and GoLoco domains. Preliminary immunocytochemistry data indicate that a mammalian GEF, Ric-8A, also resides in centrosomes (unpublished data). These results suggest that cycling between GDP- and GTP-G may also be of functional importance in the mammalian centrosomes. Like the two subunits of yeast PI3K, centrosomal RGS14 can bind both inactive and active form of G. LGN, which recruits NuMA to the cell cortex, and possibly to the spindle poles during mitosis, can bind the inactive G (; ). Centrosomal Ric-8A may dissociate the G-GDP–LGN–NuMA complex releasing G-GTP and NuMA, thereby regulating the MT function, as reported at the cell cortex (). Interestingly, forced expression of G and G gave two distinct phenotypes during cell division. Both wild-type and the GTPase-deficient form of G resulted in prolonged mitosis, although they did not affect cytokinesis. The GTPase-deficient form of G caused defective cytokinesis, but did not impact mitosis. Overexpression of the wild-type G, wild-type G, or the GTPase-deficient form of G did not reveal any apparent abnormalities during cell division. The difference between G and G may arise, in part, from their differing intracellular localization. G is present in the centrosomes early in mitosis, and it shifts to the midbody, as has been observed with the centrosome proteins Cep55 and centriolin (). It also differs slightly from G and G in being more centrally located in the midbody region. Because it is likely that multiple regulators of G are involved, the phenotypic difference may also reflect differences in binding specificities of the regulatory proteins. For example, LGN and Ric-8 can bind only the GDP-bound G, but not the QL forms. In contrast, RGS14 can bind both GDP- and GTD-bound G and G, although it is unknown whether and how these interactions are regulated. There may be spatial and/or temporal regulation determining where and when the interaction between various G proteins and their regulators occurs. Expression of G and G induced altered spindle orientation in mammalian neural progenitors and abnormal rocking motion of chromosome in MDCK cells, respectively, although how these defects affected cell division was not reported (; ). Interfering with G function by PTX or with G expression by siRNAs resulted in mainly defective cytokinesis. The mitotic spindle and spindle midzone (likely regulated by centrosome function) provide spatiotemporal control over many of the mechanical events occurring at the cleavage furrow during cytokinesis (). It is reported that depletion of the centrosome/midbody protein centriolin results in cytokinesis failure without affecting mitosis (). In Swiss3T3 and neuroepithelial cells, PTX treatment impaired cell proliferation, which was suggested to result from inhibited GPCR signaling (; ). However, the PTX-induced intercellular bridge and MT defects may be the consequence of abnormal G function in the centrosomes. PTX may also interfere with G GDP/GTP exchange by centrosomal Ric-8A. Whether G is coupled with G in the centrosome/midbody has never been examined. Whether PTX can ribosylate G proteins complexed with a protein other than G also remains to be seen. In addition, defective G function at the midbody may contribute to the observed defects. Although the exit from cytokinesis was severely delayed in G siRNA-transfected cells, many interconnected cells eventually became separated. This may be caused by the residual expression of G proteins in knockdown cells. Reduced expression of RGS14 also induced cytokinesis defects, suggesting that GAP and/or GDI activity may be required for proper cell division. The uneven thickness of intercellular bridge MTs caused by decreased RGS14 expression suggests dysregulation of MT stability/dynamics or uneven pulling force, leading to abnormal cytokinesis. It is not clear how MTs in the intercellular bridge extended from one daughter cell, but not from the other is affected. Contrary to a previous report (), we did not observe any significant reduction in tubulin staining in the cells transfected with RGS14 siRNA. The difference may be caused by our relatively modest knockdown compared with theirs. More studies are needed to model a molecular mechanism by which the G proteins and regulators exert their control on cell division via centrosome/midbody function. However, the following mechanisms are conceivable. First, the mechanism proposed for MT pulling force involving G, LGN, NuMA, and Ric-8A may be used to regulate MT stability, dynamics, or pulling force at these sites (; ; ). RGS14 may serve to regulate these processes. It may act as a GAP via RGS domain and/or sequester G away from LGN via GoLoco domain, thus interfering with NuMA interaction. Second, G proteins may regulate MT function via direct interaction with tubulins. Both G and G modulate MT assembly in vitro, and the heterotrimer inhibits the ability of G to promote MT assembly, suggesting that G protein activation is required for functional coupling between G and tubulin/MTs (). Third, analogous to the signal transduction at the plasma membrane, G proteins and regulators may activate and deactivate yet-to-be identified centrosome/midbody effectors. G proteins and regulators are present in many different cellular compartments. Therefore, the phenotypes we observe may not arise solely from perturbation of G function in the centrosomes/midbody. Dysregulation of the endogenous G subunits at the cell cortex/plasma membrane may also contribute to the observed phenotypes. However, differential localization of G proteins to centrosome and midbody, the mitotic and cytokinesis defects observed with G overexpression, G underexpression, or PTX-treated cells, in addition to reported direct role of and G and regulators in MT function strongly argue that they play a more direct role during cell division. There are many unresolved issues, such as which proteins are the downstream effectors of centrosomal and midbody G proteins, whether or how the G proteins and their regulators, such as RGS14, LGN, and Ric-8A, are regulated during cell cycle, and how these proteins are targeted. Future studies should help resolve these issues in what is a new and exciting avenue for heterotrimeric G protein research. A rat G-YFP construct was a gift of M. Lohse (University of Wurzburg, Wurzburg, Germany; ). The human G-, G-, and G-YFP constructs were provided by S. Gibson (University of Texas Southwestern Medical Center, Dallas, TX; ). The rat and human G-YFP constructs generally produced similar results in our assays. EE-tagged G constructs were purchased from Unite Mixte de Recherche cDNA Resource Center. The antibodies used were purchased as follows: anti–γ-tubulin from Sigma-Aldrich; anti-EE monoclonal from Covance; anti-ninein polyclonal from Abcam; monoclonal anti-G from Millipore; polyclonal monoclonal anti-G from EMD Biosciences; and monoclonal anti-G and -G and polyclonal anti-G from Santa Cruz Biotechnology, Inc. To construct CFP fusions, various mouse RGS14 (available under GenBank accession no. ) DNA fragments were cloned into pN1-CFP vector in frame with C-terminal CFP, and RGS14 and G point mutants were generated as previously described (). Immunocytochemistry and immunoblotting were performed as previously described (). Cells were fixed with 4% PFA/0.1% Triton-X100 for MT staining using anti–α-tubulin antibody or with 50% acetone/50% methanol for centrosome staining using antibodies raised against centrosome proteins. The method of was closely followed for G siRNA knockdown in HeLa cells. G (721–739; CCGAAUGCAUGAAAGCAUG) and G (681–699; CUUGAGCGCCUAUGACUUG) siRNAs, control siRNA, and DY547-tagged siRNA to monitor transfection efficiency were purchased from Dharmacon. For RGS14 siRNA silencing, three RGS14 siRNAs (216–226, AACGGGCGCAUGGUUCUGGCU; 347–367, AACCGAGGAGCAGCCUGUGGC; 474–494, AAGGCCUGCGAGCGCUUCCAG; available under GenBank accession no. ) were selected based on the Dharmacon siRNA user guide and purchased from Dharmacon. HeLa cells were transfected with the pool of three siRNAs, as described above. Images were collected on a TCS-SP2 or SP5 confocal microscope (Leica) and processed as previously described (), except that 561- and 405-nm diode lasers were used for Alexa Fluor 568 and Hoechst 33342, respectively. When it was not possible to eliminate cross-talk between channels, the images were collected sequentially and later merged. The exposure times were kept equal within each series of images and chosen such that all pixel intensities were within the linear range. A 63× oil lens was used, unless mentioned otherwise. Confocal zoom factors between 1 and 8 (Z1 to Z8) were used. Acceptor photobleaching module of the Leica software was used for acceptor photobleaching FRET assay. The 405-nm laser line was used to excite CFP and the 514-nm laser line was used to excite and bleach YFP. Microscopic fields containing two to three centrosomes that expressed both RGS14-CFP and G-YFP with similar levels of expression were found. A region of interest was drawn on only one of the centrosomes in the field or on a randomly chosen area in the cytoplasm for bleaching. The Leica confocal software was configured to achieve a 50 or 80% bleach of YFP only in the selected region of interest. FRET efficiency was calculated as follows: FRETeff = − / for all > when and were donor fluorescence intensity before and after photobleaching, respectively. The Leica acceptor photobleaching method collected 8-bit images (gray scale 0–256). Therefore, any fluorescent intensity <20 was regarded as background, as these were near the limits of detection. At least 20 different cells were examined for the presence of FRET signals in each experiment. The FRET efficiencies of each construct were averaged for comparison. For live cell FRET analysis, cells were imaged using sensitized emission routine of the Leica software. A CFP-YFP fusion was used as a positive control. Images of RGS14-CFP only or G-YFP only were acquired to correct for cross talk between channels. FRET efficiency was calculated as follows: FRETeff = B − b × A − (c − a × b) × C/C where A represents the fluorescence intensity of channel 1 (donor excitation/donor emission); B represents the fluorescence intensity of channel 2 (donor excitation/FRET emission); C represents the fluorescence intensity of channel 3 (acceptor excitation/acceptor emission); a represents the correction factor of acceptor only measurement (donor emission × excitation for the donor/acceptor emission × excitation for the acceptor); b represents the correction factor of donor only measurement (acceptor emission × excitation for the donor/donor emission × excitation for the donor); and c represents the correction factor of acceptor only measurement (acceptor emission × excitation for the donor/acceptor emission × excitation for the acceptor). The Leica sensitized emission module used 12-bit images (gray scale 0–4,096). Therefore, any fluorescent intensity <200 in all three channels was regarded as background, as these were at the limit of detection. A microscope (DMIRBE; Leica) equipped with a 1,376 × 1,040 cooled charge-coupled device camera (Sensicam QE; Cooke) was used to capture time-lapse images. Microscope settings such as exposure time and magnification of objective (20×) were kept constant. This microscope was also equipped with Pe-Con environmental chamber (PeCon GmbH) that maintained temperature, CO, and humidity levels for long-term imaging. DIC and fluorescence images were captured using ImagePro (Media Cybernetics) or IpLab (BD Biosciences). The collected images were processed using the Imaris (Bitplane, Inc.) and reconstructed into videos using QuickTime software. Fig. S1 shows specificity of anti-G, -G, and -G, antibodies. Fig. S2 reconfirms intracellular localization pattern of endogenous G and G using a second set of anti-G monoclonal and anti-G polyclonal antibodies, as well as cells isolated from G knockout mice. Video 1 shows normal cell division of HeLa cells expressing vector control YFP. Video 2 shows normal cell division of HeLa cells expressing G2 QL-YFP. Video 3 shows prolonged mitosis of HeLa cells expressing wild-type G3-YFP. Video 4 shows prolonged cytokinesis of HeLa cells expressing G1 QL-YFP. Video 5 shows normal cell division of NIH3T3 cells. Video 6 shows defective cytokinesis of PTX-treated NIH3T3 cell. Video 7 shows normal cell division of HeLa cells transfected with control siRNA. Video 8 shows defective cytokinesis of HeLa cells transfected with G siRNA. Video 9 shows defective cytokinesis of HeLa cells transfected with RGS14 siRNA. Video 10 shows coalescing of two daughter HeLa cells transfected with RGS14 siRNA. The online version of this article is available at .

Cyclin-dependent kinases (Cdks) and their regulatory cyclin subunits play a crucial role in cell cycle control (). In budding and fission yeast, a single Cdk, bound to different sets of cyclins, initiates DNA synthesis and centrosome duplication, suppresses re-replication of already duplicated DNA, and triggers entry into mitosis once replication is complete (; ). Higher eukaryotes have evolved a group of specialized Cdks, each of which is active in a different phase of the cell cycle (). Cdk1 together with cyclin A and B forms the maturation- promoting factor, and is required for entry into mitosis. Cdk2 bound to cyclin E and A was considered to be essential for initiation and completion of DNA replication, and the control of centrosome duplication, until several groups found that mice lacking Cdk2 develop normally (; ). This raises the question of which Cdk controls the initiation and completion of S phase in the absence of Cdk2. Although Cdk1 is an apparent candidate for this redundant S phase Cdk, as proposed, an essential function for vertebrate Cdk1 during G1 and S phase has not been directly demonstrated. In fact, Cdk4 has also been implicated recently as a back up kinase for Cdk2 in G1 phase (). Hence, we do not know to what extent different Cdks overlap in the initiation of S phase in vertebrate cells. In addition to the initiation of replication, the inhibition of endoreplication is another essential S phase function of yeast Cdk1, which ensures that each replication origin fires only once per cell cycle by inhibiting the untimely assembly of pre-replication complexes (pre-RCs) (). At the exit from mitosis, Cdk1 activity is shut down by the anaphase promoting complex, also known as cyclosome (APC/C), which triggers cyclin destruction (). This inactivation of Cdk1 by cyclin proteolysis seems sufficient for the re-licensing of origins in the next G1 phase (). This idea is supported by the observation that artificial inactivation and reactivation of yeast Cdk1 are sufficient to reset the cell cycle and induce endoreplication (). Several studies also implicate Cdk1 in the inhibition of endoreplication in flies and human cells (; ; ). However, higher eukaryotes, but not yeast, contain an additional licensing inhibitor, Geminin, which binds to and inactivates the pre-RC assembly factor Cdt1 (; ; ). Moreover Cdk-dependent and -independent proteolysis pathways regulate the stability of the licensing factor, Cdt1 during S phase (). It remains elusive how Geminin, Cdk1 activity, and proteolysis of Cdt1 are coordinated to suppress endoreplication in human cells. The following two questions arise regarding the contribution of Cdk1 to the control of S phase: Is Cdk1 involved in the initiation of DNA replication and centrosome duplication? Is Cdk1 inhibition sufficient to induce endoreplication in vertebrate cells, despite the presence of Geminin? These questions have not been sufficiently addressed, owing to the difficulty to specifically, rapidly, and effectively inactivate Cdk1. In fact, a conditional deletion of the Cdk1 promotor in a human cell line has been achieved, but the levels of the kinase drop only very slowly and incompletely (). A mouse cell line (FT210) that carries a temperature-sensitive mutation has also been isolated, but this cell line appears to maintain about 25% kinase activity at the restrictive temperature (). A variety of chemical inhibitors of Cdk1, such as Roscovitine and Olomoucine, have been used to explore Cdk1 function (; ). However, these inhibitors are likely to affect other kinases within and possibly outside of the Cdk family. To increase the specificity of chemical inhibition, Shokat and coworkers recently developed a chemical genetics approach to sensitize kinases to bulky ATP analogs by mutating a conserved bulky residue in the active site (; ). This strategy has been successfully applied to Cdk1 and other kinases in yeast (), and a similar approach has been exploited to study Jun and Trk kinase in mouse models (; ; ) and in human cells to analyze Cdk7 (). We have taken advantage of the high gene-targeting frequencies in chicken DT40 cells to disrupt the endogenous chicken gene, encoding the Cdk1 kinase, and ectopically express a mutant Cdk1 cDNA ( that is selectively sensitive to inhibition by the ATP analog 1NM-PP1. Using this system, we have investigated S phase functions of vertebrate Cdk1. We found that Cdk1 activity is essential for triggering DNA replication and centrosome duplication in cells lacking Cdk2. Conversely, if Cdk2 is present, Cdk1 inhibition does not delay S phase or block centrosome duplication. We also show that whereas inhibition of Cdk1 in G2 phase, before entry into mitosis, does not induce endoreplication, inhibition of Cdk1 during prometaphase does stimulate origin licensing and endoreplication. This depends on the proteolysis of Geminin. These results clarify the role of vertebrate Cdk1 in controlling replication and endoreplication. We initiated this study by establishing DT40 cell lines, in which we disrupted the endogenous chicken Cdk1 by gene targeting and exogenously expressed either an analog sensitive F80G mutant ( wild-type () cDNA of Cdk1 (Fig. S1, A–E; available at ). We isolated two independent cell lines, with slightly different levels of Cdk1as expression. The cell line grew with similar kinetics as DT40 WT cells, whereas cells that expressed less Cdk1 transgene showed a slight growth retardation. (Fig. S1 F). In both cases the activity of the Cdk1 kinase was reduced when compared with WT Cdk1, probably due to the F80G mutation in the active site. We continued to work with the cell line, hereafter called cells. To confirm that the mutant Cdk1 was indeed selectively sensitive to the bulky ATP analog inhibitor 1NM-PP1, we immunoprecipitated Cdk1 from extracts of Cdk1-deficient cells reconstituted with WT Cdk1 ( cells) or cells, and measured the Cdk1 kinase activity of the immunoprecipitates in the presence or absence of 1NM-PP1, using Histone H1 as a substrate. Although the amount of Cdk1 was comparable in the immunoprecipitates (unpublished data), the kinase activity of cells was reduced to about 20% of WT Cdk1 activity (). Addition of 10 μM 1NM-PP1 inhibited phosphorylation of Histone H1 by the mutant kinase but had no effect on the WT Cdk1. Likewise, 10 μM 1NM-PP1 had no effect on the growth of cells, but abolished proliferation of cells (). To investigate the effects of Cdk1 inhibition on the cell cycle, we isolated the G1 fraction of and cells by elutriation, added 10 μM 1NM-PP1 to the media, and took samples every 2 h for FACS analysis of the DNA content. shows that both cell lines initiated and completed S phase with very similar kinetics. The cells subsequently completed mitosis and re-entered the next cell cycle, whereas cells remained arrested in the 4N state (). This arrest was maintained for several days without further division and DNA synthesis (). To analyze the activity of the APC/C in the arrested cells, we measured cyclin levels in arrested cells, in which de novo protein synthesis was inhibited by cycloheximide (CHX). Both cyclins were stable during the G2 arrest for more than 6 h after CHX treatment (, left). CHX did not interfere with cyclin destruction after the release from a Nocodazole block, excluding the possibility that CHX itself affects the proteolysis of cyclins (, right). We also found that cyclin B2 localized predominantly in the cytoplasm throughout this prolonged arrest (). These observations indicate that Cdk1 inhibition blocks the cell cycle in the G2 phase before APC/C activation and translocation of cyclin B to the nucleus. To explore whether the arrest induced by Cdk1 inhibition was reversible, we removed 1NM-PP1 after 8 h incubation of cells with the inhibitor. This resulted in rapid entry into M phase, as evidenced by Histone H3 phosphorylation, and cyclin destruction, during a synchronous passage through mitosis (Fig. S2, available at ). These findings suggest that the active state of Cdk1, including association of cyclin B, is unaltered during the 1NM-PP1 mediated inhibition of , allowing the rapid activation of Cdk1 upon removal of the inhibitor during the G2 arrest. mutants (Fig. S3, available at ). Ablation of Cdk2 had little effect on DT40 WT cells, but retarded the cellular proliferation in the background even in the absence of 1NM-PP1 (). cells that were synchronized in G1 phase by elutriation took approximately 2 h longer than cells to initiate S phase (compare the FACS histogram in with ). Nonetheless, the double-mutant cells were still able to complete S phase and accumulated in G2 phase even in the presence of a low dose (1 μM) of 1NM-PP1 (). A tenfold higher dose of the inhibitor blocked the asynchronous cell cycle both in G1 and G2, as judged by the histogram of DNA content (). This suggests that Cdk1 is responsible for S phase control in the absence of Cdk2. cells. 8 h after Cdk1 inhibition, cells contained two separated centrosomes, and the centrosome number appeared to double subsequently during 8-h intervals (). cells that were treated with 1 μM 1NM-PP1 (). treated with 10 μM 1NM-PP1 (). cells indicates that Cdk1 and Cdk2 share overlapping functions in the control of both DNA replication and centrosome duplication. cells in closer detail, by examining cell cycle progression. We collected the G1 fraction of cells, treated them with 10 μM 1NM-PP1, and analyzed the subsequent progression through the cell cycle (). cells was able to increase their DNA content (), and uptake BrdU (Fig. S4, available at ) in the presence of 1NM-PP1. cells (; Fig. S4). These observations indicate functional redundancy between Cdk1 and Cdk2 for the initiation of S phase. and cells using pulse-chase BrdU labeling. After 10 min of BrdU exposure, G1, S, and G2 fractions of the cells were clearly distinguishable by dot blot analysis of BrdU/PI double staining (). BrdU was subsequently removed from the medium and the cells were further incubated in the presence of 10 μM 1NM-PP1. cells were able to complete DNA synthesis, and the entire population of the cells was blocked in the G2 phase. mutants did not initiate DNA synthesis, whereas the mid S phase cells were shifted toward G2 (). Thus, although ongoing DNA synthesis during S phase does not appear to require Cdk1/2 activity, either Cdk1 or Cdk2 activity is essential to initiate DNA replication. cells, we measured the levels of cyclin A and B in cells synchronized in G1 by elutriation (). Both cyclins accumulated even after addition of 10 μM 1NM-PP1, suggesting that Cdk1/2 inhibition does not interfere with the expressions of genes required for the G1/S transition. demonstrated that artificial inactivation and reactivation of Cdc2 in G2 phase is sufficient to reset the cell cycle and to initiate a further round of DNA replication, without previous chromosome segregation. cells in G2 phase by addition of 1 μM 1NM-PP1 for 6 h, and subsequently increased the inhibitor concentration to 10 μM for an additional hour to completely block Cdk1 activity. Afterward, we washed the inhibitor out with excess medium, and monitored the cell cycle progression by measuring DNA content by using FACS analysis (). Surprisingly, unlike , this inhibition and reactivation of Cdk1 in G2 phase did not result in endoreplication, and all cells initiated the next round of replication only after mitosis (; note the absence of cells containing DNA >4N). This observation is in marked contrast with a previous study by , who showed that a human fibroblast cell line continued to increase in ploidy, after conditional inhibition of Cdk1 expression. cells, we analyzed the recruitment of Mcm proteins, essential components of the pre-RC, to chromatin. In G2 phase, Mcm2–7 are excluded from chromatin, and need to be loaded as a hexameric ring structure onto the DNA to license origins for a new round of DNA replication. cells that were synchronized in G2 by1 μM 1NM-PP1, and then further treated with 10 μM 1NM-PP1 to fully block Cdk1 (, lane 5). Moreover, additional inhibition with the general Cdk inhibitor Roscovitine also failed to induce Mcm2–4 binding to chromatin (, lane 6). Thus, inhibition of both Cdk1 and Cdk2 during G2 phase may not be sufficient for origin licensing in chicken DT40 cells. Previous studies suggested that the general kinase inhibitor DMAP induces Mcm loading onto chromatin in the G2 phase of HeLa cells (). We repeated the same experiments by inhibiting Cdks more specifically by treating HeLa cells with Roscovitine. For this purpose, we synchronized the cells in early S phase by a double thymidine block, released them, and analyzed the chromatin binding of both the licensing factor Cdt1 and Mcms, as cells progressed through S phase. After 7 h after release, when most cells were in G2 phase (), Mcms had been largely displaced from the chromatin (, lane 8). At this time, we added Roscovitine to the cells, while a control sample was left without Cdk inhibition (, C and D; compare “9” with “9+Ros”). Both samples were treated with Nocodazole, an inhibitor of spindle formation, to avoid the entry in the next G1 phase. In accordance with the results obtained with chemical genetics in DT40 cells (), Roscovitine did not appear to induce Mcm loading onto chromatin in the G2 population of HeLa cells (, lane 10). In conclusion, Cdk inhibition in G2 phase is not necessarily sufficient to induce origin licensing in chicken and human cells. recently observed an induction of endoreplication in Nocodazole arrested human cells treated with a Cdk1 inhibitor. We wished to know whether Cdk1 inhibition by the chemical genetics method also triggers endoreplication in mitotic DT40 cells. To avoid a prolonged treatment with Nocodazole, we first synchronized cells in G2 using 1NM-PP1, and then released them into mitosis by removing the inhibitor, while adding Nocodazole to obtain cells that were briefly arrested in prometaphase (see diagram in ). In contrast to the cells arrested in G2 (), the mitotic cells exhibited DNA replication without completing mitosis, after addition of 10 μM 1NM-PP1 (). Cdk1 inhibition also led to rapid dephosphorylation of Histone H3, indicative of decondensation of chromosomes (). We also observed the quick dephosphorylation of the APC/C subunit Cdc27 (), which is hyperphosphorylated by Cdk1 in mitosis (). Moreover, we found that the APC/C was activated in response to mitotic Cdk1 inhibition, as judged by the rapid degradation of cyclin A upon addition of the inhibitor (). Nocodazole arrested cells showed no such response to 1NM-PP1 and remained unchanged in mitosis (unpublished data). These data suggest that inhibition of endoreplication may be carried out differently in G2 and M phase. A possible explanation for the differential effects of Cdk1 inhibition on DNA synthesis in G2 and M phase could be the activation of APC/C mediated proteolysis during mitosis but not G2 phase (compare cyclin stability in and ). To test this hypothesis, we treated cells that were synchronized in mitosis as shown in with the proteasome inhibitor MG132 before Cdk1 inhibition. We found that MG132 prevented the induction of DNA synthesis by 1NM-PP1 (compare with ). Furthermore, MG132 suppressed the induction of Mcm loading on chromatin after Cdk1 inhibition (, compare lane 5 with lane 6). These results suggest that Cdk1 inhibition is not sufficient to allow origin licensing and endoreplication, unless proteolysis is activated. To verify these results from DT40 cells in a human cell line, we analyzed Nocodazole-arrested HeLa cells treated with Roscovitine. We found that this Cdk inhibitor initiated origin licensing, as judged by Cdt1 and Mcm2 loading onto chromatin, as previously described by . Inhibition of the proteasome by MG132 abolished the Roscovitine induced chromatin binding of Mcm2 and Cdt1 in HeLa cells (, lane 3 and lane 6), confirming our previous results with DT40 cells (). The proteolysis dependence of origin licensing and endoreplication induced by Cdk1 inhibition, suggests that proteins other then cyclins need to be degraded to allow pre-RC formation. Geminin, which is an APC/C substrate and a licensing inhibitor (; ), is a good candidate to account for this proteolysis requirement after Cdk inhibition. Accordingly, we found that Roscovitine triggered the destruction of both cyclin B1 and Geminin (, lane 2), whereas MG132 treatment stabilized these proteins (, lane 3). shows that the APC/C targets Aurora kinase A and Cdc20 are also degraded upon Cdk inhibition during mitosis. To test if the inhibition of Geminin degradation is sufficient to prevent Roscovitine-induced origin licensing, we investigated the effects of ectopic expression of a Geminin mutant that is resistant to APC/C dependent degradation, on mitotic HeLa cells treated with Roscovitine (). We transiently expressed this stable Geminin mutant in HeLa cells, after release from a double thymidine block. As a control experiment, we expressed GFP or an APC/C-resistant mutant of mouse cyclin B1 in HeLa cells treated in the same manner. The transfected cells were arrested in prometaphase by Nocodazole treatment. In the control samples, licensing was initiated by the addition of Roscovitine, as judged by the chromatin loading of Mcms and Cdt1 (, lane 8 and lane 10). In contrast, expression of the stable Geminin mutant inhibited loading of Cdt1 and Mcm onto chromatin in response to Roscovitine treatment (, lane 12). In fact, we observed that even overexpression of WT Geminin in mitosis partially inhibited Mcm loading (unpublished data). We conclude that Geminin needs to be targeted for degradation by the APC/C, even after Cdk inactivation, to allow for origin licensing and endoreplication in vertebrate cells. To clarify the relationship of Geminin and Cdks during G2 phase, we tested if Geminin depletion and Cdk inhibition is sufficient to induce origin licensing and endoreplication in HeLa cells, synchronized in G2 phase as shown in . Surprisingly, Roscovitine treatment during the G2 phase did not cause Mcm loading onto chromatin even after Geminin depletion (, lane 7). We noticed that in these cells Cdt1 was hardly detectable, suggesting the Cdk inhibition was not sufficient to stabilize this licensing factor during the G2 phase (note that the Cdt1 antibody used in this experiment and in bottom lane of , produces a cross-reacting band, marked with an asterisk). Conversely, inhibition of proteolysis by MG132 caused an increase in Cdt1 levels, and resulted in origin licensing, however, only after Geminin depletion. This origin licensing in G2 occurred with similar efficiency with or without Cdk inhibition (compare , lanes 11 and 12). To test if these prematurely licensed origins were able to trigger re-replication we incubated cells synchronized in G2 with MG132 and/or Roscovitine. After 2 h we removed the inhibitors from the medium and measured the DNA content of the cells after further 12 h of incubation (see experimental outline in ). The FACS histograms in show that Geminin depletion and MG132 treatment but not Cdk inhibition was sufficient to induce endoreplication in the majority of these G2 cells. Our results show that proteolysis has opposing effects on endoreplication during G2 and M phase. One of these differences could to be the stability of licensing factors such as Cdt1. We compared Cdt1 levels in G2 and M phase using two different Cdt1 antibodies. For this purpose, we synchronized HeLa cells by a double thymidine release in G2 phase and inhibited cell cycle progression into M phase by Roscovitine addition, or allowed the cells to proceed into M phase by adding nocodazole. shows that Cdt1 levels decreased after Cdk inhibition by Roscovitine, but were markedly increased in the Nocodazole treated cells. These results indicate that Cdt1 levels are kept low during G2 phase independently of Cdk activity, and are rapidly increased after mitotic Cdk1 activation and entry into M phase. This study describes the use of chemical genetics to study vertebrate Cdk1. We generated a chicken DT40 cell line, in which we compensated the deleted endogenous Cdk1 with a mutant cDNA of the Cdk1 orthologue. The -mutation of the active site has varying effects on different kinases. In the case of Cdk1, the mutation of F80G significantly reduces the activity of the kinase. To overcome this obstacle, we chose two different strategies. First, we used the cDNA of Cdk1, because we found that this Cdk1 orthologue was relatively resistant to the introduction of the F80G mutation, retaining 20% of the WT kinase activity (see ). Second, we aimed to compensate this fivefold reduction by overexpression of the mutant kinase, and selected a stably transfected clone that showed a fourfold increase of Cdk1 levels when compared to WAK cells (unpublished data). However, using a PSTAIRE antibody that should cross-react to frog and chicken Cdk with the same affinity, we determined that the levels of the exogenously expressed Cdk1 were approximately the same as the endogenous chicken Cdk1 (Fig. S1 D). Our approach proved to be successful, and we were able to fully reconstitute the loss of the endogenous Cdk1 with the transgene. Compared to conventional small molecule inhibitors, the advantage of this genetic approach lies primarily in its highly increased specificity, rapid action and reversibility (). Accordingly, 1NM-PP1 had no discernable effect on cells. In cells, on the other hand, the cellular Cdk1 activity was annulled within 10 minutes after inhibitor addition, as judged by dephosphorylation of Histone H3 and the Cdk1 substrate Cdc27 (). This study shows that DT40 cells are a useful tool for chemical genetic analysis of vertebrate Cdk1. This cell line is characterized by a very stable karyotype and high gene targeting frequencies, allowing the establishment of double and even triple mutants (). The transformed character of the cell line and the use of Cdk1 might be a potential drawback of this approach. However, Cdk functions in cell cycle control are highly conserved among different species and cell lines, and the S phase functions of Cdk1 described here are likely to be of general relevance. cells reveals redundant as well as specific roles of vertebrate Cdk1 in the control of DNA replication, centrosome duplication, and mitosis (). We show that the deletion of Cdk2 renders both the initiation of DNA replication and centrosome duplication dependent on Cdk1 ( and ). This suggests that Cdk1 and Cdk2 share an essential function in the control of S phase. Our findings also suggest that the centrosome cycle and the cell cycle can be uncoupled simply by blocking entry into mitosis, and that either Cdk1 or 2 activity is required to drive this cell cycle–independent centrosome amplification. In principle, the experiment in shows that mid S phase cells do not need Cdk1/2 activity to continue replication. cells. A detailed analysis of late origin firing and the dynamics of replication elongation will be necessary to determine the roles of Cdk1/2 during ongoing replication. The functional overlap of Cdk1 and Cdk2 does not include the mitotic functions of Cdk1, which cannot be compensated by Cdk2. We also found that neither Cdk1 nor Cdk2 appear to control the events of early G1 phase such as initiation of de novo cyclin synthesis (), which is likely to be triggered by Cdk4/6. Moreover, we cannot exclude at this point that either kinase carries out other specific functions that evade our phenotypic analysis. It remains to be addressed, which of the different cyclins is primarily responsible for the S phase functions of Cdk1. A-type and B-type cyclins, the predominant binding partners of Cdk1as in the DT40 cells (Fig. S3 E) can both initiate DNA replication (; ). Cyclin A is the more likely candidate for this function, because cyclin B/Cdk1 is rapidly exported from the nucleus during S phase () and kept inactive by the Wee1 kinase (). However, a recent study suggests that cyclin E is also capable of binding and activating Cdk1, especially after deletion of Cdk2 (). The precise functions of the individual cyclins during S phase need to be addressed in future studies. Our study points to fundamentally different mechanisms in the control of endoreplication before and after mitotic Cdk1 activation and entry into M phase. The three main players of licensing control in vertebrate appear to be Geminin and Cdk1 (and possibly Cdk2), as well as proteolysis pathways such as the degradation of Geminin and Cdt1. In the following section we will discuss our findings on the function of these players comparing G2 to M phase. A summary of this discussion is presented in . Our results in and show that removal of Geminin is a necessity for origin licensing both in G2 and M phase, and that Geminin can act independently of Cdk activity. This idea is not supported by a previous study (), where Roscovitine treatment does not cause APC/C activation and Geminin degradation in mitotic HeLa cells. In contrast, we observed in consistence with previous studies () that the APC/C was rapidly activated upon mitotic Cdk1 inactivation causing the degradation of Geminin as well as other APC/C substrates (). We also confirmed that a degradation resistant Geminin mutant inhibited origin licensing, even after Cdk1 inactivation (). The discrepancy between our and Ballabeni et al.'s results may be due to the twofold lower doses of Roscovitine used in the previous study, which may trigger Geminin degradation only partially. We found that specific inhibition of both Cdk1 and Cdk2 initiated neither origin licensing nor DNA re-replication during G2 phase in DT40 (). The same result was obtained from Roscovitine-treated HeLa cells (). Paradoxically, Cdk1 inhibition in prometaphase rapidly triggered origin licensing and endoreplication (, B and G, lane 2 and lane 5; and ). Accordingly, Cdk1 inhibition during G2 phase does not lead to APC/C activation and Geminin destruction, whereas in M phase Cdk1 inactivation results in the rapid removal of Geminin and other APC/C substrates (). The presence of the APC/C inhibitor Emi1 during G2 but not during M phase (; ) is a likely explanation for this difference in APC/C activity between the two different cell cycle phases. This idea is supported by a recent report by , which demonstrates that depletion of the APC/C inhibitor Emi1 is sufficient to induce endoreplication in human cells. In contrast to our results, the APC/C appears to be activated during G2 phase in Cdk1 depleted HT2-19 human cells and cells (; ). This could explain the observed endoreplication in these cells after Cdk1 inactivation. The premature APC/C activation in G2 might be an effect of incomplete Cdk1 inactivation, which could allow Emi1 degradation, while not being sufficient to trigger mitosis. Alternatively, Emi1 levels might differ among different cell lines. When we analyzed the redundant roles of Geminin and Cdks in the control of origin licensing in the G2 phase, we made the surprising observation that Cdk inhibition did not trigger Mcm loading onto chromatin, even after Geminin depletion. This suggests the presence of an additional control mechanism that suppresses endoreplication in the G2 phase. Accordingly, we found that the licensing factor Cdt1 did not accumulate during G2 phase even after Cdk1 inhibition (, lane 1; and ). Conversely, Cdt1 appears to accumulate once cells enter M phase (). Accordingly, we found that transient inhibition of proteolysis stabilized Cdt1, and was sufficient to induce a new round of DNA replication in Geminin-depleted cells. This MG132 induced endo-replication occurred regardless of the state of Cdk activity, suggesting that Geminin and proteolysis are the major control mechanisms that block endoreplication in G2 phase. The question remains what the essential targets of these proteolysis pathways are. Degradation of the licensing factor Cdt1 is an obvious candidate (), but other players such as orc1 (), and the degradation of Cdk inhibitors such as p21 and p27 might also be involved (). Defining these essential proteolysis targets for licensing control in G2 phase will be an important challenge for future studies. Collectively, a model emerges in which differential proteolysis of licensing factors and licensing inhibitors control origin licensing during G2 and M phase (). In G2 phase the licensing inhibitor Geminin is stable, while the licensing factor Cdt1 is degraded. Once cells enter mitosis Cdt1 is stabilized and Geminin degradation by the APC/C is initiated upon induction of anaphase. In this way multiple mechanisms ensure in human cells that chromosomes are not replicated before the sister chromatids are separated. A schematic overview of the targeting construct is shown in Fig. S1 A. The upstream arm was amplified using 5′-AACGCGTAACTAGGACGGCTCCCGAGCAGG-3′ and 5′-GAAACGCACATAGCAAATACCAGTCTCAGG-3′ and cloned into pBS via SacI and BamHI. The downstream arm was amplified using the primers 5′-TCCTAAACTGCTTGTGAAGAAATAAGCAGG-3′ and 5′-CCTGCTTATTTCTTCACAAGCAGTTTAGGA-3′ cloned into pBS via EcoRI and SalI. BSR or Neomycin selection markers were cloned into the BamHI site of the construct. Gene targeting of DT40 cells was performed as described previously (). The Neomycin resistance cassette was flanked by two loxP sequences, and was removed from the chicken genome by transient expression of the Cre-recombinase. Neomycin sensitive cells were isolated by subcloning. Cdk1myc was cloned into pIRES2-EGFP (CLONTECH Laboratories, Inc.) via BamHI–EcoRI. Mutagenesis was performed using a standard PCR protocol. Stable integration of this vector in the DT40 genome was achieved by electroporating 20 μg of the linearized Cdk1 expression vector into DT40 cells, and selection of Neomycin resistant colonies as described previously (). The chicken cDNA for Cdk2 was cloned by RT-PCR using the primers 5′-ATGGAGAACTTTCAAAAGGTGGAGA-3′ and 5′-GGCTGTCCCCCACCTGCGCCTGTGA-3′, and the sequence was submitted to the NCBI GenBank (accession number ). A schematic overview of the gene disruption construct is shown in Fig. S3 A. There is no genomic information available for the chicken gene, and we were not able to amplify its genomic sequences by PCR. We screened a chicken genomic DNA pool (provided by RZPD Deutsches Ressourcenzentrum für Genomforschung GmbH) for a gene containing fragment using PCR with chicken cDNA primers 5′-GTCGTGTACAAGGCCCGGAACAAGGTCACG-3′ and 5′-CGCTGCCTTGGCCGAGATGCGCTTGTTGGG-3′, and analyzed isolated clones by Southern blot analysis with a chicken Cdk2 cDNA probe. We generated the targeting constructs by restriction enzyme digestion of the isolated cosmid clone. A BglII fragment was isolated upstream of exon 2 for the 5′ arm, and an EcoRV NotI fragment between exon 5 and 6 was isolated for the 3′ arm. The fragments were cloned into pCR 2.1-TOPO vector (Invitrogen). Selection cassettes (either His or Puro flanked with two LoxP sequences) were inserted into the construct at EcoRI–EcoRV sites by blunt-end ligation. The following primers were used to detect Cdk1 and 2 orthologues by RT-PCR: Cdk1 (5′-ATGAAGAAAATTCGATTGGAAAACG-3′ and 5′-GATCTCCGAGGAGGACCTGAACTAA-3′), Cdk1 (5′-ATGGAGGATTACACGAAGATAGAGAAGATT-3′ and 5′-CTTCCTGCTAATCTGATTAAGAAATTCTAA-3′), and Cdk2 (5′-TACCTGTTCCAGCTGCTGCAAGGCC-3′ and 5′-GGCTCAAATGCTGCACTACGATCCC-3′). Human Geminin was cloned by RT-PCR and the destruction box (amino acid residues 23–30) was deleted as described by . Mouse cyclin B1 destruction box mutant (R42A; L45A) in pEVT7 was a gift from Stephan Geley (Medical University Innsbruck, Austria). DT40 cells were cultured, and transfected as described previously (). described the synchronization of DT40 cell by elutriation. 1NM-PP1 was obtained from Cellular Genomics and used at the indicated concentrations (1–10 μM). We used a 50-μM concentration of Roscovitine (Calbiochem), and a 5-μM concentration of MG-132 (Calbiochem). Cell cycle analysis of BrdU-pulsed and PI-stained samples was performed on a FACScan (Becton Dickinson) using CellQuest software (. HeLa cells were synchronized in mitosis by incubation in 100 ng/ml Nocodazole, and in S phase by a double thymidine block. In brief, cells were incubated for 14 h in 2 mM thymidine, washed in PBS, released for 10 h, and then blocked for another 14 h in 2 mM thymidine. Transient transfection of HeLa cells by Lipofectamine (Invitrogen) was performed following the second thymidine block as described in the manufacturer's protocol. To target Geminin we transfected a QIAGEN-validated siRNA SI02653805 at a final concentration of 100 nM by oligofectamine (Invitrogen), following the manufacturer's instructions. For total cell extraction, 10 DT40 cells or 10 HeLa cells were lysed in 10 μl ECB buffer (50 mM Tris, pH 7.5, 120 mM NaCl, 0.5% NP-40, and 1 mM EDTA) containing 1 mM DTT and a 1:100 dilution of a protease inhibitor cocktail (Nakalai Tesque 25955-11). The extracts were incubated for 20 min on ice, sonicated, and suspended in 2× Laemmli buffer. For chromatin fractionation, we separated soluble and insoluble fractions as described by . 10 DT40 or 10 HeLa cells were first lysed for 20 min in 10 μl CSK buffer containing DTT and protease inhibitors as above. Following centrifugation, the supernatant was mixed with 2× Laemmili buffer, and the pellet was washed and resuspended in 10 μl CSK buffer, sonicated and suspended in 2× Laemmli buffer. Immunoblotting and immunoprecipitation, and histone H1 kinase assays were done as described earlier (; ). Immunofluorescence of cyclin B was performed with formaldehyde fixed samples that were spun on a glass slide by centrifugation in a Cytospin centrifuge. α- and γ-tubulin immunofluorescence was performed as described earlier (). All images were taken with an Olympus BX61 microscope, equipped with a Photometrics CoolSnap HQ camera, and Olympus Uplan/APO 100× lens (NA 1.35). Samples were mounted in Vectashield mounting medium with DAPI (Vector Laboratories, Inc.) and analyzed at room temperature. Image acquisition was performed using MetaMorph software. Antibodies used in this study comprised the anti Cdk1 mouse monoclonal A17 antibody (), rabbit polyclonal anti Cdk1 antibody from Upstate Biotechnologies, rabbit polyclonal anti Cdk2 antibody from Abcam (ab-7954-1), anti-PSTAIRE (), and rabbit polyclonal anti-phosphotyrosine15 Cdk1 antibody from Cell Signaling (Nr. 9111S).. Anti-chicken cyclin A () and anti-chicken cyclin B2 () antibodies were gifts from E. Nigg's laboratory (MPI of Biochemistry, Munich, Germany); monoclonal anti-α-tubulin FITC conjugate (No. F2168), and γ-tubulin polyclonal (No. T3559) antibodies were obtained from Sigma-Aldrich; Mcm2, 3, 4, and Histone H2A polyclonal antibodies were obtained from Abcam (ab 4461-50, 4460-50, 3728-50, 13923-100, respectively); and anti–human Cdt1 antibody was a gift from H. Nishitani (Kyushu University, Fukuoka, Japan; ). The Cdt1 antibody used in was purchased from Santa Cruz Biotechnology, Inc. (sc-28262). Aurora kinase A rabbit polyclonal antibodies were purchased from Abcam (ab12875), and to detect Cdc20 we used a mixture of two monoclonal anti-cdc20 antibodies from Santa Cruz Biotechnology, Inc. (SC1907 and SC1906). Human cyclin B1 monoclonal V152 antibodies (ab-72) and rabbit polyclonal anti-Geminin antibodies (ab-12147-50) were obtained from Abcam. For FACS analysis, ethanol-fixed cells were stained with anti-BrdU monoclonal antibody (BD Biosciences; Nr. 555627) and anti-phospho Ser10 rabbit polyclonal Histone H3 antibody (Upstate Biotechnology; Nr. 06-570). Alexa-labeled secondary antibodies were purchased from Molecular Probes, and HRP-labeled secondary antibodies were from Santa Cruz Biotechnology. Cdc27 was detected by a monoclonal antibody from Abcam (ab10538). Fig. S1 shows generation of cells. Fig. S2 shows synchronous mitosis after release from G2 arrest. Fig. S3 shows gene targeting of CDK2 in cells. Fig. cells.

The mitotic spindle is composed of microtubules emanating from spindle poles and has pivotal roles in high fidelity chromosome segregation during mitosis (; ; ). Kinetochores are initially captured by the lateral surface of spindle microtubules and are transported toward spindle poles during prometaphase (; ). Subsequently, sister kinetochores are captured by microtubules extending from the opposite spindle poles (called sister kinetochore biorientation or amphitelic attachment) before anaphase onset (). Poleward kinetochore transport during prometaphase is especially crucial when kinetochores are located at a distance from the mitotic spindle. In the budding yeast , centromeres are tethered to spindle poles by microtubules during most of the cell cycle (; ). Nonetheless, centromeres are released from and recaptured by microtubules during a brief period in S phase, probably as a result of kinetochore disassembly and reassembly upon centromere DNA replication (; ; our unpublished data). We have visualized this process in S phase using electron microscopy and live cell fluorescence microscopy (). Moreover, to analyze individual kinetochore–microtubule interaction with higher resolution, we displaced a selected centromere () from the spindle and other centromeres by conditionally inactivating it (). Then, during metaphase arrest, we reactivated , which was subsequently captured by the lateral surface of single microtubules (centromere reactivation system; ). Using this system, we found that Kar3, a kinesin-14 family member and microtubule minus end–directed motor, is involved in poleward kinetochore transport along the lateral surface of microtubules (kinetochore sliding; ). For instance, in the rigor mutant that can bind microtubules but does not have motor activity as a result of an ATP hydrolysis defect (; ), was captured by microtubules but frequently was not transported (). In contrast, overexpression accelerated transport along microtubules (). However, it has remained unclear whether Kar3 localizes at kinetochores and directly drives sliding along microtubules. Moreover, although Kar3 is apparently involved in kinetochore sliding, was still able to reach a spindle pole in the majority of -deleted cells (). This suggests the involvement of a redundant mechanism for kinetochore transport that still remains to be identified. The Dam1 complex, which is also called DASH or DDD and is composed of at least 10 proteins, has important roles in ensuring proper kinetochore–microtubule interaction (; ; ). However, in our reactivation system, mutants of the Dam1 complex components did not show substantial defects in capture by microtubules or in the subsequent sliding of along microtubules (). It has been recently reported that several Dam1 complexes could gather together and form rings encircling microtubules in vitro (; ). It is still unclear whether this is the case in vivo or how the complex regulates kinetochore–microtubule interaction. Here, we studied mechanisms of kinetochore transport by microtubules using our centromere reactivation system () as well as in normal cell cycles (i.e., without cell cycle arrest or regulation of centromere activity). We show that poleward movement of kinetochores can occur in two distinct ways: lateral sliding, in which kinetochores move along the side of a microtubule, and end-on pulling, in which the kinetochore is attached to the end of a microtubule and is pulled poleward as the microtubule shrinks. Our study reveals how Kar3 and the Dam1 complex regulate these processes. To visualize individual kinetochore–microtubule interactions at high resolution in , we regulated the activity of a particular centromere (). is displaced from the spindle and other centromeres by conditional inactivation using transcription from the adjacently inserted promoter (; ). Then, during metaphase arrest by Cdc20 depletion, we reactivated by turning off the promoter. , which was marked with CFP or GFP, was captured by the lateral surface of CFP- or YFP-labeled microtubules. To address whether Kar3 localizes at kinetochores and directly drives their transport along microtubules, we visualized Kar3 by fusing it with four tandem copies of GFP (Kar3-4GFP). After the reactivation, in most cells, Kar3-4GFP was visible at the CFP-labeled before its capture by CFP-labeled microtubules and also during its transport along microtubules (). Kar3 was also detected at the plus ends of growing microtubules (supplemental note 1; available at ) and at spindle poles as previously reported (; ). The amount of Kar3 at kinetochores appeared to decrease after sister kinetochore biorientation (supplemental note 1; ). Kar3 is involved in kinetochore transport toward spindle poles (). However, in the majority of cells, still reached spindle poles after being captured by microtubules (), suggesting the involvement of a redundant mechanism for kinetochore transport. To identify this mechanism, we analyzed poleward kinetochore transport in cells in greater detail. In 68% of wild-type cells, reached the spindle by sliding along the lateral surface of microtubules, whereas in cells, this occurred in only 11% of cells (, supplemental note 2, and Fig. S1 A, available at ). The sliding observed in cells might depend on other microtubule motors, but, as we discuss below, it probably depends on motor-independent one-dimensional diffusion (supplemental note 3). In many of the cells, after capture by the lateral sides of microtubules, was tethered at the distal ends of microtubules extending from spindle poles and was pulled poleward as the microtubules shrank (, microtubule end-on pulling of kinetochores; and Video 1). 81% of cells showed microtubule end-on pulling of , whereas this occurred in only 13% of wild-type cells (). Thus, in cells, kinetochores were transported mainly by sliding along microtubules, but, in the absence of Kar3, microtubule end-on pulling was the main mode for kinetochore transport. In cells, poleward sliding of occurred with frequent pausing, and the associated microtubules showed frequent rescue (conversion from shrinkage to growth; , left; ). In contrast, in both and cells, microtubule end-on pulling transported poleward on average more rapidly than sliding and without pausing ( [right] and D). During end-on pulling, microtubule rescue was rarely observed (supplemental note 4 and Fig. S1 B). In addition, occasionally (4.9%) detached from microtubules during sliding, whereas such detachment was never observed during end-on pulling (supplemental note 5). These differences suggest that the two modes of kinetochore transport are distinctly regulated processes. Nonetheless, sliding was sometimes converted to end-on pulling ( [pale green] and E; and Video 2, available at ), whereas the opposite conversion was seldom, if ever, observed. Having established that is transported poleward either by sliding or end-on pulling, we were in a position to evaluate the contribution of Kar3 specifically to sliding by excluding end-on pulling events from our analysis. To make an unbiased comparison between and cells, we studied movement for a short period, during which is associated with the microtubule lateral surface but not at the microtubule distal end after its initial capture by the microtubule lateral surface (supplemental note 6). During such a period, in cells, travelled preferentially poleward (, top left) except in a small number of cells (supplemental note 7). In contrast, in cells, moved in both directions apparently equally (, top right). Consequently, the mean displacement of from its original position (i.e., its position when initially captured by a microtubule) along a microtubule increased with time and was oriented toward a spindle pole in cells at a speed of 0.39 μm/min (, bottom left; and supplemental note 6), whereas it remained approximately zero in cells (, bottom right). Given a lack of preferential direction to motion in cells, we suspected that motility in this mutant might be caused by one-dimensional diffusion along a microtubule. To address this, we plotted the mean squared displacement (MSD) of along a microtubule against a change of time () in cells (; ; ). The MSD increased linearly as increased, which was indeed consistent with the one-dimensional diffusion of . The diffusion coefficient ( in = 2) of motility along microtubules was calculated as 0.11 μm/min. This result was consistent with our previous observations that single deletion mutants of other microtubule-dependent motor proteins (Cin8, Kip1, Kip2, Kip3, and Dyn1) had no effect on transport (). We also failed to observe a association of these other motor proteins during its transport (supplemental note 8 and Fig. S2, A and B; available at ). These data suggest that Kar3 is the main and probably the sole factor that drives poleward kinetochore sliding along microtubules. In most cells that we observed, was first captured by a microtubule lateral surface, but, when the microtubule subsequently shrank, its distal end encountered (note that the microtubule shrinkage rate exceeds the mean velocity of sliding; ; ) unless either the microtubule was rescued before this or reached a spindle pole by sliding. Immediately after the microtubule end encountered , we observed either of the following two events: microtubules were rescued, and remained on the microtubule lateral surface; or became tethered at the microtubule end and subsequently was transported poleward by end-on pulling (). In such encounters in wild-type cells, we observed microtubule rescue in 59% of cases and the establishment of end-on pulling in 41% of cases (). When microtubule plus ends reach after microtubule shrinkage, how do cells choose between these two options? We addressed whether Kar3 affects this choice because end-on pulling was more frequently observed when Kar3 was not functional (). For this, we evaluated how frequently each option was chosen when microtubule plus ends reached in cells (). We found that cells showed a more frequent establishment of end-on pulling than cells, suggesting that Kar3 can suppress the establishment of end-on pulling when microtubule plus ends reached (supplemental note 9 and Fig. S2 C). How does Kar3 partially suppress the establishment of the microtubule end-on pulling of ? We speculated that Kar3 might anchor to the microtubule lateral surface (close to microtubule plus ends) and that this might hinder from forming a specific attachment required for end-on pulling. To test this idea, we used the mutant, which can bind microtubule lateral surfaces but cannot work as a motor because of an ATP hydrolysis defect (; ; ). In contrast to cells, mutant cells showed a strong reduction in the frequency of establishing end-on pulling (, compare with ; and supplemental note 10). This result also explains why microtubule-dependent poleward transport was more severely delayed in than in (). The mutant cannot facilitate sliding along microtubules as a result of its defective motor activity yet considerably inhibits the establishment of the microtubule end-on pulling of in contrast to , thus causing a considerable delay in transport (supplemental notes 10 and 11). We next addressed which microtubule end underwent depolymerization during the microtubule end-on pulling of . We marked a microtubule region midway between (bound at the microtubule plus end) and a spindle pole by photobleaching the YFP-Tub1 signal (). The distance between the photobleached region and a spindle pole did not substantially change as was pulled poleward by microtubule shrinkage. Thus, depolymerization occurred at the plus end (i.e., bound to ) but not substantially at the minus end (i.e., at a spindle pole). This is consistent with other results indicating that microtubules are dynamic only at the plus ends in budding yeast (; ). Subsequently, we compared the microtubule depolymerization (shrinkage) rate in the presence of associated with the microtubule end (i.e., during end-on pulling) and in the absence of association with microtubules (supplemental note 12). The shrinkage rate was significantly slower during the end-on pulling (), suggesting that the presence of kinetochores at the microtubule plus ends somehow reduced the rate of microtubule depolymerization at those ends. In addition, because Kar3 and Kip3 (a kinesin-8 family member) facilitate microtubule depolymerization in vitro (; ; ; ; ) and foster microtubule disassembly in vivo (; ; ; ), we addressed whether these motor proteins affect the microtubule shrinkage rate (Fig. S3 A, available at ). Deletions of or and the double mutant did not significantly change the microtubule shrinkage rate either when was transported by microtubule end-on pulling or in the absence of association. Nonetheless, the mean length of nuclear microtubules was longer in and cells than in wild-type cells (unpublished data) as a result of a reduced frequency of microtubule catastrophe (conversion from growth to shrinkage) in these cells (unpublished data). We next studied what factors facilitate the microtubule end-on pulling of kinetochores. The Dam1 complex was a candidate for the following reasons: the Dam1 complex is not required for kinetochore sliding () but, nonetheless, becomes associated with kinetochores before sister kinetochores biorient in metaphase (; ; ); the complex forms a ring encircling microtubules in vitro (; ) and, if this is the case in vivo, it could tether kinetochores at microtubule plus ends while the complex is pushed poleward as microtubule protofilaments splay out during microtubule shrinkage (); and and mutants show a synergistic effect on chromosome loss (supplemental note 13), which could be explained if both factors are involved in microtubule-dependent kinetochore transport. To gain more clues as to the function of the Dam1 complex, we visualized the complex along microtubules by tagging both Dam1 and Ask1, two components of the complex, with four tandem copies of GFP. The Dam1 complex showed multiple punctate GFP signals along microtubules. Intriguingly, the intensity of the GFP signals at the distal ends of microtubules increased during microtubule shrinkage ( and Video 3, available at ). We also compared this increase when GFP signals were either present or absent along the extent of microtubule shrinkage (). In their presence, the GFP signal intensity at the microtubule tip increased more evidently. These results suggest that GFP spots along microtubules become collected by the microtubule ends. The simplest interpretation is that the Dam1 complexes form a ring encircling a microtubule in vivo and are collected by splaying protofilaments ( and see Discussion). Moreover, during the microtubule end-on pulling of , the Dam1 complex continuously colocalized with ( and Video 4). However, such continuous colocalization was not observed during the sliding of along microtubules ( and Video 5), which is consistent with the Dam1 complex not being required for sliding along microtubules (). We subsequently investigated the roles of the Dam1 complex in kinetochore transport by impairing the function of the complex. If the Dam1 complex facilitates the microtubule end-on pulling of kinetochores, then this, the main kinetochore transport mode in cells, would be defective when the Dam1 complex is impaired. To test this, we attempted to combine with temperature-sensitive mutants of the Dam1 complex components , , and . However, all of these combinations were lethal even at a permissive temperature for each temperature-sensitive mutant. Therefore, we used the temperature-sensitive mutant instead of (supplemental note 14). Only the double mutant was viable, whereas the other two combinations were lethal. We analyzed transport in and single mutants and in the double mutant at 35°C, which is a restrictive temperature for these mutant alleles. In the single mutant, was transported toward spindle poles by microtubule end-on pulling in the majority (74%) of cells (), which is consistent with the behavior of cells (). In the single mutant, lateral sliding was observed in 66% of the cells, similar to the wild type. However, the amount of successful end-on pulling was dramatically reduced to only 3%. The remainder of the cells ended up in a standstill state, with at the end of a microtubule that did not shrink any further (similar to the cell in ; supplemental note 15). This suggests that Dam1 is required for successful end-on pulling. This conclusion was supported by the behavior of the double mutant, in which virtually all cells ended up in the standstill state ( and Video 6, available at ). These results suggest that although Kar3 is required for lateral sliding, the Dam1 complex has an essential role in the microtubule end-on pulling of kinetochores. How could the Dam1 complex facilitate the microtubule end-on pulling of kinetochores? Given that the Dam1 complexes are able to form ring structures, it is likely that during microtubule depolymerization, the Dam1 complex has a key role in converting the splaying out of depolymerizing microtubule protofilaments into the kinetochore pulling force toward a spindle pole (see ), as recently suggested by in vitro studies (; ). A simple model would be that kinetochores are tethered to the Dam1 complex during this process. This model is supported by three pieces of evidence: in the Dam1 complex component mutant , 20% of cells showed the detachment of from microtubules (Fig. S3 B); once kinetochores were captured by microtubules, subsequent microtubule depolymerization with nocodazole did not abolish association of the Dam1 complex with kinetochores (); and the Dam1 complex directly associates with the kinetochore complex Ndc80 (). Nonetheless, only a few cells showed the detachment of from microtubules (Fig. S3 B). We suggest that any residual function of the mutant protein was sufficient to maintain -microtubule attachment. Consistent with this notion, another of the Dam1 complex components still localized at during the microtubule end-on attachment of in cells (supplemental note 16), and centromeres became detached from microtubules more frequently when was converted into a more defective mutant (supplemental note 16 and Fig. S4 A; available at ). To analyze microtubule-dependent kinetochore transport at high resolution, we have so far studied this process using centromere reactivation in metaphase-arrested cells (). Next, we wanted to study whether Kar3 and the Dam1 complex also have important roles in the transport of authentic centromeres (i.e., without regulation by an adjacent promoter) and in normal cell cycles (i.e., without cell cycle arrest). These results prompted us to compare the behavior of centromeres, which detached from microtubules in early S phase in wild-type, , and single mutant and double mutant cells. We labeled , , and microtubules by CFP, GFP, and YFP, respectively. Before cells started budding (i.e., before entry into S phase), both s localized close to a spindle pole in wild-type and cells but often somewhat more distant (>1.0 μm) from a pole while still attached to microtubules in and cells. Nonetheless, in all four kinds of cells, both s equally showed detachment from microtubules when cells started budding (i.e., during S phase; supplemental note 17) and were recaptured by microtubules extending from a spindle pole after a similar time interval (0.5–1.5 min). In both wild-type and , after s were recaptured by microtubules, most centromeres promptly (in 0.5–1.0 min) moved to the vicinity of a spindle pole (, pink). Microtubule end-on pulling was more frequently discerned in cells than in wild-type cells ( [red shaded areas] and B; and supplemental note 18), although in other cases, it was difficult in this experimental condition to distinguish transport by sliding along microtubules and by end-on pulling. In and , s were captured by microtubules but were not transported to the vicinity of a spindle pole in 53% and 93% of cells, respectively (, teal). In such cases, we often discerned that s attach at microtubule distal ends without being pulled toward a spindle pole ( [blue shaded area] and B; and supplemental note 18). We also measured the velocity of transport and found that it was distinctly altered by and mutations (supplemental note 19 and Fig. S4 B). Thus, in and single and double mutant cells, we found distinct defects in the microtubule-dependent transport of s during normal S phase, which are consistent with our results from the centromere reactivation system. Our data suggest that Kar3 and the Dam1 complex have important roles in the microtubule-dependent poleward transport of authentic centromeres during the normal cell cycle (supplemental notes 20 and 21; Fig. S4 C). We have investigated the mechanisms by which kinetochores are transported toward spindle poles by microtubules in budding yeast. We show that poleward movement can occur in two distinct ways: lateral sliding, in which kinetochores move along the side of a microtubule, and end-on pulling, in which the kinetochore is attached to the end of a microtubule and is pulled poleward as the microtubule shrinks. Kar3 is essential to drive poleward lateral sliding, whereas the Dam1 complex is crucial for end-on pulling (). It is thought that upon centromere DNA replication in budding yeast, kinetochores are disassembled, causing the release of centromeres from microtubules for a short period (; ; our unpublished data). Soon afterward, kinetochores are reassembled and captured by the lateral sides of microtubules extending from spindle poles (). Microtubule lateral surfaces can secure initial kinetochore capture by providing a much larger contact surface compared with microtubule tips. The amount of Kar3 loaded at kinetochores probably increases while kinetochores are detached from microtubules (supplemental note 1; ). Kar3 (and the Dam1 complex) is not required for the initial kinetochore capture by microtubules (supplemental note 22; ), but, once kinetochores are captured, the motor activity of Kar3 drives kinetochore sliding along microtubule lateral surfaces toward spindle poles. Kar3 is the main and probably the sole factor driving this kinetochore sliding because in the absence of Kar3, kinetochores only show one-dimensional diffusion along microtubules. Kinetochore sliding, which is promoted by Kar3, occurs toward a spindle pole with frequent pausing. This is perhaps because the Kar3 molecules loaded on kinetochores do not persistently drive their sliding along microtubules. This is a similar situation to that of Ncd, a putative Kar3 orthologue in shown to be a nonprocessive motor, whose domain is released from microtubules after each ATPase cycle and must bind microtubules repeatedly to drive motion (). Probably because of this frequent pausing, the mean velocity of kinetochore sliding is lower than that of microtubule shrinkage (). Therefore, microtubule plus ends often reach kinetochores unless microtubules are rescued and regrow before this happens, a process involving Stu2 transport from kinetochores (). If microtubule plus ends reach kinetochores, cells must choose one of the following two options: microtubules show regrowth (i.e., are rescued), probably facilitated by Stu2 and other factors loaded at kinetochores (), or kinetochores are tethered at microtubule plus ends, probably as a result of association with the Dam1 complex ring structure, which has been pulled poleward as microtubules shrink. Kar3 reduces the frequency of the second choice (i.e., establishment of microtubule end-on pulling), probably by anchoring kinetochores to the microtubule lateral surface. In addition, the establishment of microtubule end-on pulling seems to be partly affected by stochastic elements (supplemental note 9). In any case, in the first option, kinetochores still remain associated with the lateral surface of microtubules and continue to slide poleward along microtubules. In the second, kinetochores at microtubule ends are continuously pulled poleward (end-on pulling) as the attached microtubules shrink without pausing or rescue. It is currently unclear how microtubule rescue is suppressed during end-on pulling, but it is not solely caused by a lack of Stu2 loaded on kinetochores (supplemental note 4). The microtubule end-on pulling of kinetochores is facilitated by the Dam1 complex. In vitro experiments suggested that several Dam1 complexes could gather together and form a ring structure encircling a microtubule, which could move along the microtubule (; , ; ). We found that the Dam1 complexes along a microtubule were collected at the plus ends of depolymerizing microtubules in vivo (supplemental note 23). The simplest interpretation would be that the Dam1 complexes indeed form a ring encircling a microtubule in vivo, which is pushed poleward by splaying protofilaments as the microtubule depolymerizes. However, it cannot be completely ruled out that the Dam1 complexes do not form a ring in vivo () and that unknown mechanisms collect the complexes along a microtubule at its plus end during microtubule shrinkage. In any case, when Dam1 function was impaired, the microtubule end-on pulling of kinetochores became defective. Our in vivo data support the model () in which the Dam1 complex tethers kinetochores and plays a crucial role in converting microtubule depolymerization to kinetochore pulling force as initially proposed from the in vitro experiments. Kinetochore sliding and microtubule end-on pulling are two distinct modes of microtubule-dependent kinetochore transport and seem to use different energy sources to produce the force necessary for kinetochore transport. When microtubules polymerize, the curvature of GDP-bound tubulin dimers is constrained by microtubule geometry so that the polymer lattice stores energy from GTP hydrolysis (). During end-on pulling, the free energy is released and converted to kinetochore pulling force by a power-stroke mechanism as microtubule protofilaments change from a straight to curved form (). The Dam1 complex apparently has an important role in this conversion (). In contrast, kinetochore sliding is driven by the Kar3 motor activity that is dependent on its ATP hydrolysis (i.e., additional energy is consumed; ). In spite of this, kinetochore sliding is less processive and achieves less efficient kinetochore transport. Given these disadvantages, why do cells still use kinetochore sliding for their transport? Kinetochore sliding may have the following merits compared with end-on pulling: for the establishment of microtubule end-on pulling, kinetochores must wait until the associated microtubule shrinks and the microtubule plus end finally reaches the kinetochore. Therefore, depending on the situation, kinetochores may reach a spindle pole earlier by sliding than by end-on pulling. A single microtubule plus end is probably able to attach to only a single kinetochore during end-on pulling (), but, in contrast, multiple kinetochores could be transported simultaneously by sliding (supplemental note 24). Microtubule rescue, which happens during kinetochore sliding but not during end-on pulling, would increase the chance that kinetochores further afield are also captured by the same microtubule (supplemental note 25). Because kinetochore sliding is converted into end-on pulling but not vice versa, the population of kinetochores attached to microtubule plus ends increases during poleward kinetochore transport. Both sister kinetochores subsequently interact with microtubules, and the Ipl1 kinase promotes the reorientation of kinetochore–microtubule attachment (, ), in which the Dam1 complex is a crucial substrate of the kinase (). Because this reorientation happens in a tension-dependent manner (; ), sister kinetochores eventually attach to microtubules from opposite spindle poles (biorientation). To establish biorientation efficiently, kinetochores must be located within the spindle where microtubules extend from both spindle poles at high density. Because microtubule-dependent transport brings kinetochores close to the spindle, this process should facilitate efficient sister kinetochore biorientation. The stable maintenance of biorientation crucially requires Dam1 complex function (). Presumably, in metaphase, the Dam1 complex is necessary to pull sister kinetochores toward opposite spindle poles, generating tension across sister kinetochores and, in turn, stabilizing kinetochore–microtubule attachment (), thus simultaneously avoiding breakage of the attachment when this tension is applied (supplemental note 26). Metaphase is followed by anaphase A (), in which the kinetochore–spindle pole distance is shortened. We envisage that the Dam1 complex also plays the same role in anaphase A, as we found in prometaphase (i.e., tethering kinetochores at the microtubule plus ends and converting microtubule depolymerization [occurring at kinetochore sides; ] into kinetochore pulling force). Consistent with this notion, we found that the Dam1 complex colocalizes with kinetochores during anaphase A (supplemental note 27; unpublished data). Recently, the Dam1 complex orthologue was identified in fission yeast (; ). In this organism, the Dam1 complex has important roles in sister kinetochore biorientation and kinetochore congression to the spindle midzone, which is consistent with Dam1 complex function in budding yeast. Moreover, kinetochores are still transported poleward in the absence of all of the known microtubule minus end–directed motors (i.e., two kinesin-14s and dynein) in fission yeast (); thus, perhaps kinetochores are transported by end-on pulling in this organism, as we have shown directly here in budding yeast. In vertebrate cells, kinetochores are also captured by the lateral sides of single microtubules and are transported toward spindle poles in prometaphase (). How is the kinetochore transport regulated in vertebrate cells? In contrast to mechanisms in budding yeast (supplemental note 8), vertebrate dynein could be involved in fast and processive kinetochore sliding along microtubules (supplemental note 28; ; ). If this is the case, the depletion of dynein may reveal the microtubule end-on pulling of kinetochores as a possible redundant mechanism for kinetochore transport in vertebrate cells, just as it was revealed by in yeast. Although convincing orthologues of the Dam1 complex components have not yet been identified in vertebrate cells (), functional counterparts of the Dam1 complex may have an important role in microtubule end-on pulling. Kinesin-13s (mitotic centromere-associated kinesin, etc.) may be such functional counterparts because they also form rings encircling single microtubules in vitro (; ), localize at kinetochores in mitosis (), and act as important substrates of the aurora B kinase in ensuring proper kinetochore–microtubule attachment (; ; ; ). Kinetochore capture and transport by spindle microtubules is the first crucial step for proper chromosome segregation in all eukaryotic cells. Comparison of kinetochore transport between different organisms will uncover the evolution of regulatory mechanisms for this fundamental cellular process. The background of yeast strains (W303), methods for yeast culture, and the centromere reactivation system were described previously (; ; ). All tagging of yeast genes were performed at their C termini at their original gene loci except for the tagging of and constructs were integrated at an auxotroph marker locus. Mutant alleles of (), (), (), and () were reported previously. Cells were cultured at 25°C in YP medium containing glucose unless otherwise stated. For more information, see supplemental note 29. The procedures for time-lapse fluorescence microscopy were described previously (; ). Time-lapse images were collected every 15 s for 30 min at 23°C (ambient temperature) unless otherwise stated. For image acquisition, we used a microscope (DeltaVision RT; Applied Precision), a UPlanSApo 100× NA 1.40 objective lens (Olympus), a CCD camera (CoolSnap HQ; Photometrics), and SoftWoRx software (Applied Precision). We acquired three to seven (0.7 μm apart) z sections, which were subsequently deconvoluted, projected to two-dimensional images, and analyzed with SoftWoRx and Volocity (Improvision) software. CFP signals were discriminated from either YFP or GFP signals using the JP4 filter set (Chroma Technology Corp.). YFP signals were discriminated from CFP and GFP signals using the JP3 filter set (Chroma Technology Corp.). To collect GFP and YFP signals together without distinguishing the two, the YFP channel of the JP4 filter set was used. To evaluate the length of microtubules and position of centromeres, we took account of the distance along the z axis as well as the distance on a projected image. To score modes of kinetochore transport in the reactivation system ( and ), we selected cells in which was captured by microtubules (–spindle pole distance was 2 μm or longer at the initial capture) and subsequently reached spindle poles during a time-lapse observation of 30-min duration. Cells were scored for sliding when was transported along microtubules for 1 μm or longer (in most cases toward a spindle pole) and tubulin signal intensity was similar distal and proximal to (i.e., was not transported with end-on pulling by shorter overlapping microtubules; supplemental note 2). Cells were scored for microtubule end-on pulling when was transported poleward for 1 μm or longer with attached to the microtubule distal end. Standstill was scored when was almost at the same position on microtubules for considerable time (see supplemental note 30 for details). The rate of microtubule growth and shrinkage was evaluated only for approximate linear changes (R > 0.85 in linear regression analyses) of microtubule lengths >3 μm. MSD was calculated as described in . Statistical analyses were performed with the Fisher's exact test (, , and S4 A) or with an unpaired test (all others) using Prism software (GraphPad) unless otherwise stated. All p-values are two tailed. All error bars in figures represent SEM. For more information, see supplemental note 30. Supplemental notes 1–32 describe more results, discussions, and methods. Fig. S1 A shows kinetochore transport in cells in which microtubule plus ends are labeled, and B shows Stu2 localization at during microtubule end-on pulling. Fig. S2 A shows the localization of motor proteins Cin8, Kip1, Kip2, Kip3, and Dyn1, B shows kinetochore transport in and double mutants, and C shows Kar3 localization during microtubule end-on pulling. Fig. S3 A shows that Kar3 and Kip3 do not significantly affect the microtubule shrinkage rate, and B shows defects in kinetochore capture and transport in , , and mutants. Fig. S4 A shows that kinetochores sometimes detach from microtubules in , B shows transport velocity in normal S phase, and C shows Kar3 localization at kinetochores in normal S phase. Video 1 shows a cell showing the microtubule end-on pulling of (video of the cell shown in B). Video 2 shows the conversion of sliding along a microtubule into microtubule end-on pulling (video of the cell shown in E). Video 3 shows the Dam1 complexes accumulate at a microtubule plus end as it depolymerizes (video of the cell shown in B). Video 4 shows that the Dam1 complexes continuously colocalize with during microtubule end-on pulling (video of the cell shown in A). Video 5 shows that the Dam1 complexes do not continuously colocalize with during sliding along a microtubule (video of the cell shown in B). Video 6 presents a cell showing the standstill of attached at the plus end of a microtubule (video of the cell shown in B). Online supplemental material is available at .

Defects in the attachment of microtubules to kinetochores activate the spindle checkpoint to delay mitotic progression by transiently inhibiting the anaphase-promoting complex (also called the cyclosome) (). Genes involved in the spindle checkpoint were first isolated from and include , , and (mitotic arrest–deficient) (); , , and (budding uninhibited by benzimidazole [a microtubule-depolymerizing drug]) (); and (monopolar spindle) (). The spindle checkpoint proteins and their functions are highly conserved between yeast and humans, and defects in the spindle checkpoint result in substantial aneuploidy (; ). Much evidence also indicates a role of the spindle checkpoint in tumorigenesis, e.g., mutations in human homologues of Bub1 (BUB1 and BUBR1) have been found in subtypes of colorectal cancer cells that exhibit chromosome instability (CIN) (). The CIN phenotype has been associated with mutations in spindle checkpoint genes (; ; ), decreased levels of spindle checkpoint proteins (; ), and loss of spindle checkpoint activity (; ). mice frequently develop lung tumors after a long latency (). mice and / heterozygotes are prone to tumor development (; ). These results strongly suggest a close relation between altered activity of the spindle checkpoint and tumorigenesis. Also, many tumor cells have a diminished, but not absent, spindle checkpoint response (). When the function of mouse Bub1 is compromised cells appear to escape apoptosis and continue to progress through the cell cycle, despite leaving mitosis with an altered spindle (). However, opposing evidence indicates that the spindle checkpoint regulates apoptosis: mutations in cause chromosome missegregation and fail to block apoptosis in (), and null mouse embryos undergo apoptosis at embryonic day (E) 6.5 to E7.5 (). In all of these cases, apoptosis appeared to occur in the subsequent G1 phase; thus, the role of the spindle checkpoint in apoptosis remained unclear. When cells cannot satisfy the spindle checkpoint after a long mitotic delay, several cell fates can occur: some cells die during mitosis, some exit mitosis but die via apoptosis in the G1 phase, and some exit mitosis but are tetraploid and reproductively dead (). Microtubule inhibitors induce mitotic arrest by activating the spindle checkpoint; eventually, these inhibitors cause cytotoxicity. The cytotoxicity of microtubule inhibitors and resultant cell death has been described as either apoptosis in G1 or reproductive death (). However, questions about cell death during mitosis have remained. Although much evidence suggests that apoptosis occurs during mitosis (; ; ; ; ), in-depth analyses of mitotic cell death have not been performed; therefore, the mechanism involved remains obscure, especially the relation between the spindle checkpoint and cell death during mitosis. Here, we report the mechanism of the programmed cell death in early mitosis that is induced by defects in the kinetochore–microtubule attachment in BUB1-deficient cells. Treatment with nocodazole (a microtubule-depolymerizing drug), paclitaxel (Taxol, a microtubule-stabilizing drug) (), or 17-allylaminogeldanamycin (17-AAG, an HSP90 inhibitor that induces delocalization of several kinetochore proteins from kinetochores) () caused substantial mitotic delay (). We depleted HeLa cells of either BUB1 or MAD2 (both are spindle checkpoint components) by treating the cells with synthetic small interfering RNA (siRNA) (). The cells were then incubated in nocodazole, paclitaxel, or 17-AAG to induce mitotic arrest. The depletion of MAD2 (but not BUB1) substantially diminished the arrest (). These results are consistent with the finding by , i.e., depletion of BUB1 does not compromise the mitotic delay either during normal mitosis or in response to spindle damage induced by nocodazole (). Although a small remaining quantity of BUB1 may be sufficient to induce mitotic delay (), ∼90% depletion of BUB1 did not affect mitotic delay induced by defects in kinetochore–microtubule attachment. Synthetic lethality occurs between spindle checkpoint mutants and kinetochore mutants in yeast, presumably because of synergistic, substantial chromosome loss (; ). In human cells, simultaneous depletion of the kinetochore protein HEC1 and MAD2 causes premature catastrophic exit from mitosis (). Because 17-AAG causes kinetochore defects (), we examined whether 17-AAG treatment in conjunction with defects in the spindle checkpoint induces synthetic lethality. MAD2 or BUB1 was depleted from HeLa cells by siRNA treatment. The cells were then exposed to 100 nM 17-AAG, which did not kill most control cells, and viability was evaluated by a colony outgrowth assay (, top). The proportion of cells killed by MAD2 siRNA or BUB1 siRNA alone did not differ from that seen with control luciferase siRNA, but treatment with either MAD2 siRNA or BUB1 siRNA and 17-AAG caused substantial synergistic lethality (). Spindle checkpoint mutants in yeast are also sensitive to microtubule inhibitors (Hoyt et al., 1991; ), presumably because of synergistic substantial chromosome loss. Therefore, we examined whether paclitaxel treatment, in conjunction with defects in the spindle checkpoint, induces synthetic lethality. Like 17-AAG, paclitaxel with either MAD2 siRNA or BUB1 siRNA caused substantial synergistic lethality (). Death induced by MAD2 siRNA and 17-AAG or paclitaxel is presumably caused by the failure of checkpoint-induced mitotic arrest, which results in premature mitotic exit and synergistic aneuploidy (). The resulting abnormal nuclei (i.e., fragmented/aggregated nuclei, micronuclei, or chromosome bridges; ) are similar to those of cells that are MAD2 depleted for several cell divisions (); this lethal phenotype can be explained by premature mitotic exit, i.e., the current understanding of how the spindle checkpoint protects cells from aneuploidy () (). Moreover, the abnormal nuclear phenotypes are associated with the degree of MAD2 depletion (). However, BUB1-depleted cells treated with 17-AAG or paclitaxel did not appear to exit mitosis (). This finding raises a provoking question: How does simultaneous treatment with BUB1 siRNA and 17-AAG or paclitaxel cause substantial synergistic lethality, when BUB1 depletion does not cause premature mitotic exit? The spindle checkpoint appeared functional (). To evaluate the lethal phenotype caused by 17-AAG and either MAD2 siRNA or BUB1 siRNA, we used the TUNEL assay. When cells were treated simultaneously with 17-AAG and BUB1 siRNA, but not with MAD2 siRNA, most of the TUNEL cells were in prophase, prometaphase, or metaphase (). Furthermore, agarose gel electrophoresis directly detected DNA fragmentation in the mitotic cells (). To exclude the possibility of off-targets of siRNA, we used several siRNA oligos to induce DNA fragmentation (unpublished data; see Fig. S4 A, available at ). Overexpression of BUB1 suppressed DNA fragmentation when siRNA targeted the 3′ UTR region of BUB1 (Fig. S1, A and B). Although the BUB1-depleted cells appeared to be arrested in mitosis, they must have been dead or dying, because the DNA had already fragmented during early mitosis (). This finding answers the question posed above. Cells treated with BUB1 siRNA and either nocodazole or paclitaxel underwent mitotic cell death (; Fig. S1 C). Interestingly, ∼90% of the mitotic cells were TUNEL (). These drugs commonly cause defective kinetochore–microtubule attachment. Therefore, these results strongly suggest that this mitotic cell death occurs when the kinetochore–microtubule attachment is altered and BUB1 function is disrupted. We detected no caspase activity (caspases 1, 3–9) in cells exposed to 17-AAG and BUB1 siRNA (). Furthermore, caspase inhibitors BAF and zVAD did not inhibit DNA fragmentation induced by 17-AAG and BUB1 siRNA (, see Fig. S1 D for drug evaluation controls). Therefore, this mitotic cell death was caspase independent. Apoptosis caused by spindle checkpoint defects is thought to occur during the G1 phase, and the type of cell death that we identified does not meet the criteria for other defined types of cell death (); thus, we designated this type of cell death as caspase-independent mitotic death (CIMD). Because CIMD occurs in HeLa cells with compromised p53 activity (), CIMD appeared to be independent of p53. We confirmed that CIMD occurs in cells that lack p53 ( C; Fig. S1 E and Table S1). Next, we examined whether CIMD depends on p73, a homologue of p53, because a mitotic function of p73 has been suggested (; ). Overexpression of the dominant-negative mutant p73DD () suppressed CIMD ( D; Fig. S1 F), depletion of p73 diminished CIMD (), and CIMD did not occur efficiently in MEF cells () (Fig. S2 A, available at ). These results indicate that CIMD depends on p73 but not on p53. Mitochondria release apoptosis-inducing factor (AIF) and endonuclease G (EndoG) (; ; ), which are thought to regulate caspase-independent cell death (; ; ; ). Therefore, we examined whether AIF and EndoG are required for CIMD. Substantial amounts of AIF and EndoG were released from mitochondria in mitotic cells treated with 17-AAG and BUB1 siRNA (). AIF and EndoG immunostaining resulted in a pattern that resembled that of mitochondria stained with 3,3′-dihexyloxacarbocyanine iodide (DiOC) in mitotic cells, as described previously (). We confirmed that AIF and EndoG immunostaining was colocalized with MitoTracker Red CM-HsXRos staining (Fig. S2 B). The proportion of AIF- and EndoG-releasing mitotic cells was comparable to that of cells undergoing CIMD (compare , B and D with ); this similarity strongly suggests that AIF and EndoG are effectors of CIMD. Next, we examined whether CIMD depends on AIF and EndoG. Depletion of AIF and EndoG by siRNA treatment substantially reduced TUNEL signals that were induced by 17-AAG treatment and BUB1 depletion (; Fig. S2 C), whereas depletion of AIF, EndoG, or both did not affect the mitotic delay induced by 17-AAG (Fig. S2 D). These results indicate that DNA fragmentation is dependent on AIF and EndoG. We examined whether depletion of EndoG and AIF rescues the lethality caused by CIMD. Although depletion of AIF or EndoG alone did not rescue the lethality, of both EndoG and AIF substantially suppressed it (), indicating that both effectors are involved in the death-signaling pathway of CIMD. These findings lead us to conclude that CIMD is an active cell death system mediated by these apoptosis effectors. Inhibition of DNA decatenation arrests cells at metaphase, and the disruption of MAD2, but not BUB1, suppresses this metaphase arrest (). Therefore, we examined whether CIMD occurs when BUB1-depleted cells are arrested with ICRF187, a topoisomerase II inhibitor. Although we observed a substantial mitotic delay after ICRF187 treatment (unpublished data), the number of TUNEL BUB1-depleted cells was unchanged (Fig. S2 E). This finding suggests that inhibition of DNA decatenation does not induce CIMD and supports the hypothesis that CIMD occurs specifically when the kinetochore–microtubule attachment is altered. To investigate the timing of CIMD after the kinetochore–microtubule attachment is altered, we added 17-AAG or microtubule inhibitors to BUB1-depleted cells that were arrested by ICRF187. We then monitored TUNEL cells. CIMD began to occur within 20 min, and most of the mitotic cells were TUNEL within 2 h (). This finding indicates that CIMD occurs during mitosis and relatively rapidly after the kinetochore–microtubule attachment is altered, which supports our conclusion that CIMD is an active cell death system. We also tested whether cold shock induces CIMD. Cold treatment depolymerizes microtubules, which activates the spindle checkpoint (). When cells were incubated at 23°C, CIMD occurred in >90% of the mitotic cells within 3 h (). This is the fourth piece of evidence that supports the hypothesis that CIMD is caused by defects in kinetochore–microtubule attachment when BUB1 function is disrupted. To learn the fate of cells in which DNA was fragmented during mitosis, we performed time-lapse experiments. Most BUB1-depleted cells that remained in mitosis for ∼6 h after the addition of 17-AAG eventually collapsed directly from mitosis within 12 h (). In contrast, most luciferase siRNA–treated cells (a negative control) remained in mitosis up to 12 h later (). Therefore, the cells in which CIMD occurred looked normally arrested in mitosis for several hours after the kinetochore–microtubule attachment was altered. Most conventional apoptosis detection methods (i.e., annexin V assay, chromatin condensation, and other morphologic analyses by light microscopy) were not applicable to mitotic cells (unpublished data). Therefore, we performed transmission electron microscopy (TEM) to look at the ultrastructural features of cells in which CIMD had occurred. When DNA fragmentation was induced by BUB1 depletion and 17-AAG or paclitaxel, we observed increased numbers of abnormal mitochondria (condensed, whorled, or onion-skin) and autophagosomes (; Fig. S3, A and C, available at ). The mitochondria were significantly smaller than those in control cells (), suggesting that mitochondrial fragmentation occurred. These changes indicated active cell death, possibly through autophagy (; ; ; ). BUB1 depletion does not compromise mitotic delay during normal mitosis or in response to nocodazole-induced spindle damage (). Our findings support that earlier study, and we believe that we can now explain this phenomenon. Because CIMD occurred, the mitotic index appeared to be unchanged. When a small amount of BUB1 remains in the cell it is sufficient to induce mitotic delay, but when BUB1 is completely depleted, cells prematurely exit mitosis (). Therefore, we attempted to determine how much BUB1 would have to be depleted to induce CIMD. We performed an siRNA dilution experiment using BUB1 targets to deplete BUB1 almost completely. When BUB1 was nearly depleted, CIMD occurred or the mitotic index was significantly reduced (, A and B; Fig. S4 A, available at ). Therefore, complete depletion of BUB1 causes premature mitotic exit. The number of abnormal nuclei was also increased similarly to that seen after MAD2 depletion (), and partial depletion of MAD2 did not induce CIMD (; Fig. S4 B). These results indicate that CIMD does not occur when BUB1 is almost completely depleted; the remaining BUB1 appears to be required to induce CIMD. A substantial number of cells with abnormal nuclei did not result from CIMD, which raises the possibility that CIMD might kill the cells that are going to have abnormal nuclei. Furthermore, a kinase-dead BUB1 mutant failed to suppress CIMD, which suggests that the kinase activity is important for inhibition of CIMD (). CIMD depends on BUB1 depletion, which suggests that microtubule inhibitors or 17-AAG induces CIMD of tumor cells that have a deficient spindle checkpoint. We tested whether microtubule inhibitors or 17-AAG induces CIMD of cells derived from tumors with CIN in which the spindle checkpoint is compromised and of tumor cells with microsatellite instability (MIN) in which the spindle checkpoint is intact (). CIMD occurred in tumor cell lines with CIN (Caco-2, SW480, and HT29) but not in those with MIN (SW38, DLD-1, and HCT116) (). CIMD occurred in 70–90% of the tumor cells with CIN that were TUNEL (unpublished data). We did not detect any caspase (caspases 1, 3–9) activity in mitotic tumor cells with CIN (Fig. S4 D), and caspase inhibitors BAF and zVAD did not inhibit DNA fragmentation (Fig. S4 E). These results suggest that the tumor cell lines with CIN have defective BUB1 pathways. In an early study, BUB1 mutations were not found in these cells (). Therefore, we measured the BUB1 protein levels in tumor cells with CIN; the level of BUB1 expression in tumors with CIN was lower than that in tumor cells with MIN or in HeLa cells (). The BUB1 levels in the tumor cells with CIN were ∼40% of that in HeLa cells (Fig. S4 F). Partial reduction of BUB1 in HeLa cells can induce CIMD; therefore, the low level of BUB1 expression could explain why the tumors with CIN induce CIMD. To test this theory, we overexpressed BUB1 in tumor cells with CIN to see whether restoring BUB1 levels suppresses CIMD. As expected, overexpression of BUB1 suppressed CIMD in the colon tumor cell lines with CIN (; Fig. S1 B). Furthermore, the expression of the mutant allele Bub1*V400, which was found in a tumor cell with CIN, (Cahill et al., 1998) induced CIMD in HeLa cells (; Fig. S1 B). These findings suggest that CIMD is a main mechanism by which microtubule inhibitors and 17-AAG kill tumor cells with CIN. #text The siRNAs targeting MAD2 and luciferase have been described previously (; ). We used three BUB1 siRNAs: 5′-GCCUGCCAACCCCUGGGAATT-3′(BUB1siRNA#1), 5′-CAACACUAUACUAACAAGATT-3′(BUB1siRNA#2), and 5′-CCAGGCUGAACCCAGAGAGTT-3′ for the studies described in the main text. Similar data were obtained when these independent sets of siRNAs were used. The siRNA targeting AIF has been described previously (). We also designed another AIF siRNA: 5′-CUUGUUCCAGCGAUGGCAUUU-3′. Similar data were obtained when these two independent sets of siRNAs were used. We designed three sets of EndoG siRNAs: 5′-AAGAGCCGCGAGUCGUACGUG-3′, 5′-AACGCACCUGUGGAUGAGGCC-3′, and 5′-CGGGCUUCGGGGCUGCUCUUU-3′, and similar data were obtained when these three independent sets of siRNAs were used. The siRNAs targeting MAD2, BUB1, AIF, and EndoG were synthesized by the Hartwell Center for Bioinformatics and Biotechnology at St. Jude Children's Research Hospital (Memphis, TN). Table S2 lists the antibodies used in this study (available at ). The colony outgrowth assay was performed as described previously (; ; ) with a minor modification. HeLa cells were transfected with siRNAs by using Lipofectamine 2000. 24 h after transfection, the cells were incubated with 100 nM 17-AAG or 1.5 nM Taxol for 2 d, and the drug was removed by washing. Transfected cells (including mitotic cells that were recovered from the supernatant; = 2,000) were spread in one well of a six-well cluster (Corning Costar) and incubated 12–14 d to allow colony formation. Colonies stained with Giemsa solution (HEMA-QUIK stain solution; Fisher Scientific) were counted. The viability (%) was normalized; the percentage of surviving colonies of untreated cells transfected with control luciferase siRNA was arbitrarily set to 100. 48 h after siRNA transfection, HeLa cells, tumor cells with CIN, or tumor cells with MIN were incubated with 500 nM 17-AAG (A.G. Scientific) for 24 h. Cells were fixed with 4% paraformaldehyde in phosphate-buffered saline (pH 7.4), and the TUNEL assay was performed by using an in situ cell death detection system that contained TMR red (Roche). HeLa cells were transfected with siRNA, and 48 h later the cells were incubated in 500 nM 17-AAG for 24 h. The FLICA caspase assay was performed by using the carboxyfluorescein FLICA (Poly-Caspases FLICA [FAM-VAD-FMK]) apoptosis detection system (Immunochemistry Technologies, LLC). All human cell lines were purchased from American Type Culture Collection (Manassas, VA). HeLa and SW480 cells were cultured in high glucose DME (BioWhittaker) with 10% fetal bovine serum (FBS; Invitrogen); Caco-2 and RKO cells, in Eagle's minimum essential medium (ATCC) with 10% FBS; HT29 and HCT116 cells, in McCoy's 5A medium (ATCC) with 10% FBS; and DLD-1 cells, in RPMI-1640 medium (ATCC) with 10% FBS. All cell lines were grown at 37°C in 5% CO in a humidified incubator. Cells were transfected with annealed double-stranded siRNA or mammalian expression plasmids by using Lipofectamine 2000 (Invitrogen) or Fugene 6 (Roche). The method of immunoblotting has been described in detail elsewhere (; ). Cells were added to lysis buffer A (), and the mixture was frozen in liquid nitrogen, thawed, and sonicated. Before electrophoresis, cell lysates were mixed with an equal volume of 2× SDS sample buffer. Methods of indirect immunofluorescent staining have been described previously (; ), but were slightly modified. HeLa cells were grown for 48 h on coverslip slides (seeding, ∼2.0 × 10 cells). Asynchronous populations of HeLa cells were fixed in 4% paraformaldehyde in phosphate-buffered saline at 4°C for 30 min and then treated with 0.5% Triton X-100 in KB (10 mM Tris HCl, pH 7.5, 150 mM NaCl, and 0.5% bovine serum albumin) at room temperature for 30 min. The cells were then incubated with a specific primary antibody for 1 h at 37°C. After the cells were washed once with KB, they were incubated with the fluorescent secondary antibodies fluorescein isothiocyanate–conjugated AffiniPure IgG or Texas red–conjugated AffiniPure IgG (Jackson ImmunoResearch Laboratories) for 1 h at 37°C. Slides were washed once with KB and then incubated in KB containing 0.1 μg/ml DAPI (Sigma-Aldrich). Cells were observed through an Axioskop2 (Carl Zeiss MicroImaging, Inc.) motorized fluorescence microscope equipped with a Plan Apochromat 63× oil immersion lens (Carl Zeiss MicroImaging, Inc.), an HBO 100 microscope illuminator (Attoarc), and a microMAX CCD camera (Princeton Instruments, Inc.). Appropriate filters were used to photograph stained cells. Image acquisition and processing was performed with IP Lab Scientific Imaging Software (Scanalytics). Alternatively, we observed cells through a DM IRE2 motorized fluorescence microscope (Leica) equipped with an HCX PL APO 63× oil immersion lens (Leica), an ARC LAMP power supply HBO100 DC IGN (Ludl Electronic Products, Ltd.), and an ORCA-ER high-resolution digital CCD camera (Hamamatsu). Image acquisition and processing were performed using Openlab version 4 Scientific Imaging Software (Improvision). A DNA fragmentation assay was performed as described previously (, , ). In brief, cells were gently lysed for 30 min at room temperature in buffer containing 5 mM Tris-HCl (pH 7.4), 20 mM EDTA, and 0.5% Triton X-100. After centrifugation at 15,000 rpm for 15 min, supernatants containing soluble, fragmented DNA were collected and treated with RNase (20 μg/ml; Sigma-Aldrich) and then with protease K (20 μg/ml). DNA fragments were precipitated in 99% ethanol. Samples were then subjected to electrophoresis in a 2% agarose gel and visualized by staining with ethidium bromide. HeLa cells were plated on 10-mm-diameter tissue culture plates with glass bottoms (MatTek Corp.) that had been coated with poly--lysine. Cells were then transfected with either human BUB1 siRNA or luciferase siRNA using Lipofectamine 2000 (Invitrogen). After 48–54 h, cells were incubated with 1 mM ICRF-187 (cardioxane; Chiron Corp.). After 54 h, cells were transferred to L15 Leibovitz medium to which 2.05 mM -glutamine (HyClone) had been added. The medium was then supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin (both from Invitrogen). At the same time cells were coincubated with 1 mM ICRF-187 and 500 nM 17-AAG, over which Sigma mineral oil had been placed. Cells were then maintained at 37°C. Phase-contrast images were captured every half hour for 24 h (after 54–78 h of transfection). Cells were imaged on a DM IRE2 motorized microscope equipped with an HCX PL FLUOT AR 40× lens (Leica), an ARC LAMP power supply HBO100 DC IGN (Ludl Electronic Products, Ltd.), and an ORCA-ER high-resolution digital CCD (charge-coupled device) camera (Hamamatsu). Images were acquired and processed using Openlab version 4.0.4 Scientific Imaging Software (Improvision). Pictures images were saved in Openlab's LIFF format, converted to TIFF format, and then exported to Adobe Photoshop. Mitotic HeLa cells were collected by gentle pipetting and fixed briefly with a 37°C solution of 2% paraformaldehyde, 2.5% glutaraldehyde in 0.15 M sodium cacodylate (pH 7.4). Low-melting point agarose (4%) was mixed with an equal volume of fixed cells and was transferred to a box with cold packs for shipment to National Center for Microscopy and Imaging Research (NCMIR). The samples arrived cold at NCMIR and were rinsed three times for 5 min with 0.15 M sodium cacodylate plus 3 mM calcium chloride (pH 7.4) on ice and then post-fixed with 1% osmium tetroxide, 0.8% potassium ferrocyanide (Sigma-Aldrich), and 3 mM calcium chloride in 0.15 M sodium cacodylate (pH 7.4) for 2 h and then washed three times for 5 min with ice-cold distilled water. The cells were stained for 2 h with 2% uranyl acetate at 4°C, dehydrated in graded ethanol incubations, and embedded in Durcupan resin (Fluka). Ultrathin (80 nm) sections were post-stained with uranyl acetate for 10 min and Sato lead for 2 min and imaged with a JEOL 1200FX transmission electron microscope operated at 80 kV. Mitotic cells were imaged on film at 3,000–5,000 magnification on a JEOL 1200FX electron microscope. The negatives were digitized at 1,800 dpi using a Nikon Coolscan system, giving an image size of 4034 × 6009 pixel array. All reagents were purchased from TED PELLA, Inc., unless otherwise indicated. Fig. S1A shows that overexpression of BUB1 can suppress CIMD when siRNA that targets the 3′-UTR region of BUB1 and 17-AAG were used. Fig. S1 B shows a control immunoblot to confirm the transfection of the plasmids that were used in Fig. S1A and H. Fig. S1 C shows that increased concentrations of paclitaxel (Taxol; 10–1,000 nM) treatment of BUB1-depleted cells do not affect the levels of mitotic TUNEL-positive (black bars) cells substantially. Fig. S1 D shows determination of the concentration of inhibitors needed to suppress caspase activity in HeLa cells. Fig. S1 E shows that CIMD is independent of p53. Fig. S1 F shows overexpression of p73DD. Fig. S2 A shows CIMD did not occur in p73 MEF cells efficiently. Fig. S2 B shows AIF and EndoG (green) are colocalized with mitochondria (red). Fig. S2 C shows protein depletion by AIF siRNA and EndoG siRNA. Fig. S2 D shows mitotic arrest caused by 17-AAG was not affected by either AIF siRNA #1 or #2 or EndoG siRNA #1 or #2. Fig. S2 E shows TUNEL assay of mitotic BUB1-depleted and ICRF187-treated cells indicated that ICRF187 did not induce CIMD in BUB1-depleted cells. Fig. S2 F shows kinetics of Endo G release and TUNEL signals after addition of 17-AAG or microtubule inhibitors to BUB1-depleted cells. Fig. S3 (A–C) shows electron microscopy images of BUB1-depleted cells that were treated with 17-AAG. Fig. S4 A shows depletion of BUB1 by using various sets of siRNAs. Fig. S4 B shows the mitotic index is the same as that shown in C. Mitotic TUNEL-positive cells were not observed when MAD2 was partially depleted. Fig. S4 C shows immunoblotting of HeLa cells that were transfected with the indicated plasmid vectors by using anti-BUB1 antibody. Fig. S4 D shows CIMD induced by 17-AAG treatment did not activate caspases in tumor cells with CIN. Fig. S4 E shows CIMD induced by treatment with 17-AAG was not suppressed by Pan-caspase inhibitors VAD (zVAD; 50 μM) or BAF (50 μM). Fig. S4 (F and G) shows overexpression of survivin did not inhibit CIMD. Online supplemental material is available at .

Phosphatidylinositol 4,5-bisphosphate (PI4,5P) has been implicated in many biological processes, including vesicular trafficking (), secretion (), focal adhesion and cytoskeleton assembly (), regulation of ion channels (), and nuclear signaling pathways (). PI4,5P has a role not only as a substrate of PLC and phosphoinositol 3-kinase–mediated second messenger production, but also as a direct effector that binds to and regulates the function of many PI4,5P-interacting proteins (; ). The generation of PI4,5P in cells primarily occurs through the phosphorylation of phosphatidylinositol phosphate (PIP) by type I phosphatidylinositol phosphate kinases (PIPKI; ). Three isoforms of PIPKI (Iα, Iβ, and Iγ) have been characterized along with several splice variants. By associating with their unique binding partners, different PIPKI isoforms produce PI4,5P with distinct subcellular distributions, from which they perform individual biological functions (; ; ; ). Cell migration requires the coordination of many biochemical events, including organized adhesion formation and turnover as well as dynamic cytoskeletal rearrangements (; ). PI4,5P binds to and regulates many proteins that are crucial for the assembly of the migratory machinery. For example, PI4,5P regulates reorganization of the actin cytoskeleton by associating with α-actinin, WASP/N-WASP, gelsolin, cofilin, profilin, and villin (; ). PI4,5P has also been proposed to regulate adhesions by binding to and modulating talin, vinculin, ezrin/radixin/moesin, calpain, and other proteins involved in adhesion dynamics (; ). PI4,5P is therefore positioned to play key roles in migration by modulating adhesion dynamics and cytoskeleton rearrangement. Many observations indicate that PI4,5P is a key signaling molecule in the regulation of cell migration, yet the role of specific PIP kinases in the regulation of cell migration remains to be clarified. PIPKIγ is alternatively spliced in cells, resulting in at least two major variants, PIPKIγ635 and PIPKIγ661, which differ by a 26-amino-acid C-terminal extension (). Most interesting, the 26-amino-acid C-terminal extension binds to talin and targets PIPKIγ661, but not PIPKIγ635, to adhesions (; ). This specific targeting of PIPKIγ661 allows for the generation of PI4,5P at adhesions, which is known to enhance the association between integrin and talin (). The binding of talin to β-integrin enhances the affinity of integrin for its ligands and activates the integrin heterodimer (). As a result, PIPKIγ661 may manipulate the inside-out activation of integrin signaling and the adhesion formation through its association with talin. Other than facilitating the assembly of talin into adhesions, PIPKIγ661 may also influence the recruitment and activation of other adhesion components through the local generation of PI4,5P. These combined data indicate that PIPKIγ661 may play key roles in regulating adhesion dynamics that are critical for cell migration. Unlike PI3,4,5P, total levels of cellular PI4,5P are relatively high and undergo only modest changes upon stimulation of cell migration (). This suggests that PI4,5P generation is tightly controlled, both spatially and temporally, to fulfill the requirements of rapid adhesion turnover and cytoskeleton rearrangement that are critical to the process of cell migration. Here, we demonstrate that PIPKIγ661 is required specifically for growth factor–stimulated directional migration, supporting a role for PIPKIγ661 in generation of the PI4,5P required for cell migration toward an EGF concentration gradient. Previous studies have demonstrated that PIPKIγ661 is targeted very specifically to adhesions (; ; ). The localized generation of PI4,5P at adhesions has been proposed to have roles in both integrin activation and adhesion formation (), and these events are key for cell migration. To explore the possible role of PIPKIγ661 in cell migration, siRNA specifically targeting PIPKIγ668 (human homologue of mouse PIPKIγ661) was designed. As shown in Fig. S1 A (available at ), PIPKIγ668-specific siRNA could specifically knock down expression of PIPKIγ668 but had no effect on the expression of PIPKIγ640 (human homologue of PIPKIγ635). The effect of PIPKIγ knockdown on EGF-stimulated cell migration was quantified using a modified Boyden chamber transwell assay (; ). As shown in , the knockdown of global PIPKIγ by pan-PIPKIγ siRNA blocked EGF-stimulated migration of HeLa cells. To determine if this effect is specifically due to the knockdown of the PIPKIγ661 splice variant, PIPKIγ668-specific siRNA was used in the same assay. Specific PIPKIγ668 knockdown had the same effect as global PIPKIγ knockdown on attenuation of EGF-stimulated migration of HeLa cells (). The efficacy of these two siRNAs on PIPKIγ668 knockdown is shown in ; the expression of PIPKIγ668 in HeLa cells was efficiently knocked down by both of these two siRNAs without affecting the expression level of other proteins, such as talin, FAK, and actin. To distinguish the role of PIPKIγ668 in directed migration or random migration, a checkerboard assay was used. As shown in Fig. S1 B, knockdown of PIPKIγ668 attenuated EGF-induced directional migration but not random migration. These results provide the first evidence that PIPKIγ661 plays a role in directional migration. Equivalent results were observed using MtLn3 and A431 cell lines (unpublished data), confirming the observation that PIPKIγ661 is required for EGF-stimulated cell migration in these cells as well. To determine if the function of PIPKIγ661 in regulating migration is specific for EGF, the effect of PIPKIγ661 knockdown on migration induced by other chemoattractants, such as hepatocyte growth factor (HGF), LPA, and SDF1α, was investigated. HGF-induced cell migration was blocked by PIPKIγ661 knockdown (). This result further confirmed the role of PIPKIγ661 in growth factor–induced migration. Remarkably, the results from these migration assays showed that the knockdown of PIPKIγ661 did not affect LPA- or SDF1α-stimulated cell migration (; and Fig. S2, available at ). SDF1α is the ligand of CXCR4 chemokine receptor. Both CXCR4 and LPA receptor belong to the G protein–coupled receptor family. These data indicate that PIPKIγ661 plays a specific role in growth factor receptor–stimulated cell migration. In addition, the effect of PIPKIα knockdown on EGF-stimulated migration was also investigated. siRNAs specifically targeting PIPKIα were used to knock down the expression of PIPKIα in HeLa cells. As shown in , ∼85% of endogenous PIPKIα was knocked down by PIPKIα-specific siRNA in HeLa cells. However, the knockdown of PIPKIα did not affect EGF-stimulated migration (), demonstrating specificity for the PIPKIγ661 isoform. To determine if PI4,5P generation was required for PIPKIγ661 control of EGF-stimulated migration, the kinase-dead PIPKIγ661 (PIPKIγKD)–expressing cell line was used in the rescue experiment. As shown in , expression of PIPKIγKD could not rescue EGF-stimulated cell migration in PIPKIγ668 knockdown cells. However, the expression of the wild-type mouse PIPKIγ661 fully rescued EGF-stimulated cell migration. These findings demonstrate that the ability of PIPKIγ661 to produce PI4,5P is required. PIPKIγ661 binds to talin, and binding is regulated by tyrosine phosphorylation on Y644 (, ). Talin is a key protein in regulating adhesion turnover (; ). To determine if talin binding is necessary for PIPKIγ661 to mediate EGF-stimulated migration, the Y644 to phenylalanine mutant of PIPKIγ661 (PIPKIγY644F) was assayed to rescue the endogenous knockdown. PIPKIγY644F has in vivo defects in the association with talin (). As shown in , the mutant could not rescue the effect of PIPKIγ668 knockdown on EGF-stimulated cell migration. Furthermore, expression of PIPKIγ635, the short splice variant of PIPKIγ, which does not bind talin, was also unable to rescue the effect of PIPKIγ668 knockdown on EGF-stimulated migration. The same region of PIPKIγ661 that interacts with talin also associates with the μ subunits of the adaptor protein (AP) 1B and AP2 complexes. Both PIPKIγY644F and PIPKIγ635 also have in vivo defects in the association with μ subunits (; ). To exclude the possibility that PIPKIγ661 effects on EGF-induced migration is due to its binding with μ subunits of the AP1B and AP2 complexes but not with talin, an S645 to phenylalanine mutant of PIPKIγ661 (PIPKIγS645F) was used. PIPKIγS645F has in vivo defects in the association with talin but maintains the ability to associate with μ subunits of the AP1B and AP2 complexes (; ). As shown in , the mutant could not rescue the effect of PIPKIγ668 knockdown on EGF-stimulated cell migration. It is important to note that the PIPKIγY644F also loses the ability to associate with the AP1B and AP2 complexes (; ). Together, these findings indicate that PIP kinase activity, talin binding, and possibly AP complex binding, are required for PIPKIγ661 regulation of EGF-stimulated migration. Because talin binding is required for PIPKIγ661 effects on EGF- induced migration, regulating talin assembly into adhesions is a likely mechanism by which PIPKIγ661 regulates EGF-induced migration. To determine if talin assembly into adhesions is altered in PIPKIγ661-deficient cells, the impact of PIPKIγ661 knockdown on EGF-induced talin assembly of adhesions was assessed. A vector-based short hairpin RNA (shRNA) was used to knock down PIPKIγ668 expression in HeLa cells. This vector also expresses a red fluorescent protein, DsRed, alongside the expression of the PIPKIγ668-specific shRNA. In this way, the PIPKIγ668 knockdown cells can be identified as the red fluorescence positive cells. Transfection of this vector-based shRNA into HeLa cells resulted in efficient knockdown of PIPKIγ668 (). As shown in , in the absence of EGF, there are relatively few talin-containing adhesions found in either the control shRNA–transfected HeLa cells or in the PIPKIγ668 shRNA–transfected HeLa cells. Stimulation with EGF leads to talin recruitment to adhesions in DsRed-negative or control shRNA–expressing HeLa cells. However, in the PIPKIγ668 knockdown HeLa cells, the EGF-induced talin recruitment to adhesions was significantly decreased (). This result indicates that PIPKIγ668 is required for EGF-induced talin assembly into adhesions. Furthermore, in the PIPKIγ668 knockdown HeLa cells, the EGF-induced vinculin recruitment to adhesions was also decreased (Fig. S3, available at ). This result further confirmed that PIPKIγ668 is required for EGF-induced adhesion formation. At the onset of cell migration, cells extend protrusions of the plasma membrane at their leading edge and then assemble nascent adhesions serving as points of traction and help to establish cell polarity (; ). As PIPKIγ661 is required for EGF-induced talin assembly into adhesions, it is possible that PIPKIγ661 is required for EGF-induced talin recruitment to the leading edge and to regulate adhesion assembly. To determine if PIPKIγ661 is required for the polarized recruitment of talin to the leading edge in a gradient of chemoattractant, a micropipette-stimulation assay () that can selectively stimulate cells with locally released EGF was used. As shown in , a micropipette filled with EGF was placed near cells, and constant pressure was added to the micropipette to create a concentration gradient of EGF. Vector-based shRNA was used to knock down the expression of PIPKIγ668, and knockdown cells were visualized with DsRed. To assess talin turnover at the leading edge in real time, a GFP-talin construct () was transfected into HeLa cells to detect its dynamic assembly at adhesions. GFP-talin assembly rate was quantified in . Rate constants for talin assembly of individual adhesions were calculated as described in Materials and methods. GFP-talin assembly into adhesions at the leading edge at different time points of micropipette stimulation is shown in . In the cells expressing control shRNA, local stimulation with EGF by the pipette led to rapid recruitment of GFP-talin to the leading edge and assembly into nascent adhesions. In the cells expressing PIPKIγ668 shRNA, however, GFP-talin recruitment to nascent adhesions was significantly decreased (; and Videos 1 and 2, available at ). These findings indicate that EGF induces new adhesions in the direction of the growth factor concentration gradient, and this requires PIPKIγ661. The function of PIPKIγ661 is regulated by its tyrosine phosphorylation. Src-mediated phosphorylation of Y644 increases the PIPKIγ661-talin binding affinity () and blocks the PIPKIγ661 interaction with AP complexes (). To determine if EGF can stimulate PIPKIγ661 tyrosine phosphorylation, the tyrosine phosphorylation of PIPKIγ661 was quantified. As shown in , EGF stimulation noticeably increased tyrosine phosphorylation of PIPKIγ661. To determine if this effect is specifically due to the activation of EGF receptor (EGFR), cells were preincubated with EGFR-specific inhibitor PD153035, and this blocked EGF-induced tyrosine phosphorylation of PIPKIγ661. Interestingly, when Y644 of PIPKIγ661, the known Src phosphorylation site, was mutated to phenylalanine (PIPKIγY644F), the EGF-stimulated tyrosine phosphorylation was not affected (), suggesting that Y644 is not required for EGF-induced tyrosine phosphorylation of PIPKIγ661 in vivo. To identify the key tyrosine residues for the EGF-induced tyrosine phosphorylation, a series of tyrosine residues were mutated, and the Y634 to phenylalanine mutant of PIPKIγ661 (PIPKIγY634F) was found to ablate EGF-induced tyrosine phosphorylation (). These findings demonstrate that Y634 is the key tyrosine residue for EGF-induced tyrosine phosphorylation. EGFR is a tyrosine kinase, and ligand binding to the extracellular portion of the EGFR leads to autophosphorylation of specific tyrosine residues in its cytoplasmic region, which stimulate the intrinsic kinase activity of the receptor (). To determine if EGFR could directly phosphorylate PIPKIγ661, an in vitro EGFR kinase assay was used. The reconstructed wild-type PIPKIγ661 C terminus or the mutant PIPKIγY634F C terminus was purified from and subjected to in vitro EGFR kinase assay as substrates of purified EGFR. As shown in (E and F), wild-type PIPKIγ661 C terminus can be phosphorylated directly by purified EGFR. In this assay, mutant PIPKIγY634F C terminus lost EGFR-induced phosphorylation compared with wild-type PIPKIγ661. Interestingly, in the Y644 to F mutant of PIPKIγ661 (PIPKIγY644F), EGFR-induced phosphorylation was also reduced compared with wild-type PIPKIγ661. Mutation of both Y634 and Y644 in PIPKIγ661 to phenylalanine resulted in a loss of the EGFR-induced phosphorylation compared with the wild type. In vivo the mutation of Y634, but not Y644, resulted in the loss of EGF-induced tyrosine phosphorylation of PIPKIγ661. These data demonstrate that EGFR phosphorylates Y634 in vivo and in vitro and are also consistent with results showing that Y644 is preferentially phosphorylated by Src (). The results demonstrate that Y634 of PIPKIγ661 is phosphorylated upon EGF stimulation. To determine if Y634 phosphorylation is required for PIPKIγ661 effects on EGF-induced migration, the PIPKIγY634F-expressing stable HeLa cell line was used in a cell migration assay to demonstrate its effect on EGF-induced migration. As shown in , the expression of PIPKIγY634F could not rescue the effect of PIPKIγ668 knockdown on EGF-induced directional migration. The results indicate that Y634 phosphorylation is required for the role of PIPKIγ661 in EGF-induced migration. As shown in (B and D), PIPKIγY634F still retains the ability to bind talin. Quantification of PIP kinase activity demonstrated that PIPKIγY634F retains kinase activity on a level that is comparable to wild-type PIPKIγ661 (unpublished data). It is interesting to determine if Y634 phosphorylation is required for EGF-induced talin assembly to facilitate migration. As shown in , both wild-type PIPKIγ661 and talin are recruited to adhesions after EGF stimulation. However, in comparison, PIPKIγY634F was less efficiently recruited to adhesions by EGF stimulation. Also, talin recruitment to adhesions was decreased in these same cells. The expression of PIPKIγKD (kinase dead) also decreased talin recruitment to adhesions after EGF stimulation, similar to the PIPKIγY634F mutant. These data demonstrate that both Y634 phosphorylation and the kinase activity of the PIPKIγ661 are required for efficient talin assembly into adhesions induced by EGF stimulation. In addition, PIPKIγS645F, the mutant that has in vivo defects in association with talin and does not rescue directional migration, also was not recruited to adhesions after EGF stimulation. Correspondingly, cells expressing PIPKIγS645F showed decreased talin recruitment to adhesions after EGF stimulation compared with cells expressing wild-type PIPKIγ661. This result indicates that the collaboration of PIPKIγ661 and talin is required for their recruitment to adhesions induced by EGF. Furthermore, Y634 phosphorylation and the kinase activity of the PIPKIγ661 are also required for efficient vinculin assembly into adhesions induced by EGF stimulation (Fig. S4, available at ). This result further confirms that Y634 phosphorylation and the kinase activity of the PIPKIγ661 are required for EGF-induced adhesion formation. To determine if Y634 phosphorylation of PIPKIγ661 is required for the polarized recruitment of talin to the leading edge, HeLa cells expressing mCherry-tagged wild-type PIPKIγ661, PIPKIγY634F, or PIPKIγKD were stimulated with a gradient of EGF using the micropipette-stimulation assay for directional migration. A micropipette filled with 10 nM EGF was placed near the cell as in . GFP-talin was used to allow for the detection of talin dynamics at adhesions in real time. As shown in , local stimulation by EGF induced rapid co-translocation of both mCherry-tagged wild- type PIPKIγ661 (mc-PIPKIγ661) and GFP-talin to the leading edge and assembly into nascent adhesions. However, mCherry-tagged PIPKIγY634F and PIPKIγKD (mc-PIPKIγY634F and mc-PIPKIγKD) were less efficiently recruited to adhesions by EGF stimulation. Correspondingly, EGF-induced GFP-talin assembly to nascent adhesions at the leading edge was also decreased both in mc-PIPKIγY634F– and mc-PIPKIγKD–expressing HeLa cells (; and Videos 3–8, available at ). These findings further demonstrate that both Y634 phosphorylation and the kinase activity of PIPKIγ661 are required for the polarized recruitment of talin to the leading edge in a gradient of EGF. Although phosphorylation of PIPKIγ661 at Y634 does not alter the ability of PIPKIγ661 to bind talin (), Y634 phosphorylation may affect the association of PIPKIγ661 with other proteins involved in regulating migration. To explore this possibility, the binding ability of wild-type PIPKIγ661 and PIPKIγY634F with PIPKIγ-associating proteins other than talin was compared. AP1B and AP2 complexes bind to PIPKIγ661 with the same sequence that binds talin (). When PIPKIγY634F was assayed for binding to AP1B and AP2 complexes, this mutant associated with both AP complexes in a manner that was indistinguishable from the wild-type PIPKIγ661 (unpublished data). Interestingly, we have shown that PLCγ1, an enzyme also required for EGF-stimulated directional migration (), associates with the PIPKIγ661 complex. Further, PLCγ1 was found to be differentially associated with wild-type PIPKIγ661 and PIPKIγY634F. As shown in , PLCγ1 was coimmunoprecipitated with endogenous PIPKIγ and is the first evidence that PLCγ1 could associate with the PIPKIγ661 complex. Both wild-type PIPKIγ661 and PIPKIγY634F associate with PLCγ1, whereas PIPKIγY634F had a stronger interaction with PLCγ1 than did wild-type PIPKIγ661 (). Intriguingly, association of PIPKIγ661 with PLCγ1 was lost after EGF stimulation, whereas the PLCγ1 association with PIPKIγY634F could not be efficiently disrupted by EGF treatment (). These results indicate that phosphorylation of PIPKIγ661 at Y634 regulates PLCγ1 association with PIPKIγ661. Directional cell migration is critical to many biological and pathological processes, including embryogenesis, the inflammatory response, atherosclerosis, tissue repair and regeneration, and cancer metastasis (; ). PI4,5P modulates many key components of the cell migration machinery and is proposed to be synthesized in a highly spatial and temporal fashion to regulate this process (; ; ). However, the underlying mechanism for the restricted generation of PI4,5P that regulates migration is poorly understood. PIPKIγ was proposed to be the major enzyme responsible for PI4,5P synthesis at synapses (), and much of the work on the biological functions of PIPKIγ since then have been focused on this system. Recently, compelling observations revealing the critical biological roles of PIPKIγ in nonneuronal systems have emerged. PIPKIγ635 has been reported to have important functions in G protein–coupled receptor–mediated IP generation (). More interesting, PIPKIγ661 is implicated in focal adhesion assembly (), AP2-mediated endocytosis (), and the endocytosis and basolateral sorting of E-cadherin (). Our current findings demonstrate that PIPKIγ661 is required for EGF-stimulated directional migration. The role of PIPKIγ661 in regulating EGF-stimulated directional migration is unique, as neither PIPKIγ635 nor PIPKIα can compensate for the loss of PIPKIγ661. In coordination with Rac signaling, PIPKIα plays roles in dorsal membrane ruffling stimulated by PDGF (; ), and PIPKIα interacts with the LIM domain protein Ajuba and appears to coordinate the targeting to membrane ruffles (). Membrane ruffles are often found on the cell surface and at the advancing front of a lamellipodium and serve as sites of actin polymerization (). Although membrane ruffling is thought to be important for migration, it is not necessary for migration of all cells. For example, Rac1-deficient macrophages exhibit defects in membrane ruffling but show normal directional migration (). Investigation in epidermal keratinocytes demonstrated that high membrane ruffling rates correlated with low lamellipodia persistence and inefficient migration (). PIPKIα-induced membrane ruffling may be required for certain types of migration but may not be essential for the EGF-stimulated migration of epithelial cells like HeLa, MtLn3, and A431. Cell migration is an integrated process that requires the continuous, coordinated formation and disassembly of adhesions (; ). Our results demonstrate that PIPKIγ661 is required for talin assembly into nascent adhesions forming at the leading edge toward the direction of the growth factor concentration gradient. This supports the hypothesis that PIPKIγ661 regulates growth factor–mediated migration ultimately through modulation of adhesion dynamics. These data indicate that PIPKIγ661 works as an effector of growth factor signaling to provide a link to the corresponding intracellular adhesion dynamics required for migration. Talin plays key roles in adhesion turnover and cell migration by providing a link between integrin and the cytoskeleton (; ; ). Reduction of talin expression leads to defects in normal cell migration in (). Despite the critical roles in cell migration, how talin is recruited to and regulated at adhesions is poorly understood. The results presented here support a role for PIPKIγ661 in talin recruitment and regulation downstream of EGF-induced directional migration. Y634 phosphorylation of PIPKIγ661 is required for talin recruitment in response to EGF. It is interesting that the phosphorylation of Y634 does not obviously affect the kinase activity of PIPKIγ661 or its talin binding affinity. The phosphorylation of Y634 regulates the interaction between PIPKIγ661 and PLCγ. Therefore, current results provide a possible model of how Y634 phosphorylation of PIPKIγ661 is involved in EGF-induced migration (). By associating with PIPKIγ661, PLCγ1 could hydrolyze the PI4,5P produced by PIPKIγ661; the hydrolysis would diminish the PI4,5P required to regulate talin assembly to adhesions. EGF-induced phosphorylation of PIPKIγ661 at Y634 causes a disassembly of the PLCγ1–PIPKIγ661 complex, and this could enhance PI4,5P accumulation and thus enhance talin assembly into adhesions. It is established that PLCγ1 regulates EGF-induced migration (). Activation of PLCγ is required for protrusion formation at the leading edge. PLCγ1 cleaves PI4,5P and releases actin binding proteins gelsolin and cofilin to initiate protrusion and define the direction of cell migration (; ; ). Furthermore, PLCγ1 also modulates the polarized localization of m-calpain and regulates the adhesion detachment (). In this context, PIPKIγ661 and PLCγ likely work together to regulate local PI4,5P level at the leading edge to facilitate the protrusion formation and stabilization of adhesions. As the involvement of PIPKIγ661 seems specific for growth factor stimulation and not LPA and SDF1α stimulation, it is likely that G protein–coupled receptors use distinct downstream elements to modulate the formation of protrusions. Together, this unexpectedly indicates that PIPKIγ661 is a unique and key signaling component specific for growth factor–induced directional cell migration. Tyrosine 644 is another residue site on PIPKIγ661 that is phosphorylated by Src. This phosphorylation enhances the PIPKIγ661-talin binding affinity (). Src can be activated by EGFR via a Ral-GTPase–dependent mechanism (). It is possible that EGF treatment could also lead to PIPKIγ661 phosphorylation at Y644 by activating Src. Although our results show that EGF stimulation under these conditions did not lead to obvious change of cellular PIPKIγ661-talin binding affinity, EGF stimulation may modulate PIPKIγ661-talin association at specific sites, such as adhesions via Src-mediated phosphorylation of Y644. EGF stimulation may lead to an ordered set of phosphorylation events that regulate PIPKIγ661 interactions with its partners, such as PLCγ1 and talin, and, via this mechanism, regulate directional migration. By regulating adhesion dynamics at the leading edge, PIPKIγ661 may play a role in the regulation of cell protrusion toward the direction of stimulation. Cell adhesion and protrusion are highly interrelated during migration. Protrusion results primarily from actin polymerization at the leading edge of migrating cells. Nascent adhesions forming at leading edge could stabilize the new protrusion and help establish cell polarity during migration. Talin binds actin and provides a molecular linkage between adhesion and actin that inhibits retrograde flow and thus regulates the rate of protrusion by counterbalancing the forward movement of actin polymerization (; ). It is plausible that PIPKIγ661 participates in stabilization of protrusions by enhancing the recruitment of talin to the leading edge; via this mechanism, PIPKIγ661 would determine the direction of migration. The same region of PIPKIγ661 that interacts with talin also associates with the μ subunit of the AP1B and AP2 complexes (; ). Indeed, membrane trafficking also regulates the directional migration of cells (; ; ; ; ). Thus, our results are equally consistent with a role for the PIPKIγ661 in membrane trafficking in controlling EGF-stimulated directional migration. Endocytosis of receptor tyrosine kinases plays a key role in growth factor–stimulated chemotaxis (; ). In , endocytic trafficking is required for directional migration stimulated by EGF and EGFR homologues in vivo (). We have explored the possibility that PIPKIγ661 modulates EGFR endocytosis. Knockdown of PIPKIγ661 or expression of the PIPKIγ661 kinase-dead mutants did not effect EGFR internalization upon agonist binding (unpublished data). Nevertheless, this result does not eliminate a potential role for PIPKIγ661 in modulation of membrane trafficking within the EGF-stimulated migration pathway. There are other trafficking events that may be crucial for PIPKIγ661 to regulate EGF-stimulated migration, such as trafficking of integrins (; ). PIPKIγ661 also regulates the ability of epithelial cells to assemble E-cadherin based cell–cell contacts (), and here we show that it also regulates the ability of cells to migrate toward an EGF gradient. Therefore, PIPKIγ661 has been implicated to be required for two key physiological processes: cell–cell adhesion and directional cell migration. This is very significant, as these two processes are fundamental in early stages of the metastasis of cancers of epithelial origin. In breast cancer metastasis, the sequential loss of E-cadherin and cell–cell contacts allows tumor cells to migrate (). The migration toward blood vessels is stimulated in some cases by a gradient of EGF, and this facilitates a key step called intravasation, where the tumor cells migrate through the vessel wall to be transported throughout the body (; ). Further exploration of the underlying signals and mechanisms that regulate PIPKIγ661 will be crucial to understanding the complete role of this enzyme in cell migration and tumor metastasis. Site-directed mutagenesis for the PIPKIγ661 mutants was performed using PCR-primer overlap extension with mutagenic primers. The mutations were confirmed by DNA sequence analysis. The siRNA sequence targeting pan-PIPKIγ is 5′-GGACCUGGACUUCAUGCAG-3′. The siRNA sequence targeting human PIPKIγ668 is 5′-GAGCGACACAUAAUUUCUA-3′. The sequence of control scrambled siRNA is 5′-GUACCUGUACUUCAUGCAG-3′. The mCherry vector was provided by R.Y. Tsien (University of California, San Diego, La Jolla, CA). Anti-talin and anti-actin antibody were purchased from Sigma-Aldrich. Monoclonal mouse anti-PY (4G10), anti-FAK (4.47), and anti-PLCγ1 antibodies were obtained from Upstate Biotechnology. Anti-HA antibody was purchased from Covance and Roche. Polyclonal PIPKIγ anti-serum was generated as described previously (). Anti-PIPKIγ661 specific antibody was purified on an affinity column generated by coupling the 26-amino-acid C-terminal peptide of PIPKIγ661 to cyanogen bromide–activated Sepharose 4B (Sigma-Aldrich) as described previously (). Secondary antibodies were obtained from Jackson ImmunoResearch Laboratories. HeLa cells and A431 cells were cultured using DME supplemented with 10% FBS. MTLn3-EGFR cells, provided by J. Condeelis (Yeshiva University, New York, NY), were maintained in MEM α supplemented with 5% FBS. For plasmid transfection, HeLa cells were transfected by using Lipofectamine 2000 (Invitrogen) following the manufacturer's instructions. For siRNA transfection, MTLn3-EGFR and A431 cells were transfected with Oligofectamine, and HeLa cells were transfected with Lipofectamine 2000 following the manufacturer's instructions. Tet-off HeLa cells (CLONTECH Laboratories, Inc.) were stably transfected with various PIPKIγ constructs, which are under the control of the tetracycline responsive promoter. The transfected HeLa cells were maintained in DME containing 10% FBS, 200 μg/ml G418, and 100 μg/ml hygromycin B to select for stable transfection. The medium was supplemented with 2 μg/ml doxycycline to suppress transgene expression, as doxycycline withdrawal results in expression of transfected PIPKIγ. The assays were performed in modified Boyden chamber transwell (Neuroprobe) as described previously (; ). The membrane was precoated with 10 μg/ml type I collagen. 50,000 cells were applied per well. Most chemotaxis assays were done at 37°C in humidified air with 5% CO for 4 h. For a different time course, cells were allowed to migrate for 2, 4, or 8 h. For each agonist concentration tested, cells migrated through to the underside of the membrane were counted in five high-power fields, in a blinded fashion. The migration index for each experiment was calculated as the mean number of cells that migrated toward medium-containing agonist divided by the mean number of cells that migrated toward medium-containing bovine serum albumin only. Immunoprecipitation was performed as described previously (). In brief, 48 h after transfection, HeLa cells were starved with serum-free DME overnight and then stimulated with 10 M EGF for 5 min. Then cells were harvested and lysed in 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5% NP-40, 5.0 mM NaF, 2 mM NaVO, 4 mM NaPO, 1 mM EDTA, 0.1 mM EGTA, 10% glycerol, and proteinase inhibitor cocktail, centrifuged, and incubated with protein A–Sepharose and 2 μg anti-HA antibody as indicated at 4°C overnight. The immunocomplexes were separated by SDS-PAGE and analyzed as indicated. The in vitro kinase assays were performed with EGFR (Promega) and recombinant purified PIPKIγ661 C terminus, encompassing residues 439–661 in the pET28 vector, as previously described (). The reaction was performed with a half unit of EGFR (as defined by Promega) in 5 mM Hepes, pH 7.4, 50 μM NaVO, 5 mM MgCl, 2 mM MnCl, and 250 mM (NH)SO with 5 μg protein substrate. The reaction was initiated with the addition of 20 μM ATP with 5 μCi γ-[P]adenosine triphosphate (GE Healthcare) and terminated by the addition of sample buffer after 15 min. The substrates were resolved by SDS-PAGE and fixed and stained using GelCode Blue (Pierce Chemical Co.). Image analysis was performed using NIH ImageJ. Immunofluorescence was performed as described previously (). Glass coverslips were acid washed and coated with 10 μg/ml type I collagen. Cells were resuspended and plated on the coverslips in serum-free DME and allowed to adhere for 3 h. Then, cells were stimulated with 10 M EGF for different time course (from 0 to 15 min) and fixed by methanol at −20°C for 10 min. Cells were then blocked by 3% BSA in PBS at RT for 30 min, incubated with the primary antibody overnight at 4°C, washed with 0.1% Triton X-100 in PBS, incubated with fluorescence-labeled secondary antibody at RT for 30 min, and washed with 0.1% Triton X-100 in PBS. Cells were maintained and examined using a 60× Plan oil-immersion lens on an inverted microscope (Eclipse TE200-U; Nikon). Images were processed as described previously () using Photoshop 7.0 (Adobe). The micropipette assay was performed as described previously (, ). A Femtojet Micromanipulator 5171 (Eppendorf) and a pump (Femtojet; Eppendorf) were used to control the position of the micropipette and the pressure required for the chemoattractant flow. To induce the formation of nascent adhesions at the leading edge, a micropipette was filled with 10 nM EGF and was placed ∼10 μm from the edge of a cell, and a constant pressure was exerted to induce flow. Fluorescence imaging of live cells was performed using a 60× objective on the Eclipse TE200-U inverted microscope housed in a closed system to maintain the temperature at 37°C. Glass-bottomed dishes were acid washed and coated with 10 μg/ml type I collagen. Cells were plated in DME media and allowed to adhere for 1 h, after which time the media was replaced with serum-free L15 media supplemented with 0.2% fatty acid–free BSA. Fluorescent images then captured every 1 min for 30 min using MetaMorph Imaging software (Universal Imaging Corp.). The dynamics of fluorescently tagged talin was quantified according to the described protocols (; ). Fluorescence intensities of individual adhesions from background-subtracted images were measured over time using MetaMorph Imaging software. For rate constant measurements, periods of assembly (increasing fluorescence intensity) of adhesions containing GFP-talin were plotted on separate semilogarithmic graphs representing fluorescence intensity ratios over time. Semilogarithmic plots of fluorescence intensities as a function of time were generated using the following formula: Ln([I]/[I0]) for assembly, where I0 is the initial fluorescence intensity and I is the fluorescence intensity at various time points. The slopes of linear regression trend lines fitted to the semilogarithmic plots were then calculated to determine apparent rate constants of assembly. For each rate constant, measurements were made on at least 10 individual adhesions of the cell, for a total of >50 adhesions in six separate cells. All measurements shown are the mean ± SEM. P values were calculated using test. Fig. S1 shows that knockdown of PIPKIγ668 attenuated EGF-stimulated directional migration. Fig. S2 shows that knockdown of PIPKIγ does not affect LPA- or SDF1α-stimulated cell migration in a different time course. Fig. S3 shows that PIPKIγ is required for EGF-induced vinculin assembly into adhesions. Fig. S4 shows the different effects of PIPKIγ661, PIPKIγY634F, PIPKIγKD, or PIPKIγS645F on EGF-induced vinculin assembly into adhesions. Videos 1 and 2 show the polarized recruitment of GFP-talin to the leading edge in a gradient of EGF in control shRNA– or PIPKIγ668 shRNA–transfected HeLa cells. Videos 3–8 show the polarized recruitment of GFP-talin to the leading edge in a gradient of EGF in mc-PIPKIγ661–, mc-PIPKIγY634F–, or mc-PIPKIγKD–expressing HeLa cells.

Endocytosis is crucial for the reformation of functional synaptic vesicles (SVs) after exocytosis of neurotransmitters. Although multiple mechanisms have been observed during SV endocytosis (; ; ; ; ), all of these involve the detachment of the vesicle from the plasma membrane, a step that is known as fission. Current evidence indicates that the GTPase dynamin functions at the fission step. For example, endocytic vesicles are arrested at the fission step in , a temperature-sensitive mutant of dynamin, and in the presence of the nonhydrolyzable analogue of GTP (; ). As inhibition of dynamin leads to complete blockage of SV endocytosis in mutants, dynamin activity is absolutely required for SV endocytosis (; ; ; ). To further understand dynamin function, we and others have previously characterized a major binding partner of dynamin, Dap160 (dynamin-associated protein 160 kD) and shown that it maintains proper dynamin localization at synapses (; ; ; ). Although there is a complete block in endocytosis during inhibition of dynamin function, residual endocytic activity persists in the absence of Dap160. This prompted us to study other proteins that may function in conjunction with dynamin. EGF receptor pathway kinase substrate clone 15 (Eps15) has been found in protein complexes with multiple endocytic proteins, including dynamin and the vertebrate homologue of Dap160, intersectin (; ; ). Originally discovered as an EGF receptor kinase substrate and a putative oncogene, Eps15 was later implicated in endocytosis and endosomal trafficking by in vitro studies (; ; ; ; ; ; ; ). Intriguingly, RNA interference knockdown of Eps15 in a nonneuronal cell line indicated that Eps15 is redundant with other endocytic proteins during the endocytosis of EGF receptor (). This prompted several groups of investigators to examine its role in SV endocytosis. In , -null mutants undergo a temperature-sensitive paralysis, and mutant nerve terminals exhibit vesicle depletion at high temperatures, indicating a role for Eps15 in maintaining the presence of SVs at the nerve terminals (). More recently, a mutant with reduced Eps15 levels was shown to undergo impairment in neurotransmission and paralysis only at high temperatures but not at physiological temperatures (). In contrast, injection of dominant-negative peptides to inhibit Eps15 function in squid axonal terminals leads to a very subtle decrease in the number of SVs but a strong inhibition of neurotransmitter release (). The in vivo data therefore suggest that Eps15 plays either an accessory role or no role in endocytosis. We therefore decided to generate protein-null alleles and investigate the role of Eps15 in flies in more detail. Here, we describe the functional characterization of Eps15 to address the following questions: Does Eps15 function in endocytosis of SVs from the plasma membrane at the nerve terminal? If so, at which step is Eps15 involved? Is there a functional relationship between Eps15 and Dap160? Our data indicate that Eps15 is required for efficient SV endocytosis and that loss of Eps15 causes a very severe reduction in dynamin and Dap160 protein levels at neuromuscular junctions (NMJs). Furthermore, the endocytic defects of - and -null mutations are not additive, suggesting that Eps15 and Dap160 act at the same step during endocytosis. The genome contains a single homologue of mammalian Eps15 (). It has three Eps15 homology (EH) domains, a coiled coil domain, a region with multiple DPF (aspartate-proline-phenylalanine) motifs, and at least one, possibly two, ubiquitin-interacting motifs (UIMs) at the C terminus, similar to its mammalian homologue (Fig. S1 A, available at ; ). To determine its spatial and subcellular distribution, we raised two different polyclonal antibodies against epitopes shown in Fig. S1 A: a rabbit antibody and a guinea pig antibody. Both antibodies are specific, as mutants that lack the gene display no immunoreactivity on Western blots ( and Fig. S1 B) and immunohistochemically stained tissues ( and Fig. S1, C and D). Western blots of fly heads show an ∼160-kD and a 90-kD band, whereas the predicted mol wt is ∼132 kD (). Immunoprecipitation experiments with rabbit antiEps15 and fly head protein extracts show that Eps15 binds to Dap160, as detected by Western blot and mass spectrometry of Coomassie blue–stained bands ( and Table S1). The Coomassie blue–stained gel indicates that Dap160 is a major binding partner of Eps15, in agreement with the biochemical interaction data for the vertebrate homologue of Dap160, intersectin (). In addition, rabbit anti-Eps15 also coimmunoprecipitates Eps15 with α-adaptin, epsin/Liquid facets, and dynamin (), consistent with protein interaction studies using the rat and homologues of Eps15 (; ). In summary, Eps15 shows structural and biochemical properties similar to Eps15 homologues in other organisms. To determine the function of Eps15 in vivo, we disrupted the gene using two different reverse genetic approaches. We imprecisely excised , a semilethal element insertion in the 5′ untranslated region of the gene (; ). This allowed isolation of a lethal allele, . We also deleted the entire locus by inducing site-specific recombination between two FLP recognition target (FRT)–containing transposons flanking the locus, generating (; ; ). Complementation analysis revealed that , , , and (a large deficiency uncovering the genomic region encompassing the locus), all fail to complement each other's lethality, indicating that each of these mutations affects the same gene, presumably (). To determine the severity of these mutations, we examined the lethal phase of the mutants in various allelic combinations. The homozygote and transheterozygote animals die as pharate adults, homozygous animals die as first instars, and transheterozygotes die as third instars (). To ensure that the lethality is due to mutations in the gene, we introduced a genomic fragment encompassing the locus into the mutant backgrounds (; genomic rescue construct). This rescues homozygotes and transheterozygotes to normal adults but only partially rescues transheterozygotes, as the adults are uncoordinated. The lethal phase and the rescue data indicate that the lethality of is solely due to deletion of the locus and that carries a second site mutation that is not shared by . To determine whether Eps15 is required in the nervous system, we expressed the cDNA using a neuronal-specific driver (). This rescued to normal adults with no obvious morphological defects, indicating that Eps15 is predominantly required in the nervous system (). To determine which mutations correspond to null alleles, we performed Western blot analyses. The and animals do not express detectable levels of Eps15 protein on Western blots ( and Fig. S1 B). In summary, we have isolated two independent protein-null mutations, and , which cause lethality, indicating that is an essential gene in , unlike the -null mutation, which is viable (). In addition, the essential role of Eps15 is confined to the nervous system in flies. To determine where Eps15 is expressed, we performed immunohistochemistry of third instar larvae (). Both anti-Eps15 antibodies reveal the same expression profiles, and this immunoreactivity is lost in null mutant animals, indicating that the labeling is specific ( and Fig. S1, C and D). Eps15 is expressed broadly in the nervous system and is much enriched in the neuropil of the central nervous system and at the NMJs (). In addition, most, and possibly all, imaginal discs also express Eps15 (unpublished data). To determine the subsynaptic localization of Eps15 at the NMJ, we costained Eps15 with Dlg, which at this resolution is effectively a postsynaptic marker (). Eps15 is enveloped by Dlg at the NMJ, indicating that Eps15 is enriched in the presynaptic compartment (). Collectively, we conclude that Eps15 is widely expressed but is enriched in the neuropil and in the presynaptic compartment of the NMJ. At the NMJ, Eps15 is present in the central region of the boutons in a honeycomb-like pattern, where most Eps15 staining was found surrounding active zones, defined by anti-Bruchpilot/nc82 (; ; ). In addition, Eps15 shows extensive colocalization with Dap160 (). Therefore, at the level of confocal microscopy, Eps15 is enriched in areas surrounding active zones and colocalizes with Dap160, which has been reported to show a similar honeycomb-like localization pattern (). These data prompted us to investigate Eps15 localization at the ultrastructural level. Silver-enhanced immunogold labeling of resting wild-type NMJ boutons with anti-Eps15 reveals that Eps15 is associated with vesicles in the lumen of boutons but is at neither presynaptic dense bodies (T-bars) nor vesicle-free areas of the boutons (, black precipitates; see Materials and methods). This immunogold labeling pattern is not observed when a nonrelevant primary antibody was used, indicating that the anti-Eps15 label is specific (unpublished data). Upon a mild stimulation with 60 mM K, Eps15 becomes more concentrated in regions close to the plasma membrane surrounding the T-bars (). This redistribution of Eps15 is particularly evident in mutants stimulated at a restrictive temperature ( and ). Interestingly, a subset of Eps15 is localized to the rim of invaginating vesicles or pits at the plasma membrane in stimulated mutant boutons (). In summary, Eps15 is associated with vesicle in resting boutons and is mobilized to areas adjacent to active zones upon stimulation. A larval NMJ typically consists of a series of boutons connected by neurites, like beads on string. These boutons are dynamic structures that sprout and retract during larval development (). Because Dap160 mutants display numerous extra NMJ boutons and branching (; ), and because Eps15 and Dap160 interact, we quantified the number of boutons and branches in mutant NMJs. Compared with controls, mutant larvae show a pronounced increase in bouton number at three different NMJ synapses (). In addition, control NMJs typically have few branchpoints from which boutons sprout (, arrows). In contrast, mutant NMJs show more branchpoints, some of which lead to more than two branches (). For example, mutant muscle 4 synapses commonly exhibit boutons with three branches or more, but such hyperbranched boutons are infrequent in controls (). Thus, mutant NMJs show supernumerary boutons and branches when compared with controls, similar to Dap160 mutants (; ). Eps15 has been implicated in endocytosis and endosomal trafficking (; ; ; ). In addition, loss of Eps15 in leads to a defect in vesicle cycling at the restrictive temperature (). To define the role of Eps15 in the SV cycle more precisely, we performed electrophysiological assays of mutant NMJs. To determine whether the exocytic machinery is intact, we elicited low-frequency stimuli and recorded evoked excitatory junctional potentials (EJPs) in 0.5, 1, and 5 mM extracellular Ca. Under these conditions, the EJPs are similar to wild-type amplitudes (). In addition, spontaneous release or miniature EJPs (mEJPs) of mutants shows similar amplitude but higher frequency than controls (). Therefore, in the absence of Eps15, exocytosis of neurotransmitters is normal under low-frequency stimulation, but spontaneous release is altered. To determine whether mutant NMJs are able to efficiently recycle vesicles under repetitive stimulation, we applied 10 Hz of stimulation for 10 min in 5 mM extracellular Ca. Control NMJs are able to maintain EJPs at ∼90% of the initial amplitudes, but NMJs show a strong synaptic depression to <40% of initial EJP amplitudes (). In this assay, , encoding a temperature-sensitive dynamin mutant, depresses to zero as a result of a complete block in endocytosis, whereas mutant NMJs show profiles of synaptic depression similar to those of mutant NMJs (; ). Hence, mutant NMJs show a defect in vesicle recycling that is similar to mutant NMJs. Compared with our current data on and mutants, other endocytic mutants like , , or show faster synaptic depression under a milder stimulation condition (higher Mg/Ca ratio) in our previous experiments (). In summary, although exocytosis of neurotransmitters is essentially normal at mutant NMJs, vesicle recycling is impaired but not blocked. One possible cause for the vesicle recycling defect in mutants is impaired endocytosis. To assay for endocytosis at the NMJs, we performed dye uptake experiments by applying a strong stimulus of 90 mM K and 5 mM Ca in the presence of the styryl dye FM1-43FX. Compared with controls, ∼50% of dye uptake remains in the partial loss-of-function mutant , whereas only ∼20% of dye uptake remains in the -null mutants, and / (). To determine the background fluorescence level, we performed mock-labeling control experiments in which control and mutant NMJs were incubated with FM1-43FX in a Ca-free/low-K solution. Comparison with mock-labeled controls indicates that very low levels of dye uptake occur in the -null mutants. The severe reduction in dye uptake phenotype in the -null mutant can be rescued with neuronal expression of the cDNA (see ). In contrast to its role at the synapse, Eps15 is not essential for receptor-mediated endocytosis in hemocytes (Fig. S2, A–C and F, available at ) and in the S2 cell line (RNAi knockdown experiment; unpublished data). Therefore, Eps15 is required for efficient SV endocytosis during nerve stimulation. Endocytosis from the plasma membrane consists of sequential steps that can be observed with transmission EM (TEM). To further investigate the endocytic defect in mutants, we quantified vesicle density and size of boutons in the resting state, during a strong stimulation with 90 mM K, and recovery after stimulation (). The vesicle densities of mutant boutons at rest are not significantly different from those of controls (; P > 0.05), in agreement with a lack of defect in exocytosis. Although the majority of vesicles in both control and mutant boutons are <45 nm in diameter, we observe an increased proportion of abnormally large vesicles or cisternae in resting mutant boutons (). Because intense stimulations reveal reduced FM1-43FX uptake and more severe synaptic depression at NMJs compared with controls, we applied 90 mM K stimulation to control and mutant boutons before processing for TEM. As shown in , we observed a reduction in vesicle densities in mutant boutons when compared with control boutons. When allowed to recover for 1 min without stimulation, the vesicle densities in wild-type control boutons return to levels that are not significantly different from resting levels, but vesicle densities in mutant boutons show a poor recovery (; P > 0.05). Relative to control boutons, the proportion of cisternae >85 nm is increased by stimulation in mutant boutons but is reduced and returns to wild-type levels after a 1-min recovery (). In stimulated mutant boutons, some cisternae appear to be large invaginations, which are contiguous with the plasma membrane (). In summary, mutant boutons undergo vesicle depletion, accumulation of cisternae, and large membrane invaginations during stimulation. After 1 min of recovery, the relative size distributions of vesicles in mutant and control boutons become similar, but the control boutons still have approximately twice the number of vesicles observed in mutant boutons. Thus, the TEM data indicate that Eps15 is required for efficient vesicle retrieval during intense stimulation. In its absence, bulk membrane invagination is not abolished, but budding of small clear vesicles from the plasma membrane and cisternae occur with much lower efficiency than in wild- type synapses. The presence of large cisternae in mutant boutons is reminiscent of mutant boutons after a 1-min recovery from stimulation (). This prompted us to examine mutant bouton morphology during stimulation in more detail. As seen in mutants, we observe cisternae and large membraneous bodies that appear to be contiguous to the plasma membrane (). The similarity of the large membrane invaginations in and mutant boutons suggests that both Eps15 and Dap160 act at a similar step in endocytosis. The endocytic defects, in particular, the retrieval defects observed by TEM in mutants resemble those observed in mutants (; ; ) and are quite different from those associated with the loss of and (, ). This suggests that Eps15 may be functionally related to Dap160, prompting us to examine the localization of Dap160 and its partner dynamin (), as well as other synaptic proteins in mutant NMJs. Immunohistochemistry of mutant NMJs shows that dynamin and Dap160 levels are severely reduced to ∼10 and ∼25% of control levels, respectively (). In addition, we observe reductions of other proteins associated with SV cycling, including Stoned B, synaptotagmin I, α-adaptin, and endophilin (; ; ; ; ). Cysteine string protein (Csp), a cochaperone protein required for proper neurotransmitter release, and Fas II, a cell-adhesion molecule implicated in synapse formation and function, are not significantly reduced at mutant NMJs (; ; ; P > 0.05). The strong reduction of Dap160 at NMJs prompted us also to examine the Eps15 levels at mutant NMJs. Although Eps15 levels are variably affected in different boutons and synapses, there is only a slight reduction, which is not statistically significant (; P > 0.05). These data indicate that Eps15 maintains normal levels of multiple synaptic proteins at the synapse. The severe reductions of dynamin and Dap160 at mutant NMJs suggest that these two proteins are critical partners of Eps15 in endocytosis. We then asked whether the reductions of these two proteins are restricted to the synapse. Western blots of third instar larval brains indicate a subtle reduction of Dap160 levels and normal dynamin levels in the central nervous system (Fig. S3, available at ), suggesting that the reductions of these two proteins in mutants occur predominantly at the synapse. Because Eps15 possesses different domains that biochemically interact with various protein partners, it is possible that different domains are involved in distinct processes. To address this possibility, we verified the binding properties of three major protein–protein interaction domains of Eps15 using in vitro binding assays. Consistent with similar studies on the mammalian Eps15, the EH-, DPF- and the UIM-containing regions of Eps15 bind Dap160, α-adaptin, and ubiquitin, respectively (Fig. S4, available at ; ; ; ). We then tested the requirement of N- and C-terminal domains of Eps15 in SV endocytosis and synaptic bouton development by neuronally expressing three cDNA constructs in the -null background: a full-length cDNA (Eps15wt), a truncated cDNA encoding only the N-terminal EH domains and coiled coil domains (ΔDPF; Fig. S1 A), and a full-length cDNA with point mutations in the EH domains (EHmut; unpublished data). It is expected that the ΔDPF truncation protein will lose the ability to interact with α-adaptin and ubiquitinated proteins, whereas the EHmut protein will only lose the ability to bind Dap160 and other EH binding proteins (; ). EHmut is expressed but fails to localize at the NMJ and was not further analyzed (unpublished data). Eps15wt and ΔDPF are expressed and localized normally at the NMJ and rescue to normal adults (unpublished data). This allowed us to assess the function of ΔDPF at the NMJ with respect to synaptic bouton development and SV endocytosis. To determine the contributions of Eps15 domains to synaptic bouton development, we quantified bouton numbers of the muscle 6/7 synapses (). The -null mutants show approximately twice the number of NMJ boutons compared with controls. The supernumerary bouton phenotype is rescued by neuronal expression of Eps15wt and partially rescued by ΔDPF. Thus, the Eps15 C-terminal domains play minor roles in synaptic development. Given the importance of α-adaptin in endocytosis, we asked whether a truncated Eps15 protein without the α-adaptin binding motifs can function normally in SV endocytosis by performing an FM1-43FX uptake assay. We observed rescue of the FM1-43FX uptake in -null mutants by neuronal expression of Eps15wt and ΔDPF (). Interestingly, neuronal expression of ΔDPF in -null mutants show rescue of FM1-43FX uptake to a level significantly less than, but close to, control levels (P < 0.05). Therefore, the DPF-containing region plays a minor role in SV endocytosis. In summary, the roles of Eps15 in synaptic bouton development and SV endocytosis can be partially substituted by an N-terminal fragment containing the EH and coiled coil domains. Several observations suggest a functional relationship between Eps15 and Dap160. First, Eps15 and Dap160 physically bind to each other and colocalize at NMJs ( and and Fig. S4 A; ). Second, mutant NMJs, like mutant NMJs show extra boutons and branches (; ; ). Third, both and mutants exhibit endocytic defects but no exocytic defects at the NMJ ( and ). Fourth, we observe large membraneous bodies in stimulated and mutant boutons (; ), which suggest a defect in the resolution of these large invaginations into small clear vesicles. Fifth, we observe a reduction of Dap160 levels at mutant NMJs. These data suggest that the two proteins may function at very similar steps in endocytosis, namely, during vesicle budding from the plasma membrane and from large invaginations during bulk endocytosis. Because Dap160 and Eps15 share N-terminal EH domains, which are more homologous to each other than EH domains of other proteins (40–60% similarity), it is possible that the EH domains of these two proteins serve redundant roles and act in parallel during endocytosis; if so, genetic ablation of both and should lead to more severe defects than either mutant alone. Alternatively, if Eps15 and Dap160 function at the same step during endocytosis from the plasma membrane, double-null mutants should have a phenotype similar to the single mutants. To investigate the functional relationship between Eps15 and Dap160, we determined the endocytic defects of and mutants in double mutant larvae. First, we examined the lethal phases of double mutants and show that - and -null mutants live to pharate adult stage, similar to the single - and -null mutants. Second, - and -null mutants show normal evoked EJPs in response to single stimuli and during 10 Hz of stimulation; hence, these double mutants also show the same depression kinetics as the - and -null single mutants (). Thus, the loss of Dap160 does not enhance the endocytic defects in mutants. To provide further support for the electrophysiology data, we performed FM1-43FX dye uptake experiments. As shown in , we observed a reduction in dye uptake in the double-null mutant that is similar to that of and single mutants. Thus, the -null mutation does not affect the residual vesicle recycling in the -null mutant background. In summary, these data argue that Eps15 and Dap160 act at the same step in SV endocytosis. mutant NMJs exhibit a severe reduction in dynamin and Dap160 protein levels, suggesting a functional relationship between these proteins and Eps15. At the same time, we observe relatively mild reductions of α-adaptin, Stoned B, synaptotagmin I, and endophilin. The reduction of protein levels of dynamin, Dap160/intersectin, α-adaptin, and Stoned B in mutants may be accountable by the interaction of these proteins with Eps15 (; ; ; ; ). As our data suggest that Eps15 is not required to regulate global protein levels of dynamin and Dap160 (Fig. S3), Eps15 may facilitate anterograde axonal transport of dynamin and Dap160 through interaction of the EH domains with vesicular membrane proteins containing NPF motifs (). Alternatively, because Eps15 has been implicated in the endocytosis of synaptotagmin I (), Eps15 may retain high concentrations of synaptic proteins by promoting the local recycling and sorting of vesicle-/membrane-associated proteins at synapses. Interestingly, a similar role has been proposed for Stoned in recycling synaptotagmin I at the synapse (; ). The endocytic defect in can be largely ascribed to impaired vesicle budding from plasma membrane and cisternae. Compared with control boutons, stimulations of mutant boutons result in more severe vesicle depletions, concurrent with the appearance of cisternae and large invaginations, which appear to be contiguous with the plasma membrane (). Importantly, 1 min after stimulation, wild-type control boutons recover to near normal vesicle densities, whereas mutant boutons show little recovery. Note that internalization of cisternae, an early step in bulk endocytosis (; ), does occur in the mutants, but the cisternae are not efficiently resolved into properly sized SVs. The endocytic intermediates found in mutants resemble those observed when dynamin or Dap160 function is perturbed. Cisternae and invaginations contiguous with the plasma membrane have been reported at neuronal synapses during genetic or dominant-negative peptide perturbations of dynamin function in and lampreys, respectively (; , , ; ; ). We have previously reported that mutant NMJs show accumulation of cisternae during recovery from stimulation (). Here, we fixed mutants during stimulation and observed severe vesicle depletion concurrent with large invaginations from the plasma membrane, similar to those observed in . Because dynamin levels are severely reduced at and NMJs, it is likely that both Eps15 and Dap160 act in concert to maintain high concentrations of dynamin at the synapse. Indeed, double mutants show the same defects in dye uptake and in maintaining release during prolonged stimulations as the and single mutants, suggesting that both Eps15 and Dap160 act at the same step in endocytosis. Furthermore, the notion that Eps15 is localized at the site of dynamin action is consistent with the mobilization of Eps15 to the plasma membrane in the vicinity of endocytic intermediates in stimulated nerve terminals ( and ). In addition, a functional interaction between Eps15 and dynamin is further bolstered by the recent demonstration of genetic interaction between mutants and the hypomorphic mutant, (). Note that the lack of endocytic defect in mutant hemocytes also correlates with normal dynamin levels in these cells (Fig. S2), further corroborating with a link between SV endocytosis and the Eps15–Dap160–dynamin interaction. In agreement with the correlation of endocytic defects with reduced dynamin levels in and mutants, the GTPase activity of dynamin is allosterically dependent on dynamin protein concentration (). Therefore, these data suggest that the maintenance of high dynamin levels at synapses is one of the key functions of Eps15 and Dap160 in SV endocytosis. Although we observed severe reduction of dynamin levels in both and mutants, there are subtle differences in synaptic protein levels between and mutants. For example, there is a reduction of synaptotagmin I in mutants, which was not observed in mutants. This may have resulted in minor differences in phenotypes, such as an increase in mEJP amplitudes in mutants but not in mutants. Eps15 has been implicated in clathrin-coated pit assembly in nonneuronal cells, based on its interaction with α-adaptin (; ; ; ). However, our data indicate that a truncated Eps15 protein lacking the α-adaptin–interacting DPF motifs (ΔDPF) partially rescues the FM1-43FX uptake defect of the -null mutant. This suggests that a direct interaction between Eps15 and α-adaptin is not essential for SV endocytosis. Consistent with the nonessential role of the DPF domain of Eps15, the Eps15 binding region of α-adaptin is not essential for the rescue of transferrin endocytosis after α-adaptin siRNA knockdown (). Hence, we propose that the dynamin–Dap160–Eps15 interaction is important for SV endocytosis. Based on our data and previous work, we propose a two-tier model for Eps15 function during SV endocytosis and synapse development. During SV endocytosis, Eps15 and Dap160 act together to stabilize several proteins at the synapse. In particular, the Eps15–Dap160 complex maintains high concentrations of dynamin, allowing allosteric activation of its GTPase activity, which is crucial for the resolution of newly internalized membranes into SVs (). It has been proposed that actin polymerization propels endocytosed vesicles toward the interior of the cell during dynamin-mediated endocytosis (; ; ; ). Given that Eps15, Dap160, and related EH domain proteins in yeast have been implicated in regulating the actin cytoskeleton (; ; ; ; ), it is tempting to speculate that the Eps15–Dap160 complex functions as a molecular scaffold to coordinate dynamin function in SV endocytosis with actin polymerization, a role that has been previously proposed for Dap160 (). It is thus interesting that the redistribution of Eps15 and Dap160 within NMJ boutons during stimulation suggests a dynamic role of Eps15 and Dap160 during endocytosis (; and ; , ). In addition to its role in SV endocytosis, Eps15 may function in concert with Dap160 during synapse development through regulation of signal transduction pathways and/or cytoskeletal organization. Both Eps15 and Dap160 have been implicated in signal transduction pathways (; ). In addition, Dap160/intersectin has been proposed to regulate cytoskeletal organization at the NMJ and in mammalian cells through interaction with actin-associated molecules like WASp (; ; ). This is consistent with our recent data on lamprey reticulospinal synapses, which show that actin at synaptic endocytic sties is reorganized when intersectin interactions are perturbed (). Interestingly, an EH domain protein is involved in cytoskeletal reorganization during budding of yeast cells (; ), which is thought to be analogous to NMJ bouton formation (). We speculate that Eps15 and Dap160 may function as a protein complex—with Nervous wreck and WASp—that regulates cytoskeletal organization during synapse development (; ; ). In this respect, Eps15 and Dap160 may serve to bridge upstream signal transduction pathways and cytoskeletal organization during synapse development. Two fragments encoding amino acid residues 586–816 and 915–1093 from Eps15 cDNA (SD09478 from Research Genetics) were cloned into the pGEX4T1 expression vector in frame with the GST coding region. GST-Eps15 (586–816) was used to generate the rabbit polyclonal antiserum, Rb Ab, and GST-Eps15 (915–1093) was used to generate guinea pig polyclonal antiserum, Gp Ab. xref #text italic xref #text italic ext-link xref #text xref italic sup #text Mutant larvae were separated from larvae bearing balancer chromosomes and cultured on grape juice agar with yeast paste (). This allows lethal phase of mutants to be determined without complications from overcrowding and competition from larvae bearing balancer chromosomes. Labeling of third instar NMJs, third instar brains, and adult brains was performed as described by . Animals were dissected in PBS and fixed in 4% formaldehyde in PBS for 20 min. Tissue was extensively washed in PBS and permeabilized with 0.4% Triton X-100; for anti-endophilin labeling, permeabilization was performed using 0.1% Tween 20. Antibodies were used in the following dilutions: affinity-purified anti-Dap160, 1:500 (); mouse anti-Dlg, 1:150 (4F3; ); rabbit anti-Dlg, 1:2,000 (); guinea pig anti-Dlg, 1:500 (); rabbit anti-HRP, 1:1,500 (Jackson ImmunoResearch Laboratories); rabbit anti–Stoned B, 1:200 (); mouse anti-Csp mAb 49, 1:20 (); rat anti-Nwk, 1:500 (); rabbit anti-dynamin, 1:500 (); mouse anti-dynamin, 1:50 (clone 41; BD Biosciences); rabbit anti–α-adaptin, 1:50 (); mouse anti–Fas II 1D4 guinea pig anti-endophilin, 1:200 (); mouse anti-Bruchpilot nc82, 1:100 (); and anti–synaptotagmin I, 1:500 (, ). Secondary antibodies conjugated to Cy3 or Cy5 (Jackson ImmunoResearch Laboratories) or Alexa 488 or 594 (Invitrogen) were used at 1:250. Samples were mounted in Vectashield mounting medium (Vector Laboratories). For , , , and Fig. S1, images were captured with a confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.) with a 40×/1.3 differential interference contrast plan-Neofluar oil-immersion lens using the LSM 5 software (Carl Zeiss MicroImaging, Inc.) and were processed using Amira 2.2 (Mercury Computer Systems) followed by Photoshop (Adobe). For and Fig. S2, images were viewed using a confocal system (MRC-1024; Bio-Rad Laboratories) mounted on a microscope (Eclipse E800; Nikon) and were captured using a 40×/1.3 NA or 60×/1.4 NA oil-immersion objective and LaserSharp software (Bio-Rad Laboratories). Brightness and contrast levels and color channels were adjusted using ImageJ () or Photoshop. For quantification of synaptic proteins, anti-Dlg labeling was used to outline type I boutons for quantification of synaptic protein levels (). Boutons on muscle 12 were scanned with steps of 0.7 μm. Using Amira software, the Dlg-stained type I boutons in each confocal slice were highlighted, and the mean pixel intensity of all slices of the highlighted boutons of each muscle 12 NMJ were computed. Background fluorescence in muscle areas adjacent to the boutons was quantified similarly, and the background was subtracted from the bouton values to yield the mean intensity of labeling in the boutons. The mean value from at least three mutant NMJs was then expressed as a percentage of the corresponding control value. The primary antibodies against presynaptic proteins were used at concentrations that produced strong specific labeling in the NMJ boutons without giving excessive background labeling in other tissues. Immunoprecipitation experiments were performed essentially as described previously (; ). The Dap160 band was detected by in-gel trypsin digest followed by peptide determination using a Q-STAR hybrid tandem mass spectrometer (MDS Sciex). Spectra were searched against a NCBI nonredundant database with MASCOT MS/MS Ions search (Matrix Science). The only other two Coomassie blue–stained bands that were successfully identified were keratin (likely a contaminant from the experimenter) and CG5214, a mitochondrial enzyme in the tricarboxylic acid cycle. For Western blots, rabbit anti-Eps15 was used at 1:2,500, rabbit anti–α-adaptin () at 1:200, rabbit anti-dynamin () at 1:300, and rabbit anti-Dap160 () at 1:500. For pull-down assays, Histidine (His)-tagged and GST fusion proteins were expressed in BL21(DE3)pLysS (Novagen) cells and purified from soluble fraction using Ni-NTA (QIAGEN) and glutathione (GE Healthcare) beads, respectively, before being used for binding assays. Ni-NTA agarose-coupled His-tagged UIM region of Eps15 (amino acids 1177–1253) was incubated with GST or GST-Ubiquitin (G76A; ) for 2 h at 4°C, washed extensively in PBS (0.01 M phosphate buffer, 0.0027 M KCl, and 0.137 M NaCl, pH 7.4) supplemented with 0.1% Triton X-100 and Complete protein inhibitor cocktail (Roche) and prepared for electrophoresis. For pull downs from fly head extracts, glutathione Sepharose–coupled GST, GST-3xEH (amino acides 1–403), and the GST-DPF region of Eps15 (amino acids 551–1001) were incubated with extracts (frozen heads homogenized in PBS supplemented with 2 mM CaCl, 1.5 mM MgCl, 1% Triton X-100, and Complete protein inhibitor cocktail; cleared soluble fraction adjusted to 0.1% Triton X-100 concentration) for 2 h at 4°C, washed extensively in the same buffer, and prepared for electrophoresis. Third instar electrophysiology was also performed essentially as described previously (, ). Larvae were dissected, and recordings were performed in modified HL3: 110 mM NaCl; 5 mM KCl; 10 mM NaHCO; 5 mM Hepes; 30 mM sucrose; 5 mM trehalose; 10 mM MgCl; and 0, 0.5, or 5 mM CaCl (as indicated in the text and figure legends). For 10-Hz stimulations, the above medium was modified to contain 4 mM MgCl and 5 mM CaCl. Recordings of low-frequency stimulations and 10-Hz stimulations were made from muscles with resting potential lower than −60 and −65 mV, respectively. For the FM1-43FX (Invitrogen) styryl dye uptake experiments, larvae were dissected on Sylgard plates and incubated in modified HL3 with 4 μM FM1-43FX, 5 mM CaCl, 90 mM KCl, and 25 mM NaCl for 10 min. Excess dye was then washed away with zero Ca solution for 10–15 min, and labeling was imaged using a 40×/0.75w Acroplan water-immersion lens on the LSM 510 confocal microscope. Data acquisition as well as data processing and quantification were performed as described previously (). The FM1-43FX dye uptake assay in was performed as above, but the data were acquired after fixation with 4% formaldehyde in PBS, imaged using a 60×/1.4 NA oil-immersion objective on the MRC-1024 confocal system, and quantified using NIH ImageJ. Data acquired using the two methods were comparable. All electrophysiology and FM1-43FX uptake was performed at room temperature (∼21–23°C), except when indicated otherwise. Larval fillets were prepared in HL3 without Ca and fixed in 4% paraformaldehyde/1% glutaraldehyde/0.1 M cacodylic acid, pH 7.2, or 3% paraformaldehyde/0.5% glutaraldehyde/0.1 M cacodylic acid, pH 7.2. Samples were postfixed in 2% OsO and stained in 2% uranyl acetate, dehydrated in alcohol, and embedded in Spurr's resin (Electron Microscopy Sciences) or Durcupan (Fluka). Ultrathin sections were cut with a diamond knife (Diatome) and stained with 2% uranyl acetate and lead citrate on grids. The grids were then visualized with an electron microscope (JEM-1010; JEOL) fitted with a charge-coupled device digital camera (Gatan). To study the morphology of stimulated boutons, larval fillets were incubated in HL3 with 5 mM Ca and 60 or 90 mM K for 10 min and fixed immediately. In stimulation-recovery experiments, stimulated NMJs were allowed to recover for 1 min in normal HL3 without Ca before fixation. Images were quantified by NIH ImageJ. Muscles 6 and 7 were dissected from third instar larvae were fixed in 4% paraformaldehyde and embedded in agarose (Sigma-Aldrich). Vibratome slices of the agarose blocks were incubated with guinea pig anti-Eps15 antiserum followed by secondary antibodies conjugated to 1.4-nm gold particles (Nanoprobes, Inc.). The immunogold labeling was silver enhanced using IntenSE Silver Enhancement kit (GE Healthcare), and samples were embedded in Durcupan ACM (Fluka) for ultrathin sectioning. Serial ultrathin sections were counterstained with uranyl acetate and lead citrate and examined in a microscope (Tecnai 12; FEI Company). Images were quantified using NIH ImageJ, and statistical evaluation was performed using Excel (Microsoft). Note that the silver enhancement technique works by the precipitation of silver particles around the gold particles, resulting in the formation of irregularly shaped black precipitates. Fig. S1 illustrates the domain structure of Eps15 and shows that the -null alleles do not express detectable amounts of Eps15. Fig. S2 shows that mBSA endocytosis and dynamin levels are both normal in -null hemocytes. Fig. S3 shows that dynamin and Dap160 levels are not substantially altered in the larval brains of mutants. Fig. S4 shows that the EH, DPF, and UIM domains of Eps15 bind to Dap160, α-adaptin, and ubiquitin, respectively. Table S1 shows mass spectroscopy data of peptide sequences corresponding to Dap160, which coimmunoprecipitated with Eps15 (A). Online supplemental material is available at .

The asymmetric distribution of membrane proteins in different cell surface domains is a feature common to many types of polarized cells, including epithelia, neurons, and immune cells (). In epithelial cells, membrane proteins are segregated into functionally and structurally different apical and basolateral membrane domains. Considerable evidence has accumulated that the Golgi complex () and the recycling endosome () regulate sorting of apical and basolateral membrane proteins into separate vesicles in the exocytic and endocytic (recycling) pathways. However, less is known about mechanisms that specify post-Golgi vesicle delivery and fusion with the correct membrane domain. In fully polarized cells, the delivery of basolateral membrane proteins from the TGN and recycling endosomes to the plasma membrane may be regulated at several steps, including long-range vesicle delivery and membrane tethering and fusion. Vesicles travel from the region of the TGN to the plasma membrane along microtubules (). Upon arrival at the plasma membrane, vesicles are thought to interact with the exocyst (Sec6/8 complex; ) which is a multiprotein complex that may tether vesicles to the membrane before their fusion by a complex of vesicle-soluble -ethylmaleimide–sensitive factor attachment protein receptors (v-SNAREs) and target (t)-SNAREs (; ). In fully polarized MDCK cells, the exocyst is localized to the apex of the lateral membrane, and addition of function-blocking Sec8 antibodies inhibited basolateral, but not apical, vesicle delivery to the plasma membrane (). In these cells, the t-SNARE syntaxin 4 is also localized to the basolateral plasma membrane (), and inhibition of t-SNARE function using botulinum neurotoxins blocked basolateral vesicle delivery (). It has been suggested that the exocyst and t-SNAREs compose a vesicle “targeting patch” that specifies basolateral vesicle delivery to sites of cell–cell adhesion (), but this hypothesis has never been directly tested. Cell adhesion to other cells and the extracellular matrix is important in the generation of epithelial cell surface polarity. In nonpolarized fibroblasts () and single MDCK cells grown in suspension culture in the absence of cell contacts (), apical and basolateral membrane proteins are intermixed on the cell surface, though they are sorted from each other in the exocytic pathway. Upon cadherin-mediated cell–cell adhesion in fibroblasts () and suspension-grown MDCK cells (), basolateral membrane proteins are restricted to cell–cell contacts, whereas apical proteins accumulate on the unbounded membrane facing the growth medium. Cell–cell adhesion is also critical for the correct orientation of asymmetric cell divisions in the stem cells and maintenance of the stem cell–niche interface (; ; ). Adhesion to the extracellular matrix also plays a role in cell polarization, as laminin is required for correct apical pole orientation in three-dimensional epithelial cysts () and induces β-casein secretion from single mammary epithelial cells (). Although these studies are suggestive of a role for extracellular contacts in the orientation of different membrane domains in fully polarized cells, a link between these spatial cues and localized vesicle delivery, and the mechanisms involved have not been investigated directly. We have taken a direct approach to these problems by examining the distributions of aquaporins (AQP) during initial cell–cell adhesion in MDCK cells. AQPs are a structurally homologous family of channel proteins that facilitate the movement of water, glycerol, and urea across different membrane domains in polarized epithelia. AQP3 has an N-terminal basolateral sorting signal () and localizes to the basolateral membrane in multiple epithelial tissues (), whereas AQP5 has a C-terminal signal for targeting to or retention in the apical membrane () and localizes to the apical membrane of secretory tissues (; ) . We show that post-Golgi vesicles containing AQP3, but not AQP5, are targeted directly to the site of initial E-cadherin–mediated cell–cell contacts. Components of a putative lateral targeting patch localize rapidly and independently of each other to sites of cell–cell adhesion, where they function as a holocomplex that specifies basolateral vesicle delivery to cell–cell contacts. These results have broad implications for how cell polarity may be initiated by extrinsic spatial cues in a wide variety of differentiated and stem cells. Protein distributions during cell–cell adhesion were examined by high resolution time-lapse imaging in MDCK cells cotransfected with tandem-dimer red fluorescent protein (tdRFP)–tagged E-cadherin and either EGFP-tagged AQP3 or EGFP-tagged AQP5. EGFP-tagged forms of E-cadherin (), AQP3, and AQP5 (; ) localize correctly. Note that the expression of tagged forms of E-cadherin did not increase overall levels of E-cadherin in cells because of a decrease in the level of endogenous E-cadherin commensurate with the level of ectopic E-cadherin (). E-cadherin rapidly accumulated at sites of initial cell–cell adhesion and spread laterally as the surface area of the contact increased (; Video 1, right; and Video 2, right, available at ), as reported previously (). Shortly after the accumulation of E-cadherin was detected at the cell–cell contact, increased levels of AQP3 were detected at the same locations (; and Video 1, left). Thereafter, the accumulation of AQP3 was strikingly coincident with that of E-cadherin at all times during cell–cell adhesion. During final compaction of the cell–cell contact, the brightest regions of AQP3 were superimposed onto plaques of E-cadherin that coalesced at the edges of the cell–cell contact. Although some AQP5 (; and Video 2, left) was detected at the cell–cell contact with E-cadherin (; and Video 2, right) at the beginning of the time-lapse imaging, it rapidly disappeared, and as the cell–cell contact spread, AQP5 remained diffusely distributed over the entire surface and did not accumulate at the cell–cell contact. Quantitation confirmed that AQP3, but not AQP5, accumulated at cell–cell contacts (). Collectively, these results demonstrate that a basolateral protein AQP3, but not a homologous apical protein AQP5, precisely coaccumulated with E-cadherin during the very earliest stages of cell–cell adhesion. Although both AQP3 and AQP5 are normally expressed in polarized epithelial cells, we tested whether the difference in localization of ectopic AQP3 and AQP5 to nascent cell–cell contacts was the same as that of endogenous basolateral and apical membrane proteins in MDCK cells. We plated cells at low density, fixed them after 1 h when initial cell–cell contacts had formed, and stained for two endogenous membrane proteins: the basolateral membrane protein NaK-ATPase and the apical membrane protein gp135 (podocalyxin; ). NaK-ATPase (), like AQP3 (; and Video 1, left), accumulated precisely with E-cadherin at the cell–cell contact. On the other hand, gp135 (), like AQP5 (; and Video 2, left), did not accumulate at cell–cell contacts and was diffusely distributed over the cell surface. Hence, both exogenously expressed basolateral AQP3 (; and Video 1, left) and endogenously expressed NaK-ATPase () coaccumulated with E-cadherin at cell–cell contacts, whereas apical AQP5 (; and Video 2, left) and gp135 () did not. Protein accumulation at nascent cell–cell contacts is dependent on the balance between delivery from intracellular compartments and lateral diffusion in the plasma membrane. We initially designed experiments to directly observe delivery of newly synthesized AQP3 from the Golgi (; and Video 3, left, available at ). We created a stable cell line expressing AQP3 tagged with photoactivated GFP (AQP3-PAGFP), which allowed us to activate a small intracellular pool of AQP3-PAGFP in the Golgi and follow its fate by time-lapse imaging in a blank background; note that the signal from the EGFP-tagged protein is too bright at cell–cell contacts to allow visualization of increased accumulation after release of protein from the Golgi. To synchronize cell surface delivery of protein, AQP3-PAGFP was accumulated in the Golgi by a 19°C block and released by shifting to 37°C. Although photoactivation of AQP3-PAGFP at the Golgi could activate AQP3-PAGFP in other membrane compartments localized close to the Golgi (e.g., endosomes), it has been shown that the 19°C block causes the accumulation of newly synthesized protein in the TGN (). A spot of AQP3-PAGFP over the Golgi was laser activated and followed by time-lapse imaging for a short period (<10 min) in pairs of cells forming cell–cell contacts (; Video 3, left; and see ). The intensity of the activated pool of AQP3-PAGFP decreased rapidly around the Golgi, and after a short delay (<1 min), there was a concomitant increase in AQP3-PAGFP at the cell– cell contact; note that AQP3-PAGFP initially accumulated at the membrane immediately adjacent to the activated spot in the Golgi and then more distally during later times. We quantified the fluorescent intensities of equal areas of AQP3-PAGFP at the cell–cell contact (, blue) and at the noncontacting plasma membrane (, red) that were equidistant from the initial photoactivated spot. The intensity of AQP3-PAGFP fluorescence increased in the cytosol (not depicted) and at the site of cell–cell adhesion (, blue) but not at the noncontacting plasma membrane (, red); this is consistent with direct delivery of AQP3-PAGFP from the Golgi to the site of initial cell–cell contact. In contrast to AQP3-PAGFP, we found that AQP5-PAGFP activated in the Golgi did not accumulate at the site of cell–cell adhesion (; and Video 3, right). Individual post-Golgi carriers could be observed leaving the initial region of AQP3-PAGFP activation in the direction of the site of initial cell–cell contact and traveled all the way to the contact, where they disappeared (; Video 4; and Video 5, available at ). These vesicles were observed by either epifluorescence microscopy ( and Video 4) or total internal reflection fluorescent (TIRF) microscopy ( and Video 5). Because the volume of cytoplasm in thin lamellipodia forming cell–cell contacts is low, it is likely that the disappearance of these vesicles is a consequence of their fusion with the plasma membrane rather than their diffusion out of the focal plane. These post-Golgi carriers traveled in linear paths, with several pauses and changes of direction (; Video 4; and Video 5), at speeds averaging 0.2–0.3 μm/s, which correlates well with vesicle movements generated by the microtubule motor kinesin (). Approximately 15% of the loss of AQP3-PAGFP fluorescence in the Golgi was detected at the cell–cell contact after 2 min (). This correlates well with measurements by cell surface biotinylation that 20% of the low density lipoprotein receptor was delivered from the Golgi to the plasma membrane after release from a 19°C block in polarized MDCK cells (). We conclude that basolateral membrane AQP3, but not the equivalent apical membrane AQP5, is targeted directly from the Golgi to initial sites of E-cadherin–mediated cell–cell adhesion. Although the results described in the previous paragraph show direct delivery of AQP3 from the Golgi to cell–cell contacts, protein accumulation could also be affected by diffusion between the noncontacting plasma membrane and the cell–cell contact. To measure protein diffusion, AQP3-PAGFP was activated in a small spot within the cell–cell contact (; and Video 6, available at ), and the fluorescence intensity of the spot was measured over time. The diffusion coefficient of AQP3 was fast at initial contacts between cell pairs (t = 19 ± 8 s; ; and Video 6, right); for comparison, we measured AQP3 diffusion in confluent monolayers that had begun to establish full polarity over a period of 24 h and found that AQP3 diffusion was slower (t = 143 ± 46 s; ; and Video 6, left), indicating a change in AQP3 organization in the membrane during development of cell polarity. To test if AQP3 was retained at the contact or was free to diffuse into the noncontacting plasma membrane, we performed fluorescence loss in photobleaching (FLIP) of AQP3-EGFP at the plasma membrane adjacent to the contact (). In this experiment, MDCK cells stably expressing AQP3-EGFP were mixed with nonexpressing cells to examine protein diffusion in only one membrane of the cell pair at the cell–cell contact and in the presence of cyclohexamide to eliminate the addition of newly synthesized protein to the plasma membrane. The intensity of a small spot of fluorescence at the contact was quantitated over time and showed that AQP3-EGFP fluorescence dissipated rapidly (), consistent with rapid diffusion within the contact and the surrounding noncontacting plasma membrane. Collectively, these results show that AQP3 diffuses rapidly in the plane of the membrane and, hence, accumulation of AQP3 at cell–cell contacts must require rapid, direct, and sustained delivery of AQP3 from the Golgi. We tested whether components of the putative lateral targeting patch, consisting of the exocyst and SNARE complexes, colocalized to initial cell–cell contacts and functioned there in the delivery of AQP3 from the Golgi to those contacts. In single MDCK cells, Sec6 and Sec8, two core components of the exocyst (), localized in the cytosol with the cortical actin bundle (). Upon initiation of cell–cell adhesion, Sec6 and Sec8 (not depicted) localized to the plasma membrane at sites of initial cell–cell adhesion, although some intracellular staining remained (). In compacted contacts between cells, Sec6 and Sec8 (not depicted) localized along the length of the cell–cell contact and at higher concentrations at the edges of the contact (), similar to the distribution of E-cadherin (). The basolateral SNARE syntaxin 4 localized to the lateral plasma membrane in polarized cells () and in clusters at the plasma membrane in single cells. Syntaxin 4 also localized to initial cell–cell contacts (), and its distribution appeared to coincide with the distributions of E-cadherin () and the exocyst (compare with ). Thus, two components of the lateral targeting patch are recruited rapidly to the plasma membrane at cell–cell contacts after initiation of cell–cell adhesion. We tested whether recruitment of components of the lateral targeting patch and microtubules were interdependent by examining their distribution after disruption or inhibition of each of the components. We initially tested whether localization of the exocyst and syntaxin 4 was microtubule dependent. MDCK cells stably expressing E-cadherin–tdRFP were plated in media containing 5 μM Ca to inhibit cell–cell adhesion. Nocodazole was added to depolymerize microtubules (; and Fig. S1 A, available at ), and 1.8 mM Ca was added to the media to initiate the formation of cell–cell contacts (; and Fig. S1 A). The disruption of microtubules did not impair the formation of cell–cell contacts as visualized by the distribution of E-cadherin (; and Fig. S1 A), nor the accumulation of Sec8 () or syntaxin 4 () at cell–cell contacts. We next tested whether accumulation of the exocyst and SNARE complexes at forming cell–cell contacts was interdependent. The SNARE complex was disrupted in single cells by injecting tetanus toxin, which cleaves VAMP2 and VAMP3/cellubrevin (), and protein distributions were examined after the addition of 1.8 mM Ca to initiate cell–cell contact formation. Tetanus toxin did not disrupt contact formation, as visualized by E-cadherin localization at cell–cell contacts (; and Fig. S1 B), nor the accumulation of Sec8 () or syntaxin 4 () to the forming contact. Injection of function-blocking Sec8 antibodies, which caused Sec8 to relocalize from the cell–cell contact to the cytoplasm (Fig. S1 and Fig. S2, available at ), did not inhibit either E-cadherin–mediated cell–cell contact formation ( and Fig. S1 C) or syntaxin 4 accumulation at cell–cell contacts (). Because both the Sec8 and syntaxin 4 antibodies are mouse monoclonals, we could not directly discriminate between the distributions of the two proteins by immunofluorescence microscopy in this experiment. However, the plasma membrane staining in is most likely syntaxin 4 staining, as microinjection of Sec8 antibodies caused all of the plasma membrane Sec8 to be redistributed into the cytoplasm (Fig. S1 C and Fig. S2). Thus, we conclude that the components of the lateral targeting patch (the exocyst and syntaxin 4) accumulate independently of each other and of microtubules at forming cell–cell contacts. To investigate whether the exocyst plays a role in the delivery of AQP3 to the site of initial cell–cell adhesion, we injected cells forming cell–cell contacts with function-blocking Sec8 antibodies (). We then synchronized exocytosis in the Golgi with the 19°C block and laser-activated trapped AQP3-PAGFP, as described in Materials and methods. Immunofluorescence of Sec8 antibody–injected cells showed that Sec8 was localized in the cytoplasm and not at cell–cell contacts (Fig. S1 C and Fig. S2). In contrast to noninjected cells (; ; and Video 3, left) and cells injected with nonspecific IgG (Fig. S3, A and B; and Video 7, available at ), AQP3-PAGFP released from the Golgi in cells injected with Sec8 antibodies did not accumulate at sites of initial cell–cell contact (; and Video 8). Thus, inhibition of exocyst function at the plasma membrane was sufficient to block delivery of AQP3-PAGFP vesicles to cell–cell contacts even though syntaxin 4 localized to cell–cell contacts under these conditions (). To test the involvement of the lateral membrane SNARE complex in AQP3-PAGFP vesicle fusion at sites of cell–cell contact, we injected cells with tetanus toxin. Results show that AQP3-PAGFP released from the Golgi did not accumulate at the cell–cell contact in cells injected with tetanus toxin (; and Video 9, available at ). Thus, the SNARE complex is also essential for the fusion of newly synthesized AQP3-PAGFP to sites of initial cell–cell contact. Collectively, these results demonstrate that the exocyst and t-SNAREs are rapidly and precisely recruited to cell–cell contacts, and that both complexes are required for AQP3 delivery and accumulation at initial sites of E-cadherin–mediated cell–cell adhesion. Vesicles travel from the region of the Golgi via microtubules to the plasma membrane () and to sites of established cell–cell contacts (). Although microtubules undergo a complex reorganization as epithelial cells polarize (; ), microtubules in nonpolarized epithelial cells are initially organized in an array similar to that in fibroblasts, in which they extend radially from the centrosome toward the periphery, where they impinge on initial E-cadherin–mediated contacts between cells (; ). We tested whether microtubules are important in the delivery of newly synthesized AQP3 from the Golgi to sites of initial cell–cell contact. AQP3-PAGFP was accumulated in the Golgi at 19°C, and microtubules were depolymerized with nocodazole before the shift to 37°C and laser-activation of a small spot of AQP3-PAGFP in the Golgi in cells forming contacts; note that cell–cell contacts were not disrupted under these conditions (; and Fig. S1 A). We measured the fluorescence intensity of AQP3-PAGFP at the site of cell–cell contact and the noncontacting plasma membrane and did not detect an increase at either site (; and Video 10, available at ), indicating that, in the absence of microtubules, AQP3 was not transported from the Golgi region to the site of initial cell–cell adhesion, even though both the exocyst and SNARE complexes remained localized to cell–cell contacts. The establishment of cell surface polarity is common to many cell types and requires the accumulation of specific proteins in spatially restricted regions of the plasma membrane that uniquely contribute to cell and tissue functions (). In transporting epithelia, the spatial segregation of proteins to the apical and basolateral membrane domains is the basis for the formation of ion and solute gradients across the epithelial monolayer (). Similarly, the spatial restriction of subsets of proteins to neuronal (; ) and immunological synapses (; ) are critical to the functions of these cell types. Studies have shown that cell–cell adhesion coincides with the development of cell surface polarity in epithelia (; ) and other cell types () and is required to maintain stem cell–niche interactions (; ) and the correct orientation of asymmetric stem cell divisions (; ; ; ). Important problems are to identify how these spatial cues initiate formation of these cell surface domains, and the downstream machinery that regulates the type of protein that is delivered to and integrated into the membrane domain. Polarized transporting epithelia provide a useful system to approach these problems, because the mechanisms involved in protein sorting have been well described, and cell–cell adhesion can be easily manipulated and imaged. Apical and basolateral membrane proteins appear to be constitutively sorted from each other in the Golgi and/or recycling endosome (; ), whereas their distributions are intermixed at the plasma membrane in nonpolarized cells (; ). E-cadherin–mediated cell–cell adhesion appears to provide a spatial cue for cells to distinguish an unbounded (apical) from a bounded (basolateral) surface and to accumulate apical and basolateral membrane proteins in the correct surface (), but the mechanisms linking E-cadherin to protein sorting and redistribution to different plasma membrane domains are unknown. To test the role of E-cadherin–mediated cell–cell adhesion in protein trafficking and cell surface distribution, we took a direct approach by imaging the delivery of two highly homologous apical (AQP5) and basolateral (AQP3) membrane proteins to sites of initial cell–cell adhesion. Because apical and basolateral proteins are presorted before their arrival at the cell surface (), including AQPs (; ), we could ask whether plasma membrane sites at the earliest stages of cell–cell adhesion become specialized for the delivery of basolateral rather than apical vesicles and, if so, investigate the nature of the cellular machinery involved. E-cadherin–tdRFP was generated from EGFP-tagged E-cadherin (). Photoactivatable constructs were generated by subcloning PAGFP (a gift from J. Lippincott-Schwartz, National Institutes of Health, Bethesda, MD) from the N1 vector (CLONTECH Laboratories, Inc.) into AQP3-EGFP–N2 and AQP5-EGFP–N2 (gifts from A. Aperia, Karolinska Institutet, Stockholm, Sweden; ) using BSeRI. MDCK GII cells were transfected with Effectene (QIAGEN) and selected with G418 (Invitrogen). Cells stably expressing E-cadherin–tdRFP (a gift from S. Yamada, University of California, Davis, Davis, CA) were transfected with AQP3-EGFP and subjected to four rounds of FACS to generate cells stably expressing both E-cadherin–tdRFP and AQP3-EGFP. Cells stably expressing E-cadherin–tdRFP were transiently transfected with AQP5-EGFP to generate double-expressing cells. All constructs were stably expressed in MDCK GII cells without any apparent change in phenotype. Correct localization was verified by the imaging (see the following section) of cells grown on filters for 10 d. Cells were seeded on collagen-coated coverslips and allowed to attach and spread for at least 1 h before imaging. Time-lapse imaging was performed in phenol red–free DME media (Sigma-Aldrich) with 10% fetal bovine serum (Atlas Biologicals) and 25 mM Hepes (Invitrogen) using the Marianas system (Intelligent Imaging Innovations) with a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) equipped with a camera (Photometrics CoolSNAP; Roper Scientific), a laser system (MicroPoint FRAP; Photonic Instruments, Inc.), and a TIRF system (TIRF Slider; Carl Zeiss MicroImaging, Inc.). An α Plan-FLUAR 1.45 oil (for TIRF) and a 100× Plan-APOCHROMAT 1.40 oil differential interference contrast (for epifluorescence) objectives were used (both obtained from Carl Zeiss MicroImaging, Inc.). Images were analyzed using Slidebook (Intelligent Imaging Innovations) or ImageJ (available at ) software. For initial cell adhesion, cells starting to form contact were imaged every minute for 5 h. To measure the half-time of intensity recovery, AQP3-PAGFP cells were plated 24 h (forming monolayers) or 1 h (initial contacts) before analysis. Cells were imaged every 3 s for 10 min. After the first frame, the time lapse was paused, and a small area of the contact was activated. The intensity profiles were analyzed for the maximum intensity recovery (percentage) and fitted to a single exponential function up to 10 min after photoactivation to extract the half time of intensity recovery (t). For assessment of adhesion after treatments, single cells expressing E-cadherin–tdRFP were plated in low calcium media containing 5 μM Ca to inhibit cell–cell adhesion, and different manipulations (nocodazole treatment, tetanus toxin, or Sec8 antibody injection; see the following section) were performed on single cells; 1.8 mM Ca was added back to the growth medium to initiate cell–cell adhesion, and the amount of E-cadherin at cell–cell adhesion was examined. 33 μM nocodazole was added for the last 30 min before calcium readdition. Sec8 antibodies (equal mixture of 2E9, 2E12, 5C3, 10C2, and 17A10 hybridoma supernatants; ) were concentrated 16 times on a column (Microcon 50.000 MW; Millipore) and washed 5 times with microinjection buffer (10 mM Hepes, 140 mM KCl [pH 7.4]). The mixture was diluted five times for injection into microinjection buffer. 60 ng/μl tetanus toxin (needle concentration; List Biological Laboratories, Inc.), Sec8 antibodies, and 1 mg/ml rabbit IgG (needle concentration) were microinjected into one cell of a duplet using a microinjection system (Eppendorf). 0.5–1 mg/ml Texas red or FITC-labeled dextran were coinjected to identify injected cells. Newly synthesized protein was accumulated at the Golgi, released, and imaged as described in the previous section. After imaging, Sec8 antibody–injected cells were fixed, permeabilized, and stained with secondary goat anti–mouse Cy5-conjugated antibody. Cells were seeded on collagen-coated coverslips for 1 h at subconfluent density, fixed in 2% paraformaldehyde, and permeabilized with Triton X-100. For staining of microtubules, cells were fixed in ice-cold methanol at −20°C. Primary antibodies were as follows: monoclonal Sec6 (clone 9H5) and Sec8 (clone 8F12; ), Syntaxin 4 (BD Biosciences), DM1α tubulin (Sigma-Aldrich), Gp135 3F2/D8 (a gift from G.K. Ojakian, State University of New York Health Science Center, Brooklyn, NY), and polyclonal NaK-ATPase (a3NKA; ) and ZO1 (Zymed Laboratories). Images were obtained using the Marianas system, except for AQP5-EGFP images, which were obtained with a microscope (model IX-70; Olympus). The AQP5-EGFP images were processed using deconvolution software (DeltaVision; Applied Precision) on a workstation (Silicon Graphics, Inc.). Fig. S1 depicts E-cadherin localization in adhering cells after a change in media containing 5 μM Ca to 1.8 mM Ca in the presence of nocodazole (A), tetanus toxin (B), or Sec8 function-blocking antibody (C). Fig. S2 shows a retrospective stain of Sec8 localization after Sec8 antibody injection. Fig. S3 depicts the distribution of Golgi-accumulated AQP3-PAGFP released from the Golgi after a shift in temperature from 19°C to 37°C in the presence of rabbit IgG. Video 1 provides a time-lapse movie of two single cells making initial cell–cell contact. The cells are stably expressing AQP3-EGFP (green) and E-cadherin–tdRFP (red). Video 2 shows a time-lapse movie of two single cells making initial cell–cell contact. The cells are transiently expressing AQP5-EGFP (green) and stably expressing E-cadherin–tdRFP (red). Video 3 provides a time-lapse movie of cell pairs stably expressing AQP3-PAGFP (left) and AQP5-PAGFP (right) after release from a 19°C temperature block. AQP3-PAGFP and AQP5-PAGFP were photoactivated in the Golgi region, and images were captured every 3 s for 10 min. Video 4 shows a time-lapse movie of a cell pair stably expressing AQP3-PAGFP after release from a 19°C temperature block. AQP3-PAGFP was photoactivated in the Golgi region, and images were captured every 3 s. Video 5 provides a time-lapse movie using TIRF microscopy of a cell pair stably expressing AQP3-PAGFP after release from a 19°C temperature block. Video 6 shows a time-lapse movie of a cell pair stably expressing AQP3-PAGFP. Cells that had formed confluent monolayers over a 24-h time period are shown on the left, and initial cell–cell contact is shown on the right. AQP3-PAGFP was photoactivated at a small point within the cell–cell contact. Video 7 provides a time-lapse movie of a cell pair stably expressing AQP3-PAGFP after release from a 19°C temperature block. One cell was injected with rabbit IgG before a 19°C temperature block. Video 8 shows a time-lapse movie of a cell pair stably expressing AQP3-PAGFP after release from a 19°C temperature block. One cell was injected with Sec8 antibodies before a 19°C temperature block. Video 9 provides a time-lapse movie of a cell pair stably expressing AQP3-PAGFP after release from a 19°C temperature block. One cell was injected with tetanus toxin before a 19°C temperature block. Video 10 shows a time-lapse movie of a cell pair stably expressing AQP3-PAGFP after release from a 19°C temperature block. Cells were treated with nocodazole during the last 30 min of a 19°C temperature block. Online supplemental material is available at .

As a kid, I liked to mix stuff up in the sink. I grew up on a farm, and when we'd butcher the chickens, I'd get some of the internal organs and cut them up. Coming out of high school, I was really interested in molecular genetics and the ability to clone DNA. Our senior year we got to do independent research projects. So I did a genetics project. There were two reasons. One was the cytoskeleton, which I thought was just amazingly cool and an important system to study. The other was Rong herself. She was a really dynamic person, and in retrospect I couldn't have asked for a better mentor at that stage. To be honest, at first I thought it was kind of obscure, and that really appealed to me. This was pre-Arp2/3 complex, and at that point there was really nothing known about how actin was nucleated inside the cell. But then it turned out to be about the hottest area in cell biology at the time. It exploded within the next couple of years. A lot of that was happening around us and in our lab; Marc Kirschner and Tim Mitchison's labs were upstairs. So it was just this incredibly vibrant and exciting time where all these important discoveries were being made. I think we had two advantages: we had the genetics; and we could do some of the physiology of the actin assembly that other people couldn't do because we were working in yeast. We could actually look inside the cell at what was happening with gain- and loss-of-function approaches. We had identified most of these molecules from a more physiological standpoint and then went on to find their mechanism. I knew I still really liked the whole area of cytoskeleton morphogenesis, but I also wanted to be exposed to a lot more. The Fuchs lab was the ideal place for me because it did have that cytoskeletal niche, but it also has a broad range of interests in transcription, stem cells, and development. I thought it'd be an ideal place to be exposed to all of that and still have an area of comfort. And she was actually one of the few people who were okay with me taking off for six months to go to Nepal and teach. A group of Nepali doctors wanted to start a medical school in Nepal, but they needed some basic science faculty, so they recruited a bunch of us to come over. In the first six months, there were three graduate students and one post-doc from Harvard. We all knew Cliff Tabin, who was running it; he was the chair of my graduate program. We did everything you need to do to set up the school. We interviewed the students, set up the library, stuff like that, and then we taught them for the first four months. It was a problem-based learning curriculum, so there were some didactic components, but a lot of it was getting the students to start thinking and interacting and working through problems. It was fantastic. Everyone should take time off after grad school. That was a really good experience because I got to completely develop a project from scratch. I began by looking at tissues, which was new for me. I'd worked with yeast and was used to looking at single cells. I think because of that, I looked at the tissue a little bit differently; I looked a lot more at what the cells were doing. During that process, I noticed that cells were dividing in different orientations during development and thought that that was a really interesting, important process in the development of the epidermis. It's one potential way that stem cells could both renew themselves and generate new stem cells, and at the same time contribute transit-amplifying or differentiated cells. It's not necessary that stem cells divide asymmetrically, but it is a really elegant way that they can couple those two characteristics. Asymmetric divisions are also exceptional because they help control epithelial tissue morphogenesis by promoting stratification. And while developing reagents for that project, I was also able to make observations that opened up additional areas of study. The role of microtubules in differentiated cells and how they reorganized, which was not part of what I was originally doing, was a really happy extension of that. In some ways, I think I'm more interested in that story. Many people right now are interested in asymmetric cell division, but the role of cytoskeletal reorganization during cell differentiation is an area that's just beginning, and there are a lot of open questions. It's so fundamental. Form is function. That's clear in differentiated cell types, and yet we have so little idea about how the cytoskeleton reorganizes when cells differentiate, or even what the functions are of most of the cytoskeletal elements in differentiated cells. We've become a lot more interested in cytoskeletal remodeling downstream of the desmosome and trying to identify the complement of proteins that are recruited by the desmosome. What's most interesting there is that it looks like there's a group of centrosomal proteins. They're at the centrosome normally, and they're brought to the desmosome when the cells differentiate. We're interested in this family of proteins and how they coordinate the cytoskeleton. In terms of asymmetric cell division, we are interested in spindle reorientation and in questions like, What are the forces that are acting on the spindle to allow spindle orientation, and how does the spindle anchor to the cell cortex? This is really an open question. The third area is more of a tissue biology question. In most asymmetric cell division systems, the cells consistently divide asymmetrically. But in the epidermis, cells can divide either symmetrically or asymmetrically. We're interested in how they integrate chemical and mechanical signals across the epidermis to make the decision whether to divide asymmetrically or symmetrically to generate an epidermis that's the right thickness. The skin is really the only well-developed system where cells are making this choice between asymmetric and symmetric divisions. Understanding how that decision is made I think is going to be a really interesting area of biology. The other big plus is the fact that we can go back and forth between the cultured cells and the in vivo setting to understand both the mechanism and the physiology of the cytoskeletal remodeling. The intestines are a great extension of that, because they're also an epithelial cell, but they're very different morphologically and functionally. So we can take what we learn in the skin and see whether it's generalizable to other epithelial cell types. In terms of the skin, one nice thing is you've got a proliferative department, and you've got a differentiated compartment. So there's also the really interesting question of, How does one transit to the other? What are the morphological consequences of that, and how do those occur? Ultimately, it was pretty much a gut feeling. I knew I fit there. The big draws were the people. It always has to come down to that. You want to surround yourself with people that are going to stimulate you, but also people that you like and want to interact with on a daily basis. And North Carolina was a really easy adjustment. I adopted a dog, I got my house, I'm kind of set. But I get back to New York for sushi every now and then.

Cell adhesion molecules of the Ig superfamily (IgSF) typically have one or more Ig-like domains in their extracellular N-terminal region that are implicated in molecular recognition and one cytoplasmic C-terminal region that functions in signal transduction pathways. The Ig-like modules near the N terminus can form linear rods when arrayed in series and are sufficient for homophilic and heterophilic binding (). They mediate cis-interactions in the plane of the membrane and trans-interaction on opposing cell membranes. In addition, these proteins share common intracellular binding partners, which enables cross talk with other cell surface molecules. These features make IgSF proteins ideal components of cell–cell junctions and cell surface receptors. Cell adhesion molecules of the Ig family are highly conserved proteins. As they structurally resemble molecules of the adaptive immune system (e.g., antibodies and T cell receptors), genes that encode junctional adhesion molecules (JAMs), cortical thymocyte marker of (CTX), and nectins were considered to be “fossil” genes that later gave rise to essential elements of the adaptive immune system (). Indeed, members of the JAM, CTX, and nectin subgroups are expressed on circulating lymphocytes and leukocytes (; ). In vertebrates, they have the propensity to serve as virus receptors at endothelial or epithelial barriers. Their function at the cell junctional complex is not simply that of gatekeepers. They can also transduce signals at the cell membrane, maintain cell polarity, and mediate cell migration. JAM, CTX, and nectin molecules are engaged in a wide spectrum of cellular events ranging from viral infections and leukocyte transmigration to spermatogenesis. They are also essential to the central nervous system (). Here, we give a full account of the multifaceted characteristics of JAM/CTX and nectin family molecules. By integrating these facts with some of our recent findings, we bring forth a novel understanding of germ cell migration across the seminiferous epithelium during spermatogenesis, which can be tested in future experiments. italic #text Cell adhesion molecules of the Ig family localize at the subapical surface of polarized epithelial cells. Surprisingly, almost all members of the JAM/CTX and nectin family mediate viral entry and spread (). JAM-A is a receptor for reoviruses (); mammalian reoviruses of serotype 1, 2, and 3 and their respective field strains all bind to JAM-A (). Viral interaction with JAM-A triggers NF-κB activation and cell apoptosis and may be a defense mechanism or an innate immune response before the start of viral replication (; ). JAM-A also has been identified as a receptor for feline calicivirus (). Nectins also serve as viral receptors and have been reviewed recently (; ; ). Both nectin-1 and nectin-2 were originally isolated as poliovirus receptor (PVR)–related proteins, PRR-1 and PRR-2, respectively (; ). Later they were shown to be receptors for α-herpes virus rather than poliovirus and, hence, were renamed HveC and HveB (). Human nectin-like molecule-5 (hNecl-5) is known as the PVR (; ) and mediates entry of porcine pseudorabies virus as well as bovine herpesvirus 1 (). Nectin-1 mediates herpes simplex virus (HSV) infection in a wide range of cell types, including fibroblasts, primary sensory neurons, and trabecular meshwork cells of the human eyes (; ). This explains the pathogenicity of HSV-1 in these tissues. Access to nectin-1 at the apical surface of polarized cells contributed substantially to HSV infection in vitro (). The clustering of nectin-1 on the membrane protrusions of CHO cells facilitates HSV-1 attachment and a subsequent phagocytosis-like virus uptake (). CAR is best known for its role as a virus receptor (). As a member of the CTX protein family (), CAR mediates viral attachment and spread for coxsackie virus group B and adenovirus groups 2 and 5. The availability of CAR on the cell surface is a determining factor for susceptibility to adenoviral gene delivery (). However, gene delivery to differentiated epithelia is largely unsuccessful because CAR is sequestered in the intercellular junctions between columnar shaped cells, and thus the receptors are inaccessible to adenovirus entering from the apical surface (). How viruses reach receptors that are located inside intercellular junctions is an intriguing question that several studies have sought to address. Studies confirming JAM-A as a reovirus receptor were conducted on cell cultures without tight junction structures (). It remains unclear how viruses gain entry to JAM-A in the subapical regions of tight junctions in vivo (). Nectin-1, when confined to adherens junctions, is not easily accessible to virus either. Release of nectin-1 to the apical cell surface can greatly enhance cell susceptibility to HSV infection (). A novel strategy has been proposed to explain coxsackie virus invasion in the absence of cell surface receptor CAR (). To initiate an infection, group B coxsackie virus first interacts with a secondary receptor, decay-accelerating factors, on the cell surface. This in turn activates nonreceptor protein tyrosine kinases of the Abl and Src families, including Fyn. Abl then triggers Rac-dependent actin rearrangement and opens up tight junctions. Once the coxsackie virus reaches tight junctions, viral fiber knobs interact with CAR, replacing the original CAR–CAR homodimers. Viral particles are subsequently internalized by host cells via the caveolin pathway. It is possible that reoviruses, adenoviruses, and α-herpes viruses all conspire to use a secondary receptor on the cell surface to sneak into tight junctions. Interestingly, for almost all viral pathogens, the interfaces of IgSF molecules for binding to viral surface proteins overlap extensively with the domains that mediate homophilic trans-interaction inside junctional complexes (; ; ; ). It is believed that viral ligands perturb intercellular junctional complexes partly by competing with the homophilic interaction of cell adhesion molecules (). Accumulating evidence has pointed to the importance of IgSF cell adhesion molecules in mediating cell migration. They are present not only on endothelial and epithelial cells that are forming junctional barriers but also on circulating leukocytes and platelets. For example, JAM-A and -C proteins are expressed by platelets, neutrophils, monocytes, and lymphocytes (; ; ). JAM-like protein was also detected on human leukocytes (). Several functional studies support the role of JAM proteins in mediating leukocyte transmigration across endothelial tight junctions. Upon treatment with TNFα or IFNγ, JAM-A molecules found within cell junctions are redistributed onto the luminal endothelial cell surface (). Leukocytes adhere to the endothelial cells via the interaction between integrins (e.g., integrin αβ) on the leukocyte surface and JAM-A on the endothelial cell surface (). Although JAM-B is primarily restricted to endothelial cells, it is involved in trans-heterophilic interaction with JAM-C on leukocytes (). Endothelial JAM-C is a counter-receptor for leukocyte integrin Mac-1 (; ), and overexpression of JAM-C in transgenic mice enhanced leukocyte recruitment to sites of infection (). In fact, JAM-C is the first tight junction molecule reported to promote endothelial permeability (). In knockout studies, ESAM was shown to support neutrophil extravasation by destabilizing tight junctions via Rho GTPase (). JAM-like proteins on neutrophils and CAR inside epithelial tight junctions were also identified as ligand–receptor pairs that mediate neutrophil transepithelial movements (). CAR is also expressed by primary human endothelial cells derived from pancreatic islets and umbilical veins (; ). Both nectin-2 and necl-5(PVR) are also found at endothelial cell junctions. Necl-5(PVR) is observed to regulate transendothelial migration of monocytes by interacting with DNAM-1 (DNAX accessory molecule-1, which binds to both PVR and nectin-2; ; ). In culture, cell movement and proliferation are inhibited when two or more cells come into contact with each other and establish cell–cell junctions (), a phenomenon known as “contact inhibition.” Nectins are implicated in the establishment of contact inhibition. To initiate junctional formation, necl-5(PVR) trans-interacts with nectin-3 at the colliding edges of two approaching cells (). This trans-interaction activates Cdc42/Rac, which, in turn, triggers actin remodeling to promote junctional formation (). However, the trans-interaction of necl-5 (PVR) with nectin-3 is transient. Necl-5(PVR) is internalized by endocytosis from the cell surface once cell–cell junctions are established, causing reduction of cell movement and proliferation. Hence, down-regulation of necl-5 has been proposed to be one of the mechanisms underlying contact inhibition (). This finding is further supported by increased necl-5 expression during loss of contact inhibition in oncogene-transformed NIH3T3 cell lines (). The migratory behavior of cancer cells has been associated with loss of CAR expression in several tissues and cell lines (; ). CAR overexpression considerably reduced cell migration in cervical and ovarian cancer cell lines (). It also inhibited glioma cell invasion and tumor growth in vivo (). The cytoplasmic domain of CAR binds tubulin and microtubules, which possibly decreases cell motility through microtubule stabilization (). JAM-C, on the contrary, was shown to enhance both the adhesion and invasion properties of cancer cells (). The N-terminal Ig-like domains of JAM-C mediate trans-homophilic adhesion between tumor and endothelial cells (). Functional disruption of JAM-C and ESAM could inhibit pathological angiogenesis for tumor growth (; ). A mutation in the cytoplasmic tail of JAM-C also abolishes cell polarity and stimulates β1 or β3 integrin–mediated cell migration (), which converts cells from a static polarized state to a promigratory phenotype. In adult mammalian testes, the seminiferous epithelium is composed of Sertoli and germ cells. Sertoli cells create a unique environment that provides structural support and nutrients for postmeiotic germ cell development. Inter–Sertoli cell tight junctions compose the blood–testis barrier (BTB), which divides the seminiferous epithelium into two compartments: the spermatogonia-containing basal compartment and the immune-privileged adluminal compartment (). Spermatogonia differentiate into preleptotene spermatocytes, which are the germ cells that translocate from the basal to the luminal compartment for maturation without compromising the integrity of the BTB (). This requires rapid disassembly of junctional complexes ahead of migrating preleptotene spermatocytes and instant assembly of these complexes behind moving spermatocytes (for reviews see ; ; ). After traversing the BTB, germ cells rely on a series of transient junctions for anchorage onto Sertoli cells during their movement along the seminiferous epithelium. Cell adhesion molecules of the JAM/CTX and nectin family are abundantly expressed in the testis. Similar to their roles in leukocyte transmigration and viral invasion, JAM/CTX and nectin family molecules participate in germ cell migration through homophilic and heterophilic interactions. Nectin-2 is expressed on both Sertoli and germ cells, whereas nectin-3 expression is strictly limited to spermatids (). Nectin-2 also localizes at the inter–Sertoli cell junctions of the BTB. The “ectoplasmic specialization,” a testis-specific junctional structure formed by Sertoli cells, contains F-actin bundles that are arranged at regular intervals beneath the plasma membrane and a cistern of the endoplasmic reticulum connected to microtubules (). At the Sertoli cell–spermatid interface, nectin-2 and -3 form trans-heterotypic junctional complexes. The nectin-based adhesive membrane microdomains exhibit one-to-one linkage with each F-actin bundle underlying Sertoli cell–spermatid junctions (). In the absence of nectin-2 or -3, the ectoplasmic specialization at Sertoli cell–spermatid junctions does not form properly (; ). Nectin-2 and nectin-3 mice both exhibited defective sperm morphogenesis and male infertility (; ). Interestingly, nectin-3 was found on spermatids attached to Sertoli cells but not on spermatozoa released from the seminiferous epithelium (). This may imply that nectin-3 is required to confer the adhesion to germ cells that is necessary for migration across the seminiferous epithelium. Heterophilic binding of necl-5(PVR) and necl-2 was recently identified in the interaction between mouse germ and Sertoli cells (). Necl-2 is strongly expressed on the germ cell surface but not on Sertoli cells (), whereas necl-5 is only present on the Sertoli cells, as demonstrated by electron microscopy. In the seminiferous tubules of necl-2–deficient mice, round and elongating spermatids with a distorted shape failed to attach to the Sertoli cells and were sloughed off into the tubule lumen, resulting in male infertility (; ; ; ). first detected the interaction between necl-5 and necl-2 in immunoprecipitation experiments. They then used a culture system to demonstrate that overexpression of necl-5(PVR) in the Sertoli cell line (TM4) increased its capacity to adhere to Tera-2 cells expressing necl-2. This heterotypical interaction between necl-5(PVR) and necl-2 at the Sertoli–germ cell interface may partially explain the indispensable role of necl-2 in spermatogenesis. JAM-B and -C interact with each other in a manner strikingly similar to the nectin-2–nectin-3 or the necl-5(PVR)–necl-2 complex, with JAM-B coming from the Sertoli cell side and JAM-C from the spermatid side to form heterotypic interactions (). Knockout studies have shown that JAM-C is required for the assembly of the polarity complex in round spermatids and JAM-C knockout mice are infertile because of a lack of mature spermatozoa (). Although JAM-A was not found on germ cells, it colocalized with zona occludens-1 at the tight junctions of the BTB (). CAR was recently identified on the spermatozoa of rats, mice, and humans, where it was observed to occupy the acrosome membrane region (). Our group reported that CAR is concentrated at inter–Sertoli cell junctions in vitro and the BTB in vivo. Through immunofluorescent staining of isolated germ cells, we observed the presence of CAR on spermatogonia, spermatocytes, and round and elongate spermatids (). Considering that spermatogonia are nonpolarized stem cells without acrosome structures, we favor the notion that CAR is present on the germ cell plasma membrane. The presence of CAR on both Sertoli and germ cells suggests that trans-homophilic CAR–CAR interactions might take place. Like JAMs and nectins, CAR can form homodimers, which are mediated by the D1 domain of their two Ig-like loops (). During viral infection, viral fiber knobs competitively inhibit CAR–CAR interactions, which either perturb the cell junction mechanically or trigger a signaling cascade to disintegrate the entire cell junctional complex (). In a similar manner, CAR–CAR interactions between Sertoli and germ cells may compete with the original CAR–CAR trans-homodimers between Sertoli cells, thus allowing the passage of germ cells (). The migration of germ cells across the seminiferous epithelium is highly reminiscent of leukocytes squeezing through tightly apposed endothelial cells. It also bears a close resemblance to viruses traversing adjacent epithelial cells during infections. In all these scenarios, moving cells or viruses first have to break through the tight junction barriers with minimal disruption. The integrity of epithelial or endothelial barriers (e.g., the vascular endothelium with tight junctions, the BTB, or the epithelia of the pathogen host) must not be compromised. After the opening of tight junctions, the progression of these moving cells and viruses relies solely on a series of transient adherens junctions. In addition, these events engage inflammatory cytokines (e.g., TNFα) as common regulators of junctional dynamics. Our group has reported that TNFα is capable of perturbing Sertoli cell tight junction barrier assembly dose dependently in vitro (). TNFα administration to adult rat testes also reversibly perturbed the BTB, making it leaky to a small fluorescent probe such as FITC (). Localized production of TNFα from Sertoli and germ cells into the microenvironment at the basal compartment of seminiferous tubules may induce a transient “opening” of inter–Sertoli cell junctions (). It is attractive to speculate that strategies used by viruses and leukocytes are also used by developing germs cells to break through tight junctions. When germ cells move along Sertoli cells, the original JAM/CAR/nectin trans-homodimers between Sertoli cells could be replaced by the Sertoli–germ cell junctions composed of CAR–CAR/nectin-2–nectin-3/necl-5– necl-2/JAM-B–JAM-C complexes. Notably, the affinity of trans-heterodimers of nectin-3 with nectin-2 was higher than that of trans-homodimers of nectin-2 (). Likewise, the affinity of JAM-C–JAM-B heterodimers is higher than that of JAM-C–JAM-C homodimers (). As JAM-C can be coimmunoprecipitated with CAR in mouse testis lysates (), it is intriguing to speculate whether JAM-C on the germ cell surface can form a heterophilic complex with CAR on the Sertoli cell in vivo. Clearly, JAM/CTX and nectins are just a subset of the junctional molecules that participate in the cell adhesion events that are required for junction formation and stability in the seminiferous epithelium. It is impossible to break through the BTB or other junctional barriers without unlocking occludin- and claudin-based tight junction protein complexes. It is likely that JAM/CTX and nectins are perturbed by heterophilic or homophilic interactions with viral fiber knobs or with the cell surface molecules of leukocytes and germ cells. Disruption of JAM/CTX and nectins may then induce signals via Rho GTPase, Rac-1, or Cdc42 that lead to the breakdown of the entire junctional complex at cell–cell contacts, thereby allowing the passage of migrating cells at the epithelial and endothelial barriers. #text

Yeast and vertebrate nuclear pore complexes (NPCs) are structurally similar and consist of multiple copies of ∼30 different nucleoporins (; ). Approximately one third of all nucleoporins (Nups) carry phenylanine-glycine (FG) repeats of variable length. They are found at the nuclear basket, cytoplasmic fibers, and the central part of the NPC and can bind to both importins and exportins (). X-ray crystallography has mapped the contact sites between FG repeats and importin β, and mutations altering these amino acids in importin β also reduce nuclear protein import (). The extended conformation of the FG regions, their abundance in the NPC, and their differential affinity for transport receptors suggest that they are major determinants of transport through the channel. However, genetic and biochemical experiments in yeast show that half of the FG repeats can be removed without any defect in protein transport and cell viability (). The FG domain nucleoporins collectively provide a diffusion barrier to the pore. According to the virtual gating model, macromolecules are excluded from the pore by the fluctuations of unfolded peripheral FG domains. The local interaction between transport receptors and peripheral FG repeats traps the cargo, increases its residence time, and facilitates passage through the pore (). In the selective phase partitioning model, intermolecular hydrophobic interactions between the FG repeats create a selective permeability barrier that prohibits free diffusion through the NPC. The interaction of nuclear transport receptors with distinct FG nucleoporins locally breaks the mesh and allows passage through the NPC (). Are the mechanistic functions of all FG nucleoporins the same? Do individual metazoan FG nucleoporins contribute to protein transport differently than their yeast counterparts? We addressed these questions by functional analysis of the NPC using inducible GFP transport reporters in conjunction with RNAi in S2 cells. We established inducible S2 cells expressing GFP, GFP fused to a classic NLS (cNLS [cNLS-GFP]), or GFP carrying a nuclear export signal (NES [GFP-NES]). Living cells expressing native GFP showed a homogenous distribution of the fluorescent signal (). The cNLS-GFP reporter accumulated in nuclei, whereas the GFP-NES cargo was localized predominantly in the cytoplasm (). We tested whether the cNLS-GFP and GFP-NES reporters are cargoes of importin α/βs and CRM1. We first treated the cell lines with double-stranded RNA (dsRNA) against the homologues of importin α1, α2 (pendulin), α3, β (ketel), or kapβ3 (; ). Only the addition of importin α3 and β dsRNAs reduced the relative levels of nuclear cNLS-GFP. The distribution of the GFP and GFP-NES reporters was unaffected by the dsRNA treatments (Fig. S1 A, available at ; and not depicted). Thus, the cNLS-GFP reporter is transported into the nucleus by importin α3/β. In parallel, we treated the reporter cell lines with dsRNA for CRM1 (; ). The nuclear intensity of GFP-NES was increased in CRM1-depleted cells. This phenotype was comparable with the one generated by the treatment of GFP-NES cells with leptomycin, a CRM1-specific inhibitor (Fig. S1 B). Therefore, the cytoplasmic accumulation of GFP-NES provides a functional assay for CRM1- mediated export. To assess the relative contributions of the NPC components on cNLS import and NES export, we searched the genome database for nucleoporins. We identified a set of 30 putative nucleoporins and a protein export cofactor, RanBP3, in (Table S1, available at ). We did not detect any Pom121 and Nup180 homologues in the fly genome. The putative nucleoporin function of the selected genes was also predicted by the Inparanoid algorithm (), which classified them as orthologues of human genes encoding nucleoporins (Table S1). For simplicity, we will refer to the putative nucleoporins by the names of their human homologues. We generated dsRNAs targeting each candidate nucleoporin and tested gene inactivation efficiency in the reporter cell lines by RT-PCR and by immunostainings and Western blots in cases in which specific antibodies were available (Fig. S1, C–E). The dsRNA treatments considerably reduced the endogenous gene product after 4 d and allowed functional analysis of the genes in protein transport. The cellular distribution of each GFP reporter was assessed in parallel 4 d (Table S1) and 6 d (unpublished data) after the addition of dsRNA to the cultures. To avoid artifacts as a result of the potential off-target effects of the dsRNAs, we generated a second set of dsRNAs for all nucleoporins that scored positive in the primary screen. These dsRNAs generated similar defects in the distribution of the reporters, arguing for phenotype specificity (see below and Fig. S2, A–C; available at ). Cells treated with dsRNAs for Nup358, Nup153, or Nup54 exhibited a clear reduction in cNLS-GFP nuclear concentration but showed no defects in GFP-NES and GFP localization, suggesting a selective role for Nup358, Nup153, and Nup54 in cNLS-protein import (; Fig. S2 D; and Table S1). The import phenotype might be secondary to structural defects in the NPC caused by silencing of the nucleoporin genes. To assess NPC integrity, we stained dsRNA-treated cells with the nucleoporin marker mAb414 and a panel of specific antibodies against NPC components: Nup214 and Nup88 at the cytoplasmic face (), gp210 at the central core (), and TPR in the nuclear basket (). We found a pronounced reduction in mAb414 and anti-TPR rim labeling in Nup153 dsRNA-treated cells (). In addition, a substantial amount of Nup214 and its binding partner Nup88 was displaced from the pore (). Thus, both the cytoplasmic and nuclear basket nucleoporins are severely affected in Nup153 RNAi cells. We did not detect any phenotype with the gp210 antibody, suggesting that this part of the central core was intact (). None of the NPC composition defects in Nup153-depleted cells were detected in cells lacking Nup358 or Nup54, arguing that the import deficiency in these cells was not caused by major changes in pore integrity. We further examined whether the RNAi inactivations caused defects in the localization or the amount of importin β by in situ stainings and Western blots (). Untreated cells showed the characteristic rim-staining pattern of importin β. Nup358 RNAi cells exhibited a weak cytoplasmic staining. The importin β signal was also reduced in Nup54 dsRNA-treated cells, but its localization was not affected. In Nup153 RNAi cells, the levels of importin β were not appreciably affected, but a substantial fraction of the protein was displaced from the rim into the cytoplasm (). Thus, in all cases, the nuclear import deficit of the dsRNA-treated cells correlates with defects in the levels and/or localization of importin β. Neither the distribution nor the intensity of CRM1 staining was appreciably changed in these cells (Fig. S2 E), implying that Nup358, Nup153, and Nup54 are selectively required for importin β–mediated import. Our genetic analysis of Nup153 and Nup54 function in cNLS import is consistent with studies in yeast (Nup57; ), oocytes (Nup153 [] and Nup54 []), and HeLa cells (Nup153; ) using immunodepletion and overexpression experiments. However, the role of Nup358 is surprising. Nup358 is the major component of the cytoplasmic filaments, and immunodepletion of its homologue does not cause cNLS import defects in oocyte nuclei (). Nup358 is essential for importin β expression or integrity (), and the cNLS-GFP mislocalization in cells may be caused by the massive reduction of importin β levels. A common feature of Nup358, Nup153, and Nup54 is the high content of FG repeats in their primary sequence. Does the FG-rich part of Nup153 contribute to nuclear import? To address this question, we overexpressed a V5-tagged full-length (V5-Nup153) and a truncated form of Nup153 lacking the FG domain (V5-Nup153ΔFG) in Nup153 RNAi cells. Both chimeric proteins were expressed at similar levels and became localized at the nuclear envelope (). The full-length form restored both the pore composition defects and the cNLS-GFP phenotype ( and S2, F–H), indicating that Nup153, like its vertebrate homologues, contributes to both pore integrity and importin β transport (). The Nup153ΔFG fragment could rescue the defects in Nup214 and TPR localization in >98% of the expressing cells ( = 79), suggesting that it contains all of the necessary sequences for Nup153 function in NPC integrity ( andS2 H). We examined whether the Nup153ΔFG fragment is also sufficient to restore the importin β localization and cNLS import defects of Nup153 dsRNA-treated cells. In 50% of the V5-positive cells ( = 82), importin β accumulation resembled its steady-state localization in untreated cells (, bottom), suggesting that the Nup153 FG repeats are partly redundant for importin β localization. Restoration of the import receptor at the NPC may be caused by the reinstatement of other importin β–binding FG nucleoporins like Nup214 (). However, only 10% of the transfected cells ( = 66) displayed an increased nuclear cNLS-GFP accumulation (). The results argue that the role of Nup153 in protein import is independent of its function in NPC integrity. The FG region is required for importin β–mediated transport, whereas the remainder of the protein ensures an intact NPC. A direct role of the Nup153-FG part in conveying importin α3/β cargos through the pore is further supported by its localization along the entire channel () and by its highly flexible conformation (). None of the dsRNA treatments against nucleoporins caused detectable defects in GFP-NES distribution (Table S1). However, the inactivation of RanBP3 increased the nuclear accumulation of the export reporter (). The treatment had no effect on GFP and cNLS-GFP localization (Table S1 and Fig. S3 A, available at ). Yrb2, the yeast homologue of RanBP3, is also essential for CRM1-dependent export (; ). Vertebrate RanBP3 forms complexes with CRM1, RanGTP, and export substrates to stimulate NES nuclear protein export (; ). RanBP3 and CRM1 were also found in complex with the chromatin-associated protein RanGEF (). We asked whether RanBP3 inactivation impacts CRM1 distribution by staining for CRM1. Untreated cells showed a predominantly nuclear accumulation of CRM1 with only a small fraction of the protein localized at the nuclear envelope ( and S3 C). The nuclear CRM1 staining was severely reduced in RanBP3 RNAi cells. Instead, CRM1 became highly concentrated at the rim and, to some extent, in the cytoplasm of RanBP3 dsRNA-treated cells ( and S3 C). The treatment had no effect on the accumulation of Nup214, Nup88, or any of the tested nucleoporins (Fig. S3 B), suggesting a new function of RanBP3 in CRM1 localization. Reexpression of V5-tagged RanBP3 at low levels in cells restored both CRM1 depletion from the nucleus and the NES export defect (Fig. S3, E and F). The results suggest that RanBP3 directly controls CRM1 localization and protein export. CRM1 forms complexes with Nup88 and Nup214 (; ), and, in mutants lacking either of the nucleoporins, the NPC-bound CRM1 fraction accumulates in the nucleus (; ). To determine whether Nup88 or Nup214 silencing causes similar phenotypes in S2 cells, we stained cells treated with Nup88 or Nup214 dsRNA for CRM1. The treatments reduced the CRM1 signal intensity at the nuclear envelope ( and S3 C), suggesting that Nup88 and Nup214 anchor a CRM1 fraction at the NPC of S2 cells. However, unlike the defects of () and mutant larvae, the inactivation of Nup214 or Nup88 in S2 cells did not increase the cytoplasmic accumulation of the GFP-NES reporter (). Thus, Nup214 or Nup88 depletion has no impact on CRM1 activity in S2 cultured cells. This difference between larval tissues and S2 cells can be attributed to the relatively high levels of CRM1 bound to the NPCs of distinct larval tissues (; ; ). The redundancy of Nup214 for NES-GFP export in S2 cells is consistent with the lack of detectable defects in the nuclear export of NLS-GFP-NES in HeLa cells depleted for Nup214 (). Surprisingly, RNAi inactivation of Nup214 in the same cell line resulted in defects in nuclear export of the NFAT (nuclear factor of activated T cells) transcription factor and, to a lesser extent, in export of the Rev-GR-GFP reporter (). The different phenotypes may suggest specific requirements of the different export cargoes used in the two studies. CRM1 can shuttle between the nucleus and the cytoplasmic face of the NPC in an energy-independent manner (), and the inactivation of Nup214 and RanBP3 show opposing phenotypes in its localization. Therefore, we investigated CRM1 accumulation in cells treated simultaneously with both Nup214 and RanBP3 dsRNAs. In these cells, CRM1 was found inside the nucleus (), suggesting that RanBP3 and Nup214 antagonize each other to determine the nuclear concentration of CRM1. The C-terminal FG-rich region of Nup214 binds to CRM1 directly (), and we asked whether it is also required for its antagonistic role in CRM1-mediated export. We expressed V5-tagged full-length or FG-deleted versions of Nup214 in cells lacking both RanBP3 and Nup214, where CRM1 accumulates inside the nucleus. The V5-Nup214 protein complemented the Nup88 deficit at the nuclear envelope (, middle column) and prohibited the nuclear accumulation of CRM1 (, left column). The V5-Nup214ΔFG protein was expressed at similar levels as the wild-type protein and rescued the Nup88 degradation defect caused by the Nup214 inactivation (; ). Thus, the N-terminal part of Nup214 is sufficient for the interaction with Nup88 and NPC. However, the V5-Nup214ΔFG fragment only slightly increased the NPC-bound fraction of CRM1 (). This small amount of CRM1 at the rim may be attracted by Nup88, which also binds to the export receptor (). The results suggest that the antagonistic function of Nup214 on CRM1 localization is dependent on the Nup214 FG repeats. How do the opposing roles of Nup214 and RanBP3 on CRM1 accumulation influence its activity in NES export? Treatment of GFP-NES–expressing cells with dsRNAs against both Nup214 and RanBP3 resulted in the cytoplasmic distribution of the reporter, closely resembling its accumulation in untreated cells (, right column). Thus, unleashing the pore-bound fraction of CRM1 through Nup214 inactivation largely restores the GFP-NES export defect caused by the depletion of RanBP3. The results suggest that CRM1 NES export activity can be tuned by the opposing functions of Nup214 and RanBP3. Overexpression of the full-length Nup214 construct in Nup214 and RanBP3 RNAi cells resulted in a nuclear accumulation of GFP-NES closely resembling the phenotype caused by single RanBP3 inactivation (). In contrast, the distribution of GFP-NES remained unaffected in V5-Nup214ΔFG–expressing cells lacking both Nup214 and RanBP3 (). The results indicate that the ability of Nup214 to antagonize the function of RanBP3 in NES export requires the Nup214 FG repeats. CRM1 is reduced in the nucleus of cells, arguing that RanBP3 retains it inside the nucleus. To further examine the proposed new role of RanBP3, we overexpressed it in S2 cells and analyzed its effects on CRM1 and GFP-NES localization. The expression of V5-tagged RanBP3 increased the nuclear intensity of CRM1 ( and S3 D), further arguing for a dynamic equilibrium between the Nup214- and RanBP3-bound forms of CRM1 (). In parallel experiments, we assessed the effect of RanBP3 overexpression in GFP-NES distribution. Although low levels of V5-RanBP3 did not change the predominantly cytoplasmic distribution of GFP-NES, high amounts of the exogenous protein increased the nuclear intensity of the reporter (). This phenotype is consistent with in vitro experiments in which high levels of RanBP3 inhibit the assembly of CRM1 export complexes. In summary, we propose a dual function of RanBP3 (): one maintaining high nuclear levels of CRM1 and one aiding the assembly of CRM1–RanGTP cargo complexes (; ). First, Nup214 and RanBP3 antagonize each other to determine CRM1 localization and function. RanBP3 has a primary role in maintaining CRM1 inside the nucleus. This function of RanBP3 becomes redundant when Nup214 is codepleted. Second, we provide genetic evidence arguing that individual FG domains are essential for distinct transport pathways in . The importance of the Nup153 FG motif in mediating cNLS import was already suggested by overexpression experiments in permeabilized HeLa cells (). Surprisingly, the FG repeats of Nup214 do not facilitate NES-GFP export but rather inhibit it. The FG regions from Nup153 or Nup214 are indispensable for the distinct transport roles of Nup153 and Nup214, yet they are not expected to affect the total mass of FG repeats and the barrier function of the NPC. The genetic analysis of nucleoporins in argues that the Nup153 and Nup214 FG regions have specific functions in import and export, respectively, and suggest that peripheral nucleoporins have aquired additional roles during metazoan evolution. Understanding the mechanistic roles of animal nucleoporins in endogenous protein transport may provide new insights into the regulatory potential of the NPC. The GFP, cNLS-GFP, and GFP-NES constructs were described previously (). S2 cells were cultured in Schneider medium (PAN) supplemented with FCS, glutamine, and streptomycin/penicillin (Invitrogen). Stable cell lines were generated by hygromycin selection according to the manufacturer's instructions (Invitrogen). The expression of GFP-tagged cargoes was induced by 0.2 mM CuSO for 16 h. Leptomycin was used at the concentration of 10 ng/ml for 15 min. RNAi was performed essentially as described previously (). Primer pairs tailed with the T7 RNA polymerase promoter were used to amplify PCR fragments obtained from cDNA clones. PCR products with an average size of 700 bp were then used as templates for dsRNA production with the MEGAscript RNAi kit (Ambion). For transfection, 15 μg dsRNA was added to 2.5 × 10 S2 cells in six-well plates. DNA was visualized by the addition of Hoechst 33342 (Sigma-Aldrich) at a concentration of 4 μM. Images were recorded with an inverted fluorescence microscope (DM IRB; Leica) at days 4 and 6 after dsRNA treatment and were quantified using Volocity version 2.0.1 (Improvision). Cells were lysed in 10 mM Tris, 140 mM NaCl, 1.5 mM MgCl, and 1% NP-40. Lysates were resolved by SDS-PAGE and analyzed by immunoblotting. Blots were developed using the ECL Advance kit (GE Healthcare). Images were acquired with a luminescent image analyzer (LAS1000; Fuji) and quantified with Image Gauge version 3.45 (Fuji). For RT-PCR, mRNAs were isolated using magnetic oligo(dT)-coupled beads (Dynabeads). Reverse transcription was performed with SuperScript-II (Invitrogen). The inactivation of Nup153 was performed with dsRNAs generated by the primers 5′-TTAATACGACTCACTATAGGGAGACATGTGTGAACAAATACCGCT-3′ and 5′-TTAATACGACTCACTATAGGGAGAGTGTGTGTGAATCTAAACGCTA-3′. Nup214 was inactivated with dsRNAs made by the primers 5′-TTAATACGACTCACTATAGGGAGATTGGTGCTGCTGCAAAGC-3′ and 5′-TTAATACGACTCACTATAGGGAGACTGAACAAGCAAAACTATTG-3′. RanBP3 was inactivated with dsRNAs generated by the primers 5′-TTAATACGACTCACTATAGGGAGACTCGCTCTTGTTCTTTTTATACG-3′ and 5′-TTAATACGACTCACTATAGGGAGAAGAGCGTGTACGATCGATATC-3′. In all cases, products were targeting the 3′ untranslated region of the respective mRNAs. The cDNA encoding full-length Nup153 (amino acids 1–1,905) and Nup153ΔFG (amino acids 1–1,288) were introduced into the BstBI site of pAc5.1/V5-His (Invitrogen). The cDNAs encoding Nup214 (amino acids 1–1,670) and Nup214ΔFG (amino acids 1–1,080) were inserted into the NotI site of pAc5.1/V5-His. The cDNA encoding RanBP3 (amino acids 1–451) was inserted into the BstBI site of pAc5.1/V5-His. We used antibodies against CRM1 (), importin β (provided by J. Szabad, University of Szeged, Szeged, Hungary; ), Nup88 (), Nup214 (), TPR (a gift from V. Cordes, University of Heidelberg, Heidelberg, Germany), α-tubulin (Sigma-Aldrich), mAb414 (Babco), lamin (provided by G. Krohne, University of Wurzburg, Wurzberg, Germany; ), and gp210 (provided by P.A. Fisher, Developmental Studies Hybridoma Bank, University of Iowa, Iowa City, IA; ). S2 cells were attached to poly--lysine– (Sigma-Aldrich) or ConA (Sigma- Aldrich)-coated coverslips. Adherent cells were fixed with 4% PFA for 30 min in PBS, permeabilized for 5 min in 0.1% Triton X-100/PBS, and incubated overnight at 4°C with primary antibodies diluted in 0.5% BSA/PBS. Cells were washed in 0.1% Triton X-100/PBS, incubated with secondary antibody for 2 h at RT, and incubated with 0.4 mg/ml DAPI for 5 min. Cells were mounted in Vectashield (Vector Laboratories). Wide-field images were acquired with Openlab version 3.1.4 (Improvision), and ratios of nuclear envelope to cytoplasmic labeling intensities were quantified using Volocity version 2.0.1 (Improvision). Table S1 summarizes the transport phenotypes after dsRNA treatment against nucleoporins. Fig. S1 shows importin α/β and CRM1 transport assays in S2 cells (A and B) and shows the inactivation efficiency in dsRNA-treated cells (C–E). Fig. S2 shows the cNLS-GFP import defect in cells treated with the second set of Nup358, Nup153, and Nup54 dRNAs (A–C) and shows CRM1 and GFP-NES localization in cells defective in cNLS-GFP import (D and E). Fig. S2 also shows the restoration of TPR and Nup214 localization by full-length Nup153 or Nup153ΔFG (F–H). Fig. S3 shows the NPC integrity and cNLS-GFP import in RanBP3 RNAi cells (A and B) and shows that Nup214 and RanBP3 determine the CRM1 concentration at the pore (C and D). Fig S3 also shows the restoration of CRM1 localization and protein export activity in RanBP3 RNAi cells (E and F). Online supplemental material is available at .

Elimination of cells by programmed cell death (PCD) is a universal feature of development and aging (; ). In both vertebrates and invertebrates, dying cells often progress through a stereotyped set of transformations referred to as apoptosis. In this form of PCD the nucleus condenses, and the collapsing cell corpse fragments into “apoptotic bodies” that are engulfed by specialized phagocytes or neighboring cells (; ; ). Apoptosis requires autonomous genetic functions within the dying cell, and extrinsic cues that elicit apoptosis have been investigated in numerous experimental models (; ). Other forms of death are also thought to contribute during development and differ from apoptosis with respect to cellular morphology, mechanism, or mode of activation. These may include necrosis, characterized by swelling of the plasma membrane, or autophagic cell death, which is linked to extensive vacuolization in the cytoplasm (). These forms of cell death can be caspase dependent or independent and may or may not be under deliberate genetic control (). Two conserved protein families comprise central elements of the apoptotic machinery (). Orthologous proteins represented by Ced4 in the nematode, Apaf1 in mammals, and Ark (Dark) function as activating adaptors for CARD-containing apical caspases. During apoptosis, Ced4/Apaf1/Dark adaptors associate with pro-caspase partners (Ced3, Caspase 9, and Dronc) in a multimeric complex referred to as the “apoptosome”. This complex is regulated by Bcl2 proteins, but apparently through different mechanisms (for review see ). Previously, we and others genetically examined components of the apoptosome (, ; ; ; ; ; ; ). and are recessive, lethal genes. Both exert global functions during PCD and in stress-induced apoptosis. However, their roles in apoptosis are not absolute because rare cell deaths occurred in embryos lacking maternal and zygotic product of either gene (; ). Elimination of in the wing caused a unique, age-dependent phenotype associated with late-onset blemishing throughout the wing blade (). Here, we show that this progressive phenotype is characteristic for wing epithelia that lack apoptogenic functions and is caused by defects in a communal form of PCD where epithelial cells are collectively and rapidly eliminated. We leveraged these findings to discover additional genes required for PCD and recovered a limited set of loci, many of which were previously unknown to function in cell death. Here, we establish that () is essential for coordinated death in the wing epithelium and, consistent with PCD functions in earlier developmental stages, regulates proper cell number in diverse tissue types. tissue exhibit normal morphology at eclosion but develop progressive, melanized blemishes with age (). We applied similar methods to determine whether lesions in other apoptogenic genes present a similar phenotype. After eclosion, wings mosaic for , a deletion removing the apoptotic activators (), , and (), were morphologically normal at eclosion, but over 3–7 d, melanized blemishes appeared at random throughout the wing (). Likewise, homozygous adult “escapers” deficient for the effector caspase Drice () also presented normal wings at eclosion but developed blemishes with age (). Wings mosaic for a null allele of (), were indistinguishable from wild-type (WT) at eclosion (), but within 4 d developed wing blemishes (). These late-onset blemishes became markedly more severe as animals aged. Similar yet less severe wing blemishes occurred in adults homozygous for , a hypomorphic allele of (). Together, these observations establish that late-onset progressive blemishing in mosaic wings is a characteristic phenotype shared among mutants in canonical PCD pathways. In the wing of newly eclosed adults, PCD removes the epithelium that forms the dorsal and ventral cuticles (; ). To determine whether the cause of the blemish phenotype might trace to defective death in the wing epithelium, we examined this tissue in mutants. For these studies, wings of adults were prepared for light and electron microscopy. Histological analyses at the light level showed that on the first day of eclosion, the dorsal and ventral cuticles of WT animals became tightly merged with no intervening tissue evident between these layers (). However, even 14 d after eclosion, cells and cell remnants remained situated between the dorsal and ventral cuticles in mutants. This “undead” tissue was most easily visualized in lateral sections through melanized blemishes (). Further examination of the persisting epithelium at the EM level showed evidence of intact cells soon after eclosion () and ectopic cellular material 24 h after eclosion (). To directly examine the death of wing epithelial cells in vivo, we adapted a transgenic nuclear DsRed reporter driven by () (; ), which permits visualization of the fate of these cells soon after eclosion. Observations with this pan-epithelial marker in the wing confirmed earlier studies (; ). shows that within 1 h of eclosion, intact epithelial cells are clearly present and regularly patterned throughout the wing. 1–2 h later (2–3 h after eclosion), the entire intervein epithelium disappears, manifested here by the abrupt loss of DsRed throughout the wing blade (). Live, real-time imaging of the wing in newly eclosed adults revealed unexpected features associated with elimination of the intervein epithelium ( and Videos 1–3, available at ). Epithelial cells, labeled by nuclear fluorescence, were arranged in a regular, predictable pattern throughout the wing. Then, consistent with nuclear breakdown, fluorescence became redistributed throughout the cell followed by indications of blebbing and the appearance of fragmenting cells. Occasionally, weak fluorescence enclosed in cell corpses condensed to bright punctate bodies. This series of apoptogenic changes spread extremely rapidly throughout the epithelium, appearing here as a collective wave initiating from the peripheral edge and moving across the wing blade (, top panels). Within just 4 min, virtually all nuclei (∼450 cells) within a space of ∼114 mm converted from viable to apoptotic morphology. The process involved tight coordination at the group level because the likelihood of a single cell apoptosing was clearly linked to similar behaviors by nearest neighboring cells over short time frames (Video 1). Also, the direction and size of the cell death wave may not be fixed in every region of the wing, but centrally located cell groups were generally eliminated earlier (). Unlike conventional examples of PCD in development, we found no indication that overt engulfment of apoptotic corpses occurred at the site of death. Instead, DsRed-labeled cell remnants were passively swept toward the nearest wing vein (, bottom panels) where, apparently under hydrostatic pressure, cell debris streamed proximally toward the body through the wing or along the wing vein (, bottom panels; and Video 3). Together, these observations describe a communal form of PCD that rapidly eliminates the wing epithelium through coordinated group behavior. We used the reporter to track the fate of mosaic wing epithelia where mutant clones were induced. In sharp contrast to WT wings, abnormally persisting cells could be readily detected as patches of DsRed in the nuclei of epithelial cells in mosaic tissues. clones retained extensive patches of persisting DsRed-labeled cells (). Here, cells and nuclei were readily detected 4 d after eclosion (), and even at 11 d post-eclosion, extensive evidence of cell debris was seen (not depicted). We found that wings mosaic for the deletion gave identical results (). Likewise, adults mutated for exhibited persisting cells throughout the wing blade (). Consistent with this, rare escapers also showed evidence of persisting cells after eclosion (). These observations link failures in PCD to progressive melanized wing blemishes, raising the possibility that other apoptogenic mutants might also produce this phenotype. Unlike previously described wing defects, which are congenital and evident at eclosion (; ), the age-dependent phenotype described in (A–D) is characteristic of mutations in genes that function in canonical PCD pathways. Moreover, when the dosage of was reduced by half in adults () or if WT Dmp53 was removed from these same animals, melanized blemishes became far more severe. homozygotes or in heterozygotes (unpublished data). Numerous other mutants showed no such effects in combination with a hypomorph. or tissue. We screened a collection of preexisting transposon mutants to capture insertions that exhibit normal wings at eclosion but develop melanized blemishes with age. Our strategy exploits the FLP/FRT system together with wing-specific drivers to interrogate animals bearing wing genotypes mosaic for clones of P element–derived lethal mutations. Progeny with mosaic wings were examined for late-onset wing blemishes at 1, 7, and 14 d post-eclosion. We screened over 1,000 lethal insertions, representing 356 2nd chromosome mutations and 707 3rd chromosome mutations. The majority of insertions (87%) produced no visible defects as wing mosaics (see online supplemental tables, available at ). 13% of insertions tested produced abnormalities, and these were scored for the phenotypic categories shown in Fig. S1. Congenital defects including notched, blistered, or wrinkled wings occurred alone or occasionally as compound phenotypes (Fig. S1, A–D). The candidate strains that developed wing blemishing were further subdivided based on phenotypic severity. Insertions in class A developed pronounced blemishes within a week of eclosion (Fig. S1 E), whereas those in class B developed relatively light-colored patches between 1 and 2 wk after eclosion (Fig. S1 F). Mutant lines exhibiting class A phenotypes were rare (∼2%). All members of this class lacked blemishes at eclosion and displayed progressive blemishing occasionally associated with fragile and sometimes broken wings (Fig. S1 E). A new allele of (lSH0173) was recovered in this class (Table S2), providing reassuring validation for our screening strategy. Some members among these classes exhibited congenital notches or blisters, but congenital blemishes present at eclosion were not found. We applied inverse PCR to map or confirm insertion sites of many class A and B strains (Table S2). In addition to dark, we isolated several mutations associated with genes previously implicated in PCD (Table S1). For example, lSH2275 contains an insertion 2 kb upstream of , a microRNA capable of modulating Rpr-induced cell death (). Likewise, lS048915 maps to the first intron of DIAP1 and may represent a hypermorphic allele at this locus. lS055409 maps near (), a gene implicated in cell killing triggered by Rpr or Eiger, the fly counterpart of TNF (; ). Several insertions map in or near transcriptional or translational regulators that might alter the expression of cell death genes. For example, (lS146907), an Atrophin-like protein (), functions as a transcriptional repressor (), while (lS097074) belongs to the DEAD-box family of proteins () often implicated in translational regulation and RNA processing. A portion of the class A and B hits were also directly examined for defective PCD by applying the reporter in mosaic wings. Of the 29 strains tested, 14 showed obvious evidence for persisting cells in the wing epithelium (Table S2), which include dark (). Mutants identified above that exhibit both blemishing and persisting cells are likely candidates for PCD genes. One strain, lS134313, produced severe late-onset blemishing () and a persisting cell phenotype (). After mapping this insertion to the first intron of the , we produced null alleles at this locus by a customized deletion strategy illustrated in . Two FRT-containing P element insertions flanking the coding region of () were used to generate a novel deletion depicted in (see Materials and methods). PCR verified recombination between P elements (), and 8 deletion strains were recovered. These validated alleles eliminate exons 4–12, removing over 92% of coding sequence in the predicted open reading frame. Deletions at the locus were uniformly lethal before the 3rd instar stage. However, zygotic is not essential to complete embryogenesis because ∼70% of homozygotes hatch to 1st instar larvae. was recombined on the FRT79 chromosome to generate adult wings mosaic for this allele, and like the original insertion, these animals also developed robust progressive blemishes () and a persisting cell phenotype (). Both phenotypes were more severe than the original P insertion, suggesting that the lS134313 allele is hypomorphic for . These findings link loss of function to our query phenotypes, establishing that the action of HIPK is essential for post-eclosion PCD in the wing epithelium. Using general stains (acridine orange) or TUNEL methods, embryonic PCD was not overtly disturbed in mutants. To investigate the possibility of more subtle or specific phenotypes, we examined the nervous system using antibodies that label specific populations of neurons affected by the deletion (). Using α-Kruppel antibody (), we confirmed that stage 14–15 WT embryos contained 9–12 Kruppel-positive cells in the Bolwig's Organ (). However, a portion of animals lacking maternal contained as many as 15 cells per organ () at a penetrance comparable to animals, which are completely cell death defective (). We also examined neurons expressing dHb9, a homeodomain protein marking a subset of cells that persist in cell death–defective embryos (). In germline clones, distinct classes of dHb9 staining patterns emerged. A subset of animals exhibited extreme patterning defects. Other animals displayed a striking increase in dHb9-positive cell numbers () when compared with WT embryos of the parental strain (). These data establish that HIPK fundamentally regulates cell numbers in the nervous system, and because the same subpopulation of cells are affected by the mutation, they implicate HIPK as a more general regulator of PCD. The pupal eye undergoes reorganization involving cell death of interommatidial cells after pupation (). To determine if HIPK regulates cell death in the retina, we generated whole eye clones and used the α-Dlg (discs large) antibody to outline cell borders in dissected pupal eyes after pupation. The WT pattern of interommatidial cells is represented in . clones (). This phenotype is overtly similar to animals lacking the apical caspase Dronc () and consistent with an essential role in retinal PCD. Elimination of the wing epithelium in newly eclosed adults is predictable, easily visualized, and experimentally tractable. The major histomorphologic events involve cell death, delamination, and clearance of corpses and cell remnants. Recent studies established that post-eclosion PCD is under hormonal control and involves the cAMP/PKA pathway (). While dying cells in the adult wing present apoptotic features (e.g., sensitivity to p35 and TUNEL positive), elimination of the epithelium is distinct from classical apoptosis in several important respects. First, unlike most in vivo models, overt engulfment of cell corpses does not occur at the site of death (; ). Instead, dead or dying cells and their remnants are washed into the thoracic cavity via streaming of material along and through wing veins (, Videos 1–3; and ). Second, extensive vacuolization is seen in ultrastructural analyses, which could indicate elevated autophagic activity (for review see ; ; and ). Third, widespread and near synchronous death that occurs in this context defines an abrupt group behavior. The process affects dramatic change at the tissue level, causing wholesale loss of intervein cells and coordinated elimination of the entire layer of epithelium. Rather than die independently, these cells die communally, as if responding to coordinated signals propagated throughout the entire epithelium, perhaps involving intercellular gap junctions. This group behavior contrasts with canonical in vivo models where a single cell, surrounded by viable neighbors, sporadically initiates apoptosis. One study proposed that an epithelial-to-mesenchymal transition (EMT) accounts for the removal of epithelial cells after eclosion (). Although our results do not exclude EMT associated changes in the newly eclosed wing epithelium, compelling lines of evidence, presented here and elsewhere, establish that post-eclosion loss of the wing epithelium occurs by PCD in situ—before cells are removed from the wing (). First, before elimination, wing epithelial cells label prominently with TUNEL. Second, every mutation in canonical PCD genes so far tested failed to effectively eliminate the wing epithelium (), and at least two of these were recovered in our screen. Third, elimination of the wing epithelium was reversed by induction of p35, a broad-spectrum caspase inhibitor (). Fourth, using time-lapse microscopy, we clearly detected condensing or pycnotic nuclei, followed by the rapid removal of all cell debris in time frames (minutes) not consistent with active migration. Instead, removal of cell remnants occurred by a passive streaming process, involving perhaps hydrostatic flow of the hemolymph. Here, we sampled over one fifth of all lethal genes and nearly 10% of all genes in the fly genome for the progressive blemish phenotype, a reliable indicator of PCD failure in the wing epithelium. Nearly half of the mutants that produced melanized wing blemishing also displayed a cell death–defective phenotype when examined with the reporter. The precise link between these defects is unclear, but a likely explanation suggests that as the surrounding cuticle fuses, persisting cells, now deprived for nutrients and oxygen, become necrotic and may initiate melanization. Mutants could arrest at upstream steps, involving the specification or execution of PCD, or they might affect proper clearance of cell corpses from the epithelium. We recovered new alleles of (lSH0173) and a likely hypermorph of (lS048915), which provides reassuring validation of this prediction. By leveraging this distinct phenotype, we captured novel cell death genes, including the orthologue of . Though first identified as an NK homeodomain binding partner (), we found this gene to be an essential regulator of PCD and cell numbers in diverse tissue contexts. Of the four mammalian genes, , the predicted orthologue of , has been placed in the p53 stress-response apoptotic pathway (; ; , ), but whether the counterpart similarly impacts this network is not yet known. The lSxxxxxx() and lSHxxxx() FRT stocks were obtained from Szeged Stock Center. and were provided by K. Basler (University of Zürich, Zürich, Switzerland). flies are from J. Jiang (UT Southwestern Medical Center, Dallas, TX). stocks were provided by A. Gould (National Institute for Medical Research, London, UK) and J. O'Tousa (University of Notre Dame, Notre Dame, IN). To generate wing clones, 4 males of the genotype were crossed to 3 females of . F1 flies were examined at eclosion and at 1 and 2 wk of age for appearance of “melanized blemishes” on the wing. and flies were similarly screened by crossing to and , flies, respectively. was used for mutations on 2R. Adult wings were removed at different ages and fixed in paraformaldehyde for standard histology. Electron micrographs were generated using an electron microscope (TEMP2 1200 EX II; JEOL). The FLP/FRT system was used to generate mutant wing clones, and persisting cells were visualized using DsRed. was crossed to . After eclosion, adults were aged from 1 to 14 d. Wings were removed, mounted on glass slides, and visualized using a fluorescent DLM (Axioplan; Carl Zeiss MicroImaging, Inc.) and a monochrome digital camera (Hamamatsu). , , or lines with or without were also crossed to their respective FLP/FRT lines and imaged as stated above. Epithelial cell death was recorded in time-lapse experiments using the previous crosses to image adults at 1–2 h after eclosion using a stereomicroscope (SteREO Discovery V.12; Carl Zeiss MicroImaging, Inc.) with Pentafluar S. Adults were glued on their dorsal surface to glass slides and imaged while alive. lines were from Bloomington Stock Center. -element insertion sites were mapped by inverse PCR according to protocols from BDGP (). Genomic DNA of insertion lines containing the PlacW insertion element was extracted using Wizard Genomic DNA Extraction kit (Promega) and digested with HhaI, HpaII, and MboI restriction enzymes for 2.5 h at 37°C. Resulting digestions were diluted into 400 μl T4 DNA ligase (Roche) reactions and incubated overnight at 4°C. DNA was ethanol precipitated and used in Expand Long Template (Roche) PCR reactions with primers specific to the PlacW insertion. Unique PCR products were gel purified and sequenced, and insertion locations were confirmed using genomic PCR with primer sets specific to PlacW and surrounding genomic sequences. Deletions were generated using the Exelixis collection of P elements as described previously (; ). To delete the locus, insertions and were placed in trans together with and heat-shocked to generate a FRT-mediated deletion. PCR primers directed to the remaining P elements and the surrounding genomic locus were used to identify deletion alleles (5′-TACTATTCCTTTCACTCGCACTTATTG-3′ and 5′-TAGATGAGGAAGTTCTGCGTGCAAGA-3′, 5′-CCTCGATATACAGACCGATAAAAC-3′ and 5′-CGACCTTCACCGACTGATCCTGGAT-3′). Two additional primer pairs, one producing a novel PCR product spanning the deleted locus (5′-GTGTCACTCGAAATTCGCCAGTGACT-3′ and 5′-GACGACTGACTCGGTAGCCTACTTCG-3′) while another specific to the deleted locus producing a negative result (5′-CGCTACTATCGTGCTCCCGAAATCAT-3′ and 5′-CGGATGCCTTGACATTGTTGCAGT-3′), were used to confirm deletions. Germline clones were generated using the dominant female sterile technique described previously (; ). / was crossed to , and pupated animals were removed and aged for 48–55 h. After aging, pupal eyes were dissected and fixed in 4% formaldehyde in PBS. Aged matched siblings carrying were used as controls. Immunohistochemistry was performed as described in . Guinea pig α-Kruppel was used 1:600 (), rabbit α-dHb9 was used 1:500 (), and α-Dlg was used 1:500 (Developmental Studies Hybridoma Bank) at 4°C overnight. Secondary antibodies used were labeled with Texas red or Fluorecein from Vector Laboratories (1:250) or Alexa 568 from Invitrogen (1:500). Genotyping was done using anti-GFP (1:1,000) from Invitrogen recognizing GFP-labeled balancers. Confocal z-series were taken using a confocal microscope (TCS SP5; Leica) and used for counting. Z-series were stacked for presented images. Adult wings were dry mounted, and images were acquired using a microscope (Stemi V6; Carl Zeiss MicroImaging, Inc.) equipped with a 1.0× lens using a digital camera (Coolpix5000; Nikon) or a stereomicroscope (SteREO Discovery V.12; Carl Zeiss MicroImaging) with PentafluarS using 0.63× or 1.5× PlanApoS lenses and an MRm or MRc5 digital camera (Axiocam) and Axiovision Release 4.6 software. Additional fluorescent wing images were acquired with a microscope (Axioplan 2E; Carl Zeiss MicroImaging, Inc.) and a monochrome digital camera (Hamamatsu) using Plan Neofluar 10×/0.30, Plan Apochromat 20×/0.60, and Plan Neofluar 40×/0.75 objectives and OpenLab software. Confocal images of tissues stained with Fluorescein and Alexa 568 were mounted in Vectashield (Vector Laboratories), and images were acquired on a confocal microscope (TCS SP5; Leica) with Leica LAS AF software. The following lenses were used: HC PL APO 20×/0.70, HCX PL APO 40×/1.25-0.75 oil, and HCX PL APO 63×/1.40-0.60 oil objectives. All images were taken at room temperature and were processed in ImageJ or Photoshop 7.0. Occasionally, images were linearly rescaled to optimize brightness and contrast uniformly without altering, masking, or eliminating data. Fig. S1 displays the wing phenotypes characterized in the mosaic screen. Videos 1–3 are time-lapse experiments showing PCD of wing epithelial tissue taken as described in Materials and methods. Table S1 displays a summary of the P insertion wing mosaic screen, Table S2 lists loci implicated in coordinated death in the wing epithelium, and Table S3 is a summary of all strains screened. Online supplemental material is available at .

xref #text To further understand the functional role of Lgl proteins, we used biochemical affinity purification to identify novel interacting proteins of mLgl. As bait, we used the C terminus of mLgl2 tagged with maltose binding protein (MBP). By applying rat kidney lysate to the fusion protein coupled to amylose resin beads, we purified a protein with a size of 30 kD (Fig. S1 A, left, available at ). No other mLgl2 binding proteins were identified. Mass spectrometric analysis identified the 30-kD protein as p32 (gC1QR, gC1q-BP, and HABP1), which was first characterized as a splicing factor 2–associated protein () but later described as a multifunctional chaperone protein (). Western blotting with anti-p32 antibody confirmed the identity of the protein (Fig. S1 A, right). To validate the interaction, we examined whether mLgl2 and p32 were coimmunoprecipitated. GFP-mLgl2-WT (wild type) and Myc-p32 were coexpressed in human embryonic kidney (HEK) 293 cells and immunoprecipitated with either anti-GFP or anti-Myc antibody. Reciprocal coimmunoprecipitation of mLgl2 and p32 was observed, confirming the interaction between the two proteins (). Myc-p32 was not coimmunoprecipitated with GFP alone (Fig. S1 B). mLgl1, another mammalian homologue of Lgl, was also coimmunoprecipitated with p32 (Fig. S1 C). We also demonstrated that endogenous p32 and mLgl2 proteins were coimmunoprecipitated from HEK293 cell lysate (). We further showed that the interaction between mLgl2 and p32 was direct, as recombinant His-p32 protein bound specifically to MBP-mLgl2C-WT in an in vitro binding assay (). mLgl2 binds to and is phosphorylated by aPKC (PKCλ and PKCζ). p32 has also been shown to interact with and regulate PKCs, including aPKC (; ). We therefore examined whether mLgl2, p32, and PKCζ form a trimeric complex. FLAG-mLgl2-WT and p32-GFP were coexpressed in HEK293 cells, and double immunoprecipitation experiments were performed. We first immunoprecipitated with anti-FLAG antibody, which pulled down both endogenous PKCζ and p32-GFP together with FLAG-mLgl2 (). The immunoprecipitate was then eluted from the beads with FLAG peptide, and a second immunoprecipitation was performed using anti-GFP antibody. After the second immunoprecipitation, p32 remained bound to mLgl2 and PKCζ (). These results indicate that the three proteins form a trimeric complex. Using several truncation mutants, we showed that amino acids 544–1027 of mLgl2 and the C terminus of p32 are responsible for the interaction (Fig. S1, D–F). To investigate whether the phosphorylation status of mLgl2 influences its interaction with p32, we performed in vitro binding experiments (). Untreated and λ-phosphatase–treated MBP-mLgl2C-WT bound to endogenous p32 from HEK293 cell lysates to a similar extent. Interestingly, MBP-mLgl2C-WT that had been incubated with PKCζ at 4°C pulled down more p32 (). However, MBP-mLgl2C-WT incubated with PKCζ at 30°C did not show increased interaction with p32 (). These data suggest that PKCζ enhances the interaction between mLgl2 and p32, but once mLgl2 is phosphorylated by PKCζ, p32 is no longer able to efficiently interact with the complex. Comparable results were obtained using a recombinant p32 protein (Fig. S2 A, available at ). To further investigate the interaction between mLgl2, PKCζ, and p32, we performed immunoprecipitation experiments using WT (mLgl2-WT) and nonphosphorylatable (mLgl2-SA) forms of mLgl2, in the presence or absence of overexpressed PKCζ (). In the mLgl2-SA mutant, three serines in the conserved aPKC phosphorylation site were changed to alanine. Endogenous p32 from HEK293 cell lysates was coimmunoprecipitated with both mLgl2-WT and mLgl2-SA (). Coexpression of GFP-PKCζ enhanced the interaction of p32 with mLgl2-SA, but not with mLgl2-WT (). These data further support a key role of PKCζ and the phosphorylation status of mLgl2 in the transient interaction between mLgl2 and p32. Overexpression of mLgl2 did not substantially affect the interaction between p32 and PKCζ, suggesting that the binding of p32 to PKCζ is not affected by the presence of mLgl2 (Fig. S2 B). We next examined the effect of p32 binding on the activity of PKCζ using in vitro kinase assays. Approximately equal amounts of PKCζ were pulled down by immunoprecipitation with anti-p32 antibody from 3 ml of HEK293 cell lysate and with anti-PKCζ antibody from 100 μl of lysate (, top). Immunoprecipitates were incubated with a biotinylated PKCζ-specific peptide substrate and γ-[P]ATP. PKCζ immunoprecipitated with anti-p32 antibody phosphorylated the substrate ∼2.5 times more efficiently than PKCζ immunoprecipitated with anti-PKCζ antibody (), suggesting that the interaction with p32 enhances the catalytic activity of PKCζ. A specific PKC inhibitor, Gö6983, suppressed phosphorylation of the peptide by p32 immunoprecipitates (Fig. S2 C) and other types of PKCs, such as PKCα (classical PKC) and PKCδ (novel PKC), did not efficiently phosphorylate the substrate under the conditions used (Fig. S2 D). These data indicate that the increased phosphorylation of the peptide in the p32 immunoprecipitates is indeed due to enhanced aPKC activity. As mLgl2 is a crucial target for phosphorylation by PKCζ, we examined whether the interaction with p32 affected the ability of PKCζ to phosphorylate mLgl2. We first immunoprecipitated PKCζ with either anti-p32 or anti-PKCζ antibody as described in the previous paragraph. The immunoprecipitates were then incubated with MBP-mLgl2C-WT and γ-[P]ATP in the presence or absence of a PKCζ inhibitor. As seen with the peptide substrate, PKCζ bound to p32 was significantly more efficient at phosphorylating MBP-mLgl2C-WT than PKCζ alone (). Furthermore, the phosphorylation of MBP-mLgl2C- WT was shown to be mediated by PKCζ, as the addition of the PKCζ inhibitor efficiently blocked the phosphorylation of MBP-mLgl2C-WT (). In the absence of immunoprecipitates, MBP-mLgl2C-WT protein was not substantially phosphorylated, indicating that there is no contaminating kinase present with the protein (unpublished data). We also showed that the PKCζ inhibitor did not suppress phosphorylation by other types of PKC, such as PKCα and PKCδ (Fig. S2 E), confirming the specificity of the inhibitor. These data indicate that p32 enhances the activity of PKCζ to phosphorylate mLgl2. It was recently shown that one mechanism of regulating aPKC activity involves the small GTPase Cdc42; binding of GTP-bound Cdc42 and Par-6 to aPKC enhances the activity of aPKC (; ; ). In , however, Lgl is still phosphorylated by aPKC, even if the ability of Par-6 to bind to Cdc42 is abolished, indicating that Cdc42 is not required for the activation of aPKC in this system (). We propose that p32 binding to mLgl2 and PKCζ is a novel mode of aPKC regulation that enhances its catalytic activity, leading to increased mLgl2 phosphorylation. Next, we examined the subcellular localization of endogenous p32 and mLgl2 proteins in MDCK cells. mLgl2 localized at the plasma membrane as well as in the cytosol observed as small puncta, whereas p32 was accumulated along the plasma membrane with some cytosolic localization (, top). We found that both proteins partially colocalized at the cortex region of the plasma membrane (, bottom), suggesting that the interaction may occur at the cortical region rather than in the cytosol. We also examined the effect of overexpression of p32 on the localization of mLgl2. We used MDCK cell lines that stably express p32-GFP in a tetracycline-inducible manner (Fig. S3 A, available at ). In the absence of tetracycline, expression of p32-GFP was not induced and mLgl2 localized at the plasma membrane (, top), as observed in parental nontransfected MDCK cells (not depicted). Upon the addition of tetracycline, the expression of p32-GFP was induced and mLgl2 dissociated from the cell cortex and localized to the cytosol (, bottom). These results further support a role for p32 as a regulator of mLgl2. To investigate the functional role of p32 in cell polarity, we used a 3D MDCK cell culture system. First, we used the MDCK cell lines expressing p32-GFP in a tetracycline-inducible manner described in the previous paragraph. Noninduced MDCK cells grown in a collagen gel formed cysts with a regular lumen and well-defined, F-actin–rich apical membrane domains (, top). PKCζ also localized to the apical membrane domains (, top). However, upon the induction of p32-GFP expression, cysts showed striking morphological changes, with expanded apical membrane domains and irregularly shaped lumens (, bottom). PKCζ no longer localized to the apical membrane domains, but instead concentrated at apical sites of cell–cell contacts. Interestingly, an expansion of the apical membrane domains into the basal membrane domains was also observed in mLgl1/2 knockdown MDCK cells (). We next examined the effect of p32 overexpression on the localization of the adherens junction marker E-cadherin and the tight junction component zonula occludens-1 (ZO-1). Under noninduced conditions, E-cadherin localized in a well-defined strip at the basolateral membrane. ZO-1 localized at the tight junction as a distinctive spot in close proximity to the internal lumen (, top). In cells induced to overexpress p32-GFP, the linear localization of E-cadherin at the basolateral membrane domain was lost and E-cadherin localized more diffusely (, bottom). ZO-1 localization was also disturbed and no longer restricted to the tight junctions (, bottom). We quantified the cell polarity defects in p32-overexpressing cells by determining the percentage of cysts with regular apical actin enrichment and normal lumen formation (Fig. S3 B). Although 77% of noninduced cysts had this normal phenotype, only 15% of the induced cysts did, supporting an involvement of p32 in cell polarity. Because aPKC inhibition using dominant-negative mutants in epithelial cells has been shown to cause a loss of apical membrane domain identity (), we examined the effect of a cell-permeable PKCζ inhibitor on cell polarity. In the presence of the PKCζ inhibitor, cysts failed to form normally, as indicated by the lack of apical actin enrichment, mislocalization of ZO-1, and failure to form lumens (, top images). To investigate whether the effect of p32 overexpression on cell polarity involves PKCζ activation, we studied whether p32 overexpression could rescue the phenotype induced by the PKCζ inhibitor. As shown in (bottom images), overexpression of p32 partially reverted the defects in apical actin enrichment and lumen formation, but not in ZO-1 localization. We conclude that p32 regulates cell polarity at least in part by modulating PKCζ activity. Finally, we examined whether the knockdown of p32 affects epithelial cell polarity. We established MDCK cell lines stably expressing p32 short hairpin RNA (shRNA) in a tetracycline-inducible manner. In these cells, the level of endogenous p32 protein decreased by >90% in the presence of tetracycline ( and Fig. S3 C). Even in the absence of tetracycline, a slight decrease in p32 expression was observed compared with nontransfected parental cells, which can sometimes be seen with the Tet-on expression system (, top). In the absence of tetracycline, p32 shRNA cells formed cysts where ZO-1 localized at the subapical region (, second from top). In contrast, the addition of tetracycline induced cell polarity defects, as indicated by the mislocalization of ZO-1 to the basolateral and basal membrane domains (, second from top; more images are shown in Fig. S3 D). A similar mislocalization of ZO-1 to the basolateral membrane domain was also reported in MDCK cells overexpressing mLgl2 (). Statistical analysis showed a significant difference between tet (−) and tet (+) in p32 shRNA cells (, third panel). The slight increase of cell polarity defects in tet (−) p32 shRNA cells compared with those in parental cells may be due to decreased p32 expression (, third panel). The expression of p32 shRNA also induced mislocalization of PKCζ to the basolateral membrane domain or the cytosol (, bottom), further suggesting a role of p32 in regulating aPKC. Collectively, these data indicate a physiological role of p32 in cell polarity. The deregulation of cell polarity proteins often induces a more invasive phenotype in malignant tumors (; ). In this study, we show that overexpression and knockdown of p32 disturb the normal polarization of MDCK cells (). Interestingly, p32 is overexpressed in thyroid, colon, pancreatic, gastric, esophageal, and lung adenocarcinoma, but not in their nonmalignant counterparts (). Furthermore, p32 is differentially expressed during the progression of epidermal carcinoma and accumulates in metastatic islands (). Our results suggest that p32 not only serves as a marker of malignant cells but also may function as an oncoprotein. The cell polarity defects observed in p32 knockdown cells further suggest that precisely controlled levels of p32 protein may be essential for normal polarization. In conclusion, we have identified p32 as a novel mLgl2-interacting protein that forms a transient complex with mLgl2 and PKCζ. In this protein complex, p32 regulates cell polarity through its ability to enhance the kinase activity of PKCζ. For each experiment, five adult rats were killed and kidneys were dissected and homogenized in buffer A (5 mM Tris/HCl, pH 7.5, 320 mM sucrose, and 10 μM PMSF). The homogenate was fractionated as previously described (), and the detergent fraction was applied to 250 μl amylose resin beads (New England Biolabs, Inc.) coupled to 30 μg MBP-mLgl2C-WT or -SD. Beads were washed three times in buffer B (20 mM Hepes/NaOH, pH 7.4, 1 mM DTT, 5 mM MgCl, 2% NP-40, 150 mM NaCl, and 10 μM PMSF) followed by elution using buffer B with 10 mM maltose. The eluted samples were separated by SDS-PAGE, and proteins were visualized using SYPRO Ruby protein gel stain (Invitrogen). MBP-mLgl2–interacting proteins were excised from the gel, and the amino acid sequences were determined by LC-MS/MS. mLgl2-interacting protein gel bands were excised and in-gel digested with trypsin. The digest mixtures were separated by nanoscale liquid chromatography (LC Packings) on reverse-phase C18 column (150 × 0.075 mm ID; flow rate 0.15 ml/min). The eluate was introduced directly into a Q-STAR Pulsar-i-hybrid quadruple time of flight mass spectrometer (MDS Sciex). The spectra were searched against a National Center for Biotechnology Information nonredundant database with MASCOT MS/MS Ions search (Matrix Science). Immunoprecipitation was performed as described before (). When indicated, immunoprecipitates were incubated with 400 U λ-phosphatase (New England Biolabs, Inc.) in λ-phosphatase buffer (50 mM Tris/HCl, pH 7.5, 100 mM NaCl, 0.1 mM EGTA, 0.01% Brij 35, and 2 mM MgCl) or 125 ng recombinant PKCζ (Calbiochem) in phosphorylation buffer (20 mM Tris/HCl, pH 7.5, 5 mM MgCl, 1 mM EGTA, and 40 μM ATP) for 30 min at the indicated temperature. pFLAG-CMV2-mLgl1 was provided by T. Pawson (Mount Sinai Hospital, Toronto, Canada). MBP-mLgl2 pull down from HEK293 lysates was performed as described for immunoprecipitation, except that 10 μg MBP-mLgl2C-WT were conjugated to 50 μl amylose resin beads. Western blotting was performed as described previously () with the exception that blots in (A, B, and D), , Fig. S1 (A, C, and F), Fig. S2 (A and B), and Fig. S3 (A and C) were visualized using an infrared detection system (Odyssey; Licor Bioscience). In experiments with recombinant proteins, we used 1.6 μg His-p32 and His-Hakai and 10 μg MBP-mLgl2C-WT and MBP-Rac-WT proteins. To obtain MDCK cell lines stably expressing p32-GFP in an inducible manner, we used the tet-on system (Invitrogen). First, MDCK cells were transfected with pcDNA6/TR, followed by selection in a medium containing 5 μg ml blasticidin. Then, pcDNA4/TO/p32-GFP was used for the second transfection, and doubly transfected cells were selected in a tetracycline-free medium containing 5 μg ml blasticidin and 400 μg ml Zeocin. Induction of p32-GFP expression after addition of 2 μg/ml tetracycline was monitored over time and confirmed by Western blotting. Four independent clones were obtained. To obtain MDCK cell lines stably expressing p32 shRNA in a tetracycline-inducible manner, the same procedure was performed, except that pSUPERIOR-p32 shRNA was used for the second transfection and 800 μg ml G418 was added to the medium for selection. Three independent clones were obtained. Cysts were processed for immunofluorescence staining as described previously (). Rabbit anti–mLgl2-S653P-2 antibody was provided by S. Ohno (Yokohama City University, Yokohama, Japan). Cysts were mounted using ProLong Gold Antifade reagent (Invitrogen) and analyzed on a confocal microscope (TCS SP5; Leica) with a 63× NA 1.4 oil-immersion objective at room temperature (Leica). 65 z sections at 1-μm intervals were captured per cyst, and images were acquired using the Leica Application Suite. Images shown are single sections through the center of the cysts. Images were colorized and contrast was enhanced linearly using the Volocity software package (Improvision), and brightness was adjusted using Photoshop CS (Adobe). For the statistical analysis of cell polarity defects, noninduced and p32-GFP–overexpressing MDCK cell cysts were analyzed at room temperature using a microscope (DM IRB; Leica) with a 10× 0.25 air objective (Leica). Images were captured using a camera (Orca; Hamamatsu) and Openlab software (Improvision). Secondary antibodies used were goat anti-mouse Alexa 488 and goat anti-rabbit Alexa 647 (Invitrogen), as well as goat anti-rat Rhodamine Red-X (Jackson ImmunoResearch Laboratories). To visualize actin and nuclei, we used TRITC-labeled phalloidin (Sigma-Aldrich) and Hoechst 33342 (Invitrogen), respectively. 3D cell culture was performed as described previously () except that chamberslides (BD Biosciences) were used instead of filter inserts. In brief, MDCK cells were seeded into a 2 mg/ml collagen-I gel at a concentration of 2.5 × 10 cells/ml, covered with 500 μl of culture medium and cultured for 6 or 9 d. For induction of p32-GFP expression, the culture medium was supplied with 2 μg/ml tetracycline from day 1, and the medium was replaced daily. For inhibition of PKCζ, a cell-permeable myristoylated PKCζ pseudosubstrate inhibitor (Biosource International) was added to cysts at a concentration of 25 μM from day 1. Before these experiments, we titrated the concentration of the inhibitor (25, 50, and 100 μM) to obtain an optimal condition where cell polarity was disturbed without cytotoxicity. For p32 knockdown, cells were supplied with tetracycline for 3 d before seeding into a collagen gel, and the medium was replaced daily. After immunoprecipitation, samples were incubated in 60 μl phosphorylation buffer with 9 μCi γ-[P]ATP and either 40 μM biotinylated PKCζ peptide substrate or 200 ng MBP-mLgl2C-WT at 30°C for 5 min. 100 nM Gö6983 (Calbiochem) or 0.5 mM PKCζ pseudosubstrate inhibitor (Calbiochem) was added when indicated. The kinetics of mLgl2 phosphorylation by endogenous PKCζ was first determined, and a 5-min incubation time was used for all experiments. After the kinase reaction, 50 μl of the reaction buffer was incubated with either 20 μl streptavidin agarose (Sigma- Aldrich) or 50 μl amylose resin (New England Biolabs, Inc.) at 4°C for 30 min, followed by intensive washing. Radioactivity of the beads was measured using Cerenkov counting in a liquid scintillation analyzer (1900-TR; Packard Instrument Co.). For Fig. S2 E, phosphorylation by PKCα and PKCδ was performed as described previously (). After phosphorylation, the reaction was stopped by addition of 30 mM NaF and 100 mM EDTA, and the phosphorylated substrate was spotted onto P81 paper (Whatman), followed by intensive washing in 75 mM HPO and Cerenkov counting. The recombinant PKCα and PKCζ were purchased from Calbiochem, and PKCδ was obtained from Sigma-Aldrich. tests assuming equal or unequal variance were performed for statistical analysis. Fig. S1 shows characterization of the interaction between mLgl2 and p32. Fig. S2 shows that p32 interacts with mLgl2 transiently and enhances PKCζ activity. Fig. S3 shows that p32 expression level affects cell polarity of MDCK cells in 3D culture. The supplemental text provides details about the generation of plasmid constructs and sources of antibodies. Online supplemental material is available at .

Cytoplasmic dynein is a multisubunit complex that functions as a minus end–directed microtubule motor and plays critical roles in a variety of eukaryotic cellular functions, including retrograde axonal transport () and directed cell migration (). In addition, cytoplasmic dynein is involved in numerous aspects of mitosis, such as spindle pole organization, spindle orientation (; ; ), and mitotic checkpoint regulation (; ; ). Dynein behavior is mediated by numerous factors, including the dynactin complex, as well as by an additional group of regulatory proteins initially identified in . The gene in was found to be homologous to human LIS1, which causes the brain developmental disease type I lissencephaly when mutated (). This condition results from defects at several stages in the pathway of neurogenesis and migration in the neocortex and involves defects in both cell division and migration (). In vitro studies of LIS1 have revealed that it interacts physically with both cytoplasmic dynein and dynactin. Furthermore, LIS1 colocalizes with dynein at multiple subcellular sites, including the centrosome, kinetochores, mitotic cortex, and the leading edge of migrating cells (; ). In addition to LIS1, and were also identified in the dynein pathway. was first identified as a multicopy suppressor of () and has two mammalian homologues, NudE and NudEL (gene names and , respectively), which were identified in LIS1 two-hybrid screens (; ). These proteins are 55% identical in full-length sequence and are similar in both size and predicted secondary structure. Each interacts with LIS1 through a predicted N-terminal coiled-coil region (; ; ) and with cytoplasmic dynein through a globular C-terminal domain (; ; ). NudEL was reported to interact with the dynein motor domain by yeast two-hybrid and coexpression assays, suggesting a potential role in dynein motor regulation (). In spite of the structural similarities between NudE and NudEL, recent studies have revealed surprisingly different phenotypes in knockout mice. Homozygous -null mice are viable but display a microcephalic, or small brain, phenotype predominantly affecting the cerebral cortex (). An increase in mitotic index as well as spindle defects was reported in the developing brain, implying that NudE, like LIS1 (), participates in cell division. In contrast, -null mice exhibited early embryonic lethality (), a phenotype similar to that of both LIS1 () and cytoplasmic dynein (). However, hypomorphic mutants exhibited defects not in cell division but in neuronal distribution (), supporting previous RNAi data to this effect (). Both proteins have been localized to the centrosome (; ). Yeast two-hybrid data have identified interactions between NudE and several centrosomal proteins (), whereas coimmunoprecipitation studies identified γ-tubulin as an additional binding partner (; ). NudEL has also been implicated in dynein-mediated vesicular transport (), although localization of NudEL to membranous organelles has not been demonstrated. Additionally, NudE has been reported to localize to punctate structures within the mitotic spindle that were thought to be kinetochores (), although the extent to which the distribution of NudE overlapped with that of LIS1 and dynein was not examined. We initiated this study to determine the extent to which NudE and NudEL function in concert with LIS1, dynein, and dynactin and to test the degree to which NudE and NudEL differ functionally from each other. We find that both NudE and NudEL associate with mitotic kinetochores but arrive at these sites well in advance of dynein, LIS1, dynactin, and ZW10. The inhibition of NudE and NudEL function prevents dynein, dynactin, and LIS1 localization to kinetochores, leading to metaphase arrest and kinetochore misorientation. Finally, we find that NudE interacts with the dynein complex surprisingly through its tail, or stem, domain and shows no interaction with dynactin. These results identify a novel mechanism for kinetochore dynein recruitment and suggest important roles for NudE and NudEL in kinetochore assembly and microtubule attachment. NudE and NudEL have been reported to interact with cytoplasmic dynein and LIS1 (; ; ; ; ). We find that bacterially expressed GST-tagged full-length NudE pulls down LIS1 and dynein from bovine brain extract (). However, in contrast to LIS1 (; ), we observe no detectable interaction with dynactin or NudC (). Additionally, we found the dynein/dynactin-interacting protein ZW10 bound to NudE, biochemically () confirming an interaction previously identified in a ZW10 yeast two-hybrid screen (). To examine the distribution and in vivo functional properties of NudE, we produced a polyclonal antibody against the full-length protein mNudE. We find that the antibody recognizes a 43-kD band in 3T3, C6, LLC-PK1, MDCK, and COS7 cell lysates ( and not depicted). Interestingly, recombinant NudE and NudEL each reacted with the antibody (Fig. S1, available at ), which is consistent with their extended sequence homology. In addition to the major NudE/NudEL band at 43 kD, we occasionally detected additional bands at 75 and 105 kD, as has been previously reported for an anti-NudE antibody (); however, they are eliminated along with the major NudE/NudEL band by preadsorption against GST-NudE (Fig. S1), suggesting that they represent SDS-insensitive NudE and NudEL aggregates (). To determine the reactivity of the antibody under native conditions, we tested its behavior by immunoprecipitation. Despite the ability of recombinant NudE to pull down dynein, LIS1, and ZW10 (), the antibody coimmunoprecipitated only the latter two proteins (). This observation suggests that the antibody selectively blocks only the interaction of NudE and NudEL with dynein. The differential expression of NudE and NudEL during brain development has been reported (; ), but the extent to which expression overlaps in individual cell types has not been explored. To examine this issue, we conducted RT-PCR in 3T3, C6, LLC-PK1, and COS7 cell lysates. As seen in , both proteins were expressed in all cell lines examined. NudE and NudEL have been found to localize to centrosomes in interphase cells (; ; ; ) and to participate in vesicular transport (). Additionally, NudE was reported to localize to punctate structures within the mitotic spindle, which may be kinetochores (). To define the distribution of these proteins more fully and to test the extent to which they function in concert with cytoplasmic dynein, dynactin, and LIS1, we examined their distribution throughout the cell cycle, with particular emphasis on mitosis. Immunofluorescence staining using our NudE/NudEL antibody in COS7 cells showed colocalization with CREST human autoimmune serum on mitotic kinetochores (). To test whether this localization pattern reflected the behavior of NudE, NudEL, or both, each protein was expressed as a GFP fusion. GFP-NudE and GFP-NudEL each colocalized with LIS1 at mitotic kinetochores, a pattern observed in nocodazole-treated () and untreated (not depicted) COS7 epithelial cells. At high magnification, the NudE and NudEL staining pattern was slightly shifted from that of the CREST autoimmune signal, indicating that NudE and NudEL localize to the outer region of the kinetochore (, insets). Additionally, when cells were treated with nocodazole before fixation, NudE and NudEL () accumulated in a crescent pattern, which is a characteristic common to outer kinetochore proteins, including dynein and LIS1 (; ; ). Despite these localization similarities, we identified striking differences in the timing with which NudE and NudEL appeared and departed from mitotic kinetochores in comparison with dynein, dynactin, and LIS1. Immunofluorescence microscopy revealed strong NudE/NudEL kinetochore staining during prophase before either dynein, dynactin, or LIS1 ( and S2 a, available at ), suggesting that NudE/NudEL localization to mitotic kinetochores occurs independently of dynein. Furthermore, NudE and NudEL each remained at the kinetochore until early anaphase, well after dynein and LIS1 had departed ( a and S2 a). Surprisingly, GFP-NudE (unpublished data) and -NudEL ( and S2 b) also preceded ZW10 at kinetochores, although each of these proteins departed at a similar stage of mitosis. Cells were examined for cortical NudE and NudEL staining, as has previously been observed during mitosis for cytoplasmic dynein () and LIS1 (). We found NudE and NudEL to be absent from the mitotic cortex (). In contrast, NudE/NudEL showed strong localization to the cortex of partially polarized interphase MDCK epithelial cells, where we did not detect dynein or dynactin (). We also observed strong staining at the leading edge of migrating NIH3T3 fibroblasts in wounded monolayers (), as we have previously observed for LIS1, dynein, and dynactin (). We found no evidence of NudE/NudEL accumulation at microtubule plus ends as has been seen in other systems (; ). These data revealed similarities but also interesting differences in the distribution of NudE and NudEL from that of dynein, dynactin, and LIS1. Furthermore, we found that the localization patterns for NudE and NudEL were indistinguishable at mitotic kinetochores, suggesting that these proteins may have common cellular functions. NudE and NudEL RNAi has been shown to result in cell death (; ; our unpublished data). As an alternative approach, we examined the behavior of cells expressing NudE fragment (). Overexpression of an N-terminal fragment of NudE has been reported to interfere with mitosis (). As in a previous study (), we observe severe defects in mitotic spindle organization after the overexpression of an N-terminal NudE fragment (GFP-NudE-C188) but found no effect on spindle organization with a C-terminal fragment (GFP-NudE-N189) containing the dynein-binding region or with empty vector controls (). To gain further insight into the effects of these fragments on mitotic behavior, we monitored transfected cells by live imaging. Not unexpectedly in view of the fixed cell analysis, the N-terminal fragment caused a variety of effects, including multipolar divisions and prometaphase arrest (unpublished data). Surprisingly, overexpression of the C-terminal fragment had a more pronounced effect on mitotic progression. 44% of cells arrested in metaphase and subsequently underwent cell death, as indicated by nuclear condensation and cell shrinkage (), whereas an additional 44% of cells exhibited delayed anaphase onset ( = 16; ). Analysis of the mitotic index in fixed cells expressing the C-terminal NudE fragment revealed a striking decrease in mitotic index. The latter observation is similar to results of ZW10 inhibition (; ; ) and has been attributed to premature anaphase onset. Our live cell imaging data have revealed a dramatic delay in anaphase onset, and the basis for the reduced mitotic index in our system is uncertain. No effect on mitotic progression was seen in cells transfected with GFP alone ( = 6; ). In addition, we found the C-terminal fragment of NudE to localize to mitotic kinetochores, which is in contrast to the N-terminal fragment (). These results suggest that the C-terminal fragment is necessary for NudE localization to the kinetochore and could compete with full-length NudE or NudEL at this site. We note that the C-terminal fragment also contains a binding site for the early kinetochore protein CENP-F (centromere protein F), which may play a role in NudE and NudEL recruitment to mitotic kinetochores (). Despite the different patterns of inhibition we observe with the NudE fragments, they are both consistent with the inhibition of cytoplasmic dynein, which is known to participate in spindle organization () as well as the metaphase/anaphase transition. As spindle disorganization is itself likely to affect this transition indirectly, we sought a means for acute NudE and NudEL inhibition to test the specific function of these proteins at kinetochores. In view of the ability of our anti-NudE/NudEL antibody to block the interaction of these proteins with cytoplasmic dynein, we injected it into LLC-PK1 cells during prophase and followed the cells by time-lapse phase-contrast microscopy. 84% of injected cells exhibited a defect in mitotic progression ( = 13; ), whereas no effect was visible in control cells injected with preimmune serum ( = 6; ). Metaphase arrest was the most prominent phenotype, occurring in 69% of injected cells. We also examined additional cells that were fixed 60 min after injection and subsequently analyzed for kinetochore composition and orientation (see the next section). Of 80 such cells, 80% exhibited a metaphase-like chromosome configuration. Unlike the effects of LIS1 antibody injection (), only minor defects in chromosome congression and the stability of chromosome alignment were evident (15.3%). The majority of arrested cells ultimately showed evidence of cell death; however, a small subset of the cells progressed into anaphase (15.3%; ). To identify the site at which the antibody acted, injected cells were preextracted, fixed, and stained with secondary antibody. The injected anti-NudE/NudEL antibody exhibited a clear localization to kinetochores, indicating that endogenous NudE and NudEL were still present at these sites and that inhibition of their interaction with dynein occurred locally (). Although chromosome congression appeared normal by phase-contrast microscopy, more detailed immunofluorescence analysis at 60 min after injection found 61% of cells with evidence of at least one improperly oriented kinetochore pair relative to the spindle axis ( = 23; ). Of the cells containing misoriented kinetochores, 29% contained three or more misoriented pairs. Control cells treated with the proteasome inhibitor MG132 to prevent anaphase onset and subsequently injected with preimmune serum exhibited the proper orientation of almost all kinetochore pairs, although individual misoriented pairs could be detected in a small subset of injected cells (20%; = 15; ). In addition, the average distance between sister chromatids in NudE/NudEL antibody–injected cells (1.07 μm; = 33) was substantially shorter than that seen in control cells at metaphase (1.22 μm; = 46). After calculating the distance of unstretched centromeres in nocodazole-treated LLC-PK1 cells (0.58 μm; = 47), the degree of tension in anti-NudE/NudEL antibody–injected cells was found to be 77% of that of controls ([1.07 – 0.58]/[1.22 – 0.58] = 0.77), a decrease similar to that previously seen with dynein inhibition (). To determine whether the loss of tension was the result of unstable microtubule attachment at kinetochores, antibody-injected cells were subjected to cold treatment to depolymerize nonkinetochore microtubules. The number of kinetochore microtubule bundles appeared substantially decreased, although some bundles were still detected (). Examination of individual kinetochores found that many of the misoriented kinetochore pairs lacked microtubule attachments, although examples of unattached but seemingly well-oriented kinetochore pairs were also seen (). Whether the loss of kinetochore microtubules reflects a failure in initial attachment or in the stability of attachment is uncertain. The appearance of NudE and NudEL at kinetochores before dynein, dynactin, LIS1, and ZW10 suggested that NudE and NudEL might function to organize these proteins at mitotic kinetochores. To test the effect of NudE/NudEL inhibition on dynein distribution, we examined antibody-injected cells by immunofluorescence microscopy. Surprisingly, the injected cells showed a complete loss of dynein, dynactin, and LIS1 from all kinetochores, including those on chromosomes that had not yet congressed to the metaphase plate and would normally exhibit strong staining (). Quantitative analysis of kinetochore staining showed a 73%, 61%, and 72% reduction for dynein, dynactin, and LIS1, respectively, relative to preimmune-injected controls (Fig. S3, available at ). Additionally, antibody injection had no effect on ZW10 or Hec1 kinetochore localization ( and S3). We also examined kinetochore composition in cells expressing the C-terminal dominant-negative NudE fragment (GFP-NudE-N189). In contrast to the antibody-injected cells, the fragment had no apparent effect on dynactin, LIS1, or ZW10 localization to kinetochores (Fig. S4). To determine the status of the mitotic checkpoint in cells in which NudE/NudEL-dependent dynein recruitment has been inhibited, we stained cells injected with the anti-NudE/NudEL antibody for the checkpoint protein BubR1. Strong colocalization of BubR1 and CREST at mitotic kinetochores was detected in all cells examined at both properly oriented and misoriented kinetochores (). This result indicates that the NudE/NudEL-dependent loss of dynein prevents checkpoint protein removal even on kinetochores that have congressed to the metaphase plate. Although BubR1 levels have been shown to decrease on fully aligned metaphase chromosomes, Mad2 provides a more definitive indication of the mitotic checkpoint status, as it shows a greater depletion on aligned kinetochores () and is removed from kinetochores in a dynein-dependent manner (). Because of the difficulty in imaging Mad2 by immunofluorescence in anti-NudE/NudEL–injected cells, we performed sequential injections with anti-NudE/NudEL and anti-Mad2 antibodies and imaged cells using time-lapse phase-contrast microscopy. Cells were observed to enter anaphase ∼10 min after the second injection, indicating that the Mad2-dependent checkpoint was responsible for anti-NudE/NudEL–induced metaphase arrest ( = 5). In addition, cells exhibited lagging chromosomes during this precocious anaphase onset, which is consistent with our findings that kinetochore microtubule attachment was incomplete or unstable (). Although NudEL has been reported to interact with the cytoplasmic dynein motor domain (), this type of interaction seems inconsistent with a role for NudE and NudEL in anchoring the motor protein to kinetochores. To define the mechanism by which dynein, NudE, and NudEL interact, we tested the ability of recombinant NudE to bind to a well-behaved recombinant rat cytoplasmic dynein motor domain () as well as the cytoplasmic dynein complex purified from brain tissue (). Although the purified dynein complex exhibited a clear interaction with NudE (), as was found for dynein in brain cytosolic extract (), no interaction with the recombinant motor domain could be detected (). These data suggested that the interaction with the dynein complex might be through its stem, or tail, domain. To test this possibility, we performed GST-NudE pulldowns from COS7 cells transfected with cDNAs encoding individual dynein subunits. We observed a clear, strong interaction with both the dynein intermediate chain IC2C and the light chain LC8 (). No interaction was seen with the cytoplasmic dynein light intermediate chains LIC1 and LIC2 or a 1,137-aa fragment of the cytoplasmic dynein heavy chain corresponding to the entire base region of the molecule (). #text COS7, MDCK, and HeLa cells were grown in DME supplemented with 10% FBS, LLC-PK1 cells were grown in DME supplemented with 3% FBS, and NIH-3T3 fibroblasts were grown in DME supplemented with 10% bovine calf serum. Transient transfections were performed using Effectene (QIAGEN). Full-length NudE and NudEL as well as the NudE-C188 and NudE-N189 deletion constructs were each cloned from a mouse cDNA library using the Marathon cDNA Amplification kit (CLONTECH Laboratories, Inc.) into pEGFP-C1 (CLONTECH Laboratories, Inc.). The IC2C-myc, HC-C1140-myc, LIC1-HA, LIC2-FLAG, and LC8-VSVG constructs were described previously (; ,; ). We generated a polyclonal rabbit antibody against bacterially expressed full-length mNudE. Additional antibodies used included polyclonal anti-Arp1 (provided by D. Meyer, University of California, Los Angeles, Los Angeles, CA), monoclonal anti-p150 (BD Biosciences), polyclonal anti-LIS1 (Santa Cruz Biotechnology, Inc.), monoclonal anti-LIS1 (provided by O. Reiner, Weizmann Institute, Rehovot, Israel), monoclonal anti-BUBR1 (BD Biosciences), chicken anti-GFP (Chemicon), monoclonal HEC1 (Abcam), polyclonal anti–ZW10-Cter (), rat monoclonal tyrosinated anti–α-tubulin (YL1/2; ; Gomes et al., 2004), monoclonal antidynein intermediate chain (clone 70.1; Sigma-Aldrich), and human CREST autoimmune serum (Antibodies, Inc.). To generate stable cell lines, full-length NudE and NudEL were cloned into the localization and affinity purification vector pIC113. Individual colonies were screened for endogenous levels of expression and kept under stable selection. Full-length NudE was cloned into pGEX6p-1 (GE Healthcare) and purified from bacteria by glutathione-Sepharose affinity chromatography. Bovine brain lysate and cytoplasmic dynein were prepared as previously described in phosphate-glutamate buffer (). Individual dynein subunits or fragments were expressed in COS7 cells. Pulldowns were performed by adsorbing GST-NudE to glutathione-Sepharose and rocking gently for 1–2 h in the presence of bovine brain extract, 2.5 μg of purified cytoplasmic dynein (∼6 nM), 10 μg of purified motor domain (∼75 nM), or transfected cell lysate. Immunoprecipitations from bovine brain extract were performed by overnight incubation using protein A–Sepharose. NudE- and NudEL-interacting proteins were identified by immunoblotting. Total RNA was isolated from cell lysates using the Absolutely RNA RT-PCR Miniprep kit (Stratagene), processed using RETROscript (Ambion), and analyzed by agarose gel electrophoresis. Cells were preextracted using 0.1% Triton X-100 in PHEM buffer (120 mM Pipes, 50 mM Hepes, 20 mM EGTA, and 4 mM magnesium acetate) for 30 s, fixed in 3% PFA in PHEM buffer for 20 min, and fixed in methanol at −20°C for 6 min (). Coverslips were blocked for 30 min with 0.5% BSA, incubated for 1 h in primary antibody, washed, and incubated for 1 h using Cy2-, Cy3-, and Cy5-conjugated secondary antibodies (Jackson ImmunoResearch Laboratories). To stain chromosomes, cells were subsequently exposed to DAPI for 10 min and mounted using ProLong Gold antifade reagent (Invitrogen). Cells were then visualized on an inverted microscope (described below) or by confocal imaging (510 META; Carl Zeiss MicroImaging, Inc.). Analysis of kinetochore orientation and microtubule attachment was performed on 3D stacks that were acquired using a spinning disk confocal microscope (DSU; Olympus) on an inverted microscope (IX80; Olympus). For all antibody injection experiments, a purified IgG fraction of our anti-NudE/NudEL polyclonal antibody was concentrated in microinjection buffer (8–15 mg ml in 50 mM potassium glutamate containing 0.5 mM MgCl, pH 7.0). Preimmune IgG serum was used in control injections of cells treated with 10 μM MG132. Cells were visualized at 37°C in a 5% CO atmosphere using an inverted microscope equipped with an incubation chamber (DMIRBE; Leica) and injected during prophase. For live cell analysis, images were collected every 4 min after injection for up to 120 min with a CCD camera (ORCA 100; Hamamatsu) piloted by MetaMorph (Universal Imaging Corp.). Fig. S1 includes additional information on the biochemical analysis of our polyclonal anti-NudE/NudEL antibody. Fig. S2 shows the distinct mitotic localization of GFP-NudEL in comparison with LIS1 and ZW10. Fig. S3 contains images of kinetochore composition in preimmune-injected control cells. Fig. S4 demonstrates the effects of NudE-N189 overexpression on kinetochore composition. Online supplemental material is available at .

The nuclear envelope (NE) consists of an outer nuclear membrane and an inner nuclear membrane (INM). The outer nuclear membrane is continuous with the membrane system of the ER, whereas the INM contains a specific set of transmembrane proteins and is closely associated with the nuclear lamina and the chromatin. At sites where both membranes are fused, nuclear pore complexes (NPCs) are inserted, which serve the receptor-mediated exchange of macromolecules between the nucleus and the cytoplasm. The small GTPase Ran plays a pivotal role in determining the directionality of nuclear transport during interphase of the cell cycle, but it is also used to mark the position and identity of chromatin during mitosis. In interphase, Ran is enriched in the nucleus, where it is in its GTP-bound form as a result of the action of the chromatin-bound guanyl-nucleotide exchange factor RCC1. In the cytoplasm, RanGTP is readily converted to RanGDP by the RanGTPase-activating protein (RanGAP) that stimulates the GTPase activity of Ran. During mitosis, the generation of RanGTP around chromatin persists (), providing spatial information for spindle formation and NE assembly (for reviews see ; ). At the onset of mitosis, major structural reorganizations of the cell occur, including NE breakdown (NEBD), condensation of chromosomes, and formation of a mitotic spindle. NEBD involves the disassembly of the NPCs, the depolymerization and solubilization of the lamina, and the detachment and removal of the nuclear membrane from chromatin, resulting in the redistribution of NE membrane proteins to the ER network (; ). NEBD is a phosphorylation-dependent process. Phosphorylation of NE components is thought to disrupt the protein–protein interactions required for nuclear integrity. Several kinases have been implicated in the nuclear disassembly process, namely Cdk1–cyclin B, PKC (for review see ), NIMA (never in mitosis A; ; ), Cdk–cyclin A2 (), and others (). The activation of Cdk1–cyclin B leads to the mitotic hyperphosphorylation of lamins, resulting in the depolymerization of higher order lamin polymers and solubilization of the lamin proteins (; ; ; ). Besides Cdk1–cyclin B, PKC is required for NEBD, and the PKC isoform PKCβII phosphorylates lamin B (; ; ). Other constituents of the NE are also targets for mitotic phosphorylation, including INM proteins (; ; ; ) and nucleoporins (; ; ; ), which are the constituents of the NPC. Interestingly, nucleoporins might be involved in NEBD beyond being phosphorylation substrates. Certain nucleoporins have been suggested to serve as landing pads for the COPI (coat protein I) coatamer complex, which might assist NE disassembly in a yet to be defined mechanism (; ). Studies in embryos and starfish oocytes suggest that NPC disassembly is the initial step of NEBD (; ; ). When the relative timing of NPC disassembly, NE rupture, and lamina solubilization was investigated in starfish oocytes, two phases of NE permeabilization were observed. During the first phase of NEBD, NPCs became partially dismantled, allowing the influx of a 70-kD fluorescent dextran. The NE structure, including the lamina, remained intact during this first phase. Complete permeabilization of the NE during phase two resulted in a fenestration of the membrane (detected by the influx of 500-kD dextran) followed by the complete disassembly of the lamina. Fenestration is thought to represent the complete removal of the NPCs (). This two-step process is explained by the initial phosphorylation of nucleoporins facilitating partial NPC disassembly. The increasing NPC permeability would then allow kinases to enter the nucleus, to phosphorylate their targets, and to trigger lamina and final NPC disassembly. In addition to the mitotic phosphorylation of NE components, a microtubule-based tearing process assists NE disassembly in somatic cells (; ). Dynein, which is recruited to the NE at prophase (; ; ), interacts with spindle microtubules, thereby creating tension on the NE, which finally leads to its rupture (; ). Rupture starts with the formation of one to three holes in the NE, which then rapidly expand over the nuclear surface. Interference with microtubule function by microtubule-depolymerizing drugs does not inhibit but delays NEBD (; ). Although a general description of the dynamic process of NEBD is starting to emerge, very little is known about the molecular machinery behind it. We have established a visual assay to study NEBD in vitro that allows for monitoring morphological changes of the NE in semipermeabilized somatic cells. To induce NEBD, we use mitotic egg extracts, which are amenable to biochemical treatments. Importantly, the use of fully activated mitotic extracts enables the dissection of NEBD independently of signaling events leading to mitotic entry in vivo. This in vitro assay allows for the molecular characterization of events leading to mitotic nuclear disassembly. Using this system, we have investigated the molecular requirements of mitotic nuclear breakdown. Our analysis uncovered an important role for the RanGTPase system in the final steps of nuclear disassembly and expands previous evidence for a supportive function of microtubules in NEBD (; ). To study the molecular requirements of NEBD, we have established an in vitro system that allows for the observation of NEBD by 4D fluorescence microscopy. Our assay uses semipermeabilized HeLa cells derived from a cell line expressing a fusion of GFP to the transmembrane and lamin-binding domains of lamina-associated polypeptide 2β (LAP2β), a protein of the INM. GFP-LAP2β–expressing HeLa cells grown on coverslips were semipermeabilized with digitonin. The nuclei were then incubated with egg extracts either in an interphase or a mitotic state. Changes in the permeability of the NE were visualized by nuclear influx of a TRITC-labeled 155-kD dextran. Nuclei incubated in interphase extract remained intact over the 45-min time course of the experiment ( and Video 1, available at ). Neither visible changes in the structure of the NE nor changes in its permeability occurred. The incubation of nuclei in cytostatic factor (CSF)–arrested egg extracts induced nuclear disassembly. In 70–80% of the analyzed nuclei, NEBD started ∼10–15 min after the addition of the mitotic extract. A first visible indication of nuclear disassembly was the influx of the TRITC-labeled 155-kD dextran accompanied by some shrinkage of the nuclei. This first phase of NEBD was followed by the formation and expansion of holes such that roughly 40% of the nuclei contained holes 25–35 min after the addition of mitotic extract. At this time point, all nuclei were permeable for the fluorescent dextran. Complete nuclear breakdown, which is characterized by a dynamic rupture of the NE, occurred ∼35–45 min after addition of the CSF extract ( and Video 2). Variations of the kinetics of the disassembly process were observed depending on the quality of the extracts and the condition of the cells. When we compared the relative timing of influx of two differently sized (70 or 155 kD) fluorescent dextrans in the first 20 min of NEBD (Fig. S1, available at ), we noticed that both dextrans entered the nuclei with almost identical kinetics, with the entry of the 155-kD dextran being delayed by only ∼1 min. The influx of both dextrans was rapid, whereas the loss of the GFP-tagged nucleoporins Nup58 and Nup98 from the nuclear rim appeared to be a gradual process, suggesting that the change in NE permeability does not require the disassembly of all copies of these nucleoporins at once (Fig. S1). Because of the almost identical kinetics of the influx of both dextrans during the disassembly reaction, we decided to use the 155-kD species to monitor changes in NE permeability in all further experiments. Two kinases have been directly implicated in NEBD, namely Cdk1–cyclin B1 and PKCβII (; ; ; ; ; ; ). To verify that our system truly recapitulates NEBD, we tested whether NEBD in vitro was dependent on Cdk1 and PKC by using specific inhibitors of these kinases. Alsterpaullone was chosen as a Cdk1 inhibitor (). Gö6983 was used as a PKC inhibitor, as it inhibits several PKC isoforms, including PKCβ (). Inhibition of Cdk1 by treatment of the CSF extract with alsterpaullone strongly delayed the initiation of NEBD. Permeabilization of the nuclei started with the influx of dextran after ∼20–25 min. NEBD proceeded slowly, and nuclei with holes could be only rarely detected after 45 min. The complete disassembly of nuclei failed ( and Video 3, available at ). Inhibition of PKC activity by the addition of Gö6983 also inhibited nuclear disassembly, but differently (). The initiation of NEBD was not substantially delayed, as dextran started to enter the nuclei at the same time as in control cells, and the nuclei also shrunk. However, most nuclei failed to form holes and to fenestrate; none disassembled. The activity and specificity of the inhibitors were tested in a histone H1 phosphorylation assay. The addition of the inhibitors to the CSF extract showed that histone H1 phosphorylation was strongly reduced by alsterpaullone (reduction by 80%) and only moderately by the PKC inhibitor Gö6983 (reduction by 35%), showing that Cdk1 is the main histone H1 phosphorylating activity in the extract (). As expected, phosphorylation of histone H1 by recombinant Cdk1–cyclin B1 was reduced upon treatment with the Cdk1 inhibitor alsterpaullone. The PKC inhibitor Gö6983 did not affect Cdk1 activity, but it strongly reduced the phosphorylation of histone H1 by PKCβII. In reverse, alsterpaullone had no effect on PKCβII activity (). Distinct subcellular localizations of key mitotic regulators are crucial for the initiation and progression through mitosis. This includes the nuclear accumulation of Cdk1–cyclin B1 at prophase before nuclear disassembly, a process that has been suggested to be important for Cdk1 activation (; ). Cdk1 present in our mitotic extract is fully active, making the nuclear import of Cdk1 unnecessary for its activation. This enabled us to directly test whether there are other requirements for nuclear import in the in vitro NEBD process or whether NEBD can, in principle, start from the cytoplasmic side. To block nuclear import, we used a dominant-negative mutant of the nuclear import receptor importin β, Impβ, which is known to block all active transport through the NPC by stably binding to nucleoporins (). Impβ reliably blocked nuclear transport over the timespan of the experiment when tested in interphase extracts on semipermeabilized cells (Fig. S2, available at ). The addition of Impβ to the disassembly reaction did not affect the kinetics of the initiation of NEBD ( and Video 4). The influx of TRITC-labeled 155-kD dextran occurred with normal kinetics. Dextran entered 100% of the nuclei, and hole formation occurred as usual in ∼40% of the nuclei. Neither the formation nor the expansion of these holes was affected by the inhibition of nuclear transport. The only observed effect of Impβ addition was that some of the completely holey nuclei failed to undergo the final dynamic rupture process. This is supported by high magnification confocal sectioning of disassembled nuclei at the end points of these disassembly reactions (). To verify that nuclear import is not required for the onset of NEBD in the presence of fully activated mitotic extracts, we next tested the effect of RanQ69L addition. This GTPase-deficient Ran mutant stays GTP bound in the presence of RanGAP in the extracts, causing the dissociation of import cargoes from nuclear import receptors and, thereby, preventing the nuclear import of most nuclear proteins (; ). The addition of 20 μM RanQ69L(GTP) to the CSF extract did not block the initial steps of nuclear disassembly (). The timing of dextran influx was unchanged. Compared with the control experiments, fewer nuclei formed holes, and, interestingly, these holes remained static and hardly expanded over the nuclear surface (Video 5, available at ). The final dynamic disassembly of the nuclei was blocked, as none of the nuclei disassembled after 45 min, underscoring this undynamic behavior. This RanQ69L(GTP)-mediated strong delay of NEBD is likely not caused by the inhibition of nuclear import because the initial kinetics of NE permeabilization was unchanged. NPC disassembly started normally, as shown by the unchanged kinetics of Nup58 dissociation from the nuclear rim (Fig. S3). RanQ69L(GTP) most probably affects a later, dynamic step of nuclear breakdown, which normally facilitates the expansion of holes and, thus, disassembly. The phenotype of inhibition by RanQ69L(GTP) was different to treatment of the extracts with Impβ, which did not block complete disassembly. Disassembly in the presence of RanQ69L(GTP) was delayed at an earlier point, resulting in nuclei with fewer holes (). The RanQ69L(GTP) effect could indicate that the GTPase activity of Ran is required for NEBD or, alternatively, that unbalancing the RanGDP/RanGTP ratio in the extract might cause disassembly defects. When endogenous RanGDP was converted to RanGTP by adding an excess of recombinant RCC1 to the extracts (; and Video 6, available at ), NEBD was inhibited at the same stage as observed upon RanQ69L(GTP) addition. This indicates that a high RanGTP concentration in the egg extract affects NEBD and that proper function of the RanGTPase system might be required for NE disassembly. Inhibition of directed nuclear transport itself does not seem to be responsible for the RanGTP-mediated disassembly defect. However, RanGTP might influence NEBD via the characteristic feature of nuclear transport receptors to bind or release their cargo in a RanGTP-dependent manner. For instance, RanQ69L(GTP) could stimulate export factors to sequester a component required for NEBD. A simple explanation for the effect of RanQ69L(GTP) or RCC1 could thus be that their addition alters the activity of Cdk1–cyclin B in the extracts by driving the kinase heterodimer into a potentially inactive complex with an exportin (for instance, with CRM1; ; ). Therefore, we analyzed the mitotic kinase activity of egg extracts in the presence of RanQ69L(GTP) or RCC1. The histone H1 phosphorylation assay shows that mitotic kinase activity of the extracts is not influenced by these treatments (). Alternatively, the RanGTP effect on NEBD might be explained by a requirement for importins, which could aid nuclear disassembly by sequestering nuclear (envelope) components in the mitotic cytosol, a process prevented by RanQ69L(GTP). To address the involvement of importins, we used known competitors of cargo binding to different import receptors (). Strikingly, addition of the importin β–binding domain (IBB) of the NLS import adaptor importin α (; ) caused a similar block in NEBD as did RanGTP. NE permeabilization, as judged by the influx of fluorescent dextran, was normal, but NE fenestration and dynamic rupture were blocked. In contrast, the addition of two other nuclear transport competitors of distinct specificity, M9 () and BIB (), had no effect on NEBD. The effect of IBB was not the result of the inhibition of mitotic kinase activity in the extracts (). Many of the mitotic roles of the RanGTPase system are known to impinge on importin β (for review see ), and our new data suggest that importin β's capacity to associate with cargo is also required during late steps of NEBD. We further found that this role of importin β involves the binding of cargo to the NLS adaptor importin α, as NEBD was also impaired by a high concentration of a BSA-NLS conjugate (Fig. S4, available at ). The production of RanGTP normally occurs in the neighborhood of chromatin, where the RanGEF RCC1 is localized. RanGTP production around mitotic chromatin provides the positional information used for the local release of cofactors from inhibitory complexes with importins to allow for spindle assembly and NE reformation after mitosis (for reviews see ; ). Likewise, we reasoned that the activation of a disassembly-promoting factor close to chromatin aided by RanGTP might assist NEBD. To prevent RanGTP production around chromatin, we used a Ran mutant (RanT24N), which strongly binds and inhibits RCC1 because of its low affinity for guanyl nucleotides (). To block RCC1 activity, nuclei of semipermeabilized cells were preloaded with RanT24N. Then, mitotic extract supplemented with RanT24N was added, and NEBD was monitored by live microscopy. Strikingly, RanT24N did not block early steps of the nuclear disassembly process but blocked the final dynamic NE rupture ( and Video 7, available at ). Inspection of the nuclei at late time points revealed that nuclei were slightly perforated in the presence of RanT24N, indicating that NE disassembly had proceeded further than hole formation (). Control experiments showed that RanT24N was indeed blocking RCC1 activity, as Ran-dependent nuclear import was efficiently inhibited after preloading of nuclei with the Ran mutant (Fig. S2). The dynamic rupture of the nuclei in our in vitro disassembly assay hinted at a possible involvement of the microtubule cytoskeleton. To test whether complete NEBD in vitro depended on forces exerted by microtubule-associated motor proteins, the CSF extract used for the disassembly reaction was treated with nocodazole. Early steps in NEBD were not affected by nocodazole, but, strikingly, the dynamic rupture of the nuclei was blocked () even when the nuclei were incubated with the extract over time as long as 100 min (Video 8, available at ). Treatment with nocodazole had no influence on the activity of mitotic kinases in the extract, as shown by the unchanged ability of the extracts to phosphorylate histone H1 (). Previous studies have suggested that a microtubule-based tearing mechanism supports the formation of holes in the NE at the onset of NEBD (; ). Our in vitro analysis indicated that microtubules might also aid a later step, namely the removal of NE membranes from the vicinity of chromatin. To gain in vivo evidence for our observation, we analyzed the fate of GFP-LAP2β in nocodazole-treated HeLa cells in comparison with untreated cells. Time-lapse imaging revealed that in both cases, GFP-LAP2β redistributed into tubular membrane structures at the onset of mitosis, which is consistent with the redistribution of INM proteins into the ER. Strikingly, in nocodazole-treated cells, the GFP-LAP2β–containing membranes remained longer in close proximity to chromatin, which was visualized by H2B-mRFP (Videos 9 and 10, available at ). To gain a more quantitative insight into the effect of nocodazole on the timing of NE/ER removal from chromatin, we defined the starting point of NEBD by measuring changes in NE permeability () using a mPlum-GST-M9 fusion, which localizes to the nucleus of interphase cells (as a result of constant nuclear import) and leaks out into the cytoplasm at the beginning of mitosis. 6, 10, and 14 min after the detection of mPlum-GST-M9 in the cytoplasm, marking the onset of NEBD, we took serial sections through cells (). Quantitative analysis of the images by measurement of the distances between the NE/ER membrane and chromatin supported the idea that the NE/ER network remained in the vicinity of the condensing chromatin for a longer timespan (). Whereas in untreated cells, membranes had been efficiently removed from the chromatin area, this process was significantly delayed in nocodazole-treated cells (P < 0.01). After 6 min, membrane removal was less efficient by at least a factor of six, and, 14 min after the initiation of NEBD, there was still a difference by a factor of two. Together, this analysis reveals that microtubules play an important role in clearing the chromatin area from the NE/ER membrane network. The microtubule minus end–directed motor protein dynein and its regulator dynactin localize to the NE late during G2 before any visible changes in mitotic NE organization occur (). Dynein is thought to then promote the microtubule-dependent reorganization of the NE during prophase and NEBD. Overexpression of the dynactin subunits p62 or p50/dynamitin interferes with dynein function and was found to delay NEBD and to abolish NE invaginations during prophase, respectively (). To test whether microtubule-dependent steps in NEBD in vitro are dependent on cytoplasmic dynein, we added recombinant p50/dynamitin to in vitro nuclear disassembly reactions. Strikingly, the dynamic disassembly of the nuclei was inhibited, whereas permeabilization and fenestration of the NE occurred with unchanged kinetics (Fig. S4). Thus, the microtubule dependence of NEBD in vivo and in vitro appears to rely on the same molecular mechanism. To visualize microtubule-based NE disintegration in vitro, we added rhodamine-labeled tubulin to the egg extracts. The first visible sign of microtubule polymerization was seen in close proximity to the nuclei ∼5–10 min after the addition of extract and was enhanced upon the formation of holes in the NE after 15 min (). Then, spindlelike structures formed at the position of the nuclear remnants. Subsequently, NE pieces were torn apart in association with microtubules, supporting the notion that the microtubule cytoskeleton directly aids the peripheral scattering of NE membranes. We have developed a visual assay to study NEBD in vitro that allows for monitoring morphological changes of the NE in semipermeabilized somatic cells. This in vitro assay provides a powerful tool to investigate the molecular mechanisms of NEBD. The progression of NEBD in vitro resembles the nuclear disassembly observed in vivo. Nuclear breakdown starts with permeabilization of the NE, which is followed by the formation of few holes, NE fenestration, and, finally, rupture of the nuclei. Also, on the molecular level, the in vitro system faithfully recapitulates many of the known requirements for NEBD. First, in our system, early events of NEBD are dependent on Cdk1 and PKC activity. Cdk1 is the master regulator of mitotic entry, and it has long been difficult to uncouple its potential involvement in NEBD from its general requirement for progression into mitosis. Our system uses extracts containing fully activated Cdk1. The requirement for Cdk1 activity to initiate NEBD in vitro implies a direct involvement of Cdk1 in the initiation of nuclear disassembly, likely by inducing NPC disassembly. This assumption is supported by the ability of Cdk1 to phosphorylate several nucleoporins in vitro (). Interestingly, a recent study provided evidence for direct involvement of a Cdk–cyclin A complex in NEBD in vivo (). The depletion of cyclin A2 from HeLa cells by RNAi did not affect cyclin B1–Cdk1 activation but delayed NEBD (). However, it remains unclear whether cyclin A functions in NEBD in complex with Cdk1 or Cdk2. Earlier studies had implicated PKC in mitotic entry. PKCβII is a well-characterized lamin B kinase (; ). Inhibition of PKC in synchronized human promyelocytic leukemia (HL60) cells leads to an arrest in G2 without inhibiting Cdk1 (), but it is currently unclear whether this G2 arrest is solely explained by the disturbed phosphorylation of NE components or whether events upstream of NEBD are affected. Interestingly, our data revealed a differential requirement for Cdk1 and PKC activity in NEBD. The inhibition of Cdk1 delayed permeabilization of the NE and, therefore, is most probably important for NPC disassembly. However, PKC inhibition only mildly affected the kinetics of NE permeabilization, but all subsequent processes like hole formation and NE fenestration are blocked. Together, this indicates that Cdk1 might be required earlier during NEBD than PKC. Second, consistent with in vivo studies analyzing NEBD in maturing starfish oocytes (), we provide evidence that NEBD can be initiated from the cytoplasmic side independently of nuclear transport. It has been proposed that regulation of the nucleocytoplasmic localization of protein kinases is crucial for initiation and progression through mitosis. Mitotic kinases such as Cdk1–cyclin B (, ) and PKC () accumulate in the nucleus before NEBD. The nuclear accumulation of Cdk1–cyclin B has been suggested to be critical for its complete activation as well as for triggering mitotic nuclear events (for review see ). We used the dominant-negative mutant Impβ as an inhibitor of receptor-mediated nuclear transport and did not observe any delay in early events of NEBD. NE permeabilization and hole formation occurred as in control experiments, suggesting that in our in vitro system, in which mitotic kinases are fully active, NEBD can be initiated from the cytoplasmic side independently of nuclear transport. This is consistent with the notion that nuclear breakdown is initiated by nucleoporin phosphorylation, triggering NPC disassembly. Third, we observed a strong dependence on microtubule dynamics for efficient nuclear disassembly in vitro. Previous in vivo studies demonstrated that microtubules are involved in the initial events of NEBD by mechanically supporting the formation of holes in the nuclear lamina and nuclear membrane (; ). In contrast to the in vivo situation, however, in which NEBD still occurs after the treatment of cells with nocodazole, NE rupture never occurred in the presence of this drug in our in vitro assay, even when NEBD was monitored for as long as 2 h. A reason for the strict dependence on microtubule-based tearing in our in vitro system might lie in a failure to retract INM proteins into the ER. We currently do not know why our system fails to reproduce this process, as the ER appears to be intact after cell permeabilization (not depicted). Time-lapse microscopy of living cells had previously revealed that NEBD is already delayed at early steps like the initial hole formation of the NE when exposed to nocodazole (; ). In addition, microtubules may also aid subsequent steps of nuclear disintegration, as NE markers are still present in the vicinity of chromatin in prometaphase cells after nocodazole treatment, as suggested by the examination of fixed cells (). Our in vivo analysis of the INM protein LAP2β in cells entering mitosis extends these previous observations. We observed that the dispersal of GFP-LAP2β into the ER in prophase occurs with similar dynamics as in untreated cells, whereas removal of the membrane from the chromatin area was strongly delayed by nocodazole. Therefore, it seems that microtubules are not only involved in the early steps of NEBD like hole formation but are also important during a later step of NEBD when the NE/ER network is pulled away from the chromatin toward the centrosomes. Previous studies using nuclei that were first assembled in vitro in egg extracts and dismantled in the presence of mitotic egg extracts have implicated the COPI coatomer complex and ADP-ribosylation factor (ARF) in NEBD (). Analysis of NE disintegration in this experimental set-up by electron microscopy revealed ER-like membrane tubules and vesicles emanating from the NE (). Membrane disassembly in this system is sensitive to brefeldin A and inhibited by ARF peptides. In our system, which uses nuclei of semipermeabilized somatic cells, we were unable to detect any inhibition of NEBD by brefeldin A or ARF peptides (unpublished data), indicating that the two experimental systems might differ. In living somatic cells, NE membranes retract back into the ER without any obvious sign of vesiculation (). Therefore, it remains to be seen whether there is a necessity for a vesiculation pathway during the disassembly of somatic cell nuclei in vivo. One great advantage of the in vitro system lies in the opportunity to interfere with the function of components that are required throughout the cell cycle in living cells and to specifically investigate their function in NEBD. One such factor is RanGTP, which defines the identity of chromatin throughout the cell cycle. Indeed, our results show that the proper balance between RanGTP in the vicinity of chromatin and RanGDP in the mitotic extracts is required for late steps of NEBD in vitro but not for initial NE permeabilization, supporting the finding that nuclear import is not required for initial events in NEBD in this system. The RanGTPase system is involved in other mitotic events like spindle assembly and NE reformation (for reviews see ; ; ; ). Studies on spindle formation using egg extracts demonstrated that RanGTP induces the release of microtubule-associated proteins like TPX2 or NuMA from importins α and β, thereby stimulating microtubule assembly (; ). Likewise, the release of nucleoporins from complexes with importin β by RanGTP in the vicinity of chromatin has been suggested to spatially regulate the recruitment of nucleoporins to chromatin at the end of mitosis (; ). Similarly, block of the final stages of in vitro NEBD induced by RanGTP or ectopic RCC1 might be ascribed to RanGTP's role in regulating the binding of cargo to nuclear transport receptors. As NEBD was inhibited by excess of the IBB of importin α in a similar way as by RanGTP, it is tempting to speculate that sequestration of NE components by importin β in the mitotic cytosol might facilitate late steps in NEBD. Both RanGTP and IBB would prevent such a function of importin β. Candidate NE components sequestered by importin β are nuclear lamins that carry classical nuclear localization signals (). Furthermore, several INM proteins contain NLS-like sequences, and it has recently been suggested that importins α and β are directly involved in binding to and escorting INM proteins across the NPC to the nuclear interior (). Mitotic depolymerization of the nuclear lamina depends on the phosphorylation of lamins and lamin-binding proteins of the INM. However, efficient dissolving of chromatin-lamina-INM contacts might not only require the phosphorylation of structural components but also their sequestration in the mitotic cytosol by binding to importins, thereby preventing their tendency to repolymerize. Thus, high levels of RanGTP and IBB might interfere with the sequestration of these proteins; repolymerization might occur and result in a block of dynamic rupture of the nuclear membrane. Strikingly, RanGTP has been reported to induce the polymerization of lamin-containing structures in mitotic extracts in vitro (). According to a different but not mutually exclusive scenario, RanQ69L(GTP) and IBB may exert their effects on microtubules that aid NE disintegration. High levels of RanGTP or the presence of IBB, which are both known to induce chromatin- independent microtubule aster formation in egg extracts (; ; ; ; ), might hinder the normal attachment of microtubules to the NE. Notably, not only high levels of RanGTP but also RanT24N, which inhibits RanGTP production by RCC1, interfered with the final processes of NEBD. However, the phenotypes of RanT24N and RanQ69L(GTP) addition on NEBD differed in that RanT24N only inhibited the final dynamic rupture of NE, leaving the NE fully fenestrated, whereas RanQ69L(GTP) treatment essentially blocked NEBD after hole formation. Strikingly, both treatments affected the productive tearing of NE remnants along microtubules in bipolar spindlelike structures (Fig. S5, available at ). Whereas RanT24N addition reduced microtubule polymerization/stability, RanQ69L induced the formation of multiple asters around the holey nuclei, which, however, were nonproductive in exerting forces on the NE. Together, our observations can be reconciled by the existence of several distinct steps late during NEBD that differentially rely on the proper function of the RanGTPase system. Perhaps the transition from hole formation to fenestration requires the sequestration of NE components in the mitotic cytosol (inhibited by RanQ69L(GTP), ectopic RCC1, and IBB), whereas a subsequent step of microtubule-based NE tearing depends on the local production of RanGTP around chromatin, likely by aiding proper spindle assembly. Clearly, the RanGTPase system also plays a pivotal role in mitotic entry in somatic cells in vivo, not only in controlling the nucleocytoplasmic distribution of mitotic regulators. Similar to the situation in our in vitro system, microinjection of RanQ69L(GTP) into HeLa prophase cells induces ectopic microtubule asters around the NE/chromatin (), but it has not yet been addressed whether RanQ69L(GTP) injection affects NE disintegration. Such in vivo, studies are complicated by the multiple roles of the RanGTPase system. Therefore, one would benefit from knowing the molecular targets of Ran action in NEBD. Exploiting the in vitro NEBD system will certainly help their identification in the future. pcDNA3GFP-LAP2β (aa 244–453) was a gift from T. Rapoport (Harvard Medical School, Boston, MA). mRFP-LAP2β was generated by replacing EGFP with the mRFP coding sequence () in the backbone of pEGFP-C3 (CLONTECH Laboratories, Inc.) and subsequently inserting the coding sequence of LAP2β (aa 244–453) into the HindIII and BamHI sites. pEGFP-Nup58 and pEGFP-Nup98 were gifts from J. Ellenberg (European Molecular Biology Laboratory, Heidelberg, Germany; ). H2B-mRFP () in a modified pIRESpuro vector (CLONTECH Laboratories, Inc.) was a gift from D. Gerlich (Swiss Federal Institute of Technology, Zurich, Switzerland). mPlum-GST-M9 was generated by replacing the coding sequence of EGFP in pEGFP-C1 (CLONTECH Laboratories, Inc.) by the coding sequence of mPlum () and ligating the coding sequence of the GST-M9 fusion protein into the BglII and HindIII sites. Plasmids used for the recombinant protein expression of Impβ, wild-type Ran, RanQ69L, RanT24N, RCC1, GST-M9, IBB-GST, and GST-BIB have been described previously (). The plasmid for the bacterial expression of p50/dynamitin was a gift from T. Hyman (Max Planck Institute, Dresden, Germany). For the expression of Cdk1–cyclin B1 in insect cells, the coding sequences of cyclin B1 (GenBank/EMBL/DDBJ accession no. ) and Cdk1 (GenBank/EMBL/DDBJ accession no. ) were amplified by PCR from HeLa cell cDNA and were cloned into the NcoI–HindIII and BamHI–HindIII sites, respectively, of the pFastBac HTb vector from the BAC-TO-BAC Baculovirus Expression System (Invitrogen). Transposition into the bacmid, transfection of Sf9 insect cells, and amplification of viral stocks were performed as described by the manufacturer. Stable cell lines were obtained after the transfection of HeLa cells with the plasmid coding for GFP-LAP2β followed by selection for positive clones with 500 μg/ml G418. Subsequently, H2B-mRFP was introduced into this cell line, and double stable clones were selected with 500 μg/ml G418 and 0.5 μg/ml puromycin. Transfections were performed using FuGENE 6 (Roche). HeLa cells were maintained in complete DME (containing 10% FCS, 100 U/ml penicillin, and 100 μg/ml streptomycin) and the appropriate selection drugs. For in vivo imaging, cells were cultured in chambered coverglasses (LabTekII; Nunc) in complete DME without phenol red (Invitrogen) containing an additional 10% FCS. Expression and purification of wild-type Ran, RanQ69L, RanT24N, RCC1, Impβ, GST-M9, IBB-GST, and GST-BIB have been described previously (). Expression and purification of p50/dynamitin has been described previously (). Before addition to CSF extracts, the buffer of wild-type Ran, RanQ69L, RanT24N, RCC1, GST-M9, IBB-GST, and GST-BIB was exchanged to 50 mM Hepes, pH 7.4, 250 mM KoAc, and 2 mM MgOAc. The buffer of p50/dynamitin was exchanged to permeabilization buffer (PB). Cdk1 and cyclin B1 were coexpressed in baculovirus-infected Sf21 cells. 3 d after infection, the cells were harvested, washed once, and resuspended in 1 vol of ice-cold hypotonic buffer (10 mM Hepes, pH 7.5, 25 mM NaCl, 0.5 mM EDTA, 10 μg/ml cytochalasin B, 2 mM PMSF, 2 μg/ml leupeptin, 2 μg/ml pepstatin A, 5 mM NaF, and 1 mM NaVO). After Dounce homogenization, the NaCl concentration was adjusted to 150 mM, the lysate was cleared by centrifugation, and MgCl was added to a final concentration of 5 mM. Cdk1–cyclin B1 was purified via Ni-NTA-agarose chromatography. After elution, protein-containing fractions were pooled and concentrated to a volume of 100–200 μl. The Cdk1–cyclin B1 complex was separated from monomeric Cdk1 and cyclin B1 by gel filtration using a Superdex 200 HR 10/30 column (GE Healthcare) in 25 mM Hepes, pH 7.5, 150 mM NaCl, 10% (wt/vol) glycerol, and 1 mM DTT. PKCβII was purchased from Panvera. Alsterpaullone and Gö6983 were purchased from Calbiochem. Nocodazole was purchased from Sigma-Aldrich. Priming of for ovulation and preparation of interphase and CSF-arrested (mitotic) egg extracts were performed essentially as described previously (). Eggs were washed once in 2% cysteine, pH 8.0, dejellied for 5 min in 2% cysteine, pH 8.0, and washed three times in 1 mM Hepes, pH 7.8, 20 mM NaCl, 0.4 mM KCl, 0.2 mM MgSO, 0.5 mM CaCl, and 16 μM EDTA. For interphase egg extract preparation, eggs were washed three times in 10 mM Hepes, pH 7.7, 50 mM KCl, 2.5 mM MgCl, 1 mM DTT, and 250 mM sucrose. For CSF extract preparation, eggs were washed three times in 10 mM Hepes, pH 7.7, 100 mM KCl, 0.1 mM CaCl, 1 mM MgCl, 50 mM sucrose, and washed three times in the same solution additionally containing 1 mM MgCl and 5 mM EGTA. After washing, the eggs were transferred into centrifuge tubes (model 331372; Beckman Coulter), packed by spinning at 400 for 5 min in a SW41 rotor (Beckman Coulter), and the extra buffer was removed from the top of the eggs. Aprotinin, leupeptin, and cytochalasin B were added to a final concentration of 5 μg/ml each, and caspase inhibitors (caspase-3 inhibitor I and caspase inhibitor II; Calbiochem) were added to a final concentration of 20 nM. Eggs were crushed by spinning at 12,000 for 20 min, and crude extracts were harvested from the centrifuge tubes using a 19-gauge needle. The protein concentration of the crude extracts is ∼30 mg/ml. Crude extracts were supplemented with 250 mM sucrose and stored in small aliquots at −80°C. After thawing, an energy-regenerating system was added to 0.25 mM GTP, 0.25 mM ATP, 5 mM creatine phosphate, and 25 μg/ml creatine kinase. Preloading of HeLa cell nuclei with Ran or RanT24N was performed by incubation of permeabilized cells for 7 min with PB containing 7.5 μM RanT24N, 1.5 μM NTF2, 0.75 μM RanBP1, and 0.5 μM RanGAP. Preincubation of permeabilized nuclei with Impβ was performed for 7 min with PB containing 15 μM Impβ. Note that all control disassembly reactions contained a buffer control with equal amounts of solvents or buffers that have been used for the addition of inhibitors or recombinant proteins. 1 μl of egg extract was incubated with 9 μl H1 kinase buffer (80 mM β-glycerophosphate, pH 7.4, 15 mM MgCl, 20 mM EGTA, 5 mM PMSF, 10 μg/ml leupeptin, 10 μg/ml aprotinin, 5 mM NaF, 1 mM NaVO, and 0.1 mM ATP) containing 3 μg histone H1 and 2 μCi γ-[P]ATP. To assay the kinase activity of recombinant kinases, 0.5 μg of purified Cdk1–cyclin B1 and 0.04 μg PKCβII were incubated with 9 μl H1 kinase buffer containing 3 μg histone H1 and 2 μCi γ-[P]ATP. The kinase reaction was performed for 10 min at 30°C and stopped by the addition of 30 μl SDS sample buffer. One fourth of each reaction was run on a 14% polyacrylamide gel. For confocal laser-scanning microscopy of in vitro disassembly reactions, a 63× 1.4 NA differential interference contrast plan Apochromat oil immersion objective (Carl Zeiss MicroImaging, Inc.) mounted on a microscope (Axiovert 200; Carl Zeiss MicroImaging, Inc.) and a confocal scanning module (excitation at 488 and 543 nm; LSM510 META; Carl Zeiss MicroImaging, Inc.) were used. In 5-min intervals, confocal sections of selected areas were captured using macros allowing multiposition time-lapse image acquisition and image file concatenation (). Live cell imaging was performed using a 63× 1.4 NA differential interference contrast plan Apochromat oil immersion objective (Carl Zeiss MicroImaging, Inc.) mounted on a customized confocal microscope (excitation 488 and 561 nm; LSM510; Carl Zeiss MicroImaging, Inc.) equipped with a temperature- and CO-controlled incubator box. To monitor NEBD of individual cells with high time resolution, image stacks of 512 × 512 × 7 with a width of 35.7 μm and a 1-μm step size were acquired every 15 s (pixel time of 0.80 μs; line average 1). For subsequent measurements of distances between chromatin and NE/ER membranes, stacks of 512 × 512 × 4 with a width of 35.7 μm and a step size of 1 μm were acquired at the indicated times (pixel time of 0.80 μs; line average 8). Images were further analyzed and processed using LSM software (Carl Zeiss MicroImaging, Inc.) and ImageJ (National Institutes of Health; ). Distances between the chromatin and the NE/ER membrane system were determined using the ImageJ software and a customized version of the radial grid plug-in. A radial grid of 20 lines (angles between lines were 18°) was placed in the center of the chromatin mass. In all four slices of a stack, the distance between the chromatin and the NE/ER membrane was manually determined along the lines of the grid. Videos 1–7 show time-lapse videos corresponding to the experiments presented in Figs. 1 (Video 1, interphase extract; Video 2, CSF extract), 2 A (Video 3, Cdk1 inhibition), 3 (Video 4, Impβ addition), 4 (Video 5, RanQ69L addition; Video 6, RCC1 addition), and 6 (Video 7, RanT24N addition). Video 8 shows a time-lapse video of NEBD in vitro in the presence of nocodazole over a time of 100 min. Videos 9 and 10 compare the progression of GFP-LAP2β–expressing HeLa cells into mitosis in the absence or presence of nocodazole, respectively. Fig. S1 shows the time course of the nuclear influx of 70- and 155-kD dextrans as well as the kinetics of nucleoporin disassembly using GFP-Nup58 and GFP-Nup98 as markers. Fig. S2 demonstrates that nuclear accumulation of a fluorescently labeled NLS-containing substrate is blocked by Impβ, RanQ69L(GTP), and RanT24N, inhibited by RCC1, and competed by unlabeled BSA-NLS conjugates. Fig. S3 shows that the addition of RanQ69L does not interfere with the initial steps of NPC disassembly. Fig. S4 provides evidence that late steps of NEBD in vitro are inhibited by the addition of BSA-NLS conjugates or by recombinant p50/dynamitin. In Fig. S5, we compare the effect of RanT24N and RanQ69L on microtubule polymerization around nuclei during NEBD in vitro. Online supplemental material is available at .

GTPases comprise a superfamily of proteins that provide molecular switches to regulate many cellular processes, including protein synthesis, signal transduction, cytoskeletal organization, vesicle transport, nuclear transport, spindle assembly, and many more (; ; ). The classic work on signaling GTPases, such as Ras, has established a “GTPase switch” paradigm in which a GTPase alternates between two distinct conformational and functional states: an active, GTP-bound state and an inactive, GDP-bound state. Both states are kinetically stable and are thus separated from one another temporally. Interconversion between the GTP- and GDP-bound states is facilitated by external regulatory factors, such as GTPase-activating proteins and guanine nucleotide exchange factors. This allows a GTPase to switch between “on” and “off” states in temporal succession in response to extra- or intracellular signaling cues. Two homologous GTPases, one in the SRP54 subunit of the signal recognition particle (SRP) and one in the SRP receptor (SR; called Ffh and FtsY in bacteria, respectively), mediate the cotranslational targeting of membrane and secretory proteins to the eukaryotic endoplasmic reticulum membrane, or the bacterial plasma membrane. During the targeting reaction, SRP and SR switch between different functional states (; ). At the beginning of each targeting cycle, SRP binds to a nascent polypeptide that contains a signal sequence as it emerges from the ribosome (; ). The ribosome–nascent chain complex (RNC) is then delivered to the membrane via an interaction between the SRP and SR. Upon arrival at the membrane, SRP releases its “cargo,” the RNC, to the translocation channel, or the translocon (; ,). Once the RNC is released, SRP and SR dissociate from each other, allowing another cycle of protein targeting to occur (; ). Analogous to other GTPases, the switches in the functional states of SRP and SR are coordinated by their GTPase cycles. Numerous biochemical experiments have shown that formation of a stable SRP–SR complex requires both GTPases to be bound with GTP or nonhydrolyzable GTP analogues (; ; ); thus, GTP binding is critical for delivery of the cargo protein to the target membrane. However, less data are available on the role of GTP hydrolysis in the protein targeting reaction. In the classical experiment by , it was shown that a nonhydrolyzable GTP analogue, 5′-guanylylimido-diphosphate (GMPPNP), can substitute for GTP and mediate a single round of protein translocation. However, in the presence of GMPPNP, the SRP and SR GTPases are irreversibly trapped in a stable complex and cannot mediate subsequent rounds of protein targeting (; ). These observations led to the current model in which GTP hydrolysis is not important for the targeting reaction per se, but is used to drive the disassembly of the SRP–SR complex, thus regenerating free SRPs and SRs for subsequent rounds of protein targeting. Nevertheless, this classical model might be an oversimplified picture, as the behavior of SRP and SR is modeled in analogy to the canonical GTPase switch mechanism, in which regulation is exerted by the switch of GTPases between the GTP- and GDP-bound states. However, the SRP and SR GTPases exhibit biochemical properties that are distinct from those of classical GTPases, and hence are likely to use different regulatory mechanisms. Unlike many other GTPases, both SRP and SR exhibit weak nucleotide affinities and fast nucleotide exchange rates (, ; ), and thus do not require external nucleotide exchange factors. Furthermore, crystallographic analyses showed that the conformations of the SRP and SR GTPases are similar regardless of whether GTP or GDP is bound (, ; ; ); thus the exchange of GDP for GTP is unlikely to be the mechanism that switches these GTPases to the on state. In addition, SRP and SR reciprocally activate each other's GTPase activity once they form a complex with each other (); thus regulation of GTP hydrolysis by an external GTPase-activating protein is unlikely to be the mechanism that turns these GTPases to the off state. Instead, the recent crystal structure of the GTPase domains of the Ffh–FtsY complex showed that these GTPases undergo large-scale conformational changes only after they form a complex with one another when each GTPase is already bound with GTP (; ). Compared with the structures of the apoproteins, two major conformational changes are observed upon complex formation. A major rearrangement occurs at the interface between the central GTPase G domain and the N domain, a unique insertion in the SRP subgroup of GTPases that packs tightly against the G domain. The readjustment of the relative position of the N and G domains allows the N domains of both proteins to bend toward its binding partner and form additional interface interactions with one another. The other major rearrangement occurs at the insertion box domain (IBD) loops of both GTPases. This loop is highly conserved in the SRP subfamily of GTPases but is not present in other GTPases. Upon complex formation, each IBD loop brings three key catalytic residues into the GTPase site of their respective protein to position and activate the nucleophilic water molecule and to stabilize the negative charges on the γ-phosphate. Consistent with the crystal structure, multiple distinct classes of mutant GTPases have been isolated, each defective at a different step during the SRP–FtsY interaction (; ). Mutations of many residues at the N–G domain interface severely impair SRP–FtsY binding (class I), supporting the importance of this domain rearrangement for complex formation, as well as the importance of an extensive interaction surface that pays for the energetic cost of conformational changes during complex formation. Surprisingly, even after a stable complex is formed, single mutations in FtsY can block the reciprocal activation of GTP hydrolysis in both active sites (class II or “activation-defective” mutants). Thus, activation requires additional conformational changes across the interface that coordinate the positioning of catalytic residues and that are highly coupled between the two GTPase sites. Most of these mutations map to the IBD loop, supporting the importance of this loop in GTPase activation. A distinct class of mutants exhibit half-site reactivity and allows us to further uncouple the activation of the individual sites (class IV or “half-site” mutants). These mutants suggest the presence of additional conformational changes that complete the individual active sites. These distinct classes of mutant GTPases strongly suggest that the SRP–FtsY interaction is a dynamic process involving multiple, discrete conformational changes that culminate in the activation of GTP hydrolysis. These results also raise the intriguing possibility that instead of using external regulatory factors, the conformational rearrangements during SRP–SR complex formation and activation may provide critical points for regulation during the protein targeting reaction (). To test this notion, we have examined the effect of the different classes of mutant GTPases on the protein targeting reaction. Surprisingly, the class II, or activation-defective, mutants severely block protein translocation, even though assembly of the SRP–SR complex is unimpaired in these mutants. Thus the activation of GTP hydrolysis in the SRP–SR complex plays a much more important role in the protein targeting reaction than was previously thought. To test the effect of mutant GTPases on the protein targeting reaction, we first developed an assay that reports on the efficiency of translocation by the bacterial SRP and SR (FtsY). Most of the existing assays in the bacterial system are qualitative, relying either on protease protection of the protein substrate by membrane vesicles after translocation (, ) or on the ability of the protein substrate to cross-link to the SRP or to the sec translocon (, ). In contrast, a much more robust translocation assay exists for eukaryotic systems, because translocation of a eukaryotic protein across the ER microsomal membrane results in efficient cleavage of the signal sequence by signal peptidase, allowing pre- and mature proteins to be resolved by SDS-PAGE, so that both reaction substrates and products can be visualized and the fraction of translocated protein can be quantitated. A similar assay does not yet exist for bacterial systems because most of the substrates for bacterial SRP are inner membrane proteins whose signal sequences are not cleaved upon translocation. For these reasons, we decided to use a heterologous protein translocation assay pioneered by . In this assay, wheat germ (WG) translation extract is used to synthesize a mammalian SRP substrate, preprolactin (pPL). We then assessed the ability of bacterial SRP and FtsY to deliver pPL to microsomal membranes in which endogenous SRP and SR have been removed by TKRM (a high salt wash and partial trypsin digestion). To best mimic the in vivo targeting reaction, we designed the assay to report on translocation cotranslationally (). Shortly after translation is initiated, a cap analogue, 7-methyl-GTP, is added to inhibit additional rounds of translation initiation, such that translocation of only the first round of translation product is followed. SRP (the Ffh protein bound to the 4.5S SRP RNA), FtsY, and TKRM are added to allow translocation of nascent pPL. Translation is continued for 20–30 min to allow completion of pPL synthesis, at which time the reaction is stopped and analyzed. Consistent with previous results (), translocation of pPL is very robust in this heterologous system () and depends on the concentration of SRP, FtsY, and TKRM (; and see , , and ). To probe the sensitivity and dynamic range of this targeting assay, we tested the translocation efficiency of mutant FtsY GTPases that block SRP–FtsY binding and therefore are expected to compromise the delivery of pPL to the membrane (, class I mutants). Three FtsY mutants were tested. FtsY E475K specifically compromises SRP–FtsY binding by ∼20-fold, but still allows efficient activation of GTP hydrolysis when complexes with SRP form at high protein concentrations (). FtsY K399A has a more severe defect, compromising SRP–FtsY complex formation by 30-fold (). In contrast, FtsY T307A blocks both complex formation and GTPase activation by >200-fold, as Thr307 is one of the key residues that coordinate the Mg ion in the GTPase active site (). As expected, mutant FtsY T307A almost completely blocks translocation of pPL ( and Fig. S3 A, squares, available at ), whereas mutants FtsY E475K and FtsY K399A reduce the translocation efficiency more modestly (two- and threefold, respectively; and Fig. S3 A, triangles and closed circles). Other class I FtsY mutants that compromise SRP–FtsY binding by more moderate amounts (three- to fivefold) do not show a considerable translocation defect (not depicted). Thus we conclude that this assay can reliably detect translocation defects if SRP–SR complex formation is weakened by >20-fold. In contrast, more moderate defects are masked, presumably because targeting and translocation of preproteins occur much faster than protein synthesis (Fig. S1) and only become rate-limiting when the translocation efficiency is compromised beyond a certain threshold. Nevertheless, the assay reliably detects defects of GTPase mutations that substantially compromise the efficiency of protein targeting. With the assay in hand, we tested the translocation efficiency of another class of FtsY mutants. Class II (activation-defective) mutants allow a stable SRP–FtsY complex to be assembled, but specifically block the reciprocal activation of both GTPase sites (; ). The GTPase activity of these mutants was previously characterized in the context of a truncated version of FtsY, FtsY(47–497), in which the N-terminal 46 amino acids were removed to allow better expression and solubility of the protein. These amino acids are not important for FtsY's GTPase activity or for its interaction with SRP. Nevertheless, as the N-terminal A domain of FtsY (residues 1–196) has been implicated in its membrane association (; ; ; ), we reintroduced these mutations into full-length FtsY. However, cells harboring most of the class II FtsY mutants were sick and grew slowly, and only small quantities of these proteins were produced. Nevertheless, we succeeded in purifying sufficient quantities of full-length FtsY bearing two of the class II mutations (FtsY R386A and FtsY N302W; ). The other class II mutants (FtsY A335W, FtsY A336W, and FtsY R333A) could not be expressed, and were therefore only characterized in the context of FtsY(47–497) and compared with the wild-type FtsY(47–497) protein. In our assay, FtsY(47–497) exhibited only a twofold reduction in translocation efficiency compared with full-length FtsY (Fig. S2, available at ). The mildness of the translocation defect exhibited by FtsY(47–497) may stem from the fact that only a small portion of the A domain of FtsY is removed in this construct and/or that our cotranslational targeting assay can only detect relatively large defects in protein targeting as discussed in the previous section. A recent study also suggests that the majority of the A domain of FtsY is not essential for the function of FtsY, as a truncated FtsY(196–497) construct containing only Phe196 in its A domain can rescue protein targeting and cell growth in vivo (). It therefore seems reasonable that characterization of the mutants in this targeting assay yields meaningful results even in the context of the truncated FtsY(47–497). All of the class II mutants compromise the efficiency of the translocation reaction, both in the context of full-length FtsY ( and Fig. S3 B) and the FtsY(47–497) protein ( and Fig. S3 C). The translocation defect of each mutant compared with the respective wild-type protein is the same, within the margin of error, regardless of whether the mutations are introduced in the context of the full-length or truncated FtsY. Except for mutant FtsY R333A, a correlation can be found between the translocation defect of each mutant and the degree to which reciprocal GTPase activation is blocked in the mutant SRP–FtsY complex (). For example, one of the most severe mutants, FtsY A335W, which binds SRP with wild-type affinity but reduces the stimulated GTPase rate by >50-fold, almost completely blocks pPL translocation (). In contrast, the mutant FtsY R386A, which reduces the GTPase rate by sixfold, causes only a threefold reduction in translocation efficiency (). The deviation observed with mutant FtsY R333A (, triangle) potentially stems from the fact that, whereas the other mutants inhibit GTP hydrolysis primarily by blocking the conformational rearrangement that leads to GTPase activation, the FtsY R333A mutation also removes a key catalytic residue that directly participates in the chemical reaction (). As discussed later, the rearrangement of the IBD loop is critical for protein translocation, whereas GTP hydrolysis is not (see and and Discussion). Therefore, the FtsY R333A mutant is more active in protein translocation than would be expected from its reduction in GTPase rate. The data presented so far show that all of the class II FtsY mutants severely compromise the efficiency of protein targeting. Three of these mutations (FtsY R333A, FtsY A335W, and FtsY A336W) map to the highly conserved IBD loop. Because the IBD loop is conserved between the two GTPases and the catalytic interactions made by each IBD loop with the respective GTP molecule are highly symmetrical (; ), we hypothesized that the residues in the IBD loop of the SRP GTPase also play a crucial role in two aspects: reciprocal GTPase activation and protein targeting. We therefore introduced the class II mutations, so far only characterized in FtsY, into the homologous positions of the SRP GTPase Ffh (). Mutant Ffh proteins were first characterized in terms of their basal GTPase cycles and their stimulated GTPase reaction upon interaction with FtsY. Most of the mutations do not affect the basal GTP binding and hydrolysis cycle of Ffh, except for Ffh R141A and Ffh A144W, which reduce the basal GTP hydrolysis rate by 8- and 20-fold, respectively (unpublished data). Notably, of the five class II mutants in FtsY, all three mutations in the IBD loop also substantially reduce the rate of stimulated GTP hydrolysis from the SRP–FtsY complex when introduced into homologous positions in Ffh ( and Fig. S3 D). The effects of the other two mutations in this class are much milder when introduced into homologous positions in Ffh. The Ffh R194A mutation reduces the rate of stimulated GTP hydrolysis less than threefold, and the Ffh Q109A mutation has no effect on the stimulated GTPase reaction ( and Fig. S3 E). Thus, there is a break in the functional symmetry of residues not residing in the IBD loop. These results indicate that the most conserved and symmetrical feature between the two GTPases are the catalytic interactions made by the IBD loops. In FtsY, the class II mutations in this loop can still allow a stable complex to assemble. To test if this is also true for the corresponding mutations in the IBD loop of Ffh, we used a slight modification of an inhibition assay previously developed () to determine the ability of each mutant SRP to inhibit the interaction of wild-type SRP with FtsY. This assay allowed us to selectively monitor complex formation between FtsY and the mutant SRPs. The conditions of the assay were designed so that in the absence of any mutant SRP as an inhibitor, a robust GTPase reaction mediated by wild-type SRP and FtsY was observed (, ). Addition of mutant SRP, SRP(mt), which can form a complex with FtsY, will sequester the FtsY molecules into a less active SRP(mt)–FtsY complex ( ≪ ), thus inhibiting the observed GTPase reaction. All three mutants are strong competitive inhibitors (), indicating that these mutant SRPs can form a strong complex with FtsY. The observed inhibition constants range from 260 to 390 nM for the three SRP mutants. However, because at least 300 nM FtsY needs to be present to allow a sufficient amount of GTPase reaction, and at least as much mutant SRP is needed to sequester all the FtsY molecules into the SRP(mt)–FtsY complex, the measured apparent inhibition constants represent an upper limit for the actual affinity of the mutant SRPs for FtsY. Thus mutations in the IBD loop of Ffh also result in the class II phenotype, with the mutant SRPs able to form a stable complex with FtsY but failing to efficiently activate GTP hydrolysis in the complex. Analogous to the results obtained with the FtsY that belong to this class, these class II SRP mutants also exhibit a substantial defect in mediating translocation of pPL ( and Fig. S3 F). The translocation defect for each mutant correlates well with the amount of reduction in the stimulated GTP hydrolysis rates from the mutant SRP–FtsY complex (). In contrast, mutants Ffh Q109A and Ffh R194A (which map outside the IBD loop) exhibit no translocation defect ( and Fig. S3 G). This is consistent with the observation that the Ffh Q109A mutation does not affect the activated GTPase reaction in the SRP–FtsY complex, and the Ffh R194A mutant has only a mild effect on GTPase activation (), and this small effect is not sufficient to manifest itself as a substantial translocation defect given the sensitivity of our targeting assay (see first section of Results). Our analyses of the mutant GTPases indicate that blocking GTPase activation in the SRP–FtsY complex severely impairs protein translocation. This is surprising in light of the results from the pioneering studies by , which showed that a nonhydrolyzable GTP analogue, GMPPNP, can allow a single round of protein translocation to occur in the mammalian SRP system. To ensure that the translocation defect we observed with the mutant GTPases is not caused by the use of heterologous components, we reexamined the nucleotide requirement for translocation in our system. The cotranslational assay () was inappropriate for this purpose, however, as GTP is also required for ongoing protein translation in addition to translocation. We therefore used an alternative assay in which a stalled RNC was generated by translation of a truncated mRNA that encodes the first 86 amino acids of pPL (pPL). Nucleotides were removed from RNC by gel filtration chromatography (), and targeting of purified RNC to TKRM by bacterial SRP and FtsY was assayed in the presence of various nucleotides (). Both GTP and GMPPNP mediated efficient translocation of pPL, as indicated by the production of PL as pPL was translocated across the microsomal membranes and processed by signal peptidase (). In contrast, in the presence of GDP or with no nucleotide added, no more than background levels of translocation were observed (). The class II mutant FtsYs described in the previous section still exhibit a large translocation defect in this posttranslational assay (unpublished data). These results confirm the conclusions by and demonstrate that the translocation defect of the mutant GTPases arises from a block of the conformational rearrangements that lead to GTPase activation, rather than inhibition of the chemical reaction of GTP hydrolysis itself. The two GTPases in SRP and SR use their GTPase cycles to regulate cotranslational targeting of proteins to membranes. However, the regulatory mechanism of SRP and SR GTPases is a notable exception to the GTPase switch paradigm established for classical signaling GTPases. We have previously isolated different classes of mutant GTPases that block the binding and reciprocal activation between SRP and FtsY at specific stages (). Analyses of these mutants reveal a series of discrete conformational rearrangements that occur during the interaction between SRP and FtsY, culminating in the reciprocal activation of GTP hydrolysis in both proteins. Here, we have used these mutants to examine the role of these conformational changes in a complete, functional protein targeting reaction. summarizes the effect of the different classes of mutant GTPases on protein targeting and translocation. All of the mutant GTPases that compromise SRP–SR complex formation reduce the efficiency of protein targeting (, blue box). This result is consistent with the notion that formation of a stable SRP–SR complex is crucial for delivery of the cargo protein to the target membrane (, ). Contrary to expectations based on a previous paper (), we found that the class II (activation-defective) mutants (, red box) also block efficient protein translocation when introduced into either FtsY or Ffh. The translocation defect correlates well with the degree to which GTPase activation is inhibited by these mutants ( and ). In both GTPases, the class II mutations allow a stable SRP–SR complex to be assembled, but specifically inhibit the reciprocal activation of GTP hydrolysis in the complex. The fact that these mutants block the protein translocation reaction is intriguing in light of the previous results obtained using GMPPNP (). Yet these seemingly contradictory results are easily reconciled by the fact that GMPPNP and the mutant GTPases inhibit the SRP–SR interaction cycle at different stages. GMPPNP is a good mimic of GTP that allows all or most of the conformational changes in the SRP–SR complex to occur, but blocks the chemical step of GTP hydrolysis caused by substitution of the β−γ phosphate bridging oxygen with an imino group (). In contrast, the class II mutant GTPases block GTP hydrolysis at an earlier stage by inhibiting the conformational rearrangements that lead to activation of the two GTPase sites (, red box). Thus, although GTP hydrolysis is not, per se, required, the conformational changes upon the SRP–SR interaction that lead to GTPase activation play a crucial role in the protein targeting reaction. What structural changes in SRP and SR are responsible for mediating both GTPase activation and efficient targeting of the nascent protein? Analyses of the mutational effects suggest that movement of the IBD loop is the most crucial feature (; this study). In both GTPases, mutations in this loop result in pronounced class II phenotypes and block protein translocation. The similar effects of these mutations in both GTPases are consistent with the symmetrical pattern of the interaction network formed between these loops and the GTP molecules bound at the respective active site (; ). In contrast, the other two FtsY mutations in this class, when introduced into homologous positions in Ffh, do not substantially block reciprocal GTPase activation or the protein targeting reaction, indicating that the interactions made by these residues are less conserved and not symmetrical between the two GTPases. Therefore, an impairment to properly rearrange the IBD loop stands out as the likely cause leading to the functional targeting defects observed here. How do the activation-defective GTPase mutants block the protein targeting reaction? We consider it most likely that these mutants block the cargo-unloading step, at which the RNC is released from SRP and transferred to the translocon embedded in the membrane (, ). Cargo unloading has to occur after a stable SRP–SR complex is formed, but before GTP hydrolysis is activated to drive complex disassembly. Other steps in the targeting reaction are less likely to be a target of these mutant GTPases: binding of SRP to the RNC (, ) should not to be affected by mutations in FtsY, which does not participate in cargo recognition. Formation of the SRP–SR complex, which mediates delivery of cargo to the membrane surface (, ), is also unlikely because this class of mutants has been shown to form stable SRP–FtsY complexes. Finally, hydrolysis of GTP to drive the dissociation and recycling of the SRP components (, ) is unlikely because our assay monitors a single round of protein targeting and any defect on this step would not be observed. A role of the SRP–SR interaction in facilitating cargo release is also suggested by the recent cryo-EM structures of the RNC–SRP and RNC–SRP–SR complexes (, ). Comparison of the two structures shows that, upon binding of SR to the RNC–SRP complex, the electron density of the GTPase domains of both SRP and SR is no longer visible, although the other domains of SRP and SR can be identified and remain close to the RNC. Thus the interaction with the SR induces structural rearrangements that change the way SRP is positioned at the ribosome exit tunnel. Collectively, both the biochemical and structural characterizations suggest that the concerted rearrangements that occur upon formation of the SRP–SR complex serve dual purposes. First, movement of the IBD loops into close proximity to the bound GTP activates GTP hydrolysis in the complex that sets the stage for subsequent disassembly and recycling of the SRP components. Second, these movements trigger (directly or indirectly) the switch of SRP from the cargo-binding mode to the cargo-release mode, and thus help drive the transfer of the nascent chain from SRP to the translocon. In this way, cargo transfer and GTPase activation are effectively coupled to each other to ensure the maximum efficiency by which cargo protein is delivered to the translocation channel on the target membrane. Further, by using the conformational change for GTPase activation to trigger cargo release, GTP hydrolysis could also be used by the SRP to improve the fidelity of the protein targeting reaction akin to kinetic proofreading mechanisms used by elongation factor GTPases (), although no concrete evidence in support of this notion is currently available. Most importantly, our results imply that bringing SRP and SR together in a complex, and thereby juxtaposing the RNC and the translocon at the membrane surface, is not sufficient to initiate transfer of the nascent chain from SRP to the translocon. Rather, for SRP and SR to exert their roles as molecular matchmakers, an active cargo-unloading step has to take place that requires an elaborate conformational rearrangement within the complexed GTPase modules of SRP and SR. WG translation extract was obtained from Promega. Microsomal membranes from dog pancreas were prepared by J. Miller (University of California, San Francisco, San Francisco, CA) according to published procedures () and were treated with high salt and partial trypsin digestion to generate TKRM as described previously (; ; ). The in vitro transcription plasmid for pPL was provided by E. Powers (University of California, Davis, Davis, CA). The expression and purification of Ffh and FtsY have been described previously (; ). Mutant Ffh and FtsY proteins were constructed using the QuickChange mutagenesis protocol (Stratagene). Mutant Ffh and FtsY were purified using the same procedures as those for wild-type proteins. [S]methionine and γ-[P]GTP were obtained from GE Healthcare. The cotranslational protein-targeting assay was described in detail in the text () and in a previous paper (). Posttranslational protein-targeting assay of pPL () was performed with slight modifications of the procedures used by . Stalled RNCs containing pPL were generated by in vitro translation using the WG translation extract. After completion of translation, nucleotides were removed from the RNC using a 1-ml Sephacryl S-200 gel filtration column (Sigma-Aldrich; ). Fractions in the void volume containing purified RNC were pooled and incubated with TKRM for 2 min at 25°C. 2 mM puromycin was added and the reaction mixture was incubated at 37°C for 15 min; this step releases the pPL nascent chain from the ribosome to allow for its translocation across the membrane and cleavage by the signal peptidase. The reaction was then analyzed by 15% SDS-PAGE. The translocation efficiency of each mutant GTPase was measured in parallel with that of the wild-type protein, and these comparative measurements were repeated three to five times. Most of the figures presented in this paper show a representative measurement performed in parallel for all the proteins. In general, the absolute translocation efficiency for each protein can vary up to 30% from day to day and depends on several factors such as the amount and purity of TKRM and the quality of [S]methionine. Nevertheless, these are systematic rather than random errors, and the translocation efficiency of the mutant relative to the wild-type protein, measured in side-by-side experiments, is highly reproducible and independent of the aforementioned factors, with deviations of <10%. GTP hydrolysis reactions were performed and analyzed as described previously (). The use of the GTPase assay to measure the basal GTPase activity of Ffh, the stimulated GTPase reaction between SRP and FtsY, and the affinity between a mutant GTPase and its binding partner have been described in detail previously (; ). Fig. S1 shows that the targeting and translocation of pPL occurs on a faster time scale than completion of protein synthesis. Fig. S2 shows that FtsY(47–497) is reduced by about half in translocation of pPL compared with full-length FtsY. Fig. S3 and Tables S1–S7 show additional data for repetitions of the experiments shown in Figs. 2–5. Online supplemental material is available at .

Cyclin-dependent kinases (Cdks) play a key role in mammalian cell cycle regulation by ensuring that cell cycle events proceed in a stepwise fashion and produce two identical daughter cells. Several cyclin/Cdk complexes have been implicated in cell cycle control (). In mammalian cells, entry into mitosis is governed by cyclin A/Cdk1 and cyclin B/Cdk1. Another Cdk activity, cyclin A/Cdk2, is required for both DNA synthesis and mitosis (). Cyclin A/Cdk2 activity is first evident in late G1, and it persists through S and G2 phase until prometaphase (; ; ). To date, two known isoforms of mammalian cyclin A (cyclin A1 and A2) are present. Although cyclin A1 is restricted to male germ cells, cyclin A2 (hereafter referred to as cyclin A) is widely expressed in both germ cells and somatic tissues (; ). Cyclin A–deficient mice die during early embryogenesis (), demonstrating that cyclin A is an essential gene. The requirements for cyclin A in both replication and mitosis have not been fully elucidated. A number of experimental approaches have suggested potential mechanisms whereby cyclin A/Cdk2 complexes regulate the G1/S transition. Overexpression of cyclin A accelerated entry of mammalian cells into S phase, indicating that this cyclin could be a limiting factor needed to trigger replication (). Conversely, microinjection of anti-cyclin A antibodies or production of antisense RNA in G1 phase cells prevented entry into S phase (). It has been proposed that cyclin A/Cdk2 is required for the initiation of replication, activation of preexisting replication complexes, and/or prevention of rereplication (; ; ). In support of these mechanisms, cyclin A has been found to associate with DNA replication foci in human cells, and it has been shown to bind to and phosphorylate proteins in the origin recognition complex in extracts (; ). Cyclin A has also been shown to associate with and phosphorylate Cdc6, provoking its nuclear export and degradation and thereby preventing rereplication (; ). The mechanisms by which cyclin A promotes entry into, and progression through, mitosis are also unclear. This is likely due, at least in part, to the functional overlap with another mitotic kinase, cyclin B/Cdk1. Nevertheless, it is clear that cyclin A plays an essential role in mitotic entry because microinjection of anti-cyclin A antibodies into G2 phase cells can prevent progression into M phase (). Microinjection of recombinant cyclin A/Cdk2 into human G2 (but not S phase) cells accelerated entry into mitosis, suggesting that this kinase may be a rate-limiting factor for the G2/M transition (). Interestingly, cyclin A/Cdk2 was also required to progress through mitosis until the middle of prophase because the inhibition of this kinase with a Cdk inhibitor caused early or mid-prophase cells to return to interphase (). Cyclin A is destroyed by the anaphase-promoting complex in pro-metaphase, and mutants lacking a destruction box arrest in anaphase (). In light of these experiments and others, it will be critical to determine the key substrates of this kinase, how they enable progression through replication and mitosis, and whether these targets are specific to cyclin A/Cdk complexes. In addition to its role in activating Cdks, the cyclin component is critical for directing kinase activity to particular compartments. Cyclins are directed to their substrates through signals that regulate both subcellular localization and targeting to a specific protein binding site (the so-called RXL or cyclin-binding motif) (; ; ). Cyclin A had been shown previously to be predominantly nuclear by immunofluorescence (). Yet it is clear that cyclin A/Cdk2 phosphorylates both nuclear and cytoplasmic targets, including those involved in centrosome duplication (). Interestingly, recent data suggest that the cyclin A/Cdk complex is not statically maintained in the nucleus. Rather, it shuttles between nucleus and cytoplasm, consistent with its ability to modify targets in both compartments (). These and other studies also suggested that Cdk2 was required in vitro and in vivo for nuclear import of cyclin A (). However, the requirements for cyclin A nuclear export have not been defined. In an effort to understand in greater detail how cyclin A is regulated, we have performed a biochemical screen for proteins that interact with cyclin A in human cells. Our efforts have led to the identification of SCAPER, a novel protein that specifically interacts with cyclin A/Cdk2 in vivo through a cyclin-binding motif. Although a small portion is associated with the nucleus, SCAPER localizes primarily to the endoplasmic reticulum (ER), and its expression is relatively constant throughout the cell cycle. It associates with cyclin A at multiple stages of the cell cycle. Ectopic expression of SCAPER sequesters cyclin A from the nucleus and delays cell cycle progression in M phase. Furthermore, ablation of SCAPER by RNAi decreases the pool of cyclin A in the cytoplasm (manifested as a membrane-bound complex), resulting in delayed progression into S phase from quiescence (G0) in response to mitogens. Our data suggest that the isolation of SCAPER may allow us to dissect the S and M phase functions of the cyclin A/Cdk2 kinase. Further, SCAPER binding to cyclin A may represent a mechanism for retaining cyclin A transiently in the cytoplasm and directing the kinase to specific substrates that must be phosphorylated in that compartment before S phase entry. To search for cyclin A/Cdk2-interacting proteins, we took advantage of the observation that cyclins often exhibit stable binding to substrates and regulators through dedicated docking sites and that the resulting associated proteins can be subsequently phosphorylated in vitro. We biochemically isolated proteins that associate with cyclin A in vivo using anti-cyclin A immunoaffinity purification. Purified proteins were subjected to in vitro kinase assays and separated by SDS-PAGE, revealing a series of phosphorylated species (). Given that the Kip/Cip family of kinase inhibitors, p21 and p27, and the pRB family proteins, p107 and p130, have been shown to associate with cyclin A in vivo, we used a combination of immunodepletion, Western blotting, and kinase assay to verify that they were indeed present in this collection of polypeptides ( and unpublished data). These results substantiate our approach and suggest that physiological cyclin A/Cdk2-interacting proteins can be identified using this strategy. In addition to the expected complement of proteins described above, we detected a major phosphorylated polypeptide of 158/160 kD (p158/p160) that did not correspond to known cyclin A–interacting proteins. To further validate the interaction between cyclin A and p158/p160, we fractionated whole cell extracts using Q- and SP-Sepharose ion exchange columns and then performed immunopurification with anti-cyclin A antibody and kinase assays. p107 and p130 appeared in multiple fractions (), attesting to their heterogeneous representation in multiple, distinct complexes (). In contrast, p158/p160 was detected in a single fraction on either ion exchange column, suggesting the existence of stable, specific cyclin A/Cdk2 complexes. To further test the notion that p158/p160 could be isolated by an independent method that also relied on affinity for cyclin A/Cdk2, we chromatographed human cell extracts over a matrix containing recombinant GST-cyclin A/Cdk2 and performed kinase assays. GST-cyclin A/Cdk2, but not unfused GST protein, interacted with a polypeptide indistinguishable in mobility from p158/p160 (). GST-cyclin E/Cdk2 also bound p158/p160, albeit to a lesser extent, in this in vitro experiment. To identify p158/p160, we performed a large-scale purification of the protein from human cell extracts using resin containing recombinant GST-cyclin A/Cdk2 protein. The 158/160-kD band was subjected to mass spectrometric sequencing. This analysis revealed that a large number of peptides belonged to two uncharacterized proteins with similar molecular masses. One of these polypeptides has a theoretical molecular weight of 158 kD and encoded a putative zinc finger motif (sequenced previously as a cDNA designated ZNF291, GenBank/EMBL/DDBJ accession no. ; ). We termed this polypeptide SCAPER (S phase cyclin A–associated protein residing in the endoplasmic reticulum) in light of our findings described below. SCAPER shows no considerable homology with known human proteins. A motif search identified a putative C2H2-type zinc finger motif, a putative transmembrane domain, an ER retrieval signal at the C terminus, four coiled-coil domains, six potential RXL motifs that might confer binding to cyclins, and six consensus Cdk phosphorylation sites (). We note the existence of mouse, rat, bovine, and chimpanzee SCAPER orthologues. We amplified the coding region, inserted the resulting product into an expression vector with a Flag tag, and expressed the protein in human 293T cells. Flag-SCAPER was purified and could be phosphorylated in vitro with recombinant cyclin A/Cdk2. The resulting phosphoprotein co-migrated with endogenous SCAPER (), suggesting that we had amplified the full-length protein. We probed Northern blots derived from human tissue and found that SCAPER is ubiquitously expressed in each of the human tissues we examined, with highest transcript levels in testis (). Next, we determined whether SCAPER was expressed in a cell cycle–dependent manner. We synchronized human T98G cells by means of serum deprivation and restimulation and prepared total RNA from cells at each stage. Semi-quantitative RT-PCR revealed that SCAPER was expressed at each stage of the cell cycle, although the gene was expressed at somewhat higher levels in late G1 and S phase (). It is worth noting that SCAPER is expressed in G0 cells. Northern blot analysis of total RNA from synchronized T98G cells revealed a transcript of the expected size and confirmed the results we obtained with RT-PCR (unpublished data). To examine SCAPER protein levels, we produced and affinity purified a polyclonal antibody against an internal fragment of SCAPER located close to the N terminus. SCAPER was the predominant polypeptide visualized by Western blotting of crude cell lysates (), and it was expressed in every cell line we have tested, including immortal (T98G, U2OS, HeLa, 293T, Saos2) and normal diploid (IMR90, WI38) human cell lines, and mouse myoblast C2C12 cells (unpublished data). We examined SCAPER protein levels as cells progress from G0 to S phase and in cells progressing from the G1/S transition. Consistent with our analysis of RNA, SCAPER is expressed at relatively constant levels throughout the cell cycle, including G0, although it was slightly elevated in early S phase (). Our initial identification of SCAPER was based on the observation that it stably interacts with cyclin A/Cdk2. Given that Cdk substrates are generally capable of associating with multiple cyclin/Cdk complexes, we asked whether SCAPER also interacts with other cyclins or Cdks in vivo. Endogenous SCAPER could be immunoprecipitated from cell extracts with anti-cyclin A or anti-Cdk2 antibody, but not with anti-cyclin E or anti-Cdk1 antibody (). This suggests that SCAPER specifically interacts with cyclin A/Cdk2, but not with cyclin A/Cdk1, cyclin B/Cdk1, or cyclin E/Cdk2, in vivo within the detection limits of our Western blot analysis. This in vivo interaction was also detected using normal human diploid fibroblasts (IMR90) and mouse myoblasts (unpublished data). Similar results were obtained when we transiently transfected 293 cells with a Flag-SCAPER expression vector, immunoprecipitated the protein with anti-Flag antibodies, and performed Western blotting with antibodies against different cyclins and Cdks. We observed coimmunoprecipitation of Flag-SCAPER with cyclin A and Cdk2, but not with cyclin E or Cdk1 (, lane 9). Next, we investigated the interaction between SCAPER and cyclin A at different cell cycle stages in T98G cells rendered quiescent and restimulated to enter the cell cycle. Like SCAPER, Cdk2 levels were relatively constant throughout the cell cycle (). In contrast, cyclin A expression peaks in S/G2 phase, after which it decreases (), as expected. Interestingly, cyclin A associates with SCAPER whenever it is expressed (). Collectively, our results demonstrate that SCAPER specifically interacts with cyclin A/Cdk2 at multiple phases of the cell cycle, including S and G2/M. We investigated whether SCAPER/cyclin A/Cdk2 complexes also contained members of the Kip/Cip family of Cdk inhibitory proteins and/or F-box family proteins. Strikingly, we found that immunoprecipitates of p21 and p27 contained cyclin A and Cdk2, as expected, but not SCAPER (). Thus, the SCAPER/cyclin A/Cdk2 complex is separable from the pools of cyclin A/Cdk2/p21 and cyclin A/Cdk2/p27, and this finding is consistent with our immunodepletion experiments (unpublished data). Furthermore, we note that immunoprecipitates of another cyclin A–interacting protein, Skp2, do not contain SCAPER (unpublished data). To identify the region in SCAPER that confers binding to cyclin A/Cdk2, N-terminal (residues 1–717; SCAPER-N) and C-terminal (residues 717–1399; SCAPER-C) fragments of SCAPER were fused to a Flag tag and expressed in both bacteria and mammalian cells (). Subsequent immunoprecipitation of transfected human 293T cells revealed that cyclin A and Cdk2 strongly associated with Flag-SCAPER-N, but only weak interactions were evident with Flag-SCAPER-C (). Identical results were obtained when bacterially expressed Flag-SCAPER-N and Flag-SCAPER-C were incubated with recombinant GST-cyclin A/Cdk2 (). Thus, the N-terminal region of SCAPER is responsible for cyclin A/Cdk2 binding, and the interaction between SCAPER and this kinase is direct. Further, kinase assays revealed that cyclin A/Cdk2 phosphorylates Flag-SCAPER-N but not Flag-SCAPER-C (), suggesting that cyclin A/Cdk2 can both bind and phosphorylate the N-terminal half of SCAPER. We note that the addition of excessive amounts of exogenous SCAPER did not substantially inhibit cyclin A/Cdk2 kinase activity, suggesting that this fragment of SCAPER is not a Cdk inhibitor ( and unpublished data). Previous work revealed that RXL motifs in Cdk-interacting proteins are responsible for stable interactions with cyclin/Cdks (; ; ), and we examined the requirement for this motif in SCAPER. The N-terminal portion of SCAPER confers cyclin A/Cdk2 binding (), and given that there are three cyclin-binding motifs in SCAPER-N, we mutated each of them in the context of full-length SCAPER to generate mutants mA, mB, and mC (). Each mutant was expressed in 293T cells. Immunoprecipitation and Western blotting clearly show that mutants mA and mC retain the ability to bind to cyclin A/Cdk2 to the same extent as wild-type SCAPER (). In contrast, mB is unable to associate with cyclin A/Cdk2 (), indicating that this region constitutes a bona fide cyclin-binding motif. To biochemically examine the subcellular localization of SCAPER, we fractionated cell extracts into nuclear, cytosolic, and membrane compartments and verified enrichment of each fraction using several well-established markers (). Cyclin A is found predominantly in the nucleus, whereas Cdk2 is found in both the nucleus and the cytosol, in agreement with earlier reports (; ; ). Interestingly, SCAPER is most enriched in the membrane fraction, similar to an ER membrane protein calnexin, although a portion was also detectable in the nuclear fraction (). This is in contrast to a Golgi resident protein giantin, which is located exclusively in the membrane fraction. Further, cyclin A resident in these membrane fractions was associated with SCAPER (). We estimate, based on our subcellular fractionation and RNAi experiments (see ), that ∼10–20% of the total cellular cyclin A is bound to SCAPER. Given that the ER is contiguous with the nuclear envelope and that most rough ER membrane proteins are present in both the rough ER and the nuclear envelope (), these data could suggest that SCAPER is a component of the ER. Next, we used indirect immunofluorescence to further investigate the subcellular localization of SCAPER. Endogenous SCAPER was associated mostly with perinuclear and reticular structures (), with a small but detectable portion localized to the nucleus. Use of a second affinity-purified anti-SCAPER antibody directed against a different epitope gave similar results (unpublished data). We examined the localization of recombinant SCAPER by transfecting Flag-tagged SCAPER into U2OS cells, and indistinguishable results were obtained (). Comparable findings were obtained using multiple human cell lines. To more precisely localize SCAPER within the perinuclear region, we performed colocalization experiments with GFP-tagged mannosidase II and GFP-tagged sec61β in the absence or presence of nocodazole. Mannosidase II is a luminal protein found in the medial compartment of the Golgi. Its staining did not substantially overlap with that of SCAPER (). On the other hand, sec61β, a subunit of the heteromeric sec61 translocation complex found in the membrane of the ER, exhibited substantial overlap with SCAPER (), indicating that SCAPER may localize to the ER. To test this idea further, we treated cells with a low dose of nocodazole, which depolymerizes microtubules and causes the Golgi to break down into small fragments containing mannosidase II that are dispersed throughout the cytoplasm (). In contrast, ER proteins such as sec61β are not affected by nocodazole, and likewise, the distribution of SCAPER was not substantially altered by such treatment (). Use of a different drug, brefeldin A, which induces relocalization of Golgi proteins to the ER, has no effect on SCAPER localization (unpublished data). Collectively, subcellular fractionation and the immunofluorescence experiments confirm that SCAPER is an integral component of the ER. To study the effect of SCAPER expression on endogenous cyclin A, we ectopically expressed Flag-SCAPER in human cells. Immunofluorescence studies indicated that cyclin A localizes primarily to the nucleus, as expected (). However, when Flag-SCAPER was ectopically expressed, endogenous cyclin A was diffusely distributed throughout the cell. A substantial fraction of cyclin A was sequestered outside of the nucleus in at least 70% of Flag-SCAPER–expressing cells (). During mitosis, the pattern of cyclin A in Flag-SCAPER–expressing cells looked indistinguishable from control cells (unpublished data). Interestingly, the Flag-SCAPER-N, mA, and mC SCAPER mutants, each of which retains an intact RXL motif (site B), potently sequestered cyclin A in the cytoplasm ( and unpublished data). In sharp contrast, Flag-SCAPER-C and mutant mB, which lacks this site, were unable to cause cyclin A relocation ( and unpublished data). Similar results were obtained with other human cell lines. These data strongly suggest that the change in cyclin A localization was mediated by SCAPER binding to cyclin A in the cytoplasm. Thus, the expression of SCAPER can dominantly interfere with cyclin A localization. If SCAPER specifically associates with cyclin A/Cdk2, and cyclin A/Cdk2 is required at distinct stages of the cell cycle, ectopic expression of SCAPER might be expected to affect cell cycle distribution. We transiently expressed Flag-tagged SCAPER and several mutants in human cells and examined their cell cycle distribution at different intervals after transfection using flow cytometry. SCAPER expression initially caused a modest increase in the S phase population that eventually led to a prominent G2/M accumulation 72 h after transfection (). To pinpoint the stage at which the arrest occurred, we compared the mitotic index of transfected versus untransfected cells by visualizing cells that exhibit phosphorylation of serine 10 of histone H3, a well-established marker of mitosis (). Microscopic examination revealed a fourfold enrichment of mitotic cells expressing Flag-SCAPER (13.3 ± 3.6%) as compared with control mitotic cells (2.8 ± 0.7%) (). DAPI staining revealed that these cells had condensed chromosomes. To more precisely determine whether this enrichment in mitotic cells stemmed from a cell cycle block or a delay, we blocked transfected cells at the G1/S phase border with hydroxyurea (HU). After treatment with HU, the mitotic index of control (0.5 ± 0.1%) and SCAPER-expressing cells (0.7 ± 0.1%) did not differ substantially (), suggesting that SCAPER overexpression results in an M phase delay rather than an arrest. Furthermore, the proportion of prophase cells in SCAPER-expressing mitotic cells was slightly elevated compared with that in control mitotic cells, suggesting the delay may occur at this stage in mitosis (unpublished data). Studies using deletion mutants showed that Flag-SCAPER-N, but not Flag-SCAPER-C, produced a similar cell cycle phenotype (), and a striking accumulation was also evident when cells were transfected with mutants mA and mC (). In contrast, expression of mutant mB, which lacks the ability to bind cyclin A/Cdk2, had only a modest effect on cell cycle progression (). Collectively, our data suggest that expression of SCAPER can affect cell cycle progression specifically in M phase by binding and sequestering cyclin A/Cdk2. We investigated the effects of suppressing SCAPER production by RNAi-mediated down-regulation of endogenous SCAPER. siRNAs corresponding to four distinct SCAPER coding sequences were chosen to target SCAPER in human cells. Western blot analysis showed that treatment of cells with SCAPER siRNA resulted in a substantial reduction (∼75%) in SCAPER expression (). Fluorescence-activated cell sorting (FACS) analysis of transfected cells indicated that cell cycle progression was not affected in cycling cells, and SCAPER siRNA-treated cells released from an HU block progressed through S phase and G2/M phase with kinetics similar to control cells (unpublished data). Microscopic inspection of transfected cells also did not reveal obvious abnormalities. In particular, immunofluorescence studies failed to detect any changes in the levels or in the subcellular localization of cyclin A, Cdk2, γ-tubulin, or calnexin. DNA staining also appeared similar in control and knock-down samples. SCAPER expression is not cell cycle dependent (), and depletion of SCAPER had no obvious impact on cycling cells. However, the presence of abundant SCAPER in G0 cells () prompted us to investigate the effect of SCAPER depletion on the ability of cells to exit the cell cycle or to reenter the cell cycle from quiescence (). First, there was no apparent impact of SCAPER depletion on entry into quiescence (; row 1, columns 1 and 2), suggesting that ablation of SCAPER does not affect cell cycle exit in response to mitogen deprivation. Next, SCAPER-depleted cells brought to quiescence through serum deprivation were restimulated. As expected, a substantial number of cells transfected with control siRNA entered S phase ∼18 h after serum stimulation (40%, ; row 4, column 1), and by 21 h, 99% of cells were in S phase (; row 5, column 1). 60% of cells progressed into G2/M phase 3 h later (; row 6, column 1). In striking contrast, the majority of SCAPER siRNA-treated cells were still in G1 18 h after stimulation (81%, ; row 4, column 2). By 21 h after stimulation, a substantial proportion of these cells began to enter the S phase (57%) (; row 5, column 2). Only 2% of cells had progressed into G2/M phase after 24 h (; row 6, column 2). Thus, there appears to be a considerable delay (∼3 h in duration) in S phase entry associated with SCAPER depletion. Consistent with our FACS analysis, we observed a comparable delay in BrdU incorporation associated with SCAPER depletion in siRNA-treated cells (). Importantly, this cell cycle delay phenotype was also observed with three additional, distinct siRNAs targeting SCAPER (Fig. S1 A, available at ). Furthermore, similar findings were observed in normal human cell lines such as IMR90 and WI38 (unpublished data), and in mouse 3T3 cells using a mouse-specific SCAPER siRNA (Fig. S1 B). Collectively, we conclude that SCAPER plays a critical role in promoting progression from G1 to S phase after cell cycle reentry from a quiescent state. We attempted to pinpoint more precisely when the cell cycle delay takes place. We determined whether escape from the quiescence state was impaired in cells depleted of SCAPER by counting cells that expressed the Ki67 proliferation marker, apparent only in nonquiescent cells. Interestingly, both control and SCAPER siRNA-transfected cells were able to exit G0 with similar kinetics (). Furthermore, when G0 cells were released and synchronized at the G1/S boundary with HU, SCAPER depleted cells progressed into the S phase more slowly than control cells after release from the HU block (). Collectively, we believe that the S phase delay occurs in late G1/S and not earlier in the cell cycle (G0 or early G1 phase). We reported that overexpression of SCAPER results in the relocalization of cyclin A from the nucleus to the cytoplasm (). We sought to determine whether the converse is true, namely, whether cyclin A would relocalize from the cytoplasm when SCAPER function is suppressed. We and others have detected a portion of cyclin A residing within the cytoplasm and the microsomes (; ; ). This cyclin A partitioning is difficult to detect by immunofluorescence or live-cell imaging. Not surprisingly, when U2OS cells were transfected with either control or SCAPER siRNA, no difference in cyclin A staining in the cytoplasm was observed (, columns 1 and 2). To overcome this limitation, cells were first transfected with Flag-SCAPER to sequester cyclin A in the cytoplasm. They were subsequently treated with either control (, column 3) or SCAPER siRNA (, column 4) to gauge the effects of knock-down. Upon silencing SCAPER, we observed recycling of cyclin A from the cytoplasm back to the nucleus (, column 4). Thus, SCAPER can influence the balance between cytoplasmic and nuclear pools of cyclin A, and it may represent a novel cyclin A/CDK2 regulatory protein that transiently maintains this kinase outside of the nucleus. As a further confirmation that RNAi-mediated depletion of SCAPER alters cyclin A localization, we performed a series of subcellular fractionation experiments to examine the dynamic equilibrium of cyclin A in distinct cellular compartments, at a time when the cell cycle delay phenotype was first evident at late G1/S (15 and 18 h after serum stimulation as in , rows 3 and 4). For control siRNA-treated cells, we found that 17 and 26% of total cellular cyclin A was associated with the membrane compartment 15 and 18 h after stimulation, respectively (). Remarkably, the membrane-bound cytoplasmic pool of cyclin A was dramatically reduced when SCAPER was depleted. Thus, after SCAPER suppression, we observed only 2% of total cellular cyclin A associated with the membrane fraction 15 h after stimulation, and at 18 h after release, the percentage of cyclin A residing in the membrane increased to only 12% (). Consistent with decreased cyclin A retention at membranes, we observed a comparable reduction in membrane-associated Cdk2 in SCAPER-depleted cells at 15 and 18 h after stimulation (). As a further test of the notion that partitioning of cyclin A within the cytoplasm by SCAPER occurred at a critical time in late G1/S phase when cyclin A levels were limiting, we compared the proportions of cyclin A in different compartments at a time when SCAPER is not thought to play a role, namely, after S phase entry (24 h, ), and we did not observe any substantial difference between the two populations. In addition, consistent with cyclin A partitioning in SCAPER-depleted cells, no apparent difference in Cdk2 localization was observed in cells that had entered S phase (24 h, ). In conclusion, our studies strongly support the notion that ablation of SCAPER decreases the pool of cyclin A in the cytoplasm and results in a delayed S phase reentry from quiescence. We have therefore for the first time unraveled the physiological significance of SCAPER and its role in maintaining cyclin A homeostasis in the cytoplasm. One enduring, fundamental issue in the field of cell cycle control pertains to our incomplete knowledge of Cdk substrates (). We isolated and purified a novel protein, SCAPER, which specifically interacts with cyclin A/Cdk2 both in vitro and in vivo. SCAPER is expressed in quiescent cells and its levels remain relatively unchanged throughout the cell cycle, suggesting that it may have a housekeeping function. It localizes primarily to the ER and interacts with cyclin A at multiple phases of the cell cycle. Progression through two key cell cycle transitions, G1/S and M phase, is differentially affected by SCAPER. Ablation of SCAPER by RNAi leads to delay of cell cycle reentry into S phase, but overexpression provokes cell cycle delay in M phase. These data lead us to propose an intriguing hypothesis: cytoplasmic cyclin A may be a limiting factor for the G1/S transition upon cell cycle reentry from quiescence; that is, cyclin A could be tethered to the ER, retaining it in the cytoplasm, where it may be required to phosphorylate a key substrate before S phase reentry. On the other hand, nuclear cyclin A may play a more prominent role in progression through the G2/M phase transition. The presence of cyclin A/Cdk2 complexes in the cytoplasm is in accordance with the notion that cyclin A/Cdk2 phosphorylates substrates in that compartment as well as the nucleus. A requirement for cytoplasmic pools of cyclin A/Cdk2 is substantiated by recent studies implicating this kinase in mammalian centrosome duplication () and by other experiments that have identified Cdk2 targets associated with this organelle (; ; ). Recent studies have shown that the nuclear import of cyclin A is much more rapid than its export (). In light of these observations, we believe that our identification of SCAPER has several important implications for the partitioning and regulation of cyclin A/Cdk2 activity and its targets (). First, because nuclear import of cyclin A/Cdk2 is kinetically favored over its export, it may be necessary to sequester the small quantities of this kinase that enter the cytoplasm at any given time. SCAPER, or an analogous cyclin A/Cdk2 binding protein, may perform such a function by partitioning cyclin A in the cytoplasm. In this regard, SCAPER may play a more prominent role during the G1/S transition in cells reentering the cell cycle from quiescence, whereas an analogous cyclin A/Cdk2 binding protein could substitute for SCAPER function at other stages of the cell cycle. Furthermore, we note that SCAPER is present in quiescent and early G1 phase cells. It could therefore play a second, “buffering” role, to scavenge small quantities of cyclin A present in early G1 cells that could trigger premature initiation of DNA replication. In contrast, cyclin A levels rise dramatically during S phase and could thereby exceed the amount of SCAPER, allowing nuclear localization of cyclin A/Cdk2. SCAPER could thus provide a threshold function, delicately balancing the amount of cyclin A/Cdk2 partitioned between the two cellular compartments (). Although we do not yet know the precise mechanism by which SCAPER overexpression affects M phase, it does not appear to result from the direct attenuation of cyclin A/Cdk2 activity because excessive amounts of recombinant SCAPER do not substantially inhibit Cdk2 activity ( and unpublished data). It is more likely that ectopic expression of SCAPER induces a cell cycle delay by sequestering the bulk of cyclin A/Cdk2 away from the pool of cyclin A complexes that normally interact with the Kip/Cip family of inhibitors (p21, p27) and the pRB family (p107 and p130) in the nucleus, thereby preventing the phosphorylation of critical substrates necessary for progression through M phase. Indeed, SCAPER represents a unique tool for dissecting the function of cyclin/Cdk complexes. It appears to be the first protein shown to preferentially interact with cyclin A/Cdk2 to the apparent exclusion of all other cyclin/Cdk complexes. Recent evidence has implicated a role for cytoplasmic cyclin A in oncogenesis. Cytoplasmic accumulation of cyclin A has been reported in tumor cells (,). In addition, targeting of cyclin A to the ER in normal rat fibroblasts leads to oncogenic activation and results in polyploidy, abnormal centrosome duplication, and genomic instability (; ). These studies raise the intriguing possibility that an increase in cytoplasmic cyclin A levels is the direct cause of cellular transformation. As SCAPER possesses the ability to bind cyclin A in the cytoplasm, experiments are currently underway to determine precisely whether SCAPER up-regulation, which could lead to an increase of cytoplasmic cyclin A, is a common hallmark in cancer cells. If so, suppression of SCAPER function in tumor cells could be a major potential therapeutic application for SCAPER. HeLa, 293T, T98G, U2OS, IMR90, WI38, Saos2, and mouse C2C12 and 3T3 cells were grown in DME supplemented with 10% FBS at 37°C in a humidified 5% CO atmosphere. Recombinant GST protein was expressed in bacteria transformed with pGEX2TK. GST-cyclin A/Cdk2 and GST-cyclin E/Cdk2 were produced in Hi5 insect cells by co-infection with viruses carrying GST-cyclin A and Cdk2 or GST-cyclin E and Cdk2. Fusion proteins were purified by affinity chromatography with glutathione agarose (Sigma-Aldrich) as described previously (). In brief, cells were lysed in 0.1 HEMGN buffer (25 mM Hepes, pH 7.6, 100 mM KCl, 0.2 mM EDTA, pH 8.0, 12.5 mM MgCl, 0.1% NP-40, and 1 mM DTT with a cocktail of protease and phosphatase inhibitors). Cell extracts were incubated with glutathione agarose for 1 h at 4°C. The beads were collected by centrifugation, washed 4 times with 0.1 HEMGN buffer, and eluted with 20 mM glutathione in buffer containing 100 mM Tris (pH 7.9) and 120 mM NaCl. Flag-SCAPER-N and Flag-SCAPER-C were expressed in strain BL21 transformed with RSET vectors containing Flag-SCAPER-N and Flag-SCAPER-C, and Flag-SCAPER was expressed in human 293 cells transfected with CBF-SCAPER, and purified with anti-Flag agarose (M2, Sigma-Aldrich). The relative protein concentrations were estimated by silver or Coomassie staining. Human 293 cells were lysed in ELB buffer (50 mM Hepes, pH 7.0, 250 mM NaCl, 0.5 mM EDTA, pH 8.0, 0.1% NP-40, and 1 mM DTT with a cocktail of protease and phosphatase inhibitors) for 30 min on ice and then centrifuged at 20,000 for 30 min. The supernatant (1 g of total protein) was precleared three times by incubation with glutathione agarose loaded with ∼50 μg of GST for 2 h at 4°C. The precleared supernatant was incubated with glutathione agarose loaded with ∼40 μg GST-cyclin A/Cdk2 for 4 h at 4°C. The beads were collected by centrifugation and washed four times with ELB buffer. Finally, the beads were boiled in SDS sample loading buffer and analyzed by 10% SDS-PAGE. The gel was stained with silver. Immunodepletion and immunoprecipitation coupled to Western blotting with antibodies against p21, p27, p107, and p130 (Santa Cruz Biotechnology, Inc.) were used to identify each of the proteins in the eluate. The 158/160-kD band was excised from the gel, subjected to tryptic digestion, and the resulting peptides were separated by reverse-phase HPLC and analyzed with an LCQ quadrupole ion trap mass spectrometer (Finnigan) in the Harvard Microchemistry Facility. SCAPER cDNA was generated using SuperScript II H Reverse Transcriptase (Life Technologies) according to the manufacturer's protocol. The primers for the reverse transcription reactions were AAC AAG GGT ACT CAA ATA CAG and AGC TCT TTC CCG GGC TGC ATC. The entire coding region was amplified by two PCR steps with PfuTurbo DNA polymerase (Stratagene) using the two fragments of the first strand of SCAPER cDNA as templates. The product was sub-cloned into an expression vector (CBF; M. Cole, Princeton University, Princeton, NJ) in frame with the N-terminal Flag tag to produce CBF-SCAPER. Synthetic siRNA oligonucleotides were obtained from Dharmacon. Transfection of siRNAs using Oligofectamine (Invitrogen) was performed according to the manufacturer's instructions. The 21-nucleotide siRNA sequence for the nonspecific control was 5′-AATTCTCCGAACGTGTCACGT-3′. The 21-nucleotide siRNA sequence for SCAPER-1 was 5′-GAATAAACGTCATGATGTTTT-3′. Total RNA was prepared using Trizol Reagent (Life Technologies) according to the manufacturer's protocol. Semi-quantitative RT-PCR was performed using RT-PCR Superscript One Step kit (Invitrogen) as described by the manufacturer. Linear amplification was ensured in all cases. Primers for analysis of SCAPER expression were: CAG TGA TTT TTC TGC CAG CA and AGC TCT TTC CCG GGC TGC ATC; primers for analysis of β-actin levels were: ATC CTC ACC CTG AAG TAC CCC A and CTC GGC CGT GGT GGT GAA GCT GTA GCC GCG CT. T98G cells were rendered quiescent by serum deprivation for 72 h and stimulated with 20% FBS to allow cell cycle reentry. Cell cycle synchronization of T98G and U2OS cells was also performed by drug treatment. G1, G1/S, and G2/M cells were obtained by treating the cells for 24 h with 0.4 mM mimosine, 2 mM hydroxyurea, and 40 ng/ml nocodazole, respectively. S phase cells were obtained by releasing cells from hydroxyurea treatment into fresh medium for 4–6 h. Cell cycle distribution was monitored by propidium iodide staining and FACS analysis was performed as reported previously (). Transient transfections were performed using calcium phosphate (). For cell cycle analysis, the indicated plasmids (20 μg) were cotransfected with pCMV-CD20 (2 μg) into human cells and processed as described previously (). Cells were harvested with phosphate-buffered saline (PBS) containing 0.1% EDTA and stained with anti-CD20-FITC (Sigma-Aldrich) before FACS analysis. To generate anti-SCAPER antibodies, a GST fusion protein encoding residues 382–493 was produced in bacteria and used to immunize rabbits after coupling to keyhole limpet hemocyanin. Polyclonal antibodies against SCAPER were purified using a GST-SCAPER fusion protein affinity column after the serum was precleared with a GST protein affinity column. Cultured cells were washed once with PBS and lysed in ELB buffer. In a typical immunoprecipitation, whole-cell extract was incubated with antibodies for 2 h, and beads were washed four times with ELB buffer. Proteins bound to beads were eluted by boiling in sample buffer, separated by SDS-PAGE, transferred to a nitrocellulose membrane, and subjected to Western blot analysis. All kinase reactions were performed at 30°C in buffer containing 25 mM Hepes, pH 7.6, 150 mM NaCl, 0.5 mM EDTA, pH 8.0, 10 mM MgCl, 1 μM ATP, 1 mM DTT and 5 μCi 32γ-ATP (NEN Life Science Products). For immunoblotting, antibodies against cyclin A, cyclin B, cyclin E, Cdk2, Cdk1, Sp1 (each from Santa Cruz Biotechnology, Inc.), calnexin (BD Biosciences), giantin (Covance Research Products, Inc.), actin, β-tubulin, and Flag (each from Sigma-Aldrich) were used. Indirect immunofluorescence was performed as described previously (). In brief, cells were grown on glass coverslips, fixed with 4% paraformaldehyde and 0.1% glutaraldehyde in PBS for 10 min followed by incubation in −20°C methanol for 2–3 min. The cells were permeabilized with 1% Triton X-100/PBS for 10 min. Slides were blocked with 3% BSA in 0.1% Triton X-100/PBS before incubation with primary antibodies. Antibodies against SCAPER, Flag, cyclin A, p-ser10-H3 (Upstate Biotechnology), BrdU (Sigma-Aldrich), and Ki67 (Zymed Laboratories) were used. Secondary antibodies used were Cy3- or FITC-conjugated donkey anti–mouse or anti–rabbit IgG (Jackson ImmunoResearch Laboratories). Cells were then stained with DAPI and slides were mounted, observed, and photographed using a Nikon Eclipse E800 microscope (63× or 100×, NA 1.4) equipped with a Photometrics Coolsnap HQ CCD camera. Images were acquired with MetaMorph7 (Molecular Devices). Cells were incubated in hypotonic lysis buffer (10 mM Tris, pH 8.0, 10 mM NaCl, 3 mM MgCl, and 1 mM EGTA with a cocktail of protease and phosphatase inhibitors) for 10 min, followed by 20 strokes of a Dounce B homogenizer. The lysates were spun down for 5 min at 500 , and the supernatant was designated the cytosolic fraction. The nuclear pellet was washed three times in wash buffer (hypotonic lysis buffer with 0.1% NP-40) and then treated with nuclear lysis buffer (20 mM Hepes, pH 8.0, 25% glycerol, 0.42 M NaCl, 1.5 mM MgCl, 0.2 mM EDTA, and 0.5 mM DTT with a cocktail of protease and phosphatase inhibitors) for 30 min. The sample was spun at 14,000 for 30 min, and the supernatant designated the nuclear fraction. The cytosolic fraction was spun at 100,000 for 2 h, and the pellet (which was designated as the membrane fraction) was resuspended in ELB buffer with 1% NP-40. To verify the integrity of each fraction, Western blots were probed with nuclear (Sp1), cytosolic (β-tubulin), and membrane (calnexin and giantin) markers. For quantitation, 50 μg of protein from nuclear, cytosolic, or membrane fraction were resolved by SDS-PAGE and blotted with the antibodies indicated. The films were scanned with an Epson Perfection 4990 scanner and band intensities were quantitated and analyzed using Bio-Rad Quantity One 1-D analysis software. The entire analysis was done within the linear range of detection. Fig. S1 shows ablation of SCAPER with siRNA results in a delayed G1/S transition upon cell cycle reentry from G0. Online supplemental material is available at .

The intestinal epithelium is constantly and rapidly renewing throughout the lifespan of vertebrates, thereby representing a major target for tumorigenesis. This epithelium can be divided into two functionally distinct compartments. The crypt of Lieberkühn constitutes the proliferative compartment and contains stem/progenitor cells, as well as, in the small intestine, terminally differentiated Paneth cells. Multipotent stem cells, located near the bottom of crypts, generate new cells, which migrate upwards while differentiating into enterocytes, goblet, and enteroendocrine cells. Proliferation stops at the crypt–villus junction, and terminally differentiated cells are located on the neighboring villus, which constitute the differentiated compartment. In the small intestine, a fourth cell type, the Paneth cell, migrates downward and settles at the bottom of the crypts as postmitotic, differentiated cells. The balance among proliferation, differentiation, migration, and cell death must be tightly regulated to maintain homeostasis of this epithelium. We reported the expression of Sox9, an HMG-box transcription factor, specifically, in the rapidly proliferating stem/progenitor cells found at the bottom third of Lieberkühn crypts throughout the length of the intestine and in the Paneth cells of the small intestine, as well as in human tumors of the intestinal epithelium (). Sox9 was first identified as a key regulator of cartilage and male gonad development. Heterozygous mutations are responsible for the campomelic dysplasia syndrome, a skeletal dysmorphology syndrome characterized by skeletal malformation of endochondral bones and by male-to-female sex reversal in the majority of genotypically XY individuals (; ). Sox9 has also been implicated in the development of cranial neural crest derivatives (), in the neural stem cell switch from neurogenesis to gliogenesis () and in heart (), hair (), and pancreas () development. In each of these tissues, Sox9 expression is restricted to specific cell types, suggesting a complex transcriptional regulation. In addition, the currently identified Sox9 target genes, for instance, in the cartilage and in the gonad, display tissue-specific expression (; ), indicating that Sox9 may regulate distinct sets of genes in the different tissues in which it is expressed. In the intestinal epithelium, the function of Sox9 remains unresolved, although in vitro studies suggested a role in the control of cell differentiation (). In vitro and in vivo data indicate that is a transcriptional target of Wnt signaling. For instance, Sox9 expression is abrogated in Tcf4-null embryos, and it is strongly expressed in colorectal carcinoma cell lines containing activating mutations in components of the Wnt pathway (). The Wnt pathway plays a central role among the extracellular signals required to maintain the homeostasis of the intestinal epithelium. In particular, deletion of the gene encoding Tcf4, another HMG-box transcription factor (), or overexpression of the inhibitor Dickkopf (; ), resulted in a loss of the proliferative compartment and in impaired differentiation of secretory cell lineages. Conversely, mutation of the gene encoding Apc, a negative regulator of the pathway, resulted in crypt expansion, abrogation of cell migration, and amplification of the Paneth cell population (; ). In addition, deletion of the Wnt receptor Frizzled-5 revealed an essential role of the Wnt–Frizzled-5 pathway in the maturation of Paneth cells (). The sorting process of epithelial cells along the crypt–villus axis also depends on the Wnt pathway, via a modulation of Ephrin–Eph receptor interactions (). The Wnt signaling pathway can thus induce diverse cellular responses in the intestinal epithelium. In addition to these physiological functions, the Wnt pathway is centrally implicated in cancer, as mutations in components of this pathway have been identified in the majority of human colorectal carcinoma (). Such mutations mimic activation of the pathway by Wnt ligands (i.e., stabilization of β-catenin) and result in constitutive transcriptional activity of the β-catenin–Tcf4 complex and in aberrant expression of its target genes (). Despite the central importance of this pathway in the physiopathology of the intestinal epithelium, little is known about the molecular mechanisms involved in restricting this wide spectrum of potential functions to elicit a specific and adequate response from Wnt-stimulated cells. The fact that is transcriptionally regulated by the β-catenin–Tcf4 complex (), together with the particular expression of Sox9 in the compartment of the intestinal epithelium that contains Wnt-stimulated cells, suggests distinct functions in proliferating stem/progenitor cells and in the postmitotic Paneth cells (Fig. S1 A, available at ). To address the different aspects of Sox9 function during the turnover of the intestinal epithelium, including its possible role in specifying the cell response to Wnt signals, we specifically inactivated the corresponding gene in the intestinal epithelium. To analyze the function of Sox9 in the turnover of the adult intestinal epithelium, Villin-Cre (vil-Cre) mice, in which the Cre recombinase is expressed specifically in the intestinal epithelium from 10.5 d postcoitum onward () were crossed with Sox9 () mice, which have both alleles flanked by loxP sequences. This generated Sox9-vil-Cre mice, with an intestinal epithelium lacking Sox9 protein, indicating effective vil-Cre–mediated recombination of the Sox9 allele (). The control vil-Cre mice had no detectable phenotypic defect. Sox9-vil-Cre mice developed as their control littermates (Sox9 or Sox9-vil-Cre) to become healthy and fertile adult mice. No evidence for intestinal bleeding was found. Histological analysis of the intestine from 2–6-mo-old adult Sox9-vil-Cre mice revealed that, although the overall morphology of the small intestine seemed, at first sight, unaffected (), that of the colon was aberrant. The most striking feature of the Sox9-vil-Cre mice colon was the folding of the epithelium into villus-like structures, protruding into the colon lumen, reminiscent of the small intestine morphology (; and Fig. S1, B–D). Proliferation, however, was adequately restricted to the bottom half of the crypts in Sox9-vil-Cre mice, as assessed by Ki67 staining (). We then examined the differentiation pattern of the Sox9-vil-Cre mice colon epithelium into the three main types of differentiated colon epithelial cells. Among these, mucus-producing goblet cells represent the largest population and are responsible for epithelium protection and lubrication (). Most of the other cells are enterocytes, with few interspersed enteroendocrine cells. Alcian blue and Muc2 stainings showed that the goblet cell population was strongly decreased in the colon of Sox9-deficient animals (). No differences were detected in the morphology or staining intensity between individual alcian blue–positive goblet cells in Sox9 and Sox9-vil-Cre mice, and no changes in cellular representation were found either for the scarce enteroendocrine cell population () or for the Cdx2-expressing enterocyte population (). Thus, Sox9 is involved in defining the colon epithelium morphology and plays a specific role in the differentiation of the goblet cell lineage in the colon. When we examined the expression of markers representative of the four main cell lineages constituting the small intestinal epithelium, no major differences were found in enterocyte and enteroendocrine cell numbers, as shown by alkaline phosphatase and chromogranin A staining (). However, both the goblet and Paneth cell lineages were considerably affected. Alcian blue–positive goblet cells were found in the Sox9-vil-Cre intestinal epithelium and appropriately expressed Muc2, but their number was reduced by 40% compared with control mice (; and Fig. S2, A and B, available at ). Paneth cells represent the fourth cell type found in the small intestine. They secrete a variety of products, including antimicrobial peptides, growth factors, phospholipase A2, and matrilysin. These cells are involved in regulating the interactions between epithelial cells and the indigenous microorganism population, which, in turn, is essential to elaborate the microvasculature underlying the epithelium (; ). Remarkably, morphological identification coupled with staining of small intestinal sections from Sox9-vil-Cre mice for lysozyme, an early marker of Paneth cell differentiation, revealed that almost all the crypts were completely devoid of Paneth cells (). In Paneth cell–depleted crypts, the proliferative compartment expanded to occupy the whole crypt base, including the normal Paneth cell compartment (). As Paneth cells are located next to the putative stem cells, the replacement of Paneth cells by proliferating cells raises the possibility that the number of stem cells is altered in Sox9-deficient mice. Indeed, the number of cells positively stained with Musashi-1, a putative marker of stem cells in the nervous system and the intestinal epithelium (; ), was increased in Sox9-deficient mice (). Sox9 is thus required for differentiation of the Paneth cell lineage, is involved in differentiation of the goblet cell lineage, and might be involved in the regulation of the stem cell number. The observed decrease in Paneth and goblet lineages in Sox9-deficient mice was not due to increased apoptosis, as no differences were found in the apoptotic rates between Sox9-vil-Cre and Sox9 mice (Fig. S2 C). In the healthy human body, Paneth cells are also found uniquely in the small intestinal crypts of Lieberkühn. In some pathological situations, such as intestinal metaplasia, ectopic Paneth cells can also be found in the esophagus (Barrett's esophagus) or in the stomach (). Thus, if Sox9 is required for the differentiation of the Paneth cell lineage, it should be expressed in Paneth cells found in such aberrant structures. To test this, we analyzed biopsy sections from a patient with Barrett's esophagus. Paneth cells were detected using lysozyme expression, and Sox9 expression was analyzed on an adjacent section. Sox9 expression was found in most cells constituting crypt-like structures in the metaplasic area, including Paneth cells (). Thus, Sox9 expression also seems to be associated with Paneth cells in the pathological context of human intestinal metaplasia. To gain insight into the mechanism underlying the Sox9-dependent differentiation of Paneth cells, we screened by real-time PCR colon carcinoma cell lines for expression of Paneth cell markers. All the tested cell lines (SW480, HT29Cl.16E, HCT116, and DLD-1) had detectable expression of such markers, and this expression was highest in HT29Cl.16E cells (unpublished data), which were chosen for further analyses. A moderate (fivefold) overexpression of Sox9 in these cells resulted in an up-regulation of expression of several Paneth cell markers, which was most prominent for lysozyme, the matrix metalloproteinase MMP7, and Angiogenin-4 (ANG-4) mRNAs (). Thus, Sox9 may regulate the differentiation of Paneth cells, at least in part, through the transcriptional regulation of several markers of these cells. This regulation might be direct or indirect, but it likely contributes to the absence of identifiable Paneth cells in the intestinal epithelium of Sox9-deficient mice. We then asked whether the homeostasis of the epithelium would be conserved despite the extension of the proliferative compartment into the usual Paneth cell area and found that, in fact, the crypt size of Sox9-vil-Cre mice seemed increased compared with Sox9 control mice (). When crypt diameters and BrdU incorporation rates were measured, an unambiguous increase of crypt size was found in Sox9-deficient mice (), and the ratio between BrdU-labeled cells and the total number of cells found in a crypt circumference was slightly, but reproducibly, increased in the Sox9-deficient mice (). This was statistically significant (P < 0.0001). The total number of cells in any crypt circumference increased according to the crypt size, indicating that the cell size was not affected (unpublished data). This indicates that the absence of Sox9 resulted in increased cell proliferation, leading, in turn, to crypt hyperplasia throughout the small intestine. The epithelium from the proximal colon was also found to display hyperplastic features, but we were unable to perform accurate measurements because of its severely altered morphology in Sox9-vil-Cre mice. In addition to the general mild hyperplasia found throughout the intestine of Sox9-vil-Cre mice, extensive hyperplasia occurred, with occasional glandulocystic features, in the distal half of Sox9-vil-Cre mice colon. Numerous crypts were enlarged and branched, and some were extensively dilated with a cystic appearance (). Proliferation was correctly restricted to the bottom of hyperplastic crypts (). Cells constituting cystic crypts also proliferated, albeit modestly (), and were poorly differentiated (Fig. S2, D–F). In addition, tubulovillous microadenomas occasionally developed in hyperplastic areas of the epithelium with atypical tissue architecture (). Crypts with dysplastic features, including poor differentiation, pseudostratified nuclei, multiadenoid structures, and numerous mitosis, spontaneously developed in several locations along the colon of Sox9-deficient mice (). Interestingly, slight crypt hyperplasia/dysplasia was also detectable in the colon of 3-wk-old Sox9-deficient mice, indicating that these defects appear early but become more severe with time, likely as a consequence of increased cell proliferation (Fig. S2, G–J). Some Sox9-vil-Cre mice had no detectable lesions but also had a normal colon morphology and had Paneth cells in their small intestine (unpublished data). In such mice, Sox9 staining was invariably identical to that of wild-type mice (unpublished data), indicating inefficiency of the Cre recombinase. Thus, dysplastic-like lesions were always found in the colon of true Sox9-deficient mice. That we never observed true carcinoma in the intestine of Sox9-deficient mice up to 6 mo old suggests, in turn, that Sox9 deficiency may not be sufficient, per se, to induce cell transformation. Hyperplastic- and dysplastic-like crypts were found to strongly overexpress Wnt pathway–related genes, such as c-Myc and cyclin D1, suggesting an increase of Wnt-dependent transcriptional activity (, compare g with h, and i with j). This overexpression resulted from both an increase in the number of c-Myc– and cyclin D1–expressing cells and increased staining intensities in individual positive cells (unpublished data). Expression of the Ki-67 proliferation marker was also affected (, compare k with l). This finding was confirmed by Western blot analysis of extracts from Sox9 and Sox9-vil-Cre mice. Although few variations were found in the small intestine, likely because the few proliferating crypt cells are not sufficiently represented in the whole epithelial cell population, a clear increase of c-Myc and cyclin D1 was evident in the Sox9-deficient colon (). The increase in cell proliferation rate in Sox9-deficient mice (estimated as 15% from BrdU incorporation rates) was probably not sufficient to be clearly visualized with an anti-PCNA Western blot (). An alternative explanation is that, although crypt hyperplasia and increased BrdU intake are evident, the density of crypts is reduced in Sox9-deficient animals, which may compensate the increased proliferation observed in each crypt. That the overexpression of Wnt target genes is much more visible in the colon samples may reflect either the bigger size of the crypts relative to the entire epithelium in the colon compared with the small intestine or the presence of dysplastic-like lesions, which strongly overexpress Wnt target genes, in the colon samples. We then asked whether these alterations in the small intestine and colon structure may impact the mouse physiology. Indeed, the weight of Sox9-vil-Cre animals was always reduced (unpublished data), compared with the related Sox9 control mouse. This reduction was modest (mean 17%; = 10) but, despite the heterogeneity in the age and sex of the pairs of animals tested, reached statistical significance (P < 0.05, test). In addition, stools from Sox9-vil-Cre mice were more hydrated than those of control mice, indicating a partial impairment of the colonic epithelium function in Sox9-deficient animals (). To understand the molecular bases of the observed up-regulation of Wnt target genes, we compared the expression of nuclear β-catenin, the hallmark of Wnt signaling, in Sox9-deficient versus control mice. Comparable results were found in both situations, with a typical nuclear staining of some cells scattered through the crypt bottom (; and Fig. S3, A and B, available at ), indicating that the overexpression of Wnt target genes found in Sox9-deficient animals was not due to increased levels of β-catenin, in the crypt nuclei of Sox9-vil-Cre mice. We and others have shown that the level of Sox9 expression regulates the transcriptional activity of the β-catenin–Tcf4 complex in cultured HEK293 cells (; ). Physical interaction between Sox9 and β-catenin has been reported, resulting in a competition between Sox9 and Tcf4 for binding to β-catenin. Formation of the Sox9–β-catenin complex results in degradation of the two proteins (). We thus hypothesized that the absence of Sox9 in crypt cells of Sox9-vil-Cre mice, where Wnt signaling is physiologically active, might result in increased availability of the nuclear pool of β-catenin for binding to Tcf4. To test this, we analyzed the transcriptional activity of the β-catenin–Tcf4 complex after manipulation of Sox9 expression in colon carcinoma cell lines containing a constitutive activation of the β-catenin–Tcf transcriptional complex (). The basal β-catenin–Tcf4 activity present in DLD-1 cells was efficiently inhibited by Sox9 and increased after overexpression of a dominant-negative version of Sox9 (). Comparable results were obtained in other colon carcinoma cell lines, such as SW480, HCT116, and HT29Cl.16E (Fig. S3, C–E), demonstrating that even in colon carcinoma cells, in which nuclear β-catenin accumulates constitutively, the level of Sox9 expression critically modulates the level of β-catenin–Tcf transcriptional activity. We used the HT29Cl.16E-Sox9 and HT29Cl.16E-ΔCSox9 cell lines, inducibly overexpressing full-length or C-terminally truncated Sox9, respectively, to test whether Sox9 overexpression would result in a down-regulation of expression of endogenous Wnt pathway target genes. Doxycycline induction of Sox9 or ΔCSox9 expression also resulted in down- or up-regulation, respectively, of the β-catenin–Tcf complex activity (). Using real-time PCR, we found that induction of Sox9 expression resulted in a down- regulation of c-Myc and cyclin D1 mRNA expression, whereas that of ΔCSox9 resulted in an up-regulation of these two Wnt target genes (). C-Myc and cyclin D1 protein expression changed accordingly (). This result demonstrates modulation of expression of key β-catenin–Tcf target genes by Sox9, but the molecular mechanism underlying this regulation remained unclear, as the ΔC-Sox9 construct does not contain the domain thought to interact with β-catenin (). The previously reported physical interaction between Sox9 and β-catenin had been detected after overexpression of tagged Sox9 and β-catenin in COS cells (). We reasoned that if this interaction was important to modulate Wnt signaling in colon carcinoma cells, then the endogenous complex should be readily detectable in these cells. Despite repeated efforts, no β-catenin–Sox9 complex could be immunoprecipitated, although Tcf4 coimmunoprecipitated with β-catenin (unpublished data). We concluded that β-catenin–Sox9 complexes are probably not abundant in colon carcinoma cells. In addition, the subcellular localization of both β-catenin and Tcf4 was unchanged after transient overexpression of Sox9 in SW480 cells (Fig. S4, A–F, available at ), and the level of β-catenin expression was not decreased after induction of Sox9 expression in the HT29Cl.16E-Sox9 cells (Fig. S4 G). To identify another possible mechanism, we performed mutational analysis of the Sox9-mediated inhibition of the β-catenin–Tcf activity. The W143R point mutation () was originally identified in a campomelic dysplasia patient and abolishes the DNA binding properties of Sox9 (). We introduced this mutation in both the Sox9 and ΔCSox9 constructs. The resulting products did not have any transcriptional activity using Sox-luciferase reporters () and failed to modify β-catenin–Tcf activity (). This indicates that Sox9-mediated inhibition of the β-catenin–Tcf activity requires an intact DNA binding domain of Sox9, which raises the possibility that this inhibition might be at least partly due to transcriptional regulation. To test this, we used a Sox9 construct in which the C-terminal domain of Sox9, involved in transactivation, and in the interaction with β-catenin (), is removed and replaced by the unrelated VP16 transactivating domain (; ). Transient transfection of this construct in colon carcinoma cells showed that the Sox9-VP16 chimeric protein potently activates transcription of a Sox-luciferase reporter gene () and inhibits the β-catenin–Tcf activity even more efficiently than wild-type Sox9 (). We conclude that Sox9-mediated inhibition of β-catenin–Tcf activity involves transcriptional regulation. We then aimed to identify potential Wnt pathway inhibitors, such as the inhibitor of β-catenin and Tcf (ICAT; ) and Groucho-related (Grg/TLE) corepressors (; ), which might be transcriptionally regulated by Sox9. In vitro, the induction of Sox9 expression in HT29Cl.16E-Sox9 cells resulted in increased expression of the ICAT and TLE2-4 genes, whereas the expression of TLE1 was unaffected (). We then analyzed the expression of the mouse homologues of these genes in Sox9-vil-Cre mice and Sox9 mice, and we found that, again, the expression of Grg2, -3, and -4 was obviously down-regulated in Sox9-deficient mice, whereas that of Grg1 remained unchanged (). Icat expression seemed unchanged in Sox9-deficient mice, but this result varied with the samples analyzed (unpublished data). Although additional Wnt pathway inhibitors may be involved, this result provides a basis to explain the increased expression of Wnt target genes observed in Sox9-deficient mice, despite the absence of an increase in nuclear β-catenin expression. When we stained small intestinal tissue for c-Myc and cyclin D1 protein expression, we found that both genes were expressed as expected in the stem/progenitor cell compartment but that their expression was strongly decreased in Paneth cells (). In contrast, the absence of Sox9 resulted in uniform expression of both c-Myc and cyclin D1 in the whole crypt bottom (), which then lacked Paneth cells. i s s t u d y s h o w s t h a t l o s s o f S o x 9 f u n c t i o n a f f e c t s t h e i n t e s t i n a l e p i t h e l i u m p h y s i o p a t h o l o g y a t t h e l e v e l o f ( 1 ) c e l l d i f f e r e n t i a t i o n , ( 2 ) t i s s u e h o m e o s t a s i s , a n d ( 3 ) c o l o n t i s s u e m o r p h o l o g y . T h e s e d i f f e r e n t a s p e c t s o f t h e p h e n o t y p e w e r e s e x i n d e p e n d e n t . The vil-Cre strain () was in a nearly pure C57BL/6 background (at least 14 backcrosses to this background). To generate Sox9 mice (), exons 2 and 3 of the Sox9 gene were flanked by LoxP sequences. Cre recombination results in the deletion of the last two exons, which encode part of the HMG DNA binding domain and the transactivation domain, thus resulting in a likely null allele. A peptide might still be produced from the first exon, but this would not contain any known functional domain. Sox9 mice were originally on a 129P2/OlaHsd × C57BL/6 mixed genetic background and have been backcrossed to C57BL/6 for three generations. These N3 mice were made Sox9 by sister–brother mating before being crossed with the vil-Cre strain. Colorectal cancer cell lines HCT116, HT29.16E, DLD1, and SW480 were cultured at 37°C in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (Eurobio), 1% -glutamine, and penicillin/streptomycin. The HT29.16E-Sox9 and HT29.16EΔCSox9 cell lines were described previously (; ). Full-length human Sox9 and C-terminally truncated (dominant-negative) human Sox9 constructs were described previously (), as were the pTOP-FLASH and pFOP-FLASH reporter constructs (). The Sox9-VP16 construct () was a gift from H. Kondoh (Institute for Molecular and Cell Biology, Osaka University, Osaka, Japan). The full-length human Sox9 (W143R) and the C-terminally truncated human Sox9 (W143R) expression constructs were constructed from a human Sox9(1–304) construct containing the W143R mutation, provided by P. de Santa Barbara (Institut National de la Santé et de la Recherche Médicale, Montpellier, France). DLD1, SW480, HCT116, and HT29-16E cells were cotransfected (EXGEN500; Euromedex) with 0.25 μg pcDNA3, Sox9, or ΔCSox9 DNA constructs and 0.25 μg TCF/LEF-1 reporter (pTOP-FLASH) or control vector (pFOP-FLASH), using standard procedures. 0.025 μg pRLSV-Renilla was used as an internal control. Luciferase assays were performed with the Dual luciferase reporter assay system (Promega) according to the manufacturer's instructions. Luciferase activities in cell lysates were normalized relative to the Renilla luciferase activity, and the indicated activities represent the TOPFLASH/FOPFLASH ratio, indicative of the Tcf binding site-specific activity. Each experiment was performed in duplicate and repeated several times, and representative examples are shown. Immunohistochemistry was performed essentially as described previously (). Sections of human intestinal metaplasia were provided by F. Bibeau (CRLC Val d'Aurelle Lamarque, Montpellier, France). In brief, for preparation of mouse intestinal sections, the intestinal tract was dissected as a whole from 2–6-mo-old mice and flushed gently with cold PBS to remove any fecal content. The small intestine and colon were rolled up into a compact circle and fixed in 4% PFA in PBS at RT for 4 h, dehydrated, embedded in paraffin, and sectioned at 5 μm, using standard procedures. Sections were dewaxed in a xylene bath and rehydrated in graded alcohols. Endogenous peroxidase activity was quenched with 1.5% HO in methanol for 20 min and washed in PBS. Antigen retrieval was performed by boiling slides in 10 mM sodium citrate buffer, pH 6.0, except for anti–c-Myc and anti–cleaved Caspase3 antibodies, for which antigen retrieval was performed by boiling slides 20 min in 100 mM TRIS and 12.6 mM EDTA, pH 9.0. Nonspecific binding sites were blocked with 1% BSA, 3% NGS, and 0.2% Triton X-100 in PBS for 45 min at RT for all antibodies staining except for anti–c-Myc (1% BSA in PBS) and anti-Ki67 (no blocking). Slides were incubated with the primary antibodies overnight at 4°C in PBS with 0.1% BSA. In all cases, Envision+ (DakoCytomation) was used as a secondary reagent, stainings were developed with DAB (brown precipitate) or Vector Vip substrate (purple precipitate), and hematoxylin counterstain was used. After dehydration, sections were mounted in Pertex (Histolab). For alkaline phosphatase activity staining, sections were dewaxed and rehydrated as described, and the alkaline phosphatase substrate (Vector red; Vector Laboratories) was applied for 10 min. Sections were counterstained, dehydrated, and mounted as described. For BrdU countings, mice were injected with a solution (0.1 milligram per gram of mouse body weight) of Brdu diluted in PBS. Mice were killed 2 h after injection. For BrdU staining, the same method as explained above was used, except for an additional step in HCl 2N for 45 min, added after antigen retrieval. For the Musashi staining, an ABC kit (Vectastain) was used instead of the Envision+ system. For Western blotting, an equal amount of protein, measured by the Bradford assay, was loaded on each lane of the gel. Protein lysates and immunoblotting were performed as described previously (). Proteins were visualized using ECL Plus (GE Healthcare), and the bands were quantified by densitometry using ImageJ 1.32J (NIH). Immunofluorescence stainings were performed essentially as immunohistochemical stainings, but a goat anti–rabbit secondary antibody (Alexa Fluor 488; Invitrogen) was used, and nuclei were counterstained with Hoechst. Fluorescence quantifications were performed with ImageJ software. Nuclei were selected using the freehand selection tool, and integrated densities were measured. The Muc2 antibody (1:250) was provided by I. Van Seuningen (Institut National de la Santé et de la Recherche Médicale, Lille, France), the Cdx2 antibody (1:200) was provided by J.-N. Freund (Institut National de la Santé et de la Recherche Médicale, Strasbourg, France), the Musashi-biotin antibody (1:500) was provided by H. Okano (Keio University, Tokyo, Japan), and the anti-Sox9 was described previously (). β-Actin A5441 (Western blot; 1:5,000) was purchased from Sigma-Aldrich, anti-Ki-67 (1:200) and anti-lysozyme (1:1,000) were purchased from DakoCytomation, anti-PCNA (1:100) and anti–cyclin D1 (1:100) were purchased from Neomarker, anti-c-Myc (1:50) was purchased from Santa Cruz Biotechnology, Inc., anti–E-cadherin (1:150) and anti–β-catenin (1:50) were purchased from BD Biosciences, anti-Claudin2 (1:100) was obtained from Zymed Laboratories, anti–chromogranin A (1:300) was purchased from Immunostar, anti-BrdU (1:200) was obtained from Novocastra, and anti-Caspase3 (1:100) was purchased from Cell Signaling. To determine stool hydration, freshly isolated stool was weighted before and after overnight incubation at 50°C. Four mice of each genotype were analyzed. Mean and standard deviation values, as well as statistical significance ( test) were calculated using Excel (Microsoft). Fields containing crypt transverse sections were selected randomly at several locations along the rostrocaudal axis of the small intestine. Only sections with several BrdU-positive cells and an apparent lumen were considered to avoid large variation in the position of the section in the crypt–villus axis. Six to nine fields, each containing 6–40 measured crypts, were analyzed by two individuals, who were blinded to the mouse genotypes, and two mice of each genotype were analyzed. Mean and standard deviation values, as well as statistical significance ( test) were calculated using Excel. Immunohistochemistry images were acquired at RT using either an Axiophot microscope (Carl Zeiss MicroImaging, Inc.) with 10× 0.3 Plan Neofluar or 40× 1.0 Plan Apochromat lenses (Carl Zeiss MicroImaging Inc.) and a camera (DXM1200; Nikon) or an Eclipse 80i microscope (Nikon) with Plan Fluor 10× 0.3, 20× 0.5, 40× 0.75, and 60× 0.5–1.25 lenses (Nikon) and a camera (Q-Imaging Retiga 2000R with a Q-Imaging RGB Slider). Images were acquired with ACT-1 or Q-Capture Pro softwares (Nikon) and manipulated with Photoshop (Adobe), using the crop, levels, curves, brightness/contrast, and image size commands. For acquisition of immunofluorescence experiments, an Axiophot2 (Carl Zeiss MicroImaging, Inc.) microscope was used with a 63× 1.4 Plan Apochromat objective (oil) and a Coolsnap (Photometrics) camera driven by the Metavue software. Fig. S1 shows Sox9 expression in Paneth cells fand a general view of a Sox9-deficient colon, showing the localization and extent of the described phenotypic features. Fig. S2 provides an analysis of the goblet cell population and of apoptosis rates in Sox9-deficient mice. Fig. S3 shows β-catenin expression in Sox9-deficient mice and consequences of Sox9 and ΔCSox9 overexpression on the β-catenin–Tcf4 activity in a panel of colon carcinoma cell lines. Fig. S4 shows no modification in the subcellular localization of β-catenin or Tcf4 or in the degradation of β-catenin after induction of Sox9 expression in the HT29Cl.16E-Sox9 cell line. Fig. S5 shows that in Sox9-deficient mice, the c-Myc and cyclin D1 genes are expressed by an increased number of cells, and individual cells express higher levels of the c-Myc and cyclin D1 proteins. Online supplemental material is available at .

Commitment of a cell to develop along a prescribed pathway is thought to occur in two stages (). The first stage, which is called specification, is characterized by the ability of a cell to differentiate autonomously in a neutral environment; however, the fate of the cell can still be redirected. During the determination phase, a cell will differentiate according to its prescribed fate even when placed in an environment that induces the development of other tissue types. In this case, commitment is considered to be stable and irreversible under normal circumstances. Previous studies suggested that commitment to the skeletal muscle lineage occurs in the somites in response to factors released by surrounding structures and those produced within the somites themselves (; ; for reviews see ; ; ). Stable commitment was operationally defined as the maintenance of myogenic potential when the muscle-forming region of the somite was exposed to factors that promote chondrogenesis (; ; ; ; ; ; ; ; ). At the molecular level, commitment is driven by up-regulation of the skeletal muscle–specific transcription factors Myf5, MyoD, and Mrf4 in progenitor cells expressing the paired box transcription factors Pax-3 and/or Pax-7 (; ; ; ; ; ; ; ; ; ; , ). Genetic manipulations of the mouse embryo revealed that Myf5 and Mrf4 are responsible for the initial activation of MyoD, whereas Pax-3 assumes this function later in development (; ). Analyses of the chick embryo have yielded conflicting results regarding the sequence of expression of skeletal muscle transcription factors. In situ hybridization with enzymatic probes revealed that Myf5 but not MyoD was expressed in the presomitic mesoderm (; ; ; ). Low levels of Myf5 were detected in the primitive streak and adjacent epiblast of the stage 3 embryo (). In contrast, in situ hybridization with fluorescent dendrimer probes and RT-PCR demonstrated the presence of MyoD mRNA in the presomitic mesoderm, stage 3 embryo, and stage 1 epiblast (; ; ). Given that epiblast cells expressing low levels of MyoD mRNA represent a small subpopulation within the presomitic mesoderm and somites (), it is possible that they were not clearly visible with enzymatic probes. These MyoD-positive (MyoD) cells may correspond to the small subpopulation of presomitic mesoderm cells that are capable of differentiating in cultured explants of presomitic mesoderm tissue (; ). The importance of MyoD expression in the epiblast of the chick embryo was demonstrated in vitro and in vivo. When epiblast cells that express MyoD mRNA were isolated from the embryo and cultured in serum-free medium, nearly all differentiated into skeletal muscle (). In vivo, most MyoD epiblast cells were incorporated into the somites and synthesized Noggin (). Noggin promotes skeletal muscle differentiation in the somites by blocking the bone morphogenetic protein (BMP) signaling pathway (; ; ; ; ; ; ; ; ; ). Ablation of MyoD cells in the epiblast resulted in a decrease in Noggin in the somites and a dramatic reduction in skeletal muscle in the trunk and limbs (). The inhibition of muscle differentiation after ablation was averted with the addition of exogenous Noggin. Thus, cells that express MyoD mRNA in the epiblast regulate skeletal myogenesis in the somites by releasing Noggin. The ability to differentiate in vitro and to promote muscle differentiation in vivo does not necessarily indicate that cells expressing MyoD mRNA in the epiblast are stably committed to the skeletal muscle lineage. Several studies have demonstrated that skeletal muscle transcription factors are transiently expressed during the early stages of development. Myf5 and MyoD mRNAs were initially expressed in nonmyogenic tissues of the chick and embryo, respectively, and gradually became restricted to muscle-forming regions of the somite (; ; ). Separation of the somites from surrounding tissues resulted in a down-regulation of MyoD and Myf5 and a failure of cells to differentiate (; ; ; ; ; ; ). Furthermore, some cells that express skeletal muscle transcription factors in adult muscle can be induced to differentiate into bone and adipocytes (for review see ). These studies suggest that low levels of expression of skeletal muscle transcription factors may be the hallmark of specification but not of stable commitment to the skeletal muscle lineage. However, small numbers of cells in the presomitic mesoderm and myogenic precursors in the limb that have not up-regulated skeletal muscle–specific transcription factors do undergo skeletal myogenesis even when challenged with cartilage-promoting factors from the notochord (, ). The following studies were designed to determine whether MyoD epiblast cells are stably committed to the skeletal muscle lineage or whether their fate can be altered in environments that induce the differentiation of nonskeletal muscle tissues. The environment of the developing heart is particularly challenging to skeletal myogenesis because BMPs are required for specification of the heart-forming fields and cardiomyocyte differentiation (; ), but they inhibit skeletal muscle differentiation (; ; ; ; ; ; ; ; ). BMPs also play a role in the induction of dorsal cell fates in the neural tube (; ; ; ; ; ; ). Although BMP inhibitors are required for establishing ventral neuronal fates, neural tube closure, and anterior brain formation (; ; ), skeletal muscle differentiation is not induced in the nervous system. The state of commitment of MyoD epiblast cells was examined in two ways. First, cells that express MyoD in the epiblast were tracked into the heart and nervous system. Second, MyoD cells were isolated from the epiblast and microinjected into the precardiac mesoderm and neural plate. In both types of experiments, MyoD epiblast cells continued to express MyoD mRNA, some synthesized MyoD protein, and none differentiated into cardiac muscle or neurons. In contrast, cells that did not express MyoD in the epiblast differentiated according to their location. MyoD mRNA is expressed in a small subpopulation of cells in the pregastrulating epiblast (; , ; ). To determine whether these cells produce MyoD protein, stage 1–4 embryos were double labeled with mAbs to MyoD and the G8 antigen. The G8 mAb binds to a cell surface antigen and is a specific marker for cells that express MyoD mRNA in the epiblast (; , ; ). MyoD protein was not detected in stage 1 or 2 embryos, and only a single G8-positive (G8) cell in the stage 4 epiblast was labeled with the MyoD antibody (). These results were confirmed with a rabbit polyclonal antiserum to MyoD. Therefore, MyoD mRNA either is not translated or the protein does not accumulate to detectable levels in the early epiblast. MyoD epiblast cells were further characterized by determining whether they express Myf5 mRNA. In agreement with the results of , Myf5 mRNA was not detected by in situ hybridization in the stage 1 or 2 epiblast (). The lack of detection of Myf5 mRNA in the pregastrulating epiblast suggests that MyoD is expressed before Myf5 in the chick embryo, although analyses of the embryo before laying would be required to demonstrate this definitively. Myf5 mRNA was detected in only a subpopulation of G8 cells in the stage 4 epiblast (). Interestingly, a few stage 4 epiblast cells that did not appear to express the G8 antigen expressed Myf5 (). Therefore, epiblast cells of gastrulating embryos may be heterogeneous with respect to their expression of muscle regulatory factors. Cells expressing MyoD mRNA in the stage 2 epiblast were tracked into the heart and nervous system by fluorescently labeling them with the G8 mAb and incubating the embryos for 3 d in ovo (stage 16). Although the majority of G8 cells were incorporated into the somites (56%), ∼9 labeled cells were found in the heart, 10 were found in the neural tube, 9 were in the brain, and 27 were found in other nonsomitic tissues of the embryo, including the mesenchyme of the head. Most were present as single cells surrounded by G8-negative (G8) cells. Within the somites, all of the G8 cells contained MyoD protein (), and the majority (73%) had synthesized sarcomeric myosin. All of the G8 cells tracked into the heart continued to express MyoD mRNA, and a subpopulation (59%) was stained with the MyoD mAb (). Although most cells lacking the G8 label had synthesized cardiac troponin T, this marker for cardiac muscle differentiation was not detected in any G8 cells in the stage 16 heart (). Within the central nervous system (CNS) of the stage 16 embryo, all G8 cells that originated in the epiblast continued to express MyoD mRNA, and ∼50% were stained with the MyoD mAb (). Only a single G8 cell contained sarcomeric myosin (). None of the G8 cells were labeled with an antibody to neurofilament-associated antigen (). The CNS () and heart contained a few cells that expressed MyoD but lacked the G8 tag that had been applied in the stage 2 embryo. This is consistent with the observation that more G8 cells (∼36 cells) were found in hearts directly labeled with the G8 mAb than the number of G8 epiblast cells tracked into the heart (nine cells). Some cells with MyoD mRNA but lacking the G8 tag applied in the epiblast were present in clusters containing G8-labeled cells, whereas others were surrounded by G8 cells (). Although it is possible that the G8 signal was lost in some cells as a result of proliferation, the expression of MyoD may have been initiated in a separate population after application of the antibody (). Importantly, no cell that expressed MyoD mRNA and the G8 antigen within the epiblast and were tracked into the heart or nervous system or any cells that may have initiated G8 synthesis sometime after stage 2 of development contained detectable levels of cardiac troponin T or neurofilament-associated antigen. These results indicate that G8 epiblast cells are not induced to differentiate into cardiac muscle or neurons in the developing heart and nervous system. In the aforementioned tracking experiments, it is possible that the restriction of developmental potential may have occurred in cells on route to their final destination. Therefore, a second approach was taken to challenge the behavior of MyoD epiblast cells that involved microinjecting them directly into the precardiac mesoderm and neural plate. Stage 1 epiblasts were removed from the embryo, dissociated, labeled with the G8 mAb, and the G8 and G8 populations were isolated by magnetic cell sorting. The purity of both sorted populations was >97% (). Sorted cells were labeled with Hoechst dye, a procedure that did not affect their viability or ability to differentiate in vitro (unpublished data). 60 G8 or G8 Hoechst-labeled epiblast cells were microinjected into six sites (10 cells per site) of the precardiac mesoderm of stage 4–5 embryos () or the neural plate of stage 6–7 embryos (). The microinjection procedure did not appear to affect morphogenesis of the heart or nervous system during the course of the experiment (; and ). The expression of cell type–specific markers was analyzed in tissue sections and after the dissociation of tissues and centrifugation of the cell suspensions onto slides. After microinjecting cells into the precardiac mesoderm of stage 4–5 embryos, ∼94% of the Hoechst-labeled cells were later found in stage 12–14 hearts (). Microinjected G8 cells increased in number to a greater extent than G8 cells (2.5- and 1.9-fold, respectively; P ≤ 0.05). Some clusters of two to four G8 cells were found within the myocardium, although most were present as single cells surrounded by host cells (). The majority of microinjected G8 cells was present within the middle of the myocardium (), whereas G8 epiblast cells often were found toward the periphery of the myocardium (), suggesting that some cell sorting may have occurred. None of the Hoechst-labeled G8 cells contained detectable levels of cardiac troponin T (; and ). Instead, 99% contained MyoD mRNA (), and most were labeled with the MyoD mAb (; and ). Some of the microinjected G8 cells differentiated into skeletal muscle, as indicated by staining with the 12101 mAb (), although the percentage of these cells that expressed 12101 in vivo varied greatly between experiments (). This may reflect a delay in the accumulation of this antigen after terminal differentiation because the majority of G8 cells contained sarcomeric myosin (). G8 epiblast cells microinjected into the precardiac mesoderm displayed a greater tendency to differentiate into skeletal muscle than those that were tracked from the epiblast into the heart. This suggests that the procedure for isolating and dissociating the epiblast in preparation for sorting and microinjection may have enhanced the ability of MyoD cells to differentiate in foreign environments. Cell–cell interactions within the epiblast epithelium and a factor produced in the mesoderm are inhibitory for skeletal myogenesis (). The procedure for isolating epiblast cells is not sufficient to trigger skeletal myogenesis in epiblast cells that lack MyoD mRNA. Less than 1% of microinjected G8 cells or their progeny contained detectable levels of MyoD mRNA () or MyoD protein, and none appeared to synthesize the 12101 antigen (). Unlike the G8 cells, nearly all of the G8 cells that were microinjected into the precardiac mesoderm differentiated into cardiomyocytes (; and ). A greater percentage of G8 epiblast cells were labeled with the cardiac troponin antibody than host cells (P ≤ 0.03), illustrating their proclivity for differentiation. A small decrease was found in the number of host cells that differentiated into cardiac muscle in embryos microinjected with G8 cells than G8 cells (). Slightly more host cells were stained with mAbs to MyoD and 12101 in hearts implanted with G8 cells than G8 cells (). This raises the possibility that microinjected G8 cells influence the pathway of the differentiation of host cells in the heart. The results obtained when epiblast cells were microinjected into the developing heart were consistent with the cell-tracking experiments. That is, cells expressing MyoD mRNA in the epiblast continued to do so in the heart and were not redirected to the cardiac muscle lineage. In contrast, cells that lacked MyoD mRNA in the epiblast were capable of differentiating into cardiac muscle. The fate of epiblast cells that express MyoD mRNA was also tested in the developing nervous system. As was the case with implantations into the heart, the microinjected G8 cells increased in number to a greater extent than G8 cells (10-fold and ninefold, respectively). The higher rates of proliferation of cells microinjected into the neural plate than in the precardiac mesoderm is consistent with the observation that cardiomyocyte differentiation was nearly complete by the time the embryos were fixed for analysis (). Greater than 80% of the microinjected cells were found in the head (). Within the head, 75% of Hoechst-labeled cells were present in the brain. Approximately 60% of the Hoechst-labeled cells in the trunk were found in the neural tube (). Most of the microinjected cells were surrounded by host cells, although some were present in clusters of two to four cells in the brain and neural tube (). Neurofilament-associated antigen was not detected in any of the G8 cells microinjected into the neural plate even when they were surrounded by host cells containing this marker of neuronal differentiation (; and ). Instead, all G8 cells found within the CNS () and other embryonic tissues contained MyoD mRNA. Within the CNS, some G8 cells were labeled with mAbs to MyoD protein (∼70%; ), sarcomeric myosin (∼13%; ), and the 12101 antigen (∼7%). The majority (∼55%) of Hoechst-labeled G8 cells present in the mesenchyme of the head or myogenic region of the somite had differentiated into skeletal muscle. A few host cells were found to express MyoD mRNA in the brain (). In contrast to the behavior of G8 cells, ∼70% of G8 cells that were microinjected into the neural plate and incorporated into nervous tissue expressed neurofilament-associated antigen (). Only 1% of the Hoechst-labeled G8 cells contained MyoD mRNA (), and none were stained with mAbs to MyoD protein or the 12101 antigen. All Hoechst-labeled G8 cells found in the myogenic region of the somite did express MyoD mRNA (unpublished data). A precise determination of the percentages of cells microinjected into the neural plate that later differentiated into neurons or skeletal muscle throughout the embryo was calculated by separating the head from the trunk, dissociating the tissues to produce a single-cell suspension, centrifuging the cells onto slides, and staining with antibodies (). Significantly more Hoechst-labeled G8 than G8 cells were stained with the MyoD mAb (head, P ≤ 0.000005; trunk, P ≤ 0.0003) and differentiated into skeletal muscle (head, P ≤ 0.02). Importantly, no microinjected G8 epiblast cells were stained with the antibody to neurofilament-associated antigen, whereas approximately one third of the microinjected G8 cells in the head and 10% in the remainder of the embryo expressed this marker of neurogenesis. The results of experiments involving the microinjection of epiblast cells into the neural plate were consistent with those in which cells were tracked into the nervous system. They also mirrored the data obtained when epiblast cells were microinjected or tracked into the heart. Regardless of whether MyoD cells were incorporated into the nervous system or heart, they either remained as skeletal muscle precursors or formed skeletal muscle. They did not differentiate into neurons or cardiac muscle. Although G8 epiblast cells microinjected into the precardiac mesoderm and neural plate continued to express MyoD mRNA in the heart and nervous system, only a subpopulation synthesized sarcomeric myosin and the 12101 antigen. To test whether the population of microinjected G8 epiblast cells that remained undifferentiated was capable of undergoing skeletal myogenesis, cell cultures were prepared from hearts, heads, and trunks. Because epiblast cells lacking MyoD mRNA in vivo do not form skeletal muscle in this culture system, the conditions are permissive and not instructive for skeletal myogenesis (; ). 4 d after plating, 90–100% of Hoechst-labeled G8 epiblast cells that had been microinjected into the embryo synthesized the 12101 antigen (90 ± 5% from the heart, 95 ± 4% from the head, and 100% from the trunk; = 3 cultures per region; ). Therefore, MyoD epiblast cells microinjected into the precardiac mesoderm and neural plate are able to differentiate into skeletal muscle in a permissive environment. MyoD mRNA is expressed in a subpopulation of cells in the epiblast before the onset of gastrulation (; , ; ). When these cells are isolated from the stage 1 chick embryo and cultured in serum-free medium, >95% differentiate into skeletal muscle (). In this study, we tested whether MyoD epiblast cells are stably committed to the skeletal muscle lineage by examining their behavior in environments that promote cardiomyogenesis and neurogenesis. Cells expressing MyoD mRNA in the epiblast were either labeled in vivo and tracked into the heart and nervous system or isolated from stage 1 embryos and microinjected into the precardiac mesoderm and neural plate of gastrulating embryos. Both types of experiments revealed that MyoD epiblast cells survived, proliferated, and continued to express MyoD in the heart and nervous system. A subpopulation of these cells differentiated into skeletal muscle in these ectopic locations. Nearly all MyoD epiblast cells that were microinjected into the heart and nervous system differentiated into skeletal muscle when placed in culture. These findings indicate that epiblast cells expressing MyoD mRNA but lacking detectable MyoD protein or mRNA for Myf5 (this study), myogenin, or sarcomeric myosin (; ) are stably committed to the skeletal muscle lineage before the onset of gastrulation. Therefore, incorporation into the mesoderm and somites is not required for the specification and determination of this enigmatic population of skeletal muscle stem cells. Most cells that do not express MyoD in the epiblast appear to be uncommitted and multipotent because they differentiate into either cardiac muscle or neurons when microinjected into the precardiac mesoderm or neural plate. Their proclivity for differentiating according to environmental signals was revealed within the context of a developing tissue. Multipotent stem cells used for tissue regeneration may not display the same ability to differentiate into the desired cell type because the environment within diseased tissues of the adult may lack the molecules required for inducing specification, determination, and differentiation. Premixing multipotent cells with stably committed cells capable of producing factors that recruit unspecified cells to the appropriate lineage may improve the efficiency of tissue regeneration. Our cell-tracking experiments revealed that the majority of MyoD epiblast cells were incorporated into the somites (; and this study). Although many of these cells differentiated into skeletal muscle, some were retained in the dermomyotome region of the somite that contains replicating myogenic precursors. Therefore, MyoD epiblast cells may undergo self-renewal and produce offspring that differentiate in the somite. Some MyoD epiblast cells were integrated in the heart, nervous system, and other structures. This observation is consistent with the locations of MyoD cells in the epiblast that correspond to areas fated to give rise to somites as well as nonsomitic tissues (). Given their location in the epiblast, it is likely that cells with MyoD mRNA followed similar routes to those taken by other epiblast cells destined for integration into the heart and nervous system. In this case, MyoD cells would be exposed to factors that regulate early stages in the specification of cells to the cardiogenic or neurogenic lineages. Specification of the nervous system begins before gastrulation in the central epiblast, as indicated by the expression of the preneural markers ENRI and SOX3; however, the definitive neural marker SOX2 does not appear until neural plate formation (; ; ; ). Specification of cardiomyocytes also begins before the onset of gastrulation in the posterior epiblast (; ), although the process continues in an anterior to posterior direction in the primitive heart (; ). Exposure to early inducers of heart specification in the posterior epiblast may not be required for recruitment to the cardiomyocyte lineage because >95% of MyoD cells located throughout the epiblast differentiated into cardiac muscle when microinjected into the precardiac mesoderm. We previously determined that small numbers of MyoD cells are present in the heart, brain, and other organs of the fetal chicken (). Ectopically located skeletal muscle precursors of the fetus may be the direct descendants of MyoD epiblast cells that were incorporated into all three germ layers during early stages of development. Murine embryos also contain ectopically located skeletal muscle precursors, as indicated by the expression of Myf5 in the nervous system and the emergence of skeletal muscle in cultures prepared from mouse neural tubes (; ). It remains to be determined whether precursor cells expressing MyoD survive in ectopic locations of the adult. If so, they may be capable of proliferating in response to inflammatory cytokines and growth factors. Skeletal muscle precursors may be vulnerable to mutations leading to the development of rhabdomyosarcoma tumors, which often arise outside of skeletal muscle (). MyoD epiblast cells incorporated into seemingly ectopic locations may be involved in the formation of nonskeletal muscle tissues by serving as a local source of a BMP inhibitor. Cells that express MyoD in the epiblast produce Noggin both inside and outside of the somites (). Within the somites, the release of Noggin from MyoD epiblast cells is critical for skeletal muscle differentiation (). Noggin derived from MyoD epiblast cells also may be important for the development of other structures because ablation of these cells in the epiblast results in facial and eye defects (). The population of MyoD epiblast cells appears to be distinct from those cells that become committed to undergo myogenesis within the somites. Uncommitted myogenic precursors in the somites express Pax-3 and Pax-7 but not Myf5 or MyoD (; ; ; ; ). Pax-3–positive precursor cells are present in the somites after MyoD cells are ablated in the epiblast, and they differentiate in response to exogenous Noggin (). Therefore, myogenic precursors that arise in the somites do not appear to be the direct descendants of cells that express MyoD in the stage 2 epiblast. Another difference between MyoD epiblast cells and the majority of myogenic precursors in the somite is that the latter express Myf5 before MyoD (; ; ; ; ; ). In contrast, MyoD but not Myf5 mRNA is present in stage 1 and 2 chick embryos (; and this study). By stage 3, only a subpopulation of epiblast cells with MyoD coexpresses Myf5 and vice versa. In conclusion, the expression of MyoD mRNA in the pregastrulating epiblast defines a unique population of stem cells that are committed to the skeletal muscle lineage and are capable of self-renewing and differentiating. These cells promote the differentiation of a separate population of skeletal muscle precursors that arise within the mesoderm (). MyoD epiblast cells are also integrated into nonsomitic tissues of the embryo. In these seemingly ectopic locations, they retain their identity as skeletal muscle stem cells (this study) and produce Noggin (). Stable programming within the epiblast may ensure that MyoD cells express similar regulatory molecules in a variety of environments. White Leghorn chick embryos (B E Eggs) were staged according to the method of . Three stage 1, four stage 2, four stage 3, and three stage 4 whole embryos were double labeled for the G8 antigen and MyoD protein. G8 is a cell surface antigen that is specifically expressed in epiblast cells that contain MyoD mRNA (, ,; ). Embryos were incubated with 35 μg G8 mAb diluted in 100 μl of 10% goat serum in PBS and goat anti–mouse IgM antibodies conjugated with AlexaFluor488 (Invitrogen) diluted 1:1,000 in 10% goat serum in PBS. After permeabilizing with 0.5% Triton X-100, embryos were incubated with NCL-MyoD1 mAb to MyoD (Novacastra) diluted 1:1,000 and with goat anti–mouse IgG Fab′2 fragments conjugated with rhodamine (Jackson ImmunoResearch Laboratories) diluted 1:400 in 10% goat serum in PBS. The NCL-MyoD1 mAb labels pectoralis skeletal muscle and G8 epiblast cells but not G8 epiblast cells, cardiac muscle, or fibroblasts in culture. This mAb also stains the dermomyotome, myotome, and small numbers of cells that express MyoD mRNA outside of the somites in vivo. Stage 2 and 4 whole embryos also were labeled with the rabbit 6975B polyclonal antiserum to MyoD (a gift from S. Tapscott, Fred Hutchinson Cancer Research Center, Seattle, WA) and a goat anti–rabbit IgG conjugated with rhodamine (Chemicon). Nuclei were counterstained with 1 μg/ml of Hoechst dye in deionized water (Sigma-Aldrich) for 10 min. Labeling of embryos was performed at room temperature. Three stage 1, two stage 2, two stage 3, and four stage 4 whole embryos were double labeled for the G8 antigen and Myf5 mRNA as described previously (, ,; ). The G8 mAb was tagged with secondary antibodies conjugated with AlexaFluor488 as described in the previous paragraph. After permeabilizing in 0.1% Triton X-100 and 0.1% pepsin (Sigma-Aldrich), embryos were incubated in Cy3-labeled 3DNA dendrimers (Genisphere, Inc.) conjugated with an antisense cDNA sequence for chicken Myf5 mRNA (S53719; 5′-ATATAGTGGATGGCAGAGCTGAGGATTTCG-3′; ). Dendrimers lacking cDNAs for a specific mRNA were used as a negative control for background fluorescence. Nuclei were counterstained with Hoechst dye as described in the previous paragraph. The tracking of MyoD epiblast cells was performed as described previously (). In brief, stage 2 embryos were removed from the shell on the yolk and incubated for 45 min at room temperature in 35 μg G8 mAb/100 μl HBSS (Invitrogen) and rinsed three times in PBS. 100 μl of secondary antibodies conjugated with rhodamine diluted 1:400 in Hanks buffer or AlexaFluor488 diluted 1:1,000 in Hanks buffer was added for 45 min, and the embryos were rinsed three times in PBS. Labeled embryos still on the yolk were placed in an empty shell, incubated at 37°C, fixed in formaldehyde, embedded in paraffin, and sectioned transversely at 10 μm. Paraffin sections were labeled with mAbs to sarcomeric myosin heavy chain (MF20 mAb diluted 1:60; ), the skeletal muscle–specific 12101 antigen (12101 mAb diluted 1:10; ), neurofilament-associated antigen (3A10 mAb diluted 1:400; ), MyoD1 (NCL-MyoD1 diluted 1:150), or cardiac troponin T (AB-1 mAb diluted 1:400; Neomarkers). The 12101 mAb stained skeletal muscle in cultures of pectoralis muscle and G8 epiblast cells and in sections through the myotomes and limbs. 12101 did not stain cardiac muscle in vivo or in vitro. The cardiac troponin T mAb labeled cardiac muscle in culture and in sections through the heart but did not stain skeletal muscle in sections through the somites and limbs or cultures of pectoralis muscle or G8 epiblast cells (). Primary antibodies were labeled with secondary antibodies conjugated with rhodamine or AlexaFluor488. All antibodies were diluted in 10% goat serum in PBS. The MF20, 12101, and 3A10 mAbs were obtained from the Developmental Studies Hybridoma Bank. Most sections were double labeled with antibodies and Cy3-labeled dendrimers conjugated with an antisense oligonucleotide sequence to chicken MyoD (5′-TTCTCAAGAGCAAATACTCACCATTTGGTGATTCCGTGTAGTA-3′ [L34006; ]). Nuclei were counterstained with Hoechst dye. The sites of incorporation of G8 epiblast cells were determined in two stage 16 embryos. G8 and G8 cells were separated by magnetic cell sorting as described previously (; ) In brief, epiblasts were isolated from the stage 1 embryo, dissociated in 0.25% trypsin-EDTA (Invitrogen) for 10 min, incubated with the G8 mAb for 45 min, and labeled with rat anti–mouse IgM microbeads (Miltenyi Biotec). Nuclei were counterstained with 1 μg Hoechst dye/ml Hanks buffer. Sorting of G8 and G8 cells was performed in a MiniMACS column (Miltenyi Biotec). After centrifugation, cells were resuspended in PBS (1,000 cells/μl). 2 μl of fast green solution (0.1 μl of fast green/μl of 70% glycerol; both were obtained from Sigma-Aldrich) was added to the cell suspension. Embryos were rinsed in Hanks buffer and placed on nucleopore filters. Microinjections were performed with a microinjector (Nanoject II; Drummond Scientific). 10 cells were microinjected into each of six sites in the precardiac mesoderm (). Bilateral injections were made just above the rostral end of the head process, lateral to the head process, and lateral to the rostral end of the primitive streak. Implantations into the neural plate of stage 6–7 embryos consisted of six microinjections along the midline of the developing neural folds above the rostral end of the primitive streak (). Embryos on the nucleopore filters were transferred to nine-well 2.5-cm plates containing thin egg albumen and a piece of ashless filter paper (Whatman) with a hole cut in its center. The nine-well plate was placed in a Petri dish on filter paper moistened with PBS and incubated at 37°C with 5% CO for 48 h. Embryos receiving microinjections into the precardiac mesoderm (two embryos with G8 cells and two embryos with G8 cells) were fixed at stages 12–14, embedded in paraffin, and serially sectioned. Embryos that were microinjected into the neural plate (two embryos with G8 cells and two embryos with G8 cells) were fixed at stages 15–16. Sections were labeled with dendrimers to MyoD mRNA and/or mAbs to MyoD, MF20, 12101, cardiac troponin T, or 3A10 as described above (see Double labeling of whole embryos). Hearts were obtained from seven stage 10–14 embryos that received microinjections of G8 or G8 cells into the precardiac mesoderm (three experiments; two to three hearts pooled per experiment). Head and trunk cells were obtained from seven stage 15–17 embryos that received microinjections into the neural plate (two experiments; three to four embryos pooled per experiment). A test was used to compare populations. 2 d after Hoechst-labeled G8 and G8 stage 1 epiblast cells were microinjected into the precardiac mesoderm or neural plate, the hearts, heads, and trunks from two embryos were dissociated in trypsin-EDTA. Cells were plated at a density of 20,000/15 μl of medium on gelatin- and fibronectin-coated dishes as described previously (, ). Dishes were flooded with 1.5 ml DME/F12 (Invitrogen). Cells were fixed in 2% formaldehyde after 4 d in culture and stained with the 12101 mAb.

Barrier to autointegration factor (BAF) was first discovered as a host component of retroviral preintegration complexes, which is required for integrase-mediated retroviral DNA insertion into target DNA in vitro (; ; ). BAF is highly conserved in metazoan evolution (52% identical between human and ; ) and typically localizes at the nuclear periphery, nucleoplasm, and cytoplasm (). BAF binds directly to many different partners, including double-stranded DNA, histone H3, and certain linker histones, and to all characterized LAP2–emerin–MAN1 (LEM) domain proteins plus lamins and selected homeodomain transcription activators (; ). These interactions are regulated, at least in part, by the posttranslational modification of BAF (; ; ). In , the nuclear envelope localization of BAF-1 protein (previously termed Ce-BAF) depends on its interaction with two inner nuclear membrane LEM domain proteins encoded by (Ce-emerin) and (Ce-lem2; formerly known as Ce-MAN1; ). In turn, BAF-1 is required to localize both LEM domain proteins and Ce-lamin in embryonic cells, suggesting mutual structural interdependence (). RNAi-mediated down-regulation in of either or or double down-regulation of both and caused chromosome segregation defects and failure to properly assemble daughter nuclei (). In mammalian cells, the ectopic expression of mutant BAF that cannot bind DNA or LEM domain proteins dominantly blocked the recruitment of lamin A and the LEM domain proteins emerin and LAP2β but had no effect on B-type lamins or on LBR (lamin B receptor protein; ). Structural roles were also seen in egg extracts, in which nuclei can assemble in vitro; adding recombinant BAF either inhibited or enhanced nuclear assembly, depending on the amount of BAF added (), suggesting important roles for BAF in organizing chromatin and the nucleus. Indeed, BAF is essential in both () and (). BAF-null die at the larval-pupal transition, when they run out of maternally deposited BAF, with phenotypes that include arrest at various stages of the cell cycle, chromatin clumping, abnormal lamin distribution, aberrant nuclear morphology, small brains, and missing imaginal discs (). RNAi-mediated down-regulation of in revealed that the loss of both maternal and zygotic BAF-1 caused anaphase chromatin bridges, abnormal chromatin morphology, and chromosome missegregation as early as the two-cell stage, and embryos died at or before the ∼100-cell stage (; ). The few embryos that escaped () lethality grew into sterile adults with misplaced distal tip cells and gonads (). To better understand the potential cellular and developmental roles of BAF, we studied a loss of function mutation in (). In animals homozygous for the allele, the maternal supply of BAF-1 was sufficient to allow these animals to complete embryogenesis and larval stages, bypassing BAF-1's mitotic roles and allowing us to focus on later stages of development. This genetic analysis reveals several novel tissue-specific roles for BAF-1 and sheds new light on the disease mechanisms of Emery-Dreifuss muscular dystrophy, which is caused by mutations in each of BAF's direct binding partners emerin and lamin A. In early embryos, BAF-1 is enriched near the nuclear inner membrane (). To follow BAF-1 expression, localization, and dynamics in larval and adult stages, we prepared a construct in which the ORF was fused to the 5′ end of the complete ORF and driven by the promoter (). Microparticle bombardment () was used to create three independent stable transgenic lines expressing the GFP–BAF-1 fusion protein. GFP–BAF-1 localized primarily at the nuclear envelope, with weaker signals in both the nucleoplasm and cytoplasm (), as seen previously for endogenous BAF-1 (). GFP–BAF-1 was expressed ubiquitously throughout development, as seen in gonads (, arrowhead), early embryos (, arrows), late embryos (not depicted), all larval stages (; shows L1), and adults (; adult vulva indicated by an arrow). The localization of GFP–BAF-1 at different stages of the cell cycle was similar to that of endogenous BAF-1 (; and unpublished data) and human BAF (), including its localization at the core region of chromosomes during late anaphase/telophase (unpublished data). The mobility of GFP–BAF-1 was measured by FRAP in worms lacking endogenous BAF-1 (; see the following section). GFP–BAF-1 recovered rapidly with a half-time of 2.24 ± 0.66 s ( = 4), which is somewhat less mobile than human BAF (0.26 s; ). One possible explanation for this difference is that the mobility of human BAF was measured in the presence of endogenous BAF. GFP–BAF-1 was mislocalized in embryos with RNAi–down-regulated Ce-lamin expression (, arrowheads) and associated with anaphase-bridged chromatin (, arrow), which is similar to BAF-1 behavior in embryos down-regulated for both and () and is consistent with the behavior of endogenous BAF-1 during mitosis (). GFP–BAF-1 localized normally in control animals fed with the empty L4440 vector (). Together with its ability to rescue most phenotypes caused by the lack of endogenous BAF-1 (see the following section), we conclude that the expression, localization, and dependence on lamins for its assembly of GFP–BAF-1 were comparable with those of endogenous BAF-1. The VC699 strain contains the allele in which the promoter and ORF are deleted (Fig. S1 A, available at ). This allele was outcrossed three times and balanced with the hT2 balancer, which carries an integrated pharyngeal GFP element. PCR analysis was used to confirm the 766-bp deletion in the allele (Fig. S1 B). mRNA was undetectable by RT-PCR analysis of RNA from 4-d-old homozygous animals (Fig. S1 C). Controls showed that an unrelated transcript encoded by was detected at similar levels in homozygous, heterozygous, and wild-type animals (Fig. S1 C). BAF-1 protein was not detected in Western blots of extracts from 4-d-old animals, whereas their Ce-lamin signal was comparable with wild type (Fig. S1 D). Heterozygous animals were indistinguishable from wild type (N2), with similar body size and normal brood size (unpublished data) despite their ∼40% reduced levels of BAF-1. In contrast, homozygous animals, which apparently used maternally supplied BAF-1 to complete embryogenesis and larval stages, were ∼50% thinner and ∼35% shorter and arrested at late L4/early adult stage with several tissue-specific phenotypes, as detailed in the following sections. The short/thin phenotype could be the result of normal numbers of smaller cells, fewer cells, and/or abnormal gonads. Homozygous animals frequently exploded when touched, suggesting cuticle defects. Worms contain two syncytial rows of seam cells that interrupt the hypodermis and form alae on the cuticle surface during certain developmental stages (). During the L1–3 stages, the epithelial seam cells positioned along each side of the worm proliferate to create two cell types: epidermal daughter cells that fuse to the main body hyp7 hypodermis and ectoblastic daughter cells that remain unfused (; ). During the L4 stage, these unfused ectoblastic seam cells, which extend from the tail to the head, fuse to each other laterally to form two syncytia on each side of the animal that eventually create cuticular structures (; ). To visualize the borders of the ectoblastic seam cells, we crossed the apical junction marker AJM-1–GFP into heterozygous worms () and examined progeny that were homozygous or heterozygous for the deletion. Until the late L3 stage, seam cells in both homozygous and heterozygous worms behaved like wild-type animals, forming one row of ∼12 cells on each side of the worm (unpublished data). In heterozygous animals, the ectoblastic seam cells remained unfused at the late L3/early L4 stage (; arrows), which is similar to wild-type animals (). In contrast, the ectoblastic seam cells of homozygous animals fused prematurely with the epidermis, which was visualized by loss of the GFP signal at the apical borders of the seam cells (; arrows). Differential interference contrast (DIC) analysis revealed that at L3, after ectopic fusion, the number of seam cells remained at ∼12 (unpublished data). By late L4/early adult stage, all seam cells had fused to the epidermis instead of to each other in all tested worms ( = 20; ; arrows). DIC analysis showed that after that stage, the seam cells disappeared. These cells did not stain for SYTO 11 (Invitrogen; ), indicating that their disappearance probably did not involve apoptosis (unpublished data). In addition, DIC analysis revealed no apoptotic bodies. This phenotype was specific for BAF-1 because in animals that expressed both GFP–BAF-1 and AJM-1–GFP transgenes, the seam cells remained unfused until the late L4 stage, as in heterozygous worms, and later fused correctly to each other to form lateral syncytia (; arrow). We conclude that BAF-1 is required to prevent the premature fusion of seam cells to the epidermis. The premature fusion of the seam cells implies that BAF-1 represses the expression of key genes involved in cell fusion. EFF-1 is a cell surface protein that directly mediates most somatic cell fusion events in (; ). Therefore, we tested whether the loss of BAF-1 expression causes the early expression of in ectoblastic daughter seam cells. We generated a strain expressing GFP driven by the promoter () and followed GFP expression in the seam cells of and offspring animals. At the L3 stage, the epidermal daughter cells of the seam cells in worms that fused to the hyp7 hypodermis expressed (, arrow), whereas the ectoblastic daughter seam cells that remained unfused did not show a detectable GFP signal (, arrowhead), which is similar to wild-type animals (). In contrast, the –driven GFP expression was strong (>3.5-fold) in the presumed prematurely fused ectoblastic seam cells of L3-stage animals (, arrow). The ability of BAF-1 to bind DNA and chromatin () and to repress expression (see previous section) suggested that BAF-1 regulates expression by binding to its promoter. To test this hypothesis, we used a GFP antibody and the chromatin immunoprecipitation (ChIP) assay on worms expressing GFP–BAF-1 (; ). We specifically analyzed the binding of GFP–BAF-1 to the promoter () compared with the coding and intronic regions of . A GFP antibody and the ChIP assay on wild-type worms and worms expressing AJM-1–GFP were used as controls for nonspecific binding to the same regions. The binding of BAF-1 to promoter regions spanning from −2,165 to −2,057, −894 to −795, or −289 to −180 (, regions 4, 3, and 1) was 5.3–8.5-fold higher than its binding to coding and intronic regions, 3.7–6.5-fold higher than nonspecific binding to the same region in wild-type animals, and 5.3–8.5-fold higher than its binding to the same region in worms expressing AJM-1–GFP. GFP–BAF-1 bound to two other regions in the promoter (−2,696 to −2,560 and −656 to −530; , regions 5 and 2) to the same extent as nonspecific antibody controls or to GFP control. Furthermore, GFP–BAF-1 did not show increased binding to two regions in the () promoter as compared with either coding and intronic regions of or with wild-type control (). We conclude that BAF-1 represses EFF-1 expression probably by binding directly to the promoter. We next wanted to test whether the premature fusion of seam cells in animals depends only on EFF-1. Therefore, we introduced an –null deletion, (), into animals expressing the apical junction marker AJM-1–GFP and tested AJM-1–GFP expression in and offspring animals. At the L2 stage, the seam cells remained unfused in both heterozygous and homozygous worms for the allele (; arrows). During the L3 stage, the seam cells of double homozygous worms did not migrate but probably fused to the hypodermis (, arrow), and, during the L4 stage, the seam cells became very small and then disappeared (, arrow). These cells were SYTO 11 negative, suggesting that they did not undergo apoptosis (unpublished data). Likewise, DIC analysis could not find apoptotic bodies. In contrast, the seam cells of the L3- or L4-stage heterozygous worms did not fuse and continued to migrate, forming several rows of cells (; arrows; ). We concluded that BAF-1 inhibits premature fusion of the seam cells probably by repressing both EFF-1 fusion–dependent and –independent pathways. AFF-1 fusogen was recently found to also be active in seam cell fusion independently of EFF-1 (). To test whether BAF-1 can bind promoter regions, we used a GFP antibody and the ChIP assay on worms expressing GFP–BAF-1 (). The binding of BAF-1 to promoter regions spanning from −535 to −390 or from −329 to −209 (regions 2 and 1, respectively; ) was 4–4.6-fold higher than its binding to intronic and coding regions. We concluded that BAF-1 inhibits premature fusion of the seam cells to the hypodermis probably by regulating both EFF-1 and AFF-1 pathways (). During the L3 stage, the vulva begins to form in the ventral epidermis when the anchor cell induces three (P5.p, P6.p, and P7.p) of the six vulval precursor cells (VPCs) to divide and to adopt vulval fates (; ). Strikingly, all animals were vulvaless ( = 50; ; normal vulva in the control worm is indicated by an arrow in ). To determine whether the lack of vulva formation is caused by a missing anchor cell, we introduced the anchor cell–specific marker CDH-3–GFP () into worms. During the L3 and L4 stages, CDH-3–GFP expression was detected in the anchor cell in both heterozygous and homozygous animals (). We next followed the VPCs in the offspring of AJM-1–GFP; animals and wild-type worms expressing AJM-1–GFP. In early L3 stage, the six VPCs, P3.p to P8.p, were present in both wild-type and homozygous worms (; arrows). During the L3 and L4 stages, the VPCs continued to divide in wild-type worms and formed a normal vulva structure (; arrow indicates vulva at the late L3 stage). However, in 73% of worms, the second VPC division did not occur, and, in 27% of the worms, only one division occurred, usually that of P6.p (, arrowhead). During the L4 stage, all VPCs degenerated but did not stain for SYTO 11 (unpublished data). The vulvaless phenotype of animals was specific to the loss of BAF-1 expression because vulva formation was fully rescued in strains expressing the GFP–BAF-1 transgene, as seen in L4 larvae (; arrows) and adults (, arrow). We conclude that BAF-1 is required for VPC divisions and vulva formation either at the level of anchor cell signaling, defects in responding cells, or both. Homozygous animals became sterile. The gonads of early L3 animals were similar in size to wild-type gonads, with similar numbers of germ cells compared with or wild-type animals (mean of 32 ± 5 germ cells in worms [ = 12] and 30 ± 9 germ cells in worms [ = 12]). However, during the L4 and adult stages, both gonad size and germ cell numbers were substantially reduced in animals (102 ± 30 germ cells in at L4 [ = 12], 58 ± 4 germ cells in worms at the L4 time stage [ = 12], 212 ± 26 germ cells in young adult [ = 12], and 36 ± 19 germ cells in worms at the young adult time stage [ = 12]; ; DAPI stain) compared with animals (; DAPI stain). Fewer germ cells in adult gonads suggested germline cell proliferation defects, germ cell degeneration, or both. These findings were consistent with essential roles for BAF-1 in cell proliferation, as reported in (), and also with potentially essential roles in meiosis. Gametes were never seen in animals (unpublished data). The few surviving germ cells expressed the germline marker protein matefin/SUN-1 (; ), suggesting that BAF-1 was not essential for early germ cell differentiation. These phenotypes were specific for the loss of BAF-1 expression because animals expressing the GFP–BAF-1 transgene had normal sized gonads and apparently normal numbers of germ cells () plus sperm cells that appeared to be wild type but were less confined to a specific area in the gonads. There were cells at the pachytene stage, but oocytes were not formed (unpublished data). The gonad size phenotype independently supported our previous conclusion based on animals that escaped the lethal consequences of RNAi-mediated down-regulation that BAF-1 is required for gonad development (). The position of the gonads was abnormal in homozygous animals (unpublished data) and closely resembled the abnormal gonad position in animals under conditions of incomplete RNAi of BAF-1 (). To specifically test distal tip cell (DTC) migration, we prepared a strain expressing lag-2–GFP, which expresses in DTC cells, and followed GFP expression at different stages in animals (). At the late L2 and early L3 stages, migration of the two DTCs away from the midbody occurred normally in both and animals (). At the late L3 stage, the ventral to dorsal migration of DTCs was normal in animals () but failed to occur in animals (). The DTCs migrated back toward the midbody in animals but not in animals (unpublished data). The DTCs in animals had normal morphology (, arrow). DTCs migrated normally in animals expressing the GFP–BAF-1 transgene (unpublished data). These results confirmed our previous suggestion that BAF-1 is required for the second and third steps of DTC migration () and further demonstrated that this phenotype is specific to BAF-1. BAF-1 is required to properly localize Ce-lamin at the nuclear envelope of early embryos (). To investigate this potential role for BAF-1 at later stages, we introduced Ce-lamin–GFP () into animals and localized Ce-lamin–GFP in somatic postembryonic nuclei. Until the L4 stage, Ce-lamin–GFP localization was similar to that of wild-type worms (unpublished data). However, afterward, the Ce-lamin–GFP signal coalesced into one to three strong patches at the nuclear periphery of body muscle cells (, inset), epidermal cells, and pharyngeal cells (not depicted). This redistribution was not seen in wild-type worms (, inset) and differed from the bright foci of the Ce-lamin–GFP signal that accompany normal aging in (). In the germ cells (, inset) but not in control germ cells (), the Ce-lamin signal was weak and localized primarily in the nuclear interior. This phenotype was fully rescued in animals expressing the GFP–BAF-1 transgene (, inset). We conclude that BAF-1 is required to organize Ce-lamin in both germline cells and adult somatic nuclei, including muscle nuclei. The disorganization of Ce-lamin in the germ cell nuclei of animals suggested potential gross defects in nuclear architecture. Therefore, we used the thin section transmission electron microscopy method to visualize the nuclear membranes and chromatin of germ cells derived from and control animals (). Gonads of control animals had normal-appearing nuclei (, arrows) and were indistinguishable from wild-type (N2) gonads (not depicted). In contrast, the gonads of animals had only a few large nuclei (, arrowhead), whereas most nuclei were small and lobulated (, arrows). About 65% of these small nuclei ( = 26) had gaps in their nuclear envelope (, arrowheads), and 27% had extra layers of nuclear membranes (, arrow). In some cases, the chromatin was condensed in electron-dense patches (, arrowheads). We speculate that the more drastic phenotype of gapped nuclear envelopes seen in germline cells might be caused by proliferation-linked defects in nuclear assembly. Thus, for germline cells, we conclude that BAF-1 is required not only to organize Ce-lamin but also for nuclear envelope and chromatin organization. Loss of nuclear envelope integrity could account for germline failure, but our results did not distinguish whether this integrity was lost during germ cell mitotic proliferation, meiosis, interphase, or a combination of these events. An unexpected finding was that –null animals had an uncoordinated () phenotype. When grown at 20°C, the movement of animals was similar to that of or wild-type animals up to day 4 (Video 1, available at ). From day 5 onwards, all homozygous animals ( = 146) developed an uncoordinated movement (Video 2 shows homozygous animals at day 5). In addition, a gradually increasing fraction of animals became paralyzed in the mid- and tail regions but not the head (Video 3 shows homozygous animals at day 12). The fraction of animals that could only move their head was 2.4% on day 7, increasing to 21.7% on day 11 and 88.5% on day 18 ( = 130, = 104, and = 76, respectively). Control animals did not develop uncoordinated movement, and the fraction of paralyzed animals was 0%, 0%, and 32.6% on days 7, 11, and 18, respectively ( = 144, = 108, and = 49, respectively). The paralysis phenotype of the controls differed from the animals because the control worms could still move their tail and were classified as class C aging animals, in which worms do not move even after prodding and can only move their head and/or tail or twitch in response to touch (). Both the uncoordinated and paralysis phenotypes of the animals were completely rescued by the GFP–BAF-1 transgene (unpublished data). The Unc and paralyzed phenotypes could be caused either by nerve degeneration or muscle cell–intrinsic deterioration. To test the latter possibility, we used the thin section transmission electron microscopy method to examine the morphology of muscle cells in the head and tail regions of wild-type (N2), heterozygous , and homozygous animals grown at 20°C on days 4, 8, and 12. In both control and homozygous animals, the head muscle tissues were similar to wild-type animals in each age group (). Muscle cells in the tail region of animals had normal morphology in all examined days, including day 12 (). In striking contrast, muscle cell morphology in animals was normal only at day 4 () and deteriorated considerably by days 8 and 12 (). By day 12, the thin and thick filaments in the tail muscles became misorganized and contained dark aggregates (, arrows), which appeared already in day 8 (, arrow). Tail muscle cells in homozygous worms also had deteriorated nuclei (unpublished data). A previous study had shown that bodywall muscle cells are necessary for the normal distribution of myotactin, a protein that maintains the association between the muscle contractile apparatus and hypodermal fibrous organelles (). To further study the muscle deterioration, we immunostained and worms at day 12 with MH46, an antibody against myotactin. Myotactin distribution was normal at the tail and head of (Fig. S2, A and C; available at ) and at the head of worms (Fig. S2 B). However, myotactin distribution at the tail of worms was abnormal (Fig. S2 D). We conclude that BAF-1 is required to maintain the integrity of specific muscles in the body. All phenotypes seen in homozygous –null () animals appeared after the larval L2 stage. These phenotypes affected specific cell types, including seam cells, vulva precursor cells, germ cells, DTCs, and selected muscle cells in the midbody and tail regions. These phenotypes were specific for BAF-1 because a transgene expressing GFP–BAF-1 fully rescued all phenotypes but one: transgenic BAF-1 did not restore the production of oocytes. We speculate that this failure might be caused by insufficient expression of transgenic BAF-1 in germline cells, incomplete activity of the GFP fusion protein, or a second mutation in a gene located near , which was not removed by three to five outcrosses of the strain. Interestingly, heterozygotes with a single copy of , which expressed ∼60% the normal level of BAF-1 protein, were apparently normal with brood sizes similar to wild-type animals (unpublished data). Furthermore, two extra copies of the GFP–BAF-1 transgene (seen in the YG1001-3 strains) also had no apparent effects, although protein levels were not measured. Thus, in contrast to nuclear assembly extracts derived from eggs, may tolerate a wider range of BAF-1 protein levels. The mitotic and chromatin phenotypes seen in () embryos () were seen to a limited extent at later stages of development, specifically in germ cells and in the VPCs, which might reflect a failure to assemble nuclei after mitosis. However, the successful embryonic development of homozygous embryos indicates that maternally contributed BAF-1 is sufficient for early development, when most somatic nuclear divisions occur. In this respect, our findings are consistent with findings in , wherein maternal D-BAF was sufficient to complete all larval stages in flies homozygous for a deletion (). At later stages of development, the D-BAF deletion caused defects in chromatin organization, including clumped heterochromatin, which is similar to our findings for germ cell nuclei of –null . Perhaps there were defects in anaphase at postembryonic cells, but we could not see anaphase bridges in any cells examined either by DAPI staining, thin section EM, or DIC microscopy. The lack of anaphase chromatin bridges in somatic nuclei of post-L2 animals can be explained by the fact that most somatic cells are not dividing or by the activity of checkpoints that block entry into mitosis. Consistent with the first possibility, we saw severe nuclear morphology defects in proliferating germline cells after the L3 stage. Furthermore, the lack of mitotic figures in germline cells was consistent with the activity (in germline cells) of one or more checkpoints that are inactive in embryos (). We cannot rule out that many of the phenotypes of mutants, including the lack of proliferation of germ cells and VPCs and the thin and uncoordinated phenotype, could be the results of defects in overall postembryonic cell divisions. On the other hand, it is likely that most phenotypes are probably caused by the aberrant regulation of BAF-1–regulated genes because these phenotypes were either absent or different in worms containing mutations in the cell cycle genes (; ). In addition, the specific binding of BAF-1 to the promoter confirms that BAF-1 is directly involved in gene regulation, as suggested previously in mammalian retinal cells (). Given the differences in the timing of various developmental stages in and and the uncertain rate of loss of maternally provided BAF-1 protein, we are struck by the general similarities between the null phenotypes of these two organisms. In both organisms, BAF is required for efficient mitosis, chromatin organization, and nuclear envelope formation and also has partners involved in regulating tissue-specific roles during development. These results strongly support the hypothesis that many of BAF's roles are conserved in evolution. Our results suggest that BAF has additional partners involving the regulation of tissue-specific functions that remain to be discovered. In –null animals, the seam cells fused prematurely at the L3 stage. Most known genes involved in fusion and patterning of the epidermis are transcription factors (). Ectopic and premature fusion of seam cells is also seen in worms deleted for /engrailed (), which represses the transcription of a key gene, , encoding a cell surface protein that directly mediates most somatic cell fusion events in (), including the ventral cell fusions required for vulva formation. The EFF-1 protein must be expressed in both cells for fusion to occur (). Mutations in /engrailed derepress and lead to abnormal fusion events during embryogenesis (). /engrailed normally blocks seam cell fusion to the syncytial hypodermis during embryogenesis. BAF is also involved in gene regulation. In mouse retinal cells, BAF inhibits gene expression by binding directly to Otx2, Crx, and other paired-like homeodomain transcription activators and blocks Crx-dependent gene expression in vivo (). Our findings demonstrate that BAF-1 has more extensive roles in tissue-specific gene regulation because BAF-1 was required to prevent the premature fusion of seam cells. Our results suggest that BAF-1 normally represses during embryogenesis and later stages of development by binding to the promoter. It is also worth noting that BAF-1 is currently the only factor known to bind the promoter directly. Although expression was altered in worms, premature seam cell fusion still occurred in worms homozygous for mutations in both and . The seam cells in these worms are smaller and disappear later. Thus, the mutant background may cause additional deleterious effects. We speculate that there are other proteins mediating cell fusion, which are repressed by BAF-1. One such protein is AFF-1, which has fusogenic activities that do not involve EFF-1 activity, especially in anchor cell fusion and fusion between the lateral seam cells (). Our ChIP analysis suggests that BAF-1 also binds the promoter of and, therefore, prevents seam cell fusion by regulating both the and promoters (). One of the most intriguing phenotypes of animals homozygous for a deletion was the accelerated deterioration of specific muscles in aging animals. This suggests a role for BAF-1 in adult muscle integrity. This finding is consistent with the idea that BAF is required to efficiently localize emerin at the nuclear envelope of mammalian cells () and Ce-lamin in cells (this study). The observed deterioration of muscle cells in differed from normal muscle aging () in several ways. First, the –null muscular dystrophy–like phenotype appeared only in midbody and tail muscles (which are innervated by the ventral or dorsal nerve cords), whereas head muscles (which synapse to the nerve ring) remained functional. Consequently, the head muscles often behave differently. During normal aging, suffers from loss of muscle mass and the function of muscles in the body wall and pharynx (). We cannot rule out the possibility that the midbody and tail muscle phenotypes result from problems with the ventral or dorsal nerve cords. However, we disfavor this model because the morphology of the affected muscles was visibly defective on day 8, when the animals could still move, and, therefore, nerve cells were still functional. Furthermore, similar phenotypes of the disorganization of muscle cell filaments and gradual paralysis were previously reported in worms with mutations in muscle-related genes such as (myosin heavy chain; ), (Troponin I; ), or (perlecan; ), which also influence DTC migration (). These genes might be regulated directly or indirectly by BAF-1. In addition, muscle attachment to hypodermal cells was also aberrant in the tail of the animals. Therefore, we hypothesize that loss of function directly disrupts the function of selected muscle cells in . In humans, mutations in A-type lamins cause several forms of muscular dystrophy, including autosomal dominant Emery-Dreifuss muscular dystrophy, Limb-Girdle muscular dystrophy, and dilated cardiomyopathy with conduction system defects (). Mutations in emerin, an inner nuclear membrane LEM domain protein that directly binds to lamin A, cause X-linked recessive Emery-Dreifuss muscular dystrophy, which is clinically indistinguishable from autosomal dominant Emery-Dreifuss muscular dystrophy (). Both proteins (lamin A and emerin) directly interact with BAF (), and these interactions are conserved in : Ce-emerin and Ce-lamin (B type) each bind BAF-1 directly (; ; unpublished data). The selective muscular dystrophy–like phenotype of –null animals strongly suggests a novel role for BAF in muscle cell integrity, potentially at the level of muscle-specific gene regulation. Future work will aim to identify putative BAF-regulated genes in muscles to shed new light on the mechanisms of Emery-Dreifuss muscular dystrophy and other laminopathies. strains were handled as described previously (). Strains N2 (wild type), JK2049, PS3352, PS3729, , and FC121 were obtained from the genome center. Strain VC699 containing the deletion allele of (() III/hT2[] (I;III)) was prepared by the Reverse Genetics Core Facility at the University of British Columbia. It was outcrossed three times before balancing with hT2. Strain VC699 was crossed with the following strains: PS3352, JK2049, jcls, PS3729, PD4810, YG1001, YG1002, FC121, and . The hT2 balancer was reintroduced to the strains by crossing to males. The genotype was determined by single-worm PCR analysis, and GFP expression was assessed by fluorescence microscopy. The three independent GFP-BAF–expressing strains (YG1001, YG1002, and YG1003) were generated by bombarding the construct pYG1001 (). Transmission electron microscopy analysis of was performed as described previously (). DIC and immunofluorescence images were taken either with a CCD camera (Axiocam; Carl Zeiss MicroImaging, Inc.) mounted on a microscope (Axioplan II; Carl Zeiss MicroImaging, Inc.) equipped for fluorescence and DIC or with a confocal scanhead (MRC-1024; Bio-Rad Laboratories) coupled to an inverted microscope (Axiovert 135M; Carl Zeiss MicroImaging, Inc.) equipped with a 63× NA 1.3 oil immersion objective (Carl Zeiss MicroImaging, Inc). For FRAP analysis, worms expressing GFP–BAF-1 were imaged using a confocal microscope (FV-1000; Olympus) equipped with an inverted microscope (IX81; Olympus) and a 60× NA 1.4 oil immersion objective (Olympus). GFP–BAF-1 fluorescence was photobleached by a 405-nm laser in a defined region of each cell and was imaged with a 488-nm laser line for excitation and a 505–525-nm filter for emission before, during, and after the photobleach. For FRAP analysis, fluorescence intensity in the bleached area, the backgroup area, and the total cell area were measured as a function of time after bleaching and were normalized essentially as described previously (). Adult were fixed and stained by indirect immunofluorescence as described previously (). MH46 () was used at a 1:10 dilution. 135 4-d-old young adults were collected in 30 μl M9 buffer, mixed with 15 μl of 2× SDS sample buffer, boiled for 10 min, and subjected to protein blot analysis as described previously (). For extract preparations, N2, PS3729, YG1001, or YG1002 asynchronous population worms grown in six 9-cm plates were collected. Worms were washed twice with M9 and fixed with 2% formaldehyde for 30 min at room temperature, washed once with 100 mM Tris, pH 7.5, twice with M9 buffer, and once with homogenization buffer (50 mM Hepes/KOH, pH 7.5, 1 mM EDTA, 140 mM KCl, 0.5% NP-40, 10% glycerol, and 5 mM DTT with protease inhibitors), and frozen in liquid nitrogen. Worms were sonicated on ice 10 times for 30 s each with a sonicator (Sonic; Heat Systems Ultrasonic, Inc.) and centrifuged at 6,500 rpm for 20 min at 4°C. The supernatant was sonicated again to shear the DNA on ice five times for 30 s each and was centrifuged at 14,000 rpm for 20 min. The supernatant was collected and tested for the presence of GFP–BAF-1 or AJM-1–GFP by immunoblotting and frozen in liquid nitrogen. Lysates were incubated with 5 μg anti-GFP antibody (Roche) for 2 h, and cellular debris was removed by centrifugation at 6,500 rpm for 15 min at 4°C. Lysates were then centrifuged at 14,000 rpm for 10 min, and 50 μl of protein G–Sepharose (Roche) was added to the supernatant. Immunocomplexes were washed twice with each buffer: ChIP buffer (50 mM Hepes/KOH, pH 7.5, 1 mM EDTA, 0.5%, NP-40, and 5 mM DTT with protease inhibitors) with 100 mM KCl and ChIP buffer with 1 M KCl and Tris-EDTA. Complexes were eluted with elution buffer (1% SDS and 10 mM Tris-HCl, pH 8), and 16 μl of 5 M NaCl was added to the elution and heated at 65°C overnight. DNA was then isolated using a standard procedure (phenol-chloroform extraction) and was resuspended in 20 μl Tris-EDTA. The amount of eluted DNA was quantified using locus-specific primers. Quantitative PCR was used to monitor ChIP results. 20 μl of quantitative PCRs contained 1:2 SYBR green Mix (ABgene), 250 nM of each primer, and an appropriate amount of DNA. The quantitative PCR results were analyzed essentially as described previously (). Single-worm PCR analysis using primers 5′-AACCGAAATTCTCAGCCCTT-3′ and 5′-GATCGCGGCCGCCTTAGAAACACTCTTCAGGATCG-3′ to distinguish between wild-type, , and worms (Fig. S1, A and B) was performed essentially as described previously (). For RT-PCR, 100 wild-type (N2), , or worms were collected from each strain in 700 μl of extraction mixture (0.1 M NaOAc, 50% phenol, 2 M guanidinium thiocyanate, 12 mM sodium citrate, pH 7.0, 0.25% Sarkosyl, and 50 mM β-mercaptoethanol), immediately frozen in liquid nitrogen, and incubated at −80°C for at least 20 min. Samples were thawed at 60°C, vortexed, and incubated for 1 min on ice. Total RNA was isolated using a standard procedure (phenol extraction) and digested with RNase-free DNase I (Promega), and cDNA was synthesized from ∼400 ng RNA using a 15-nt oligodT primer using Moloney murine leukemia virus reverse transcriptase (Promega) according to the manufacturer's instructions. From each 20 μl cDNA, 2 μl was analyzed by PCR using the following primers: forward (5′-GATCGAATTCATGTCGACTTCTGTTAAGCATCG-3′), reverse (5′-GATCGCGGCCGCCATGAACTGATCTGCCCACTCG-3′), forward (5′-CACTTCCATTGGGGAGAGAA-3′), and reverse (5′-ACAACGCCTTTCCCTCTTTT-3′). N2, VC699, and -expressing GFP–BAF-1 worms were collected, washed with M9 buffer, treated for 5 min with hypochlorite solution (1.1% hypochlorite and 0.62 M NaOH), and washed with M9, and the embryos were collected and grown on nematode growth medium plates at 16°C for 3 d until they reached the L4/young adult stage. The VC699 and worms that expressed GFP–BAF-1 were sorted into two classes representing the and genotypes based on the presence of GFP fluorescence in the pharynx. For each experiment, 120 worms from each line were transferred to new nematode growth medium plates at 20°C (∼40 worms per plate), and movement was classified essentially as described previously (). Worms were also processed for transmission electron microscopy at days 4, 8, and 12 from synchronized embryos. Fig. S1 describes the allele and provides evidence that is not expressed in L4 worms homozygous for the 4 allele. Fig. S2 shows that muscle attachment to hypodermal cells is aberrant in the tail region of animals by staining animals with MH46 antibody. The supplemental text gives further details on primers that were used in this study. Videos 1–3 show movement of heterozygous and homozygous worms for the 4 allele at days 5 and 12. Online supplemental material is available at .

Cellular c-Src (Src) is a nonreceptor protein tyrosine kinase associated with the plasma membrane, cell–matrix and cell–cell adhesions, and endosomal vesicles. It mediates signaling by a variety of receptors (; ). Constitutively active Src can elicit cell transformation in vitro (; ; ), and Src expression and activity are elevated in many human epithelial cancers (). The first 16 N-terminal residues of Src (residue numbers refer to chicken c-Src) contain an -myristoylation site and a series of basic residues, both required for membrane association (). Residues 17–84 constitute a unique domain, followed by SH3 and SH2 domains connected by a short linker. Another linker connects the SH2 domain to the kinase domain, which is required for most biological functions of Src. Tyr527 (or Tyr530 in human Src) undergoes an inhibitory phosphorylation by C-terminal Src kinase. In the inactive or “closed” form of Src, the SH2 domain interacts with pTyr527, positioning the SH3 domain to interact with a polyproline type II helix in the kinase-SH2 linker region. This causes inactivating conformational changes in the N-lobe of the kinase. Activation can occur as a result of dephosphorylation or mutation of Tyr527 or by binding of the SH2 or SH3 domains to activating ligands (; ; ; ). Src initiates intracellular signaling by binding to protein substrates via its SH3 and SH2 domains, often inducing their processive phosphorylation. For example, Src SH3-dependent binding to p130Cas initiates a tyrosine phosphorylation cascade, in which the initially formed phosphotyrosyl (pTyr) residue binds to the Src SH2 domain, followed by phosphorylation at additional sites (). Thus, Src SH3 and SH2 domains have dual roles, mediating inactivating intramolecular interactions as well as promoting substrate phosphorylation. Membrane association is critical for Src signaling, as indicated by the finding that a mutation eliminating myristoylation blocks cell transformation by activated Src (; ). Other Src-family kinases (SFKs), such as Fyn, Yes, and Lck, are thought to reside in cholesterol-enriched assemblies (commonly referred to as lipid rafts) and caveolae by virtue of their dual acylation (; ; ; ). However, the single myristoyl in Src is not considered a strong raft-targeting signal (), and reports on the raft association of Src were ambiguous (; ; ; ; ; ). Despite the functional importance of Src membrane association, little is known about the dynamics of these interactions, and how different elements (the kinase, SH2, and SH3 domains) control Src membrane targeting and association with substrate proteins. We tackled these questions by studying a series of chicken c-Src mutants fused to GFP, combining FRAP beam-size analysis () to determine their membrane association with biochemical studies to characterize their interactions with pTyr protein targets. Our data demonstrate that wild-type (WT) Src displays lipid-like membrane association, whereas the open-conformation constitutively active Src-Y527F shows a substantial contribution of interactions with slower- diffusing membrane-associated proteins. These interactions depend mutually on Src kinase activity and SH2 domain binding. The results suggest distinct and novel roles for Src kinase activity and SH2 and SH3 domains in regulating Src membrane interactions. To study the roles of Src kinase activity and SH3 and SH2 domains in its interactions with the plasma membrane, we combined biochemical studies with FRAP beam-size analysis in live cells (; ). The latter approach characterizes the membrane interactions of inner-leaflet proteins based on their lateral diffusion rates and on the relative contribution of exchange between membrane and cytosolic pools to their FRAP. These studies used a series of mutants of chicken c-Src with GFP fused at the C terminus (). To investigate whether Src activation affects its distribution between cytosolic and membrane fractions, we transfected COS-7 cells with Src-WT-GFP or constitutively active Src-Y527F-GFP and compared their membrane association by cell fractionation () and FRAP ( and ). Src-WT-GFP and Src-Y527F-GFP displayed a twofold higher percentage in the particulate (membrane) fraction (). The percentage of Src-Y527F-GFP in the membrane was slightly higher, possibly reflecting interactions with membrane-associated protein structures (e.g., focal adhesions) and peripheral membranes (; ); this is in accord with the appearance of Src-Y527F-GFP in more distinct clusters, which partially colocalize with vinculin (). To characterize the interactions of Src-GFP proteins with the plasma membrane, we conducted FRAP studies on live cells expressing Src-WT-GFP and Src-Y527F-GFP. Typical experiments are shown in ; quantitative results on multiple cells using two different laser beam sizes (beam-size analysis) are depicted in . The beam-size analysis () explores membrane interactions of proteins interacting with the inner membrane leaflet, where FRAP can occur not only by lateral diffusion but also by exchange between membrane and cytoplasmic pools. If FRAP occurs solely by lateral diffusion, the characteristic fluorescence recovery time τ ( for recovery) is the characteristic diffusion time, τ, proportional to the area illuminated by the beam (τ = τ = ω/4, where ω is the Gaussian radius of the beam and is the lateral diffusion coefficient; ). When FRAP occurs by exchange, τ is the chemical relaxation time, which is independent of the beam size (). The τ(40×)/τ(63×) ratio expected for the two beam sizes generated using the 40× and 63× objectives is 2.56 (the measured ratio between the illuminated areas) for recovery by pure lateral diffusion, versus 1 for exchange; intermediate values suggest a mixed recovery mode, where the faster process has a higher contribution (). In the current studies, we focused the laser beam on the nonadherent plasma membrane away from potential cell-substrate contacts, although for the coverslip-attached COS cells (few focal adhesions), similar results were obtained on the adherent membranes. FRAP of a free cytoplasmic protein (GFP) was faster than the experimental time scale, ensuring that fast cytoplasmic diffusion does not contribute to the measurement (). In accord with the fractionation experiments, which demonstrated both membrane and cytoplasmic pools, the beam-size analysis yielded τ(40×)/τ(63×) = 1.8 for Src-WT-GFP (FRAP by mixed lateral diffusion and exchange; ). The contribution of exchange precludes an accurate translation of τ to , but can be estimated from τ(63×), because the smaller beam area reduces the characteristic diffusion time τ; as τ for exchange does not depend on the beam size, this results in a higher contribution of lateral diffusion to the recovery with the smaller beam size (). This yields = 0.57 μm/s ( and ) for Src-WT, close to determined by fluorescence correlation spectroscopy () for the dually acylated SFK Lyn-GFP (0.3 μm/s) and to the value we have measured for a lipid probe (DiIC; 1 μm/s) in COS-7 cells (). The membrane association of Src-WT-GFP is therefore mediated mainly by lipid-like interactions, presumably via its myristoyl residue and N-terminal region. On the other hand, constitutively active Src-Y527F-GFP exhibited markedly slower FRAP rates and a minor increase in the τ(40×)/τ(63×) ratio to 2.2 ( and ). An elevated τ ratio suggests that the exchange rate is slowed relative to the lateral diffusion, in line with the slightly higher membrane fraction of Src-Y527F-GFP (). Yet, both τ(63×) and τ(40×) of Src-Y527F were strongly elevated, indicating that both exchange and lateral diffusion are inhibited, with a somewhat stronger effect on the exchange rate. Calculating of Src-Y527F-GFP from τ(63×) yields = 0.20 μm/s ( and ), approximately fivefold slower than the lipid probe. This suggests that interactions with transmembrane (TM) and/or membrane-associated proteins (whose lateral diffusion is slower than lipids) considerably contribute to the membrane association of Src-Y527F. The observed differences between Src-Y527F and Src-WT are not due to effects of Src activity on the membrane, as overexpression of constitutively active human c-Src-Y530F (10-fold excess plasmid DNA over Src-WT-GFP) did not affect the FRAP parameters of Src-WT-GFP (unpublished data). The aforementioned experiments were conducted on cells grown with 10% FCS, conditions that may activate part of the Src-WT population. However, the percentage of activated Src-WT molecules is low, as indicated by the small (although significant; P < 0.02, test) effect of serum starvation on τ(63×) and τ(40×) of Src-WT-GFP (). The slight increase in the FRAP rate of Src-WT under these conditions (to = 0.72 μm/s, as calculated from τ[63×]), together with the lack of effect on the τ values of constitutively active Src-Y527F-GFP (unpublished data), suggests that only a minor fraction of Src-WT is activated in cells grown continuously in the presence of serum. Accordingly, all the FRAP parameters of Src-WT in the presence of serum were similar to those of the kinase-dead Src-K295M mutant (P > 0.2). Therefore, as serum starvation may induce apoptosis, further comparisons between Src-GFP mutants were conducted on unstarved cells. Importantly, activation of Src-WT-GFP in serum-starved cells by PDGF significantly increased the τ values of Src-WT-GFP, suggesting a retardation in the FRAP kinetics (). This effect is in the same direction as that induced by the constitutively activating mutation (Y527F) but is considerably milder; this is not surprising considering the modest activation of Src-WT in PDGF-treated cells (; ). To explore the roles of Src kinase activity and SH2 and SH3 domains in the membrane association of Src-Y527F, we used FRAP beam-size analysis to investigate a series of Src-GFP mutants (). Inactivating Src kinase by a second mutation in Src-Y527F (Src-Y527F/K295M) strongly reduced τ(63×) and τ(40×), bringing them close to those of Src-WT (). Analogous effects followed SH2 inactivation (Src-Y527F/R175A) or deletion of the kinase domain (Src-251). Calculation of from τ(63×) shows that these mutants revert to lipid-like diffusion ( ∼0.5 μm/s). The triple mutant Src-Y527F/K295M/R175A, which lacks both kinase activity and SH2 binding, exhibited somewhat faster FRAP rates than Src-Y527F/K295M or Src-251 ( = 0.72 μm/s); the latter two possess an exposed SH2 domain that can bind pTyr residues generated by endogenous SFKs. In contrast, SH3 domain inactivation (Src-Y527F/W118A) did not enhance the FRAP rates and mildly increased τ(63×) and τ(40×) (). The further retardation ( = 0.13 μm/s; calculated from τ[63×]) suggests enhanced interactions with slower-diffusing membrane proteins. Collectively, these findings suggest that the mobility-restricting membrane interactions of Src-Y527F depend on its kinase activity and SH2 binding, whereas SH3 binding is dispensable. This does not imply that the SH3 domain cannot bind Src target proteins (); however, the generally lower binding affinities of Src SH3 relative to SH2 (; ) may render SH3 binding ineffective in retarding the diffusion of Src-Y527F. Moreover, the conformational change mediated by the SH3-inactivating mutation enhances Src kinase activity (; ; ), potentially compensating for the loss of SH3 binding by generating more pTyr residues that can enhance Src binding via the SH2 domain. This could lead to the additional increase in τ(63×) and τ(40×) observed for Src-Y527F/W118A (). The SH2 mutation (R175A), in contrast, does not affect the catalytic activity of Src-Y527F because the intramolecular interaction between the SH2 domain and the C-terminal region is already disrupted by the Y527F mutation (). All Src-GFP mutants except Src-Y527F/W118A exhibited τ ratios not significantly different from the 2.56 ± 0.3 ratio expected for recovery by lateral diffusion, suggesting low contribution of exchange. This contrasts with Src-WT, whose τ ratio was significantly below 2.56 (P < 0.005; ). Thus, shifting to an open conformation may have a minor effect on the insertion of the N-terminal anchor of Src in the membrane. Unlike Src-Y527F, the τ ratio of Src-Y527F/W118A was significantly lower than 2.56 (P < 0.005; ), suggesting a higher relative contribution of exchange. A simple explanation may be that SH3 domain binding (lost in this mutant) retards mainly the exchange kinetics. However, because the τ ratio is sensitive to the relative rate of exchange versus lateral diffusion, an alternative explanation is that the retardation of exchange by SH3 binding is negligible, and the higher contribution of exchange is due to the slower diffusion mediated by the SH3-inactivating mutation. To study the interactions of Src with its target proteins, we constructed GST fusion proteins with specific Src domains; the fusion proteins were coupled to glutathione–Sepharose beads and used to precipitate Src binding proteins from cells expressing different Src-GFP mutants (). The precipitated proteins were quantified by Western blotting using anti-pTyr antibodies. The specificity of precipitation by such fusion proteins, established in many previous studies, was confirmed by a comparison of the pTyr proteins in lysates and in precipitates, by the inability of GST alone to precipitate pTyr proteins, and by the ability of free pTyr to compete with precipitation by the GST-SH2 fusion (Fig. S1, available at ). Comparison between lysates of cells transfected by different Src-GFP mutants measures the abundance of pTyr proteins generated in vivo by each Src mutant. The precipitated pTyr proteins were generated mainly by the expressed Src-GFP proteins, as suggested by the low pTyr levels in cells transfected with GFP or with Src-WT-GFP (). The results demonstrate a requirement for Src kinase activity (compare cells expressing Src-Y527F with K295M mutants). The SH2-inactivating mutation (R175A) reduced the levels of precipitated pTyr proteins to about half of the level generated by Src-Y527F, consistent with previous findings on SH2-dependent multisite processive phosphorylation (; ), consistent with the FRAP results () showing a mutual dependence on Src kinase activity and SH2 domain. On the other hand, Src-Y527F/W118A (inactive SH3) generated amounts of pTyr proteins similar to those of Src-Y527F (), in line with the reports that SH3 inactivation increases Src kinase activity (; ; ) and thus can compensate for loss of SH3 binding to substrates. The high pTyr levels generated by this mutant provide multiple pTyr sites that it can bind via its SH2 domain, providing increased interactions that can further retard the recovery (). In the affinity precipitation studies, the lysate from cells transfected by a given Src-GFP mutant was divided into aliquots, each precipitated by beads coupled to a specific GST-Src domain. The highest pTyr level was in cells transfected by Src-Y527F; GST-SH2 and GST-SH3 precipitated similar amounts of pTyr proteins (, third histogram). GST-SH3/SH2 was twice as effective, whereas inactivating SH3 or SH2 compromised the pull-down capacity. This distribution was retained in cells transfected by all the Src mutants, except Src-Y527F/W118A. The pTyr proteins generated by the latter mutant are precipitated more effectively by GST-SH2 (, compare the blue bars). This may reflect preferential phosphorylation of target proteins to which the SH3-inactive Src mutant binds via its SH2 domain. We next examined the ability of paxillin, a peripheral membrane-associated Src substrate, to compete for binding to activated Src. We chose to overexpress paxillin, because although it is found in focal adhesions, it has a large cytoplasmic pool (see ) and exhibits fast turnover between the pools (; ). Cells were cotransfected with Src-GFP and an excess of paxillin-DsRed2 and subjected to FRAP beam-size analysis (). None of the FRAP parameters of Src-WT were significantly altered, in line with the low level of pTyr proteins in cells expressing Src-WT-GFP. On the other hand, τ(63×), τ(40×), and the τ ratio of Src-Y527F were strongly reduced to values resembling those of Src-WT (). Thus, excess paxillin appears to compete with the mobility- restricting binding of Src-Y527F to membrane protein targets, leaving Src-Y527F anchored in the membrane mainly by lipid-like interactions. This interpretation is supported by the finding that overexpression of paxillin-DsRed2 reduces the coimmunoprecipitation of Src-Y527F with FAK, a known Src substrate (Fig. S2, available at ). Moreover, mutation of the tyrosine phosphorylation sites on paxillin (paxillin-Y31F/W118F; ) significantly compromised the ability of the mutated paxillin-Y31F/Y118F-DsRed2 to affect the FRAP of Src-Y527F (). The incomplete loss of the effect is most likely due to the fact that paxillin contains not only SH2 but also SH3 binding sites (; ), which can contribute to Src binding via its SH3 domain. This notion is supported by the demonstration () that inactivation of the SH3 domain in Src-Y527F/W118A, which retains an active SH2 domain, results in a small but distinct reduction in the ability of paxillin to affect the FRAP parameters of this Src mutant; this is shown by the lesser effect of overexpressed paxillin-DsRed2 on the τ values of Src-Y527F/W118A as compared with Src-Y527F () and the insignificant effect on its τ(40×)/τ(63×) ratio (). The τ values of all the other Src-Y527F double mutants, which contain either kinase-dead or SH2-inactivating mutations, were only weakly affected by paxillin overexpression (), in line with their mainly lipid-like interactions with the membrane even in the absence of paxillin overexpression. The high sensitivity of Src-Y527F and Src-Y527F/W118A, which are the most effective in generating pTyr proteins and possess SH2 binding, to competition by paxillin is consistent with the major role of Src kinase activity and SH2 binding in the mobility-restricting interactions with membrane-associated proteins. In accord with the requirement for Src kinase activity, kinase-dead Src-Y527F/K295M, which exhibited lipid-like membrane interactions to begin with, did not show a significant modulation in τ(63×), suggesting that is not altered, upon expression of paxillin-DsRed2 (). However, its τ(40×) and τ(40×)/τ(63×) ratio were significantly reduced (), suggesting a higher contribution of exchange. This may reflect an ability of this mutant to bind paxillin via its SH3 domain, with a potential contribution of SH2 binding to pTyr-paxillin-DsRed2 generated to some degree by endogenous SFKs. The FRAP rates of Src-Y527F/R175A (inactivated SH2 domain) were only weakly affected (), consistent with SH3 binding to target proteins being much less effective than SH2 binding both in mediating mobility-restricting interactions and in competing for them. We next examined the effects of paxillin-DsRed2 on the membrane-cytosol distribution of Src-WT and Src-Y527F (). Overexpression of paxillin-DsRed2 had no significant effects on the percentage of Src-WT-GFP in either fraction and induced a modest increase in the pellet fraction of Src-Y527F-GFP. This indicates that paxillin competition shifts Src-Y527F to membrane interactions that are mainly lipid-like, resembling Src-WT (). Yet, these interactions are sufficient to retain a high percentage in the particulate fraction (as observed also for Src-WT). At the same time, the ability of overexpressed paxillin to reverse the retarding interactions of activated Src with membrane proteins suggests that paxillin itself does not associate tightly with the membrane and is mainly cytoplasmic in these cells. We therefore combined subcellular fractionation with paxillin immunoprecipitation to investigate the distribution of paxillin-DsRed2, endogenous paxillin, and their pTyr levels in cells transfected with Src-WT-GFP or Src-Y527F-GFP alone or together with paxillin-DsRed2 (). A higher percentage of paxillin and paxillin-DsRed2 was in the cytosolic fraction in cells expressing Src-WT (84% of the total paxillin and paxillin-DsRed2) and Src-Y527F (82%; , middle). pTyr-paxillin and pTyr-paxillin-DsRed2 had a similar distribution (76–81% cytosolic) in these cells. Importantly, expression of Src-Y527F-GFP as compared with Src-WT-GFP significantly increased the levels of all pTyr-paxillin proteins (P < 0.001; ). Thus, although a fraction of paxillin is membrane associated, a major fraction is cytoplasmic, and both are subject to tyrosine phosphorylation by the coexpressed Src-Y527F-GFP. This is in accord with the ability of paxillin-DsRed2 to compete with membrane-associated Src substrates for binding Src-Y527F, and with the partial colocalization observed by confocal microscopy between the subpopulation of paxillin-DsRed2 found in clusters and Src-Y527F-GFP but not Src-WT-GFP (). The studies reported here indicate that Src association with the plasma membrane is dynamic and that there are major differences between the membrane interactions of Src-WT (closed conformation) and Src-Y527F (activated open conformation). We demonstrate that these differences are mediated by transient binding of Src-Y527F to target proteins in the membrane, which depend on Src kinase activity and its SH2 domain, and propose a mechanism for the regulation of Src membrane association by its specific domains. A major fraction of Src-WT-GFP and Src-Y527F-GFP is membrane associated, but they both display significant cytosolic fractions (). This is in accord with a recent study on the SFK Lyn-GFP, which detected a fast-diffusing cytoplasmic population both before and after stimulation (). Although the membrane fraction of Src-Y527F is only slightly higher then that of Src-WT (), their association with the membrane is very different, as indicated by the FRAP studies ( and ). The lateral diffusion of Src-WT in the plasma membrane ( ∼0.6 μm/s) is close to the values measured for a lipid probe (see Results) and for the membrane-associated population of Lyn-GFP (), suggesting that, at steady state, the large majority of Src-WT has no considerable interactions with slower-diffusing TM proteins. The Src N-terminal domain contains an -myristoyl anchor and a polybasic cluster (), resembling the C-terminal farnesylated lipid anchor domain of K-Ras (). In contrast, the lateral diffusion of Src-Y527F is threefold slower, implying that it is retarded by interactions with TM and/or membrane-associated proteins ( and ). The biological relevance of this difference is supported by the observation that PDGF, an established activator of Src (; ), induces a significant retardation in the FRAP kinetics of Src-WT-GFP (). This effect is smaller than the mobility retardation of the constitutively active Src-Y527F-GFP, presumably because only a fraction of the Src-WT-GFP molecules are activated by PDGF. This is in line with the modest activation of Src-WT after PDGF stimulation (; ). The threefold slower FRAP kinetics of Src-Y527F indicate that its mobility-retarding interactions are transient (dynamic), as stable binding to target membrane proteins, which typically diffuse 10–100-fold slower than lipids (; ), would inhibit to a much higher degree. In line with these findings, only a mild retardation of was observed also for Lyn after antigen stimulation of IgE receptors () and for Lck after cross-linking of T cell receptors (). Moreover, because the mobility-restricting interactions of Src are largely mediated by its SH2 domain ( and ; discussed in the following paragraph), their dynamic nature is in accord with the report that high-affinity binding of SH2 domains to pTyr peptides is accompanied by rapid dissociation and exchange kinetics (; ). The interactions of Src-Y527F with membrane proteins retard its exchange rate to an even higher degree than its lateral diffusion, as suggested by the mild increase (from 1.8 to 2.2) in the τ(40×)/τ(63×) ratio (; see Results). This correlates with the minor increase in Src-Y527F membrane association relative to Src-WT (). To identify the Src domains responsible for the mobility-retarding interactions of constitutively active Src, we determined the effects of second mutations in Src-Y527F on the FRAP parameters (). The mobility retardation required both an intact SH2 domain and Src kinase activity, as shown by the shift to lipid-like diffusion (resembling Src-WT) after functional inactivation of the SH2 domain, Src kinase inactivation, or deletion of the entire catalytic segment. The SH3 domain has only a minor contribution to the mobility-restricting interactions, as its inactivation (Src-Y527F/W118A) did not eliminate the mobility retardation, which was even modestly enhanced (). These observations suggest that the mobility-restricting interactions of activated Src are mediated mainly by the binding of SH2 domain to pTyr residues in membrane-associated proteins. Formally, we cannot exclude the possibility that some of these pTyr residues are phosphorylated by Src-activated tyrosine kinases rather then directly by Src. However, studies with modified forms of Src that accept chemically modified nucleotides indicated that many pTyr proteins in Src-transfected cells are indeed the result of direct phosphorylation by Src (). Thus, it appears that the modulation in Src-Y527F FRAP kinetics is mediated largely by its binding to membrane-associated protein substrates that it has itself directly phosphorylated (). The effects of the aforementioned mutations on Src-Y527F FRAP parameters were paralleled by their effects on the abundance of pTyr proteins that can be recognized in vitro by the Src noncatalytic domains (). In particular, inactivation of the catalytic domain reduced the abundance of these proteins below the level observed in cells overexpressing Src-WT, and inactivation of the SH2 domain strongly reduced their level. In vitro the Src SH3 domain can bind to Src substrates (). However, in agreement with the dispensability of the SH3 domain for the mobility retardation of Src-Y527F, its inactivation in vivo hardly affected the abundance of the pTyr proteins (). The failure of the SH3 mutation to reverse the mobility inhibition indicates that in vivo the interactions of SH3 with target proteins, which are weaker than those of SH2 domains (; ), are too weak and labile to strongly retard the dynamic parameters. In fact, the SH3 mutation (Src-Y527F/W118A) was not simply ineffective but induced a further modest retardation in Src mobility. This can be explained by two opposing effects of SH3 inactivation. First, it can increase Src kinase activity by decreasing the inhibitory intramolecular interactions of the SH3 domain with SH2 kinase linker (; ; ). On the other hand, it decreases intermolecular interactions with SH3 binding domains in target proteins. These effects largely cancel out, but the modest additional mobility retardation after SH3 inactivation suggests that the increase in the activation of Src kinase has a higher contribution to the mobility-restricting interactions. This is in line with our conclusion that these interactions depend primarily on SH2-pTyr binding. These findings reflect what is known about the affinities of SH2 and SH3 domains. The affinity of SH2 domains to pTyr peptides is between 100 nM and 1 μM, whereas SH3 domains have weaker (micromolar) affinities for their target sites (; ; ). A model for processive phosphorylation by Src was proposed based on studies with defined Src target proteins (; ). In this model, the relatively weak binding via the Src SH3 domain promotes initial phosphorylation of the substrate protein, leading to binding of SH2 to the resulting pTyr residue, stabilization of Src substrate interaction, and further substrate phosphorylation (). Here, we expand this model to explain the roles of Src domains in the dynamic regulation of its association with membrane sites (). We propose that Src-WT is dynamically associated with the inner membrane leaflet through lipid-like interactions (). Activated Src retains the lipid anchor interactions, but in addition, binds transiently to TM and peripheral membrane protein substrates, inhibiting its lateral diffusion and exchange. The Src SH2 domain promotes multisite processive phosphorylation (); this generates a positive feedback loop, with the SH2 domain successively promoting Src substrate binding and further phosphorylation. Although weak SH3 domain interactions may promote initial binding to substrates (; ), it is apparent that in cells expressing Src-Y527F, both the FRAP kinetics of active Src and the overall abundance of pTyr proteins are primarily determined by the Src SH2 domain and its interactions with pTyr sites. Activated Src binds not only to TM proteins but also to peripheral protein substrates, which may possess a considerable cytoplasmic population. According to the model (), overexpression of such a protein should compete with the binding of activated Src to target membrane proteins. Indeed, overexpression of paxillin-DsRed2, which has a high cytosolic fraction similar to that of endogenous paxillin (), selectively interfered with the mobility retardation of Src-GFP mutants. Only Src mutants with both active kinase and intact SH2 domain were highly affected by paxillin (), in line with the major role of Src kinase and SH2 domain in the interactions with membrane proteins. However, although Src-Y527F reverted to lipid-like membrane association, Src-Y527F/W118A (inactivated SH3 domain) retained some mobility-retarding interactions in the presence of overexpressed paxillin, suggesting less effective competition. Because paxillin contains both SH2 and SH3 binding sites (; ), this suggests that, at least under overexpression conditions, SH3 binding partially contributes to the competition by paxillin. It should be noted that competition by paxillin did not increase the cytoplasmic fraction of Src-Y527F, in line with the retained lipid interactions of the Src lipid anchor, which effectively targets Src-WT to the membrane (). The membrane interactions of Src–paxillin complexes may be somewhat enhanced, as indicated by the small increase in the membrane fraction of Src-Y527F after paxillin overexpression. This may reflect a contribution by the complexed paxillin to the targeting of Src to specific structures/clusters, as shown by their partial colocalization in the cells (). The behavior of activated Src may have general implications for signaling by SFKs activated by physiological signals. Thus, the SFK Lyn was found both in the plasma membrane and the cytoplasm, and its lateral diffusion rate in the membrane was retarded after IgE receptor stimulation (). The lateral diffusion of Lck, another SFK, was also retarded in response to cross-linking of T cell receptors (). In addition, recent findings support the idea that enhanced formation of Src complexes with targets such as integrins and receptor tyrosine kinases is important for tumorigenesis (). Thus, modulation of Src membrane interactions by binding to membrane-associated target proteins is relevant to its biological functions, including cell migration and tumorigenesis. Further characterization of these complexes by both biophysical and biochemical approaches will therefore be of considerable interest. Murine anti-vinculin (clone hVin-1) and anti-lactate dehydrogenase (clone HH-17) ascites were obtained from Sigma-Aldrich. Murine anti-paxillin and anti-FAK were obtained from BD Biosciences. Rabbit anti-GFP and anti- caveolin1 were obtained from Santa Cruz Biotechnology, Inc. Murine anti-pTyr mAb (clone 4G10) was obtained from Upstate Biotechnology. Cy3 goat anti–mouse IgG and rabbit anti–mouse IgG were obtained from Jackson ImmunoResearch Laboratories. Goat anti–mouse IgG conjugated to IRDye 800 was obtained from Rockland Immunochemicals, and Alexa 680 goat anti–rabbit IgG was obtained from Invitrogen. -phospho--tyrosine was obtained from Sigma-Aldrich. Expression vectors for Src-WT-GFP, Src-Y527F-GFP (both in pEGFP-N1), and Src-251-GFP (in pBabePuro) (; ) were donated by M. Frame (The Beatson Institute for Cancer Research, Glasgow, Scotland). Additional Src-GFP mutants () were generated by site-directed mutagenesis using QuikChange (Stratagene). The mutagenic primers used are given in Table S1 (available at ). Constitutively active human c-Src-Y530F (prepared from molecularly cloned human c-Src; ) in pCI vector was a gift from D.J. Fujita (University of Calgary, Alberta, Canada). Chicken paxillin-DsRed2 in pDsRed2 () was donated by A.F. Horwitz (University of Virginia, Charlottesville, VA). This vector was used as a template to generate the paxillin-Y31F/Y118F-DsRed2 double mutant by site-directed mutagenesis (QuikChange), using primers 5′-GAGGAAACGCCTTCCTACCCAACTGG-3′ (forward) and 5′-CCAGTTGGGTAGGAAGGCGTTTCCTC-3′ (reverse) for the Y31F mutation (underlined); the Y118F mutation was added to this mutant using primers 5′-GTGAGGAGGAACACGTGAGCTTCCC-3′ (forward) and 5′-GGGAAGCTCACGTGTTCCTCCTCAC-3′ (reverse). COS-7 cells (American Type Culture Collection) were grown in DME containing 10% FCS as described previously (). For FRAP and confocal microscopy, COS-7 cells grown on glass coverslips were transfected using DEAE-dextran () with plasmid DNA encoding one of the Src-GFP derivatives alone or together with a 10-fold excess of paxillin-DsRed2 or empty pDsRed2. For biochemical studies, the procedure was similar, except that cells were grown in 10-cm dishes and transfected using Lipofectamine 2000 (Invitrogen). COS-7 cells expressing Src-GFP proteins were taken for FRAP studies 24–48 h after transfection. Measurements were in HBSS supplemented with 20 mM Hepes, pH 7.2. In some experiments, cells were first subjected to starvation in serum-free DME (16–20 h), followed by a 10-min incubation at 37°C with or without 50 ng/ml PDGF (rhPDGF-BB; R & D Systems). To minimize internalization, FRAP measurements were at 22°C, replacing samples within 15 min. An argon ion laser beam (Innova 70C; Coherent) was focused through a fluorescence microscope (Universal; Carl Zeiss MicroImaging, Inc.) to a Gaussian spot of 0.85 ± 0.02 μm (63× oil-immersion objective) or 1.36 ± 0.04 μm (40× objective), and experiments were conducted with each beam size (beam-size analysis; ; ). The ratio between the illuminated areas was 2.56 ± 0.30 ( = 39). After a brief measurement at monitoring intensity (488 nm, 1 μW), a 5-mW pulse (5–10 ms) bleached 60–75% of the fluorescence in the spot, and recovery was followed by the monitoring beam. The apparent characteristic fluorescence recovery time τ and the mobile fraction () were extracted from the FRAP curves by nonlinear regression analysis, fitting to a lateral diffusion process (). Transfected cells were fixed with 4% PFA in PBS. For antibody labeling, they were permeabilized with 0.2% Triton X-100 and immunostained as detailed in the specific figure legend. The cells were mounted with Prolong Antifade (Invitrogen). Fluorescence images were collected on a confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.) with plan-apochromat 100×/1.4 objective using LSM 510 (version 3.2 sp2) software and exported to Photoshop (Adobe). COS-7 cells were grown and transfected with Src-GFP vectors (alone or together with paxillin-DsRed2) as described. After 24 h, the cells were suspended in hypotonic buffer (10 mM Hepes, pH 7.4, 10 mM KCl, 3 mM MgCl, 1 mM EDTA, 20 mM NaF, and 10 mM β-glycerophosphate) containing protease inhibitors (174 μg/ml PMSF, 1.5 μg/ml benzamidine, 1 μg/ml phenanthroline, 0.5 μg/ml antipain, 0.5 μg/ml leupeptin, 0.5 μg/ml pepstatin, 0.5 μg/ml aprotinin, and 0.5 μg/ml chymostatin) and subjected to Dounce homogenization. After removal of nuclei and cell debris by low-speed centrifugation, particulate and soluble fractions were separated by centrifugation (106,000 , 1 h, 4°C). Particulate fractions were solubilized in a high-salt buffer (50 mM Hepes, pH 7.4, 250 mM KCl, 1.0% NP-40, 1 mM EDTA, 20 mM NaF, 10 mM β-glycerophosphate, and 5% glycerol) containing the protease inhibitors; 4× concentrated high-salt buffer was added to the soluble fractions (3:1, vol/vol). Equal proportions of the particulate and soluble fractions (5%, vol/vol) were analyzed by SDS-PAGE and immunoblotting. Blots were quantified using an imaging system (Odyssey IR; LI-COR) as detailed in the Immunoblotting section. To generate GST-Src domain DNA constructs, the DNA sequences encoding each domain were amplified by PCR and cloned into the pGEX-KG vector (). For plasmids encoding GST-SH3 (amino acids 79–147 of chicken c-Src), GST-SH2 (amino acids 148–251), and GST-SH3/SH2 (amino acids 79–251), Src-Y527F-GFP in pEGFP served as template. Plasmids encoding GST-SH3*/SH2 (containing a W118A SH3-inactivating mutation) and GST-SH3/SH2* (containing an R175A SH2-inactivating mutation) were generated similarly, except that Src double mutants in pEGFP (Src-Y527F/W118A-GFP and Src-Y527F/R175A-GFP, respectively) served as template. The primers used are detailed in Table S2 (available at ). Forward and reverse primers contained a BamHI site and a HindIII site, respectively (Table S2). After digestion of the PCR product and the pGEX-KG vector with these enzymes, they were ligated to generate the pGEX-Src domain vectors. All constructs were confirmed by sequencing. Bacteria transformed by the above plasmids were grown, sonicated, and lysed as described previously (). The lysate from 8 ml bacteria was rocked (1 h; 4°C) with 120 μl (packed volume) of glutathione–Sepharose 4B beads (GE Healthcare). The beads were pelleted, washed five times, and stored for short periods on ice in PBS containing 0.25% Tween 20 and the protease inhibitor cocktail described in the Cell fractionation section. At 48–72 h after transfection, COS-7 cells expressing specific Src-GFP proteins were lysed in NP-40 lysis buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 5 mM EDTA, 1% NP-40, 50 mM NaF, and 200 μM NaVO) containing the protease inhibitors. Lysates were clarified by centrifugation, and aliquots (0.5 mg protein) were incubated for 1.5 h at 4°C with glutathione–Sepharose beads bearing ∼30 μg of GST fusion protein (GST, GST-SH3, GST-SH2, GST-SH3/SH2, GST-SH3*/SH2, or GST-SH3/SH2*; described in Table S2). After four washes in lysis buffer, proteins bound to the beads were solubilized by boiling in SDS sample buffer, resolved by 8.5% SDS-PAGE, and analyzed by immunoblotting using anti-pTyr antibodies. COS-7 cells cotransfected by Src-WT-GFP or Src-Y527F-GFP and paxillin-DsRed2 (or empty vector) were fractionated as described. 5% (vol/vol) of each fraction (soluble and particulate) in the high-salt buffer was taken to determine Src-GFP levels. The remaining 95% was immunoprecipitated by 1.8 μg mouse anti-paxillin essentially as described previously (). Immunoprecipitation of endogenous FAK was performed similarly on whole cell lysates using 2 μg mouse anti-FAK mAb. The beads were washed three times with the high-salt buffer. Bound material was solubilized in SDS sample buffer, resolved by SDS-PAGE, and immunoblotted with anti-pTyr or anti-paxillin antibodies. Blots were blocked in Odyssey blocking buffer (OBB; LI-COR) diluted 1:1 with PBS and incubated (12 h; 4°C) with primary antibodies (1 μg/ml anti-pTyr, 0.4 μg/ml anti-GFP, 0.25 μg/ml anti-paxillin, 0.25 μg/ml anti-FAK, and 1:500 dilution of anti-lactate dehydrogenase ascites or 0.2 μg/ml anti-caveolin1) in OBB containing 0.2% Tween 20. They were then incubated (1 h, 22°C) with secondary antibodies (0.1 μg/ml goat anti–mouse or 0.2 μg/ml goat anti–rabbit IgG coupled to IR dyes) in OBB containing 0.2% Tween 20 and 0.01% SDS. The blots were quantified using the Odyssey IR imaging system (LI-COR). Table S1 lists the primers used to generate double and triple chicken c-Src mutants. Table S2 depicts the pGEX-KG plasmids encoding GST-Src domain fusion proteins and the primers used to generate them. Fig. S1 shows that GST-Src domain fusion proteins specifically precipitate pTyr proteins from lysates of cells expressing Src-GFP mutants. Fig. S2 shows that overexpression of paxillin-DsRed2 reduces the coimmunoprecipitation of Src-Y527F with FAK. Online supplemental material is available at .

Dendritic spines are small, actin-rich protrusions scattered along dendrites, and they carry the postsynaptic components of >90% of excitatory synapses. Despite of being tiny in size, spines are highly dynamic structures that continuously undergo changes in shape and size over time (; ). A multitude of molecules has been implicated in dendritic spine development and remodeling. Among these, the neurotransmitter receptors, especially -methyl--aspartic acid (NMDA) receptor (NR) and α-amino-3-hydroxy-5-methylisoxazole-propionic acid (AMPA) receptor (GluR), which are well-known inducers of spine formation, regulate spine maturation and stabilization via calcium-dependent regulation of filamentous actin turnover (; ). Besides, other cell surface molecules also influence spine properties in response to external signals by mediating cell adhesion and regulating the networks of interconnected signaling pathways, which converge to regulate actin dynamics in spines (; ). Cell adhesion molecules (CAMs) and ECM molecules are instrumental in providing physical connections and generating cellular signaling events. Importantly, several recent studies have suggested the involvement of CAMs and ECMs in dendritic spine remodeling () and synaptic plasticity (; ). These include N-cadherin (), syndecan-2 (), neural CAM (), integrins (), laminins (), and reelin (). Intercellular adhesion molecule-5 (ICAM-5; Telencephalin) belongs to the Ig superfamily (; ). It is specifically expressed in the postnatal excitatory neuronal cell bodies, dendritic shafts, and dendritic filopodia of the telencephalon (; ). The expression of ICAM-5 temporally parallels dendritogenesis and synaptogenesis (). In agreement, ICAM-5 has been shown to promote dendritic elongation and branching of hippocampal neurons in vitro (). ICAM-5–deficient mice exhibited decreased density of dendritic filopodia, accelerated maturation of dendritic spines (), and changes in long-term potentiation (LTP) in the hippocampus (). Soluble ICAM-5 (sICAM-5) has been detected in physiological fluids under several pathological conditions (; ; ). However, the nature of candidate proteinases and the physiological meaning of the proteolytic cleavage of membrane-bound ICAM-5 have remained unknown. Matrix metalloproteinases (MMPs) form a large family of mostly secreted, zinc-dependent endopeptidases, which are important for the regulation of cellular behavior through proteolytic cleavage of ECMs and cell surface proteins (). Although a large body of data has connected MMPs to brain injury and pathology, accumulating evidence has extended their roles into the normal physiological functions of the brain (; ; ). Among MMPs, MMP-2 and -9 are most abundantly expressed in the developing brain. MMP-2 is found mainly in astrocytes, whereas MMP-9 is highly expressed in neuronal cell bodies and dendrites (; ). The expression and activity of MMP-9 have been shown to depend on NR activation and LTP (; ). Growing data also suggest the association of MMP-9 (; ) and other ECM-degrading enzymes (; ; ) with dendritic spine remodeling, synaptic plasticity, learning, and memory formation. Although several ECMs and cell surface proteins, which play important roles in the aforementioned functions, have been identified as MMP substrates, the target molecules for MMP-2 or -9 in the brain have remained largely elusive. In the present study, we examined the possibility of ICAM-5 acting as a substrate for MMP-2 or -9, using a variety of experimental approaches, and investigated the role of MMP-mediated ICAM-5 proteolytic cleavage in the regulation of dendritic spine development. Because sICAM-5 has been detected under various pathological conditions, we studied whether ICAM-5 cleavage takes place during physiological neuronal maturation. For this purpose, we examined the release of sICAM-5 from cultured hippocampal neurons at different developmental stages in vitro (3–21 d in vitro [DIV]). The expression of full-length ICAM-5 of 130 kD was low in the 3-DIV hippocampal neurons, but starting from 7 DIV, when dendrites extensively develop in hippocampal neurons, the expression of full-length ICAM-5 dramatically increased and reached a plateau thereafter (, right). In comparison with this, sICAM-5 of 85–110 kD was most strongly released at 14–21 DIV (, left), which parallels the period of dendritic spine maturation and synaptic formation. This indicates that the cleavage of membrane-bound ICAM-5 may play an important role during spine maturation. The NRs and GluRs are key regulators of spine formation and maturation. Therefore, we tested the effects of NMDA and AMPA on ICAM-5 cleavage from hippocampal neurons. In 14-DIV hippocampal neurons, 5 μM NMDA or AMPA treatment caused significant release of the sICAM-5 fragments of 110 and 80–85 kD, with concomitant reduction of the membrane-bound ICAM-5 (). Moreover, the cleavage of ICAM-5 was inhibited by the NMDA antagonist MK801 and the non-NMDA antagonist DNQX (6,7,-dinitroquinoxaline-2,3[1H,4H]-dione; ). To further investigate the mechanism of ICAM-5 proteolytic cleavage, we used various chemical or peptide inhibitors of MMPs. As shown in , ICAM-5 cleavage was almost completely inhibited by the broad-spectrum MMP inhibitor GM6001, but not by its negative control. A variety of MMP-2 and -9 inhibitors also partially or completely blocked the cleavage (). These data provide the first evidence that MMP-2 and -9, especially when activated by NMDA, are mainly responsible for the proteolytic cleavage of ICAM-5. Because the expression and activity of MMP-9 have been shown to be up-regulated in response to stimuli that induce NR activation and LTP (; ), we tested the conditioned culture media from treated hippocampal neurons and found increased levels of both active MMP-2 and -9 upon NMDA stimulation (Fig. S1, available at ). We then used the RNA interference technique to temporarily decrease the expression of MMP-2 or -9 in hippocampal neurons (). After transfection into neurons, the MMP-2 and -9 siRNAs substantially decreased the protein levels of MMP-2 and -9, respectively (, right), with the concomitant inhibition of ICAM-5 cleavage from the transfected neurons (, left). We further incubated the recombinant mouse ICAM-5 D1-9-Fc protein with activated MMP-2 or -9 and studied the cleavage using anti–ICAM-5 polyclonal antibody 1000J () or anti-Fc polyclonal antibody () in Western blots. The cleaved proteins were then analyzed by matrix-assisted laser desorption/ionization–time of flight (MALDI-TOF) peptide mass mapping to identify the individual fragments (C-terminal fragment [CTF] 130-Fc, N-terminal fragment [NTF] 80–85, and NTF40; Fig. S2, available at ). We found that both MMP-2 and -9 cut at similar sites in the second and ninth ectodomains of ICAM-5, as depicted in . Interestingly, they seemed to show selectivity for the cleavage sites, as shown by the intensities of the fragments they produced. The minor bands detected within 30–75 kD in may be caused by protein degradation. Because the NRs and GluRs are known to regulate actin dynamics, we examined whether the NR-promoted cleavage of ICAM-5 is dependent on its anchorage to actin filaments. Interestingly, we found that cytochalasin D and latrunculin A, which prevent actin polymerization by capping the barbed end of actin filaments and monomeric actin, respectively, significantly increased the cleavage of ICAM-5 (). This indicates that the association between ICAM-5 and actin filaments may affect the MMP-mediated proteolytic cleavage. Furthermore, we found that treatment of hippocampal neurons with 20 μM NMDA for 1 h resulted in a considerable release of ICAM-5 from the cytoskeletal fraction into the soluble fraction of neuronal lysates (). These findings strongly support the hypothesis that dissociation of ICAM-5 from the actin cytoskeleton promotes its cleavage. To further study this phenomenon, we compared Paju-ICAM-5-fl and Paju-ICAM-5-ΔCP cell lines, which express the full-length ICAM-5 or the truncated ICAM-5 without the cytoplasmic domain (), respectively. We have shown that truncation of the cytoplasmic tail of ICAM-5 results in a more diffuse distribution of the molecule on the plasma membrane and reduced colocalization with the subcortical actin filaments (). The cell surface expression level of ICAM-5 on both cell lines was comparable (Fig. S3, available at ). Compared with Paju-ICAM-5-fl, Paju-ICAM-5-ΔCP cells showed a considerable increase of ICAM-5 cleavage (). To further clarify the involvement of MMP-2 and -9 on ICAM-5 cleavage, we studied the expression of ICAM-5 in MMP-deficient mice at different postnatal developmental stages (from postnatal 1 d to 10 wk). ICAM-5 expression was increased in all MMP-deficient mice after birth, whereas L1CAM showed a decreased expression (). Moreover, it is noteworthy that, compared with the wild-type (WT) mice, the MMP-2 and MMP-9 deficient mice contained significantly more full-length ICAM-5 at the early postnatal stage (postnatal day 1). However, the MMP-2/MMP-9 double-deficient mice expressed lower levels of ICAM-5 (). The changes in the MMP-2, MMP-9, and MMP-2/MMP-9 double-deficient mice tended to disappear after 1 wk postnatally, except in the MMP-2/MMP-9 double-deficient mice, where ICAM-5 expression still remained slightly but significantly lower than in the WT mice (). Studies on the enzymatic activities of MMP-2 and -9 by gelatinase zymography verified the identity of the respective deficient mice () and indicated that the activity of MMP-2 in both the MMP-9 deficient mice and the WT mice decreased during the postnatal development. However, there seemed to be some compensating up-regulation of the proMMP-9 level in the adult MMP-2 deficient mice, which was not obvious in the younger mice (, right). Similar findings have been reported (). Our histological analysis of the brains of MMP-2– and MMP-9–deficient mice showed that both, especially the MMP-2–deficient mice, had abnormal cerebral cortical and hippocampal structures. The cortical layers 2–3 seemed to have increased number of cells (Fig. S4, available at ). Similar findings have been reported in the cerebellar cortex of MMP-9–deficient mice, which showed an abnormal accumulation of granular precursors in the external granular layer (). We believe that the increased level of ICAM-5 in MMP-2–deficient mice is not due to the increased number of neurons in the cortex, because we carefully controlled the protein load per sample and monitored the amount of loaded actin during Western blotting. Dendritic spines occur in a range of sizes and in a variety of shapes, commonly classified as thin, stubby, and mushroom (; ). There is a strong correlation between the size of the spine head and the strength of the synapse, presumably related to the higher levels of GluRs in larger spines (). There is also evidence that the smaller weaker spines preferentially undergo LTP, whereas larger spines are more stable and show less plasticity (). Such observations have led to the view that thin spines represent “plasticity spines” and large mushrooms “memory” spines (). ICAM-5 was earlier found to be mainly expressed in dendritic filopodia, and its expression inversely correlated with synapse maturation (). To further clarify whether ICAM-5 is distinctively expressed in spines at different maturation stages, we studied the dendritic protrusions of 10–17-DIV hippocampal neurons. Such neurons contain various dendritic protrusions, including filopodia, small-head thin spines, and large-head mushroom spines. The dendritic fine structures were visualized by transfection of the EGFP into the 12-DIV neurons. The overlapping of immunostained ICAM-5 with EGFP was measured by calculation of the Pearson's colocalization efficiency (). Our result showed that ICAM-5 was more abundantly expressed in filopodia (, arrow) and thin spines (, arrowheads) but much less in the mushroom spines (, asterisks), particularly when the heads of the two different types of spines were compared (). These data suggest that ICAM-5 may play a more active role in the “plastic” thin spines than in the more “stable” mushroom spines. Because we found that activation of the NRs led to dissociation of membrane-bound ICAM-5 from the actin cytoskeleton, it became important to study ICAM-5 and F-actin distributions in neurons upon activation of NRs. We treated the 10-DIV hippocampal neurons with 5 μM NMDA and triple stained the neurons for the ICAM-5 cytoplasmic tail, the F-actin, and the postsynaptic density (PSD) 95 protein (). Treatment of hippocampal neurons with NMDA resulted in reduced colocalization of ICAM-5 with F-actin not only along dendritic shafts () but also in thin and mushroom spines (, arrowheads and asterisks, respectively). The reduced colocalization of ICAM-5 with F-actin was partially counteracted by pretreatment of neurons with the NR antagonist MK801 (). NMDA treatment led to a significantly increased amount of mushroom spines (). Because we found that NMDA and AMPA reduced anchorage of ICAM-5 to the actin cytoskeleton and promoted its cleavage through MMP-2 and -9, these facts could be the reasons for the exclusion of ICAM-5 from maturating spines. Therefore, it was important to study whether the ICAM-5 cleavage from spines, as a result of activation of the NRs, could be blocked by MMP inhibitors. For this purpose, we studied the EGFP-transfected 17-DIV hippocampal neurons that were treated with 5 μM NMDA and various MMP inhibitors (). To measure the effects of MMP inhibitors more accurately, ICAM-5 was visualized by immunostaining with a mAb that recognizes its extracellular region. We found that NMDA reduced the localization of ICAM-5 in both the dendritic shafts and thin spines in 17-DIV neurons (), similar to what was observed in younger neurons (). Blocking the NRs with MK801 clearly recovered the localization of ICAM-5 in both shafts and thin spines. When various MMP inhibitors, including the general inhibitor GM6001, the MMP-2/9–specific peptide inhibitor CTT, and the inhibitor II, were applied together with NMDA, most ICAM-5 along dendritic shafts was recovered (). However, the recovery of ICAM-5 in thin spines, especially in spine heads, was much less efficient by inhibition of MMPs, as compared with the NR antagonist MK801 (). This may be due to the fact that ICAM-5 was more vulnerable for MMPs in spine heads, which contain highly motile and dynamic actin filaments, as compared with the shafts that contain a more stable actin cytoskeleton. We further found that pretreatment with the MMP-2/9 inhibitor II significantly decreased the number of spines induced by NMDA (). To further study the function of ICAM-5 in thin spines, we compared the response of WT and ICAM-5 hippocampal neurons to NMDA stimulation. The EGFP-transfected 17-DIV fixed neurons were first immunostained for ICAM-5 and PSD95. ICAM-5 immunostaining verified the identity of ICAM-5 neurons (). In addition, the size of mushroom spines in ICAM-5 neurons seemed to be larger than those in WT neurons, which was similar to an earlier report (). To monitor the growth of thin spines in these neurons, we studied the 15-DIV EGFP-transfected neurons with a time-lapse fluorescence microscope. The neurons were treated with 20 μM NMDA for 1 h with or without a 1-h pretreatment with MK801, and monitored for 1 h. We found that thin spines in WT neurons showed increased growth of spine heads in response to NMDA stimulation. In contrast, spine heads in ICAM-5 neurons seemed to be retracting (, arrowheads). Spine numbers were increased in WT neurons, but not in ICAM-5 neurons after treatment with 5 μM NMDA for 8 h (). These data indicate that ICAM-5 is important for the motility of thin spines. Interestingly, we also found that mushroom spines in ICAM-5 neurons respond positively toward NMDA stimulation, with increased size of spine heads (, asterisks). To study functions of the sICAM-5, we cultured the EGFP-transfected 9-DIV WT and ICAM-5 neurons in the presence of 10 μg/ml recombinant sICAM-5 D1-4-Fc protein for 3 d. The neurons were then immunostained for ICAM-5 and microtubule-associated protein-2 (MAP-2; ). sICAM-5 D1- 4-Fc protein induced a significantly higher number of filopodia from the WT neurons, compared with the ICAM-5 neurons (). The filopodial length of WT neurons also significantly increased in the presence of sICAM-5 D1-4-Fc protein, as compared with the ICAM-5 neurons (). ICAM-5 has been shown to be gradually excluded from mature synapses, but the mechanism was not elucidated (). Here, we show that activation of the NRs induced cleavage of ICAM-5 (), which evidently is mediated by active MMP-2 and -9 (). The association of ICAM-5 with the actin cytoskeleton was decreased in dendritic spines in response to activation of the NRs, which affected the ICAM-5 cleavage ( and ). ICAM-5 deficiency led to the retraction of spine heads and a decreased number of spines in response to NMDA stimulation (). sICAM-5 protein increased the number and length of filopodia in WT neurons but not in ICAM-5–deficient neurons (). Combining these data with the earlier findings on ICAM-5 (; ; ), we present a schematic model depicting the NR-mediated spine development in which ICAM-5 is involved (). NMDA or AMPA stimulation causes increased MMP-2 and -9 activities in neurons and neighboring glial cells (not depicted), resulting in cleavage of the ectodomains of ICAM-5 from immature nascent spines. The reduced membrane level of ICAM-5 may facilitate local membrane and cytoskeleton reorganization, and thereby morphological remodeling of dendritic spines. We have shown that ICAM-5 promotes dendritic elongation through homophilic interaction (). Our current data further indicate that the increased number and length of filopodia from WT neurons is mediated by the homophilic interaction of sICAM-5 D1-4-Fc protein with membrane-bound ICAM-5. These data extend our knowledge on functions of ICAM-5 in the context of the NR-regulated dendritic development. Indeed, blocking the ionotropic glutamate receptors has been demonstrated to result in an ∼35% decrease in the density and turnover of shaft filopodia, whereas focal glutamate application leads to a 75% increase in the length of shaft filopodia (). ICAM-5 has been postulated to be a negative regulator of filopodia-to-spine transition (). In this sense, the promoted cleavage by NRs implies an important mechanism for a transformation of immature spines toward maturation. Moreover, the cooperative performance of ICAM-5 together with the NRs and MMPs may fine-tune the process of spine remodeling. The phenomenon that thin spines in ICAM-5 neurons retracted in response to NMDA treatment () seems to be contradictory to the fact that ICAM-5 neurons have eventually larger mature spines (). As the expression of ICAM-5 is the lowest in mature spines, we suspect that the eventual increase in size of mature spine heads is either not directly ICAM-5 related or resulted from secondary effects of ICAM-5 deficiency, which needs further clarification. We provide several lines of evidence that MMP-2 and -9 are responsible for the proteolytic processing of ICAM-5, leading to the production of the sICAM-5. First of all, we detected a steady-state cleavage of ICAM-5 from the cultured primary neurons, which was increased by NMDA or AMPA stimulation (). As earlier shown, MMP-9 gene expression is up-regulated in response to extracellular stimuli, like growth factors, cytokines, and neurotransmitters, whereas there is lack of transcriptional regulation of MMP-2 expression (; ; ). These facts suggest that MMP-2 is involved in the basal processing of ICAM-5 and MMP-9 in the activity-dependent cleavage of ICAM-5. In addition, NMDA-induced ICAM-5 cleavage was efficiently prevented by various MMP-2 and -9 inhibitors and siRNAs (). Abnormally high expression levels of ICAM-5 were found in the newborn MMP-2– or MMP-9–deficient mice (), supporting the finding of involvement of MMP-2 and -9 in ICAM-5 proteolytic processing. Interestingly, the difference in ICAM-5 expression between the MMP-2– or MMP-9–deficient mice and the WT mice gradually disappeared during postnatal brain development (), which may partially be due to the decrease of MMP-2 enzymatic activity () with the simultaneous increase of ICAM-5 expression during the later postnatal period. In contrast to ICAM-5, another important CAM, L1CAM, did not show changes in the expression levels in the MMP-2– or MMP-9–deficient mice as compared with the WT mice. Furthermore, L1CAM showed a gradual decrease in expression during the postnatal period (), indicating a shift of roles between the two molecules during brain maturation. An earlier report on MMP-2–deficient mice has shown that MMP-9 activity is up-regulated (). Here, we found a similar phenomenon in the brains of adult MMP-2–deficient mice (). The expression of ICAM-5 in the MMP double-deficient mice was reduced during the early postnatal period (), which may be due to compensating effects of other proteases. Another possibility could be that protein synthesis is deficient in these mice because of developmental defects. We found that the cytoskeletal anchorage of membrane-bound ICAM-5 was critical for controlling its proteolytic cleavage by MMPs. Disruption of actin filaments by cytochalasin D or latrunculin A, or deletion of the cytoplasmic tail of ICAM-5, significantly promoted its cleavage. Activation of the NRs resulted in dissociation of ICAM-5 from the actin cytoskeleton. The actin cytoskeleton determines the shape, motility, and stability of dendritic spines and provides the substrates for the Rho family small GTPases, which are the key regulators of actin polymerization and spine motility (; ). The NRs and GluRs have been shown to promote formation and stabilization of dendritic spines, respectively, by inhibiting the actin-based protrusive activity from the spine heads () and increasing the turnover time of dynamic actin in spines (). Inhibition of actin motility caused spines to round up so that spine morphology became more stable and regular (). These facts indicate that the cytoplasmic part of ICAM-5 participates in the NR-dependent morphological change of spines by exerting a regulatory role on MMP-mediated ICAM-5 cleavage. We have shown that ICAM-5 associates with the actin filaments via α-actinin and promotes neuritic outgrowth (). The NR subunit (NR2B) has been shown to interact with α-actinin (; ). α-Actinin has been implicated in the regulation of spine morphology (). Therefore, it is plausible that the NR may directly compete with ICAM-5 for interaction with actin filaments (). The role of MMPs in the normal brain development is gradually becoming apparent (; ; ). However, little is known concerning the effects of MMPs on dendritic spine development, even though both their ECM and non-ECM substrates in the brain have been found to be important for spine formation and remodeling (). Recently, MMP-7 (), MMP-9 (; ), and MMP-24 () were shown to be involved in dendritic filopodia elongation or synaptic remodeling. Particularly, MMP-9–deficient mice show impaired LTP and behavioral impairments in hippocampus-dependent associative learning (), suggesting the potential importance of MMP on dendritic spine development. Although mutant mice lacking individual MMPs have been generated (; ), no obvious defects in embryogenesis have been reported. In particular, MMP-2–deficient mice seemed to be healthy and fertile (), although they exhibited defects in bone metabolism (). We found that MMP-2–deficient mice seemed to have an increased number of cells in the cerebral cortex, especially in layers 2–3 (Fig. S4). Similar findings have been reported in the cerebellar cortex of MMP-9–deficient mice, which showed an abnormal accumulation of granular precursors in the external granular layer (). Thus, our findings on MMP-2–deficient mice deserve more careful and detailed study in the future. In summary, we have defined a physiological mechanism for the proteolytic processing of ICAM-5 by MMP-2 and -9, and the importance of its cleavage on regulation of dendritic spine development. Our results will help elucidate the functions of both MMPs and adhesion molecules on dendritic development, which is still poorly understood. AMPA, DNQX, gelatin, MK-801, NMDA, and poly--lysine were obtained from Sigma-Aldrich. Cytochalasin D, Latrunculin A, GM6001, GM6001 Neg. Ctrl, MMP-2/MMP-9 inhibitor II, and MMP-9 inhibitor I were obtained from Calbiochem. ProMMP-2 and -9 were obtained from Roche. CTT and CTT peptides were gifts from E. Koivunen (Division of Biochemistry, University of Helsinki, Helsinki, Finland; ). MMP-2–, MMP-9–, and ICAM-5–deficient mice were generated by gene targeting (; ; ). All animals were backcrossed at least six generations into a homogenous C57BL/6 genetic background and were bred as homozygous lines. Mice deficient in both MMP-2 and -9 were obtained by intercrossing mice that were heterozygous for both mutations. All experiments were approved by and performed according to the guidelines of the local animal ethical committee. Paju-Mock, Paju-ICAM-5-fl, and Paju-ICAM-5-ΔCP cell lines and hippocampal neurons were prepared as described earlier (). The CHO cell line stably expressing ICAM-5 D1- 4-Fc recombinant protein, a gift from J. Casasnovas (Universidad Autonoma, Madrid, Spain), was grown as recommended (). During the 3-wk period of in vitro cultivation of hippocampal neurons, the culture media were replaced with HBSS with 1.8 mM CaCl buffer for 16 h on days 3, 7, 14, and 21. The 14-DIV hippocampal neurons were treated for 16 h with 5 μM NMDA or AMPA, with or without a 2-h pretreatment with 20 μM MK-801 or DNQX, respectively, in HBSS/Ca buffer. When MMP inhibitors were applied, 20–25 μM chemical inhibitors or 100 μM peptide inhibitors were used together with NMDA. Paju-ICAM-5-fl and Paju-ICAM-5-ΔCP cells were incubated in serum-free culture media for 18 h. Then, 1-ml aliquots of the conditioned culture media were concentrated 20-fold by Vivaspin centrifugal concentrators (Sartorius Ltd.), and the cells were stripped off. All samples were suspended in Laemmli sample buffer for Western blotting. 9-DIV rat hippocampal neurons were transfected with 50 nM predesigned siRNAs against rat MMP-2, MMP-9, or negative control siRNA (Ambion) using Lipofectamine RNAiMAX reagent (Invitrogen) for 48 h. The culture media were changed into HBSS/Ca buffer for 16 h and then collected and concentrated for Western blotting. Forebrains from the postnatal 1 d ( = 4), 1 wk ( = 2), and 10 wk ( = 2) MMP-deficient or WT mice were homogenized with buffer containing 0.32 M sucrose, 10 mM Hepes, pH 7.4, 2 mM EDTA, 50 mM NaF, 1 mM NaVO, and 1× protease cocktail inhibitors, using a glass-teflon homogenizer. The homogenates were then centrifuged at 1,000 , and the supernatants were centrifuged at 50,000 rpm to separate the membrane fractions from the soluble fractions. The membrane fractions were suspended in lysis buffer (1% Triton X-100, 50 mM Hepes, pH 7.4, 2 mM EDTA, and protease/phosphatase inhibitors). For Western blotting, 20 μg of protein from each sample was suspended in Laemmli sample buffer. 14-DIV hippocampal neurons were either left untreated or treated with 20 μM NMDA for 60 min, and cells were then lysed in lysis buffer. Lysates were centrifuged at 5,000 rpm to get rid of the nuclear fractions, and the supernatants were further centrifuged at 100,000 rpm for 2 h at +2°C to separate the cytoskeletal fractions from the soluble fractions, and each sample was suspended in Laemmli sample buffer for Western blotting. Recombinant human ICAM-5 D1-4-Fc and mouse ICAM-5 D1-9-Fc proteins were purified from cell culture supernatants by affinity chromatography with protein A–Sepharose and ÄKTAprime system (GE Healthcare). ProMMP-2 and -9 were activated with p-aminophenylmercuric acetate and trypsin, respectively, and 40 ng of activated enzymes was incubated with 2 μg ICAM-5 D1-9-Fc protein in 50 μl enzyme buffer (20 mM Hepes, 150 mM NaCl, 0.2 mM CaCl, 1 mM MnCl, and 1 μM ZnCl) at 37°C for 18 h. 5 μl of the enzyme-substrate mixtures were suspended in sample buffer. Samples were separated by 4–12% SDS-PAGE (Invitrogen) and transferred to nitrocellulose membranes (Whatman GmbH). After blocking, membranes were incubated with anti–ICAM-5 pAb 1000J, anti–ICAM-5cp pAb, anti-L1CAM mAb, anti-actin pAb, or horseradish peroxidase–conjugated anti-human pAb, respectively, followed by peroxidase-conjugated secondary antibodies. Membranes were washed with TBS and 0.05% Tween 20 after each incubation and developed with an ECL kit (GE Healthcare). Band intensity was quantified by the software Tina 2.09c (Raytest). 20 μl of 60-fold–concentrated serum-free cell culture media or 50 μg of protein from brain membrane fractions was suspended in sample buffer and separated by 8% SDS−PAGE containing 0.2% gelatin. Gels were then washed with 2.5% Triton X-100 to remove SDS and incubated in substrate buffer (50 mM Tris, pH 8, and 5 mM CaCl) for 18 h at 37°C, followed by staining with 0.5% Coomassie blue. About 1 μg ICAM-5 D1-9-Fc fragments after the MMP-2 or -9 digestion were separated by 4–12% SDS-PAGE (Invitrogen), silver stained, and analyzed in the Protein Chemistry Unit of the Institute of Biotechnology, University of Helsinki. The bands of interest were cut out, reduced with dithiothreitol, alkylated with iodoacetamide, and “in-gel” digested with trypsin (Sequencing Grade Modified Trypsin; V5111; Promega). The recovered peptides were, after desalting using μ-C18 ZipTip (Millipore), subjected to MALDI-TOF mass spectrometric analysis. MALDI-TOF mass spectra for mass fingerprinting and MALDI-TOF/TOF mass spectra for identification by fragment ion analysis were obtained using an Ultraflex TOF/TOF instrument (Bruker-Daltonik GmbH). Protein identification with the generated data was performed using Mascot Peptide Mass Fingerprint and MS/MS Ion Search programs. Paju-Mock, Paju-ICAM-5-fl, or Paju-ICAM-5-ΔCP cells were incubated with 5 μg/ml mAb TL-3 and then with Alexa488-conjugated anti-mouse pAb (Invitrogen). Cells were washed with PBS after each incubation. Samples were analyzed with FACScan and CellQuest software (Becton Dickinson). Hippocampal neurons were transfected with pEGFP-N1 plasmid using Lipofectamine 2000 reagent (Invitrogen) at 8–9 DIV and cultured until 12–17 DIV. For filopodia elongation assay, the 9-DIV neurons were treated twice with 10 μg/ml of recombinant ICAM-5 D1-4-Fc protein or control mIgG for 72 h. The 10–12-DIV neurons were then fixed with 4% paraformaldehyde and permeabilized with 0.1% Triton X-100. After blocking with 2% BSA in PBS, neurons were stained with pAb anti–ICAM-5cp plus Alexa488- or Cy3-conjugated anti-rabbit IgG, Cy3-conjugated phalloidin, anti-PSD95 mAb, or anti-MAP-2 mAb plus Cy5-conjugated anti-mouse IgG. The 17-DIV neurons were left untreated or treated with 5 μM NMDA with or without a 2-h pretreatment with 20 μM MK801 or MMP inhibitors in HBSS/Ca buffer for 6 h. The neurons were fixed, permeabilized, and blocked afterward, and stained with mAb 127E plus Cy3-conjugated anti-mouse IgG and anti-PSD95 pAb plus Cy5-conjugated anti-rabbit IgG. The fluorescent images were taken with a confocal laser-scanning microscope under 63× magnification (TCS SP2 AOBS, HCX PL APO 63×O/1.4-0.6; Leica) using a charge-coupled device camera (Leica) and the LCSLite software. Four to five neurons per sample were randomly imaged for each experiment. At least three proximal dendritic segments (∼60 μm per segment) were analyzed for each neuron. Dendritic filopodia (>2 μm long with pointy tip), thin spines (0.5–2.5 μm long with bulbous tip and <0.1 μm thick in neck), or mushroom-shaped spines (0.5–2.0 μm long and 0.3–0.6 μm wide in head) were quantified and presented as numbers per 100 μm dendritic length. For live imaging, 14-DIV EGFP-transfected neurons were treated with 20 μM NMDA in HBSS/Ca buffer, with or without pretreatment with 20 μM MK801 for 1 h, in 5% CO/10% O at 37°C, and monitored with an inverted fluorescent microscope under 60× magnification (IX-71; UPlanSApo 60×W/1.2; Olympus) using an electron multiplying charge- coupled device camera (DV885; Andor Technology) and the TillVision software (Till Photonics GmbH). Images were processed with Photoshop and ImagePro plus. Pearson's coefficients were used for colocalization analysis. Brains from 8-wk-old mice were fixed with 4% paraformaldehyde in PBS and embedded in paraffin wax. Coronal paraffin sections 10 μm thick were cut and mounted on glass slides. Brain sections were stained with cresyl violet and visualized with a light microscope (IX71; Olympus). Images were processed with Photoshop (Adobe). test was used to compare different groups of data. Fig. S1 shows that NMDA increases the expression and activities of MMP-2 and -9. Fig. S2 shows peptide mass mapping of MMP-cleaved ICAM-5-Fc proteins. Fig. S3 shows flow cytometry analysis of transfected Paju cell lines. Fig. S4 shows abnormal cortical and hippocampal development in MMP-2–deficient mice. Online supplemental material is available at .

Fibronectin (Fn) is required for mammalian development and for blood vessel formation (, ; ; ; ). Fn exists in both a soluble form in the plasma and as insoluble disulfide-bonded multimers in the extracellular matrix. The insoluble matrix form of Fn is essential for most of its biological functions in events such as wound healing, embryogenesis, and formation of the tumor microenvironment (; ; ). The soluble form of Fn is converted to the insoluble form by a process termed Fn matrix assembly, an active cellular process in which the soluble, dimeric Fn molecules are assembled into an insoluble, fibrillar pericellular matrix. Thus, an understanding of Fn matrix assembly is of broad biological significance. Pioneering studies (, ; ; ; , ) have defined the regions of Fn important in the assembly process. The N-terminal 70-kD domain of Fn plays a pivotal role in matrix assembly. It does so by interacting with several other sites within the Fn molecule. These sites, which are generally contained within the type III repeats, are cryptic in soluble Fn and exposed as a consequence of conformational changes in Fn (). These conformational changes may involve extension of the Fn or unfolding of particular repeats. Studies from several laboratories show that Fn conformation can be changed by mechanical deformation (; ; , ; ). Indeed, the hypothesis that matrix assembly site exposure is initiated by integrin-dependent mechanical deformation of Fn () has been experimentally validated (). Thus, the assembly of an Fn matrix results from Fn–Fn interactions initiated by mechanical deformation of the Fn molecule. The binding of Fn to integrins, such as α5β1, initiates the matrix assembly process (; ). In particular, forces generated by the cytoskeleton and conveyed to Fn via integrins cause deformation of the Fn molecule (). For assembly to begin, these integrins must physically associate with Fn, usually by binding it with high affinity (operationally defined as an “activated” integrin; ), and physically associate with the cytoskeleton (). Several signaling enzymes, including Src family kinases, pp125, Rho GTPases, and PI3 kinase, are also involved in the matrix assembly process (). These same signaling enzymes can be regulated by integrin ligation in a process referred to as outside-in integrin signaling (; ; ). Thus, it is possible that integrin signals that regulate kinases such as pp125 could participate in the assembly process. We recently found that CD98hc, a type II transmembrane protein, mediates outside-in integrin signaling (). CD98hc has two distinct functions: it can associate with and regulate the function of selected integrins and it can regulate the expression and distribution of CD98 light chains to modulate amino acid transport function. Each function depends on distinct domains within CD98hc, with the intracellular portion being required for interaction with integrins (). Deletion of CD98hc in embryonic stem (ES) cells impaired the formation of teratocarcinomas in vivo and many integrin-dependent functions in vitro because CD98hc is a contributor to integrin-dependent biochemical signals (). The role of CD98hc in integrin signaling suggested that it might participate in the Fn matrix assembly process. Here, we report that CD98hc is required for efficient Fn matrix assembly both in vitro and in vivo. Furthermore, deletion of CD98hc has little effect on Fn biosynthesis or integrin activation; instead, lack of CD98hc impairs outside-in integrin signals that result in RhoA-mediated cellular contractility necessary for Fn matrix assembly. Finally, the portion of CD98hc that interacts with integrins is necessary and sufficient to support cellular contractility and Fn matrix assembly. Thus, we find that CD98hc participates in Fn matrix assembly by mediating the outside-in integrin signals required for the contractile events that initiate and sustain Fn matrix assembly. To determine whether CD98 regulates matrix assembly in vivo we examined the distribution of Fn in teratocarcinomas formed in nude mice after subcutaneous injection of ES cells. As previously reported, injection of wild-type (WT) ES cells into nude mice led to the formation of large tumors. In contrast, CD98hc-null cells either did not form tumors or formed very small tumors (). The CD98hc-null tumors that formed exhibited profound reduction in Fn fibrils, as judged by staining with a mAb against Fn (). In contrast, a prominent network of Fn was detected in the tumors formed by WT ES cells. Thus, CD98hc is involved in Fn matrix formation in vivo. Importantly, these CD98hc-null teratocarcinomas, like those formed from WT ES cells contain tissues from all three germ layers. Furthermore, absence of CD98hc is compatible with differentiation of multiple cellular lineages (); hence, changes in cell lineages in the tumors are unlikely to account for the observed defect in matrix assembly. Thus, these results strongly suggest that lack of CD98hc resulted in a marked reduction in Fn matrix formation in vivo. Because Fn is required for vascular development, we examined the blood vessels formed in the teratocarcinomas () by whole-mount staining with endothelial cell–specific fluorescent-labeled lectin. We also visualized pericytes/smooth muscle cells and endothelial cells on these tumors by detection of the specific markers α-smooth muscle actin and platelet/endothelial cell adhesion molecule 1 (PECAM-1), respectively (). In the CD98hc-null tumors, endothelial cells were present but failed to form intact blood vessels () or to invest with α-smooth muscle actin–expressing cells (). Instead of associating with endothelial cells, α-smooth muscle actin–expressing cells were dispersed throughout the tumors. In sharp contrast, WT tumors exhibited an organized network of blood vessels, including larger vessels with an endothelial lining and smooth muscle cell–containing intima and media (). Thus, the absence of CD98hc impairs Fn matrix deposition and vascular development, a process known to depend on Fn. To learn whether CD98hc deficiency by itself leads directly to defective cellular Fn matrix formation and to understand the mechanisms by which CD98hc supports Fn matrix formation, we developed conditional CD98hc-null mouse embryonic fibroblasts (MEFs; ). CD98hc conditional–null mice were generated as described in Materials and methods. We deleted CD98hc in vitro by infecting these cells with adenoviruses expressing Cre recombinase, resulting in 95% excision of exons 1 and 2 of CD98hc (). These cells were then mass sorted with immunomagnetic beads to select fibroblasts lacking detectable CD98hc (). Importantly, the CD98hc-null fibroblasts and parental WT cells expressed similar quantities of integrins α5, β1, α6, and αv, as judged by flow cytometry () and showed defects in integrin signaling similar to those seen in CD98hc-null fibroblasts derived from ES cells (Fig. S1, available at ). None of the cells expressed a detectable amount of α1, α2, α4, or β3 integrins. Thus, these conditional CD98hc MEFs provide a tool to analyze the effect of CD98hc deletion on Fn matrix formation and integrin signaling by fibroblasts. To examine Fn matrix assembly, we cultured WT or CD98hc-null fibroblasts and assessed matrix deposition by immunofluorescence. 48 h after plating, WT cells exhibited abundant Fn fibrils, whereas CD98hc-null cells (, green) were devoid of Fn fibrils (). Similar differences were observed regardless of whether the cells were subconfluent () or confluent (not depicted). The lack of matrix was confirmed biochemically, by assaying the formation of deoxycholate (DOC)-insoluble Fn (), which was dramatically reduced in the confluent CD98hc-null fibroblasts. Because reduced assembly was observed in confluent CD98hc-null cells, the lack of matrix in these cells cannot be ascribed to reduced cell density (). Cell-associated Fn is present in fibroblast cultures in two separate pools, distinguishable on the basis of their solubility in DOC. One pool contains cell-bound DOC-soluble Fn, whereas the other one contains matrix-associated DOC-insoluble Fn. Fn matrix assembly proceeds by the binding of Fn to cells in the DOC-soluble pool with subsequent transfer and accumulation of assembled Fn in the DOC-insoluble pool (). Although there was at least a 20-fold reduction in the DOC-insoluble Fn with the CD98hc-null cells (, left), there was only a modest, approximately twofold reduction in the DOC-soluble cell-bound Fn (, middle) and no decrease Fn in the conditioned medium (, right) compared with WT cells. Thus, CD98hc deletion leads to defective Fn matrix assembly, and the defect appears to be at the step of transfer from the DOC-soluble to -insoluble pool. Because CD98hc participates in amino acid transport (), we considered the possibility that CD98hc deficiency reduced the biosynthesis of Fn. To assess the availability of endogenous soluble Fn, we cultured the cells in medium containing Fn-depleted FBS (). Roughly twofold greater quantities of secreted soluble Fn were detected in the conditioned medium of CD98hc-null cells cultured in Fn-depleted medium (, left), and a profound reduction in DOC-insoluble Fn matrix was evident (, middle). This result suggests that both cell types produce similar quantities of Fn, but CD98hc-deficient cells cannot assemble the Fn, resulting in the increase in the conditioned medium. Furthermore, when we supplemented the cell cultures with an excess of plasma Fn, the CD98hc-null cells still failed to assemble a matrix (). Thus, CD98hc is required for Fn matrix assembly, but not for Fn synthesis or secretion, and the lack of assembly in CD98hc-null cells is not due to reduced Fn availability. Matrix assembly requires that Fn first binds to high affinity (operationally defined as activated) integrins, such as α5β1, αIIbβ3, or αVβ3 (,; , ). To assess the effect of CD98hc deletion on the affinity of Fn binding integrins in these MEFs, we examined their binding to a recombinant-soluble cell binding domain of Fn, composed of type 3 repeats 9–11 (3Fn[9–11]). The absence of CD98hc had no effect on the ability of the cells to bind to 3Fn(F9–11), a direct measure of the affinity of Fn binding integrins, such as integrin α5β1 (). Importantly, when incubated with an activating anti–β1 integrin antibody, 9EG7, there was a marked increase in 3Fn(9–11) binding to both the WT and the null cells (); however, this enforced activation of integrins did not rescue Fn assembly by the CD98hc-deficient cells (). Thus, the CD98hc-null cells' failure to assemble matrix is not due to a defect in their ability to synthesize or secrete Fn or to bind to soluble Fn with high affinity. Fn must be conformationally altered to initiate assembly into fibrils (). Once Fn binds to integrins, and the integrins connect to the cytoskeleton, RhoA-mediated cellular contractility exerts force on the Fn, leading to these conformational changes (; ). We therefore examined the role of CD98hc in cellular contractility by assaying the contraction of an Fn-fibrin provisional matrix. Fibrin clots formed from blood plasma contain covalently cross-linked Fn that serves as a ligand for integrin α5β1–dependent matrix contraction (). Incorporation of WT MEFs in these matrices led to a 52 ± 0.8% contraction after 2 h (). In contrast, contraction was markedly reduced when CD98hc-null cells were incorporated in the clot (24 ± 0.55% clot contraction). These cells lack β3 integrins, suggesting that the contraction was mediated by Fn binding β1 integrins, such as α5β1. Indeed, clot retraction was strongly inhibited by a function-blocking anti–mouse β1 antibody (clone Ha2/5; 18 ± 0.2 vs. 52 ± 0.8% clot contraction; unpublished data). Thus, CD98hc mediates the β1 integrin–dependent contraction of an Fn-fibrin matrix. The finding that CD98hc was required for efficient clot contraction suggests that it mediates the generation of traction force on the extracellular matrix, a critical requirement for Fn assembly (). To examine this question, we used displacement of embedded fluorescent microbeads in flexible polyacrylamide substrates as a direct measure of traction forces (). Subconfluent MEFs were seeded for 2 h on Fn-coated polyacrylamide substrates in serum-free medium. Deformation of substrate caused by cellular traction forces was detected by tracing the displacement of fluorescent beads, and displacement vector maps, or strain maps, were generated (, green). The resulting strain maps demonstrated a marked reduction in traction forces exerted by CD98hc-deficient MEFs compared with WT cells (). There was about a threefold reduction in integrated strain for CD98hc-null cells compared with WT (0.77 vs. 2.5 pixels/area [arbitrary units], respectively; see Materials and methods). The small GTP binding protein RhoA is a major regulator of cell contractility (; ) that provides the traction forces to initiate Fn matrix assembly (). Matrix assembly within tissues occurs when cells are in contact with surrounding 3D extracellular matrix, leading us to examine the effect of CD98hc deletion on adhesion-mediated alterations in RhoA activity when cells interact with 3D matrix (). 30 min after adhesion of WT MEFs to a 3D matrix, RhoA activity decreased, followed by a secondary increase as previously described (). In sharp contrast, the CD98hc-null cells did not show this late increase in RhoA activity (). Strikingly, both CD98hc-null and WT MEFs exhibited similar increases in RhoA activity in response to lysophosphatidic acid (LPA), an agonist that activates RhoA via G protein–coupled receptors (; ). Thus, CD98hc is required for RhoA activation in response to cell adhesion to a 3D Fn matrix. As noted, LPA induced RhoA activation in CD98hc-null MEFs, and activation of RhoA stimulates contractility, leading to the assembly of Fn by a variety of fibroblastic cells (; ; ). Consistent with the capacity of LPA to induce RhoA activation in CD98hc-null MEFs, addition of LPA enabled CD98hc-deficient MEFs to contract a Fn-fibrin matrix (matrix contraction: WT, 32 ± 1.1%; CD98hc null, 10 ± 0.2%; LPA-treated CD98hc null, 52 ± 0.7%; ). Similarly, LPA treatment enabled CD98hc-deficient cells to assemble an Fn matrix as assessed by quantifying formation of DOC-insoluble Fn (CD98hc null, 38 ± 6.3%; LPA-treated CD98hc null, 159 ± 3.3%, relative to WT MEFs; ). Collectively, these data show that the reduced capacity of CD98hc-null cells to activate RhoA in response to extracellular matrix leads to their failure to assemble an Fn matrix. The foregoing studies established that CD98hc participates in assembly of the Fn matrix by mediating adhesion-dependent RhoA activation that leads to traction on the extracellular matrix. Importantly, the CD98hc-null cells were able to activate RhoA in response to LPA, suggesting that matrix-driven RhoA activation is important in the assembly process. Integrins are the principal receptors that lead to matrix-initiated biochemical signals (; ), and we previously found that the physical interaction of CD98hc with integrins is required for efficient integrin signaling (). To examine the role of CD98hc–integrin interaction in matrix assembly, we reconstituted CD98hc-null MEFs with retroviruses encoding chimeras () formed between CD98hc and another type II transmembrane protein, CD69 (); each was well expressed as judged by flow cytometry (). The C98T98E69 (cytoplasmic domain CD98hc, transmembrane domain CD98hc, and extracellular domain CD69) chimera interacts with integrins, whereas C69T98E98 and C98T69E98 do not associate with integrins. Only the integrin binding chimera (C98T98E69) rescued the defect in Fn matrix assembly () and contractility (). CD98hc-deficient cells expressing the chimera that binds to integrins (C98T98E69) were able to induce a clot contraction (56 ± 1%) even more efficiently than WT cells (37 ± 0.8%). Thus, the capacity of CD98hc to mediate contraction of an Fn-fibrin clot depends on its ability to interact with integrins. Fn matrix assembly can be conveniently divided into three steps. First, Fn binding integrins must be activated to bind Fn with high affinity (; ). Recent studies (; ) establish that integrin activation is mediated by talin binding to the integrin β cytoplasmic domain. Second, the integrins must connect to the actin cytoskeleton (; ; ), a connection that can be mediated by talin (; ). Third, cellular contractility exerts force on the integrin-bound Fn, deforming it and leading to the initiation of Fn matrix assembly (; ). Here, we show that CD98hc participates in Fn matrix assembly and that it does so by physically associating with integrins to mediate matrix-driven activation of RhoA and the resulting cellular contractility that exerts force on the matrix. The variable expression of CD98hc and its contribution to Fn assembly have implications for the relationship between cell proliferation and formation of the extracellular matrix. CD98hc expression is tightly regulated coordinately with cell proliferation (), a pivotal event in wound healing, tumorigenesis, and development. Repair of skin wounds depends on precisely controlled formation of a fibrous extracellular matrix. The extracellular matrix deposition correlates with proliferation of fibroblasts and endothelial cells within the damaged area (). The dramatic up-regulation of CD98hc in proliferating cells could thus serve to promote Fn matrix assembly in healing wounds; conversely, a failure to down-regulate CD98hc could lead to excess matrix deposition and cell proliferation, as seen in fibrosis and keloid scars (). Similarly, the marked up-regulation of CD98hc in tumors () could contribute to enhanced formation of Fn extracellular matrix, thus altering the microenvironment of cells during tumorigenesis (; ; ; ; ). Moreover, we show that CD98hc mediates cell contractility, an event strongly associated with progression of some tumors (). Finally, the deposition of Fn along the blastocoele roof is reported to be a critical step in gastrulation () and later for morphogenetic movements driven by convergent extension (). This spatiotemporally regulated matrix assembly is controlled by the activity of cellular integrins, as manifested by their ability to promote cell spreading (; ), a CD98hc-dependent process (). Thus, the importance of Fn matrix assembly in development and the role of CD98hc in matrix assembly shown here provide a cogent explanation for the early embryonic lethality of CD98hc gene disruption ( ) and suggest that changes in CD98hc expression, or its association with integrins, may influence a wide range of developmental processes through the regulation of Fn matrix assembly. A suspension of ES cells (1.5 × 10 cells per site) was injected subcutaneously into athymic BALB/c wehi nude mice. After 33 d, tumors were fixed in 10% formaldehyde, paraffin embedded, sectioned, and stained for Fn with a rabbit polyclonal antibody against human Fn (Sigma-Aldrich). Alexa Fluor 546 goat anti–rabbit IgG (H+L; Invitrogen) was used as a secondary antibody. Sections were counterstained with the nuclear dye YOPRO-1 (Invitrogen). Tumor samples were fixed in 10% formaldehyde for 30 min at room temperature, washed in 1× PBS containing 0.1 g/liter CaCl, and cut into small (2 mm) tissue blocks. The tissue blocks were then incubated in the calcium-containing PBS in the dark for 48 h at 4°C in the presence of 10 μg/ml isolectin GS-IB4 Alexa Flour 568 conjugate (Invitrogen). After washing, the stained tissues were analyzed using a laser-scanning confocal microscope (1024 MRC; Bio-Rad Laboratories, Inc.). Z-serial images were merged using Lasersharp software (Bio-Rad Laboratories, Inc.) to obtain a 3D impression of the tumor samples. 4-μm tissue sections were deparaffinized and treated with methanol containing 0.3% HO for 30 min at room temperature. Antigen retrieval and staining were performed using standard procedures. In brief, sections were first stained for α-smooth muscle actin using mouse anti–human α-smooth muscle actin antibody (clone 1A4; DakoCytomation) diluted at 1:50 in PBS containing 3% BSA and 0.01% Tween 20 for 60 min at room temperature. After washing, sections were incubated with secondary antibody goat anti– mouse Alexa 546 (Invitrogen) at 1:500 in PBS containing 0.01% Tween 20. Second, the TSA kit (PerkinElmer) was used for PECAM-1 staining using purified rat anti–mouse PECAM-1 mAb (BD Biosciences). The bound antibodies were detected using biotin-conjugated rabbit anti–rat (Vector Laboratories) in combination with Streptavidin-FITC antibody (Vector Laboratories). Sections stained as above but in the absence of the primary antibodies served as the negative controls. A P1 mouse ES cell clone containing the gene was isolated from a 129Sv/J mouse library by PCR screening (Genome Systems, Inc.). The targeting vector consisted of a 1.4-kb 5′ homologous region and a 4.9-kb 3′ homologous region. The region of exon 1, encoding the transmembrane domain of CD98hc, was flanked with loxP site. A neomycin selection cassette flanked by Flp sites (provided by H. Beggs and G. Martin, University of California, San Francisco, San Francisco, CA) was inserted in intron 2. The linearized targeting construct was electroporated into R1 ES cells. G418-resistant colonies were selected for 7 d. Two homologous ES cell recombinants were identified by PCR analysis (forward primer A, 5′-GATAGACGGGAGTATTCAGCGAGGC-3′; and reverse primer B, 5′-CTCATGGTGCCTGCAGAAACGG-3′) and confirmed by Southern blotting. PCR products were obtained at the expected size: 248 bp for WT allele and 304 bp for conditional knockout allele. These clones were karyotyped and subsequently injected into E3.5 C57BL/6 host blastocysts. The blastocysts were then transferred into pseudopregnant foster females. A total of seven chimeric males (distinguished by coat color) were obtained and bred to WT C57BL/6 females. Germ line offspring were genotyped by preparing DNA from tail biopsy for the presence of the targeted allele. Heterozygote males were identified by PCR analysis. Homozygote CD98hc-conditional knockout mice were bred onto human β-actin FLPe deleter strain (The Jackson Laboratory) to excise the neomycin selection cassette. Mice were housed in the University of California, San Diego, animal facility, and experiments were approved by the University of California, San Diego, Institutional Animal Care and Use Committee. MEFs were derived from CD98hc-conditional knockout homozygote embryos. MEFs were cultured in complete DME high glucose (Invitrogen), supplemented with 10% FBS (HyClone), 20 mM Hepes, pH 7.3 (Invitrogen), 0.1 mM nonessential amino acid (Invitrogen), 0.1 mM β-mercaptoethanol (Invitrogen), and 2 mM -Glutamine (Invitrogen). The CD98hc-null MEFs were generated by infecting CD98hc-conditional knockout MEFs with adeno-CRE encoding CRE recombinase. Viral titers ranged from 0.6 to 1.2 × 10 U/ml. CD98hc deletion was detected by PCR (forward primer A, see the previous section; and reverse primer C, 5′-CAGGGTTCTGTGTATGTGGGCGG-3′) and confirmed by flow cytometry. Cells were stimulated with 2 μg/ml (unless otherwise mentioned) LPA (1-oleoyl-2-hydroxy-sn-glycero-3 phosphate, monosodium salt) as described elsewhere (). In brief, MEFs were plated at 1 million per 10-cm Petri dish and treated every 4–5 h for 18 h (starting 24 h after plating). The reconstituted cells were generated by infecting CD98hc-null MEFs with pBabe-Puromycin retrovirus encoding or CD98hc/CD69 chimeras (). Viruses were generated in EcoPack 293 cells (CLONTECH Laboratories, Inc.), and viral titers ranged from 0.9 to 1.7 ×10 U/ml. After puromycin selection, chimera expression was confirmed by flow cytometry. MEFs were seeded in 1% FBS at 10,000 cells/well in 24-well plates onto 12-mm coverslips coated with 4 μg/ml Fn and blocked with 0.5% BSA. Where indicated, either 25 μg/ml of exogenous plasma Fn or 10 μg/ml of activating anti-β1 mAb, 9EG7, was added to the culture medium. After 48 h in culture, the cells were fixed with 2% PFA in PBS for 20 min and permeabilized for 10 min with 0.05% Triton X-100. Fixed cells were washed with PBS, and nonspecific sites were blocked with 3% BSA before the addition of primary antibody (rabbit anti-Fn [Sigma-Aldrich]; 1:2,000) for 2 h at 37°C. Specific binding was detected using goat anti–rabbit antibody conjugated with Alexa 488 (Invitrogen). The actin cytoskeleton was visualized by incubating with rhodamine-conjugated phalloidin while the nuclei were stained with DAPI (Invitrogen). Individual coverslips were mounted in aqueous mounting media with anti-fade (Gel Mount; Sigma-Aldrich). Digital images were captured at room temperature with a charge-coupled device camera (Orca ER; Hamamatsu) using a standard upright fluorescent microscope (Axioplan 2 [Carl Zeiss MicroImaging, Inc.; Plan-Neo Fluar 20× objective with 0.50 NA) controlled by the image capture and processing program Openlab (Improvision). Photoshop (Adobe) was used to increase the γ of images. To assess the Fn assembly biochemically, the DOC-soluble and -insoluble portions of the cell matrix were analyzed as described previously (). In brief, MEFs were plated at 3 million cells per 10-cm Petri dish in complete medium (Fn+) or Fn-depleted medium (Fn−). Fn-depleted medium was generated by depleting Fn from the FBS using gelatin Sepharose before preparing the medium (). The cells were incubated for 48 h, at which time the conditioned medium was collected. The cell monolayer was washed with 3× PBS and lysed in 2 ml 3% DOC, 50 mM Tris, pH 8.8, and 0.1 mM EDTA, with protease inhibitors (complete, mini protease inhibitor tablets; Roche) for 15 min at 4°C. The lysate was subsequently passed through a 23-gauge needle five times and spun at 35,000 for 20 min to remove the insoluble material. The soluble component was kept for further analysis. The insoluble pellet was washed once with DOC lysis buffer and spun again. The pellet was solubilized in 50 μl reducing Laemmli buffer (62.5 mM Tris-HC1, pH 6.8, 1.5% SDS, 9% glycerol, 50 mM dithiothreitol, and 0.005% bromophenol blue). Samples were separated by SDS-PAGE and analyzed by Western blotting using rabbit anti-Fn antibody (Sigma-Aldrich) followed by an HRP-conjugated goat anti–rabbit IgG (Jackson ImmunoResearch Laboratories), and bands were visualized by chemiluminescence (Pierce Chemical Co.). Intensities of bands were quantified by scanning densitometry using ImageJ software. Anti–mouse α (clone Ha 31/8), anti–mouse α (clone HMα2), anti–mouse α (PS/2), anti–mouse α (clone HMα5-1), anti–mouse α (clone GoH3), anti–mouse α (clone RMV-7), anti–mouse β (clone 9EG7) integrin, and anti–mouse CD98 (clone H202-141) were purchased from BD Biosciences and used at the recommended concentrations. Goat FITC-conjugated anti–rat IgG and goat FITC-conjugated anti–hamster IgG were obtained from Biosource International and were used as secondary antibodies for α, α, α, and β; murine CD98; and α, α, and α detection, respectively. Fn 9–11 binding was assayed by two-color flow cytometry as previously described () in the presence or absence of 10 μg/ml of activating anti-β1 mAb 9EG7. This assay was performed by modification of a published method (). In brief, WT or CD98hc-deficient MEFs were harvested with trypsin-EDTA (Invitrogen), quenched with complete medium, washed twice with PBS one time, and resuspended at 8.5 million cells per milliliter in serum-free DME (Invitrogen) buffered with 25 mM Hepes. 350 μl of this cell suspension was added to 7 × 45 mm siliconized glass cuvettes (Sienco, Inc.). Then, 200 μl of human platelet–poor plasma anticoagulated with ACD (85 mM sodium citrate, 65 mM citric acid, and 104 mM glucose) was added, followed by 200 μl Hepes-buffered DME containing 28 mM CaCl and 5 U/ml human thrombin (Sigma-Aldrich). Cuvettes were incubated at 37°C with 5% CO. Images were acquired with a digital camera at 1 or 2 h, and subsequently the 2D area of the clot was measured using ImageJ software. The percentage of clot contraction was calculated according to the following equation: percentage of clot contraction (t = 1 h) = 100 ([area at t = 1 h/area at t = 0] × 100). Polyacrylamide gels with embedded fluorescent beads on coverslips were prepared using previously described protocol () with some modifications (see the supplemental text, available at ). Cell culturing, image acquisition, and strain map construction were performed as follows. WT and CD98-deficient MEFs were harvested with trypsin-EDTA (Invitrogen), quenched with complete medium, washed twice with 1× PBS, and resuspended at 0.1 million cells per milliliter in serum-free DME (Invitrogen) buffered with 25 mM Hepes. Cells were kept in suspension for 1 h at room temperature, and 20,000 cells were plated on a polyacrylamide sheet. After culturing on a polyacrylamide sheet for 2 h, images of both cell types were acquired within 10 min from each other. A microscope (TE2000; Nikon) equipped with environmentally controlled enclosure (37°C and 5% CO), 60× 1.2 NA water objective (Olympus), motorized stage (Ludl), excitation filter wheel (Ludl), filter set (Semrock), and camera (CoolSnap HQ; Roper Scientific) was used to acquire images of fluorescent beads embedded in polyacrylamide and brightfield images of cells in six different fields. After addition of 200 μl RIPA buffer to 2 ml of media, cells detached and new images of the same fields were taken to collect reference data on beads' location in nonstressed gel. Images of fluorescent beads were analyzed using image analysis software written in Matlab (courtesy of G. Danuser's and C. Waterman-Storer's groups, The Scripps Research Institute, La Jolla, CA). Images were divided to 1.6-μm areas, and displacement vectors between were calculated (). Vectors (, green) on the strain map were increased five times for visualization purposes. Brightfield and strain map images were superimposed using Photoshop (Adobe). Integrated strains were calculated by summing up the magnitudes (WT, 7,400 pixels; and CD98hc null, 1,444 pixels) of the strain vectors inside the cell mask and dividing them by the cell area (WT, 2,945 arbitrary units; and CD98hc null, 1,873 arbitrary units). WT and CD98hc-null MEFs were grown in 0.5% serum for 24 h and then in serum-free medium for another 17 h. Cells were detached and kept in suspension at room temperature for 1 h in serum-free medium. Serum-starved cells were then plated on 3D Fn plates (3 million cells/plate), and RhoA activity was assayed at the indicated time points. For LPA treatment, cells were grown and serum starved as described, 1 μg/ml LPA was added to each plate, and RhoA activity was measured after a 5-min incubation. 3D Fn matrix was prepared as described previously () using NIH3T3 cells cultured in a 10-cm Petri dish. RhoA activity was measured using a commercially available ELISA-based assay (G-LISA; Cytoskeleton, Inc.) according to the manufacturer's protocol. Lysates were also resolved by SDS-PAGE and immunoblotted with rabbit anti-RhoA (67B9) antibody (Cell Signaling), followed by a IRDye 800CW goat anti–rabbit IgG (LI-COR Biosciences), and bands were visualized by scanning blots using an infrared imaging system (Odyssey; LI-COR Biosciences). Fig. S1 shows that CD98hc contributes to the integrin-dependent activation of FAK and p130 via its integrin binding domain by reconstitution experiments. The supplemental text provides additional methodological details for the cellular tension measurements and for the assessment of tyrosine phosphorylation. Online supplemental material is available at .

xref italic #text Microtubule dynamics rely on the ability of β-tubulin to bind and hydrolyze GTP (). In an effort to obtain strains that have altered microtubule dynamics, we changed to alanine each of 12 amino acids in yeast β-tubulin (Tub2) that interact directly with GTP (; ). Each of these alleles was used to replace one of the copies of in a diploid yeast strain (). The diploids were then sporulated to obtain haploid segregants. Seven of the haploid strains containing only the mutated allele of were viable (). Four of these displayed altered sensitivity to the microtubule-destabilizing drug benomyl, but only one, , showed a substantial change in microtubule dynamics. The dynamics of individual cytoplasmic microtubules were measured in live yeast cells expressing GFP-tagged Tub1, the major α-tubulin in yeast ( and ). Microtubules in cells spent the majority of their time in the paused state neither growing nor shrinking to any detectable degree (83% in G1 cells and 65% in preanaphase cells). In contrast, microtubules in cells paused <10% of the time. In addition, rates of microtubule growth and shrinkage were two- to threefold lower in cells. Overall, the mutation reduced cytoplasmic microtubule dynamicity by 18- and 7-fold in G1 and preanaphase cells, respectively. Because individual kinetochore microtubules cannot be visualized in yeast by fluorescence microscopy, we used FRAP to assess their dynamics (). Half of a preanaphase spindle labeled with GFP-Tub1 was selectively photobleached. Then, the fluorescence intensities of both the bleached and unbleached half spindles were measured over time. From these values, we calculated the extent of redistribution and a time to half-maximal redistribution (t), assuming that redistribution was caused by a first-order kinetic exchange of fluorescent molecules between the two half spindles. In cells, ∼90% of the fluorescence redistributes with a t of 48 s (). In cells, only ∼40% of the fluorescence redistributes with a t of 109 s. Thus, the mutation severely decreases the dynamics of kinetochore microtubules. The structure of yeast tubulin has been derived as a homology model from the solved structure of bovine brain tubulin (). V169 of yeast β-tubulin is predicted to contact the ribose moiety of bound GDP (). We hypothesize that the V169A mutation lowers microtubule dynamics by altering the binding and/or hydrolysis of GTP. Another mutation in , , has also been reported to severely decrease microtubule dynamics (). C354 is located at the intradimer contacts () and may affect interactions within or between protofilaments (). We measured the effect of the mutation on microtubule dynamics in our strain background. The results confirmed that this mutation leads to dramatic decreases in cytoplasmic and kinetochore microtubule dynamics that are similar to the effects of ( and ). Any differences between our results and those previously reported for () are likely the result of variations in strain background and methods of measurement. We decided to use both and to characterize the roles of microtubule dynamics in mitosis. Given that these mutations likely work through different mechanisms, we should be able to attribute any common phenotypes to their similar effects on microtubule dynamics. Both and strains grow slowly: generation times are 280 and 220 min, respectively, versus 90 min for cells. Cultures of and cells contain ∼60% large-budded cells, and ∼60% of these contain an ∼1.5-μm preanaphase spindle. This high percentage of preanaphase cells is indicative of spindle assembly checkpoint activity that responds to defects in chromosome attachment to the mitotic spindle. To examine chromosome attachment in and cells, we visualized kinetochores using Mtw1-3GFP and spindle pole bodies (SPBs) using Spc42-RFP. In preanaphase cells, sister chromatids are attached to microtubules originating from opposite poles, which is a configuration referred to as bipolar (amphitelic) attachment. Tension exerted by the microtubules pulls sister kinetochores and their associated centromeric DNA (centromere [CEN]) toward the two spindle poles (). Thus, in >75% of preanaphase cells, Mtw1-GFP has a bilobed appearance with a region of staining adjacent to each spindle pole (, top). However, in and mutants, only 20–30% of preanaphase cells show this bilobed Mtw1-GFP configuration. In ∼70% of these cells, Mtw1-GFP is distributed in a disorganized fashion along the spindle (, middle and bottom). In addition, a small percentage of the mutant cells have extra Mtw1-GFP foci located off the spindle and presumably denoting unattached kinetochores (, bottom). These results indicate that the mutants have difficulty establishing proper kinetochore attachments to the spindle. To gain a more precise measure of kinetochore localization, we visualized a single CEN by integrating a LacO array 1.2 kb from in cells expressing LacI-GFP. SPBs were visualized by expressing Spc42-RFP. 81% of preanaphase cells contained two GFP dots representing the separation of sister CENs as a result of tension on the bioriented chromosomes. In contrast, 60% of the cells and 56% of the cells contained unseparated CENs, indicating monopolar attachment or bipolar attachment with a lack of tension. To distinguish between these possibilities, we examined the positions of unseparated CENs relative to the spindle. Monopolar CENs will remain close to the spindle pole to which they are attached, whereas bipolar CENs are more likely to reside in the middle of the spindle (; ). In cells, only 14% of unseparated CENs are located closer to one SPB than they are to the middle of the spindle (). In contrast, in 58% of and 62% of cells, unseparated CENs are located closer to one SPB. In each mutant, these numbers include ∼10% of unseparated CENs that reside on the far side of one SPB relative to the spindle, which is a position inconsistent with biorientation. Overall, using SPB proximal localization of unseparated CENs as a criterion for monopolar attachment, has a monopolar attachment in 35% of preanaphase and cells versus 3% of cells. Next, we examined the movement of unseparated CENs by time-lapse microscopy. In wild-type cells, CENs transiently separate as they oscillate along the spindle. However, mutants, which form primarily monopolar (syntelic) attachments, contain unseparated CENs that remain close to one SPB (). In contrast, in mutants, which form bipolar attachments lacking tension, unseparated CENs are found near the spindle midregion (). As expected, we found that unseparated CENs in cells moved back and forth along the spindle (). In 12/14 of the cases, CENs that were unseparated at the beginning of the recording separated, at least transiently, during the 10-min period of observation. Both the midspindle location and transient separation indicate bipolar attachment. In contrast, CEN separation was observed in only 4/15 cells and 3/15 cells during the 10-min periods of observation. In 6/11 cells and 8/12 cells that did not separate CENs, the unseparated CENs remained close to one SPB. Thus, time-lapse imaging indicates that 40 and 53% of the unseparated CENs in preanaphase and cells, respectively, have a monopolar attachment. An additional assay for monopolar attachment of chromosomes is to measure CEN segregation in the absence of the spindle assembly checkpoint. Normally, the presence of this checkpoint keeps cells with monopolar chromosomes from entering anaphase. In its absence, monopolar chromosomes will missegregate at anaphase. If both sister CENs are attached to one SPB (syntelic attachment), both will segregate with this SPB into one of the daughter cells. If only one of the sister CENs is attached to the SPB (monotelic attachment), random segregation of the unattached CEN will cause both CENs to segregate into one of the daughter cells about half of the time, although the unattached CEN would not likely reside near the SPB. On the other hand, bipolar attachment of CENs (amphitelic attachment), even the absence of tension, would not result in missegregation. We eliminated the spindle assembly checkpoint by the overexpression of Cdc20 from the promoter (). After 3 h of induction with galactose, sister CEN segregation was examined in anaphase cells. In 94% of anaphase cells, sister CENs segregated correctly into the two daughter cells (). In and cells, sister CENs segregated correctly in only 57 and 53% of cells, respectively. In all cases of missegregation, both CENs were located close to the SPB, suggesting syntelic attachment. In summary, the examination of position and movement indicates that at least 25% of the and preanaphase cells contain monopolar attachments. Assuming that each of the 16 yeast CENs behaves similarly to , each preanaphase cell contains on average four monopolar attached chromosomes. Although the mutants biorient chromosomes much less efficiently than cells, these cells are viable because the spindle assembly checkpoint holds them in mitosis until the biorientation of all chromosomes has been achieved. Thus, preanaphase cells in asynchronously growing cultures represent a mixed population; those that have been arrested in mitosis the longest will likely contain the fewest monopolar chromosomes. As expected, in the absence of the checkpoint, these cells missegregate chromosomes at a high rate. Assuming that chromosome IV is representative, about half of the 16 yeast chromosomes missegregate in each mitosis. To assess the rate of biorientation, we arrested cells at the start of the cell cycle with α factor and then allowed them to proceed synchronously into mitosis. The cells were held in metaphase by blocking the transcription of at the time of release from α factor. The gene was under the control of the promoter, allowing it to be shut off by the addition of methionine. These strains also contained GFP-labeled (in this case, a TetO array was integrated near in cells expressing TetR-GFP) and YFP-Tub1. The cells formed preanaphase spindles ∼50 min after release from α factor. At this time, >75% of the preanaphase spindles contained bioriented chromosomes, as indicated by the separation of CEN dots (). This value remained relatively constant over the next 90 min. In and cells, only 21 and 15% of the preanaphase cells, respectively, contained separated CENs at the 50-min time point. This value increased steadily over the next 50 min but leveled off at ∼55% after 100 min. To examine chromosome capture and biorientation directly, we used a system developed by . In this system, the promoter is placed adjacent to . Transcription through the CEN interferes with kinetochore assembly; thus, can be conditionally activated or inactivated by switching between media containing glucose and galactose. This is also marked by TetO/TetR-GFP. The cells again have the gene under the control of the promoter and express Tub1-YFP. To visualize the capture of by an individual microtubule, cells were synchronized with α factor and released into medium containing methionine and galactose for 3 h. This caused cells to arrest in metaphase with inactivated generally located away from the spindle. The cells were then switched to glucose medium to reactivate . The kinetics of capture and biorientation were determined by fixing aliquots of cells at different time points and determining the fraction of cells with free , on a captured microtubule, on the spindle but not separated, and on the spindle and separated. In cells, was captured in ∼90% of the cells within 10 min of its reactivation (). In contrast, was captured in <50% of and cells in 10 min. Even after 60 min, 24 and 18% of the and cells, respectively, contained free . In addition, captured CENs became bioriented much more slowly in the mutants. For cells, the ratio of separated to unseparated spindle-associated CENs increased steadily to ∼4.5 in 40 min (, bottom). At the same time, this value was 1.1 for cells and 1.2 for cells. Finally, we used live cell imaging to measure the time between the arrival of on the spindle and biorientation as indicated by its separation into two dots along the spindle (). In cells, this time interval was 163 ± 98 s ( = 14). For 9/12 cells observed, the time interval for CEN separation was 533 ± 275 s. In the other three cells, CENs failed to separate within periods of observation averaging 1,030 s. For 6/10 cells observed, the time interval for CEN separation was 725 ± 302 s. In the other four cells, CENs failed to separate during a mean period of observation of 900 s. The slow rates of initial CEN capture in this experiment might appear to be at odds with our aforementioned results indicating that most CENs in the mutants are attached to at least one SPB. However, kinetochores likely maintain microtubule attachments throughout the cell cycle, eliminating the need for their initial capture early in mitosis (; ). On the other hand, the slow rates of CEN separation agree with the low percentage of bioriented chromosomes in growing cultures of these mutant strains. The yeast strains used in this study are listed in . alleles were generated by site-directed mutagenesis using overlapping PCR () and were integrated into yeast as described previously (). All other constructs were made by plasmid integration or the one-step PCR method for gene modification (). Mtw1-GFP was a gift from S. Biggins (Fred Hutchinson Cancer Research Center, Seattle, WA). Images that did not involve time-lapse microscopy were obtained with a microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) equipped with a 100× plan-Apo NA 1.4 objective, camera (CoolSNAP fx; Photometrics), and Openlab software (Improvision). All images were captured using 2 × 2 binning except those in , which were not binned. Time-lapse studies in and were obtained with a spinning disk confocal imaging system (PerkinElmer) equipped with a microscope (TE2000; Nikon), a 100× plan-Apo NA 1.4 objective, camera (Orca ER; Hamamatsu), and UltraVIEW software (PerkinElmer) using 2 × 2 binning. Time-lapse images in were obtained using a confocal microscope system (TCS SP2; Leica) with a 100× plan-Apo NA 1.4 objective. The excitation wavelength was 488 nm; the GFP signal was collected at 493–518 nm, and the YFP signal was collected at 527–607 nm. All images were obtained at room temperature and are maximum intensity projections of z-series stacks (Δz = 0.5 μm). Linear contrast enhancement was performed using Photoshop CS (Adobe). Analysis of cytoplasmic microtubule dynamics () and FRAP of spindle microtubules () were performed as described previously. In the latter experiments, the extent of redistribution is defined as 1 − f. Videos 1–3 show the movement of CENs in , , and yeast, respectively. Videos 4–6 show CEN capture and biorientation in , , and yeast cells, respectively. Online supplemental material is available at .

xref italic #text For this study, we made kinesin-13 protein constructs that include the MD only or the MD and additional amino acids (called the neck) N-terminal to the MD (). A previous study has indicated that only the MD of kinesin-13 is necessary for microtubule depolymerization activity (), but the additional neck sequences have been shown to be important for microtubule interaction and efficient depolymerization (). We found that in the presence of AMPPNP, all the kinesin-13 constructs investigated form rings and spirals around microtubules (). Individual or clusters of kinesin-13 rings can be situated anywhere along the microtubule without apparent preference for the microtubule ends. These structures have not been reported in any previous structural study of the interactions of kinesins with microtubules. The rings are specific for kinesin-13s. Control experiments with conventional kinesin (human Kif5b, MD construct) performed under exactly the same conditions produced only regular microtubule lattice decoration (unpublished data), similar to what has been observed previously by many laboratories (). The fact that we found these rings with constructs from three different kinesin-13s from two different animal species strongly suggests that ring formation is a general characteristic of the kinesin-13 family. The minimal construct investigated in this work, the KLP10A MD-only construct, forms rings. Comparing constructs with and without the neck, we found that after mixing KLP10A MD + neck protein with microtubules (1:1 molar ratio kinesin/tubulin heterodimer), ∼20% of the microtubules have at least one ring. In similar conditions with the KLP10A MD-only protein, 80% of the microtubules have at least one ring and many have numerous rings (). Rings formed on either taxol- or guanosine-5′-([α,β]-methyleno)triphosphate (GMPCPP)–stabilized microtubules (). Thus, ring formation is independent of the nucleotide condition of the polymerized tubulin. Rings only formed in the presence of both microtubules and kinesin-13 proteins. However, ring-type structures are also found when unpolymerized tubulin is incubated with kinesin-13s (), suggesting that the observed rings may be oligomers of tubulin and kinesin-13 MDs. The rings formed with free tubulin appear as one or more concentric rings with an outside diameter of 42 ± 3.4 nm (mean ± SD; = 32). Occasionally, long regular spirals formed around the microtubule (). These spirals extend ∼12 nm from the microtubule surface (). In contrast, in the typical kinesin–microtubule complex, the microtubule-bound MD extends radially only ∼4 nm from the microtubule surface (). The spirals follow the shallow tubulin heterodimer helical path on the microtubule lattice (∼0.9 nm rise between adjacent protofilaments), creating an axial periodicity of 8 nm (), typical of microtubules decorated with kinesin proteins. This result suggests that the kinesin-13 MD is an integral part of the ring structure, likely to be involved in microtubule binding. However, unlike other kinesins, this interaction is not very tight or stereospecific, as rings are found at variable angles relative to the microtubule (). Immunogold labeling against kinesin-13 also forms ring-like structures on the microtubule, confirming that kinesin-13 is part of the ring (Fig. S1, available at ). The kinesin-13 spirals sometimes follow only one of the several tubulin helical path starts (1–3, depending on the number of protofilaments in the microtubule; ), forming an axial periodicity with a spacing that is a higher multiple of 8 nm (16 nm in the case shown in ). These isolated spirals indicate that the strongest interactions holding the spiral together are lateral (along the shallow tubulin helical path) and not axial (along the microtubule axis). shows a microtubule with 15 protofilaments. The spirals following the two-start tubulin helical paths are indicated by the yellow arrows. This type of microtubule is suitable for helical 3D reconstruction because of the lack of discontinuities or seams (). A lateral projection of a 3D reconstruction calculated from the microtubule of is shown in . To obtain further insights into the structure and mechanism of ring formation, we calculated a 3D reconstruction of spirals formed on 15-protofilament microtubules like those shown in (spiral formed by the KLP10A MD-only construct). shows surface representations of the calculated 3D density, color coded according to the radial position from the helical axis. The outermost structure (blue) forms a relatively continuous structure that resembles a tubulin protofilament. In end-on views (), the map outer region (blue and green) closely resembles end-on projections of kinesin-13 interacting with isolated protofilament rings (), indicating that the rings may be formed by kinesin-13 MDs interacting with an isolated tubulin protofilament. We investigated this possibility by fitting the atomic structures of a kinesin MD and the tubulin heterodimer into the 3D electron microscopy density map. For the fitting, we used the coordinates of the complex formed by a kinesin MD and the tubulin heterodimer in a microtubule (Protein Data Bank accession no. 1IA0). We found that a very good fit to each asymmetric unit in the 3D map can be obtained with two of these complexes (). For the fitting, only the relative position between the two complexes was changed, keeping constant the relative positions of the proteins within each complex. shows several orientations of the molecular model inside the 3D electron density map (transparent gray). The atomic structures fit very well into the electron density map, particularly in the outer parts of the rings (blue tubulin and green kinesin MD). The inner region of the model also fits relatively well with the density map, but a small electron density remains unaccounted for (visible in above and below the yellow kinesin MD). Also, the densities corresponding to α and β tubulin in the microtubule are different even though they are expected to be similar at the current map resolution (∼3 nm). Differences between the α and β tubulin have been observed previously in negatively stained specimens (), so this may represent a staining artifact. Despite these small discrepancies, the two kinesin MD–tubulin complexes oriented as shown in are in very good agreement overall with the 3D map of the kinesin-13 spiral complexes. The molecular model shown in has several noteworthy features. Contacts along the protofilament in the outside ring (blue) must stabilize the spiral because there are no contacts between adjacent axial levels. The innermost part of the ring is a kinesin MD (yellow) interacting with tubulin in the microtubule lattice (red) in the same configuration found in many kinesin–microtubule complexes. The contacts between kinesin and tubulin are similar in the outer part of the ring and in the microtubule lattice. Interactions between two kinesin MDs bridge the inner and outer ring regions. Features 1 and 2 nicely explain why the spirals follow the tubulin lattice helical path (see the previous section). Features 2 and 3 are consistent with previous structural work that has shown kinesin-13 MDs interacting with the microtubule lattice () or isolated protofilaments () in similar configurations. Feature 4 points to interactions between kinesin molecules as part of the mechanism leading to ring and spiral formation. These interactions are mediated by residues on the kinesin-13 MD away from the ones involved in the kinesin–tubulin interface. We observed rings in the presence of AMPPNP or the slowly hydrolyzable ATP analogue ATP-γ-S. We did not find rings in the presence of ATP, ADP, or ADP+AlF (used to mimic the ADP-Pi state). Thus, ring formation is favored specifically by the ATP-bound state. During steady ATP hydrolysis, the rings would be expected to be transient structures, unless the ATP-bound state is prolonged. Interestingly, the ATPase activity of kinesin-13s is stimulated preferentially by the microtubule ends but not the lattice (). Furthermore, recent work has shown that kinesin-13 in solution has a γ-phosphate–bound nucleotide (ATP or ADP-Pi) instead of ADP as in other kinesins (). Therefore, kinesin-13s, still in the ATP-bound state, could interact with each other on the microtubule lattice to form rings. If only the kinesin-13s at the very end of the microtubule are engaged in ATP hydrolysis and microtubule depolymerization (; ), then other kinesin-13s forming rings along the microtubule will be pushed as the depolymerizing end advances. In support of this idea, we have observed kinesin-13 accumulation at the depolymerizing end of microtubules in vivo (see the next section). shows an example of a depolymerizing microtubule decorated with EGFP-KLP59C in a live S2 cell. The fluorescence intensity at the tip of the depolymerizing microtubule increases steadily as the microtubule depolymerizes, whereas the intensities at points on the microtubule away from the end remain relatively constant (). Only when a bright punctum (containing many EGFP-KLP59C molecules) is released from the microtubule end (, green arrows) does the fluorescence at the depolymerizing end decrease. Protein accumulation at the depolymerizing end is not an obligatory event. Shortening microtubules in cells expressing EGFP-tubulin but not overexpressing kinesin-13s (so that the observed microtubules will have few or no kinesin-13s) showed no increase in fluorescence at the depolymerizing microtubule end (). We find that the depolymerizing end of EGFP-KLP59C increases in intensity by 79 ± 16% (mean ± SEM; = 8) per 1 μm of microtubule length decrease (excluding events in which a sudden drop in intensity was associated with the release of a bright punctum from the microtubule). In the case of EGFP-tubulin microtubules, the depolymerizing ends change fluorescence by −6 ± 3% (mean ± SEM; = 10) per 1 μm of microtubule length decrease. The difference between EGFP-KLP59C– and EGFP-tubulin–labeled microtubules was statistically significant (P < 0.01). These results show that KLP59C specifically accumulates at the depolymerizing microtubule end and slides along the tubulin lattice as the depolymerizing end advances. Another intriguing possibility is that the rings may be able to slide along the microtubule lattice like a loose sleeve. Our in vivo data with EGFP-KLP59C () support this possibility. Recently, the yeast Dam1–DASH kinetochore complex was shown to form rings around microtubules that work as movable sleeves. Based on these data, it was proposed that a Dam1–DASH sleeve at the kinetochore allows an associated chromosome to be pulled toward the spindle pole while the attached microtubule end is depolymerizing (; , ). However, homologues of the Dam1–DASH complex in higher eukaryotes have not been identified (). Mitotic KLP10A and KLP59C are located at spindle poles and kinetochores () and so are properly positioned to perform an analogous function to the yeast Dam1–DASH complex. Thus, an interesting possibility is that kinesin-13s in higher eukaryotes have the dual mitotic function of controlling microtubule depolymerization and forming a sleeve at microtubule ends. Further studies will be required to test these possibilities. The KLP10A MD + neck construct contained residues T198–I615 of the KLP10A amino acid sequence fused with a His tag at the N-terminal end in the pRSET B vector. The KLP10A MD-only construct contained residues 279–615 fused with a His tag at the N-terminal end in the pRSET B vector. The KLP59C MD + neck construct encoded residues V113–T539 of the KLP59C amino acid sequence fused with a GST tag at the N-terminal end. The recombinant plasmids were transformed into BL21 (DE3) host cells (Stratagene). Cells were grown to O.D. 1 (O.D. 0.6 for GST-KLP59C construct), and protein expression was induced by addition of 0.1 mM IPTG overnight at 20°C. The MCAK MD + neck construct consisted of residues 182–583 fused with a His tag at the C-terminal end in the pET30 vector. To purify His-tagged proteins, lysates from construct-expressing bacteria were clarified by centrifugation and the supernatant was applied to Ni-NTA agarose resin (QIAGEN). Further purification was performed on a HiPrep 16/60 Sephacryl S-200 size exclusion column (GE Healthcare). GST-KLP59C was purified using glutathione–Sepharose 4 Fast Flow (GE Healthcare), and the GST tag was cleaved by Presicion Protease (GE Healthcare). Pure proteins fractions were concentrated, aliquoted, and flash frozen. The KLP10A MD + neck constructs were incubated with microtubules in the presence of AMPPNP on carbon-coated electron microscope grids and then incubated with either of two primary and secondary gold labeled antibody pairs: a polyclonal rabbit antibody raised against the KLP10A N-terminal sequence (M1-A229) and 5 nm colloidal gold–labeled anti-rabbit IgG (GE Healthcare) or a mouse anti-His antibody (GE Healthcare) and 5 nm colloidal gold–labeled anti-mouse IgG (GE Healthcare). The grids were then negatively stained with 1% uranyl acetate and imaged in the electron microscope. Microtubules were polymerized from purified tubulin from bovine brain (Cytoskeleton) according to standard protocols. Taxol-stabilized microtubules were prepared as in . To increase the frequency of microtubules with 15 protofilaments, suitable for helical 3D reconstruction, some microtubules were polymerized in the presence of DMSO according to . GMPCPP-stabilized microtubules were prepared by incubating tubulin at 37°C for 30 min in BRB80 buffer (80 mM Pipes, pH 6.8, 2 mM MgCl, and 1 mM EGTA) supplemented with 2.5 mM GTP to form microtubules. Microtubules were then pelleted at 239,000 for 15 min at 28°C, resuspended in cold BRB80 buffer, and allowed to depolymerize on ice for 20 min. The solution was centrifuged at 239,000 at 4°C for 5 min to remove insoluble aggregates, and GMPCPP (Jena Bioscience) was added to 4 mM final concentration. After a 20-min incubation on ice, GMPCPP-tubulin was diluted to 4–5 mg/ml in BRB80 buffer and incubated at 37°C for 2 h. The resulting GMPCPP-stabilized microtubules were then spun down, and the pellet was resuspended in BRB80 buffer + 2 mM GMPCPP. The different kinesin-13 constructs were incubated with microtubules (∼3 μM tubulin and 1:1 to 1:2 molar ratio kinesin MD/tubulin) in BRB80 buffer and one of the following according to the experimental nucleotide condition: AMPPNP: 1 mM AMPPNP (Sigma-Aldrich); ATP-γ-S: 1 mM ATP-γ-S (Qbiogene); ATP: 1 mM ATP; no nucleotide: no nucleotides added; ADP: 1 mM ADP (Sigma-Aldrich); or ADP + AlF : 4 mM ADP, 2 mM AlCl, and 10 mM KF. Incubation was performed at room temperature for 20 min followed by ultracentrifugation (217,000 , 15 min, 30°C). The pellets were resuspended in BRB80 (+ 20 μM taxol and nucleotide according to the experiment) at room temperature and loaded on freshly glow-discharged 400-mesh carbon-coated grids for negative staining. In some cases, microtubules and kinesins were mixed and incubated directly on the grids. Rings on microtubules were observed with both methods of grid preparation. All experiments using GMPCPP-stabilized microtubules were performed by mixing kinesin and microtubules directly on the grid. For the experiments with unpolymerized tubulin, equivalent molar amounts of unpolymerized tubulin dimers (purified tubulin kept cold in BRB80 buffer + 1 mM GTP) and KLP10A MD-only proteins were mixed in BRB80 supplemented with 3 mM MgCl, 2 mM AMPPNP, and 1 mM GTP on ice. After a 1-min incubation on ice, the mixture was absorbed on the glow-discharged carbon-coated grid. Different amounts (3, 5, 10, and 20 μM) of tubulin and KLP10A motor proteins were tested. Rings were observed in all cases. Microtubules were not observed in any case. Rings were also observed when the KLP10A MD + neck construct was used. A 3D reconstruction was calculated using the standard Fourier-Bessel algorithm (; ). The software packages Suprim () and NIH ImageJ () were also used for preparing the images for the helical processing programs (low-pass filtering, reinterpolation, rotation, centering, and padding). Display of the calculated 3D map and manual fitting of atomic structures into the 3D map was performed using UCSF Chimera (). A long and straight 15-protofilament (3 μm) microtubule with a regular spiral formed by the KLP10A MD-only construct was selected for the reconstruction. The number of protofilaments was determined by the diameter of the microtubule and typical moiré pattern caused by the projection of supertwisted protofilaments.The microtubule image was reinterpolated down to a pixel size of 0.549 nm/pixel and low-pass filtered to eliminate frequencies beyond the first CTF zero (approximately at 1/2 nm). Layer lines were collected (Fig. S2, available at ), and a 3D map was calculated by Fourier-Bessel inversion. The final reconstruction included ∼600 averaged asymmetric units. Clear layer lines were visible up to a resolution of 1/3.2 nm. The dynamics of EGFP-KLP59C–decorated microtubules or EGFP–α-tubulin microtubules were observed in live Schneider S2 cells using fluorescence confocal microscopy. For the EGFP-KLP59C experiments, S2 cells were transiently transfected with the pMT/V5-HisC expression plasmid (Invitrogen) encoding EGFP fused with full-length KLP59C (). Stably transfected S2 cells with a plasmid encoding EGFP–α-tubulin in the pAc5.1/V5-HisB vector were used for the EGFP-tubulin experiments (). Time-lapse movies were acquired using an Ultraview spinning disc confocal microscope system (PerkinElmer). 1–4-μm z sections were obtained with a piezo-electric z-axis controller for 4D data collection (x, y, z, and time). Images were acquired at 1 s/frame for the EGFP–α-tubulin–expressing cells and 1.6–2 s/frame for the EGFP-KLP59C–expressing cells. In both cases, the spatial resolution was 0.129 μm/pixel. Microtubules undergoing depolymerization at the periphery of the cells were chosen to measure their fluorescence intensity. Only microtubules that were clearly separated from other microtubules were chosen for analysis. The mean intensity in two regions (3 × 3 pixels; 0.387 × 0.387 μm) on each microtubule was measured using NIH ImageJ. The mean intensity of two other regions adjacent to these but outside the microtubule were also measured. These background intensities were subtracted from the intensities in the microtubule region to yield the mean intensity (minus background) on the microtubule. The microtubule position and its end were tracked manually on each video frame. For each microtubule end, we calculated the percentage change of fluorescence per depolymerization length aswhere I and I are the mean intensities at the first and last frame, respectively, within the analysis interval. ΔLength is the change in microtubule length between the first and last interval frame. In the case of EGFP-KLP59C–decorated microtubules (where bright puncta were often seen releasing from the microtubule as shown in ), the image sequence was divided into intervals. The end of each interval was defined as the frame before puncta detachment. The change in fluorescence for a microtubule divided into two or more intervals was calculated as the weighted mean of all intervals (weighted by the number of frames in each interval). Data plotting and statistical tests were done using Prism4 (GraphPad Software, Inc). Fig. S1 shows anti–kinesin-13 immunogold labeling. Fig. S2 shows the layer line dataset used to generate the 3D reconstruction. Online supplemental material is available at .

Tetraspanins form a family of small proteins that are expressed in virtually all cell types and tissues (; ). They consist of short intracellular termini, four transmembrane domains, and one small and one large extracellular loop that contain two highly conserved cysteine motifs. Tetraspanins oligomerize into tetraspanin-enriched microdomains, in which they associate with integrins, Ig superfamily members, growth factor receptors, and proteoglycans. Tetraspanin-enriched microdomains modulate diverse cellular activities, such as adhesion strengthening, migration, signal transduction, and proliferation (). The importance of proper tetraspanin function is demonstrated by several human diseases; distinct mutations in () and () cause X-linked mental retardation and retinal dystrophy, respectively (; ). A nonsense mutation in () leads to end-stage hereditary nephropathy associated with pretibial epidermolysis bullosa and sensorineural deafness (). CD151 is expressed in epithelia, endothelia, muscle cells, renal glomeruli, proximal and distal tubules, Schwann cells, platelets, and dendritic cells (; ). Extensive biochemical studies have shown that CD151 is the primary tetraspanin associated with the laminin-binding integrins α3β1, α6β1, α6β4, and α7β1 (, ; ; ). The interaction of CD151 with α3β1 is particularly strong and occurs at high stoichiometry (). Integrins are αβ heterodimeric cell surface proteins that dynamically link the extracellular matrix and/or adjacent cells to the intracellular cytoskeleton (; ). In epithelial cells, the integrins α3β1 and α6β4 are mainly present in the basolateral compartment, where they bind to the basement membrane component laminin-5. Although much more severe, the phenotypes associated with mutations in , , and share features with the phenotype of patients with truncated CD151, indicating that the CD151–α3β1 and CD151–α6β4 heterotrimers are functionally important. Mice that lack the β4 subunit suffer from extensive detachment of the epidermis, and patients without functional α6β4 display junctional epidermolysis bullosa (; ). Mice without α3 exhibit the mild skin blistering associated with ruptured basement membranes and die shortly after birth because of severe abnormalities in the epithelia of lung and kidney (; ). In the renal glomerulus, podocytes are anchored to the glomerular basement membrane (GBM) via α3β1 and dystroglycans (). The interdigitating foot processes (FPs) of podocytes are connected by glomerular epithelial slit diaphragms (GESDs) consisting of nephrin, podocin, P-cadherin, and other proteins, which are linked directly or indirectly to the cytoskeleton. Both disturbed podocyte–GBM anchoring and podocyte–podocyte interaction at the level of GESDs lead to the loss of FPs, a dysfunctional filtration barrier, and, ultimately, to glomerulosclerosis and renal failure (). We report the generation of knockout mice for that show severe renal failure caused by progressive abnormalities of the GBM, loss of podocyte FPs, glomerulosclerosis, and cystic tubular dilation. Furthermore, we show that mice with a targeted deletion of the α3 subunit in podocytes have a similar, although more severe, phenotype. -null mice were generated (Fig. S1, available at ), born at the expected Mendelian ratio (0.28 : 0.49 : 0.23 ; = 86), and appeared to be healthy and normal at first observation. mice by SDS-PAGE revealed that all mice developed proteinuria before 3 wk of age, which is indicative of kidney dysfunction (). Both the onset and the degree of proteinuria were variable. Nevertheless, all knockout mice had to be killed before the age of 9 mo because of substantial loss of body weight. mice that increased with age and reached a plateau at 6 wk (3 mo follow up; unpublished data). mice (). littermates also varied considerably (), and there were animals with mildly or severely affected kidneys in the same litter. Histological examination of the mildly affected kidneys showed focal glomerulosclerosis, and interstitial fibrosis and inflammation (). The GBM of some capillary loops was abnormally thick and formed extensive spikes (). EM revealed that the GBM was laminated and that FPs in contact with the abnormal GBM were effaced (). On the vascular side, the endothelium was swollen and fenestrations were occasionally lost (). Severely affected kidneys were contracted and their capsules granulated because of cortical degeneration (). Light microscopy showed extensive glomerulosclerosis in several stages, tuft adhesions to Bowman's capsule with extracapillary cell proliferation and fibrosis, and marked expansion of the mesangial matrix. Proximal tubuli were either dilated and contained PAS-positive protein casts or they displayed degeneration of varying severity (). Furthermore, we observed periglomerular fibrosis and focal interstitial inflammation in close proximity to interlobular blood vessels (unpublished data). Silver staining showed the glomeruli to be segmentally or globally sclerosed with extensive deposits of basement membrane components (), an observation that could be confirmed by ultrastructural analysis (unpublished data). mice to EM. mice already have a laminated GBM, it seems that spike formation does not occur until the mice are 1-wk-old (unpublished data). Together, these observations suggest that the mild abnormalities found in some of the mice represent early stages of the severe phenotype in the other mice. Tubular changes may be secondary to massive glomerular protein leakage, but may also reflect dysfunction of the tubules themselves. To investigate whether the progressive GBM abnormalities are correlated with glomerular injury, we stained for tenascin-C and fibulin-2. Both proteins have been shown to be up-regulated in response to glomerular and vascular lesions (; ). We observed an increased glomerular expression of tenascin-C in both mildly and severely affected kidneys, but fibulin-2 was up-regulated only in the latter (). Whereas fibulin-2 is important for the migration of smooth muscle cells, tenascin-C regulates migration of a variety of cell types, including fibroblasts (; ). The finding that tenascin-C is already present in the early stages of the disease suggests that fibrosis precedes the pathology of the vasculature. mice is thickened, staining for nidogen and laminin-10 revealed an abnormally strong presence of these GBM components in peripheral capillary loops and extracapillary spaces ( and P–R). To exclude an arrest of the normal developmental switch from α1.α1.α2 (IV) collagen to the α3.α4.α5 (IV) and α5.α5.α6 (IV) collagen networks, as seen in the X-linked form of Alport's syndrome (), we stained for all six chains of collagen IV. The results showed the mature collagen IV pattern to be present in all glomeruli (unpublished data). Only the α2 (IV) chain was strongly up-regulated in podocytes of the severely affected kidneys (). We also checked the integrity of the filtration barrier by staining for the GESD components podocin and nephrin ( and U–W). Both proteins appeared to be down-regulated in the mildly affected kidneys, and are almost absent in the severely affected kidneys, demonstrating a complete loss of GESD architecture (). Together, these results support the hypothesis that in -null mice disorganization of the GBM precedes the effacement of FPs and the loss of GESDs. mice by histological and immunofluorescent analysis and measurements of auditory brainstem responses (ABR), respectively (Fig. S2, available at ). The highly stoichiometric binding of CD151 to the α3β1 integrin, and the fact that mice lacking the α3 subunit show severe renal defects, led us to hypothesize that the absence of CD151–α3β1 complexes is responsible for the renal pathology seen in our -null mice. In knockout mice, glomerular capillary branching is impaired, leading to a decreased number of capillary loops. Furthermore, podocytes fail to form normal FPs and lose lateral cell junctions (). To study the deletion of after glomerular capillary branching, and to investigate possible similarities with the phenotype in our -null mice, we generated conditional knockout mice and crossed them with 2.5P-Cre mice that express the Cre recombinase under the control of the human podocin promoter (). As shown by immunofluorescence, α3 is, indeed, almost absent in the glomeruli of these 2.5P-Cre; mice (), indicating that this integrin subunit is mainly expressed by differentiated podocytes. 2.5P-Cre; mice show massive proteinuria starting within the first week of age (), develop abdominal edema when 5–6-wk-old and, subsequently, have to be killed. Structurally, the milky, discolored kidneys contain partially sclerosed glomeruli with a disorganized GBM and prominent protein casts in dilated proximal tubuli (). EM revealed a complete effacement of podocyte FPs in newborns (unpublished data) and, in addition, widespread lamination and protrusions of the GBM in 6-wk-old mice (). mice, i.e., up-regulation of tenascin-C and fibulin, strong staining of the GBM components nidogen and laminin-10, and down-regulation of the GESD proteins podocin and nephrin (). Notably, sclerosis of the glomeruli, up-regulation of the collagen α2 (IV) chain, and down-regulation of podocin and nephrin is much less prominent in the 2.5P-Cre; mice than in the knockout mice. 2.5P-Cre; mice do not develop structural or functional renal abnormalities (, , and ); neither do 6-wk-old compound heterozygotes (2.5P-Cre; ; ), which may suggest that a simultaneous reduction of α3 and CD151 does not result in renal defects (). However, we cannot prove that the expression of these proteins is actually reduced in the compound heterozygotes, nor can we exclude that renal pathology develops upon aging. Because -null mice can form normal GBMs with a regular FP pattern () and expression of both α3 and α6 in the glomeruli and tubules appeared to be normal (), we suggest that the function of α3 in development is not affected by the absence of CD151. mice. The filtration of plasma exerts considerable mechanical stress on the filtration barrier. Podocyte FPs, thus, have to withstand substantial mechanical forces. Indeed, CD151 appears to be involved in adhesion strengthening because adhesions mediated by CD151–α6β1 complexes tolerate stronger mechanical forces than those mediated by α6β1 alone (). A similar effect has been suggested for CD151–α3β1 complexes (). Thus, FPs without CD151 might not be able to withstand prolonged transcapillary pressure, a phenomenon that is also responsible for the renal manifestations in the Alport's syndrome after several years of life in patients with an abnormal GBM caused by collagen type IV mutations (). Epithelial cells may become partially detached, leading to a compensatory production of new basement membrane components. As a result, the GBM thickens and spikes are formed, as has also been described in membranous nephropathy (). Correct reassembly of basement membranes upon injury is impaired in the skin of -null mice (). If the absence of CD151 is similarly important for GBM repair, this vicious circle would indeed result in glomerular malfunction. As shown by ultrastructural and immunofluorescent analysis, glomerulosclerosis precedes the loss of GESD components, leading to the aforementioned renal pathology. The assumption that CD151 is important for maintaining glomerular architecture upon mechanical stress is in accordance with the observation that glomeruli develop normally and that abnormalities only occur after several weeks or months. Furthermore, it explains differences in the rate of progression among littermates, as the degree of intraglomerular hydrostatic pressure depends on several genetic and epigenetic factors. Differences in genetic background might also explain why -null mice that have been previously described did not show renal failure as observed in our mice (). In conclusion, we show that the renal manifestations in our mice lacking CD151 are similar to those in patients with a mutated . Our data support in vitro studies pointing to a role of CD151 in adhesion strengthening, thus, suggesting that CD151–α3β1 complexes have an essential function in vivo. Urine from mice younger than 3-wk-old were collected by applying gentle dorsal pressure to the caudal area of the animal. Older mice were placed in metabolic cages for 24 h. Samples were either analyzed by SDS-PAGE, followed by Coomassie Brilliant blue staining, or by competitive ELISA using the Albuwell M kit that was obtained from Exocell. Sections of kidneys were prepared, fixed for 1 d in EAF (ethanol, acetic acid, and formaldehyde), and stained with PAS-D, HE, or Jones' methenamine silver. Images were taken with PL APO objectives (10×/0.25 NA, 40×/0.95 NA, and 63×/1.4 NA oil; Carl Zeiss MicroImaging, Inc.) on an Axiovert S100/AxioCam HR color system using AxioVision 4 software (Carl Zeiss MicroImaging, Inc.). After fixation in Karnovsky buffer for 48 h, the material was post-fixed with 1% osmiumtetroxide, the tissue samples were block-stained with 1% uranyl acetate, dehydrated in dimethoxypropane, and embedded in epoxyresin LX-112. LM sections were stained with toluidine blue. EM sections were stained with tannic acid, uranyl acetate, and lead citrate, and then examined using a transmission electron microscope (Philips CM10; FEI). Images were acquired using a digital transmission EM camera (Morada 10–12; Soft Imaging System) using Research Assistant software (RvC). Rat mAbs used in this study were 4G6 against laminin-10 (provided by L. Sorokin, University of Münster, Münster, Germany), GoH3 against α6, MB1.2 against β1 (a gift from B.M. Chan, University of Western Ontario, London, Canada), LAT-2 against tenascin-C (), and 346-11A against β4 (Abcam). Y. Sado (Shigei Medical Research Institute, Yamada, Japan) provided the rat mAbs H11, H22, H31, RH42, M54, and B66 against the mouse collagen IV chains α1, α2, α3, α4, α5, and α6, respectively. Rabbit polyclonal antibodies (pAbs) directed against mouse nidogen and mouse fibulin-2 were generous gifts from T. Sasaki (Max Planck Institute for Biochemistry, München, Germany); pAbs against nephrin and podocin were from H. Holthöfer (University of Helsinki, Helsinki, Finland), and C. Antignac (Cochin Biomedical Research Institute, Paris, France). The pAbs against keratin 1 and 14 were purchased from BabCO. Immunization of New Zealand rabbits with the cytoplasmic tail of human α3A fused to GST and the peptide CKENLKDTMVKRYHQSGHEGVSSAVDKLQQEFH coupled to KLH (Pierce Chemical Co.) yielded pAbs 141742 against α3A and 140190 against CD151, respectively. Texas red– and FITC-conjugated secondary antibodies were purchased from Invitrogen. Tissues from adult mice were collected and embedded in cryoprotectant (Tissue-Tek O.C.T.; Sakura Finetek Europe). Cryosections were prepared, fixed in ice-cold acetone, blocked with 2% BSA in PBS, and incubated for 45 min with primary antibodies undiluted (LAT-2, GoH3, MB1.2, 4G6) or diluted 1:2 (H22), 1:50 (346-11A), 1:100 (anti–fibulin-2, anto-podocin, anti-nephrin, and 141742), 1:250 (anti-nidogen), and 1:300 (anti-keratin 1 and 14), followed by incubation with secondary antibodies diluted 1:200 for 45 min. Samples were analyzed at 37°C using a 63×/1.4 HCX PL APO CS objective on a TCS SP2 AOBS confocal microscope (both from Leica). Images were acquired using LCS 2.61 (Leica) and processed using CorelDRAW Graphics Suite 12 (Corel). Fig. S1 shows the targeting strategy and molecular analysis of recombinant embryonic stem cells and knockout mice. Fig. mice, as well as auditory hearing thresholds of wild-type, , and mice upon click and tone burst stimuli, as determined by ABR measurements. The Supplemental materials and methods describes the generation of transgenic mice, immunoblotting, and ABR measurements. Online supplemental material is available at .

Faithful segregation of chromosomes during mitosis requires a dynamic interaction between spindle microtubules and the kinetochore, which is a macromolecular complex that localizes at the centromere of mitotic chromosomes. The kinetochore was originally identified by EM as a trilaminar stack of plates that formed along the outer surface of the centromere region of each sister chromatid. As part of the centromere–kinetochore complex, the inner centromere is defined as the region between the inner plates of apposing sister kinetochores, and it is occupied largely by centromeric heterochromatin. Extending away from the surface of the outer plate is an electron-dense cloud that is termed the fibrous corona (; ). Early EM studies revealed that trilaminar kinetochore structure is only visible from late prophase until the end of mitosis, suggesting that the kinetochore undergoes an assembly/disassembly cycle during each mitosis (; ). Molecular studies over the past two decades have confirmed the notion that the centromere–kinetochore complex undergoes cell cycle–dependent changes. Currently, >100 proteins, many of which are evolutionarily conserved, have been reported to associate with the centromere–kinetochore complex in human cells (; ; ; ). Some of the proteins are constitutively associated with centromeres throughout the cell cycle, whereas others are only transiently detected at the centromere–kinetochore complex from late G2 to telophase. The cell cycle–dependent localization of proteins to the kinetochore is consistent with the EM data that showed kinetochores are assembled and disassembled during mitosis. Moreover, observations made in human cells showing that different proteins exhibit a distinct temporal order of appearance at the kinetochore suggested that the trilaminar plates may be assembled in a stepwise fashion. In human cells, CENP-A, -C, -H, and -I (hMis6) are constitutive centromere proteins that become part of the inner plate of the mature kinetochore. These proteins are therefore likely to participate in the earliest steps of kinetochore assembly, which is thought to initiate during G2, after centromeric chromatin have replicated. Although kinetochore structures are not visible at this time, proteins such as hZwint-1, BUB1, Aurora B, MCAK, and CENP-F begin to accumulate at the nascent centromere–kinetochore complex. Immediately after nuclear envelope breakdown, when trilaminar plates are first visible, a host of proteins that are important for microtubule binding and checkpoint control, including the dynein–dynactin complex, CENP-E, CDC20, MAD1, MAD2, BUBR1, hMPS1, hZW10, and hROD assemble onto the mature kinetochore (; ). Despite the fact that proteins exhibit a temporal order of assembly at kinetochores, the cumulative data does not support a single linear assembly pathway. It appears that kinetochore assembly in human cells, as well as in a variety of model organisms, is complex (; ; ; ; ; ; ; ). For example, CENP-I is a constitutive protein that specifies the assembly of CENP-F, MAD1, and MAD2, but not BUB1, BUBR1, hZW10, and hROD (). Interestingly, kinetochore localizations of CENP-F and MAD2 (as with CENP-E and BUBR1) were reported to depend on BUB1 (), although BUB1 localization is independent of CENP-I. This suggests that multiple pathways may be necessary to recruit some proteins to the kinetochore. Although there are numerous studies describing the relationships amongst selected kinetochore proteins, our current understanding of how proteins come together to construct the kinetochore trilaminar structure remains fragmentary (; ; ). No single study of a large set of proteins has been conducted to achieve a global view of kinetochore assembly. To integrate findings reported by different labs and to advance our understanding of how kinetochores are assembled, we examined the relationships amongst twenty human centromere–kinetochore proteins and their contributions toward their organization at the ultrastructural level in HeLa cells. We chose to examine the relationships amongst twenty of the best-characterized proteins as the first step toward mapping the pathways for kinetochore assembly. These proteins include the constitutive centromere proteins CENP-A, -B, -C, -H, and -I; the inner centromere protein Aurora B; the microtubule-interacting proteins dynein–dynactin complex (represented by p150); CENP-E, -F (), and MCAK; and the mitotic checkpoint proteins BUB1, BUBR1, MAD1, MAD2, hMPS1, hZW10, and hROD, as well as HEC1, NUF2, and hMis12. From this group, Aurora B was classified as a “chromosomal passenger protein” or “inner centromere protein” (), but its role in kinetochore–microtubule attachment, mitotic checkpoint signaling, and recruitment of kinetochore proteins (; ; ; ; ) led us to include it in our analysis. The relationships between the aforementioned proteins were established by depleting specific proteins by siRNA and examining the effects on the localization of other proteins. Representative results of RNAi-induced knockdown of specific proteins and the specificities of several new antibodies developed during this study are shown in Fig. S1 (available at ). summarizes our epistasis analysis of the 20 kinetochore proteins. A genetic interaction map () was constructed based on and other studies (; ; ; ; ; ). The proteins are organized based on their relative temporal order of appearance (top to bottom), as documented in previous studies (; ; ) or uncovered in this study; BUB1 and Aurora B, HEC1, and CENP-F, respectively, appear at prekinetochores around the same time during late G2 (unpublished data). The reader should refer to this map as we present the details of our experiments. A more comprehensive map may be obtained upon request. Our map shows that the centromere-specific histone H3 variant CENP-A occupies the top of a hierarchy that directs three major assembly pathways that are specified by CENP-C, -I, and Aurora B. The pathways are not linear, but contain multiple branches that intersect to form a network that defines the spatial and temporal relationships amongst the proteins that were examined. No direct physical interaction between proteins is implied, although that is certainly possible, as in the cases of CENP-H–HEC1–NUF2 and hMis12–HEC1–NUF2 (; ; ). The contribution of seven proteins to the ultrastructure of the kinetochore was examined by EM ( and Table S1, available at ). CENP-I was previously shown to specify the assembly of CENP-F, MAD1, and MAD2, but not the localization of BUB1, BUBR1, hZW10, and hROD (). As the localization of MAD1, MAD2, and hMPS1 to kinetochores was reported to depend on HEC1 (), we tested and found that HEC1 localization depends on CENP-I (). Examination of cells that differed in the amounts of CENP-I depletion showed that the fluorescence intensities between CENP-I and HEC1 exhibited a relationship near 1:1, which was held over a 20-fold range (). Consistent with the fact that HEC1 is part of a complex with NUF2 (; ), we found that the localization of NUF2 was also sensitive to CENP-I level (). These results are consistent with those reported by , who showed in chicken DT-40 cells that the localization of HEC1-GFP and NUF2-GFP at kinetochores requires CENP-I. We also found that the localization of hMPS1 at kinetochores was dependent on CENP-I (Fig. S2 A, available at ). Thus, CENP-I specifies the following assembly pathway: CENP-I → HEC1–NUF2 complex → hMPS1 → MAD1 → MAD2, which links the inner kinetochore with the outer kinetochore. The position of the HEC1–NUF2 complex in this pathway is consistent with our observation that HEC1 was first detected at kinetochores in late G2, before the appearance of hMPS1, MAD1, and MAD2 (unpublished data). Although we previously showed that the localization of CENP-F and p150 also depend on CENP-I (Fig. S2 B; ), our studies and those reported by others (; ) indicated that neither belongs to HEC1–NUF2 pathway. Nonetheless, CENP-F may regulate the kinetochore localization of the dynein–dynactin complex (). Thus, CENP-I specifies at least two separate assembly branches for recruiting CENP-F and the HEC1–NUF2 complex. EM studies showed that depletion of NUF2 from HeLa cells resulted in disorganized kinetochores with poorly defined outer plates (). We made similar observations, as >50% of the kinetochores (43 out of 83) examined in NUF2-depleted cells lacked trilaminar plates, but instead displayed a “fuzzy ball”–shaped structure with few, if any, bound microtubules ( and ). Consistent with Deluca et al.'s proposition that NUF2 may contribute to the end-on attachment of microtubules to kinetochore, we observed microtubules in longitudinal profile associated with the surface of the fuzzy ball–shaped kinetochore masses in NUF2-depleted cells (). Importantly, similar fuzzy ball structures or expanded kinetochore structures were detected in 30 out of 38 kinetochores scored in CENP-I siRNA–transfected cells (, ). Kinetochores with defects of different extent were observed even in the same cell (Table S1), which was attributed to the variability in the extent or timing of protein depletion mediated by siRNA. These EM data nonetheless provide further confirmation that CENP-I and NUF2 lie in the same assembly pathway and contribute to the formation of the higher order trilaminar plate structure. We also examined kinetochore structures in CENP-F–depleted cells (Fig. S3 A, available at ) to see if this portion of the CENP-I pathway contributed toward trilaminar plate formation. Consistent with light microscopy data from recent CENP-F siRNA studies (; ; ; ), microtubules were attached to most of these kinetochores. Kinetochores displayed a variety of morphologies, including an uncondensed fuzzy ball or pulled out and fibrillar appearance (Fig. S3 A, a), a trilaminar structure (Fig. S3 A, b), and a trilaminar structure associated with strands of material emanating from the kinetochore region and intertwined with the kinetochore-associated microtubules (Fig. S3 A, c). These observations suggest that plate development can proceed further without CENP-F than without NUF2, but is nevertheless compromised. CENP-I localization at kinetochores was reported to depend on CENP-A (). However, we found that cells with significantly diminished levels of CENP-A still displayed bright CENP-I staining (compare the left and middle columns in ). Therefore, we quantitated the intensity of CENP-A signals in interphase cells that were transfected with CENP-A siRNA and then compared them with the intensity of CENP-I staining at the same kinetochores. To eliminate experiment-to-experiment variations in absolute staining intensities, we normalized the signal intensity to the brightest CENP-A signal in mock-transfected cells. We observed that when the reduction in CENP-A was <10-fold, CENP-I intensities were clustered at 37 and 75% of the brightest CENP-I signal ( [middle column] and ). The twofold difference in signal intensity probably reflects the relative abundance of CENP-I in cells before and after replication of their centromeres. We independently confirmed that cells in G2 with duplicated, but unseparated, centromeres exhibited a twofold higher intensity of CENP-I staining compared with cells with unreplicated centromeres (unpublished data). Near-normal levels of CENP-I were also seen in mitotic cells whose CENP-A levels were reduced by <10-fold (unpublished data). A dramatic drop in CENP-I levels was seen only when CENP-A was reduced by >10-fold ( [right column] and [circled datapoints]). Fisher's exact test showed that CENP-I levels at kinetochores with CENP-A intensity below or above 10% of normal level are significantly different (P = 0.00029). When CENP-A is reduced by ∼20-fold, CENP-I levels can be reduced to as low as 3% of controls. These cells were difficult to find, and the precise level of reduction was difficult to determine, as we were approaching the limits of detection. In addition, the loss of other centromere markers (i.e., ACA staining) in cells with extremely low levels of CENP-A made it more difficult to identify kinetochores with certainty. Thus, we can only estimate that substantial reduction of CENP-I was achieved when there was a 10–20-fold reduction in CENP-A. We conclude that only 10% of the normal level of CENP-A is sufficient to assemble near normal levels of CENP-I to the kinetochore. In addition to the dependence on CENP-A, we found that the localization of CENP-I depended on CENP-H (), as reported for chicken CENP-I (). The localization of CENP-H also depended on CENP-I (Fig. S2 C), which is consistent with a recent study that found that they can form a complex with several other proteins (). hMis12 was reported to be a constitutive kinetochore protein that specified CENP-I localization independently of CENP-A (). However, examination of HeLa and immortalized normal hTERT-RPE1 cells showed that hMis12 is not constitutively localized to kinetochores, as its signals started to decline during late anaphase and were not detectable (<10% of metaphase signals) in late telophase or early G1 cells (). This pattern was confirmed by real-time analysis of cells expressing GFP-hMis12 (unpublished data). Thus, hMis12 localization to kinetochores is cell cycle dependent, although its pattern is unique amongst the transient kinetochore proteins. Detailed characterization of hMis12 dynamics will be reported elsewhere. Of relevance to this study, we found normal levels of CENP-I at kinetochores that were devoid of hMis12 (). This observation demonstrates that the presence of hMis12 is not a prerequisite for localization of CENP-I to interphase centromeres. In mitotic cells, kinetochores depleted of hMis12 did exhibit a slight reduction in the level of CENP-I staining (). Quantitative analysis showed that even when hMis12 levels were reduced by 10-fold, there was only a twofold reduction in the level of CENP-I at kinetochores. The relationship was puzzling until we discovered connections between hMis12, HEC1, and CENP-I. When HEC1 was depleted from kinetochores by 10-fold, CENP-H and -I levels were reduced by ∼40% ( and not depicted). This suggests a negative feedback loop between HEC1 and CENP-H/-I. As depletion of hMis12 was shown to reduce HEC1 at kinetochores (; ), the twofold reduction of CENP-I in hMis12-depleted cells is likely to be indirectly caused by the loss of HEC1 from kinetochores. Regardless, the combined data suggest that hMis12 is not essential for kinetochore localization of CENP-I in interphase or mitosis. The localization of hMis12 to kinetochores in both and human cells was reported to be independent of CENP-A (; ). In contrast, Mis12 has been placed downstream of CENP-A (). Our finding that hMis12 is not constitutively localized to kinetochores led us to reexamine its relationship with CENP-A. As shown in , localization of hMis12 was clearly affected by the loss of CENP-A. Quantitative analysis showed an ∼2–3-fold reduction of hMis12 when CENP-A levels were reduced to between 5- and 10-fold (). This moderate reduction may be caused by the possibility that there is a separate pool of hMis12 whose localization at centromeres is specified by HP1α and HP1γ (; ). It is also possible that kinetochores in HeLa cells can tolerate a 10-fold reduction in CENP-A, and thus, a >10-fold depletion of CENP-A would be required to see significant loss of downstream proteins. As CENP-C lies downstream of CENP-A (), we tested and found that depletion of CENP-C resulted in a greater than fivefold reduction of hMis12 from kinetochores (). Thus, hMis12 localization can be specified by CENP-A and lies downstream of CENP-C. We next examined other proteins whose assembly at kinetochores depended on CENP-C. As CENP-C and -I specify separate pathways (), we focused on proteins that did not depend on CENP-I (). Cells depleted of CENP-C exhibited severe chromosome missegregation and micronucleation that are fully consistent with those reported for antibody injection experiments (). Kinetochores depleted of CENP-C were found to lack nearly all the transiently localized kinetochore proteins except the inner centromere protein Aurora B. Thus, BUB1, BUBR1, hROD, hZW10, and CENP-E all failed to assemble onto kinetochores that lacked CENP-C (; Fig. S4 A, a, available at ; and ). HEC1 and p150, whose localization at kinetochores depends on CENP-I, were also found to depend on CENP-C (; Fig. S4 B). We next examined the relationships amongst the proteins that lie downstream of CENP-C. BUB1 was given special attention, as there were conflicting reports about its checkpoint and recruitment functions (; ; ). We confirmed that BUB1 is, indeed, essential for the mitotic checkpoint, as microinjection of BUB1 antibodies or transfection of BUB1 siRNA prevented HeLa cells from delaying mitosis in response to nocodazole (unpublished data). As for its role in recruiting proteins to kinetochores, we found that BUBR1, MCAK, CENP-F, and CDC20 depended on BUB1 for localization, whereas hMis12, hROD, HEC1, hMPS1, MAD1, MAD2, CENP-E, p150, and Aurora B did not (; ; and Fig. S4, C–G). Within this latter group, we found that hMis12 specified kinetochore localization of HEC1 ( and Fig. S4 B). However, hROD and CENP-E were not dependent on either BUB1 or hMis12 for their kinetochore localization (Fig. S4 E and not depicted). The combined data suggest that CENP-C specifies three or four subbranches (). EM analysis showed that all cells depleted of CENP-C ( = 6) contained kinetochores with discernible laminar plate structure ( and Table S1). However, they were usually deformed, as they appeared smaller (4/50; ), pulled away from the underlying centromere heterochromatin (8/50; ), or exhibited thinner outer plates that often displayed a beaded appearance (17/50; ). Microtubules were usually absent from these kinetochores, and prominent fibrillar extensions were sometimes observed extending from the outer plates (). Both the kinetochore morphology and microtubule pattern observed in these cells were comparable to those seen in cells injected with antibodies to CENP-C (), but were distinct from kinetochores depleted of CENP-I. EM analysis of kinetochores in BUB1- or hMis12-depleted cells did not recapitulate all the ultrastructural defects found in CENP-C–depleted cells. For example, we did not find kinetochores with smaller plates. However, kinetochores with plate structure but no microtubule binding and kinetochores with pulled out, thinner, or punctate plates were observed (Fig. S3, B–C, and ). In some cells depleted of BUB1 or hMis12, we noticed undercondensed chromatin, either in centromere regions or elsewhere along the chromosomes (Fig. S3, B [a] and C [b]). The reasons for this are unclear, but we rarely observed such changes in CENP-C– depleted cells. Despite these discrepancies (see Discussion), our observations suggest that some of the ultrastructural anomalies resulting from the loss of CENP-C may be attributed to the disruption of the BUB1 and hMis12 branches. Aurora B and the microtubule depolymerase MCAK are both concentrated at the inner centromere from late G2 until metaphase (). Both the localization and enzymatic activity of MCAK are regulated by Aurora B kinase (; ). Although MCAK was reported to interact with CENP-H in vitro (), neither Aurora B nor MCAK localization was affected when CENP-H or -I was depleted from cells ( and not depicted). Furthermore, Aurora B or MCAK localization was not dependent on CENP-C ( and not depicted). Conversely, the localization of CENP-I and -C also did not depend on Aurora B (). We found, however, that the localization of Aurora B was dependent on CENP-A (). Therefore, CENP-A specifies a third assembly branch that specifies the localization of Aurora B and MCAK to the inner centromere (). Consistent with its reported roles in chromatin condensation and kinetochore–microtubule interactions (; ; ; ; ; ), two major types of kinetochore defects were observed in cells depleted of Aurora B (). The first type showed an expanded, C-shaped outer plate accompanied by undercondensed subjacent chromatin and no discernable inner plate (). The second type of defect displayed both inner and outer plates that appeared more electron dense, and the outer and inner plates often appeared to be fused at one end (). The chromosomes in these cells also appeared undercondensed, similar to those seen in BUB1-depleted cells (Fig. S3 B; Table S1). We have performed an epistasis analysis of 20 kinetochore proteins with respect to their assembly at the centromere–kinetochore complex and constructed a genetic interaction map based on our analysis and previously published data (). The map shows CENP-A occupying the top of a hierarchy that consists of three major branches that are specified by CENP-C, -I, and Aurora B. Each of the three branches form subbranches that intersect with one another to form a network that we believe describes the temporal and spatial relationship of these proteins. If kinetochores are assembled from repetitive units (), this map depicts the organization of a single unit that, in concert with other similar units, would specify the structure that is seen by EM. Depletion of specific components of the epistasis groups differentially affects the integrity of the unit and the unit–unit interactions. Each of the three major branches appears to play distinct roles in the formation of the trilaminar kinetochore structure. The CENP-I branch is clearly essential for trilaminar plate formation because, in its absence, we saw a structure that is reminiscent of the fuzzy balls that were described early on as prekinetochores (; ; ). However, the molecular composition of the fuzzy balls described in this paper differs from prekinetochores because they contain proteins (e.g., CENP-E and BUBR1) that normally assemble after nuclear envelope breakdown. Fully consistent with the fact that NUF2 belongs to the CENP-I pathway, its depletion also produced the fuzzy balls. However, the CENP-F subbranch does not participate in this organizational step. The fuzzy ball structure is also reminiscent of the structures seen when cells were depleted of the nuclear pore protein Nup358/RanBP2 (). Interestingly, depletion of HEC1 and NUF2 are known to affect the kinetochore targeting of Nup358/RanBP2, suggesting the latter may contribute to the maturation of kinetochore plates (). Consistent with this, depletion of Nup358/RanBP2, indeed, affected the localization of several kinetochore proteins (). However, Nup358/RanBP2 seems not to depend on CENP-I for its kinetochore targeting, and it is unlikely that Nup358/RanBP2 acts as a scaffold for assembly because its own localization at kinetochores depends on microtubule attachments, whereas the proteins that rely on Nup358/RanBP2 do not (; ). Given that Nup358/RanBP2 has SUMO-conjugating activity (), it may act independently of kinetochores by modifying proteins to facilitate their assembly into kinetochores. Disruption of the CENP-C and Aurora B pathways affected the structure of the trilaminar plate in a qualitative way. The CENP-C pathway appeared to specify the compaction and dimensions of the kinetochore plates, as its loss resulted in small-sized kinetochores and kinetochores with a thin or expanded and beaded outer plate. As CENP-C is involved in recruiting a substantial number of proteins, its loss would reduce the number of fully assembled unit modules. As siRNA depletes CENP-C to different extents in any given cell (), sufficient modules might be available to form a trilaminar plate, albeit one of reduced dimension (). Depletion of proteins downstream of CENP-C, such as BUB1 and hMis12, partially recapitulates the thin plates, but does not produce small-sized kinetochores that were seen in cells depleted of CENP-C (Fig. S3, B–C). As the localization of proteins like hROD and CENP-E were affected by CENP-C, but not by BUB1 or hMis12, it is likely there are unidentified proteins that are downstream of CENP-C and contribute to this aspect of trilaminar plate organization. Two candidates may be the human homologues of KNL-1 and -3, which are two proteins in that lie downstream of CENP-C. The human counterpart of KNL-1, AF15q14, has already been shown to be a kinetochore protein (). Defects associated with the Aurora B pathway seem to affect the relationship between the kinetochore and the subjacent chromatin, as seen by the formation of C-shaped kinetochores and kinetochores with seemingly continuous outer and inner plates. There is no easy explanation for how the inner and outer plates become fused, but C-shaped kinetochores have been seen in prematurely condensed chromosomes induced by either cell fusion or caffeine treatment (; ). In both situations, the condensation state of the underlying heterochromatin may be altered, thus, impacting the relationship between the kinetochore and the chromosome. Despite the impact of Aurora B on the inner regions of the kinetochore, proteins such as CENP-C and -I, which normally localize to the inner plate, are still retained at kinetochores after depletion of Aurora B. This suggests that neither of these proteins are able to form or maintain a distinct inner kinetochore plate ultrastructure in the absence of Aurora B. As Aurora B is also present throughout the cell, its actions on kinetochores do not have to be restricted to the pool that is localized at the inner centromere. Indeed, the undercondensation of chromosomes in both Aurora B– and BUB1-depleted cells may reflect additional roles of these kinases that are not linked to the kinetochore. Many of the centromere–kinetochore proteins examined here are evolutionarily conserved, and thus, similarity in their assembly amongst other organisms was expected (; ; ). However, we would like to point out that there are distinctive features of human kinetochore assembly that distinguish it from some model organisms. For example, both budding and fission yeasts perform closed mitosis, and thus, most if not all of the kinetochore proteins are present at kinetochores throughout the cell cycle. This is in contrast to the distinct temporal order of assembly for a large number of human kinetochore proteins. In addition, there are important human centromere–kinetochore components that are missing in some other species and vice versa. For instance, CENP-I and -H homologues are not found in and , and ZW10 or ROD homologues are absent from fungal genomes (; ). These differences suggest some flexibility in how the conserved proteins are assembled. At least six proteins or protein complexes were found to occupy nodes (HEC1–NUF2, MAD1, CENP-F, BUBR1, dynein–dynactin, and MCAK) that linked various branches within the network that is depicted in . We believe that these intersections reflect coordination among different assembly pathways so that no single pathway outpaces the other during the construction of a functional kinetochore. It is interesting that BUB1 occupies a position that links three pathways. Although earlier siRNA results on BUB1 challenged its role as a bona fide checkpoint protein (; ), a recent study confirmed its status as a functional mitotic checkpoint kinase and proposed a new role for BUB1 to resolve improper lateral attachment and mediate the bi-orientation of sister chromatids (). Along this line, we found BUB1 controls the localization of BUBR1 and CDC20, two essential components of the mitotic checkpoint. Previous works claimed that MAD2 localization at kinetochores depends on BUB1 (; ). We found no evidence that MAD2, or any of its upstream components (MAD1, hMPS1, and HEC1), to depend on BUB1 (; Fig. S4, D and G). In regard to its role in correcting the improper kinetochore–microtubule attachment, we have shown that BUB1 is required for the localization of MCAK, the microtubule depolymerase that, together with Aurora B kinase, is involved in correcting syntelic attachment (; ). Considering recent reports that BUB1 also affects centromeric cohesion through Sgo1 (; ; ), it will be interesting to understand why a single protein is assigned to coordinate these diverse functions. We were initially surprised to learn that only a fraction (<10% by intensity) of the CENP-A that is normally present at kinetochores is sufficient to maintain near-normal levels of CENP-I (). Further studies showed that this relationship was extended to include CENP-C, hMis12, and Aurora B (, and ; unpublished data). Although we did not directly test whether 10% levels of CENP-A supported normal kinetochore functions, we routinely observed normal metaphase cells whose kinetochores exhibited 10% levels of CENP-A. It remains to be tested whether this level of CENP-A can support a normal mitotic checkpoint response. We suggest that kinetochores in HeLa cells contain an amount of CENP-A that is in excess to that required for kinetochore assembly. Examination of extended chromatin fibers in and human cells has revealed that CENP-A (CID) and histone H3 are interspersed along the chromatin fiber. However, 3D reconstruction of their localization at kinetochores shows that domains of CENP-A and histone H3 are spatially separated (). This suggests that the CENP-A nucleosomes that are interspersed along the chromatin fiber can contact one another so that the intervening H3-containing chromatin loops out to form a spatially separated domain (). Although we cannot say with certainty the minimal number of CENP-A nucleosomes that are required for a normal kinetochore structure, our data in HeLa cells would suggest that this higher order structure can still be established when the level of CENP-A at kinetochores is reduced by 10-fold. Below this level, the density of CENP-A along the chromatin may be too low to be reliably organized into a domain that can promote the assembly of the subsequent layer of proteins that include Aurora B, CENP-C, and -I. In conclusion, we show that CENP-A plays a central role in kinetochore structure, as it specifies the assembly of proteins that form not only the trilaminar plates but also components of the inner centromere. Three major assembly pathways that contribute to different aspects of the higher order organization of the kinetochore were defined. A network of intersecting branches suggests some level of coordination between the assembly pathways. The biochemical interactions that link the assembly steps remain to be determined. The recent isolation of discrete kinetochore protein complexes consisting of hMis12, CENP-A, -I, and -H have revealed a constellation of novel kinetochore proteins whose characterization will provide insights into this question (; ; ; ; ). As new kinetochore proteins are identified, they will be assigned to our map to see how they are related to the pathways that we have identified here. The website address to an interactive map is available upon request. HeLa cells were grown in a humidified 37°C incubator with 5% CO level in DME supplemented with 10% fetal bovine serum, nonessential amino acids, and 2 mM -glutamine. All siRNAs (synthesized by Dharmacon; sequences can be provided on request) were transfected at the final concentration of 100 nM using Oligofectamine (Invitrogen) according to the manufacturer's instructions. Usually, HeLa cells were grown on coverslips in 24-well plates and transfected at 50% confluency. Coverslips were processed 4 d later, except those transfected with BUB1, Aurora B, HEC1, or hMPS1 siRNA, which were fixed 48 h after transfection. Fresh medium was added or cells were split when necessary. For BUB1 and Aurora B, double-thymidine synchronization was used to increase the number of mitotic cells at the harvesting time point; cells were incubated with 2 mM thymidine for 15 h after siRNA transfection, released for 8 h, and blocked again for 15 h with 2 mM thymidine. They were harvested 10–11 h after release, usually after 1 h of treatment of MG132 at 1 μM final concentration. In some early experiments, CENP-I siRNA1 was transfected as previously described (). Usually, several different siRNAs against the same target were used to validate the knockdown results. Rabbit, rat, or mouse monoclonal MAD1 antibody, Rabbit polyclonal MAD2 antibody, and rat or mouse monoclonal antibodies against CENP-A, -I, -E, -F, BUB1, BUBR1, hZW10, and hROD have been previously described (). Other antibodies used include monoclonal CENP-B antibody 2D8D8 (H. Masumoto, Nagoya University, Nagoya, Japan; ), rabbit anti-hMPS1 antibody Ag3 (), rabbit anti-MCAK antibody (a gift from Linda Wordeman, University of Washington, Seattle, WA), anti-p150 Glued (subunit of dynactin) rabbit polyclonal (from Linda Wordeman), and mouse monoclonal antibody (BD Biosciences), mouse monoclonal HEC1 antibody (BD Biosciences), anti-Aurora B monoclonal antibody (BD Biosciences), and rabbit polyclonal antibody (Zymed). Anticentromere antibody (ACA) serum was provided by J.B. Rattner. Image processing and fluorescence intensity measurement were performed basically as previously described using ImagePro Plus 5.0 software (Media Cybernetics) and 12-or 16-bit raw image stacks obtained from the aforementioned microscopes (). To eliminate variations of staining between experiments and coverslips, we normalized the absolute intensity values to the intensity of the brightest stained kinetochore, which was set at 100%. The relative intensities at kinetochores of different proteins were plotted to examine the relationship. AutoDeblur software (Media Cybernetics) was used to perform image deconvolution. HeLa cells were grown in 35-mm cell culture dishes and transfected with control and target siRNAs. 2–4 d after transfection, the cells were fixed in 3% glutaraldehyde and 0.2% tannic acid in 200 mM sodium cacodylate buffer for 1 h at room temperature. Post fixation was in 2% OsO4 for 20 min. The cells were dehydrated in ethanol, and then infiltrated with Polybed 812 resin. Polymerization was performed at 60°C for 24 h. Silver-gray sections were cut with an ultramicrotome (Leica) equipped with a diamond knife (Diatome), and sections were stained with uranyl acetate and lead citrate and examined in an EM (H-7000; Hitachi). Fig. S1 shows representative results of siRNA knockdown and specificity test of several antibodies. Fig. S2 shows that CENP-I is required for kinetochore localization of hMPS1, p150, and CENP-H. Fig. S3 shows representative EM images of chromosomes or kinetochores in cells depleted of CENP-F, BUB1 or hMis12. Fig. S4 shows the effects of CENP-C and BUB1 depletion on the localization of other kinetochore proteins. Table S1 shows the details of EM examination on kinetochore structures in cells transfected with siRNAs targeting CENP-I, NUF2, CENP-F, -C, BUB1, hMis12, and Aurora B. Online supplemental materials are available at .

The promyelocytic leukemia (PML) protein and PML nuclear bodies (NBs) are implicated in several cellular processes, including transcriptional regulation, tumor suppression, apoptosis, DNA repair, and the replication of both viral and cellular DNA (for reviews see ; ; ; ). How they contribute to these nuclear activities, however, has remained elusive. In normal mammalian cells, the PML protein coaccumulates in 5–30 NBs () with as many as 75 other proteins (listed in the Nuclear Protein Database; ). Rather than just sequestering these proteins, there is compelling evidence that the bodies serve as sites for the posttranslational modification of nuclear proteins. For example, the coaccumulation of p53, CBP, and HIPK2 in PML NBs contributes to the regulated phosphorylation (by HIPK2) and acetylation (by CBP) of p53 in response to DNA damage (; ). The structural and dynamic behavior of PML NBs is intimately linked to the cell's chromatin integrity (; ). Extensive chromatin contacts on the periphery of the protein cores of the NBs may account for their positional stability through extended periods in interphase of the cell cycle. Physical contacts with chromatin may be important for their proposed role in DNA replication. For example, early transcription and replication of the genomes of several DNA viruses occur immediately adjacent to PML NBs (). A link between PML NBs and chromatin is also demonstrated in the maintenance of telomeres through a recombination mechanism called alternative lengthening of telomeres, whereby a subset of PML NBs in late S/G2 phase become associated with nascent DNA synthesis, DNA repair factors, and telomere proteins (; ). The connection between PML NBs and chromatin also extends to a possible role for PML NBs in DNA repair mechanisms. For example, after DNA damage, several DNA repair factors transit to and from PML NBs, and the bodies themselves have been reported to colocalize with sites of unscheduled DNA synthesis in damaged cells (). PML may also function in DNA damage signaling because PML-null cells fail to fully activate p53 in response to DNA damage () and the PML protein is phosphorylated in response to DNA double-strand breaks (DSBs) by Chk2 () and ataxia telangiectasia and Rad3-related (ATR) kinase (). It is unclear whether these modifications of PML or PML NB composition are critical for DNA repair to proceed or are a consequence of ongoing repair. Regardless, PML NBs are clearly more than passive accumulations of nuclear proteins. We propose in this study that PML NBs can be used to monitor the topological state and integrity of chromatin in mammalian cells. In so doing, they act as sensors of DNA damage. Previously, we have shown that when the topological state of chromatin is altered during early S phase by the replication of DNA, PML NBs lose both radial symmetry and integrity, fragmenting into “microbodies” by a fission mechanism (). We demonstrate a similar response of PML NBs after the introduction of DNA DSBs, thereby providing a basis for previous observations of increases in PML NB number after DNA damage with ionizing radiation (IR; ; ). We demonstrate that PML NB breakdown occurs in two components. The first is a rapid biophysical response, occurring in cells damaged at 4°C, a state in which ongoing DNA repair is inhibited, and a second component associated with repair mechanisms. Inactivation or loss of repair factors, such as Nbs1 or the checkpoint kinases ataxia telangiectasia mutated (ATM), Chk2, and ATR, inhibits PML microbody formation in response to DSBs. We suggest that the PML NBs are highly sensitive DNA damage sensors whose dynamic behavior reflects both the degree of DNA damage and the integrity of the DNA repair pathways involved in maintaining the mammalian genome. The mechanism responsible for an increase in PML NBs after DNA damage (; ) has not been elucidated. To address specifically how DSBs might contribute to this process, we characterized the response of PML NBs to DSBs in the normal human diploid fibroblast (NHDF) cell line GM05757 using IR, etoposide (VP16), and doxorubicin (). IR generates both single-strand breaks and DSBs in DNA, whereas the topoisomerase II inhibitors VP16 and doxorubicin primarily create DSBs (for review see ). PML NBs were counted in maximum-intensity Z projections of individual cells. In agreement with previous work, we found that PML NB number increased after DSB induction (). Furthermore, we found that the time point associated with the highest number of PML NBs coincided with the peak of H2AX phosphorylation (γ-H2AX; ), an event that occurs on chromatin surrounding DSBs (). Maximum PML NB number correlated with peak γ-H2AX signal regardless of the method of DSB induction, suggesting that the increase in PML NB number is coupled to DSB formation. PML NB induction was most rapid for IR, peaking at 30 min after IR (). In contrast, γ-H2AX signal and PML NB number peaked later, 3 h after treatment with VP16 or doxorubicin (). Consistent with previous studies of PML NB association with γ-H2AX and components of the Mre11–Rad50–Nbs1 (MRN) complex (; ), we observed foci of γ-H2AX and Nbs1 that partially colocalized with or were juxtaposed to PML NBs between 6 and 18 h after DSB induction ( and Fig. S1 A, available at ). In contrast, we observed a much earlier colocalization and juxtaposition between the foci of RPA and PML NBs at 1.5 h after DSB induction, which persisted for up to 18 h (Fig. S1 B). After etoposide treatment, only a subpopulation of cells in S and G2 phase develop replication protein A (RPA) foci in NHDFs. Therefore, the association of PML NBs with RPA foci after DNA damage is restricted to S and G2 phase of the cell cycle (Fig. S1 C). We then tested whether the increase in PML NB number in response to DSBs is dose dependent by treating cells with doses of IR varying from 0–10 grays (Gy; ). We found that at doses as low as 1 Gy (i.e., producing ∼35 DSBs; ), PML NBs increased in number in NHDFs, and the response was dose dependent, based on analysis of variance (ANOVA) between our datasets (Table S1, available at ). In contrast to low doses of IR, where the number of PML NBs returned to baseline levels by 24 h after irradiation, at higher doses of 5 and 10 Gy, PML NB numbers remained elevated for an extended period of time (). Therefore, after low doses of IR, the increase in PML NB number after DNA damage appears to be reversible. When PML NB number is plotted versus dose of IR, NB number appeared to reach a plateau at doses of 5 Gy or above for all time points, with the exception of 24 h (). Therefore, PML NB number varies with the power of the dose of IR and can be described by the modified power function, y = a × b + c, where y is the number of PML NBs and x is the dose of IR in grays (Fig. S2 A). Thus, the increase in PML NB number in response to DSBs is rapid, sensitive to sublethal levels of DNA damage, and dose dependent. The dynamics of PML NBs after DSB induction with VP16 was examined by live cell analysis of U-2 OS cells stably expressing PML isoform IV (). We found that within 5 min after addition of VP16, new and smaller PML-containing structures began to appear adjacent to the larger PML NBs that were present before treatment (). These new bodies, which we term microbodies, arise from preexisting PML NBs by a supramolecular fission mechanism, as confirmed by spinning-disc confocal microscopy ( and Video 1, available at ). This fission mechanism is similar to that observed for new PML NB formation in early S phase (), as PML NB's biochemical composition was initially indistinguishable between microbodies and the larger parental PML NBs, in respect to Sp100 and small ubiquitin-like modifier (SUMO-1) content (Fig. S3). However, although Sp100 levels at PML NBs did not change over the time course observed (Fig. S3 A), we did notice a reproducible drop in SUMO-1 levels in PML NBs at 3 h after VP16 treatment (Fig. S3, B and C). Overexpression of SUMO-1 dramatically reduced PML NB number (Fig. S3 D), resulting in enlarged bodies that showed reduced or delayed increase in PML NB number in response to DSBs (Fig. S3 E). It is unclear if overexpression of SUMO-1 is directly or indirectly responsible for the stabilization of PML NBs in our experiments because sumoylation is implicated in many biological pathways, including DNA repair (). PML microbodies also formed immediately after irradiation with doses as low as 1 Gy of IR (unpublished data), and an increase in PML NB number was seen even when cells were irradiated and fixed on ice to prevent diffusional movement of PML protein or ongoing DNA repair (). At temperatures <15°C, PML protein diffusion is very limited, as confirmed by FRAP analysis (). Interestingly, we also observed a 10% difference in the maximum fluorescence recovery between control and etoposide-treated cells, consistent with a larger immobile fraction of PML protein in bodies after DNA damage (). We next examined the behavior of PML NBs in the vicinity of site-specific DSBs induced by UV laser irradiation (; and Video 2, available at ). Within 5 min of the induction of DSBs, PML NBs in the vicinity of the laser track began to move and coalesce (, arrowheads). This process continued for over 20 min, resulting in a drop in PML NB number from 21 to 17 NBs, but did not affect PML NBs distal to the laser track (). At later time points, however, even PML NBs far from the laser track lost their positional stability. Imaging of cells in the absence of UV laser microbeam irradiation did not affect the mobility or number of PML NBs (Video 3). Continuous imaging by laser scanning confocal microscopy (LSM) after DNA damage did not reveal microbody formation, likely because of photobleaching and the loss of visibility of small, PML-containing structures. However, after fixation and immunofluorescence (IF) detection of PML and γ-H2AX by wide-field microscopy 1 h after photoinduction of DNA DSBs, it was apparent that the DNA damage was confined to the laser track and that PML NB number had increased from 17 to 36 PML NBs (). Although wide-field microscopy is generally more sensitive than LSM in the detection of PML microbodies, we found that LSM was sufficient to detect >90% of bodies within a focal plane (unpublished data). Therefore, the increase in the number of PML NBs at 1 h after DSB induction is primarily caused by microbody formation. To address the ultrastructural changes in PML NBs after DNA damage, we used immunogold detection of PML with correlative light microscopy (LM) and electron spectroscopic imaging (ESI; ; ). Using LM/ESI we observed that, in control NHDFs, PML NBs exhibit radial symmetry and make extensive contacts with the surrounding chromatin (). Upon treatment with VP16, we found that PML NBs lose their radial symmetry and make fewer contacts with the surrounding chromatin fibers. We also observed “microbody-like” structures, which were identified by immunogold detection of PML, adjacent to chromatin in the vicinity of larger “parental” PML NBs (). A much larger interchromatin domain space was also apparent in cells treated with VP16 (black spaces outside of chromatin in ). These changes in both chromatin and PML NBs are reminiscent of those seen in cells entering S phase (). Based on these results, we suggest that the introduction of DSBs results in topological changes in chromatin linked to PML NBs, which destabilizes the PML NB core. PML protein levels can increase after treatment with IR in a p53-dependent manner (). Therefore, we examined the PML NB response to DNA DSBs in NHDFs with inhibition of PML protein synthesis by treatment with cycloheximide, and in cells that lack a functional p53 pathway (i.e., null p53 human Saos-2 osteosarcoma cells and paired HCT116 cell lines, which were isogenic save for p53 protein; ). We found that PML protein levels increased slightly, by 1.3-fold at 4 h after VP16 treatment, and that they reached 1.8-fold by 12 h (). As expected, cycloheximide treatment inhibited the DNA damage–dependent increase in PML protein levels at 12 h, but had little effect at 4 h, suggesting that posttranslational regulation of PML protein levels may occur at this earlier time point. We found that inhibition of protein synthesis or loss of p53 function did not prevent the initial increase in PML NB number (at 30 min and 3 h) in response to DSBs (). Loss of p53 in the HCT116 cell background actually appeared to enhance the increase in PML NB numbers at 30 min after VP16 treatment (), perhaps because further genome instability from a concurrent loss of the mismatch repair factor MLH1 (). Cycloheximide-treated NHDFs exhibited a higher number of PML NBs initially, compared with untreated cells, and NB number returned to control levels much earlier than in untreated NHDFs (). In contrast, PML NB number in VP16-treated Saos-2 cells continued to increase over time (). PML NB number is affected by cell cycle progression and increases in early S phase (). FACS analysis revealed that after VP16 treatment, NHDFs showed a marked accumulation in G1 and G2 phase of the cell cycle by 18 h (Fig. S4 A, available at ). Therefore, differences in the number of cells in S phase at the late time points (6 and 18 h) might account for the continued increase in PML NBs observed in the G1/S checkpoint-deficient Saos-2 cells. We examined this possibility by detecting BrdU incorporation at 18 h after VP16 treatment to determine the fraction of cells replicating DNA (). We estimate that ∼40% of the Saos-2 cells were in S phase, compared with only 5–6% of NHDFs, and no BrdU incorporation was observed in NHDFs treated with cycloheximide. Therefore, at early time points, the increase in PML NB number is independent of both new protein synthesis and p53. However, at later time points after DNA damage, PML NB number is sensitive to loss of p53 caused by abrogation of the G1/S checkpoint, and as a result, PML NB number continues to increase as cells enter S phase inappropriately. Because ATM, ATR, and Chk2 kinase are key regulators of the cellular response to DNA damage (; ), we examined whether chemical inhibition of these kinases might affect the response of PML NB to DSBs in NHDFs (). Inhibition of ATR and ATM kinase by 5 mM caffeine had an inhibitory effect on the increase of PML NB number after VP16 treatment at all time points (P < 0.0001; ). Similarly, caffeine significantly reduced the response of PML NBs to 5 Gy of IR (P < 0.001; Fig. S4 B). This effect was not caused by caffeine-dependent changes in the cell cycle profile because PML NB number did not change when cells were pretreated with caffeine for 30 min before induction of DSBs (), and only prolonged treatment with caffeine had an effect on the cell cycle profile of NHDFs, with or without VP16 treatment (Fig. S4, A and C, 18 h). Pretreatment of NHDFs with the Chk2 kinase inhibitor II () did not affect the initial increase in PML NBs in response to VP16, but did significantly reduce the number at 3 h, compared with cells treated with VP16 in the absence of inhibitor (P < 0.0003; ). Similarly, 20 μM wortmannin, which strongly inhibits DNA-PK and ATM kinase, but weakly inhibits ATR, had a significant effect on PML NB number only at 3 h after VP16 treatment (P < 0.001; ). In contrast, the DNA-PK inhibitor LY2942002 had little effect on PML NB number in response to DSBs. We also treated several repair-deficient cell lines with VP16 to compare the PML NB response after DNA damage. As with chemical inhibition of ATM, AT cells, which are deficient in ATM, showed a significant inhibition of PML NB number increase only at 3 h after VP16 treatment (P < 0.02; ), after which PML NB number actually increased beyond that expected for NHDFs at 6 h. AT-like disorder (ATLD) cells expressing mutant Mre11 (), which is a component of the DNA damage sensor known as the MRN complex (), showed an initial increase in PML NB number after VP16 treatment at 30 min, which is similar to NHDFs. At the 3-h time point and thereafter, however, the increase in PML NB number was inhibited by loss of Mre11 function (P < 0.0001; ). The increase in PML NB number after induction of DSBs was significantly inhibited at all time points observed in Nijmegen breakage syndrome (NBS) cells, which are deficient in Nbs1, which is also a member of the MRN complex (), and was profoundly inhibited in Seckel syndrome cells, which are deficient in ATR kinase (P < 0.0001; ; ). Because the concentration of Chk2 inhibitor used in our experiments could have residual effects on other kinases (<25% inhibition of a panel of 35 kinases, ), we wished to confirm the role of Chk2 in regulating the response of PML NBs to DSBs using a genetic mouse model. As predicted from our inhibitor data, Chk2 −/− murine embryonic fibroblasts (MEFs) had an abrogated PML NB response to DSBs, compared with isogenic wild-type MEFs at 3 h after VP16 treatment (P < 0.001; ). Similarly, reconstitution of NBS cells with wild-type human Nbs1 by retroviral transduction resulted in a robust increase in NB number at 3 h after VP16 treatment, confirming a role for Nbs1 in regulating the PML NB response to DSBs (P < 0.02; ). Finally, we further characterized the role of ATR kinase in regulating the response of PML NBs to DSBs by VP16 treatment of U-2 OS cells expressing an inducible dominant-negative mutant of ATR (kinase-dead ATR-DN; ; ). Induction of ATR-DN in these U-2 OS cells for 24 h before VP16 treatment significantly inhibited PML NB induction at 30 min, 3 h (P < 0.0001), and 6 h after treatment (P < 0.001; ). Interestingly, PML NB numbers continued to rise in U-2 OS cells expressing the ATR-DN protein, possibly because of extensive genome instability and eventual apoptosis associated with prolonged expression of this protein. Even within the population of U-2 OS cells expressing the ATR-DN protein, high expression correlated with reduced PML NB number, compared with low-expressing cells at 3 h after VP16 treatment (). #text Cell lines used in this study are as follows: NHDFs (GM05757; Coriell Cell Repository); human AT fibroblasts (AT5B1 and GM05823; Coriell Cell Repository); human NBST fibroblasts (gift from J. Lukas, Danish Cancer Society, Copenhagen, Denmark); human ATLD fibroblasts (gift from Y. Shiloh, Tel Aviv University, Tel Aviv, Israel); Saos-2 (American Type Culture Collection [ATCC]); HCT116 and p53-null HCT116 isogenic cells (gift from B. Vogelstein, Johns Hopkins University, Baltimore, MD); ATR-DN and ATR-WT cells (gift from Paul Nghiem, University of Washington Medical School, Seattle, WA); Seckel syndrome cells (GM18366; Coriell Cell Repository); Chk2 −/− MEFs and isogenic WT MEFs (gift from T. Mak, University of Toronto, Toronto, Canada); and isogenic and U2OS cells stably expressing GFP-PML IV (gift from J. Taylor, University of Wisconsin, Milwaukee, WI). NBST-pBabe and NBST-pBabe-NBS1 cell lines were generated by retroviral transduction of NBST fibroblasts using either pBabe-Puro alone or encoding full-length human Nbs1 (gift from J. Lukas), respectively. To generate DSBs, cells were treated with 20 μM etoposide (VP16; Sigma-Aldrich) or 1.5 μM doxorubicin (Sigma-Aldrich) for 30 min, washed two times in PBS (WISENT, Inc.), and left to recover for the indicated time. We determined that 20 μM VP16 for 30 min was equivalent to ∼2 Gy of IR by the neutral comet assay (Fig. S2 B). Alternately, asynchronous cell cultures were exposed to whole-cell IR (dose range, 0–20 Gy) using a Cs irradiator (MDS Nordion) at 1 Gy/min (on ice, aerobic conditions). For kinase inhibition studies, cells were incubated with growth medium supplemented with 20 μM wortmannin (Sigma-Aldrich), 5 mM caffeine (Sigma-Aldrich), 50 μM LY294002, or 10 μM Chk2 inhibitor II (EMD Biosciences, Inc.; ) for 30 min, before addition of VP16 or exposure to IR. Cells were maintained in growth medium containing kinase inhibitors for the indicated time. For the inhibition of protein synthesis, cells were treated with 150 μg/ml of cycloheximide (Sigma-Aldrich) for 30 min before treatments and maintained in cycloheximide until processed for LM. Cells grown on coverslips were treated with or without kinase or protein synthesis inhibitors before DSB induction (etoposide or IR), fixed, and processed for IF as previously described (). Primary antibodies used in this study are as follows: rabbit anti-PML (CHEMICON International, Inc.); rabbit anti-Sp100 (CHEMICON International, Inc.); mouse anti-RPA (RPA34-20; Calbiochem); mouse anti–phospho-Histone H3 (ser10; clone MC463; Millipore); rabbit anti–cyclin A (sc-751; Santa Cruz Biotechnology, Inc.); mouse anti–SUMO-1 (GMP-1; Invitrogen); mouse anti–γ-H2AX (JBW301; Millipore). Secondary antibodies conjugated to Cy3 and Cy5 were obtained from Jackson ImmunoResearch Laboratories, and secondary antibodies conjugated to Alexa Fluor 488 were obtained from Invitrogen. DNA was stained with DAPI (Sigma-Aldrich) in mounting media containing 90% glycerol and 1 mg/ml paraphenylenediamine (Sigma-Aldrich). Fluorescence micrographs of fixed cells were collected using a 63×, 1.32 NA, oil-immersion objective lens (HCX PL APO CS; Leica) on an upright fluorescence microscope (DMR2; Leica) fitted with a camera (Orca; Hamamatsu). OpenLab 3.5.1 software (Improvision) was used for image acquisition. Live-cell imaging and FRAP analysis of GFP-PML IV in U-2 OS cells was performed as previously described (). ImageJ v1.33 (National Institutes of Health) and Photoshop 7.0 (Adobe) software were used for image processing and analysis. To determine mean PML NB number, maximum intensity projections of multiple focal planes were generated for the IF localization of PML using OpenLab 3.5.1 software (Improvision). PML NBs were counted in a minimum of 30 cells per time point and the NB number per cell was normalized for nucleus size. Normalization of NB number was accomplished by multiplying the ratio of the area of each nucleus divided by the mean area of a nucleus in a given dataset. This calculation was necessary to account for the variability in PML NB number caused by cell cycle phase or ploidy (). However, in normal diploid cell lines, this calculation will not affect the mean NB number per cell, but will reduce statistical variability between datasets. Each experiment was repeated in triplicate, and the mean PML NB number was used directly or divided by the mean number of bodies in the control (untreated) to give the fold induction of PML NBs. Error analysis for triplicate experiments is expressed as the SEM, where SEM = SD ÷ √3. For all other experiments, error analysis was expressed simply as standard error. Datasets of PML NB number per cell exhibit a normal distribution; therefore, statistical significance between datasets was derived using the test for pair-wise analysis using Excel software (Microsoft) and by ANOVA for testing the significance of IR dose on PML NB number using online statistical tools available from . Curve fitting for PML NB induction versus dose of IR was accomplished online using tools available from . Cells grown on coverslips were incubated for 5 min in growth medium containing 0.5 μg/ml of Hoechst 333258 to sensitize cells to the UV laser–induced damage. A confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.), equipped with an argon laser (488 nm) and tunable multiphoton laser (Chameleon; Coherent Inc.) capable of effective wavelengths in the UV range, was used to image cells and to generate UV laser–induced damage using a 63×, 1.40 NA, oil-immersion objective lens (Plan-Apochromat; Carl Zeiss MicroImaging, Inc.). Laser damage was accomplished by selecting a region of interest (ROI) within a cell and bleaching the ROI using the tunable laser set at 790 nm (effective λ = 390 nm) and 20% power for a 200-ms pulse. At 20% power, the laser generates 7−8 mW, which translates to a cellular dose of ∼80 Gy (). Images were collected immediately after the bleach, using the argon laser at 50% power and 10% transmittance. Cells were maintained at 37°C during live-cell imaging, using a heated stage (Bioptechs). Samples were prepared and sectioned for correlative microscopy and ESI as previously described (; ). Nitrogen and phosphorus maps were collected using a transmission electron microscope (Tecnai 20; FEI) fitted with an electron imaging spectrometer (Gatan). Immunogold labeling was accomplished using a secondary antibody conjugated to Ultrasmall nanogold (donkey anti–rabbit; Electron Microscopy Sciences). Fig. S1 shows the relative localization of PML NBs in respect to foci containing γ-H2AX, Nbs1, and RPA over time, after etoposide induced DSBs. Fig. S2 shows the mathematical modeling of PML NB number in response to IR-induced DSBs and compares the number of DSBs induced by 20 μM etoposide versus varying doses of IR by neutral comet assay. Fig. S3 shows the biochemical composition of PML NBs over time after etoposide induced DSBs, in respect to SP100 and SUMO-1 content. Fig. S4 shows the effects of etoposide and caffeine, alone or in combination, on both the cell cycle profile and PML NB number of NHDFs. Table S1 shows the ANOVA analysis of PML NB number as a function of the dose of IR in NHDFs. Video 1 shows the fission of a PML microbody from a preexisting PML NB in response to DSBs induced by etoposide. Video 2 shows the loss of positional stability of PML NBs after UV laser induction of DSBs. Video 3 shows a control cell where, in the absence of UV laser–induced DSBs, PML NBs are positionally stable. Online supplemental material is available at .

mRNA localization is a critical step in the development of cellular asymmetry. This suggests that controlling sites of translation restricts target proteins to specific subcellular compartments (; ). For instance, mRNA localization is required during development to establish morphogen gradients in the oocyte and for cell lineage specification in the early embryo (). mRNA localization to distal regions of neurons has been observed, and local translation of these mRNAs can be initiated in response to various extracellular cues (). The ability to determine where translation occurs may provide clues to the spatial and temporal organization of cells and tissues by elucidating the relationship between localized mRNAs and the distribution of protein products of translation. Where translation occurs relative to extracellular cues is critical to understanding how cells chemotax in response to their environment, a process that is critical to understanding how cells within a tumor make the decision to metastasize (). During neuronal development, determining the sites of translation of specific proteins may be critical in understanding how neurons find their proper targets for synapse formation through the regulation of growth cone assembly and movement. In addition, monitoring the sites of translation of specific mRNAs in the pre- and postsynaptic regions of neurons may elucidate how cells convert electrical stimulation into stored memory, the process of synaptic plasticity. One of the better characterized models of mRNA localization is the process used by β-actin. The localization of β-actin mRNA is correlated with the localization of β-actin protein to apical structures such as filaments in microvilli of epithelia and auditory hair cells and the leading edge of lamellipodia and filopodia of crawling cells (; ). Zipcode sequences are components of localized mRNAs that are necessary and sufficient for the proper targeting of transcripts. The β-actin zipcode is responsible for the localization of β-actin mRNA to the periphery of fibroblasts or the growth cone of neuronal cells (; ). The concept that β-actin contributes to leading edge dynamics is supported by the observation that β-actin mRNA targeting to the front of crawling cells is necessary for directed motility of fibroblasts and carcinoma cells (; ; ). Inhibition of β-actin mRNA targeting, by using antisense oligonucleotides directed to the zipcode or by deleting the zipcode, resulted in a decrease in cell speed and directionality (; ). Inhibition of β-actin mRNA targeting by using antisense oligonucleotides also resulted in the randomization of the location of free barbed ends in the cell cortex, suggesting that the sites of β-actin polymerization are determined in part by the localization of its mRNA (). Although these reports emphasize the fundamental role of RNA localization, it has not been established how cell physiology is mechanistically related to the compartmentalized synthesis of proteins. To do so will require a means to follow newly synthesized proteins from their sites of translation to their sites of utilization. Therefore, we developed a method called translation site imaging for identifying polypeptides as they are synthesized in living cells and released to their sites of utilization. The approach can simultaneously detect β-actin mRNA, translation sites, and mature protein in living cells based on genetically encodable tetracysteine (TC) tags that associate with the biarsenical dyes FlAsH and ReAsH (). To identify sites of β-actin translation, β-actin promoter–driven constructs were generated to ensure the proper copy number of total actin transcripts. A TC tag (CCPGCC) was inserted at the N terminus, followed by a GFP fusion domain and the β-actin coding sequence, with or without the 3′UTR zipcode sequence. The construct that contains the β-actin 3′UTR and, thus, the zipcode will be referred to as the full-length (FL) construct, whereas the constructs that lacks the β-actin 3′UTR containing the zipcode will be referred to as the ΔZIP construct. After ReAsH staining, C2C12 cells transfected with the FL construct exhibited a staining pattern nearly identical to that of GFP–β-actin (). C2C12 myoblasts were used for these investigations based on their ability to target β-actin mRNA (). In addition, these cells differentiate into muscle through fusion of cell contacts, a process that we were interested in investigating with regard to a potential role for localized translation at contacts. By using the concentration of nascent polypeptide chains at polysome sites, we expected to identify translation sites using ReAsH staining and the translation elongation inhibitor cycloheximide (). Cycloheximide is known to accumulate ribosomes and nascent polypeptide chains on mRNA (; ). ReAsH staining of C2C12 cells transfected with the FL construct resulted in the detection of discrete uniformly sized bright puncta generally localized at the cell periphery, even without cycloheximide (). Their distribution was similar to that of β-actin mRNA in mouse myoblast cells (). Untransfected cells did not exhibit punctate staining (not depicted), indicating that the observed ReAsH puncta resulted specifically from the binding of dye to the transgene protein product. When cells containing the ΔZIP construct were stained with ReAsH, puncta were observed in the perinuclear region and very few were observed at the cell periphery, suggesting that the information contained within the zipcode region is responsible not only for the proper targeting of mRNA but also for proper targeting of the translation sites (). These data also demonstrate that translation sites can be detected after ReAsH staining by quick destaining and fixation () or by the use of cycloheximide during staining and destaining (). We report the novel use of labeled TC-tagged proteins to image specific nascent polypeptides during their synthesis in situ. We have used the TC system to detect polysomes in the process of synthesizing a specific polypeptide in real time with a spatial resolution of ∼0.2 μm. Our data demonstrate that peripheral translation of β-actin is zipcode dependent, as predicted by the localization of its mRNA (; ). Mislocalization of β-actin mRNA using the ΔZIP construct resulted in the mislocalization of translation sites and, consequently, the newly synthesized protein. β-Actin translation sites were anchored and exhibit very little movement, providing a stable molecular framework for the observed accumulation of mature protein to specific subcellular compartments. In noncontacting cells, localized translation is seen at the leading lamella. In contrast, in contacting cells, localized translation is observed at sites of cell contact. When the trafficking of newly synthesized β-actin was monitored, accumulation of the protein at points of cell contact was only observed when the mRNA had a zipcode. This particular compartmentalization may have relevance because the actin cytoskeleton is actively involved in cell interactions, particularly in the fusion of myoblasts (). The TC system has also been used to determine the temporal changes occurring at gap junctions using pulse-chase staining of newly synthesized connexins that were delivered to the plasma membrane containing junctional plaques composed of older connexin molecules (). In addition, pulse-chase staining with FlAsH and ReAsH has been used to confirm that there is dendritic synthesis of GluR1 in response to extracellular stimuli using cultured neurons with physically isolated dendrites (). However, the translation site imaging technique allows the identification of the site where a specific mRNA is translated. We have been able to correlate the site of translation with the accumulation of the newly synthesized protein. This bridges the fields of mRNA localization and protein trafficking. FlAsH and ReAsH were a gift from R. Tsien (University of California, San Diego, La Jolla, CA) and were later purchased from Invitrogen under the trade names lumio green and lumio red, respectively. A PCR primer was synthesized containing a HindIII restriction site, the coding sequence for a TC tag, and sequences complementary to the GFP coding region called TC1, 5′-tggaagcttccaccatgtgggattgttgtccaggatgttgtaaaatggtgagcaagggcgaggagctgttc-3′. An additional primer called TC2, 5′-tggaagcttccaccatgttcctcaactgctgcccaggatgttgtatggagccatggtgagcaagggcgaggagctgttc-3′, was synthesized. A third PCR primer was synthesized containing a BamHI restriction site and sequence complementary to the GFP coding region 5′-ccggatcccttgtacagctcgtccatgc-3′. The primers were used to amplify the GFP coding sequence from plasmid pβ-actin EGFP () with a TC motif MWDCCPGCCKM (from the TC1 primer; – and ) or the improved TC motif MFLNCCPGCCMEP (from the TC2 primer; and ) at the N terminus (). The EGFP coding sequence was removed from plasmid pβ-actin EGFP using the restriction enzymes BamHI and HindIII and replaced with the TC-containing EGFP PCR product to produce the plasmid TC–GFP–β-actin with no zipcode (ΔZIP). To produce the FL β-actin plasmid, the ΔZIP plasmid was cut with BamHI and XbaI to remove the β-actin coding sequence. A second plasmid containing the β-actin coding sequence with a zipcode was also cut with BamHI and XbaI, and this product was ligated into the cut ΔZIP plasmid to produce the TC–GFP– β-actin FL plasmid. C2C12 mouse myoblast cells were cultured in α-MEM media with 10% FBS using standard techniques. For imaging experiments, C2C12 cells were plated directly onto 35- × 10-mm plastic tissue culture dishes or onto acid-washed glass coverslips 24 h before transfection. The TC–GFP–β-actin constructs were transfected into the C2C12 cells using Fugene 6 transfection reagent (Roche Diagnostics Corporation) according to the manufacturer's instructions for 24 h at 37°C. After the incubation, the Fugene 6 solution was removed and replaced with α-MEM media supplemented with 10% FBS for 1 h at 37°C before cells were used in experiments. Transfected C2C12 cells were treated with 100 μg/ml cycloheximide ready-made solution (Sigma-Aldrich) for 30 min at 37°C. The cells were then treated with 1× staining solution (1 μM FlAsH or 2.5 μM ReAsH in 1 ml of Opti-MEM medium (Invitrogen), 100 μg/ml cycloheximide solution, and 10 μM ethanedithiol (EDT) for 15 min to 2 h at room temperature. The cells were washed in Opti-MEM followed by destaining in 250 μM EDT, 100 μg/ml cycloheximide, and 1 ml of Opti-MEM for 5–30 min at room temperature. The cells were washed in Opti-MEM, and digital images were acquired using an epifluorescence microscope (BX61; Olympus) with a UPlanApo 60×/1.2 NA W PSF (water immersion; Olympus) objective for live cell imaging or a PlanApo 60×/1.4 NA oil-immersion objective (Olympus) for fixed cells and a 100-Watt mercury arc lamp (Olympus), equipped with a camera (CoolSNAP HQ; Photometrics) using IPLab software (Windows v3; BD Biosciences) and filter sets 41001 (FITC), 41007 (Cy3), 41004 (ReAsH), 41008 (Cy5), and SP104v1 (Cy5 narrow band pass; Chroma Technology Corp.). C2C12 cells containing the FL construct were treated with 200 μg/ml puromycin for 30 min followed by 30 min of ReAsH staining in the presence of 100 μg/ml cycloheximide. As a control, C2C12 cells containing the FL construct that were not treated with puromycin were stained for 30 min in ReAsH in the presence of 100 μg/ml cycloheximide. The number of translation sites per cell was determined for the untreated and puromycin-treated populations of C2C12 cells. C2C12 cells containing the FL plasmid were stained with ReAsH in the absence of cycloheximide. After destaining, the cells were imaged using a Cy3 filter set at 1 image/min. As a control, C2C12 cells containing the FL plasmid were stained with ReAsH in the presence of cycloheximide, washed, and imaged using a Cy3 filter set. C2C12 cells containing either the FL or the ΔZIP plasmids were hybridized with a Cy5-labeled antisense probe to the GFP coding sequence (). The antisense GFP probe is a mixture of three oligonucleotides with the following sequences: GFP-1, GGGTCTTGTAGTTGCCGTCGTCCTTGAAGAAGATGGTGCG; GFP-2, GGCTGTTGTAGTTGTACTCCAGCTTGTGCCCCAGGATGTT; and GFP-3, TCTTTGCTCAGGGCGGACTGGGTGCTCAGGTAGTGGTTGT. Images were obtained using a Cy5 filter set. The intracellular distribution of FISH signal was determined using software written to measure the total fluorescence intensity as a function of distance from the nucleus. The software identified the nuclear boundary using an overlaid and registered image of the DAPI-counterstained nucleus. The software identified the cellular boundary using an edge-detection routine with user-adjustable parameters. The cytoplasmic area was defined as the region between the nuclear and cellular boundaries identified. The FISH signal in the entire cytoplasmic area was analyzed by the software, which outputted a histogram of FISH intensity as a function of distance from the nucleus. The data were normalized by total FISH intensity and longest distance from the nuclear to cellular boundary. C2C12 cells containing the FL or ΔZIP constructs were stained with ReAsH, fixed in 4% paraformaldehyde for 30 min at room temperature, and processed for FISH using a Cy5-labeled probe to the GFP coding sequence (). Images were obtained using the FITC, ReAsH, and Cy5 narrow filter sets and deconvolved using Huygens Professional version 2.6.4 (Scientific Volume Imaging). Colocalization between the ReAsH and FISH images was determined by overlaying individual planes from each channel using IPlab software. 50% confluent C2C12 cells containing the FL or ΔZIP constructs were pulsed with 1 μM FlAsH in 1 ml of Opti-MEM and 10 μM EDT for 1 h at 37°C. The cells were then chased with 2.5 μM ReAsH in 1 ml of Opti-MEM and 10 μM EDT for 15 min at 37°C. A 5-min destain in 250 μM EDT in 1 ml of Opti-MEM was performed, and cells were imaged for live cell experiments or fixed in 4% paraformaldehyde in 1% PBS. For each cell, an image was obtained using the FITC and ReAsH filter sets. Ratio image analysis was performed using software that generated a binary mask of the denominator image by taking all pixel values greater than a fixed value. To this binary mask, we apply a grayscale closing operation (dilation followed by erosion) using a 4 × 4 matrix where each element is 1. The resulting mask defines the region of the cell that has an adequate signal/noise ratio. The ratio of the original background-subtracted images is then calculated in regions defined by the binary mask. The ReAsH image is the numerator, and the FlAsH image is the denominator. ReAsH staining was performed followed by fixation in 4% paraformaldehyde in 1% PBS. The samples were blocked in 3% BSA, stained in a 1:50 dilution of anti–N-cadherin antibodies for 3 h at room temperature (BD Biosciences), and stained in a 1:250 dilution of Cy5-labeled anti-mouse secondary antibody. Images were obtained using the FITC, ReAsH, and Cy5 narrow filter sets. Fig. S1 demonstrates that translation sites are resistant to detergent extraction, suggesting an interaction with the cytoskeleton. Video 1 shows a C2C12 cell that was stained with ReAsH; an image was collected every 5 min for 1 h at room temperature. Video 2 shows a C2C12 cell transfected with the FL construct contacting an adjacent cell pulse labeled with FlAsH for 1 h and chase labeled with ReAsH for 15 min. Online supplemental material is available at .

Cell commitment and differentiation involves the carefully regulated expression of genes associated with terminal differentiation, and many different biological systems have been used to identify the factors and networks that regulate these genes. Skeletal myogenesis represents a particularly well-studied model system (). Vertebrate skeletal myogenesis is regulated by a family of four related basic helix-loop-helix transcription factors: Myf5, MyoD, Myog, and Mrf4 (; ; ; ). Recent studies indicate that MyoD, and perhaps by analogy Myf5, binds to promoters of genes expressed throughout the program of myogenesis, establishing temporal patterning through feed-forward mechanisms with other regulatory factors (; ; ). An interesting and relatively unstudied feature of the MyoD-mediated differentiation program, however, is the suppression of the nonmuscle phenotype in cells that are converted to skeletal muscle by MyoD. For example, expression of MyoD in melanocytes, adipocytes, and chondrocytes results not only in the positive regulation of skeletal muscle genes, but also in the suppression of molecular phenotypes specific to the melanocyte, adipocyte, and chondrocyte, respectively (; ). In this regard, conversion of a cell to skeletal muscle by MyoD is accompanied by the suppression of the nonmuscle phenotype of that cell. This is a specific example of a general feature of cell differentiation; establishing a new cellular phenotype requires suppression of features associated with the prior phenotype or other related phenotypes. Although the activation of skeletal muscle gene expression by MyoD has been intensively studied, very little is known about the ability of MyoD to suppress gene expression. Several studies have indicated that MyoD can form repressive complexes, and these complexes have been suggested to be a mechanism of directly recruiting transcriptional repressors to specific promoters. For example, MyoD has been shown to recruit HDAC1 to the Myog promoter in myoblasts with an associated hypoacetylation of regional histones (; ), and a similar MyoD-mediated recruitment of the Sir2 HDAC to suppress gene expression has been previously described (). When cells differentiate, MyoD has been shown to recruit histone acetylases and chromatin-remodeling complexes to some of the same promoters shown to be suppressed by HDAC recruitment in myoblasts (, ; ), suggesting that a switch between a repressive complex and an activating complex occurs at the initiation of terminal differentiation. In addition, components of the repressive Polycomb complex have been shown to be associated with repressed muscle genes before differentiation and replaced by MyoD and other activators during differentiation (). Therefore, there is precedent for a regulated transition from repressive promoter complexes to activating complexes during skeletal muscle differentiation, and some evidence that MyoD can recruit either activators or repressors to the promoter regions, depending on cellular context, i.e., whether a cell is a replicating myoblast or differentiating myotube. These developmental transitions can account for a general switch from repression to activation, but do not necessarily account for the simultaneous activation and repression of sets of genes during MyoD-mediated myogenesis. Our expression array study with an inducible MyoD in fibroblasts showed that MyoD activates the expression of several distinct temporal clusters of genes and simultaneously suppresses the expression of other gene sets (). In our current study, we use this model system of MyoD-mediated myogenesis to determine how a single transcription factor simultaneously activates and suppresses different sets of genes during myogenic differentiation. We demonstrate that MyoD directly regulates the transcription of microRNA (miRNA) expression that suppresses specific targets during myogenic differentiation. In addition, our demonstration that a MyoD-induced miRNA targets the Utrn RNA and can posttranscriptionally suppress expression through this sequence suggests that therapies of Duchenne muscular dystrophy based on increasing Utrn expression should include modulation of these posttranscriptional mechanisms. Previously, we have used microarray studies to show that MyoD both induces expression of a large number of genes and simultaneously elicits a reduction in the RNA abundance of a subset of genes (). To study the mechanism of decreased RNA abundance by MyoD, we focused on follistatin-like 1 (). Although does not have a defined biological role in muscle physiology, we chose to study this gene because it showed a robust decrease in its message during the first 24 h of MyoD induction. As in our previous studies, we used mouse embryonic fibroblasts (MEFs) derived from mice with both the endogenous MyoD and Myf5 genes disrupted (M+M cells) that contain a constitutively expressed fusion protein between mouse MyoD and the hormone-binding domain of the estrogen receptor (MyoD estrogen receptor [MDER]; ; ). Transition of the cells to low-serum differentiation medium (DM; see Materials and methods) with the addition of β-estradiol synchronously induces MyoD activity and initiates differentiation in these cells (). Consistent with our previous findings, Fstl1 RNA decreased several fold at 24 h after the induction of MyoD activity by β-estradiol in the M+M cells, whereas the abundance of Myog, which is a downstream transcriptional target of MyoD, increased substantially (). The decreased abundance of Fstl1 RNA in response to MyoD induction and myogenic differentiation was also evident in the myoblast cell line C2C12 that expresses endogenous MyoD and in the 10T1/2 MEF cell line that constitutively expresses MDER (). Notably, the abundance of Fstl1 RNA decreases when MyoD is induced in the presence of the protein synthesis inhibitor cycloheximide, indicating that MyoD causes a decrease in this RNA in the absence of new protein synthesis (). MyoD is a transcription factor with a well-defined DNA-binding domain and an N-terminal acidic activation domain. To determine whether the DNA-binding or activation functions of MyoD are necessary to suppress the abundance of Fstl1, we tested the activity of two MDER mutant proteins. MDproER has an inactivating point mutation (A114P) in the MyoD DNA-binding domain (), whereas MDΔNER has a deletion of the MyoD activation domain (aa 3–56; ). These mutant proteins are expressed at levels comparable to wild-type MDER in M+M cells (unpublished data). Consistent with prior studies showing that the activation functions of E-protein heterodimers and other recruited factors can partially compensate for the absence of the MyoD activation domain on many promoters (), we found that MDΔNER had a modest but discernable effect on Fstl1 mRNA levels (, compare lanes 9 and 10), whereas the induction of the DNA binding–deficient MDproER mutant did not affect the abundance of Fstl1 (, compare lanes 7 and 8). Together, these data suggest that DNA binding by MyoD is required to diminish the abundance of Fstl1. We next used chromatin immunoprecipitation (ChIP) to determine whether MyoD was down-regulating Fstl1 through direct binding to regulatory regions. We identified conserved regions within 5 kb of the promoter and used primers targeted to these regions (Fig. S1, available at ); however, because of the size of the fragmented DNA for the ChIPs it is likely that we effectively screened the entire region 2.5 kb upstream of the promoter. Using antisera to the MyoD protein, we looked for enrichment of MyoD at the regulatory regions, compared with negative () and positive () internal controls where we have previously determined MyoD binding (; ; ). ChIP assays showed robust MyoD binding at the regulatory region at 24 h after MyoD induction, but none of the regions in the vicinity of the promoter showed enrichment relative to the internal negative control (Fig. S1), indicating that MyoD was not directly binding near the promoter of the gene. To determine whether RNA transcription is necessary for MyoD to decrease the abundance of Fstl1 RNA, we treated cells with α-amanitin, which, at low concentrations, is an RNA polymerase II inhibitor (). At doses of α-amanitin that are sufficient to prevent the transcription of MyoD target genes, we found that Fstl1 RNA was no longer down-regulated upon MyoD induction (). These results indicate that MyoD decreases the abundance of Fstl1 in a manner that requires DNA binding and transcription, but not protein translation, and is consistent with MyoD-regulated expression of a regulatory RNA, such as a miRNA or siRNA, which posttranscriptionally regulates the abundance of the Fstl1 mRNA. Although in many cases miRNAs block translation, RNA degradation can also be induced by miRNA (for review see ; ). We used the miRanda algorithm to identify several miRNAs that are predicted to potentially bind sites in Fstl1 (). Within this candidate set of miRNAs, Northern blot analysis demonstrated that the abundance of miR-206 was dramatically up-regulated by the induction of MyoD activity (). Up-regulation of miR-206 was dependent on active MyoD, as cells that did not express MDER did not show miR-206 expression (Fig. S2, available at ). These data are consistent with recently published studies indicating an up-regulation of miR-206 in differentiated C2C12 myoblasts (; ). In addition, miR-206 was not induced by the MDproER mutation in the DNA-binding domain (). The locus encoding mmu-miR-206 is on chromosome 1 and the 5′ end of the pre–miR-206 closely coincides with the 5′ end of a longer transcript (AK132542). A conserved E box is near the predicted transcription start of the putative miR-206 hairpin and the AK132542 transcript. ChIP assays show robust MyoD binding to this putative regulatory region of miR-206 (). (In addition, while this manuscript was under review, identified miR-206 as a MyoD target by ChIP.) Northern and RT-PCR analysis demonstrate that the longer AK132542 transcript is also induced by MyoD (Fig. S3, available at ), suggesting that a larger RNA might be transcribed and processed to produce miR-206. It is interesting to note that the miR-133b sequence is also contained in the AK132542 transcript. miR-133 has recently been shown to be induced during differentiation of C2C12 muscle cells (; ), and miR-133 is induced by MyoD together with miR-206 in M+M MDER cells (unpublished data). Therefore, it is possible that the induction of AK132542 by MyoD might lead to the production of both miR-206 and -133. Our data indicate, therefore, that MyoD directly binds the promoter region of the miR-206 precursor, and the binding is correlated with transcription of a longer precursor RNA and the appearance of the mature miR-206. To determine whether the miR-206 can target the 3′UTR of the gene, we cloned three copies of the putative target sequence () into the 3′ UTR of a luciferase reporter gene (CS2-luc-Fstl1) and, as a control, we cloned similar multimers of the target sequence containing mutations that should disrupt miR-206 binding (MutA and MutB). Compared with the parent vector, the expression of the CS2-luc-Fstl1 reporter was significantly suppressed by MyoD induction, whereas the MutA and MutB were not suppressed by MyoD (). These data suggest that MyoD activation results in specific repression targeted to the Fstl1 3′UTR. To show a direct effect of miR-206 on the CS2-luc-Fstl1 reporter, we cotransfected cells with a miR-206 hairpin precursor (pre–miR-206), which is processed by the M+M MDER cells to produce the mature, 22 nucleotide miR-206 (). Consistent with the effect of MDER induction, we see a decrease in CS2-luc-Fstl1 reporter activity, but not in MutA or MutB reporter activity, when cells are cotransfected with pre–miR-206. Together, these data demonstrate that the conserved sequence motif in the 3′-UTR of Fstl1 is targeted by miR-206. To confirm that miR-206 can target the Fstl1 mRNA, and to develop a method to identify additional miR-206 targets, we used a miR-206 oligonucleotide as a primer in a reverse transcriptase reaction to see if it would prime the Fstl1 mRNA, as detected by subsequent PCR. An oligonucleotide of the miR-206 sequence primed first strand synthesis from the Fstl1 mRNA, whereas neither desmin nor Timm17b (a constitutively expressed control RNA) were primed by miR-206 (). Cloning and sequencing the miR-206–primed cDNA confirmed that miR-206 binds to the predicted target site in Fstl1 and to an additional site further 3′ in the RNA, suggesting that multiple miR-206–binding sites might exist in the Fstl1 3′ UTR (unpublished data). In considering the aforementioned data, we conclude that MyoD down-regulates Fstl1 by transcriptionally activating the miR-206 gene, which posttranscriptionally regulates Fstl1 through the conserved motif in its 3′UTR sequence. Although little is known about the potential function of Fstl1 in myogenesis, other predicted targets of miR-206 have been extensively studied. For example, Utrn is a predicted target of miR-206 (). Consistent with the predicted interaction, a miR-206 oligonucleotide can prime the Utrn mRNA in a reverse transcriptase reaction (), and cloning and sequencing of the miR-206–primed cDNA confirmed that the miR-206 oligonucleotide binds to the predicted target site in Utrn (not depicted). The predicted miR-206–binding site in Utrn is highly conserved in mouse, human, and dog and shares the same “seed” sequence as the miR-206 site in Fstl1 (). In our model system of MyoD-induced myogenesis, there is a small decrease in the abundance of the Utrn mRNA after 12 h of MyoD induction () and a more marked decrease in the abundance of the protein at 48 and 96 h of MyoD induction, which is dependent on the presence of an active MyoD (, compare the pBABE control cells with the MDER cells). A CS2-luciferase reporter containing a multimer of the putative miR-206–binding site from the Utrn mRNA shows substantial inhibition when cotransfected with the pre–miR-206 construct (), demonstrating that this site is indeed targeted by miR-206. Another muscle-specific miRNA, miR-1, has high sequence homology to miR-206, and it is also induced in the M+M MDER cells (unpublished data). Transfection of pre–miR-1 had a significant, but more modest, suppression of the CS2–luc–Utrn construct, and cotransfection of both pre–miR-206 and -1 did not show synergistic activity. Therefore, we conclude that the induction of miR-206 by MyoD can alter the stability and/or the translation of subpopulations of RNAs, some of which are biologically relevant to the process of terminal myogenic differentiation. Furthermore, the demonstration that at least a portion of the suppression of Utrn expression occurs posttranscriptionally suggests specific therapies aimed at increasing or maintaining Utrn expression in Duchenne muscular dystrophy. Expression of MyoD in a wide variety of cell types is sufficient to activate the program of skeletal muscle differentiation (). Although different sets of genes are expressed in specific temporal patterns during muscle differentiation, we have shown that MyoD directly binds to the regulatory regions of genes expressed both early and late in the program (). Temporal patterning is achieved through a feed-forward mechanism, where transcription factors activated by MyoD early in the myogenic program feed-forward to cooperate with MyoD to activate genes expressed later in the program (). In this manner, a single transcription factor can directly orchestrate the temporal program of gene activation during skeletal muscle differentiation. Our findings in this study demonstrate that the same transcription factor that has a central role in activating muscle-specific genes simultaneously induces miRNAs that repress gene expression during myogenic differentiation. Specifically, we demonstrate that MyoD acts as a transcriptional activator of the miR-206 pre–miRNA transcript and the induced high levels of the mature miR-206 result in the down-regulation of specific target genes, such as Fstl1 and Utrn. We chose to investigate the regulation of Fstl1 because it was identified as a suppressed RNA in our time-course study of MyoD-regulated genes (). Fstl1 was originally identified as TSC-36 (TGF-β stimulated clone 36) in a screen for genes induced by TGF-β in mouse osteoblasts, and was observed to have some amino acid homology to Fst (). It is not currently known whether Fstl1 has overlapping or distinct functions from Fst, however, the different developmental expression patterns of these two genes () suggest distinct biological functions. Additionally, a functional domain important for Fst activity, the activin-binding domain (), is not conserved in Fstl1. Therefore, although Fst has been shown to antagonize activin and myostatin during myogenesis (), the role of Fstl1 in this process remains unknown. In contrast, the decreased expression of Utrn protein in skeletal muscle is thought to be an important process in differentiation. During muscle differentiation, dystrophin, which is a Utrn paralog, replaces the Utrn protein in the dystrophin-associated glycoprotein complex. Importantly, the loss of dystrophin protein is the etiologic basis for Duchenne muscular dystrophy. The demonstration that constitutive expression of Utrn protein in the skeletal muscle of transgenic mice () can partially compensate for the loss of dystrophin has led to attempts to transcriptionally induce gene expression in mature muscle cells, in the hope that Utrn up-regulation might prove therapeutic to DMD patients. Our demonstration that miR-206 can suppress expression posttranscriptionally through a sequence in the Utrn RNA suggests that therapies based on posttranscriptional regulation of the Utrn RNA should also be explored. Since the initial discovery that lin-4 acts as a regulatory RNA in (), there has been growing interest in understanding how miRNAs, acting posttranscriptionally, interface with well-characterized transcriptional regulatory networks to enforce precise regulation of gene expression programs in animals (). As a result, we are beginning to learn that miRNAs play important roles in embryonic development and cell fate (for review see ). Several groups have demonstrated tissue-specific distribution of miRNAs in developing embryos and adult animals (; ; ). Further, it seems that several miRNAs act in the specification of cell lineage. For example, miR-181, -223, and -142 are preferentially expressed in mouse hematopoietic tissues, where mir-181 specifies the B cell lineage and miR-223 promotes granulocyte differentiation (; ). Recently, the role of miRNAs in skeletal and cardiac muscle biology has been the focus of intense interest. To date, three muscle-specific miRNAs have been identified: miR-1, -133, and -206. miR-1 expression is strictly limited to cardiac and skeletal muscle in zebrafish, and mouse, although species-specific variations in expression patterns have been noted (; ; ; ; ; ). Expression of miR-1 is up-regulated in response to differentiating signals in cardiac and skeletal muscle in vivo and differentiated myoblasts in vitro, and it has been shown to be activated by several factors, including SRF, Twist, myocardin, and Mef2 (; ; ). Another muscle-specific miRNA, miR-133, has been shown to promote proliferation of myoblasts by antagonizing SRF (). Recently, a role for miR- 181 in muscle differentiation and regeneration was also described (). We now show that miR-206 is a direct transcriptional target of MyoD and functions to suppress targets during muscle differentiation. Although our study has focused on a single miRNA and its targets, we assume that multiple miRNAs will be similarly regulated by MyoD. Similarly, it is likely that the related myogenic bHLH proteins (Myf5, Myog, and Mrf4) will also induce miR-206. Indeed, our recent ChIP study identified four additional miRNA promoter regions bound by MyoD: miR-100, -138-2, -191, and -22 (), and a study published while this paper was under review showed that MyoD and Myog bind the putative regulatory regions of miR-206, -1, and -133 (). Because many of the targets of miRNAs can be translationally inhibited in the absence of RNA degradation, the expression array studies we used as the basis for investigating the regulation of Fstl1 are likely underestimating the extent of genes suppressed during myogenesis. Additionally, because it has been demonstrated that transfection of a single miRNA (e.g., miR-1) into HeLa cells decreases nearly 100 mRNA transcripts (), it seems likely that future studies will show multiple additional genes to be targeted by miRNAs in response to MyoD. MEFs null for both and (M+M cells) were infected with a retrovirus containing the pBABE vector expressing a puromycin-resistance gene and MDER (). Mutant MDER proteins (MDproER and MDΔNER) were constructed in the same vector context. Control cells (M+M pBABE) were infected with virus containing only vector and selectable marker. Cells were maintained in growth medium, which consisted of DME containing 1% -glutamine and 10% bovine calf serum (Hyclone). Infected cells were selected in 1.2 μg/ml puromycin. To induce expression of MDER, cells were switched to DM, which consisted of DME containing 1% -glutamine, 0.5% horse serum, 10 μg/ml insulin, 10 μg/ml transferrin, and 10 M β-estradiol. All transient transfection/luciferase assays were performed in triplicate on 35-mm tissue culture dishes (Corning). For transfections, each plate was seeded with 10 cells and incubated overnight at 37°C. Cells were transfected using Superfect reagent (QIAGEN) as specified by the manufacturer. The constructs used for transfections contained the indicated regions of Fstl1 or Utrn 3′UTR sequence inserted 3′ of the firefly luciferase gene, under the control of the CMV promoter in the CS2 vector background. Each plate was transfected with 100 ng of CS2-luc reporter vector and 2 μg of empty CS2 vector. Where indicated, cells were cotransfected with pre–miR-206, -1, or -Negative Control #1 (Ambion) to a final concentration of 25 nM, and luciferase activity was measured 24 h after transfection. Where indicated, MDER activity was induced in transfected cells at 16–24 h after transfection by switching cells to DM with 10 M β-estradiol (uninduced cells were switched to DM without β-estradiol), and luciferase activity was measured at 24 h after induction. For all experiments, cells were cotransfected with 200 ng of CS2-β-gal, and assayed for β-galactosidase activity as an internal control for transfection efficiency using the MUG assay. Luciferase assays and MUG assays were performed as previously described (), using AutoLumat LB 953 (BERTHOLD TECHNOLOGIES) and MicroFluor (Dynatech) instrumentation. Cultured cells were harvested by scraping, and RNA was prepared using the RNEasy kit (QIAGEN). Northern blot analysis was performed according to standard techniques, which were previously described (). Probes were generated by PCR amplification of cDNA reverse transcribed from total M+M MDER RNA using an oligo-dT primer and SuperScript II (Invitrogen), and were radioactively labeled using Ready-to-Go DNA labeling beads (GE Healthcare). Images were captured on film, digitized, and if needed, minor linear adjustments in contrast were made using Adobe Photoshop software. Total RNA was prepared using Trizol (Invitrogen) extraction, per the manufacturer's instructions. Trizol purification was followed by acid phenol extraction to remove any DNA contamination. Each sample contained 20 μg RNA. Samples were separated electrophoretically in a 20% polyacrylamide/8 M urea/1× TBE gel. RNA was electroblotted onto Nytran SPC nylon membrane in 1× TBE at 250 mA for 45 min, and was fixed to the membrane by UV cross-linking. Blots were hybridized overnight at 35°C in Ultrahybe Oligo buffer (Ambion) with radiolabeled oligonucleotide probes complementary to the mature miRNA sequences of interest. Images were captured on film, digitized, and if needed, minor linear adjustments in contrast were made using Photoshop software (Adobe). ChIP analysis was performed as previously described (; ). Precipitations from 750–1,000 μg M+M MDER cell lysate were incubated overnight at 4°C with 5 μl anti-MyoD antiserum (). Duplex PCR was performed by coamplifying test control regions from MyoD target genes with an internal control region from the enhancer. Amplification reactions included P-dCTP for incorporation into PCR product that was then detected and quantified by PhosporImager analysis using ImageQuant software (Molecular Dynamics). For input samples, a titration of 0.03–3.00 ng of genomic DNA was used as a template for duplex PCR to establish relative ratios of PCR product in the absence of any asymmetry in target abundance. For MyoD IPs, 5% of the IP sample was used per reaction and amplified over 32 PCR cycles. Results were analyzed for enrichment at the muscle-specific target sequence relative to the target sequence/control sequence ration established using input DNA sample. PCR linearity was confirmed by titrating the input DNA over a 30-fold range. Efficiency of MyoD IP was confirmed by analysis of MyoD enrichment at the promoter, as described previously (). Cultured cells were harvested by scraping, resuspended in RIPA lysis buffer with 5% SDS (150 mM NaCl, 10 mM Tris, pH 7.2, 1% Triton X-100, 1% deoxycholate, and 5 mM EDTA), and homogenized by repeated manipulation through a 22-gauge needle. Protein concentration for each lysate preparation was determined by BCA assay. Samples were concentrated by methanol-chloroform extraction () and resuspended in loading buffer. Each sample analyzed contained 70 μg total protein. Samples were electrophoresed on a 6% SDS-PAGE and transferred to nitrocellulose for 3.5 h at 500 mA. Utrophin was detected using goat anti-utrophin (E-16) polyclonal antibody (Santa Cruz Biotechnology, Inc.); Hsp70 was detected using mouse anti-Hsp70 monoclonal antibody (Stressgen). Images were captured on film, digitized, and if needed, minor linear adjustments in contrast were made using Photoshop software. Fig. S1 shows ChIP analysis of MyoD binding at the Fstl1 promoter, where MyoD binding was not detected. Fig. S2 shows that M+M pBABE cells (M+M cells that do not express MDER) do not express miR-206 upon switching to DM. Fig. S3 shows that MyoD induces the expression of the AK132542 transcript. Online supplemental material is available at .

Skeletal muscle mass is determined by both muscle fiber number and muscle fiber size. Skeletal muscle fibers are syncytial in nature, and an increase in myonuclear number is required for myofiber growth not only during muscle development (; ) but also during postnatal muscle growth (; ; ). Postnatal muscle growth is dependent on myonuclei addition via myoblast proliferation and fusion to preexisting myofibers (). However, the signaling cascades that regulate myofiber growth during development and postnatal myofiber growth remain to be fully defined. Protein tyrosine phosphatases (PTPs) play an essential role in regulating the balance of cellular protein tyrosyl phosphorylation. PTPs have been implicated in processes such as cell proliferation, differentiation, and development (). SHP-2 is a ubiquitously expressed cytoplasmic PTP with two Src-homology 2 domains, a catalytic domain, and a C terminus containing two tyrosyl phosphorylation sites (; ). SHP-2 participates in signaling events downstream of growth factor receptors, cytokines, hormones, and integrins to control cell proliferation (; ), cell adhesion (; ), and cell survival (; ). SHP-2 is essential for development because its mutation or deletion in mice results in embryonic lethality (; ). SHP-2 transduces positive signals through the Ras (; ), phosphatidylinositol 3′-kinase (; ; ), and Src pathways (; ). Although a complete understanding of how SHP-2 propagates intracellular signals remains to be determined, the basis for some of its positive signaling effects has been uncovered. SHP-2 controls the localization of C-terminal Src kinase (CSK), a negative regulator of c-Src, by dephosphorylating CSK-binding proteins such as Pag/CSK-binding protein and paxillin, thereby positively regulating c-Src activation (; ). SHP-2 can control the recruitment of p120 Ras–GTPase-activating protein to the plasma membrane in order to affect Ras activation (; ; ). In myoblasts, SHP-2 positively regulates RhoA signaling by dephosphorylating p190-B Rho–GTPase-activating protein, leading to the activation of RhoA-dependent muscle-specific gene expression (). These observations provide important mechanistic insights into the pleiotropic signaling effects of SHP-2. The nuclear factor of activated T cells (NFAT) family is comprised of five members (NFAT1–5) that are implicated in a variety of developmental and disease processes (; ). NFAT1, 2, and 4 are highly expressed in skeletal muscle. The NFATs appear to play differential roles during skeletal muscle development, as indicated by the distinct skeletal muscle defects displayed by mice lacking individual NFAT family members. mice have reduced muscle mass as a result of decreased fiber number (), whereas mice exhibit reduced muscle mass because of smaller myofibers (). During myogenesis, the NFATs translocate into the nucleus (). Nuclear translocation of NFAT is mediated by the calcium-activated serine/threonine phosphatase calcineurin. When activated, calcineurin dephosphorylates NFAT, causing it to translocate to the nucleus to stimulate gene expression. NFAT has been proposed to mediate skeletal muscle growth by activating the transcription of interleukin-4 (IL-4), which acts as a myoblast recruitment factor to facilitate myotube multinucleation (; ). Although these observations provide insight into the growth-promoting effects of NFAT on skeletal muscle, how NFAT is regulated in skeletal muscle remains to be fully defined. Because of the early embryonic lethality observed in mice lacking SHP-2 (; ), we have used the Cre-loxP system to disrupt the gene in order to study its role in skeletal muscle function. We show that SHP-2 coordinates signals that stem from the extracellular matrix to target the NFAT signaling pathway, which promotes myofiber-type formation and skeletal muscle growth. To investigate the function of SHP-2 in skeletal muscle, we used the Cre-loxP system to conditionally inactivate in the myogenic lineage. SHP-2 (lox/lox) mice were generated by engineering loxP sites flanking exon 11 of SHP-2, which encodes its catalytic motif (). We bred SHP-2 (lox/lox) mice with transgenic mice expressing Cre recombinase from the muscle creatine kinase (MCK) promoter () to generate mice with a skeletal muscle–specific disruption of , designated herein as MCK–SHP-2 null. We found that MCK–SHP-2–null mice at 1 wk of age were phenotypically indistinguishable from their SHP-2 (lox/lox) littermates. Skeletal muscle extracts from the hind limb showed that SHP-2 was expressed to comparable levels in MCK–SHP-2–null and SHP-2 (lox/lox) control mice at 1 wk of age (). These results demonstrated that SHP-2 is not deleted embryonically. Consistent with this, MCK–SHP-2–null mice exhibited neither embryonic lethality nor developmental skeletal muscle defects. MCK–SHP-2–null mice also appeared to grow normally (). As MCK–SHP-2–null mice aged, SHP-2 expression in skeletal muscles declined, whereas in SHP-2 (lox/lox) mice, SHP-2 expression remained constant (). By 11 wk of age, the expression of SHP-2 in the soleus, tibialis anterior (TA), and gastrocnemious muscles was undetectable in MCK–SHP-2–null mice (). SHP-2 was expressed at equivalent levels in tissues other than striated muscles, such as the liver, as compared with SHP-2 (lox/lox) controls (). Thus, MCK–SHP-2–null mice exhibit the postnatal skeletal muscle–specific deletion of SHP-2. At the histological level, skeletal muscle from MCK–SHP-2–null mice did not reveal any striking abnormalities (). Immunofluorescence staining of skeletal muscle sections also revealed no apparent differences in the expression of either β-dystroglycan or dystrophin between MCK–SHP-2–null and SHP-2 (lox/lox) mice (). However, in skeletal muscle sections of MCK–SHP-2–null mice, the integrity of laminin staining was slightly irregular (). The altered laminin staining did not appear to be indicative of muscle damage because there were no differences in the number of centrally located myonuclei between MCK–SHP-2–null and SHP-2 (lox/lox) mice (unpublished data). We noted that the myofibers from MCK–SHP-2–null mice appeared smaller as compared with SHP-2 (lox/lox) controls (). Therefore, we measured the cross-sectional area (CSA) of myofibers in skeletal muscles from male MCK–SHP-2–null and SHP-2 (lox/lox) mice. MCK–SHP-2–null male mice had a significant (P < 0.01) reduction in CSA of the TA and diaphragm myofibers as compared with SHP-2 (lox/lox) mice (). Similar reductions in myofiber CSAs were also observed in the TA and diaphragm of MCK–SHP-2–null female mice (). The fiber number in the TA muscles of male MCK–SHP-2–null mice was also reduced (unpublished data). The size distribution of myofibers in the TA muscles of male MCK–SHP-2–null mice showed an increase in small myofibers (CSA < 1,200 μm) and a reduction in large myofibers (CSA > 2,400 μm; , left). An increase in small myofibers (CSA < 500 μm) and a decrease in large myofibers (CSA > 1,000 μm) in the MCK–SHP-2–null diaphragm was also observed (, right). The reduced myofiber CSA seen in MCK–SHP-2–null mice suggests that impaired myofiber growth occurs through a defect in myoblast fusion. To examine this in more detail, we quantitated myonuclear number in cross sections derived from the soleus muscle of MCK–SHP-2–null and SHP-2 (lox/lox) mice. shows that MCK–SHP-2–null mice contained ∼25% less myonuclei than SHP-2 (lox/lox) controls. Collectively, these results suggest that SHP-2 is required for skeletal muscle growth by promoting myotube multinucleation. We next examined whether muscle fiber type composition was perturbed in MCK–SHP-2–null mice. Depending on the oxidative capacity, muscle fiber types are classified as either type I slow or type II fast fibers. We examined fiber type composition by performing immunostaining on skeletal muscle sections derived from SHP-2 (lox/lox) and MCK–SHP-2–null mice with antibodies directed to either type I slow or type II fast myosin heavy chain (MHC). MCK–SHP-2–null mice exhibited a marked reduction (∼25%) in the number of type I slow fibers as compared with SHP-2 (lox/lox) mice (), whereas no differences were observed in the proportion of type II fast fibers (). These results demonstrate that SHP-2 contributes to the formation and/or maintenance of type I slow fibers. In adult mice, fiber type specification is regulated by the calcium-activated serine/threonine protein phosphatase calcineurin (; ). Our data indicating reduced type I slow fibers in MCK–SHP-2–null mice raised the possibility that SHP-2 might regulate calcineurin-dependent signaling in skeletal muscle. Interestingly, the reduction in myofiber size in MCK–SHP-2–null mice was similar to that of skeletal muscle defects described in mice lacking the calcineurin substrate NFAT1 (). NFAT1 regulates IL-4 expression, which promotes muscle growth by stimulating fusion (; ). We hypothesized that the smaller myofibers in MCK–SHP-2–null mice might be caused by the reduced NFAT-mediated transcriptional expression of IL-4. RT-PCR analysis showed that IL-4 mRNA expression was virtually undetectable in MCK–SHP-2 null as compared with SHP-2 (lox/lox) muscle (). Neither the regulatory nor catalytic subunits of calcineurin or NFAT1 were altered in their expression levels in MCK–SHP-2–null mice (unpublished data). The observation that IL-4 levels were reduced in MCK–SHP-2–null mice suggests that the defect in myofiber size occurs because of a failure of myoblasts being recruited to the nascent myofiber to fuse. To test this, we compared the ability of myoblasts from MCK–SHP-2–null and SHP-2 (lox/lox) mice to form multinucleated myotubes. SHP-2 expression in proliferating myoblasts isolated from MCK–SHP-2–null mice was equivalent to that of myoblasts from SHP-2 (lox/lox) mice (). Proliferating myoblasts have low levels of MCK activity (; ), resulting in an insufficient level of Cre-mediated SHP-2 deletion. Similar results have been reported previously using this MCK-Cre transgene (). Even when myoblasts from MCK–SHP-2–null mice were induced to differentiate, the level of MCK-mediated Cre expression was still too low to result in the deletion of SHP-2 (unpublished data). In vivo SHP-2 would be expected to be deleted from MCK–SHP-2–null myoblasts during myogenesis. To recapitulate the in vivo elimination of SHP-2 during differentiation, we infected myoblasts with an adenoviral vector expressing Cre recombinase concurrent with the initiation of myoblast differentiation. As shown in , Ad-Cre infection of myoblasts isolated from floxed SHP-2 mice concurrent with the initiation of myoblast differentiation resulted in the maximal expression of Cre recombinase within 48 h. Concomitant with the increased expression of Cre recombinase, SHP-2 was deleted during myogenesis, and, by 72 h, its expression was reduced by ∼80% as compared with Ad-GFP–infected myoblasts (). Using these conditions, SHP-2 floxed myoblasts were infected with Ad-GFP as a control or with Ad-Cre concurrent with the onset of myoblast differentiation. After 8 d of differentiation, control myoblasts formed prominent multinucleated myotubes (). However, SHP-2–deleted myoblasts were dramatically deficient in their ability to form myotubes (). Of the myotubes formed in the Ad-Cre–infected cultures, ∼9% contained more than five nuclei, whereas ∼58% of control myotubes contained more than five nuclei (). Concomitantly, differentiated SHP-2–deleted myoblasts accumulated myotubes with two to four nuclei to a significantly (P < 0.05) greater extent than control myotubes (). These data demonstrate that SHP-2 is required to mediate myotube multinucleation. Because the expression of IL-4 is diminished in MCK–SHP-2–null mice (), we tested whether NFAT transcriptional activity is reduced in the absence of SHP-2. An adenoviral NFAT-luciferase reporter (Ad-NFAT-luc) that contained the NFAT-binding site from the IL-4 promoter () was used to determine NFAT activity in differentiating myoblasts. SHP-2 expression was eliminated during myoblast differentiation as described in . After 2 d in differentiation medium (DM), NFAT activity was significantly (P < 0.05) lower by up to 60% as compared with Ad-GFP–infected myoblasts (). These results indicate that SHP-2 stimulates NFAT activation during myoblast differentiation. We examined whether SHP-2 promotes NFAT activity through its phosphatase domain by determining the effects on NFAT transcriptional activity in differentiating myoblasts upon the overexpression of a dominant-negative (DN) mutant of SHP-2 that lacks catalytic activity. For these experiments, we used C2C12 myoblasts because primary myoblasts were not conducive to transient transfection. C2C12 myoblasts were transiently transfected with vector control, wild-type (WT) SHP-2, or a catalytically inactive mutant of SHP-2 (SHP-2 cysteine to serine [CS]). NFAT transcriptional activity was assessed 48 h later during differentiation using the IL-2 minimal promoter containing NFAT-binding sites fused to the luciferase gene. The expression of SHP-2–CS in differentiating C2C12 myoblasts suppressed NFAT transcriptional activity significantly (P < 0.05) by ∼50% as compared with SHP-2 WT (). Because SHP-2 only partially interfered with NFAT activity, it suggests that additional signaling components are required to provide the optimal activation of NFAT. SHP-2 is required for integrin-induced activation of the Src family kinases (SFKs; ; ). In addition, we have shown that SHP-2 catalytic activity increases during C2C12 differentiation (). We hypothesized that SHP-2 positively regulates NFAT via c-Src to promote myotube multinucleation. First, we examined c-Src activity during C2C12 differentiation. Cell lysates obtained from C2C12 myoblasts cultured in growth medium (GM) for 24 h or in DM for 24, 48, and 72 h were subjected to immunoprecipitation with c-Src antibodies, and immune complexes were immunoblotted with antibodies against total c-Src and the activating tyrosine 416 phosphorylation site (c-Src–pY416). We found that c-Src expression was up-regulated, and, when normalized to total c-Src, c-Src–pY416 levels were induced by approximately sixfold during C2C12 myogenesis (). Thus, c-Src–specific activity and total expression levels increase during myogenesis. To determine whether c-Src activates NFAT during myogenesis, we used a DN mutant of c-Src that renders it kinase inactive (Src-DN). C2C12 myoblasts were differentiated for 24 h and were then transfected with Src-WT and Src-DN. When overexpressed in differentiating C2C12 myoblasts, Src-WT stimulated NFAT transcriptional activity (). In contrast, the overexpression of Src-DN inhibited NFAT transcriptional activity by ∼40% when transfected into differentiating C2C12 myoblasts as compared with vector control transfectants (). Moreover, c-Src–pY416 phosphorylation was markedly decreased in C2C12 myoblasts infected with a catalytically inactive/nonsubstrate-trapping mutant of SHP-2 (Ad–SHP-2 arginine to methionine [RM]) but not with SHP-2–WT (). Therefore, in myoblasts, SHP-2 acts upstream of c-Src. To establish the relationship between SHP-2 and c-Src for NFAT activation in differentiating myoblasts, we tested whether a gain of function mutant of c-Src could rescue the inhibitory actions of a catalytically inactive mutant of SHP-2 on NFAT activity. Again, differentiated C2C12 myoblasts were transfected, and an assessment of NFAT activity was performed 48 h later. Compared with vector control, constitutively active c-Src (c-Src–Y527F) stimulated NFAT activity approximately fourfold (). The inhibitory effect exerted by SHP-2–CS on NFAT was rescued upon coexpression with c-Src–Y527F but not to levels equivalent to that of c-Src–Y527F alone (). In light of the results shown in , this partial rescue of NFAT activity by c-Src–Y527F suggests that c-Src lies downstream of SHP-2 in the NFAT pathway. Failure to completely restore NFAT activity to levels obtained by c-Y527F alone again raises the possibility that multiple pathways cooperate with SHP-2 to promote NFAT activity in differentiating myoblasts. It has been suggested that the signal regulatory protein-1α (SIRP-1α; ; ; ; ), a transmembrane glycoprotein, participates in a positive-feed forward pathway to promote SHP-2 signaling in response to integrins (). Moreover, SIRP-1α becomes tyrosyl phosphorylated and forms a complex with SHP-2 during myogenesis (). We hypothesized that in differentiating myoblasts, SIRP-1α may serve to recruit SHP-2 in an integrin-dependent manner to propagate SHP-2/c-Src signaling. We first determined whether SIRP-1α tyrosyl phosphorylation and association with SHP-2 were dependent on SFK activity in differentiating myoblasts. C2C12 myoblasts induced to differentiate for 24 h were treated with the SFK inhibitor PP2 for an additional 24 and 48 h (). We found that PP2 treatment dramatically decreased SIRP-1α tyrosyl phosphorylation and association with SHP-2, demonstrating that SHP-2 complexes with SIRP-1α in myoblasts in an SFK-dependent manner. If our hypothesis that c-Src–mediated SIRP-1α tyrosyl phosphorylation and subsequent recruitment of SHP-2 are critical for propagating SHP-2/c-Src signaling, multinucleated myotube formation should be reduced upon inhibition of the SFKs. To test this, C2C12 myoblasts were induced to differentiate for 24 h and were treated either with or without PP2 for an additional 48 h. PP2 inhibited multinucleated myotube formation both phenotypically (, top) and quantitatively (, bottom). These data imply that SFKs stimulate the recruitment of SHP-2 to tyrosyl-phosphorylated SIRP-1α, resulting in the initiation of SHP-2 signaling to promote myoblast fusion. Next, we investigated whether integrin-dependent engagement couples SIRP-1α to SHP-2 to promote myotube formation. SIRP-1α tyrosyl phosphorylation in myoblasts was induced in response to integrin engagement (). We next determined whether the DN mutant of SHP-2 disrupts integrin-mediated multinucleated myotube formation. C2C12 myoblasts were infected with adenoviruses encoding for Ad-GFP, Ad–SHP-2–WT, or the catalytically inactive/nonsubstrate-trapping mutant of SHP-2, Ad–SHP-2–RM. Adhesion to fibronectin was essential for C2C12 myogenesis because C2C12 myoblasts were unable to form either multinucleated myotubes or express MHC when plated on poly--lysine (unpublished data). Although SHP-2–WT expression did not affect multinucleated myotube formation, the expression of SHP-2–RM prevented both MHC expression and multinucleated myotube formation on fibronectin (). Altogether, these data argue that in differentiating myoblasts, integrin engagement promotes SIRP-1α tyrosyl phosphorylation through the SFKs, resulting in the recruitment of SHP-2, which facilitates c-Src–mediated multinucleated myotube formation. Skeletal muscle growth occurs during embryonic development and through the early stages of life. In mammals, two waves of myogenesis termed primary and secondary myogenesis occur (; ). In mice, by day 11 of embryogenesis, primary myogenesis takes place. Muscle growth during embryogenesis after day 14 then proceeds through secondary myogenesis, in which secondary myofibers form. Postnatally, myoblasts continue to fuse with the existing myofiber, thus contributing to muscle growth (; ). The signaling pathways that participate in controlling muscle growth postnatally remain poorly defined. In this study, we have examined the in vivo role of SHP-2 in skeletal muscle by generating a skeletal muscle–specific deletion of SHP-2. We found that skeletal muscle development in MCK–SHP-2–null mice was normal. The lack of a skeletal muscle developmental defect in MCK–SHP-2–null mice is attributable to the observation that MCK–SHP-2–null mice ablate SHP-2 expression postnatally. In postnatal skeletal muscles of MCK–SHP-2–null mice, we observed that myofiber size was reduced, demonstrating that SHP-2 is required for the growth of skeletal muscle. Postnatal skeletal muscle growth is mediated by a muscle somatic stem cell called a satellite cell. Satellite cells remain quiescent, but, when activated, they proliferate, differentiate, and fuse with the existing myofiber. Presumably, satellite cells in MCK–SHP-2–null mice, when activated, first proliferate and then differentiate, where upon they express MCK (), resulting in the deletion of SHP-2. Therefore, it is reasonable to conclude that the skeletal muscle phenotype of MCK–SHP-2–null mice is reflective of SHP-2 deficiency during myogenesis. Our results demonstrate that MCK–SHP-2–null mice accumulate smaller myofibers concomitant with a reduction in larger myofibers as well as a decrease in myonuclear number. These results are consistent with the interpretation that MCK–SHP-2–null mice have smaller myofibers as a consequence of a failure of nascent myofibers to grow into larger ones by recruiting myoblasts to fuse. This interpretation is supported by the result that cultured myoblasts in which SHP-2 was deleted during differentiation fail to undergo myotube multinucleation. There are other potential explanations for our results: a reduction in myoblast proliferation could result in a failure to fuse. However, we think this less likely because MCK is only expressed in differentiating myoblasts. Another possibility is that postnatal myofibers undergo apoptosis, resulting in a decrease in myofiber size in MCK–SHP-2–null mice. However, no observable differences in apoptosis were observed in skeletal muscles of MCK–SHP-2–null mice (unpublished data). Thus, the most plausible explanation for the reduced myofiber size in MCK–SHP-2–null mice is a failure of differentiating myoblasts to form multinucleated myotubes. Interestingly, MCK–SHP-2–null mice and NFAT1-deficient mice bear some similarities. Like NFAT1-deficient mice, MCK–SHP-2–null mice were found to have smaller myofibers and reduced levels of IL-4 expression in skeletal muscle. These results support the idea that NFAT1 is a target for positive regulation by SHP-2 in skeletal muscle. However, it remains to be determined whether the phenotype of MCK–SHP-2–null mice is a consequence of disrupting NFAT1 solely as opposed to other (or additional) NFAT family members. PTPs have been reported previously to regulate NFAT activity (). However, these studies were performed in hematopoietic cells and demonstrated the negative regulation of NFAT activation by PTPs (). Our results provide evidence for a positive role of PTPs in NFAT signaling. The precise mechanism through which SHP-2 regulates NFAT in skeletal muscle remains to be defined. SHP-2 plays an important role in the oscillatory control of calcium and NFAT in cardiomyocytes (). It is conceivable that in skeletal muscle, SHP-2 also participates in the regulation of calcium signaling, raising the possibility that it controls NFAT via activation of the calcium-dependent serine/threonine phosphatase calcineurin. Both the magnitude and duration of calcium signals in skeletal muscle are thought to be important for the regulation of calcineurin activity, which is thought to control muscle fiber type composition. A combination of genetic and biochemical data support a role for calcineurin in slow muscle fiber type formation (; ). Therefore, our results, which show a reduction in type I slow muscle fiber in MCK–SHP-2–null mice, suggest that SHP-2 may act upstream of calcineurin. SHP-2 indirectly activates c-Src by controlling Pag/CSK-binding protein or paxillin tyrosyl phosphorylation and, thus, the recruitment of CSK (; ). Engagement of the SH2 domains of SHP-2 is essential for its activation (). Previously, we demonstrated that SIRP-1α becomes tyrosyl phosphorylated and associates with SHP-2 during myogenesis (). Moreover, the kinetics of SIRP-1α–SHP-2 complex formation correlates with the induction of SHP-2 activation during C2C12 myogenesis (). In this study, we found that c-Src was active basally in proliferating myoblasts, and its activity increased during differentiation. Furthermore, c-Src activation was dependent on the catalytic activity of SHP-2. We propose that during myogenesis, SHP-2 is required for the activation of c-Src–dependent NFAT activation (). We have also shown that the c-Src–mediated activation of NFAT is reduced by the pharmacological inhibition of mitogen-activated extracellular-regulated kinase (Erk) kinase, suggesting that c-Src promotes NFAT transcriptional activity in myoblasts via Erk (unpublished data). Recent findings support both direct and indirect roles of Erk in the regulation of NFAT activity (). We have also found that SHP-2 promotes NFAT activation by stimulating the RhoA pathway in myoblasts (unpublished data). Thus, calcineurin-independent pathways exist through which SHP-2 stimulates NFAT activity in myoblasts (). The importance of the SFKs in myogenesis has been suggested previously (; ), and, here, we show that the inhibition of SFKs impairs myoblast fusion. We did observe that inhibition of the SFKs induced myoblast death, although the extent of myoblast death was not dramatic (unpublished data). Nevertheless, we cannot exclude the possibility that in addition to activating NFAT, c-Src promotes myotube formation by also maintaining myoblast survival. Although it is likely that NFAT may be only one of the downstream targets of c-Src involved in myogenesis, it is the first target of c-Src, which was defined here to be important for skeletal muscle growth. A considerable amount of data supports a critical role for integrin-mediated signals in the regulation of muscle development and myofiber integrity (). We have established a requirement for SHP-2 in integrin-mediated myogenesis. These results implicate SHP-2 as an early conduit linking extracellular matrix cues to downstream pathways involved in regulating muscle function (). When c-Src activity is abrogated in myoblasts, SIRP-1α tyrosyl phosphorylation and association with SHP-2 are inhibited, NFAT activation is diminished, and multinucleated myotube formation is blocked. These data predict that NFAT activation should be regulated in response to integrin engagement. In support of this, it has been shown that integrins can activate the NFATs (). One pathway through which the integrin-dependent activation of c-Src/SHP-2 signaling and, subsequently, NFAT-mediated gene expression might occur is through the activation of Erk, but, additionally, calcium may also play a role. In summary, we have identified SHP-2 as a novel regulator of skeletal muscle growth and slow type I skeletal muscle fiber formation. We have established a pathway through which SHP-2 regulates muscle growth by stimulating NFAT transcriptional activity. Furthermore, we show that SHP-2 stimulates c-Src in differentiating myoblasts to link extracellular matrix stimuli to intracellular signaling cascades that converge on NFAT. These results may provide the basis for further investigation into whether these signaling pathways are targets for dysregulation in skeletal muscle pathogenesis. The floxed allele was generated by introducing two loxP sites that flanked exon 11 encoding 55 amino acids containing the PTP signature motif. This floxed allele was generated in a 129Sv/B6 mixed background, crossed with MCK-Cre transgenic mice to obtain SHP-2 (lox/lox):MCK-Cre mice, and maintained by intercrossing with SHP-2 (lox/lox) mice. SHP-2 (lox/lox) littermates were used as controls in all experiments. All genotyping was performed by PCR using genomic DNA isolated from the mouse tail using primers for identifying mice carrying the Cre transgene (5′-ATGTCCAATTTACTGACC-3′ and 5′-CGCCGCATAACCAGTGAAAC-3′) or primers for the SHP-2 floxed allele (5′-TAGCTGCTTTAACCCTCTGTGT-3′ and 5′-CATCAGAGCAGGCCATATTCCG-3′). All histological and morphometric analyses were performed on male mice aged 11–13 wk unless otherwise indicated. The following antibodies were used: mouse monoclonal antibodies to SHP-2 and caveolin-3 (BD Transduction Laboratories) and to Src and phosphotyrosine (Upstate Biotechnology); rabbit polyclonal antibodies to Src, Fyn, laminin, c-Met, and Erk (Santa Cruz Biotechnology, Inc.); rabbit polyclonal SIRP-1α (described previously; ); and rabbit polyclonal to phospho-Src (Tyr416; Cell Signaling Technology). Rabbit polyclonal antibodies to β-dystroglycan and dystrophin were gifts from K. Campbell (The University of Iowa, Iowa City, IA). A mouse monoclonal Cre recombinase (clone 7.23) antibody and a rabbit polyclonal glyceraldehyde-3-phosphate dehydrogenase (GAPDH)–linked HRP antibody were purchased from AbCam. Type II MyHC (Fast) clone MY-32 and Type I MyHC (Slow) clone NOQ7.5.4D were purchased from Sigma-Aldrich. Human plasma fibronectin was purified as described previously (), poly--lysine was purchased from Sigma-Aldrich, and PP2 was purchased from Calbiochem. Replication-deficient adenoviral constructs encoding WT SHP-2 (Ad–SHP-2–WT), SHP-2 R465M mutant (Ad–SHP-2–RM), and GFP (Ad-GFP) have been described previously (). The pLXSH retroviral vectors containing Src-WT, the Src-Y527F mutant, and the K295R/Y527F double mutant (Src-DN) were provided by J.A. Cooper (Fred Hutchinson Cancer Research Center, Seattle, WA; ). pCA plasmids containing the RhoA V14 and N19 (Rho N19) mutants have been described previously (). The pIRES-GFP plasmids encoding SHP-2–WT or the catalytically inactive mutant of SHP-2 C459S (SHP-2–CS) have been described previously (). Replication-deficient adenoviral constructs encoding Cre (Ad-Cre) or control GFP (Ad-GFP) were provided by F.J. Giordano (Yale University, New Haven, CT). The adenoviral NFAT-luciferase reporter (Ad-NFAT-luc) construct was provided by J.D. Molkentin (Children's Hospital Medical Center, Cincinnati, OH; ). C2C12 myoblasts were maintained as described previously (). All tissue culture reagents were obtained from Invitrogen. For histological analysis, muscles from MCK–SHP-2–null and SHP-2 (lox/lox) controls were dissected and fixed in formalin, embedded in paraffin, and stained with hematoxylin and eosin. Transverse muscle sections were visualized by using a microscope (Axiovert 25; Carl Zeiss MicroImaging, Inc.) and photographed by using a charge-coupled device camera (DAGE 330; Meyer Instruments). The CSA of myofibers was measured using ImageJ software (version 1.36b; National Institutes of Health [NIH]). 200–550 muscle fibers per muscle group were measured. For immunofluorescence staining, isolated muscles were embedded in optimal cutting temperature (Tissue-Tek) and frozen in liquid nitrogen–cooled isopentane. 10-μm transverse cross sections were dried for 2 h at 45°C, fixed in 2% PFA in Dulbecco's PBS for 5 min at room temperature, and permeabilized in methanol at −20°C for 5 min. Sections were blocked with 10% normal goat serum (NGS) in PBS containing 1 mg/ml RNase A for 30 min at 37°C and were incubated with anti–β-dystroglycan (1:25) or antidystrophin (1:50) antibodies diluted in 10% NGS/PBS for 1 h. Sections were washed with PBS and incubated for 1 h with 1 μg/ml goat anti–rabbit antibody conjugated to AlexaFluor488 (Invitrogen) diluted in 10% NGS/PBS containing 2 μM TOTO-3 (Invitrogen). After washing, sections were mounted with VectaShield (Vector Laboratories) and visualized using a confocal microscope (510 META; Carl Zeiss MicroImaging, Inc.). Myonuclear number was determined by counting the nuclei within the dystrophin-stained sarcolemmal of at least 100 myofibers per animal using ImageJ software (version 1.36b) and is expressed as the number of myonuclei per myofiber. For fiber type analysis, sections of frozen soleus muscle were prepared as described in the previous section. Sections were fixed with 4% PFA in PBS for 10 min at room temperature and immunostained with anti–type I and anti–type II MHC antibodies for 1 h at room temperature. Primary antibodies were detected using the Vectastain Elite ABC reagent and 3,3-diaminobenzidine tetrachloride (Vector Laboratories) according to the manufacturer's instructions. Sections were photographed using a microscope (Axiovert S100; Carl Zeiss MicroImaging, Inc.) fitted with a digital camera (Spot; Diagnostic Instruments) and analyzed by counting the number of positively staining fibers in >300 fibers from each animal using ImageJ software (version 1.36b). Primary myoblasts were prepared from the fore and hind limb skeletal muscle of 5–8-wk-old SHP-2 (lox/lox) mice. Muscle tissue was isolated, minced, and enzymatically digested in Ham's F-10 medium containing 1.25 mg/ml protease type XIV and 2.5 mg/ml trypsin type IX-S for 2 h at 37°C. The digested tissue was strained through 40 μm of nylon mesh and centrifuged at 1,500 for 3 min at 4°C. The resulting cell pellet was washed and resuspended in Ham's F-10 medium, layered on a Percoll gradient consisting of 40 and 70% Percoll (GE Healthcare) in PBS, and centrifuged at 2,500 for 20 min at 4°C. The purified cells were washed and resuspended in GM consisting of Ham's F-10 with 20% FBS, 5 ng/ml FGF-2, 50 U/ml penicillin, and 50 ng/ml streptomycin. Myoblasts were expanded and enriched as described previously (), and the purity was assessed by MyoD and c-Met staining. To assess the effect of SHP-2 deletion on primary myoblast differentiation, the cells were plated at 10 cells per 35-mm dish and infected with either Ad-GFP or Ad-Cre at an MOI of 50 for 2 h at 37°C in Ham's F-10 containing 2% FBS. Cultures were then switched to DM containing DME supplemented with 0.1% FBS, 1% sodium pyruvate, 5 μg/ml insulin, and 5 μg/ml transferrin. To determine the fusion index, cells were fixed with cold methanol for 5 min at −20°C, washed with PBS, and stained with modified Wright Giemsa (Sigma-Aldrich) stain. The number of nuclei in individual myotubes was counted for 40–100 myotubes. For the NFAT-luciferase assays, primary myoblasts were infected sequentially with Ad-NFAT-luc for 2 h followed by infection with either Ad-GFP or Ad-Cre for an additional 2 h. Luciferase activity was determined using the Luciferase Assay System (Promega) and normalized to the total protein concentration for each sample as determined by Coomassie Protein Assay Reagent (Pierce Chemical Co.). RNA was isolated from hind limbs of 11-wk–old mice using Triazol reagent. IL-4 amplification was performed by incubation at 94°C for 2 min followed by 35 cycles at 94°C for 30 s, 72°C for 45 s, and 1 cycle at 72°C for 10 min using the following primers: sense, 5′-AACCCCCAGCTAGTTGTCATC-3′; and antisense, 5′-CATCGAAAAGCCCGAAAGAGTC-3′. GAPDH was used as a control for each sample and was amplified at 60°C using the following primers: sense, 5′-GGGTGGAGCCAAACGGGTC-3′; and antisense, 5′-GGAGTTGCTGTTGAAGTCGCA-3′. NFAT activity was measured using the NFAT-GL3 plasmid (). C2C12 myoblasts were plated at a density of 4–8 × 10 cells per well of a 12-well plate and were cotransfected with 0.1 μg NFAT-luc and 4 ng pRL-Renilla (Promega) along with 1–2 μg of the indicated expression plasmids using LipofectAMINE 2000 (Invitrogen). Cells were initiated to undergo differentiation for 48 h. Cells were harvested, and luciferase activities were measured using the Dual Luciferase Assay System Kit (Promega). Tissues were homogenized in radioimmunoprecipitation assay buffer (1% Triton X-100, 0.1% SDS, 0.5% sodium deoxycholate, 150 mM NaCl, 20 mM Hepes, pH 7.4, 10% glycerol, 10 mM NaF, 2 mM NaVO, 2 mM EDTA, 5 μg/ml leupeptin, 1 mM benzamidine, 5 μg/ml aprotinin, 1 μg/ml pepstatin A, 1 mM PMSF, and 1 mM DTT). Primary and C2C12 myoblasts were lysed in NP-40 lysis buffer (1% NP-40, 150 mM NaCl, 50 mM Tris-HCl, pH 7.4, 10 mM NaF, 1 mM NaVO, 1 μg/ml leupeptin, 1 mM benzamidine, 1 μg/ml aprotinin, 1 μg/ml pepstatin A, and 1 mM PMSF). Lysates were subjected to centrifugation at 20,800 for 15 min at 4°C. Protein concentrations were determined using the BCA Protein Assay Kit (Pierce Chemical Co.) before immunoblotting. For immunoprecipitations, supernatants were precleared for 15 min at 4°C with pansorbin (Calbiochem) and were incubated overnight at 4°C with primary antibodies. Immunoprecipitates were recovered by using either protein A– or protein G–Sepharose, washed five times with lysis buffer, and resuspended in sample buffer. Immunoblotting was performed as described previously (). Densitometric analysis on immunoblots was performed by using the LabWorks 4.0 Image Analysis software (UVP BioImaging Systems). C2C12 myoblasts were detached using trypsin/EDTA, washed three times in serum-free DME, held in suspension for 30 min at 37°C, and either maintained in suspension or plated on tissue culture plates coated with 5 μg/ml fibronectin at 37°C for the indicated times. To analyze the role of integrins in muscle-specific gene expression, C2C12 myoblasts were detached using trypsin/EDTA, washed three times in serum-free medium, plated onto petri dishes coated with either 5 μg/ml fibronectin or 5 μg/ml poly--lysine, and cultured in serum-free DME for 48 and 72 h. For experiments involving SHP-2 adenoviral infections, C2C12 myoblasts were infected for 2 h, switched to fresh GM, and incubated for 24 h before plating on fibronectin-coated petri dishes. C2C12 myotubes were visualized using AlexaFluor545-conjugated phalloidin (Invitrogen) as previously described () except that nuclei were stained with TOTO-3 and coverslips were mounted in Vectashield before imaging by confocal microscopy as described previously. C2C12 myoblasts maintained in GM were plated at a density of 1.25 × 10 cells per 35-mm plate. Cells were induced to differentiate by transfer to DM for 24 h and were treated with 2 μM PP2 or DMSO in DM for an additional 48 h. Cells were fixed in 2.5% glutaraldehyde in PBS for 10 min at room temperature, mounted in Vectashield with DAPI, and visualized by glutaraldehyde-induced autofluorescence. Fusion index was calculated as the percentage of cells that had two or more than two myonuclei. The number of nuclei within individual myotubes was counted for 70–150 myotubes. Data were assessed for statistical significance by either the application of a two-way test assuming unequal variances or by analysis of variance.

In skeletal muscle, satellite cells have long been described as “reserve” or “stem” cells. They are located between the basal lamina and the sarcolemma of myofibers and are able to self-renew and differentiate into mature muscle (; ; ). Satellite cells are the most efficient cell type in skeletal muscle repair after acute injury (; ; ; ; ; ; ). However, contribution of other stem-like cells to muscle regeneration has been reported (). One of these additional cell populations is the so-called side population (SP), which has been isolated from skeletal muscle using the FACS based on its greater ability to efflux the fluorescent dye Hoechst 33342, when compared with the main population (MP; ; ; ; ). In vivo mouse muscle SP cells can fuse to dystrophic myofibers after systemic delivery () and can give rise to -positive myogenic cells after intramuscular injection in acutely injured muscle (). In vitro, muscle SP cells can adapt myogenic specification after coculture with C2C12 () or are able to express the myogenic marker after intravenous injections into mice (). Despite studies supporting the ability of SP cells to give rise to differentiated progeny in vitro and in vivo, the molecular pathways that define their phenotype remain unclear. We hypothesized that the specific molecular networks responsible for the phenotype of SP cells could be identified by global gene expression analysis. During embryonic development, a morphogen gradient of bone morphogenetic protein 4 (BMP4) and its antagonists plays important roles in mesoderm induction, establishment of dorso-ventral polarity, ectodermal differentiation, somite formation, and myogenesis induction (; ; ; ; ). In the paraxial mesoderm, local variation of BMP4 concentrations created by interaction between BMP4 and its antagonists are known to differentially affect the induction of and (). In developing somites, BMP4 is expressed by ventral cells that give rise to the sclerotome, whereas BMP4 antagonists, such as Noggin, Chordin, Gremlin, and Follistatin are expressed in the dorsal part, which gives rise to the dermomyotome. A depletion of the BMP antagonists Noggin, Chordin, and Follistatin leads to a catastrophic loss of dorsal structures in (). In the present study, microarray analysis revealed that BMP4, a known repressor of myogenic differentiation (; ), is highly expressed in muscle SP cells, whereas its antagonist, Gremlin (; ; ; ), is up-regulated in MP cells. Functional studies demonstrate that BMP4 expressed by muscle SP cells induces proliferation of BMP receptor 1a (BMPR1a)–positive MP cells, and this effect can be reversed by Gremlin. Detection of BMP4 and BMPR1a cells by immunohistochemistry in human fetal skeletal muscle revealed that BMP4 cells are located near BMPR1a cells in the interstitial spaces, supporting the hypothesis that interactions between these cells occur in vivo. Gremlin is expressed by mature myofibers and interstitial muscle cells, which are separate from BMP4-expressing cells. Our results propose a functional role for BMP4 and Gremlin, which are expressed by muscle SP and MP cells, respectively, as regulators of proliferation and differentiation of myogenic progenitors in human fetal skeletal muscle. SP cells have been isolated from multiple tissues (, ), including murine skeletal muscle (; ; ; ). To identify SP cells in human skeletal muscle, dissociated mononuclear cells from discarded muscle samples of individuals aged 14 gestational weeks to 63 yr were stained with the vital DNA dye Hoechst 33342. The dye concentration used for each sample was individually optimized because of observed interindividual variability in sensitivity to Hoechst 33342 dye (Fig. S1, available at ). After the initial optimization, independent SP cell isolations from the same individual demonstrated minimal variability (Fig. S1, C–F). For fetal samples, the optimal Hoechst dye concentration ranged from 3–9 μg/ml, whereas for adults it ranged from 7.5 to 12.5 μg/ml (Fig. S1, G and H). A control sample stained in the presence of reserpine allowed definition of the appropriate SP gate. Because fetal samples contained the highest proportion of muscle SP cells (Fig. S1), all subsequent studies were performed on human fetal muscle SP cells. To identify muscle SP-specific cell surface markers, the expression of two antigens present on stem cells from other tissues, CD34 (; ) and CD133 (; ), was analyzed. As shown in , only 3.6% of CD34-positive cells and 0.56% CD133-positive cells were detected within the SP gate. The high dye efflux ability of SP cells has been previously attributed to the function of the ABCG2 transporter, which is expressed in multiple tissues, including skeletal muscle (; ; ). In our experiments, the ABCG2 transporter was detected in 2% of the cells in human fetal muscle (), with 76% of these cells present within the MP gate and only 0.25% in the muscle SP gate (). In control samples of mouse 3T3 cells transfected with either a mock vector () or with a vector encoding the human ABCG2 cDNA (), 0.37 and 60.7% of ABCG2-positive cells were detected, respectively. Thus, our studies indicate that human fetal skeletal muscle–derived SP cells are largely ABCG2-negative. Previous studies reported that a variable percentage of murine skeletal muscle SP cells are of hematopoietic origin and express the pan-hematopoietic marker CD45 (). We found that only 0.44% of human fetal skeletal muscle SP cells expressed CD45, suggesting that >99% of these cells are not of hematopoietic origin (Fig. S2, available at ). To characterize the repertoire of expressed genes and identify a developmental hierarchy among mononuclear cells within human muscle, microarray analyses were performed on sorted, noncultured human fetal skeletal muscle SP ( = 10) and MP ( = 9) cells. A detailed description of the samples used in the microarray studies is provided in Table S1 (available at ). The percentage of SP cells in fetal muscle samples derived from 15 different individuals ranged between 0 and 6.57%, with no significant correlation to gestational age or gender. A geometric fold change analysis of the microarray data () identified 222 unique genes/ESTs as differentially regulated between human fetal skeletal muscle SP and MP cells (Table S2, available at ). Of these genes, 162 known genes and 4 ESTs were significantly overexpressed in SP cells compared with MP cells and were called SP genes, whereas 60 genes were significantly underexpressed in the SP population and were called MP genes (Table S2). Gene ontology analysis of the differentially expressed genes using DAVID () indicated that SP cells express high levels of transcriptional repressors and negative cell-cycle regulators, whereas MP cells are enriched for genes involved in metabolism, DNA replication, and cell surface membrane proteins (). Two members of TGFβ signaling pathway, BMP4 and ID4, were found among the transcriptional repressors up-regulated in human muscle SP cells. Intriguingly, muscle MP cells expressed significantly higher levels of Gremlin, a known antagonist of BMP4 signaling (; ; ). The presence of high levels of BMP4 in SP cells and its inhibitor, Gremlin, in MP cells led to the hypothesis that these secreted factors could create antagonistic effects in the cellular environment, regulating the proliferation and differentiation of mononuclear cells in human fetal skeletal muscle. To test this hypothesis, we first confirmed the differential expression of BMP4 (overexpressed in SP cells) and Gremlin (overexpressed in MP cells) by quantitative real-time RT-PCR using total RNA extracted from SP and MP cells ( and Table S3, available at ). The results confirmed that SP cells express eightfold higher levels of BMP4 compared with MP cells. Expression of Gremlin mRNA was twofold higher in MP cells compared with SP cells (), whereas other known BMP4 antagonists such as Chordin, Noggin, and Follistatin were not up-regulated in MP compared with SP cells at the mRNA level (not depicted). Next, BMP4 and Gremlin protein expression was assessed by immunohistochemistry on cytospins of SP and MP cells (). All images were acquired at the same exposure time to enable a quantitative analysis of protein expression levels. This analysis demonstrated that SP cells expressed significantly higher levels of BMP4 protein than MP cells, with the mean pixel intensity per unit area for SP cells measured at 506 ± 30 versus 113 ± 8 pixels/μm for MP cells (mean ± SEM; P < 0.0001; ). Gremlin was significantly more abundant in MP cells (455 ± 22 pixels/μm) compared with SP cells (261 ± 17 pixels/μm [mean ± SEM]; P < 0.0001; ), confirming the results obtained by quantitative real-time RT-PCR. Because both RNA and protein studies demonstrated that human fetal skeletal muscle SP cells express high levels of BMP4, whereas MP cells express high levels of Gremlin, it was important to investigate the functional effects of these factors on purified human muscle SP and MP cells. Both BMP4 and Gremlin are secreted morphogens and can act in the immediate cellular environment as regulators of cellular proliferation or differentiation. To study these effects, purified human fetal muscle SP and MP cells were cultured for 14 d under the following four conditions: control (no factors), in the presence of BMP4, in the presence of BMP4 and Gremlin, and in the presence of Gremlin (). The factors were added at the beginning of the culture and exchanged every 48 h. BMP4 was added at 25 ng/ml according to the previously reported data (), and Gremlin was added at 2 μg/ml based on ED inhibitory concentration reported by the manufacturer (R&D Systems). Although MP cells differentiated and gave rise to myotubes, SP cells maintained their undifferentiated phenotype under all four conditions, indicating that BMP4 and/or Gremlin do not induce differentiation of SP cells (). The degree of myogenic differentiation of MP cells was measured by assessing the percentage of myogenin-positive cells (). When assessed by expression of myogenin at day 7 of culture in proliferation medium, control MP cultures had 9.07 ± 2.92% (mean ± SD) of myogenin-positive cells (), whereas in the presence of BMP4, the percentage of myogenin-positive nuclei decreased to 0.1 ± 0.31% (mean ± SD; P < 0.001; ). Addition of Gremlin to BMP4 reversed expression of myogenin in 14 ± 2.93% (mean ± SD) of cells (P < 0.001; ). Addition of Gremlin alone did not show a significant difference with control cultures, as 13.91 ± 3.55% (mean ± SD) of cells were myogenin-positive ( ). Therefore, BMP4 had a strong inhibitory effect on the differentiation of MP cells, whereas addition of Gremlin led to a complete loss of this inhibitory effect. The effect of BMP4 and Gremlin on proliferation of SP and MP cells was studied by [3H]thymidine incorporation assay (). In cultured SP cells, exposure to Gremlin had a significant inhibitory effect on their proliferation (, left) compared with untreated control SP cell cultures (100 ± 62 vs. 300 ± 103 [mean ± SD] cpm; P < 0.05), and this effect was reversed by the addition of BMP4. BMP4 alone, however, had no effect on proliferation of SP cells compared with untreated control cultures (252 ± 79 vs. 300 ± 10 [mean ± SD] cpm; NS). In contrast, addition of BMP4 to MP cultures (, right) resulted in a significant increase in their proliferation compared with untreated controls (5,174 ± 340 vs. 2,875 ± 1,088 [mean ± SD] cpm; P < 0.05). This effect was reversed by the addition of Gremlin (4,336 ± 46 vs. 5,174 ± 340 mean ± SD] cpm; P < 0.05), and Gremlin alone inhibited proliferation of MP cells compared with the untreated control (1,579 ± 178 vs. 2,875 ± 1,088 [mean ± SD] cpm; P < 0.05). To identify the population within muscle MP cells responsive to BMP4-induced proliferation, the expression of the BMP4-specific receptor BMPR1a (; ) was studied by flow cytometry (). 4–11% of human muscle cells expressed BMPR1a (; = 5), with >98% of BMPR1a cells located outside the SP gate (). To study the molecular mechanism of BMP4-induced proliferation, BMPR1a and BMPR1a cells were cultured in growth medium alone or in the presence of BMP4, and [3H]thymidine incorporation assay was performed on culture day 4 (). Results demonstrated that addition of BMP4 to BMPR1a cultures resulted in a significant increase in their proliferation compared with untreated controls (4,620 ± 63 vs. 3,672 ± 124 [mean ± SD] cpm; P < 0.05). Addition of BMP4 to BMPR1a cells had no effect on their proliferation compared with untreated cells. Untreated BMPR1a cells demonstrated significantly higher proliferation than BMPR1a cells (3,672 ± 124 vs. 2,553 ± 161 [mean ± SD] cpm; P < 0.05). These results demonstrate that BMP4 induces proliferation of BMPR1a cells, whereas it has no effect on BMPR1a cells, suggesting that the proliferation induced by BMP4 is mediated via BMPR1a. Further, BMPR1a cells have a higher intrinsic proliferative capacity than BMPR1a cells. The activation of myogenic cells is a highly regulated process that includes down-regulation of the satellite cell–specific marker (; ) and the up-regulation of the myogenic-specific factors and (; ; ; ). Muscle cells obtained from −/− mice show severe proliferation deficiency and undergo premature differentiation (). Thus, , which is normally expressed by satellite cells and activated myoblasts, is important for muscle cell proliferation (; ; ). Expression of , , and was assayed in BMPR1a and BMPR1a cells by real-time quantitative RT-PCR () and by immunofluorescence on cytospins (). The baseline expression of was twofold higher in BMPR1a cells than in BMPR1a cells, whereas and were expressed at higher level in BMPR1a cells (). and protein expression assessed by cytospins demonstrated that significantly more BMPR1a than BMPR1a cells expressed and proteins (). By total cell count, 1.04% of BMPR1a cells ( = 192) and 12.9% of BMPR1a cells ( = 743) expressed . None of BMPR1a cells ( = 205) and 7.4% of BMPR1a cells ( = 594) expressed . , , and cell population, and may provide an explanation for their increased proliferation activity. Our studies using synthetic morphogenic factors showed that addition of BMP4 induced proliferation and delayed differentiation of muscle MP and BMPR1a cells. To investigate whether BMP4-expressing SP cells also inhibit differentiation of MP cells, freshly purified SP cells were irradiated at 25Gy () and cocultured with an equal number of MP cells. Myogenic differentiation was assessed by immunostaining for myogenin or myosin heavy chain (MHC) at culture day 7 (). In control MP cultures (MP cells alone), 22.9 ± 4.4% (mean ± SD) myogenin-positive cells and 22.4 ± 6.9% MHC-positive cells were detected, whereas in cocultures of MP cells with irradiated SP cells, no myogenin or MHC expression was observed (). These results demonstrate that BMP4-expressing SP cells inhibited differentiation of muscle MP cells. To investigate whether SP cells exert a similar proliferative effect as recombinant BMP4 on BMPR1a cells and whether this effect can be abrogated by specific BMP4 blockade, BMPR1a and BMPR1a cells were cultured for 5 d under the following conditions: no treatment, cocultured with irradiated SP cells, cocultured with irradiated SP cells and 2 μg/ml BMP4 blocking antibody, and cocultured with irradiated SP cells and 2 μg/ml isotype control antibody (). Proliferation was assessed using a [3H]thymidine incorporation assay, and proliferation indexes were established by calculating the ratios of thymidine incorporation in treated versus untreated cells. BMPR1a cells cocultured with irradiated SP cells proliferated significantly more than control BMPR1a cells (proliferation index: 1.89 ± 0.27 [mean ± SEM]; P < 0.05). This effect was significantly reversed when 2 μg/ml of BMP4 blocking antibody were added (proliferation index: 1.03 ± 0.05 [mean ± SEM]; P < 0.05). IgG2b isotype control antibody had no significant inhibitory effect on the proliferation of BMPR1a cells cocultured with SP cells (proliferation index: 1.47 ± 0.4 [mean ± SEM]; NS). No significant effects were observed on proliferation of BMPR1a cells compared with untreated controls by coculture with irradiated SP cells (proliferation index: 1.22 ± 0.28 [mean ± SEM]; NS), coculture with irradiated SP cells in the presence of BMP4 blocking antibody (proliferation index: 0.843 ± 0.10 [mean ± SEM]; NS), or coculture with irradiated SP cells in the presence of IgG2b isotype control antibody (proliferation index: 0.802 ± 0.08 [mean ± SEM]; NS). These results demonstrate that SP-induced proliferation of BMPR1a cells is mediated primarily by BMP4 signaling, as it is abrogated by specific BMP4 blockade. As BMP4 expression was found significantly elevated in the majority of human muscle SP cells compared with MP cells, immunostaining of human fetal skeletal muscle tissue sections obtained from six individuals was performed using an anti-BMP4 antibody in an attempt to localize muscle SP cells in vivo. Dystrophin staining was performed simultaneously to detect the sarcolemma of myofibers (). Rare interstitial BMP4 cells were observed in muscle tissue sections (). Small clusters of BMP4 cells were also found in areas distinct from myofibers (). In eight random microscopic fields obtained from different samples, the frequency of BMP4 cells was estimated to be 1 for every 58 nuclei, corresponding to 1.8% of the cells. To define the spatial relationships between BMP4- and BMPR1a-expressing cells, immunohistochemistry was performed on sequential sections of a 20-wk gestation fetal skeletal muscle sample (). BMP4 cells () were found in the vicinity of BMPR1a cells () in the interstitial spaces surrounded by myofibers (). BMP4-reactive cells are rounded in contours and have smaller cross-sectional diameters than BMP4 skeletal muscle fibers. They focally rim the interstitial spaces formed by mature muscle fibers. Occasional cells within the interstitial spaces show variable membrane reactivity for BMPR1a, where they can be also found as clusters surrounded by BMP4 cells. Furthermore, staining for BMP4 and Gremlin () was mutually exclusive. Gremlin was highly expressed in mature myofibers () and scattered rounded cells in the interstitium (, arrows), whereas staining for BMP4 was present only in rounded cells that are most prominent in the interstitium (, arrows). Double labeling for BMP4 and Gremlin () confirmed the reciprocal staining pattern of interstitial BMP4- or Gremlin-expressing cells. Skeletal muscle SP cells represent a primitive stem cell population capable of myogenic differentiation in vitro and in vivo (; ; ). Our data show that human fetal skeletal muscle SP cells express high levels of BMP4, whereas MP cells express high levels of the BMP4 antagonist Gremlin. As a member of the TGFβ signaling family, BMP4 induces a variety of cellular responses, initiated by binding to its specific cellular receptor, BMPR1a (; ,). We found by flow cytometry that >98% of BMPR1a cells are located outside of the SP gate. Our data demonstrate that BMP4 specifically stimulates proliferation of BMPR1a cells, whereas it has no effect on BMPR1a cells, suggesting that BMP4-induced proliferation is likely mediated by the BMPR1a receptor. BMP4 has been shown to inhibit transcription of (), a muscle-specific transcription factor associated with withdrawal from the cell cycle and terminal differentiation. The myogenic transcription factor is thought to be developmentally “upstream” of (; , ), and it is expressed in actively proliferating myogenic cells (; ; ). Our studies demonstrated a twofold increase in expression of in uncultured, untreated BMPR1a cells compared with BMPR1a cells, suggesting that these cells may have intrinsic high proliferative potential. Coculture of irradiated, BMP4-secreting muscle SP cells with BMPR1a cells significantly induced proliferation of the latter compared with control BMPR1a cells cultured alone, and this effect was specifically abrogated by BMP4 blockade. In contrast, BMPR1a cells did not increase proliferation when cultured in the presence of irradiated SP cells. These results suggest that muscle SP cells are able to induce proliferation of other myogenic precursors and that this effect is mediated via BMP4 signaling. In support of this hypothesis, BMP4 and BMPR1a cells were found in vivo in close proximity of each other, with BMP4 cells surrounding small clusters of BMPR1a cells, located in the interstitial spaces between myofibers. From the current study, muscle MP cells express high levels of the BMP4 inhibitor Gremlin (), which is known to bind BMP4, thus reducing BMP4 concentration in the cellular microenvironment. In vivo data indicate that Gremlin is highly expressed by mature myofibers and interstitial cells, which are distinct from BMP4-expressing cells. The observed Gremlin-induced growth inhibition of SP and MP cultures may play an important role in counteracting the stimulatory effects of BMP4 and in regulating the proliferation of BMPR1a MP cells. In addition to its function as a BMP4 antagonist, Gremlin has been shown to suppress tumor growth by mechanisms that are independent from BMP4, which involve both up-regulation of p21Cip1 and down-regulation of p42/44 MAPK (). Remarkably, in our studies, only Gremlin is differentially expressed between SP and MP cells, whereas other BMP4 antagonists, such as Chordin, Noggin, and Follistatin, are not up-regulated in MP cells. We therefore hypothesize that secretion of BMP4 by SP cells and of Gremlin by MP cells generate antagonistic effects, influencing the rate of proliferation and differentiation of myogenic cells in human muscle. High concentrations of BMP4 have been shown to induce osteogenic differentiation of muscle progenitor cells (; ) and are counteracted by BMP4 inhibitors under normal conditions (). Overexpression of BMP4 and the inability to mount an appropriate antagonist (including Gremlin) response have been implicated in the pathogenesis of a human disease of heterotopic osteogenesis, fibrodysplasia ossificans progressiva (FOP; ). The process of ectopic ossification in this disease is often precipitated by muscle injury, i.e., biopsy or trauma, when progenitor cells are being activated for muscle repair. It is conceivable that the BMP4 SP cells participate in muscle repair after acute injury in response to trauma, creating an unopposed local increase in BMP4, which may contribute to the formation of islands of heterotopically ossified muscle tissue in patients afflicted by FOP. The data presented here suggest that SP cells regulate proliferation of more committed cells in a paracrine fashion by secreting BMP4, which is antagonized by its inhibitor Gremlin, secreted by MP cells. The functional counteraction of BMP4 signaling by Gremlin may play a role in preventing uncontrolled proliferation, thus maintaining the appropriate number of progenitors within the tissue. Based on our in vitro findings, we propose that expression and secretion of BMP4 by quiescent muscle SP cells () induces proliferation of BMPR1a Myf5 muscle MP cells (). This proliferation is counteracted by secretion of Gremlin by BMPR1a Myf5 committed MP cells (). The demonstrated tissue localization of BMP4 and BMPR1a cells in close proximity of each other and the mutually exclusive staining pattern of BMP4 and Gremlin suggest that these interactions also occur in vivo. This study provides additional insights into the complex functional hierarchy among the mononuclear cells in skeletal muscle and leads to important correlations with other stem cell niches, such as the skin, bone marrow, and intestine, where stem cells use BMP4 signaling to regulate the proliferation of their daughter cells, which in turn differentiate into committed cells and tissues (; ; ). Mononuclear cells were isolated from discarded human fetal skeletal muscle of gestational age 14–18 wk (Advanced Bioscience Resources). Tissues were collected in sterile HBSS with 1% glucose and 1% penicillin/streptomycin, finely minced using sterile scalpels, and digested using dispase (0.6 U/ml) and collagenase (0.5 mg/ml; Worthington) diluted in PBS. Digested samples were filtered through a 40-μm cell strainer before staining and analysis (). Mononuclear cells were centrifuged at 365 for 10 min and resuspended at a concentration of 10 cells/ml in prewarmed (37°C) PBS/0.5% BSA. Hoechst 33342 dye (Sigma-Aldrich) was added at a final concentration of 3–9 μg/ml for fetal samples and 7.5–12.5 μg/ml for adult samples. In parallel, as a negative control for SP cell gating, 5 × 10 muscle cells were stained with Hoechst 33342 in the presence of 5 μM reserpine (Sigma-Aldrich). Cells were incubated for 60 min at 37°C, protected from light in a waterbath, and washed by adding 3.5–5 volumes of ice-cold PBS/0.5% BSA. For labeling with specific antibodies, after Hoechst 33342 staining, samples were pelleted, resuspended at a concentration of 10–10 cells/ml, and incubated with the following antibodies: anti-human CD34-FITC (BD Biosciences), anti-human CD133-PE (Miltenyi Biotec), anti-CD45-PE (Miltenyi Biotec), rat IgG-PE (BD Biosciences), anti-ABCG2 (Stem Cell Technologies), or anti-BMPR1a (Santa Cruz Biotechnology, Inc.). Control samples of 3T3 cells transfected with either mock- or human ABCG2–encoding vectors were provided by B. Sorrentino (St. Jude Children's Research Hospital, Memphis, TN). Cells were incubated for 15 min on ice, washed, and resuspended in ice-cold PBS/0.5% BSA containing 2 μg/ml propidium iodide (PI). For detection of ABCG2 and BMPR1a cells, incubation with FITC-conjugated secondary anti-mouse (BD Biosciences) or anti-rabbit (Jackson ImmunoResearch Laboratories) antibody was performed for 10 min on ice. Subsequently, the cells were washed with cold PBS/0.5% BSA and resuspended in ice-cold 1× PBS/0.5% BSA containing 2 μg/ml PI. Flow cytometric analyses and cell sorting were performed on a dual-laser FACSVantage SE flow cytometer (Becton Dickinson) equipped with two lasers: one with 200 MW power (488 nm) and one with 150 MW power (UV). Both the Hoechst and PI dyes were excited at 350 nm, and their fluorescence was measured at 400 and 600 long pass, respectively, with a 550 short pass dichroic mirror. For each sample analyzed, 20,000 total cell counts were acquired using CellQuest software version 3.3 (Becton Dickinson). Total RNA was extracted from sorted human fetal muscle SP and MP cells using the RNeasy kit (QIAGEN). Total RNA (100 ng) from 10 SP and 9 MP samples were subjected to two rounds of linear amplification using a MessageAmp kit (Ambion). After amplification, 20 μg of fragmented, biotin-labeled cRNA was prepared and hybridized onto the Affymetrix human HG-U133A GeneChip (as per the manufacturer's instructions). Microarrays were optically scanned, and images were processed using the Affymetrix GeneChip MAS5.0 analysis software, which were then summarized in a CHP file. Each surveyed gene transcript was given a mean fluorescence intensity value (signal) indicating the amount of the transcript detected and a call (present, absent, or marginal) indicating the robustness of its detection. Each Affymetrix HG-U133A microarray contains 22,283 individual probes for gene/RNA transcripts. Using a geometric logarithmic fold analysis (; ), 244 probes had a statistically significant differential reading between the SP ( = 10) and MP ( = 9) groups. Two criteria had to be satisfied for significant differential expression: AbsoluteValue(AvgLF) − StdLF > 0.4 × Local_Fold_Noise and AbsoluteValue(AvgLF) > max(Local_Fold_Noise, Global_Fold_Noise). AvgLF and StdLF denote the mean and standard deviation of the logarithmic fold change of a probe in the SP relative to the MP group, respectively. For each probe, Local_Fold_Noise is the largest logarithmic fold change for the probe within the MP or the SP groups. Global_Fold_Noise is the mean of Local_Fold_Noise over all 22,283 probes. The false discovery rate (FDR) for this approach was <1% after a permutation analysis with 10,000 independent shuffling iterations. In each iteration, the 19 sample labels (MP or SP) of each of the 22,283 probes were first randomly shuffled. Then, the number of probes from this shuffled dataset that came up as significant (i.e., false positives) using the two criteria above were counted. The FDR is the median of these 10,000 false positive counts over 244, the number of probes called significantly differentially expressed in the original unshuffled dataset. These 244 probes (corresponding to 222 unique genes) were functionally annotated using the web-based Database for Annotation, Visualization, and Integrated Discovery 2.0 (DAVID 2.0; ; ). Total RNA (100 ng) isolated from freshly sorted human fetal skeletal muscle SP and MP cells or from BMPR1a and BMPR1a cells was reverse transcribed using the Taqman reverse transcription kit (Applied Biosystems) in a 100 μl reaction volume according to the manufacturer's protocol. Approximately 20 pcg of cDNA (1 μl of RT reaction product) was used for PCR amplification. Samples were assayed using Sybergreen chemistry and kinetic PCR (ABI 7700 Sequence Detector; Applied Biosystems). Samples were amplified using the Sybergreen PCR reagent kit (Applied Biosystems) according to the manufacturer's protocol. Sense and antisense primers were used at a final concentration of 300 nM. The primer sequences for the genes tested are listed in Table S3. The cDNA samples were amplified under “default” conditions: 50°C for 2 min and 95°C for 10 min, followed by 40 cycles of amplification at 94°C for 15 s and 60°C for 1 min. For data analysis, the cycle threshold (Ct) number was computed for each sample using the Sequence Detection Software (Applied Biosystems). The Ct baseline was preset at 10 standard deviations above the fluorescence background, which was determined in the first 3–15 cycles. All quantitations were normalized to the endogenous control gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH) to account for variability in the initial concentration and quality of the total RNA, and the efficiency of the reverse transcription reaction. Differential Ct values (Ct for the gene of interest minus Ct for GAPDH) were calculated for each amplified gene. Logarithmic fold change of gene expression between samples was calculated by subtracting gene-specific differential Ct values. For cytospin preparation, human fetal skeletal muscle SP, MP, BMPR1a, and BMPR1a cells were purified by flow cytometry. Cells (1,000 cells/slide) were placed on glass slides by centrifugation in a cytospin centrifuge (Thermo Electron Corp.). Slides were fixed in absolute methanol for 3 min at room temperature, blocked for 30 min in blocking solution (10% fetal bovine serum and 0.1% Triton X-100), and incubated with mouse anti- monoclonal antibody (diluted 1:15; Developmental Studies Hybridoma Bank) or mouse anti- monoclonal antibody (diluted at 1:50; BD Biosciences) overnight at 4°C. Slides were washed in 1× PBS/0.1% Triton X-100 and incubated with goat anti–mouse or goat anti–rabbit FITC-conjugated secondary antibody (diluted 1:100; Jackson ImmunoResearch Laboratories) for 1 h at room temperature. For BMP4 and Gremlin staining, cytospin slides were fixed in 4% paraformaldehyde for 20 min at room temperature, blocked for 30 min in 10% fetal bovine serum and 0.1% Triton X-100, and incubated with mouse anti-BMP4 monoclonal antibody (diluted 1: 20; Novocastra) or rabbit anti-Gremlin polyclonal antibody (diluted 1:50; Abgent) overnight at 4°C. Slides were washed in 1× PBS/0.1% Triton X-100 and incubated with goat anti–mouse Alexa Fluor 488 or goat anti–rabbit Alexa Fluor 488 secondary antibody (diluted 1:200; Invitrogen) for 1 h at room temperature and washed in 1× PBS/0.1% Triton X-100. Slides were then mounted in Vectashield (Vector Laboratories) supplemented with 100 ng/ml DAPI to visualize nuclei. Images from - and -stained BMPR1a and BMPR1a cells were acquired in five random microscopic fields for each sample using 200 and 400 magnification objective on a microscope (E1000; Nikon), photographed using a camera (Orca ER; Hamamatsu), and processed using OpenLab software version 3.1.5 (Improvision). For quantitative analysis, images from BMP4- and Gremlin-stained SP and MP cells were acquired in 10 random microscopic fields for each sample on an automated upright fluorescent microscope (Eclipse 90I; Nikon) using 200 magnification objective and processed using MetaMorph 7.0 software (Molecular Devices). Images were photographed using a camera (Orca II; Hamamatsu). The intensity of fluorescence labeling between the SP and MP cells was compared using the mean integrated intensity per unit area, to correct for differences in cell size. Using MetaMorph software, a region of interest was drawn around each cell to be measured and the integrated intensity of the region was calculated. To correct for local changes in background intensity, the region was then moved to a background area of the field of view close to the cell of interest and integrated intensity was recorded. The integrated intensity of the background was subtracted from the integrated intensity of the cell being measured. Integrated pixel intensity was calculated for each cell in the field and then divided by the area of the cell to calculate mean pixel intensity per unit area. Mean pixel intensities for SP and MP cells were then statistically compared using the nonparametric Wilcoxon test. For immunofluorescence studies on human fetal skeletal muscle, 10-μm tissue sections were collected on sialanazed slides (Polysciences, Inc.), fixed in absolute methanol, and blocked as described for cytospins before incubation with mouse anti-BMP4 (diluted 1:20; Novocastra) and anti-dystrophin CAP6-10 polyclonal antibody diluted at 1:500 in PBS (; ). Slides were subsequently processed as described for cytospins. Sorted human fetal muscle BMPR1a and BMPR1a cells, as well as SP and MP cells, were seeded in 96-well plates at 3 × 10 cells/well in triplicates and incubated in growth medium consisting of DME with 4% glucose, 20% fetal bovine serum (vol/vol), 1% (vol/vol) penicillin-streptomycin (10,000 UI/ml–10,000 μg/ml; Invitrogen) for 4–14 d. For BMP4 and Gremlin stimulation, 25 ng/ml human recombinant BMP4 (R&D Systems) or 2 μg/ml mouse recombinant Gremlin (R&D Systems) were exchanged every 48 h. Control wells contained sorted SP and MP cells maintained in growth medium without the addition of BMP4 or Gremlin. For coculture experiments, muscle SP cells were irradiated at 25 Gy and seeded together with MP, BMPR1a, or BMPR1a cells at 1:1 ratios. Blocking anti-BMP4 antibody (R&D Systems) and isotype control IgG2b antibody (R&D Systems) were added to cocultured SP with BMPR1a or BMPR1a cells at 2 μg/ml every 48 h of culture. Cell proliferation was assessed for SP/MP, BMPR1a, and BMPR1a cells after a culture period of 4–7 d by [3H]thymidine incorporation (1 μCi/well) for the last 18 h of culture (). Cells were harvested using an automated cell harvester, and incorporated radioactivity was assessed by a BetaMax counter (Beckman Coulter). The results of proliferation assays were compared using the Kruskal-Wallis test for comparing populations with unknown distributions (results are shown as p-values). Differences with P < 0.05 were considered statistically significant. Fig. S1 illustrates Hoechst/PI profiles of human skeletal muscle cells obtained from different individuals. Fig. S2 illustrates expression of CD45 by human fetal skeletal muscle SP cells. Table S1 describes SP and MP samples subjected to microarray studies. Table S2 lists 222 genes differentially expressed between human fetal skeletal muscle SP and MP cells. Table S3 provides sequences for primers used in real-time PCR reactions. Online supplemental material is available at .

TGFβs are cytokines with manifold key roles in modulating cell proliferation and differentiation, apoptosis, immune responses, tissue repair, and ECM formation (; ; ; ). TGFβ1, the first of 3 isoforms (TGFβ1–3) to be identified, is the prototype of a superfamily of cytokines, related in domain structure and sequence homology, that is divisible into subfamilies based on degree of sequence homology (). The TGFβ1–3 subfamily is closely related to the activin subfamily and to growth and differentiation factors (GDFs) 8 (aka myostatin) and 11 (). All superfamily members act by binding cell surface type I and II receptor heterotetramers, with signal transduction to the nucleus via Smad-dependent or -independent pathways (). Signaling by TGFβ-like ligands is tightly controlled by various intracellular, cell surface, and extracellular inhibitory proteins (; ). Signaling by TGFβ1–3 and GDF8 and -11 is also blocked by formation of a noncovalent latent complex between the functional ligand and its cleaved prodomain (; ; ; ), designated latency-associated peptide (LAP) for TGFβ1–3 (). TGFβ1–3 are produced as large latent complexes (LLCs) in which LAP is disulfide bonded to a latent TGFβ-binding protein (LTBP), of which mammals have four (). It remains unclear which processes activate TGFβ latent complexes in vivo, although cleavage within LAP sequences by plasmin and/or matrix metalloproteinases (MMPs; ; ; ; ; ; ) and/or nonproteolytic dissociation of LAP from active TGFβ, caused by interactions with thrombospondin () or integrin αβ (), may be involved. Although overall in vivo roles of LTBPs remain obscure, they serve to bind LLCs to ECM, perhaps via covalent linkage of LTBPs to ECM components (). Ability of LTBP1 to fix LLC to ECM appears involved in its ability to promote αβ-mediated TGFβ activation, whereas the apparent ability of LTBPs to facilitate TGFβ activity in other situations is by as-yet-unclear mechanisms (; ; ). Bone morphogenetic protein 1 (BMP1)–like proteinases affect morphogenesis by biosynthetic processing of precursors to mature functional proteins necessary to ECM formation. Such proteins include collagens I–III (), V, VII, and XI; proteoglycans biglycan and osteoglycin; basement component laminin 5; and the cross-linking enzyme lysyl oxidase (). BMP1-like proteinases also affect morphogenesis by activating BMP2 and -4 via cleaving the extracellular antagonist Chordin and thus regulating patterning (; ; ). In addition, BMP1-like proteinases activate GDF8 and -11 by cleaving within noncovalently bound prodomains to release mature GDF8 and -11 from latent complexes (; ), thus affecting negative growth control of skeletal muscle and neural tissues, for which GDF8 and -11, respectively, are responsible (; ). Here, we use in vitro and in vivo approaches to demonstrate that BMP1-like proteinases serve to activate TGFβ by direct cleavage of LTBP1, resulting in liberation of LLCs from ECM and consequent MMP-dependent LAP cleavage. Because BMP1 is potently induced by TGFβ1 (), its role in TGFβ1 activation completes a novel amplification loop in vertebrate tissue remodeling. Moreover, BMP1-like proteinases are identified as regulators capable of orchestrating TGFβ signaling with ECM deposition, patterning, and negative feedback control of muscle and neural tissue growth. To determine whether BMP1 might play a role in proteolytic activation of TGFβ1, BMP1 was incubated with TGFβ1 LLC. A Western blot probed with antibody to a C-terminal His tag demonstrated dose-dependent cleavage of an ∼17-kD fragment from the C-terminal portion of ∼190-kD LTBP1 (). A similar blot probed with anti-LTBP1 () demonstrated a corresponding reduction in size of the remainder of LTBP1 to produce a doublet or, at higher BMP1 concentrations, a single band of ∼150 kD. N-terminal amino acid sequencing of the 150- and 17-kD LTBP1 fragments demonstrated that cleavages had occurred at peptide bonds L-D and Q-D, respectively. The former of these sites occurs between the most N-terminal 8-Cys motif and the second most N-terminal non-Ca binding EGF-like domain, within the flexible hinge region of LTBP1 (). The latter cleavage occurs between the most C-terminal 8-Cys motif and the most C-terminal non-Ca binding EGF-like domain. Both sites have P1′ Asp residues and other features characteristic of cleavage sites of previously identified BMP1 substrates (). Control experiments showed both sites to be efficiently cleaved, with cleavage at the N- and C-terminal sites occurring within 10 and 30 min of incubation with BMP1, respectively (Fig. S1, available at ). Control experiments also demonstrated that BMP1 must be catalytically active for LTBP1 cleavage to occur, as even prolonged incubation with catalytically inactive BMP1, in which the protease domain active site Glu had been substituted for with Ala, did not alter LTBP1 electrophoretic mobility (Fig. S1). In contrast to its ability to cleave LTBP1, active BMP1 did not cleave ∼37-kD LAP to a smaller size, nor did it cleave within the ∼50-kD precursor molecule consisting of LAP covalently bound to the mature portion of TGFβ1 (). Under nonreducing conditions, an immunoblot analysis with anti-LAP antibodies showed ∼290-kD intact LLC to be cleaved to an ∼245-kD form by BMP1 (). Interestingly, mobility corresponding to 245 kD, rather than the 37 kD expected for dissociated LAP, indicates that LAP remains associated with LTBP1 subsequent to cleavage of the latter by BMP1 (). To determine whether BMP1-like proteinases are used for in vivo LTBP1 cleavage, we compared processing of LTBP1 by wild-type mouse embryo fibroblasts (MEFs) to processing of LTBP1 by MEFs derived from embryos doubly homozygous null for the gene, which encodes alternatively spliced mRNAs for proteinases BMP1 and mammalian tolloid (), and the gene, which encodes the BMP1-like proteinase mammalian tolloid–like 1 (). Simultaneous ablation of the functionally redundant products of and has previously been shown to remove detectable processing activity for various substrates of BMP1-like proteinases (). Cells produce short and long forms of LTBP1 (LTBP1S and -L, respectively), resulting from alternative splicing and multiple transcriptional promoters (), and both LTBP1S and -L were found to differ in electrophoretic mobility in wild-type and doubly null MEF media samples (). Moreover, sizes of LTBP1S in wild-type (150 kD) and -null (190 kD) samples corresponded, respectively, to sizes of BMP1-cleaved and uncleaved recombinant LTPB1S (). It has been reported that removal of sequences N-terminal to the hinge region is sufficient to release LTBP1 from transglutaminase-dependent covalent association with ECM (; ). Because such sequences are N-terminal to the L-D BMP1 cleavage site demonstrated here, the decreased amounts of LTBP1S and -1L in -null, compared with wild-type MEF medium (), is consistent with the probability that cells use BMP1-like proteinases to process the LTBP1 hinge region and that this is sufficient to release TGFβ1 LLC from ECM. Buttressing this probability was the observation that markedly greater amounts of LTBP1 fragments are released by plasmin from -null than from wild-type MEF ECM (). Surprisingly, although BMP1 does not cleave LAP in vitro (), the electrophoretic mobility of LAP was increased in wild-type compared with -null MEF media (). Because the mobility of LAP from -null MEF media is equal to that of full-length LAP, these results are consistent with the probability that LAP is cleaved in wild-type, but not -null, MEF cultures. To determine the type of proteinases responsible for LAP cleavage, wild-type MEFs were cultured in the presence of inhibitors of metallo- (TNF-α processing inhibitor 2 [TAPI-2]), cysteine- (E64), aspartic- (pepstatin A), and serine- (4-[2-aminoethyl]-benzenesulfonyl fluoride [AEBSF]) proteinases. Only TAPI-2, which inhibits metalloproteinase sheddases and a range of MMPs, inhibited LAP processing in MEF cultures, whereas none of the inhibitors interfered with LTBP1S or -L processing (). Importantly, BMP1 is not inhibited by even 200 μM TAPI-2 in vitro (unpublished data), consistent with the interpretation that LAP in wild-type MEF cultures is cleaved by metalloproteinases not of the BMP1 subgroup. Tissue inhibitor of metalloproteinase 3 (TIMP-3) has inhibitory activity toward all MMPs and against some members of the ADAM (a disintegrin and metalloproteinase) family of proteinases () but has no inhibitory activity against BMP1 (). When wild-type MEFs were cultured in the presence of the N-terminal inhibitory domain of TIMP-3, which has the same inhibitory activity as the full-length protein (), processing of LAP but not LTBP1 was inhibited (), consistent with the conclusion that LAP is cleaved by non–BMP1-like metalloproteinases. MMP2, - 9, and -14 have previously been identified in vitro as metalloproteinase activators of TGFβ (; ; ). Thus, RNAi knockdown was performed in wild-type MEFs, to test whether any of these might play a role in cleaving LAP in LLCs released from ECM by BMP1. RNAi knock down of each MMP was effective at RNA () and protein () levels. Interestingly, MMP9 RNA and protein levels were also reduced upon knock down of MMP2. Analysis of LAP from conditioned media showed that knock down of MMP2, but not of the other two MMPs, resulted in inhibition of LAP cleavage, such that electrophoretic mobility was the same as for LAP isolated from -null MEF medium (). Consistent with the latter results, Smad2/3 phosphorylation, a direct downstream indicator of TGFβ-specific signaling (), was much reduced in MMP2 RNAi-treated MEFs, compared with untreated MEFs or MEFs treated with MMP9 or -14 RNAi (). Thus, MMP2 is necessary for LAP cleavage and TGFβ activation in MEFs, subsequent to BMP1 liberation of LLC from ECM. MMP2-dependent TGFβ activation probably explains the decrease in MMP9 RNA/protein levels upon MMP2 RNAi treatment, as MMP9 RNA/protein levels are elevated by TGFβ activity (). In , it is demonstrated that MMP2 is capable of in vitro cleavage of LAP in small latent complexes (SLC), yielding a fragment similar in size to the LAP fragment in wild-type MEF media (; ). In contrast, MMP2 cleavage of LLC-associated LAP was extremely inefficient, without prior incubation of LLC with BMP1 (). Thus, in vitro, as in vivo, prior LLC processing by BMP1 greatly facilitates subsequent MMP2 cleavage of LAP. Under nonreducing conditions, immunoblot analysis with anti-LAP antibodies showed LLC to be cleaved to an ∼230-kD form upon incubation with BMP1 and MMP2 (). Mobility corresponding to a 230-kD form, rather than the 35 kD expected for cleaved and dissociated LAP, indicates that LAP remains associated with LTBP1 subsequent to BMP1 cleavage of LTBP1 and MMP2 cleavage of LAP. It should be noted that the aforementioned studies do not formally rule out roles for MMP9 and -14 in TGFβ activation in MEFs, as RNAi knock down of MMP2 also resulted in a knock down of MMP9 protein that was more dramatic than that achieved with MMP9 RNAi (), whereas levels of MMP14 protein could not be ascertained in knockdowns because of the unavailability of antibodies capable of detecting murine MMP14 via immunoblotting. Indeed, the role of MMP14 in activating MMP2 () ensures some level of contribution of MMP14 to TGFβ activation in MEFs. It has previously been shown that LTBP1-containing LLCs can be deposited in ECM as extracellular fibrillar structures associated with fibrillin microfibrils (; ). To determine whether ablation of and affects the appearance of LTBP1 in ECM, anti-LTBP1 immunofluorescence was performed for comparison of wild-type and -null MEF cell layers. Consistent with the results of , in which levels of ECM-associated LTBP1 were markedly increased in -null MEF cell layers compared with wild type, detectable fibrillar structures containing LTBP1 were greatly increased in number and thickness in -null MEF cell layers compared with wild type (). In contrast, fibrillin 1 microfibrils were similar in numbers and thickness in mutant and wild-type cell layers. Cleavage of LAP in wild-type but not -null MEF cultures suggested that TGFβ activity might be reduced in the latter. To test this, levels of active TGFβ in wild-type and -null MEF cultures were compared by adding conditioned media to a reporter gene assay consisting of mink lung epithelial cells stably transfected with a luciferase gene driven by TGFβ-responsive plasminogen activator inhibitor 1 promoter sequences (T-MLEC [transfected mink lung epithelial cell]). Assay results clearly demonstrated -null MEF media to have lower TGFβ activity levels than wild type (), even though the two types of media had similar levels of TGFβ protein, as demonstrated by the similar levels of activity obtained upon heat activation (). Thus, although the two types of MEF culture media have similar levels of TGFβ, much more is in a latent form in -null than in wild-type medium. In a control experiment, the use of either of two different TGFβ-neutralizing antibodies (one specific for TGFβ1 and the other a pan-specific TGFβ antibody that reacts with TGFβ1–5) was capable of reducing both wild- type and -null MEF media activity levels in the T-MLEC reporter gene assay to baseline levels (), thereby demonstrating such activity to be due predominantly to TGFβ1, rather than to other factors that act through the Smad2/3 signaling pathway. In another control experiment, to ensure that latent TGFβ complexes in harvested wild-type and -null MEF conditioned media were not differentially activated during in vitro handling before addition to T-MLEC cultures; MEFs were cocultured with T-MLECs, directly followed by harvesting of monolayers and determination of amounts of luciferase activity. As shown in , coculturing confirmed wild-type MEFs to have intrinsically higher levels of TGFβ signaling than -null MEFs. We have found T-MLECs to secrete active BMP1 (unpublished data), which may release some LLC from ECM during coculturing, thus accounting for the somewhat smaller difference in TGFβ activity levels between wild-type and -null MEFs in the coculturing experiment (), compared with differences found in experiments in which conditioned media were added to T-MLEC cultures (). We next assayed for evidence of effects of decreased TGFβ activity on -null MEF cells. It has been reported that TGFβ1 induces increased levels of MMP2 and -9 via transcriptional, posttranscriptional, and posttranslational mechanisms (; ; ; ). Thus, we compared MMP2 and -9 levels in -null and wild-type MEF conditioned media. Results showed levels of MMP2 RNA () and protein () to be markedly higher in wild-type than in -null MEF cultures and showed differences in MMP9 RNA and protein levels to be even more pronounced. Similarly, levels of CYR61, an established marker for TGFβ activity (), were markedly higher in wild-type than in -null MEF cultures. In contrast, levels of the protein PCOLCE1, expression of which has been shown not to be affected by TGFβ1 (), were similar in -null and wild-type MEF media (). Next, levels of phosphorylated/activated Smad2/3 were compared in -null and wild-type MEFs. As shown in , phosphorylated Smad2/3, readily detectable in wild-type MEFs, was difficult to detect in -null MEFs under the same conditions. Thus, -null MEFs bear multiple hallmarks of cells with reduced levels of TGFβ signaling. We next assayed for differences in levels of detectable LTBP1 and levels of TGFβ signaling in tissues of 13.5-d post conception (dpc) wild-type and -null embryos. As previously reported (), labeling of wild-type tissues with anti–LTBP-1 was sparse. However, markedly higher levels of LTBP1 fibrillar networks were detected in -null tissues (). In contrast, wild-type tissues had markedly higher levels of phosphorylated Smad2/3 than did -null tissues, consistent with markedly higher levels of TGFβ signaling (). In contrast to both these results, levels and distributions of fibrillin 1 microfibrils were similar in the two types of tissues. That the overabundant LTBP1 in -null tissues, which predominantly colocalizes with fibrillin fibrils (unpublished data), does not alter the appearance of fibrillin fibrils compared with wild type is consistent with previous biochemical evidence that LTBP1 is a microfibril-associated protein, rather than an integral structural component of microfibrils (). Interestingly, it has previously been shown that impairment of TGFβ signaling () and ablation of the gene () both result in retardation in formation of the intramembranous frontal bone of the skull. doubly homozygous null embryos die at 13.5 dpc (), before formation of the frontal bone. Thus, we assayed for possible differences in LTBP1 accumulation and TGFβ signaling in frontal bones of 17.5-dpc / and wild-type embryos. 17.5-dpc embryos (), the gap between ossified portions of the forming frontal bones is greatly increased in / compared with wild-type 17.5-dpc embryos (). / compared with wild-type frontal bone, with localization predominantly to the periosteum lining the ossified bone and nonossified mesenchyme of the frontal bone primordium (). / tissues. Levels of fibrillin 1 microfibrils were similar in wild-type and mutant tissues. Much of the TGFβ in tissues is in the form of LLCs linked to ECM via LTBPs. As such, they constitute a reservoir of key cytokines that can be rapidly mobilized in response to tissue perturbations. Interestingly, identified activators of TGFβ may all represent readouts of ECM perturbation (). For example, MMP2 and -9 and plasmin play central roles in ECM degradation, whereas thrombospondin-1 expression is induced in response to tissue damage and is involved in the early stages of ECM regeneration (). Thus, in many cases, ECM perturbation may be an important signal causally preceding activation of TGFβ in ECM-bound LLCs. Indeed, ablation of fibrillin1, the protein via which LTBP1 binds LLCs to ECM, is sufficient to yield raised TGFβ activity in tissues (); similarly, overexpression of truncated LTBP1 that is capable of binding SLC but incapable of binding ECM also yields raised TGFβ activity (). Both results support the concept that activation of LLC-associated TGFβ is greatly enhanced by, and may depend on, prior release from ECM. Here, we describe a novel mechanism for TGFβ activation, involving LLC liberation from ECM via specific LTBP1 cleavage by BMP1-like proteinases. This cleavage does not free LAP from LTBP1, nor does it free active TGFβ from LAP. Rather, the effect of this cleavage in both cell cultures and tissues appears to be consequent cleavage of LAP and activation of TGFβ by non–BMP1-like metalloproteinases. In MEF cultures, we have demonstrated this consequent cleavage to be dependent on MMP2 and perhaps on other non–BMP1-like metalloproteinases as well and have demonstrated that prior BMP1 cleavage of LLC in vitro results in greatly enhanced susceptibility of LLC LAP to cleavage by MMP2. In other cell types, additional proteases may be involved in LAP cleavage, subsequent to BMP1 liberation of LLCs from ECM. TGFβ potently induces net ECM formation by effecting decreased production of some ECM-degrading proteases and increased production of endogenous inhibitors for such proteases, ECM structural proteins, enzymes that stabilize ECM via cross-link formation (i.e., lysyl oxidase; ), and metalloproteinases necessary for biosynthetic processing of ECM structural proteins and lysyl oxidase to their mature functional forms (e.g., the BMP1-like proteinases; ; ). Thus, demonstration here that BMP1-like proteinases effect activation of TGFβ completes a novel feed-forward loop for net ECM deposition. This loop, illustrated in , is likely to feature in various morphogenetic events in which both BMP1-like proteinases and TGFβ and have been implicated as key players, including development, wound repair, angiogenesis (; ), and synaptic plasticity (). Data presented herein are consistent with the possibility that this loop is necessary, in a nonredundant way, for normal formation of the frontal bone of the skull. MMP2 and -9 and various other MMPs are capable of playing key roles in the tissue remodeling associated with the growth, angiogenesis, and invasiveness of tumors (; ). The same is true for TGFβ (; ), whereas high throughput screens have identified BMP1 RNA sequences as among the most up-regulated in activated endothelia associated with tumor angiogenesis (). Thus, the fast-forward loop involving activation of TGFβ by TGFβ-inducible BMP1, with subsequent roles played by MMPs (), is of potential importance to the tissue remodeling associated with morphogenesis and to the pathogenesis of cancers as well. Clearly, mammals possess a variety of molecular mechanisms for activating latent TGFβ, each of which may be suitable to a limited set of circumstances (). Thus, BMP1 activation of TGFβ may be limited to some subset of cell types, to responses to only certain stimuli, and/or to the etiology of only some pathologies involving TGFβ activity. The range of cells and situations in which BMP1 participates in TGFβ activation in vivo remains to be determined. Finally, BMP1-like proteinases are also responsible for activating BMP2 and -4 via cleavage of Chordin () and GDF8 and -11 via cleavage of prodomain sequences (; ). Thus, they may serve to orchestrate signaling by these different morphogenetic TGFβ superfamily ligands and perhaps contribute to coordination between R-Smad2/3 and R-Smad1/5/8 signaling pathways, used by TGFβ/GDF8/GDF11 and BMP2/4, respectively (). They may even contribute to antagonism between the two pathways if, as in the case of Smad4 (), these proteinases occur in limiting amounts. Importantly, the previously characterized roles of the BMP1-like proteinases in ECM formation () and their newly identified roles as activators of TGFβ, make these proteinases potential targets for anti-fibrotic therapeutic interventions. Confluent cells were washed twice with PBS and incubated for 15 min in serum-free DME at 37°C. Cells were then washed once with PBS and placed in serum-free DME/1 μg/ml tetracycline/40 μg/ml soybean trypsin inhibitor. Conditioned medium was harvested every 24 h, and protease inhibitors were added to final concentrations of 1 mM phenylmethylsulfonyl fluoride, 1 mM -ethylmaleimide, and 1 mM p-aminobenzonic acid. Conditioned medium was centrifuged to remove debris, and LLC was purified by sequential affinity purification on Ni-NTA (QIAGEN) and anti-FLAG M2 matrix (Sigma-Aldrich). 500 ng LLC was incubated for 3 h alone or with 25 or 250 ng Flag-tagged BMP1 in 20 μl 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 10 mM CaCl at 37°C. An additional 25 or 250 ng BMP1 was then added and the reaction continued another 3 h. Reactions were stopped with 2× SDS sample buffer/1% β-mercaptoethanol and boiling for 5 min. For MMP2 cleavage, 50 ng SLC or 150 ng LLC preincubated with/without BMP1 was incubated 16 h with/without 60 ng p-aminophenylmercuric acetate–activated proMMP2 (R&D Systems) in 20 μl 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 10 mM CaCl at 37°C. Reactions were stopped with 2× SDS sample buffer/1% β-mercaptoethanol and boiling for 5 min. 2 μg purified recombinant LLC were cleaved as above. Products were resolved on a 12% polyacrylamide SDS-PAGE gel, electrotransferred to Sequi-Blot polyvinylidene difluoride membrane (Bio-Rad Laboratories), and stained with 0.025% Coomassie brilliant blue R-250. N-terminal amino acid sequencing was done by Edman degradation at the Harvard Microchemistry Facility. Confluent MEFs, isolated from 13.5-dpc embryos and immortalized as described previously (), were washed twice with PBS, incubated in serum-free DME for 15 min at 37°C, and incubated for 24 h in serum-free DME and 40 μg/ml soybean trypsin inhibitor. Conditioned media were harvested, and protease inhibitors were added to final concentrations of 5 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 1 mM -ethylmaleimide, and 1 mM p-aminobenzonic acid for media used for Western blots. Inhibitors were not added to media for reporter gene assays. All media were centrifuged to remove debris. For inhibition profiles, 80% confluent wild-type or doubly null MEFs were cultured 24 h in DME/10% FBS with/without 20 μM E-64, 20 μM pepstatin A, 100 μM AEBSF (Sigma-Aldrich), or 200 μM TAPI-2 (BIOMOL Research Laboratories, Inc.). Cells were then switched to inhibitor-containing serum-free DME, and conditioned media were harvested 24 h later. For TIMP-3 inhibition, 90% confluent wild-type MEFs were cultured for 24 h in serum-free DME with/without 100 ng/ml recombinant N-terminal inhibitor domain of TIMP-3, which has inhibitory activity comparable to that of full-length TIMP-3 (). Plasmin extraction of cell-associated ECM was as described previously (). 1.6 × 10 T-MLECs, stably transfected with a luciferase gene driven by TGFβ-responsive plasminogen activator inhibitor-1 promoter sequences (; a gift from D. Rifkin, New York University, New York, NY), were allowed to attach for 3 h in 96-well plates. MEF conditioned medium (for active TGFβ estimation) or heat-treated conditioned medium (for total TGFβ estimation) were harvested and immediately (without storage) added to the T-MLECs, which were harvested 24 h later. Coculture experiments were performed as previously described (). 4 × 10 wild-type or doubly null MEFs were cultured for 16 h in DME/10% FBS in 96-well plates to allow adhesion and cell spreading. 1.6 × 10 T-MLECs were then added and allowed to attach for 3 h. Cells were then switched to serum-free medium containing 0.1% BSA. After 16 h, TGFβ activity was assessed by measuring luciferase activity in coculture cell layer lysates. Luciferase activity was measured using the Luciferase assay system (Promega). 13.5-dpc embryos were harvested, embedded in OCT (Tissue-Tek), frozen, and cut at −30°C into 10-μm sections. Frozen sections were thawed, fixed for 20 min at 4°C with 2% paraformaldehyde in PBS, and washed for 15 min at 4°C with PBS. Fixed cryosections were blocked for 4 h with 10% goat serum in PBS at room temperature and incubated overnight at 4°C with primary antibodies diluted in blocking solution. Rabbit polyclonal anti–phospho-Smad2/3 (Santa Cruz Biotechnology, Inc.), Ab39 anti-LTBP1 (; a gift from C.-H. Heldin, Ludwig Institute for Cancer Research, Uppsala, Sweden), and 9543 anti–fibrillin 1 (; a gift from L. Sakai, Shriners Hospital for Children, Portland, OR) primary antibodies were diluted 1:1,000. Sections were washed with PBS and incubated for 1 h at room temperature with Alexa Fluor 555 goat anti–rabbit secondary antibodies diluted in blocking solution. After washing with PBS, sections were mounted in Immu-Mount (Thermo Electron Corporation), and viewed with a microscope (Axiophot 2; Carl Zeiss MicroImaging, Inc.) with a Plan-NEOFLUAR 40× objective/0.75 aperture. Images were captured with a digital camera (ZVS-3C75DE; Carl Zeiss MicroImaging, Inc.) and Digital Acquire software (DEI 750; Optronics). 4.5 × 10 MEFs were cultured 2 d in Lab-Tek chambers, washed once with PBS, fixed for 10 min with methanol at –20°C, rinsed with PBS, blocked for 4 h with 3% BSA/PBS at room temperature, and incubated overnight at 4°C with primary antibodies diluted in blocking solution. Cells were washed with PBS, incubated for 1 h at room temperature with Alexa Fluor 555 goat anti–rabbit secondary antibody diluted in blocking solution, washed with PBS, and mounted in Immu-Mount. Images were deconvoluted using AutoDeblur and AutoVisualize version 9.3 (AutoQuant Imaging, Inc.), and contrast was adjusted using Photoshop version 7.0 (Adobe) with the same parameters. 5 × 10 wild-type MEFs/well on a 6-well plate were transfected with 250 pmol Stealth RNAi duplexes for MMP2 (sense, 5′-UAUUCCCGACCGUUGAACAGGAAGG-3′), MMP9 (sense, 5′-UAUACAGCGGGUACAUGAGCGCUUC-3′), MMP14 (sense, 5′-AAACUUAUCCGGAACACCACAGCGA-3′), or Stealth medium GC RNAi negative control (Invitrogen), using Lipofectamine. After 6 h, cells were placed in DME/10% FBS, and 18 h later, they were changed into serum-free medium. Conditioned media were harvested after 24 h, as above. Cell layers were washed twice with ice-cold PBS and scraped into hot SDS sample buffer for Western blotting. For RT-PCR, RNA was isolated with TRIzol (Invitrogen), and cDNA was synthesized using 1 μg RNA, random primers, and Super-Script II reverse transcriptase (Invitrogen). PCR was performed at 95°C/3 min, followed by 25, 30, or 35 cycles of 95°C/1 min, 60°C/1 min, and 72°C/1 min, and final extension at 72°C/10 min. Primers were as follows: MMP2, 5′-ACCCATTTGATGGCAAGGAT-3′ (forward) and 5′-TTGTTGCCCAGGAAAGTGAA-3′ (reverse); MMP9, 5′-GGAGAAGGCAAACCCTGTGT-3′ (forward) and 5′-AGGCTGTACCCTTGGTCTGG-3′ (reverse); MMP14, 5′-TCCTGGCTCATGCCTACTTC-3′ (forward) and 5′-GGTGTCAAAGTTCCCGTCAC-3′ (reverse); CYR61, 5′-TCACCCTTCTCCACTTGACC-3′ (forward) and 5′-AGGGTCTGCCTTCTGACTGA-3′ (reverse); GAPDH, 5′-TGGCCAAGGTCATCCATGAC-3′ (forward) and 5′-ATGTAGGCCATGAGGTCCAC -3′ (reverse). Fig. S1 presents a control showing that LTBP1 cleavage occurs only with a form of BMP1 that is catalytically active and demonstrates that the LTBP1 N-terminal site is somewhat more efficiently cleaved than the C-terminal site. Online supplemental material is available at .

(; also known as , or ) encodes a 200-kD protein (1591 aa) characterized by a large coiled-coil region (CC; residues 860–1391) containing a leucine zipper motif (residues 1371–1391; ; ). FIP200 was originally identified by us as a Pyk2 interacting protein through a yeast two-hybrid screen (). It was shown to interact with Pyk2 kinase domain and inhibit its kinase activity in in vitro kinase assays (). FIP200 was subsequently also shown to interact with FAK and function to inhibit FAK activity and autophosphorylation as well as FAK-promoted cellular functions, including cell spreading, cell migration, and cell cycle progression (). Shortly after these studies, FIP200 was also independently identified by as a potential regulator of the RB1 gene. Recent studies suggest that the tuberous sclerosis complex (TSC)–mammalian target of rapamycin (mTOR) signaling network plays an essential role in the regulation of cell growth (; ; ). and are both tumor suppressor genes responsible for tuberous sclerosis, which is characterized by the formation of hamartomas in a wide range of tissues. TSC1 and -2 can form a physical and functional complex in vivo () and function as potent negative regulators of cell growth mainly by their inhibition of mTOR and its targets ribosomal S6 kinase (S6K) and eukaryotic initiation factor 4E binding protein 1 (4EBP1), which play essential roles in the regulation of protein synthesis and cell size. Recent studies suggested that TSC2 functions as the GTPase-activating protein of the small G protein Rheb, an upstream activator of mTOR, and that the TSC1–TSC2 complex antagonizes the mTOR signaling pathway via stimulation of GTP hydrolysis of Rheb (; ). Interestingly, we have recently found a potentially novel function for FIP200 in the regulation of cell growth through its interaction with TSC1 and inhibition of TSC1–TSC2 complex function (). During embryonic development, cell survival/death is tightly regulated by both intrinsic and extrinsic factors. The intrinsic death pathway is activated by the release of cytochrome from mitochondria in response to various stress and developmental death cues, whereas the extrinsic death pathway is mainly activated by the binding of death receptors of the TNF receptor (TNFR) superfamily to their ligands. One of the ligands of death receptors is TNFα. The binding of TNFα to its receptor TNFR1 triggers several intracellular events that regulate both cell survival and cell death. TNFα-induced cell death is mainly mediated by the activation of caspase-8, whereas cell survival effect of TNFα is mainly mediated by the NF-κB pathway (; ). TNFα stimulation can also activate JNK through TNFR1–TNFR-associated factor 2 (TRAF2)–apoptosis signal–regulating kinase 1 (ASK1)–MAPK kinase (MKK) 4/7–JNK signaling cascade (; ). However, the exact role of JNK in TNFα-stimulated cell death signaling is complicated, as JNK has been found to play both antiapoptotic and proapoptotic roles in TNFα signaling in different cellular contexts. A recent study showed that JNK1 and -2 double-knockout (KO) mouse embryo fibroblasts (MEFs) exhibited increased TNFα-stimulated apoptosis, suggesting, at least in MEFs, that JNK could mediate a survival response in TNFα signaling (). Mice KO studies highlight the important role of TNFα signaling in the regulation of cell survival/death during embryonic development. Deletion of some of the genes involved in TNFα signaling, such as Rel A (a subunit of NF-κB), IκB kinase β, and IκB kinase γ, leads to mid/late gestational lethality associated with increased apoptosis in liver, indicating the role of TNFα signaling in the regulation of cell survival and death in the liver development during embryogenesis (; ; ). FIP200 is widely expressed in various human tissues () and is an evolutionarily conserved protein present in human, mouse, rat, , , and . The high degree of conservation during evolution suggests that FIP200 plays important functions in vivo. Despite these recent studies suggesting a role of FIP200 in the regulation of a variety of cellular functions in vitro, however, the function of FIP200 in vivo remains totally unknown. In the present study, we generated FIP200 KO mouse to study the physiological role of FIP200 in vivo. The analyses of FIP200 KO embryos and isolated cells reveal the important function of FIP200 in cell size control and cell survival during embryogenesis and highlight a previously unappreciated role of FIP200 in TNFα-JNK signaling. To study the potential function of FIP200 in vivo, we generated a mouse strain that contains three loxP sequences that flank the exons 4 and 5 of mouse gene and a neo cassette ( ) via a standard homologous recombination technique (). mice were crossed with EIIa-Cre mice (), which led to the generation of three types of offspring: flox allele with neo cassette deleted ( ), neo allele with exons 4 and 5 deleted ( ), and total deletion allele with both exons 4 and 5 and neo cassette deleted ( ; ). Cre-mediated deletion of exons 4 and 5 leads to a frame-shift mutation because of direct splicing from exon 3 (containing ATG codon) to exon 6, producing a small truncated and nonfunctional peptide. and mice were identified by PCR analysis of tail DNA (, left), and the PCR results were confirmed by Southern blotting (, right). mice are normal and fertile and were intercrossed to generate FIP200 KO mice. Genotyping of >300 weaning-age pups identified no homozygous mice, whereas a Mendelian ratio of nearly 1:2 was found for wild-type (WT) and heterozygous pups, suggesting that deletion of the gene results in embryonic lethality. mice yielded homozygous floxed FIP200 mice () at the expected Mendelian ratio, and mice were viable and fertile and did not show any gross or histological abnormalities (unpublished data), further confirming that the embryonic lethality for mice is caused by gene ablation. Extensive timed matings were then performed to determine when FIP200 KO embryos die and to characterize phenotypic defects in the KO embryos. Embryos were analyzed from embryonic day (E) 9.5 to birth (). The normal Mendelian ratio of 1:2:1 of WT, heterozygous, and homozygous embryos was observed until E13.5. Approximately 25 and 60% of homozygous embryos were found dead at E14.5 and E15.5, respectively, and the total number of homozygous embryos (including both alive and dead embryos) at each day were very close to the number of WT embryos. No live FIP200 KO embryos were identified at E16.5 and thereafter. The small size and autolysis of the dead KO embryos recovered after E16.5 are compatible with embryonic lethality between E13.5 and E16.5. Analysis of extracts from whole embryos verified the absence of FIP200 in homozygous KO embryos and reduced expression of FIP200 in the heterozygotes (). Together, these results show that homozygous deletion of FIP200 leads to embryonic lethality at mid/late gestation. Gross examination of the embryos showed normal morphology for FIP200 KO embryos until E12.5 (). At E13.5, ∼20% of the FIP200 KO embryos were pale, and at E14.5 and E15.5 the majority of them were pale compared with WT littermates. Histological studies of the FIP200 KO embryos showed severe cardiac abnormalities characterized by ventricular dilation, sparsely cellular myocardium (), and generalized edema () in a majority of the FIP200 KO embryos analyzed at E14.5 and E15.5. The heart ventricular wall in KO embryos has lost the normal trabecular and external compact myocytes, and the wall was significantly thinner and contained fewer cells when compared with the WT littermates (). In the most severely affected mutants, the myocardium appears to be composed almost entirely of the trabecular layer with no appreciable compact layer underneath the epicardium. In a majority of the FIP200 KO embryos analyzed at E14.5 and E15.5, we also observed liver lesions characterized by loosely arranged hepatocytes mixed with numerous red blood cells (). Hepatocytes were separated from each other with disrupted architecture because of dissecting hemorrhage in between the hepatic cords (). Histological examination of FIP200 KO embryos at E14.5 and E15.5 showed apparently normal morphogenesis of the other major organs as well as extraembryonal tissues, including the placenta (unpublished data). Together, these results suggest that the significant defects in the formation and development of the myocardium and liver are the most likely cause of the embryonic lethality observed in FIP200 KO embryos. Our previous studies showed FIP200 interaction with TSC1 and inhibition of TSC function in cell size control in vitro (). Interestingly, TSC1 KO embryos also exhibited defects in both heart and liver, with thickened ventricular walls (; ). Therefore, deletion of a putative inhibitor of TSC, FIP200, could lead to the thin ventricular walls in the FIP200 KO embryos. To test this possibility and investigate the cellular and molecular mechanisms of FIP200 function in embryogenesis, we examined the effect of FIP200 deletion on TSC functions and cell size regulation in vivo. Embryo sections were analyzed for the activity of S6K, a major target of the TSC–mTOR signaling pathway, by immunohistochemical staining with antibodies against phosphorylated S6. We found that FIP200 KO embryos showed generally weaker staining for phospho-S6 compared with WT littermate embryos (unpublished data). In particular, significantly decreased staining for phospho-S6 was found in heart and liver (). Western blotting analysis of protein extracts from heart and liver confirmed reduced phosphorylation of S6, S6K, and 4EBP1 in FIP200 KO embryos compared with the WT littermates (). Furthermore, we did not observe any change in the expression levels of TSC1 or -2 or the phosphorylation of Akt or TSC2 in the heart and liver from FIP200 KO embryos compared with that from the WT embryos (). Our previous study showed that FIP200 KO MEFs exhibited decreased S6K activation (). Consistent with this, we also found decreased cell size in FIP200 KO MEFs compared with WT MEFs (). We then examined the cell size in FIP200 KO embryos by staining with wheat germ agglutinin (WGA)–tetramethyl rhodamine isothiocyanate (TRITC), which outlines the cell membranes in the tissue sections (). shows a decreased cell size in both heart and liver in FIP200 KO embryos compared with WT littermates. Quantitative analysis of multiple samples suggested an ∼33 and 25% reduction of mean cell size in heart and liver, respectively (). The decreased S6 phosphorylation and cell size were found in FIP200 KO embryos at E14.5 as well as E12.5 and E13.5 when there was no apparent histological defect in FIP200 KO embryos. Therefore, the decreased S6 phosphorylation and cell size were unlikely to be an indirect consequence of other defects (such as apoptosis; see the following paragraph) in FIP200 KO embryos. Together, these results suggested that, consistent with our previous findings in vitro (), FIP200 functions as an inhibitor for TSC, and deletion of FIP200 could lead to increased TSC activity and corresponding decreases in S6K activation and cell size in the embryos. Although the decreased cell size could contribute to the thin ventricular wall defect in the FIP200 KO embryos, we also observed a reduced number of cells in the mutant embryo heart that could not be explained by reduction in cell size. This could be caused by either a decreased cell proliferation or an increased apoptosis or both in the FIP200 KO embryos. Analysis of heart and liver sections from FIP200 KO embryos at E13.5–15.5 by immunohistochemical staining of proliferating cell nuclear antigen showed no apparent difference from WT littermate samples (unpublished data). In contrast, significantly increased apoptosis was found in the mutant heart as measured by immunohistochemical staining for cleaved caspase-3 at E14.5, whereas no significant signals were detected at E12.5 or the WT littermates at E12.5 or E14.5 (). Consistent with the increased apoptosis in the mutant heart, signs of cell death were also observed in the liver of FIP200 KO embryos at both E14.5 and E15.5 (). Immunohistochemical staining for cleaved caspase-3 of liver sections confirmed significantly increased apoptosis in FIP200 KO embryos at E14.5 () and E15.5 (not depicted) compared with WT littermates. The majority of the apoptotic cells were identified as hepatocytes based on their larger size with large eosinophilic cytoplasm and large, vesicular, single, round nucleus possessing a single, prominent, central nucleolus, compared with hematopoietic cells present in the liver. Interestingly, the increased apoptosis in the mutant embryos appeared to be restricted to the liver and heart, as other organs of the FIP200 KO embryos did not show an obvious increase in cleaved caspase-3 staining (unpublished data). As observed in the heart, no increased apoptosis was found for the liver of mutant embryos at E12.5. Together, these results suggest that increased apoptosis, rather than decreased proliferation, is responsible for the decreased cell number of the thin ventricular walls as well as liver lesions in the FIP200 KO embryos. To understand the mechanism of regulation of apoptosis by FIP200, we isolated primary MEFs from FIP200 KO and WT littermate embryos. Under normal culture conditions, FIP200 KO MEFs did not show increased apoptosis compared with the WT MEFs. Furthermore, FIP200 KO and WT MEFs showed a similar extent of apoptosis upon several apoptotic stimuli, such as glucose depletion and sorbitol and anisomycin treatment. In contrast, FIP200 KO MEFs showed an elevated sensitivity to TNFα-induced apoptosis compared with WT MEFs (). Consistent with previous studies (; ; ), TNFα treatment did not cause significant apoptosis in WT MEFs (). Interestingly, it induced significant apoptosis in the FIP200 KO MEFs compared with the untreated cells, suggesting an increased susceptibility to TNFα-induced apoptosis. In contrast, deletion of FIP200 did not sensitize cells to FAS-L– and TRAIL-induced apoptosis in MEFs (), suggesting that FIP200 plays a specific role in TNFα-induced apoptosis. Furthermore, it appears that TNFα-induced necrosis was not increased in FIP200 KO MEFs (). To verify that the increased cell death was caused by apoptosis, lysates from FIP200 KO and WT MEFs that had been treated with TNFα or untreated cells were analyzed by Western blotting for cleaved caspase-3, a marker of apoptotic cell death. shows that TNFα treatment induced an increased amount of cleaved caspase-3 in FIP200 KO but not WT MEFs. Furthermore, reexpression of FIP200 in FIP200 KO MEFs significantly abolished increased caspase-3 cleavage and apoptosis by TNFα treatment in these cells (). Together, these results suggest that FIP200 functions as a suppressor of TNFα-induced apoptosis, and increased susceptibility to TNFα-induced apoptosis may play an important role in the heart and liver defects and embryonic lethality in FIP200 KO mice. To investigate the mechanisms by which FIP200 deletion leads to increased susceptibility to TNFα-induced apoptosis, we first examined potential changes in the NF-κB pathway, which is one of the major cell survival pathways activated by TNFα (). We tested whether TNFα induced IκBα phosphorylation and degradation is affected in FIP200 KO MEFs, which is the biochemical indicator for NF-κB activation by TNFα stimulation. shows rapid induction of phosphorylation (very significant given the dramatically reduced levels of IκBα protein upon stimulation) and degradation of IκBα at 10 min after stimulation in both FIP200 KO and WT MEFs. At 30 min after TNFα treatment, IκBα protein reappeared as expected because of translocation of freed NF-κB to the nucleus and its activation of the transcription of IκBα itself in both cells. Furthermore, TNFα stimulation increased NF-κB transcription activity to similar levels in both FIP200 KO and WT MEFs (unpublished data). These results suggest that deletion of FIP200 did not affect TNFα- induced NF-κB activation. TNFα stimulation also leads to JNK activation, which can promote either cell survival or death, depending on the cellular context (). Interestingly, JNK1 and -2 double-KO MEFs also exhibit more sensitivity to TNFα-induced apoptosis (). We therefore examined whether JNK signaling is affected in FIP200 KO MEFs. shows that induction of JNK phosphorylation (both the transient and robust JNK activation within the first 30 min, and the later sustained and weaker JNK activation) in response to TNFα stimulation was significantly reduced in FIP200 KO MEFs compared with WT MEFs. Consistent with this, phosphorylation of c-Jun on both serine 63 and 73 upon TNFα stimulation was also lower in FIP200 KO MEFs than that in WT MEFs (). We also observed that TNFα stimulation increased c-Jun protein level, which is consistent with the previous observations that JNK-mediated c-Jun phosphorylation protected c-Jun from ubiquitin- dependent degradation, resulting in increased c-Jun protein levels (; ). The increase of c-Jun protein level in response to TNFα stimulation was reduced in FIP200 KO MEFs, consistent with the reduced JNK phosphorylation in response to TNFα stimulation in FIP200 KO MEFs (). Furthermore, we found that reexpression of FIP200 in the FIP200 KO MEFs restored JNK activation in response to TNFα stimulation (), suggesting the JNK phosphorylation defect observed in FIP200 KO MEFs is a direct consequence of FIP200 deletion. To determine whether the reduced JNK activation in FIP200 MEFs is responsible for the increased TNFα-induced apoptosis, we examined the effect of restoration of JNK activity on TNFα-induced apoptosis in these cells. Transient transfection of FIP200 KO MEFs with an expression vector encoding JNK restored JNK activity as measured by the increased c-Jun phosphorylation in these cells (). This increased JNK activity significantly suppressed TNFα-induced apoptosis in FIP200 KO MEFs (), suggesting that the increased sensitivity of FIP200 KO MEFs to TNFα-induced apoptosis is caused by the defective JNK activation in these cells. The aforementioned results from analyses in MEFs raised the possibility that the increased apoptosis in the liver and heart of FIP200 KO embryos is also due to reduced JNK activity in these embryos. To test this possibility directly, we first examined JNK activation status in embryonic liver and heart of FIP200 KO. Consistent with results in MEFs, Western blotting analysis of liver and heart protein extracts showed that JNK phosphorylation was decreased in FIP200 KO embryos compared with WT littermate embryos (). Immunohistochemical analysis of liver sections by anti–phospho-JNK also showed reduced JNK activity in FIP200 KO embryos (unpublished data). We then isolated primary liver cells from FIP200 KO and control WT embryos and examined their JNK activation and apoptosis in response to TNFα stimulation. Similar to the results in MEFs, we found that TNFα-induced JNK activation was reduced in liver cells from FIP200 KO embryos compared with that from control WT embryos (). Also consistent with results in MEFs, liver cells from FIP200 KO embryos exhibited increased sensitivity to TNFα-induced apoptosis and overexpression of JNK suppressed the increased apoptosis in these cells (). Collectively, these results suggest that FIP200 is required for TNFα-induced JNK activation, and reduced JNK activation and its prosurvival function upon FIP200 deletion may be responsible for the increased susceptibility to TNFα-induced apoptosis in FIP200 KO MEFs and hepatocytes, which may contribute to the increased apoptosis and embryonic lethality observed in FIP200 KO embryos. To explore potential mechanisms by which FIP200 participates in the stimulation of JNK signaling by TNFα, we first examined whether FIP200 could affect JNK signaling through its effects on FAK and/or Pyk2. We found that TNFα stimulation did not affect FAK and Pyk2 activation, whereas under the same condition, JNK was potently activated (). Furthermore, similar phosphorylation levels of FAK and Pyk2 were observed in FIP200 KO and WT MEFs, although previous studies showed inhibition of FAK and Pyk2 by overexpression of FIP200 in other cells (; ). These results suggested that FAK and Pyk2 were not involved in FIP200 regulation of TNFα-induced JNK activation. Previous studies documented that TNFα activates JNK through TNFR-TRAF2-ASK1-MKK4/MKK7-JNK signaling cascade (). Thus, we examined whether FIP200 was capable of interacting with any of the components in the cascade. Interestingly, we observed that FIP200 could interact with both ASK1 and TRAF2. shows coimmunoprecipitation of Myc-FIP200 with HA-ASK1 by anti-HA precipitation (left) and reciprocal coimmunoprecipitation of HA-ASK1 with Myc-FIP200 by anti-Myc precipitation (right), when both were transfected into 293T cells. Similar coimmunoprecipitation studies showed association of HA-FIP200 with Myc-TRAF2 in 293T cells (). Consistent with these transfection studies, we also detected the interaction of endogenous FIP200 with the endogenous ASK1 and TRAF2 in MEFs (). We next determined the regions of FIP200 responsible for its association with ASK1 and TRAF2. We found that the FIP200 C-terminal region (CT), but not the N-terminal 859 residues (N1-859) or the CC (), could associate with ASK1 (). In contrast, CC, but not CT or N1-859, coprecipitated with TRAF2 (). These results demonstrated that FIP200 could interact with TRAF2 and ASK1 through two separate regions of the molecule. Previous studies have shown that TRAF2 stimulates TNFα-induced JNK activation through TRAF2 interaction with ASK1 and TRAF2-mediated ASK1 activation (). Thus, our identification of FIP200 interaction with both TRAF2 and ASK1 through different regions raised the interesting possibility that FIP200 may play a role in TNFα-induced JNK activation by serving as a scaffold to facilitate TRAF2-ASK1 signaling to JNK activation. To test this possibility directly, we first evaluated whether TNFα-induced TRAF2–ASK1 interaction and ASK1 activation is reduced in FIP200 KO MEFs compared with WT control MEFs. Consistent with previous reports (; ), we found that TNFα treatment increased TRAF2 interaction with ASK1 and stimulated activation of ASK1 as measured by phosphorylation of ASK1 at threonine 845 in WT MEFs (). Deletion of FIP200 significantly reduced TNFα-stimulated TRAF2–ASK1 interaction and ASK1 phosphorylation in FIP200 KO MEFs compared with that in WT MEFs, which could be rescued by reexpression of FIP200 in FIP200 KO MEFs. We then determined whether overexpression of FIP200 CC or CT segment, which is expected to compete with endogenous FIP200 to reduce its interaction with TRAF2 or ASK1, respectively, will affect TRAF2–ASK1 interaction and ASK1 activation. shows that overexpression of either of these two segments, but not N1-859 segment, functioned in a dominant-negative manner to inhibit TRAF2 interaction with ASK1. Furthermore, overexpression of FIP200 CC or CT, but not N1-859 segment, could reduce ASK1 phosphorylation (, left). Consistent with the reduced ASK1 activation, they also reduced JNK activation in these cells (, right). Finally, we found that overexpression of FIP200 CC or CT segment increased TNFα stimulation–induced apoptosis in MEFs () and hepatocytes (not depicted). Together, these results identified novel interactions of FIP200 with ASK1 and TRAF2, which might mediate FIP200 regulation of ASK1 and JNK activation in response to TNFα stimulation. Our analysis of the FIP200 KO embryos showed their defective cardiac and liver development and mid/late embryonic lethality caused by heart failure and liver degeneration. A role for FIP200 in cardiac formation and development is indicated by the severely affected myocardium with sparsely cellular myocardium forming thinner ventricular walls, as well as ventricular dilation and generalized edema, suggesting cardiac failure in the FIP200 KO embryos. Interestingly, on the other hand, TSC1 and -2 KO embryos also exhibited severe heart defects with thickened ventricular walls (, ). The opposite defective cardiac phenotypes suggested that FIP200 might function to antagonize TSC in the regulation of the thickness of the ventricular wall during heart development. This is consistent with our recent findings showing FIP200 interactions with TSC1 that inhibit TSC1–TSC2 complex function in vitro (). They also suggest that a balance between positive and negative regulation of ventricular wall thickness by FIP200 and TSC, respectively, is critical for the normal cardiac development in embryogenesis. Consistent with the opposite cardiac phenotype of the FIP200 and TSC KO embryos and our previous in vitro findings of inhibition of TSC functions by FIP200 (), we found increased TSC activity in the FIP200 KO embryos as measured by a decreased activation of S6K in cardiomyocytes and hepatocytes as well as when the whole FIP200 KO embryo extracts were analyzed. Furthermore, we observed reduced cell size in the heart and liver of FIP200 KO embryos (), as well as isolated FIP200 KO MEFs (). We noted that the reduction in the size of FIP200 KO MEFs (∼5%) is not as pronounced as liver and heart cells in the FIP200 KO embryos. The relatively modest changes in the size of MEFs may account for their apparently normal functions in vivo. It is also possible that the functions of fibroblasts in vivo are less dependent on the changes in these signaling pathways (thus apparently lack of any defective phenotypes in earlier embryogenesis) than cells in the heart and liver, which are organs characterized by high metabolic activities in mid/late gestation that may be more tightly regulated by protein synthesis and cell growth. Consistent with this notion, both TSC1 and -2 KO embryos also showed major defects in heart and liver development (, ; ; ). Therefore, a role of FIP200 in the regulation of cell size/cell growth may contribute to its potential function in heart and liver development as revealed in the current study. It has been shown that cell size increase is a prerequisite for cell proliferation during normal organ growth (). Thus, it would be expected that the cell size decrease observed in FIP200 KO embryos would also lead to a cell proliferation defect. However, we did not detect any defects in cell cycle progression in the heart or liver in the FIP200 KO embryos (unpublished data). Although cell size and proliferation are coordinately regulated in many cases, these two cellular processes have also been shown to be uncoupled under some conditions where change in cell size does not necessarily affect cell proliferation. For example, deletion of reduces the size of myoblasts without affecting their proliferation (). In other cases, cell size and proliferation are regulated in an opposite manner. For example, it was shown that Rb triple-KO MEFs (lacking all three Rb family proteins pRb, p107, and p130) showed a significantly reduced cell size but an increased cell proliferation, when compared with control MEFs (). Thus, it is likely that FIP200 function in these two cellular processes might be uncoupled during embryonic development. Large truncation deletion of the gene has been observed in ∼20% of primary breast cancers screened in a recent study, implicating a potential role of FIP200 as a tumor suppressor (). However, heterozygous deletion of has not led to development of mammary or any other tumors within 1 yr of age so far (unpublished data). It is possible that one WT allele remaining in FIP200 heterozygous mice is sufficient to maintain its potential tumor suppressor function. It has been shown that, for many tumor suppressor genes, the germline single-allele loss in combination with stochastic somatic loss would lead to an increased tumor incidence in certain organs. For example, total deletion of tumor suppressor gene leads to embryonic lethality phenotype, and Brac1 heterozygous mice don't develop tumors. Notably, introduction of a p53-null allele significantly enhances mammary gland tumor formation in Brca1 mammary gland conditional KO mice (). Therefore, it is possible that combination with another allele that enhances breast cancer development might be synergistic in breast cancer development in FIP200 heterozygous mice. Future studies using FIP200 conditional KO mice combined with crossing with other mice tumor models will be necessary to overcome the embryonic lethality of homozygous deletion of FIP200 and clarify its potential role as a tumor suppressor in vivo. Our results suggested that increased apoptosis in FIP200 KO embryos may be caused by an increased susceptibility to TNFα-induced apoptosis. Furthermore, defective JNK signaling was shown to be responsible for increased TNFα-stimulated apoptosis in FIP200 KO cells, which may contribute to the increased apoptosis and embryonic lethality observed in FIP200 KO embryos. The exact role of JNK in TNFα-regulated cell survival/death is complex and remains somewhat controversial. Some studies suggest an essential role for JNK in TNFα-induced cell death (; ). Other studies indicate that JNK is not essential for TNFα-induced cell death () and may suppress TNFα-stimulated apoptosis (; ; ; ). These studies together suggest a model in which JNK regulates TNFα-induced cell death in a temporal fashion such that early transient JNK activation upon TNFα stimulation suppresses TNFα-stimulated apoptosis, whereas late sustained JNK activation promotes TNFα-induced cell death (). Consistent with this model, our data showed that FIP200 KO cells exhibited decreased transient JNK activation () and increased apoptosis upon TNFα stimulation (). Our study thus suggests that the increased apoptosis in FIP200 KO cells is mainly mediated by the reduction of transient JNK activation and that overexpression of JNK in FIP200 KO cells can suppress TNFα-induced apoptosis possibly by restoration of transient JNK activation. Based on the available mouse genome sequence in the database, a DNA fragment in the intron between exons 4 and 5 was obtained by PCR. This fragment was then used to isolate mouse genomic clones from BAC library derived from isogenic 129SvJ mice (Genome Systems, Inc.) by PCR. A targeting vector was then constructed that contains a 1.7-kb left arm, with exons 4 and 5 flanked by two loxP sites followed by a neo cassette gene and a third loxP site, a 5-kb right arm, and a HSV-TK gene. The linearized targeting vector was transfected into the E14.1 embryonic stem (ES) cells, and homologous recombinant clones were identified by Southern blotting analysis using a fragment outside of the left arm as probe (). The recombinant clone was used to generate chimeric mice by injection into C57BL/6 blastocytes. Male chimeras (identified by the presence of agouti coat color) were then mated with C57BL/6 females to determine germline transmission. mice. All mice used in the study were bred and maintained at Cornell University Transgenic Animal Core Facility (Ithaca, NY) under specific pathogen-free conditions in accordance with institutional guidelines. The following primers were used in PCR genotyping: P1, 5′-GGAACCACGCTGACATTTGACACTG-3′; P2, 5′-CAAAGAACAACGAGTGGCAGTAG-3′; and P3, 5′-CATCAGATACACTAGAGCTGG-3′. The combination of primers P1 and P3 amplifies an ∼800-bp fragment from FIP200 allele. The combination of primers P2 and P3 amplifies 262- and 225-bp fragments from WT and FIP200 alleles, respectively. The PCR condition is as follows: 3 cycles at 94°C for 3 min, 60°C for 1 min, and 72°C for 2 min, followed by 33 cycles at 94°C for 1 min, 60°C for 1 min, and 72°C for 2 min, and 1 cycle at 94°C for 1 min, 60°C for 1 min, and 72°C for 10 min. The rabbit antiserum against FIP200 and Pyk2 have been described previously (; ). The mouse monoclonal α-vinculin and Tyrosine397 phospho-FAK antibodies were obtained from Upstate Biotechnology. The rabbit polyclonal α-HA (Y11) antibody, the mouse monoclonal α-c-Myc-tag (9E10) antibody, rabbit polyclonal α-FAK (C20) antibody, rabbit polyclonal and mouse monoclonal ASK1 antibodies, rabbit polyclonal and mouse monoclonal TRAF2 antibodies, and rabbit polyclonal S6K antibody were obtained from Santa Cruz Biotechnology, Inc. JNK1/2 antibody was obtained from BD Biosciences. Cleaved caspase-3, Tyr402 phospho-Pyk2, phospho-JNK, c-Jun, Ser63 phospho-c-Jun, Thr845 phospho-ASK1, Ser73 phospho-c-Jun, IκBα, Ser32-phospho-IκBα, Ser473 phospho-Akt, Akt, Thr1462 phospho-TSC2, Ser240/244 phospho-S6, S6, Thr389 phospho-S6K, Thr37/46 phospho-4EBP1, and 4EBP1 antibodies were purchased from Cell Signaling Technology, Inc. Sorbitol, acridine orange, and anisomycin were obtained from Sigma-Aldrich. TNFα was obtained from Calbiochem, TRAIL was provided by J. Yuan (Harvard Medical School, Boston, MA), and Fas-L was provided by Z. Liu (National Cancer Institute, Bethesda, MD). -glucose–free DME and ethidium bromide were obtained from Invitrogen. The FIP200 KO and WT MEFs were isolated from E12.5 embryos and cultured in DME supplemented with 10% FBS. 293T cells were cultured in DME supplemented with 10% FBS. Vectors encoding HA-FIP200, HA-FIP200 N1-859, and HA-FIP200 CT were described previously (; ). The CC of FIP200 (860–1401 aa) was amplified by PCR and then subcloned into HA tag–containing vector pKH3 to generate plasmid HA-FIP200 CC. Full-length FIP200, FIP200 N1-859, CC, and CT fragments were then subcloned from corresponding pKH3 vectors into Myc tag–containing pHAN vector (), resulting in Myc-FIP200, Myc-FIP200 N1-859, Myc-FIP200 CC, and Myc-FIP200 CT. The HA-ASK1 construct was provided by H. Fu (Emory University, Atlanta, GA), and the NF-κB reporter construct was provided by A. Lin (University of Chicago, Chicago, IL). Full-length mouse JNK1 and TRAF2 cDNA were amplified by RT-PCR and then subcloned into Myc tag–containing pHAN vector to generate the plasmids Myc-JNK1 and -TRAF2, respectively. The recombinant adenovirus Ad-FIP200 was generated as previously described (). Tissue samples and embryos were homogenized, and extracts were used for Western blotting analysis as described previously (). Preparation of whole cell lysates, immunoprecipitation, and Western blotting were performed as previously described (). Histological and immunohistochemical analysis were performed as previously described (). The histological and immunohistochemical slides were examined under a microscope (model BX41; Olympus) with UplanF1 10×/0.3 objective lens at RT, and the images were captured using a camera (model DP70; Olympus) with DP Controller version 1.2.1.108. To determine cell size by WGA staining, after deparaffinization and rehydration, tissue sections were stained with 150 μg/ml WGA tetramethylrhodamine conjugate (Invitrogen) at 37°C for 30 min and washed with PBS three times, each for 4 min. The samples were mounted for analyses by immunofluorescent microscopy. The area of cross section of cells was quantified by NIH Image program. To determine the cell size of cultured cells, FACS analysis with Cell Quest software (BD Biosciences) was performed as previously described (). Apoptosis assays were done as previously described (). The numbers of apoptotic and necrotic cells were determined by fluorescent dye costaining with acridine orange and ethidium bromide as described previously ().

Elucidation of the mechanisms controlling the shape of cells is an important cell-biological issue. In epithelial layers, cells are arranged in an orderly honeycomb-like pattern, in which cell–cell boundaries exhibit a more or less stretched morphology that is similar to the interfaces of soap bubbles, implying that some physical tension is operating for the shaping of cell outlines (; ). The molecular mechanisms that produce such tension, however, remain unresolved. In retinal cells, cadherin cell-adhesion molecules were shown to be required for minimizing the surface areas of the cells in contact with each other (), suggesting that this molecular family controls the tensile property of cell junctions. Cadherin organizes the complex machinery for cell adhesion by interacting with several cytoplasmic components, including catenins. This complex is concentrated at the adherens junction (AJ), which is located directly under the tight junction (TJ; ), and these two types of junctional structures are positioned at the apical-most margin of the entire cell–cell junction, although the cadherin–catenin complex is also distributed throughout the lateral cell–cell contacts. The AJ is lined with actin fibers, and cadherin requires α-catenin, which is one of the catenins that is known to interact with F-actin (; ; ), for its full adhesive activity (), suggesting that cadherin and F-actin cooperate in cell junction organization. In turn, regulators of cadherin or actin are assumed to be involved in the modulation of cell–cell boundary morphology. Rho family small GTPases (Rho GTPases) are pivotal regulators of the actin cytoskeleton (). In epithelial cells, these GTPases have also been implicated in cadherin activities (; ); conversely, the GTPase activities are modulated by cadherin-mediated adhesion (; ; ), suggesting that these enzymes are important for cadherin–actin interplay. The activities of Rho GTPases are regulated by guanine nucleotide exchange factor (GEF), which exchanges GDP for GTP (). Once activated, the small GTPases can interact with various downstream effectors, acting as a molecular switch (). Among several classes of protein identified as Rho GEFs, the Dbl family proteins are best characterized. The Dbl homology domain (DH domain) has been shown to be necessary and sufficient for the GEF activity of Dbl family Rho GEFs (). It is thought that localized activation of GEFs contributes to the spatiotemporal activation of Rho GTPases. Among the Rho GEFs identified, Tiam1, which is a Rac-specific GEF, has been best characterized as a modulator of cell junctions, and its depletion impairs both AJ and TJ formation (; ; ). GEF-H1/Lfc, which is a GEF for Rho, was also implicated in TJ functions (). On the other hand, the role of Cdc42 and its GEFs in cell junction assembly is poorly understood, although the activation of Cdc42 was shown to increase actin accumulation at cell junctions (). Also, nectin, which is an AJ component, could activate Cdc42 through FRG, which is a Cdc42 GEF (). We show that Tuba, which is a Cdc42-specific GEF that belongs to the Dbl family (), plays an important role in the regulation of cell junction configuration by becoming localized at the apical junctions of simple epithelia. We used Caco-2 cells throughout the present experiments, as they exhibited typical simple epithelial morphology. Immunostaining with a mAb specific for Tuba () showed that this molecule was sharply concentrated at cell junctions (). When Tuba cDNA had been introduced into Caco-2 cells, the intensity of junctional Tuba signals increased (), justifying the aforementioned observation on the endogenous Tuba localization. A similar distribution was observed in other simple epithelial cells, such as DLD-1 and MTD-1A (unpublished data). Immunostaining of Caco-2 cells at high densities showed that Tuba was strictly localized at the apical-most margin of cell junctions, colocalizing with ZO-1, which is a TJ component (). Tuba also colocalized with l-afadin (), which is an AJ component (). However, although the l-afadin was also detectable at basal regions of the cell–cell contacts, to some extent, Tuba did not follow this localization of l-afadin. Tuba also overlapped with the apical margin population of αE-catenin, although the latter was distributed throughout the cell–cell contacts (). Other populations of Tuba molecules were localized in the cytoplasm, which were most abundant at perinuclear regions. Because Tuba closely colocalized with ZO-1, we investigated their potential interactions. First, we observed the localization of these molecules in low-calcium medium, in which cadherin molecules became dispersed but ZO-1 remained as clusters. Even under these conditions, Tuba maintained its colocalization with ZO-1 (). We then overexpressed Tuba and ZO-1 together in Caco-2 cells. Excessive Tuba molecules were localized not only along cell junctions, but also became clustered in the cytoplasm, and ZO-1 tightly associated with all these Tuba signals (). As a control, we coexpressed Tuba and β-catenin, but these two molecules did not colocalize (), indicating that the Tuba–ZO-1 colocalization was caused by their specific interactions. We next performed immunoprecipitation using lysates of cells cotransfected with Tuba and ZO-1 cDNAs and found that Tuba coprecipitated with ZO-1 (). Tuba was also able to coprecipitate with ZO-2, but not with ZO-3 (Fig. S1, A and B, available at ). To determine the sites on Tuba responsible for its interaction with ZO-1, we coexpressed a series of deletion mutants of Tuba with ZO-1, tested their binding by immunoprecipitation, and found that their interaction required the C-terminal domain of Tuba (). However, the C-terminal fragment alone was unable to bind ZO-1 (unpublished data), suggesting that some cooperation of the C-terminus with other sites on Tuba may be required for the Tuba–ZO-1 interaction. The importance of the C-terminal domain was confirmed by immunostaining of Tuba deletion-mutant transfectants (). Constructs in which the DH or N-terminal half region had been deleted, designated as ΔDH or ΔN, respectively, could become localized at cell junctions, although ΔN was largely cytoplasmic. In contrast, ΔC was completely cytoplasmic. Finally, we depleted ZO-1 expression by using RNAi, and found that Tuba could not become localized at the cell junctions that had lost ZO-1 (), despite the normal appearance of the apical junctions in ZO-1–deficient cells (; ). These results confirmed that ZO-1 was required for the cell junction localization of Tuba. On the other hand, ZO-2 depletion did not affect Tuba localization (Fig. S1 C), indicating that ZO-1 plays the predominant role in targeting Tuba to cell junctions. Whether Tuba directly binds ZO-1 or -2 or requires a mediator remains to be determined. To elucidate the function of Tuba in Caco-2 cells, we examined the effect of its RNAi-mediated depletion, either by isolating stable transfectants expressing small hairpin Tuba RNAs or by treating cells with a siRNA specific for Tuba, and the results consistently observed in these transfectants (Tuba-RNAi cells) are described in this study. Efficient and specific knockdown of Tuba was confirmed by Western blotting and immunofluorescence (). Immunostaining for ZO-1 and l-afadin revealed that, in Tuba-RNAi cells, the overall outlines of TJ and AJ were distorted; their cell–cell boundaries were overly bent and less strained, compared with those in the controls (), suggesting that those junctions had acquired reduced tension. We quantified this difference by measuring the ratio of junction length to the distance between vertices, defining a “linearity index,” and confirmed that the junctions of Tuba-RNAi cells were excessively curved (). Concomitantly, some of the Tuba-RNAi cells exhibited unusually small apical areas (, arrows). Quantification showed that the variation in the apical surface area increased after Tuba depletion (). We also tested the effects of overexpression of Tuba deletion mutants in Caco-2 cells (). ΔDH lacking the catalytic domain perturbed the junction linearity, as found in Tuba RNAi cells, and ΔC expression showed an effect similar to that of ΔDH, suggesting that these mutant molecules compete with endogenous Tuba for interactions with partners required for their functions. These results corroborated the observation that Tuba inactivation led to the distortion of junctional morphology. Despite the deformed outlines of the apical cell–cell contacts, E-cadherin and F-actin appeared normally concentrated along the AJ in Tuba RNAi cells (). The expression levels of E-cadherin and other junctional proteins were also not changed (). However, their distributions in lower (lateral) portions of cell–cell boundaries were affected. In normal Caco-2 cells, E-cadherin was concentrated at these areas in a peculiar networklike pattern, overlapping with actin fibers with a similar network (). Close-up views showed that the lateral populations of E-cadherin or overlapping F-actin organized strands that were linked with their AJ components. In Tuba-RNAi cells, E-cadherin strands became fragmented and discontinuous with the AJ, with a concomitant disturbance of actin filaments, which appeared less polymerized than in the controls. Quantification of F-actin density at cell junctions confirmed that the lateral actin fibers less densely distributed in the RNAi cells, although the apical actin filaments were only slightly affected (). In addition, Tuba depletion reduced the colocalization of E-cadherin and actin fibers, although the Triton X-100 solubility of E-cadherin was not particularly different between control- and Tuba-RNAi cells (). We also generated stable transfectants expressing Tuba deletion mutants, and examined E-cadherin distribution in them (). In ΔDH- and ΔC-expressing cells, E-cadherin at the lateral membranes appeared more diffuse or fragmentary, as compared with that of FL-expressing cells. Thus, these two constructs not only impaired apical junctional linearity (see above) but also perturbed the lateral distribution of E-cadherin, as in the case of Tuba-depleted cells. On the other hand, in ΔN-expressing cells, E-cadherin became more highly concentrated at the cell–cell contacts. This implies that the N-terminal half region has an inhibitory activity for Tuba. To see the effects of Tuba depletion on more dynamic phases of cell contact, we observed the behavior of E-cadherin and F-actin during the recovery process of cell–cell adhesion by conducting a “calcium-switch” experiment (). In low-calcium medium, E-cadherin disappeared from cell–cell boundaries, and actin fibers irregularly ran along cell peripheries in both control and Tuba-RNAi cells. By 1 h after calcium restoration, E-cadherin and actin became sharply colocalized at cell–cell boundaries. In Tuba-RNAi cells, however, this E-cadherin accumulation was retarded. At 1 h, although we could detect junctional E-cadherin, its signals were fragmented, and actin fibers did not sharply delineate cell–cell boundaries. Furthermore, E-cadherin signals did not fully colocalize with F-actin ones. The retardation of E-cadherin accumulation was confirmed by quantifying the junctional fluorescence intensity of E-cadherin at 5 min after calcium switch (). In contrast to E-cadherin, ZO-1 recruitment to cell junctions was not severely affected in Tuba-RNAi cells (), although the outlines of these cells appeared to be less tense compared with those of control cells. These results suggest that the junctional recruitment of ZO-1 occurs independently of the Tuba signaling, whereas the organization of E-cadherin and actin at cell junctions is under the control of Tuba. Meanwhile, we found no difference in the composition of catenins coprecipitating with E-cadherin between the control and Tuba-RNAi cell lysates (unpublished data), indicating that Tuba deficiency did not affect cadherin–catenin complex formation. As ZO-1 knockdown depleted Tuba from cell junctions, we expected that ZO-1–RNAi and Tuba-RNAi cells would display similar phenotypes. However, this was not the case. The outlines of apical junctions appeared rather straighter in ZO-1–RNAi cells than in the controls (Fig. S2 A, available at ), which is opposite to what we had anticipated. We noticed that myosin IIA or IIB became strongly up-regulated along the apical junctions of ZO-1–depleted cells (Fig. S2 B); furthermore, phosphorylation of serine 19 of myosin light chain (MLC), which reflects its activation, was also up-regulated at ZO-1–depleted cell junctions (Fig. S2 C). In Tuba-RNAi cells, on the other hand, neither myosin localization nor MLC phosphorylation was up-regulated at cell junctions (Fig. S2, E and F). From these observations, we reasoned that ZO-1 depletion up-regulated myosin II activities via unknown mechanisms, and enhanced the contractility of junctional actin fibers, hiding the opposite effect of Tuba depletion. To test this possibility, we examined the effect of Y-27632, which is an inhibitor of Rho kinase () that can suppress MLC activation (), and found that this reagent abolished the up-regulation of myosin II in ZO-1–RNAi cells (Fig. S2 D). Importantly, in the presence of Y-27632, the cell junctions now became more severely distorted in ZO-1–RNAi cells than in the controls (Fig. S2 A), indicating that, provided myosin II is inactive, ZO-1 and Tuba depletions exhibit similar junctional phenotypes. These findings suggest that there are at least two pathways downstream of ZO-1, a Rho–Rho kinase–myosin pathway and a Tuba pathway, and that ZO-1 depletion affected both pathways. As Y-27632 alone can distort cell junctions to some extent (Fig. S2 A), a balance between these pathways may determine the overall junction morphology. Collectively, we can conclude that the phenotypes observed in ZO-1–depleted cells do not contradict with the proposed role of the interaction between ZO-1 and Tuba in cell junction configuration. It is of note that cell junction formation is delayed in ZO-1–depleted cells (; ), which is similar to the properties of Tuba-RNAi cells, supporting the idea that ZO-1–dependent localization of Tuba at cell junctions is required for its functions. Next, we studied the functions of Tuba as a Cdc42 GEF. We first confirmed that Tuba preferentially activated Cdc42 by conducting an in vitro guanine nucleotide exchange assay (Fig. S3, available at ). As the activation of Rho GTPases has been correlated with its membrane association (), we examined the subcellular localization of Cdc42 in Caco-2 cells by immunostaining. In confluent Caco-2 cells, immunostaining signals for Cdc42 were weakly, but consistently, detected at the apical cell–cell junctions, and this localization of Cdc42 became less visible in Tuba-RNAi cells (). When cells had been subjected to calcium-switch experiments, Cdc42 was rapidly recruited to cell–cell contact sites upon the restoration of normal calcium concentration in the control cells, and this redistribution of Cdc42 was retarded in the Tuba-RNAi cells (). The Cdc42 immunofluorescence signal per cell junction was reduced by ∼30% in the Tuba-RNAi cells, when measured at the 30 min incubation point (). These observations suggest that Tuba has the ability to facilitate the recruitment of Cdc42 to cell junctions. Because Cdc42 was shown to be activated during cell junction assembly (; ), we measured its activity by pull-down assays. The results showed that Cdc42 was transiently activated at 5 min after calcium switch, but this activation did not occur in Tuba-RNAi cells (). We could not detect any other differences in the overall activity of Cdc42 between control and Tuba-RNAi cells by this biochemical method; the Tuba-mediated Cdc42 regulation might represent only a small portion of the entire regulatory systems for this small GTPase. To test if the cellular phenotypes observed in Tuba-RNAi cells were caused by reduced Cdc42 activities, we examined if dominant-negative Cdc42 expression could mimic Tuba depletion. Caco-2 cells transfected with Cdc42 mutants were unable to grow to form cell colonies, and for this reason, we observed only early cell–cell contact events by using calcium-switch assays. In the cells transfected transiently with dominant-negative N17-Cdc42, E-cadherin and actin accumulation were perturbed in a fashion similar to those found in Tuba-RNAi cells. E-cadherin remained punctate and F-actin irregularly delineated cell junctions at 1 h after the calcium restoration (). Conversely, when dominant-active V12-Cdc42 had been expressed in Tuba-RNAi cells, the E-cadherin and actin reassembly was dramatically facilitated, resulting in the formation of junctions with a tightened appearance (). Measurement of the junction linearity confirmed that dominant-negative N17-Cdc42 was able to loosen the cell boundaries of the control cells, but could not significantly enhance the phenotype of Tuba-RNAi cells, and dominant-active V12-Cdc42 enhanced the rectilinear organization of cell junctions in both control and Tuba-RNAi cells (). Thus, Cdc42 inactivation mimicked that of Tuba, and activation of Cdc42 could rescue the Tuba RNAi phenotypes. The effect of V12-Cdc42 appeared excessive, as it facilitated even the control cell junction formation, reflecting its dominant-active nature. Previous studies showed that Tuba interacted with neural Wiskott-Aldrich syndrome protein (N-WASP; ). As N-WASP is an effector of Cdc42 (), we studied its potential role in the Tuba–Cdc42 signaling system. We first confirmed that Tuba could coprecipitate with N-WASP (). Next, we knocked down N-WASP by using RNAi (). In N-WASP–RNAi cells, the outlines of cell junctions became distorted in a pattern comparable to that in Tuba-RNAi cells (). For comparison, we also knocked down the expression of IQGAP1, which is another Cdc42 effector implicated in cell junctions (; ), and found that IQGAP1 depletion had no effects on the junction linearity (), suggesting that these effectors have distinct roles. Tuba and Cdc42 were able to localize normally to cell–cell junctions in the N-WASP–depleted cells (), suggesting that these molecules function upstream of N-WASP. We also examined E-cadherin and actin organization in N-WASP–RNAi cells. In them, E-cadherin strands became fragmented at the lateral cell–cell contacts, as seen in Tuba-RNAi cells, and their colocalization with F-actin was dramatically reduced (). The fibrous organization of actin itself appeared to have been suppressed at the lateral cell–cell contact regions, although the apical AJ-associated bundles were not affected. In calcium-switch assays, N-WASP depletion perturbed the organization of E-cadherin and F-actin at early cell–cell contacts (), which is a phenocopy of the effects of Tuba depletion. Finally, by using the calcium-switch assay, we examined whether N-WASP overexpression could rescue the defects of junction assembly that are induced by Tuba depletion. The results showed that overexpression of full-length N-WASP facilitated the accumulation of E-cadherin and F-actin at cell–cell contacts in Tuba RNAi cells (Fig. S4, A and C, available at ), although this effect was most evident at 5 min after calcium restoration, and became less clear at 1 h. In contrast, overexpression of Δcof-N-WASP, which cannot activate Arp2/3 complex–dependent actin nucleation in vitro (), did not facilitate E-cadherin or F-actin accumulation at cell–cell contacts (Fig. S4, B and D). We have demonstrated that Tuba is concentrated at the apical-most margin of the epithelial cell junctions by binding to the TJ protein ZO-1 (or its associated proteins), suggesting that this molecule is localized around the TJ. Depletion of Tuba distorted the outlines of the apical TJ–AJ complex. Closer observations of F-actin and E-cadherin revealed that although the AJ structure itself appeared normal, the distributions of these molecules located lower than the AJ were disorganized. In normal Caco-2 cells, F-actin organized a fibrous network at the lateral portions of the cell junction, and this actin network was linked with the apical actin fibers forming the AJ, and E-cadherin showed an overlapping distribution. As a result of Tuba depletion, the E-cadherin networks became fragmented, with a concomitant reduction of thick, lateral actin filaments. These observations suggest a scenario in which Tuba functions at the level of the TJ, but its effects spread to the lower portions of the junction. As such apical-specific localization has never been reported for other Cdc42 or Rac GEFs, Tuba may play a unique, position-specific role in junctional organization. Our results suggest that Tuba is required for the regulation of geometrical configuration of the apical junctions. Although the mechanism that generates the linear morphology of the apical junctions is poorly understood, we can speculate that the surface tension supported by the cortical actin cytoskeleton is important, as the actin cytoskeleton is a critical regulator of the viscoelasticity of the cell cortex (). The role of Tuba is likely to regulate actin polymerization via Cdc42 and its effectors, and E-cadherin follows the actin distribution, as suggested by their colocalization. The lateral networks of F-actin and E-cadherin strands, thus formed, may increase the surface tension for minimizing the surface area of cell junctions, or stabilize the linear morphology of cell junctions, and this molecular status is disrupted by the loss of Tuba. On the other hand, through the experiments of ZO-1 knockdown, we found that ZO-1 functioned not only as a partner for Tuba but also as a regulator of myosin activity, and the latter activity influenced the morphology of cell junctions. In contrast to the case of ZO-1 depletion, Tuba-depletion did not affect myosin distribution or activity, suggesting that actomyosin contractility may not primarily be involved in the loosened appearance of Tuba-depleted cell junctions. It is likely that cooperations of the myosin-dependent contractility and Tuba-mediated actin organization determine the overall shape of cell junctions. As an alternative role of Tuba, it may regulate the dynamics of cadherin molecules, such as their trafficking and endocytosis, thus affecting junction morphology; this possibility remains to be tested in the future. Supporting the hypothesis that Tuba functions via Cdc42, Tuba promoted recruitment of Cdc42 to cell junctions, transiently activating it during cell junction recovery processes. Dominant-negative and -active Cdc42 mimicked the Tuba depletion and rescued the Tuba RNAi phenotypes, respectively, which is consistent with the idea that Cdc42 acts downstream of Tuba signaling. Furthermore, Tuba was shown to interact with proteins to regulate the actin cytoskeleton, including N-WASP, which is a well-known effector of Cdc42 (). We demonstrated that N-WASP depletion induced distorted apical junctions, just like Tuba depletion, and also that N-WASP was required for normal accumulation of E-cadherin and F-actin at cell junctions, confirming earlier observations (). In addition, N-WASP overexpression restored rapid recruitment of E-cadherin and F-actin to the junctions in Tuba RNAi cells. These results suggest that N-WASP may be one of the components working downstream of the Tuba–Cdc42 pathway; this idea is also supported by a recent study with melanoma cells (). N-WASP activates the Arp2/3 complex, which creates branched actin filaments (; ); and the Arp2/3 complex was shown to be necessary for cells to assemble cadherin-based cell contacts (; ). When considering all of the data together, we can imagine a scheme in which N-WASP, activated by Tuba–Cdc42, enhances actin polymerization at the level of the apical junctions. This process then leads to the delivery of polymerized actin filaments to the lower portions of cell junctions, and these actin filaments may serve as a scaffold on which the E-cadherin–catenin complex may be anchored. Although other mechanisms, such as trafficking of E-cadherin might also be involved, this scheme accounts for the putative mechanism as to how Tuba, which is confined to the apical region, can organize F-actin and E-cadherin throughout the lateral cell–cell contacts. Meanwhile, the rescuing effect of N-WASP overexpression on Tuba-depleted cells appeared to have been transient, indicating that other factors are also involved in the Tuba–Cdc42 signaling system. Calcium-switch experiments showed that Tuba was also required for the initial processes of junction formation. Both E-cadherin and F-actin assemblies were impaired during the reestablishment process of cell–cell contacts. ZO-1 has been shown to form a complex with the cadherin–catenin complex at the early stages of cell junction assembly (; ), implying that, in nascent cell–cell contacts, Tuba may be in closer proximity to cadherins than in mature junctions, and thus could directly control cadherin-mediated adhesion. Even at early cell–cell contacts, however, cell boundaries always displayed a less-tensed appearance, not only in Tuba-RNAi cells but also in N-WASP–RNAi cells or those expressing dominant-negative Cdc42. These observations suggest that a common function of Tuba governs the initial, as well as the mature, phases of cell–cell contact. Several GEFs other than Tuba have been implicated in cell–cell adhesion, including Tiam1, which is a Rac-specific GEF (; ), and Asef, which is also a Rac GEF (). FRG participates in nectin-induced activation of Cdc42 (), and GEF-H1 regulates TJ permeability (). It is likely that many GEFs are sequentially involved in cell junction assembly in a redundant or independent manner. Tiam1 deficiency was reported to disrupt epithelial junctions (; ). Not only Tuba depletion, but also Tiam1 depletion impairs both AJ and TJ organization. In future studies, it is therefore important to define how Rac and Cdc42 GEFs share their roles in the regulation of cell assembly. A partial cDNA clone of human Tuba (KIAA1010) was obtained from the Kazusa DNA Institute (Chiba, Japan). The 5′ sequence was determined by 5′-RACE, using a SMART RACE cDNA amplification kit (CLONTECH Laboratories, Inc.), with cDNA from Colo205 cells used as the template. The 5′ fragment was subsequently cloned by RT-PCR. The 40–amino acid deletion in the original KIAA1010 cDNA clone () was also amplified by RT-PCR. pSUPER vectors were generated as previously described (). pGK-neo was further inserted at the EcoRI–NotI site of pSUPER to obtain stable transfectants. The RNAi target sequence of Tuba was designed and inserted into the BglII–HindIII site of pSUPER, as previously described (). The target sequences were as follows: Tuba-RNAi-1, 5′-AGTCAAGACCTCGTCAAAG-3′; Tuba-RNAi-2, 5′-ACCTTGATGCTCACTAGAA-3′; and cytokeratin 19 (as a specificity control), 5′-GCTAACCATGCAGAACCTC-3′. Stealth siRNAs with the following target sequences were synthesized by Invitrogen: siZO1-1, 5′-GCAGCTCCAAGAGAAATCTTCGAAA-3′; siZO1-2, 5′-GGCAAGAGAAGAACCAGATATTTAT-3′; siZO1-3, 5′-CCCTGGATTTAAGCCAGCCTCTCAA -3′; siZO2-1, 5′- CCCTAAAGGTGAAATGGTGACCATT-3′; siZO2-2, 5′-CCCATAGCTGATATAGCAATGGAAA-3′; siZO2-3, 5′-GGCTAATGAGTTACCTGACTGGTTT -3′; siTuba, 5′-GAGCTTGAGGGAACATACAAGATTT -3′; siNWASP-1, 5′-TCAAATTAGAGAGGGTGCTCAGCTA-3′: siNWASP-2, 5′-TCTGTGGCTGATGGCCAAGAGTCTA-3′; siNWASP-3, 5′-CCCTCTTCACTTTCCTCGGCAAGAA-3′; siIQGAP1-1, 5′-GGCCCTACAGATTCCTGCAGCTAAA -3′; siIQGAP1-2, 5′-GACAGGAAATCCTACGGTTATTAAA -3′; and siIQGAP1-3, 5′-CCAATAAGATGTTTCTGGGAGATAA -3′. Negative-control stealth siRNAs were also obtained from Invitrogen. ΔDH (amino acids 1–734 and 993–1,577), ΔN (amino acids 747–1,577), and ΔC (amino acids 1–1,276) with or without stop codons were generated by PCR and, subsequently, subcloned into the pCA-Sal-Flag-IRES-hygromycin vector to obtain stable transfectants. Mouse ZO-1 cDNA (a gift from S. Tsukita, Kyoto University, Kyoto, Japan) was subcloned in a pCA-Sal-HA vector. The ZO-2 expression vector CAG-NHA-ZO2-Ipuro and ZO-3 expression vector pME18S-ZO3-7myc were also gifts from S. Tsukita. pCA-β-catenin-HA was constructed by K. Tanabe in our laboratory. The N-WASP expression vectors pEF-BOS-myc-N-WASP and pEF-BOS-myc-Δcof-N-WASP were donated by T. Takenawa (University of Tokyo, Tokyo, Japan). Recombinant adenoviruses expressing myc-N17-Cdc42 and myc-V12-Cdc42 were provided by H. Bito and S. Narumiya (Kyoto University, Kyoto, Japan). Caco-2 cells were obtained from the American Type Culture Collection. All cell lines were cultured in a 1:1 mixture of DME and Ham's F12 supplemented with 10% FCS. To enhance cell spreading, we cultured the cells on collagen-coated dishes. Cells at 70% confluence were transfected by use of Effectene (QIAGEN), according to the manufacturer's protocol. For siRNA treatments, cells were transfected by the use of Lipofectamine 2000 (Invitrogen), according to the manufacturer's protocol. We obtained >90% reduction in the protein level for all siRNAs at 3–4 d after transfection, and all experiments were performed during this period. Adenovirus infection was performed by incubating cells with a viral solution for 3–6 h. More than 80% of the cells were infected, and the maximal expression of the transgene was observed at 24–48 h after infection, the timeframe in which all experiments were conducted. For isolating stable transfectants, transfected cells were selected by exposure to 400 μg/ml G418 or 100 μg/ml hygromycin. In the case of RNAi-stable lines, the surviving colonies were picked up and cloned, and then examined for the expression of targeted proteins by Western blotting. Multiple clones were isolated for each target sequence, in which >90% knockdown of protein expression was achieved. For the stable transfectants of deletion mutants, the cells were maintained as uncloned populations. For calcium-switch experiments, cells were cultured overnight in calcium-free DME (Invitrogen) supplemented with 10% dialyzed FCS. Cell adhesion was initiated by adding 1.8 mM CaCl to the medium. Y-27632 (Calbiochem) was added to cultures at a final concentration of 10 μM, and the cells were fixed after 30 min. The following primary antibodies were used: rabbit polyclonal antibody against α-catenin (C-2081; Sigma-Aldrich), rabbit polyclonal antibody against ZO-1 (61–7,300; Invitrogen), mouse mAb T8-754 specific for ZO-1 (a gift from S. Tsukita; ), rabbit polyclonal antibody against ZO-2 (H-110; Santa Cruz Biotechnology, Inc.), rabbit polyclonal antibody against l-afadin (Sigma-Aldrich), mouse mAb 5H10 against β-catenin (a gift from M.J. Wheelock, University of Nebraska, Omaha, NE; ), mouse mAb HECD-1 specific for human E-cadherin (), mouse mAb against p120-catenin (BD Biosciences), mouse mAb against Cdc42 (BD Biosciences), rabbit polyclonal antibody against N-WASP (H-100; Santa Cruz Biotechnology, Inc.), mouse mAb against MLCs (Sigma-Aldrich), mouse mAb specific for phospho-MLC 2 (Ser19; Cell Signaling Technology), rabbit polyclonal antibody against IQGAP1 (H-109; Santa Cruz Biotechnology, Inc.), mouse mAb DM1A against α-tubulin (Sigma-Aldrich), mouse mAb AC-15 against β-actin (Sigma-Aldrich), rabbit polyclonal antibody reactive with myosin IIA (Sigma-Aldrich), rabbit polyclonal antibody against myosin IIB (Sigma-Aldrich), mouse mAb M2 against Flag (Sigma-Aldrich), rabbit polyclonal antibody specific for Flag (Sigma-Aldrich), rabbit polyclonal antibody against myc (for immunofluorescence; Santa Cruz Biotechnology, Inc.), rabbit polyclonal antibody against myc (for immunoprecipitation; Sigma-Aldrich), mouse mAb 16B12 reactive with HA (CRP), and rabbit polyclonal antibody against HA (Millipore). The following secondary antibodies were used: goat Alexa Fluor 488/594–conjugated anti–mouse or anti–rabbit IgG (Invitrogen) and sheep HRP-conjugated anti–mouse or anti–rabbit IgG (GE Healthcare). F-actin was visualized by using Alexa Fluor 488–conjugated phalloidin (Invitrogen). For staining Tuba, ZO-1, ZO-2, and l-afadin, cells were fixed with 100% methanol at −20°C for 20 min. For staining Cdc42, cells were fixed in 10% TCA at 4°C for 15 min, and permeabilized with 0.2% Triton X-100 for 15 min at RT (). Otherwise, cells were fixed with 4% PFA for 20 min or 1.25% PFA for 5 min (for phospho-S19-MLC) by directly adding a 1/4 or 1/16 volume of prewarmed 20% PFA/HBSS to the medium, and permeabilized with 0.1% Triton X-100 for 15 min. Blocking was done by incubating the fixed cells with 5% skim milk in TBS for 10 min at RT. After the antibodies had been diluted with the blocking solution, the cells were incubated at RT for 1 h with the primary antibody, and then for 30 min with the secondary antibody. Samples were mounted in FluorSave (Calbiochem), and imaged by use of a laser scanning confocal microscope (LSM510) mounted on an inverted microscope (Axiovert 100M), using Plan-Neofluar 40×/1.30 NA and Plan-Apochromat 63×/1.40 NA objectives (all Carl Zeiss MicroImaging, Inc.). All images were processed by use of Photoshop (Adobe) software. Quantification of immunofluorescent signal intensity was done by Scion Image (Scion Corp.). In brief, the junctional regions were manually encircled (see for example), and the signal density of each region was measured. The measurement of linearity index was done by LSM 5 Image Browser (Carl Zeiss MicroImaging, Inc.). All statistical analysis was performed by using Excel (Microsoft). For Cdc42 pull-down assays, cells were lysed in HS-buffer (20 mM Tris-HCl, pH 7.4, containing 500 mM NaCl, 5 mM MgCl, and 1% Triton X-100). 10 μg of GST-PAK CRIB domain (a gift from S. Narumiya) was added to the lysate. After 1 h incubation, the lysate was incubated for another 1 h with glutathione–Sepharose 4B beads (GE Healthcare). The beads were subsequently washed three times in HS buffer. For the Triton X-100 solubility experiments, cells were lysed in lysis buffer (50 mM Tris-HCl, pH 7.5, containing 150 mM NaCl, and 0.5% Triton X-100), and incubated for 30 min on ice. Lysates were centrifuged at 17,000 for 10 min, and the supernatants were recovered (soluble fraction). The remaining pellet was resuspended to Laemmli sample buffer (insoluble fraction). Fig. S1 shows that Tuba interacts with ZO-2, but not with ZO-3. Fig. S2 shows the effects of ZO-1 knockdown on junction morphology and myosin localization in comparison with Tuba depletion. Fig. S3 shows that the DH domain of Tuba preferentially activates Cdc42 in vitro. Fig. S4 shows the effects of N-WASP expression on Tuba-RNAi cells. Online supplemental material is available at .

The basic function of neurons is to receive, integrate, and transmit signals. To do so, most neurons develop polarity by forming a single axon and multiple dendrites (; ; ; ). Neurons have the remarkable ability to polarize even in symmetrical in vitro environments (; ). The processes of their polarization have been extensively studied using hippocampal neurons. These cells first form several immature neurites that are capable of becoming either axons or dendrites. One of the neurites then acquires axonal characteristics, whereas the others later become dendrites. Hippocampal neurons must use a robust internal mechanism that guarantees polarization, as they generate a single axon and multiple dendrites even when polarity is altered by axonal amputation (; ). Recent studies have begun to define the signaling pathways involved in neuronal polarization. reported that the extracellular signals laminin and neuron-glia cell adhesion molecule can specify which neurite will become an axon. As effectors of spatial signals, rearrangements of the cytoskeleton are important, as actin filament instability () and tubulin assembly by collapsin response mediator protein-2 (; ) are reported to initiate axon formation. Recent work has shown that spatially localized intracellular signaling pathways, including phosphoinositide-3-kinase (PI 3-kinase), phosphatidylinositol triphosphate, the mPar3–mPar6–aPKC complex (with the exception of some neurons in ; ), Cdc42, Rap1B, STEF/Tiam1, Rac, Akt, adenomatous polyposis coli, and glycogen synthase kinase-3β, are involved in axon specification for neuronal polarity formation (, ; ; ; ; ; ), and PI 3-kinase is implicated as an upstream molecule in these events (; ; ). In spite of this progress, the mechanism and logic of how the polarized distribution of intracellular signals originates in the absence of external asymmetric cues remain elusive. During the polarization of cultured hippocampal neurons, undifferentiated neurites undergo competitive elongation with each other. When one of them exceeds the others by a critical length, it rapidly elongates to become an axon (). This observation led to the proposal that a positive feedback loop and negative regulation among neurites are necessary for neuronal polarization (; ; ). A locally acting positive feedback loop may amplify a small stochastic increase in signals until it exceeds a threshold to induce an axon, and negative regulation may also be important to prevent the formation of surplus axons. However, little is known about the molecular basis of such regulation. To approach this problem, we performed proteome analyses of cultured hippocampal neurons using highly sensitive large- gel 2D electrophoresis (2DE), which can detect ∼11,000 protein spots over a dynamic range of 1–10 (). We describe a novel brain-specific protein, named shootin1. Our data suggest that shootin1 organizes its own polarized distribution to break neuronal symmetry through the PI 3-kinase pathway. Cultured hippocampal neurons are a well-established system to study spontaneous neuronal polarization (; ). They extend several minor processes during the first 12–24 h after plating (stages 1–2). One of these processes then begins to elongate continuously to become an axon (stage 3). The transition from stage 2 to 3 is the initial step of polarization (). To identify proteins involved in neuronal polarization, we performed two separate proteome analyses of cultured rat hippocampal neurons using a 93- × 103-cm large-gel 2DE (). One was to detect proteins up-regulated during neuronal polarization (): we screened ∼6,200 protein spots on 2DE gels and detected 277 that were consistently up-regulated during the transition from stage 2 to 3 ( ≥ 3). The second analysis screened proteins enriched in axons (). Hippocampi dissected from embryonic day (E) 18 rat embryos were cut into ∼1-mm blocks and cultured on plastic dishes, where they formed complicated networks of radial axons in 2 wk. The explants' somatodendritic parts were then separated from the axon networks, and both were compared by 2DE. By screening ∼5,200 protein spots, we detected 200 spots enriched in the axon samples ( ≥ 3). A total of 23 spots were detected by both screenings. Tryptic digestion and mass spectrometry of one of them, located at a molecular mass of 60 kD and p = 5.3 in gels (), identified 10 peptides whose sequences corresponded to the human cDNA sequence KIAA1598 encoding a 5′-truncated ORF of 446 amino acids. A BLAST search identified four human EST clones (BI598285, BG720033, BE568283, and BI457767) and suggested that 10 additional amino acids are present in the complete ORF. We then cloned the cDNAs for the rat and human ORFs and termed them . Rat and human encode proteins of 456 amino acids and predicted molecular masses of 52.4 and 52.6 kD, respectively (). Domain searching revealed that shootin1 contains three coiled-coil domains and a single proline-rich region (). It does not show significant homology to previously known polypeptides, however, suggesting that it belongs to a novel class of proteins. Database searches also identified a mouse orthologue of () and partial ORFs in , chick, zebrafish, and . Invertebrate homologues of were not found in the databases. Thus, shootin1 is probably a late addition to the genome during the evolution of animals. We raised an antibody against recombinant shootin1. of native and recombinant shootin1, in immunoblots of rat cultured hippocampal neurons (, arrowhead). Consistent with the 2DE data for the metabolically labeled protein (), the level of shootin1 expression increased remarkably during stage 2/3 transition (14.4-fold increase; = 4; P < 0.005) and remained high until day in vitro (DIV) 14, thereafter returning to a low level by DIV28 when expression of the presynaptic protein synaptophysin increased (). Immunoblot analysis of various rat tissues detected shootin1 in postnatal day (P) 4 and adult brains but not in other tissues, suggesting that shootin1 is a brain-specific protein (). Expression of shootin1 was relatively low on E15, peaked around P4, and decreased to a low level in the adult brain (). Thus, the expression of shootin1 is up-regulated, both in hippocampal neurons and in brain, during the period of axon formation and elongation. Next, we examined the localization of shootin1 in cultured hippocampal neurons. Immunocytochemical analysis showed a faint and diffuse staining of endogenous shootin1 in early stage 2 neurons (18–24 h in culture; unpublished data). In late stage 2, moderate amounts of shootin1 appeared in some growth cones of minor processes (). We used a volume marker, 5-chloromethylfluorescein diacetate (CMFDA), to measure the relative concentration of shootin1: it was calculated by using CMFDA as an internal standard (shootin1 immunoreactivity/CMFDA staining). The relative concentration of shootin1 accumulated in the growth cones of late stage 2 neurons was 2–4 times higher than that in the cell body (, arrowheads). In stage 3, shootin1 accumulated strongly in axonal growth cones (, arrows): 100% of axonal growth cones showed accumulation ( = 19). The relative concentration of shootin1 in the axonal growth cones of stage 3 neurons was ∼10 times higher than that in the other regions. Notably, the accumulation seen at late stage 2 in minor processes mostly disappeared in stage 3 (, arrowheads), with only 12% of the processes showing accumulation ( = 68). Shootin1 concentration in the cell body remained low throughout stages 2 and 3 (, asterisks). The accumulation of shootin1 in axonal growth cones was observed until stage 5 (unpublished data). To analyze the localization of shootin1 in living neurons, we monitored fluorescent images of EGFP-shootin1 expressed in hippocampal neurons under the cytomegalovirus promoter every 5 min. Although relatively high levels of EGFP-shootin1 appeared in the soma, indicating that the expression exceeds the endogenous levels, its distribution in neurites was virtually identical to that of endogenous shootin1 (see the following paragraph). Consistent with the immunocytochemical data, we observed accumulation of EGFP-shootin1 in the growth cones of minor processes in late stage 2 neurons (). As reported previously (), minor processes showed competitive extension and retraction before polarization. Surprisingly, “hotspots” of EGFP-shootin1 accumulation repeatedly appeared and disappeared in the growth cones of individual neurites ( = 11 cells; and Video 1, available at ). Most of the neurites elongated in conjunction with EGFP-shootin1 accumulation and, conversely, retracted as EGFP-shootin1 disappeared (). To measure relative concentration of EGFP-shootin1 in growth cones, we used the volume marker monomeric red fluorescent protein (mRFP): it was calculated by using mRFP as an internal standard (EGFP-shootin1/mRFP). By quantifying EGFP-shootin1 and mRFP in growth cones and neurite elongation speed, we found a clear dose dependency of neurite elongation rate on shootin1 concentration in the growth cones of stage 2 neurons (). We continued observations until the neurons entered stage 3. Because long exposure to UV light damaged the cells, images were recorded every 30 min ( = 3; ; and Video 2, available at ). After stage 2, when EGFP-shootin1 accumulation fluctuated in individual neurites (), the neurons entered a phase in which one of the neurites was 10–15 μm longer than the others (; and Fig. S1). In most cases, this neurite would later become an axon (, ). In the longest neurites, accumulation of EGFP-shootin1 stabilized in the growth cone ( and Fig. S1, neurite 1, arrows). Simultaneously, the level of EGFP-shootin1 in its sibling neurites decreased dramatically (neurites 2–5). In this period, the mean number of neurites that showed EGFP-shootin1 accumulation decreased to 1.13 ( = 30). The longest neurites then underwent rapid elongation and the cells entered stage 3 (; and Fig. S1). Consistent with the immunocytochemical data (), EGFP-shootin1 remained highly concentrated in axonal growth cones during stage 3, whereas it disappeared from the growth cones of minor processes (; and Fig. S1). The dynamic shift of shootin1 accumulation into the nascent axon raises the possibility that it provides an intracellular asymmetric signal for neuronal polarization. To examine whether the asymmetric accumulation of shootin1 in a single neurite is important for neuronal polarization, we overexpressed EGFP-shootin1 or myc-tagged shootin1 (myc-shootin1) in hippocampal neurons under the stronger β-actin promoter. A high level of EGFP-shootin1 was detected in the soma, with its frequent transport from the soma to growth cones (, arrowheads; and Video 3, available at ). This in turn resulted in more continuous accumulation of EGFP-shootin1 in multiple growth cones (, arrows; compared with the dynamic fluctuation of a lower level of EGFP-shootin1 in , and Video 1) and ectopic accumulation of myc-shootin1 in minor process growth cones in stage 3 neurons (, arrowheads). These results suggest that the limited amount of shootin1 is essential for its asymmetric accumulation in a single neurite. We further cultured the neurons with overexpressed myc-shootin1 until DIV7. Remarkably, 47 ± 2.1% ( = 3; 71 neurons examined; P < 0.0001, compared with myc-GST) of the neurons bore more than one (two to four) axons that were immunostained by the axon-specific markers anti–tau-1 () and anti-synaptophysin (Fig. S2 A, available at ) antibodies but were immunonegative for the dendrite-specific marker anti-MAP2 antibody (Fig. S2 B). In contrast, only 2.5 ± 1.4% ( = 3; 81 neurons examined) of control neurons with overexpressed myc-GST formed supernumerary axons. On DIV4, 32 ± 1.8% of neurons overexpressing shootin1 bore multiple axons ( = 3; 209 neurons examined; P < 0.002, compared with GST) that were immunoreactive for tau-1 and anti-synaptophysin antibodies but were immunonegative for anti-MAP2 antibody. On the other hand, 10 ± 2.5% of control neurons overexpressing GST bore multiple axons ( = 3; 191 neurons examined). At 50 h in culture, 21 ± 0.9% of neurons overexpressing shootin1 bore multiple axon-like neurites ( = 3; 226 neurons examined; P < 0.001, compared with GST) that were immunoreactive for tau-1, whereas only 6 ± 1.2% of neurons overexpressing GST bore multiple axon-like neurites ( = 3; 173 neurons examined). We also quantified the length of the neurites. Neurites labeled by axonal markers were markedly longer than dendrites (). Interestingly, the sum of the length of neurites in neurons overexpressing shootin1 was similar to that in control neurons on DIV7 () and DIV4 (not depicted). Hippocampal neurons elongate axons rapidly (43 μm/d) from stages 3 to 5 (DIV7; ). We consider that the limited amount of structural components produced in cell bodies similarly limits the total neurite elongation in shootin1-overexpressing and control neurons. A similar limitation of neurite growth in neurons with multiple axons was reported previously (). Multiple axons were also induced by nontagged shootin1 cotransfected with EGFP (43 ± 2.6%; = 3; 67 neurons examined; P < 0.001, compared with EGFP), whereas a small population (1.6 ± 1.6%; = 3; 61 neurons examined) of control neurons expressing EGFP formed supernumerary axons, thereby ruling out the possibility that tagging myc to shootin1 influences the effects. These results suggest that the asymmetric accumulation of shootin1 is involved in neuronal polarization. We next suppressed shootin1 expression using a vector-based RNAi system that expresses microRNA (miRNA). To ensure a high level of expression of miRNA before polarization, hippocampal neurons prepared from E18 rat embryo and transfected with the expression vector of a miRNA designated against shootin1 or a control miRNA were plated on polystyrene plates without any coating. After 20 h for the induction of the miRNA expression, the cells were collected and cultured on coverslips coated with polylysine and laminin. The shootin1 miRNA reduced the level of neuronal shootin1 (, arrows), in comparison to control neurons (arrowheads) and neurons transfected with the control miRNA. Repression of shootin1 expression by the miRNA led to significant suppression of neuronal polarization at 50 and 70 h in culture, whereas the control miRNA had no such effect (). On the other hand, 100% of neurons transfected with the shootin1 miRNA ( = 25) became polarized on DIV7. As the 20-h delay in neuronal plating might affect time course of neuronal polarization after plating, we also performed similar experiments using E17 rat embryo. Essentially equivalent data were obtained with E17 rat embryo (). The significant suppression of neuronal polarization by shootin1 RNAi provides evidence that shootin1 is involved in neuronal polarization. As described, shootin1 showed fluctuating accumulation in growth cones concurrent with neurite elongation in stage 2 neurons, raising the possibility that shootin1 accumulation in growth cones stimulates neurite elongation. During the stage 2/3 transition, neurites of hippocampal neurons show dynamic elongation and retraction without a remarkable increase in total neurite length (). In addition, the stage 2/3 transition is a critical period of neuronal polarization. Therefore, we examined the effect of shootin1 overexpression and RNAi during this period (24 and 48 h in culture). In contrast to the data of DIV7 () and DIV4, shootin1 overexpression induced a significant increase in total neurite length during this period (). Furthermore, repression of its level by RNAi resulted in a significant decrease in it (). Along with the time-lapse data, these results suggest that shootin1 accumulation in growth cones stimulates neurite elongation during the transition from stage 2 to 3. We next asked how shootin1 accumulates asymmetrically in hippocampal neurons. As already noted (, arrowheads), the series of time-lapse imaging revealed active transport of shootin1 from the cell body to the growth cones in stages 2 and 3 neurons (). The shootin1 transport was observed along minor processes and axons. , ) reported wave-like anterograde movement of growth cone–like structures along minor processes and axons of cultured hippocampal neurons. The transport rate of these “waves” was ∼3 μm/min, similar to that of slow axonal transport component b, which transports actin (; ). In addition, waves were enriched in F-actin and their movement was reversibly blocked by the actin-disrupting agent cytochalasin. Therefore, , ) suggested that actin and other cytoskeletal components are transported as waves from the cell body to neurite tips via an actin-dependent mechanism. Shootin1 traveled as discrete boluses with growth cone–like structures at a mean rate of 1.0 ± 0.1 μm/min ( = 12), which is similar to the speed of wave transport. We occasionally observed transient retrograde transport of GFP-shootin1. However, as in the case of the wave, retrograde transport was rare and short-lived, quickly reverting to anterograde movement. In addition, the boluses of shootin1 were enriched for F-actin () and the transport was arrested by the actin-disrupting agent cytochalasin D within 5 min (Fig. S3 A, available at ), as reported for the waves. Blebbistatin, an inhibitor of myosin II (), also stopped shootin1 transport (). These results suggest that shootin1 is anterogradely transported with the wave-like structure by an actin- and myosin-dependent mechanism. Within 2 h of the cessation of the transport by blebbistatin or cytochalasin D, shootin1 accumulation in the axonal growth cones of stage 3 neurons disappeared ( and Fig. S3 B, arrows) and a relatively high level of shootin1 was observed in the soma, axonal shaft, and minor processes (arrowheads). To examine whether shootin1 returned back from the axonal growth cones to the cell bodies by diffusion or was locally degraded in the growth cones and newly synthesized in the cell body, we used the photoconvertible reporter Kaede () to distinguish old shootin1 from newly synthesized shootin1. Kaede-shootin1 expressed in stage 3 hippocampal neurons was converted from green to red using UV light, and shootin1 transport was blocked by blebbistatin (). 1 h after the cessation of shootin1 transport, the accumulation of the red Kaede-shootin1 in the axonal growth cones decreased (, yellow arrows), whereas the red fluorescence of Kaede-shootin1 increased in the soma and shaft (yellow arrowheads). On the other hand, we could not detect new synthesis of green Kaede-shootin1 in the soma (, blue arrowhead). These data suggest that shootin1 passively diffuses back from the growth cones to the cell bodies. We next asked whether the anterograde transport of shootin1 is involved in its asymmetric accumulation in hippocampal neurons. As shown in and Fig. S3 C, cessation of shootin1 transport in stage 2 neurons by blebbistatin or cytochalasin D prevented accumulation of shootin1 in multiple growth cones. Stage 2 neurons were cultured for 36 h in the presence of blebbistatin or cytochalasin D. As described, in control neurons, shootin1 accumulates asymmetrically in growth cones of nascent axons during this period. On the other hand, shootin1 did not accumulate in single neurites in the presence of these drugs ( and Fig. S3 D). Cessation of shootin1 transport in already polarized stage 3 neurons also prevented accumulation of shootin1 in axonal growth cones, as described ( and Fig. S3 B). These data indicate that the actin- and myosin-dependent anterograde transport of shootin1 is necessary for its asymmetric accumulation in single growth cones. Recent studies indicate that PI 3-kinase is located at a critical upstream position in signaling pathways for neuronal polarization (; ). We finally examined whether shootin1 interacts with the PI 3-kinase pathway. The physiological association of shootin1 and PI 3-kinase was examined by coimmunoprecipitation assay. When shootin1 was immunoprecipitated from P5 rat brain lysates, coimmunoprecipitation of the p85 subunit of PI 3-kinase was detected (). Shootin1 was also reciprocally coimmunoprecipitated with p85, indicating that it associates with p85 in vivo. PI 3-kinase activity, indirectly visualized by the phosphorylation of Akt at Ser473 (P-Akt), was enriched in the axonal growth cones of stage 3 neurons (, arrows) as reported () and preferentially colocalized there with shootin1 (, insets). We exogenously coexpressed shootin1 and p85 in HEK293T cells but could not detect coimmunoprecipitation between shootin1 and p85 (not depicted). Thus, shootin1 may interact with PI 3-kinase through unidentified neuronal proteins. As shown recently (), overexpression of constitutively active PI 3-kinase (Myr-PI 3-K p110) induced formation of multiple axons ( and Fig. S4 B, available at ), as in the case of shootin1 overexpression. also reported that overexpression of constitutively active Akt (Myr-Akt), a downstream kinase of PI 3-K, induced formation of multiple axons. On the other hand, inhibition of PI 3-kinase activity by 20 μM LY294002, a specific inhibitor of PI 3-kinase, delayed neuronal polarization (), as reported previously () and as in the case of shootin1 RNAi. These results suggest that shootin1 interacts with PI 3-kinase and is involved in a similar pathway mediating neuronal polarity. Next, we examined whether shootin1 functions upstream of PI 3-kinase or vice versa. Shootin1 RNAi decreased its level in axonal growth cones, which in turn inhibited accumulation of PI 3-kinase activity there (, arrows), suggesting that shootin1 in axonal growth cones is required for accumulation of PI 3-kinase activity there. Conversely, myc-shootin1 overexpression induced its ectopic accumulation in the growth cones of minor processes, which in turn resulted in ectopic accumulation there of P-Akt (, arrowheads), thereby suggesting that accumulation of shootin1 can recruit PI 3-kinase activity. On the other hand, inhibition of PI 3-kinase activity by LY294002 did not affect the accumulation of shootin1 in axonal growth cones (, arrows). Shootin1 overexpression or RNAi did not change the activity of PI 3-kinase in hippocampal neurons (), ruling out the possibility that the expression level of shootin1 changes the total activity of PI 3-kinase in neurons. These results suggest that shootin1 regulates subcellular localization of PI 3-kinase activity in hippocampal neurons. We further examined the functions of shootin1 and PI 3-kinase within the cell polarity pathways. Inhibition of PI 3-kinase activity by LY294002 led to a reduction in the percentage of neurons with multiple axons induced by shootin1 overexpression ( and Fig. S4 A). On the other hand, multiple axon formation by overexpression of constitutively active PI 3-kinase was not inhibited by shootin1 RNAi ( and Fig. S4 B). Collectively, these results provide evidence that shootin1 functions upstream of PI 3-kinase and is required for spatially localized PI 3-kinase activity, which is essential for neuronal polarization (). h a v e i d e n t i f i e d a n o v e l b r a i n - s p e c i f i c p r o t e i n , s h o o t i n 1 , u s i n g h i g h l y s e n s i t i v e 2 D E - b a s e d p r o t e o m i c s . T h e s p a t i o t e m p o r a l l o c a l i z a t i o n o f s h o o t i n 1 i n h i p p o c a m p a l n e u r o n s c h a n g e d d y n a m i c a l l y d u r i n g p o l a r i z a t i o n : i t b e c a m e u p - r e g u l a t e d , b e g a n f l u c t u a t i n g a c c u m u l a t i o n a m o n g m u l t i p l e n e u r i t e s , a n d e v e n t u a l l y a c c u m u l a t e d a s y m m e t r i c a l l y i n a s i n g l e n e u r i t e , w h i c h l e d t o a x o n i n d u c t i o n f o r p o l a r i z a t i o n . D i s t u r b i n g t h e a s y m m e t r i c o r g a n i z a t i o n o f s h o o t i n 1 b y e x c e s s s h o o t i n 1 i n d u c e d f o r m a t i o n o f m u l t i p l e a x o n s , w h e r e a s r e p r e s s i n g s h o o t i n 1 e x p r e s s i o n i n h i b i t e d p o l a r i z a t i o n . T h e s e r e s u l t s s u g g e s t t h a t s h o o t i n 1 p l a y s a c r i t i c a l r o l e i n n e u r o n a l p o l a r i z a t i o n . Hippocampal neurons prepared from E18 rat embryos were cultured as described previously (). For hippocampal explant culture, the hippocampi dissected from E18 rat embryos were cut into blocks (∼1 mm), carefully washed to remove dissociated cells, and cultured on polylysine- and laminin-coated plastic dishes. The explants started to extend radial axons on the dishes within 12 h. On DIV14, the radial axons formed complicated networks around the explants. The axonal networks were usually devoid of cell bodies, dendrites, and nonneuronal cells (), although we occasionally observed migration of neurons from blocks onto axonal networks. Such cells were rigorously removed under a microscope using pipette tips. Explants containing somatodendritic parts were separated from radial axons by applying streams of medium to the explants with a pipette, and the explants were then collected in microcentrifuge tubes. Removal of the explants and dissociated cells from axon networks was verified by microscopy. For quantitative 2DE, stages 2 and 3 neurons were metabolically labeled with the culture medium containing 13% of the normal levels of methionine and cysteine plus Pro-mix -[S] in vitro cell labeling mix (containing ∼70% -[S]methionine and ∼30% -[S]cysteine; GE Healthcare) for 4 h. Hippocampal explants were labeled with the same medium for 24 h. 2DE was performed as reported previously (), using a 93- × 103-cm large-gel system (). For differential 2DE, neurons or explants were metabolically labeled and protein spots separated by 2DE gels were visualized by autoradiography. For protein identification, unlabeled protein samples from 2-wk-old rat brains were separated by the 2DE gel and visualized by silver staining. The protein spots corresponding to the radio-labeled ones were then excised from gels and in-gel digested as described previously (). Matrix-assisted laser desorption/ionization mass spectrometry was performed using a Voyager Elite equipped with delayed extraction (Applied Biosystems). Database searches were conducted using the Mascot program (Matrix Science) and National Center for Biotechnology Information databases. cDNA encoding KIAA1598 was provided by T. Nagase and O. Ohara (Kazusa DNA Research Institute, Chiba, Japan). Full-length cDNA of human was obtained by PCR of KIAA1598 with the primers 5′-GCGGATCCATGAACAGCTCGGACGAAGAGAAGCAGCTGCAGCTCATTACCAGTCTGAAG and 5′-GCGGATCCCTACTGGGAGGCCAGTATTC. cDNA encoding rat was amplified by PCR from a rat brain cDNA library (CLONTECH Laboratories, Inc.) with the primers 5′-CCGCTCGAGATGAACAGCTCGGACGAGGAGAAG and 5′-CCGCTCGAGTTACTGGGAGGCCAGGATTCCCTTCAG. The cDNAs were then subcloned into pCMV (Stratagene), pCAGGS with a β-actin promoter (provided by J. Miyazaki, Osaka University, Osaka, Japan; ), pEGFP (CLONTECH Laboratories, Inc.), pGEX (GE Healthcare), and pKaede-MC1 (MBL International Corporation) vectors. Recombinant shootin1 was expressed in as a GST fusion protein and purified on a glutathione–Sepharose column (GE Healthcare), after which GST was removed from shootin1 by PreScission protease (GE Healthcare). Rabbit polyclonal anti-shootin1 antibody was raised against the recombinant shootin1 and affinity purified before use. Immunocytochemistry, CMFDA staining, Rhodamine phalloidin staining, and immunoblot were performed as described previously (). For immunoprecipitation, P4 or P5 rat brains were extracted by addition of lysis buffer (50 mM Tris-HCl, pH 8.0, 1 mM EDTA, 150 mM NaCl, 1% Triton X-100, 0.1% SDS, 0.1% sodium deoxycholate, 2 mM phenylmethylsulfonyl fluoride, 5 μg/ml leupeptin, 10 mM NaF, 1 mM NaVO, and 10 mM β-glycerophosphate) and centrifuged at 100,000 for 30 min at 4°C. The supernatants were incubated with antibodies overnight at 4°C, and immunocomplexes were then precipitated with protein G–Sepharose 4B (GE Healthcare). After washing out beads with RIPA buffer, immunocomplexes were analyzed by immunoblot. Fluorescent and phase-contrast images of neurons were acquired at room temperature using a fluorescent microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) equipped with a Plan-NEOFLUAR 40×, 0.75 NA, or 20×, 0.50 NA, objective, a charge-coupled device camera (AxioCam MRm; Carl Zeiss MicroImaging, Inc.), and imaging software (AxioVision 3; Carl Zeiss MicroImaging, Inc.). Time-lapse microscopy was performed at 37°C using a fluorescent microscope (Axiovert S100; Carl Zeiss MicroImaging, Inc.) equipped with a Plan-NEOFLUAR 40×, 1.3 NA oil iris objective, CSNAP, and Deltavision 2 (Applied Precision) software or Axiovert 200M (Carl Zeiss MicroImaging, Inc.) equipped with a Plan-NEOFLUAR 40×, 0.75 NA objective, LSM 510 scan module (Carl Zeiss MicroImaging, Inc.), and LSM 510 META software (Carl Zeiss MicroImaging, Inc.). The acquired images were analyzed with Multi Gauge (Fujifilm) or LSM510 META software. Neurons or HEK293T cells were transfected with cDNA or RNA by the calcium phosphate method (), Nucleofector (Amaxa), or Lipofectamine 2000 (Invitrogen) before or after plating. For vector-based RNAi analysis, we used BLOCK-iT Pol II miR RNAi expression vector kit (Invitrogen). The targeting mRNA sequence TGAAGCTGTTAAGAAACTGGA corresponds to nucleotides 138–158 in the coding region of rat shootin1, whereas the control vector pcDNA 6.2-GW/EmGFP-miR-neg encodes an mRNA not to target any known vertebrate gene. Antibodies against myc, tau-1, synaptophysin, MAP-2, α-tubulin, the p85 subunit of PI 3-kinase, and monoclonal (587F11) phospho-Akt (Ser473) were obtained from MBL International Corporation, Boehringer, Progen, Sigma-Aldrich, Sigma-Aldrich, Upstate Biotechnology, and Cell Signaling Technology, respectively. CMFDA, Rhodamine phalloidin, blebbistatin, cytochalasin D, and LY294002 were obtained from Invitrogen, Invitrogen, BIOMOL Research Laboratories, Inc., Calbiochem, and Calbiochem, respectively. cDNA encoding Myr-PI 3-K p110 was obtained from Upstate Biotechnology. mRFP was provided by R. Tsien (University of California, San Diego, La Jolla, CA). Fig. S1 shows serial time-lapse images of EGFP-shootin1 accumulation in neurites 1 and 2 of . Fig. S2 shows DIV7 hippocampal neurons overexpressing myc-shootin1, which are immunostained by anti-synaptophysin or anti–MAP-2 antibody. Fig. S3 shows the effects of cytochalasin D on shootin1 distribution in hippocampal neurons. Fig. S4 shows that inhibition of PI 3-kinase activity suppresses formation of shootin1-induced multiple axons, but repression of shootin1 expression by RNAi does not inhibit formation of PI 3-kinase–induced multiple axons. Video 1 is a time-lapse video of a stage 2 hippocampal neuron expressing EGFP-shootin1 as described in . Video 2 is a time-lapse video of a hippocampal neuron expressing EGFP-shootin1 taken from stages 2 to 3 as described in . Video 3 is a time-lapse video of a hippocampal neuron overexpressing EGFP-shootin1 as described in . Online supplemental material is available at .

A subpopulation of cortical neurons (γ-aminobutyric acid transmitting [GABAergic] interneurons) originates from an extracortical forebrain region during development, i.e. the ganglionic eminences that constitute the presumptive basal ganglia. Tracing experiments using thymidine and lipophilic dyes (; ,), as well as fate-mapping experiments in vitro (; ) and in vivo (), have provided direct evidence for the migration of these interneurons from the ventral developing striatum into the major dorsal region of the forebrain—the cortex. We provide evidence for another migratory stream traveling in the opposite direction between the developing cortex and striatum; i.e., the migration of Emx1-lineage cortical neural stem cells from their original dorsal location ventrally into the striatal germinal zone (GZ). This study was prompted by the finding that neural stem cells have been isolated from the embryonic cerebral cortical GZ (), but do not appear to persist in the adult cortex (; ). In contrast, stem cells from the striatal GZ persist into adulthood and senescence () in the adult remnant of the striatal GZ, i.e., the subependyma that surrounds the forebrain lateral ventricles. Given that stem cells are defined by their long-term self-renewal capacity (), we sought to determine the fate of embryonic cortical neural stem cells by taking advantage of region-specific molecular marker expression in transgenic mice that express Cre recombinase under the control of the Emx1 gene (). Emx1 is a homeobox gene that is expressed in a dorsal-restricted pattern in the developing forebrain () and persists in many adult cortical cells (). Emx1 mice have been shown to express Cre recombinase in a spatial and temporal pattern that recapitulates known endogenous Emx1 expression (). Emx1 mice were mated to the Cre-dependent reporter strain LacZ/EGFP (Z/EG), which expresses EGFP upon Cre-mediated excision (). In this way, we were able to indelibly mark and follow the fate of all cortical cells and their progeny that express Emx1 at any point during development. At embryonic day (E) 15.5, the pattern of GFP expression in coronal forebrain sections from Emx1;Z/EG double-transgenic embryos recapitulated the pattern previously described for Emx1 mRNA () and protein (); i.e., that GFP cells were confined to the developing cortex. No GFP cells were detected ventrally in the striatal GZ at this stage (), or in the “border” zone of the striatal GZ, which is defined by expression of Dbx1 and SFRP2 (,). A small number of GFP cells were present in the postmitotic region of the striatum at E15.5, as has been reported previously (), but these cells were clearly absent from the striatal GZ at this stage (). RT-PCR analysis of E15.5 striatal GZ revealed that Emx1 is not expressed in this tissue at this age. To further confirm that the GFP striatal GZ did not contain any GFP neural stem cells, single cells from this region were cultured, and the resultant clonal neurosphere colonies were examined for GFP expression. The E15.5 striatal GZ yielded an average of 80.9 ± 2.7 colonies per 10,000 cells, and all were GFP. Cortical cell cultures from E15.5 Emx1;Z/EG double-transgenic forebrains yielded an average of 34.1 ± 2.8 colonies per 10,000 cells, all of which were GFP (). Each of these separate clonal cortical and striatal neural stem cell colonies were multipotential in that they generated neurons, astrocytes, and oligodendrocytes and were able to generate secondary colonies as shown previously (; ). Surprisingly, examination of coronal sections from postnatal day (PND) 1 Emx1;Z/EG mice revealed that some GFP cells were now scattered along the striatal and septal aspects of the lateral ventricle (). This is the first time that Emx1-lineage cells have been reported in the striatal GZ. GFP cells were particularly numerous in the dorsolateral portion of the striatal GZ, which is an area well populated with colony-forming neural stem cells (; ; ), and were also scattered less densely throughout the ventral extent of the striatal periventricular tissue. Note that in addition to the cortex, cortically derived corpus callosum and anterior commissure fibers, as well as internal capsule axonal fibers that traverse the striatum, were also GFP (), as expected. As reported previously (), GFP cells were also present in the rostral migratory stream and the olfactory bulbs. To determine whether GFP or GFP cells in the striatal GZ were neural stem cells, single cells from dorsal and ventral regions of the striatal GZ were assayed for clonal colony formation. For all experiments at PND1 and adult, “dorsal” refers to the most dorsal 1/3 of the striatal GZ or subependyma, respectively, and “ventral” to the ventral 1/3 (). Both GFP and GFP cells from the PND1 striatal GZ generated separate (GFP or GFP, respectively) clonal colonies () that were self-renewing and capable of generating multilineage progeny, including GABA-expressing neurons (). In addition, immunocytochemical analysis of striatal GZ slices revealed that a subpopulation of GFP and GFP cells within this region expressed the neural stem cell marker nestin. The cortical GZ cultures yielded only GFP sphere colonies and did not yield any GFP sphere colonies (). There was no difference in the number of secondary colonies generated upon dissociation of single GFP and GFP primary colonies of either dorsal or ventral striatal GZ origin (P > 0.05; ), or in the relative percentages of neuronal and glial progeny generated upon differentiation. To confirm that only GFP primary cells generated GFP colonies and rule out the possibility that exposure to culture conditions caused changes in transgene expression, primary PND1 striatal GZ cells were sorted by FACS into GFP and GFP subpopulations () and cultured. In all cases, GFP cells generated only GFP colonies, and GFP cells generated only GFP colonies. Also, GFP primary colonies only generated GFP secondary clonal colonies, and GFP primary colonies generated only GFP secondary clonal colonies. In addition to demonstrating that culture conditions did not affect the integrity of transgene expression, this result suggested that GFP striatal cells do not generate GFP Emx1-lineage cells. Thus, the GFP neural stem cells isolated from the PND1 striatal GZ were derived from the cortical Emx1-lineage, suggesting that cortical neural stem cells had specifically migrated ventrally into the striatal GZ between E15.5 and PND1. Striatal GZ tissue analyzed by RT-PCR at PND1 demonstrated Emx1 expression at this stage in this region; this finding was confirmed by the immunocytochemical detection of Emx1 in the GFP cells in the PND1 striatal GZ (). Moreover, the finding that there were no GFP sphere colonies isolated from the cortical GZ suggests that the migration only occurs in one direction; striatal neural stem cells do not migrate dorsally into the cortex. To further investigate whether Emx1-lineage cortical neural stem cells were indeed migrating ventrally into the striatal GZ, we performed an experiment to better characterize the time course over which these cells migrated. Thin (15-μm) cryosections of forebrains from E16.5, 17.5, and 18.5 mice were analyzed, and they demonstrated a progressive pattern of stem cell migration over this time period, with the majority of migration being completed between E16.5 and 18.5 (). In an effort to quantify the number of migratory cells, the number of GFP cells present in the striatal GZ at PND1 was determined using the optical dissector method. At the most dorsal level of the striatal GZ, Emx1-lineage GFP cells represent up to 19% of the total number of cells present. This proportion decreases in a dorsal-to-ventral manner along the striatal GZ, such that at the most ventral extent of the GZ only 1% of cells are Emx1-lineage GFP cells (). To rule out the possibility that indigenous striatal GZ cells were up-regulating Emx1 over this time course, and to show more definitely that Emx1-lineage cortical stem cells were migrating, we used a technique that was previously effectively used to definitively demonstrate the ganglionic eminence origins of cortical interneurons (,). Live coronal slices were obtained from E15.5 Emx1;Z/EG embryos, and one hemisphere of each slice was surgically transected along the presumptive cortical/striatal GZ boundary (). The other hemisphere was left intact throughout the period of in vitro culture. We predicted that if Emx1-lineage GFP cortical neural stem cells were indeed migrating ventrally into the striatal GZ, we should be able to isolate GFP stem cell colonies from the striatal GZ of intact cultured hemispheres, but not from the striatal GZ of hemispheres that had been transected at the beginning of the culture period. Indeed, GFP neurosphere colonies were isolated from 91% ( = 11) of intact hemispheres. In contrast, no GFP colonies were isolated from surgically transected hemispheres ( = 12). This difference cannot be attributed to cellular damage at the cortical/striatal boundary, by itself, because there was no difference in the number of GFP colonies generated from intact (35.9 ± 2.5) versus transected (32.2 ± 2.7) slices (P > 0.05), although the overall number of colonies obtained from these cultured slices was decreased compared with the number obtained from fresh slices, as might be expected (). Together, these data provide strong evidence that the GFP cortex is the source of migratory GFP neural stem cells in the perinatal striatal GZ. To investigate the possibility that in cultured slices a critical signal from the adjacent cortex was responsible for up-regulating Emx1 in indigenous striatal GZ cells, a DiI labeling experiment was performed. Single DiI crystals were placed on the E15.5 cortical GZ in a position just dorsal to the dorsolateral aspect of the lateral ventricle, but carefully avoiding the striatal GZ (). After culture for 3 d (i.e., E15.5–18.5), analysis of these slices clearly revealed the migration of Emx1-lineage GFP/DiI double-labeled cells from the cortical GZ into the striatal GZ (). Double-labeled cells demonstrated a typical bipolar morphology characteristic of migrating cells (). Importantly, near the migrating cells, other GFP cells were detected that did not display bipolar morphology and were not labeled with DiI (). This clearly demonstrates that DiI was not simply diffusing throughout the slice, but instead was specifically labeling cells migrating ventrally from the cortex. To obtain direct evidence of cell migration, time-lapse video recordings of E16.5 coronal forebrain slices were performed. Analysis of the 4-d videos (E16.5–20.5) clearly showed the dorsal-to-ventral migration of GFP cells originating from the immediately cortex-adjacent dorsal striatal GZ (). The GFP cells migrated dorsoventrally at an average speed of 9.8 ± 2.1 μm/h ( = 6), whereas adjacent GFP cells moved at an average speed of 1.5 ± 1.8 μm/h ( = 7). To demonstrate that neural stem cell migration occurs in vivo and is not merely an artifact of slice culture, GFP cells from the cortical GZ of PND1 Emx1 mice were homotopically transplanted into the cortical GZ of wild-type GFP PND1 mice. Mice in which cortical GZ cells were injected into the lateral ventricle were excluded from further analysis ( = 10). Of the remaining mice ( = 4), there were two independent cases in which the transplanted GFP cells integrated into the host cortical GZ and, after 5 d, were found within the striatal GZ (). In contrast, there were no examples of migration from control experiments in which GFP cells from the postmitotic cortical plate of PND1 Emx1 mice were transplanted into host cortical GZ ( = 4). In summary, these data obtained from multiple experimental techniques demonstrate a novel dorsal-to-ventral migration pattern of neural stem cells in the perinatal forebrain. Dlx2 is a homeobox gene that is characteristic of the developing striatum (), adult subependyma (), and striatal neural stem cells (). To determine whether migrating GFP Emx1-lineage cortical neural stem cells maintained Emx1 gene expression characteristic of their host tissue of origin or whether they acquired Dlx2 expression characteristic of their striatal neighbors, nested RT-PCR was performed on single GFP and GFP clonal colonies generated from dorsal and ventral aspects of the striatal GZ. GFP clonal neural stem cell colonies of both dorsal and ventral striatal origin expressed Dlx2, but not Emx1, as is expected of cells indigenous to the striatum. Interestingly, GFP Emx1-lineage clonal colonies from the ventral striatal GZ similarly did not express Emx1, but did express Dlx2. This result was confirmed by immunocytochemical analysis of forebrain slices; 70 ± 2% of GFP cells in the striatal GZ also express Dlx2. Most importantly, GFP clonal colonies of dorsal striatal GZ origin expressed both Emx1 and Dlx2 (), which is suggestive of a transitional gene-expression state. That these transitioning, migrating stem cells continue to express Emx1 was confirmed by the immunocytochemical detection of Emx1 in GFP cells in the PND1 striatal GZ (). The only other clonal colonies to express Emx1 were cortically derived GFP neural stem cell colonies. Real-time RT-PCR indicated that resident cortical neural stem cell colonies expressed a higher level of Emx1 than those that had migrated into the dorsal striatal GZ (). Although not detected by the nested RT-PCR performed on single clonal colonies (), minute levels of Emx1 were also found in pooled samples of dorsal striatal GFP or ventral striatal GFP colonies (). A subpopulation of cortical neural stem cell colonies (6/20 or 30%) also expressed very low levels (<50% the level expressed by striatal colonies in pooled samples) of Dlx2, as reported previously (). Together, these data suggest that as Emx1-lineage cortical neural stem cells migrate ventrally they acquire the striatal characteristic of Dlx2 expression and, ultimately, down-regulate Emx1. One of the criteria that differentiate neural stem cells from more restricted progenitor cells is that they persist throughout life (; ). Experiments were performed on adult animals to test if Emx1-lineage cortical neural stem cells do, indeed, persist throughout life in the adult remnant of the striatal GZ, the lateral ventricular subependyma. Examination of coronal sections from adult Emx1;Z/EG mice revealed that some GFP cells remained scattered along the striatal ventricular subependyma (). Cells were cultured from the striatal subependyma of adult Emx1;Z/EG double-transgenic animals at 6 wk and 8 mo of age. GFP Emx1-lineage neural stem cell colonies formed from striatal cultures at both ages (), indicating that Emx1-lineage cortical neural stem cells persist in the adult striatal subependyma. The data presented () underestimate the total number of Emx1-lineage cortical neural stem cells surviving into adulthood, as ∼33% of the GFP neurospheres isolated from the adult striatal subependyma were β-galactosidase by X-gal histochemistry, indicating some in vivo suppression of the GFP transgene over time. RT-PCR analysis demonstrated that Emx1 was no longer expressed by single GFP clonal adult colonies from dorsal or ventral striatal GZ () and that all colonies expressed the striatal marker Dlx2. Adult cortical cultures did not yield any GFP or GFP sphere colonies (). Interestingly, immunolabeling with a GFP antibody and GAD65/67 antibody (which labels GABAergic neurons) revealed the presence of double-labeled cells in the adult striatum, indicating that Emx1-lineage migratory neural stem cells generate GABAergic neurons in this region (). Cells that were GFP/GAD65/67 were further analyzed by confocal microscopy; >90% of these single cells were confirmed to be double-labeled by this method. These data demonstrate that ventral migration is the postnatal fate of a subpopulation of cortical neural stem cells. However, it remains possible that another subpopulation of these cells may remain dormant in the adult cortex and be transformed into stemlike cells after long periods of culture (). Indeed, other subpopulations of adult progenitors may reside in the white matter; progenitors have been isolated directly from adult human subcortical white matter (), postnatal mouse optic nerve (), and rat postnatal optic nerve after “reprogramming” in vitro (). The dorsal-to-ventral migration of cortical stem cells could be the result of a passive mechanism whereby callosal cortical projection fibers separate the cortical plate from the cortical GZ and push cortical neural stem cells into the striatal GZ. Alternatively, stem cell migration may occur by an active push or pull mechanism involving Slit, Netrin, or Eph/ephrin signaling, as has been shown for migrating neuroblasts (). Importantly, dorsal-to-ventral neural stem cell migration has not been previously described; this study represents not only the first study of in vivo neural stem cell migration but also the first study of a dorsal-to-ventral migration phenomenon of any cell type between the developing cortex and striatum. In addition to the well-established ventral- to-dorsal migration pathway of postmitotic neurons (; ,, ; ; ), the other route described has been a lateral-to-medial route of “inward migration” of early and postmitotic preplate neurons from the piriform cortex to the striatum (). This migration takes place exclusively through the postmitotic mantle tissue, and does not involve the GZ. Interestingly, this mechanism of migration may explain the presence of the small number of previously described postmitotic Emx1 neurons identified in the mantle zone (), but never in the striatal GZ. We previously demonstrated that neural stem cells from different embryonic brain regions maintain expression of molecular markers characteristic of their region of origin, even after passaging in vitro (). Nevertheless, neural stem cells are able to alter their regional gene expression profile in response to their environment (; ). This data demonstrate a dramatic in vivo example of neural stem cells altering their regional identity as they migrate from the cortical GZ to the striatal subependyma in situ. It is possible that this type of plasticity is a property specific to neural stem cells, as it has been demonstrated that migrating Dlx2-lineage GABAergic interneurons originating from the developing striatum never up-regulate Emx1, even though they persist in the cerebral cortex (). Interestingly, a subpopulation of Emx1-lineage cells in the adult striatum has been found to express markers of striatal neurons (). We demonstrate directly that a subpopulation of GABAergic neurons in the striatum are generated by Emx1-lineage neural stem cells that migrate perinatally from the cortex. Further, adult Emx1-lineage stem cells generate progeny that label with BrdU and migrate to the olfactory bulb; these GFP/BrdU cells represent up to 20% of the BrdU population in both the adult subependyma and adult olfactory bulb (unpublished data). It is not known at this time whether Emx1 expression is required for the migration of cortical neural stem cells into the striatal GZ. Mutations of the Emx1 gene have been demonstrated to result in lack of corpus callosum formation (), although this may depend on the genetic background of the mice (). However, in Emx1 mutants the cortex remains intact, according to histological and molecular analyses (; ), with only few subtle cortical defects reported in adult mice (). A more severe cortical phenotype, featuring reduced cortical size, absence of the hippocampus and dentate gyrus, and olfactory bulb defects, is noted in double mutant when both Emx1 and Emx2 gene expression are disrupted (). Similar defects have been described in mice with mutations of the Gli3 gene, which is essential for Emx1 and Emx2 expression (). Defects in the striatum or striatal GZ have not been reported in Emx1 mutants. However, because migratory Emx1-lineage neural stem cells constitute a maximum of 19% of the cells at the dorsal limit of the striatal GZ at PND1, an analysis of early postnatal Emx1 mutants specifically investigating potential striatal GZ stem cell defects may be required to determine the possible requirement for Emx1 gene expression in this cortical neural stem cell migration phenomenon. It is of further interest to note that a related transcription factor, Emx2, has been shown to play a role in the symmetric division of both embryonic cortical neural stem cells () and adult striatal neural stem cells (). Intriguingly, cortical neurogenesis has been demonstrated after induced cell death in layer 6 pyramidal neurons () and, more recently, in layer 5 corticospinal neurons (). These studies suggested that the newly born neurons originated from the adult periventricular striatal subependyma and migrated into cortical regions (; ). In addition to subpopulations of striatal and olfactory bulb cells (), we hypothesize that the Emx1-lineage cortical neural stem cell pool that takes up residence in the adult periventricular striatal subependyma may be responsible for the generation of new adult cortical neurons. Although the role of Emx1-lineage migratory neural stem cells is not well understood at this point, what emerges from our studies is the identification of a novel dorsal-to-ventral perinatal neural stem cell migration phenomenon that differs in many ways from the well-established ventral-to-dorsal migration pathway of postmitotic neurons (; ,, ; ; ). These differences include the time period of migration (postmitotic neurons are migratory by E12.5; stem cells not until E16.5), the direction of migration, and the identity of the migratory cells. Emx1 mice were generated and genotyped as previously described (). Z/EG mice, which express before Cre excision and after Cre excision, were a gift from C. Lobe (Sunnybrook and Women's College Health Science Centre, Toronto, Ontario, Canada; ). Matings were timed such that noon of the day vaginal plugs appeared was considered E0.5. The day of birth was counted as PND0. Pups were killed by decapitation, and adult mice were killed by cervical dislocation. Brains were removed under a dissecting microscope (Carl Zeiss MicroImaging, Inc.) and sectioned using a Vibratome, as previously described (). Live Vibratome sections were dissected, and the tissue fragments were mechanically dissociated into single cells and plated in defined serum-free media containing EGF and FGF2, as previously described (; ). Clonal stem cell neurosphere colonies have previously been demonstrated to arise from single cells, and not by cellular aggregation, under these conditions (; ). The number of GFP and GFP sphere colonies was determined by visualizing fluorescence with an inverted microscope (Diaphot; Nikon) after 7 d in vitro. That Cre-mediated excision had not occurred in PND1 GFP colonies was confirmed by the presence of β-galactosidase activity using X-gal histochemistry, as previously described (). Individual clonal sphere colonies were picked for passaging or differentiation, as previously described (), or for single-colony RT-PCR analysis (see RT-PCR analysis). Colonies that were differentiated were immunostained after 5 d in vitro for the presence of neurons, astrocytes, and oligodendrocytes using antibodies against β-tubulin, glial fibrillary acidic protein, and O4, respectively, as previously described (), as well as rabbit polyclonal anti-GABA (Sigma-Aldrich) used at 1:500. For some experiments, primary PND1 striatal cells were isolated as described in the previous paragraph, and cells were sorted with an EPICS Elite Cell Sorter (Beckman-Coulter) based on GFP fluorescence. GFP and GFP populations after FACS were determined to be 99.5% pure. Live coronal Vibratome sections (∼200 μm thickness) of E15.5 forebrains were obtained and cultured essentially as previously described (,). One hemisphere of each forebrain slice was transected at the presumptive cortical/striatal border, and the other hemisphere was left intact. Slices were cultured for 5–10 d with media supplemented every 2 d. For the direct labeling experiments, single DiI crystals (Invitrogen) were carefully placed on the cortical GZ of intact E15.5 Vibratome forebrain slices, just dorsal to the dorsolateral corner of the lateral ventricle. These slices were cultured for 3 d and analyzed for the presence of double-labeled DiI/GFP migrating cells. Time-lapse video microscopy was performed on live coronal Vibratome sections (∼300 μm thickness) of E16.5 forebrains that were obtained and cultured essentially as previously described (,). The slices were maintained on the microscope stage in a humidified chamber at 37°C and 5% CO. Video recording was initiated on the day of dissection. Slices were imaged at 10× magnification using an inverted microscope (Axiovert 200; Carl Zeiss MicroImaging, Inc.). Brightfield images were acquired at 2-min intervals to track the same fluorescent cell at each time point to correspond with the fluorescent images taken at 1-h intervals. Images were captured with a digital camera (XCD-SX900; Sony) using ImageJ software (National Institutes of Health). For immunohistochemistry, brains were fixed, embedded, cryosectioned into 15-μm sections, and immunostained essentially as previously described () using rabbit polyclonal anti-GFP (Abcam) at 1:1,000, mouse monoclonal anti-GFP (CHEMICON International, Inc.) at 1:500, rabbit polyclonal anti-GAD65/67 (CHEMICON International, Inc.) at 1:200, mouse monoclonal anti-nestin (CHEMICON International, Inc.) at 1:100, rabbit polyclonal anti-Dlx2 (a gift from D. Eisenstat, University of Manitoba, Winnipeg, Manitoba, Canada) at 1:100, rabbit polyclonal anti-Emx1 (Santa Cruz Biotechnology, Inc.) at 1:500, and anti–rabbit Alexa Fluor 488, anti–mouse Alexa Fluor 488, anti–rabbit Alexa Fluor 568 (all from Invitrogen) at 1:300 dilution. Cell nuclei were counterstained with DAPI. Fluorescent images were visualized using a motorized inverted research microscope (IX81; Olympus) or a confocal microscope (LSM 410; Carl Zeiss MicroImaging, Inc.) and were captured using Microsuite Version 3.2 image analysis software (Olympus/Soft Imaging System Corp.). To ensure correct placement of cells into the host cortical GZ, experiments were performed using Trypan blue as an injection site marker into PND1 CD1 mice. Upon confirmation that the injection technique reliably delivered Trypan blue to the cortical GZ, transplant experiments were initiated as follows: GFP cells were obtained from either the cortical GZ or postmitotic cortical plate of PND1 Emx1 mice, as described in Materials and methods above. A suspension of 5,000 GFP cells in 0.5 μl of D-PBS was injected into the cortical GZ of PND1 wild-type CD1 mice (Charles River Laboratories). After 5 d, brains were fixed, embedded, cryosectioned into 15-μm sections, and analyzed for the presence of GFP cells, as described in the previous section. Levels of mRNA were quantified in pooled stem cell colony samples by real-time PCR using the sequence detection system (ABI/Prism 7000; Applied Biosystems). Reverse transcription was performed with random hexamers on DNase I–treated total RNA with Superscript II (Invitrogen). cDNA samples were amplified using the Brilliant SYBR Green QPCR master mix (Stratagene). Primer pairs were used at 100 nM and were validated for amplification efficiency against the HPRT or GAPDH internal control genes by carrying out a comparative standard curve analysis. Relative quantitation of transcript levels in unknown samples was performed by the [delta][delta]C method using HPRT as internal control. Amplification cycles comprised of 15 s at 95°C and 1 min at 59°C. All primers (5′ and 3′) were designed using the Primer Express software (Applied Biosystems) with melting temperatures as follows: Dlx2, CCA GTT CGT CTC CGG TCA AC (59°C) and TCG GAT TTC AGG CTC AAG GT (58°C); Emx1, GAA GAA GAA GGG TTC CCA CCA T (59°C); and CCG TTG GCC TGC TTC GT (59°C). All analyses used two-tailed tests. P < 0.05 was considered to be a significant difference between groups. The ‘’ for each comparison and the statistical test used (either test or single variable analysis of variance), is noted in the respective figure legend.

The general signaling mechanisms by which the cross-linking of membrane determinants induces linkage to the cytoskeleton is a long-standing issue dating back to the original patching and capping observations () and the ideas of Singer (; ). More recently, such attachments have assumed clearer physiological and pathological importance. For example, receptor-induced dimerization () causes retrograde transport off the filopodia to distal sites for further processing. Bead-induced clustering of integrins and cell adhesion molecules causes retrograde transport of these molecules away from the leading edge, and considerable effort has been devoted to the manner by which different sized ligand-coated beads induce clusters of cell adhesion molecules to link to the retrograde actin flow (; ; ). After binding to membrane receptors, viral particles are eventually associated with the cytoskeleton in different ways (; ). T cell activation, which is initiated by ligation, is mediated by T cell receptor–containing microclusters that reorganize in an actin-dependent manner (). Even lipids and glycosyl-phosphatidylinositol–anchored proteins (GPIAPs), when cross-linked, undergo patching and capping (; ), and GPIAPs can signal across the plasma membrane. The binding of antibody to several GPIAPs was shown early on to induce an association with Src family kinases (SFKs; ). Cross-linking the GPIAP Thy-1 on T lymphocytes results in mitogenesis (; ). Group B coxsackieviruses begin the process of infection of epithelial cells by binding to and clustering the GPIAP coreceptor decay-accelerating factor on the apical surface (). Transmembrane signaling has been speculated to occur in nanodomains such as lipid rafts when clusters are induced via receptor ligation and cross-linking (), and such signaling may serve to link the cluster to the cytoskeleton (). However, the precise mechanisms of how GPIAPs signal and link to the cytoskeleton remain to be elucidated. This issue remains central in the study of the functionality of membrane microdomains (). In this study, we use a novel feature of single-particle tracking (SPT) trajectories as an assay to begin a dissection of how the linkage of certain GPIAPs and transmembrane proteins to the membrane-associated cytoskeleton may be regulated. SPT has been used to study membrane heterogeneity on various time and distance scales. Using video rate SPT, gold particles bound to membrane lipids and proteins were found temporarily corralled in transient confinement zones (TCZs; ; ; ; ). With much higher time resolution, gold particles that bound to lipids and GPIAPs undergo compartmentalized hop diffusion on the millisecond time scale (). Most previous experiments were aimed at producing pauci- or univalent gold to minimize the number of membrane molecules bound to gold so as to minimize artifacts caused by cross-linking membrane molecules (). In contrast, in this study, we deliberately used the gold particle to form clusters of GPIAPs, mimicking the clusters formed under physiological conditions. The size of clusters associated with gold particles is much smaller than the size of clusters that were seen by immunostaining in previous studies (i.e., patches), which may represent ∼1,000 molecules (; ). This protocol produced a unique nanoscale signature in the SPT trajectories, termed transient anchorage, that depends on SFKs, PI3 kinase, cholesterol, and caveolin-1. In some respects, our study confirms and extends the findings of using the GPIAP CD59. A transmembrane protein, the cystic fibrosis transmembrane conductance regulator (CFTR), also exhibits transient anchorage that strictly depends on its C-terminal PDZ-binding domain, but it is regulated differently than the GPIAP anchorage. Mild cross-linking of membrane molecules by paucivalent gold is most likely the reason for transient confinement (; ). However, in our hands, this type of transient confinement was not sensitive to inhibitors of SFKs (unpublished data), which is in contrast to results reported by . We reasoned that perhaps our level of cross-linking was insufficient to induce the involvement of SFKs. Therefore, we used three different layers of cross-linking antibodies () to collect the GPIAPs of interest together under and proximate to a single gold particle. The trajectories of these clusters on the cell membrane reveal how signal transduction influences the lateral motion of GPIAPs that are sufficiently cross-linked. We chose to investigate two GPIAPs, Thy-1 and CD73 (a 5′ exonucleotidase), both of which are endogenously expressed. The trajectories of both molecules exhibited a surprising feature: a considerable number of temporary anchorage events could be observed in the video record ( and Video 1, available at ). To automatically identify the transient anchorage, a detection program was developed. Transient anchorage, in which particles stop (no displacement within experimental error) transiently, is distinguished from transient confinement, in which single particles still diffuse but in a slower and confined manner (). The periods of transient anchorage varied from several hundred milliseconds to >10 s as shown in an example from cross-linked CD73 on IMR 90 cells and are bimodally distributed (). The longer anchorage times are directly visible in the video record. shows how the total relative confinement, combining both TCZs and transient anchorage for both Thy-1 (top) and CD73 (bottom), was increased by the maximal cross-linking protocol up to seven times compared with the control (left), in which the tertiary cross-linking antibody was not applied. Note that despite different definitions of transient anchorage and transient confinement, transient anchorage points sometimes overlapped with TCZs. Therefore, to avoid the overestimation of relative confinement time (RCT) + relative anchorage time (RAT), the time segments in the trajectory in which transient anchorage was detected were removed, and the trajectory was concatenated. This operation resulted in a decreased RCT in trajectories in which transient anchorage was detected. Transient anchorage only occurred after the addition of the tertiary antibodies (, bottom right) and showed titration behavior () so that there was an optimum concentration at a dilution of ∼1:100 with the final concentration of 20 μg/ml. The dependence of transient anchorage on the concentration of tertiary antibody suggests that a critical size and/or number of cross-linked GPIAPs is required for transient anchorage (). Having too many tertiary antibodies will result in a competition for the available binding sites offered by the primary antibodies; this will lead to monovalent binding of the tertiary antibody with diminished cross-linking and transient anchorage. The number of molecules cross-linked in one cluster is important for constructing a detailed mechanism of transient anchorage. In the maximal cross-linking protocol, primary antibodies are first bound to specific GPIAPs (); antibiotin gold is then added, decorating the surface at a very low concentration. Finally, tertiary antibody is added. Under these conditions, it is difficult to estimate the size of the cluster aggregated under one particle. In addition, the possibility of distal patches of cross-linked GPIAPs globally affecting the behavior of the gold particles cannot be excluded. Therefore, we preassembled gold particle–antibody complexes to test whether transient anchorage would still occur. These complexes contain antibiotin gold with biotinylated anti–mouse IgG antibodies and anti–Thy-1 antibodies. An approximation to the number of primary antibodies bound to each gold particle in the preassembled complexes was made by estimating the contact area between the gold particle and cell membrane (see Gold conjugation to cells, procedure II). The estimate suggests that the maximal number of gold-bound Thy-1 molecules is likely to be <135 and is probably much less. These preassembled complexes exhibited transient anchorage, although the RAT was somewhat reduced (20%) compared with the original protocol (28%; Fig. S2 b, available at ). The bimodal distribution of anchorage times was similar to that seen in the maximal cross-linking SPT experiments (). Thus, a cluster of <135 GPIAPs is sufficient to induce transient anchorage. GPIAPs such as CD59 and Thy-1 are found to couple with SFKs, which are distributed in the inner leaflet of the cell membrane, to effect signal transduction (). It has also been shown that filamentous actin accumulates under patches of GPIAPs (). The actin enrichment requires the activities of Fyn and Lck kinases (), both of which belong to the Src family. A plausible hypothesis is that SFKs mediate the process of tethering the cross-linked GPIAP clusters to the membrane-associated cytoskeleton. Indeed, a form of transient confinement with the GPIAP CD59 has been shown to be mediated by Lck (). Thus, we treated cells with the specific SFK inhibitor PP2 before the cross-linking. Transient anchorage was completely suppressed for both GPIAPs tested (), indicating that SFKs are critical for stabilization of the clusters. However, transient confinement was not substantially affected. Because transient anchorage occurs with preassembled particles, the possibility that global activation of SFKs by the maximal cross-linking protocol is responsible for transient anchorage is excluded. We used the Src, Yes, and Fyn (SYF) triple knockout mouse fibroblast cells to confirm these results. Our anti–human CD73 antibody bound to CD73 on SYF cells, and Src rescued SYF cells as assessed by immunostaining (not depicted) and Western blots from these cells (). SYF defective cells showed no transient anchorage and reduced confinement similar to PP2-treated cells. However, transient anchorage was restored in Src-rescued SYF cells (). Thus, SFK activity is crucial for the transient anchorage phenotype. xref #text xref #text To further test whether the critical role of SFKs in the transient anchorage of GPIAPs is related to phosphorylation-enabled cytoskeletal association of clustered GPIAPs, we examined the diffusion behavior of a transmembrane protein, the CFTR. The C-terminal cytoplasmic domain of CFTR associates with the actin cytoskeleton via Na/H exchanger regulatory factor PDZ proteins (EBP50 and E3KARP) and ezrin (; ; ). If transient cytoskeletal association is the cause of transient anchorage and cross-linking is necessary to recruit SFKs for clustered GPIAPs to be linked to the cytoskeleton, CFTR might be expected to demonstrate transient anchorage in a cross-linking and SFK-independent manner. Furthermore, if cholesterol involvement in the transient anchorage of GPIAPs is required because SFKs can only be recruited under the clusters through cholesterol-mediated nanodomains, CFTR might also be expected to demonstrate transient anchorage independent of cholesterol. CFTR tagged with an extracellular HA epitope was expressed in C3H cells and labeled with biotinylated anti-HA antibodies and antibiotin antibody–conjugated gold. We found that CFTR displayed transient anchorage in the absence of tertiary cross-linking antibody (). Upon PP2 or cholesterol depletion treatment, gold-conjugated HA-CFTR demonstrated only a modest reduction in RAT (), whereas both Thy-1 and CD73 showed a marked decrement in RAT after the same treatments (). Also, deletion of the PDZ-binding domain in CFTR, the Δ4 mutant (), caused transient anchorage to be almost completely eliminated (), indicating that Na/H exchanger regulatory factor (EBP50) PDZ proteins and ezrin provide the key linkage to the cytoskeleton that is required for the transient anchorage of CFTR to occur. Thus, GPIAPs exhibit a much more marked dependence of transient anchorage on SFK and cholesterol than does transmembrane CFTR. This suggests a mechanism by which cross-linked GPIAPs could exploit (via a cholesterol-mediated nanodomain followed by SFK regulation) the normal linkages of transmembrane proteins to the cytoskeleton (see Discussion). Robust transient anchorage was observed with the maximal cross-linking protocol. The preassembled complexes clustering 135 or fewer Thy-1 GPIAPs are shown to be sufficient to trigger transient anchorage comparable with the clusters formed by maximal cross-linking. Cholesterol is essential for gold particle–bound clusters to be transiently anchored on the cell surface, suggesting that a cholesterol-dependent nanodomain is formed under maximally cross-linked gold particles. The cross-linking–triggered signaling events involve SFK cascades, PI3 kinase, and caveolin-1. What molecular mechanisms might account for this phenomenon? The broad outlines of how different modes of transient anchorage might occur can be gleaned from recent reviews. suggested that the cross-linking of receptors might be a key to effecting signal transduction by either altering partitioning into existing raft domains or bringing smaller rafts together. suggested that one mode of coupling clustered GPIAPs may involve an unspecified transmembrane protein and recruitment of small inner leaflet rafts containing lipid-linked signaling molecules to the site of the cluster raft on the outer leaflet. To focus future experiments, we propose a somewhat more specific working hypothesis () using these ideas as a starting point. In this model, cross-linking induces cholesterol-dependent nanodomains (cluster rafts) to form on both the inner and outer leaflet. Such clusters would include key transmembrane proteins bridging the inner and outer leaflets so that signals could be transmitted across the membrane in discrete locations. These proteins would be included from the outset and/or incorporated after collision of the cluster with the transmembrane protein. Whether initial anchoring is assisted by oligomerization-induced trapping () is an open issue. When an activated SFK randomly partitions into such a nanodomain on the inner leaflet, phosphorylation by SFK induces a resident transmembrane molecule to attach to the actin cytoskeleton through adaptor proteins. The attachment results in a transient anchorage event that continues until the SFK is deactivated and a recruited phosphatase dephosphorylates the resident linking molecule. Specific molecular players can be accommodated by this generic hypothesis. A novel SFK substrate, C-terminal Src kinase–binding protein (Cbp; or PAG), was identified recently as a transmembrane protein-binding partner for C-terminal Src kinase, a kinase that regulates SFKs by phosphorylation at their C-terminal regulatory site (). There is also evidence indicating that Thy-1 is associated with Cbp (). Furthermore, after phosphorylation by proximate SFKs, Cbp binds the ERM (ezrin/radixin/moesin)-binding protein (EBP50) via its PDZ domain and, thus, associates with the cytoskeleton through ERM proteins (). Therefore, Cbp is a plausible candidate for the transmembrane protein component of the hypothesis. Indeed, we have obtained results that are at least consistent with the transient anchorage of GPIAPs proceeding via the EBP50-ERM-actin cytoskeleton linkage: an EBP50-binding transmembrane protein, CFTR, exhibits transient anchorage without using the maximal cross-linking protocol, and removal of the C-terminal PDZ-binding domain of CFTR, which binds EBP50, in the Δ4 mutant () abrogates transient anchorage (). Therefore, our hypothesis suggests that there is a common mode of cytoskeletal binding for both CFTR and GPIAPs (via EBP50-ezrin) but two different ways to couple to these adaptors, either directly (CFTR) or indirectly (GPIAP) via the formation of nanodomains and SFK regulation. Indeed, a regulatable ezrin linkage to Cbp was recently hypothesized to link rafts containing the B cell receptor to the lymphocyte cytoskeleton (). Previous studies have shown that caveolin-1 deficiency diminishes both the number of caveolae () and caveolin-2 expression (). Therefore, the diminution of transient anchorage in caveolin-1 cells is consistent with some of it occurring in caveolar structures. Alternatively or additionally, other cellular functions of caveolin-1 and -2 may be associated with transient anchorage. In their studies of the mobility of Simian virus 40 (SV40) on the cell surface before and just after its entry into the cell, also found a partial dependence on caveolin-1. Their findings can be compared and contrasted with our study. After diffusing, SV40 often stops as opposed to being transiently anchored as a prelude to internalization via caveolin-1–dependent or –independent pathways. However, only the caveolin-1–independent internalization pathway of SV40 requires tyrosine kinase activity and cholesterol. In our study, transient anchorage does not exclusively depend on caveolin-1, suggesting at least two pathways to anchorage, but both pathways depended on cholesterol and SFK. also used SPT to study the diffusion behavior of mouse polyoma viruslike particles (which are 45 nm in diameter and similar to the size of gold particles) bound to live cell membranes and found cholesterol-dependent but SFK-independent confinement (). Viruslike particles exhibited confined diffusion in small zones 30–60 nm in diameter, which differs from transient anchorage in the fact that neither SFKs nor caveolin is required for confinement. Confinement requires the cross-linking of viral receptor (ganglioside), which promotes linkage to the actin cytoskeleton in an as yet undefined way. In general, based on the extensive study of viral interactions with the cell membrane, it could be anticipated that multiple modes of transient anchorage would exist (). Unexpectedly, inhibition of the conversion of PIP into PIP by PI3 kinase enhances transient anchorage and induces directional motion in a substantial fraction of trajectories. It is possible that PIP may preferentially partition into the inner leaflet microdomains induced by the cross-linked clusters of GPIAPs, and transient anchorage might then be formed via PIP to adaptor proteins that are associated with the actin cytoskeleton underneath the cell membrane, such as the ERM family proteins (). The bidirectional movements observed in some of the trajectories suggests that PIP clusters that participate in transient anchorage might also bind to motor proteins through their FERM or pleckstrin homology domains, thus contributing to the directed transport of cross-linked clusters by walking along cytoskeletal filaments. Binding between clustered PIP and motor proteins also provides a possible explanation for the disappearance of longer stopping periods because the motor proteins are active most of the time and might be expected to move the cluster directionally with short pauses (). Collectively, our data suggests that the transient anchorage phenotype may be regulated in different ways depending on the biological context. Moreover, the transient anchorage assay presented here should be valuable in defining precise linkages to the cytoskeleton and how they are regulated. C3H 10T1/2 mouse fibroblasts (American Type Culture Collection [ATCC]) and IMR 90 human fibroblasts were maintained in basal medium Eagle and DME, respectively, and both were supplemented with 10% FBS, 100 units/ml penicillin, and 100 μg/ml streptomycin. 2–4 d before an SPT experiment, fibroblasts were plated onto sterile 22 × 22-mm coverslips (glass) that were placed into 35-mm petri dishes at an appropriate cell density that yielded single cells for SPT measurements. Cells were depleted of cholesterol by treatment with 5 mM MβCD (Sigma-Aldrich) at 37°C for 30 min in unsupplemented medium or with 0.5 units of cholesterol oxidase (C8273; Sigma-Aldrich) at 37°C for 1 h in complete medium, respectively. The Src, Yes, and Fyn mouse embryonic cell line (SYF; ATCC) and the SYF cell line with restored c-Src expression by the retroviral vector pLXSH (SYF; ATCC) were used to examine the effect of SFKs on transient anchorage. The caveolin-1–deficient mouse embryonic cell line (caveolin-1) and its wild-type parental cell line, which were gifts from M. Schwartz (University of Virginia, Charlottesville, VA), were used for experiments investigating the role of caveolin-1 in transient anchorage. HA-tagged CFTR protein was expressed in C3H cells by LipofectAMINE (Invitrogen) transfection. sup #text sup #text Computer-enhanced video microscopy, which was described previously (), was used to image colloidal gold bound to the plasma membrane of fibroblast cells. In brief, the cell lamella with bound gold was imaged in brightfield mode and recorded with a video camera (Newvicon; Hamamatsu). After real-time background subtraction and contrast enhancement with an image processing unit (Argus 20; Hamamatsu), video frames were recorded in time-lapse mode (1,800 frames with 30 frames/s) on the hard disk of a computer (O2; Silicon Graphics). Recorded videos were analyzed by the commercial software package ISee (Inovision Corp.), which identifies relative changes of gold particle positions on the cell lamella with a precision of ±20 nm. All trajectories were visually inspected to ensure the correct tracking of gold particles. Trajectories of gold particles obtained from SPT videos were analyzed for both transient confinement and transient anchorage. The detection for TCZs was performed as described previously (). Transient anchorage detection software was designed to detect regions of trajectories where no displacement within experimental error occurred, as defined by all aspects of measurement stability. By recording videos of gold particles firmly glued on coverslips using nail polish and then detecting particle centroids over the time recorded, it was determined that 95% of centroids fell within ±25 nm along both the x and y axis (Fig. S1, available at ). A program was developed to recognize fragments from a trajectory as potential transient anchorage events when the displacement in successive frames was <25 nm in both the x and y dimensions. To qualify as transient anchorage, this condition must have persisted for more than four video frames (132 ms). Note that transient anchorage differs from immobilization, in which the particle does not move during the experimental observation time of a particular cell. To further characterize transient anchorage, we defined the RAT as the sum of durations in which a single particle is transiently anchored/total time of trajectory. To avoid the overestimation of total confinement given that transient anchorage sometimes overlaps with TCZs, transient anchorage segments were removed from the trajectories, and the remaining trajectory fragments were concatenated before TCZ detection. We calculated this estimate using the following considerations. Gold particles almost appear to float on the membrane and can diffuse rapidly. This suggests that they are not nearly enveloped by the plasma membrane. The degree to which they can bend the membrane around them will determine the maximum number of membrane antigens the preassembled complex can bind. The bending modulus determines how facilely the membrane can be molded. However, the bending modulus will depend on the length scale examined. For example, if the particle is over an actin or spectrin filament, bending will not be as easy as if the particle is in the region between filaments. The maximum bending may be estimated in the following way. Let us say the compartment dimension of the cytoskeleton meshwork underlying the membrane is ∼100 nm. As shown in Fig. S2 a, taking the thickness of a single layer of antibodies to be 10 nm (), the overall preassembled complex diameter would maximally be 100 nm. For the sake of argument, say each membrane (bilayer plus peripheral proteins) is ∼10 nm thick, so the gold particle complex in which a membrane partially wraps it has a radius of 60 nm. The furthest the membrane-wrapped particle could sink between adjacent filaments (Fig. S2 a) would result in the membrane covering the gold–antibody complex with a coverage area (CA), given by the equationwhere α represents the half angle corresponding to the contact area and r represents the radius of the whole complex (Fig. S2 a). For the geometrical situation depicted in Fig. S2 a, α ≈ 56°. In , the number of available anti–Thy-1 antibodies on the particle was multiplied by two because of the bivalency of antibodies (total number bound = 600). Given that r = 60 nm and 1 − cosα ≈ 0.45, the maximum number of bound Thy-1 was calculated as being ≤135. Considering the possible inactivation by partial denaturation of antibiotin antibodies during the coating process on gold particles, the steric and orientation considerations preventing antibody binding, and whether the particle actually sinks to the extent postulated in calculation, the actual number of Thy-1 bound to the preassembled complexes could be much less. In equilibrium, the free energy gained from the antibody–antigen interaction should be greater than or equal to the energy required to bend the membrane to conform the membrane to the particle (E). To bend an area of the membrane with an effective radius of curvature of 60 nm will require an energy given bywhere kc represents the bending modulus of the cell membrane, r is the radius of curvature, and the area is given by the calculation in (). The local bending modulus will most likely lie somewhere between that for a bilayer, in which kc ≈ 2 8 kT at 37°C (), and that for the red cell membrane, in which kc ≈ 23kT (). This leads to E values ranging from 40 kT, if pure bilayer intervened between the cytoskeletal filaments (), to 33 kT, if the red cell membrane modulus is more appropriate. The energy released from antibody–antigen bond formation ranges from 11 to 22 kT at 37°C, corresponding to K values from 10 and 10 M (). Therefore, the free energy released from only four or fewer antibody–antigen bonds would be sufficient to bend the membrane to the extent required by the aforementioned extreme model. While this paper was being reviewed, a study concluded that so much energy is gained from ligand–receptor bonds that the plasma membrane would completely envelope a 100-nm–diameter HIV particle (). Our calculation assumes that the complete envelopment of a coated 40-nm gold particle would be frustrated by the presence of immediately subjacent cytoskeletal filaments. Then, between transient anchorage points, gold particles also diffuse rapidly, which seems unlikely if they were completely enveloped. In (a–d), we used a microscope (IX-81; Olympus) with a 100× NA 1.25 oil immersion objective (Olympus). FluorSave reagent (Calbiochem) was used as the imaging medium, and AlexaFluor488 (, a and c) and Texas red (, b and d) were used as the fluorochromes. Images were captured with a dual-mode cooled CCD camera (C4880; Hamamatsu) and MetaMorph image acquisition software (Molecular Devices). Photoshop (Adobe) was used to crop the whole cell images into small fractions. Fig. S1 shows determination positional noise in the SPT system. Fig. S2 shows that preassembled complexes bound to ≤135 GPIAPs still demonstrate transient anchorage. Video 1 shows that the gold-bound cross-linked Thy-1 cluster is diffusing on the C3H membrane with transient anchorage. Online supplemental material is available at .

During angiogenesis, a complex coordination of cues from cytokines, growth factors, proteinases, and integrins mediate cellular changes to control the processes of sprouting, lumen formation, and proliferation (; ; ). Once networks of endothelial cell (EC)–lined tubes are formed, the stabilization of these structures is regulated by support cells such as pericytes (; ; ). In PDGF-B and -β receptor knockout mice, the lack of pericyte recruitment results in vascular instability and embryonic lethality (; ; , ; ). A molecular understanding of how pericyte–EC interactions lead to EC tube stability is not well understood and is an emerging field in vascular biology (; ; ). Matrix metalloproteinases (MMPs) regulate many biological processes, including ECM degradation, proteolysis of cell surface proteins, proteinase zymogen activation, liberation of growth factors, and regulation of tissue morphogenesis (; ; ), which includes vascularization (; ). Membrane-type (MT) MMPs but not soluble MMPs have been shown to play a critical role in cellular invasion through 3D matrices by degrading ECM proteins at the cell surface–ECM interface while maintaining the integrity of the surrounding ECM scaffold (, ; ; ; ). MMPs are controlled by various inhibitors, including tissue inhibitor of metalloproteinases-1–4 (TIMPs-1–4; ). TIMPs have been shown to regulate angiogenesis, wound repair, and tumor metastasis (; ; ; ; ), and a balance of MMPs and TIMPs appears to be critical during these events. Interestingly, MMPs appear to contribute to tissue regression in the mammary gland (), vasculature (; ; ), and during the menstrual cycle (). In this study, we present the novel concept that EC-derived TIMP-2 and pericyte-derived TIMP-3 coregulate capillary tube stabilization by the inhibition of key EC targets such as MT1-MMP, ADAM-15 (a disintegrin and metalloproteinase-15), MMP-1, and MMP-10, which normally control EC tube formation and/or regression. Using an in vitro model of angiogenic sprouting, human ECs invade ∼500 μm into 3D collagen matrices over a 48-h period (). This invasion response is completely inhibited by TIMP-2 and -3 () but not by TIMP-1. Although control and TIMP-1–treated invading ECs form lumenal structures, no lumen formation is seen from TIMP-2– or -3–treated invading ECs (). Similar results using ECs transfected with lentiviral vectors expressing control GFP, TIMP-1, or TIMP-3 were observed (Fig. S1 A, available at ). Time-lapse videos were performed with ECs suspended as individual cells within collagen matrices (vasculogenesis-like assay; ) over a 48-h period and treated with TIMP-1, -2, or -3 versus the control (Videos 1–4, available at ). Photographs from these videos are shown at 48 h of culture (). ECs treated with TIMP-2 and -3 fail to form tubes (i.e., no lumen structures observed) and instead send out fine processes over this time course (, arrowheads; and Videos 1–4). Addition of the chemical MMP inhibitor GM6001 shows essentially identical results, suggesting that these effects are likely the result of proteinase inhibition (Video 5). In contrast, obvious lumenal and tube structures are observed in the control and TIMP-1–treated cultures (). Quantitation of the sprouting assay as well as quantitation of the vasculogenesis assay reveals marked blockade of invasion and lumen formation by both TIMP-2 and -3 (). These results demonstrate that both TIMP-2 and -3 completely inhibit EC invasion and lumen formation (perhaps indicating a common molecular target), whereas TIMP-1 has no effect. To determine the ability of TIMP-3 to affect vessel formation in vivo, mouse embryo cultures were established from embryonic day (E) 7.5 to E9.5 in the presence of TIMP-1, TIMP-3, GM6001, and DMSO control (). Untreated, DMSO, and TIMP-1–treated cultures were photographed live at 48 h and exhibited the remodeling of capillary networks, large vitelline vessels at the origin in the embryo, and large vessels leading to capillaries. The large vitelline vessels (, insets) along with capillary networks stained positively for vascular endothelial (VE) cadherin. TIMP-3– and GM6001–treated embryos lacked vitelline vessels, exhibited an immature capillary plexus, and failed to turn. No large vessel structures were formed in GM6001– (unpublished data) or TIMP-3–treated explants as compared with the untreated or TIMP-1–treated explants. These data indicate that TIMP-3 treatment blocked the assembly of vascular structures in vivo, which directly correlated with our in vitro findings. RT-PCR analysis of embryonic region (E6.5 and E7.5), extraembryonic region (6.5× and 7.5×), and yolk sac (8.5 and 9.5 d) tissues during vascular development revealed that TIMP-3 was expressed abundantly in both embryonic and extraembryonic tissue as well as by the yolk sac at the time points indicated (). TIMP-3 showed increased expression up to E8.5, whereas TIMP-1 expression decreased over this time period. Immunofluorescence imaging of visceral endoderm and mesoderm sections of yolk sac at E8.5 and E9.5 revealed the colocalization of TIMP-3 with angiopoietin-1 in perivascular mesenchymal cells (pericyte and vascular smooth muscle precursors), which surround the early vascular tubes (). Also, strong staining for TIMP-3 was observed in the visceral endoderm, an important supporting structure that controls development of the early vasculature (). These staining distributions were confirmed by in situ hybridization (unpublished data). Thus, the visceral endoderm and mesenchyme surrounding EC tubes (stained with VE-cadherin; ) in the mesoderm may serve to control and stabilize EC tubular morphogenesis by delivering TIMP-3 to newly formed vascular tubes. Because TIMP-3 completely inhibits EC tube morphogenesis, we asked whether TIMP-3 could induce the collapse and regression of preexisting networks of capillary tubes. A previous study revealed that α2-integrin–blocking antibodies interfere with the formation of tubes but also cause the rapid collapse (within several hours) of previously formed tube networks (). EC cultures were treated at different times with TIMP-1, -3, or blocking antibodies directed to the α2- and α5-integrin subunits (Fig. S2, available at ). The treatment of EC cultures from 0–8 or 0–24 h with α2-integrin–blocking antibodies or TIMP-3 strongly inhibited intracellular vacuole and lumen formation (Fig. S2, A and B). TIMP-1 addition or blocking antibodies directed to the α5-integrin subunit had no effect and were identical to the control. The addition of these agents to preestablished EC networks (48 h) revealed that although α2-integrin–blocking antibodies completely collapsed lumenal structures and tubular networks, TIMP-3 did not have this effect (Fig. S2 C). From these data, we hypothesized that TIMP-3 may serve as a stabilizing signal by preventing further morphogenic events and blocking the natural tendency of EC tubes to regress. A time-lapse experiment shown in Fig. S3 supports this conclusion in that TIMP-3 addition to preformed tubes does not cause collapse but retards further morphogenic progression, whereas TIMP-1 addition had no effect on either process. We also tested the ability of TIMP-3 to block proteinase-mediated regression of endothelial networks. We used a previously established model for EC network regression and collagen gel contraction (; ) in which ECs were allowed to assemble into tube networks for 24 h before plasminogen (Plg) addition to activate pro–MMP-1 and -10. Cultures were either left untreated (control) or were treated with TIMP-1 or -3. After 48 h, cultures were fixed, and the percentage of collagen gel contraction was quantitated (collagen gel contraction represents the end stage of capillary tubular network collapse after proteolysis of the collagen gel matrix scaffold). These data are shown in Fig. S2 D, and they support the concept that TIMP-3 functions to stabilize and prevent the regression of EC networks. TIMP-1 also inhibits regression events that are regulated by MMP-1 and -10, which is consistent with our previous observations (; ). TIMP-2 and GM6001 show essentially identical effects compared with TIMP-3 in these latter assays (unpublished data). Pericytes are known to stabilize the developing or wound vasculature (). We hypothesized that perhaps pericytes deliver TIMP-3 to EC tubes to stabilize them. To evaluate TIMP-3 production by vascular cells, including EC, bovine retinal pericytes (BRPs), and coronary artery smooth muscle cells (CASMCs), cells were cultured on plastic substrates and treated with or without TGF-β for 24 h (Fig. S1 B). Western blot analyses of cell lysates revealed that TIMP-3 was not produced by ECs. However, marked production of TIMP-3 was observed from pericytes and to a lesser degree by CASMCs. Because pericytes appear to be a strong source of TIMP-3, 3D cocultures were established with ECs and pericytes at varying pericyte/EC ratios using constant levels of ECs (). In addition, we performed the reverse experiment whereby pericytes were added at a constant level and ECs were added at varying levels (). As shown in , increasing the levels of pericytes in the presence of constant ECs or increasing the levels of ECs with constant pericytes leads to a strong increase in TIMP-3 production. Interestingly, TIMP-3 is strongly produced by pericytes on plastic dishes (Fig. S1 B), but this production markedly decreases when they are cultured alone in 3D collagen matrices (). Importantly, the coculture of ECs with pericytes strongly induces TIMP-3 production by pericytes (). In contrast, it is clear that ECs are by far the predominant source of TIMP-2, TIMP-1, and PAI-1 in this 3D model. Little production of these factors is seen from pericytes alone, whereas the strong production of each of these factors is observed with ECs alone. To evaluate this conclusion further, siRNA suppression experiments were performed using both ECs and pericytes in 3D culture to assess which cell type is producing each proteinase inhibitor. As shown in , siRNAs directed to these proteins show that ECs are the predominant source of TIMP-2, TIMP-1, and PAI-1, whereas pericytes are the source of TIMP-3. Thus, it appears that TIMP-3 production is markedly regulated by EC–pericyte interactions, whereas the other proteinase inhibitors do not appear to be altered. Interestingly, pro–MMP-1 levels, which control capillary tube regression, are progressively decreased with increasing levels of pericytes in the cocultures (). In contrast, pro–MMP-10 levels do not change under the same conditions. To address whether pericytes are able to contribute to capillary tube stabilization through the blockade of MMP-1– and -10–dependent tube regression, cocultures were prepared, and quantitative analysis of tube regression over time was examined. To illustrate the coculture system, GFP-labeled ECs were cocultured with mRFP-labeled () pericytes, which clearly shows the ability of pericytes to associate with developing EC tubes (). As shown in , the addition of pericytes induced a strong delay in regression until the addition of 20 or 30% pericytes relative to ECs where tube regression is completely inhibited. To examine whether the tube regression event correlates with pro–MMP-1 or -10 activation, Western blots were performed (). The addition of 30% pericytes completely blocks MMP-1 and -10 activation in response to Plg addition, indicating that proteinase inhibition is the most likely mechanism by which pericytes interfere with the capillary tube regression response (). To address whether pericytes were unique in their ability to block EC tube regression, 30% human fibroblasts or CASMCs were added, and neither cell type was able to block the regression response like the pericytes at this cell/EC percentage (). To address the role of the different TIMPs and PAI-1 in this coculture model of pericyte-induced capillary tube stabilization, siRNA suppression was used. As shown in , siRNA suppression of specific TIMPs in ECs versus pericytes was observed relative to an actin control. In this experiment, ECs were treated with luciferase control, TIMP-1, or -2 siRNAs, whereas pericytes were treated with luciferase control or TIMP-3 siRNA. The different cell types were mixed in different combinations and were cultured in 3D collagen matrices at two pericyte/EC ratios of 30 or 50% pericytes relative to ECs (which previously completely inhibited capillary tube regression; ). As shown in , at two different time points (3 or 6 d), siRNA suppression of EC TIMP-2 and pericyte TIMP-3 resulted in tube regression responses of the EC–pericyte cocultures. At later times of culture at 6 d, the suppression of EC TIMP-2 alone resulted in a tube regression response in a coculture containing 30% pericytes. At 3 d, the combination of EC TIMP-2 and pericyte TIMP-3 strongly induced capillary tube regression. In 50% pericyte cocultures, the only combination of siRNAs that affected tube regression was EC TIMP-2 combined with pericyte TIMP-3. This data strongly argues that both EC-derived TIMP-2 and pericyte-derived TIMP-3 are required for EC–pericyte cocultures to stabilize capillary tubes. A key question in our studies is what molecular targets are blocked by TIMP-2 and -3 that interfere with both tube formation () as well as the tube regression response (). To address this issue, we performed experiments with a series of potential targets of these TIMPs using an siRNA suppression approach. As shown in , we have suppressed the protein expression of a variety of MT-MMPs as well as ADAM proteinases and performed tube formation experiments using the vasculogenic assay system. Western blots show strong protein suppression of key enzymes (along with actin control) with biological activity in our system ( and ). There was complete specificity in this suppression in that only the indicated siRNA suppressed the expression of the appropriate protein and had no influence on any of the other proteinases using Western blot analysis. The MT1-MMP siRNA is shown to markedly block tube and lumen formation compared with luciferase control and MT3-MMP siRNAs (). Interestingly, the morphology of MT1-MMP siRNA–treated cells looks remarkably similar to ECs treated with TIMP-2 or -3 as shown in and in Videos 1–5. MT1-MMP siRNA–treated ECs extend fine processes into the 3D collagen matrices but fail to form lumenal structures or tube networks. Quantitation of these results reveals that only MT1-MMP siRNA had a strong blocking influence (). In contrast, no other siRNA had a blocking influence, and ADAM-9 siRNA actually stimulated the EC lumen formation process, suggesting that it may normally be an inhibitor of these events. Overall, these data strongly show that the likely key target of TIMP-2 and -3 in vasculogenesis assays is MT1-MMP, which is known to be blocked by these two TIMPs (). The inability of TIMP-1 to block in these assays also supports this conclusion because it has no ability to block MT1-MMP (). To further confirm this conclusion, we performed ELISA capture assays using GFP epitope-tagged MT1-MMP or TIMP-3 to demonstrate binding interactions between the two proteins. As shown in , the strong binding of TIMP-2 and -3 is detected to GFP epitope-tagged MT1-MMP by capturing the complex on anti-GFP–coated wells in HEK293 cell cotransfection experiments. The reverse experiment also shows the binding of MT1-MMP to TIMP-3 in ECs when a TIMP-3–GFP fusion protein is expressed in ECs using a recombinant adenovirus. TIMP-3–GFP is captured onto wells, and the wells were probed with anti–MT1-MMP antibodies to reveal the binding interaction (). To further analyze the influence of these metalloproteinase siRNAs, we performed two additional assays that mimic the process of angiogenesis whereby ECs invade into collagen matrices from a matrix surface (; ). For these assays, we used two distinct stimuli to induce EC invasion and tube morphogenesis in 3D collagen matrices. We previously reported that sphingosine-1 phosphate (S1P) induces EC invasion (), and, here, we report for the first time that the chemokine stromal-derived factor-1α (SDF-1α) strongly induces invasion and tube morphogenesis in our system when it is incorporated into a 3D collagen matrix. This latter chemokine is known to synergize with VEGF in both developmental and pathological vascular morphogenic events (; ). It may represent a more physiologically relevant stimulus than S1P in this type of assay system. As shown in , MT1-MMP siRNA dramatically blocks the invasion stimulated by either S1P or SDF-1α. This is shown quantitatively in (B and C) and through photographs in Figs. S4 and S5 (available at ). Interestingly, MT2-MMP siRNA also inhibits SDF-1α invasion while partially inhibiting S1P invasion. Interestingly, ADAM-15 siRNA suppresses invasion selectively with SDF-1α–induced responses. ADAM-9 siRNA stimulates invasion responses like the aforementioned vasculogenesis assay, whereas ADAM-17 siRNA partially stimulates the SDF-1α invasion response. Collectively, these data show that MT1-MMP is required for EC tube formation in vasculogenic assays as well as in invasion and tube morphogenesis in response to two strong invasion-promoting stimuli, SDF-1α and S1P. Both MT2-MMP and ADAM-15 appear to play a lesser but detectable role selectively in SDF-1α–induced EC invasion responses. To address the role of various metalloproteinases in the capillary tube regression response, we performed regression time courses with the siRNA-treated ECs over time (). Marked delays in the regression response occurred with three siRNA treatments: MMP-1, MMP-10, and ADAM-15 (). Previously, we reported that MMP-1 and -10 both control capillary tube regression (). ADAM-15, a known TIMP-3 target, appears to play a strong role in this process as well. A partial delay occurred with MT1-MMP–treated cells, which may relate to the fact that these cells fail to undergo substantial tube formation. EC tube formation is not blocked with MMP-1–, MMP-10–, or ADAM-15–treated cells. Interestingly, ADAM-9– and -17–treated ECs show accelerated regression responses compared with control luciferase as well as ADAM-10 and -12 treatments. To address how these treatments affect pro–MMP-1 or -10 activation, Western blots were performed (). MMP-1 and -10 siRNA knockdown shows marked suppression of the respective proteins, whereas ADAM-15 knockdown appears to suppress the overall levels of both pro–MMP-1 and -10 as well as the activated forms of these two enzymes, which appears to account for the regression delay. The mechanism by which ADAM-15 suppression affects MMP-1 or -10 levels and activation is unclear. Interestingly, ADAM-9 and -17 suppression appears to accelerate the regression response by affecting MMP-1 activation, whereby a greater proportion of activated pro–MMP-1 is observed in both cases. Overall, these data suggest that three primary enzymes appear to play a central, required role in the MMP-dependent regression response (MMP-1, MMP-10, and ADAM-15). These metalloproteinases are all blocked by TIMP-3, whereas MMP-1 and -10 are known to be blocked by TIMP-2. To further confirm that TIMP-3 acts as a stabilizing factor through blocking metalloproteinase targets, we performed site-directed mutagenesis of TIMP-3 using known mutations that interfere with its proteinase inhibitory activity (; ). We mutated either cysteines 1, 3, or both and made lentiviral vectors that express these constructs in comparison with TIMP-1. Stable EC cell lines expressing these recombinant proteins were selected, and each of the respective recombinant proteins were produced at appropriate levels (). To address whether the recombinant proteins influence tube formation or regression, we performed various biologic assays (). As shown in , TIMP-3 wild type is able to block lumen formation (), invasion responses (), and capillary tube regression (). In contrast, increased TIMP-1 expression fails to block tube formation or invasion but completely inhibits tube regression through its known ability to inhibit MMP-1– and -10–dependent proteolytic events (; ). The TIMP-3 mutants failed to block EC lumen formation or invasion, suggesting that the proteinase inhibitory function of TIMP-3 is required for the ability of TIMP-3 to stabilize tubes through interference with EC tube morphogenesis. Surprisingly, the TIMP-3 C1S mutant retained its ability to block tube regression responses, whereas the TIMP-3 C3S mutant or C1SC3S double mutant did not. Overall, these data demonstrate that the proteinase inhibitory function of TIMP-3 is central to its ability to stabilize tubes through the blockade of tube morphogenesis and tube regression responses. There is considerable interest in the mechanism by which pericytes stabilize capillary tubes during development and other vascular morphogenic events (; ; ). Much data support the concept that pericytes stabilize EC-lined tubes as a result of observations that tubes without associated pericytes are more susceptible to regression (; ). However, the molecular mechanisms that regulate these phenomena remain unclear. The data presented here propose a new concept in that pericyte-induced capillary tube stabilization can be mediated, in part, by proteinase inhibition (). Our work suggests that proteinase inhibition of both EC tube morphogenesis (leading to the cessation of morphogenesis) and tube regression is necessary to control tube stabilization. Previous data from our laboratory support the concept that ECs undergoing the lumen formation process and assembling into 3D networks induce a gene program to regulate tube formation but also establish the conditions necessary for tube regression (; ; ). In studies of EC network collapse of 3D collagen matrices, ECs secrete high levels of a latent collagenase, pro–MMP-1, as well as pro–MMP-10 during the assembly of 3D tubular networks (; ), which we have shown are proregression agents. In the presence of serine proteinases, MMP-1 and -10 zymogen activation occurs, and active MMP-1 controls collagen and ECM dissolution and collapse of capillary networks. This regression event is completely blocked by the presence of sufficient pericytes (i.e., to deliver TIMP-3; – ) or the addition of exogenous TIMP-3 or -2 ( and Fig. S2). These observations all support the concept that the regulation of proteinase activity during angiogenesis (e.g., during wound healing) must be tightly controlled to prevent the regression of newly formed vascular networks. A similar conclusion was reached in studies of a mouse knockout of the MMP inhibitor RECK. An embryonic lethal phenotype occurs at E10.5 secondary to excessive collagenase activity, vascular collapse, and hemorrhage (). Marked destruction of the collagen scaffold was observed in vivo and was associated with defects in vascular tube networks in a manner that was remarkably similar to our previous studies (; ). Our work here shows that distinct metalloproteinases, including members of the MMP and ADAM families, influence promorphogenic versus proregression events (). Furthermore, this study shows that EC-derived TIMP-2 and pericyte-derived TIMP-3 play direct roles in regulating these stabilizing effects by inhibiting specific MMPs and ADAMs. EC-derived molecular targets for TIMP-2 and -3 include MT1-MMP, which appears to be the major regulator of EC tube morphogenesis by controlling EC invasion and lumen formation. In addition to MT1-MMP, we show that MT2-MMP and ADAM-15 appear to contribute to tube morphogenesis by regulating the EC invasion of collagen matrices. Other targets include MMP-1, MMP-10, and ADAM-15, which act together to regulate capillary tube regression. TIMP-3 has previously been reported as an antiangiogenic agent and also inhibits the growth of solid tumors (; ; ). Interestingly, TIMP-3 is frequently down-regulated by various tumors through methylation of its promoter (). In this study, we show that TIMP-3 prevents EC tubular morphogenesis in 3D matrices in vitro as well as vasculogenesis and vascular remodeling in vivo ( and ). Importantly, TIMP-3 expression from pericytes was strongly induced by the presence of ECs () in 3D collagen matrices. We show that siRNA suppression of TIMP-3 in pericytes markedly diminishes their ability to stabilize EC-lined tubes that are susceptible to MMP-1–, MMP-10–, and ADAM-15–dependent tube regression (– ). This suppression was selective for pericytes because ECs in our system do not express TIMP-3, and this siRNA also had no effect when ECs were treated with it. Site-directed mutagenesis of TIMP-3 using treatments known to eliminate the proteinase activity of various TIMPs () interfered with its ability to block both tube formation and regression, showing that these activities depend on its proteinase inhibitory function. Pericytes produce TIMP-3 to inhibit EC MT1-MMP, MT2-MMP, and ADAM-15, which we have identified as important targets controlling tube morphogenesis. Previous studies have revealed important roles for MT1-MMP and ADAM-15 in vascular morphogenic events (; ). A final point concerns the known ability of TIMP-3 to strongly interact with ECM, such as heparan sulfate in vascular basement membrane matrices (; ). This unique ability of TIMP-3 among the TIMPs () is likely to be important in vascular stabilization because of the fact that it will be presented as a component of the basement membrane matrix in direct contact with endothelium and pericytes in stabilized vessels. Previous studies indicate that TIMP-2 can block angiogenesis through a nonproteinase inhibitory mechanism by binding the α3β1 integrin and activating signaling events leading to the inhibition of VEGF receptor-2 (VEGFR-2) signaling (; ). In this study, TIMP-2 appears to block in a similar manner to TIMP-3 () or to chemical inhibitors of surface metalloproteinases such as GM6001 (Videos 1–5). In our model, the inhibitory functions and vascular stabilizing functions of TIMP-2 appear to occur through its proteinase inhibitory function. Supporting this conclusion, the siRNA suppression of MT1-MMP expression, a key TIMP-2 and -3 target, markedly blocks EC tube formation (), and the morphology of MT1-MMP siRNA–treated ECs appears remarkably similar to ECs treated with either TIMP-2 or -3 in 3D collagen matrices ( and ). Furthermore, siRNA suppression of TIMP-2 expression in ECs can lead to destabilization and MMP-1–dependent regression of EC–pericyte cocultures (). Overall, our work with TIMP-2 as well as TIMP-3 supports the concept that both TIMPs play a role in the stabilization of newly formed vessels in 3D matrices. In addition, TIMP-3 (like TIMP-2) has been reported to inhibit VEGFR-2 signaling (). The inhibition of VEGFR-2 is a major therapeutic target in antiangiogenic protocols to inhibit excess vascularization in tumors and other contexts (). Interestingly, recent data suggest that treatment of angiogenic fields with VEGFR-2 antagonists results in the regression of select vessels followed by the normalization of vasculature, which is a step toward stabilization (, ). Thus, both TIMP-2 and -3 have been reported to regulate VEGFR-2 in addition to their proteinase inhibitory activity. Our work extends these observations in that TIMP-2 and -3 can act as vascular stabilizing agents by blocking distinct proteinases required for tube morphogenesis and regression (). Thus, this study, coupled with previous studies (; ), suggests that EC-derived TIMP-2 and pericyte-derived TIMP-3 act through several distinct mechanisms to facilitate vascular stabilization. Finally, we observe that ECs will continuously undergo morphogenesis and lumen formation in 3D matrices unless they are provided with a stabilization or stop signal. Without this signal, proteinase activities eventually result in the regression of EC networks. This concept is consistent with a proposed target hypothesis () whereby ECs (in a similar fashion to neurons) search for or recruit a target cell that provides a stabilizing signal. Pericytes appear to provide this target function for EC tube networks, which shifts the molecular balance from morphogenesis or regression toward tube stabilization. Purified human Plg was obtained from American Diagnostica, Inc. VEGF and bFGF were purchased from Upstate Biotechnology. Activated human plasmin was obtained from Calbiochem. Purified TIMP-1 (CC3328) and -2 (CC3327) were obtained from Chemicon. Recombinant human TIMP-3 (973-TM) and human SDF-1α (CXCL12) were obtained from R&D Systems, and the metalloproteinase inhibitor GM600 was obtained from Calbiochem. -erythro-S1P was obtained from Avanti Polar Lipids, Inc. A blocking antibody targeting α2 integrin (BHA2.1) was purchased from Chemicon, and a blocking antibody against α5 integrin (mAb 16) was purchased from Becton Dickinson. Antibodies targeting MMP-10 (mAb 9101), TIMP-1 (AF970), TIMP-2 (AF971), ADAM-9 (AF949), and ADAM-15 (AF945) were obtained from R&D Systems. Antibodies targeting ADAM-12 (AB 19133), ADAM-17 (AB 19027), and TIMP-3 (mAb 3318) were obtained from Chemicon. A polyclonal antibody against human MT1-MMP (RP1–MMP-14) was purchased from Triple Point Biologics. The PAI-1 polyclonal antibody, MMP-1 mAb (IM35L), and actin mAb (CP01) were purchased from Calbiochem. Oligonucleotide primers were obtained from Sigma-Genosys. All other reagents were purchased from Sigma-Aldrich unless otherwise noted. The pAdEasy adenoviral system was obtained from B. Vogelstein (Johns Hopkins Medical School, Baltimore, MD), and the mRFP construct was obtained from R. Tsien (University of California, San Diego, La Jolla, CA). Human umbilical vein ECs (HUVECs) were purchased from Cambrex and were used from passage 2–6. BRPs were cultured in DME containing 10% FBS in gelatin-coated flasks. Bovine eyes were obtained from the Rosenthal Center (Texas A&M University), transported on ice, and submerged in Betadine for 10 min before retinas were dissected and placed into sterile buffer. The isolation protocol was essentially as previously described (). At confluence, pericytes were screened for the expression of 3G5 antigen and α-smooth muscle actin. ECs were placed in 3D invasion assays as described previously (), in which ECs invade into 3.75 mg/ml of collagen matrices containing either 1 μM S1P or 200 ng/ml SDF-1α. Alternatively, lumen formation assays were used where individual ECs were suspended within 3.75 mg/ml of collagen type I matrices and allowed to assemble into 3D networks over time (). Cultures were fixed at indicated time points with 3% glutaraldehyde in PBS, pH 7.5, for at least 30 min before additional manipulation. In some cases, cultures were stained with 0.1% toluidine blue in 30% methanol and were destained before visualization and photography. Conditioned media were collected to examine differential protein expression at various time points. Conditioned media were run on SDS-PAGE gels, transferred to polyvinylidene membranes, probed, and developed. 3D collagen gels were also extracted in some cases to examine protein expression. Regression assays were performed as described previously (). For quantitative analysis of gel contraction, 15-μl mixtures of ECs and collagen matrix ( = 8 cultures per condition) were placed in 384-well tissue culture plates (VWR Scientific Products). Cultures were allowed to equilibrate for 30 min before the addition of media with or without the serine proteases Plg or plasmin. The addition of culture media denoted time zero, and cultures were monitored every 4 h for gel contraction. Upon initiation of gel contraction, gel area was recorded using an ocular grid, and percent collagen gel contraction was calculated as follows: [(original area − current gel area)/original area] × 100. When contraction was complete, cultures were fixed or extracted, and media was examined for MMP expression and activity. Data are reported as the mean percent gel contraction ± SD. Confluent ECs were resuspended in 3.75 mg/ml of 3D collagen matrices at a final concentration of 10 cells/ml. BRPs were incorporated into collagen matrices at varying ratios as indicated. EC or EC/PC cocultures (15-μl vol) were evaluated in EC tubular network formation or quantitative regression assays performed in 384-well plates as previously described (). Custom smart pool siRNAs for TIMP-3 and luciferase controls were purchased from Dharmacon and were resuspended (40 μM) in universal buffer and stored at −80°C. siGenome siRNAs targeting TIMP-1, -2, -4, and PAI-1 were obtained from Dharmacon and prepared. Transfection of ECs with siRNAs before use in experiments was performed essentially as described previously (). Transfection of BRPs with siRNAs was performed in a similar manner except that LipofectAMINE 2000 (Invitrogen) was used for transfection. Cells were transfected at a final siRNA concentration of 200 nM. CD1 wild-type mouse embryos were dissected at E7.5 and cultured for 48 h in a continuous gas (5% CO) rolling embryo culture apparatus. Media (0.5 ml/embryo) was changed after 24 h and contained high glucose (4.5 g/L) DME (Invitrogen) with 75% immediately centrifuged rat serum (Harlan) and glutamine-penicillin-streptomycin (Invitrogen). Recombinant TIMP-1 (Chemicon) and -3 (R&D Systems) were added to culture media at 5 μg/ml, and GM6001 (Calbiochem) was added at 5 μM. Embryos were photographed live before whole mount staining with VE-cadherin antibodies (BD Biosciences). RT-PCR analysis and immunohistochemistry were performed as previously described (; ). Primers used for this analysis were as follows: HPRT, upstream primer (UP) (GCTGGTGAAAAGGACCTCT) and downstream primer (DN) (CACAGGACTAGAACACCTGC); TIMP-1, UP (CGAATCAACGAGACCACCTT) and DN (TGCAGGCATTGATGTGAAAT); and TIMP-3, UP (CACGGAAGCCTCTGAAAGTC) and DN (CCCAAAATTGGAGAGCATGT). TIMP-1 was amplified from HUVEC cDNA () using the primers TIMP-1 HindIII UP (5′-AGAAGCTTCGAGATCCAGCGCCCAGAG-3′) and TIMP-1 EcoRV DN (5′-AGGATATCGCAGGCTTCAGTTCCACTC-3′). TIMP-3 was amplified from human placental cDNA (BD Biosciences and CLONTECH Laboratories, Inc.) using the primers TIMP-3 XhoI UP (5′-AGCTCGAGGCAATGACCCCTTGGCTCGGGCTCATC-3′) and TIMP-3 XbaI DN (5′-AGTCTAGACGCTCAGGGGTCTGTGGCATTG-3′). Both inserts were initially cloned into the pAdTrack-CMV vector using similar methods that we have described previously (). Positive clones were sequenced, and protein expression was verified via Western blotting. The production of recombinant lentivirus was accomplished using the pLenti6/V5 TOPO cloning system (Invitrogen). TIMP-1 was amplified from the pAdTrack-CMV vector using the primers TIMP-1 pLenti UP (5′-CACCATGGCCCCCTTTGAGCCCCTG-3′) and TIMP-1 pLenti DN (5′-AGGATATCGCAGGCTTCAGTTCCACTC-3′). TIMP-3 was amplified from the pAdTrack-CMV vector using the primers TIMP-3 pLenti UP (5′-CACCATGACCCCTTGGCTCGGGCTC-3′) and TIMP-3 pLenti DN (5′-AGTCTAGACGCTCAGGGGTCTGTGGCATTG-3′). Enhanced GFP was amplified from the pEGFP-N2 vector (CLONTECH Laboratories, Inc.) using the primers GFP pLenti UP (5′-CACCATGGTGAGCAAGGGCGAGG-3′) and GFP pLenti DN (5′-AGTCTAGATTACTTGTACAGCT-CGTCCATGC-3′). The sequence encoding a modified mRFP was amplified from the vector pRSETB () using the primers mRFP pLenti UP (5′-CACCATGGCCTCCTCCGAGGACGTC-3′) and mRFP pLenti DN (5′-AGTTAGGCGCCGGTGGAGTGGCG-3′). Once amplified, each insert was TOPO cloned into the pLenti6/V5 TOPO vector. After TOPO cloning, One Shot Stbl3 Chemically Competent (Invitrogen) were transformed with DNA according to the manufacturer's protocol. Individual clones containing TIMP-1, TIMP-3, GFP, and mRFP inserts were verified via PCR and sequence analysis. TIMP-3 pAdTrack-CMV was used as a template for the amplification of TIMP-3 via standard PCR for cloning into the pEGFP-N2 fusion vector (CLONTECH Laboratories, Inc.) using the primers TIMP-3 N2 XhoI UP (5′-AGCTCGAGGCAATGACCCCTTGGCTCGGGCTCATC-3′) and TIMP-3 N2 BamHI DN (5′-AGGGATCCCGGGGTCTGTGGCATTGATGATG-3′). A positive TIMP-3–GFP clone was used as a template to amplify TIMP-3–GFP for standard cloning into the pAdShuttle-CMV adenoviral vector using the primers TIMP-3–GFP pAd Shuttle XhoI UP (5′-AGCTCGAGGCAATGACCCCTTGGCTCGGGCTCATC-3′) and TIMP-3–GFP pAd Shuttle EcoRV DN (5′-AGGATATCCGGGGTCTGTGGCATTGATGATG-3′). At each step, positive clones were verified using PCR, restriction digestion, and sequence analysis. Additionally, clones were transfected, and visualization of GFP fusion proteins was verified via fluorescent microscopy, transfection of plasmids into HEK293 cells, and Western blot analysis of TIMP-3 and TIMP-3–GFP proteins. Recombinant adenoviruses encoding TIMP-3 and TIMP-3–GFP were generated from expression plasmids TIMP-3 pAdTrack-CMV and TIMP-3–GFP pAdShuttle-CMV via recombination with the pAdEasy adenoviral vector as described previously (). A mAb (3E6) against GFP used to capture GFP fusion proteins was obtained from Qbiogene. Polystyrene microwell plates (VWR Scientific Products) were precoated with or without 2 μg/ml anti-GFP antibody in PBS overnight at 4°C. The following morning, plates were warmed to room temperature, aspirated to remove residual GFP antibody, and blocked in Tris–Tween 20 detergent solution for 20 min. HEK293 cell cotransfection lysates were diluted 1:2 with Tris–Tween 20 containing 1% BSA and incubated at 100 μl/well vol for 2 h at room temperature. Lysates were aspirated, and plates were washed four times in Tris–Tween 20 before incubation with polyclonal antibodies (50 μl/well) against TIMP-2 or -3 (R&D Systems) diluted at 1:200 and 1:100, respectively, in Tris–Tween 20 and 1% BSA. In separate wells, a control antibody targeting MT1-MMP (Triple Point Biologics) was used to ensure the proper capture of MT1-MMP–GFP and not MT1-MMP pAdTrack. Additionally, conditioned media and lysates were assessed via Western blotting to ensure the proper protein expression of each transfection condition. Primary antibodies were incubated for 1 h at room temperature, aspirated, and plates were washed four times before the addition of HRP-conjugated secondary antibodies (50 μl/well; DakoCytomation) diluted 1:2,000 in Tris–Tween 20 and 1% BSA. Secondary antibodies were incubated for 30 min at room temperature before washing three times in Tris–Tween 20 and were washed twice in ddHO. ELISA plates were developed using a typical -phenylenediamine substrate solution (100 μl/well) and stopped with 100 μl/well of 1 M sulfuric acid before the measurement of absorbance at 490 nm. Data are reported as the mean absorbance for duplicate wells and have been background corrected against control wells, which were not precoated with the capturing GFP mAb. HUVECs were cultured to confluence in T25 flasks. Viral transformation of ECs with recombinant adenoviruses encoding TIMP-3 or TIMP-3–GFP fusion was performed as described previously (). The following morning, transformation efficiency was assessed using fluorescent microscopy. Flasks were washed twice with 3 ml of warm M199, and detergent lysates of TIMP-3– and TIMP-3–GFP-producing ECs were prepared using 1.5 ml of cold lysis buffer per T25 flask. The ELISA capture technique was performed as described above except that for HUVEC lysates, TIMP-3–GFP was captured, and lysates were probed for the EC-derived target MT1-MMP. The MT1-MMP polyclonal antibody (Triple Point Biologics) was used at a 1:200 dilution. Data are reported as the mean absorbance for duplicate wells and have been background corrected against control wells, which were not precoated with the capturing GFP mAb. A fluorescence inverted microscope (Eclipse TE2000-U; Nikon) was used to visualize EC invasion, vasculogenesis, and regression as well as immunofluorescence microscopy. Lenses used for time lapse included a CFI plan-Fluor 20× with an NA of 0.45, whereas for immunofluorescence, a CFI plan-Apo 60× oil objective with an NA of 1.4 was used. FITC and rhodamine were used for immunofluorescence, whereas GFP and mRFP were used for still photographs of EC–pericyte cocultures. Imaging media included Slow Fade (Invitrogen) for immunofluorescence and M199 culture media without phenol red for all time lapses of EC cultures and still photography of EC–pericyte cultures. Imaging of cocultures used an inverted microscope (CKX41; Olympus) and 20× NA 0.45 Luc plan FLN or 40× NA 0.6 Luc Plan FL objectives. Images were acquired using a camera (DP70; Olympus) and DP manager software version 2.1.1.163 (Olympus). Time-lapse imaging of living cells used the Eclipse microscope and was performed with a temperature-controlled chamber (Solent Scientific) set to 37°C with continuous flow of 5% CO. Time-lapse two-photon imaging was performed at the lowest possible excitation levels. Images were captured every 10 min in single z planes with a monochromatic camera (CoolSNAP HQ; Photometrics) and a 6.45 × 6.45-μm pixel pitch (Photometrics) using MetaMorph software (Molecular Devices). After image acquisition, stacks of each stage position were assembled using MetaMorph software. Statistical analysis of selected EC invasion, vasculogenesis, and regression data was performed using SPSS 11.0 software (SPSS, Inc.). Analysis of variance was used to compare means when analyzing more than two groups. The Tukey's post-hoc test was used to determine groups of similarity. Statistical significance was set at P < 0.01. Unpaired tests were used when analyzing two groups within a single experiment. Fig. S1 shows that EC invasion of 3D collagen matrices is markedly inhibited by ECs expressing TIMP-3 but not by ECs expressing TIMP-1 and that pericytes are an abundant source of TIMP-3. Fig. S2 shows that TIMP-3 stabilizes ECs undergoing EC tubular network formation or regression, whereas anti–α2-integrin–blocking antibodies cause the rapid collapse of EC tubular networks. Fig. S3 shows that TIMP-3 blocks further lumen development and network remodeling when added to preexisting EC tubular networks without causing the collapse of these structures. Fig. S4 shows that the suppression of membrane-associated metalloproteinases via siRNAs in ECs indicates a critical role for MT1-MMP in the S1P-mediated invasion of 3D collagen matrices. Fig. S5 shows that the suppression of membrane-associated metalloproteinases via siRNAs in ECs indicates a critical role for MT1-MMP, MT2-MMP, and ADAM-15 in the SDF-1α–mediated invasion of 3D collagen matrices. Video 1 is a time-lapse video of control EC tube morphogenesis in 3D collagen matrices. Video 2 is a time-lapse video of normal EC tube morphogenesis in the presence of TIMP-1 in 3D collagen matrices. Videos 3 and 4 are time-lapse videos of the blockade of EC tube morphogenesis by TIMP-2 and -3, respectively, in 3D collagen matrices. Video 5 is a time-lapse video of the blockade of EC tube morphogenesis by the MMP inhibitor GM6001 in 3D collagen matrices. Online supplemental material is available at .

In this study, we investigate the localization of microtubule (MT) minus ends within the meiotic spindle. This localization is important because it may reflect the location of MT nucleation within the spindle. Since the discovery of centrosomes, models for the assembly and maintenance of mitotic and meiotic spindles have included a dominant role for spindle poles as MT nucleation centers (; ). The “search-and-capture” model suggested that poles dominate spindle morphogenesis, anchoring the minus ends of MTs, whereas plus ends polymerize and depolymerize until some are stabilized by kinetochores (). Later models proposed that MTs could be stabilized “at a distance” by chromosomes, presumably via diffusible factors such as RanGTP (; ; ). In anastral spindles, which are typified by oocyte/egg meiotic spindles, centrosomes are unnecessary for spindle morphogenesis (). Spindles assemble in an “inside-out” manner, with initial formation of MTs near chromatin, followed by condensation of minus ends into poles (; ; ; ). In some meiotic spindles, density tapers off toward the poles in a manner suggesting that many MTs terminate before reaching the poles (). Studies in X. laevis egg extracts, which recapitulate assembly of the anastral meiosis II spindle, show that chromosomes trigger an exchange of GTP on Ran, promoting MT nucleation in the absence of centrosomes, thereby probably explaining early steps in spindle assembly (for review see ). Continued production of RanGTP is also required for maintenance of the metaphase steady-state in anastral spindles (; unpublished data), but it is unknown whether this is caused by stabilization or nucleation activity downstream. Steady-state anastral spindles might be dominated by nucleation at chromatin, like during assembly, or at poles assembled in response to Ran activation (; ). Knowing the localization of nucleating sites is, thus, central to understanding spindle morphogenesis. The search-and-capture picture is based on spatial separation between nucleating and stabilizing centers, and new models would be required to account for morphogenesis by other mechanisms. To this end, we sought to measure the localization of minus ends within the spindle. Previous work localized minus and plus ends using serial-section electron microscopy (; ; ), but this method is difficult to apply to large spindles and lacks reliable markers for end polarity. MT ends nearest to centrosomes were assumed to be minus ends (; ; ), which is an unreliable criterion if MTs are nucleated throughout the spindle. “Hook decoration” (; ) allows identification of polarity, but is unsuitable for localizing ends because MTs elongate under the hook decoration conditions. γ-Tubulin complex is probably involved in nucleation, but our knowledge of its function is limited, so we cannot equate its localization with that of minus ends. NuMA and other spindle pole proteins probably move to the most distal minus ends in the spindle via dynein-mediated transport (). Instead of using any of these to locate minus ends, we developed a quantitative optical method combining analysis of oriented MT distributions with localization of plus ends by tubulin incorporation. Our analysis shows that MT minus ends are present everywhere in the spindle, with a minimum density near the chromosomes. Our method to calculate the density of plus and minus ends at a single location within a extract spindle is shown in . Although we could not directly measure the density of minus ends, we could calculate the density of plus ends and the difference between the densities of plus and minus ends. The sum of these two quantities was the density of minus ends. Our technique required three steps. First, to obtain the end densities, we observed the flow of MTs in a portion of the spindle (, i [dashed box]). The amounts of leftward and rightward flow were proportional to the local numbers of MTs with their minus ends toward each pole. Second, we looked at how the numbers of MTs varied in space to find the difference between the local densities of minus versus plus ends (). Third, we measured the local density of plus ends by observing incorporation of labeled tubulin into the spindle (). We summed the results of steps two and three to find the local density of minus ends. The process was repeated at many locations on the spindle-pole axis to find the spatial distributions of plus and minus ends. We prepared egg extracts and assembled spindles after one cycle of DNA replication (). We performed florescence speckle microscopy () using X-rhodamine–labeled tubulin (Invitrogen) at 25 μg/ml. Images were acquired at 20°C on a microscope (either E800 or 90i; Nikon) with 60×/1.4 NA or 100×/1.4 NA objectives (Plan Apo DIC; Nikon), immersion oil (Deltavision), and a cooled charge-coupled device camera (MicroMAX; Princeton Instruments [or ORCA-ER; Hamamatsu]) using Metamorph imaging software (Universal Imaging Corp.). 4–5 μl of spindle reactions were squashed under 18 × 18 mm coverslips and imaged by wide-field microscopy, with the focal plane in the middle of each spindle. We typically acquired 18 frames per spindle at 5-s intervals and 400-ms exposures. Each spindle was rotated to align its pole–pole axis with the x axis. Cross-correlations were calculated between sequential frames as a function of the x and y displacements Δx and Δy. These were averaged over the temporal sequence, as described in ; , ii). A profile of the resultant surface was calculated along a line, near parallel with the Δx axis, passing through the two peaks that represented leftward and rightward flow. To estimate the volumes of the two peaks, this profile was fit using Matlab (Mathworks) to a sum of two Gaussians plus a background term. The numbers of left- and right-pointing MTs were obtained from the integrated intensities under the peaks, up to an unknown proportionality constant (see Supplemental materials and methods). Cross-correlations were found for windows 22 pixels wide (∼3 μm), which were spaced every 1 μm along the length of the spindle, to obtain distributions of oriented MT number as a function of position (). Oriented MT number distributions were smoothed in Matlab using a 20-pixel-wide moving-average filter. Left- and right-end number differences were obtained from the derivatives, computed moving left to right for right-pointing MTs and right to left for left-pointing MTs. The left and right end number differences were summed and then divided by the total MT number at each point to obtain the fractional end density difference. We focused on the midplane of each spindle, where the mean angle of MTs in the z direction was minimal, to minimize the effects of MTs entering or departing the plane of focus. 3 μl of preformed spindles in extract that had been assembled with speckle-level X-rhodamine–labeled (red) tubulin were mixed on the slide with 2 μl of extract preequilibrated with 50 μg/ml green Alexa Fluor 488–labeled (Invitrogen) tubulin, squashed under a coverslip, and imaged as soon as possible (within 10–30 s) using a dry 40×/0.95 NA lens and ORCA-ER camera. After observation of incorporation of the green tubulin until near steady-state (∼3 min), the objective was switched to a 60×/1.4 NA oil lens, and a speckle sequence of the same spindle was recorded for use in calculating oriented MT distributions. The median intensity, calculated in a region outside the spindle, was subtracted from each frame. New tubulin incorporation was measured as the intensity, recorded as a function of position along the spindle-pole axis () and the time elapsed after mixing (), and summed along the direction perpendicular to the spindle-pole axis (y). At each point, the initial tubulin incorporation rate as a fraction of the final intensity was calculated from the increase of the intensity over the first 10 frames. This was divided by the published velocity of MT plus end growth, which was 10 μm/min (), and the fraction of growing plus ends, 0.75 (Supplementary materials and methods), to obtain the fractional plus end density (, dotted lines). The fractional plus end densities were multiplied by the total MT number found from the cross-correlation to obtain plus end density distributions in the same (arbitrary) units as the latter (, dotted lines.) For more information, see Supplemental materials and methods. At each point, the fractional end density difference was added to the fractional plus end density to obtain the fractional density of minus ends (, solid lines.) Minus end density distributions, in arbitrary units, were found by multiplying the fractional density by the total MT number as calculated from cross-correlation analysis (, solid lines.) To test our analysis, we created data using Matlab, simulating MT creation, growth, shrinkage, flux, and the addition of labeled tubulin at a given time point. The images produced were analyzed using the same methods as for real data to calculate the distributions of plus ends, fractional end fractional density differences, and the distribution of minus ends. The calculated distributions agreed well with the real distributions (Supplemental materials and methods; Fig. S2, available at ). The Supplemental materials and methods describe the calculation of dynamical cross-correlations, plus end density measurements and simulations, and internal consistency checks. Fig. S1 compares our cross-correlation method with the speckle-tracking method described in and frames from an image sequence of a spindle after green tubulin addition. Fig. S2 shows the results of the computer simulations to test the analysis techniques. Online supplemental material is available at .

Phosphatidylinositol 4,5-bisphosphate (PtdIns ) is the major phosphoinositide species in mammalian cells and has been associated with numerous molecular events critical for cellular signaling. is hydrolyzed by PLC enzymes to generate diacylglycerol and inositol 1,4,5-triphosphate, two pivotal second messengers (), and it is also converted by class I phosphoinositol 3-kinases to PtdIns (). directly interacts with several ion channels, transporters (; ), and actin binding proteins () and regulates enzymes such as PLC and PLD (; ). Several molecules within the receptor internalization machinery also contain inositide binding domains, but the exact lipid species that regulates them in the cell has not been firmly established (). It is a major challenge to understand how a single type of molecule is able to regulate so many processes simultaneously and perhaps independently within the plasma membrane (PM). is that it is difficult to manipulate phosphoinositide levels within the cells. formation (; ). in intact cells by expressing either phosphatidylinositol 4-phosphate 5-kinase or 5-phosphatase (5-ptase) enzymes (; ). levels initiate several trafficking and signaling events that will alter the disposition of the cells by the time the effects are analyzed (). This makes it difficult to draw firm conclusions regarding direct effect of the lipid on any single process. levels by a drug-inducible membrane targeting of a type IV 5-ptase enzyme (; ) based on the heterodimerization of the FRB (fragment of mammalian target of rapamycin [mTOR] that binds FKBP12) and FKBP12 (FK506 binding protein 12; ). In this approach, the phosphatase is fused to the FKBP12 protein, and upon addition of rapamycin (rapa; or an analogue that does not interact with endogenous mTOR protein) the enzyme rapidly translocates to the membrane where its binding partner, the FRB domain, is targeted. This method has been successfully used to manipulate small GTP binding proteins at the PM () and to study the effects of β-arrestin membrane recruitment (). levels and demonstrate how these manipulations affect selected processes that are regulated by this phosphoinositide species. shows the concept and the constructs used for rapa-induced targeting of the type IV 5-ptase to the PM. For membrane targeting of the FRB domain of mTOR, the palmitoylation sequence of the human GAP43 protein was used (). To follow their localization, the FRB protein was also tagged with either CFP or monomeric red fluorescent protein (mRFP). The 5-ptase (either full-length or only the 5-ptase domain) was mutated (C641A) to eliminate its C-terminal lipid modification and membrane targeting and was fused to FKBP12 and also tagged with mRFP (). A mutant form of FRB (T2098L of mTOR) that can be heterodimerized with FKBP with a rapa analogue (AP21967; rapalogue) that does not bind to endogenous mTOR has been recommended. However, because of its easier availability and faster action, we mostly used rapa and the wild-type FRB protein in the present studies. Nevertheless, the mutant FRB and the rapalogue have also been tested and their use is recommended for applications where rapa itself could affect the process being investigated. levels in the PM, these constructs were transfected together with the PLCδ1PH-GFP construct in COS-7 cells. levels. However, a small number of cells expressing high levels of the truncated 5-ptase domain showed no PLCδ1PH-GFP localization, indicating that high concentrations of the truncated enzyme could decrease lipid levels even without membrane targeting. 100 nM rapa caused rapid translocation of the fusion protein containing the 5-ptase domain to the PM causing a prompt and complete loss of PLCδ1PH-GFP localization in most cells (). Using the full-length phosphatase, however, caused only incomplete translocation of the PLCδ1PH-GFP reporter to the cytosol in some of the cells, and many cells showed no detectable change in PLCδ1PH-GFP localization in spite of efficient enzyme recruitment to the membrane (unpublished data). This important finding is consistent with the notion that a full-length enzyme contains regulatory regions that keep the enzyme activity under control. No changes were observed in PLCδ1PH-GFP distribution upon rapa-induced translocation of the mRFP-FKBP fusion protein that did not contain the 5-ptase. FRET measurements between the CFP- and YFP-tagged PLCδ1PH domain () used either in single cells (not shown) or in a population of COS-7 cells () have clearly demonstrated the lipid changes evoked by this approach. depletion on Ca signaling evoked via the endogenous P receptors in COS-7 cells. Cells were transfected with the PM-targeted FRB-CFP (or -mRFP) together with either the full-length or truncated 5-ptase mRFP-FKBP fusion construct for 1 d. The expression as well as the movements of the 5-ptase were monitored in the red channel simultaneously with single-cell cytoplasmic Ca ([Ca]) measurements with fura-2. Addition of 50 μM ATP evoked a Ca signal in many cells expressing the 5-ptase in the cytosol, but several cells expressing a high level of the phosphatase showed impaired response to ATP. shows averaged Ca recordings from single cells where the truncated 5-ptase domain was expressed and the cells showed a response to ATP. decrease. Notably, these cells failed to respond to a subsequent stimulation with another Ca-mobilizing agonist, lysophosphatidic acid. Translocation of the full-length 5-ptase with 100 nM rapa also caused a rapid inhibition of the ATP-induced Ca signal. levels than the initial response of Ca mobilization (). in the membrane (). To do this, the effect of lipid depletion on thapsigargin (Tg)-induced Ca response was examined. Tg depletes Ca stores by inhibition of the sarcoplasmic and ER Ca ATPase that keeps Ca stores filled and therefore activates Ca influx without the need for InsP. shows that the sustained [Ca] increase after Tg treatment was not affected by the same manipulations of PtdIns levels that eliminated the ATP-induced sustained Ca elevations. depletion interferes with the sustained generation of InsP rather than with the capacitative Ca influx mechanism itself. A more detailed analysis of the relationship between these mechanisms is currently under way. Several controls were used to rule out that the observed effects are caused by rapa itself or by the transfected constructs and/or their translocation to the membrane. First, the response of cells in the same field of view not expressing the phosphatase were monitored and found to show no change in response to rapa. Second, the Ca response of cells expressing both the targeting construct and mRFP-FKBP12 without the 5-ptase also showed no change in response to rapa addition (). for its activity (; ). dependence by monitoring either their activity in patch-clamp recordings using the whole-cell configuration or by following the [Ca] signal evoked via their activation by menthol. shows that human embryonic kidney 293 (HEK293) cells expressing TRPM8 channels respond to menthol stimulation with a large increase in an outwardly rectifying current. TRPM8 channel activation by menthol is also reflected in the rapid and sustained [Ca] elevation observed in transfected COS-7 () or HEK293 cells (not depicted). In cells also expressing the truncated FKBP12−5-ptase construct together with the PM-targeted FRB domain, addition of 100 nM rapa caused a prompt decrease in the menthol-induced current (). A rapid drop in [Ca] was also observed in COS-7 cells () and HEK293 cells (not depicted). None of these changes were observed when rapa was added to cells expressing the PM-targeted FRB and the FKBP12 construct without the phosphatase (). for their activity and can report on rapid changes in the level of this lipid in intact cells. Interestingly, in spite of the apparently complete inhibition of the menthol-induced membrane conductance, [Ca] did not return to baseline after rapa addition. This remaining [Ca] elevation observed in both cell types could be explained by a stimulated store-operated Ca entry pathway if functional TRPM8 channels in the ER release ER Ca in response to menthol, as suggested by a recent study (). This and other possibilities will require further analysis, as do questions on the role of the lipid in determining the menthol sensitivity and gating behavior of these channels. depletion affects internalization of the transferrin (Tf) and EGF receptors. regulates the recruitment of numerous proteins to the PM that are required for receptor endocytosis (). depletion. We also determined the uptake of the fluorescent analogues in cells in which rapa induced the translocation of an mRFP-FKBP12 construct without the phosphatase. As shown in , both Tf and EGF appeared in intracellular vesicular compartments in all cells regardless of their transfection with the constructs. This compartment corresponds to early and recycling endosomes in the case of Tf and to the multivesicular body/late endosomal compartment in the case of EGF (; ). with rapa addition completely prevented the uptake of either fluorescent cargo into the cells. These effects were not observed in rapa-treated cells that expressed the same constructs without the 5-ptase domain. A more quantitative assessment of this process was obtained by FACS analysis in the case of Tf. Here, the mean green fluorescent intensity of the cells (a measure of internalized Tf) in the population of cells expressing the red (5-ptase) construct showed the changes observed in the confocal pictures (). is not available in the PM. levels by themselves without the generation of second messengers can have multiple consequences on a wide range of cellular processes. or other inositol lipids in specific cellular processes. Although this approach has considerable potential, caution and the use of appropriate controls are essential to avoid possible artifacts. Numerous cellular processes are based on FKBP12 interactions, and overexpression of an FKBP12 construct could alter their properties. Similarly, rapa is an inhibitor of mTOR that can exert several effects on its own. This problem is alleviated with the use of the rapalogue that does not bind to the endogenous protein. Lastly, the targeting of the FRB by itself can have its own effects on selected cellular functions. However, if these possibilities are kept in mind this technique can permit further exploration of the complex regulatory features of phosphoinositides. AP21967 was obtained from Ariad Pharmaceuticals. Rapa and Tg were purchased from Calbiochem. Alexa 488−Tf and Alexa 488−EGF were obtained from Invitrogen. All other chemicals were purchased from Sigma-Aldrich and were of highest analytical grade. The PLCδ1PH-GFP construct and its color variants have been previously described (; ). For PM tethering, the N-terminal localization sequence (MLCCMRRTKQVEKNDDDQKI) of the human GAP43 (residues 1–20) was fused to the N terminus of the FRB domain of human mTOR1 (residues 2019–2114 amplified from a human EST available from GenBank/EMBL/DDBJ under accession no. ) through a short linker. To visualize the fusion protein, the construct was tagged with CFP or mRFP (mRFP provided by R.Y. Tsien, University of California, San Diego, San Diego, CA). The T2098L mutant version of FRB was generated by exchanging the FRB portion from the plasmid obtained from the Argent heterodimerization kit (Ariad Pharmaceuticals). Three constructs were designed that contained FKBP12 (amplified from a human EST available from GenBank/EMBL/DDBJ under accession no. ). All of them contained mRFP fused to the N terminus of FKBP12. The human type IV 5-ptase enzyme (available from GenBank/EMBL/DDBJ under accession no. ; provided by P.W. Majerus, Washington University, St. Louis, MO) was then fused to the C terminus of the FKBP12 either as the full-length protein or only its 5-ptase domain (residues 214–644). In both cases, the C641A mutation was introduced to destroy the C-terminal CAAX domain. A construct containing a stop codon at the end of the FKBP12 was also created (FKBP only). COS-7 cells were cultured on glass coverslips (3 × 10 cells/35-mm dish) and transfected with the various constructs (2 μg of total DNA/dish) using Lipofectamine 2000 for 24 h as described elsewhere (). For Ca measurements, cells were loaded with 3 μM fura-2/AM (45 min, room temperature). Ca measurements were performed at room temperature in a modified Krebs-Ringer buffer containing 120 mM NaCl, 4.7 mM KCl, 1.2 mM CaCl, 0.7 mM MgSO, 10 mM glucose, and 10 mM Na-Hepes, pH 7.4. An inverted microscope (IX70; Olympus) equipped with an illuminator (Lambda-DG4; Sutter Instrument Co.) and a digital camera (MicroMAX-1024BFT; Roper Scientific) and the appropriate filter sets were used for Ca analysis. Data acquisition and processing was performed by the MetaFluor software (Molecular Devices). Confocal analysis was performed in the same solution at 35°C using a confocal microscope (LSM 510-META; Carl Zeiss MicroImaging, Inc.). Patch-clamp experiments were performed on HEK293 cells after 2 d of transfection with the respective DNA constructs. Recordings were made using an amplifier (Axopatch 200B; Axon Instruments, Inc.) in an extracellular solution containing 137 mM NaCl, 5 mM KCl, 1 mM MgCl, 10 mM Hepes, and 10 mM glucose, pH 7.4. The pipette solution contained 135 mM K gluconate, 5 mM KCl, 1 mM MgCl, 5 mM EGTA, 10 mM Hepes, and 2 mM ATP (Na), pH 7.2. To assess TRPM8 channel activity, voltage ramps were applied from −100 to +100 mV every second. The current values measured at the +100 and −100 mV potential are shown in the recordings. Menthol was used at a concentration of 500 μM and rapa at 100 nM. COS-7 cells were cultured in 10-cm dishes (3 × 10 cells) and transfected with equal amounts of PLCδ1PH-CFP and -YFP, as well as the mRFP version of the appropriate FRB and FKBP constructs (10 μg of total DNA/dish) using Lipofectamine 2000 for 24 h. Cells were then trypsinized, centrifuged, and resuspended in the same modified Krebs-Ringer solution used in the Ca experiments. Measurements were performed at 35°C using a Deltascan fluorometer (PTI Technologies, Inc.) with excitation of 425 nm. To monitor the FRET signal, the ratio of the 525- and 475-nm emission was calculated. COS-7 cells were cultured in 10-cm dishes (3 × 10 cells) and transfected with equal amounts of the appropriate FRB and FKBP constructs (10 μg of total DNA/dish) using Lipofectamine 2000 for 24 h. Cells were then trypsinized, centrifuged, and resuspended in the same solution that was used in the Ca experiments (10 cells/ml). After treating the cells with rapa (3 min) and then with fluorescent transferrin (5 min) they were fixed with 2% PFA. FACS measurements were performed using a FACScan instrument (Becton Dickinson). To monitor the internalization in the transfected cell populations, the red channel was set to analyze the transfected cells and the mean green fluorescence of these cells was calculated.

MAPKs function as the terminal components of three-tiered cascades of kinases comprised of a MAPK kinase kinase (MAP3K), MAPK kinase (MAP2K), and MAPK and are important signal transducers in development, homeostasis, and disease (). For example, the p38 subfamily of MAPKs is involved in a wide variety of biological processes, including inflammation, stress responses, and cell differentiation (). The myriad roles of MAPK cascades indicate that the specificity of MAPK activation and function must be regulated. One mechanism by which this occurs is via MAPK scaffold proteins, which are thought to provide specificity between distinct MAPK subfamilies by assembling individual MAPK modules and precise spatial and temporal regulation to MAPK signaling (). How this latter function is accomplished is unclear, but it suggests that scaffold proteins may interact with cell-type specific factors. Differentiation of cells in the skeletal muscle lineage is coordinated by the family of myogenic bHLH factors (Myf5, MyoD, myogenin, and MRF4; ). During differentiation, these proteins work together with additional transcription factors, notably MEF2, to drive muscle-specific gene expression and promote myoblast fusion into myofibers (; ). Tight control of these transcription factors during myogenesis is required, and their activities are regulated by signal transduction pathways. Much evidence indicates that the p38α/β MAPK pathway plays an important role in myogenesis (; ; ; ). p38α/β activity increases and persists in differentiating myoblasts, and differentiation is blocked by the p38α/β inhibitor SB203580. p38α/β phosphorylates several proteins involved in muscle-specific gene expression, including MEF2 isoforms, the myogenic bHLH heterodimeric partner E47, the SWI–SNF complex subunit BAF60, and the RNA decay–promoting factor KH-type splicing regulatory protein (; ; ; ). MAPKs are generally activated in response to extracellular stimuli, and many such cues that activate p38 during inflammatory and other responses have been identified (). However, despite the attention p38α/β has received as a modulator of myogenesis, the signaling mechanisms by which it is activated during this process are largely unknown. Cdo is a cell surface Ig superfamily member with a long intracellular region (). Cdo promotes myogenesis in vivo and in vitro; mice lacking Cdo display delayed skeletal muscle development, and primary myoblasts (satellite cells) obtained from such animals differentiate defectively in vitro, expressing reduced levels of muscle-specific proteins and producing myotubes very inefficiently (). Cdo functions in myoblasts as a component of multiprotein complexes that also include the closely related factor Boc, the Ig superfamily receptor neogenin and its ligand netrin-3, and the adhesion molecules N- and M-cadherin (). Experiments with myoblast cell lines and reporter assays in fibroblasts indicate that one way Cdo promotes myogenesis is to signal to posttranslationally activate myogenic bHLH factors in a fashion that requires its intracellular region (; ). We report that the Cdo intracellular region binds JLP, a scaffold protein for the p38α/β MAPK pathway (; ). Cdo and JLP cooperate to activate p38α/β in transfectants, and endogenous Cdo, JLP, and p38α/β form complexes during myoblast differentiation. satellite cells are deficient in their ability to activate p38α/β, and their defective differentiation phenotype is rescued by the expression of an activated form of a p38 MAP2K, MKK6. Thus, one way p38α/β is activated during myogenesis is through the interaction of a pathway-specific scaffolding module with a Cdo-containing cell surface complex. To identify proteins that interact with the Cdo intracellular region, a yeast two-hybrid screen was performed with a construct containing the Cdo transmembrane plus intracellular region as bait (). Several positive clones corresponded to a portion of JLP. The transmembrane plus intracellular region of a different Ig protein, Necl-2, was used as a control bait, and JLP did not interact with Necl-2. Conversely, a Necl-2–binding protein, Pals2 (), bound the Necl-2 bait but not the Cdo bait. JLP, JIP4, and SPAG9 are alternatively spliced products of a single gene (), and RT-PCR analysis suggested that JLP is the major product in C2C12 myoblasts (unpublished data). and satellite cells were immunoprecipitated with antibodies to Cdo and blotted for the presence of JLP (). but not lysates, indicating that Cdo was required to bring down JLP and that the antibody was specific. In a reciprocal experiment, C2C12 myoblasts were transiently transfected with an expression vector encoding S epitope-tagged JLP or a control vector, and lysates were precipitated with anti-S agarose (). Full-length endogenous Cdo was coprecipitated from the JLP transfectants but not the control transfectants. We concluded that Cdo and JLP interact in myoblasts. Coimmunoprecipitation experiments in COS cells were then used to identify regions of JLP and Cdo involved in binding. A series of S-tagged fragments of JLP were coexpressed with Cdo and lysates pulled down with anti-S agarose (). Only a fragment encoding amino acids 465–1,008 coprecipitated Cdo; because a positive yeast two-hybrid clone contained JLP amino acids 317–647 (), the major Cdo-binding region of JLP resides between amino acids 465–647. p38α/β binds to two sites within JLP (amino acids 1–110 and 160–209; ), neither of which overlaps the Cdo-binding region (), suggesting that Cdo, JLP, and p38 could form a ternary complex. An analogous experiment was performed with Cdo deletion mutants that lack portions of the intracellular region. Loss of JLP binding was seen in each case (Fig. S1, available at ). It is possible that multiple regions of the Cdo cytoplasmic domain are required to provide structural integrity sufficient for JLP binding or that the Cdo deletion mutants may not be targeted to an appropriate subcellular compartment for interaction. To begin to determine whether Cdo–JLP interaction has a positive effect on p38α/β activity, transfections in heterologous systems were used. 293T cells were transfected with a vector encoding T7-tagged p38α and various combinations of other expression vectors, and lysates were blotted with antibodies to the dually phosphorylated (active) form of p38α/β (pp38α/β; ). The expression of p38α alone resulted in very low levels of pp38α, but the coexpression of ASK1 (a p38 MAP3K) produced abundant pp38α. The expression of Cdo increased the levels of pp38α above that seen with p38α alone, and this was further increased in a dose-dependent manner by the coexpression of JLP. The expression of JLP without Cdo was less effective. Cdo–JLP interaction was also tested with a MyoD-dependent reporter gene assay in fibroblasts, in which an activated form of the p38 MAP2K, MKK6 (MKK6EE), enhances MyoD activity (). Although the cotransfection of Cdo or JLP separately each increased MyoD-dependent reporter activity above what p38α alone produced, cotransfection of the two together produced ∼80% above what would be expected from a purely additive response (). Although the effects of Cdo and JLP coexpression on p38α activity in these heterologous systems are relatively modest, they are clearly stimulatory. We next asked whether endogenous Cdo, JLP, and p38α/β could be found in complexes during myogenesis. C2C12 cells were harvested while proliferating in growth medium (GM), at the time of transfer to differentiation medium (DM), and 48 h after transfer, when they were differentiating. Lysates were immunoprecipitated with Cdo antibodies and blotted for JLP, p38α/β, and Cdo; whole lysates were also probed for the expression of total p38, JLP, pp38α/β, and the differentiation markers myogenin and myosin heavy chain (MHC; ). JLP and p38α/β coprecipitated with Cdo only when the cells were actively differentiating despite the fact that total levels of all three proteins were unchanged in the various conditions examined. Furthermore, the formation of this complex correlated with pp38α/β production. It is likely that p38α/β coprecipitates with Cdo via binding to JLP, which interacts with Cdo, as p38α and Cdo did not interact in the yeast two-hybrid system (Fig. S2, available at ). To address this notion more directly, an RNAi approach was used. The pSilencer vector containing a sequence corresponding to mouse or an irrelevant sequence (as a control) was cotransfected into C2C12 cells with a GFP expression vector, and the cultures were sorted for the presence of GFP. Sorted cultures were replated and subsequently transferred into DM for 48 h, at which point whole lysates and lysates immunoprecipitated with Cdo antibodies were blotted for JLP, p38α/β, and Cdo (). RNAi-mediated knockdown of JLP led to a substantial decrease in the amount of p38α/β that coprecipitated with Cdo even though total levels of p38 were unaffected. Collectively, these results argue in favor of a model in which Cdo binds to JLP and JLP binds to p38α/β. Importantly, C2C12 cells that expressed RNAi to differentiated less efficiently than control transfectants as measured by myotube formation (fusion indices: control cells = 61.7 ± 4.2% and RNAi-expressing cells = 31.3 ± 5.9%; ). and satellite cells in GM and DM (). Unlike C2C12 cells, satellite cells must be cultured in basic FGF to remain in a proliferative nondifferentiated state, and this is associated with production of a pool of pp38α/β that is thought to target substrates distinct from those involved in differentiation (). and satellite cells produced roughly similar amounts of pp38α/β when cultured in GM/basic FGF. cells increased their relative pp38α/β levels when cultured in DM; in contrast, cells failed to increase or even maintain pp38α/β levels in DM, resulting in an approximately sixfold lower level of pp38α/β than cells. cells, it would be predicted that the restoration of p38α/β activity at a point downstream of Cdo would rescue the differentiation of these cells. and cells were infected with recombinant adenoviruses encoding either MKK6EE or, as a control, GFP. cells infected with the GFP virus had higher levels of pp38α/β than did GFP vector–infected cells; however, infection with the MKK6EE virus drove production of abundant pp38α/β in both cell types without altering total p38α/β levels (). These cultures were then scored for the expression of MHC and production of multinucleated myotubes when cultured under differentiation conditions for 72 h (). cells that were MHC versus MHC was ∼80/20, whereas the percentages of similarly infected cells were ∼50/50. cells restored the 80/20 percent ratio of MHC versus MHC cells seen in cells but had little effect on cells (). cells formed elongated myotubes and had a fusion index more than fourfold higher than cells, which failed to elongate (). cells with the MKK6EE virus led to the production of elongated myotubes by these cells and raised their fusion index to a level similar to that seen with cells, which were much more modestly affected by the expression of MKK6EE (). Similar results were obtained in C2C12 myoblasts. Differentiation of these cells was inhibited by the expression of RNAi to , but production of MHC and multinucleated cells was restored by the coexpression of MKK6EE (Fig. S3, available at ). It is concluded that the expression of MKK6EE specifically rescues the defects in myogenic differentiation caused by Cdo deficiency, presumably via restoration of p38α/β activity. p38α/β MAPK is established as a promyogenic kinase, but the mechanisms by which it is activated during differentiation are not well understood. Certain soluble signaling factors, including ATP and amphoterin, stimulate p38α/β activity and enhance myogenesis when added exogenously to cultured myoblasts; likewise, the expression of a dominant-negative amphoterin receptor blocks production of pp38α/β and differentiation (; ). Additionally, MyoD activity stimulates a feed-forward pathway that involves activation of p38 via induction of target genes (). However, in general, the signaling mechanisms underlying p38α/β activation by these factors are not clear nor are they confirmed by genetic loss of function data. The results described here reveal a novel mechanism of p38α/β activation during myogenesis: the interaction of JLP, a p38α/β MAPK scaffold protein, with Cdo, which is a component of multiprotein cell surface complexes comprised of promyogenic signaling receptors and adhesion molecules (). The binding of a MAPK scaffold protein to the intracellular region of a transmembrane receptor protein is unusual; furthermore, this mechanism is distinct from other known receptor-mediated signaling mechanisms, such as intrinsic enzyme activity (e.g., receptor tyrosine kinases) or direct coupling to nonreceptor tyrosine kinases (e.g., cytokine receptors). JIP3, which is structurally related to JLP, binds the cytoplasmic tail of Toll-like receptor 4 (), suggesting that direct interaction with transmembrane receptors may be a feature of this class of scaffold protein. It is anticipated that in its role as a scaffold (; ), JLP brings additional components of the pathway, such as MAP3Ks and MAP2Ks, to these complexes. Furthermore, the interaction of a pathway-specific scaffolding module with Cdo- containing cell surface complexes may allow the coordination of additional signals required for p38α/β activity via the actions of other membrane components of such complexes (e.g., regulation of small GTPases by cadherins; ). Cdo is also expressed in and promotes the differentiation of neuronal precursors (), and similar signaling mechanisms may be involved in myogenesis and neurogenesis. Assembly at sites of cell–cell contact of higher order structures comprised of multiprotein cell surface complexes and intracellular signaling modules is an appealing mechanism for coordinating changes in gene expression and cell morphology during cell differentiation in general. A cDNA encoding the transmembrane and intracellular regions of mouse Cdo (amino acids 948–1,250) was fused in frame to the Gal4 DNA-binding domain in the pGBDU-C1 vector. The yeast strain PJ69-4A was sequentially transformed with this vector and a library containing mouse embryo cDNAs fused to the Gal4 activation domain via lithium acetate. Approximately 4.2 × 10 transformants were obtained and screened as described previously (). Four identical clones encoding amino acids 317–647 of JLP were isolated. and mice were cultured as described previously (; ). To induce differentiation, C2C12 and satellite cells were transferred into medium containing 2% horse serum or 5% FBS, respectively. Quantification of myotube formation was performed as described previously (). For RNAi studies shown in , the sequence 5′-AGATGCGTCTATGAAGCTG-3′ was inserted into the pSilencer 2.0-U6 vector (Ambion). pSilencer 2.0-U6 expressing an irrelevant sequence (Ambion) was used as a control. These vectors were transfected along with a GFP expression vector. Cells were sorted for the presence of GFP and analyzed by coimmunoprecipitation and Western blotting as described in the next paragraph and for the ability to form myotubes. RNAi-mediated knockdown of was as described previously (), and transient assay of its effects on C2C12 differentiation was performed as described previously (). Replication-deficient adenoviruses encoding GFP or HA-tagged MKK6EE were provided by M. Meseck (Mount Sinai School of Medicine, New York, NY) and L. Puri (Burnham Institute, La Jolla, CA; ), respectively, and amplified in 293T cells. Satellite cell cultures were infected at an MOI of 50 and assessed for the expression of MKK6EE, pp38α/β, and MHC and for myotube formation. Western blot analyses were performed as described previously by . For immunoprecipitations, cells were lysed in extraction buffer (20 mM Hepes, pH 7.5, 150 mM NaCl, 1.5 mM MgCl, 10 mM NaF, 2 mM DTT, 1 mM NaVO, and 0.5% Triton X-100 supplemented with 1 tablet/40 ml of Complete protease inhibitor cocktail [Roche]). 2 mg of whole cell extract from each sample was precleared with protein G–Sepharose (GE Healthcare) conjugated with 2 μg of normal rabbit IgG (Santa Cruz Biotechnology, Inc.) for 1 h at 4°C followed by immunoprecipitation with 2 μg anti-Cdo antibody for 2 h at 4°C. Immunocomplexes were washed three times with and suspended in extraction buffer, and samples were analyzed by Western blotting. For pull-down experiments, 4 × 10 cells were seeded onto 100-mm plates 1 d before transfection with plasmids encoding a series of S-tagged JLP proteins (). 2 d after transfection, cells were harvested and lysed in extraction buffer. Whole cell extracts were incubated with 20 μl of 50% slurry S-protein agarose beads (Novagen) for 3.5 h at 4°C. Beads were washed three times with and suspended in extraction buffer, and samples were analyzed by Western blotting. For reporter assays, 3 × 10 10T1/2 cells were seeded onto individual wells of a six-well plate 1 d before transfection with FuGene6 (Roche). For each well, 200 ng 4RTK-luc and, as an internal control, 100 ng CMV- plasmids were transfected along with expression vectors for MyoD (50 ng), E47 (10 ng), p38α (100 ng), Cdo (200 ng), and JLP (400 ng) as indicated in . Cells were incubated for 48 h after transfection and harvested to determine luciferase and β-galactosidase activity as described previously (; ). Cultures were processed as described previously () and examined on a phase-contrast microscope (Eclipse TS100; Nikon) with plan Fluor 10× NA 0.3 and 20× NA 0.45 objectives (Nikon) at room temperature. Images were captured with a camera (model 2.2.1 Spot RT Color; Diagnostic Instruments) using Spot software (version 3.5.9; Diagnostic Instruments) and Photoshop 7.0 (Adobe). Fig. S1 shows that Cdo deletion mutants do not bind to JLP, and Fig. S2 shows that p38α does not bind to Cdo. Fig. S3 shows that MKK6EE rescues the block to C2C12 cell differentiation imposed by RNAi to . Online supplemental material is available at .

In eukaryotic nuclei, DNA is wrapped around a protein octamer containing two copies of each of the core histones H2A, H2B, H3, and H4, forming nucleosomes, the fundamental units of chromatin (). Nucleosomes are assembled with the assistance of chaperones or assembly complexes. During de novo nucleosome assembly, DNA is first wrapped around the H3–H4 tetramer before the addition of two H2A–H2B dimers (). Once assembled, these core histones are tightly bound to DNA, and the interactions must be loosened or remodeled to allow the access of molecular machineries (e.g., polymerases) to DNA. In living cells, the histone–DNA interaction and chromatin structure are expected to be continually altered during transcription, genome duplication, and damage recovery, and the remodeling of chromatin is often associated with histone exchange (; ; ; ; ). On the other hand, nucleosome contexts on specific loci must be preserved to maintain epigenetic marks on histone tails (). The modification of histones, including acetylation, methylation, and phosphorylation, plays essential roles in chromatin functions such as gene expression and chromosome segregation. Thus, fluidity and stability seem to be well balanced in nucleosomes in living cells. Early studies of histone deposition and exchange in living cells used radiochemical labeling. The stable association of [H]arginine-labeled H3–H4 with chromatin was demonstrated by cell fusion (). In a series of studies analyzing the deposition of the newly synthesized radio-labeled histones into nucleosomes, it was found that linker histone H1 and H2A–H2B undergo exchanges independently of DNA replication and transcription (; ). Such different behaviors of different histone species are also seen in living cells using GFP-tagged proteins (). The linker histone H1 is rapidly exchanged within a few minutes, and core histones are more stably bound. Long-term observation and cell fusion experiments further revealed that a substantial fraction of H2B-GFP exchanges slowly in euchromatin, whereas most H3-GFP (which is the H3.1 variant and is referred to below as H3.1-GFP) and H4-GFP remain bound to chromatin. In addition to the slowly exchanging fraction of H2B-GFP, which exchanges independently of DNA replication and transcription, another rapidly exchanging fraction, probably coupled to transcription, has been observed (), which is in agreement with the dimer eviction observed during transcription (; ). The exchange and assembly of histones are regulated in a development- and differentiation-specific manner () by chaperones or assembly factors that are distinct for each histone variant. Histone H3 has several variants, which are deposited into specialized chromatin loci mediated differentially through the action of deposition complexes (; ; ; ). The modification pattern in the conserved tail is also distinctive in each variant (). H3.3 has modifications associated with transcriptionally active chromatin, which is consistent with its localization on active genes; in contrast, replication-coupled H3.2 has mostly silencing modifications. Although the modification pattern of each histone is established after its assembly into nucleosomes (influenced by the surrounding chromatin state), some specific modifications are associated with nucleosome-free deposition forms. Such modifications are typically found in H4, whose deposition form is diacetylated throughout eukaryotes, and some acetylation is associated with newly synthesized H3 in human cells (). In addition to H3, variant-specific deposition and modification are found in H2A. Nucleosome assembly protein 1 (Nap1) and the related proteins are known as somatic H2A–H2B chaperones after Nap1's purification from HeLa cells on the basis of nucleosome assembly activity in vitro (; ; ). Although Nap1 is not essential for yeast viability, its disruption affects the expression level of ∼10% of genes in clusters, suggesting a nucleosome maintenance function of Nap1 by depositing H2A–H2B (). Although Nap1 assists nucleosome assembly without ATP in vitro, complexes containing ATP-dependent chromatin remodeling activity have recently been shown to mediate the exchange of H2A–H2B dimers (; ; ; ). A complex containing yeast SWR1 (Swi2/Snf2-related ATPase 1) exchanges canonical H2A with H2AZ in nucleosome arrays, and SWR1 and H2AZ regulate an overlapping subset of genes. Another complex containing Tip60 (Tat-interacting protein 60) is involved in the exchange of phosphorylated H2Av (a histone H2A variant homologous to H2AX) with the unphosphorylated form at DNA lesions in (). Thus, multiple mechanisms appear to exist to control the exchange of H2A variants at appropriate chromatin loci and in response to various stimuli, including DNA damages. To understand the molecular mechanisms that regulate the assembly and exchange of histones in higher eukaryotes, we set out to establish an in vitro system that mimics in vivo histone dynamics using permeabilized cells. When cells are treated with nonionic detergents such as Triton X-100 or saponin, cellular membranes are permeabilized and some proteins are extracted, but many nuclear functions remain active (; ), and some nuclear structures can be manipulated by adding exogenous factors (; ). Therefore, we expected that exogenously added histones might be incorporated into chromatin in permeabilized cells by exchange or replication-coupled assembly. As expected, GFP-tagged histones were indeed incorporated into chromatin in permeabilized cells with the assistance of cellular factors. By purifying the factors assisting GFP-H2A–H2B incorporation, we identified the type 2C protein phosphatase (PP) 2Cγ/PPM1G (; ) in addition to the Nap1 family members. PP2Cγ directly bound to and dephosphorylated H2A–H2B, and its disruption in chicken DT40 cells caused hypersensitivity to checkpoint abrogation. Although PP2Cγ did not appear to be the major phosphatase for H2AX and H2B, dephosphorylation and exchange via PP2Cγ may function to allow full recovery from DNA damage. As illustrated in , HeLa cells were permeabilized and incubated in cell extracts prepared from cells stably expressing GFP-tagged core histones whose expression levels were <10% of their endogenous counterparts (). After washing out the unincorporated materials, GFP-H2A localized to euchromatin, which was devoid of DAPI-dense heterochromatin (, inset), in most permeabilized cells (). In contrast, H3.1-GFP highlighted replicated chromatin, which was labeled with Cy3-dUTP. The euchromatic localization of GFP-H2A was confirmed by the overlapping signals with specific antibodies recognizing acetylated histone H3 (not depicted) and H4 (), which are associated with transcriptionally active chromatin (). In contrast, GFP-H2A was excluded from inactive chromatin rich in K20-trimethylated H4 (; ). To confirm that GFP-H2A replaced the endogenous H2A in chromatin in permeabilized cells, GFP-containing mononucleosomes were prepared by immunoprecipitation using antibody directed against GFP, and the ratio of core histones was analyzed by SDS-PAGE and Coomassie staining (). The amount of H2A was roughly halved in GFP-H2A nucleosomes (), suggesting the incorporation of a dimer of GFP-H2A and H2B (GFP-H2A–H2B) into chromatin by exchange. This stoichiometry is unlikely to be created by the nonspecific aggregation of a GFP-H2A–containing histone octamer from the extract onto chromatin, as H2A–H2B and H3–H4 are present in different complexes in the chromatin-free fraction (see ; ; ). We next used specific inhibitors to examine whether the incorporation of histones depends on transcription and/or DNA replication in permeabilized cells. Most H2A–H2B appeared to be exchanged independently of ongoing RNA polymerase II transcription and DNA replication, as the incorporation of GFP-H2A and H2B-GFP was still observed in the presence of α-amanitin and aphidicolin, respectively ( and not depicted). The incorporation of H3.1-GFP into chromatin was coupled to DNA replication, as the signal almost disappeared in the presence of aphidicolin (). These results are reminiscent of the different behaviors of H2A–H2B and H3.1–H4 observed in living mammalian cells (; , ; ). To analyze whether GFP-H2A–H2B dimer alone can be incorporated into chromatin, permeabilized cells were incubated with GFP-H2A–H2B purified from HeLa cells expressing GFP-H2A (). Although GFP-H2A–H2B alone failed to be incorporated, its incorporation was restored when supplemented with HeLa cell extract (), suggesting the presence of soluble factors required for H2A–H2B exchange in the extract. By following the incorporation of GFP-H2A under a fluorescent microscope, we purified the activity required for H2A–H2B exchange using column chromatography (). The purest active fraction consisted of three major bands by SDS-PAGE (). Mass spectrometry analysis identified these polypeptides as PP2Cγ/PPM1G (; ), Nap1/Nap1L1 (), and Nap2/Nap1L4 (). It was not surprising to find Nap1 and Nap2 in the active fractions, as they have been described as histone chaperones that assist nucleosome assembly in vitro (; ). In contrast, no link between PP2Cγ and histones was previously established, which prompted us to focus on the function of PP2Cγ in histone H2A–H2B exchange. The phosphatase might regulate the chaperone activity of Nap1/2 by altering their phosphorylation state. On the other hand, PP2Cγ might also mediate the H2A–H2B exchange as such because it has a unique acidic domain () that could potentially interact with histone H2A–H2B. To examine the relationship between H2A–H2B exchange activity and the individual proteins, we incubated permeabilized cells with the purified GFP-H2A–H2B and each recombinant protein fused to a histidine hexamer (His) tag expressed in and purified from (). GFP-H2A was incorporated into chromatin in the presence of either His-Nap1, -Nap2, or -PP2Cγ (), and similar results were obtained when ATP was omitted from the system (Fig. S1, available at ), suggesting that PP2Cγ itself possesses ATP-independent chaperone activity, as do Nap1 and Nap2. Although the cofractionation of these three proteins by gel filtration chromatography () suggests their presence in a complex, only additive effects on the incorporation of GFP-H2A–H2B were observed when these three recombinant proteins were mixed (). As the acidic domain is unique to PP2Cγ among PP2C family members (; ), this domain might be essential for the chaperone function. Indeed, a phosphatase mutant lacking the acidic domain (ΔAcDo) did not support GFP-H2A incorporation (). Furthermore, coimmunoprecipitation analysis confirmed that the physical interaction of PP2Cγ with H2A–H2B requires this domain (). When FLAG-tagged phosphatase was transiently expressed in human 293T cells and recovered using anti-FLAG agarose beads, substantial amounts of endogenous H2A and H2B were coprecipitated with FLAG-PP2Cγ but not with FLAG-ΔAcDo (). The interaction between basic proteins like histones and the acidic domain could occur through nonspecific binding as a result of the positive and negative charges. However, the immunoprecipitation experiments show specific binding of the phosphatase to H2A–H2B because only these two, but not the other histones (i.e., H1, H3, and H4), were coprecipitated even though all histone subtypes are positively charged. The complex formation between GFP-H2A–H2B and PP2Cγ in the cell extract (used in ) was observed by immunoprecipitation using anti-GFP agarose beads (). The presence of PP2Cγ as well as Nap1 in the immunoprecipitates was confirmed by mass spectrometry () and immunoblotting (). A two-hybrid cDNA library screen also yielded histone H2B as an interactor with PP2Cγ, and the interaction required the phosphatase's acidic domain (unpublished data). The aforementioned results showing the physical interaction between PP2Cγ and H2A–H2B suggest that these histones could be substrates for the phosphatase. Therefore, we analyzed the phosphorylation state of FLAG-PP2Cγ–bound histones using acid-urea-Triton (AUT) gel electrophoresis and immunoblotting with specific antibodies directed against phosphorylated histones (). As expected, histones coprecipitated with the wild-type phosphatase were poorly recognized by antiphosphohistone antibodies. In contrast, histones bound to a phosphatase-inactive mutant (D496A) comprised detectable levels of phosphorylated molecules, including those related to DNA damage response and apoptosis such as Ser139-phosphorylated H2AX (called γ-H2AX; ) and Ser14-phosphorylated H2B (; ), although the overall migration pattern was similar to those bound to the wild-type phosphatase. As bulk nucleosomal histones were still phosphorylated in cells overexpressing the wild-type phosphatase ( and by immunofluorescence; not depicted), only nucleosome-free H2A–H2B may be dephosphorylated by the phosphatase. Consistently, the purified His-PP2Cγ efficiently dephosphorylated nucleosome-free histones, including γ-H2AX in vitro (). As the D496A mutant still supported histone exchange in permeabilized cells (), the histone exchange and dephosphorylation do not appear to be coupled. These results suggest that a nucleosome-free H2A–H2B that binds to PP2Cγ may be dephosphorylated before its deposition into a nucleosome. Although we did not obtain positive data indicating the dephosphorylation of nucleosomal histones by PP2Cγ in overexpression and in vitro assays, it is also possible that additional cellular factors, which may be limited in the assays, stimulate the phosphatase activity or targeting toward the nucleosomal histones. We next tested whether PP2Cγ has in vitro nucleosome assembly activity using a supercoiling assay (Fig. S2, available at ) in which the assembly of nucleosomes can be assessed by the formation of supercoils from relaxed circular DNA (; ). Most of the plasmid DNA became supercoiled in the presence of Nap1, but only some supercoiled molecules accumulated even in the presence of high levels of PP2Cγ (Fig. S2), indicating that PP2Cγ has only weak de novo nucleosome assembly activity. To investigate whether PP2Cγ is involved in the regulation of H2A–H2B kinetics in living cells, we knocked down PP2Cγ in HeLa cells expressing histone-GFP using RNAi; 3 d after the transfection of a specific siRNA, the level of PP2Cγ decreased substantially to <5% of the normal level (). The mobility of H2A–H2B was analyzed by fluorescence recovery after photobleaching (). The recovery kinetics of both GFP-H2A and H2B-GFP decreased in cells transfected with PP2Cγ-specific siRNA compared with those with the control siRNA (), whereas the mobility of the linker histone H1c-GFP was unaffected (). These observations in living cells reflect the results from in vitro assays, suggesting that at least a part of H2A–H2B exchange is mediated by PP2Cγ as a histone chaperone in HeLa cells. To gain further insights into the biological function of PP2Cγ in vertebrate cells, we established PP2Cγ-deficient chicken DT40 cells by gene targeting (Fig. S3, available at ). As the deficient cells were generated by a simple knockout strategy to disrupt both alleles, PP2Cγ does not appear to be essential for cell growth. However, substantial growth defects were observed when DNA replication and damage checkpoints were abrogated by caffeine, which preferentially inhibits ataxia telangiectasia mutated– and ataxia telangiectasia and RAD3 related–dependent pathways, although its exact interfering points remain elusive (; ). As shown in , PP2Cγ-deficient cells were more sensitive to caffeine compared with the wild type in a growth rate assay () and in a colony formation assay (). In 2 mM caffeine, the wild-type cells continued to grow for 3 d, but PP2Cγ-deficient cells stopped growing at day 2. At a higher concentration (4 mM), the number of live cells (judged by the exclusion of trypan blue) became considerably lower after day 2 in PP2Cγ-deficient cells (). The colony formation assay revealed that the survival rate after 22 h of incubation in 4 mM caffeine was 35 ± 8 and 8.2 ± 0.3% in the wild-type and PP2Cγ-deficient cells, respectively (). As caffeine is known to sensitize cells to DNA double-strand breaks induced by ionizing radiation (; ), we compared the sensitivity of these cells with γ-ray irradiation in the presence or absence of caffeine. Although PP2Cγ-deficient cells showed a similar radiation sensitivity to the wild type without caffeine, they became more sensitive when 1 mM caffeine was present in the colony-forming medium (). These results indicate that PP2Cγ is not essential for DNA double-strand break repair but suggest its involvement in recovery from damage. As H2AX is known to be phosphorylated around damaged chromatin, its dephosphorylation is required for full recovery from the damage response (; ). Even though PP2A is likely to be the major γ-H2AX phosphatase in higher eukaryotes (), PP2Cγ could be involved in a backup dephosphorylation and deposition pathway. To assess the role of PP2Cγ in γ-H2AX dephosphorylation, the phosphorylation level of H2AX (i.e., the signal detected with antibody directed against γ-H2AX) was compared between the wild-type and PP2Cγ-deficient cells in response to DNA damage combined with treatment with calyculin A, an inhibitor of PP1 and PP2A (; ). In both cells, γ-H2AX appeared at a similar level 2 h after irradiation (8 Gy) and disappeared by 8 h (, lanes 1–6); faint signals of apoptosis-associated H2B (S14) phosphorylation appeared by 8 h. When cells were incubated with calyculin A, γ-H2AX was accumulated by 8 h even in the wild-type cells, probably as a result of spontaneous or replication-associated damages, which is consistent with the involvement of PP2A in γ-H2AX dephosphorylation (). The levels of γ-H2AX and phospho-H2B (S14) were higher in PP2Cγ-deficient cells in the presence of calyculin A (, lanes 9 and 12), suggesting that PP2Cγ is also one of the phosphatases that regulate γ-H2AX and phosphorylated H2B. The difference between wild-type and mutant cells became more evident when the cells were irradiated and incubated in calyculin A, as more γ-H2AX and phospho-H2B (S14) were accumulated in PP2Cγ-deficient cells (, lanes 13–18). These results suggest that PP2Cγ dephosphorylates γ-H2AX and phosphorylated H2B in wild-type DT40 cells. To understand the biological function and molecular mechanisms of histone dynamics, we established a permeabilized cell-based assay for histone assembly and exchange. GFP-H2A and H2B-GFP were incorporated into euchromatin in permeabilized cells. This is consistent with the exchange of H2A–H2B in living cells, which can occur independently of DNA replication and transcription (; ), preferentially in chromatin-containing acetylated H4 (). H3.1-GFP assembled into replicated chromatin but contrasted to H2A–H2B, which is also reminiscent of the behavior in living cells (). By purifying the activity that assists GFP-H2A–H2B incorporation into chromatin in permeabilized cells, we identified three proteins—Nap1, Nap2, and PP2Cγ—in the purest fraction. Finding these Nap1-related proteins in our active fractions reassures us that the permeabilized cell-based assay has physiological relevance. The third protein we found was PP2Cγ, which harbors a unique acidic domain () and was purified as a factor that stimulates spliceosome assembly in vitro (). Our analyses indicated that the phosphatase as such can assist the incorporation of H2A–H2B into chromatin in permeabilized cells and that it binds to and dephosphorylates histone H2A and H2B subtypes. Although the acidic domain of PP2Cγ could potentially mediate nonspecific electrostatic binding to basic proteins such as the histones, the fact that H2A–H2B was exclusively coimmunoprecipitated among all of the histones using FLAG-tagged phosphatase suggests that the interaction between PP2Cγ and H2A–H2B is specific. These histone chaperones do not require ATP for assisting H2A–H2B incorporation into chromatin in permeabilized cells as well as for in vitro nucleosome assembly with naked DNA. Because we followed the most active fractions that support GFP-H2A incorporation globally in euchromatin, other H2A–H2B exchange factors that are probably less abundant and act on more specific loci, including facilitating chromatin transcription (FACT; ) and ATP-dependent remodeling factors (), were not found in the final preparation. Although Nap1/2 and PP2Cγ may mediate global H2A–H2B exchange independently of transcription and DNA replication, FACT may participate in transcription-coupled exchange. Future studies may reveal whether FACT supports H2A–H2B incorporation in a transcription-dependent manner in permeabilized cells. A recent study revealed that ATP-dependent chromatin remodeling complexes can mediate histone exchange in addition to their remodeling function without the displacement of histone octamers (). Therefore, it is also possible that the function of ATP-independent chaperones like Nap1/2 and PP2Cγ is solely to escort H2A–H2B and transfer the dimer to the ATP-dependent machineries, such as the yeast SWR1 complex that catalyzes the exchange between a canonical dimer and an H2AZ–H2B dimer (). However, several lines of evidence suggest that the chaperones might also mediate H2A–H2B incorporation by themselves in addition to their escorting function. First, yeast Nap1 has the ability to exchange H2A–H2B in mononucleosomes in vitro (). Second, additional ATP is not required for H2A–H2B incorporation supported by Nap1/2 and PP2Cγ in permeabilized cells. Third, a substantial H2B-GFP recovery was still observed in living cells by FRAP even when the cellular ATP pool was depleted by treatment with sodium azide (unpublished data). Thus, although ATP-dependent factors might be required for the exchange of a dimer containing H2AZ at specific loci or during gene activation, ATP-independent chaperones may participate in the basal level of exchange of the major H2A and other variants. Alternatively, the major role of ATP-independent chaperones may be to deposit an H2A–H2B dimer into an incomplete nucleosome lacking a dimer, which can result from positive torsional stress () or through ATP-driven eviction. This may account for the slow exchange rate of H2A–H2B in living cells despite the presence of a large pool of PP2Cγ (∼10 molecules/HeLa cell) diffusing almost freely in the nucleus (unpublished data). To understand the biological function of PP2Cγ at the cellular level, we used chicken DT40 cells to create knockout cells by gene targeting. Although the deficient cells are viable, they show subtle growth retardation and a remarkable hypersensitivity to caffeine, which abrogates DNA replication and damage checkpoints. One possible mechanism to explain these phenomena is that the chaperone function together with the phosphatase activity plays a role in completing chromatin formation after DNA repair and/or replication by depositing dephosphorylated H2A–H2B molecules (). H2AX is phosphorylated around damaged chromatin (), and its dephosphorylation is required for full recovery from damage responses. Also, H2AX molecules outside the damaged area are kept from undergoing phosphorylation for several hours. Although PP2A seems to play a major role in γ-H2AX dephosphorylation on chromatin (), we showed that PP2Cγ likewise mediates γ-H2AX and H2B dephosphorylation, as PP2Cγ-deficient cells showed a greater accumulation of γ-H2AX and phosphorylated H2B (S14) compared with wild-type cells when PP1 and PP2A were inhibited by calyculin A. Although the eviction of γ-H2AX or phosphorylated H2B may be mediated by other proteins such as the Tip60-containing complex (), PP2Cγ may passively deposit dephosphorylated H2A–H2B or H2AX–H2B to incomplete nucleosomes lacking one dimer. This view is consistent with the observed uncoupling of the chaperone function and phosphatase activity of PP2Cγ; histone dephosphorylation can occur at any time after the binding of PP2Cγ until deposition (). Most H2A–H2B that bound to PP2Cγ but away from nucleosomes was indeed dephosphorylated. The lack of PP2Cγ in the DT40 knockout cells may thus delay the recovery from damage. When checkpoints are functional, such a subtle repair defect would not be critical and might only cause a subtle delay in cell growth. However, when checkpoints are abrogated, more cells with damaged chromatin would enter into mitosis for catastrophe. An alternative possibility is that the substrate specificity or phosphatase activity of PP2Cγ is regulated by binding to H2A–H2B (); the level of nucleosome-free H2A–H2B could be altered by damage or replication fork arrest. The type 2C phosphatase family members are indeed involved in checkpoint responses (; ), and the γ subtype in particular might take part in inactivating checkpoints by sensing the free H2A–H2B level in the nucleus. Finally, a link between chromatin-remodeling factors and alternative pre-mRNA splicing was recently reported (). Consistent with this observation, PP2Cγ was previously identified as a factor that stimulates pre-mRNA splicing in vitro (), raising the interesting possibility that PP2Cγ coordinately regulates stress responses in mammalian cells at the level of chromatin and RNA splicing. It is now widely acknowledged that histone modification is key for the regulation of chromatin functions. Recent studies further indicate that the deposition and exchange of appropriate histone variants to specific chromosome loci are also important for gene expression and genome integrity (; ). A connection between the histone modification and deposition has been shown typically in the case of histone H4; before replication-coupled assembly, the newly synthesized molecules are diacetylated by HAT1 histone acetylase in the H3.1–H4 deposition complex (; ). Although diacetylation is not a prerequisite for assembly (), this modification contributes to the recovery from replication block-mediated DNA damage (). Similarly, in the case of H2A–H2B and H2AX–H2B, the deposition of unphosphorylated forms mediated by PP2Cγ appears to play a role in DNA damage responses. Thus, controlling the incorporation of appropriately modified histones seems to be important for maintaining genome integrity. Future studies should reveal how individual ATP-independent chaperones and ATP-dependent remodeling complexes function in distinct exchange processes in different chromatin contexts. Although differences in histone exchange kinetics in vivo were shown decades ago (; ), the biological significance of the exchange and the underlying molecular mechanisms are just emerging. The approach presented in this study may contribute to bridging the gap between live cell observations and biochemical analyses. In typical experiments, HeLa cells were plated in a 12-well plate containing 15-mm coverslips and were grown up to subconfluence. Cells were chilled on ice, washed twice in ice-cold physiological buffer (PB; 100 mM CHCOOK, 30 mM KCl, 10 mM NaHPO, 1 mM DTT, 1 mM MgCl, and 1 mM ATP; ) containing 5% Ficoll (PBF; pH 7.4; 1 ml per well; Nacalai Tesque), permeabilized in PBF containing 0.1% Triton X-100 (1 ml; for 5 min on ice), and washed twice in 1 ml PBF on ice. Cells were incubated for 1 h at 30°C in a reaction mixture containing cell extract (40%) or purified proteins supplemented with 100 μM each of NTP and dNTP (GE Healthcare), 0.4 μM Cy3-dUTP (PerkinElmer), and 800 μM MgCl in PBF. For incubation, a coverslip was overlaid (cell side down) on a 100-μl drop of the reaction mixture on Parafilm covering a flat aluminum block in a water bath at 30°C. After washing twice in 1 ml PBF for 5 min on ice in a 12-well plate, cells were fixed in 4% PFA (Electron Microscopy Sciences) in 250 mM Hepes-NaOH, pH 7.4 (Wako), for 20 min at room temperature, washed three times in 1 ml PBS, and DNA was counterstained with DAPI (12.5 ng/ml in PBS; 1 ml for 15 min; Nacalai Tesque). After washing twice in 1 ml PBS, coverslips were mounted using Prolong Gold (Invitrogen). In some cases, ATP and the other nucleotides were omitted from PBF and the reaction mixture. For immunolabeling (), permeabilized cells were incubated in the reaction mixture containing 40% GFP-H2A extract and 2 μM Cy5-dUTP instead of Cy3-dUTP for 30 min at 30°C. After fixation, cells were treated with 1% Triton X-100 in PBS for 20 min, washed five times in PBS, and incubated in blocking buffer (0.2% gelatin, 1% BSA, and 0.05% Tween 20 in PBS, pH 8.0) for 30 min and then with rabbit polyclonal antibodies directed against hyperacetylated H4 (1:1,000; Upstate Biotechnology) or H4-trimethylated K20 (1:500; Abcam) in the same buffer for 3 h. Cells were washed in PBS containing 0.05% Tween 20 (PBST) three times for 10 min, incubated in Cy3-conjugated donkey anti–mouse IgG (1:500; Jackson ImmunoResearch Laboratories) overnight at 4°C, and washed with PBST three times for 10 min before DAPI staining. Fluorescence images were sequentially collected using a confocal microscope featuring 405-, 488-, 543-, and 633-nm laser lines with the optimized pinhole setting operated by the built-in software: either a microscope (LSM510 META; Carl Zeiss MicroImaging, Inc.) with a C-Apo 40× NA 1.2 objective lens (for and ) or a microscope (FV-1000; Olympus) with a UPlanSApo 60× NA 1.35 lens (for and – ). Image files were converted to tiff format using the operating software, merged, linearly contrast stretched (with the same setting in each set of experiments) using Photoshop version 7.01 (Adobe), and imported into Canvas 8 (Deneva) for assembly. For chromatin immunoprecipitation, cells were centrifuged at 1,300 for 10 min at 4°C after each step for buffer replacement. After the incubation and washing, nucleosomes were prepared, and GFP-containing nucleosomes were precipitated as described previously (; ). HeLa cells and derivatives expressing H2B-GFP () and H3.1-GFP were grown as described previously (), and lines expressing GFP-H2A and H1c-GFP were established by transfecting the expression vectors (; ). Cell extracts were prepared based on the study by with modifications. The S100 extract was prepared using a 1.5 cell-packed volume of 10 mM CHCOOK, 3 mM KCl, 1 mM NaHPO, 1 mM MgCl, 1 mM ATP, 1 mM DTT, 10 mM Hepes-KOH, pH 7.4, and Complete protease inhibitor cocktail (EDTA-free; Roche) and dialyzed against PB plus inhibitors (1.5 μg/ml leupeptin, 2.5 μg/ml aprotinin, and 1 μg/ml pepstatin A; Wako). The nuclear pellet was extracted using an equal volume of 20 mM Hepes-KOH, 0.6 M KCl, 0.2 mM EDTA, 25% glycerol, 1 mM DTT, 1.5 mM MgCl, and protease inhibitor cocktail (Roche) to yield the nuclear extract, which was also dialyzed against PB. Histone H2A–H2B and H3–H4 were separately purified from the nuclear pellet essentially according to , and the GFP-H2A–H2B fraction was separated from untagged H2A–H2B using gel filtration column chromatography (HiLoad Superdex 75; GE Healthcare). To purify the activity assisting histone H2A–H2B incorporation in permeabilized cells, S100 extract was first fractionated through a histone H2A–H2B column, which was prepared by coupling 3 mg of the purified H2A–H2B to 1 ml -hydroxysuccinimide ester–activated Sepharose (GE Healthcare) according to the manufacturer's instructions. Approximately 6 mg/ml HeLa S100 extract was mixed with 5 M NaCl to yield a final salt concentration of 0.5 M before applying to the column (2 ml per run). After washing with five column volumes of PB containing 0.5 M NaCl, bound proteins were eluted with a linear gradient of NaCl (0.5–2 M in PB; 20 column volumes). Each fraction was concentrated, and the buffer was substituted to PB using Ultrafree 0.5 (Millipore) before use in the permeabilized cell assay with 10–20 μg/ml of purified GFP-H2A–H2B. The activity supporting the nuclear localization of GFP-H2A except in nucleoli was followed under a fluorescence microscope (Axiovert2; Carl Zeiss MicroImaging, Inc.). The active fractions eluted in ∼1 M NaCl were further separated using a MonoQ 5/50 GL column (GE Healthcare) with a linear gradient of NaCl (0–2 M in PB; 20 column volumes). The most active fraction eluted in ∼0.6 M NaCl was concentrated and separated (0.2 ml in each fraction) on a Superose 6 gel filtration column (GE Healthcare). Proteins were identified by mass spectrometric analysis using a mass spectrometer (Ultraflex TOF/TOF; Bruker Daltonics) and by comparison between the determined molecular weights and theoretical peptide masses from the proteins registered in the National Center for Biotechnology Information. The cDNAs encoding human Nap1 (Nap1L1) and Nap2 (Nap1L4) were amplified by reverse transcription (Revertra Ace and oligo-dT18; TOYOBO) of HeLa RNA and PCR (high fidelity PCR master; Roche) using the following primers designed from GenBank/EMBL/DDBJ accession no. (Nap1) and (Nap2): Nap1 forward, TTACCATATGGCAGACATTGACAACAAAGAACAGTC; Nap1 reverse, CATCAAGCTTCACTGCTGCTTGCACTCTGCTGGGTT; Nap2 forward, CCTTCATATGGCAGATCACAGTTTTTCAGATGGGGT; and Nap2 reverse, GACAAAGCTTACACCTTGGGGTTAATTTCCGCATCA. The amplified products were purified (QIAGEN), digested with NdeI and HindIII, and ligated into pHIT51 (containing T7 promoter, D box, His sequence, and multicloning site; provided by H. Tabara, Kyoto University, Kyoto, Japan) digested with the same enzymes. The resulting plasmids were verified by nucleotide sequencing. The expression plasmids for PP2Cγ and the mutants were also constructed by inserting the PP2Cγ sequence () into pHIT51. Each plasmid was introduced into BL21-Gold (Stratagene), and expression was induced with 1 mM IPTG at 30°C. The His-tagged proteins were purified using Ni-agarose beads (Sigma-Aldrich) according to the standard protocol by the manufacturer followed by MonoQ chromatography (GE Healthcare) and elution with a linear NaCl gradient (0.1–1 M NaCl in 50 mM Tris-Cl, pH 8.0). For the phosphatase assay of bulk histones, histones purified from 20 μg HeLa cells were phosphorylated using 1 μg MSK1 (Upstate Biotechnology) in 12 mM MOPS, pH 7.0, 15 mM MgCl, 0.2 mM EDTA, 1 mM EGTA, 0.2 mM DTT, 0.1 mM ATP, and 7.4 MBq/ml γ-[P]ATP for 10 min at 30°C. The unincorporated ATP was removed, and the buffer was substituted with TMD (10 mM Tris-Cl, pH 8.0, 10 mM MgCl, and 1 mM DTT) using Ultrafree filters 0.5 (Millipore). The phosphorylated histones (2 μg in 10 μl) were mixed with a serial dilution of His-tagged proteins (2.5 μl) and incubated for 1 h at 37°C. After stopping the reactions by adding 12.5 μl of 2× SDS gel loading buffer () and boiling, the samples were separated by SDS-PAGE and stained with Coomassie. The radioactivity was detected using an imaging analyzer (BAS2000; Fujifilm). For the γ-H2AX dephosphorylation assay, the γ-H2AX–containing H2A–H2B fraction was prepared from HeLa cells irradiated (12 Gy) using a Cs source at a dose rate of 1.13 Gy/min (Gammacell 40 Exactor; MDS Nordion). The H2A–H2B sample (1 μg in 10 μl) was mixed with 2.5 μl phosphatases in TMD buffer, incubated for 30 min at 37°C, and the level of γ-H2AX was analyzed by immunoblotting using antiphospho-H2AX Ser139 antibody (1:1,000; Upstate Biotechnology). To immunoprecipitate GFP-H2A and its binding proteins, 1 ml S100 extract from HeLa cells (control) or cells expressing GFP-H2A was mixed with 50 μl anti-GFP agarose beads (Nacalai Tesque). After incubation for 1.5 h at 4°C with rotation, the beads were collected by centrifugation at 1,600 for 5 min at 4°C. After washing four times for 10 min at 4°C in PB containing 0.05% Tween 20, 0.2 M NaCl, and protease inhibitor cocktail (Nacalai Tesque), the immunoprecipitates were eluted from the beads by boiling for 10 min in 60 μl of 2× SDS gel loading buffer (). For , the FLAG-PP2Cγ, -ΔAcDo, and -D496A plasmids were generated by inserting the corresponding cDNAs into a modified version of pcDNA3.1/Hygro (Invitrogen) that contains N-terminal FLAG and V5 tags. 293T cells (4 × 90-cm dishes; 20% confluent) were transfected with these constructs using a calcium phosphate precipitation method (). 3 d later, cells were washed with ice-cold PB, lysed in 2 ml PB containing 0.1% Triton X-100 and protease inhibitor cocktail, incubated for 5 min on ice, and cleared by centrifugation at 1,600 for 10 min at 4°C. The supernatant was collected and mixed with 100 μl anti-FLAG agarose M2 beads (Sigma-Aldrich). After incubation and washing in the same buffer four times for 10 min at 4°C, the immunoprecipitated material was eluted with 100 μg/ml 3× FLAG peptide in PB (three times at 150 μl). The elution was pooled and either mixed with 2× SDS gel loading buffer for SDS-PAGE or with 20 mg/ml Casamino acids (final concentration of 100 μg/ml; Difco) and 100% trichloroacetic acid (final concentration of 20%) for AUT gel electrophoresis. After incubation for 1 h on ice and centrifugation at 20,000 for 30 min at 4°C, the pellet was washed with acetone chilled at −20°C and dissolved in AUT sample buffer (). Immunoblotting was performed as described previously () using the following primary antibodies: rabbit antiphospho-H2A/H4 Ser1 (1:1,000; Upstate Biotechnology), mouse antiphospho-H2AX Ser139 (1:1,000; Upstate Biotechnology), and mouse antiphospho-H2B Ser14 (clone 6C9; 1:20 hybridoma supernatant). To produce antiphospho-H2B Ser14, mice were immunized with a synthetic peptide KSAPAPKKG(phospho-S)KKAVTKAQKC (Sigma-Genosys) coupled to keyhole limpet hemocyanin (), and a hybridoma clone 6C9 was obtained by ELISA screening using the phosphorylated and unphosphorylated peptides. As H2B Ser14 is phosphorylated during apoptosis (), the specificity was then checked by the specific appearance of positive signals in apoptosis-induced (etoposide treated) HeLa cells by immunoblotting and immunofluorescence. PP2Cγ-specific Stealth RNA (Invitrogen; nucleotide number 351-376 or 642-667 of GenBank EMBL/DDBJ accession no. ) and the control RNA (Invitrogen; number 12935-300) were transfected using LipofectAMINE2000 (Invitrogen). Total cellular proteins were prepared 1–3 d after transfection, separated on an 8% SDS-polyacrylamide gel, and immunoblotted () with mouse monoclonal antibody directed against PP2Cγ (1:10,000; ) or α-tubulin (1:1,000; Oncogene Research Products) as a control. Cells grown on coverslips were transfected with Stealth RNA and fixed for immunofluorescence using the mouse anti-PP2Cγ (1:30,000) and Cy3-conjugated anti–mouse IgG (1:500; Jackson ImmunoResearch Laboratories). For photobleaching studies, HeLa cells expressing GFP-H2A, H2B-GFP, or H1c-GFP () grown on glass-bottom dishes (Mat-Tek) were transfected with Stealth RNA. 3 d later, the dish was set on an inverted microscope (LSM510 META; Carl Zeiss MicroImaging, Inc.) in an air chamber at 37°C, and the mobility was analyzed by photobleaching using the inverted microscope with a plan-Neofluar 40× NA 1.3 objective. For H2A and H2B, five confocal images were collected (512 × 512 pixels, zoom 3, maximum scan speed, pinhole 3.7 airy unit, LP505 emission filter, and 0.3% transmission of 458-nm Ar laser with 75% output power), one half of a nucleus was bleached using 100% transmission of 458 and 488 nm (eight iterations), and images were collected using the original setting every 5 min. For H1c, five images were collected (256 × 256 pixels, zoom 8, and scan speed 12), a 2-μm spot was bleached using 100% transmission of 458 and 488 nm (eight iterations), and images were collected every 5 s (the graph in shows the points of every 10 s for ease of comparison). The fluorescence intensity of the bleached area was measured using MetaMorph software (Molecular Devices). After subtracting the background, the intensity was normalized to the initial intensity before bleaching. PP2Cγ-deficient DT40 cells were established using standard methods (Fig. S3; ) and grown at 37°C. To measure the cell density, cells were mixed with trypan blue solution (Invitrogen), and the number of live cells excluding the dye was counted. To determine the sensitivity to caffeine and irradiation, serially diluted cells were plated in methylcellulose plates with or without 1 mM caffeine (Sigma-Aldrich) and irradiated using a Gammacell 40 Exactor (Nordion). Colonies were counted 10–12 d after plating. For immunoblotting (), 4 × 10 cells/ml were irradiated, and calyculin A (Sigma-Aldrich) was immediately added (final concentration of 10 ng/ml). A 1-ml aliquot was taken at each time point, and cells were collected (600 g for 2 min) and lysed in 100 μl of 2× SDS gel loading buffer. Fig. S1 shows that ATP is not required for GFP-H2A incorporation into chromatin in permeabilized cells assisted by PP2Cγ or Nap1. Fig. S2 shows that PP2Cγ has weak de novo nucleosome assembly activity. Fig. S3 shows evidence for the generation of PP2Cγ knockout DT40 cells. Online supplemental material is available at .

The mammalian cell nucleus is a complex structure containing distinct nuclear bodies (NBs), such as nucleoli, PML bodies, splicing speckles, and Cajal bodies (CBs; ; ; ). These bodies are typically enriched for specific proteins and nucleic acids, reflecting their function. Dynamic changes in NBs occur under both physiological and pathological conditions. For example, both the number and size of nucleoli vary between metabolically active and inactive cells, and PML bodies are altered in leukemic blasts and during virus infection (). The molecular events triggering such changes are not well characterized. We examine the dynamic behavior of CBs and how their composition changes under stress conditions. CBs were discovered in 1903 by Santiago Ramón y Cajal () and are involved in the assembly and maturation of small nuclear ribonucleoproteins (snRNPs; ; ). Indeed, snRNPs are thought to accumulate in CBs upon their initial entry into the nucleus (), and a class of CB-specific modification guide RNAs (scaRNAs) are important for the sequence-specific modification of the snRNAs within CBs (; ). CBs also contain survival of motor neuron (SMN), a protein linked to the neurodegenerative disease spinal muscular atrophy (). The SMN complex plays an important role in the cytoplasmic assembly of Sm core RNPs, () and in their nuclear reimport and targeting to CBs (; ). Other CB components include fibrillarin and NOPP140, proteins that localize in CBs before their subsequent accumulation in nucleoli. CBs likely have other functions besides snRNP maturation. For example, NPAT and PTFγ, which are proteins regulating histone and snRNA gene transcription, respectively, are found in CBs (; ). CBs can indeed associate with histone and snRNA gene loci (), and they may also play a more general role in coordinating assembly of large multiprotein complexes in the nucleus (). Interestingly, the presence in a subset of CBs of ZPR1 and FGF-2 suggests that CBs could be involved also in transducing proliferative signals to the nucleus (; ). Genetic evidence suggests that coilin, a nuclear phosphoprotein widely used as a marker for CBs, plays a role in the structural organization of CBs. Thus, in coilin knockout cells, CBs are disrupted and fail to accumulate snRNPs and SMN, whereas other CB components, such as fibrillarin, NOPP140, and scaRNAs, are redistributed in distinct subsets of remnant structures (; ). Posttranslational modifications of coilin can affect CB integrity. For example, changes in the phosphorylation state of coilin affect the number and integrity of CBs in mitotic and interphase cells (; ), and the extent of symmetrical dimethylation of arginine residues on coilin influences the targeting of SMN and, consequently, the accumulation of newly imported snRNPs in CBs (; ). However, coilin modification is not always linked with CB disassembly or turnover. For example, adenovirus infection causes fragmentation of CBs () without causing changes either in the levels or in the electrophoretic mobility of coilin. UV irradiation represents a complex, multicomponent stress stimulus that subverts the metabolic activity of the cell nucleus. It affects different nuclear domains including nucleoli (; ) and PML bodies (; ). UV light irradiation causes an immediate ligand-independent activation of receptor tyrosine kinases (i.e., EGF and PDGF receptors) caused by the inactivation of receptor-directed tyrosine phosphatases (; ). Subsequently, it triggers DNA damage caused by the formation of cyclobutane pyrimidine dimers, (6–4) photoproducts (), and reactive oxygen species generation (), and a complex transcriptional response involving modulation of genes associated with cell proliferation and repair of damaged DNA (). The next phase is characterized either by repair of DNA lesions, or by apoptosis of the cells that have not been able to start an appropriate response. In this study, we show that CBs are also responsive to UV-C and characterize the molecular mechanism underlying this effect. We have identified PA28γ (proteasome activator subunit 28S γ) as a factor whose stable association with coilin-containing complexes is increased by UV-C treatment and show that PA28γ plays an important role in the mechanism underlying the disruption of CBs upon UV-C irradiation. The effect of UV-C treatment on CB integrity was assessed by immunofluorescence labeling using antibodies specific for coilin (, top left). HaCaT (human immortalized keratinocytes) cells were immunolabeled at 6 h after brief exposure to UV-C (30 J/m, 254 nm; see Materials and methods). This changed the number and appearance of CBs, with coilin redistributed to hundreds of microfoci clustered throughout the nucleoplasm (, bottom left). Splicing speckles were also affected by UV-C, although less dramatically than CBs; thus, UV-C caused splicing speckles to become more rounded, but with little or no change in the mean number per nucleus (, compare middle images). We noticed a similarity between the redistributed coilin labeling pattern and the rounded splicing speckles present after UV-C treatment. This was unexpected because coilin does not normally colocalize with splicing speckles (). To investigate this further, we double labeled UV-C–treated cells with anti-coilin antibodies (, left) and antibodies specific for known splicing speckle components, i.e., anti-SC35, anti-Sm (Y12), anti– trimethylguanosine (TMG)-cap, and anti-U1A (, middle). All four of these splicing speckle components showed a similar change into more rounded structures after UV-C treatment, and we observed a partial overlap between the reorganized coilin microfoci and the rounded splicing speckles (, compare left and middle). These experiments involved immunolabeling cells at 6 h after exposure to UV-C because this was empirically observed as causing the maximum effect on coilin redistribution (). Moreover, the UV-C effect is at least partially reversible, with recovery of the normal coilin distribution evident in ∼40% of the irradiated cells within 10–12 h. A subset of the irradiated cells, at 8–12 h after UV-C, formed coilin-containing perinucleolar caps (unpublished data). Further analysis confirmed that the majority of cells remained viable after UV-C exposure, with no correlation between coilin redistribution and UV-C–induced cell death (unpublished data). We conclude that UV-C treatment has a specific and reversible effect on subnuclear organization affecting CBs and splicing speckles. Next, we evaluated the effects of UV-C irradiation on a variety of cell lines, differing in tissue of origin, growth properties, and transformation status, to determine if the UV-C–induced effect on CB structure is either general, or specific for HaCaT cells (). This shows a similar UV-C–induced redistribution of coilin in all the cell lines tested, including WI-38, HeLa, COS-7, HaCaT, MCF-7, SAOS-2, 293T, HCT116, and SW480, with little variation in the percentage of responsive cells between each cell type (). A functional p53 gene is not required for the coilin redistribution in response to UV-C because whereas the SAOS-2 cells are p53 null, the HaCaT cells have a mutated, functionally impaired p53 gene, and the MCF-7 cells have a functional, wild-type p53 gene. In subsequent experiments, we concentrated on the HaCaT cells as the main experimental model to investigate the mechanisms of CB disassembly after UV-C irradiation because those cells were the most responsive to UV-C treatment and also because, among those tested, they represent the tissues most exposed to sunlight in vivo. To better characterize the effect of UV-C irradiation on CBs, we next addressed whether UV-C irradiation affected other CB proteins, in both HaCaT and MCF-7 cells, namely SMN, snRNP components, fibrillarin, NOPP140, and NPAT (). After UV-C irradiation, both SMN and snRNP factors are no longer concentrated in CBs (). SMN and the U6 snRNP-associated factor SART3 show a widespread diffuse staining, whereas TMG-capped snRNAs and snRNP proteins show both nucleoplasmic staining and accumulation into rounded splicing speckles (; arrows and arrowheads indicate intact and rearranged CBs, respectively). Surprisingly, after UV-C irradiation, CB-like structures are still evident as bright, coilin-negative nucleoplasmic bodies when detected with antibodies against either NPAT, fibrillarin, or NOPP140 (, left, arrows). Thus, UV-C irradiation selectively affects a subset of CB components. The small CB-like structures, which remain after UV-C and lack coilin, SMN, and snRNPs, resemble in appearance and composition the residual CBs observed in coilin −/− mouse cells (; ). It is known that transcription inhibitors disrupt CBs and cause speckles to enlarge and round up (for review see ). This led us to examine whether the effect of UV-C exposure on transcription levels could explain the observed changes in CBs and in the distribution of coilin in the irradiated cells. A 5-fluorouridine (5-FU) incorporation assay () confirmed, as expected, that there was a time-dependent general reduction in nuclear RNA synthesis after exposure to UV-C. The UV-C inhibition affected primarily nucleoplasmic transcription, with 5-FU incorporation within nucleoli still visible. Double-labeling with anti-coilin antibodies, again, showed a time-dependent disruption of CBs after UV-C treatment, with most coilin relocalized to nucleoplasmic microfoci. In some cells, a minor fraction of coilin also localizes in dots within nucleoli, distinct from perinucleolar caps (, insets). We compared the UV-C response directly with the effect of the transcription inhibitors 5,6-dichloro-1-β-D-ribobenzimidazole (DRB) and actinomycin D ( and not depicted). Both inhibitors cause coilin to relocalize and accumulate in prominent perinucleolar caps (, bottom, arrowheads). However, we do not observe accumulation of coilin in nucleoplasmic microfoci or associated with splicing speckles after exposure to transcription inhibitors, which differs from the effect of UV-C exposure. Therefore, although some of the effects seen after UV-C exposure, such as the formation of large, rounded splicing speckles and possibly the appearance of coilin dots within nucleoli, may be attributed to the down-regulation of transcription, our data suggest that the UV-C disruption of CBs is not simply an indirect result of transcription inhibition. Because genetic depletion of murine coilin causes a similar change in CB appearance and composition to that detected here after UV-C treatment, we tested whether UV-C irradiation affected either the levels or modification state of coilin. Protein blotting analysis using anti-coilin antibodies to probe either whole-cell lysates, or soluble nuclear extracts prepared from both untreated and UV-C–treated cells, revealed little or no change in the relative levels of coilin, even under different extraction conditions (, compare lanes 1–5 and 6–10). Moreover, the absence of obvious changes in the electrophoretic mobility of coilin suggests that it does not undergo a major change in its phosphorylation status. In contrast, a clear shift in electrophoretic mobility of coilin was observed upon its hyperphosphorylation when cells enter mitosis (). Symmetric dimethylarginine (sDMA) modification of coilin has been linked to the ability to recruit SMN to coilin-containing CBs. Either drug-induced hypomethylation of coilin (; ), or transient expression of mutant forms of the protein (), prevented SMN recruitment to CBs but, notably, did not cause CBs to fragment. Therefore, we investigated whether sDMA modification of coilin was affected by UV-C irradiation. Coilin was immunoprecipitated from nuclear extracts prepared from either control or UV-C irradiated cells, and the eluted proteins were blotted and probed with both anti-coilin and anti-sDMA antibodies (). This showed a minor reduction in sDMA levels. However, when whole-cell lysates obtained in denaturing conditions (9 M urea) from irradiated and mock-treated cells were separated by SDS-PAGE, transferred to nitrocellulose filters, and probed with anti-coilin and anti-sDMA antibodies, any differences in the extent of sDMA modification were less evident (). Given the absence of CB fragmentation in cells with hypomethylated coilin (), we infer that although minor changes in the sDMA status of coilin may take place in UV-C–treated cells, this does not likely play a major role in determining the fragmentation of CBs. We tested whether UV-C irradiation causes a change in the protein composition of nuclear complexes containing coilin. Coilin complexes were affinity purified from cells expressing FLAG-tagged coilin (see Materials and methods). Silver staining of the eluted, affinity-purified material from both mock-treated and UV-C–treated cells revealed that a protein band of ∼32 kD was enriched specifically in the eluate from the irradiated cells (, arrow). Peptide mass fingerprinting was performed on the enriched band isolated from a large-scale preparation of UV-C–treated cells and identified as proteasome activator subunit γ (PA28γ). To validate the identification of PA28γ as a novel component of coilin complexes, and to test whether coilin and PA28γ associate in the absence of UV-C, immunoprecipitation experiments were performed with anti-PA28γ polyclonal antibodies and extracts from nonirradiated HaCaT and MCF-7 cells. The immunoprecipitated material was separated by SDS-PAGE, transferred to a nitrocellulose membrane, and probed with anti-coilin antibodies. Coilin was coimmunoprecipitated with PA28γ (, compare lanes 2 and 3), confirming that coilin and PA28γ are present in a common complex in vivo, even in the absence of UV-C treatment. Furthermore, the complex containing PA28γ and coilin was resistant to 1 M KCl, indicating a stable interaction (,). Gel filtration analysis of nuclear extracts reveals that, under steady-state conditions, coilin is present in two different regions of the column eluate (). The first region corresponds to very high molecular weight complexes, of ∼2 MD (, lanes 8–10). The second region corresponds to complexes ∼200–300 kD in size (, lanes 13–16). Consistent with previous studies (), we found PA28γ in molecular weight fractions corresponding to a size range of ∼250 kD (, lanes 12–17). Therefore, PA28γ cofractionates with the smaller-sized pool of coilin complexes. Immunoprecipitation with anti-PA28γ antibodies in extracts prepared from mock-treated and UV-C–treated cells confirmed the result shown in , i.e., that the level of PA28γ isolated in a stable complex with coilin is significantly increased by UV-C treatment (, compare lanes 5 and 7; and Fig. S1, available at ). Thus, the amount of coilin complexed with PA28γ is maximal at 6 h after UV-C treatment and decreases at later time points (12 h; , lane 9), when the majority of the cells show complete or partial recovery of the normal coilin labeling pattern. Under these conditions we observe only a small increase in the total cellular level of PA28γ after UV-C, as judged by immunoblotting (), and infer that the enhanced PA28γ association with coilin is unlikely to arise only through changes in PA28γ protein levels (see Discussion). The observed association between coilin and PA28γ in control, nonirradiated cells likely does not occur within CBs, as we observe by immunofluorescence no accumulation of PA28γ in CBs (, top). However, upon UV-C treatment, coilin and PA28γ show enhanced colocalization in the irradiated cells, supporting the aforementioned biochemical evidence for a UV-C–induced increase in the association between coilin and PA28γ (, bottom, arrowheads). Consistent with our observation that coilin partially colocalizes with rounded splicing speckles in UV-C–treated cells (), a double-immunofluorescence labeling with anti-PA28γ and –SC35 antibodies revealed that PA28γ also shows a UV-C–induced partial colocalization with splicing speckles (, arrowheads). This is further supported by double-labeling experiments showing partial colocalization of PA28γ with SC35, U1A, and TMG cap RNAs in UV-C–treated cells (Fig. S2, available at ). Collectively, the data indicate that coilin and PA28γ are present in a common complex in vivo whose levels are increased by UV-C treatment. We hypothesized that the increased formation of complexes containing coilin and PA28γ upon UV-C treatment could contribute to the mechanism of UV-C–induced CB disruption and coilin redistribution. Therefore, we studied whether altering PA28γ expression levels could affect CB integrity and the distribution of coilin in the absence of UV-C treatment. Strikingly, transient overexpression of exogenous PA28γ in both HaCaT and MCF-7 cells triggered a similar redistribution of coilin to that observed in UV-C–irradiated cells (). Thus, the more brightly stained cells overexpressing PA28γ (, arrowheads) have lost the bright coilin foci (arrows) seen in untransfected cells. To test whether overexpression of PA28γ also induces a reduction in transcription, similar to that caused by UV-C treatment, we analyzed 5-FU incorporation in both transfected and untransfected cells (). This showed clearly that cells overexpressing PA28γ had similar levels of nuclear transcription to untransfected cells (, bottom). This thus demonstrates that the PA28γ-mediated disruption of CBs and relocalization of coilin is independent of changes in transcription levels. First, because growth stimuli can increase PA28γ levels (), we treated serum-starved MCF-7 cells for 72 h with either PDGF or EGF at the indicated doses for 30 h ( and not depicted). The majority (>80%) of treated cells showed a brighter signal for PA28γ in the nucleus (, left), and almost all cells with an increased PA28γ signal showed a redistribution of coilin identical to that elicited by UV-C irradiation (, bottom). Second, because PA28γ levels are increased in thyroid neoplasms and correlate with the tumor stage (being low in differentiated and very high in anaplastic thyroid carcinoma; ), we tested whether PA28γ levels were also elevated in a thyroid tumor–derived cell line, FTC133 (follicular thyroid carcinoma; ). Immunofluorescence analysis revealed high levels of endogenous PA28γ in FTC133 nuclei in unstressed control cells (, top). Intriguingly, in most FTC133 cells CBs are fragmented and coilin is distributed in a widespread nucleoplasmic pattern similar to that observed in other cell lines only after UV-C irradiation. Furthermore, treatment of the FTC-133 cells with interferon γ (IFNγ), a stimulus known to down-regulate PA28γ levels (), for 72 h caused down-regulation of nuclear PA28γ and the reappearance of intact CBs in >70% of the treated cells (, bottom). Overexpression of PA28γ recapitulates many, but not all, of the changes in subnuclear structure induced by UV-C treatment. Thus, TMG-capped snRNAs, SMN and SART3 (), were each relocalized out of CBs upon transient overexpression of PA28γ, whereas NPAT, fibrillarin, and NOPP140 remained localized in residual CBs (). However, it did not cause snRNPs to accumulate in enlarged splicing speckles. A likely explanation for this difference is that PA28γ overexpression, unlike UV-C, does not cause inhibition of transcription (). We observe that the effect of PA28γ is specific for CBs. Thus, transient overexpression of PA28γ had little or no effect on the number, integrity, or morphology of PML bodies, paraspeckles, or nucleoli in MCF-7 cells (). In summary, multiple independent lines of evidence strongly suggest that PA28γ can specifically affect CB integrity and coilin subnuclear localization. If PA28γ is the mediator of the UV-C effect on CBs, we postulated that RNA interference (RNAi)–mediated knock down of PA28γ should attenuate this effect. Therefore, we performed PA28γ knockdown experiments in both MCF-7 and HaCaT cells before and after UV-C treatment ( and not depicted). Initial experiments identified two double-stranded RNA oligonucleotides able to reduce the levels of PA28γ in transfected cells after 72 h. To monitor the transfection efficiency, a plasmid vector expressing YFP alone was cotransfected together with the RNAi duplexes. Protein blotting analysis showed that both RNAi duplexes specifically decreased PA28γ protein levels, but not the levels of either endogenous lamin B1 or transiently expressed YFP (). Immunofluorescence analysis with anti-coilin antibodies showed that the degree of UV-C–induced disassembly of CBs and coilin redistribution was reduced (>50% of the transfected cells do not show CB disassembly upon UV-C) in the cells transfected with either of the specific RNAi oligos (, arrows). In contrast, cells transfected with the control oligonucleotides are indistinguishable from the adjacent untransfected cells in terms of their UV-C response (, bottom). We note that neither the specific nor the control siRNAs altered the integrity of CBs in mock-treated cells (unpublished data). We conclude that PA28γ contributes to the mechanism that modulates the response and/or the sensitivity of CBs to UV-C irradiation. In this study, we have detected a novel effect of UV-C irradiation, showing that it disrupts CBs, causing a selective redistribution of a subset of CB components. Coilin, snRNPs, and other CB proteins involved in snRNP assembly and/or maturation are displaced from CBs after UV-C treatment, whereas other CB factors, including fibrillarin, NOPP140, and NPAT, remain in residual CB-like bodies. This effect is reversible, with maximum disruption evident within 6 h after treatment of cell lines in culture with UV-C and partial recovery within 12 h. We identify a novel coilin-associated factor, PA28γ, whose association with coilin complexes is increased by UV-C treatment, and we show that it is an important mediator of the molecular mechanism leading to UV-C–induced CB fragmentation. Furthermore, a transient increase in the level of PA28γ expression, in the absence of UV-C, is sufficient to trigger a similar CB disruption and relocalization of the same subset of CB components to that seen in response to UV-C. PA28γ was originally discovered as a target of human autoantibodies in serum from patients suffering from systemic lupus erythematosus and as a proliferation-associated antigen (Ki antigen; , ). It was named PA28γ because of its homology to the known proteasome activators PA28α and β, which share ∼40% amino acid identity. This similarity was also supported by the independent observation that PA28γ enhances the trypsin-like activity of the proteasome toward small peptides in vitro (). However, it remains unclear whether the only in vivo role of PA28γ is to activate the proteasome. For example, purified PA28γ activates the isolated proteasome to a lesser extent than PA28α and β, (). Moreover, although expression of both PA28α and β is up-regulated by IFNγ, the protein levels of PA28γ, in contrast, drop dramatically after treatment with this chemokine (). This effect of IFNγ reducing the levels of PA28γ was also observed in this study using FTC133 cells. Another difference is provided by the phenotype of both PA28 α and β −/− mice, which have major immunological problems, including impaired antigen processing, cytotoxic T cell activation, and assembly of the proteasome subunits (). This phenotype is not evident in either of the two PA28γ knockout mice models generated, only one of which showed relatively modest immune defects, namely a slight reduction in the levels of CD8+ T cells and slower clearance of an experimental lung infection (); in both cases, PA28γ −/− mice are viable (; ). We provide evidence demonstrating both a physical and a functional link between coilin and PA28γ. Evidence for a physical link came from biochemical affinity purification and proteomic analysis, which revealed PA28γ as the most prominent protein whose copurification with FLAG-tagged coilin was increased by UV-C irradiation. The endogenous forms of both proteins are also coimmunoprecipitated with antibodies specific for either coilin or PA28γ, and they form a stable, salt-resistant complex in vivo, even in the absence of UV-C irradiation. We examined whether PA28γ binds directly to coilin using several independent in vitro assays and purified recombinant proteins (unpublished data). In all cases, we see no evidence for direct binding. Therefore, although we cannot exclude that these negative results are caused by technical limitations in the binding assays, it is also possible that the association between coilin and PA28γ in vivo involves additional nuclear components and/or posttranslational modifications of one or both proteins. Although PA28γ associates with coilin-containing complexes in the absence of UV-C treatment, we do not observe colocalization or accumulation of PA28γ in CBs. Instead, PA28γ in untreated cells is widely distributed throughout the nucleoplasm. Despite the obvious concentration of coilin in bright CB foci when viewed by immunofluorescence, it is known that the major fraction of coilin is also present in a diffuse nucleoplasmic pool (). We infer from this that the complexes containing both coilin and PA28γ are present in this diffuse nucleoplasmic pool. Considering that our gel filtration chromatography experiments showed that only the smaller size class of coilin complexes cofractionates with PA28γ, it is possible that these smaller complexes may form at least part of this pool. The level of complexes containing coilin and PA28γ clearly increases after UV-C treatment. Interestingly, the kinetics of this complex formation tightly parallel the kinetics of CB disruption, which is consistent with a functional connection. Further evidence for a functional role for PA28γ in CB integrity was provided by our discovery that RNAi-mediated knockdown of PA28γ either prevents or reduces the effect of UV-C treatment on CBs in irradiated cells. Although loss of PA28γ by RNAi prevents the UV-C response, we also show that an increase in the levels of PA28γ, either through transient overexpression, growth factor treatment, or in the transformed cell line FTC133, is sufficient to trigger a similar disruption of a subset of CB components to that induced by UV-C. However, it is unlikely that the effect of UV-C irradiation on CBs results from a large increase in intracellular PA28γ levels. Indeed, we observe only a small increase in PA28γ levels during the time course of the UV-C response. Instead UV-C may lead to an increase in affinity of PA28γ for the coilin complexes. For example a UV-C–induced phosphorylation mechanism could be involved. In this regard, it is interesting that a physical and functional interaction of PA28γ with MEKK3 has been reported (). MEKK3 is a kinase involved in transducing both mitogenic (serum and FGF-2) and stress signals (UV-C and osmotic stress; ; ; ). Our preliminary results indicate that overexpression of MEKK3 increases the UV-C–induced interaction of PA28γ and coilin (unpublished data). Furthermore, we have also observed that hyperosmotic stress, induced by treating both HaCaT and MCF7 cells with 0.5 M sorbitol, resulted in a similar disruption of coilin from CBs to that caused by UV-C (Fig. S3, available at ). The fact that MEKK3 is known to be activated by both UV-C and hyperosmotic shock may not be a coincidence. Therefore, future studies will examine in more detail the role and mechanism of PA28γ in regulating CB integrity and the potential role of MEKK3 in signaling to CBs. Interestingly, we observe that overexpression of PA28γ causes CB disruption without reducing nuclear transcription levels. It also acts specifically on CBs without changing the morphology or number of PML bodies, paraspeckles, or nucleoli. Based on our results, we infer that the novel effect of UV-C we have detected on CBs involves a specific PA28γ-dependent pathway. Although UV stress causes a complex set of cellular responses, affecting multiple targets and down-regulating nuclear transcription, our analysis of PA28γ should allow us to dissect a subset of these cellular events and the mechanisms involved. WI 38, HeLa, COS-7, HaCaT, SAOS-2, and 293T cells were cultured as monolayers in DME (Invitrogen) supplemented with 10% fetal bovine serum, nonessential amino acids, penicillin-streptomycin, and -glutamine (Invitrogen) in a humidified incubator at 37°C with 5% CO. HCT116 and SW480 cells were cultured as monolayers in McCoy 5A medium (Invitrogen), supplemented like the aforementioned cells. MCF-7 cells were grown in RPMI medium (Invitrogen) supplemented as like the aforementioned cells. PDGF and EGF (Millipore) were added to the medium of serum-starved MCF-7 cells to a final concentration of 25 and 100 ng/ml, respectively. DRB was added to the cells at a final concentration of 30 μg/ml, for 3 h. Actinomycin D was added to the cells at a final concentration of 10 μg/ml for 2 h. The PCDNA3-FLAG-PA28γ expression vector was a kind gift of M. Rechsteiner (University of Utah School of Medicine, Salt Lake City, UT). For PA28γ RNAi experiments, the following oligonucleotides targeting PA28γ have been used: OLIGO 1 (5-GAA GCC UUC CAA GGA ACC ATT-3) and OLIGO 2 (5-ACA UCC AUG ACC UAA CUC ATT-3; MWG Biotech). As a control (“off-target”), a dsRNA oligonucleotide targeting the luciferase gene was used (5-CGU ACG CGG AAU ACU UCGA-3). dsRNA oligonucleotides were transfected by using the RNAiFect transfection reagent (QIAGEN). dsRNAs were cotransfected with a highly purified (CsCl gradient) plasmid vector encoding YFP (p-EYFP-C1; CLONTECH Laboratories, Inc.), at a molar ratio of 1:40 (YFP: RNAi oligonucleotide) to facilitate the identification of transfected cells and to evaluate the effect of the transfection on the viability of the cells. Except when stated otherwise, all of the chemicals used were from Sigma-Aldrich. Subcellular fractions were obtained from either mock- or UV-C–treated cells as follows: in brief, cell nuclei isolated in hypotonic/detergent containing buffer were subjected to DNase/RNase treatment and high salt extraction with (NH)SO and NaCl, respectively. The remaining material was solubilized with urea-containing buffer (10 mM Tris-Cl, pH 8.0, 9 M urea, IGEPAL AC-630, 100 mM KCl, 50 mM DTT, 50 mM NaF, 1 mM NaVO, 1 mM MgCl). Before SDS-PAGE analysis, the final salt concentration of the samples was normalized by the addition of 2 M (NH)SO. In brief, semiconfluent cells were washed with PBS and incubated for 8–12 h in DME (or RPMI or McCoy 5A) supplemented with 0.5% FBS (serum starvation). Medium was collected and kept at 37°C and the PBS-washed cells irradiated in a UV Stratalinker 2400 oven with 254-nm bulbs (Stratagene) at 30 J/m. After that, the old medium was quickly added back into the dishes and the cells were incubated for the indicated times. Semiconfluent cells were washed with PBS and incubated for 8–12 h in DME (or RPMI or McCoy 5A) supplemented with 0.5% FBS (serum starvation). After that, medium was collected and cells were washed once with PBS and either mock-treated (old medium added back) or incubated with 0.5 M sorbitol (Sigma-Aldrich) dissolved in the previously collected medium for the indicated times. Polyclonal antibodies used were as follows: anti-coilin 204/10 (1:300; ; [top and middle] and C; [top and middle] and C [middle and bottom]; ; ; ; ), anti-NPAT (1:500; a gift from J. Zhao, University of Rochester, Rochester, NY; ; ) anti–SART3 (1:200; a gift from D. Stanek and K. Neugebauer, Max Planck Institute, Dresden, Germany; ; ); anti-PSP1 (1:400; ; ); and anti-PA28γ (1:400; Affiniti Research Products; ; ; Fig. S2, top and bottom). Monoclonal antibodies used were as follows: anti-coilin 5P10 (1:300; bottom; [bottom] and C [top]; ); anti-fibrillarin 72B9 (1:25; ; ; ), anti-SC35 (1:400; Sigma-Aldrich; , A and B; ; Fig. S2); anti PSME3 (PA28γ; 1:100; BD Biosciences; ; ; Fig. S2, bottom); anti-SMN (1:100; BD Biosciences; ; ); anti-2,2,7 TMG (1:300; Oncogene; ; ; Fig. S2); anti-PML (1:100; Santa Cruz Biotechnologies, Inc.; ). HaCaT or MCF-7 cells, either mock-treated or treated with UV-C or DRB for the indicated length of time, were incubated with 2 mM 5-FU (F5130) for 30 min at 37°C. Subsequently, cells were fixed, permeabilized, and incubated with primary anti-BrdU antibody (B2531; 1:500). Immunofluorescence microscopy was performed as indicated (see previous section). For transfected cells, 5-FU labeling was performed at 24 h after transfection. In brief, enzymatically detached cells were pelleted and resuspended in 2× reducing loading buffer (Invitrogen) and denatured at 80°C for 5 min. When indicated, cells were previously resuspended in urea-containing buffer (see Biochemical fractionation of isolated nuclei). For Western blot analysis, proteins resolved by SDS-PAGE were transferred to nitrocellulose filters (Schleicher and Schuell) and probed with specific primary antibodies and the corresponding horseradish peroxidase-conjugated secondary antibodies (GE Healthcare). An enhanced chemiluminescence reagent (ECL; GE Healthcare) was used to visualize protein bands, according to manufacturer's instructions. Red Ponceau staining of the transferred proteins was used to assess equal loading of the samples. Monoclonal antibodies used for Western blot analysis were as follows: anti-coilin 5P10 (1:100; ; ; ); anti-LaminB1 (1:200; Zymed Laboratories; ); anti-GFP (1:500; Roche; ), anti-PSME3 (PA28γ; 1:500; BD Biosciences; ; ). Polyclonal antibodies used for Western blot analysis were as follows: anti-coilin 204/10 (1:2,000 ; ; ), anti-PA28γ (1:2,000; Affiniti Research Products; ; ); and anti-sDMA (SYM10; 1:500; Millipore; ). Enzymatically detached cells were pelleted and resuspended in hypotonic buffer (10 mM Tris-Cl, pH 7.1, 10% glycerol, 4 mM DTT, 50 mM NaF, 1 mM NaVO, 1 mM MgCl, and Roche protease inhibitors). IGEPAL AC-630 was added dropwise while mixing to a final concentration of 0.5%. Cells were gently resuspended for 3 min on ice and crude nuclei were pelleted by centrifugation (3,000 rpm for 5 min at 4°C) and resuspended in two bead volumes of ice-cold digestion buffer (2 mM Tris-Cl, pH 8.5, 20% glycerol, 10 mM DTT, 50 mM NaF, 1 mM NaVO, 1 mM MgCl, 5 mM CaCl, Roche protease inhibitors, and 75 U/ml micrococcal nuclease; GE Healthcare). Samples were allowed to digest at 25°C for 15 min with continuous mixing, and then ice-cold extraction buffer (2 mM Tris-Cl, pH 8.5, 50 mM NaF, 1 mM NaVO, 1 mM MgCl, 20 mM EDTA, pH 8.0, 0.84 M KCl, and Roche protease inhibitors) was added vol/vol. Samples were then incubated on ice with frequent mixing for 20 min. Finally, the nuclear lysate was clarified by ultracentrifugation at 50,000 rpm for 30 min at 4°C in a fixed-angle rotor (TLA 100.3; Beckman Coulter). Supernatant was collected and used immediately or flash frozen in liquid nitrogen. For size fractionation of nuclear extracts, cleared nuclear extract (0.5 ml) was loaded into a Superose 6 (GE Healthcare) gel filtration column equilibrated with 2 mM Tris-Cl, pH 8.5, 50 mM NaF, 1 mM NaVO, 1 mM MgCl, 20 mM EDTA, pH 8.0, 0.5 M KCl, and Roche protease inhibitors by using a FPLC system (GE Healthcare). 1-ml fractions were collected, and the size of the eluted complexes calculated according to the elution profile of known molecular weight markers (). For immunoprecipitation studies, in brief, unconjugated antibodies were added to the cleared nuclear extracts on ice for 45 min. After that, protein A– or G–agarose beads (GE Healthcare) preblocked for 30 min at RT with 2% BSA in hypotonic buffer, were added to the solution, and the mixture was incubated for 3 h at 4°C with slow agitation. The beads were then collected by low-speed centrifugation (1,000 rpm for 3 min at 4°C) and washed three times with high-salt washing buffer (2 mM Tris-Cl, pH 8.0, 10% glycerol, 50 mM NaF, 1 mM NaVO, 1 mM MgCl, 0.3 M KCl, 0.05% IGEPAL AC-630, protease inhibitor cocktail [Roche]) and twice with low-salt washing buffer (2 mM Tris-Cl, pH 8.0, 10% glycerol, 50 mM NaF, 1 mM NaVO, 1 mM MgCl, 0.1 M KCl, 0.2% IGEPAL AC-630, and Roche protease inhibitor cocktail). Immunoprecipitated material was eluted by incubating the beads with reducing loading buffer at 70°C for 5 min. Primary antibodies used for immunoprecipitation studies were as follows: a mixture (1:1) of mouse monoclonal anti-coilin antibodies (1:50 [BD Biosciences] and 1:50 [Sigma-Aldrich], respectively: ); anti-FLAG M2 agarose conjugated (Sigma-Aldrich; ), rabbit polyclonal anti-PA28γ antibodies (a mix of MBL [1:100] and Affiniti Research [1:150], respectively; and ). For control immunoprecipitations, equivalent amounts of a rabbit polyclonal anti-GST antibody (GE Healthcare; ) or a mouse monoclonal anti-GFP antibody (Roche; and ) were used, respectively. All images were acquired with a Deltavision Restoration Microscope (Applied Precision) and a camera (Micromax KAF1400; Kodak). Imaging was performed at RT using either a 40× Plan-Neofluar or a 63× Plan-Apochromat (Carl Zeiss MicroImaging, Inc.) objective lens. Images were acquired as TIFF files using SoftWorX (Applied Precision) and Photoshop (Adobe) used for composing the panels shown in the respective figures. Semi-confluent 293 cells transiently transfected with a PCDNA3-FLAG-coilin expression vector, were either mock- or UV-C–treated. The derived cleared nuclear extracts (typically 50 mg of protein) were loaded into a 5-ml heparin–Sepharose CL 4B column (GE Healthcare) by using a FPLC system. The retained protein complexes were eluted with a gradient of increasing ionic strength (10 mM Tris-Cl, pH 7.1, 10% glycerol, 4 mM DTT, 50 mM NaF, 1 mM NaVO, 1mM MgCl, 0.15% IGEPAL AC-630, 50–600 mM KCl, and Roche protease inhibitors cocktail) and the coilin-containing fractions (eluted in a peak at ∼200–250 mM KCl) were subsequently incubated with M2 anti-FLAG antibody conjugated agarose beads (Sigma-Aldrich) for 3 h at 4°C with gentle agitation. Immunoprecipitated material was processed as previously described (see Nuclear extract preparation and immunoprecipitation) and collected by a three-step elution with two bead volumes of 100 mM glycine-HCl, pH 3.0., 1 M Tris-Cl, pH 8.5, and 100% glycerol (Sigma- Aldrich) were added to the eluted fractions to a final concentration of 100 mM and 10%, respectively, before further use or flash freezing in liquid nitrogen. Purified protein complexes were separated on SDS-PAGE gels and stained with Colloidal Coomassie (Invitrogen). The chosen protein bands were excised and subjected to trypsin digestion, and the derived peptides were analyzed by matrix-assisted laser desorption/ionization time-of-flight (Perspective Biosystems) at the Peptide Mass Fingerprinting facility at the University of Dundee (). Protein identification was made with the ProteinProspector software MS-FIT () using the NCBInr (nonredundant) and Swissprot databases. Fig. S1 shows that UV-C enhances PA28γ in coilin complexes. Fig. S2 shows that PA28γ redistributes to splicing speckles in irradiated cells. Fig. S3 shows that hyperosmolar shock triggers fragmentation of CBs. Online supplemental material is available at .

RNA regulons have been proposed as a means by which eukaryotic cells coordinate gene expression (; ). In contrast to prokaryotes, in which the coordinated regulation of genes is achieved by genomic organization, eukaryotes coordinate the regulation of subsets of mRNAs involved in the same biological processes at the posttranscriptional level by manipulating compositions and activities of discrete subsets of RNPs. It has been postulated that related RNA sequences termed untranslated sequence elements for regulation (USER) codes, which are similar to zip codes for RNA localization, are used for specific association with a variety of regulatory proteins involved in different levels of posttranscriptional regulation (). mRNA nuclear export is one level of control that could be coordinated in this way. Initially, mRNA export was thought to be a general process by which all mRNAs were transported from the nucleus to the cytoplasm irregardless of sequence-specific features. More recent findings indicate that mRNA export can be coordinated with other events in RNA metabolism, particularly transcription and splicing, and, thus, that nuclear history of transcripts can modulate the cytoplasmic fate of targeted mRNAs (, ; ; ). This way, nuclear export can be coordinated through compartmentalization via mRNP organization, coupling the coordinated export of functional classes of mRNAs with their functions in biological processes such as proliferation, differentiation, and development. Studies with eukaryotic translation initiation factor eIF4E provide an example of a factor that differentially affects the expression of a subset of mRNAs. Although it associates with all transcripts through the common 5′ methyl-7-guanosine (mG) cap structure (), many groups showed that eIF4E overexpression does not lead to global increases in protein expression (; ). In the cytoplasm, mRNAs deemed eIF4E sensitive have their protein levels modulated by eIF4E more so than other mRNAs. This sensitivity is attributed to the complexity of the 5′ untranslated regions (UTRs) in these transcripts (). Up to 68% of eIF4E is found in the nucleus in a broad variety of species ranging from yeast to humans (; ). Here, eIF4E overexpression leads to the increased export of cyclin D1 but not glyceraldehyde-3-phosphate dehydrogenase (GAPDH) mRNA (; ; ). Specific association of eIF4E with cyclin D1 mRNA in the nucleus requires the mG cap and a small element in its 3′ UTR referred to as an eIF4E sensitivity element (4E-SE; ). Overexpression of eIF4E is correlated with oncogenic transformation in tissue culture, cancers in animal models, and poor prognosis in several human cancers (; ). Several lines of evidence suggest that the mRNA export function of eIF4E contributes to its oncogenic potential. For instance, cyclin D1 mRNA export is up-regulated in specific subtypes of human leukemia (). These specimens contain unusually high levels of eIF4E, the vast majority of which is located in the nucleus (). Also, inhibitors of eIF4E-dependent mRNA export, the promyelocytic leukemia protein (PML), and proline-rich homeodomain (PRH) bind eIF4E in the nucleus, inhibit eIF4E-dependent mRNA export, and inhibit eIF4E-mediated oncogenic transformation (; ; ). Furthermore, mutagenesis studies link the activity of eIF4E in mRNA export to its ability to oncogenically transform cells (; ). Although cyclin D1 plays a key role in the cell cycle that links eIF4E's proliferative properties and its mRNA export function, it is possible that eIF4E coordinately alters the expression of some other growth-promoting mRNAs as well to drive its proliferative potential. This study shows that several mRNAs involved in cell cycle progression are also targets of eIF4E-dependent mRNA export and that the subsets of mRNAs regulated at the level of eIF4E-dependent mRNA export are distinct from those that are preferentially translated in the cytoplasm. We identified an underlying USER code for the export of eIF4E-sensitive transcripts. This code is required for the subnuclear distribution of these RNAs as well as for the formation of relevant eIF4E RNPs. Interestingly, the 4E-SE USER code is a structurally conserved element rather than a sequence-based one. eIF4E-dependent mRNA export can be decoupled from translation. Finally, eIF4E-dependent mRNA export occurs via an alternative mRNA export pathway than bulk mRNA. These results provide the basis for a novel paradigm of eIF4E-mediated tumorigenesis. eIF4E-dependent mRNA export is potentially a broadly based mechanism by which eIF4E controls gene expression and, thereby, modulates growth and proliferation. We sought to determine whether mRNAs other than cyclin D1 might be regulated in an eIF4E-dependent manner. Using nuclear lysates, we isolated mRNAs associated with endogenous eIF4E via immunoprecipitation (IP) or with recombinant eIF4E using a GST pull down–based method (the SNAAP method of ) and identified these by differential display. Given that many of the identified genes are involved in cell cycle progression, eIF4E-immunoprecipitated fractions were also tested for other genes known to be involved in these processes as well as for known growth-inhibitory mRNAs (). All target identification was confirmed by eIF4E IP and quantitative or semiquantitative RT-PCR analysis ( and Fig. S1 A, available at ). Importantly, the list provided in is not intended to be totally inclusive but rather to represent a sampling of the target mRNA population because results from differential display data suggest that hundreds of mRNAs are likely regulated in this manner; in this study, we identified only a subset (Fig. S2). Many of the mRNAs that physically associate with the nuclear fraction of eIF4E code for gene products that act in cell cycle progression and survival, which is consistent with the physiological functions associated with eIF4E (). Importantly, eIF4E does not bind all mRNAs tested (). For instance, eIF4E does not associate with the mRNAs corresponding to negative regulators of growth such as PML or p53 or housekeeping genes such as GAPDH, β-actin, or α-tubulin. Also, this specificity is not a simple reflection of the sensitivity of mRNAs for regulation at the translational level, as mRNAs sensitive only at the translation level (such as VEGF; ) are not associated with the nuclear fraction of eIF4E (). It is important to note that mRNAs that were not found in the eIF4E-immunoprecipitated fractions were readily detected in our nuclear lysates ( and Fig. S1 A). Note that that the estimated efficiency of IP with anti-eIF4E mAb is up to 80%. Because eIF4E associates with the mG cap of mRNAs, we examined whether this was required for the association of eIF4E with mRNAs in the nuclear fraction (). eIF4E was immunoprecipitated from the nuclear fraction, and mRNAs were treated with excess mGpppG or an analogue that does not bind eIF4E (GpppG). All mRNAs tested associate with eIF4E in a cap-dependent manner (i.e., mGpppG competes for binding, whereas GpppG does not). These data indicate that the association of eIF4E with mRNAs in the nucleus is mG cap dependent. To test whether there is a correlation between the ability of eIF4E to associate with mRNAs in the nuclear fraction and the ability of eIF4E overexpression to enhance eIF4E-dependent mRNA export, the subcellular distribution of identified mRNAs as a function of eIF4E overexpression was analyzed ( and Fig. S1 B). U937 and NIH3T3 cells overexpressing eIF4E or appropriate mutants were fractionated, and mRNAs levels were monitored by real-time PCR or Northern analysis. eIF4E overexpression increases the amount of eIF4E-sensitive mRNAs in the cytoplasmic fraction versus vector controls ( and Fig. S1 B). Conversely, transcripts that did not associate with eIF4E in the nuclear fraction did not have their export altered by eIF4E overexpression ( and Fig. S1 B). As expected, the subcellular distribution of β-actin, GAPDH, U6snRNA, and tRNA were unaffected ( and Fig. S1 B). There is no alteration in total mRNA levels (Fig. S1 C). Consistently, when eIF4E could not bind these mRNAs because of a mutation in its cap-binding site (W56A), the subcellular distribution of these mRNAs was not altered ( and Fig. S1 B). Furthermore, the dorsal surface mutant W73A, which does not act in translation but promotes cyclin D1 mRNA export (; ), also promotes the export of other eIF4E-sensitive mRNAs ( and Fig. S1 B). Thus, it is likely that all sensitive mRNAs will require the mG cap-binding activity of eIF4E but not W73 on the dorsal surface for their interaction with eIF4E in the nucleus. Importantly, a circular dichroism study indicates that both W73A and W56A mutants have structures indistinguishable from wild-type eIF4E (). One of the consequences of the eIF4E-dependent promotion of mRNA export is increased availability of these mRNAs to the translation machinery, leading to increased protein levels. Thus, we examined whether protein levels for a subset of identified genes are elevated by eIF4E. Consistent with enhanced mRNA export, the overexpression of wild-type eIF4E or the W73A mutant leads to increased protein levels of a subset of examined genes ( and Fig. S1 D), whereas there is no increase in protein levels when the cap-binding mutant (W56A) is overexpressed. Importantly, wild-type eIF4E and the W73A and W56A mutants were expressed to similar levels (about threefold overexpression) for all experiments ( and Fig. S1 D). To determine whether these mRNAs are regulated through the same mechanism, it was important to examine the effect of PML. Our previous studies showed that PML colocalizes and coimmunoprecipitates with nuclear eIF4E and that this interaction is important for the ability of PML to inhibit eIF4E-dependent cyclin D1 mRNA export and eIF4E-mediated transformation (; ). Thus, we examined whether PML inhibits the export of a range of target mRNAs. In this way, we could determine whether PML was a general inhibitor of eIF4E-dependent export or whether these activities were limited to cyclin D1 mRNA. We observed decreased mRNA export (Fig. S1 B) and reduced protein levels of ornithine decarboxylase (ODC), c-myc, cyclin D1, and cyclin E1 ( and Fig. S1 D) in cells overexpressing PML. Also, PML did not reduce levels of eIF4E, β-actin, or GAPDH proteins ( and Fig. S1 D), and there was no alteration in total mRNA levels for any of these transcripts when PML was overexpressed (Fig. S1 C). Thus, PML acts as an inhibitor of eIF4E-dependent mRNA export, not just as an inhibitor of cyclin D1 mRNA export. Because we previously identified a 100-nucleotide 4E-SE in the 3′ UTR of cyclin D1, which sensitizes cyclin D1 and corresponding chimeric lacZ constructs to regulation by eIF4E at the mRNA export level (), we performed an extensive bioinformatics analysis to identify 4E-SE–like elements in the other target RNAs identified in . Sequence analysis indicated that the 4E-SE was well conserved in cyclin D1 transcript (from birds to humans; ), but a comparison of cyclin D1 with the other eIF4E-sensitive transcripts identified here failed to reveal any shared sequence homology. Therefore, we examined the possibility that the 4E-SE is a structurally conserved element. Using the PatSearch algorithm (), we found that the cyclin D1 4E-SE had a predicted pattern of stem loop pairs (SLPs). Using this same strategy, we found that Pim-1 also contained a similar putative element in its 3′ UTR. As for the cyclin D1 4E-SE mapping (), we made the corresponding constructs of Pim-1–lacZ fusions and showed that one of these elements (Pim-1–p4E-SE; ) was a functional 4E-SE (see next paragraph). Note that a construct derived from a similar element in Pim-1's 3′ UTR (SLP) did not associate with eIF4E (). To best identify the common structural elements in the target mRNAs, we decided to compare the cyclin D1 4E-SE with the 4E-SE from one of the newly identified target mRNAs, Pim-1. We mapped the 4E-SE from cyclin D1 and the 4E-SE from Pim-1 to a minimal ∼50-nucleotide region (). These minimal domains, when fused to heterologous lacZ mRNA, immunoprecipitate with eIF4E and have their mRNA export promoted by eIF4E (; ). Thus, these are functional 4E-SEs. In summary, although there was no sequence homology observed, both elements contain putative SLPs as predicted by PatSearch. We used nuclease digestion methods to determine whether these two functional 4E-SEs had conserved secondary structural features such as the predicted stem loops. Importantly, these studies revealed that both elements fold into similar secondary structures. We refer to this element as an adjacent SLP (, A and B; and Fig. S3 A, available at ). Consistently, biophysical analysis indicates that Pim-1 and cyclin D1 4E-SEs have similar biophysical properties. For instance, circular dichroism analysis of thermal melting curves using purified RNA oligomers for cyclin D1 and Pim-1 4E-SEs revealed multiphase behavior consistent with the presence of multiple structural elements with different melting temperatures (unpublished data). Thus, both Pim-1 and cyclin D1 4E-SEs have similar secondary structures, which is consist with two adjacent stem loop elements. An initial problem we encountered in these studies is that the presence of stem loop elements is common in the 3′ UTRs of cyclin D1 and Pim-1. In cyclin D1 alone, the PatSearch program () predicts 10 potential SLPs, but our previous findings indicate that the only part of the cyclin D1 3′ UTR that can impart eIF4E sensitivity is the aforementioned 4E-SE (). Similarly, the Pim-1 3′ UTR contains two predicted adjacent SLPs, whereas only one is a functional 4E-SE (). Thus, we compared the secondary structures of Pim-1 and cyclin D1 4E-SEs to determine features that would enable us to distinguish functional 4E-SEs from other SLPs. Visual inspection of the secondary structures reveal the conservation of a set of A and U nucleotides (UXUXA, highlighted in ). Importantly, these patterns of nucleotides were not found in any of the other SLPs found in cyclin D1 or Pim-1 3′ UTRs. Thus, these are features that can be used to distinguish functional 4E-SEs from other elements that have the potential to fold into similar secondary structures. Further bioinformatics analyses showed that the predicted SLP structure with the conserved pattern of nucleotides is also present in other eIF4E-sensitive targets identified here (, top). Importantly, none of the mRNAs that are not eIF4E sensitive contain SLPs with the conserved pattern of nucleotides found in the functional 4E-SEs. In summary, we have identified a structural motif consisting of two adjacent SLPs, which impart eIF4E sensitivity. Importantly, there exist in this motif sequence features of 4E-SEs that can be used to distinguish functional 4E-SEs from other paired stem loop structures. To assess whether the 4E-SE acted as an RNA zip code for eIF4E nuclear bodies, lacZ chimeric constructs with either Pim-1 or cyclin D1 4E-SE were expressed in U2OS cells. Both chimeric mRNAs colocalize with eIF4E nuclear bodies (). In the absence of the 4E-SE, no localization of lacZ transcripts to eIF4E nuclear bodies is observed (). Importantly, lacZ–4E-SE does not associate with eIF4E bodies that contain the negative regulator PML. This is consistent with our previous studies showing that there are two classes of eIF4E nuclear bodies: those that colocalize with PML and those that colocalize with endogenous cyclin D1 mRNA (; ; ). Thus, endogenous cyclin D1 mRNAs colocalize with eIF4E nuclear bodies that do not contain PML (). In this way, lacZ–4E-SE transcripts and endogenous mRNAs behave similarly. To establish whether the 4E-SE functions simply as a localization signal or whether it acts in the formation of eIF4E-dependent mRNPs, we performed electromobility shift assays (EMSAs). Studies were performed with both the lacZ–cyclin D1–4E-SE (c4E-SE) and the lacZ–Pim-1–4E-SE (p4E-SE) to ensure that the assembly of these complexes is dependent on the 4E-SE itself and not on features specific to either 4E-SE. RNA probes were P 3′ end labeled and mG capped. The addition of either mouse eIF4E with a 6-kD solubility tag (m4E) or untagged human eIF4E (h4E) led to the formation of slower migrating species for both lacZ–4E-SE constructs (). Importantly, the addition of nuclear lysates led to the formation of substantially higher molecular weight complexes, indicating that proteins other than eIF4E are likely to be present. Complex sizes were approximately the same for both 4E-SE constructs. The addition of cold competitor 4E-SE RNAs led to a reduction in signal, which is consistent with the 4E-SE element competing for the labeled 4E-SE–containing transcripts (). The addition of nuclear lysates to lacZ transcripts lacking the 4E-SE did not lead to the formation of these complexes ( and Fig. S3 B). Although the addition of purified eIF4E to the nuclear lysate supplemented with 4E-SE–containing transcripts led to a considerable increase in the amount of RNA undergoing the shift, it did not alter the position of the shift (). To determine whether the 4E-SE complexes formed from nuclear lysates were dependent on eIF4E, EMSAs were performed with nuclear lysates depleted of eIF4E via IP. We estimated that lysates were at least 80% depleted of eIF4E (unpublished data). Lysates immunodepleted of eIF4E did not produce high molecular weight complexes (). The addition of purified tagged eIF4E to immunodepleted lysates led to a partial restoration of the complex, which could be expected because only eIF4E but not other factors that were depleted during the anti-eIF4E IP were reintroduced. Thus, eIF4E and associated factors are required for the formation of these RNPs. In addition, an antibody to eIF4E leads to a super shift of complexes formed from nuclear lysates (). Identical results are observed for lacZ–p4E-SE. Finally, a mutant that disrupts the first stem loop (GCG mutated to CAC; ) in the p4E-SE is defective in complex formation (). Thus, the 4E-SE element forms complexes that are dependent on eIF4E and on the structure of the 4E-SE. To further characterize these complexes, lacZ–4E-SE constructs were UV cross-linked followed by RNase digestion and SDS gel electrophoresis (). As for the EMSA studies, transcripts were mG capped and 3′ end labeled, and the effects of addition of purified eIF4E or nuclear lysates to the size of cross-linked complexes was monitored. Because mRNAs were 3′ end labeled, binding of the cap only by purified eIF4E was not sufficient to protect the rest of the RNA from RNase digestion. The addition of the nuclear lysate leads to substantial shifts in molecular weight. Importantly, the lacZ–c4E-SE and the lacZ–p4E-SE form complexes similar in size. Three discrete species between 75 and 90 kD are observed (, arrows). The same complexes are absent in eIF4E-depleted nuclear lysate, indicating that these require eIF4E to form. Consistently, treatment of the nuclear lysate with the mGpppG cap analogue (nuclear lysates + cap) also disrupts 75–90-kD range complexes. These species are absent from the lacZ controls, which lack the 4E-SE. A lower band at ∼64 kD is present in all of the experiments, likely indicating the formation of some general RNP not directly involved with eIF4E and the 4E-SE. In summary, we observe two types of complexes: those that can form in the absence of eIF4E and are cap and 4E-SE independent (, asterisk) and the second type that depends on eIF4E, the mG cap, and a structurally intact 4E-SE. The UV cross-linking experiments together with the EMSA results indicate that the 4E-SE acts both as a zip code localizing mRNAs to bodies () and as a USER code for the eIF4E nuclear mRNP (). We examined the importance of new protein synthesis and transcription for eIF4E-dependent mRNA export. To inhibit protein synthesis, cells were treated with 100 μg/ml cycloheximide for 1 h. Under these conditions, eIF4E-dependent mRNA export of lacZ–c4E-SE is not altered (). Also, the export of endogenous cyclin D1 mRNA was not modulated by cycloheximide treatment (unpublished data). Similarly, actinomycin D treatment (10 μg/ml) did not affect the export of these mRNAs (). Although cycloheximide treatment did not modify export, it is still possible that the 4E-SE could modulate polysomal loading in an eIF4E-dependent manner. Thus, we monitored polysomal profiles of lacZ as a function of the 4E-SE and of eIF4E overexpression. The profiles of lacZ and lacZ–c4E-SE are indistinguishable and are not altered by eIF4E overexpression (unpublished data). This is consistent with the finding that eIF4E overexpression does not change cyclin D1 mRNA polysomal loading (). Given that eIF4E-dependent mRNA export is independent of ongoing protein synthesis and that the 4E-SE does not alter polysomal loading, the functions of eIF4E in mRNA export and translation appear to be decoupled. We previously demonstrated that lacZ–c4E-SE transcripts did not have altered stability relative to lacZ transcripts using actinomycin D over the course of several hours (). However, it is still possible that mRNA turnover could be substantially more rapid than several hours. Thus, we constructed lacZ and lacZ–4E-SE TetON-inducible cell lines and examined the stability of these mRNAs immediately upon doxycycline addition. The presence of the 4E-SE does not substantially alter the stability of the lacZ transcripts in either the short (minutes) or long term (hours; unpublished data). We reasoned that if the 4E-SE is required for export, the overexpression of lacZ–c4E-SE or lacZ–p4E-SE should specifically inhibit the export of other (endogenous) 4E-SE–containing mRNAs by competing for the 4E-SE–specific export machinery (). Using our TetON-inducible lacZ, lacZ–p4E-SE, or lacZ–c4E-SE constructs, we monitored the export of chimeric mRNAs as a function of total mRNA levels. The expression of lacZ–p4E-SE and lacZ–c4E-SE mRNAs as a function of time is shown in . At early time points, when levels of lacZ mRNAs are low, 4E-SE export is more efficient with higher ratios of cytoplasmic to nuclear chimeric mRNAs. As the levels of these mRNAs increase, 4E-SE export becomes saturated, and the ratio of cytoplasmic to nuclear chimeric mRNAs decreases (). At the same time, the export of endogenous cyclin D1 mRNA was impaired by the expression of 4E-SE chimeric mRNAs (). Furthermore, the export of VEGF mRNA was not affected, which is consistent with its insensitivity to eIF4E at the mRNA export level (). Thus, the overexpression of 4E-SE leads to competition for the 4E-SE–specific export machinery. Because the best-described cellular mRNA export pathway involves the NXF1/p15 heterodimer, which appears to mediate bulk mRNA export (, ), the dependence of 4E-SE mRNA export on NXF1 was examined (). Consistent with previous findings as well as our own, eIF4E does not immunoprecipitate with NXF1 in the nuclear fraction of cells (; and unpublished data). However, this does not preclude an NXF1-dependent mechanism in which eIF4E does not need to physically associate with NXF1. To further investigate NXF1 involvement in 4E-SE export, Flag-tagged NXF1- or NXF1/p15-overexpressing cells were immunoprecipitated with anti-Flag antibodies, and the presence of lacZ or lacZ–c4E-SE mRNAs was monitored by real-time PCR (). In contrast to lacZ mRNA that is enriched in the NXF1 fractions, lacZ–c4E-SE mRNA appears to be almost completely excluded. These results are independent of the presence or absence of p15 (unpublished data). We extended these studies to examine the effects of knocking down NXF1 expression on lacZ–c4E-SE export (). Overexpression of eIF4E enhanced the export of lacZ–c4E-SE transcripts even when NXF1 levels were substantially reduced, indicating that the export of lacZ–c4E-SE in the presence of overexpressed eIF4E is independent of NXF1. In the absence of the 4E-SE, the lacZ mRNA cytoplasmic/nuclear ratio was substantially reduced by NXF1 depletion. Analysis of lacZ protein levels confirmed the aforementioned findings (). As expected, siRNA treatment led to a reduction in NXF1 levels, whereas treatment with scrambled controls did not (). Furthermore, levels of eIF4G were not altered, which is consistent with a study showing that longer siRNA treatments (>72 h) are needed to reduce eIF4G levels (). Thus, the export of 4E-SE–containing transcripts is independent of the NXF1 pathway. This does not rule out the possibility that a subset of 4E-SE transcripts do transit through this pathway but simply that they do not require this pathway to be exported. Because many RNAs can be exported through the CRM1 pathway, we examined this possibility by using leptomycin B (LMB), a specific inhibitor of CRM1 (,). The export of lacZ or lacZ–c4E-SE mRNAs as a function of overexpressed eIF4E and LMB treatment was monitored using real-time PCR (). Strikingly, LMB suppressed the export of lacZ–4E-SE constructs but not of lacZ or β-actin transcripts. LMB leads to the retention of 18S ribosomal RNA (rRNA; ), which is consistent with previous studies showing that rRNA export requires CRM1 (, ). Since it was described, no underlying mechanism for eIF4E-dependent export has been determined (). There are several characteristic features that differentiate eIF4E-mediated export from the pathway used for bulk mRNA (summarized in ): 4E-SE saturates export of the eIF4E pathway but does not affect the export of bulk mRNA (); LMB inhibits eIF4E-dependent export (); and the mG cap is required for the eIF4E pathway ( and ). Interestingly, there are many parallels between the eIF4E pathway and U small nuclear RNA (UsnRNA) export: both are CRM1 dependent, and both require the mG cap. However, in contrast to the eIF4E pathway, UsnRNA export depends on RNAs being cap-binding complex bound in complex to PHAX, which acts as an adaptor for CRM1 (; , ,; ). In general, CRM1-mediated mRNA export requires cofactors that depend on the type of RNA being exported (i.e., large rRNA, small rRNA, 5S rRNA, or UsnRNA; ,). Our previous studies indicate that eIF4E overexpression does not modulate the export of 18S or 28S rRNA, which is CRM1 dependent, or tRNA, which is exported using the exportin-t receptor (; , ). Thus, we hypothesize that eIF4E or some subset of factors associated with the 4E-SE RNP require CRM1 adaptor proteins specific to the eIF4E-dependent pathway. Furthermore, these adaptors are found in limiting amounts and are titratable by high 4E-SE levels or by immunodepletion of eIF4E. Identifying such adaptor proteins will be an area of intense future work. A conundrum in understanding eIF4E-dependent mRNA export results from the observation that eIF4E stimulates the export of mRNAs that can be still exported under physiological eIF4E levels. Thus, eIF4E-dependent mRNA export is a means by which the cell rapidly up-regulates gene expression by stimulating the export of mRNAs that can be exported through other pathways, albeit less efficiently. When eIF4E levels are low, or in the absence of the mG cap or 4E-SE, transcripts are exported (presumably) through the NXF1 pathway. This idea is consistent with previous suggestions that the NXF1 pathway is a default RNA export pathway for those RNAs that do not have any special features associated with them (). In this way, eIF4E levels can act as a cellular rheostat. As levels increase, eIF4E-sensitive mRNAs are exported much more efficiently and in a coordinated fashion through the eIF4E-dependent CRM1-sensitive pathway described in this study. A recent study indicates that CRM1-dependent mRNA export can occur during T cell activation, indicating that external cellular signals can lead to alterations in mRNA export pathways (). The studies reported here suggest the possibility that the proliferative and transforming properties associated with eIF4E are, at least partially, a result of the dysregulation of eIF4E-dependent mRNA export. These studies indicate a role for eIF4E in coordinating the export and expression of transcripts involved in cell cycle progression, proliferation, and survival. Importantly, eIF4E does not promote the expression of negative regulators of itself (i.e., PML). eIF4E also promotes the expression of c-myc, a factor that up-regulates the transcription of eIF4E in some cellular growth conditions (). Thus, eIF4E modulates the expression of many genes involved in multiple points of cell cycle progression. The 4E-SE provides a USER code for targeting these transcripts for export in an eIF4E-sensitive manner. Other transcripts may be regulated by eIF4E at the translation level using USER codes that are different from the 4E-SE. Furthermore, the 4E-SE may associate with other, as yet unidentified, RNPs. In this way, the effects of eIF4E and regulation of 4E-SE–containing transcripts are likely to be complex and combinatorial. For instance, the translation of export-sensitive mRNAs does not depend on the 4E-SE but rather on the complexity of the 5′ UTR. Transcripts such as Pim-1 and ODC (; ) serve as examples of the combinatorial use of USER codes for modulating gene expression and support the idea of the use of such a network. Consistently, our experiments indicate that the translation and export functions of eIF4E can be decoupled based on the composition of the 3′ and 5′ UTRs (i.e., eIF4E enhances the export of cyclin D1 but enhances the translation of VEGF). Several key regulators of eIF4E-dependent mRNA export have been identified, including PML () and several homeodomain proteins that contain conserved eIF4E-binding sites (, ). These regulators are positioned to modulate the entire RNA regulon, potently modulating cell cycle progression and cell survival. Our studies demonstrate that PML and PRH impede the eIF4E-dependent export of cyclin D1 and other 4E-SE–containing transcripts (; ; and this study). Further in vitro, PML can also inhibit eIF4E-sensitive translation () and, thus, is positioned to control the regulon at different levels. Stimulators of this growth regulon include HOXA9, which promotes both the mRNA export and translation of genes in the regulon (). The far-reaching activities of these regulators, particularly those that regulate multiple eIF4E functions simultaneously, likely lies in their ability to modulate eIF4E, a key nexus in this regulon. The physiological importance of this regulation is clear. In primary specimens from acute myeloid leukemia patients, PRH is both down-regulated and delocalized from eIF4E nuclear bodies (). At the same time, HOXA9 is up-regulated and becomes associated with eIF4E in both the nuclear and cytoplasmic compartments, leading to the up-regulation of both eIF4E-dependent mRNA export and translation (). Chimeric constructs in pcDNA3.1lacZ vector (Invitrogen) were positioned 3′ of the coding region of lacZ. Cyclin D1 minimal 4E-SE (c4E-SE) was amplified using primers containing EcoRI or XbaI restriction sites at the 5′ ends and the lacZ 3′ UTR construct as a template (). The same approach was used for the cloning of Pim-1 constructs, in which pRBK–Pim-1 (a gift from N. Magnuson, Washington State University, Pullman, WA; ) was used as a template. Primer sequences are given in Table S1 (available at ). For the TetON system, chimeric lacZ constructs were cloned into pTREMyc vector (CLONTECH Laboratories, Inc.) using EcoRI and XbaI. pcDNA2Flag-eIF4E, pMV, pMV-eIF4E wild type or mutants, pLINKSV40-PML, MSCV, MSCV-eIF4E wild type or mutants, and bacterial expression constructs were previously described (; ; ). Reagents used were all analytical grade from Sigma-Aldrich unless otherwise stated. Antibodies for immunoblotting were as follows: mAb anti-PML (5E10; ), mAb anti-eIF4E (BD Biosciences), mAb anticyclin D1 (BD Biosciences), mAb anti-Xpress (Invitrogen), rabbit pAb anticyclin E1 (M20; Santa Cruz Biotechnology, Inc.), mAb anti-GAPDH (MAB374; Chemicon), mAb anti–c-myc (9E10; Santa Cruz Biotechnology, Inc.), rabbit pAb anticyclin A (C-19; Santa Cruz Biotechnology, Inc.), rabbit pAb antinibrin (Cell Signaling), mAb anti–Pim-1 (19F7; Santa Cruz Biotechnology, Inc.), mAb cyclin B1 (GNS1; Santa Cruz Biotechnology, Inc.), mAb anti-eIF4G (BD Biosciences), and mAb anti-NXF1/TAP (BD Biosciences). Stably eIF4E- and PML-transfected NIH3T3 and U937 cells were as described previously (, ). U937 cells were used to analyze endogenous Pim1, which is not expressed in NIH3T3 cells. LacZ/lacZ–4E-SE with or without 2Flag-eIF4E as well as the TetON lacZ system were stably transfected in U2OS cells. For NXF1 depletion, U2OS cells were transfected with LipofectAMINE 2000 and 10 nM siRNA duplex HSC.RNAI.N006362.1.3 (Integrated DNA Technologies) according to the manufacturer's instructions. Cells were analyzed 72 h after transfection. Actinomycin D, cycloheximide, and LMB were all cell culture grade (Sigma-Aldrich). Time and concentrations used in treatments are described in the Results and and . Immunopurification was performed as previously described (). Real-time PCR analyses were performed using Sybr Green PCR Master mix (Applied Biosystems) in Mx3000P thermal cycler (Stratagene), and data were analyzed with MxPro software (Stratagene). All conditions were described previously (). Primer sequences are listed in Table S1. All calculations were performed using the relative standard curve method described in User Bulletin

cAMP is a ubiquitous second messenger and is responsible for a plethora of cellular effects and biological functions (). The generation of cAMP occurs upon ligand binding to G protein–coupled receptors (GPCRs) and the consequent activation of a family of transmembrane adenylyl cyclases (ACs) localized at the plasma membrane (). An alternative source of cAMP is the soluble AC, an enzyme that has been shown to localize in different subcellular compartments, the activity of which is independent of GPCR stimulation and is regulated by bicarbonate and calcium ions (). cAMP exerts its cellular functions via the activation of three different effectors: the cAMP-dependent PKA, cyclic nucleotide-gated ion channels (), and exchange proteins directly activated by cAMP (Epac; ). The action of cAMP is terminated via its degradation by phosphodiesterases (PDEs), a large superfamily of enzymes grouped in 11 families with >40 isoenzyme variants (; ; ; ). Individual PDE enzymes exert specific functional roles as a consequence of the unique combination of regulatory mechanisms, intracellular localization, and enzyme kinetics (; ; ) and play an important role in shaping intracellular gradients of cAMP (; ; ). The notion of the spatial regulation of cAMP as a means of generating specific downstream responses is now well accepted (; ). Such a paradigm is grounded on increasing evidence that cAMP/PKA signaling is compartmentalized in discrete subcellular domains in which PKA is anchored to A kinase–anchoring proteins (AKAPs; ) in close proximity to its specific targets and is activated by restricted pools of cAMP (; ; ; ; ; ). However, the concept of the restricted intracellular diffusion of cAMP contrasts with the evidence that the diffusion rate of this second messenger in the cytosol appears to be unrestrained (∼500–700 μms; ; ). To explain the paradox of a signaling pathway organized in spatially segregated transduction units, the selective activation of which is committed to a freely diffusible second messenger, the hypothesis of either a physical or an enzymatic barrier restricting intracellular diffusion of cAMP has been formulated (, ; ; ; ; ; for review see ). Recently, new methods for monitoring intracellular cAMP concentration in single living cells have been developed. One approach using cyclic nucleotide-gated ion channel–based sensors relies on ion influx measurements as a readout of cAMP changes near the plasma membrane (). Another approach uses fluorescence resonance energy transfer (FRET)–based indicators in which cAMP binding to PKA () or Epac (; ; ) proteins leads to a change in fluorescence emission that correlates with the intracellular concentration of cAMP. Such optical biosensors allow the detection of cAMP changes with submicrometer resolution in different intracellular compartments. These new technologies led to the identification in human embryonic kidney (HEK) 293 cells of a subplasma membrane compartment showing a larger rise in cAMP concentration in response to prostaglandin 1 (PGE) receptor stimulation as compared with the bulk cytosol (; ), but the molecular and structural components responsible for such compartmentalization largely remain to be defined. Here, we set out to study local cAMP dynamics by using FRET-based biosensors that are selectively targeted to distinct subcellular compartments in HEK293 cells. We found that compartmentalized PDEs, rather than acting as barriers to cAMP diffusion from the plasma membrane to the bulk cytosol, act as a sink that drains cAMP concentration in defined domains by locally degrading the second messenger. As a result, intracellular cAMP concentration is not bound to change along a uniform gradient from the plasma membrane to the deep cytosol, but multiple contiguous domains with different concentrations of cAMP may coexist within the volume of the cell. We recently generated a FRET-based sensor for real-time imaging of cAMP. The sensor, called PKA-GFP, includes the regulatory (R) and catalytic (C) subunits of the PKA tagged with the CFP and YFP variants of GFP, respectively (). When overexpressed in HEK293 cells, PKA-GFP shows a uniform distribution in the cytosol (). To compare cAMP dynamics in the bulk cytosol and in the subplasma membrane compartment, we generated a variant of the PKA-GFP sensor that is targeted to the plasma membrane. To this end, we fused to the N terminus of the R subunit a short polypeptide () corresponding to the N-terminal targeting signal from the Lyn kinase (). This sequence is posttranslationally myristoylated and palmitoylated and targets to the plasma membrane (). HEK293 cells cotransfected with R-CFP and C-YFP (PKA-GFP) show the predicted localization of both the R and C subunits at the plasma membrane (). To verify that modification of the R-CFP sequence with the peptide does not affect the sensitivity for cAMP, we determined activation constant (K) values both for PKA-GFP and PKA-GFP. As shown in , K values were identical for the two sensors (K = 0.77 μM). Modification at the N terminus of the R subunit does not affect its dynamic interactions with the C subunit, as shown by its complete release into the cytosol upon cAMP increase and binding to R-CFP. C-YFP is effectively resequestered to the plasma membrane upon cAMP removal (supplemental material and Video 1; available at ). To verify whether GPCR stimulation generates different cAMP signals in the subplasma membrane compartment as compared with the bulk cytosol, we stimulated HEK293 cells transfected with either PKA-GFP or PKA-GFP with 10 μM PGE and measured cAMP-induced FRET changes in the two compartments. As shown in , the cAMP response in the subplasma membrane compartment was almost twice as large as the response in the bulk cytosol (ΔR/R = 18.6 ± 2.1 [mean ± SEM; = 30] vs. 9.6 ± 1.2% [ = 34]; P = 1.4 × 10). Furthermore, the time to reach half-maximal response to PGE was almost twice as fast in the subplasma membrane compartment as compared with the bulk cytosol ( = 58.9 ± 7.3 [ = 30] and 110.6 ± 12.5 s [ = 34], respectively; P = 1.7 × 10; ). We found that in ∼50% of the analyzed cells, PGE generated a transient response in both the subplasma membrane compartment and the bulk cytosol, with the level of cAMP returning to baseline levels in 5.8 ± 0.84 and 7.6 ± 1.04 min, respectively, after application of the stimulus (). In the other 50% of the cells, the response reached a plateau value and remained sustained for at least 10 min (unpublished data). Application of the nonselective PDE inhibitor isobutyl-methyl-xanthine (IBMX; 100 μM) in the continuous presence of PGE raised the cAMP level in both compartments, indicating that termination of the cAMP response to PGE was caused by PDE activity. We excluded any overt role for receptor desensitization in the termination of the cAMP response to PGE based on the observation that it is possible to repeatedly stimulate the PGE receptor and obtain comparable rises in intracellular [cAMP] (supplemental material and Fig. S1; available at ). The application of 1 μM PGE also generated different cAMP levels in the two compartments, with a response in the subplasma membrane compartment of 8.2 ± 1.5% ( = 34) and of 3.3 ± 0.6% ( = 33) in the bulk cytosol (P = 9.5 × 10; ). When using the PKA-GFP sensor, the binding of cAMP to R-CFP causes subunit dissociation, releasing the C-YFP subunits to diffuse away from the RII-CFP FRET partners anchored at the plasma membrane. Conversely, when cAMP binds to the R-CFP subunits of PKA-GFP in the bulk cytosol, the dissociating C-YFP subunits remain in close proximity to their cytosolic R-CFP FRET partners. In our experiments, FRET changes are measured as CFP intensity/YFP intensity (480/545-nm fluorescence emission; see FRET imaging section). Therefore, as a result of the diffusion of C-YFP away from the plasma membrane, the FRET changes measured upon probe dissociation in the subplasma membrane compartment may result in artifactually larger values being observed than the FRET changes measured in the cytosol. To exclude this possibility and confirm that the subplasma membrane compartment and the cytosolic compartments show a distinct cAMP response to PGE, we used a unimolecular FRET sensor for cAMP based on Epac1 (H30; ). As H30 is a single polypeptide chain, CFP and YFP do not diffuse apart upon cAMP binding. We modified H30 by fusing the plasma membrane–targeting sequence (mpH30; ) to its N terminus, which, as with RII, allowed for the effective targeting of this unimolecular sensor to the plasma membrane (). Membrane targeting of H30 did not substantially modify its sensitivity to cAMP, as shown by the apparent dissociation constants measured for H30 and H30 (EC = 12.5 and 20 μM, respectively; ). Using these unimolecular probes, we show in that the stimulation of HEK293 cells expressing H30 or H30 with 1 μM PGE generated a mean ΔR/R of 23.5 ± 1.4 ( = 40) and 31.4 ± 1.2% ( = 36), respectively (P = 10). These results confirm that PGE stimulation generates a compartmentalized cAMP response, with a higher level of cAMP being generated in the subplasma membrane compartment as compared with the bulk cytosol. Interestingly, the steepness of the cAMP gradient between the subplasma membrane and bulk cytosol compartments is smaller (35% higher cAMP response at the plasma membrane vs. the cytosol) as compared with the steepness of the gradient recorded with the PKA-based sensor (148% higher cAMP response at the plasma membrane vs. the cytosol; compare with ). The PKA-based and Epac1-based sensors both clearly reveal a gradient of cAMP between the plasma membrane and the cytosol upon PGE stimulation. However, such a gradient appears steeper when detected by the PKA-GFP probe as compared with H30. One possible explanation for this is that overexpression of the PKA-based sensor, the catalytic subunit of which is enzymatically active, may affect the steepness of the cAMP gradient. To test this hypothesis, we measured cAMP levels in HEK293 cells transfected with either H30 or H30 alone or in combination with untagged PKA. As shown in , in the presence of overexpressed PKA, the FRET change recorded at the plasma membrane was ΔR/R = 9.5 ± 1% ( = 48), whereas the FRET change recorded in the bulk cytosol was ΔR/R = 5.8 ± 0.5% ( = 60; P = 0.0009), indicating that the level of cAMP is ∼62% higher at the plasma membrane as compared with the cytosol. Thus, PKA overexpression increases the steepness of the cAMP gradient between the two compartments around twofold. In further support of this, we found that the higher the level of PKA overexpression, the larger the effect is on the steepness of the cAMP gradient (supplemental material and Fig. S2; available at ). PKA is known to activate PDE3 and PDE4 families of PDEs, thereby stimulating the degradation of cAMP (; ). Interestingly, PDE4 and PDE3 are the major cAMP PDE activities represented in HEK293 cells, with PDE4 accounting for ∼68% of the total PDE activity and PDE3 accounting for a remaining 30% (). Therefore, we asked whether PDEs may be the effectors of PKA in modulating the steepness of the cAMP gradient. We found that the inhibition of PDEs with 100 μM of the nonselective PDE inhibitor IBMX completely abolished the effect of PKA overexpression on the cAMP gradient (cytosol ΔR/R = 26.03 ± 1.73% [ = 64] in the cytosol and 32.46 ± 1.72% [ = 56] at the plasma membrane; P = 0.009; ), reestablishing a difference in the cAMP level present in the two compartments of ∼25% (compare with ). These results confirm that PDEs mediate the effect of PKA on the cAMP gradient. In agreement with this finding and in support of the key role played by PKA in shaping the cAMP gradient, the inhibition of overexpressed PKA activity with 10 μM H89 completely abolished the cAMP gradient (ΔR/R = 12.7 ± 2.05% [ =35] in the cytosol and 16.18 + 2.7% [ =22] at the plasma membrane; P = NS; ). Similarly, the inhibition of endogenous PKA with H89 was sufficient to completely dissipate the cAMP gradient generated upon PGE stimulation (). Interestingly, the cAMP response to PGE in the presence of H89 was invariably more sustained in time, with the level of cAMP being reduced by only ∼25% at 10 min after application of the stimulus. To determine the contribution of different PDE families in shaping the cAMP gradient, we selectively inhibited either PDE4 with 10 μM rolipram () or PDE3 with 10 μM cilostamide (). We found that the sole inhibition of PDE4 reproduced the effect of total PDE inhibition with IBMX (ΔR/R = 30.88 ± 3.9% [ = 14] in the cytosol and 41.8 ± 2% [ = 16] at the plasma membrane; P = 0.01; compare with ). Rolipram showed a similar effect in cells transfected with H30 or H30 in the absence of overexpressed PKA (unpublished data). Conversely, selective PDE3 inhibition with 10 μM cilostamide did not substantially raise the cAMP level generated by PGE stimulation in either compartments both in the presence (ΔR/R = 5.05 ± 0.6% [ = 16] in the cytosol and 10.8 ± 1.7% [ = 12] at the plasma membrane; P = 0.00095; ) and in the absence of overexpressed PKA (not depicted). These results strongly indicate that PDE4 is the key regulator of the intracellular cAMP gradient in HEK293 cells. Our results are in agreement with studies indicating that PDEs may be responsible for the generation of intracellular cAMP gradients by acting as an enzymatic barrier that by rapidly degrading cAMP limits its diffusion from the site of synthesis (the plasma membrane) to the deep cytosol (; ; ). According to this view, the inhibition of PDE activity would abolish the barrier to cAMP diffusion and should result in a fast reequilibration of cAMP concentration within the cell. Surprisingly, however, our results show that PDE inhibition with either IBMX or rolipram does not dissipate the cAMP gradient between the plasma membrane and the bulk cytosol elicited by PGE stimulation (; , , and ). We reasoned that as both IBMX and rolipram are competitive PDE inhibitors, in the high [cAMP] subplasma membrane compartment, they may be less efficient in inhibiting PDEs than in the bulk cytosol (lower [cAMP]), thus sustaining the cAMP gradient between the two compartments. To test the effect of noncompetitive inhibition of PDE4 on the intracellular cAMP gradient, we performed genetic ablation of PDEs by an RNA silencing approach to reduce enzyme concentration. The most represented PDE4 subfamilies in HEK293 cells are PDE4B and PDE4D, with PDE4B representing ∼30% of the total PDE4 activity and PDE4D representing ∼65% in these cells (). Therefore, we focused on these two subfamilies. The design of siRNA oligonucleotides as well as the time and means of transfection had been previously optimized by us to achieve ∼95% selective knockdown of each PDE subfamily in HEK293 cells (). We found that the genetic knockdown of PDE4B did not affect the cAMP gradient (ΔR/R = 13.9 ± 2.2 [ = 27] and 8.5 ± 1.2% [ = 23] in the subplasma membrane and bulk cytosol, respectively; P = 0.04; ). Surprisingly, when we cotransfected the cAMP sensor with siRNA sequences that selectively knock down PDE4D, we found that the cAMP response was higher in the cytosol as compared with the subplasma membrane compartment (ΔR/R = 22.8 ± 2.4 [ = 25] and 14.6 ± 2.5% [ = 23], respectively; P = 0.04; ). The finding that the ablation of PDE4D results in the inversion of the cAMP gradient suggests that a PDE different from PDE4D is active in the subplasma membrane compartment and maintains the cAMP concentration low in that domain. In fact, when we ablated both PDE4B and PDE4D, we found that the cAMP gradient was completely abolished (ΔR/R = 22.8 ± 2.2% [ = 20] in the cytosol and 20.3 ± 2.4% [ = 22] in the subplasma membrane compartment; ). Overall, our results suggest that PDE4B mainly regulates the subplasma membrane compartment, whereas PDE4D mainly regulates the cytosolic pool. The ability of individual PDE4 subfamilies to control cAMP concentration independently in defined compartments implies that they are selectively tethered within such compartments. To test this, we overexpressed catalytically inactive point mutants of PDE4B and PDE4D in HEK293 cells. Such mutants have been shown to exert a dominant-negative (dn) effect by displacing the cognate endogenous active PDE4 isoforms from their functionally relevant anchor sites (). As shown in , upon the overexpression of either dnPDE4B1 or dnPDE4B2, the cAMP response to PGE was higher in the subplasma membrane compartment than in the cytosol. This finding indicates that the displacement of PDE4B from its anchor sites and its consequent distribution inside the cell do not affect the gradient dictated by the prevalent PDE4D activity localized in the cytoplasm. On the contrary, the overexpression of dnPDE4D3 or dnPDE4D5 completely abolished the cAMP gradient (), as expected upon displacement of the prevalent PDE4D activity from its anchor sites within the cytoplasm. Overall, these results confirm that the different cAMP concentrations in these two compartments are dependent on compartmentalized PDE4 isoforms. The aforementioned effect of the genetic manipulation of PDE4B and PDE4D implies a distinct localization of the two enzyme subfamilies. To confirm this, we performed immunocytochemistry by using monoclonal antibodies specific for either subfamilies and found that, as expected, PDE4B shows a prevalent localization at the plasma membrane, whereas PDE4D localizes mainly in the bulk cytosol (). Such membrane sequestration of PDE4B is consistent with a previous study showing that in HEK293 cells, PDE4B2 is unavailable to interact with cytosolic β-arrestin (). The results presented in and indicate that PDE4 enzymes, rather than acting as a barrier to cAMP diffusion from the site of synthesis to the bulk cytosol, function as a local drain to degrade freely diffusible cAMP and maintain the low concentration of the second messenger in discrete compartments. Thus, rather than a continuous gradient from the plasma membrane (higher [cAMP]) to the deep cytosol (lower [cAMP]), we may expect to find contiguous compartments with higher or lower [cAMP] independent of their distance from the site of synthesis but depending on the localization and activity of PDEs. According to this hypothesis, a subcellular compartment deeper inside the cell as compared with the bulk cytosol and in which PDE activity is low should show a higher [cAMP] as compared with the bulk cytosol. The nucleus is such a compartment. To monitor [cAMP] inside the nucleus, we fused a nuclear localization sequence () to the C terminus of the H30 sensor () that selectively targets the probe to the nucleus (). When we challenged HEK293 cells transfected with H30 with 1 μM PGE, we found that the FRET change was higher in the nucleus as compared with the bulk cytosol (ΔR/R = 28.5 ± 1.7 [ = 25] and 23.5 ± 1.4% [ = 28], respectively; P = 0.03) and was similar to the FRET change found in the subplasma membrane compartment (). The different response recorded in the nucleus was not caused by a higher sensitivity of the H30 sensor to cAMP changes, as indicated by similar EC (17.5 μM for nlsH30 vs. 12.5 μM for H30; ). In apparent contrast with our data showing that there is a delay in the cAMP changes in the bulk cytosol as compared with the subplasma membrane compartment (), the kinetics of the nuclear cAMP changes appear to be as fast as those occurring at the plasma membrane (). This is caused by differences in probe kinetics in that the velocity of H30 FRET change upon [cAMP] rise is significantly faster than the velocity of H30 (P = 0.004) and H30 (P = 0.0001) FRET changes at the same [cAMP] (supplemental material). The difference in the FRET change between the nucleus and the bulk cytosol was completely abolished when PDE4D was ablated by siRNA duplex cotransfection (ΔR/R = 17.7 ± 1.5% [ = 26] in the nucleus and 22.8 ± 2.5% [ = 28] in the cytosol; P = 0.12; ). These results confirm that multiple contiguous, subcellular domains with diverse [cAMP] may coexist within HEK293 cells irrespective of their distance from the site of cAMP synthesis and depending on PDE4B and PDE4D activity and localization. The pleiotropic effects of cAMP pose the pressing question of how signaling specificity is achieved. In the past few years, compartmentalization of the cAMP signal transduction pathway has emerged as an important mechanism to ensure the necessary specificity of response (; ). A particular focus has been placed on the organization of macromolecular complexes, including receptors, effectors, modulators, and targets effectively organized in restricted domains, within which the molecular components of the complex only affect each other appropriately. Some of these domains are localized at the plasma membrane, one example of which is the assembly of the β adrenergic receptor, heterotrimeric G proteins, AC, and PKA and its target, the L-type Ca channel Ca1.2 (). Within the complex, activation of the receptor leads to the synthesis of cAMP and activation of PKA, which, in turn, can regulate the activity of the channel in a highly localized manner. AKAP/PKA signaling domains have also been found to be located deep in the cytosol and away from the site of cAMP synthesis, as is the case, for example, for the centrosome-associated AKAPs () or the nuclear membrane-associated muscle AKAP (). Specific activation of the PKA pools, which are spatially segregated deep inside the cell, requires that cAMP is made available selectively in such compartments. How the highly hydrophilic, freely diffusible cAMP molecule can selectively activate deep intracellular targets without affecting PKA enzymes located closer to the site of cAMP synthesis remains to be defined. One example of the limited diffusion of cAMP is the generation of a subplasma membrane pool of the second messenger in response to PGE stimulation of HEK293 cells. To explain such compartmentalization, the hypothesis was formulated of a physical barrier that is possibly formed by elements of the endoplasmic reticulum and localizes underneath the plasma membrane, thereby limiting cAMP diffusion from the site of synthesis to the deep cytosol (). However, an important contribution to cAMP compartmentalization appears to derive from the activity of PDEs, as pointed out by a large body of indirect evidence based on altered cell functioning after selective PDE inhibition (; ; ; ; , ). PDEs can be targeted to subcellular compartments, including the plasma membrane, and can be recruited into multiprotein signaling complexes (), thereby providing a means to terminate the cAMP signal in a spatially restricted manner (). Such features suggested a role for PDEs as an enzymatic barrier to cAMP diffusion (for review see ; ; ). The model of a barrier to cAMP diffusion implies the generation of a gradient of cAMP with a higher concentration of the second messenger at the plasma membrane and a progressively lower concentration toward the deep cytosol. However, the direction of such a gradient is incompatible with the selective activation of a subset of PKA anchored deep inside the cytosol without concomitant activation of those PKA enzymes that are localized at the plasma membrane and closer to the site of cAMP synthesis. In this study, we set out to study the cAMP response to GPCR stimulation of the transmembrane AC with PGE in HEK293 cells with the aim of defining the role of PDEs in the generation of a subplasma membrane pool of high cAMP. We applied an imaging approach for real-time monitoring of cAMP dynamics in single living cells and used FRET-based biosensors selectively targeted to different subcellular domains. We found that the cAMP response in the subplasma membrane compartment is higher as compared with the bulk cytosol because of a compartmentalized PDE4D located in the cytosol that keeps the level of the second messenger low in this compartment. Indeed, genetic ablation of PDE4D inverted the direction of the cAMP gradient, resulting in a higher concentration of the second messenger in the cytosol as compared with the subplasma membrane compartment. On the contrary, genetic ablation of PDE4B did not affect the direction of the gradient. The cAMP gradient was completely abolished only when the expression of both PDE4B and PDE4D was blocked. These results demonstrate that the concerted activity of PDE4B and PDE4D is sufficient to generate an intracellular cAMP gradient in response to PGE in HEK293 cells. We also found that dislocation of PDE4D from its anchoring sites by means of the overexpression of catalytically inactive enzymes dissipated the cAMP gradient, indicating that the compartmentalization of PDE4D is responsible for shaping such a gradient. Intriguingly, we found that the inhibition of endogenous PKA with H89 completely abolished the cAMP gradient between the plasma membrane and bulk cytosol. On the contrary, the overexpression of PKA via PDE activation increased the steepness of the subplasma membrane cAMP gradient generated by PGE. These results indicate that PKA not only has a role in shaping the intracellular cAMP gradient upon receptor stimulation but also that PKA itself can reinforce the boundaries between the intracellular cAMP compartments and can self-regulate the specificity of its own activity. In this study, we show for the first time that a compartment deep inside the cell may accumulate higher levels of cAMP as compared with the bulk cytosol. These findings demonstrate that multiple and contiguous domains with different concentrations of cAMP can be generated simultaneously inside the cell irrespective of their distance from the site of synthesis of the second messenger. To explain such findings, we must assume that PDEs, rather than acting as enzymatic barriers to cAMP diffusion from the plasma membrane to the bulk cytosol, are organized to generate local drains that dump cAMP in defined compartments. Such a mechanism of control of cAMP diffusion may be more general and may also apply to the production of nonuniform intracellular cAMP domains in response to soluble AC activation. Therefore, we propose a new model whereby cAMP is free to diffuse from the site of synthesis and to accumulate in the cell to levels effective for PKA activation except in those domains in which localized PDEs degrade it to protect sensitive targets from inappropriate activation. DME, Opti-MEM, FBS, -glutamine, penicillin, trypsin/EDTA, PBS, and LipofectAMINE 2000 were purchased from Invitrogen. PGE and IBMX were obtained from Sigma-Aldrich. Restriction enzymes, T4 ligase, and shrimp AP were purchased from New England Biolabs, Inc. FuGENE-6 transfection reagent was obtained from Roche. The membrane-targeted version of the R-CFP subunit of the PKA-based cAMP sensor () was generated by inserting the N-terminal targeting signal MGCIKSKRKDNLNDD () from Lyn kinase at the N terminus of R-CFP. This becomes posttranslationally myristoylated and palmitoylated, resulting in its targeting to membrane rafts. The cAMP Epac1-based sensor, called H30, was provided by K. Jalink (The Netherlands Cancer Institute, Amsterdam, Netherlands) and corresponds to the CFP-Epac(δDEP-CD)-YFP sensor (). The membrane-targeted version of H30 was generated by fusion at the N terminus of the MGCIKSKRKDNLNDD plasma membrane–targeting sequence. The nuclear-targeted version of H30 was generated by fusion of the nuclear localization signal PKKKRKVEDA () at the C terminus (). Untagged PKA was generated by cloning both the RIIβ subunit of PKA and the Cα subunit of PKA into pCDNA3.1 (Invitrogen). Constructs for the expression of PKA subunits in were generated by cloning RIIβ, RIIβ, and Cα into pRSETB (Invitrogen). To determine apparent activation constants for PKA-GFP and PKA-GFP, the pRSETB vectors carrying the RIIβ, RIIβ, or Cα subunits were introduced into BL21(DE3) (Stratagene) for heterologous protein expression in bacteria. Individual subunits were subsequently purified with Ni–nitrilotriacetic acid resin according to the manufacturer's instructions (QIAGEN). Purified proteins were buffer changed by gel filtration to 20 mM MOPS, 150 mM NaCl, 5 mM MgCl, 100 μM ATP, and 1 mM β-mercaptoethanol, pH 7.0 (dialysis buffer), to remove imidazole. For PKA holoenzyme formation, PKA-R (PKA-GFP and PKA-GFP) and PKA Cα subunits (; ) were mixed in a molar ratio of 1.5:1.0, and dialysis was performed overnight with three buffer changes against dialysis buffer (). For the determination of activation constants, a coupled spectrophotometric assay () was performed using the peptide kemptide (LRRASLG; Biosynthan) as the substrate. Phosphotransferase activity was measured in an assay mixture consisting of 100 mM MOPS, pH 7.0, 10 mM MgCl, 1 mM phosphoenol pyruvate, 1 mM ATP, 200 μM NADH, 1 mM DTT, 15 U/ml lactate dehydrogenase (Roche), and 70 U/ml pyruvate kinase (Roche). For the reaction, 100 μl of assay mixture, 20 nM PKA holoenzyme, and 1 μl kemptide peptide (200 μM) were mixed in a quartz cuvette. The enzymatic activity (ΔE × min) was monitored at room temperature using a spectrophotometer (Lambda Bio UV/Vis; PerkinElmer). EC values for activation were determined by preincubating the PKA holoenzyme with increasing concentrations of cAMP for 1 min in assay mixture before starting the reaction with kemptide peptide. Data points were determined in duplicates, and experiments were repeated at least twice with similar results. Statistical analyses were performed using Prism 4.0 software (GraphPad Software). To determine in vivo apparent dissociation constants for H30, H30, and H30, HeLa cells expressing the cAMP sensor were injected with known concentrations of cAMP via a patch pipette, and FRET changes were recorded as described in the FRET imaging section. Cells were continuously superfused at 2 ml/min in a standard extracellular solution, ECS-1, which contained 150 mM NaCl, 5 mM KCl, 1 mM MgCl, 10 mM Hepes, 2 mM CaCl, 2 mM pyruvate, and 5 mM glucose, pH 7.4. Patch pipettes were filled with an intracellular solution, ICS-1, containing 120 mM K- aspartate, 10 mM TEA-Cl, 1 mM MgCl, 10 mM Hepes, 10 mM CsCl, 0.3 mM GTP-Na, 3 mM ATP-K (adjusted to pH 7.2 with KOH), 5 mM BAPTA, 1 mM thapsigargin, 0.1 mM IBMX, and 0.007–0.18 mM cAMP as indicated and filtered through 0.22-μm pores (Millipore). Only cells that displayed a seal resistance of >2 GΩ before achieving the whole cell configuration were retained. Cells admitted to the analysis after achieving the whole cell configuration were further characterized by an access resistance of <12 MΩ and an apparent membrane resistance of >300MΩ. At least six cells were analyzed for each concentration of cAMP. HEK293 cells were grown in DME containing 10% FBS supplemented with 2 mM -glutamine, 100 U/ml penicillin, and 100 μg/ml streptomycin in a humidified atmosphere containing 5% CO. For transient expression, cells were seeded onto 24-mm diameter round glass coverslips, and transfections were performed at 50–70% confluence with FuGENE-6 transfection reagent according to the manufacturer's instructions using 1–2 μg DNA per coverslip. Imaging experiments were performed after 24–48 h. To achieve the selective knockdown of PDE4B or PDE4D subfamilies, we used double-stranded 21-mer RNA duplexes (Dharmacon) targeted at regions of sequence that are unique to each of these subfamilies as described previously (). Each siRNA duplex was delivered into target cells via the reagent LipofectAMINE 2000 (Invitrogen). Specifically, 5 μl LipofectAMINE 2000 (1 mg/ml) was diluted in 100 μl Opti-MEM, and, separately, 125 pmol of each siRNA sample and 1 μg cAMP sensor DNA were diluted in 100 μl Opti-MEM. 200 μl siRNA–DNA transfection complexes were added to each well, and the plates were incubated for 3–4 h at 37°C (5% CO). These complexes were then removed and replaced with DME. Imaging experiments were performed after 48 h. FRET imaging experiments were performed 24–48 h after transfection. Cells were maintained in Hepes-buffered Ringer-modified saline containing 125 mM NaCl, 5 mM KCl, 1 mM NaPO, 1 mM MgS0, 5.5 mM glucose, 1 mM CaCl, and 20 mM Hepes, pH 7.5, at room temperature (20–22°C) and imaged on an inverted microscope (IX50; Olympus) with a 60× NA 1.4 oil immersion objective (Olympus). The microscope was equipped with a CCD camera (Sensicam QE; PCO), a software-controlled monochromator (Polychrome IV; TILL Photonics), and a beam-splitter optical device (Multispec Microimager; Optical Insights). Images were acquired using custom-made software and processed using ImageJ (National Institutes of Health). FRET changes were measured as changes in the background-subtracted 480/545-nm fluorescence emission intensities on excitation at 430 nm and expressed as either R/R, where R is the ratio at time t and R is the ratio at time = 0 s, or ΔR/R, where ΔR = R – R. Cells were stained with anti-PDE4B and -PDE4D monoclonal antibodies (FabGennix). AlexaFluor488-conjugated anti–mouse antibody was used as the secondary antibody. Confocal images were acquired with an inverted microscope (Eclipse TE300; Nikon) equipped with a spinning disk confocal system (Ultraview LCI; PerkinElmer) and a 12-bit CCD camera (Hamamatsu; Orca ER). Cells were excited using the 488-nm line of a krypton-argon laser (643-Ryb-A02; Melles Griot) for imaging YFP and the AlexaFluor488 fluorophore and using the 405-nm diode laser (iFlex2000; Point Source) for imaging CFP. Fig. S1 shows an evaluation of PGE receptor desensitization. Fig. S2 shows an evaluation of the dose-response relationship between PKA overexpression and the steepness of the cAMP gradient. Video 1 shows the dynamics interactions between RII-CFP and C-YFP upon cAMP binding and release. Supplemental material also provides information about determination of the velocity of FRET change for the cAMP sensors. Online supplemental material is available at .

, the cause of food poisoning and typhoid fever, has evolved sophisticated mechanisms to manipulate host cell functions. Central to this ability is a specialized organelle, the type III secretion system (TTSS), which mediates the transfer of a battery of bacterial proteins into host cells (). Through molecular and functional mimicry of host cell proteins, these bacterial effectors of virulence can stimulate or interfere with a variety of cellular activities (). For example, a subset of these effectors stimulates Rho family GTPases, triggering actin cytoskeleton rearrangements, macropinocytosis, and bacterial internalization into host cells (). The stimulation of Rho family GTPases also leads to activation of the MAPK signaling pathways as well as transcription factors, resulting in the reprogramming of host cell gene expression and the production of proinflammatory cytokines (; ). Two of these bacterial effectors are SopE and SopE2, which are highly related bacterial proteins with the capacity to catalyze nucleotide exchange on Rho family GTPase members (; ; ). Another of these effector proteins is SopB, a phosphoinositide phosphatase that can stimulate Rho family GTPase-dependent actin cytoskeleton rearrangements by unknown mechanisms (; ). After bacterial internalization, the activation of Rho family GTPases and the actin cytoskeleton rearrangements are reversed by the TTSS bacterial effector SptP, a GTPase-activating protein for several Rho family members (). Therefore, through a carefully orchestrated “yin and yang,” reversibly activates Rho family GTPases to induce its own uptake and the production of proinflammatory cytokines. Previous studies have reported the requirement of Cdc42 and Rac1 for -induced actin cytoskeleton remodeling, macropinocytosis, and MAPK activation (; ). However, it is not known whether these two GTPases act redundantly or whether their activation leads to specific responses. In vitro, SopE and SopE2 exhibit similar specificity, catalyzing exchange on several members of the Rho family of GTPases, including Rac1 and Cdc42 (; ; ). However, the relative contribution of these GTPases to the different responses induced by has not been determined. Evidence indicates that the SopB-mediated actin cytoskeleton rearrangements are also dependent on Rho family GTPases (). The observation that a noncatalytic mutant of SopB is unable to induce actin remodeling suggests that it must exert its function by activating an endogenous exchange factor, presumably through phosphoinositide fluxes. However, the actual GTPases targeted by the SopB activity or the identity of the exchange factor that this bacterial effector stimulates has not been defined. In this study, we describe distinct roles for different Rho family GTPases in -induced cellular responses. In addition, we report that SopB stimulates cellular responses by activating SH3-containing guanine nucleotide exchange factor (SGEF), an exchange factor for RhoG, which we found plays a central role in the actin cytoskeleton remodeling stimulated by . Together, these results reveal a remarkable level of complexity in the manipulation of Rho family GTPases by a bacterial pathogen. Previous studies have established that the cytoskeletal remodeling driving entry into nonphagocytic cells is triggered by three bacterial effector proteins (SopE, SopE2, and SopB) that, upon delivery into host cells, target the Rho family GTPases Cdc42 and Rac1 (; ; ; ). Despite strong evidence linking these bacterial effectors to entry, their specific contribution to the in vivo activation of Cdc42 and/or Rac1 is not known. Therefore, we examined the speci ficity of GTPase activation by serovar Typhimurium () mutant strains lacking and () or (). Cultured COS-2 cells were infected with these different mutant strains, and Cdc42 and Rac1 activation was assessed 20 min after infection, which was the time shown to result in their maximal stimulation by wild-type (). The absence of had little effect on the ability of to activate Cdc42 or Rac1 because cells infected with a induced similar stimulation levels of these GTPases as did wild type (). In contrast, the mutant, which solely relies on SopB for Rho family GTPase stimulation, did not activate Rac1 despite its ability to induce Cdc42 activation in a manner similar to wild type (). As expected, in the absence of a functional TTSS () or in the combined absence of , , and (), was unable to stimulate Cdc42 or Rac1. The lack of Rac1 activation by the strain was apparent even up to 30 min after infection (unpublished data), which is much later than the time when the actin cytoskeleton rearrangements induced by this strain are apparent. Therefore, these results suggest that SopB can mediate actin remodeling in a Rac1-independent manner. Collectively, these results show that in vivo, the effector proteins SopE, SopE2, and SopB differentially activate Rho family GTPases. Wild-type can robustly activate both Cdc42 and Rac1 in vivo. This is consistent with the observation that its Rho family GTPase exchange factors SopE and SopE2 can catalyze exchange on these GTPases in vitro (; ). However, it is not known whether the activation of Rac1 in vivo occurs through the direct action of the bacterial proteins on this GTPase or indirectly through the activation of Cdc42. Although there is no evidence that Cdc42 can be activated downstream of Rac1, it is well established that Rac1 activation can also occur subsequent to the activation of Cdc42 through the stimulation of downstream exchange factors (; ; ). To address the mechanisms of GTPase signaling stimulation by in vivo, we examined Cdc42 and Rac1 activation after the selective inhibition of either GTPase. As expected and as previously shown (), the expression of a dominant-negative mutant of Cdc42 (Cdc42) abolished GTP loading of endogenous Cdc42 after wild-type infection (unpublished data). However, the expression of Cdc42 also prevented the -induced activation of Rac1 (). These results would suggest that -induced Rac1 activation in vivo requires Cdc42 and, thus, may not be the consequence of a direct action of the bacterially encoded effectors (e.g., SopE and SopE2) on Rac1. However, the expression of a dominant-negative mutant of Rac1 (Rac) also inhibited -induced Cdc42 activation (unpublished data), suggesting that either of these GTPase mutants sequesters the bacterial exchange factors, thereby preventing the unambiguous interpretation of these results. To address this issue, we used RNAi to selectively inhibit Cdc42 and Rac1 expression. Delivery of specific RNAi constructs caused >90% depletion of Cdc42 or Rac1 in both intestinal Henle-407 and COS-2 cells (). Cells depleted of Cdc42 and/or Rac1 remained viable with normal cytoskeletal morphology for up to 3 d after treatment (unpublished data). We examined the effect of Cdc42 or Rac1 depletion on –induced Rac1 and Cdc42 activation, respectively. In contrast to what we observed when expressing dominant-negative mutants, the depletion of one GTPase had no effect on the activation of the other ( and not depicted). These results indicate that consistent with the in vitro specificity of the exchange factors SopE and SopE2 (; ), can independently activate Cdc42 and Rac1 in vivo. We have previously shown that the transient expression of SopE in cultured cells induces profuse actin rearrangements and membrane ruffling at the cell periphery and that these responses can be completely abrogated after the coexpression of Cdc42 or Rac (). However, our observation of the lack of specificity of these dominant-negative constructs () prompted us to reexamine the contribution of Cdc42 and Rac1 to SopE-mediated actin cytoskeleton rearrangements. COS-2 cells were cotransfected with plasmids expressing SopE, GFP (to mark transfected cells), and RNAi constructs to deplete either Cdc42 or Rac1. 2 d after transfection, cells were stained with rhodamine-phalloidin to visualize polymerized actin. Depletion of Rac1 by RNAi effectively abrogated SopE-mediated ruffling (). The inhibition resulting from Rac1 depletion was equivalent to that observed after the expression of dominant-negative Cdc42 (Cdc42; ) or Rac1 (Rac; ). In contrast, the depletion of Cdc42 by RNAi had no effect on the ability of SopE to induce actin cytoskeleton rearrangements (). These results indicate that actin rearrangements initiated by the effector SopE and presumably its highly related paralogue SopE2 require the GTPase Rac1 but not Cdc42. The observation that SopB-mediated actin remodeling and entry proceeds concomitant with the activation of Cdc42 but in the absence of Rac1 activation () suggested that must possess alternative mechanisms to induce Rho family GTPase-dependent actin remodeling than those used by SopE or SopE2. Therefore, we examined the ability of wild-type to induce actin cytoskeleton rearrangements and its uptake into cells depleted of Cdc42 or Rac1. Depletion of Cdc42 had no effect on the induction of actin reorganization () or bacterial entry () in cultured intestinal epithelial cells. In contrast, the depletion of Rac1 markedly reduced but did not completely abrogate bacteria-induced actin cytoskeleton rearrangements and entry into host cells (). Together, these results strongly suggest that Cdc42 is dispensable for the actin remodeling events generated during entry. In addition, the failure of Rac1 siRNA to reduce bacteria-induced ruffling and internalization to the same levels observed after the expression of Cdc42 coupled with the Rac1-independent actin remodeling seen upon SopB-mediated bacterial entry () indicate that additional host components must contribute to the stimulation of these cellular responses. The observation that depletion of Cdc42 did not result in any measurable defect in the ability of to stimulate actin cytoskeleton rearrangements prompted us to investigate the potential involvement of this GTPase in other cellular responses induced by these bacteria. In addition to inducing bacterial internalization into nonphagocytic cells, the TTSS is also essential for the stimulation of a rapid reprogramming of host gene expression, particularly the induction of proinflammatory cytokines such as IL-8 and TNF-α (). Stimulation of these nuclear responses is the consequence of the TTSS-dependent activation of MAPK pathways and transcription factors such as AP-1 and NF-κB and can be inhibited by the expression of dominant-negative Cdc42 (; ). Because the expression of Cdc42 nonspecifically blocks the activation of Rac1 (), we reevaluated the specific contribution of Cdc42 and Rac1 to the -mediated nuclear response using RNAi. Henle-407 cells were cotransfected with a plasmid encoding an IL-8 luciferase reporter and RNAi constructs to deplete either Cdc42 or Rac1. Transfected cells were then infected with wild-type , and the induction of IL-8 expression was assessed by measuring the luciferase levels in infected cells. Depletion of Rac1 by RNAi had no effect on the ability of to stimulate IL-8 transcription (). In contrast, the depletion of Cdc42 completely abrogated the stimulation of IL-8 expression upon bacterial infection (). Similar results were obtained using an NF-κB–dependent reporter cell line (unpublished data). These results indicate that the stimulation of Cdc42 is essential for the bacteria-induced nuclear responses. Furthermore, these results indicate that the different cellular responses induced by proceed through the activation of distinct Rho family GTPases: Rac1 for the actin cytoskeleton remodeling and Cdc42 for the nuclear responses. We then examined the specific contribution of the different effectors that activate Rho family GTPases (i.e., SopE/SopE2 and SopB) to the stimulation of nuclear responses. Henle-407 cells were transfected with the IL-8 reporter plasmid and were subsequently infected with wild-type , the isogenic , , or a strain lacking the three effectors (, and ). Both the and the mutant strains stimulated IL-8 expression but at lower levels than wild type (). In contrast, the triple mutant did not induce the expression of IL-8. These results indicate that SopE, SopE2, and SopB act redundantly to stimulate nuclear responses, which is consistent with the observation that all of them can activate Cdc42. Infection of cultured epithelial cells with an mutant, which induces actin remodeling through the activity of the effector protein SopB, resulted in the activation of Cdc42 but not Rac1 (). However, Cdc42 does not contribute to bacterial uptake (). These results suggested the possibility that SopB may stimulate actin-cytoskeleton responses through the stimulation of another Rho family GTPase. A potential candidate is RhoG (), a Rho family GTPase that has been previously shown to mediate dorsal ruffling and macropinocytosis in fibroblasts (). We first investigated whether infection of cultured epithelial cells resulted in the activation of RhoG. COS-2 cells were infected with wild-type , and the activation of RhoG was monitored by pull-down assays using the RhoG effector protein ELMO as an affinity probe for activated RhoG (). RhoG activation by wild-type was observed as early as 15 min after infection (unpublished data), although maximal activation was not seen until at least 30 min after infection (). The activation of RhoG was not observed in cells infected with a TTSS-deficient strain (), indicating that stimulation of this GTPase requires the activity of TTSS effector proteins. Importantly, RhoG stimulation was observed in cells infected with a mutant strain (), which induces signaling exclusively through SopB. SopB-mediated activation of RhoG was dependent on its phosphatase activity because the strain encoding a catalytically inactive SopB (SopB) was unable to stimulate RhoG (). A ΔsopB mutant, which signals though SopE and SopE2, was also able to activate RhoG (). This is consistent with the observation that SopE can catalyze exchange on RhoG in vitro (). A strain lacking , , and was not able to stimulate RhoG activation (), indicating that these three effector proteins can account for all of the activation of this GTPase by wild-type . GFP-RhoG was recruited to the site of entry shortly after infection ( and Videos 1–4; available at ) and remained associated with the -containing vacuole as well as with empty vacuoles generated during the stimulation of membrane ruffling (, inset; and Videos 1–3). This pattern of recruitment and localization closely resembled the localization of GFP-Rac1 during infection (Video 5), although it differs from that of GFP-Cdc42 (Video 6). Collectively, these results indicate that upon infection, elicits the transient activation of at least three different Rho family GTPases—Cdc42, Rac1, and RhoG—through the coordinated activities of its TTSS effector proteins SopE, SopE2, and SopB. We then examined the potential role of RhoG in -induced and, more specifically, SopB-mediated actin remodeling. RhoG was depleted from intestinal Henle-407 cells by the transfection of RNAi constructs designed to target this GTPase (), and the ability of and mutant strains to stimulate actin remodeling in these cells was examined by fluorescence microscopy (). The depletion of RhoG resulted in a marked inhibition of the ability of wild-type to induce actin-cytoskeleton rearrangements (). However, the inhibition was reproducibly not as pronounced as that observed by the expression of dominant-negative Cdc42 () or Rac1 (; ), which nonspecifically blocks signaling through several GTPases. However, simultaneous depletion of Rac1 and RhoG achieved the maximum level of inhibition (), indicating that –induced actin remodeling is the result of signaling through both GTPases. Importantly, the depletion of RhoG completely abolished the ability of the mutant to induce actin remodeling (). This mutant strain signals to Rho GTPases exclusively through SopB; therefore, these results indicate that this effector protein stimulates actin remodeling through the activation of RhoG. Consistent with this hypothesis, the depletion of Rac1 had no effect on the ability of the mutant to induce actin remodeling (). Because the expression of SopB results in strong toxicity (), the effect of RhoG depletion on the actin cytoskeleton rearrangements induced by transiently expressed SopB could not be evaluated. Collectively, these results indicate that induces actin remodeling by targeting both Rac1 and RhoG via distinct TTSS effector proteins. SopB-mediated actin cytoskeleton rearrangements require its phosphoinositide phosphatase activity () and are dependent on RhoG. Because SopB lacks measurable in vitro nucleotide exchange activity toward Rho family GTPases (unpublished data), these observations suggest that SopB, presumably through its ability to induce phosphoinositide fluxes, must exert its function by activating an endogenous RhoG exchange factor. Therefore, we investigated the potential involvement of several cellular exchange factors that have been shown to have the ability to catalyze nucleotide exchange on RhoG. Cells were cotransfected with plasmids expressing siRNAs directed to Dbl (), Dbs (), Duo (Kalirin; ; ), SGEF (), Vav, and Vav3 () along with a plasmid expressing GFP to identify transfected cells. Cells were infected with the mutant strain, which induces actin cytoskeleton rearrangements exclusively through the activity of SopB. Depletion of Dbl, Dbs, Duo (Kalirin), Vav, or Vav3 had little or no effect on the ability of the mutant strain to induce actin cytoskeleton rearrangements (). However, the depletion of SGEF markedly impaired actin cytoskeleton rearrangements induced by the strain (). Furthermore, the depletion of SGEF abolished the ability of the strain to activate RhoG (Fig. S1, available at ). The effect of SGEF depletion on the ability of this strain to stimulate actin remodeling was equivalent to the effect of RhoG depletion (), suggesting that most of the SopB-mediated signaling to RhoG is transduced through its exchange factor SGEF. Consistent with its involvement in –induced actin remodeling, SGEF-GFP was efficiently recruited to the site of bacterial entry into host cells ( and Videos 7 and 8; available at ). Previous studies have indicated that the SopB-mediated actin cytoskeleton rearrangements are morphologically distinct from those induced by SopE or SopE2 (; ). Although the expression of SopE in cultured cells resulted in actin cytoskeleton rearrangements and membrane ruffling throughout the cell () similar to that observed upon the overexpression of constitutively active Rac1 (), the expression of SopB within cultured cells, although toxic, induced much more localized membrane ruffles and actin remodeling in the few cells that can be visualized in a standard transfection (; ; ). Consistent with its involvement in SopB-mediated actin remodeling, the overexpression of SGEF induced localized membrane ruffles that closely resembled those induced by SopB (; ). Collectively, these results indicate that SGEF is the exchange factor activated by SopB to induce actin cytoskeleton rearrangements and mediate bacterial entry. Many bacterial pathogens target Rho family GTPases with a variety of toxins (). These toxins most often introduce covalent modifications on these critical signal transduction molecules, which usually result in their irreversible modification, leading to their constitutive activation, their complete inactivation, or their degradation. On the other hand, modulates Rho family GTPases with bacterial proteins that precisely mimic the function of endogenous modulators of this family of small GTP-binding proteins (; ). Previous studies have defined the precise mechanisms by which two -encoded exchange factors, SopE and its highly related paralogue SopE2, catalyze the activation of Rho GTPase family members (; ; ). However, the contribution of the specific Rho family members to –host cell interactions has remained elusive. In our studies, we have defined the specific contribution of Rho family GTPases to different -induced cellular responses, thus revealing a hitherto unknown complexity in the functional consequences of the modulation of these molecular switches by this bacterial pathogen. Our studies revealed that different Rho GTPases are required for different responses induced by during its interaction with intestinal epithelial cells. Previous studies using the dominant-negative mutant of different Rho GTPases have indicated that Cdc42 was central for the ability of to induce actin cytoskeleton rearrangements and to enter nonphagocytic cells (; ). However, using RNAi, we found that Cdc42 is dispensable for these responses. Instead, we found that Rac1 is required for efficient induction of the actin remodeling that leads to bacterial uptake into nonphagocytic cells. Our studies also showed that Cdc42 is essential for the transduction of signals induced by the TTSS that lead to nuclear responses, a process in which Rac1 seems to play an unimportant role. Consequently, these studies have uncovered a remarkable mechanism by which a bacterial pathogen has the capacity to generate a rather diverse set of cellular responses that are essential for its pathogenicity through the activity of even a single bacterial effector. In addition to SopE and SopE2, a previous study has shown that another bacterial effector, SopB, can also induce actin cytoskeleton rearrangements and bacterial uptake into cells (). Although it has been shown that the phosphoinositide phosphatase activity of SopB is required for the stimulation of these responses (), the mechanisms by which the signal generated by this bacterial effector protein is transduced to the actin cytoskeleton has remained unknown. In this study, we showed that the stimulation of actin cytoskeleton rearrangements by SopB is dependent on RhoG, a Rho family GTPase not previously implicated in –host cell interactions. RhoG has been previously implicated in modulation of the actin cytoskeleton as well as in the induction of membrane ruffling and macropinocytosis (; ). A mutant strain, which relies solely on SopB to induce actin cytoskeleton rearrangements, was able to induce the activation of RhoG but not Rac1. More importantly, cells that were depleted of RhoG by RNAi were markedly impaired in their ability to rearrange their actin cytoskeleton upon infection with this mutant strain. We have previously shown that SopE can catalyze RhoG exchange in vitro (), and, consistent with this observation, we have shown in this study that SopE and its highly related paralogue SopE2 can activate RhoG in vivo. The observation that SopB lacks in vitro exchange activity on Rho family GTPases combined with the observation that SopB-mediated actin remodeling requires its phosphoinositide phosphatase activity suggested that this bacterial effector must activate RhoG through the stimulation of an endogenous cellular exchange factor. We found that such an exchange factor is SGEF because its depletion from cells effectively prevented the ability of to signal to the actin cytoskeleton through the activity of SopB. SGEF is ubiquitously expressed and has been implicated in the induction of localized dorsal membrane ruffles and macropinocytosis (), which closely resembles the cellular responses induced by and SopB in particular. We have previously found that the ability of to induce macropinocytosis is strictly dependent on the phosphoinositide phosphatase activity of SopB (). Indeed, an mutant strain lacking SopB or expressing a catalytically inactive mutant was fully capable of inducing membrane ruffling (through SopE and SopE2) but failed to induce macropinocytosis (). The observation that the expression of constitutive active RhoG or its exchange factor SGEF but not Rac1 leads to profuse macropinocytosis () is consistent with these observations and with the role of SGEF and RhoG in –host cell interactions. It is unknown how SopB may activate SGEF. The requirement of the phosphatase activity of SopB to induce actin rearrangements () as well as to activate RhoG (this study) suggests that the phosphoinositide fluxes induced by its catalytic activity may directly activate SGEF. Consistent with this hypothesis, SGEF possesses a phosphoinositide-binding pleckstrin homology domain, which is essential for its membrane localization and activity (). In summary, our study has revealed a remarkable complexity in the stimulation of cellular responses by through the specific and selective activation of Rho family GTPases (). Both directly through action of the bacterially encoded exchange factors SopE and SopE2 and indirectly through activation of the endogenous RhoG exchange factor SGEF by SopB, this bacterial pathogen orchestrates the coordinated activation of specific Rho family GTPase members to stimulate specific cellular responses. These constitute remarkable examples of pathogen adaptations to modulate specific cellular functions. The wild-type strain of serovar Typhimurium () SL1344 and its isogenic derivatives used in this study, (SB1120; ), (SB1301), (SB1302; ), and (SB136; ), have been previously described. encodes an essential component of the invasion-associated TTSS, and, therefore, strains lacking this gene are completely defective in their ability to stimulate actin cytoskeleton rearrangements, membrane ruffling, bacterial uptake, and nuclear responses (; ; ,). A derivative expressing the catalytic mutant SopB was constructed by allelic exchange as previously described (). All bacterial strains were cultured under conditions that stimulate the expression of the pathogenicity island-1–encoded TTSS (). In brief, overnight cultures were diluted 1:25 in L-broth containing 0.3M NaCl, incubated on a rotating wheel for 3 h at 37°C until an OD of ∼0.9, and used immediately for infection. Where appropriate, 0.1% l-arabinose was added to cultures at the early logarithmic phase of growth (OD of 0.4) to induce expression of the dsRed gene under the control of the pBAD promoter. COS-2 cells were maintained in DME (Invitrogen) supplemented with 10% heat-inactivated FCS (Gemini), 100 U/ml penicillin, and 50 μg/ml streptomycin. Henle-407 cells were grown in DME supplemented with 10% heat-inactivated bovine calf serum (Hyclone) and penicillin/streptomycin. 293T/NF-κB–luc (Panomics) cells were passaged in DME containing 10% FBS (Gemini), penicillin/streptomycin, and 100 μg/ml hygromycin (Sigma-Aldrich) as recommended by the manufacturer. For infection experiments, cells were washed and passaged into media lacking penicillin/streptomycin overnight. Eukaryotic expression vectors encoding GFP-tagged wild-type Cdc42 and Rac1 as well as their dominant-negative (Cdc42 and Rac1) and constitutively active (Cdc42 and Rac1) mutants have been previously described (). Wild-type RhoG and SGEF were subcloned into vectors pRK5myc and pEGFP-C1 (CLONTECH Laboratories, Inc.) by PCR using pECEFL-Au5 RhoG (a gift from X. Bustelo, University of Salamanca, Salamanca, Spain) and clone 10624662 (American Type Culture Collection), respectively, as templates. To fluorescently label , the plasmid vector pdsRed.T3S4T (a gift from D. Bumann, Max-Planck Institute, Berlin, Germany; ), which encodes the bacterially optimized dsRed protein under the control of an arabinose-inducible promoter, was transformed into the different bacterial strains by electroporation. To measure IL-8 transcription, a firefly luciferase reporter plasmid (pSB2805) was constructed by replacing the CAT gene of pIL-8-CAT () with the firefly luciferase gene (Promega). To standardize transfection experiments, a plasmid was constructed (pSB2806) in which the Renilla luciferase, which was derived from pRL-TK (Promega), was cloned into the eukaryotic expression vector pCDNA3.1 (Invitrogen). Plasmids pSB1141, a bicistronic vector encoding SopE and GFP, and pSB2807 encoding YFP-tagged SopB have been previously described (). Plasmid DNA was purified using the endotoxin-free Maxiprep kit (QIAGEN) and used for the transfection of cells with LipofectAMINE 2000 (Invitrogen) according to the manufacturer's instructions. Depletion of endogenous Cdc42 was performed using a short hairpin sequence targeting nucleotides 296–318 of human Cdc42. Oligonucleotide pairs A (ACAGTGGTGAGTTATCTCAGGAAGCTTGCTGAGATAACTCACCACTGTCCATTTTTT) and B (GATCAAAAAATGGACAGTGGTGAGTTATCTCAGCAAGCTTCCTGAGATAACTCACCACTGTCG) were annealed and ligated into the vector pSHAG (a gift from G. Hannon, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY; ). Attempts to knockdown endogenous Rac1 and RhoG using a similar vector-based approach proved unsuccessful. Thus, silencing of Rac1 and RhoG gene expression was achieved using synthetic SMARTpools (Dharmacon), each comprising four proprietary siRNA sequences. Three independent RNAi constructs (denoted as 1–3) targeting the RhoG GEFs Dbl, Dbs, Duo (Kalirin), SGEF, Trio, Vav1, and Vav3 comprising short hairpin RNA sequences cloned into pSUPER Retro (Oligoengine) were provided by A. Schmidt and A. Hall (Memorial Sloan Kettering Institute, New York, NY). In all cases, siRNA and short hairpin RNA constructs were transfected into COS-2 and Henle-407 cells using LipofectAMINE 2000 (Invitrogen). Where appropriate, RNAi constructs were cotransfected with pEGFP-C1 at a ratio of 5:1 to mark transfected cells. Activation of endogenous Cdc42 and Rac1 GTPases was measured by p21-activated kinase (PAK)–Cdc42–Rac1 interaction binding (CRIB) pull-down assays as previously described (). In brief, COS-2 cells transfected as appropriate in 10-cm dishes were washed twice with prewarmed HBSS and infected with strains at an MOI of 100. At different time points, COS-2 cells were washed twice in cold HBSS and lysed by scraping on ice in cold radioimmunoprecipitation assay buffer (50 mM Tris, pH 8, 150 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 10 mM MgCl, 0.2 mM PMSF, and Complete protease inhibitor cocktail; Roche). Lysates were cleared by centrifugation at 14,000 rpm for 10 min at 4°C, an aliquot was saved to assess total GTPase levels, and the remaining lysates were divided in two aliquots to assess levels of active Cdc42 and Rac1. Lysates were incubated for 1 h at 4°C with ∼20 μg GST fused to the CRIB domain of PAK (GST-PAK-CRIB) that was prebound to glutathione agarose beads (50% slurry; prepared as previously described by ) to precipitate GTPases. Beads were subsequently washed twice in cold wash buffer (50 mM Tris, pH 8, 150 mM NaCl, 1% Triton X-100, 10 mM MgCl, 0.2 mM PMSF, and protease inhibitor cocktail). Equal amounts of beads and total cell lysate were resolved by SDS-PAGE, transferred to polyvinylidene difluoride membranes (Immobilon-P; Millipore), and immunoblotted using anti-Cdc42 or Rac1 antibodies. For RhoG, GTP loading was assessed on exogenously expressed myc-tagged RhoG using GST-ELMO 2 (a gift from X. Bustelo) coupled to glutathione agarose beads as bait. The GST fusion protein was expressed and purified as for GST-PAK-CRIB, and the pull-down assays were performed as described for Cdc42 and Rac1 with the exception of the cell lysis buffer (20 mM Tris, pH 7.6, 150 mM NaCl, 5 mM MgCl, 0.5% Triton X-100, 1 mM DTT, 0.1 mM sodium vanadate, 1 mM PMSF, and protease inhibitor cocktail). Western blots were probed with mouse anti-myc (clone 9E10; Santa Cruz Biotechnology, Inc.) antibody. In all cases, blots were visualized using chemiluminescence reagents (Supersignal chemiluminescence substrate; Pierce Chemical Co.), and, where appropriate, bands were scanned and their intensities were quantified using the public domain ImageJ program (National Institutes of Health; ). induced ruffling assays in COS-2 and Henle-407 cells have been previously described (). In brief, semiconfluent cells were infected with different strains for various time intervals (cells were infected with wild-type for 20 min and with for 30 min with an MOI of 40 and 100, respectively). After infection, cells were washed twice with HBSS, fixed with 4% PFA for 20 min, permeabilized with 0.05% Triton X-100 for 5 min, and stained with FITC- or rhodamine-labeled phalloidin to visualize the actin cytoskeleton. To visualize internalization into host cells, we used an assay that takes advantage of the inaccessibility of internalized bacteria to gentamicin, an antibiotic that kills bacteria by inhibiting protein synthesis. In brief, cells were infected with an strain (MOI of 20) containing a plasmid that expresses the dsRed fluorescent protein under an arabinose-inducible promoter. 45 min after infection, cells were washed twice with warm HBSS, and noninternalized bacteria were killed by the addition of gentamicin (100 μg/ml in DME). 45 min after gentamicin addition, the media was replaced with DME containing 100 μg/ml gentamicin and 0.1% arabinose to induce dsRed expression only in internalized bacteria, as bacterial protein synthesis is inhibited by gentamicin in extracellular bacteria. Internalized were visualized 2 h after the addition of arabinose by fluorescence microscopy. This assay was validated by carrying out parallel experiments using an antibody-based differential staining of internalized versus external bacteria as previously described (), obtaining equivalent results. The SopE-mediated ruffling assay has been previously described (). Immunofluorescence studies used DAPI (Sigma-Aldrich) to detect bacteria (in some experiments), TRITC-conjugated phalloidin (Sigma-Aldrich) to stain the host actin cytoskeleton, AlexaFluor596- and -488–conjugated goat anti–rabbit IgG and goat anti–mouse IgG antisera (Invitrogen), and rabbit anti lipopolysaccharide (Difco Laboratories). Fluorescence microscopy was performed using a confocal laser-scanning microscope (LSM 510; Carl Zeiss MicroImaging, Inc.) or an inverted microscope (Eclipse TE2000-U; Nikon) equipped with oil immersion plan Apo 60× NA 1.4, 100× NA 1.4, or dry plan Fluor 40× NA 0.6 objectives (Nikon) and a CCD camera (MicroMAX RTE/CCD-1300Y; Princeton Instruments). Acquisition and analysis of still images were performed using MetaMorph Imaging Software (version 6.1; Universal Imaging Corp.). For live cell imaging, cells were maintained at 37°C. Henle-407 cells were transiently transfected with the dual luciferase reporter constructs pSB2805 and pSB2806 together with different RNAi constructs. 2 d after transfection (including overnight serum starvation), cells were infected with wild-type or an isogenic TTSS-defective mutant at an MOI of 25 for 40 min. Cells were washed with HBSS and further incubated with DME containing 100 μg/ml gentamicin for 4 h. After passive cell lysis, firefly and Renilla luciferase levels were determined using the Dual Luciferase Reporter assay (Promega) according to the manufacturer's instructions. Transfection efficiency was normalized by the comparison of IL-8–induced firefly luciferase levels with that of constitutively expressed Renilla luciferase. Induction of the IL-8 reporter by strains was expressed relative to that of an isogenic mutant strain, which is defective for type III secretion and the induction of cellular responses. To determine the activation of NF-κB by , we used a 293T NF-κB luciferase reporter cell line (Panomics). Cells were transiently transfected with different RNAi constructs and assayed as indicated above for IL-8. Induction of the reporter by wild-type was expressed relative to that of an isogenic mutant. Fig. S1 shows that RhoG activation by strains requires SGEF. Video 1 is a time-lapse video showing RhoG localization during infection. Video 2 shows that GFP-RhoG labels vacuoles generated during entry. Video 3 is a confocal time-lapse video showing GFP-RhoG localization during infection, and Video 4 is a 3D projection showing GFP-RhoG recruitment to a -induced ruffle. Video 5 is a time-lapse video showing Rac localization during infection. Video 6 is a time-lapse video showing Cdc42 localization during infection. Videos 7 (wide field) and 8 (confocal) are time-lapse videos showing GFP-SGEF localization during infection. Online supplemental material is available at .

is the leading cause of dysentery (shigellosis) worldwide, often provoking severe colitis and diarrhea that quickly dehydrates afflicted patients, causing high morbidity and mortality, with up to 160 million cases, and 2.6 million deaths, annually (). thrives in the human intestine, where it utilizes a type III secretion system typical of many Gram-negative pathogens to inject invasin proteins (IpaA–D) into intestinal epithelial cells (). The IpaA invasin is a 70-kD (633-residue) protein required for pathogenesis, and it facilitates entry into the host cell by binding to the N-terminal domain of vinculin (; ), which is an essential protein that links the actin cytoskeleton to adhesion receptors in focal adhesions and adherens junctions (). However, the precise mechanism by which subverts vinculin functions is unknown. The crystal structure of vinculin in its resting, inactive state revealed that the protein contains five loosely packed α-helical bundle domains clamped together through intramolecular hydrophobic interactions of vinculin's N-terminal, seven-helical bundle domain (Vh1) with its five-helical bundle tail (Vt) domain (; ; ). Vinculin activation requires severing of the Vh1–Vt interaction (; ), and it was thought to be directed by the binding of phosphatidylinositol-4,5-bisphosphate (PIP) to the tail domain of vinculin (; ). However, the PIP binding site is occluded in inactive vinculin (), and vinculin mutants defective in binding to PIP localize to focal adhesion complexes (). More recently, the vinculin-binding sites (VBSs) of the adhesion proteins talin and α-actinin have been revealed as triggers that are sufficient to activate vinculin by binding to its Vh1 domain and provoking remarkable changes in its structure that displace Vt from a distance (; ; , ; ). Vinculin plays an important role in pathogenesis, as entry of the pathogen is markedly impaired in vinculin-deficient cells (). Entry of the pathogen also depends on its invasin protein IpaA, which binds to vinculin through an interaction with vinculin's Vh1 domain (). IpaA is also capable of activating the latent F-actin– and vinexin-β–binding functions of vinculin (; ). The amphipathic α-helical VBSs of talin and α-actinin are sufficient to alter the overall conformation of vinculin and to activate its ability to bind to F-actin (). Therefore, we tested the hypotheses that 's IpaA invasin would also harbor a VBS that is capable of triggering vinculin activation, and that this VBS plays an essential role in pathogenesis. The VBSs of talin and α-actinin are amphipathic α helices (; ; ). IpaA binds to vinculin (; ) through its C-terminal domain (IpaA-Cterm; IpaA residues 557–633; unpublished data). Therefore, we predicted that the ability of IpaA to bind to vinculin would be directed by an amphipathic α-helical VBS within this domain. Indeed, structure-based predictions revealed two potential amphipathic α helices in the IpaA-Cterm domain that might serve as VBSs. One, which we coined IpaA-VBS, was located at the C terminus of IpaA (residues 611–633; ), whereas the second, IpaA-VBS2 (residues 565–587), is located only 24 residues N-terminal to IpaA-VBS (). To initially evaluate the role of the potential VBSs of IpaA in binding to vinculin, we generated strains of the deletion mutant of that were engineered to re-express wild-type IpaA protein (), or mutant IpaA proteins lacking IpaA-VBS (; deletion of residues 589–633) or IpaA-VBS2 (; deletion of residues 565–586), or the entire C terminus of IpaA (; lacking residues 559–633). All IpaA mutant proteins were secreted at levels similar to that of wild-type IpaA, as determined by Western blot analysis (). Wild-type and mutant IpaA proteins were tested for their ability to bind to vinculin in an overlay assay (). IpaA-ΔCterm failed to bind to vinculin, whereas binding to vinculin was detected for both IpaA-ΔVBS and -ΔVBS2 proteins (). Therefore, the C terminus of IpaA has two potential VBSs for vinculin, and either of them is sufficient to bind to vinculin. To determine whether these potential VBSs of IpaA could also bind to vinculin in solution, we tested whether IpaA-VBS, -VBS2, or -Cterm could form stable complexes with recombinant Vh1 protein, as detected by native PAGE assays (). Indeed, IpaA-VBS, -VBS2, or -Cterm readily formed stable, single complexes with the Vh1 domain of vinculin that were easily distinguishable from unbound Vh1 (). In contrast, a peptide comprised of the 23 residues linking the two IpaA-VBSs (IpaA-NB, for “nonbinder”; residues 588–610; ) failed to bind to the Vh1 domain of vinculin (). Therefore, the two VBSs of IpaA can also bind to vinculin in solution. To determine the affinity of IpaA-VBS versus -VBS2 for vinculin, and of the entire IpaA-Cterm domain, surface plasmon resonance (SPR) binding assays were performed. IpaA-VBS bound to the Vh1 domain of vinculin with very high-affinity ( of 0.11 nM; ), 16–300-fold higher than that of the VBSs of talin and α-actinin (with s ranging from 1.8 to 32.8 nM; ; , ). of 6.61 nM (), which is comparable to the highest affinity VBS of talin (talin-VBS3; ) and which approaches that estimated for the intramolecular head–tail interaction that clamps vinculin in its inactive state (). Nonetheless, the affinity of IpaA-VBS2 for Vh1 was still 60-fold lower than that of IpaA-VBS for the N-terminal domain of vinculin. of the Vh1–IpaA-Cterm interaction was 6.2 femtomolar, and the of the IpaA-Cterm interaction with full-length vinculin was 35.2 picomolar (). Furthermore, off-rate analysis demonstrated that, once bound to the Vh1 domain of vinculin, IpaA-Cterm essentially did not dissociate (), and that it also had a very slow off-rate for full-length vinculin (). Indeed, the affinity of the IpaA-Cterm domain for vinculin is approximately two orders of magnitude greater than that reported for any other binding partner of vinculin, including the low nanomolar affinity of the intramolecular interaction of vinculin's head and tail domains, which clamp vinculin in its inactive state (). To test the effects of 's IpaA-VBSs on IpaA function, we analyzed the invasion and cell–cell dissemination of wild-type (strain M90T; ) and mutant strains complemented with full-length IpaA ), or IpaA derivatives lacking the IpaA-Cterm domain (), the IpaA-VBS (), or the IpaA-VBS2 (). We first tested if IpaA translocation into infected HeLa cells was comparable between the strain and mutant IpaA strains by immunoprecipitating translocated IpaA derivatives from lysates of HeLa cells infected with the various strains. As negative controls, we used the deletion mutant strain, as well as the ( deficient type III secretion apparatus) mutant strain, which, as expected, did not show any translocated into HeLa cells. Although, as observed for other translocated type III effectors (), some degradation of IpaA proteins was observed, the overall levels of the IpaA-ΔVBS, -ΔVBS2, and -ΔCterm proteins in infected HeLa cells were similar to those of wild-type IpaA protein in HeLa cells infected with the strain (). To determine the roles of the VBSs of IpaA in the recruitment of vinculin during bacterial entry, we assessed vinculin localization at sites of entry by immunofluorescence assays. As expected, vinculin was efficiently recruited to bacterial-induced actin foci in HeLa cells infected with wild-type or with the strains, but was not recruited to actin foci in cells infected with the mutant (). Interestingly, vinculin was also not recruited to actin foci in cells infected with the , , or strains (), indicating that both VBSs are required for efficient recruitment at bacterial entry sites. Although quantification of the average intensity of vinculin staining at sites of actin foci confirmed the deficiency of vinculin recruitment in cells infected with the , , or strains (), there were also differences in the sublocalization of vinculin observed at entry foci induced by the , , and strains (). This difference was most clearly apparent in the formation of the “actin cup” that normally surrounds the internalized pathogen (), and which is also characterized by intense vinculin staining around the internalized bacterium (, WT and ). Notably, actin–vinculin cups around the internalized bacterium were not observed in cells infected with the strain, and these were reduced, by 10- or 4-fold, in cells infected with the and strains, respectively (). Therefore, IpaA-VBS and -VBS2 of are both essential for efficient recruitment of vinculin to bacterial actin foci at entry sites (), whereas the IpaA-VBS plays a more important role in recruiting vinculin to the actin cups that surround the internalized pathogen (). We next tested whether 's IpaA mutants were compromised in their ability to invade the host cell by performing gentamicin protection assays (). Notably, the and deletion mutants were markedly impaired in their ability to invade HeLa cells compared with wild-type or the deletion strain engineered to reexpress IpaA (/; ). In contrast, the / strain was as effective as wild-type and the / strain in cell entry (). The reduction in bacterial invasion linked to these mutations was, however, not as drastic as that observed for the mutant strain, indicating that in addition to the IpaA C terminus, other domains of IpaA also contribute to bacterial invasion. During replication in epithelial cells, disseminates from cell to cell using actin-based motility (). To further address the role of the VBSs of IpaA in invasion and also in cell–cell spread, we performed a modified plaque assay on polarized monolayers of Caco-2 colon epithelial cells. In this assay, exposure of the monolayers to low bacterial multiplicities of infection leads to discrete entry events, which are visualized as bacterial plaques after a 6-h incubation in the presence of gentamicin. This prevents de novo invasion yet allows for bacterial dissemination to neighboring cells. Thus, the frequency of plaque formation reflects bacterial invasion, whereas bacterial dissemination is assessed by quantifying the number of infected cells per plaque (). When invasion was analyzed by counting plaques induced by the various strains on polarized monolayers, results similar to those in HeLa cells were obtained, with the / strain forming essentially as many plaques as the / strain, whereas the / and strains showed a similar twofold reduction in plaque frequency (). Remarkably, analysis of bacterial dissemination provided results that paralleled invasion. Plaques formed by the mutant contained two times less infected cells than the strain (). Importantly, the or mutant strains were as defective in their dissemination as the strain lacking IpaA. In contrast, the spread of the / strain was similar to that of the / strain that expresses wild-type IpaA (). Therefore, IpaA-VBS, but not IpaA-VBS2, plays an important role for entry into the host cell, and for cell–cell spread of . The high affinity of IpaA-VBS or -VBS2 for vinculin, and their sequence similarity with the VBSs of talin and α-actinin (), suggested that, like the VBSs of talin and α-actinin (; ; , ), 's IpaA-VBSs might also be sufficient to sever vinculin's head–tail interactions, which normally clamp the molecule in its inactive conformation (; ; ). Indeed, native gel analyses demonstrated that relatively low molar ratios of either IpaA-VBS, -VBS2, or -Cterm were sufficient to completely displace Vt from preexisting Vh1–Vt complexes (). Therefore, the VBSs of IpaA are sufficient to disrupt the intramolecular head–tail interactions of vinculin, which is the initiating event of vinculin activation (; ). The VBSs of talin and α–actinin bind to vinculin in a mutually exclusive manner and can efficiently displace one another from preexisting complexes with Vh1 (). The very high affinity of 's IpaA-VBSs for vinculin (), however, suggested that the interaction of the VBSs of IpaA with vinculin in solution might be resistant to displacement by the VBSs of talin or α-actinin. Indeed, the VBSs of talin (VBS1 and VBS3) and α-actinin (αVBS) were not capable of displacing IpaA-VBS from preexisting Vh1–IpaA-VBS complexes (). Similarly, despite a 60-fold lower affinity of IpaA-VBS2 for vinculin (), talin-VBS3 and αVBS were not capable of displacing IpaA-VBS2 from preexisting Vh1–IpaA-VBS2 complexes (). Therefore, once bound, IpaA would be predicted to compromise contacts of vinculin with its endogenous partners. Surprisingly, IpaA-VBS and -VBS2 were relatively ineffective at displacing talin-VBS1, talin-VBS3, or αVBS from preexisting Vh1–talin-VBS1,–talin-VBS3, or –αVBS complexes (). These results suggest that 's IpaA protein would preferentially interact with pools of free vinculin, rather than with vinculin bound by talin or α-actinin at sites of focal adhesions or adherens junctions. A hallmark of vinculin activation is its ability to bind to F-actin (), and IpaA activates this latent binding activity (). To address the role of IpaA-VBS and IpaA-VBS2 in this process, we used F-actin cosedimentation assays (). Although native vinculin or vinculin incubated with the IpaA-NB peptide failed to bind to F-actin, IpaA-VBS–, IpaA-VBS2–, or IpaA-Cterm–bound vinculin efficiently cosedimented with F-actin (). Finally, neither IpaA-VBS or IpaA-Cterm contained a cryptic F-actin–binding domain, as IpaA-VBS or IpaA-Cterm bound to a vinculin mutant that lacks its actin-binding tail domain (hVΔVt; human vinculin residues 1–840; ) failed to cosediment F-actin (). Therefore, the VBSs of IpaA, and its C-terminal domain, are sufficient to promote binding of vinculin to F-actin. The amphipathic α-helical nature of the VBSs of IpaA, their high-affinity binding to vinculin, and their similarity to the VBSs of talin and α-actinin in displacing the head–tail interaction of vinculin, suggested they might bind to vinculin in a manner akin to talin or α-actinin. The VBSs of talin and α-actinin insert between Vh1 helices α1 and α2, and make extensive hydrophobic contacts with Vh1 helices α1, α2, and α4 that provoke drastic changes in its structure, displacing Vt from a distance (; ; ; ). To determine the precise nature of the IpaA–vinculin interaction, we solved the crystal structure the Vh1 domain in complex with IpaA residues 602–633 (containing the IpaA-VBS; , and Tables S1 and S2, available at ), as well as with IpaA residues 565–587 (IpaA-VBS2; and Tables S3 and S4). These structures established that IpaA-VBS and -VBS2 are indeed amphipathic α helices (), confirming their sequence alignments () and predictions of their secondary structure (). Both IpaA-VBS and -VBS2 bind to the Vh1 domain of vinculin in an intimate fashion, largely through van der Waals interactions, where the hydrophobic faces of these VBSs interact with the hydrophobic core of the N-terminal four-helical bundle subdomain of Vh1 (), events that would displace Vt's interactions with the Vh1 domain from a distance. The structural alterations provoked by the binding of the IpaA-VBSs to vinculin were reminiscent of those induced by the binding of the VBSs of talin and α-actinin, which produce dramatic changes in the structure of the Vh1 domain, yet bind to Vh1 in an inverted orientation and provoke unique conformational changes in full-length vinculin (; ; ). Superposition revealed that IpaA-VBSs behaved as striking and selective mimic of talin's VBSs in altering the structure of vinculin (). Therefore, applies molecular mimicry of talin to bind to and activate vinculin. The unusual behavior of 's IpaA-VBSs in displacement assays with talin– or α-actinin–VBS–bound vinculin, and their higher affinity for vinculin, suggested there might be some unique features of their hydrophobic interactions with the Vh1 domain of vinculin. Indeed, inspection of their structures established that Phe-126 of vinculin interacts with Tyr-613 or -567 of IpaA-VBS or -VBS2, respectively (Fig. S2 A, available at ), whereas the interaction of vinculin Phe-126 with the VBSs of talin are through interactions with small, aliphatic sidechains (). To test our hypothesis that the presence of Tyr-613 (IpaA-VBS) and -567 (IpaA-VBS2) strengthened their binding to Vh1, we synthesized mutant IpaA-VBS and -VBS2 peptides where these tyrosine residues were replaced by isoleucine residues as seen in talin-VBS3 (; Ile-1950; Fig. S2 A). of IpaA-VBS for Vh1 was reduced from 0.11 (WT) to 0.27 nM (Y613I), and where the of IpaA-VBS2 was reduced from 6.61 (WT) to 15 nM (Y567I). Furthermore, comparison of the Vh1–IpaA-VBS and Vh1–IpaA-VBS2 crystal structures revealed that the sidechain of Lys-569 was near the IpaA-VBS2–vinculin interface, where it could potentially sterically interfere with binding. Indeed, mutating lysine-569 to an alanine (as seen in IpaA-VBS Ala-615; , , and Fig. of IpaA-VBS2 K569A was 0.53 nM, which is an order of magnitude higher than wild-type IpaA-VBS2 (6.61 nM). Therefore, at least two residues of the IpaA-VBS, Tyr-613 and Ala-615, contribute to its unique, very high-affinity interactions with the hydrophobic interface with vinculin. The Gram-negative pathogens and gain passage into the host cell through the agency of a type III secretion apparatus that jettisons invasin proteins into the host cell, which then act in concert to reorganize the actin cytoskeleton to direct entry and intercellular spread of these pathogens. invasion requires IpaA and its cellular target vinculin (), which together reorganize the actin cytoskeleton to allow completion of the entry process (). This strategy differs from that used by , where its SipA actin-binding protein stabilizes actin filaments to promote formation of filopodial extensions and membrane ruffles that together engulf the pathogen (; ; ). Nonetheless, it is now clear that a shared strategy of all Gram-negative pathogens is to subvert the actin cytoskeleton to gain purchase within the host (; ). The data presented clearly demonstrate that, using evolutionary convergence, has engineered two VBSs that modulate the structure and function of vinculin by using a remarkable level of mimicry of the VBSs of talin. This is most strikingly shown by the crystal structures of the VBSs of IpaA in complex with vinculin, which also bind to vinculin in the same orientation as the VBSs of talin (). However, our structural and biochemical studies have indicated that there are unique features of the IpaA–vinculin interaction that might explain the high affinities of the VBSs of IpaA () and their rather unusual behavior in displacement assays (), and it is possible that these differences could be exploited for deriving new therapeutics to treat and/or protect humans from this devastating pathogen. Although the proteins of other pathogens have been shown to be mimic other components and regulators of the actin cytoskeleton (), thus far is unique in that one of its invasins functions as a talin mimic to invade the host cell and to recruit vinculin to sites of bacterial entry. Furthermore, the requirement for the IpaA-VBS–vinculin interaction in cell–cell dissemination suggests that IpaA alters vinculin functions in unique ways that allow for intercellular motility of , possibly through influencing vinculin–Arp2/3 complexes that are required for spread of the pathogen (; ). The most C-terminal VBS of IpaA, IpaA-VBS, functions by all measurable criteria as a bona fide trigger that activates vinculin. First, IpaA-VBS has a remarkably high affinity for vinculin (), at least 10-fold higher than any other VBS (e.g., the αVBS of the α-actinin–vinculin interaction has a of ∼1.8 nM; ), and the femtomolar–picomolar affinities of the C terminus of IpaA for vinculin indicate that once vinculin is bound by IpaA, it cannot be displaced by the binding of vinculin's other partners, a notion supported by VBS displacement assays (). Furthermore, IpaA-VBS can efficiently sever vinculin's head–tail interaction and trigger the latent F-actin–binding activity of vinculin (), both hallmarks of vinculin activation (; ). Finally, and most importantly, the IpaA-VBS plays critical roles in directing host cell entry and dissemination of (). At face value, the second VBS of IpaA, IpaA-VBS2, would also appear to have all the hallmarks of a trigger that should activate vinculin in cells, especially given its ability to sever the vinculin head–tail interaction and to promote F-actin binding (). However, in tests of its biological activity, IpaA-VBS2 appeared less critical than IpaA-VBS for cell entry and dissemination of . However, possible clues to the potential function and relevance of having these two VBSs have come from further revelations regarding the overall scope of the interactions between and vinculin (). Although IpaA and vinculin were known to interact locally at sites of pathogen–host contact, and to provoke changes in the actin cytoskeleton to form the pseudoadhesion complex or the actin cup that forms around the internalized bacterium (Trans Van Nhieu and ; ), this interaction clearly requires the functions of both of the VBSs of IpaA. Strikingly, however, IpaA-VBS appears to play a more crucial role than IpaA-VBS2 in recruiting vinculin, a finding that is consistent with the estimated 60-fold difference in affinity. Thus, the two VBSs of 's IpaA protein appear to facilitate both the recruitment of vinculin at sites of bacterial entry, and in forming the actin-vinculin cup that surrounds the membranes of the internalized pathogen. The precise role that this highly effective recruitment of vinculin by the VBSs of IpaA is unclear because recruitment mediated by IpaA-VBS alone appears sufficient for bacterial invasion and cell–cell spread in epithelial cells in vitro. It is possible, however, that more efficient vinculin recruitment mediated by both of the VBSs of IpaA becomes critical for invasion of enterocytes during in vivo infection. Human full-length vinculin (residues 1–1,066), vinculin lacking its tail domain (residues 1–840), and the Vh1 (residues 1–258) and Vt (residues 879–1,066) domains of vinculin were purified as previously described (; ; ). The IpaA-Cterm (residues 559–633) protein was expressed in as a GST fusion protein using a pGEX expression construct (Novagen). Cells were lysed in PBS and PMSF, and the protein was purified using a glutathione–Sepharose affinity column, GST cleavage, and anion exchange chromatography (Q–Sepharose). Human αVBS (residues 731–760), talin-VBS1 (residues 607–636), or talin-VBS3 (residues 1,949–1,970), the six wild-type IpaA-VBSs of varying lengths, and a nonbinding IpaA peptide, as well as the three mutant IpaA-VBSs (IpaA-VBS-Y613I [NIKAAKDVTTSLSKVLKNINKD], IpaA-VBS2-Y567I [AIEKAKEVSSALSKVLLSKIDDT], and IpaA-VBS2-K569A [AIYEAKEVSSALSKVLSKIDDT]) were synthesized and HPLC purified in our in-house facility. Binding studies were performed by SPR using a biosensor (Biacore 2000; Biacore) equipped with a carboxymethyldextran-coated gold surface (CM-5) sensor, as previously described (). 65 nM Vh1 protein was incubated with IpaA-VBS2, -NB (nonbinder), -VBS, or -Cterm () in PBS at room temperature using an ∼1:2 molar ratio of Vh1/IpaA. Complexes were run on 20% homogeneous native polyacrylamide PhastGels (GE Healthcare). Protein detection was performed using Coomassie blue staining. Only a single Vh1–IpaA-Cterm complex was detected, indicating that IpaA-Cterm protein, at this molar ratio, interacted with Vh1 with a 1:1 stoichiometry. All binding assays were performed in PBS buffer. 130 nM Vh1 protein was added to Vt at room temperature using a ∼1:2 molar ratio of Vh1/Vt (i.e., Vt at 260 nM), to ensure that all Vh1 was in complex with Vt. Analysis on native polyacrylamide gels confirmed formation of the Vh1–Vt complex. IpaA-VBS, -VBS2, or -Cterm were added to the Vh1–Vt complex. IpaA-VBS was titrated at ∼2-, 3.5-, 10-, and 20-fold molar ratios, and IpaA-Cterm was titrated at 0.9-, 1.7-, 3.4-, and 6.8-fold molar ratios. Complexes formed were resolved on native polyacrylamide PhastGels and stained with Coomassie blue. Vinculin (in PBS) was incubated at ambient temperature with IpaA-VBS, -VBS2, or -Cterm, and the ability to cosediment with F-actin was assessed as previously described (). Levels of IpaA proteins expressed by wild-type and mutant strains of were determined using an anti-IpaA–specific antibody (). Binding of IpaA mutant proteins to vinculin was also assessed in a vinculin overlay assay using I-labeled vinculin, as previously described (). Levels of IpaA protein translocated inside HeLa cells were assessed as previously described (). Immunoprecipitation was performed using the anti-IpaA2 serum (). Western blot was performed with anti-IpaA (working dilution 1:10,000) as a primary antibody and protein A-HRP (working dilution 1:10,000; BioRad Laboratories) as a secondary antibody. Details regarding crystallization of the Vh1–IpaA-VBS and –IpaA-VBS2 complexes can be found in Table S1 and S2, respectively. Details regarding the collection of x-ray data (; ) and the determination and refinement (; ; ) of the structures of the Vh1–IpaA-VBS and –IpaA-VBS2 complexes are provided in the captions of Table S1 and S2 and S3 and S4, respectively. HeLa cells were grown in DME supplemented with 10% FCS, Hepes, penicillin, streptomycin, and -glutamine in 9% CO. Caco-2/TC-7 cells were cultured in DME supplemented with 20% FCS and allowed to form polarized monolayers, as previously described (). Wild-type serotype 5 M90T and the mutant were cultured as previously described (). The mutant was generated by cutting pBad:ipaA with I and I, followed by religation to generate a deletion of the last 81 C-terminal residues. The mutation was created using the QuickChange II (Stratagene) method using the primer 5′-GACGATACCTCTGCAGAATAACTTACAGATGATATATCTG-3′, which generates a deletion of the last 44 C-terminal residues. The mutation was created using the primers 5′-GATACAATTGATAAAAATCATTCTGCAGAATTACTTA CAGATG-3′ and 5′-CATCTGTAAGTAATTCTGCAGAATGATTTTTATCAATTGTATC-3′ and PCR, which generates a deletion of residues 564–586 of IpaA. Plasmid constructs were subcloned into the vector pCR2.1 (Invitrogen) and transformed into the mutant strain to generate , , and strains (Fig. S1 A). Wild-type , , , , and strains were tested for internalization in HeLa cells using the gentamicin protection assay, as previously described (). Bacterial entry and dissemination in polarized Caco-2 cells was determined as follows. Bacteria grown in exponential phase were resuspended in DME containing 20 mM Hepes, pH 7.3, at a final optical density of 0.06 and used to challenge cell monolayers for 90 min at 37°C. Samples were rinsed and incubated for 6 h in DME containing 10% FCS and 50 mg/ml gentamicin to prevent bacterial extracellular growth and to allow for intercellular dissemination. Samples were fixed and stained for F-actin and LPS, as previously described (, ). Quantification of bacterial invasion and cell–cell spreading in Caco-2 cells monolayers was performed using an epifluorescence microscope (BX50; Olympus). For invasion, the number of infection foci per field was scored with a 20× objective (Olympus) on at least 100 fields in two independent experiments. To assess cell–cell spread of bacteria, the number of infected cells/foci was scored using a 40× immersion objective (Olympus) on at least 30 foci in three independent experiments. Statistical quantification of the surface area and the intensity of vinculin recruitment at entry sites was performed from images reconstructed from z stacks corresponding to planes spaced by 0.2 mm, acquired with an epifluorescence microscope (DRMIBe; Leica) equipped with a 63× immersion objective and a piezo (P-721.17, Physik Instrumente), connected to a charge-coupled device camera (CoolSNAP HQ; Princeton Instruments) under the Metamorph software (version 6.3 r1; Universal Imaging Corp.). The plotted data are representative of at least 30 foci in two independent experiments. The assay used to evaluate cell–cell spreading of wild-type and mutant strains of () was derived from the plaque assay that has been extensively used to assess dissemination (). Translocation and immunofluorescence assays to determine levels of intracellular IpaA proteins and vinculin localization were performed as previously described (). The atomic coordinates of the Vh1–IpaA-VBS and Vh1–IpaA-VBS2 crystal structures have been deposited in the Protein Data Bank under accession codes 2GWW and 2HSQ, respectively. Fig. S1 shows a schematic of the various IpaA and vinculin constructs used in this study. Fig. S2 shows a stereo view of Tyr-613 of IpaA-VBS and Tyr-567 of IpaA-VBS superimposed onto Ile-1950, as seen in the vinculin–talin-VBS3 crystal structure (), as well as details of the interaction of Lys-569 or Ala-615 of IpaA with Leu-54 and Val-57 of vinculin. Tables S1–S4 show the data reduction and crystallographic refinement statistics for the Vh1–IpaA-VBS and–IpaA–VBS2 structures. Online supplemental material is available at .

The contractile cortex is a 50-nm–2-μm-thick layer of cytoskeleton under the plasma membrane that is rich in actin filaments, myosin II, and actin-binding proteins (). Assembly dynamics and contractility of this layer are thought to generate cortical tension, drive cytokinesis, and play a central role in cell locomotion and tissue morphogenesis (; ). Despite its importance, the structural organization, dynamics, membrane connections, and assembly pathway of contractile cortex are not well understood. The classic unit of contractile actomyosin organization is the sarcomere, a structure that is well characterized in muscle cells and present in stress fibers of nonmuscle cells (; ). However, typical contractile cortex does not contain well-organized sarcomeres by light or electron microscopy, and how its actin and myosin filaments are structurally organized is unclear. Understanding the structure and dynamics of the cortex is important because it determines how cells respond to mechanical force or generate force for shape change and movement. Cells are composite materials, with each constituent conferring different mechanical properties. The membrane bilayer enables the maintenance of a specific microenvironment, but cannot expand or retain a stable shape when subjected to environmental forces (). The plasma membrane of red blood cells is stiffened by a submembranous cytoskeleton consisting of a meshwork of spectrin tetramers tethered both to plasma membrane proteins and to short actin filaments by linking proteins, notably ankyrin and protein 4.1 (). Motile cells contain all of these proteins, but the extent and function of a submembranous cytoskeleton is unclear. They also have a much thicker and stiffer cortex under the plasma membrane, consisting of a shell of cross-linked actin filaments oriented tangential to the cell surface, which enables cells to better resist mechanical deformation (). This shell can produce force either through myosin II–driven contraction or actin polymerization. Myosin-driven contraction generates cortical tension that can be converted into different types of motility by appropriate symmetry breaking (). Biochemically, the proteinaceous composition of the cortex is dominated by actin, actin-bundling proteins, and myosin II. How cortex is regulated and attached to the plasma membrane is unclear. The small GTPase RhoA is probably the most important regulator of contractile cortex assembly (). Its activation leads to both actin polymerization and myosin II recruitment through several pathways in cytokinesis and chemotaxis (; ; ), but its role in the regulation of generic contractile actin cortex is less well understood. RhoA directly activates formins, which are actin-nucleating proteins that hold onto growing barbed ends (), and activates myosin II by regulation of its phosphorylation state through Rho-kinase (). The nature of the attachment of cortical actin to the membrane is poorly understood, despite identification of several protein candidates. In red blood cells, protein 4.1 links short actin filaments to integral membrane proteins, but its role in motile cells is less clear (). ERM (ezrin-radixin-moesin) proteins are natural candidates because they can bind both actin and integral membrane proteins (). ERM proteins switch from an inactive closed conformation to an active open conformation that exposes an actin-binding site (at the tail) and a four one-ezrin-radixin-moesin (FERM) membrane–targeting domain (at the head; ). Perturbation experiments in cells are consistent with ERMs functioning in actin-membrane linkage (). One challenge in dissecting assembly of the contractile cortex in cells has been finding a starting point. The plasma membrane is almost always underlain by a cortex, and in situations like the assembly of the cytokinetic furrow, it is difficult to evaluate the relative contributions of de novo assembly at the furrow site versus sliding of cortical elements from neighboring membrane. One natural situation where cortex-free membrane is transiently generated is blebbing. This is a dramatic type of cell motility involving the contractile cortex that is often observed during apoptosis () and cytokinesis () and takes part in migration of some embryonic cells (). Bleb nucleation is initiated by the rapid detachment of a patch of membrane from the cortex. The patch is inflated over ∼30 s to form a spherical protrusion that is 1–10 μm in diameter and filled with cytosol (). When expansion stops, contractile cortex reassembles under the bleb membrane, and the bleb is retracted. Retraction finishes as the bleb cortex reintegrates the bulk cell cortex. Membrane detachment and bleb inflation are thought to be driven by intracellular pressure transients generated by myosin II contraction of the actin cortex. Indeed, drugs that relax the cortex by inhibiting actin or myosin II inhibit blebbing. We take advantage of blebbing as a window into the process of cortical assembly and dissect its assembly pathway. Bleb expansion occurs when the membrane detaches from the actin cortical cytoskeleton. Imaging of constitutively blebbing M2 cells expressing both GFP-actin and a membrane marker, the PH domain of phospholipase C δ (PH-PLCδ), tagged with monomeric red fluorescent protein (mRFP) confirmed the absence of large amounts of actin in growing blebs and the persistence of an actin cortex in the cell body beneath the growing bleb (). Abundant cortical actin remains at the site of membrane detachment in the cell body during bleb expansion (, arrows). Actin is progressively recruited to the membrane during retraction ( and ). Thus, bleb dynamics provide a system to distinguish cortical proteins that associate constitutively with the plasma membrane from those that associate only when an actin cortex is present or assembling. We investigated the dynamics of 24 different cortical proteins in this system (), selecting candidate proteins implicated in cortex–membrane attachment, actin organization, and contractility. Ankyrin B () and protein 4.1 (nonerythroid isoform; ), which are proteins of the erythrocyte cytoskeleton, formed a distinct rim of fluorescence at the bleb membrane, with no difference in intensity between expanding and contracting blebs. Myr 2 and 3, which are myosin I isoforms that target to the membrane (), exhibited similar dynamics ( and not depicted). Thus, these proteins are constituents of a protein network that underlies the membrane at all times during bleb dynamics. We were unable to express full-length GFP-spectrin or GFP-adducin, but immunostaining revealed a distinct rim of spectrin () and adducin (Fig. S1 B, available at ) at the bleb cortex in fixed cells. Because contracting blebs are better preserved than expanding blebs during fixation, these images most likely reveal association during or after actin assembly. We have no data on spectrin or adducin in expanding blebs, but suspect they must be present because proteins that associate with them (ankyrin B and protein 4.1) are present. In summary, a submembranous cytoskeleton is present at all times during the blebbing cycle, and the cytoskeleton of cortex-free expanding blebs may resemble the rudimentary cytoskeleton that underlies the erythrocytic membrane. The transition from bleb expansion to stasis and, finally, contraction occurs over ∼30 s. This transition involves the recruitment of a new cortex to the membrane and provides a time window for measuring the relative timing of recruitment of cortical components, and testing dependency relationships. We imaged the recruitment dynamics of 18 different actin associated proteins. Eight did not localize to the bleb rim, and 10 transiently associated with the bleb rim ( and – ). We investigated the relative timing of the appearance of actin and six representative proteins that belonged to the three main classes of actin-binding proteins (linker proteins, bundling proteins, and proteins of the contractile apparatus) by measuring when the protein was first detectable above background at the bleb rim, and then normalized this time to the time when actin recruitment was first detected at the rim (). The first protein recruited was ezrin (), which links the actin cytoskeleton to the membrane. Ezrin was not enriched at the membrane during expansion, but rapidly localized to the bleb membrane as expansion slowed, forming a continuous rim colocalized with the membrane (). In two color videos, the actin rim was displaced toward the center of the bleb compared with the ezrin rim (, inset [t = 42 s], and Fig. S1 D), which is consistent with ezrin's role as a membrane linker (). Recruitment of ezrin preceded recruitment of actin by 4.87 ± 3 s (P < 0.01; = 23; and ). The delay between actin and ezrin recruitment could be visualized on two color kymographs (), where the newest cortex appears greener (ezrin rich/actin poor) compared with the older cortex (yellow). Moesin, another ERM family protein, had a localization similar to ezrin (). Three actin-binding proteins were recruited shortly, but significantly, after actin itself. These were α-actinin (2.4 ± 4 s relative to actin, = 20, P < 0.01; , , and Fig. S1 A), coronin (1.5 ± 1.5 s relative to actin, = 23, P < 0.01; ), and tropomyosin-4 (1.3 ± 1.5 s relative to actin, = 19, P < 0.01; and Fig. S1 H). All formed a uniform shell, exactly colocalized with actin, and presumably coassembled with new actin polymer. The timing of arrival of these proteins was not significantly different from one another (pairwise comparisons, P > 0.29). Fimbrin also assembled in a continuous rim colocalized with actin at the bleb periphery, but significantly later than the previous three (8.17 ± 4.9 s relative to actin, = 17; and Fig. S1 E). Myosin II recruitment to the bleb cortex is thought to drive bleb retraction. Myosin regulatory light chain (MRLC) was recruited significantly later than the early actin-binding proteins (7.3 ± 2.4 s relative to actin, = 27, P < 0.01 for each pair-wise comparison with the early actin-binding proteins; , , and ). Unlike other actin-binding proteins, MRLC and myosin heavy chain (MHC) assembled into small puncta that may represent contractile foci (, , Fig. S1 F, and not depicted). Anillin ( and Fig. S1 C) and tropomodulin (not depicted), proteins that regulate or interact with myosin II and actin (; ), were recruited to retracting blebs as a continuous rim, suggesting that they bind primarily to actin rather than myosin II. Three different electron microscopy techniques were used to probe bleb ultrastructure. Scanning electron microscopy (SEM) examination of blebbing cells fixed without permeabilization () confirmed that fully extended blebs were approximately spherical and that retracting blebs displayed a crumpled morphology (, inset). Live observation of cells during triton permeabilization, with or without simultaneous aldehyde fixation, showed that only retracting blebs survive permeabilization (unpublished data). Thin-section transmission electron microscopy (TEM) of saponin-permeabilized aldehyde-fixed cells revealed that the cortex of retracting blebs comprises a 10–20-nm-thick shell of actin filaments (verified by immunoTEM; not depicted) and shows clear signs of crumpling (). The best views of actin organization were provided by SEM examination of cells permeabilized in the presence of phalloidin to stabilize filamentous actin. Actin in retracting blebs formed an interconnected network resembling a cage (). Incubation of extracted cells with the S1 fragment of myosin revealed that all of the fibers within the cagelike structures were actin filaments (). Distinct knots were apparent at the vertices, but actin filaments' polarities showed no specific order around these (, arrow). Typical distances between adjacent vertices were 200 nm, or ∼80 actin subunits, but this varied considerably between blebs, perhaps reflecting the age of the cortex, the degree of cross-linking, or how advanced contraction was. To test the generality of this organization, we examined blebs that occur naturally during cytokinesis in HeLa cells. Actin ultrastructure in HeLa blebs was identical, except that the actin mesh was tighter (). Interestingly, the cortex of HeLa cells arrested in metaphase had a similar morphology, but with a much tighter mesh (∼20 nm; ) and a thicker shell (50–100 nm; not depicted). Treatment of blebbing cells with cytochalasin D, a drug that blocks actin polymerization at the barbed end, had two effects. At short times (∼30 s), bleb initiation and expansion was stimulated; at longer times, blebbing was inhibited. This dual effect may arise because initially, the attachment between actin and the membrane is destabilized by cytochalasin D treatment, promoting bleb nucleation. At longer times, contraction of the cortex was compromised and new blebs ceased to appear. Blebs that had already formed an actin cortex before the onset of treatment were not affected and retracted normally (, white arrows). Conversely, blebs that emerged shortly after treatment did not assemble an actin rim and did not retract (, red arrows). This last effect allowed us to determine which proteins depend on actin for their recruitment to the bleb membrane. To test the actin dependence of cortical protein recruitment, we imaged M2 cells transfected with GFP-tagged ankyrin B, ezrin, α-actinin, or coronin during cytochalasin D treatment. Cytochalasin D treatment did not affect either the localization or level of ankyrin B at the bleb membrane (). This was expected, given its constitutive targeting to the membrane. α-Actinin was not recruited to the rim of blebs that emerged after the beginning of drug treatment and remained cytoplasmic (), which is consistent with coassembly of α-actinin and F-actin. Ezrin was still recruited to the membranes of blebs that formed after treatment (), and accumulated progressively at these membranes, giving a signal identical to that seen during the normal bleb cycle (cytochalasin, 76 ± 36%, = 20 vs. control, 66 ± 31%, = 23, P = 0.34). Thus, ezrin can target to membranes independent of actin and its localization, and kinetics are similar with and without actin. Coronin-3 was also recruited to blebs that formed after onset of treatment (unpublished data). The kinetics of ezrin recruitment, its response to cytochalasin D, and reports of ERM-mediated actin nucleation (), suggested an important role for this protein in cortex reassembly. To assess the role of ezrin in actin nucleation, we incubated cells expressing mRFP-actin and GFP-ezrin with drugs targeting reported regulators of ezrin (). The only drug to have an effect on ezrin recruitment to the cell membrane was the broad-specificity kinase inhibitor staurosporine. Blebs that formed after staurosporine addition did not recruit ezrin to the bleb membrane, did form an actin rim, and did not retract (). Hence, ezrin localizes to the bleb rim earlier than actin but does not nucleate actin. Lack of retraction is presumably caused by inhibition of kinases that phosphorylate myosin. In blebs formed before staurosporine addition, ezrin fluorescence at the rim gradually decreased until it was no longer detectable above background. Drugs that targeted Rho-kinase (Y27632), protein kinase A (H89, K252C, and HA1077), protein kinase C (Go6976, K252C, and H9), tyrosine kinases (Genistein), G proteins (pertussis toxin and suramin), or phosphotidylinositol bisphosphate (PIP2; PBP-10; ) had no effect on ezrin localization to the bleb rim (unpublished data). To further investigate the role of ezrin in bleb retraction, we perturbed ezrin function by overexpressing or microinjecting mutant forms of ezrin into cells expressing actin-GFP. Microinjection of recombinant FERM domain of ezrin-mRFP caused acute defects in bleb retraction. Actin was still recruited to the rim of blebs, but as retraction initiated the actin rim tended to tear away from the membrane, resulting in an inward flow of actin but complete or partial failure of membrane retraction ( and ). Conversely, when recombinant ezrin-T567D-GFP was microinjected into blebbing cells, it localized to the submembranous cortex and inhibited blebbing, suggesting stabilization of actin–membrane links ( and ). Stable expression of mutant forms of ezrin had effects consistent with membrane–actin attachment, depending on ezrin function. M2 cells bleb profusely after plating, and the proportion of blebbing cells decreases over a period of days (). Cells expressing dominant active ezrin-T567D-GFP stopped blebbing earlier than wild-type cells; whereas cells expressing either the FERM domain of ezrin-GFP or dominant-negative ezrin-T567A-GFP blebbed in higher proportion than wild-type cells (Fig. S2, available at ; ). Studying the signaling events leading to bleb retraction is challenging because drugs that might inhibit retraction also inhibit general contractility of the actin cortex, making the effects on retraction alone difficult to interpret. To unambiguously examine regulation of retraction, we decoupled bleb retraction from expansion using a brief application (<2 min) of cytochalasin D, followed by washing out. This creates a population of stationary blebs devoid of actin that retract upon washout by recruiting actin and MRLC (). The steps leading to actin nucleation and retraction can be investigated by inclusion of inhibitors in the washout medium. Inclusion of the myosin II ATPase inhibitor blebbistatin in the washout medium had no effect on actin rim formation, but inhibited retraction (). Upon inactivation of blebbistatin by exposure to blue light (), retraction resumed rapidly. In cells transfected with MRLC– tandem dimer RFP (TDRFP), blebbistatin did not inhibit recruitment of MRLC to the bleb rim (). The inhibitory effect of Rho-kinase inhibitors on blebbing suggested a role for the small GTPase RhoA. Immunostaining for active RhoA after TCA fixation (; this preserves both expanding and retracting blebs; unpublished data) and expression of GFP-tagged RhoA revealed the presence of RhoA colocalized with the bleb membrane during all stages of blebbing (). We investigated the effect of direct RhoA inhibition using proteins that affect RhoA activity. Microinjection of Rho guanine nucleotide dissociation inhibitor α (GDIα), the catalytic domain of p50Rho GTPase activating protein (GAP), the RhoGTPase-binding domain of rhotekin, or C3 exoenzyme lead to a significant decrease in the proportion of blebbing cells compared with controls (). Unfortunately, in all cases, RhoA inhibition was too slow to assess its effect on actin nucleation or bleb retraction. Consistent with the microinjection results, overexpression of GFP-tagged RhoGDIα, rhotekin-binding domain, or p50RhoGAP lead to a significant decrease in the proportion of blebbing cells (). Localization of active RhoA at the cell membrane and the presence of ezrin within blebs directed our attention to Rho guanine nucleotide exchange factors (GEFs) reported to associate with ezrin, dbl and Net1 (; ). GFP-tagged Net1 localized to the nucleus of blebbing cells (not depicted), but GFP-tagged KIAA0861, a GEF closely related to dbl (), colocalized with the cell membrane throughout all phases of blebbing (). Blebbing offers a window into the sequence of events leading to the reassembly of a contractile cortical actin cytoskeleton. One limitation of our study is that the accumulation of small quantities of proteins at the bleb cortex may be obscured by background fluorescence emanating from cytosolic protein. Indeed, we estimate that proteins must be ∼10% more concentrated at the cortex than in the cytosol for us to reliably image localization (unpublished data). We show that bleb retraction is the result of the sequential assembly of actin–membrane linker proteins, actin, actin-bundling proteins, regulatory proteins, and finally motor proteins. Two proteins of the erythroid submembranous cytoskeleton, protein 4.1 and ankyrin B, were present at the membrane during all phases of blebbing, and both spectrin and adducin were present in retracting blebs. This suggested that an erythroid-like submembranous cytoskeleton may protect the cell membrane from lysis during the repeated cycles of bleb expansion and retraction. Nonerythroid cells have a spectrin-based network that extends over the entire cell surface; however, its function is unclear because microinjection of antibodies to spectrin that precipitated the network had no deleterious consequences in cells (). This may be because, under normal circumstances, the cell membrane is tethered to the actin cortex by ERM proteins independently from its attachment to the spectrin meshwork. Ezrin played an important role in the stabilization of actin–membrane attachment in retracting blebs. Microinjection of the FERM domain of ezrin, which inhibits attachment of cellular ezrin to actin, weakened the actin–membrane attachment causing transient detachment of the actin cortex from the membrane during bleb retraction. Conversely, microinjection of a dominant active ezrin caused cessation of blebbing. These effects suggest that ezrin contributes strongly to actin–membrane adhesion energy. Decreasing the adhesion energy by impeding actin–membrane attachment increases bleb nucleation, whereas increasing the adhesion energy decreases bleb nucleation. Hence, two independent systems may coexist to protect the membrane from lysis. One, an erythroid-like submembranous cytoskeleton, is present during all phases of blebbing; the other tethers the membrane to the actin cytoskeleton via ERM proteins, is reassembled shortly after expansion ceases, and is regulated by an unknown staurosporine-sensitive kinase. The actin cortex in newly formed blebs assembles into a cagelike structure cross-linked by actin-bundling proteins. The mechanism of actin nucleation beneath the membrane of newly formed bleb remains unclear, as neither Arp2/3 nor mDia1 localize to the bleb membrane. The role of RhoA in blebbing regulation suggests a role for a different formin in actin nucleation. The new actin cortex confers resistance to further expansion, as well as an essential framework for the transduction of forces generated by myosin-based movement. The cortical shell consisted of long criss-crossing strands of actin that intersected at ∼200-nm intervals, was 3–4 filaments thick, and was devoid of internal structures. Blebs from mitotic HeLa cells and the cortex of cells rounded in metaphase also had similar, but tighter, ultrastructures. Actin-bundling proteins rigidify actin gels by reducing their degrees of freedom. In blebs, actin-bundling proteins appeared later than actin, colocalized with the gel and presumably coassembled with it. Some actin-bundling proteins localized to the bleb rim shortly after actin (α-actinin and coronin), and significantly earlier than others (fimbrin). This may be caused by a difference in cross-linking distance (α-actinin, 40 nm vs. fimbrin, 14 nm) that enables the longer ones to participate in early network assembly through a more efficient capture of surrounding actin filaments. As expected, the recruitment of α-actinin was dependent on the presence of an actin gel, but, surprisingly, the recruitment of coronin 3 was not, consistent with the hypothesis that it may play a role in reinforcing links between the actin cortex and the membrane (). The final step was the assembly of the contractility control apparatus and the recruitment of motor proteins. Tropomyosin and tropomodulin, two proteins involved in the control of myosin contractility, are recruited to the bleb rim. Finally, myosin II localizes to discrete foci, which may correspond to minifilaments or ribbons of myosin (similar to the organization observed in lamellipodium; ) and these provide the force needed for retraction. A central question in bleb dynamics is how cortex reassembly is triggered. We envisage two possibilities. First, cortex assembly is constitutive, but slow relative to bleb expansion. In this model, the expanding bleb is cortex free because of its rapid growth in area. When expansion slows, constitutive cortex assembly simply catches up. Cortex assembly may require a signaling pathway, but no special detection mechanisms to differentiate bleb membrane from generic membrane. The presence of both RhoA and a RhoGEF at the cell membrane at all stages of blebbing supports this model. In addition, ezrin may be recruited to the membrane via direct attachment to the RhoGEF (). In the second model, cortex assembly is triggered by an active signaling process downstream of some sensor that detects a change in the bleb membrane, such as membrane tension or exposed lipid head groups. In this model, cortex reassembly is locally triggered by a special property of the bleb membrane. For example, as the membrane tears from the actin cortex, PIP2 gets freed from its interaction partners, and this may provide an upstream signal for cortex reassembly. Another possibility is that the tension change concomitant with bleb expansion () could be detected by tension-sensitive mechanisms. Our data do not support either of these possibilities because treatment with chelators of PIP2 (neomycin sulfate and PBP10) or blockers of mechanosensitive channels (gadolinium chloride and GsMTx-4; unpublished data) had no effect on bleb retraction. Though, at present, our data seem to favor the constitutive assembly model, we cannot provide definitive evidence and more in-depth studies will be necessary. Finally, Do our observations provide any clues as to the function of blebbing? Most animal cells bleb during cytokinesis and apoptosis, and some bleb during cell migration. Blebbing might simply be an epiphenomenon caused by strong activation of cortical contractility, or it might have a real function in these processes. Our observations reveal that blebbing efficiently triggers assembly of new cortex that can integrate into the bulk cortex as the bleb contracts. It might, thus, function as a pathway for new cortex generation when cells need to rapidly expand their cortical surface area, such as during cytokinesis. Consistent with this hypothesis, inhibition of actin polymerization at the poles of dividing cells inhibits cytokinesis more efficiently than inhibition at the furrow (). As blebbing normally occurs at the poles during cytokinesis, it may be an efficient way of generating new cortex to increase cortical surface area while maintaining the furrow stable. Filamin-deficient M2 cells () were cultured in MEM with Earle's salts (Invitrogen) with penicillin/streptomycin, 10 mM Hepes, and 10% 80:20 mix of donor calf serum/fetal calf serum. All imaging was done in Leibovitz L-15 media (Invitrogen) supplemented with 10% 80:20 mix of donor calf serum/fetal calf serum. HeLa cells were cultured in DME (Invitrogen) with penicillin/streptomycin and 10% fetal calf serum. Detailed information about all of the plasmids used in this study is summarized in Table S1 (available at ). Unless otherwise noted, the full length of each gene was cloned. MRLC and MHC tagged with GFP were gifts from A. Straight (Stanford University, Stanford, CA). Anillin-GFP was a gift from Field. mRFP and TDRFP were gifts from R. Tsien (University of California, San Diego, La Jolla, CA). The membrane was visualized by transfecting the cells with the PH domain of PLCδ tagged with GFP (a gift from T. Balla, National Institutes of Health, Bethesda, MD) or mRFP. Ankyrin B-GFP was a gift from V. Bennett (Duke University, Durham, NC). Moesin-GFP was a gift from H. Furthmayr (Stanford University). Myr2 and myr3-GFP were gifts from T. Lechler (Duke University). tropomyosin 4-GFP, tropomodulin 3-GFP, and mDia1-GFP were gifts from N. Watanabe (University of Kyoto, Kyoto, Japan). Arp3-GFP and capping protein-GFP were gifts from D. Schafer (University of Virginia, Charlottesville, VA). Fascin-GFP was a gift from P. McCrea (University of Texas, Houston, TX). Sept6-GFP was a gift from M. Kinoshita (University of Kyoto). Vimentin-GFP was a kind gift from R. Goldman (Northwestern University, Chicago, IL). 6xHis-FERM domain of ezrin-mRFP was a gift from V. Gerke (University of Muenster, Muenster, Germany). Rhotekin-binding domain-GFP was a gift from W. Bement (University of Wisconsin, Madison, WI). 6xHis-Rhotekin binding domain-mRFP was a gift from R. Grosse (University of Heidelberg, Heidelberg, Germany). GST-tagged RhoA, RhoGDIGα, and p50RhoGAP were gifts from A. Hall (Memorial Sloan-Kettering Cancer Center, New York, NY). Actin-GFP and tubulin-GFP were purchased from CLONTECH Laboratories, Inc. Actin localization was visualized by transfecting cells with an adenovirus containing GFP-tagged human β-actin (). Alternatively, we used a melanoma cell line stably expressing actin-mRFP derived from wild-type M2 cells infected with actin-mRFP retrovirus in the retroviral vector pLNCX2 (CLONTECH Laboratories, Inc.). Ezrin point mutations T567A (impaired actin binding and head-to-tail association) and T567D (constitutively active actin binding and impaired head-to-tail association; ) were performed using the one-step mutagenesis kit (Stratagene) on wild-type ezrin in pcDNA3.1-topo-GFP-CT. 6xHis-tagged ezrin T567D GFP was created by directly ligating the full-length PCR product of ezrin T567D GFP into pET100D (Invitrogen). Recombinant protein expression and purification were effected using standard methods for His-tagged or GST-tagged protein purification. The purified proteins were either eluted directly or dialyzed overnight in either microinjection buffer (50 mM K-glutamate and 0.5 mM MgCl, pH 7.0) or in 50 mM KCl and 20 mM Tris-HCl, pH 7.0. For Western blotting, cells were scraped off the tissue culture dish in PBS, pelleted, resuspended in an equal volume of Laemmli buffer with β-mercaptoethanol, and boiled for 15 min. The samples were then loaded onto SDS-PAGE gels. Standard Western blotting techniques were used. Antibodies used (all monoclonal) were as follows: filamin A (1:1,000; CHEMICON International, Inc.), ezrin (1:2,000; Sigma-Aldrich), GFP (1:2,000; CLONTECH Laboratories, Inc.). HeLa cells were used as a positive control for filamin expression. For all antibodies except RhoA, the cells were fixed for 1 min at room temperature in a solution containing fixation buffer (137 mM NaCl, 5 mM KCl, 1.1 mM NaHPO, 0.4 mM KHPO, 2 mM MgCl, 2 mM K-EGTA, 5 mM Pipes, pH 6.8, and 5.5 mM Glucose) with 0.1% glutaraldehyde, 1% formaldehyde, and 0.3% Triton X-100. The cells were then fixed for 10 min further at room temperature in fixation buffer with 0.5% glutaraldehyde. For active RhoA staining, the cells were fixed with 10% TCA for 15 min on ice and treated as above. Monoclonal anti-RhoA antibody was purchased from Santa Cruz Biotechnologies, Inc. and used at 1:100 dilution. Cells were then stained using standard immunostaining techniques (). Rhodamine-labeled phalloidin, monoclonal anti–α-actinin, polyclonal anti-vimentin, monoclonal anti-ezrin, monoclonal anti-MRLC, monoclonal anti-tropomyosin, and monoclonal anti-tubulin were purchased from Sigma-Aldrich and used at a 1:200, 1:200, 1:40, 1:100, 1:100, 1:100, and 1:500 dilutions, respectively. Polyclonal anti-spectrin, T-Plastin (fimbrin), and adducin antibodies were purchased from Santa Cruz Biotechnologies, Inc., and they were all used at 1:100 dilution. Polyclonal anti-myosin antibody was purchased from Biomedical Technologies and used at a 1:400 dilution. Monoclonal anti-anillin, polyclonal Arp3, and polyclonal VASP antibodies were gifts from C. Field (Harvard Medical School, Boston, MA), C. Egile (Harvard Medical School), and F. Southwick (University of Florida, Gainesville, FL), respectively, and they were used at 1:500, 1:500, and 1:100 dilutions. Monoclonal anti-fascin was from DakoCytomation and used at 1:100 dilution. All fluorescent imaging was done using a 1.3 NA 100× oil-immersion objective on an inverted microscope (Nikon TE-2000; Nikon) interfaced to a spinning disk confocal microscope (Perkin-Elmer) equipped with a heating stage heated to 37°C. Images were captured on a charge-coupled device camera (Orca ER; Hamamatsu) and acquired on a PC using Metamorph software (Molecular Devices). Images were acquired either 488-nm wavelength for GFP-tagged proteins and FITC-labeled secondary antibodies or with 568-nm wavelength for RFP-tagged proteins and TRITC-labeled secondary antibodies. For display, images were low pass filtered and scaled such that background fluorescence was minimal. To assess the time of recruitment to the bleb rim of different proteins in relationship to actin, we cotransfected cells with actin-GFP or -mRFP and one of MRLC-TDRFP, α-actinin-mRFP, fimbrin-mRFP, tropomyosin-mRFP, coronin-GFP, or ezrin-mRFP. The next day, cells were imaged for 120 s at 1-s intervals. The intensity of both reporter constructs in retracting blebs was evaluated along a line drawn through the diameter using Metamorph (Molecular Devices). When one of the proteins appeared at the bleb rim, a clear peak in intensity above background could be observed. Actin was taken as the time reference and we measured the time of apparition of a fluorescence intensity peak at the bleb rim for each protein in comparison to actin. We collected peak appearance times for a maximum of 5 blebs per cell, for a minimum of 5 different cells for each protein, and a minimum of 17 blebs. For each protein, the time it appeared was compared with actin using a test. Proteins were compared pairwise with a test. Tests were considered significant if P < 0.01. To assess the dependence of actin recruitment on the presence of ezrin, we treated cells stably expressing actin-mRFP and transfected with ezrin-GFP with inhibitors to known regulators of ezrin, following a protocol identical to the one used with cytochalasin D. For ease of visualization, actin is shown in green in the figures and ezrin in red. The inhibitors used were Y27632 (25 μM; Calbiochem), H89 (20 μM; Calbiochem), K252C (50 μM; Calbiochem), HA1077 (20 μM; Sigma-Aldrich), Gö6976 (15 μM; Calbiochem), H9 (50 μM; Tocris), Genistein (100 μM; Calbiochem), pertussis toxin (1 μg/ml), suramin (300 μM; Calbiochem), PBP-10 (10 μM; Calbiochem), and staurosporine (5 μM; Calbiochem). To investigate the proteins and signaling events important for actin nucleation and bleb retraction, we transiently treated M2 cells transfected with actin-GFP and MRLC-TDRFP with 2.5 μM cytochalasin D for 2 min. This created a population of blebs devoid of actin that could reform an actin rim and recruit myosin once cytochalasin was removed. To remove cytochalasin, we exchanged medium 4 times. In experiments to determine how blebs retracted, we included blebbistatin (100 μM; Tocris) in the washout medium and treated for 10–50 min, then inactivated blebbistatin by exposure to 488-nm light for 600 ms (). During the whole procedure, images were acquired at 488 nm (except during blebbistatin treatment) and 568 nm every 10 s. For SEM imaging, cells were processed as described in with minor modifications. 2 h before treatment, cells were plated onto 12-mm glass coverslips. 10 min before, blebbing was stimulated by addition of fresh medium (Leibowitz L15 medium with 10% 80:20 DCS/FCS). The coverslips were washed 3 times with L15 without serum and transferred to extraction buffer for 20 min (50 mM imidazole, 50 mM KCl, 0.5 mM MgCl, 0.1 mM EDTA, 1 mM EGTA, 1% CHAPS, 2% Triton X-100, 2% PEG 35000, 10 μM phalloidin, and 1 μl/ml RNase cocktail (RNase T1 and A; Ambion), pH 6.8). For myosin S1 decoration of the actin filaments, the cells were incubated with 1 mg/ml myosin S1 (Sigma-Aldrich) in cytoskeleton buffer (50 mM Imidazole, 50 mM KCl, 0.5 mM MgCl, 0.1 mM EDTA, and 1 mM EGTA, pH 6.8) for 30 min. The remainder of the protocol was identical to . The cells were then dehydrated by exposure to serial ethanol dilutions, dried in an autosamdri-815 (Tousimis) critical point dryer, coated with 5–6 nm platinum-palladium and imaged using the in-lens detector of a SEM (Leo 892; Carl Zeiss MicroImaging, Inc.). For surface examination, cells were fixed for 10 min in 3% glutaraldehyde in cacodylate buffer, followed by a post-fix in tannic acid and uranyl-acetate. The sample was then prepared as above. For TEM examination, the cells were first permeabilized with 0.025% saponin in the same fixation buffer used for immunostaining for 1 min to release cytosol. They were fixed in fixation buffer with 1.5% glutaraldehyde and 50 mM lysine in 50 mM cacodylate, pH 7.0, for 6 min and post-fixed in 3% glutaraldehyde in cacodylate buffer for 6 min. For labeling experiments, the cells were then incubated with phalloidin-XX biotin (Invitrogen) for 30 min, followed by incubation with streptavidin–gold beads (EY laboratories). The remainder of the procedure was performed as described in . Microinjections were performed using standard procedures. In brief, borosilicate glass needles were pulled using a Sutter P-89. Before microinjection, the protein solution was centrifuged at top speed in a tabletop centrifuge to remove aggregates. Cells were microinjected with an Eppendorf 5242 microinjector (Eppendorf AG) using 30 hPa backpressure. For proteins that were not fluorescently tagged, 100 μg/ml 3-kD dextran-FITC (Invitrogen) was added to the protein solution. Recombinant C3-exoenzyme was a gift from K. Burridge (University of North Carolina, Chapel Hill, NC). Cells were transiently transfected with GFP (control), p50RhoGAP-GFP, Rhotekin binding domain-GFP, or RhoGDIα-GFP. The proportion of blebbing cells within a transfected population was evaluated by manually counting the total number of transfected cells and the number of transfected blebbing cells. The effect of protein expression was evaluated by comparing the proportion of blebbing cells expressing the protein of interest to the proportion of blebbing cells expressing GFP only, using a χ-square test with Yates correction. Fig. S1 shows protein localization in fixed blebbing cells using immunofluorescence. Fig. S2 shows that ezrin stabilizes cell shape. Fig. S3 shows localization of GFP-tagged ezrin mutants in spreading cells at different times after plating. Detailed information about all of the plasmids used in this study is summarized in Table S1. Online supplemental material is available at .

Synapses are specialized intercellular junctions that are required for the transfer of information between neurons. At chemical synapses, the presynaptic neuron forms a specialized membrane domain, termed the active zone, which contains the molecular machinery required for calcium-dependent synaptic vesicle fusion and recycling. A postsynaptic density, consisting of concentrated neurotransmitter receptors, forms in direct apposition to the active zone. A fundamental feature of chemical synapses is that they can be modulated to alter the transfer of information between neurons (). There are several general mechanisms by which the strength of synaptic connections can be influenced, including the following: (a) a change in the probability of synaptic vesicle fusion in response to a presynaptic action potential, (b) a change in the density or sensitivity of postsynaptic receptors, and (c) a change in synapse size (defined as the area of opposed presynaptic active zone membrane and postsynaptic density; ). A strong correlation between synapse size and the probability of presynaptic release has lead to speculation that the regulation of synapse size could participate in the mechanisms of neural development and activity-dependent plasticity (). Although the mechanisms that modulate vesicle fusion and postsynaptic receptor trafficking have received considerable attention, little is known about the molecular mechanisms that control synapse size. In and , genetic studies have uncovered several signaling molecules that, when mutated, affect synapse size. These signaling molecules include Liprin-α/Syd-2, Dlar, and Dally-like (; ; ). It remains to be determined how these signaling molecules mediate their influence on synapse size. We investigate the role of the postsynaptic Spectrin skeleton in the regulation of synapse size. The Spectrin skeleton consists of heterotetramers of two α- and two β-Spectrin subunits that, together with short actin filaments, form a cytoplasmic Spectrin–actin network that parallels the plasma membrane. The Spectrin skeleton can be recruited to the plasma membrane in several ways, including interactions with the adaptor protein Ankyrin, direct interactions with integral membrane proteins, or through interactions with plasma membrane phospholipids (). In the vertebrate nervous system, Spectrin and Ankyrin are required for the organization of ion channels and cell adhesion molecules into discrete domains. For example, at the nodes of Ranvier and at axon initial segments, Ankyrin G assembles a protein complex of voltage-gated sodium channels, L1-family cell adhesion molecules, and βIV-Spectrin. The organization of these membrane domains is severely impaired in and mutant mice, demonstrating that Ankyrin G and βIV-Spectrin mutually stabilize these integral membrane clusters (; ; ; ). Spectrin is found at synaptic connections throughout the central and peripheral nervous system () and has been hypothesized to participate in the clustering of postsynaptic neurotransmitter receptors (; ). However, direct genetic evidence documenting such a role for Spectrin is lacking. In , Spectrin is required for neuronal outgrowth and sarcomere stabilization in muscle. Interestingly, an ultrastructural analysis did not reveal obvious defects at the neuromuscular junction (NMJ; ; ). In , α- and β-Spectrin are present at the NMJ and null mutations in or die at the late embryonic/early larval stages (; ; ; ). In - null mutant embryos, synaptogenesis proceeds, but synaptic efficacy is impaired and the localization of both presynaptic vesicle proteins and postsynaptic Discs-large (PSD-95 homologue) is altered. Despite these defects in protein localization, iontophoretic application of glutamate evoked normal postsynaptic currents, suggesting that glutamate receptors are normally recruited to the nascent embryonic NMJ in the absence of α- or β-Spectrin (). Embryonic lethality precluded further analysis of synapse maturation and stability. To analyze the function of α- and β-Spectrin during postembryonic synapse development, we have previously used a transgenic RNA interference (RNAi) approach that allows us to circumvent the embryonic lethality associated with and mutations. We demonstrated that we can efficiently knock down α- or β-Spectrin at either the pre- or the postsynaptic side of the larval NMJ, and that the presynaptic Spectrin skeleton is essential for the stability of the NMJ (). We document a separable and unique requirement of the postsynaptic Spectrin skeleton for the specification of active zone size, spacing, and function during postembryonic development. We then extend our observations to include an analysis of postsynaptic Ankyrin, demonstrating that Ankyrin participates in the Spectrin-dependent regulation of synapse development. Together, our data suggest the existence of a postsynaptic, submembranous Spectrin–actin network that imposes a transsynaptic organization upon synapse development at the NMJ. We first confirm our ability to knock down Spectrin protein by expressing double-stranded RNA (dsRNA) in muscle, beginning in the first larval instar stage (, E and J; see Materials and methods). Interestingly, the expression of dsRNA eliminates both β- and α-Spectrin protein from the postsynaptic muscle membrane (), whereas the expression of dsRNA eliminates only α-Spectrin protein (). We tested the specificity of our dsRNA constructs and show that ubiquitous dsRNA expression knocks down α-Spectrin protein without substantially affecting β-Spectrin protein levels (). Similarly, dsRNA expression strongly reduces β-Spectrin protein levels, with only minor effects on α-Spectrin protein levels (). These experiments demonstrate that transgenically expressed dsRNA can knock down α- or β-Spectrin protein in the muscle below levels detectable by light microscopy. We conclude that β-Spectrin is required for the localization and/or stabilization of α-Spectrin to the postsynaptic muscle membrane, as has been observed in other tissues (). We next tested the time required for Spectrin protein knockdown after dsRNA expression in the muscle. We find that β-Spectrin protein is eliminated from larval muscle by the second instar stage, ∼24 h after the onset of dsRNA expression (). An identical result is observed for α-Spectrin (unpublished data). Because we conduct our assays in wandering third instar larvae, we estimate that the NMJ has developed in the near absence of postsynaptic α- or β-Spectrin for at least 3 d. d e m o n s t r a t e t h a t t h e p o s t s y n a p t i c S p e c t r i n s k e l e t o n i s n e c e s s a r y f o r t h e n o r m a l s i z e a n d s p a c i n g o f s y n a p s e s a t t h e N M J , w i t h b o t h p r e - a n d p o s t s y n a p t i c c o m p a r t m e n t s o f t h e s y n a p s e b e i n g a l t e r e d i n p a r a l l e l . A l t h o u g h s y n a p s e s i z e a n d p o s i t i o n i n g a r e c h a n g e d , w e s t i l l o b s e r v e a p e r f e c t a l i g n m e n t o f t h e p r e s y n a p t i c a c t i v e z o n e a n d t h e p o s t s y n a p t i c d e n s i t y a n d s e g r e g a t i o n b e t w e e n a c t i v e z o n e a n d p e r i a c t i v e z o n e c o m p o n e n t s o n b o t h s i d e s o f t h e s y n a p s e . M e c h a n i s t i c a l l y , β - S p e c t r i n i s r e q u i r e d f o r t h e l o c a l i z a t i o n o f α - S p e c t r i n a n d A n k y r i n t o t h e p o s t s y n a p t i c m e m b r a n e . T h e s e d a t a s u g g e s t t h a t t h e p o s t s y n a p t i c S p e c t r i n s k e l e t o n i m p o s e s a t r a n s s y n a p t i c o r g a n i z a t i o n t h a t i s n e c e s s a r y f o r t h e n o r m a l d e v e l o p m e n t o f b o t h t h e p r e s y n a p t i c a c t i v e z o n e a n d t h e o p p o s i n g p o s t s y n a p t i c d e n s i t y . Flies were maintained at 25°C on normal food. The following strains were used in this study: w (wild type), -GAL4 (muscle expression from mid–first-instar on; ), -GAL4 (ubiquitous expression; ), -GFP (), UAS- dsRNA, and UAS-dsRNA (). We recapitulated the expression pattern of BG57-Gal4 using a UAS-mCD8-GFP transgene (). We observed strong expression in postembryonic muscle and some expression in peripheral neurons that send projection axons into the CNS, as previously reported (). We did not observe any expression in motoneuron axons. To ensure that there is no expression in the motoneurons projecting to muscles 6/7 that might not be detectable at the light level, we expressed UAS–tetanus toxin () using BG57-Gal4. The expression of UAS–tetanus toxin in neurons abolishes evoked neurotransmission (; ). We did not observe any significant decrease in the amplitude of EPSPs at muscles 6/7 in these animals, demonstrating the tissue- specificity of the BG57-Gal4 driver line. We used the pWIZ-Vector (R. Carthew, Northwestern University, Evanston, IL; ) to generate the UAS--dsRNA construct. The target sequence was selected to avoid significant homology with any other gene to ensure specificity of the resulting dsRNA. We introduced XbaI restriction sites (underlined) to allow direct cloning into the pWIZ vector (). The following primers were used to amplify a 593-bp fragment of the open reading frame starting at position 3,835 of the cDNA: 5′GGGCGGGTAAAACGAATTTCCCAAACGGAAGC-3′ and 5′-GGGCGGGTAAAGGCCAGTCACTTCCTAAGTGGC-3′. The DNA fragment was amplified from wild-type genomic DNA and cloned into pWIZ following the methods used by . The construct was confirmed by sequencing. Transgenic flies were generated by standard methods. At least two independent transgene insertions were established for chromosomes 2 and 3. Wandering third instar larvae were dissected in HL3 saline and fixed either with 4% paraformaldehyde/PBS for 15 min or in Bouin's fixative (Sigma-Aldrich) for 2 min. Primary antibodies were applied at 4°C overnight. Primary antibodies were used at the following dilutions: anti-Bruchpilot (nc82) 1:100 (gift from E. Buchner, Theodor-Boveri-Institut für Biowissenschaften, Würzburg, Germany); anti–α-Spectrin (3A9) 1:50; anti–D-GluRIIA (8B4D2) 1:10; anti–Fasciclin II (1D4) 1:10 (all provided by the Developmental Studies Hybridoma Bank, Iowa); rabbit anti–D-GluRIIB (1:2,500), rabbit anti–D-GluRIIC (1:5,000; both antibodies were gifts from A. DiAntonio, Washington University, St. Louis, MO); rabbit anti-Dlg 1:5,000 (gift from V. Budnik, University of Massachusetts, Worcester, MA); rabbit anti–α-Spectrin 1:500, rabbit anti–β-Spectrin 1:500, rabbit anti-Ankyrin 1:500 (all gifts from R. Dubreuil, University of Illinois, Chicago, IL); rabbit anti-Pak 1:500 (gift from L. Zipursky, University of California, Los Angeles, CA); rabbit anti-synaptotagmin 1:500; rabbit anti-Dap160 1:200; rat anti–Nervous wreck 1:1,000 (gift from B. Ganetzky, University of Wisconsin, Madison, WI). All secondary antibodies and Cy3- and Cy5-conjugated anti-HRP were obtained from Jackson ImmunoResearch Laboratories and Invitrogen and used at a 1:200–1:1,000 dilution and applied for 1–2 h at RT. Larval preparations were mounted in Vectashield (Vector Laboratories). Images were captured at RT using an inverted microscope (Axiovert 200; Carl Zeiss MicroImaging, Inc.), a 100×/1.4 NA Plan Apochromat objective (Carl Zeiss MicroImaging, Inc.), and a cooled charge-coupled device camera (CoolSNAP HQ; Roper Scientific). Intelligent Imaging Innovations (3i) software was used to capture, process, and analyze images. Glutamate receptor cluster size and number were analyzed in the original 3D images to ensure precise measurements and distinction between neighboring clusters. The percentage of colocalization between different synaptic proteins was analyzed in single median focal planes using automated routines in the 3i software. For each genotype, 10 larval body-wall muscle preparations were homogenized in 2% SDS buffer (50 mM Tris-HCl, pH 6.8, 25 mM KCl, 2 mM EDTA, 0.3 M sucrose, and 2% SDS) on ice. After centrifugation at 5,000 , samples were boiled for 5 min in sample buffer (includes DTT and β-mercaptoethanol) and proteins were separated in a 7.5% SDS-PAGE gel and immunoblotted with primary antibodies overnight at 4°C. Protein bands were visualized with HRP-conjugated secondary antibodies and enhanced chemiluminescence reagents (GE Healthcare). Third instar larvae were selected and dissected according to previously published techniques (). Whole-muscle recordings were performed on muscle 6 in abdominal segment A3 using sharp microelectrodes (12–16 MΩ). Recordings were selected for analysis only if resting membrane potentials were more hyperpolarized than –60 mV and if input resistances were greater than 5 MΩ. The mean spontaneous mepsp amplitude was quantified by measuring the amplitude of ∼100–200 individual spontaneous release events per synapse. The mean per-synapse mepsp amplitudes were then averaged for each genotype. Measurement of mepsp amplitudes was semi-automated (Synaptosoft). The mean super-threshold–evoked EPSP amplitude was calculated for each synapse, ensuring that both motor axons innervating muscle 6 in segment A3 were recruited. Quantal content was calculated as the mean EPSP amplitude divided by the mean mepsp amplitude. Quantal content was determined for each synapse and averaged across synapses to generate the mean quantal content for each genotype. Data for steady-state synaptic transmission were acquired in HL3 saline (0.3 mM Ca and 10 mM Mg). Data were collected using an amplifier (Axoclamp 2B; Axon Instruments), analogue-to-digital board (Digidata 1200B; Axon Instruments), a Master-8 stimulator (AMPI,) and PClamp software (Axon Instruments). Data were analyzed offline using Mini-Analysis software (Synaptosoft, Inc.). Third instar larvae were prepared for electron microscopy as previously described (). For the analysis of active zone length, SSR phenotypes, and vesicle diameters, the largest diameter section of 1b boutons corresponding to the bouton midline were selected. To determine SSR thickness, the distance between the presynaptic membrane and the distalmost SSR membrane was measured. Four different measurements at 90° angles from each other, starting orthogonal to the muscle surface, were performed for each bouton. The two orthogonal and the two parallel measurements were averaged and analyzed as separate datasets. To analyze the density of the SSR, we counted the number of membrane segments crossed by the line used to measure the SSR thickness. Four measurements at 90° angles were performed and averaged as above. This number was divided by the SSR thickness to calculate the number of layers per μm. Average synaptic vesicle diameters were measured for vesicles within 250 nm of the active zone. The average vesicle diameter was determined for each active zone, and then measurements per active zone were averaged for each genotype. Fig. S1 shows low magnification images of the ultrastructural analysis of synaptic boutons in wild type and in animals lacking postsynaptic β-Spectrin. Fig. S2 shows that postsynaptic Ankyrin is not required for the localization of α- or β-Spectrin to the postsynaptic membrane (SSR). Fig. S3 shows that postsynaptic Ankyrin is required for the control of glutamate receptor cluster size. Online supplemental material is available at .

In epithelial cells, autocrine growth factor signaling requires coordinate input from β1 integrins for cell cycle progression (; ; , ; ). Conditional deletion of β1 integrins in skin and mammary epithelia confirmed the dependence of proliferation in these tissues on β1 integrin–mediated cell adhesion in vivo (; ; ). However, the role of β1 integrins in the regulation of epithelial cell proliferation in vivo has only been tested in a limited number of tissues. Within the digestive tract, the intestinal epithelial cells (IECs) express several α/β1 integrin heterodimers that mediate binding to ECM proteins (; ; ) and mediate adhesion to the underlying basement membrane. During late embryogenesis and the early postnatal period, the intestine becomes highly compartmentalized, which leads to the formation of the crypts of Lieberkühn and villi. The villi greatly increase the surface area for nutrient absorption and are generated by the stroma (mesodermal origin), which pushes the overlying IECs (endoderm origin) into the gut lumen. Stem cells take up residence near the bottoms of the crypts and give rise to four lineages of IECs (absorptive/enterocytic, enteroendocrine, goblet, and Paneth) that, with the exception of Paneth cells, migrate up the villi, differentiate, and are sloughed off within 3–5 d (). Over 90% of the IECs are absorptive enterocytes. The huge turnover of IECs requires vigorous intestinal crypt proliferation, which is mediated by key growth factor signaling pathways. Wnt signaling is a major driving force behind IEC proliferation (; ), whereas Hh and Bmp signaling normally inhibit IEC proliferation (, ; ; ; ). The role of ECM–integrin interactions in the regulation of intestinal epithelial proliferation through these or other growth factor signaling pathways in vivo is not well understood. Some of the epithelial α/β1 integrin heterodimers and the ECM proteins within the basement membrane that they bind are expressed in gradients along the crypt–villous axis, as are various growth factors and their cognate receptors, and may provide important positional cues for the IECs (; ; ). Although the intestinal epithelial stem cells comprise a subset of the β1 integrin–expressing IECs (; ), the role β1 integrins in intestinal epithelial proliferation has not been tested in vivo. We have used a conditional gene deletion system to determine the role of β1 integrins in intestinal epithelial proliferation. Mice carrying () and () transgenes () were crossed to generate / mice. Intestinal epithelial–specific recombination of the loxP sites was confirmed by PCR (), and loss of β1 integrin protein expression was verified by immunodetection ( and Fig. S1, A and B, available at ). Consistent with previous results, Cre-mediated recombination was detected by PCR only in genomic DNA isolated from the small intestine and the proximal large intestine, but not from the stomach or kidneys (; ). / mouse pups were born at the expected frequencies. At embryonic day 18 and birth, the appearance and size of the pups and intestinal organs were similar in all littermates (unpublished data). However, by postnatal day (P) 4, the / mice were less than half the body weight of their control littermates (/ or ; ) and died between P7 and P14 from severe malnutrition. The early deaths were not due to a lack of feeding, as the stomachs of all the newborn mice were full of colostrum at the time of death (unpublished data). The intestinal epithelium not only carries out nutrient absorption but also presents a barrier to the environment. Because of previous reports that β1 integrins mediate IEC survival ex vivo (), the / mice were carefully examined for evidence of mucosal defects resulting from a loss of IECs. Cleaved caspase 3 immunohistochemistry revealed the presence of rare apoptotic IECs in the / mice and control littermates, but the prevalence of apoptosis was similar (<2% of total IECs counted) in both types of mice (). There was no evidence of mucosal defects or inflammation (specifically, crypt abscesses, intraepithelial leukocytes, and loss of crypts) that would have arisen from barrier defects in the intestines of the / mice (). Conditional deletion of in skin disrupted basement membrane formation (; ), but ultrastructural examination revealed no differences in basement membrane structure between the / mice and their control littermates (). These results suggest that other adhesion molecules mediate epithelial adhesion, survival, and basement membrane formation in the intestine. The distal small and proximal large intestines of the / mice were noticeably larger in external diameter compared with their control littermates (). This was found to be due in part to a dramatic expansion of the intestinal stroma (Fig. S1, C and D), muscularis (Fig. S1, E and F), and ECM (Fig. S1, G–N). The epithelium in the / mice was markedly expanded compared with the control littermates as well (). The intestinal crypts and villi of P14 / mice () were much larger than those of their control littermates (). In addition, the majority of the crypts in the P14 / mice were dysplastic, as indicated by pseudostratified, enlarged and crowded nuclei, and abnormal crypt architecture in both the small () and large intestines (). The crypt expansion, enlargement, and dysplasia were much more pronounced in the cecum, a specialized portion of the large intestine (). Villous enlargement with expansion of the stroma was apparent, and there were multiple polypoid structures in the small intestinal mucosa of the / mice () but not in their control littermates (). The mucosa overlying the polyps was not dysplastic, indicating normal maturation of the villous epithelium. The polyps had the appearance of juvenile type polyps because of the stromal expansion and cystic dilation of the crypts (; ). Formed stool was found in the large intestines of the / mice but was absent in the large intestines of their control littermates, suggesting diarrhea (). The intestinal contents of the / mice stained positively for large fat droplets (unpublished data), which was not observed in the control littermates and indicated the presence of steatorrhea and fat malabsorption in the former mice. Fat is absorbed by enterocytes along the length of the small intestine, and examination of intestinal epithelium from / mice revealed large lipid inclusions within the villous enterocytes that were not present in their control littermates (). Total serum lipid levels were significantly reduced in the / mice compared with their control littermates (unpublished data), confirming the presence of fat malabsorption. Because the / mice appeared to die from severe malnutrition between P7 and P14, the absorptive lineage of IECs (enterocytic) was examined. Expression of the enterocytic marker sodium hydrogen exchanger 3 (NHE3) was detected in the vast majority of the IECs of the villi of the / mice and their control littermates (), demonstrating the abundance of enterocytes in both mice. However, ultrastructural examination of the small intestinal epithelium of / mice by electron microscopy revealed a severely defective microvillus brush border on the apical surfaces of the villous enterocytes (). Microvilli greatly increase the surface area of the intestine for nutrient absorption, are essential for proper nutrition, and express nutrient transporters and digestive enzymes (). The intestinal microvilli were diminished in size and poorly formed in the / mice compared with their control littermates (), indicating defective enterocyte differentiation (). Other IEC lineages were examined as well. The Paneth cell marker, Defensin/Cryptdin5, which was properly restricted to Paneth cells in control mice, was markedly increased in expression by the IECs along the entire crypt–villous axis of the / mice (Fig. S2, A and B, available at ). However, electron microscopy revealed that the villous IECs of the / mice expressed microvilli and lacked the secretory granules characteristic of true Paneth cells (unpublished data). The secretory goblet cell lineage was preserved in the / mice and their control littermates (Fig. S2, C and D) as well. Thus, proper cell fate determination occurred in the absence of β1 integrin expression. The enlarged and dysplastic crypts in the / mice suggested that the aberrant epithelial proliferation could be responsible for the defective enterocytic differentiation. In the control mice, IECs with nuclear immunostaining for the cell cycle progression marker Ki-67 were confined to the crypt bases () as expected. However, the number of IECs with nuclear Ki-67 was greatly increased in the crypts of the / mice () compared with their control littermates. The / mice also demonstrated ectopic foci of Ki-67–positive IEC nuclei in the villi (), which were absent in the control littermates (). Immunodetection of Musashi-1, a putative intestinal stem cell marker () revealed an expansion of intestinal stem cells in the crypts of the / mice compared with their control littermates (). Musashi-1 was not detected in the villi of the / mice, suggesting that the ectopic foci of proliferating IECs were not stem cells mislocalized to the villi and were instead properly retained in the bottoms of the crypts. Thus, β1 integrin expression is not necessary for proper intestinal epithelial stem cell localization to the crypt bases. During the first two postnatal weeks, crypt development in the mouse intestine occurs. Through the use of chimeric mice, it was previously shown that the nascent crypts are initially polyclonal and become monoclonal by P14, suggesting that stem cell selection occurs and yields a single pluripotent progenitor cell in each mature crypt (). Genetic deletion of Tcf-4 in mice resulted in early postnatal lethality because of a complete lack of IEC proliferation in the nascent crypts, which led to intestinal failure shortly after birth (), demonstrating the essential role of Tcf-4 in intestinal stem cell proliferation and maintenance. Although the expression of nuclear Tcf-4 was limited to a few IECs in the bases of the nascent crypts in P6 control mice (), it was expressed by many more IECs in the nascent crypts and even the villi of P6 / mice (). Immunoblotting of nuclear lysates of the IECs confirmed the greater nuclear Tcf-4 protein expression in IECs of the / mice compared with IECs of their control littermates (). In addition, quantitative RT-PCR showed a significantly greater expression of Tcf-4 mRNA in the IECs from the / mice compared with IECs from their control littermates (). Thus, β1 integrin deletion in the intestinal epithelium causes increased and mislocalized expression of Tcf-4 along the crypt–villous axis. β1 integrins are well known to mediate cell proliferation through extracellular signal–regulated kinase (ERK) activation (for review see ). Examination of intestinal epithelial lysates from P6 / mice and their control littermates failed to show differences in ERK activation that could explain the large differences in IEC proliferation (). This result suggested that other proliferative signaling pathways mediated the hyperproliferation observed in the intestinal epithelium of the / mice. The phenotypic changes observed in the / mice (crypt hyperplasia, defective enterocyte differentiation, severe malnutrition, lipid inclusions, juvenile-like polyps, ectopic intestinal epithelial proliferation, and stromal expansion) were similar to those described in mice with defective Hedgehog signaling (; ; ). Thus, Hedgehog expression was examined. Immunodetection and quantitative RT-PCR demonstrated large reductions of Shh and Ihh in the IECs of P6 / mice compared with their control littermates ( and ). In a previous study, neonatal mice with defective intestinal Hh signaling displayed the most severe changes in the distal ileum and cecum, the latter of which was enlarged (). The crypt hyperplasia in the / mice was most dramatic in the distal small intestine, proximal large intestine, and cecum, which were enlarged in diameter as well (). To determine if defective Hh signaling in the intestine could be a cause of aberrant expression of Tcf-4, neonatal mice were randomized to vehicle or the Hh inhibitor cyclopamine treatment for 7 d. Treatment with cyclopamine resulted in increased and mislocalized expression of Tcf-4 along the entire crypt–villous axis, suggesting that the dysregulation of Tcf-4 expression in the / mice may be due to diminished Hh expression and signaling (Fig. S2, E and F). To further investigate the regulation of Hh expression by β1 integrins, studies on IECs were performed in vitro. Caco-2 cells differentiate into enterocyte-like and electrically resistant monolayers when postconfluent (). As Caco-2 cells differentiated in culture, β1 integrin as well as Shh and Ihh expression increased (). Overexpression of β1 integrin in subconfluent Caco-2 cells in the presence of fibronectin caused increases in Ihh and Shh protein expression above constitutive levels (). β1 integrin overexpression in a rat intestinal epithelial (RIE) cell line with relatively low levels of Shh expression also increased Shh expression (). These results show that intestinal epithelial Hh expression requires an intact β1 integrin signaling pathway. Because regional expression in the central nervous system is dependent on HNF-3β (; ) and the promoter contains consensus binding sites for HNF-3β (; ; ; ), the role of HNF-3β in mediating β1 integrin–induced Shh expression was examined. HNF-3β protein levels were greatly decreased in nuclear lysates of IECs from P6 / mice compared with their control littermates (). Furthermore, HNF-3β mRNA levels were reduced approximately eightfold in / mice compared with their control littermates (). These results suggest a dependence of HNF-3β expression on β1 integrin expression and signaling. Transfection of subconfluent Caco-2 cells () with full-length human induced SHH expression. Because β1 integrins can modulate cell proliferation through activation of MAPK and PI3-kinase signaling, confluent RIE cells, which express relatively low constitutive levels of HNF-3β, were cultured in the presence of fibronectin and in the presence or absence of inhibitors of MEK-1 (PD98059) and PI3-kinase (LY294002) to determine how β1 integrins might regulate HNF-3β expression. Although MEK-1 inhibition failed to change HNF-3β expression (unpublished data), PI3-kinase inhibition caused HNF-3β expression to decrease (). When RIE cells with low levels of β1 integrin expression were transiently transfected with a full-length human β1 integrin construct, HNF-3β expression increased (). Furthermore, the PI3-kinase inhibitor LY294002 abrogated this β1 integrin–induced increase in HNF-3β expression (). These studies show that β1 integrin–PI3-kinase signaling stimulates HNF-3β expression, which in turn increases transcription in IECs. Here, we show through the use of a genetic mouse model the key role of β1 integrins and, by inference, their ECM protein ligands in the regulation of IEC proliferation. The rapid turnover of enterocytes requires an early commitment to enterocytic cell fate and rapid differentiation of crypt IECs to achieve the vital function of nutrient absorption before the enterocytes are shed. As the IECs migrate out of the crypts and onto the villi, they undergo cell cycle arrest and differentiation (). The lack of regulation of intestinal epithelial proliferation in the / mouse likely contributed to defective enterocytic differentiation and fat malabsorption, which contributed to their postnatal lethality. However, we cannot conclude from our data that deletion of β1 integrins negatively affected IEC differentiation independent of IEC proliferation. The epithelial crypt hyperproliferation and dysplasia in the / mice were unexpected because β1 integrins were previously shown to promote anchorage-dependent cell proliferation in a variety of cells and tissues (for review see ) and growth factor–induced nuclear translocation of activated ERK (; , ; ). Conditional deletion of β1 integrins in the epidermal and mammary epithelia, but not in intestinal epithelium, caused decreased epithelial stem cell proliferation and ERK activation (; ; ). A major difference between β1 integrin deletion in the intestinal epithelium compared with epidermal or mammary epithelium was that the structure of the basement membrane and epithelial cell adhesion were disrupted in the epidermal and mammary epithelia (; ; ) but maintained in the intestinal epithelium. That conditional deletion of β1 integrins in the intestinal epithelium of mice failed to cause defects in IEC adhesion or survival was somewhat surprising because inhibition of β1 integrin–mediated adhesion in IECs ex vivo resulted in anoikis (; ). A large disruption of IEC adhesion during the neonatal period, a time when the intestine becomes rapidly colonized with bacteria, would have resulted in necrotizing enterocolitis because of a loss of the mucosal barrier (), which was not observed. Thus, the maintenance of anchorage of the IECs to a normal basement membrane in the / mice suggests that the principal function of β1 integrins in the intestinal epithelium during intestinal development is not cell anchorage but ECM-induced regulation of epithelial proliferation. The phenotypic changes of the / mice and neonatal mice with defective Hh expression or signaling (; ; ) are strikingly similar. Although the expression of the Hh receptor Patched and downstream signaling proteins Smoothened and Gli were unaffected (unpublished data), the expression of Shh and Ihh was greatly reduced in the / mice. Furthermore, transient expression of β1 integrin in IECs increased Shh expression. These novel results demonstrate that Hh expression is dependent on β1 integrin expression and signaling in the intestinal epithelium. Although the intensity of Hh signaling is regulated through several important posttranslational steps (, ; ), what governs transcriptional regulation and expression of Hhs is poorly understood. The expression of the Forkhead family transcription factor HNF-3β (Foxa2), which is involved in Shh expression (; ; ; ), was significantly reduced in the / mice, and overexpression of HNF-3β in IECs stimulated Shh expression. Thus, β1 integrins may mediate Shh expression via HNF-3β in IECs. Which α/β1 integrin heterodimers contribute to Hh expression in the intestinal tissues is presently unknown. Fibronectin is expressed in the intestinal crypts and α5/β1 integrin, the classical fibronectin receptor, along the crypt–villous axis (; ). Although fibronectin expression was increased in the stroma of the / mice, α5/β1 integrin expression was lost in comparison with their control littermates (Fig. S2, G and H). Expression of α5/β1 integrin in Caco-2 and HT-29 intestinal cells, which lack α5/β1 integrin expression, increased Shh protein expression in the presence of fibronectin (Fig. S2 I). These studies suggest that α5/β1 integrin expression may be important for Shh expression in the intestinal epithelium. The intestinal crypt hyperplasia and dysplasia in the / mice were similar to findings in mice in which was conditionally deleted in the intestinal epithelium (). Epithelial dysplasia is the earliest neoplastic change preceding macroscopic adenomatous polyp formation in the intestine and is most commonly associated with mutations in human and mouse intestines (; ). Crypt dysplasia occurred within 3–4 d after induction of deletion in the intestinal epithelium but, similar to the / mice, the overlying villous epithelium was not dysplastic (). The unexpected finding of crypt dysplasia after the conditional deletion of intestinal epithelial β1 integrins may relate to the previous observations that β1 integrins are decreased in expression during human intestinal carcinogenesis and can abrogate tumorigenesis when expressed in colon cancer cells (, ; ; ). It is presently unknown if the dysplastic crypts would have given rise to adenomas in the / or -deleted mice because they all died rapidly after birth and APC deletion, respectively. The increased expression of the putative intestinal stem cell marker Musashi-1 in the crypt IECs of the / mice compared with their control littermates suggests that β1 integrins are important in regulating intestinal stem cell proliferation and prompted examination of Tcf-4 expression in the / mice. Tcf-4 expression is normally restricted to IECs in the crypt bases in neonatal mice (); however, its expression was increased and mislocalized along the entire crypt–villous axis of the / mice. Inhibition of Hh signaling increased Wnt gene expression in the mouse intestinal epithelium (), and Ihh was shown to inhibit Tcf-4 expression in colon cancer cells (). Furthermore, Tcf-4 protein expression was increased and mislocalized along the entire crypt–villous axis in neonatal mice treated with the Hh signaling inhibitor cyclopamine (Fig. S2, E and F). These findings suggest that Hh signaling regulates Tcf-4 expression in the intestinal epithelium. β-Catenin protein expression levels and nuclear localization were similar between the IECs of the / mice and their control littermates by Western blot and immunofluorescence (Fig. S2, J–L), suggesting that canonical Wnt signaling was not increased in the former mice. Recently, it was shown that Tcf-4 can mediate proliferative and noncanonical Wnt signaling (). Thus, it is possible that the increased Tcf-4 expression in the intestinal epithelium of / mice may heighten the sensitivity of IECs to key proliferative signaling pathways. In summary, this study shows that β1 integrins regulate Hh expression in the intestinal epithelium and are required for the proper compartmentalization and regulation of intestinal proliferation. Because the Hh receptor Patched has been shown to be localized to stromal cells in the intestine (; ), it is likely that the ability of Hhs to regulate intestinal proliferation is stromal dependent. Finally, the dispensability of β1 integrins for intestinal basement membrane formation and epithelial anchorage, but requirement for intestinal epithelial and stromal regulation, exemplifies their strategic role as mediators of epithelial–stromal cross talk. The and mice were previously described (; ). and mice were mated and the offspring were backcrossed to generate / mice. Genotyping was performed on genomic DNA isolated from tail snips or whole intestine as previously described (; ). The pups were killed when they displayed lethargy and inability to feed. The intestines and other organs were harvested, washed in PBS, and fixed in formalin overnight or frozen in an ethanol–dry ice bath. All animal studies were approved by the Institutional Animal Care and Use Committees at the University of Utah and Salt Lake City Veterans Affairs Health Care System. The mice were a gift from E. Fuchs (The Rockefeller University, New York, NY). The following antibodies were used: HNF-3β, Ihh, Shh (N-terminal), and Tcf-4 pAb (Santa Cruz Biotechnology, Inc.) polyclonal antibodies; integrin α5 mAb (BD Biosciences); Ki-67 mAb (BD Biosciences); integrin β1 mAb and collagen I, collagen IV, fibronectin, and laminin polyclonal antibodies (Chemicon); and actin mAb (Neomarkers). Fixed tissues were embedded in paraffin as described previously (). The samples were deparaffinized in xylene and rehydrated in a 30–100% ethanol series and ddHO. Antigen retrieval was performed by boiling the samples in 10 mM Citrate Buffer, pH 6.0, in a microwave oven. The slides were then washed with 1× PBS for 5 min at RT. The samples were blocked in 3% horse serum, 3% bovine calf serum, or 3% goat serum in 0.1% Triton X-100/1% BSA in PBS for 30 min at RT in a humidity chamber. Primary antibody dilutions in the blocking buffer were incubated with the samples overnight in a humidity chamber at RT. For immunohistochemical detection, the samples were first deparaffinized and rehydrated as above. The endogenous peroxidases were quenched with 3% HO in PBS or 1× TBS for 10 min at RT. The slides were then washed with 1× TBS or 1× PBS. Antigen retrieval (which must be optimized for each tissue) was performed by adding Target Retrieval Solution (DakoCytomation) for 30 min at 90°C. The slides were washed with PBS. The samples were blocked in 3% bovine calf serum, 3% goat serum, or 3% horse serum in 1% BSA in PBS depending on the species used to generate the primary antibody. All incubation steps were done in a humidified chamber. At this point, we used the immuno or pap pen to mark an aqueous barrier around each tissue. Antibody incubation was performed as above except that a horseradish peroxidase–conjugated secondary antibody was used. The slides were incubated in ABC reagent (Vector Laboratories) for various times up to 30 min. The slides were washed in 1× TBS. The reaction was stopped by placing the slides in ddHO. The slides were counterstained with hematoxylin for 1 min. The slides were dehydrated in a 70–100% ethanol series and xylene. The slides were mounted with coverslips. IECs from mouse intestines were isolated as previously described (). Nuclear lysates were prepared from IECs using the following protocol: the IECs obtained above were washed in PBS and pelleted at 300 rpm at RT for 3 min. The PBS was removed, and the cells were resuspended in ice-cold 200 μl Buffer A (20 mM Hepes, pH 7.4), 1 mm EDTA, 1 mM EGTA, 10% glycerol, and 0.2% NP-40; just before use, a protease inhibitor cocktail (aprotinin, 10 μg/ml leupeptin, 10 μl NaVO [0.2 mM], 50 μl Dithiothreitol [1 mM], and 50 μl PMSF [0.5 mM]) was added. The cells were vortexed five times for 10 s each. The lysates were centrifuged at 500 for 5 min at 4°C, and the supernatants (cytoplasmic fraction) were removed. 50 μl of Buffer B (20 mM Hepes, pH 7.4), 1 mM EDTA, 1 mM EGTA, 10% glycerol, 0.2% NP-40, and protease inhibitor cocktail (see above) were added just before use. The cells were vortexed five times for 10 s each, and the lysates were incubated on ice for 30 min. The lysates were centrifuged at 13,000 for 10 min at 4°C, and supernatants (nuclear preparations) were removed. Whole intestine was homogenized in lysis buffer (50 mM Hepes, 150 mM NaCl, 1.5 mM MgCl, 1 mM EGTA, 100 mM NaF, 10 mM NaPO, 1 mM NaVO, 10% glycerol, 1% Triton X-100, and 1 μg/ml each of aprotinin, leupeptin, chymostatin, and pepstatin) on ice. The homogenates were then sonicated for 10 s and clarified by centrifuging at 14,000 for 15 min at 4°C. The Triton soluble and insoluble pellets were boiled in sample buffer (125 mM Tris-HCl, pH 6.8, 20% glycerol, 4% sodium dodecyl sulfate, 2% β-mercaptoethanol, and 10 μg/ml bromophenol blue) for 3 min. Western blots were generated as previously described (). Total RNA was isolated in RNeasy kit (QIAGEN) from sections (∼3 × 3 × 3 mm) of freshly resected intestine or purified IECs (). First-strand cDNA was synthesized from 1 μg of total RNA using M-MLV reverse transcriptase (Invitrogen). Quantitative RT-PCR was performed using SybrGreen incorporation or Taqman primer probe sets on a Sequence Detection System (ABI PRISM 7900HT; Applied Biosystems). Threshold cycles for TaqMan primers were normalized to threshold cycles for actin, and threshold cycles for SybrGreen primers were normalized to glyceraldehyde-3-phosphate dehydrogenase (G3PDH). Quantitative PCR results were compared by analyzing the differences in the midpoints of the linear phases of the appearance of double-stranded DNA products using Sybr Green for six pairs of knockout and control mice. Each gene expression product was normalized against G3PDH for that sample before the comparisons. A two-sided test with unequal variance was used to statistically compare the results. The mean nonfasting cholesterol and triglyceride levels were determined (Anilytics) on serum levels and compared using a test. Caco-2 and IEC-6 RIE cells (American Type Culture Collection) were cultured in DME supplemented with 10% fetal bovine serum, glutamine, penicillin, and streptomycin to ∼70% confluency on plain or fibronectin-coated dishes. Transient transfections of a full-length human pcDNA4 (Invitrogen), pLacZ, construct (pcDNA4-), or pRc/CMV- (a gift from V. Besnard, Cincinnati Children's Medical Center, Cincinnati, OH) were performed with Lipofectamine (Invitrogen) reagent as previously described (). The full-length human gene was digested from pECE-α5 (a gift from E. Ruoslahti, Burnham Institute, La Jolla, CA) with EcoRI and ligated into the EcoRI site of pcDNA4 (Invitrogen). Fig. S1 shows ECM protein expression patterns in the intestines of control and conditional β1 integrin knockout mice. Fig. S2 shows intestinal epithelial lineage markers, α5/β1 integrin expression, and β-catenin expression in control and conditional β1 integrin knockout mice, as well as Tcf-4 expression in mice treated with the Hedgehog signaling inhibitor cyclopamine. Online supplemental material is available at .

I had really good science teachers in high school. I took biology and chemistry, and I got really excited by that. Between my junior and senior years of high school, I took a botany course sponsored by the National Science Foundation, and that gave me a lot of confidence. If I'd had wonderful English teachers instead, I might be quite different. My parents weren't scientists or anything like that. My father has an engineering background, my mother was a musician. I went to Yale, and originally I tried out chemistry, but I switched to biology. I really loved biology, and so I stuck with it. I liked just about everything I studied but particularly evolutionary biology. I never wanted to go to medical school or any of that stuff, like all my friends, just cutting up another corpse. I was studying ecology and evolution, and I was very much influenced by a guy named Charles Remington; he worked on butterfly evolution. I love this stuff. I thought I could do something like that in plants. So I got excited, and I went to work with Peter Ray at Stanford—an exceedingly good plant physiologist—but the project was way ahead of its time. There weren't the necessary molecular tools. It was clear as soon as I started that none of that was going to work. So I did a much more conventional developmental biology/plant hormone Ph.D. thesis. I like unusual organisms. I've always liked plants, I've always liked microbial organisms. Different types of organisms that aren't mammalian/metazoan tissue culture cells can give you an edge on certain problems. I don't think I would've thought of it that way back then, but I just liked being a little bit different. This was '67. It was the hippy era! There was a small element of cell biology in my Ph.D. thesis, which was to think about whether cytoplasmic streaming was involved in hormone movement, and so I got interested in cell motility, streaming, and cellular architecture. At that time, it wasn't even clear that things like actin and myosin were important for plants. People were thinking that things like actin were much more muscle-based and metazoan and not present in every single eukaryotic cell. So I was going to work on cytoplasmic streaming with Dick McIntosh in Boulder (Colorado). But instead we thought we would work on mitosis. This all makes sense in hindsight, but none of it was predictable. It's one of these things where you wander to the top of a mountain, and you can look down when you're there, but you would never imagine that was going to be your path. Most of the things cells do are in interphase, not at cell division, but visually, that's their most dramatic time. So I suspect that it was seeing the chromosomes move that really got me fired up —it's just a thrill. It still is. Yeah, to my surprise, I got hired as an assistant professor! I arrived at Berkeley in January '76, right on my 30th birthday. This is my only job. I've never moved. I've been here for 30 years. I've worked on a variety of model organisms. But the thing that was really wonderful about maize is that it has the most beautiful chromosomes. The cytology is unbelievable. Getting into this was really the product of three people. One was Mike Freeling, a really excellent maize geneticist here at Berkeley who convinced me to start looking at maize. Another was Inna Golubovskaya, a wonderful Russian scientist, now at Berkeley with me, who's one of the foremost geneticists working on maize meiosis. She had this incredible mutant collection. And the third was John Sedat at UCSF, who was into advanced light microscopy. He had developed deconvolution methods, so we applied this imaging technology to maize and looked at the mutants and chromosome behavior. It was serendipity. A lot of my life has been that way. You meet somebody interesting, and they tell you a great story, and you get fired up. I've always been interested in evolution, and in the last five or six years my lab has gone back to thinking about how the cytoskeleton and cell division evolved in eukaryotes. I guess the thing that fascinates me is you've got to go from one to two, and how do you do it faithfully? It's one of the most amazing things that cells do. It's clear that forming the spindle in mitosis is a dramatic and elegant event. It involves sorting and movement and this and that. In meiosis, it's even more difficult, because first the homologous chromosomes have to find each other. Evolution has made it all work! It's fun to get back to this after 30 years. Yeah, in most of my studies I picked the organisms not because of the evolutionary twist, but because they had some aspect of their biology that made it much easier to look at, for example, chromosome structure. In fission yeast, cytology of the chromosomes is easy because they're big and there are 3, as opposed to 16, in budding yeast. Plus you have all the molecular biology tools for yeast. Maize is incredible for cytogenetics, and it also has a wonderful mutant collection. I started working in diatoms because they have highly ordered spindles. I've worked with mammalian cells. You have to take what the systems will give you. But , that's a real attempt to pick an organism not because it's a little different, but because it really might illuminate evolution. Ultimately, plants, animals, and yeast may be very different in how they do certain things, but the underlying mechanism is going to be very similar. Whereas , who knows? It is so divergent. Evolutionarily, plants, animals, and yeast shared a common ancestor maybe three-quarters of a billion years ago. last shared a common ancestor with man or yeast maybe two billion years ago. The logic of the spindle is similar to that of metazoans. Many aspects of its architecture are similar. Scott Dawson, a former post-doc, started this project and has been investigating the various motors at the kinetochore that are involved in chromosome movement. He found that many of the motors in metazoans are also in , and most seem to be playing similar roles. Some of the motors are absent from yeast but are present in metazoans and in . It's not that yeast never had them, they just threw them away. Also, has flagella and basal bodies, just like humans. Yeast doesn't have any of that. If you compare yeast and man, without an out group, it's hard to know what's happening. Are you really adding on or are you losing? tells you that, for example, the complexity of spindle motors is very ancient. Not that it evolved subsequent to the divergence of yeast and man. What I'm going to argue for the future is that more of these very divergent eukaryotic microbes should be looked at, because it's here that you're going to begin to understand the impact of the evolution of these very basic processes. I think the cell biology community recognizes the importance of evolution. But their idea of evolution and thinking about evolution with respect to fundamental cellular processes is to compare, let's say, yeast and man. And maybe they throw in plants. This is sort of like comparing houses in the same neighborhood with similar building codes. Some of them have a garage, some of them don't, some of them have two stories, some have one. But these are not profound differences compared with, let's say, housing in Africa or India. The basic problem for most cell biologists now is they're not using microbial diversity. We've started to, and we hope others will as well.

Centrosomes comprise a pair of centrioles and a surrounding pericentriolar matrix (PCM). They are the major microtubule (MT) organizing centers (MTOCs) in animal cells and are thought to play an important part in organizing many cell processes, including cell polarity, cell migration, and cell division (). Centrioles are also required for the formation of cilia and flagella (), and they have important roles in many developmental processes (). Not surprisingly, centriole or centrosome dysfunction has been implicated in a wide variety of human genetic diseases (). Although much is known about the protein composition of the centrosome, it remains unclear how centrosome structure and organization are maintained. In flies, the centrosomal protein Centrosomin (Cnn) is required to recruit several proteins to the centrosome (, ; ; ). In mutant embryos, and in somatic cells lacking Cnn, the centrosomes fail to function as MTOCs during mitosis and anastral spindles assemble through a centrosome-independent pathway. This leads to dramatic mitotic defects in embryos (; ) but only to subtle mitotic defects in somatic cells (; ), presumably because centrosomes are not essential for cell division in somatic cells (; ). Cnn is a member of a family of structurally related proteins that have been implicated in organizing MT arrays. In the yeast , the Cnn-related protein Mto1 recruits the γ-tubulin complex to several types of MTOCs (; ). In human cells, the Cnn-related proteins CDK5RAP2 and Myomegalin/PDE4-DIP are concentrated at centrosomes, but their function is unknown (; ). Mutations in the gene encoding CDK5RAP2, however, cause autosomal recessive primary microcephaly, in which the brain is small at birth and thereafter (). The underlying cause of microcephaly is unknown, but it has been proposed that a failure of the centrosomes to function as efficient MTOCs in mitosis might lead to defects in asymmetric neuroblast (NB) divisions during fetal development (; ; ). Here, we have used live confocal imaging to examine how Cnn functions to ensure the proper organization of the centrosome in flies, and to test whether Cnn is required for asymmetric divisions in larval NBs. In fixed embryos and somatic cells that lack Cnn, PCM components are barely detectable at the poles of the mitotic spindles (, ; ). Centrioles are still present in mutant cells (), but their function and positioning within the centrosome have not been analyzed. To understand better how Cnn normally recruits PCM components to the centrioles, we generated transgenic lines expressing an mRFP-centriolar marker (either mRFP-Fzr or mRFP-PACT), together with one of three PCM markers fused to GFP: Aurora A–GFP, Grip75-GFP (a component of the γ-tubulin ring complex), and GFP- D-TACC. It has previously been shown that Cnn can interact with both the γ-tubulin ring complex and Aurora A (Terada et al., 2003), but we found no evidence for an interaction between Cnn and D-TACC in coimmunoprecipitation experiments (unpublished data). In wild-type (WT) syncytial embryos, centrioles recruited approximately equal amounts of PCM at all stages of the rapid mitotic cycles, and they remained well centered within the PCM throughout the cell cycle (; Fig. S1; and Videos 1 and 2, available at ). During interphase, the centrioles were always closely associated with the nuclear envelope, whereas in mitosis, they were always closely associated with the spindle poles (Videos 1 and 2). In embryos laid by homozygous females (hereafter, embryos), we were surprised to observe that the centrioles were associated with appreciable amounts of PCM, but they were often not properly centered within it (; Fig. S1; and Videos 1 and 2). In video recordings of embryos, the centrioles appeared to be constantly nucleating PCM but seemed unable to maintain their connection to it. The centrioles often exhibited irregular, stochastic movements, leaving a trail of PCM behind them as they moved away. This PCM trail was most easily seen in embryos expressing GFP–D-TACC (Video 1), as this protein was recruited in particularly large amounts to the centrioles, and large clusters of GFP–D-TACC often remained in the cytoplasm for some time after the centrioles had moved away. Smaller amounts of Aurora A–GFP and Grip75-GFP were recruited to the centrioles (Video 2), and so only small amounts of these proteins remained associated with the centrioles as they moved around the embryo. As a result of this abnormal centriole behavior, the centrioles in embryos often lost their attachment to the nuclear envelope in interphase and to the spindle poles in mitosis. We refer to this behavior of the centrioles as “centriole rocketing” (see the following section). Previous studies suggested that centrosomes lacking Cnn fail to function as MTOCs during mitosis (, ; ; ). We therefore examined whether the PCM organized by the centrioles in embryos was capable of nucleating MTs. As shown in (A–D) and Videos 3 and 4 (available at ), the centrosomes in embryos organized astral MT arrays but seemed unable to maintain their connection with them. When the embryos entered mitosis, many nuclei were not associated with centrioles, and anastral spindles assembled around the mitotic chromatin (not depicted). Many nuclei, however, were close enough to a centriole for the astral MTs to contribute to spindle assembly (, arrow). Often, however, these centrioles failed to maintain their position at the spindle pole and either wandered around within the spindle (, red arrowhead) or lost their connection to the spindle altogether (, yellow arrowhead). We conclude that the dramatic mitotic defects observed in embryos do not result from a failure of the centrioles to recruit PCM, or of the centrosomes to nucleate astral MTs, but instead result from the failure of the centrioles to maintain a stable connection to the PCM and MTs that they organize. The centriole rocketing appeared to be driven by the asymmetric organization of the PCM and MTs around the centrioles (Videos 1–3). To test whether the rocketing was MT dependent, we injected the MT-depolymerizing drug colchicine into embryos. In WT embryos in late interphase, the centrioles had already migrated around the nuclei and, when we plotted their movement over time, the centrioles moved regularly across the embryo cortex ( and Video 5, available at ). This regular movement of the centrosomes and their associated nuclei across the cortex is driven by actin- and myosin-dependent cortical contractions, and it continued after colchicine injection (Video 5). In contrast, when we plotted centriole movement in embryos in late interphase, we observed the rocketing behavior described previously ( and Video 5). The rocketing ceased after colchicine injection, and the centrioles reverted to a regular movement across the embryo cortex ( and Video 5). Thus, centriole rocketing in embryos depends on intact MTs. The injection of colchicine into embryos also enabled the centrioles to remain associated with the PCM (), suggesting that it is the MT-dependent rocketing of the centrioles that ultimately breaks the link between the centrioles and the PCM in embryos. Intriguingly, however, the injection of colchicine into embryos did not correct the positioning defect of the centrioles within the PCM: whereas the centrioles were usually (>90%) well centered within the PCM in colchicine-injected WT embryos (between 50 and 100 centrioles observed with each of the three different PCM markers; and Video 6, available at ), they were very rarely centered within the PCM in colchicine-injected embryos (<10%) and were usually positioned at the very edge of the PCM (>100 centrioles observed with each of the three different PCM markers; and Video 6). This last observation was unexpected, and we are unaware of any other perturbation to the centrosome that results in this very specific displacement of the centrioles from the center of the PCM. This observation may have important implications for understanding how Cnn functions to maintain the link between the centrioles and the PCM. One interesting possibility is that the MT-dependent centriole rocketing we observe in embryos may be mechanistically related to the actin-dependent rocketing of certain pathogenic bacteria (; ). These bacteria are coated with proteins that initially stimulate the polymerization of an actin “cloud” symmetrically around the surface of the bacteria. If the actin surrounding the bacteria is structurally weak, it can “fracture”, allowing the bacteria to move to the edge of the actin cloud and rocketing to begin (; ). Thus, we propose that the primary function of Cnn may be to mechanically strengthen the PCM: in the presence of Cnn, the PCM is structurally strong and the centrioles can maintain their position at the center of the PCM; in the absence of Cnn, the PCM is weakened and the centrioles move to the edge of the PCM. This then initiates centriole rocketing, although the exact mechanism of this MT-dependent rocketing remains unclear. Maintaining the proper connection between the centrioles and the PCM is clearly crucial in syncytial embryos, as a lack of Cnn results in catastrophic failures in mitosis. In contrast, somatic cells that lack Cnn have few mitotic defects, and mutant flies are viable (; ). To test whether Cnn was required to maintain the proper connection between the centrioles and the PCM in somatic cells, we treated third instar larval brain cells with colchicine to depolymerize the MTs and then fixed and stained them to examine the distribution of the centrioles and the PCM. We found that hardly any PCM was detectable around the centrioles in brain cells that had not been treated with colchicine (unpublished data). In cells treated with colchicine, however, considerable amounts of PCM accumulated around the centrioles, but, as in embryos, the centrioles were displaced from the center of the PCM (Fig. S2, available at ). To further investigate whether the centrioles in somatic cells behaved in the same way as the centrioles in embryos, we examined living third instar larval NBs expressing the centriole marker DSas-4–mRFP and GFP–α-tubulin. In WT NBs entering mitosis, the centrioles were always centered within astral MT arrays, and the centrioles remained tightly associated with the poles of the spindle throughout mitosis (). In contrast, the centrioles in NBs were often not associated with prominent astral MTs and exhibited irregular movements throughout the cell during mitosis. As a consequence, they were often abnormally displaced from the poles of the mitotic spindles (). Nevertheless, we could transiently detect astral MTs associated with some of the “rocketing” centrioles in some NBs (, arrows; see the next section). In fixed larval NBs, the centrioles were often randomly positioned around the cell (), and we noticed that 20–30% of brain cells had either too few or too many centrioles (). Taken together, these findings suggest that the centriole behavior is similar in embryos and somatic cells; while these defects do not lead to dramatic errors in somatic cell division, they do lead to errors in centriole segregation. These findings support the hypothesis that centrioles have evolved the ability to recruit PCM to ensure the equal partitioning of the centrioles during cell division, rather than to ensure the efficient assembly of the mitotic spindle (; ). The Cnn-related protein CDK5RAP2 has been implicated in human microcephaly (), and several recent studies have shown that centrosomes exhibit an asymmetric behavior during the asymmetric divisions of male germline stem cells (GSCs) and larval neural stem cells (NBs) (; ; ). During interphase in these cells, only one centrosome is initially associated with PCM and MTs, and this centrosome becomes anchored on one side of the cell (near the stem cell niche in GSCs, or near cortically localized cell polarity markers in NBs). When the second centrosome eventually associates with PCM and MTs, it localizes to the opposite side of the cell, thus ensuring that the forming mitotic spindle is correctly oriented relative to these positional cues. This asymmetric centrosome behavior appears to be important in male GSCs, as the asymmetric division of these cells is dramatically perturbed in mutants (). Although mutant NBs have defects in aligning their spindles with cortical determinants early in mitosis (), it is not clear that this ultimately leads to failures in asymmetric division: early mitotic spindle alignment defects are often corrected in these cells by the time the cells divide (). To determine whether mutant NBs ultimately divide asymmetrically, we analyzed living WT and third instar larval NBs expressing only GFP–α-tubulin. As reported previously (; ), a single, anchored MTOC was usually visible in WT NBs before the entry into mitosis (not depicted). After nuclear envelope breakdown (NEB), however, both centrosomes nucleated prominent arrays of MTs, and spindle assembly occurred primarily by a centrosomal pathway ( and Video 7, available at ). As expected, the cells divided asymmetrically to produce a large NB and a small ganglion mother cell (GMC). In most NBs, no prominent MTOC was detectable before NEB, and spindle assembly occurred largely by an acentrosomal pathway ( and Video 8). Nonetheless, ∼95% of NBs ultimately divided asymmetrically ( = 81; and Video 8), whereas ∼4% divided symmetrically ( and Video 9) and ∼1% failed in cytokinesis (not depicted). Although this failure rate is modest, we believe it is considerably higher than in WT, as we observed only one symmetric division in >100 WT central brain NBs examined (unpublished data; Basto, R., and C.I. Dix, personal communication). We previously showed that mutations in , which encodes the homologue of the human microcephaly protein CenpJ/CPAP (), also lead to defects in the asymmetric divisions of larval NBs. The defects were much more severe in mutants, which completely lack centrioles/ centrosomes (∼15% of NBs divided symmetrically, whereas ∼15% failed in cytokinesis; ). The much milder defects in asymmetric division that we observe in NBs suggest that centrosomes are partially functional as MTOCs in mutant somatic cells, consistent with our previous observations ( and Fig. S2). Indeed, we frequently observed relatively well-focused astral MT arrays forming and disassembling in the cytoplasm, and these were often transiently associated with the spindle poles in NBs (, arrow; and Video 10, available at ). Taken together, our observations on and mutant NBs reveal that, unlike the situation in male GSCs, the asymmetric behavior of the centrosomes is not essential for the accurate asymmetric division of larval NBs. Nevertheless, mutations in the homologues of two of the three human centrosomal proteins implicated in microcephaly do lead to relatively subtle defects in NB divisions in flies. and mutants do not have small brains, suggesting that flies are able to compensate for defects in these divisions in a way that perhaps humans cannot. Oregon R or flies were used as the WT stock, and flies as the parental stock for the generation of all transgenic lines. The null allele has been described previously (; ), and the allele, a insertion in the middle of the gene (Exelixis stock no. f04547), was obtained from the Exelixis Stock Centre (Harvard Medical School, Boston, MA). In Western blotting experiments, embryos laid by females homozygous for this allele had no detectable Cnn protein. GFP fusions to the following proteins were used in this study: Aurora A–GFP (Lau, J., personal communication), GFP–D-TACC (), Grip75-GFP (), GFP–α-tubulin (), and DSas-4–GFP (). We also generated fusions between mRFP and the full-length cDNA, the PACT domain of D-PLP (), and DSas-4 (Basto, R., personal communication). Most of these fusions were subcloned into the pWR-Ubq transformation vector that drives the ubiquitous expression of the fusion protein at moderately high levels. For Grip75-GFP and GFP–α-tubulin, we used previously established lines in which the expression of these proteins is under the control of the UASp promoter (); we drove their expression in embryos using the maternal 67C α-tubulin–GAL4 promoter (), and in brains using the 69B enhancer trap line (). Transgenic lines were generated using standard methods. Most studies were performed using the allele, but we obtained similar results with either the allele or transheterozygous combinations of the two alleles. Live embryos expressing fluorescent fusion proteins were examined as described previously (). The embryos were observed on an ERS spinning disc confocal system (PerkinElmer), mounted on an inverted microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) that was equipped with a charge-coupled device camera (Orca ER; Hamamatsu), using a 63×/1.25 NA objective. For each combination of GFP and mRFP fusions, 4–15 WT and mutant syncytial blastoderm stage embryos were examined. For drug injections, live embryos were injected at the desired stage with 5 mg/ml colchicine (Sigma-Aldrich). All images were captured and made into videos using the Ultraview ERS software (PerkinElmer). Third instar larval brains were prepared as described previously (Basto et al., 2006), and central brain NBs were followed by time-lapse confocal microscopy as described, with a 100×/1.3 NA objective. 12–20 focal planes spaced by 0.5 μm were acquired every 25 or 45 s (0.5–1 s/frame, respectively). All images shown are maximum intensity projections of z stacks at selected time points. Third instar larval brains were fixed and stained as described previously (). For drug treatment, dissected brains were incubated for 2 h at 25°C in PBS containing 1 μg/ml colchicine before fixation. The following primary antibodies were used: rabbit anti–D-PLP at 1–2 μg/ml (); mouse anti–α-tubulin at 1:1,000 (DM1α; Sigma-Aldrich), mouse anti–γ-tubulin at 1:1,000 (GTU88; Sigma-Aldrich), and mouse anti–phospho-Histone H3 at 1:2,000 (Abcam). All secondary antibodies coupled to the appropriate fluorophore (Alexa 488 or 568; Invitrogen) were used at 1:1,000 in PBT. Fixed preparations were examined on a widefield upright microscope (Axioskop II; Carl Zeiss MicroImaging, Inc.), equipped with a camera (CoolSnap HQ; Photometrics) and MetaMorph software (Molecular Devices), using a 100×/1.3 NA objective. Images of fixed brain cells are all maximum intensity projections of optical sections acquired at 0.1– 0.2-μm intervals. Individual images were imported into Photoshop 7.0 (Adobe) and adjusted to use the full range of pixel intensities. In panels for some figures, pixel intensities were adjusted using the “curves” control panel, and an unsharp mask and despeckle filter were applied to the whole image. In all cases, the images from control and experimental embryos were adjusted in the same way. To quantify the amount of PCM recruited to centrioles, individual images were imported into MetaMorph. The centrosome was circled, and the integrated intensity was calculated for independent centrosomes after subtraction of cytoplasmic background fluorescence. The integrated intensity per pixel area was determined from at least three different WT and embryos per centrosomal marker. A total of 45 centrosomes were scored for each marker. Error bars represent the SD. The data were analyzed for statistical significance using a two-tailed test. To quantify the number of centriole dots per mitotic cell, fixed preparations of third instar larval brains were stained with anti–D-PLP and anti–phospho-Histone H3 antibodies. We examined a minimum of 70 mitotic (phospho–Histone H3 positive) cells per larval brain from at least six different brains. A total of 718 mitotic cells from WT and 976 mitotic cells from homozygous mutants were scored. Error bars represent the SD. Semiautomated tracking software (Imaris 4.5.2; Bitplane AG) was used to identify and track DSas-4–GFP trajectories over time (150 time frames; 1 s/frame). Image segmentation was performed to convert pixel intensities above a given threshold into computerized spots, and this method was applied equally to all the images in the time series. Semiautomatic track building was based on autoregressive motion algorithms. Fig. S1 shows the localization of Aurora A–GFP, Grip75-GFP, and centrioles in WT and embryos. Fig. S2 shows the centriole positioning defects in larval brain cells after colchicine treatment. 10 additional videos are also included, showing the behavior of centrioles and GFP-D-TACC in WT and embryos (Video 1); centrioles and Aurora A–GFP or Grip75-GFP in WT and embryos (Video 2); centrioles and GFP– α-tubulin in a embryo (Videos 3 and 4); centrioles in WT and embryos before and after colchicine injection (Video 5); centrioles and PCM in WT and embryos after colchicine injection (Video 6); a WT larval NB expressing GFP–α-tubulin, dividing asymmetrically (Video 7); a larval NB expressing GFP–α-tubulin, dividing asymmetrically (Video 8); a larval NB expressing GFP–α-tubulin, dividing symmetrically (Video 9); a larval NB expressing GFP–α-tubulin with prominent astral MTs, and a small focus of MTs moving around the cell (Video 10). Online supplemental material is available at .

Small nuclear RNPs (snRNPs) are core components of spliceosomes and are required for the catalytic steps of splicing. Most spliceosomal snRNPs, with the exception of U6, contain a common set of Sm proteins that associate with the Sm site of the small nuclear RNA (snRNA; ). The biogenesis of Sm-class snRNPs is a highly orchestrated process that takes place in multiple cellular compartments (). Subsequent to transcription and nuclear export of the snRNA, the survival of motor neurons (SMN) complex mediates the assembly of snRNPs by loading Sm proteins onto the snRNA (; ). The core factor within this large oligomeric complex is the SMN protein (; ). Importantly, mutations that reduce the level of SMN protein result in the inherited human neuromuscular disorder spinal muscular atrophy (). The molecular etiology of spinal muscular atrophy is currently unknown. Recent work suggests that the perturbation of snRNP biogenesis may be a contributing factor (). However, these findings do not rule out tissue-specific functions of SMN as factors contributing to the disease pathology (; ; ). Three of the seven Sm proteins, B/B′, D1, and D3, contain symmetric dimethylarginine (sDMA) modifications within their C-terminal tails (, ). This posttranslational modification is catalyzed by type II protein arginine methyltransferases (PRMTs; for review see ). Type I enzymes catalyze the more common asymmetric dimethylarginine (aDMA) modification. PRMT5 and PRMT7 have each been shown to possess type II methyltransferase activity and to symmetrically dimethylate Sm proteins in vitro (; ; ). Reduction of PRMT5 levels using RNAi correlates with a decrease in the level of Sm protein sDMA modification (). Furthermore, in cytoplasmic lysates, PRMT5 is found in a complex with MEP50/WD45, iCln, and Sm proteins (, ; ). PRMT7 was identified more recently as a type II methyltransferase (Lee et al., 2005). Consequently, very little is known about this enzyme. The precise role of Sm protein sDMA modification in snRNP biogenesis remains unclear. Recruitment of Sm proteins to the SMN complex is thought to be facilitated by sDMA modification. Consistent with this notion, SMN binds with a much higher affinity to sDMA-modified Sm proteins (; ). However, a loss of function mutation in , the fly orthologue of PRMT5, is homozygous viable (). The mutants displayed no overt defect in snRNP levels despite expressing Sm proteins that were not recognized by two sDMA-specific antibodies, SYM10 and Y12 (). To gain a better understanding of the role of Sm protein sDMA modification in snRNP biogenesis in humans, we depleted HeLa cells of PRMT5 or PRMT7 using RNAi. Surprisingly, we found that both PRMT5 and PRMT7 were required for efficient Sm protein sDMA modification. Both enzymes independently associated with Sm proteins but not with each other. In addition, we demonstrate that PRMT5 and PRMT7 do not function in an additive or redundant manner, thus suggesting a unique requirement for each methyltransferase in the Sm protein methylation pathway. Finally, we show that the symmetric dimethylation of Sm proteins is required for cytoplasmic snRNP assembly in human cells. To understand the specific requirement for Sm protein sDMA modification in mammals, we examined the in vivo functions of PRMT5, MEP50, and PRMT7. HeLa cells were depleted of these proteins using RNAi (). siRNAs targeting SMN or GFP were used as controls (). The siRNAs directed against PRMT5, PRMT7, and SMN were able to deplete >80% of their respective target proteins (). Treatment of cells with siRNAs targeting MEP50, a PRMT5 complex member, caused a slight codepletion of PRMT5 (). This finding is consistent with our previous results in : the mutation of resulted in a loss of Dart5 expression (). In contrast, only specific siRNA treatments reduced the level of PRMT7 (). We next analyzed the methylation status of Sm proteins in the depleted lysates using the sDMA-specific antibodies SYM10, SYM11, and Y12 (). Unmodified and asymmetrically dimethylated Sm proteins are not recognized by these antibodies (; , ). Consistent with previous findings (), the knockdown of PRMT5 resulted in a reduction in Sm protein sDMA modification (, lane 2). A similar effect was also observed when cells were treated with siRNAs targeting MEP50 (, lane 4). However, because MEP50 RNAi treatment codepletes PRMT5, we cannot conclude whether this defect in methylation is direct. Curiously, we found that PRMT7 knockdown also caused a reduction in Sm protein sDMA modification (, lane 3). Because the depletion of PRMT7 does not codeplete PRMT5 (), this effect is likely to be direct. In addition to Sm proteins, SYM10 and SYM11 also recognize several uncharacterized sDMA-modified proteins (, ). The same SYM10- and SYM11-reactive proteins were hypomethylated in the PRMT5 and MEP50 siRNA–treated lysates (). This finding was expected because PRMT5 and MEP50 are part of the same complex and also because the depletion of MEP50 codepletes PRMT5. In contrast, PRMT7 siRNA treatment resulted in the hypomethylation of only a subset of these proteins (). Thus, PRMT5 and PRMT7 do not always act on the same substrates in vivo. Consistently, we observed that the protein– protein interaction profile of PRMT5 and PRMT7 was largely nonoverlapping (Fig. S1, available at ). We next determined whether Sm proteins were sDMA modified in an additive manner by depleting cells of both PRMT5 and PRMT7 (). Interestingly, double depletion did not disrupt Sm protein sDMA modification to a greater extent than either single depletion alone. Thus, the two enzymes do not function additively to produce the full complement of methylated Sm proteins. We also attempted to determine whether PRMT5 and PRMT7 could functionally compensate for each other with respect to Sm protein sDMA modification (). We found that the overexpression of PRMT7 was not able to restore sDMA modification of the Sm protein in cells that were depleted of PRMT5. Thus, PRMT7 cannot functionally substitute for PRMT5. At this time, however, we cannot conclude whether the converse is also true. Although cells depleted of PRMT7 alone show relatively little cytotoxicity, they do not survive the subsequent DNA transfection procedure. Transfection of PRMT7 siRNA–treated cells with empty vector or a PRMT5-expressing plasmid resulted in rapid and pronounced cell death. PRMT5 has been shown to associate with the C-terminal arginine-glycine (RG)–rich tail of SmD3 (; ; ). Therefore, we tested for a similar association between SmD3 and PRMT7. We found that the GST-tagged C terminus of SmD3 (GST-D3tail) was able to specifically purify both PRMT5 and PRMT7 from cell lysates (). Interestingly, the association between PRMT7 and GST-D3tail was unaffected by prior treatment of the cells with the methyltransferase inhibitor 5′-deoxy-5′-(methylthio)adenosine (MTA; ). In contrast, a similar treatment disrupted the association between PRMT5 and GST-D3tail (). In addition to GST-D3tail, PRMT7 is able to bind to full-length GST-SmD3 and -SmB (unpublished data). Because PRMT7 is able to associate with Sm proteins but not with PRMT5 (), these observations suggest a direct role for PRMT7 in Sm protein methylation. During the cytoplasmic phase of snRNP biogenesis, the SMN complex loads Sm proteins onto the Sm sites of snRNAs (; ). SMN has a much higher affinity for sDMA-modified Sm proteins in comparison with unmodified or aDMA-modified Sm proteins (; ). Consistent with the finding that both enzymes are required for the sDMA modification of Sm proteins, both siRNA treatments interfered with the SMN–Sm interaction (). Previous studies demonstrated that the SMN–Sm interaction was disrupted by treatment with general methyltransferase inhibitors (; ). Our current results extend these findings and demonstrate that the specific depletion of either PRMT5 or PRMT7 is able to mimic the drug treatment. To determine whether the methylation of Sm proteins is a prerequisite for efficient snRNP assembly, we treated HeLa cells with MTA. Subsequently, a pulse-chase experiment using a mixture of [S]methionine and [S]cysteine was used to examine the in vivo kinetics of snRNP assembly. The newly assembled snRNPs were immunoprecipitated using anti-trimethylguanosine (TMG)–coated beads. Anti-TMG antibodies recognize the cap structure present on mature Sm-class snRNPs. More precisely, this assay monitors the step in snRNP biogenesis that is immediately downstream of Sm core assembly (). Interestingly, we found that MTA treatment substantially disrupted snRNP assembly (), suggesting that methylation is required for efficient snRNP biogenesis. Because MTA treatment reduced but did not completely abolish Sm protein sDMA modification (Fig. S2, available at ), we conclude that the residual snRNPs assembled under these conditions contain methylated Sm proteins. To specifically narrow down the snRNP assembly defect to sDMA modification of Sm proteins, HeLa cells depleted of PRMT5 or PRMT7 were used in the pulse-chase assay. HeLa cells treated with siRNAs targeting SMN served as a control. Consistent with previous results (; ; ), the depletion of SMN severely disrupted snRNP assembly (, C and D; and Fig. S3, available at ). A similar disruption of snRNP assembly was also observed in cells depleted of either PRMT5 or PRMT7 (, C and D; and S3). It is worth noting that snRNP assembly was disrupted to a somewhat lesser extent upon the depletion of either PRMT5 or PRMT7 as compared with SMN (). These findings are consistent with the observation that PRMT5 and PRMT7 depletion reduces but does not eliminate the SMN–Sm interaction (). The most likely function for PRMT5 and PRMT7 in the snRNP biogenesis pathway is to enable the efficient association of SMN with Sm proteins. Therefore, we directly examined the assembly of Sm cores in control or depleted lysates. For this experiment, a HeLa strain stably expressing CFP-SmB () was used. The cells were pulsed with [P]orthophosphate. Subsequently, tagged SmB was immunoprecipitated from the labeled lysates. The associated RNAs were examined by autoradiography. As expected, the depletion of SMN considerably inhibited Sm core assembly (). Likewise, the depletion of either PRMT5 or PRMT7 also resulted in a similar Sm core assembly defect (). The depletion of core snRNP assembly factors results in the breakdown of Cajal bodies (CBs) and the redistribution of coilin to the nucleolus (; ; ). Therefore, we examined whether the depletion of PRMT5 or PRMT7 also leads to CB breakdown (). Consistent with these earlier studies, CBs were undetected in ∼60% of SMN-depleted cells (). In contrast, roughly 30% of cells depleted of PRMT5 or PRMT7 lacked visible CBs (). Also consistent with these earlier studies, coilin localized within the nucleolus in a subset of SMN-depleted cells (unpublished data). This phenotype was not observed in cells depleted of PRMT5 or PRMT7. However, in a subset of cells depleted for PRMT5, PRMT7, or SMN, coilin localized at the nucleolar periphery (). Interestingly, in cells depleted of PRMT5 or PRMT7, we found that these nucleolar caps also contained SMN ( and not depicted). Thus, in contrast to SMN depletion, PRMT5 or PRMT7 depletion produced a milder CB phenotype. One explanation for this finding is that SMN depletion results in a more severe snRNP defect than either methyltransferase deletion (). Alternatively, in contrast to PRMT5 and PRMT7, SMN may play additional roles in targeting imported snRNPs to CBs (). Finally, we tested whether Sm protein methylation was required for snRNP import using the S pulse-chase assay. Labeled cells were harvested, and nuclear and cytoplasmic fractions were prepared. Each fraction was then subjected to immunoprecipitation using anti-TMG antibody–coated beads. TMG-positive RNPs produced during the 1.5-h pulse-chase period were nearly all located in the nuclear fraction (), suggesting that in HeLa cells, snRNP biogenesis and nuclear import occur relatively rapidly. As observed previously (), MTA treatment disrupted snRNP assembly. However, the residual snRNPs that were assembled were present almost exclusively in the nuclear fraction (). Similar to MTA treatment, RNAi-mediated depletion of SMN, PRMT5, or PRMT7 resulted in snRNP assembly defects (). As with MTA treatment, the residual snRNPs that were assembled were also imported into the nucleus (). Thus, sDMA modification of Sm proteins does not play a major role in the nuclear import of snRNPs. During the preparation of this manuscript, PRMT9 was shown to possess type II methyltransferase activity and to methylate a variety of targets in vitro, including SmB (). Therefore, it will be interesting to determine whether PRMT9 plays a role in Sm protein methylation and snRNP biogenesis in vivo. In addition, CARM1/PRMT4, a type I methyltransferase, was recently shown to asymmetrically dimethylate SmB in vivo (). Because aDMA residues are found exclusively on nuclear Sm proteins (), this modification is not likely to be required for the cytoplasmic phase of snRNP assembly. The aDMA modification of Sm proteins may be important for subnuclear targeting of snRNPs or for the regulation of pre-mRNA splicing (). In conclusion, we have shown that two distinct methyltransferases, PRMT5 and PRMT7, are required for Sm protein sDMA modification and snRNP assembly. We envision that both enzymes function in the snRNP pathway by sDMA modification of Sm proteins, thus increasing their affinity for the SMN complex. A previous study showed that the activity of the SMN complex in Sm core assembly is stimulated by phosphorylation (). Here, we demonstrate that sDMA modification of Sm proteins also serves an important regulatory function. Thus, mammalian snRNP biogenesis is controlled by multiple posttranslational events. may differ from mammals in this regard. Whereas PRMT5 is required for snRNP biogenesis in human cells, the loss of Dart5 does not result in decreased snRNP levels in . However, flies also express an orthologue of PRMT7 called Dart7. Therefore, it will be interesting to test whether snRNP assembly in is independent of Sm protein sDMA modification. GST-tagged SmD3 C-terminal tail (GST-D3tail) was a gift from G. Dreyfuss (University of Pennsylvania, Philadelphia, PA). The FLAG-PRMT7 overexpression plasmid was constructed by cloning PRMT7 cDNA into the p3XFLAG-myc-CMV-23 expression vector (Sigma-Aldrich). Because the PRMT7 construct contains its endogenous stop codon, the C-terminal myc tag is not translated. The siRNAs used in these studies were obtained from Ambion. The PRMT5 siRNA sequence is GGCCAUCUAUAAAUGUCUG, and the PRMT7 siRNA sequence is GCUAUUUCCCAUCCACGUG. HeLa cells were cultured in DME supplemented with 10% FBS. For each RNAi transfection, 275 pmol siRNAs were transfected into a 20% confluent T25 flask using the DharmaFECT1 reagent (Dharmacon). For those experiments in which the steady-state methylation status of the Sm proteins was analyzed (), the cells were treated with siRNA twice—once on day 1 and once on day 3. The cells were harvested and analyzed on day 5. This was done to accommodate the long half-life of Sm proteins. For the rest of the RNAi experiments, the cells were treated with siRNA just once. The cells were harvested and analyzed on day 3. DNA was transfected using Effectene (QIAGEN). HeLa lysates were prepared by resuspending the cells in radioimmunoprecipitation assay buffer (50 mM Tris-Cl, pH 7.5, 150 mM NaCl, 1% NP-40, and 1 mM EDTA) containing protease inhibitors (Halt protease inhibitor cocktail kit; Pierce Chemical Co.) and passing several times through a 25-gauge needle. The lysate was cleared by centrifugation at 10,000 for 5 min at 4°C. Nuclear and cytoplasmic HeLa fractions were prepared using the N-PER fractionation kit (Pierce Chemical Co.) as directed. Bacterial lysates were prepared using sonication in 1× PBS/1% Triton X-100/protease inhibitor cocktail. The PRMT5 and PRMT7 antibodies were obtained from Upstate Biotechnology. The SMN antibody (clone 7B10) was a gift from U. Fischer (University of Wuerzburg, Wuerzburg, Germany). The tubulin monoclonal antibody used as a loading control was obtained from Sigma-Aldrich. The methylation status of the depleted lysates was analyzed using Sym10 (Upstate Biotechnology), Sym11 (Upstate Biotechnology), and Y12 anti-sDMA antibodies (gift from J. Steitz, Yale, New Haven, CT). CFP-SmB was immunoprecipitated from cytoplasmic lysates using polyclonal GFP antibodies (Abcam). The precipitates were probed using SMN and GFP monoclonal antibodies (Roche). In the pulse-chase experiments, the newly synthesized snRNPs were precipitated using anti-TMG antibody–coated beads (Calbiochem). For the Sm core assembly assay, CFP-SmB was immunoprecipitated using the polyclonal GFP antibody. The distribution of coilin was examined using a previously generated polyclonal antibody R124 () and a goat anti–rabbit AlexaFluor594 secondary antibody (Jackson ImmunoResearch Laboratories). The localization of fibrillarin was determined using the 72b9 antibody (gift from E. Chan, University of Florida, Gainesville, FL), and the distribution of SMN was verified using the 7B10 monoclonal antibody. A goat anti–mouse FITC antibody (Jackson ImmunoResearch Laboratories) was used to detect fibrillarin and SMN. Western blots and autoradiography film were quantified using the Quantity One program (Bio-Rad Laboratories). Immunofluorescence images were captured using a microscope (DM6000; Leica) interfaced with Volocity software (Improvision). Images were captured at room temperature using a 63× oil immersion objective (Leica). The image stacks were deconvolved using Volocity software. Images were cropped using Photoshop 7.0 (Adobe). The [S]methionine and [S]cysteine pulse-chase assay was performed as described previously () with a few modifications. The chase time was reduced from 1 h to 30 min. The newly synthesized snRNPs were purified using anti-TMG antibody–coated beads. Cells were treated with 750 μM MTA for 24 h. An equivalent volume of DMSO was added to the control cells. In the snRNP import experiment, nuclear and cytoplasmic fractions were prepared subsequent to the pulse chase using the N-PER kit (Pierce Chemical Co.). Subsequently, each fraction was incubated with anti-TMG–coated beads. The P-labeling experiment was performed by first depleting cells of intracellular phosphate by growing in 3 ml phosphate-free DME media for 2 h. Subsequently, 100 μCi [P]orthophosphate (PerkinElmer) was added to the cells. The cells were labeled for 2 h and harvested. For the Sm core assembly reaction, HeLa cells stably expressing CFP-SmB were used. The tagged Sm protein was immunoprecipitated using polyclonal GFP antibodies. The associated snRNAs were then examined using autoradiography. The P-labeled cells were also used in a separate TMG-capping assay. The TMG-capped snRNPs were purified using the anti-TMG antibody–coated beads. The RNA was extracted and examined using autoradiography. Fig. S1 shows that PRMT5 and PRMT7 associate with different proteins in vivo. Fig. S2 shows that residual snRNPs assembled under conditions of MTA treatment contain symmetrically dimethylated Sm proteins. Fig. S3 shows that SMN, PRMT5, and PRMT7 are required for TMG capping. Online supplemental material is available at .

Skp2 is the limiting component of the E3 ligase controlling the proteosomal degradation of p27 (p27) in late G1/S phase (; ). Skp2 can also stimulate p27 degradation and S phase entry in serum-deprived cells (). Although Skp2 has several other substrates (), the importance of p27 as a Skp2 substrate is emphasized by the finding that Skp2-p27 double-null mice lose most of the Skp2 knock out phenotype (). It was therefore surprising that knockin mouse embryo fibroblasts (MEFs) expressing p27, a Skp2-resistant p27 mutant, did not show a pronounced defect in mitogen-stimulated S phase entry (). Skp2 levels are inhibited posttranscriptionally by retinoblastoma protein (Rb) through its effects on anaphase-promoting complex/cyclosome and its activator Cdh1 (APC/C)–mediated Skp2 degradation (; ; ; ; ). In this report, we describe a parallel regulation of Skp2 by Rb that results in the formation of a transcriptionally based Skp2 autoinduction loop. Interference with this loop selectively affects the transition to mitogen-independent cell cycle progression, also called the restriction point. Transcript profiling indicates that E2F controls Skp2 gene expression (; ), and we found, in agreement with this data, that ectopic expression of human papilloma virus–E7 (E7), which inactivates pocket proteins and releases E2Fs, rescued Skp2 mRNA and protein expression in serum-starved MEFs (). We used zPicture (available at ) to identify conserved domains in the mouse versus the human, chimp, and dog Skp2 promoters, and rVista (available at ) to look for putative E2F binding sites within the conserved domains. This analysis () revealed an evolutionarily conserved E2F binding site in human, chimp, and dog Skp2 promoters that matches the E2F consensus (SCGSSAAA; ). The homologous mouse sequence (GCGCTAAA) differs from the consensus by one base (), but this same sequence acts as a functional E2F site in the E2F1 promoter (). The mouse sequence begins at position +114 relative to the transcription start site, between the transcription and translation start sites. It is the only E2F site we could identify in the mouse Skp2 promoter. have reported that the Skp2 promoter contains a binding site for the transcription factor GABP. Interestingly, GABP can cooperate with E2F1 to regulate the transcription of target genes (). We generated luciferase constructs with the wild-type mouse Skp2 promoter, including one with a mutation in the conserved E2F site (), and transiently transfected them into MEFs. Chromatin immunoprecipitation (ChIP) performed with amplicons specific for the transfected genes showed that E2F1 bound to the transfected wild-type mouse Skp2 promoter but not to the transfected promoter with the E2F site mutation (C). Moreover, mutation of the single E2F site completely blocked the activity of a Skp2 promoter–luciferase construct in serum-stimulated or E7-expressing MEFs (). Skp2 mRNA levels fluctuate during cell cycle progression (). We found that Skp2 mRNA and protein expression are low in G0, gradually increase in early G1 phase, and further increase ∼15 h after mitogenic stimulation (, E and F, respectively). This second, late G1/S phase induction of Skp2 coincided with the hyperphosphorylation of Rb (). Moreover, this late G1/S phase induction of Skp2 mRNA closely matched the time-dependent increase in cyclin E1 mRNA (12–18 h; ), a prototypic E2F1-regulated gene (). ChIP was then used to examine the time-dependent binding of endogenous E2F1 to the mouse Skp2 promoter (). Indeed, the binding of endogenous E2F1 to the conserved site on the endogenous Skp2 promoter increased in late G1/S phase. Thus, the mid-to-late G1 phase induction of Skp2 (12–18 h after mitogen stimulation) is regulated by E2F activity. Others () have reported that the human Skp2 promoter contains three E2F-like sequences, one of which is the human homologue of the mouse E2F site reported in this paper (). However, these investigators concluded that a distinct, nonconserved E2F site (TTGCGCGCG) accounted for E2F-stimulated luciferase activity of the human Skp2 promoter. Although we cannot exclude the possibility that the human promoter relies on this nonconserved E2F site, it is curious that a consensus E2F site (CGCGCAAA) did not contribute to E2F-stimulated luciferase activity or interact with E2F in electrophoretic mobility shift assays in . We also note that the amplicon used to show binding of E2F1 to this nonconserved site in the human promoter includes the conserved E2F site described in this report. Our identification of a conserved E2F site in the mouse Skp2 gene allowed us to assemble Rb-E2F, Skp2, p27, and cyclin E–Cdk2 into a self-amplifying loop (). In this loop, the stimulatory effect of E2F on Skp2 gene expression would feed back to sustain Rb inactivation, E2F release, and further induction of the Skp2 gene in late G1 phase. Thus, this model predicts that Skp2 should induce itself and that this autoinduction should be detected as increased mRNA. Moreover, the autoinduction of Skp2 should occur in serum-deprived cells, because the loop is self-amplifying and does not require the presence of mitogens. To test this model, we infected serum-free cultures of MEFs with an adenovirus encoding human Skp2 (Ad-hSkp2) and used a species-specific Skp2 primer probe set with quantitative real-time RT-PCR (QPCR) to detect the induction of mouse Skp2 mRNA. Ectopic expression of human Skp2 induced endogenous mouse Skp2 mRNA, as well as other known E2F target genes (cyclins E1 and A; and Fig. S1, A and B, available at ). Induction of mouse Skp2 mRNA and down-regulation of p27 by Ad-hSkp2 were seen when the serum-starved MEF were expressing ectopic Skp2 at near normal levels (compare FCS to 60 MOI Ad-hSkp2; Fig. S2) and when the infection was performed during or after the serum-starvation period (Fig. S3). Conversely, inhibition of the loop by RNAi-mediated knock down of Skp2 reduced serum-stimulated Skp2 promoter activation (). As expected, this Skp2 requirement was lost after pocket protein inactivation with E7 (). The loop shown in predicts that the autoinduction of Skp2 requires Cdk activity, and we indeed found that the induction of endogenous Skp2 mRNA seen in response to ectopic Ad-hSkp2 in serum-starved MEFs was blocked by the Cdk inhibitor roscovitine (). Conversely, expression of cyclin E induced Skp2 mRNA and protein in serum-deprived MEFs ( and Fig. S1 C). The autoinduction of Skp2 mRNA was efficiently inhibited in serum-deprived MEFs when transit through the loop was precluded by knockin of a Skp2-resistant p27 mutant, p27 (). A characteristic of positive feedback loops is that they yield “all-or-nothing” responses. Indeed, we observed stepwise increases in cyclin E1 gene induction (, top) and S phase entry (, middle) upon infection of serum- starved MEFs with increasing MOIs of Ad-hSkp2. The maximal responses occurred with near-normal levels of Skp2 protein and were clearly distinguishable from the gradual increase in Skp2 expression (, bottom). The induction of endogenous Skp2 mRNA and degradation of p27 also occurred in a stepwise fashion (Fig. S2, available at ). have reported that Skp2 degrades E2F1 in G2/M phase cells. However, Ad-hSkp2 did not decrease E2F1 levels in serum-deprived MEFs (, bottom), which is consistent with the fact that Skp2 stimulates S phase entry under these conditions. Collectively with , this result suggests that Skp2 may switch from a positive to a negative regulator of E2F activity as cells progress from G1/S to G2/M. Ad-hCyclin E also had no effect on E2F1 levels (unpublished data). The Skp2 autoinduction loop has the potential to regulate S phase entry because it can perpetuate the down-regulation of p27 and, thereby, the activation of cyclin E–Cdk2, phosphorylation of Rb and release of E2Fs. However, , using knockin of p27, reported that Skp2-mediated p27 degradation is not required for S phase entry in serum-stimulated MEFs. Because the Skp2 autoinduction loop can function in mitogen-deprived cells (), we reasoned that these results could be reconciled if Skp2-mediated p27 degradation was essential only for the transition to mitogen independence, also called the restriction point (). We therefore used p27 MEFs to interrupt the Skp2 autoinduction loop and look for consequences on the restriction point. To measure passage through the restriction point, serum-starved p27 or wild-type MEFs were stimulated with 10% FCS for selected times. The serum was removed, and the cells were incubated with serum-free medium and BrdU. S phase entry in the wild-type MEFs required mitogens for the first 10 h after serum stimulation and then quickly became mitogen independent (, WT). This rapid transition to mitogen independence was defective in primary p27 MEFs; these cells did not become mitogen independent until 16 h (, T187A). Importantly, this defect in restriction-point control was not caused by a general decrease in the rate of cycling because, as previously reported (), the kinetics of S phase entry were nearly identical when wild-type and p27 MEFs were continuously exposed to mitogens (). Thus, Skp2-dependent degradation of p27, and probably the Skp2 autoinduction loop, regulates progression through the restriction point. A positive feedback loop should accelerate the transition to mitogen independence, and we indeed find that the rate of progression through the restriction point is decreased when transit through the loop is blocked in p27 MEFs (). Previous papers have proposed that passage through the restriction point is regulated by a positive feedback loop comprised of cyclin E and Rb-E2F (; ). In this model, cyclin E activation of Cdk2 would stimulate Rb phosphorylation and E2F-dependent transcription of the cyclin E1 gene, thereby furthering Rb inactivation and cyclin E1 induction. However, we reasoned that the repeated autoinduction of cyclin E might also require Skp2-mediated p27 degradation to allow for the activation of cyclin E–Cdk2. Based on this reasoning, the cyclin E autoinduction loop would be within, rather than separate from, the Skp2 autoinduction loop. To test this notion, we used a species-specific QPCR primer-probe set to mouse cyclin E1 mRNA to compare the efficiency of cyclin E autoinduction in wild-type and p27 MEFs infected with an adenovirus encoding human cyclin E. Indeed, we found that the autoinduction of cyclin E1 mRNA was barely detected in p27 MEFs (). This result strongly suggests that the major effects of cyclin E and Rb/E2F in restriction point control depend on their inclusion in the Skp2 autoinduction loop. Others have reported that APC/C stimulates Skp2 protein degradation (; ), and that APC/C activity is inhibited by the E2F-dependent induction of Emi1 (). We therefore envision that the inactivation of pocket proteins and release of E2F controls the restriction point through the coordinated effects of the transcriptionally based Skp2 autoinduction loop described in this report and the posttranscriptionally based APC/C pathway. Both of these effects would converge to increase the expression of Skp2 and degradation of p27. We note that p27 MEFs are not completely restrictionless, indicating that other Skp2 targets may also contribute to restriction point control. Alternatively, an independent positive feedback loop, perhaps in which Rb-E2F induces cyclin D1 (), may cooperate with the loop described in this report. In addition to its effect on cell cycle progression, the Skp2 autoinduction loop may contribute to cell cycle exit associated with pocket protein activation. Others have reported that Rb and p107 regulate p27 levels posttranscriptionally by acting as a scaffold for Skp2 and Cdh1 and thereby facilitating APC/C-dependent Skp2 proteolysis (; ; ). Interestingly, this rapid posttranscription down-regulation of Skp2 should inhibit the Skp2 autoinduction loop, which would in turn prevent Skp2 gene transcription and thereby enforce the quiescent state. Thus, coordinated transcriptional and posttranscriptional pocket protein effects on Skp2 levels may contribute to both the transition to mitogen independence and the G1 phase arrest that follows mitogen withdrawal. Skp2 knock down or p27 overexpression inhibit S phase entry in serum-stimulated cells (; ; ), whereas S phase entry is nearly normal in p27 MEFs (; ) cultured under similar conditions. These results imply that p27 MEFs (which have gone through mouse development in the absence of wild-type p27) may have acquired a compensatory mechanism that bypasses the need for Skp2-mediated p27 degradation in mitogen-bathed cells. In contrast, the restriction point defect is clearly seen in p27 MEFs, emphasizing that the role of Skp2-mediated p27 degradation in the transition to mitogen independence is essential. The results in show that autoinduction of Skp2 is linked to efficient progression through the restriction point. However, this loop might become dispensable if Rb is inactivated by oncogenes during cellular transformation. To explore this possibility, we inactivated pocket proteins by ectopic expression of E7, knocked down Skp2 with siRNA, and determined the consequence of reduced Skp2 expression on S phase entry in serum-deprived MEFs (conditions where Skp2- mediated p27 degradation is required for the feedback loop and S phase entry; – ). Interestingly, S phase entry was not inhibited by knock down of Skp2 (), and a distinct Skp2 siRNA gave similar results (not depicted). We also found that Skp2 siRNA inhibited S phase entry in serum-stimulated MEFs, but even this Skp2 requirement was lost upon expression of E7 (). Elevated Skp2 expression is observed in cancers and has been considered a causative factor due, in part, to its effects on p27 (; ; ). However, our data indicate that high Skp2 expression may simply be a consequence of aberrant Rb inactivation and E2F release. Because many cancers constitutively activate upstream mitogenic signaling pathways that sustain Rb inactivation, the ability of Skp2 to sustain Rb inactivation through the feedback loop shown in may not be needed in cancer cells. Indeed, our results indicate that S phase entry induced by E7 is independent of Skp2. In fact, the notion that Skp2-mediated degradation of p27 may not be required for tumor development is supported by the finding that the p27 mutation does not delay tumorigenesis in mouse models of Ras-dependent lung and colon cancer (). Spontaneously immortalized MEFs were maintained in DME (Invitrogen) with 5% FBS. For experimentation, near confluent cells were serum starved in DME with 1 mg/ml heat-inactivated, fatty acid–free BSA (DME-BSA; Sigma-Aldrich) for 48 h, trypsinized, reseeded at subconfluence, and stimulated largely as described previously (). Stimulated cells were washed once with cold PBS, scraped, collected by centrifugation, quick frozen, and stored at −80°C before analysis. S phase entry was determined by incorporation of BrdU (). Primary MEFs from wild-type and p27 knockin mice were maintained in 10% FCS and used at passages 2–5. To study progression through the restriction point, serum-starved wild-type and p27 MEFs were replated at subconfluence in 35-mm dishes and stimulated with 10% FCS for selected times. The FCS-containing medium was removed, and the cells were washed once with a 50-mM glycine, 150-mM NaCl, pH 2.8, acid wash buffer and twice with cold DME and were then incubated in DME-BSA for a total incubation of 17 h. BrdU was added to all of the plates, and the incubation continued for an additional 12 h before determining BrdU incorporation (relative to total DAPI-stained nuclei) by immunofluorescence microscopy. BrdU results show mean ± SD from multiple fields of view. MEFs (∼60% confluent) in a six-well plate were transiently cotransfected as described previously (), but using 1 μg Skp2-pGL3, 0.1 ng cytomegalovirus– luciferase, 4 μl Lipofectamine (Invitrogen), and 6 μl Plus reagent (Invitrogen) per well. For experiments using E7, the cells were cotransfected with 0.5 μg of the firefly luciferase vector driven by wild-type (S) or E2F-mutated (S) Skp2 promoter, 0.1 ng CMV– luciferase, and either 0.5 μg of E7 plasmid or empty vector. After a 24-h incubation in 10% FCS, the cells were collected in passive lysis buffer, and luciferase activity was determined using the dual-luciferase reporter assay system (Promega). Measurements were performed in duplicate and recorded as mean ± SD. Skp2 luciferase activity was normalized to luciferase activity. Unless noted otherwise in the figure legends, confluent MEFs were infected with adenoviruses after a 12-h incubation in serum-free DME-BSA. The cells were infected overnight at 100 MOI using adenoviruses encoding GFP, Ad-LacZ, Ad-E7 (provided by J. Meinkoth, University of Pennsylvania, Philadelphia, PA), human cyclin E1 (Ad–hCyclin E; provided by J. Albrecht, University of Minnesota, Minneapolis, MN, and S. Reed, University of California, San Diego, San Diego, CA), or Ad-hSkp2 (provided by K. Nakayama, Kyusu University, Fukuoka City, Fukuoka, Japan), and then incubated in fresh serum-free medium to obtain a total serum-free medium incubation time of 48 h. For plasmid transfections, MEFs were transiently transfected as described previously () using 5 μg pCDNA3.1 (vector control), pCDNA3.1-based E7, or pcDNA3.1-based human E2F1. Transfected cells were incubated overnight in DME containing 10% FCS before use or serum starvation for 24–36 h. Transfections of siRNAs were performed as described previously (), except that 100 nM irrelevant (human E cadherin; GAGUGAAUUUUGAAGAUUGtt) or mouse Skp2 (UUUGUCACUCCCUUUGCCCtt) siRNAs were used. When adenoviral infection was combined with siRNA, near confluent MEFs were serum starved for 12 h, infected with either Ad-E7 or Ad-LacZ, and incubated for 24 h in DME-BSA. The medium was removed, and the infected cells were transfected with the irrelevant control siRNA or Skp2 siRNA. After an additional 24 h, the siRNA-containing medium was replaced with DME-BSA or 10% FCS DME with BrdU, and the incubation was continued for another 24 h. After a total of 84 h in serum-free medium, coverslips were collected for analysis of BrdU incorporation. In some experiments, the siRNA transfection also contained 0.05 μg of the wild-type Skp2 promoter–luciferase and 0.05 ng of luciferase vectors. Collected cell pellets were lysed in 0.5–1 ml of TRIzol (Invitrogen) to extract total RNA. Real-time PCR for mouse Skp2 and Cdk4 were performed as previously described (). Controls (unpublished data) demonstrated that the mouse Skp2 primer probe set did not detect human Skp2 mRNA. Mouse cyclin E1 mRNA, mouse cyclin A mRNA, and 18S rRNA levels were determined using assay-on-demand primer probe sets Mm00432367_ml, Mm00438064_ml, and Hs99999901_s1 (Applied Biosystems), respectively. Skp2 and cyclin E1 mRNAs were normalized to Cdk4 mRNA or 18S rRNA, neither of which varied reproducibly in response to any of the treatments used. Duplicate PCR reactions were run for each sample, and results are plotted as mean ± SD. Results shown in the figures are typically representative of three independent experiments. ChIPs were performed as described previously () using 10 MEFs per sample and 5 μg of either anti-E2F1 (C-20X; Santa Cruz Biotechnology, Inc.) or preimmune antibody control. One tenth of the final immunoprecipitated DNA (5 μl) was analyzed by QPCR with SYBR green to quantify the amount of immunoprecipitated Skp2 promoter. Primer sequences for mouse Skp2 were 5′-TGGTGATGGAACGTTGCTAGT-3′ (forward) and 5′-GGTGTCCACTGATTCAGGA-3′ ( reverse). ChIPs on MEFs transiently transfected with Skp2 promoter–luciferase constructs were performed as previously described () and analyzed by PCR using 5′-TGGTGATGGAACGTTGCTAGT-3′ (forward) and 5′-CTTTATGTTTTTGGCGTCTTCCA-3′ (reverse; encoding plasmid backbone sequence within the promoter–luciferase construct). The amplified PCR product (300 bp) was detected on a 1.5% agarose gel. Western blotting was performed as described previously () using 30–40 μg of total cellular protein and the following antibodies: Skp2 (SKP2-2B12; Invitrogen), Cdk4 (C-22 [Santa Cruz Biotechnology, Inc.] or DCS-31 [Invitrogen]), p27 (clone 57; BD Biosciences), Rb (Mab1; Invitrogen), E2F1 (C-20; Santa Cruz Biotechnology, Inc.) cyclin E (M-20; Santa Cruz Biotechnology, Inc.), and actin (1616R and C-2; Santa Cruz Biotechnology, Inc.). The resolved proteins were detected using ECL (GE Healthcare). Autoradiographs were digitized by scanning, and figures were assembled using Photoshop (Adobe). Fig. S1 complements the mRNA analysis in Figs. 2 B and 3 B to show that infection with Ad-hSkp2 or Ad–hCyclin E leads to protein expression of the E2F targets, cyclin A, cyclin E, and Skp2. Fig. S2 shows that near endogenous levels of Skp2 can initiate the Skp2 autoinduction loop. Fig. S3 shows that Skp2 expression can initiate the Skp2 autoinduction loop even when cells are fully quiescent. Online supplemental material is available at .

Mitochondrial fusion plays an important role in controlling the shape and function of mitochondria (; ). In mammalian cells, the dynamin-related GTPase OPA1 is essential for mitochondrial fusion (; ). The yeast OPA1 orthologue Mgm1 is also essential for fusion (, ; ) and has been shown to form oligomers important for tethering and fusion of the inner membranes (). OPA1 is associated with the inner membrane and protects cells from apoptosis by regulating inner membrane dynamics (; ). Mutation of OPA1 causes the disease dominant optic atrophy, a degeneration of the retinal ganglion cells (; ). Both OPA1 and Mgm1 undergo proteolytic processing, and, in the case of Mgm1, such processing has been shown to be essential for mitochondrial fusion activity (; ; ; ). Yeast Mgm1 is produced as a precursor with a mitochondrial leader sequence that is cleaved by the mitochondrial processing peptidase (MPP). Mgm1 processed by only MPP leads to the long isoform, l-Mgm1. The mitochondrial rhomboid Pcp1/Rbd1 further cleaves a subset of Mgm1 to form the short isoform, s-Mgm1 (; ; ). Loss of Pcp1 greatly reduces mitochondrial fusion because a mixture of both the long and short Mgm1 isoforms is essential for normal activity (). Several issues concerning OPA1 processing remain enigmatic. In contrast to the two isoforms produced by Mgm1, OPA1 produces many more isoforms. OPA1 is encoded by a complicated set of at least eight mRNA splice forms that are produced by differential splicing (). In addition to the MPP processing site, the polypeptides encoded by each mRNA splice form contain an S1 cleavage site, and some also contain a more C-terminal S2 cleavage site (; ). In principle, therefore, each mRNA splice form can produce a long isoform (produced by cleavage with MPP alone) and one or more short isoforms (produced by cleavage at S1 or S2). The proteases acting at sites S1 and S2 are poorly understood. There is evidence for the involvement of both the rhomboid protease presenilin-associated rhomboid-like (PARL) and the m-AAA protease paraplegin in OPA1 processing (; ). However, cells lacking PARL or paraplegin have normal OPA1 processing (), suggesting that other proteases remain to be identified. In addition, when reconstituted in yeast cells, OPA1 cleavage appears to depend on two other m-AAA proteases, Afg3L1 and Afg3L2 (). In contrast to Mgm1, it has been suggested that the long form of OPA1 isoform 1 is the fusion-active species (). As a result, it is unclear whether proteolytic processing is simply a way to inactive OPA1 or whether the short isoforms have another function. To clarify these issues, we developed a cellular system to study the role of specific OPA1 isoforms in mitochondrial fusion. We find that a combination of long and short OPA1 isoforms is important for mitochondrial fusion activity. Moreover, we find that the i-AAA protease Yme1L regulates OPA1 processing. As a result of eight mRNA splice forms () and subsequent proteolytic processing, OPA1 isoforms migrate as a complex mixture of at least five bands (a–e) on gel electrophoresis (). The bands a and b are thought to be a mixture of long isoforms of OPA1, whereas the shorter bands (c–e) are thought to result from additional proteolytic processing (). The complexity of this mixture prevents the definitive assignment of function to specific mRNA splice forms and their processed polypeptides. To circumvent these complications, we analyzed the mitochondrial fusion activity of the eight known human mRNA splice forms upon expression in OPA1-null mouse embryonic fibroblasts (MEFs; ). The parental OPA1-null cells contain no OPA1 protein and have completely fragmented mitochondria as a result of the lack of mitochondria fusion (unpublished data; ). Of the eight OPA1 mRNA splice forms, the expression of isoforms 1, 2, 4, and 7 result in the robust tubulation of mitochondria in OPA1-null cells (). In contrast, splice forms 3, 5, 6, and 8 have modest or barely detectable tubulation activity. The ability of OPA1 splice forms to tubulate mitochondria in OPA1-null cells correlates well with their mitochondrial fusion activity in a polyethylene glycol (PEG)–induced cell hybrid assay ( and Fig. S1, available at ). Interestingly, all of the mRNA splice forms with high levels of fusion activity produce a long form of OPA1 in addition to one or more further processed short forms (). All mRNA splice forms encode a protease processing site in exon 5 (S1 site), and some contain a second site (S2) in the alternative exon 5b (). mRNA splice forms 1 and 2 encode only site S1 and yield a single long and a single short form. mRNA splice forms 4 and 7 encode sites S1 and S2 and therefore yield two short forms in addition to a single long form. These results illustrate the complexity of analyzing endogenous OPA1 bands from wild-type cells. The top two bands (a and b) are generally thought to be the two long forms of OPA1, but our analysis indicates that there are at least four species of long forms. In contrast, all of the mRNA isoforms with little fusion activity are processed with greater efficiency to yield only short forms of OPA1 (). These results suggest that the short isoforms produced by cleavage at S1 or S2 have little activity on their own. Interestingly, all of the mRNA splice forms producing highly processed polypeptides (3, 5, 6, and 8) contain exon 4b (). Polypeptides produced from mRNA splice forms 1 and 2 are partially processed, whereas those from splice forms 5 and 3, which are identical, respectively, except for the 4b insertion, are fully processed. Similarly, polypeptides from splice forms 6 and 8 are identical to those from splice forms 4 and 7, respectively, except for the 4b insertion. These results indicate that peptide sequences encoded by exon 4b stimulate proteolytic processing. In future experiments, it will be interesting to determine whether exon 4b–containing transcripts are preferentially expressed under specific cellular conditions. In yeast, production of the short isoform of Mgm1 requires adequate ATP levels in the mitochondrial matrix, suggesting that Mgm1 processing is regulated by cellular metabolism (). It should be noted that in our expression system, the total amount of exogenously expressed OPA1 is comparable with the total endogenous OPA1 in wild-type MEFs (). In contrast, gross overexpression of exogenous OPA1 is unable to rescue mitochondrial fusion activity in OPA1-null cells (unpublished data). Previous experiments suggested that the long form of isoform 1 is the fusion-active species (). This conclusion was based on analysis of an isoform 1 construct in which the S1 cleavage site was removed by the deletion of 10 residues surrounding alanine 195 (ΔS1 mutation). To reexamine this issue, we analyzed the function of polypeptides containing the ΔS1 mutation. As expected, OPA1 1ΔS1 and 2ΔS1 produce only the long form (). Splice forms 1 and 2 are the only ones lacking both the S2 cleavage site and exon 4b, thereby simplifying the analysis of processing. Whereas wild-type splice forms 1 and 2 are able to induce tubulation in >50% of OPA1-null cells, the same splice forms lacking the S1 site, especially OPA1-1ΔS1, have barely detectable activity (). Because the long forms of isoforms 1 and 2 are ineffective for mitochondrial fusion, these results indicate that proteolytic processing is important for OPA1 function. The aforementioned experiments indicate that neither the long forms nor the short forms in isolation are sufficient for mitochondrial fusion. Therefore, we tested whether a combination of long and short forms might be necessary. We analyzed whether 1ΔS1, which produces only a single long form with essentially no fusion activity (), could be complemented by the additional expression of splice forms 3, 5, 6, or 8, which produce fully processed isoforms that also have little fusion activity (). As expected, coexpression produces a mixture of long and short forms (). All four combinations of coexpression resulted in a synergistic increase in mitochondrial tubulation in OPA1-null cells (). These results demonstrate that 1ΔS1 and short forms work together to promote fusion. To test this idea further, we analyzed ΔS1 versions of splice forms 3, 5, 6, and 8. Because of the 4b exon, the polypeptides produced from these splice forms are normally fully processed to yield only short forms that have little fusion activity. We hoped that by inhibiting cleavage at site S1, we could produce a mixture of long and short forms. In each case, the ΔS1 versions yield only a modest amount of the long form because OPA1 is still mostly processed (). In the case of 3ΔS1 and 5ΔS1, processing is caused by increased cleavage at sites other than S1. In the case of 6ΔS1 and 8ΔS1, cleavage still occurs efficiently at S2 based on the size of the processed bands. Remarkably, in each case, the ΔS1 mutants contain more mitochondrial fusion activity than the wild-type proteins (). These effects were most dramatic for 3ΔS1 and 5ΔS1. These results again indicate that a combination of long and short OPA1 isoforms is needed for efficient fusion activity. Disruption of mitochondrial membrane potential by carbonyl cyanide -chlorophenyl hydrazone (CCCP) has been shown to enhance the processing of OPA1, leading to accumulation of the short isoforms (; ). Consistent with this idea, we found that CCCP treatment of OPA1-null cells expressing splice form 1 or 2 leads to accumulation of the S1-cleaved short isoform (). However, this effect appears to involve not only increased processing at S1 but also induced degradation of the long isoform. Upon treatment with CCCP, cells expressing 1ΔS1 show a complete loss of the long isoform (). Under these conditions, a minor fraction of OPA1 is cleaved at a site other than S1, and this minor species is the only OPA1 that remains. With 2ΔS1, most of the long isoform is also degraded. It should be noted that we observed this degradation only with a 4-h CCCP treatment, which is longer than in previous studies (; ). Because CCCP treatment leads to the accumulation of short isoforms, we tested directly whether short isoforms are stable in the presence of CCCP. Expression of the exon 4b–containing splice forms (3, 5, 6, and 8) leads to fully processed isoforms that are stable after CCCP treatment (). Given the enhanced cleavage at site S1 in the presence of CCCP, we also examined its effect on cleavage at S2. Variants 4ΔS1 and 7ΔS1 produce only a single short isoform as a result of cleavage at site S2 (). Upon CCCP treatment, the long isoform disappears, but without an increased accumulation of the S2-cleaved isoform. In addition, there is a small accumulation of a novel cleavage product. Collectively, these results indicate that the loss of membrane potential enhances OPA1 processing at S1 but not S2, suggesting that these sites are under differential regulation. In addition, the loss of membrane potential leads to degradation of the long isoforms but not the short ones. The different response of S1 versus S2 cleavage to the loss of membrane potential suggests that the two sites may be regulated by different proteases. Previous studies have implicated both m-AAA proteases and the rhomboid protease PARL in OPA1 processing (; ; ). However, the depletion of these proteases had only limited effects on OPA1 processing (), indicating that additional proteases might be involved. Yme1L is an i-AAA protease anchored in the inner membrane with its protease domain located within the intermembrane space (; ). Short hairpin RNA (shRNA) against Yme1L effectively reduces protein levels (), and we examined its effect on OPA1 processing. Knockdown of Yme1L has no effect on isoform 1 processing, which occurs exclusively at site S1 (). In contrast, for isoforms 4 and 7, which contain both sites S1 and S2, Yme1L shRNA reduces the level of the S2-cleaved product and increases the level of the S1-cleaved product. By densitometry of the Western blots, the ratio of the S1-cleaved band to the S2-cleaved band was 0.4 and 0.6 for isoforms 4 and 7, respectively, without treatment. With the Yme1L knockdown, this ratio rose to >3. Likewise, 7ΔS1, which produces only the S2-cleaved product, shows reduced processing after shRNA treatment. In this case, the ratio of the long isoform to the S2-cleaved isoform rose from 1 to 3. Similar effects on the S1 and S2 cleavage products were observed with the exon 4b–containing splice forms 6 and 8 (). These results indicate that Yme1L is important for proteolytic processing at site S2. Isoform 5 has no S2 cleavage site, but shRNA resulted in the slight accumulation of a long isoform. It may be that Yme1L is also involved in cleavage at sites other than S2. As with paraplegin and PARL, it is not clear whether Yme1L directly cleaves OPA1 or affects another protease. Thus far, we have not been able to demonstrate a direct physical interaction between Yme1L and OPA1 by coimmunoprecipitation. Cleavage of site S2 by YmeL1 is topologically sensible because the S2 region would be expected to reside in the intermembrane space, where the protease domain of Yme1L is located. In contrast, the protease activity of PARL resides within the inner membrane, and the protease activities of m-AAA proteases reside within the matrix. If Yme1L does directly cleave OPA1, it would be the first example of a protein-processing function through site-specific cleavage for Yme1L. Previously, the role of Yme1L was thought to be in degrading misfolded mitochondrial proteins in the intermembrane space (). Analogously, the m-AAA proteases have a role in protein quality control through protein degradation but have recently been found to also process proteins at specific sites (; ; ). MEFs were maintained in DME containing 10% heat-inactivated FBS, 1 mM -glutamine, and penicillin/streptomycin. For transfection of cells with expression constructs, LipofectAMINE 2000 (Invitrogen) was used according to the manufacturer's specifications. In the experiments in , CCCP was used at 20 μM, and cells were analyzed at 4 h. To test the fusion activity of OPA1 mRNA splice forms, OPA1- null cells stably expressing mitochondrially targeted GFP or DsRed were infected with retrovirus encoding each OPA1 mRNA splice form. Pairwise PEG fusion assays were performed as described previously (). The eight isoforms of human OPA1 cDNA were amplified by PCR from first-strand cDNA and subcloned into the retroviral vector pMSCV-puro. Retrovirus production and infection were performed as described previously (). Western blots of cell lysates were probed with an affinity-purified anti-OPA1 antibody raised against the C-terminal region of human OPA1 (gift of L. Griparic and A. van der Bliek, University of California, Los Angeles, Los Angeles, CA; ). shRNAi against Yme1L was performed using a modified retroviral vector with the H1 promoter to drive the expression of shRNAs (). Three shRNAi constructs were tested, with the best results obtained with the vector constructed with the oligonucleotides 5′-GATCCCCGTGGCAGAGGAATTCATATTTCAAGAGAATATGAGTTCCTCTGCCACTTTTTGGAAA-3′ and 5′-AGCTTTTCCAAAAAGTGGCAGAGGAACTCATATTCTCTTGAAATATGAATTCCTCTGCCACGGG-3′. MEFs were grown on poly--lysine–treated coverslips and fixed in 10% formalin. Coverslips were mounted with GelMount (Biomeda). Images were acquired at room temperature with a camera (ORCA-ER; Hamamatsu) attached to a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) controlled by Axiovision version 4.5 software (Carl Zeiss MicroImaging, Inc.). A 100× plan Apochromat NA 1.4 objective was used. Mitochondria were identified by mitochondrially targeted GFP or DsRed. Fig. S1 shows that OPA1-null cells expressing a single OPA1 RNA splice form have extensive fusion activity in the PEG cell hybrid assay. Online supplemental material is available at .

Dominant optic atrophy is a progressive eye disease resulting from degeneration of the retinal ganglion cell layer with ascending atrophy of the optic nerve. The disease is most often caused by mutations in the OPA1 gene, which encodes a human homologue of yeast Mgm1 (). Mgm1 and Opa1 are members of the dynamin family of GTP-binding proteins (). The mitochondria of yeast Mgm1 mutants are fragmented because of a defect in mitochondrial fusion (; ). Transfection of mammalian cells with Opa1 siRNA or with dominant interfering constructs also causes mitochondria to fragment. In addition, the loss of Opa1 causes cell death, suggesting an antiapoptotic function that may contribute to dominant optic atrophy (; ; ). Yeast Mgm1 and mammalian Opa1 have N-terminal mitochondrial targeting signals that are cleaved by the mitochondrial matrix processing protease. Yeast Mgm1 and mammalian Opa1 reside in the mitochondrial intermembrane space, where they are tightly bound to or embedded in the inner membrane (). A substantial fraction of yeast Mgm1 is further processed by a rhomboid protease called Pcp1, which is required for normal Mgm1 function (). Yeast Pcp1 mutants can be rescued by the heterologous expression of a mammalian Pcp1 homologue called presenilin-associated rhomboid-like (PARL; ). However, only 4% of Opa1 is processed by PARL (; ), whereas a much larger fraction of Opa1 is cleaved by a protease that is induced with the loss of mitochondrial membrane potential or apoptosis (; ). Moreover, PARL knockout cells have normal mitochondrial morphologies (), whereas cells with an induced proteolysis of Opa1 have fragmented mitochondria (), suggesting different roles for PARL in mammals and yeast. Experiments with the heterologous expression of mammalian Opa1 in yeast suggest that a mitochondrial matrix protease called paraplegin can process Opa1, but paraplegin siRNA has only modest effects on Opa1 in mammalian cells (). The N terminus of mammalian Opa1 also undergoes extensive alternative slicing near the point of cleavage (). Cleavage could conceivably affect the attachment of Opa1 to the membrane because Opa1 is released from mitochondria coincident with the cleavage that occurs during apoptosis (). The mechanism of release and its functional consequences are not yet known. In this study, we confirm the release of Opa1 during apoptosis and show that this coincides with the proteolytic cleavage of Opa1. Transfection with PARL siRNA and PARL knockout did not affect Opa1 maturation nor did it affect induced proteolysis, suggesting that other proteases are responsible for these cleavages. Investigation of known mitochondrial proteases by transfection with siRNA shows that the mitochondrial intermembrane space protease Yme1 affects the constitutive processing of Opa1. Yme1 also affects mitochondrial morphology, but this property is independent of Opa1. Mature Opa1 can be resolved into five bands ranging in size from 85 to 100 kD on Western blots (). The bands, which are named a through e, correspond to different isoforms (). Treatment with apoptosis-inducing agents causes the larger isoforms to disappear from mitochondrial fractions, whereas shorter isoforms appear in the cytosol (unpublished data). This shift is not affected by cycloheximide, indicating that Opa1 undergoes inducible proteolysis and release from mitochondria rather than altered protein synthesis. Release of Opa1 occurs in the presence of the caspase inhibitor zVAD-fmk like release of cytochrome but unlike that of apoptosis-inducing factor, which is inhibited by zVAD-fmk (Table S1, available at ). Opa1 proteolysis also occurs within 12 min after adding carbonyl cyanide -chlorophenyl hydrazone (CCCP), which induces the loss of mitochondrial membrane potential. The mitochondria of CCCP-treated cells convert to an abundance of tiny fragments similar to the fragmentation that occurs during apoptosis (). However, CCCP does not cause the release of Opa1 or cytochrome into the cytosol nor does it trigger apoptosis even after 5.5 h of incubation, as shown by the lack of caspase 9 cleavage (). CCCP-induced cleavage is also not enough to release Opa1 from the inner membrane because a substantial fraction of the cleaved protein remains membrane bound, as determined with carbonate extraction (unpublished data). Removal of CCCP allows for the gradual recovery of wild-type mitochondrial morphology and of the larger Opa1 isoforms (Fig. S1). This recovery is blocked by cycloheximide, showing that it requires the de novo synthesis of Opa1. To find out which proteases affect Opa1, we conducted knockdown experiments with siRNA directed against known mitochondrial proteases. The rhomboid protease PARL was considered a likely candidate because it is very similar to yeast Pcp1. PARL siRNA was highly effective (). However, transfection with PARL siRNA had no effect on Opa1 maturation or inducible proteolysis. Because a small amount of residual PARL could still account for inducible cleavage, we also tested a PARL knockout mouse embryonic fibroblast cell line (). We find that PARL cells already show a substantial amount of Opa1 proteolysis even without CCCP. This cleavage may reflect the increased susceptibility to apoptosis that was previously observed with PARL mice () and by us with PARL siRNA (unpublished data). However, this cell line is still capable of inducible cleavage (). The existence of inducible cleavage in these cells was confirmed by transfection with selected Opa1 isoforms and treatment with CCCP. Overexpression of Opa1 isoform 7 in PARL knockout cells showed some changes in the intensities of bands c and d (, bottom), but these intensities varied between experiments, suggesting that PARL is not required for constitutive cleavage nor is it required for inducible cleavage. Another candidate is HtrA2/Omi, which faces the mitochondrial intermembrane space (). HtrA2/Omi siRNA was highly effective (>99.4% reduction as judged by densitometry of the Western blot), but this did not noticeably affect Opa1 proteolysis (). Lastly, we tested mitochondrial AAA proteases. Mammals have three of these, namely Yme1, paraplegin, and AFG3L2 (). The catalytic domain of Yme1 faces the mitochondrial intermembrane space, whereas those of paraplegin and AFG3L2 face the mitochondrial matrix. Transfection with paraplegin and AFG3L2 siRNA did not affect Opa1 proteolysis, but knockdown was incomplete (10 and 34% residual RNA), so paraplegin or AFG3L2 might still mediate inducible cleavage. In contrast, Yme1 siRNA did affect Opa1 even without CCCP. Western blots show shifts in banding patterns in treated and untreated cells (). CCCP still induces proteolysis in Yme1 siRNA–transfected cells, whereas not all bands are equally affected by Yme1 siRNA and CCCP (), suggesting that the cleavage patterns are governed by alternative splicing. We conclude that Yme1 affects the constitutive proteolysis of a subset of Opa1 isoforms but is not required for inducible proteolysis. Opa1 has extensive alternative splicing near the sites of cleavage (). To determine which splice variants are affected by inducible proteolysis and which are affected by Yme1-mediated proteolysis, we transfected cells with representative expression constructs (isoform 1 has no extra exon, isoform 5 has 4b but not 5b, and isoform 7 has 5b but not 4b; ). All three isoforms are relatively abundant in HeLa cells (; ). In addition, we transfected cells with expression constructs in which the major inducible cleavage site S1, which is contained by exon 5, was deleted (). The transfected cells were treated with or without CCCP to determine their sensitivity to inducible proteolysis and were cotransfected with or without Yme1 siRNA to determine their sensitivity to constitutive proteolysis. Our results confirm that isoform 1 is cleaved by CCCP-inducible proteolysis (). Mutations in S1, the cleavage site that was previously shown to exist in exon 5, render this isoform insensitive to inducible cleavage (). However, this isoform is not affected by Yme1 siRNA. In contrast, isoform 5 and isoform 7, which contain exon 4b and exon 5b, respectively, are affected by Yme1 siRNA. In both cases, there is a substantial shift to the uncleaved band. Residual cleavage to the lower bands in uninduced cells and further cleavage upon induction with CCCP can be attributed to the inducible protease. Isoform 5, which contains exon 4b, is fully cleaved even without CCCP induction (). This cleavage is reduced but not fully eliminated by deleting S1, and it is reduced but not fully eliminated by transfection with Yme1 siRNA. However, the uncleaved protein that accumulates under these conditions can be cleaved by inducible proteolysis even when S1 is deleted. We conclude that isoform 5 has an alternative cleavage site similar to the alternative cleavage site in isoform 7. Sizing by SDS-PAGE suggests that this site is in exon 4b (unpublished data), which is analogous to the alternative cleavage site S2 in exon 5b of isoform 7. Isoform 7 yields three bands (uncleaved protein and two cleavage products), which is consistent with previous results (). The longer cleavage product reflects a cleavage site (S1) in exon 5, and the shorter product reflects a cleavage site (S2) in exon 5b. The deletion of S1 abolishes the longer cleavage product, and it causes an accumulation of uncleaved protein. This uncleaved protein is poorly cleaved upon treatment with CCCP, suggesting that the S2 site is less efficiently cleaved by the inducible protease. Similar results were obtained with Yme1 siRNA. The uncleaved protein that accumulates with Yme1 siRNA is preferentially cleaved at S1 upon induction with CCCP (, bottom; lanes 5 and 6). To confirm that exon 4b– and 5b-containing isoforms are affected by Yme1, we transfected cells with siRNA oligonucleotides that are specific for exons 4b and 5b as described previously (). Exon 4b and 5b siRNA both reduce the intensity of band d (Fig. S2 A, available at ). Exon 5b siRNA also reduces the intensity of bands a and c, although these are already faint in wild-type cells. Transfection with Yme1 siRNA produces a pattern that appears to be the sum of exon 4b and 5b siRNA alone. We conclude that the shorter products in uninduced cells consist of isoforms with exon 4b or 5b. These isoforms require Yme1 for constitutive cleavage, whereas the other isoforms remain susceptible to inducible cleavage. To our surprise, the mitochondria of cells transfected with Yme1 siRNA are unusually connected (). 2 d after transfection, 40% of cells had more connected mitochondria, 56% appeared normal, and 4% had fragmented mitochondria ( = 400), whereas in untransfected cells, 100% of the mitochondria had normal morphologies ( = 100). Similar morphological defects were observed with two other Yme1 siRNA oligonucleotides (Fig. S2, B–D), confirming that these effects were specific for the knockdown of Yme1. The increased connectivity is strikingly similar to the connectivity observed with Drp1 siRNA (). At later times, a substantial fraction of cells undergo cell death (Fig. S2 E), and the mitochondria fragment (3 d after transfection, 23% of cells had connected mitochondria and 45% were normal, whereas 31% had fragmented mitochondria; = 300). These fragments are more irregular than the mitochondrial fragments in Opa1 siRNA cells (). Similar mitochondrial disintegration is observed with strong Drp1 mutants at later time points (). To determine whether the increased connectivity of Yme1 siRNA cells is caused by an increase in fusion or a decrease in division, we induced mitochondrial fragmentation by incubating transfected cells with CCCP and allowing them to recover (similar to the cycloheximide experiment; Fig. S1). We find that transfection with Yme1 siRNA does not prevent CCCP-induced fragmentation nor does it prevent recovery after washout (). Quantitation of these results suggests that there is some delay in recovery (), but the fact that recovery can occur at all suggests that Yme1 is not necessary for mitochondrial fusion. The observation that CCCP can still induce mitochondrial fragmentation in Yme1 siRNA cells unlike cells transfected with dominant-negative mutants of Drp1 () suggests that mitochondrial fission can still occur as well. The morphological abnormalities that we observe in Yme1 siRNA cells must therefore result from a subtle shift in the balance between mitochondrial fission and fusion rather than from strong effects on either process. Comparable morphological abnormalities were also observed with Opa1 exon 4b– and 5b-specific siRNA (). Transfection with these oligonucleotides increases mitochondrial connectivity (), but there are small differences between the morphologies observed with exon 4b and 5b siRNA. Exon 4b siRNA gives rise to thin and more connected mitochondria, whereas exon 5b gives rise to connected mitochondria with localized swellings, similar to Yme1 siRNA. In addition, Opa1 exon 4b– and 5b-specific siRNA increase the number of apoptotic cells (unpublished data), which is consistent with previously published data (). To determine whether Opa1 is required for the increased connectivity observed with Yme1 siRNA, we cotransfected Yme1 siRNA and pan-Opa1 siRNA. Western blotting showed an 82% reduction of Opa1 levels, which would ordinarily affect mitochondrial morphology (), and a 99% reduction of Yme1 levels (see Fig. S3 A for Western blots; available at ). To our surprise, almost all of the cells that were cotransfected with Opa1 and Yme1 siRNA had more connected mitochondria (), similar to Yme1 siRNA alone, whereas none had fragmented mitochondria ( = 200). In contrast, 99% of cells that were transfected with Opa1 siRNA alone had fragmented mitochondria, as seen previously (). These results suggest that Yme1 is epistatic to Opa1. We used electron microscopy to examine the effects of Yme1 siRNA on the internal structure of mitochondria (Fig. S3, C–F). For comparison, cells were also transfected with Drp1 siRNA and with Yme1 and Opa1 siRNA together. None of these transfections had drastic effects on cristae morphology. Yme1 siRNA did cause some cristae to appear disorganized, but the results were not strong nor were they readily quantified. There were also no obvious defects in Yme1/Opa1 double siRNA cells. Opa1 siRNA can affect cristae morphology (), but this might have been suppressed by the increased connectivity caused by Yme1 siRNA. Similar suppression of cristae defects was observed in a yeast double mutant of Mgm1 and Dnm1 (homologues of Opa1 and Drp1; ). We conclude that the internal structures of mitochondria are not notably disrupted by Yme1 siRNA. Although better known for its role in the proteolytic degradation of misfolded proteins (), Yme1 has been shown to assist the import and maturation of specific intermembrane space proteins (). However, cells transfected with Yme1 siRNA or with exon 4b or 5b siRNA do not show an obvious fusion defect. Instead, these transfections lead to novel mitochondrial morphologies that are more similar to the morphologies in preapoptotic cells, which is consistent with the previously proposed antiapoptotic function of exons 4b and 5b (). It is not yet clear how Yme1 siRNA affects mitochondrial morphology. Yme1 could directly influence mitochondrial fission or fusion, or it could act on a different protein with a dominant effect on mitochondrial morphology. However, increased numbers of connections were also observed in cells treated with ceramide, suggesting that this might be a prelude to apoptosis (). Yme1 siRNA might then preempt the loss of fusion caused by Opa1 siRNA in cotransfected cells by increasing the rate of mitochondrial fusion or decreasing the rate of fission at an earlier time point. These effects are not absolute, however, because they are overridden by treatment with CCCP. A similar dichotomy was observed with BH3 proteins such as Bax and Bak, which direct mitochondrial fission during apoptosis but are required for mitochondrial fusion in normal cells (). The accompanying paper by Song et al. (see p. of this issue) and previous work with yeast () show that mitochondrial fusion requires both cleaved and uncleaved forms of Opa1. show that Yme1 can mediate this cleavage, but our data indicate that fusion also occurs without Yme1. The requirement for cleavage, which is shown by , might instead be met by the inducible protease. This possibility is intriguing because the inducible protease and mitochondrial fusion are both regulated by the mitochondrial inner membrane potential (). We conclude that there are three types of Opa1 cleavage: Yme1-dependent processing of Opa1 isoforms with antiapoptotic functions; cleavage of a fraction of Opa1 protein to meet the requirement for fusion; this type of cleavage can be mediated by Yme1, but it might also be mediated by the inducible protease; and inducible cleavage of the remaining Opa1 protein after the loss of membrane potential as it occurs during apoptosis. This cleavage probably inactivates Opa1 because it coincides with mitochondrial fragmentation. HeLa cells were cultured as described previously (). Apoptosis was induced with 2 μM staurosporine (Sigma-Aldrich) or 10 μM actinomycin D (Calbiochem). Where indicated, 100 μM zVAD-fmk (Qbiogene) was added 30 min before adding actinomycin D. Mitochondrial membrane potential was disrupted with 10 μM CCCP (Sigma-Aldrich), and, where indicated, 20 μg/ml cycloheximide (Calbiochem) was added. For the washout experiments, cells were rinsed twice with PBS at 37°C and incubated with fresh culture medium. Controls were treated with solvent (DMSO). Cells were prepared for immunofluorescence as described previously (). Opa1 antibody was described previously (). Apoptosis-inducing factor antibody was obtained from Santa Cruz Biotechnology, Inc., and cytochrome antibody was purchased from BD Biosciences. Secondary antibodies were purchased from Jackson ImmunoResearch Laboratories and Biomeda. B. De Strooper (Katholieke Universiteit Leuven, Leuven, Belgium) provided PARL knockout mouse embryonic fibroblast cells, which were grown as described previously (). Fluorescence microscopy was performed with a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) using an α plan-Fluar 100× NA 1.45 oil objective (Carl Zeiss MicroImaging, Inc.). Images were acquired with a CCD camera (ORCA ER; Hamamatsu) and Axiovision software (Carl Zeiss MicroImaging, Inc.). Image files were processed with Photoshop software (Adobe). To isolate mitochondria for the Yme1 blots, cells were pelleted by centrifugation for 5 min at 600 and washed with mitochondrial isolation buffer (). Cell membranes were disrupted by sonication. Nuclei and unbroken cells were removed by centrifugation for 5 min at 1,000 . Mitochondria were pelleted by further centrifugation for 30 min at 10,000 . Pellets were resuspended in laemmli sample buffer, size fractionated by SDS-PAGE, and blotted to nitrocellulose. The expression construct containing wild-type Opa1 isoform 1 was described previously (). A C-terminal myc tag was added to this construct by PCR. Isoform 7 was generated by amplifying exon 5b from HeLa cDNA and cloning this fragment with BclI and BstBI sites in the myc-tagged isoform 1 construct. Isoform 5 was similarly generated by amplifying exon 4b and surrounding sequences from HeLa cDNA and cloning this fragment with BclI and BstBI sites in the myc-tagged isoform 1 construct. Deletions in the S1 cleavage site were made with fusion PCR using AGGTTCTCCGGAAGAATCTGAAAGTGACAAGCATTTTAG and AATGCTTGTCACTTTCAGATTCTTCCGGAGAACCTGAGGT primers for site-directed mutagenesis. The mutated fragments were cloned into the isoform 1, 5, and 7 myc-tagged constructs with BclI and BstBI sites. New clones were sequenced to ensure fidelity. Fig. S1 shows the CCCP-induced cleavage of Opa1 and recovery after washout (A) and that mitochondria also recover their filamentous morphology, but not with cycloheximide (B–E). Fig. S2 shows the effects of Yme1 siRNA on Opa1 and compares this with the effects of exon 4b and 5b siRNA (A). Fig. S2 also shows the morphological effects of two additional Yme1 siRNA oligonucleotides (B–D) and shows the effects of Yme1 siRNA on apoptosis (E). Fig. S3 shows Opa1 and Yme1 in single and double siRNA transfections by Western blotting and electron microscopy. Table S1 presents data on the percentage of cells releasing Opa1, cytochrome , or apoptosis-inducing factor from mitochondria 7 or 10 h after inducing apoptosis with actinomycin D. Online supplemental material is available at .

Cytokinesis in animal and primitive eukaryotic cells is executed by contraction of the contractile ring formed underneath the plasma membrane at the division site (; ; ), which is composed mainly of actin filaments (F-actins) and myosin-II. It has been shown by decoration with heavy meromyosin or myosin S1 that the contractile ring F-actin consists of two populations with opposite directionalities, respectively (; ), which supports the idea that the contractile ring contracts by sliding of F-actins over each other via myosin filaments (; ). How myosin and actin assemble into the ring has frequently been studied with the fission yeast because many mutant strains that show defects in ring formation have been obtained (; ). cells are cylindrical, and grow during interphase by elongation at cell ends where F-actin forms patch structures () and longitudinal F-actin cables originate (; ; ; ). These F-actin structures are considered to function in polarized growth of the cell (). During early mitosis, the novel aster-like structure of F-actin cables is formed near duplicated spindle pole bodies through reorganization of the interphase F-actin structures. From the aster the leading F-actin cables that encircle the cell at the equator elongate, which have been considered to represent the primary contractile ring, and the contractile ring is established during anaphase from these structures (). Cytokinesis progresses by constriction of the ring followed by septum formation (; ). Participation of myosin-II (; ; ), the formin Cdc12 (), and the actin-depolymerizing factor Adf1 () is requisite for assembly of the contractile ring. This suggests that polymerization of actin may be a crucial step in assembly of the ring because all of these proteins from this or other organisms can induce or accelerate actin polymerization in vitro (; ; ), and are localized at the division site at very early stage of mitosis (; ; ; ). However, it has not been known how these proteins actually function in the course of the ring assembly including the timing and precise site of function. The main reason for this is that all of the localization studies of these and other relevant proteins have so far been performed with fluorescence microscopy. Ultrastructural analyses of the process of ring assembly are now required in order to elucidate spatial organization of the assembly at a molecular level. Here, we investigated arrangements of F-actin in the ring by electron microscopy in order to understand basic structure of the ring and how actin is assembled into the ring structure. We used both wild-type cells and mutant () cells synchronized at M phase. Cell wall materials were enzymatically digested and the cells were permeabilized with Triton X-100. Myosin S1 was added to the cells to decorate F-actin and the cells were processed for examination by transmission electron microscopy. It has been confirmed that the structure of actin cytoskeleton in these cells is preserved through this procedure (; Fig. S1, available at ). Both the wild-type cells and the cells at M phase showed a bundle of microfilaments at the division site often associated with ingressions of plasma membrane in longitudinal grazing sections. S1 decoration to form arrowhead structures showed that these filaments were composed of F-actin (; Fig. S2, available at ). In , the F-actins whose pointed ends faced the top asterisk in are shown in red, whereas those showing the opposite directionality are shown in blue. It is apparent that the ring was composed of F-actins of opposite orientations from each other. A longitudinal F-actin cable attached to the ring is seen in . Such cables have been found by fluorescence microscopy both before and after establishment of the ring (; ). The S1 decoration showed that these cables consisted of F-actin of opposite orientations. Another feature of these cables was that they were fragmented at the end close to the ring and then seemed to be incorporated into the ring structure. The F-actin ring of a cell sectioned parallel to the division plane around the equator was observed as loose bundles of F-actin lying underneath the cell membrane and encircling the cytoplasm (). We examined the entire structure of six contractile rings by serial sectioning of the rings; these rings were confined in 14–33 serial sections (55 nm/section; ). Stages of the rings in the process of cytokinesis were determined as follows. Dividing nuclei were seen adjacent to edges of rings #1 and #2, but they were not seen in the vicinity of rings #3 to #6 (unpublished data). Therefore, it was considered that rings #1 and #2 were in cells at early anaphase B. Diameters of the median circles of rings #1 to #4 were similar and they were 2.3–2.5 μm, while those of rings #5 and #6 were significantly smaller (). Thus, rings #1 to #4 were not yet contracting, while rings #5 and #6 were contracting when the cells were permeabilized. This suggests that the cells possessing rings #3 and #4 were at late anaphase B. The total number of F-actin in the rings was between 1,100 and 2,100 (). F-actins whose pointed ends were clockwise on the photographs are shown in red, and those whose pointed ends were counterclockwise are shown in blue. F-actins of unclear directionality or those oriented perpendicular to the plasma membrane are shown in yellow (). The ratio of clockwise-oriented F-actins to counterclockwise-oriented ones in the contractile ring was roughly 1:1 for all the rings analyzed (). The average length of randomly extracted 100 F-actins in rings #1 to #4 was 0.6 μm, while that in rings #5 and #6 was 0.45 μm (; Fig. S3 A, available at ). This suggests that F-actins in the ring shortened as it contracted. It was also found that F-actins having opposite directionalities were not mixed homogeneously in the ring, but those of a same directionality were seen to form clusters (). In addition, branching of F-actin was hardly seen in both longitudinal grazing sections (4 rings) and equatorial sections (rings #1 to #6). Next, we performed a three-dimensional reconstruction of the contractile rings in which directionality of each F-actin is indicated (; Fig. S3, B and C; and Videos 1 and 2, available at ). It is remarkable in both of the early stage rings #1 and #2 that F-actins with the same directionality occupied about a half of the ring on the whole, as if the F-actin ring was composed of two semicircles of uniform, but opposite polarity. On the other hand, F-actins in the rings of later stages seemed to be mixed homogeneously on the whole (; Fig. S3, B and C; and Videos 1 and 2). To quantitatively show this feature, we divided all serial images of the 6 contractile rings into 12 equal segments, respectively, counted F-actins in each segment, calculated percentages of F-actins of the major two directionalities, and then obtained absolute value of difference in the percentages for every segment (; Fig. S3 C). For rings #1 and #2, the differences were minimum in the bottom (segment 6) and the top (segment 1) segments, indicating that numbers of F-actins of opposite directionalities were close in these segments, while the differences were maximum in the right and left segments, indicating that F-actins of same directionalities were dominant in these regions. Furthermore, the differences were larger in many of the segments of rings #1 and #2 than in the segments of rings #3 to #6 (; Fig. S3 C). This suggests that overlapping of the filaments of opposite directionalities progress from early anaphase B to late anaphase B/cytokinesis. The fact that the fission yeast contractile ring is composed of antiparallel actin filaments of 1:1 number ratio strongly suggests that this structure is contractile, providing that myosin filaments are present in the ring. Though we did not see myosin filament-like structures in the ring in thin sections, it has been known that the myosin-II heavy chain Myo2 is colocalized to the ring, and is necessary in establishment of the ring (; ; ). The present finding is important because no evidence for actual contraction of the ring in fission yeast cell has been obtained, although it shrinks during cytokinesis. We estimated the average length and the number of actin filaments in the contractile ring to be around 0.6 μm and 1,100–2,100, respectively. Taking the filament number as 2,000, total actin monomers comprising one ring would be 444,000. The actin concentration in the cytoplasm including all the organelles in the wild-type fission yeast cells has recently been estimated to be 8.7 μM (), which is significantly smaller than that reported previously (31.3 [in a minimal medium] or 63.2 [in a rich medium] μM; ). Supposing that actin concentration in the cell is same as that in the wild-type cell, and the length and diameter of the cytoplasmic compartment of a mitotic cell are 28.6 and 2.84 μm, respectively (; this paper), number of actin molecules in the mitotic cell is roughly 920,000. Therefore, ∼50% of the actin molecules are used to form the contractile ring. This agrees with the observation by fluorescence microscopy that a majority of actin filaments in the mitotic cell seems to form the contractile ring (). We found that the actin filaments in the contracting ring were shorter than those in the ring before contraction. It has been considered that the filaments must be disassembled during contraction because the contractile ring retains its width, depth from the plasma membrane, and density of filaments during contraction and finally disappears when contraction is complete (). Our observation demonstrated for the first time that this is the case. The shortening of the filaments is likely to be due to depolymerization of actin () especially induced by Adf1, which is localized to the ring throughout the course of cytokinesis and necessary for disassembly of the ring (). The overall structure of the F-actin contractile ring revealed here leads us to predict important features of formation and establishment of the contractile ring in fission yeast cells. Polymerization of actin seems to be a crucial step in the ring assembly. It has been believed that actin polymerizes in the cell preferably from the barbed end (). Therefore, the fact that the ring at early anaphase B seemed to roughly comprise two semicircles of unidirectional F-actins, the directionalities of which were opposite from each other, suggests that formation of the ring contains a step where polymerization of actin may take place mainly at one site in the medial cortex of the cell and elongation occurs in both directions in the equatorial plane (). A strong candidate for such a site is the center of the aster-like structure of F-actin cables, which is formed near spindle pole bodies during prophase and from which the leading cable(s) elongate (). An intriguing possibility is that the formin Cdc12 may localize to the center of this structure. This formin is necessary for elongation of the leading cable(s) from this region (). It functions in vitro to promote nucleation of actin polymerization at barbed ends (). It is localized to a single spot in the mid-region of the cell during early mitosis (; ; Yonetani, A., and F. Chang, personal communication). Furthermore, it is localized at the center of an aster-like structure formed in mutant cells at a restrictive temperature (). gene encodes the myosin-II essential light chain () and its mutation affects myosin-II functions, one of which is likely to accomplish ring formation together with F-actins after formation of the aster-like structure (). Therefore, ring formation would not proceed beyond the aster-like structure in the mutant cells. In addition, presence of clusters of F-actin of a same directionality in the contractile ring may also reflect a process of formation of the ring. It has been reported that the F-actin cables accumulate as rays of the aster-like structure are somehow fused with the leading cable to form the contractile ring during anaphase (). The ring thus may contain remnants of the cables, each of which probably consists of unidirectional F-actins. It was also shown that in late anaphase B contractile ring, which was not contracting yet, F-actins of opposite directionalities were mixed homogeneously on the whole. Therefore, there must be mechanism(s) by which F-actins are rearranged in the ring during anaphase before contraction initiates. There may be two possible mechanisms for this rearrangement: First, sites for polymerization of actin may be scattered in the division plane at the later stage of contractile ring formation, although polymerization may occur mainly at one site at the beginning as mentioned above. This idea seems to be supported by localization of Cdc12. Although Cdc12 mainly localizes to a single spot during early mitosis, it then spreads in the division plane (; Yonetani, A., and F. Chang, personal communication). Thus, this protein would be able to polymerize actin at various spots in the division plane during anaphase. Actually, we have observed that Cdc12 forms a ring containing 10–20 strong spots during anaphase (unpublished data). It has recently been published that Cdc12 is scattered as a broad band of nodes at the division site during early mitosis (). We do not know what the reasons for the discrepancy are, but this observation might have been done after its spreading. Myosin-II (Myo2) that accumulates as multiple spots () at the division site through binding to Mid1 () is also capable of organizing actin because mutant Myo2s, which prematurely accumulate at the presumptive division site even during G2 phase, induce F-actin accumulation at the same site during G2 (). These proteins could produce antiparallel F-actins in each semicircle during anaphase. If this type of rearrangement takes place, actin in the original semicircles would be turning over. It has been suggested that the contractile ring actin turns over even after establishment of the ring in eggs (), and demonstrated in cultured cells (). A FRAP study on GFP-Cdc8 (tropomyosin) also suggests that actin turns over in the contractile ring in fission yeast cells (). The fact that the actin-depolymerizing factor, Adf1, which promotes actin turnover, is necessary for maintenance of the contractile ring structure in addition to its formation in fission yeast (), supports the idea of turnover of actin in the ring. Second, F-actins could slide over each other being mediated by Myo2 without accompanying contraction of the ring, and therefore the mixed directionality is produced during late anaphase B. It may also be responsible for redistribution of Cdc12 nucleation sites in the division plane. This sliding without contraction would occur if the F-actins are only weakly or not attached to the cell membrane. Unfortunately, it has not been known when the contractile ring F-actin firmly attaches to the cell membrane during formation of the ring. Actin dynamics and actin–membrane interaction in the contractile ring must be investigated in order to clarify mechanism of contractile ring formation. Mitoses in fission yeast cells and animal cells are different in that the nuclear membrane persists and the mitotic spindle is formed inside the nucleus in fission yeast, while the mitotic apparatus is formed after breakdown of the nuclear membrane in animal cells, which has a symmetric microtubular structure. In fission yeast, a signal for contractile ring assembly is thought to be transferred from the nucleus to the cell cortex () through the spindle pole bodies (); a strong candidate for a division signal molecule, Plo1, which is a fission yeast orthologue of the Polo kinase that is first localized to the spindle pole bodies and then transferred to the mid-ring structure during early mitosis (). This agrees well with the present observation that the contractile ring actin assembly initiates at one site in fission yeast. On the other hand, microtubules emanating from the separated centrosomes converge at the equatorial cortex in anaphase animal cells, and are thought to transmit cleavage signals to the cortex. Thus, contractile ring formation is expected to initiate simultaneously at multiple sites in the mid-cortex. It has recently been shown that cleavage signaling molecules are localized to tips of astral microtubules at the equatorial cortex in HeLa cells (). The signal may be transmitted to the cortex through these molecules at multiple sites, each of which may correspond to the single site in fission yeast cells, and basically similar assembly and rearrangement of actin may occur at these sites. Yeast strains and growth conditions were described previously (). temperature-sensitive mutant cells were synchronized at M phase. Preparation of cells and electron microscopy were performed as described previously (). In brief, cells were treated for 30 min with 0.5 mg/ml Zymolyase 100T (Seikagaku Corp.) at 30°C to produce spheroplasts, then with 0.5% Triton X-100 dissolved in 20 mM Pipes, 20 mM MgCl, 10 mM EGTA, pH 7.0, protease inhibitors, and 1 M sorbitol for permeabilization in the presence of 4.2 μM phalloidin (Sigma-Aldrich) for 1 min, and incubated at 4°C overnight with 1 mg/ml myosin S1. They were fixed with 2% glutaraldehyde (Electron Microscopy Sciences) and 0.2% tannic acid for 1 h at 4°C, postfixed in 2% OsO for 1 h at 4°C, and embedded in Quetol812 (Nisshin EM). Three-dimensional reconstructions of the F-actin contractile rings decorated with S1 were performed using serially sectioned images. Micrographs originally taken at 25,000× were enlarged to 100,000×. After determining the directionality of each F-actin in the ring, images of the F-actin were traced from the individual electron micrographs. The tracings of F-actin were aligned and superimposed to produce reconstructed images using the TRI/3D SRF II program (Ratoc System Engineering) and Windows NT. Staining of fixed cells with Bodipy-phallacidin and permeabilization of unfixed cells in the presence of rhodamine-phalloidin were performed as described previously (; ). Both of these cells were finally washed with a solution of 20 mM Pipes, 20 mM MgCl, 10 mM EGTA, pH 7.0, and protease inhibitors, and observed in the same solution with an Axioskop fluorescence microscope (Carl Zeiss MicroImaging, Inc.) equipped with a Plan Apochromat 63×/1.40 oil objective lens and appropriate filters. Images were acquired using a CCD camera (SPOT; Diagnostic Instruments). Fig. S1 shows fluorescent microscopic evidence that no artificial F-actin structure was formed after permeabilization followed by S1 decoration in wild-type and cells. Fig. S2 shows an electron microscopic image of a contractile ring of an S1-decorated wild-type cell. Fig. S3 shows that the ring F-actin became short in contracting contractile ring (A), three-dimensional reconstructed images of the F-actin contractile rings #2, #4, and #6 in cells (B), and degree of unidirectionality of F-actins in segments of these rings (C). This figure suggests that overlapping of F-actins of opposite directionalities progresses during cytokinesis. Video 1 shows that F-actins with the same directionality occupied about a half of the ring on the whole in the early stage ring #1. Video 2 shows that F-actins of opposite directionalities are mixed on the whole in the later stage ring #5. Online supplemental material is available at .

The challenge of mitosis is to faithfully transmit chromosomes in each cell division. Errors in this process cause aneuploidy (abnormal numbers of chromosomes), which is frequently found in cancers and is believed to promote the growth and progression of diseases (). Accurate chromosome segregation requires proper mitotic spindle formation and successful chromosome movement along the spindle. Chromosomes capture spindle microtubules through a dynamic search and capture mechanism by their kinetochores. Numerous proteins, including motors and nonmotor proteins, have been implicated in stable kinetochore microtubule attachment, although mechanistically, the role for the majority of these proteins has yet to be identified (; ; ). A group of microtubule-associated proteins has emerged to participate in kinetochore microtubule attachment. Adenomatous polyposis coli (APC) and its binding partner EB1 () localize to kinetochores during mitosis in a microtubule-dependent manner (; ). Defects in spindle formation and chromosome missegregation have been shown in mammalian cultured cells harboring a colon cancer–related dominant APC mutant (; ). APC and EB1 depletion has also been reported to give rise to similar defects in chromosome congression without arresting cells in mitosis based on a fixed cell characterization (; ). Using a combination of RNAi-mediated protein depletion and live cell imaging, it has been shown that chromosomes in APC- or EB1-depleted cells do congress to the spindle equator, although APC-depleted cells exhibited transient mitotic checkpoint–dependent anaphase delay (). The depletion of APC from cytostatic factor (CSF)–arrested egg extracts has been shown to cause a decrease in spindle microtubule density (), although it is not well understood how bipolar spindles are formed in noncycled egg extracts with sperm nuclei (haploid). Spindles formed in cycled egg extracts were not sensitive to APC depletion (Dikovskaya et al., 2004). APC associates with the microtubule-destabilizing protein mitotic centromere-associated kinesin (MCAK) in egg extracts (). Codepletion of APC and MCAK from cycled egg extracts results in large, dense microtubule structures surrounding the chromosomes (). Furthermore, APC has been shown to form a complex with mitotic checkpoint proteins Bub1 and Bub3 and is a substrate for Bub1/BubR1 kinases in vitro (); however, the physiological role of the APC–Bub1/BubR1 interaction is not known. The mitotic checkpoint (spindle assembly checkpoint) inhibits premature anaphase onset until every chromosome has successfully attached to microtubules. Mitotic kinases BubR1 (; ; ) and Bub1 () are essential for the metazoan mitotic checkpoint. Recent studies have suggested that BubR1 () and Bub1 (; ) are also necessary for the stabilization of kinetochore microtubule capture; however, it is not known whether their actions are direct or not and whether the kinase activity is required for this. Taking advantage of egg extracts that are naturally arrested in metaphase of meiosis II by CSF, we report here that the complex formation between BubR1 and two microtubule plus end–interacting proteins, APC and EB1, is essential for chromosome alignment at the metaphase plate, providing a potential link in stable kinetochore microtubule attachment. To investigate the role of BubR1 in kinetochore microtubule attachment and metaphase chromosome alignment, we first perturbed BubR1 function in egg extracts by the addition of a specific BubR1 antibody, which inhibits BubR1 kinase activity and the mitotic checkpoint (). egg extracts were cycled through interphase to generate duplicated sister chromatids and were arrested at the subsequent M phase upon the addition of an aliquot of egg extracts containing both active Cdk1 and CSF. Although the addition of a control IgY antibody to egg extracts yielded predominantly bipolar spindles with chromosomes aligned at the metaphase plate (), the addition of an anti-BubR1 antibody produced normal bipolar spindles with misaligned chromosomes (). Prolonged incubation failed to produce bipolar spindles with properly aligned chromosomes. Similar results were obtained in three independent experiments. Addition of the BubR1 antibody does not prevent the kinetochore binding of BubR1 or its binding partner centromere protein E (CENP-E; ). This clearly indicates that BubR1 and its kinase activity are essential for chromosome alignment. To further test the requirement for BubR1 and its kinase activity in chromosome alignment, BubR1 was quantitatively immunodepleted from egg extracts (). Mock-depleted egg extracts produced bipolar spindles with chromosomes congressed at the metaphase plate (); BubR1-depleted egg extracts generated bipolar spindles with severe chromosome alignment defects (). Purified recombinant wild-type (WT) BubR1 or a kinase-inactive point mutant (kinase-dead [KD] BubR1; ) efficiently bound to kinetochores when added to depleted egg extracts (). The addition of recombinant kinase-competent BubR1 () but not the KD-BubR1 mutant () to a level comparable with that of endogenous BubR1 completely restored metaphase chromosome alignment. Although CENP-E kinetochore association is BubR1 dependent in egg extracts, kinetochore-bound KD-BubR1 is able to stimulate the binding of a normal level of CENP-E (). As an additional test of stable kinetochore microtubule attachment, we treated egg extracts with a low dose of nocodazole (100 nM), which causes a preferential disassembly of astral microtubules (). In control egg extracts, the mitotic spindles remained intact 10 min after nocodazole treatment, although the spindle length was reduced (). In BubR1-depleted egg extracts, however, only a few microtubules were observed (). Altogether, these findings support a model in which BubR1 and its kinase activity are essential for positioning chromosomes at the metaphase plate in egg extracts. To investigate potential binding partners of BubR1 that could account for its function in chromosome alignment, we performed coimmunoprecipitation experiments with a specific anti-BubR1 antibody in egg extracts with the addition of sperm nuclei. In BubR1 immunoprecipitates, besides the known binding partner CENP-E (, CENP-E in lane 4), immunoblotting analysis also identified two microtubule-associated proteins, APC and EB1, as binding partners of BubR1 (, APC and EB1 in lane 4). To verify the interaction between APC/EB1 and BubR1, APC was quantitatively immunoprecipitated from egg extracts (, APC in lane 4) with a specific APC antibody (). EB1 was detected in the immunoprecipitates (, EB1 in lane 4) as expected. Furthermore, a portion of endogenous BubR1 was also found in the immunoprecipitates (, BubR1 in lane 4), suggesting that BubR1 and APC/EB1 could exist in a complex. Also, a portion of BubR1 was cofractionated with APC and EB1 after sucrose density gradient centrifugation (), supporting the conclusion that these proteins form a physical complex. To address whether APC works together with BubR1 to regulate kinetochore microtubule capture, we first examined their localization in cycled egg extracts. egg extracts were cycled through interphase to allow DNA and kinetochore replication and then were cycled back into mitosis. As shown in , APC was detected along spindle microtubules and colocalized with BubR1 at attached kinetochores. If the BubR1–APC/EB1 interaction is required for stable kinetochore microtubule attachment, the complex formation of BubR1–APC/EB1 should be enhanced when microtubules are attached to kinetochores. To test this possibility, we conducted APC immunoprecipitation analysis in CSF-arrested egg extracts with or without added sperm nuclei (kinetochores) and nocodazole (to disassemble microtubules; as indicated in ). Immunoprecipitates obtained with an APC antibody were found to contain comparable levels of APC and EB1 under all conditions (, APC and EB1 in lanes 1–3), suggesting that the mitotic APC/EB1 interaction is independent of kinetochores and spindle microtubules. In contrast, there was considerably greater amounts of BubR1 (∼50% increase) in APC precipitates from egg extracts containing spindle structures and attached kinetochores (, BubR1 in lane 2) than that from control CSF-arrested egg extracts (, BubR1 in lane 1) or egg extracts with unattached kinetochores (in the presence of nocodazole; , BubR1 in lane 3). This strongly suggests that complex formation between BubR1 and APC/EB1 is enhanced in the presence of spindle microtubules and attached kinetochores. We have demonstrated that BubR1 and its kinase activity are necessary for metaphase chromosome alignment (). Because BubR1 cannot directly bind to microtubules, it is more likely that BubR1 regulates the activity of a microtubule-binding protein. Thus, we examined the effects of APC and EB1 immunodepletion on metaphase chromosome alignment in egg extracts. The depletion of either APC or EB1 with specific antibodies () produced bipolar spindles with misaligned chromosomes (). Again, extended incubation failed to produce bipolar spindles with properly aligned chromosomes. To confirm that the chromosome alignment defect is caused by the loss of APC or EB1 rather than other associated proteins, we reconstituted APC or EB1 in APC- or EB1-depleted egg extracts with purified recombinant APC () or EB1 () proteins to approximately the normal level, and proper metaphase chromosome alignment was restored (). On the basis of this apparently complete reconstitution, we believe that the disruption of chromosome alignment in APC- or EB1-depleted egg extracts truly reflects APC or EB1 depletion and not the depletion of some associated partners. Codepletion of APC and EB1 from egg extracts produced similar chromosome alignment defects as APC or EB1 single depletion (). Furthermore, a majority of spindle microtubules in APC- or EB1-depleted egg extracts was not stable in the presence of a low dose of nocodazole (). These results demonstrate that the microtubule-associated proteins APC and EB1 are necessary for metaphase chromosome alignment. Because BubR1 is an essential mitotic checkpoint kinase, we decided to examine whether its binding partners APC and EB1 are also necessary for mitotic checkpoint signaling in egg extracts. Sperm nuclei and the microtubule assembly inhibitor nocodazole were added to APC- or EB1-depleted egg extracts to assemble condensed mitotic chromosomes and produce unattached kinetochores. In mock-depleted egg extracts, this yielded an activated mitotic checkpoint as judged by the absence of nuclear envelope reassembly, a continuous high level of Cdc2 kinase activity after the calcium-mediated inactivation of CSF (, blot b), and kinetochore recruitment of BubR1 and Mad2 (). After BubR1 depletion, no BubR1 and Mad2 could be detected at unattached kinetochores (), and the mitotic checkpoint was not activated (, blot c) as expected (; ). The depletion of APC or EB1 neither prevented the kinetochore association of BubR1 and Mad2 () nor abrogated the mitotic checkpoint response in the presence of nocodazole (, blots d and g). Furthermore, even after spindle assembly in the absence of nocodazole, APC- or EB1-depleted egg extracts produced chronic mitotic arrest, as revealed by continued Cdc2 kinase activity and condensed chromosomes (, blots e and h). On the other hand, APC- or EB1-depleted egg extracts supplemented with recombinant APC or EB1, respectively, showed no checkpoint arrest, inactivating Cdc2 kinase and reassembling nuclei around decondensing chromosomes within 60 min (, blots f and i). BubR1 kinase activity is essential for the mitotic checkpoint (, ; ). After the depletion of endogenous BubR1 and the restoration of normal BubR1 levels with purified GST-tagged BubR1 (, ), BubR1 kinase activity was measured by affinity recovery of BubR1 with GST antibody beads from CSF-arrested egg extracts. BubR1 kinase was activated only when microtubule assembly was inhibited (, lane 2) and was silenced, as expected, after spindle assembly and chromosome alignment (, lane 1). However, in APC- or EB1-depleted egg extracts, BubR1 kinase activity was undiminished even after spindle assembly (, lanes 4 and 6). Altogether, these results are consistent with the conclusion that APC and EB1 are necessary for proper kinetochore microtubule attachment. In mammalian cultured cells, it has been shown that the BubR1 RNAi phenotype is partially rescued by an Aurora kinase inhibitor: many chromosomes aligned at the metaphase plate, with a few chromosomes remaining near the poles (), suggesting that BubR1 and Aurora B might regulate opposite activities. To directly test this hypothesis, we coimmunodepleted APC and Aurora B from egg extracts. In APC and Aurora B codepleted extracts, only ∼50% of mitotic figures were bipolar spindles (), which is consistent with an earlier report that the chromosomal passenger complex is required for bipolar spindle formation in egg extracts (); however, the number of bipolar spindles with aligned chromosomes increased (). Codepletion of BubR1 and Aurora B has shown similar results (). In contrast, codepletion of BubR1 and APC (and EB1) resulted in similar chromosome alignment defects as BubR1 or APC depletion alone (). Furthermore, the spindles formed in Aurora B–depleted egg extracts were shorter, probably as a result of the increased activity of MCAK because Aurora B can phosphorylate MCAK and reduce its microtubule-depolymerizing activity in vitro (; ; ). Indeed, egg extracts coimmunodepleted of APC and MCAK produced large microtubule structures surrounding the chromosomes instead of bipolar spindles (), which is consistent with an earlier study using an APC antibody depleted of both APC and MCAK from egg extracts (). It has been shown that an N-terminal fragment (APC or N-APC) similar to the mutant protein expressed in many colorectal cancers dominantly compromises microtubule attachment in mitosis in mammalian cultured cells, resulting in errors in chromosome segregation (). We have expressed and purified the corresponding truncated form of APC from (, lane 2) and have tested its effect on metaphase chromosome alignment. In contrast to the GST control, mitotic spindles that formed in the presence of GST–N-APC displayed severe chromosome alignment defects (), which is a phenotype similar to APC depletion (). This result provides further evidence that the chromosome alignment defect observed upon APC depletion is specific. Immunoprecipitation analysis revealed that the N-APC fragment indeed oligomerized with endogenous APC proteins (, lane 1) as expected (; ). However, APC and EB1 interaction was decreased upon the addition of N-APC to egg extracts (; compare EB1 in lanes 1 and 3), as has been shown in mammalian cells (), although purified APC without the N terminus can efficiently bind to EB1 in vitro (). Furthermore, the addition of N-APC in egg extracts almost completely eliminated BubR1 interaction with endogenous APC (; compare BubR1 in lanes 1 and 3). Our results suggest that APC/EB1 interaction is essential for BubR1–APC/EB1 complex formation at attached kinetochores. One possibility is that EB1 may enhance the interaction of microtubules and APC at plus ends of microtubules. To test this possibility, a purified recombinant C-terminal EB1 fragment EB1C (), which cannot bind to microtubules, was added to egg extracts. The EB1C still contains a conserved domain that is known to interact with APC (). EB1C was indeed coimmunoprecipitated with APC (, EB1C in lane 5). However, this fragment substantially decreased APC association with endogenous EB1 (, compare EB1 in lanes 3 and 5) and BubR1 (, compare BubR1 in lanes 3 and 5). Furthermore, metaphase chromosome alignment was disrupted upon the addition of EB1C (). These results suggest that the complex formation of BubR1–APC/EB1 is important for its function in chromosome alignment. We have shown that BubR1 can form a complex with APC and EB1 in egg extracts (). To determine whether the BubR1–APC/EB1 association is direct, full-length recombinant BubR1, APC, and EB1 proteins were purified from and/or insect Sf9 cells (using the Baculovirus system). Recombinant C-terminal APC (with the known EB1-binding domain; ) effectively pulled down full-length recombinant protein EB1 (; compare lane 2 with lane 5), whereas BubR1 protein did not (, lane 3), demonstrating that BubR1 cannot directly bind to EB1. To test whether BubR1 can directly bind to APC, purified GST-BubR1 was incubated with His-APC. As shown in , a considerable amount of APC was precipitated with GST-BubR1 but not GST alone (compare lane 3 with lane 4). This demonstrates that BubR1 can directly bind to APC. We next determined whether APC is a substrate for BubR1. An in vitro kinase assay was performed using purified APC, BubR1, and CENP-E (, lanes 7–9), the activator of BubR1 kinase (, ). This revealed that BubR1 phosphorylated APC efficiently (, lane 5). Furthermore, APC was a much better substrate for BubR1 compared with CENP-E and BubR1 itself (, compare lane 4 with lane 5). This was not the result of a contaminating kinase because in a parallel assay, BubR1 containing a K787R point mutation in the ATP-binding pocket was much less active in APC phosphorylation (, lane 6). This is consistent with a previous study that showed that BubR1/Bub1 can phosphorylate an APC fragment in vitro (). We conclude that BubR1 can phosphorylate APC efficiently and with very high specificity in vitro. To determine whether microtubule-bound APC can still interact with BubR1, pure BubR1 and APC proteins were incubated with unpolymerized tubulin or taxol microtubules. The microtubules and proteins bound to them were recovered by sedimentation through a sucrose cushion. BubR1 did not pellet on its own with taxol-stabilized microtubules (, lane 6). In contrast, after the addition of full-length APC, a protein known to bind to microtubules (; ), all APC and a substantial amount of BubR1 cosedimented with microtubules (, lane 8). Thus, the interaction between BubR1 and microtubule-bound APC produced a ternary complex, and this is essential for stable kinetochore microtubule attachment. We have concluded that complex formation between BubR1 and APC/EB1 is necessary for their functions in kinetochore microtubule attachment and metaphase chromosome alignment. To further confirm whether APC–BubR1 interaction is essential for metaphase chromosome alignment, we examined the amount of APC at kinetochores in BubR1-depleted egg extracts supplemented with either WT or KD-BubR1 because only WT- but not KD-BubR1 can restore the chromosome alignment defect in BubR1-depleted egg extracts (). Immunofluorescence analysis has revealed that the amount of APC but not BubR1 at kinetochores was considerably reduced in BubR1-depleted egg extracts supplemented with KD-BubR1 (, A and B). Furthermore, APC was immunoprecipitated from egg extracts depleted of endogenous BubR1 and supplemented with a similar amount of purified WT- or KD-BubR1 recombinant proteins. Immunoprecipitates obtained using the APC antibody contained comparable levels of APC and EB1 under both conditions (, compare APC with EB1 in lanes 2 and 4). However, ∼40% less KD-BubR1 was coimmunoprecipitated with APC compared with WT-BubR1 (, compare BubR1 in lanes 2 and 4), indicating that APC phosphorylation by BubR1 might be necessary for APC function in metaphase chromosome alignment in egg extracts. The mitotic checkpoint functions for BubR1 kinase are well established: one is an evolutionary conserved Mad3-like role as part of a diffusible inhibitor (; ; ); added to this is a CENP-E–dependent kinase activity that is essential for one or more local roles as a kinetochore-bound catalytic facilitator (, ; ). Using antibody addition, immunodepletion, and reconstitution with purified recombinant proteins in egg extracts, which are naturally arrested in metaphase, we have demonstrated that BubR1 is necessary for metaphase chromosome alignment, which is consistent with recent studies regarding the involvement of BubR1 in kinetochore microtubule capture and attachment in mammalian cultured cells (; ). Furthermore, we show that BubR1 kinase activity is essential for its role in kinetochore microtubule attachment. The alignment defects we observed do not appear to result from a general kinetochore assembly defect: the addition of inhibitory antibody or reconstitution with KD-BubR1 disrupts chromosome alignment (this study) without removing kinetochore-associated motor CENP-E () as the immunodepletion of BubR1 does; using an immunodepletion and reconstitution approach in egg extracts, it has been shown that the depletion of BubR1 has no effect on the kinetochore localization of Mps1 and CENP-A, and the addition of either WT- or KD-BubR1 to BubR1-depleted egg extracts restores BubR1, Plx1, and 3F3/2 signals at kinetochores (); and immunodepletion of BubR1, APC, or EB1 from egg extracts does not affect kinetochore localization of the Ndc80 complex (unpublished data). Thus, based on a combination of the in vitro results presented in this study and prior in vivo BubR1 RNAi phenotypic analysis (), we propose that BubR1 and its kinase activity are directly involved in regulating stable kinetochore microtubule attachment and metaphase chromosome alignment. This also resolves a difference in the involvement of CENP-E in metaphase chromosome alignment between the and mammalian systems. Using immunodepletion from and antibody addition to egg extracts, the kinetochore-associated kinesin-like motor protein CENP-E has been shown to be essential for positioning chromosomes at the metaphase plate (). In contrast, most mammalian cells without CENP-E have a robust metaphase plate with only a few chromosomes abnormally close to the spindle poles (; ; ). The key underlying difference between two systems could simply be the activation of BubR1 kinase by CENP-E. The mammalian BubR1 has a basal level kinase activity in the absence of CENP-E (), whereas its homologue does not have any kinase activity without its activator CENP-E (). Indeed, in the presence of a motorless CENP-E fragment that can constitutively activate BubR1, the majority of chromosomes congress to the metaphase plate in the egg extracts depleted of endogenous CENP-E (). The plus end–tracking protein EB1 and its binding partner APC have previously been implicated in chromosome behavior in mammalian cultured cells (; ; ; ). In the present study, we show that APC and EB1 cooperate with the mitotic checkpoint kinase BubR1 to maintain stable kinetochore microtubule attachment and metaphase chromosome alignment (). BubR1 is recruited onto unattached kinetochores (; ; ). The microtubule-associated proteins APC and EB1 bind to plus ends of microtubules (). After the initial capture of microtubules by kinetochores (through an as yet unknown mechanism), the interaction between BubR1 and APC/EB1 stabilizes kinetochore microtubule attachment, in which BubR1 might directly phosphorylate APC. Evidence for this model includes the following: BubR1 forms a complex with APC and EB1 in egg extracts; the codepletion of BubR1, APC, and EB1 has no additive effect on mitotic spindle defects; replacing endogenous BubR1 with KD-BubR1 in egg extracts considerably decreases the amount of APC at kinetochores and produces a chromosome alignment defect; N-APC and EB1C fragments, which disrupt the complex formation of BubR1–APC/EB1, produce a chromosome alignment defect in egg extracts; BubR1 forms a ternary complex with APC and microtubules in vitro; and compared with another BubR1-binding partner, CENP-E, APC is a much better substrate for BubR1, indicating that APC might be one of the physiological substrates of BubR1. Finally, BubR1 localization to kinetochores in APC/EB1-depleted egg extracts was not impaired, suggesting that chromosome alignment defects in the absence of BubR1 or APC/EB1 reflect the formation of a functional complex among these proteins. APC has been shown to interact with mitotic checkpoint proteins Bub1/Bub3 () and BubR1 (this study). Whether APC has a role in mitotic checkpoint signaling is controversial. The depletion of APC has been reported to cause chromosome misalignment in metaphase and missegregation in anaphase (), indicating a defect in the mitotic checkpoint response. In addition, a recent study has shown that cells transfected with APC-targeting siRNA accumulated 1.84-fold and 1.66-fold less Bub1 and BubR1, respectively (). However, have shown high levels of kinetochore-bound Mad1, Mad2, Mps1, Bub1, and BubR1 checkpoint proteins in control-, EB1-, or APC-depleted prometaphase cells, which is consistent with an intact mitotic checkpoint. They suggest that instead of a checkpoint defect, chromosome missegregation in APC- and EB1-depleted cells is caused by misorientation and reduced stretching of aligned sister kinetochores, errors that are detected poorly, if at all, by the mitotic checkpoint (). We have shown that the mitotic checkpoint is activated in APC- and EB1-depleted egg extracts with high levels of BubR1 kinase activity and kinetochore-bound BubR1 and Mad2 (). Furthermore, APC- and EB1-depleted egg extracts have a chronically activated mitotic checkpoint response even in the presence of spindle microtubules (in the absence of nocodazole), supporting our conclusion that there is improper kinetochore microtubule attachment in the absence of APC and EB1. However, our results do not exclude the possibility that APC might have a nonessential enhancement role in mitotic checkpoint signaling. Our current studies confirm by completely independent means and extend the evidence that both BubR1 and APC/EB1 individually play a role in metaphase chromosome alignment. In addition, our egg extracts and in vitro results have important implications that complex formation between BubR1 and APC/EB1 is essential for the successful formation of stable kinetochore microtubule attachment, providing a potential link in stable kinetochore microtubule attachment. Consistent with this idea, BubR1 APC compound mice have been shown to develop 10 times more colonic tumors than APC mice (), demonstrating that there is indeed a functional interaction between BubR1 and APC in vivo. Furthermore, APC has been shown to associate with MCAK at the centromere/ kinetochore region in egg extracts (). MCAK is thought to be involved in the depolymerization of improperly attached microtubules (merotelic and syntelic microtubule attachment at kinetochores) to ensure accurate chromosome segregation (). MCAK is one of the known substrates of Aurora B (; ), and its microtubule-depolymerizing activity is reduced upon Aurora B phosphorylation in vitro (Andrews et al., 2004; ; ). The kinetochore-microtubule destabilization phenotype upon BubR1 and APC depletion can be partially suppressed by Aurora kinase inhibition in both mammalian cultured cells () and egg extracts (). These results indicate that BubR1, APC/EB1, Aurora B, and MCAK might cooperate to ensure proper kinetochore microtubule attachment. The full-length cDNA clone for APC was a gift from B. Gumbiner (Memorial Sloan-Kettering Cancer Center, New York, NY), and the full-length EB1 cDNA clone was provided by Y. Tsuchiya (Toho University School of Medicine, Tokyo, Japan). APC and EB1 have 75.8% and 83.3% identities to human homologues, respectively. APC fragment (APC) and full-length EB1 proteins were expressed in with a His tag at the N terminus and were purified over Ni–nitrilotriacetic acid (NTA) columns. Purified proteins were used to raise antibodies in rabbits (for APC) and chickens (for EB1), and sera were affinity purified over immobilized APC and EB1 fusion proteins. CSF-arrested extracts were prepared from unfertilized eggs as previously described (). Checkpoint extracts were prepared from these by incubation with ∼9,000 demembranated sperm nuclei/μl and 10 μg/ml nocodazole (). Exit from CSF arrest was induced by the addition of 0.4 mM CaCl. Demembranated sperm was added to a portion of each extract, rhodamine-labeled bovine brain tubulin was added at 1/300 μl of extract, and exit from metaphase arrest was induced by the addition of Ca. Cell cycle progress of egg extracts was followed by fluorescence microscopic examination of 1-μl aliquots squashed under a coverslip. 80 min after exiting from metaphase, 0.5 vol of the appropriate egg extracts was added and incubated for an additional 60–120 min. M-phase structures accumulating in egg extracts were scored in squashed samples. Bipolar spindles with all chromosomes aligned at the spindle equator were scored as bipolar aligned, whereas bipolar spindles with scattered chromosomes were counted as bipolar misaligned. Monopolar and other mitotic structures were scored as other. At least three independent depletion experiments were performed. Immunofluorescence with egg extracts was performed as previously described (). For immunofluorescence, all images in each experiment were collected on the same day using identical exposure times. Image acquisition and data analysis were performed at room temperature using an inverted microscope (IX81; Olympus) with a 60× NA 1.42 plan Apo oil immersion objective lens (Olympus), a monochrome CCD camera (Sensicam QE; Cooke), and the Slidebook software package (Olympus). The APC kinetochore fluorescence intensity was quantified using Image software (Scion). Affinity-purified antibodies were coupled to Dynal beads, and the beads were washed twice with lysis buffer. The beads were then incubated at 4°C for 1 h with egg extracts that had been incubated at room temperature for 30 min with or without sperm nuclei and nocodazole. After mixing, the beads were washed twice with PBS buffer and twice with PBS plus 0.5 M NaCl. The immunoprecipitates were then solubilized in SDS sample buffer and subjected to immunoblot analysis. 200 μl of high speed supernatant of CSF-arrested egg extracts was loaded onto a 10-ml 5–15% continuous sucrose gradient and spun for 16 h at 32,000 rpm in a SW41Ti rotor (Beckman Coulter) at 4°C. 500 μl of fractions were collected from each gradient. APC was expressed in insect cells infected with a baculovirus encoding APC and purified over immobilized Ni-NTA agarose. GST– N-APC and His-EB1 and -EB1C were expressed in and purified over glutathione–Sepharose beads or Ni-NTA agarose beads, respectively. For in vitro kinase assays, recombinant BubR1 and CENP-E were incubated with or without purified APC at room temperature for 30 min with 25 mM Hepes, pH 7.5, 10 mM MgCl, 200 μM ATP, and 1 μCi γ-[P]ATP and were assayed for kinase activity. 10 nM of purified BubR1 and/or APC was mixed with 1 μM taxol-stabilized microtubules or unpolymerized tubulin in BRB80 and incubated in room temperature for 30 min. Binding reactions were centrifuged over 40% sucrose/BRB80 at 10,000 rpm for 30 min at 35°C. Recombinant proteins in the pellet and supernatant were analyzed by immunoblotting.

The nuclear envelope (NE) is the selective barrier that defines the interface between the nucleus and the cytoplasm (; ). Because it mediates molecular trafficking between these two compartments, it plays an essential role in the maintenance of their biochemical identities. In addition to its transport function, the NE is also a key determinant of nuclear architecture, providing anchoring sites at the nuclear periphery for chromatin domains as well as for a variety of structural and regulatory molecules. A corresponding contribution to cytoplasmic structure has been described in which NE components may also influence cytoskeletal organization and mechanotransduction (; ). The NE is composed of several structural elements, the most prominent of which are the inner nuclear membranes (INMs) and outer nuclear membranes (ONMs). These are separated by the perinuclear space (PNS), a gap of 30–50 nm. Annular junctions between the INM and ONM create aqueous channels that traverse the NE and that are occupied by nuclear pore complexes (NPCs). It is the NPCs that endow the NE with its selective transport properties (). The final major feature of the NE is the nuclear lamina, a thin (20–50 nm) protein layer that is associated with both the INM and chromatin. The lamina is composed primarily of A- and B-type lamins, which are members of the intermediate filament protein family (). The lamins interact with components of the INM and NPCs as well as with chromatin proteins and DNA (). In this way, the lamina plays an important structural and organizational role at the nuclear periphery. Despite their continuities, the INM and ONM are biochemically distinct. The ONM features numerous junctions with the peripheral ER, to which it is functionally and compositionally similar. In contrast, the INM contains its own unique selection of integral proteins. Clearly, the INM, ONM, and ER represent discrete domains within a single continuous membrane system. Accordingly, the PNS is a perinuclear extension of the ER lumen. Localization of integral proteins to the INM involves a process of selective retention (; ; ). Although proteins that are mobile within the ER and ONM may gain access to the INM via the continuities at each NPC, only proteins that interact with nuclear or other NE components are retained and concentrated. Recent studies suggest that additional mechanisms may overlie this basic scheme. showed that movement of integral proteins through the NPC membrane domain is energy dependent. Other studies suggest a role for the nuclear transport receptor adaptor karyopherin/importin-α in the transit of proteins to the INM (; ). Recognition of ONM-specific membrane proteins raises the question of what prevents these proteins from escaping to the peripheral ER. In , the localization of Anc-1, an ONM protein involved in actin-based nuclear positioning, requires Unc-84, an INM protein whose retention is lamin dependent (; ). These observations led to a model in which Unc-84 and Anc-1 interact across the PNS via their lumenal domains, providing a mechanism for the tethering of ONM proteins. In mammals, two large actin-binding proteins, nesprin 1 Giant (nesp1G; 1,000 kD) and nesprin 2 Giant (nesp2G; 800 kD), reside in the ONM (; ; ; ). The nesprins (also known as Syne 1 and 2) are related to both Anc-1 and a ONM protein, (; ), in that they contain an ∼50–amino acid C-terminal KASH (Klarsicht, Anc-1, Syne homology) domain consisting of a single transmembrane (TM) anchor and a short segment of ∼30–40 residues that resides within the PNS. A third ONM KASH domain–containing protein, nesprin 3, interacts with plectin, which is a large (466 kD) cytolinker (). Unc-84 contains an ∼200–amino acid C-terminal region that shares homology with Sad1p, a spindle pole body protein (). This sequence, which is known as the SUN (Sad1p, Unc-84) domain, resides within the PNS. The human genome encodes five SUN domain proteins. Two of these, Sun1 and 2, are lamin A–interacting proteins of the INM with topologies similar to that of Unc-84 (; ; ). Both Sun1 and 2 cooperate in tethering nesp2G in the ONM (; ; ; ). This tethering involves interactions that span the PNS (), similar to that suggested for Unc84 and Anc-1. Unc84 also tethers Unc-83, another ONM KASH domain protein (). Competition between nesprin 1 and 2 KASH domains () suggests that nesp1G is similarly tethered. In this way, Sun1 and 2 function as links in a molecular chain that connects the actin cytoskeleton via nesprins to lamins and other nuclear components. We have termed this assembly the LINC (linker of nucleoskeleton and cytoskeleton) complex (). The fact that nesprin 3 binds plectin, a diverse cytolinker (), indicates that there may be multiple isoforms of the LINC complex responsible for integrating the nucleus with different components of the cytoskeleton. As alluded to in the previous paragraphs, the NE can influence cytoplasmic mechanics and the responses of cells to mechanical stress. Cells depleted of either A-type lamins or emerin, an INM protein, exhibit reduced cytoplasmic resilience and an inability to activate mechanosensitive genes (; , , 2006). In humans, the loss or mutation of either A-type lamins or emerin is associated with several diseases (), including Emery-Dreifuss muscular dystrophy. It is not hard to imagine that the LINC complex might be the mediator of these effects given its proposed role in nucleocytoplasmic coupling. Less clear is the extent and nature of the interactions of LINC complex components and how these might affect LINC complex function. In the case of nesprins 1 and 2 versus nesprin 3, there are obvious differences in terms of actin versus plectin association. At the INM, the situation with the Sun proteins is more ambiguous. We know that there is some degree of functional redundancy between Sun1 and 2 with respect to nesprin 2 tethering. Furthermore, we know that Sun1 and 2 can associate with lamin A but that this interaction is not required for their localization. In this study, we further explore the interactions of SUN proteins at the nuclear periphery. In doing so, we have been able to describe discrete regions within Sun1 that function both in localization to the INM and in oligomerization. Most importantly, we are able to demonstrate that Sun1 and 2 are segregated within the INM. Although Sun2 displays a roughly uniform distribution across the NE, Sun1 is concentrated at NPCs. Elimination of Sun1 or overexpression of Sun1 mutants leads to NPC clustering. The inference is that Sun1 but not Sun2 functions in maintenance of the uniform distribution of NPCs. It also follows that certain LINC complex isoforms may mediate the differential association of cytoskeletal elements with NPCs versus NPC-free regions of the NE. Mammalian SUN proteins are encoded by at least five genes (). Of these, only Sun1 and 2 are widely expressed in somatic cells (; ). Sun3 (; ; ), Sun4 (SPAG4; ; ), and Sun5 (SPAG4Like; unpublished data) seem to be restricted largely to the testis. Each of the SUN proteins conforms to the same basic structure featuring an N-terminal domain followed by a block of hydrophobic amino acid residues, likely representing a TM domain, and a C-terminal SUN domain. The relationships between these proteins are displayed in . In the case of Sun1, the largest of the mammalian SUN proteins, the nucleoplasmic N-terminal domain is composed of 350–400 amino acid residues (). All of the sequence variants of Sun1 that arise through alternative splicing involve changes in this segment of the molecule (). A prominent feature of Sun1 is the presence of four hydrophobic sequences, H1–H4, any one of which could potentially function as a TM domain. Previously, we showed that H1 at least does not span the INM (). This conclusion was based on a naturally occurring splice isoform that is missing sequences encoded by exons 6, 7, and 8 (Sun1Δ221–344; Δexons 6–8). This isoform lacks H1 yet displays appropriate NE localization and has the same topology within the INM as full-length Sun1. Nevertheless, as will be expanded upon, although not a TM domain, H1 may still contribute to membrane association. Mouse Sun1 also contains a predicted C2H2 zinc finger. Because this is absent from primate Sun1, its significance remains unclear. The C-terminal region of Sun1, like that of Sun2, consists of roughly 450 amino acid residues and resides within the PNS. It contains a membrane-proximal predicted coiled coil and a conserved distal SUN domain. The junction between these two features contains the KASH-binding site and is therefore essential for the tethering of ONM nesprins. Although Sun1 and 2 have obvious similarities in terms of domain organization and topology, we noticed differences in their localizations within the NEs of multiple cell types. In particular, the distribution of Sun1 appears more punctate than that of Sun2 (). This was evident by both immunofluorescence microscopy using antibodies against the endogenous proteins as well as by observations on exogenous Sun1 or 2 carrying a C-terminal GFP tag. A direct comparison of the two proteins indicates that they are largely segregated within the plane of the NE. Double-label experiments using an antibody against Nup153, an NPC component, revealed that Sun1 but not Sun2 was concentrated at NPCs. This localization was confirmed by immunoelectron microscopy of HeLa cells expressing either Sun1- or Sun2-GFP (). Quantitative analysis revealed that in the case of Sun1, gold particles were all clustered within 120 nm of the NPC eightfold axis with a peak at 66 ± 20 nm (±SD; ). Sun2 displayed a far broader distribution with a median distance from the NPC eightfold axis of 240 ± 120 nm (). Only 8% of gold particles were within 120 nm, and none were within 66 nm. Of necessity, the scale in is five times that in . The differential localization of Sun1 versus Sun2 is reflected in their behavior during mitosis. After NE breakdown, both proteins are dispersed throughout the cytoplasm, most likely in ER membranes. During late anaphase/early telophase, as NE reassembly commences, the resegregation of Sun1 and 2 occurs. Sun2 concentrates at a region of each newly separated chromatin mass adjacent to one of the spindle poles. This core region () of chromatin is typically deficient in NPC reformation. The behavior of Sun2 mirrors that of another INM protein, emerin, which also concentrates at the chromatin core at roughly the same time (). In contrast, Sun1 tends to be excluded from the core region and instead concentrates on the lateral margins of the chromatid masses where NPC assembly is initiated. Localization of Sun1 to the INM involves determinants in both N- and C-terminal domains (; ; ; ; ), although their relative contribution to localization and role in NPC association remain unknown. In addition, although previous studies (; ) have demonstrated that the TM domains of Sun1 are contained within the H2–H4 region, they provide an ambiguous view of the targeting properties of this segment of the molecule. We and others (; ) had proposed that Sun1 might be a multispanning protein with three TM sequences corresponding to H2, H3, and H4. However, direct evidence to support such a model is lacking. To better define Sun1 topology, we prepared a series of mutants containing different combinations of the H2, H3, and H4 hydrophobic sequences, all of which were found to confer some degree of membrane association. These mutants contained all or part of the N-terminal domain (residues 1–355) followed by one or more hydrophobic sequences and terminating with GFP (). At the N and C termini of certain chimeras, we placed HA and myc epitope tags, respectively. These constructs were expressed by transfection in HeLa cells, which were subsequently treated with low concentrations of digitonin. Under these conditions, the plasma membrane is permeabilized, but the ER and nuclear membranes remain intact. The permeabilized cells were then incubated with proteinase K. During the course of this incubation, Western blot analysis revealed that cytoplasmic (tubulin) and nuclear (lamin A/C and Nup153) proteins were degraded (proteinase K may enter the nucleus by degrading NPC proteins), whereas ER lumenal and PNS proteins such as protein disulfide isomerase (PDI) were protected (). Triton X-100 (TX-100) permeabilization permitted the digestion of all of these proteins. In the case of the Sun1 chimeras, we determined the latency of the HA and myc epitope tags by both SDS-PAGE analysis of immunoprecipitated, radiolabeled proteins and immunofluorescence microscopy (the latter on permeabilized but not proteinase-treated cells; , A and B, respectively). In some experiments, we used specific antibodies to monitor the latency of the GFP moiety (). The results reveal that the N-terminal HA tag is always exposed to the cytoplasm or nucleoplasm. In contrast, the myc tag or GFP becomes latent (i.e., it resides within the ER lumen/PNS) whenever the H4 sequence is present within the chimera. No combination of H2 and H3 (either singly or together) would confer such latency. Conversely, neither H2 nor H3 could affect the orientation of H4 and, therefore, the latency of the myc tag or GFP. The only reasonable conclusion is that although H2 and, to a lesser extent, H3 may confer membrane association (), they do not cross the bilayer. Therefore, rather than being a multispanning protein as previously suggested, Sun1 would appear to be a type II membrane protein with a single TM domain represented by H4 (see ). With a better understanding of Sun1 topology, we next wished to identify NE and NPC retention domains. To this end, we generated an extensive family of chimeric Sun1 proteins. Because H4 appears to represent the sole TM domain, we sought to clarify the role of the other hydrophobic sequences. Sun1N355, which contains the H1 sequence but lacks H2 and H3, was found to localize to the NE (Fig. S1, available at ; ). This is in contrast to the nucleoplasmic localization of Sun1N355Δ221–343 (this corresponds to the exon 6–8 deletion; ) or of Sun1N220 (), both of which lack any hydrophobic motif. When Sun1N355Δ221–343 was extended to include the H2 domain (Sun1N380Δ221–343), NE association was rescued (). Taking all of this data together, we can conclude that H1, H2, and, to a certain extent, H3 are each sufficient to confer membrane association. However, because the Sun1N220 region itself will accumulate readily within the nucleoplasm (), these experiments do not reveal whether any of the hydrophobic sequences themselves have intrinsic INM-targeting activity. To address this question, we next examined the behavior of two additional H2-containing fusion proteins, which are both tagged at the N terminus with the myc epitope (). The first of these represented an N-terminal truncation lacking the initial 220 Sun1 residues but containing H1 and H2 (myc-Sun1 221–380), whereas the second was missing H1 in addition to the N-terminal 220 residues (myc-Sun1 261–380). The former localized to the NE and, to a lesser extent, to the peripheral ER. In contrast, the latter was primarily ER associated with little concentration in the NE. The implication of these results is that an NE localization motif is encoded by Sun1 residues 221–380. This region of the molecule must therefore share interactions with other nuclear or NE components. We next examined whether the H2–H4 region alone has a role in NE targeting. When this sequence was fused to the N terminus of GFP, it localized predominantly to the Golgi apparatus and cell surface with little, if any, associated with the NE (). In contrast, a Sun1 N-terminal truncation consisting of the H2–H4 region followed by the Sun1 lumenal domain (H234Sun1L-GFP) localized efficiently to the NE (Fig S1; ). However, we already know that a soluble form of the Sun1 lumenal domain that is appropriately localized to the ER lumen and PNS is itself insufficient for NE targeting (SS-HA-Sun1L-KDEL; ). There are at least two explanations for these results. The first is that the lumenal domain does contain targeting information but that it is only functional when the domain is appropriately oriented or tethered to the ER or nuclear membranes. The second is that the H2–H4 hydrophobic region can direct localization to the INM but that this only occurs in the context of the Sun1 lumenal domain. In other words, the Sun1 lumenal domain can modify the behavior of the H2–H4 sequences. What we can rule out, however, is any suggestion that localization of H234Sun1L-GFP to the INM occurs by virtue of oligomerization with endogenous Sun1. Overexpression of H234Sun1L-GFP leads to the displacement of endogenous Sun1 from the NE while itself concentrating in the NE (; unpublished data). Additionally, depletion of Sun1 by RNAi has no effect on H234Sun1L-GFP localization (unpublished data). To further address these issues, we replaced the H2–H4 hydrophobic region of both full-length Sun1 and H234Sun1L-GFP with the unrelated TM domain of Sun3 (to yield HA-Sun1(S3TM) and S3TMSun1L-GFP, respectively; ). As shown in Fig. S2 A (available at ), when expressed in HeLa cells, HA-tagged Sun3 localizes to the NE but does not associate with NPCs. Sun3 contains a single predicted TM domain that resides between residues 46 and 65. As will become evident below (), this sequence contains no intrinsic NE-targeting activity. Translation of Sun3 in vitro in the presence of microsomes confirms that this sequence must function as a TM domain with a type II orientation (Fig. S2 B). HA-Sun1(S3TM) behaved exactly like full-length Sun1 in that it concentrated in the NE () in association with NPCs (Fig. S3, available at ). In contrast, S3TMSun1L-GFP displayed little or no NE localization and instead was found in the Golgi apparatus and at the cell surface (). Evidently, it is not retained in the nuclear membrane/ER system. Deletion of H2–H3 from H234Sun1L (H4Sun1L-GFP) also resulted in the loss of NE association (). The implication, then, is that H2–H3 encodes an NE localization function. If this is the case, attaching H2–H3 to the N terminus of S3TMSun1L-GFP should lead to its accumulation at the NE. Indeed, we do observe a partial restoration of NE localization (). It is evident from these results that although the H2–H3 sequence promotes localization to the NE, its activity is strongly influenced by the Sun1 lumenal domain. This is despite the fact that these regions of Sun1 reside on opposite sides of the INM. A possible explanation for this result is that the targeting activity of H2–H3 may be activated or enhanced by dimerization (or oligomerization), perhaps leading to increased avidity for nuclear or INM-associated binding partners. A prediction here is that the lumenal domain of Sun1 should mediate dimerization (or oligomerization). This is borne out in transfection experiments in which full-length Sun1 was coexpressed in HeLa cells with a variety of epitope-tagged Sun1 deletion mutants (). Immunoprecipitation analyses resulted in the efficient coprecipitation of full-length and mutant Sun1 only when the mutant form contained an intact lumenal domain. Further compelling evidence for lumenal domain–mediated oligomerization was provided by immunofluorescence observations of SS-HA-Sun1L-KDEL and S3TMSun1L. As described previously (), neither of these chimeric proteins concentrates to any great extent in the NE. However, the overexpression of full-length Sun1 will recruit both of these proteins to the NE. Collectively, these data clearly demonstrate that Sun1 homooligomerizes via lumenal domain interactions, most likely involving the predicted membrane proximal coiled coil. Which Sun1 sequence elements are required for association with NPCs? Analysis of all of the Sun1 constructs that we have prepared revealed that apart from wild-type Sun1, only Sun1Δ221–343 and Sun1(S3TM) associated with NPCs ( and S3 A). Evidently, association with NPCs does not involve the H1 and H234 hydrophobic sequences acting in concert. To take these analyses further, we prepared a pair of chimeras in which we swapped the Sun1 and 2 lumenal domains. In neither case could we observe NPC association (Fig. S3 B). Instead, both recombinant proteins behaved like Sun2. Evidently, both nucleoplasmic and lumenal domains of Sun1 cooperate in conferring NPC association. So far, we have shown that there are multiple determinants within the Sun1 nucleoplasmic domain that can confer localization to the INM. used FRAP analysis to show that wild-type Sun1 is relatively immobile within the INM. We performed similar analyses on a subset of our Sun1 deletion mutants that localize to the NE (). In all cases, these mutants display enhanced mobility relative to wild-type Sun1. Even deletion of the lumenal domain, which appears to contain no intrinsic targeting function but does promote oligomerization, leads to increased mobility within the INM. Thus, although Sun1 does contain multiple autonomous features involved in localization, stable localization to the NE requires that all be present. These findings are reminiscent of our conclusion that multiple features within the Sun1 molecule are required for NPC association. What is the functional relevance of Sun1 association with NPCs? Proteomic studies provide no evidence that Sun1 is an intrinsic component of the NPC (). However, to determine whether Sun1 might contribute to NPC functionality, we examined nuclear transport in HeLa cells that had either been depleted of Sun1 by RNAi or that expressed Sun1 fragments, some of which resulted in a loss of endogenous NE-associated Sun1 (Fig. S5, available at ). To accomplish this, we took advantage of a GFP fusion protein bearing nuclear import and export signals (NLS-GFP–nuclear export sequence [NES]) and that shuttles between the nucleus and cytoplasm (). We also used a hormone-inducible nuclear import substrate consisting of β-galactosidase fused to the glucocorticoid receptor (grβ; ). Our results indicate that Sun1 has no substantial role in the nuclear transport of proteins, either import or export. Similarly, the distribution of poly A RNA revealed by in situ hybridization suggests that Sun1 makes little or no contribution to mRNA export (unpublished data). However, Sun1 depletion was not without effect on pore complexes. We noticed that the loss of Sun1 was always associated with an altered distribution of NPCs () as well as altered nuclear shape (). In wild-type cells, NPCs tend to be uniformly distributed across the nuclear surface. After Sun1 depletion, NPC aggregates or clusters could be observed leaving NPC-free areas of varying sizes. This effect was Sun1 specific because the depletion of Sun2 left NPC distribution unchanged. This effect of Sun1 depletion on NPC aggregation could be emulated by the overexpression of nucleoplasmic Sun1 deletion mutants in HeLa cells. The expression of these mutants often leads to a diminution in the amount of full-length Sun1 at the NE (and thus at NPCs). A quantitative analysis of NPC aggregation induced by both Sun1 depletion and Sun1 mutant expression is displayed in . Because Sun1 may act as a tether for ONM nesprins, it is possible that NPC aggregation is a function of the loss of nesprins rather than a loss of Sun1. To determine whether this might be the case, we overexpressed a protein consisting of GFP fused to the KASH domain of either nesprin 1 or 2 (GFP-KASH1 or 2) in HeLa cells. Overexpression of GFP-KASH1 or 2 leads to the displacement of nesprins 1 and 2 from the NE (). Treatment of cells in this way was found to have no discernible effect on NPC distribution (unpublished data). These data suggest that Sun1 has a nesprin-independent role in maintenance of the uniform distribution of NPCs across the NE. Sun1 and 2 are of a pair of ubiquitous INM proteins that tether nesprins within the ONM. Nesp1G and nesp2G contain N-terminal actin-binding domains (; ), whereas nesprin 3 binds plectin, a versatile cytolinker (). Thus, the SUN proteins represent links in a molecular chain that connects elements of the cytoskeleton to components within the nucleus. We have previously referred to translumenal Sun–nesprin pairs as LINC complexes (). Multiple LINC complex isoforms likely exist given the apparent redundancy of Sun1 and 2 in tethering nesprins. In addition, we can identify at least four or five splice isoforms of Sun1 alone, further increasing the LINC complex repertoire. The nesprins themselves (including nesprin 3) are also represented by dozens of splice isoforms. Aside from nesp1G and nesp2G, the number of these that may be tethered by Sun proteins at the ONM remains unknown. Previous studies indicated that KASH domain proteins play an important role in nuclear positioning in certain cell types (; ; ; ; ; ). However, the existence of links spanning the NE have far broader implications than mere nuclear location and present us with a mode (or modes) of nucleocytoplasmic coupling that may bypass NPCs. This notion is highlighted by biomechanical studies on -null fibroblasts, which exhibit impaired mechanotransduction and decreased viability under mechanical strain (; , , ). Induction of the mechanosensitive genes and is attenuated, as is nuclear factor κB–regulated transcription in response to either cytokine or mechanical stimulation. Although nuclei in -null cells are both mechanically fragile and highly deformable, a surprising finding of is that these cells also feature reduced cytoplasmic resilience. Given that both Sun1 and 2 interact with A-type lamins, it is possible that the LINC complex might mediate mechanotransduction and the lamin-dependent changes in cytoplasmic organization. Retention of Sun1 and 2 in the INM is independent of A-type lamins in some cell types (; ; ; ). This implies that there have to be other nuclear or NE components that interact with and retain the SUN proteins. Logically, based on our studies here, there have to be at least two discrete regions within Sun1 that are sufficient for INM localization. Evidence for this can be seen in the differential effect of exogenous Sun1 and 2 on each other. Sun2 will not substantially displace Sun1 in HeLa cells. However, Sun1 can efficiently displace Sun2 from the INM (), presumably by competition for a common binding partner or anchor . Therefore, Sun1 likely has an additional binding partner that is not shared with Sun2. This is perhaps most obvious when considering that both of these proteins are segregated within the plane of the NE. Although Sun2 predominates in NPC-free regions, Sun1 is concentrated in the vicinity of NPCs, possibly forming a halo around each NPC. The mechanisms of interaction with NPCs remain unknown. However, it clearly requires contributions from both the nucleoplasmic and lumenal domains. Our analyses suggest that there are at least two separate INM-targeting regions within the Sun1 nucleoplasmic domain. The first lies between residues 1–260 and includes the H1 hydrophobic sequence. The second is immediately downstream of the H1 sequence and includes the H2 and H3 hydrophobic sequences. The bulk of this second targeting region is absent in the Sun1Δ221–343 (i.e., the Δexon 6–8) splice isoform, although the H2 and H3 sequences are retained. Because Sun1Δ221–343 still colocalizes with NPCs, the bulk of this second targeting region cannot have an essential role in NPC association. The same is also true of the entire H234 region, which can be substituted by the Sun3 TM domain without affecting NPC association. Although the H2–H3 hydrophobic sequence exhibits INM-targeting activity, it is only functional in the context of molecules containing the lumenal domain. The lumenal domain has no intrinsic targeting properties but does promote oligomerization, most likely based upon coiled-coil homodimers. We would suggest that manifestation of the INM localization function of H2–H3 requires dimerization/oligomerization, perhaps leading to increased avidity for an NE or nuclear binding partner. The Sun1 TM domain is contained within the region of the molecule defined by the H2–H4 hydrophobic sequences. We and others had suggested that these might represent three TM domains (; ). Our current studies suggest that H2 and H3 do not, in fact, span the INM, leaving H4 as the only TM sequence within the Sun1 molecule (). This view is reinforced by the existence of an apparent human Sun1 splice isoform (GenBank/EMBL/DDBJ accession no. ) lacking sequences encoded by exons 6–9 and missing H2. Our conclusion is that Sun1 has the topology of a type II membrane protein. Although Sun1 has only a single membrane-spanning domain (H4), the three other hydrophobic sequences, H1, H2, and H3, can confer membrane association. Nucleoplasmic domain constructs that contain at least one of the three all become associated with the INM, whereas their absence leads to nucleoplasmic localization. It remains unclear whether these hydrophobic sequences interact directly with the INM lipid bilayer or whether association is mediated by other INM proteins. The former would appear more likely because regardless of expression level, H1-, H2-, or H3-containing proteins always appear membrane associated. The interaction of extended hydrophobic sequences such as H2–H3 with the lipid bilayer is not without precedent. For instance, the tubular ER protein reticulon 4 contains a 30–40-residue hydrophobic sequence that forms a hairpin, which dips into the cytoplasmic face of the ER lipid bilayer without crossing it (). The segregation of Sun1 and 2 within the plane of the NE and the association of Sun1 with NPCs is quite striking. Could Sun1 be an NPC component? The complement of mammalian NPC subunits identified by using proteomic approaches does not include Sun1. However, the same is also true of the authentic vertebrate NPC membrane protein Ndc1 (). Perhaps Sun1's additional associations with the nuclear lamina and possibly chromatin limits its coextraction with NPC proteins. Regardless, we can find no evidence that Sun1 contributes to nucleocytoplasmic transport, and, consequently, we feel that Sun1 is unlikely to represent an intrinsic NPC component. Instead, a more reasonable scenario is that Sun1 is associated with the NPC periphery and may define a novel microdomain within the nuclear membranes, which, in turn, could blur the boundary between NPCs and the bulk of the NE (). The presence of Sun1 and, by implication, nesprins at NPCs could provide a basis for older ultrastructural observations that cytoskeletal elements, particularly intermediate filaments, frequently seem to contact pore complexes (). The distribution of NPCs across the NE is not random. Rather, they are arrayed in a uniform (although not regular) fashion that is constrained by a minimum NPC separation (). We have observed that the depletion of Sun1 (but not Sun2) or overexpression of truncated forms of Sun1 lead to the formation of NPC aggregates or clusters. This suggests that Sun1 has a role in the maintenance of uniform NPC distribution across the nuclear surface. In mammalian cells, NPCs are largely immobile and maintain their relative positions over many hours (; ). The implication is that NPC clusters in Sun1-depleted cells may arise during de novo NPC assembly as well as during postmitotic reassembly. A-type lamins are also determinants of NPC distribution because -null mouse embryonic fibroblasts (MEFs) frequently feature clustered or aggregated NPCs (). Furthermore, have shown that A-type lamins strongly influence the distribution of NPCs and pore-free regions of the NE. It is important to bear in mind, however, that cells that normally lack A-type lamins (early embryonic cells, for instance) do not display obviously clustered NPCs. It follows that there must be additional mechanisms to define NPC distribution that predominate in certain cell types. Such mechanisms might potentially involve B-type lamins (). Because Sun1 interacts with lamin A via the N220 region of its N-terminal domain (), it could function as an adaptor between the nuclear lamina and the NPC (). Of more significance is our observation that Sun1 has a preference for farnesylated pre–lamin A. Given that pre–lamin A exists only transiently in normal cells, this raises the possibility that Sun1 might function in the targeting and assembly of newly synthesized lamin A at the nuclear face of the INM. If this is the case, given the localization of Sun1, NPCs could actually function as nucleation sites of A-type lamina assembly. Ultimately, this may help define the distribution of NPCs. Our next goal will be to test this notion by determining whether there is a spatial relationship between A-type lamina assembly and NPCs. Regardless of the outcome of these studies, it is becoming increasingly clear that there are complex networks of interactions at the nuclear periphery involving NPCs, INM and ONM proteins, and nuclear lamins. These interactions appear to define not only the organization of the NE but also determine cytoskeletal mechanics and perhaps mediate signaling between the nucleus and cytoplasm. HeLa cells and MEFs, both and (), were maintained in 7.5% CO and at 37°C in DME (Invitrogen) plus 10% FBS (Hyclone), 10% penicillin/streptomycin (Invitrogen), and 2 mM glutamine. Plasmid DNA was introduced into HeLa cells and MEFs by using the Polyfect reagent as described previously () or with the LipofectAMINE 2000 reagent (Invitrogen). To transfect a 3.5-cm tissue culture dish of cells with LipofectAMINE 2000, 6 μl of transfection reagent or 2 μg of plasmid DNA were each added to separate 100-μl vol of Optimem (Invitrogen) and were combined and incubated at RT for 20 min. Subsequently, the cell medium was replaced with serum-free DME and the 200-μl transfection mix added dropwise and was incubated at 37°C for 1 h, after which time the medium was replaced with DME/10% FCS. Cells were analyzed 1–2 d later. A HeLa cell line stably expressing a tetracycline repressor protein from a pcDNA6/TR plasmid (TRex-HeLa; Invitrogen) was transiently transfected with pcDNA4/TO plasmid (Invitrogen) containing murine Sun1GFP or human Sun2GFP. After transfection, cells were selected with 200 μg/ml zeocin, and stably expressing subclones were isolated. Stable cells were analyzed 24 h after the addition of 1 μg/ml tetracycline. The following antibodies were used in this study: the monoclonal antibodies 9E10 and 12CA5 against the myc, HA, and GFP epitope tags were obtained from the American Type Culture Collection, Covance, and Abcam, respectively. Rabbit antibodies against the same epitopes were obtained from Abcam. Rabbit antibodies against Sun1 and 2 were previously described (). Mouse monoclonal anti-nup153 (clone SA1) and anti-nucleoporin (clone QE5) were described previously (; ). Rabbit anti-emerin was a gift from G. Morris (Robert Jones and Agnes Hunt Orthopaedic Hospital, Oswestry, UK). Mouse anti–β-galactosidase was purchased from Promega. Mouse anti-PDI and antitubulin were obtained from Abcam. Goat anti– lamin A/C was obtained from Santa Cruz Biotechnology, Inc. Secondary antibodies conjugated with AlexaFluor dyes were obtained from Invitrogen. Peroxidase-conjugated secondary antibodies were obtained from Biosource International. For immunofluorescence microscopy, 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% TX-100. Cells labeled with anti-Sun1 were fixed for 6 min in 3% formaldehyde and permeabilized for 15 min in 0.4% TX-100 in PBS. The cells were then labeled with the appropriate antibodies plus the DNA-specific Hoechst dye 33258. For experiments involving selective permeabilization, the cells were first fixed in 3% formaldehyde. This was followed by permeabilization in 0.001% digitonin in PBS on ice for 15 min (). The cells were then labeled with appropriate primary and secondary antibodies. Specimens were observed using a fluorescence microscope (DMRB; Leica). Images were collected using a CCD camera (CoolSNAP HQ; Roper Scientific) linked to a computer (G4; Macintosh) running IPLab Spectrum software (Scanalytics). Image quantification was performed using IPLab software. FRAP experiments were performed on a confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.) with a 63× 1.4 NA oil objective (Carl Zeiss MicroImaging, Inc.). GFP was excited with the 488-nm line of an Ar laser, and GFP emission was monitored using a 505-nm longpath filter. Cells were maintained at 37°C using an incubator (ASI Air Stream; Nevtek). In transfected cells, a rectangular region typically 2–4 μm in height was bleached in three iterations using a 488-nm laser line at 100% laser power. Cells were monitored at 5-s intervals for up to 300 s. Data were normalized as described previously (; ) to take into account bleaching during the imaging phase. Recovery values are means ± SD from at least five cells from at least two independent experiments. Immunoelectron microscopy of Sun1 was performed by preembedding and labeling of the tetracycline-inducible cell line expressing Sun1-GFP. After fluorescence examination to verify GFP expression, cells were trypsinized and pelleted by centrifugation. The pellet was permeabilized with 0.1% TX-100 for 1 min in PBS and fixed with 3% PFA in PBS for 10 min followed by three washes with PBS. The fixed cells were then incubated with 2% BSA in PBS for 10 min followed by incubation with the primary polyclonal anti-GFP antibody ab290 (Abcam) for 1 h. After washing three times with 0.1% BSA in PBS, the cells were incubated with a secondary anti–rabbit IgG antibody conjugated to 10-nm gold particles (Ted Pella Inc.) for 1 h followed by washing three times in PBS. Cells were then fixed and prepared for embedding/thin section electron microcopy (). HeLa cells were transfected in triplicate for each construct. After transfection (24 h), the cells were incubated in Met/Cys-free media for 45 min followed by incubation in medium containing 50 μCi [S]Met/Cys (MP Biomedicals) for 1 h. After two rinses with ice-cold PBS, one well was incubated in 4 μg/ml proteinase K (Sigma-Aldrich) in KHM buffer (110 mM KOAc, 20 mM Hepes, pH 7.4, and 2 mM MgCl) for 45 min. Another well was permeabilized with 24 μM of ice-cold digitonin in KHM for 15 min followed by 4 μg/ml proteinase K digestion in KHM for 45 min. The third well was incubated with 4 μg/ml proteinase K for 45 min in 0.5% TX-100/KHM. Subsequently, PMSF was added to all wells to a final concentration of 40 μg/ml. The first two wells were gently washed in KHM buffer with 40 μg/ml PMSF again to remove excess proteinase K. Cells were lysed in 0.4% SDS, 2% TX-100, 400 mM NaCl, 50 mM Tris-HCl, pH 7.4, 40 μg/ml PMSF, 1 mM DTT, plus 2 μg/ml pepstatin A and 1 μg/ml leupeptin and passed through a 23-gauge needle five times before centrifugation for 10 min at 16,000 . Soluble proteins were immunoprecipitated with rabbit anti-myc or GFP by protein A–Sepharose. After three washes, proteins were incubated with sample buffer before separation by SDS-PAGE. Gels were stained with Coomassie brilliant blue and incubated with Amplify (GE Healthcare) for 20 min before drying. Autoradiographs were obtained from dried gels. In parallel, a nontransfected cell lysate was analyzed by Western blotting to validate the permeabilization and digestion conditions. In vitro translations and proteinase K digestions were performed as described previously (). To observe nuclear export, an NES-GFP-NLS (NES-MGNELALKLAGLDI and NLS-PKKKRKV) construct was transfected into HeLa cells either alone or with other expression vectors. CRM1-dependent nuclear export was inhibited by a 2-h incubation with 10 ng/ml leptomycin B. To assay nuclear import, HeLa cells were transfected with a glucocorticoid receptor β-galactosidase fusion protein (grβ) 24 h before a 30-min incubation with 10 μg/ml dexamethasone to induce nuclear import of the fusion protein (). Full-length murine Sun1 and human Sun2 were cloned as previously described (; ) and were used as a template for PCR mutagenesis to generate the Sun1 constructs described in this study. Sun1- and Sun2-GFP were cloned into the tetracycline-responsive pcDNA4/TO (Invitrogen). All other GFP-tagged Sun1 constructs were cloned into pEGFP-N1 (BD Biosciences). HA-Sun1, HA-Sun1(Sun3TM), HA-Sun1(1–221), myc-Sun1(221–380), and myc-Sun1(261–380) were all generated in pcDNA3.1. NES-GFP-NLS was created by PCR mutagenesis and inserted into pEGFP-N1. Sun2-RNAi plasmids were obtained from Open Biosystems. Sun1 RNAi was accomplished by cloning the sequences 5′-GATCCGACCGGGATGGTGGACTTTCTCAAGAGAAAAGTCCACCATCCCGGTCTTTTTTGGAAA-3′ and 5′-AGCTTTTCCAAAAAAGACCGGGATGGTGGACTTTTCTCTTGAGAAAGTCCACCATCCCGGTCG-3′ into pSilencer 3.1-H1 neo (Ambion), which were derived from the open reading frame of human Sun1 using the pSilencer expression vectors insert design tool (Ambion). Fig. S1 shows that large regions of Sun1 are sufficient for NE targeting. Fig. S2 shows that human Sun3 is a type II membrane protein that localizes to the NE but not NPCs when expressed in HeLa cells. Fig. S3 shows the NPC association of various Sun1 isoforms and chimeras. Fig. S4 shows that all Sun1 mutants containing the H4 TM domain exhibit the topology of a type II membrane protein. Fig. S5 shows that Sun1 does not contribute to the functionality of the NPC. Online supplemental material is available at .

In eukaryotic cells, all nucleocytoplasmic transport occurs through nuclear pore complexes (NPCs), which are large macromolecular assemblies (∼44 MD in yeast) that span the nuclear envelope (NE; for review see ). NPCs show eightfold rotational symmetry in a plane perpendicular to the NE and are constructed using multiple copies of ∼30 proteins, which are termed nucleoporins (Nups). Remarkably, more than half of yeast Nups are individually dispensable for growth, although strains lacking some are temperature sensitive for growth and nucleocytoplasmic transport. The nuclear pore itself can be divided roughly into three domains: the nuclear basket, the central core, and the cytoplasmic filaments. The basket and cytoplasmic filaments are composed of Nups that are found solely in those structures, whereas most other Nups are localized symmetrically on both the nuclear and cytoplasmic sides of the plane of the NE. Three integral membrane proteins are components of NPCs and have been implicated in both the organization and proper assembly of NPCs. Although genetic and biochemical analyses have advanced the identification of the Nups as well as their localizations and interactions within the NPC, the mechanism of NPC biogenesis is poorly understood. Most nucleocytoplasmic transport is mediated by members of the karyopherin family of receptors. These receptors recognize localization signals in their cargoes and move with their cargoes through the central channel of the NPC. mRNA export is not mediated by karyopherins, and the actual complex exported consists of the mRNA in a complex with proteins, forming a messenger RNP complex. The NPC plays a mechanistic role in transport of molecules between the nucleus and the cytoplasm by providing docking sites for these complexes. FG repeat domains are found in approximately one third of yeast Nups and contain FG repeat domains that have multiple copies of GLFG, XFXFG, or XXFG separated by spacers rich in polar amino acids. Structural studies indicate that FG domains are natively unfolded and are able to bind karyopherins and karyopherin–cargo complexes. It is not known how these complexes selectively penetrate the FG repeat milieu of the NPC channel (). Screening the collection of ∼4,500 yeast strains each disrupted for one nonessential gene led to the observation that cells lacking Apq12p have defects in both nuclear 3′ pre-mRNA processing () and mRNA export (). Apq12-GFP localizes to the nuclear periphery and the ER, but it is not a Nup because its distribution is unaffected by mutations that cause NPCs to cluster in one or a few regions of the NE (). More recently, Apq12 was postulated to have a role in cell division, as loss of led to synthetic growth defects when combined with mutations affecting genes coding for spindle pole body (SPB) proteins and other proteins involved in cell division. When Apq12p was not present, anaphase was delayed, and re-replication of DNA before completion of cytokinesis was also observed (). Because of our interest in mRNA biogenesis and export, we examined how the absence of Apq12p affected various aspects of nucleocytoplasmic transport. We report that Apq12p is an integral membrane protein found within the NE and in the ER (). In addition to a partial block in mRNA export, the absence of Apq12p led to cold-sensitive defects in the growth and localization of a subset of Nups, particularly those asymmetrically localized to the cytoplasmic fibrils. In addition, cells lacking Apq12 displayed defects both in NPC biogenesis and in the morphology of the NE. We suggest that these defects are caused by alterations in the dynamics and properties of the NE because the proper localization of Nups in cells was restored upon addition to the medium of benzyl alcohol (BA), which is thought to increase the fluidity and flexibility of membranes (; ). Thus, it is likely that the reported defects in mRNA export, pre-mRNA processing, and cell division of cells are indirect consequences of altered membrane dynamics. Collectively, our results demonstrate the dependence of NPC biogenesis and function on the physical properties of the nuclear membranes. A previous study demonstrated that Apq12 localizes to the NE/ER and that the protein is not associated exclusively with NPCs (). Hydropathy analyses using the program TMHMM () revealed two predicted transmembrane domains (amino acids 40–62 and 69–91). To determine biochemically whether Apq12 is a transmembrane protein, a lysate was prepared from a strain that produces a C-terminally tagged Apq12p-GFP from the genomic locus. Lysates were treated with either Triton X-100, high pH, or buffer alone and subsequently separated into supernatant and pellet fractions by centrifugation. Transmembrane proteins will be found predominantly in the pellet fraction after treatment with high pH or buffer alone. In contrast, high pH treatment causes the release from the membrane of peripherally associated proteins, and they will be found in the supernatant. Immunoblotting of the different fractions with α-GFP antibodies revealed that Apq12-GFP remained in the pellet fraction during high pH treatment and only shifted to the supernatant when lysates were treated with detergent (). As controls, Sec23, a peripherally associated ER protein, was found in the supernatant fraction after high pH treatment, whereas the integral membrane ER protein Bos1, like Apq12-GFP, remained in the pellet fraction after high pH treatment. These results prove that Apq12 is an integral membrane protein. Although is not essential, cells lacking it grow more slowly at 23°C than do wild-type (WT) cells (). We compared the growth behavior of and WT at 16, 23, 30, and 37°C. cells grew as well as WT at both 30 and 37°C but were cold sensitive and barely able to grow at 16°C. Notably, the previously described apical cell morphology () and defect in mRNA export (; ) were observed in cells grown at 23°C but were not seen in cells maintained at 37°C (see ). Because the deletion of led to defects in mRNA export, we investigated whether there were genetic interactions between and mutations affecting genes required for nucleocytoplasmic transport. We crossed the strain with haploid strains harboring deletions of genes encoding nonessential Nups or ts alleles of genes encoding essential Nups and mRNA export factors, including , , , , and . Heterozygous diploids were sporulated, tetrads were dissected, and haploid progeny were scored for the presence of both mutations. Growth of double mutant haploid strains was analyzed at temperatures ranging from 23 to 37°C. Of the five mutants listed above, had the most severe effect on the growth of (). Nup120 is a nonessential structural component of the NPC (). Strong synthetic growth defects were seen with and , but no enhanced growth defect was seen when was combined with . Mex67 is the mRNA export receptor and mediates interactions between the messenger RNP and NPCs during mRNA export (). We expanded our genetic analyses to include additional strains in which a gene encoding a nonessential Nup was disrupted. The data are summarized in . Of the others tested, the most severe growth defects were seen when was combined with disruptions of , , , , and . A strong growth defect was also seen in an strain. Ndc1 is an essential integral membrane protein and is the only protein found in both NPCs and the SPB. Note that little or no defect was seen when was combined with either or . Nup2 and Nup60 are components of the nuclear basket of the NPC. Together, these results suggest that Apq12 is important for NPC function or biogenesis. Because of these genetic interactions, we examined the localization of several Nup-GFP fusion proteins in cells at 23°C (). We also examined the localization of Nup159/Rat7 and Pom152 by indirect immunofluorescence (IF) using antibodies directed against each protein. Nuclear basket components Nup1 and Nup60 were not mislocalized in cells nor were Sac3 or Mlp1, two proteins that associate with nuclear basket Nups (; ). Similarly, we observed little or no mislocalization of Nup170, Gle2, Nsp1, or Nup57 in cells, all of which are thought to be components of the central structural framework of the NPC. Normal localization was also seen for two integral membrane Nups, Pom152 and Ndc1. Although Nup188, Nup49, and Nic96, which are also core components of the NPC, were not entirely lost from the nuclear periphery, there were subtle differences in their localization in cells compared with WT. For example, in cells lacking Apq12, there were studs or bright foci of Nup188- and Nup49-GFP distributed around the nuclear periphery in contrast to their relatively uniform punctate distribution around the nuclear periphery in WT cells. In contrast to the majority of nuclear basket and core Nups, we observed dramatic defects in the localization of all Nups that are components of the cytoplasmic filaments of the NPC. In the absence of Apq12, Nup42/Rip1, Gle1/Rss1, Nup82, and Nup159/Rat7 mislocalized to foci. Some foci were adjacent to the nuclear periphery, and others were cytoplasmic and appeared to be completely detached from the NE/NPC. We used Western blotting to compare the levels of several Nups in and WT cells and saw no differences in those analyzed (Fig. S1, available at ). In addition to Nup mislocalization, we observed the dramatic mislocalization of Dbp5/Rat8-GFP, an essential mRNA export factor that shuttles between the nucleus and cytoplasm and performs key functions during mRNA export when bound to the cytoplasmic filaments of the NPC (, ). To ensure that the mislocalization observed did not result from a synthetic growth defect caused by the presence of a Nup-GFP fusion and the absence of Apq12, we compared the growth of cells with several strains expressing Nup-GFP fusions. All grew at similar rates (unpublished data). Furthermore, using antibodies directed against the GLFG repeats, we also saw the mislocalization of Nups recognized by this antibody as well as an overall decrease in staining at the nuclear periphery (unpublished data). Because cells grow as well as WT at 37°C (), we analyzed at 37°C both mRNA export and the localization of Nups that were mislocalized at 23°C. At this optimal growth temperature, the localization of the cytoplasmic filament Nups (Nup159/Rat7 and Nup82) was normal, and no defect in mRNA export was seen (; also see ). Both Nup60- and Nup170-GFP were properly localized in cells at both 23 () and 30°C (). We shifted WT and cells expressing Nup60- or Nup170-GFP from 30 to 16°C and examined their localization after 1 and 2 d. Both Nup170- and Nup60-GFP became abnormally distributed in cells while retaining normal distribution in WT cells (). Punctate Nup-GFP foci were seen, and the fluorescent signal became more diffuse, with a notable increase in the intranuclear signal. Thus, even Nups whose distribution was normal in slow growing cells at 23°C became mislocalized in cells shifted to a more restrictive temperature (16°C). In some cells, the nucleus itself became misshapen at 16°C (), which is consistent with cold-sensitive defects in the NE. Collectively, the data indicate that Nup localization defects increased in severity and extent as the temperature was reduced. Because Apq12 localizes to the NE and cells lacking Apq12 have NPC defects, we examined the NE by light and electron microscopy. Many ER proteins are present in the outer nuclear membrane (ONM) because the two are continuous. Therefore, we assayed for nuclear membrane deformities by examining the distribution of a GFP fusion to Sec63, a resident ER protein (), and also by staining live cells with 3,3′- dihexyloxacarbocyanine iodide (DiOC), a fluorescent lipophylic dye that permits easy visualization of ER and nuclear membranes in yeast (). In WT cells maintained at 23°C, both Sec63-GFP () and DiOC staining () formed continuous rings surrounding the nucleus. In cells, there were several abnormalities, including membranous divisions within the nucleus, extra protrusions of ER membrane not normally seen in WT cells, and studs of fluorescent signal adjacent to the NE. To gain further insight into the defect, we performed electron microscopy to examine NE ultrastructure (). In WT cells, NPCs appear as electron-dense material extending from the inner nuclear membrane (INM) to the ONM (, asterisks). We observed a range of defects in both NPCs and the NE in cells. In cells grown overnight at 23°C, >90% of NPCs examined contacted only the INM (, arrows). Groups of NPCs associated with the NE were seen in some cells (, arrow), and these probably correspond to the bright fluorescent foci seen in and with some Nup-GFPs. Invaginations and/or extensions of the NE () were seen in 30–40% of the nuclei examined. However, because cells are able to grow at 23°C, cells must have some normal nuclear pores, and these were seen, although rarely (unpublished data). We also observed many cases in which large electron-dense inclusions extended into the lumen of the NE (, arrow). Sometimes, these were located entirely within the lumen (, arrow). Because growth at 37°C prevented the Nup82 localization defect seen at 23°C, we also performed electron microscope on cells grown continuously at 37°C. Although NPCs that were unable to associate with the ONM were still seen (unpublished data), >85% of NPCs appeared normal. Protrusions of the NE that contained electron-dense material (, asterisks) were still seen in ∼50% of nuclei examined, and only rarely did these contain multiple inclusions, which is in contrast to what was seen at 23°C (). More than 200 NPCs were examined in determining these percentages. Nuclear membrane herniations and membrane seals covering some NPCs are phenotypes associated with the deletion of nonessential (). Because cells also showed membrane seals covering some NPCs, we compared the localization of Nup159/Rat7 in and cells. The data in show that there was a much more severe defect in the localization of Nup159/Rat7 in than in . We also constructed an double mutant strain, and none of the morphological defects seen in either single mutant were exacerbated by combining the mutations (unpublished data). However, the double mutant strain grew considerably less well than either single mutant at 23 or 30°C and did not grow at 37°C (). Although there were no readily observed morphological defects in the peripheral ER in cells (), we analyzed cells for defects in ER function because Apq12 is present throughout the ER. No defect was observed in the maturation of carboxy-peptidase Y, a recognized cargo of ER protein transport that is modified within the Golgi and is ultimately delivered to the vacuole (Fig. S2 A, available at ). The accumulation of unfolded proteins in the ER activates the unfolded protein response (UPR; ). In yeast strains defective in ER protein trafficking and secretion, cells typically display a constitutively active UPR. The UPR leads to the activation of several genes that encode proteins needed for addressing the increased level of unfolded proteins in the ER. A conserved DNA sequence element (the UPR element [UPRE]) is a feature of promoters activated by the UPR (; ). Thus, the expression of GFP from a promoter containing a UPRE has been used as a measure of the extent to which the UPR has been activated in different genetic backgrounds (; ). We saw no induction of a UPRE-GFP reporter in cells, although the reporter was activated in the control strain (Fig. S2 B). Mislocalization of Nups could reflect defects in NPC biogenesis, NPC stability, or both. We took advantage of the cold sensitivity of cells to ask whether Apq12 is required for proper NPC biogenesis. First, we determined how the distribution of Nup82-GFP changed over time in cells shifted from 37 to 23°C. We diluted cells grown overnight at 37°C to restore exponential growth and shifted them to 23°C. Mislocalization of Nup82-GFP was detectable but minimal 2 h after the shift (unpublished data) and was complete by 8 h (). No mislocalization was seen in cells maintained at 37°C. To distinguish between defects in NPC assembly and stability, we treated cells with the translation inhibitor cycloheximide to prevent the synthesis of new Nups. We reasoned that if mislocalization did not occur in cycloheximide-treated cells, this would indicate that NPCs produced at 37°C were stable at 23°C and that the cytoplasmic foci of GFP-tagged Nups resulted from a defect in NPC assembly. This approach was used earlier by , ) to investigate the genetic requirements for yeast NPC biogenesis. Cells were grown overnight at 37°C, diluted to restore exponential growth, and shifted to 23°C with or without the addition of cycloheximide. As shown in , the normal localization of Nup82-GFP was retained in cycloheximide-treated cells shifted to 23°C but not in the untreated control. We conclude that cells are defective in NPC biogenesis. As another approach to examine whether Apq12 plays a role in NPC biogenesis, we examined the localization of Nup159/Rat7 (using anti-Nup159/Rat7 antisera) and the various Nups (including Nsp1) recognized by the RL1 monoclonal antibody () in an double mutant strain (). In cells, NPCs cluster to one area within the NE (), and both antibodies recognized these clusters (; see merge in the row). In contrast, in cells, RL1 antibody stained the nuclear periphery as well as cytoplasmic foci, presumably as a result of the antibody recognizing some cytoplasmic filament Nups. In the double mutant, very few cells had NPC clusters, and the mislocalization defects seen in were enhanced: cytoplasmic foci (stained using RL1) were much brighter in than in cells, indicating that additional Nups became mislocalized in double mutant cells, and some accumulated in cytoplasmic foci. Although both Nup159/Rat7 and Gle1/Rss1 are located asymmetrically on the cytoplasmic side of NPCs, a direct interaction between them has not been detected. However, indirect IF indicated that these two Nups colocalized to cytoplasmic foci in cells (unpublished data). Further investigation of these foci using indirect IF also demonstrated the colocalization of Nup82-GFP with Nup159/Rat7 and with a fraction of Nup170-GFP (Fig. S3, available at ). We suggest that the foci contain Nup subcomplexes that are unable to be assembled into NPCs. The aforementioned results suggest that the observed mislocalization of some Nups and the defect in NPC biogenesis might result from altered physical properties of the NE. Cold sensitivity is a phenotype sometimes observed with defective membrane proteins (; ), and cells are cold sensitive for the growth and proper organization of NPCs within the NE. One mechanism cells use to cope with a reduction in growth temperature is to modify the composition of membranes by increasing the abundance of phospholipids that contain shorter and unsaturated acyl chains (). This allows the maintenance of membrane fluidity and flexibility at lower temperatures. Alterations in the protein components of membranes may also contribute to the maintenance of normal membrane properties after temperature shifts. Recently, it was demonstrated that ethanol-induced changes in fluidity of the NE in yeast cells caused defects in both NPC organization and nuclear transport (). This suggests that the -associated phenotypes might be caused by the mutant strain's inability to modulate membrane composition at lower temperatures. To test this hypothesis, we added low levels of BA to the media to increase membrane fluidity (; ). cells expressing Nup82-GFP were incubated overnight at 37°C and shifted to 23°C for 10 h with or without 0.4% BA. In cells cultured at 23°C with BA, the mislocalization of Nup82-GFP was prevented (). This suggests that cells are defective in adjusting the composition of the NE so as to maintain proper flexibility or fluidity at lower growth temperatures. We extended this experiment by adding BA to cells that had been grown overnight at 23°C so that Nup82-GFP was already mislocalized when BA was added (). Within 4 h of BA addition, cells had regained the normal localization for Nup82-GFP at the nuclear periphery, suggesting that modifying membrane properties could correct Nup mislocalization. This normal pattern was also restored if cycloheximide was added simultaneously with BA. This indicates that restoration of the normal pattern reflects assembly into NPCs of Nups or Nup complexes already present and mislocalized in cells grown overnight at 23°C. Cycloheximide added without BA had no effect on previously mislocalized Nup82-GFP. We repeated this assay with an strain expressing both a Nup82-GFP fusion and a Sec63-RFP (fusion from their respective chromosomal loci) so that NPC localization and ER morphology could be monitored simultaneously. As shown in Fig. S4 A (available at ), the morphology of the NE was restored to near normal, but some defects in Sec63-RFP localization were still observed. We next wanted to determine whether the properly localized NPCs in cells treated with BA were functional. To do this, we examined mRNA export in BA-treated cells. Only a low percentage of these cells exhibited a defect, suggesting that BA treatment had enhanced the formation of functional NPCs in a cells (). Likewise, BA corrected the mislocalization of Nup188-GFP in cells (Fig. S4 B). Because extended exposure to 0.4% BA had deleterious effects on the growth of WT cells, we were unable to analyze the ability of BA to suppress the cold-sensitive growth defect of apq12Δ cells. Nup mislocalization has been observed in other mutant yeast strains. Both and cells accumulate Nups in cytoplasmic foci, similar to what was seen with . was identified in a screen for cold-sensitive mutants defective in mRNA export and, like , encodes an integral membrane protein of the NE and ER (). is a novel allele of , encoding the guanine nucleotide exchange factor for Gsp1 (Ran), and, unlike other alleles, this mutation is unique in that it causes defects in NPC biogenesis. In contrast to other mutant alleles, does not affect nuclear transport in general (). The addition of BA was able to partially correct for the mislocalization of Nic96-GFP and Nup170-GFP in the G282-S strain (Fig. S4 C). Little change was seen in the location of Nup159 in cells shifted to 16°C (Fig. S4 D). The observation that the addition of BA suppresses the NPC biogenesis and mRNA export defects of cells suggests that the properties and composition of the NE are affected directly by the absence of Apq12. The allele of (encoding acetyl-CoA carboxylase) is known to have defects in the lipid composition of its membranes. Acc1/Mtr7 catalyzes the rate-limiting step in synthesis of fatty acids, and the mutant was particularly defective in the synthesis of very long chain fatty acids. We constructed a strain carrying both and and examined its growth at several temperatures (). At 23°C, the double mutant strain grew less well than either single mutant, but at both 30 and 34°C, the double mutant strain showed enhanced growth in comparison with the single mutant. At 37°C, both the double mutant and cells could not grow. The enhanced defects at lower temperature suggest that double mutant cells are less able to respond to a reduction in temperature and support the hypotheses that both Apq12 and Acc1/Mtr7 affect the composition and, therefore, the properties of the NE. Interestingly, at 23°C, the double mutant had a more severe mRNA export defect than did cells (unpublished data). During interphase, the nucleus maintains a relatively stable morphology, which is in contrast to the dynamic changes it undergoes in location and structure during cell division. These alterations can be observed easily by staining live cells with DiOC during various stages of the cell cycle (). cells are known to have cell shape and cell division defects (). We wondered whether these phenotypes might be related to and were a consequence of defects in the NE and NPCs. Nup170 is a structural component of NPCs (). Much like Apq12, Nup170 has also been implicated in playing a role in cytokinesis, as the mutation of leads to defects in chromosome segregation and kinetochore integrity (). Cells carrying deletions of both and grew considerably less well at 23°C than either single deletion mutant, and synthetic lethality between and was seen at 37°C (). Because the absence of either Apq12 or Nup170 leads to defects in cell division and nuclear transport/structure, we saw this dramatic synthetic growth defect as an opportunity to determine whether or not there was interdependence between the two defects. We assayed both single mutants and the double mutant for defects in mRNA export and Nup159/Rat7 localization. Surprisingly, double mutant cells were no more defective for mRNA export at 23°C than cells (), and the mislocalization of Nup159/Rat7 seen in cells was also reduced in double mutant cells at 23°C (unpublished data). We next assessed whether or not the deletion of enhanced the cell division defects of . shows that double mutant cells were morphologically more defective than either of the parental mutants, and acquired a pseudohyphal appearance. DiOC staining of double mutant cells showed that the deletion of enhanced the nuclear dynamic defects during division, and many cells appeared to have multiple nuclei (unpublished data). Collectively, the data indicate that the deletion of enhanced the nuclear membrane and cell division defects of cells, had relatively little effect on the mRNA export defect of cells, and partially suppressed the Nup mislocalization defect of cells. These results suggest that the defects in cell division and NPC function do not depend on each other and are both consequences of NE abnormalities. Because of our long-standing interest in mRNA export, the reported defect in mRNA export in cells (; ), and the synthetic lethality we observed between and , we decided to investigate the function of Apq12. Apq12 is an integral membrane protein () present in both the ER and NE (; ). showed synthetic growth defects when combined with the deletion of any one of several nonessential Nups and with ts alleles affecting several essential Nups (). The absence of led to defects in NPC biogenesis and the mislocalization of several Nups, including those that comprise the cytoplasmic filaments of the NPC (Nup159/Rat7, Nup82, Gle1/Rss1, and Nup42/Rip1; ). Dbp5/Rat8, which binds to these filaments, was also mislocalized in cells. Nups that were not NPC associated in cells could be detected in foci that contained multiple Nups and Dbp5/Rat8 ( and not depicted). Strikingly, the Nups in these aggregates retained the ability to be incorporated into NPCs, as the addition of BA restored their localization to NPCs under conditions in which no new Nups could be synthesized (). Yeast strains and plasmids used for these studies are listed in Table S1 (available at ). All strains were grown and media was prepared using standard methods (). For growth assays, strains were grown overnight and diluted back to OD = 0.3. Strains were then serially diluted 1:10, and 3 μl were plated of each dilution. Plasmid pCSNup49-GFP-1 was created by the digestion of GFP-NUP49 plasmid (gift of V. Doye, Institut Curie, Paris, France) with SacI and BamHI and ligation into YIplac211. The plasmid was then linearized for genomic integration with SwaI. All live cell fluorescent microscopy was performed using cells grown and mounted in synthetic complete plus dextrose media (). Images were acquired using a microscope (TE2000-E; Nikon) fitted with a 100× NA 1.4 plan Apochromat oil objective (Nikon), CCD camera (Orca-ER; Hamamatsu), and Phylum Live software version 3.5.1 (Improvision). GFP and RFP were visualized by using an X-cite 120-UV lamp and Chroma filter sets. Images were processed using Photoshop 7.0 (Adobe). DiOC staining was performed as previously described (). In brief, cells in midlog phase were stained with 1 μg/ml DiOC (Invitrogen) using a 0.1-mg/ml ethanol stock. Indirect IF using α-Nup159/Rat7, α-Gle1/Rss1, or α-RL1 and FISH were performed as described previously (; ). α-Nup159/Rat7 antibody was used at a 1:3,000 dilution, and α-RL1 (Affinity BioReagents, Inc.) was used at a 1:500 dilution. Indirect IF using α-Pom152 antibodies () was performed as described previously () using a 1:2 antibody dilution. 100 ml Apq12-GFP cells were grown to an OD = 0.6 and were used for semiintact cell preparation. In brief, cells were washed once in 100 mM Tris-HCl, pH 9.4, and 5 mM DTT and resuspended in 10 mL lyticase buffer (0.7 M sorbitol, 0.5% dextrose, 10 mM Tris-HCl, pH 7.5, and 1 mM DTT). 250 μl lyticase was added, and cells were gently rotated for 15 min, at which time the reaction was stopped by adding 90 mL lyticase buffer and spinning down cells. The pellet was resuspended in 1.8 ml lysis buffer (0.4 M sorbitol, 20 mM Hepes, 150 mM KOAc, and 2 mM MgOAc). 35 μl of the semiintact cells were treated with 115 μl of buffer A (20 mM Hepes, pH 7.0, 150 mM KOAc, and 2 mM EDTA), buffer A plus 1% Triton X-100, or 0.1 M sodium carbonate, pH 11.0, for 5 min and spun at 60 K for 12 min. The supernatant was then removed, the pellet was resuspended, and the fractions were analyzed via Western blotting. Electron microscopy was performed as previously described () with some modifications. In brief, the cells were grown to an OD of 0.5–1.0 in YPD media, pelleted, and resuspended in 0.1 M cacodylate buffer, pH 6.8. Primary fixation was performed with 3% glutaraldehyde and 0.1% tannic acid in 0.1 M cacodylate buffer, pH 6.8, at room temperature for 1 h and then overnight at 4°C. Cells were washed twice with 0.1 M cacodylate buffer, pH 6.8, and twice with 0.1 M phosphate buffer, pH 7.5, and treated with zymolyase (10 mg/ml 100T) to produce spheroplasts. After washing with phosphate buffer and cacodylate buffer, pH 6.8, the cells were retreated with 3% glutaraldehyde and 0.1% tannic acid in 0.1 M cacodylate buffer, pH 6.8, for 1 h at room temperature, washed three times with 0.1 M cacodylate buffer, pH 6.8, and embedded in low melting temperature agarose (SeaPrep; FMC Corp.). Postfixation was performed with 2% osmium tetroxide in 0.1 M cacodylate buffer, pH 6.8, for 1 h on ice. Subsequently, the cells were washed in cacodylate buffer, pH 6.8, and deionized water, en bloc stained with 0.5% uranyl acetate overnight, dehydrated with ethanol, and embedded in Spurr's resin (medium grade). Thin sections were cut on an ultramicrotome (MT5000; Sorvall) with a section thickness of 100 nm. Sections were poststained with uranyl acetate and Venable and Coggleshell's lead citrate and examined on a transmission electron microscope (JEM 1010; JEOL) at 100 kV. Fig. S1 shows growth assays for some of the double mutants whose growth behavior is summarized in Table I and a comparison of the protein levels of various Nup-GFP fusion proteins in WT and cells. Fig. S2 documents that cells do not have an ER-trafficking defect, as analyzed by comparing both the maturation of carboxypeptidase Y and the induction of the UPR in WT and cells. Fig. S3 contains fluorescence micrographs that document the degree of colocalization of multiple Nups that mislocalize to cytoplasmic foci in the absence of Apq12. Fig. S4 shows how BA affects the distribution of Sec63-RFP and Nup82-GFP in cells, Nup188-GFP in cells, Nup170-GFP and Nic96-GFP in cells, and Nup159/Rat7 in cells. Table S1 lists the strains and plasmids used in these studies. Online supplemental material is available at .

Covalent attachment to polypeptide modifiers is a common means of regulating protein function (). Ubiquitin is the prototypical member of this group of small protein modifiers, called ubiquitin-like proteins (Ubls). Small ubiquitin-related modifier (SUMO) is the most extensively studied Ubl other than ubiquitin itself. SUMO attachment to a protein can change its distribution, activity, and/or binding partners, and sumoylated proteins function in processes as diverse as cytokinesis, transcription, DNA repair, and chromosome segregation (; ; ). Whether or not a given substrate becomes sumoylated is influenced by a variety of factors, including subcellular localization of the substrate and enzymes responsible for attaching and removing SUMO from the substrate, phase of the cell cycle, redox state, or DNA damage (; ; ; ). The levels of sumoylated substrates reflect a balance between rates of SUMO conjugation and deconjugation. Conjugation of SUMO to proteins requires a series of enzymes related to the E1, E2, and E3 enzymes that activate and transfer ubiquitin to its substrates. SUMO protein modification is highly dynamic. Removal of SUMO from proteins as well as SUMO precursor processing requires specialized proteases called SUMO proteases or desumoylating enzymes. The known desumoylating enzymes are conserved from yeast to humans, forming part of a clan of specialized cysteine proteases (). In yeast, SUMO (encoded by the gene) is cleaved from its substrates by one of two desumoylating enzymes, Ulp1 or Ulp2 (Smt4) (, ). Ulp1 is also the primary SUMO precursor processing enzyme. Many of the components of the SUMO pathway, including Ulp1, are essential for viability. Protein–SUMO ligation can be regulated by alteration of the substrate, such as phosphorylation, or by alteration of the SUMO-modifying enzymes, such as the control of E1 and E2 activity through redox signaling (). Regulation of SUMO protease activity is less well understood. Subcellular localization appears to be a key determinant. Ulp2 and its known substrates are found within the nucleus (; ). In contrast, Ulp1 is localized primarily to the nuclear pore complex (NPC) (; ), and this localization is crucial for proper control of protein desumoylation (; ; ). Sequestration of Ulp1 at NPCs appears to prevent it from desumoylating Ulp2 substrates in vivo inasmuch as it can readily cleave SUMO from many of these proteins in vitro (, ). The NPC is a large protein complex that spans the nuclear envelope and allows the regulated passage of macromolecules between the nucleus and cytoplasm (). It also helps organize chromatin and various protein complexes to facilitate gene expression, DNA repair, and other nuclear functions (; ). Ulp1 is among a small group of proteins that concentrates at a subset of NPCs: unlike most NPC proteins, it is largely excluded from the nuclear envelope region abutting the nucleolus (). Ulp1 associates with the NPC through its noncatalytic N-terminal domain, which includes a coiled coil and distinct binding sites for two karyopherin nuclear transport factors. The upstream karyopherin-binding site in Ulp1 interacts with Kap121 (Pse1), a karyopherin involved in mRNA export from the nucleus (). The second site binds Kap60-Kap95, the karyopherin α-β heterodimer responsible for import of proteins bearing a classical nuclear localization signal (NLS). Removal of a single karyopherin-binding element affects Ulp1 localization only moderately, whereas removal of both elements leads to an enzymatically active C-terminal fragment that localizes throughout the nucleus and can be toxic (; ; ). Other proteins that contribute to Ulp1 localization are the myosin-like proteins Mlp1 and Mlp2 and the nucleoporin Nup60 (). In the absence of Mlp1/2 or Nup60, levels of Ulp1 are significantly decreased and the remaining protein, while still partially localized to the NPC, is no longer excluded from the nucleolar region, indicating a change in Ulp1–NPC interaction (). Nup60 and Nup1 are the only nucleoporins (Nups) that are found exclusively on the nuclear side of the NPC, and they contribute to a filamentous structure called the nuclear basket, which extends into the nucleoplasm (; ). The nuclear basket provides binding sites for many proteins entering and exiting the NPC. Notably, Nup60 binds the mobile nucleoporin Nup2 that is responsible for recycling Kap60 (karyopherin α) from the nucleus back to the cytoplasm (; ; ). Mlp1 and Mlp2 are also associated with the nuclear basket. Mlp1 has recently been linked to a novel RNA quality-control pathway that prevents leakage of unspliced pre-mRNA from the nucleus (; ). Normally, incompletely spliced mRNA precursors are retained in the nucleus until splicing is completed; failure to do so can lead to translation of aberrant proteins (). Cells lacking Nup60 are also defective in this pathway. In this study, we demonstrate that the nuclear envelope protein Esc1 is required for the proper assembly and localization of the nuclear basket. Esc1 is a large coiled-coil protein that associates with the inner nuclear membrane but is not part of the NPC (; ). Esc1 binds and localizes the silencing factor Sir4. Genetic loci subject to chromatin-mediated silencing, such as telomeres, localize to the nuclear periphery and bind a complex of Sir (silent information regulatory) proteins (). Concentration of Sir4 at the nuclear periphery can facilitate silencing. Sir4 is anchored at the nuclear membrane by binding either Esc1 or another protein complex, Yku70-Yku80; these proteins play redundant roles in tethering Sir4 and silent chromatin to nuclear envelope sites, and both contribute to telomeric silencing (; ). We find that Esc1, by controlling nuclear basket assembly, also modulates Ulp1 localization at the nuclear envelope. In cells, Nup60 and Ulp1 colocalize in aberrant, bright-staining foci at the nuclear periphery. The and mutations have similar effects on SUMO conjugate accumulation and interact similarly with and mutations. Deletion of suppresses , but this does not require Sir4 or the peripheral anchoring of silent chromatin. Collectively, our data show that Esc1 is essential for normal nuclear basket assembly and helps functionally segregate the Ulp1 and Ulp2 proteases. Loss of Esc1 also causes a defect in the retention of unspliced pre-mRNA in the nucleus similar to that seen in cells. The function of Esc1 and Mlp1 in preventing such aberrant RNA export is genetically linked, and in cells, Mlp1 is mislocalized to perinuclear aggregates. Notably, Ulp1 is also required for normal nuclear pre-mRNA retention, by a mechanism genetically linked to Esc1 and Mlp1. These results suggest that Ulp1 activity at the NPC is integral to the surveillance of mRNA export. Deletion of the Ulp2 SUMO protease causes a variety of defects, including slow growth, sensitivity to high temperature, and permanent cell cycle arrest in response to DNA or spindle damage (). Previous studies had shown that defects could be suppressed by mutations in other SUMO pathway components, including mutations in the Ulp1 SUMO protease that decrease its activity or prevent it from concentrating at the NPC (, ). Ulp1 mislocalization may allow it to gain access to substrates that are normally desumoylated only by Ulp2. To identify additional proteins that are components or regulators of the SUMO pathway, including factors required for the proper localization of Ulp1, we devised a genomic screen based on suppression of . The screen was performed using the synthetic genetic array (SGA) method in which the mutant of interest, , was crossed to the ordered array of ∼4,800 nonessential gene deletion mutants assembled in the Genome Project (). After sporulation, haploid double mutants were selected and screened for suppression of defects. Double mutants were tested under three conditions: high temperature (37°C), DNA damage by hydroxyurea (0.1 M HU), and microtubule destabilization by benomyl (10 μg/ml). The single mutant shows little or no growth under each of these conditions. After a series of secondary screens, a total of 49 potential suppressors were identified. These suppressors affected proteins that participate in a variety of cellular processes, including protein translation, lipid synthesis, mitochondrial energy production and chromatin regulation. We decided to focus on the eight mutants affecting proteins linked to chromatin function. By tetrad analysis, four of the eight had suppressor mutations unlinked to the original deletion allele. Of the remaining four mutants, only showed consistent suppression of defects in all double mutant segregants. To verify that suppression was due specifically to loss of Esc1, a new mutant was constructed in the W303 genetic background and was mated to a congenic strain. The resulting diploids were sporulated and dissected, and tetrads were evaluated for growth. As expected, the single mutant grew poorly or not at all under the tested conditions, and cells grew similarly to wild type (WT) (). cells showed enhanced growth at 37°C and on benomyl when compared with . However, the irregular colony size and slow growth at 24°C and HU sensitivity associated with were not suppressed. Esc1 is one of two proteins that bind and concentrate the silencing factor Sir4 at the nuclear periphery, thereby enhancing telomeric silencing (; ). Were suppression caused by a defect in silencing, then , which causes a complete loss of silencing, should suppress at least as well as . double mutant showed no suppression of the temperature-sensitive (ts) phenotype as compared with the single mutant and even slightly enhanced the growth defect at 25°C (). Thus, loss of silencing does not suppress . It was still possible that delocalization of Sir4 was the source of the suppression. To test this, we inactivated the other factor involved in Sir4 peripheral localization, the Yku70-Yku80 heterodimer. double deletion, Sir4 is diffusely distributed throughout the nucleus (). The mutant cannot grow at 37°C, so suppression of was tested at 35°C, where cells grow similarly to WT (). cells did not show suppression of either the ts growth or benomyl sensitivity of (). In fact, the double mutant grew worse than at 35°C, presumably due to the loss of other functions of Yku80 (). cells did not enhance suppression beyond what was seen with (). We conclude that the role of Esc1 in the SUMO pathway is independent of its function in silencing or Sir4 localization. Our earlier work had shown that mutations in Ulp1 which prevented it from localizing correctly to the NPC could suppress some defects associated with (). Given the position of Esc1 at the nuclear envelope, we examined the possibility that deletion might affect anchoring of Ulp1 to the NPC. Cells lacking were complemented with a low-copy plasmid expressing under the control of the promoter. In WT cells, Ulp1-GFP localized in a broadly distributed punctate pattern around the nuclear periphery but was excluded from the presumptive juxta-nucleolar region, as expected (). Unexpectedly, in cells, most Ulp1-GFP accumulated in a small number of bright foci that still appeared to be at the nuclear periphery (this was also observed with chromosomally integrated ; not depicted). Subsequent experiments confirmed that the foci were on the nuclear envelope, and in many cases, some residual staining could be seen around the nucleus outside of the bright foci (see ). Foci were prominent in many, but not all of the optical sections of nuclei. Therefore, foci frequency was quantified. Nuclei were scored as having foci if they included at least one dense Ulp1-GFP spot on the nuclear periphery. Using this measure, 64% of Ulp1-GFP–expressing cells had foci compared with 28% in WT (). Esc1 does not colocalize with the NPC (; ), raising the question of how Esc1 can influence the distribution of Ulp1, which is mostly NPC bound. The yeast nucleoporin Nup60 localizes strictly to the nuclear side of the NPC and is a major constituent of the nuclear basket; Nup60 is known to be required for proper localization of Ulp1 (). These observations suggested that Esc1 might influence Ulp1 (and Ulp2) function by altering the nuclear basket of the NPC. We first asked whether proper nuclear basket assembly or localization depends on Esc1. In WT cells, Nup60-GFP (expressed from the endogenous locus) localized broadly around the nuclear periphery, as expected (). Strikingly, in cells, Nup60-GFP concentrated in bright foci on the nuclear periphery, very similar to the foci seen with Ulp1 (; also see ). Because Ulp1 was mislocalized in both the and mutants, we asked if and had comparable genetic interactions with mutations in the Ulp1 and Ulp2 SUMO proteases. grew even better at high temperature than did (). The irregular colony growth of , measured at 24°C, was not altered by either or . In contrast, when combined with a ts-allele of , both and caused an enhanced growth defect at the semi-permissive temperature of 33°C (). We do not know why was not identified in our original suppressor screen. Mutation of either the Ulp1 or Ulp2 SUMO protease causes an increase in the sumoylation levels of specific substrates; prominent changes are detectable by anti-SUMO immunoblot analysis of whole cell lysates (; ). If loss of Nup60 or Esc1 were affecting either SUMO protease activity, changes in the cellular SUMO protein pattern might also be seen. However, unlike cells, neither nor displayed a strong increase in the general level of sumoylated proteins, suggesting that both Ulps retained substantial activity in these nuclear envelope mutants (, lanes 1–3, 10–12). On the other hand, weak but reproducible changes in the intensity of several bands were detected (marked by asterisks; some only clearly seen with longer film exposures). Interestingly, the small changes observed in the and SUMO conjugate profiles were generally amplified when these gene deletions were combined with , but not when combined with (). This implied that the majority of detectable sumoylated proteins affected by loss of either Nup60 or Esc1 were Ulp1 substrates, consistent with a primary role for these proteins in Ulp1 localization. The alteration of only a subset of sumoylated proteins and the observation that specific sumoylated species either increased or decreased in abundance suggested that the and mutations changed the ability of Ulp1 to act on particular substrates. No obvious changes to the profile of high-molecular mass SUMO conjugates in cells were observed when either or was introduced, although it is possible that changes to individual substrates might have been obscured by other intensely stained species. Levels of polysumoylated species detected in the stacking gel from -derived extracts (lanes 7–9, 16–18) varied between experiments but were always present. Conceivably, the phenotypic suppression of by loss of either Nup60 or Esc1 might be a consequence of changes in the levels of Ulp1-specific SUMO protein substrates. In summary, the data in and indicate first, that mutations in Nup60 and Esc1 have similar interactions with Ulp1 and Ulp2 mutations, consistent with a close link between Esc1 and nuclear basket function, and second, that Ulp1 and Nup60 both accumulate in similar structures when Esc1 is deleted. These results suggest that mislocalization of Ulp1 likely underlies the genetic interactions of and with mutations in the Ulp2 protease. Loss of Esc1 caused both Nup60 and Ulp1 to concentrate in bright foci at the nuclear periphery. If Esc1 was affecting Ulp1 through changes in Nup60 assembly or localization, then Nup60 and Ulp1 should colocalize to the same foci in the mutant. This was investigated in cells coexpressing Nup60-CFP and Ulp1-YFP fusion proteins (). In WT cells, Ulp1 appeared to colocalize with Nup60, as predicted if both are part of, or associate with, the NPC. In cells, all the bright Ulp1-YFP foci also stained strongly with Nup60-CFP. Controls using each of these tags individually showed no fluorescence in the opposite channel (not depicted). Mislocalization of Nup60 to foci in cells could result from either specific aggregation of nuclear basket components and associated proteins or a general mislocalization of the entire NPC. To distinguish between these two possibilities, we examined the localization of Nup49, a core nucleoporin found within the central rings of the NPC (). In an strain with the endogenous locus tagged with either or , localization of Nup49 was perturbed but to a lesser extent than either Nup60 or Ulp1. Specifically, Nup49 showed an increase in the level of bright foci on the nuclear periphery, but it also continued to show a broad punctate localization to the nuclear periphery ( and ). GFP fusions of Nup159 or Nsp1, two other nucleoporins, gave similar results, with Nup159, a component of the cytoplasmic NPC fibrils, showing only modest redistribution (Fig. S1 A, available at ). The degree of mislocalization of Nup60 and Nup49 was quantified, and we found that 58% of cells had Nup60-GFP foci and 39% had Nup49-GFP foci (). The accumulation of other nucleoporins in the -induced foci, albeit not to the same extent as Nup60 or Ulp1, suggested that loss of Esc1 might cause a general clustering of NPCs. To investigate this idea, we examined nuclear pore distribution in electron micrographs of permanganate-fixed cells from WT and cells; the conditions were chosen to enhance visualization of membranes. shows a representative cell of each cell type. No gross nuclear membrane abnormalities were visible in . A limited analysis of pore distribution was performed in which distances between pores were measured for at least eight cells of each strain (180 pores). By comparing the distances between pores in the two strains, the distribution of pores was not significantly altered in the cells (p = 0.29). Therefore, full NPCs appear not to undergo gross redistribution, even though some of their components aggregate into peripheral foci. Future studies using different types of EM sample preparation might identify the foci detected by fluorescence microscopy and allow their ultrastructural characterization. Given the more complete redistribution of Nup60 to foci when compared with other nucleoporins, it was possible that aggregated nuclear basket components (Nup60) misrouted other nucleoporins to the foci by protein–protein interactions that normally occur in NPCs. This hypothesis predicts that eliminating Nup60 from cells would prevent accumulation of core nucleoporins in foci. double mutant, we found that Nup49-GFP was localized normally to the nuclear periphery and no longer accumulated in foci (). The single mutant did not change Nup49-GFP localization, nor did it alter Esc1-GFP distribution (). These results suggest that Esc1 is not directly required for proper nuclear membrane distribution of NPCs, but instead prevents formation of Nup60-containing aggregates that can recruit other NPC components. To examine the ability of Nup60 in foci to bind one of its direct NPC-binding partners, we looked at the localization of Nup2. Nup2 is a mobile nucleoporin that is required for the recycling of Kap60 from the nucleus to the cytoplasm (; ). On the nuclear side of the NPC, Nup2 binds to Nup60. In mutants, Nup2 was diffusely distributed in the nucleus, as expected (). double mutant, it was dispersed within the nucleus, as in the single mutant. This suggests that Nup60 is still able to contact its normal interacting partners in the foci, consistent with the requirement for Nup60 in misrouting nucleoporins such as Nup49 to these foci. cells was diffusely distributed with residual nuclear rim staining (). Nup60 is known to interact with the Kap60-Kap95 heterodimer via Nup2 (; ). As noted earlier, the Kap60-Kap95 complex can bind to Ulp1, so Nup60-dependent Ulp1 localization to foci might be mediated by Nup2. However, we found that deletion of did not reduce localization of Ulp1 to the dense perinuclear foci (), indicating that Nup2 is not essential for Ulp1 localization to these sites. The related Mlp1 and Mlp2 proteins also associate with the nuclear basket (). When we examined Mlp1-YFP localization in cells, the protein concentrated in foci that were qualitatively similar to Nup60 and Ulp1 foci in this mutant (). In cells lacking Nup60, Mlp1-YFP is known to aggregate into 1–2 large perinuclear foci (; ). These foci are fewer in number and larger than those seen in cells (). double mutant, the foci are indistinguishable from the foci in the single mutant, indicating that is epistatic to for Mlp1 localization as well as for Ulp1 localization. Notably, Nup60 localization to the NPC did not depend on Mlp1/2 (Fig. S1 B). Ulp1 has two distinct karyopherin-binding sites and a coiled-coil region in its N-terminal noncatalytic domain (; ). We used Ulp1 derivatives lacking one or more of these elements to determine their importance for Ulp1 localization to foci in cells (). The proteins were fused to a 9-myc epitope tag and expressed from the promoter on low-copy plasmids; the full-length tagged protein has WT function under these conditions (). Levels of Ulp1 were similar in and cells based on immunoblotting (Fig. S3, available at ). and , which were expressed in WT cells because these derivatives lack the essential catalytic domain (UD) and thus are unable to complement lethality. As seen previously, the catalytic domain of Ulp1 was dispensable for NPC localization: the ulp1Δ418-621 protein lacking the UD localized to the nuclear envelope and was recruited to foci in cells (). However, ulp1Δ347-621, which differs from ulp1Δ418-621 by the absence of the coiled-coil domain, while still mostly concentrated at the nuclear periphery, did not form foci in cells. Persistent envelope localization of ulp1Δ347-621 in cells () and of full-length Ulp1 in cells () is consistent with a second, Nup60-independent binding mechanism of Ulp1 to the NPC (; ). cells suggest that localization to foci provides a stringent test for Nup60-dependent NPC localization and that the Ulp1 coiled coil is crucial for this. Ulp1 constructs containing only the karyopherin Kap121 or the Kap60-Kap95 binding site are partially delocalized from the nuclear periphery (). Despite their inability to localize fully to NPCs in cells, Ulp1 derivatives missing one or the other of the karyopherin-binding sites (ulp1Δ2-144 and ulp1Δ150-346) still localized at least partially to foci in cells (). All constructs lacking the coiled coil (ulp1Δ150-403, ulp1Δ346-403, and ulp1Δ347-621) failed to incorporate into foci. Notably, the NPC localization defect of ulp1Δ150-403 in cells was more severe than that of ulp1Δ150-346, suggesting a contribution of the coiled coil to NPC localization in cells as well. Conversely, ulp1Δ2-346, which has only the coiled-coil and the catalytic domain, was diffusely localized, mostly in the nucleus, in both and cells, indicating that the coiled-coil domain, while necessary, is not sufficient for localization to foci or for full localization to NPCs. The myosin-like protein Mlp1 functions in an RNA quality-control pathway that retains unspliced pre-mRNAs in the nucleus (). We asked whether , which causes mislocalization of Mlp1 to perinuclear foci (), also compromises this specific RNA retention mechanism. Using a plasmid (pJCR1) encoding an inefficiently spliced intron-containing reporter designed to allow translation of β-galactosidase (β-Gal) only if the unspliced pre-mRNA is transported to the cytoplasm () (), we found that the relative level of β-Gal activity in cells reached a level similar to β-Gal activity in an mutant (). Interfering with splicing by mutation of the branchpoint sequence (mutBP) in the intron leads to increased leakage of unspliced pre-mRNA from the nucleus in WT cells, and leakage of this mutant pre-mRNA is greatly enhanced in cells (). We observed a similar enhancement of leakage of the splicing-defective message in the mutant (). Using a reporter (pJCR51) in which β-Gal is produced only from the correctly spliced mRNA (), no defect in pre-mRNA splicing was observed in either or (Fig. S2, available at ). When and were combined, the level of aberrant pre-mRNA export in the double mutant did not exceed what was observed in the single mutants, arguing that Esc1 acts in the same RNA surveillance pathway as Mlp1 (). Finally, we asked if mislocalization or reduced activity of Ulp1 might be linked to aberrant pre-mRNA leakage as well. Mislocalization of the active Ulp1 catalytic domain (ulp1-C204) to the nuclear interior () did not cause increased leakage (not depicted). However, cells, which have substantially reduced Ulp1 activity even at permissive temperature (), suffer levels of pre-mRNA leakage comparable to cells (). When the mutation was combined with , the extent of pre-mRNA leakage was not significantly higher than in the single mutants (). These data are consistent with the possibility that aggregation of Ulp1 in foci impairs Ulp1 function and that a reduction of SUMO protease activity at the NPC contributes to the RNA surveillance defect of the and mutants. We note that fusion of the Ulp1 catalytic domain to three different nucleoporins (Nup60, Nsp1, or Nup42; ) failed to suppress pre-mRNA leakage in either or cells (unpublished data). There are various potential explanations for these negative results; for example, elements of Ulp1 in addition to its catalytic domain might be necessary for this pathway. h a v e f o u n d t h a t t h e n o n - N P C p r o t e i n E s c 1 i s r e q u i r e d f o r t h e p r o p e r a s s e m b l y o f N P C n u c l e a r b a s k e t s a t t h e y e a s t n u c l e a r e n v e l o p e . N u c l e a r b a s k e t a s s e m b l y , i n t u r n , i s n e c e s s a r y f o r p r o p e r l o c a l i z a t i o n a n d r e g u l a t i o n o f t h e U l p 1 S U M O p r o t e a s e . M i s l o c a l i z a t i o n o f U l p 1 p a r t i a l l y o v e r c o m e s d e f e c t s a s s o c i a t e d w i t h l o s s o f t h e n u c l e a r S U M O p r o t e a s e U l p 2 . O u r r e s u l t s r e v e a l a n u n e x p e c t e d n e t w o r k o f i n t e r a c t i o n s a t t h e n u c l e a r e n v e l o p e t h a t m o d u l a t e b o t h t h e d i s t r i b u t i o n o f N P C c o m p o n e n t s a n d t h e d y n a m i c s o f S U M O p r o t e i n c o n j u g a t i o n . C o m p a r i s o n t o f i n d i n g s w i t h t h e h u m a n S U M O p r o t e a s e S E N P 2 / h U L P 2 a n d n u c l e a r e n v e l o p e f a c t o r s s u g g e s t s c o n s e r v a t i o n o f t h e s e n u c l e a r e n v e l o p e i n t e r a c t i o n s . F i n a l l y , b o t h E s c 1 a n d U l p 1 a r e r e q u i r e d f o r p r e v e n t i n g a b e r r a n t e x p o r t o f u n s p l i c e d p r e - m R N A s f r o m t h e n u c l e u s , s u g g e s t i n g a n i m p o r t a n t r o l e f o r E s c 1 - d e p e n d e n t n u c l e a r b a s k e t o r g a n i z a t i o n a n d U l p 1 - d e p e n d e n t p r o t e i n d e s u m o y l a t i o n i n t h i s R N A s u r v e i l l a n c e p a t h w a y . The screen was performed as described previously (). The gene deletion library, which was assembled by the Genome Project and contained ∼4,800 nonessential gene deletions, was purchased from Open Biosystems. The deletion strain was purchased as a heterozygote from the American Type Culture Collection. Yeast strains and plasmids used in this study are listed in and . Standard media and techniques were used for the growth and construction of yeast strains (). For fluorescence microscopy, strains were grown in SD minimal media containing 0.04 g/L adenine. . Each was generated by homologous recombination in yeast using cotransformation of a linearized copy of YCplac22-ULP1-9myc along with a PCR product that contained sequences that flanked the deleted sequence and spanned the restriction site used to linearize YCplac22-Ulp1-9myc. and , the plasmid was linearized with MscI, and for , the linearized plasmid was gel-purified from a partial SacI digest. Cells were grown overnight in YPD, diluted to an OD of 0.2 and grown for an additional 4 h at either 30°C or 37°C. A volume of culture corresponding to 2.0 OD equivalents of cells was centrifuged for 30 s at 13,000 rpm. Cells were washed once in 20% TCA (Sigma-Aldrich) and resuspended in 400 μl 20% TCA and an equal volume of glass beads. Cells were vortexed on a Turbo Mix bead beater (Scientific Industries Inc.) for 4 min at 4°C. Supernatant was removed from the beads and centrifuged at 14,000 rpm for 10 min at 4°C. The pellet was washed with 2% TCA and resuspended in SDS gel-loading buffer. One OD equivalent of sample was loaded per lane on a 6–15% gradient polyacrylamide gel. After electrophoresis, proteins were transferred to a PVDF membrane and blocked with 10% nonfat dry milk in TBST (150 mM NaCl, 10 mM Tris- HCl pH 8.0, and 0.1% Tween 20) for 30 min, then incubated with anti-Smt3 (SUMO) antibody (1:5,000 in 1% nonfat dry milk in TBST; ) or anti-PGK 22C5 antibody (1:7,000 in 1% nonfat dry milk in TBST; Molecular Probes) for 2 h at room temperature. Membranes were washed three times with TBST for 10 min each, and incubated with donkey anti–rabbit antibody (1:6,000 in 1% nonfat dry milk in TBST; Amersham Biosciences) or goat anti–mouse IgG (γ1) antibody (1:10,000 in 1% nonfat dry milk in TBST; Roche) for 1 h. Membranes were washed three times in TBST for 10 min each and ECL detection reagents were added as directed by the manufacturer (Amersham Biosciences) and exposed to film. Cells were grown in SD media to an OD of 0.3–0.8, placed on a slide and viewed on a microscope (Axioskop; Carl Zeiss MicroImaging, Inc.) with a plan-Apochromat 100× NA1.4 objective lens at room temperature using Zeiss immersion oil IMMERSOL 518F. Pictures were taken on a Zeiss Axiocam camera using a Uniblitz shutter driver (model VMM-D1; Vincent Associates) and the program Open Lab 3.1.5 (Improvision). For indirect immunofluorescent staining with anti-myc antibodies, cells were grown overnight in SD medium lacking tryptophan. They were then diluted and grown to an OD of 0.6–1.0. Ten ml of culture were collected by centrifugation, resuspended in 1 ml 3.7% formaldehyde, and incubated while shaking at 30°C for 90 min. Cells were pelleted at 2,000 rpm for 2 min, washed twice in Buffer B (0.1 M KPO), and resuspended in Buffer C (1.2 M sorbitol, 0.1 M KPO). Cells were often kept at 4°C overnight at this point. Cells were incubated while rotating at 30°C for 40 min in 1 ml Buffer C with 2.5 μl β-mercaptoethanol (14 M) and 10 μl of Zymolyase 100T (5 mg/ml in Buffer C from Seikagaku Corp. Japan). Coverslips were prepared by incubation for 1 h in 0.1 M HCl followed by ten successive washes in water to a volume of 10× and storage in ethanol. Before cell application, all ethanol was removed and coverslips were treated with poly--lysine 70,000–150,000 M.W. (1 mg/ml in water; from Sigma-Aldrich) for 5 min and washed with water. Cells were then washed twice with 1 ml Buffer C, placed on the coverslips and incubated at room temperature for 7 min. Coverslips were washed once with PBS and then incubated for 10 min in blocking buffer (1× PBS, 5% fetal goat serum [Sigma-Aldrich], 1% BSA, 0.2% NP-40, and 0.002% sodium azide). Coverslips were incubated with anti-myc monoclonal antibody 9E10 (1:1,000 in 1× PBS and 1% BSA; Babco Covance) at 30°C for 90 min. They were subsequently washed four times with 1× PBS and two times with blocking buffer, and then incubated in Alexafluor anti–mouse antibody (1:1,000 in 1× PBS and 1% BSA; Molecular Probes) in the dark at room temperature for 60 min. Finally, coverslips were washed two times in blocking buffer and once in PBS, then placed on slides with Antifade (Molecular Probes). Coverslips were sealed with nail polish and viewed as above. Yeast cells were fixed, stained, embedded and sectioned as described previously (); the procedure was performed by Marc Pypaert in the Yale Cell Biology Electron Microscopy Facility. Cells were visualized on a Tecnai 12 Biotwin (FEI) at 80 kV and images were captured using a Morada CCD camera (Olympus Soft Imaging Solutions) with the help of Dr. Pypaert. Random sections were used for measuring contour distances between pores. For each nucleus, pores were identified as a break in the nuclear membrane, and the relative distances between all nearest pairs of individual pores in a section were measured from the center of one pore along the contour of the membrane to the center of the second one. Distances were normalized to total contour length, and pore–pore clustering was compared between strains using a paired test. Cells transformed with the appropriate reporter plasmid (; ) were grown overnight in sucrose media lacking uracil, diluted to early log-phase the next morning, and transferred to raffinose media lacking uracil. Induction of was achieved by addition of 2% galactose for 2–3 h. The OD of each sample was measured, and aliquots were centrifuged, resuspended in 300 μl lysis buffer (0.6% Triton X-100 [vol/vol], 0.75% ONPG [wt/vol], 2.25% β-mercaptoethanol [vol/vol], and 0.15 M Tris-HCl, pH 7.5), and kept at −80°C overnight (). Samples were incubated at 37°C for varying times, 150 μl 1 M NaCO was added to stop the reaction, and debris was removed by centrifugation at 13,200 rpm. Absorbance at 405 nm was measured. Strains carrying and deletions were from the Genome Project. Three supplementary figures are included as online supplemental data. Fig. . Fig. S2: a β-galactosidase–based splicing reporter assay shows neither nor is defective in splicing a reporter construct (pJCR51) compared with WT. Fig. cells. Online supplemental material is available at .

Alzheimer's disease (AD) affects ∼4.5 million Americans, and this number will continue to grow. By 2050, the number of individuals with AD could range from 11.3 to 16 million (). The pathogenesis underlying AD remains unclear, and it is controversial whether AD results from a primary abnormality in amyloid precursor protein (APP) or deregulation of the inflammatory system (), although these two possibilities are not mutually exclusive. Several lines of evidence implicate abnormal processing of APP, which is cleaved by two enzymes, β-secretase 1 (BACE1) and γ-secretase, to generate excessive amyloid β protein (Aβ), as a potential cause of AD (; ). In the past decade, transgenic mice have been generated that overexpress mutant APP and display Aβ-related lesions (). Many of these mouse models exhibit amyloid plaque-predominant aspects of AD (; ), including Aβ plaque formation, cerebral amyloid angiopathy (CAA), and inflammation, but not τ pathology. The TNF death receptor belongs to the superfamily, which includes >20 cell surface receptors. When the TNF type 1 death receptor (TNFR1) binds to its ligand, TNFα, the ligand–receptor complex triggers apoptotic pathways by recruiting a TNFR-associated death domain protein and/or a Fas-associated death domain protein/mediator of receptor-induced toxicity, two intracellular adaptor proteins (). The receptor-induced multimerization of a Fas-associated death domain protein leads to caspase activation, which causes degradation of specific target proteins, ultimately damaging cell integrity (). To find out whether TNFR1 could have effect on Aβ production as well as APP processing, we specifically chose transgenic APP23 mice in our experiments, which express a mutant APP that results in extensive Aβ plaque formation. Here we show that fewer Aβ plaques and Aβ-related lesions develop in Alzheimer's transgenic mice with genetic deletion of . Detailed analyses showed decreased Aβ generation, less neuronal loss, and alleviated Aβ-related memory deficits. Our data indicates that might be a potential and novel therapeutic target for AD. APP23 transgenic mice (), a mouse model for AD with a plaque-predominant type (; ), overproduce Aβ, Aβ40, and Aβ42, and develop significant amyloid deposits by the age of 14 mo. To determine whether genetic inactivation of delays or diminishes Aβ plaque formation, we crossed APP23 mice with mice lacking the gene ( ; ) to generate APP23/ mice. To evaluate Aβ pathology in the brains, we first used Congo red staining to see if reduced protein aggregation could be observed in the brains of APP23/ mice. Congo red has been shown to be affinitive for binding to fibril proteins enriched in β-sheet conformation, and it is commonly used as a histological dye for amyloid detection (Frid et al., 2006). Results showed much less Congo red staining in APP23/ mice (), indicating that protein aggregation was alleviated in APP23/ mice. To further confirm that the aggregated protein we found was aggregated Aβ peptide, we next examined the brain section with anti-Aβ antibody 6E10 (; ). At 12 mo of age, APP23 mice displayed numerous Aβ plaques throughout the entorhinal cortex and hippocampus, consistent with a previous paper (). In APP23/ mice, however, we found only minor Aβ plaques in the entorhinal cortex () and the hippocampus (). At 24 mo of age, APP23/ mice also displayed reduced plaques (). The number of plaques was reduced by 73 and 80% in the entorhinal cortex () and hippocampus () in APP23/ mice at 12 mo of age, indicating that Aβ pathology in APP23/ indeed alleviated plaque formation compared with age-matched APP23 mice. The size of Aβ plaques also indicates the severity of Aβ pathology (). We used morphometric analyses on the brain sections immunostained with Aβ antibody 6E10. Results showed that both large (>20 μm diameter)- and medium (10–20 μm diameter)-sized Aβ plaques in the entorhinal cortex () and hippocampus () were significantly reduced in APP23/ mice at 12 and 24 mo of age, indicating that in APP23/ mice, Aβ pathology was alleviated not only by reducing the overall Aβ plaque number, but also by decreasing plaque size. CAA has been reported to have both positive and negative correlations with AD pathology (; ; ). It has been shown that CAA in APP23 transgenic mice is strikingly similar to that of human CAA (). To find out whether genetic deletion of can relieve CAA in APP23 mice, the deposition of Aβ in the vascular wall was examined by double immunostaining with antibodies against β-smooth muscle actin (a vascular smooth muscle marker; ) or von Willebrand factor (vWF; an endothelial cell marker; ) and anti-Aβ40 antibody. We found that at 24 mo of age, APP23 mice display CAA predominant in cortical, hippocampal, and thalamic vessels; Aβ40 formed a continuous ring-like shape within the vessel wall (), consistent with . However, there were very few Aβ40 deposits within the vessels of APP23/ (). Deposition of Aβ on the vascular wall could not only increase the vulnerability of cerebral vessels but also increase the possibility of intracerebral hemorrhage (; ; ; ). Our results showed little CAA progression in the brains of APP23/ mice, suggesting that deletion of could reduce the risk of CAA. Microglia activation is also a hallmark of Aβ pathology progression ( ;; ). CD11b and CD45 are two well-characterized markers for microglia activation in the brains (; ; ).To examine whether deletion of could alleviate the massive microglia activation of APP23 mice, we studied the microglia activation in APP23 and APP23/ mice. Consistent with a previous paper (), APP23 mice showed strong immunoreactivity with antibodies against CD11b (Mac-1) and CD45 in the entorhinal cortex () and hippocampus (), indicating that a massive amount of microglia had been activated along with the appearance of Aβ pathology. In contrast, APP23/ mice showed significantly less CD11b and CD45 immunoreactivity in the entorhinal cortex () and hippocampus (), indicating that genetic deletion of alleviated massive microglia activation in APP23 mice. Because we found reduced Aβ pathology in APP23/ mice, our next question was whether genetic inactivation of reduces Aβ pathology by affecting Aβ generation. The Aβ level was analyzed by immunoprecipitation followed by Western blotting ( = 3 for each group). shows a representative urea Western blot () comparing the 4-kD Aβ species ( = 3 for each group). Both Aβ40 and Aβ42 can be detected, and Aβ42 migrated ahead of Aβ40, consistent with a previous paper (). Compared with APP23 mice, APP23/ mice showed a significant reduction in both Aβ40 and Aβ42 levels (). We further measured total Aβ, Aβ40, and Aβ42 levels by sandwich ELISAs ( = 10 for each group). The results bolstered our Western blot findings and confirmed that APP23/ mice have much less total Aβ, Aβ40, and Aβ42 (). Quantitatively, total Aβ decreased by 69% (37.39 ± 12.71 ng/mg) and 30% (463.87 ± 189.83 ng/mg) in 12- and 24-mo-old APP23/ mice, respectively, compared with total Aβ in APP23 mice (12 mo old: 120.80 ± 39.74 ng/mg; 24 mo old: 693.40 ± 270.27 ng/mg; ). Both Aβ40 and Aβ42 were reduced in APP23/ mice. However, the most significant difference is at 12 mo of age, when Aβ40 decreased by 80% in APP23/ mice (20.45 ± 4.7 ng/mg) compared with APP23 mice (103.87 ± 21.81 ng/mg; ), and Aβ42decreased by 70% in APP23/ mice (4.12 ± 1.2 ng/mg) compared to APP23 mice (15.78 ± 4.7 ng/mg; ). These results indicated that reduction in Aβ40 and Aβ42 levels could account for the alleviated Aβ pathology in APP23/ mice. One of the mechanisms that could influence Aβ production is through altering APP holoprotein levels. We next analyzed by Western blot to see if deletion of affects APP protein levels ( = 3 for each group). In contrast to the reduction in Aβ levels, Western blotting did not reveal any differences in full-length APP levels between APP23/ and APP23 mice (), indicating that the decrease in Aβ levels was not caused by altering APP protein expression in APP23/ mice. To examine whether the reduced amyloidosis in APP23/ mice was caused by a reduction in abnormal APP metabolism, we examined the activity and expression levels of one key enzyme in APP processing, BACE1. We first used an MCA-labeled BACE1 substrate (; ) to examine BACE1 activity and found that BACE1 activity was significantly decreased in APP23/ mice (). To find out whether the decreased BACE1 activity was due to a decrease in BACE1 levels, we measured BACE1 levels by sandwich ELISA () and Western blot ( = 3 for each group). We found that BACE1 levels in APP23/ mice were indeed reduced in both Western blot and ELISA results (), indicating that reduced BACE1 activity in APP23/ mice was caused by a reduction in the protein level. To further investigate whether reduced BACE1 protein level is caused by reduced BACE1 mRNA transcription, we performed RT-PCR to measure BACE1 mRNA levels and found that BACE1 mRNA was also decreased in APP23/ mice (), indicating that the genetic deletion of reduced BACE1 mRNA levels and caused BACE1 activity to be down-regulated in APP23/ mice. Our RT-PCR results showed that BACE1 mRNA levels decreased in APP23/ mice; the next question was what signal transduction pathway leads to the decreased BACE1 mRNA level. To clarify how deletion of affects BACE1, we transfected 293 cells with a pB1P-A vector containing a promoter (−1941 to +292) that was upstream of a luciferase reporter gene (; ), and then treated these cells with different concentrations of TNFα. We found that BACE1 promoter activity increased in a concentration-dependent manner (). Blocking the interaction of TNFα with the extracellular domain fragment of TNFR1 inhibited such elevation in BACE1 promoter activity (), indicating that TNFα activates BACE1 promoter through . NF-κB is one of the major mediators of TNFα-activated signaling (; ). A recent finding that multiple NF-κB binding sites are located in the vicinity of BACE1 promoter () suggests that NF-κB may play an important role in regulating BACE1 transcription. To determine whether activates the BACE1 promoter through this pathway, we used the potent NF-κB activation inhibitor 6-amino-4(4-phenoxyphenylethylamino) quinazoline () to block NF-κB signaling in TNFα- treated 293 cells transfected with a BACE1 promoter luciferase reporter vector. Treating a pB1P-A transfected cell with an NF-κB inhibitor significantly reduced TNFα-induced BACE1 promoter activity (). A high concentration of NF-κB inhibitor inhibited not only TNFα-induced BACE1 promoter activity but also basal promoter activity (), indicating that NF-κB may play a central role in regulating BACE1 transcription. Thus, the TNFα-mediated activation of NF-κB through represents a key part of this regulatory pathway. These findings indicate that one mechanism underlying the regulation of BACE1 transcription may be through -mediated activation of NF-κB. We found significantly lower Aβ as well as BACE1 levels in APP23/ in older specimens (12 and 24 mo). One possible explanation is that when Aβ deposits are lower (at both time points), there is less Aβ to extract, therefore lower Aβ levels might not be caused by the reduced BACE1 level. To examine whether the reduced Aβ level is caused by the reduced BACE1 level, we measured Aβ and BACE1 levels in APP23/ mice at 6 mo of age, before Aβ pathology can be observed. If affects the BACE1 level, it should also reduce the BACE1 level at this age. We first found that total Aβ in APP23/ mice was much lower than in APP23 mice (). Both Aβ40 and Aβ42 levels in APP23/ mice were also reduced. A Western blot showed a reduction of the BACE1 protein level in APP23/ mice (). BACE1 RT-PCR showed a similar result to that of 12-mo-old mice; the BACE1 mRNA level was significantly lower in APP23/ mice than in APP23 mice (). Together, these findings indicate that indeed regulates the BACE1 mRNA level, and that Aβ reduction in APP23 mice after genetic deletion of is caused by decreased BACE1 levels. Aβ reduction could also be caused by an increase in Aβ degradation/clearance activity, which is not relevant to Aβ production. To determine whether deletion of reduces Aβ deposition by affecting enzymes involved in Aβ degradation, we assessed the protein levels of IDE and NEP, two enzymes that play an important role in Aβ degradation and clearance (). Western blot analyses did not show significant differences in either IDE or NEP levels between APP23/ and APP23 mice (), suggesting that deletion of did not interfere with IDE and NEP expression. To find out whether deletion of could have an effect on IDE or NEP activity, we further compared both IDE and NEP activity between APP23 and APP23/ mice. Again, no significant difference was observed (). Therefore, IDE and NEP were not responsible for the reduction of the Aβ level associated with the deletion. We recently reported that plays a critical role in Aβ-induced neuronal death (). To determine whether the deletion of protects neurons, we compared neuronal loss in wild-type, APP23/ , and APP23 mice. Compared with wild-type mice, APP23 mice show a 30% reduction in NeuN-positive cells in the entorhinal cortex, whereas APP23/ mice show no significant reduction at 24 mo of age (). Results were similar in the hippocampus, where APP23 mice had 15% fewer NeuN-positive cells in the CA1 area of the hippocampus (; ), whereas little neuronal loss was seen in APP23/ mice at 24 mo of age (). We found similar results using Nissl-stained brain sections (not depicted). mice with a hole-board memory test, which is a behavioral task widely used to assess exploratory learning and memory (; ). The majority of APP23 mice showed significant deficits in the spatial component of the test. mice (). This suggests that deletion of , which could lead to a reduction of Aβ levels as well as neuronal protection, might have an effect on improving spatial learning performance in APP23/ mice (). The correlation between Aβ reduction/neuronal protection and improved learning and memory behaviors remained strong when the percentages of correct pokes were averaged over all hidden hole-board trials (). Specifically, we found that 6-mo-old APP23 mice made more errors than did age-matched wild-type mice across 3 d of testing (group effect, F1,15 = 22.198, P < 0.001). However, percentage correct scores of APP23/ mice were markedly higher than those of APP23 mice. APP23/ mice made significantly fewer errors than did APP23 mice on days 2 and 3 of testing (group effect, F1,20 = 8.957, P < 0.05). The object recognition task is based on the spontaneous exploration of novel and familiar objects. Mice will spend more time exploring a novel object than a familiar one (; ; ). We further examined whether deletion of could rescue object recognition deficits in APP23 mice. Object recognition performance was much better in APP23/ mice than in APP23 mice, as the recognition indexes differed significantly between these groups (group effect, F3,31 = 24.947, P < 0.001; ). APP23/ mice performed comparably to age-matched wild-type mice. In addition, within-group -test analyses confirmed that APP23/ mice ( = 8.872, P < 0.001) and wild-type mice ( = 10.024, P < 0.001) performed above chance values (50%), whereas APP23 mice did not ( = 0.092). This result suggests that deletion of in APP23 mice not only improves learning and memory in the hole-board behavioral test but also enhances performance in objective recognition. is a family of TNF receptors, and (), both of which bind soluble Aβ40 (). Overexpression of promotes Aβ-induced neuronal death (). It has been reported that a higher inflammatory response was observed in MCI and AD patients (; ;), and inflammatory cytokines and free radicals can up-regulate BACE1 expression (; ; ; ; ). TNFα is one of the up-regulated inflammation factors in APP transgenic mice (). Here we found that could directly regulate BACE1 transcription through NF-κB, which is one of the major mediators of TNFα-activated signaling (; ). Recent findings showed multiple NF-κB binding sites located in the vicinity of BACE1 promoter (), suggesting that NF-κB may play an important role in regulating BACE1 transcription. This is confirmed by our results that the NF-κB inhibitor inhibits BACE1 promoter activity. At 24 mo of age, we found significant neuronal loss in APP23 mice. However, did not observe neuronal loss in 16-mo-old APP transgenic mice (Tg2576) expressing the APPK670N/M671L mutation, the same mutation harbored by APP23 mice. This might be because the APP transgene is controlled by different promoters in APP23 and Tg2576 mice (; ). deficiency ameliorates neuron loss in APP23 mice, consistent with our previous findings that overexpression increases the vulnerability of cultured hippocampal neurons to Aβ-induced death and promotes neuronal degeneration (). We also noticed that TNFα plays different roles in neuronal death and survival via its distinct receptors, and . Neuron loss in APP23 mice caused by “endogenous” Aβ may be conducted through different signal transduction pathways. Moreover, showed that TNFα can protect neurons derived from fetal brains against Aβ toxicity. Our unpublished data show that is expressed at a low level, whereas is expressed at a high level in fetal neurons. This may explain why TNFα is trophic in fetal neurons. Interestingly, discovered that neurons from mice with a deficiency of both and are more sensitive to excitotoxic injury. This result is interesting because the finding suggests that there is a balance between and expression levels in neurons, and seems to be more critical and more sensitive to neurons. Our behavioral analyses revealed that inactivation of rescued hippocampal-dependent learning and memory deficits displayed by young APP23 mice (). A previous study reported that disruption of the gene or PS1 in APP transgenic mice rescues memory deficits measured by social recognition and spatial alternation tasks (). This is consistent with our findings in APP23/ mice, presumably because depletion decreases Aβ production and deposition, thereby reducing Aβ-related memory deficits. The relatively normal performance of hippocampal-dependent memory tasks by APP23/ mice is age related. At 6 mo of age, APP23/ mice already performed hippocampal-dependent memory tasks better than APP23 mice. Furthermore, knockout mice exhibited normal synaptic transmission and plasticity in the Schaffer collateral pathway (unpublished data). Our results allow us to determine whether treating APP23 mice with anti-TNFR1 antibody or inhibitors of the signal transduction pathway could reduce BACE1 and cerebral Aβ. knockout mice ( ) were constructed on a C57BL/6 background as previously described (). APP23 transgenic mice were provided by Novartis Institute for Biomedical Research; these mice express mutant human βAPP (Swedish double mutation, KM670/671NL) under the control of a brain- and neuron-specific murine Thy-1 promoter element. APP23 transgenic mice develop senile plaques in the cerebral cortex and hippocampus and show neuronal loss at 12–18 mo of age; this pathology is most evident in area CA1 of the hippocampus (). APP23 mice were also constructed on a C57BL/6 background. mice were crossed and their progeny were genotyped. mice to produce APP23/ mice. To maintain the heterozygous APP transgene in our mice, we crossed APP23 mice with wild-type C57BL/6 mice. mice for three to five generations. . We used APP23/ mice of the F3–F5 generation in our experiments. Mice homozygous for the targeted mutation (formerly , p55 deficient) show defects in resistance to intracellular pathogens and are resistant to the lethal effects of lipopolysaccharide administration in conjunction with D-galactosamine. Pulmonary inflammatory responses are diminished in p55-deficient mice. There are also defects in splenic architecture, formation of germinal centers, and liver regeneration. - deficient mice display increased susceptibility to atherosclerosis when maintained on a high-fat diet (). No observations regarding any syndromes of the central nervous system have been made. APP23, APP23/ , and wild-type mice ( = 10 per group) were killed at 12 and 24 mo of age, and one hemisphere of the brain was homogenized in homogenization buffer (250 mM sucrose, 20 mM Tris-HCl, pH 7.4, 1 mM EDTA, and 1 mM EGTA). An aliquot of the homogenate was dissolved in formic acid and neutralized with a neutralization buffer (1 mM Tris and 0.5 M NaHPO). Protein concentration was measured by protein assay (Bio-Rad Laboratories). For total Aβ ELISA, the capture antibody was monoclonal anti-Aβ antibody 4G8 (Chemicon), and the detection antibody was biotinylated monoclonal antibody anti-Aβ 6E10 (AbD Serotec). Aβ40 and Aβ42 were measured with an Aβ40 and Aβ42 ELISA kit (Biosource International). The ELISA system has been extensively tested and no cross-reactivity between Aβ40 and Aβ42 was observed. Data are presented as means ± SD of four experiments. BACE1 protein levels were measured by ELISA as described previously (). The capture antibody was anti-BACE1 polyclonal antibody P1 () and the detection antibody was biotinylated anti-BACE1 polyclonal antibody P2 (). TMB substrate was used to visualize the reaction product, which was read at OD with a microplate reader (Sigma-Aldrich). BACE1 protein (Amgen) was used as a standard. Data are presented as means ± SD of four experiments. Aliquots of brain homogenates from APP23, APP23/ , and wild-type mice were further lysed with 1× RIPA buffer, and 50–150 μg of total protein was subjected to SDS-PAGE (8–12% acrylamide). Separated proteins were then transferred onto polyvinylidene fluoride membranes. The blots were probed with the following antibodies: anti-BACE1 monoclonal antibody (R&D Systems), anti-Aβ (1–17) monoclonal antibody (clone 6E10, 1:2,000; Chemicon), anti-IDE polyclonal antibody (Oncogene Research Products), anti-NEP polyclonal antibody (Chemicon), and anti– β actin antibody (Sigma-Aldrich). Western blotting for Aβ was performed as described previously (). To detect minute levels of Aβ, formic acid–dissolved brain tissue was immunoprecipitated with anti-Aβ polyclonal antibody (Zymed Laboratories) and subjected to SDS-PAGE using 10% acrylamide gels containing 8 M urea. Separated proteins were transferred onto polyvinylidene fluoride membranes. Aβ40 and Aβ42 were detected with monoclonal anti-Aβ antibody 6E10. Synthetic Aβ40 and Aβ42 (Biosource International) were used as standards. An aliquot of brain homogenates from APP23, APP23/ , and wild-type mice was further lysed with a lysis buffer (10 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1 mM NaVO, 10% glycerol, and 0.5% Triton X-100). BACE1 enzymatic activity assays were performed by using synthetic peptide substrates containing BACE1 cleavage site (MCA-Glu-Val-Lys-Met-Asp-Ala-Glu-Phe-[Lys-DNP]-OH; Biosource International). BACE substrate was dissolved in DMSO and mixed with a 50-mM Hac and 100-mM NaCl, pH 4.1, reaction buffer. An equal amount of protein was mixed with 100 μl of substrate, and fluorescence intensity was measured with a microplate reader (BioTek) at an excitation wavelength of 320 nm and an emission wavelength of 390 nm. IDE enzyme activity was measured as described previously (). In brief, brains were homogenized in 50 mM potassium phosphate buffer, pH 7.3, containing 200 μm PMSF and a proteinase inhibitor mix (Sigma-Aldrich). Samples were centrifuged and the supernatant fraction was used for IDE activity measurement. The hydrolysis of fluorogenic substrate peptides (2 μm Abz-GGFLRKHGQED-Dnp as substrate in 20 mM potassium phosphate buffer, pH 7.3) was measured by following an increase in fluorescence (excitation at 318 nm and emission at 419 nm) that occurred upon peptide bond cleavage. The max velocity of IDE activity was calculated by the first 20 min and indicated as fluorescence unit/min microgram protein. For the in vitro NEP activity assay, mouse brains were homogenized in 100 mM MES buffer (pH 6.5) with proteinase inhibitors (Sigma-Aldrich). Homogenate was centrifuged at 20,000 for 45 min to separate the membrane fraction and the supernatant was removed. The membrane pellet was resuspended in MES buffer and directly used in NEP activity assay as previously described (). To compare BACE1 expression levels, we used the following primers for RT-PCR: mouse BACE1 forward primer, 5′-AGACGCTACACATCCTGGTG-3′, and backward primer, 5′-CCTGGGTGTAGGGCACATAC-3′. The amplified BACE1 fragment was 146 bp. Mouse s18 was used as a loading control: forward primer, 5′-CAGAAGGACGTGAAGGATGG-3′, and backward primer, 5′-CAGTGGTCTTGGTGTGCTGA-3′. The amplified mouse s18 fragment was 159 bp. Total RNA was extracted from the brains of 12-mo-old APP23 and APP23/ mice ( = 5) using an RNA mini column kit (Invitrogen). RT-PCR was performed using a One- Step RT-PCR kit (Invitrogen) and the following PCR cycles: 50°C for 30 min, 94°C for 2 min, followed by 25 cycles at 94°C for 15 s, 49°C for 30 s, and 68°C for 1 min. We transfected 293 cells with pB1P-A vector containing a promoter (−1941 to +292) upstream from a luciferase reporter gene () using lipofectamine (Invitrogen). After transfection, cells were treated with different concentrations of TNFα (R&D Systems), extracellular domain of TNFR1 (R&D Systems), or NF-κB inhibitor 6-amino-4(4-phenoxyphenylethylamino) quinazoline (Calbiochem; ). Cells were collected 12 h after treatment, and a luciferase assay (Promega) was performed, according to the manufacturer's instructions. Luminescence intensity was measured with a microplate reader, normalized according to protein amount, and plotted as relative luminescence units per milligram of protein. Immunohistochemistry was performed as previously described (). In brief, paraformaldehyde-fixed brains were quickly frozen, and then sectioned at 30 μm. Sections were incubated with either anti-Aβ (6E10 clone or 4G8 clone, 1:1,000; Chemicon), anti-NeuN (MAB377, 1:400; Chemicon), anti-CD11b (MCA711, 1:500; AbD Serotec) and CD45 (MCA1388, 1:500; AbD Serotec), anti-α-smooth muscle actin (α-SM actin, A2547, 1:400; Sigma-Aldrich), or anti-vWF (AB7536, 1:200; Chemicon). Secondary antibodies were applied with horse anti–mouse (for 6E10, NeuN detection, 1:1,000) and goat anti–rat (for CD45 or CD11b, 1:1,000) followed by a DAB substrate (Vector Laboratories). For immunofluorescence, fluorescent-labeling 488 (green) or 594 (red) secondary antibodies against rabbit IgG or mouse IgG were used (1:1,000; Invitrogen). A microscope (DMLS; Leica) with a 10× N PLAN and 20× and 40× PL FLUOTAR was used. Digital images were captured and processed by digital camera (Optronics) and MagnaFire software (version 2.1C; Optronics). 30-μm serial sagittal sections through the entire rostrocaudal extent of the hippocampus were cut on a cryostat. Every 10th section was immunostained with anti-NeuN antibody. On all sections containing the hippocampus, we delineated the pyramidal cell layer CA1. The total number of neurons were obtained using unbiased stereology (; ) and a microscope equipped with a digital camera (DEI-470; Optronics). For each section, we delineated a 400-μm area in CA1 and in the entorhinal cortex and counted all NeuN-immunoreactive cells within that 400-μm box. The mean sum of neurons was counted per animal ( = 10). We used the same method to count Aβ-immunoreactive plaques (stained with 6E10) in the hippocampus and entorhinal cortex in a double blind test. We also measured the diameter of each counted plaque. Differences between groups were tested with Image-pro Plus Analysis (Media Cybernetics). As previously reported (), this task measured a mouse's ability to remember which one out of four equidistant holes was baited with food. Two photobeam apparatuses were used with a hole board for assessing directed exploration in mice for behavioral tests. A tested mouse ( = 10 for each group) was placed in the center of the hole-board and the number of nose pokes was automatically registered for 5 min. After 20 min, each animal was placed in a corner of the hole board and allowed to freely explore the apparatus for 5 min. The number of head dips, time spent head-dipping, and the number of rearings were recorded. A comprehensive cognitive performance was determined by calculating the mean number of correct pokes per trial that mouse made each day. Cognition was expressed as the percentage of correct pokes. The measurements in the hole-board test were analyzed by unpaired test. In all cases the significance level was considered to be P < 0.05, and the very significant level was considered to be P < 0.01. The day before training, an individual mouse ( = 10 for each group) was placed into a training apparatus (a box the same size as described for the hole-board test) and allowed to habituate to the environment for 15 min. Training was initiated 24 h after habituation. A mouse was placed back into the training box containing two identical objects A and B (die or marble) and allowed to explore these objects. Among experiments, training times varied from 3.5 to 20 min. For each experiment, the same set of animals was used repeatedly with different sets of objects for each repetition. Five repetitions were performed on each set of mice. Each mouse was trained and tested no more than once per week, with a 1-wk interval between testing. Moreover, each experimental condition was replicated independently four times. In each experiment, the experimenter was blinded to the subjects during training and testing. To test memory retention, mice were observed for 10 min, 6 h, and 24 h after training. Mice were presented with two objects, one that was used during training, and thus was “familiar,” and one that was novel. The test objects were divided into 10 sets of “training” plus “testing” objects, and a new set of objects was used for each training session. A recognition index was calculated for each mouse, expressed as the ratio (100TB)×(TA + TB), where TA and TB are the time spent during the second trial on subject A and subject B, respectively. To ensure that the discrimination targets did not differ in odor, the apparatus and the objects were thoroughly cleaned with 90% ethanol, dried, and ventilated for a few minutes after each experiment. In general, analysis of variance models (ANOVA) were used to analyze behavioral data. Typically, the statistical models included two between-subjects variables, the genotype of mice (APP23 vs. APP23/ ) and age, and one within-subjects variable, such as blocks of trials. When ANOVAs with repeated measures were conducted, the Huynh-Feldt adjustment of α levels was used for all within-subjects effects containing more than two levels to protect against violations of the sphericity/compound symmetry assumptions underlying this ANOVA model.

Formation and extension of axons and dendrites, so-called neurite outgrowth, is a crucial event in neuronal differentiation and maturation during development of the nervous system (; ). These morphological changes require reorganization of the cytoskeletal networks and its accompanying membrane expansion and contraction (; ). Rap1 small G protein has been shown to be involved in neurite outgrowth and neuronal polarization through regulating the MAPK cascade and the cytoskeletal network (; ; ; ). Rho small G protein has been shown to be involved in the morphological development of neurons through regulating actin dynamics (; ; ; ). We previously showed that inactivation of Rho is required for the Rap1-induced neurite outgrowth (). Rap1-activated Rho GTPase-activating protein, RA-RhoGAP, transduces a signal from Rap1 to Rho and inactivates Rho. The precise temporary and spatially activation of Rap1 is important for regulating the activity of RA-RhoGAP to produce the neurite. However, the mechanism underlying how and where Rap1 is activated remains unclear. Neurite outgrowth begins with the activation of membrane receptors of neurotrophins, such as nerve growth factor (NGF) and brain-derived neurotrophic factor (BDNF) (; ; ). NGF is the prototypic neurotrophin, a group of structurally related signaling proteins that are crucial for the survival, differentiation, and maintenance of specific neuronal population (). First, NGF binds to TrkA receptor and induces its dimerization, which then activates its own tyrosine kinase and gathers a signaling complex consisting of SOS, a GDP/GTP exchange factor (GEF) specific for Ras small G protein, Grb2, an adaptor protein for SOS, and Shc, a linker protein between Grb2 and TrkA receptor, on the plasma membrane (; ). Ras is transiently activated on the plasma membrane and induces activation of the c-Raf-MEK-ERK pathway. On the other hand, activated TrkA receptor gathers another signaling complex consisting of C3G, a GEF specific for Rap1, CrkL, an adaptor protein for C3G, and FRS2, a linker protein between CrkL and TrkA receptor, on the plasma membrane (; ). TrkA receptor and the FRS2-Crk-C3G complex associating with the receptor are internalized into clathrin-coated vesicles and transported to early endosomes (). Rap1 is activated at the early endosomes and induces activation of the B-Raf-MEK-ERK pathway. It has been believed that Rap1 is also activated on the cytoplasmic side of the plasma membrane, but the direct evidence has not been obtained. Then, TrkA receptor proceeds further along the degradative pathway in a microtubule- and motor-dependent fashion, reaching late endosomes, followed by the physical destruction of TrkA receptor by proteolysis in lysosomes (; ,). Thus, transport of TrkA receptor to late endosomes has long been recognized as a means to terminate the signaling via degradation of the activated TrkA receptor complex after their internalization from the cell surface. Recent studies suggest that TrkA receptor, after binding NGF at the cell surface, passes through early endosomes and reaches to late endosomes within 30 min (). However, there is still sustained activation of Rap1 after 30 min (; ). Like the activated TrkA receptor complex at early endosomes, TrkA receptor at late endosomes might be competent in the sustained activation of Rap1, but it remains unknown whether, or if so how, Rap1 is activated at late endosomes. Here, we addressed these issues and showed that activated TrkA receptor, which was transported to late endosomes, recruited PDZ-GEF1, a GEF specific for Rap1, by forming a complex through Trk receptors associating ankyrin repeat-rich membrane spanning (ARMS) and synaptic scaffolding molecule (S-SCAM). This tetramer complex then induced the sustained activation of Rap1 and ERK, eventually causing neurite outgrowth. ARMS is a tetraspan transmembrane protein and directly interacts with Trk receptors through each transmembrane domain (; ). S-SCAM is a synaptic scaffolding molecule with six PDZ domains (). As PDZ-GEF1 has a Rap1-binding RA domain, we first examined whether the binding of GTP-Rap1 to the RA domain affects the GEF activity of PDZ-GEF1. We generated recombinant proteins of FLAG-tagged full-length PDZ-GEF1 (FLAG-PDZ-GEF1) and its RA domain–deficient mutant FLAG-PDZ-GEF1-ΔRA and examined their GEF activity. FLAG-PDZ-GEF1 indeed showed GEF activity toward Rap1 as reported (; , ; ), while FLAG-PDZ-GEF1-ΔRA showed much less effect on Rap1 (). The constitutively active form of GST-Rap1A enhanced the PDZ-GEF1 activity twofold more than control GST did (), whereas it did not affect the GEF activity of FLAG-PDZ-GEF1-ΔRA (Fig. S1, available at ). In addition, GST-Rap-binding domain (RBD) of RalGDS, a specific inhibitor of GTP-Rap1, inhibited the PDZ-GEF1 activity (). These results indicate that PDZ-GEF1 is activated by GTP-Rap1 and suggests that once PDZ-GEF1 is activated, its activation is amplified by GTP-Rap1 in a positive feedback mechanism. As PDZ-GEF1 has an auto-amplification GEF activity toward Rap1, we examined whether PDZ-GEF1 is involved in the NGF-induced sustained activation of Rap1 and ERK. We first performed loss-of-function experiments by use of the RNA interference (RNAi) method for PDZ-GEF1. Immunoblotting showed that the amount of PDZ-GEF1 was markedly reduced by the knockdown of PDZ-GEF1 (). The amounts of other proteins, including TrkA receptor, another Rap1 GEF C3G, Rap1, Ras, and ERK remained unchanged (unpublished data). The knockdown of PDZ-GEF1 significantly decreased the activation of Rap1, but not Ras, after NGF stimulation as estimated by the pull-down assay ( and unpublished data). The knockdown of PDZ-GEF1 did not inhibit the activation of ERK at 5 min after NGF stimulation, but it significantly decreased it at 30 and 60 min. The reason why we measured the activation of Rap1 and ERK at 30 and 60 min was that it was reported that activation of Rap1 at 5 min does not contribute to transient activation of ERK at 5 min (; ; ; ); that a substantial portion of the transient activation of ERK at 5 min mainly depends on Ras, but not on Rap1, whereas the activation of Rap1 at later time points, such as 30 and 60 min, but not at 5 min, contributes to the sustained activation of ERK at 30 and 60 min; and that this sustained activation of ERK is essential for neurite outgrowth. Consistent with the results of the pull-down assay, the fluorescence resonance energy transfer (FRET) imaging assay showed that the knockdown of PDZ-GEF1 reduced the activation of Rap1 at the perinuclear region of the cells after NGF stimulation (). To rescue the knockdown of PDZ-GEF1, we generated siRNA-resistant PDZ-GEF1 expression vector (). Expression of siRNA-resistant PDZ-GEF1 potently rescued the activation of Rap1 at the perinuclear region of the cells at 30 min after NGF stimulation (). Rap1 has been shown to localize at various intracellular organelles, such as the Golgi complex, early endosomes, and late endosomes, in cultured fibroblasts and epithelial cells (; ). To examine the localization of endogenous Rap1 in PC12 cells, we performed subcellular fractionation from the post-nuclear supernatant (PNS) of PC12 cells. A significant amount of Rap1 was recovered in the EEA1-positive early endosomal fraction and the Rab7-positive late endosomal fraction (). To examine the localization of GTP-Rap1 in PC12 cells, we performed the subcellular fractionation in the presence of GST-RalGDS-3xRBD, which preferentially binds to GTP-Rap1 (). GST-RalGDS-3xRBD was recovered in the late endosomal fraction at 30 min after NGF stimulation, indicating that Rap1 was activated at late endosomes at 30 min after NGF stimulation (). The huge amount of GST- RalGDS-3xRBD, which did not bind to Rap1, remained in the bottom PNS and heavy membrane fractions. The NGF-induced neurite outgrowth was reduced in the PDZ-GEF1-knockdown PC12 cells (). This knockdown effect was rescued by expression of siRNA-resistant PDZ-GEF1 (). Moreover, PC12 cells overexpressing PDZ-GEF1 or PDZ-GEF1-ΔRA displayed marked neurite outgrowth at 12 h after NGF stimulation as compared with control cells (Fig. S2, available at ). Consistent with the difference in the GEF activities between PDZ-GEF1 and PDZ-GEF1-ΔRA shown in , the effect of PDZ-GEF1-ΔRA on neurite outgrowth was less effective than that of PDZ-GEF1. Collectively, these results indicate that PDZ-GEF1 is involved in the NGF-induced sustained activation of Rap1 and ERK, which eventually causes neurite outgrowth. Before NGF stimulation, FLAG-PDZ-GEF1 diffusely localized at the cytoplasm, but it was recruited to LBPA-positive late endosomes at 30 min after NGF stimulation (). FLAG-PDZ-GEF1 was not recruited to EEA1-positive early endosomes at either 5 or 30 min. In contrast to PDZ-GEF1, C3G was recruited to EEA1-positive early endosomes at 5 min after NGF stimulation (Fig. S3, available at ). C3G was not recruited to LBPA-positive late endosomes at either 5 or 30 min. These results indicate that PDZ-GEF1 is recruited to late endosomes in response to NGF and suggest that this Rap1 GEF is involved in the NGF-induced sustained activation of Rap1 at late endosomes. We next examined whether transport of TrkA receptor to late endosomes recruits PDZ-GEF1 there. Bafilomycin is known to be an inhibitor of V-ATPase-regulating protein trafficking from early to late endosomes (). The immunofluorescence signal for TrkA receptor became concentrated at EEA1-positive early endosomes at 5 min and at LBPA-positive late endosomes at 30 min after NGF stimulation (Fig. S4, available at ). Bafilomycin did not inhibit the NGF-induced transport of TrkA receptor to EEA1-positive early endosomes at 5 min, but it significantly decreased the transport of TrkA receptor to LBPA-positive late endosomes at 30 min ( and Fig. S4). Bafilomycin inhibited the recruitment of PDZ-GEF1 to LBPA-positive late endosomes at 30 min after NGF stimulation (). These results indicate that transport of TrkA receptor to late endosomes is required for the recruitment of PDZ-GEF1 to late endosomes. Bafilomycin inhibited the activation of Rap1 at 5 min and more potently at 30 and 60 min as estimated by the pull-down assay (). In contrast, bafilomycin did not inhibit the activation of Ras at 5 min (). Bafilomycin did not inhibit the activation of ERK at 5 min, but significantly decreased it at 30 and 60 min. Bafilomycin reduced the NGF-induced activation of Rap1 at the perinuclear region of the cells as estimated by the FRET assay (). Bafilomycin reduced the NGF-induced neurite outgrowth (). Collectively, these results indicate that the NGF-induced internalization and transport of TrkA receptor to late endosomes is involved in the NGF-induced sustained activation of Rap1 and ERK and neurite outgrowth by recruiting PDZ-GEF1 to late endosomes. We next attempted to understand the molecular mechanism how PDZ-GEF1 induces the activation of Rap1 at late endosomes in response to NGF. We previously showed that PDZ-GEF1 directly binds S-SCAM (), whereas it was shown that TrkA receptor directly binds an endosomal protein ARMS (; ). Consistent with our earlier observation, S-SCAM was coimmunoprecipitated with PDZ-GEF1 before or after NGF stimulation (). ARMS and TrkA receptor were also coimmunoprecipitated with PDZ-GEF1, but this coimmunoprecipitation was strengthened after NGF stimulation. However, presumably due to the reconstitution of the complex during the procedure of cell lysis, we did not get the strong difference in the amount of the complex between 5 min and 60 min after NGF stimulation. These results suggest that PDZ-GEF1 forms a tetramer complex with S-SCAM, ARMS, and TrkA receptor after NGF stimulation. As ARMS has a PDZ-binding motif, RESIL, at the cytoplasmic tail, we examined by the pull-down assay whether ARMS directly binds to S-SCAM or PDZ-GEF1. The GST fusion protein containing the C-terminal region of ARMS (GST-ARMS- C-terminal) bound to S-SCAM, but not to PDZ-GEF1 ( and unpublished data). GST-ARMS-C-terminal lacking the last three amino acids in the RESIL motif (GST-ARMS-C-terminal-ΔSIL) did not bind to S-SCAM. The PDZ4 domain of S-SCAM bound to GST-ARMS-C-terminal (). Under the conditions where Myc-S-SCAM was coimmunoprecipitated with FLAG-ARMS, Myc-S-SCAM lacking the PDZ4 domain (S-SCAM-ΔPDZ4) was not coimmunoprecipitated (). These results indicate that S-SCAM directly binds ARMS and that this binding is mediated by the PDZ4 domain of S-SCAM and the PDZ-binding motif of ARMS. Collectively with the earlier observations that PDZ-GEF1 binds to the PDZ1 domain of S-SCAM and that ARMS directly binds to TrkA receptor in a NGF-dependent manner, these results indicate that PDZ-GEF1 binds to S-SCAM to form a binary complex in a NGF-independent manner, which binds to TrkA receptor through ARMS to form a tetramer complex in a NGF-dependent manner. We then examined whether this tetramer complex of PDZ-GEF1, S-SCAM, ARMS, and TrkA receptor is formed at late endosomes in intact PC12 cells. The immunofluorescence signal for ARMS was mainly concentrated at LBPA-positive late endosomes, not at early endosomes, before and after NGF stimulation, indicating that ARMS is an intrinsic late endosomal protein (). The signals for PDZ-GEF1 and S-SCAM localized diffusely at the cytoplasm before NGF stimulation (). The signal for TrkA receptor localized at the cell surface plasma membrane and a vesicular-like structure, presumably, early endosomes, before NGF stimulation (). The signals for PDZ-GEF1, S-SCAM, and TrkA receptor became concentrated at ARMS-positive late endosomes after NGF stimulation. As the signals for PDZ-GEF1 and S-SCAM colocalized well before and after NGF stimulation, these two molecules seem to form a binary complex, consistent with the above coimmunoprecipitation results (Fig. S5A, available at ). Collectively, these results suggest that upon binding of NGF, TrkA receptor is internalized and transported to ARMS-positive late endosomes, interacts with ARMS, and then recruits the PDZ-GEF1-S-SCAM complex mainly to ARMS-positive late endosomes, resulting in the tetramer complex formation. To further examine the recruitment of PDZ-GEF1 to late endosomes biochemically, we performed immunoisolation of PDZ-GEF1-containing vesicles using PDZ-GEF1 pAb-coated magnetic beads from the PNS fraction from NGF- nonstimulated and -stimulated PC12 cells. Rab7 was recovered in the immunoisolated PDZ-GEF1-containing vesicles, whereas EEA1 was not recovered in the immunoisolated vesicles (). Rab7 and ARMS were more predominantly recovered after NGF stimulation. These results have provided another line of evidence that the tetramer complex of TrkA receptor, ARMS, S-SCAM, and PDZ-GEF1 is formed mainly at late endosomes in a NGF-dependent manner. To confirm that ARMS and S-SCAM are required for the recruitment of PDZ-GEF1 to late endosomes, we performed loss-of-function experiments by use of the RNAi method for ARMS or by use of a dominant negative mutant for S-SCAM. The recruitment of FLAG-PDZ-GEF1 to late endosomes was reduced in the ARMS-knockdown PC12 cells (, and Fig. S5 B). Expression of Myc-S-SCAM did not inhibit the recruitment of FLAG-PDZ-GEF1 to late endosomes, but expression of Myc-S-SCAM-ΔPDZ4 reduced the recruitment of FLAG-PDZ-GEF1 to late endosomes (). Expression of Myc-S-SCAM, but not Myc-S-SCAM-ΔPDZ4, slightly induced the recruitment of FLAG-PDZ-GEF1 to late endosomes even in the absence of NGF (Fig. S5A). We next examined whether ARMS and S-SCAM are indeed involved in the NGF-induced neurite outgrowth. The NGF- induced neurite outgrowth was reduced in the ARMS-knockdown PC12 cells (). The NGF-induced neurite outgrowth was reduced in the PC12 cells expressing Myc-S-SCAM-ΔPDZ4 (). Collectively, these results indicate that ARMS and S-SCAM are essential for the NGF-induced recruitment of PDZ-GEF1 to late endosomes and neurite outgrowth. To understand the relationship between C3G and PDZ-GEF1 in the NGF-induced neurite outgrowth, we first performed loss-of-function experiments by use of the RNAi method for C3G (). The inhibitory effect of the knockdown of C3G on neurite outgrowth was less effective than that of the knockdown of PDZ-GEF1 (). Consistently, overexpression of C3G enhanced the NGF-induced neurite outgrowth less effectively than that of PDZ-GEF1 (Fig. S2). The effect of the knockdown of C3G was rescued by expression of siRNA-resistant C3G (). The double knockdown of C3G and PDZ-GEF1 reduced neurite outgrowth more potently than the knockdown of PDZ-GEF1 or C3G alone (). In addition, the inhibitory effect of the double knockdown of C3G and PDZ-GEF1 on neurite outgrowth was similar to that of the knockdown of Rap1 (). These results indicate C3G and PDZ-GEF1 are main Rap1GEFs in the NGF-induced neurite outgrowth in PC12 cells. The inhibitory effect of the knockdown of C3G on the NGF-induced sustained activation of Rap1 was less effective than that of the knockdown of PDZ-GEF1 (). The knockdown of PDZ-GEF1 alone did not completely abolish the NGF-induced sustained activation of Rap1. The double knockdown of C3G and PDZ-GEF1 more potently abolished the NGF-induced sustained activation of Rap1 and ERK. The inhibitory effect of the double knockdown of C3G and PDZ-GEF1 on the sustained activation of ERK was similar to that of the knockdown of Rap1 (). These results indicate that C3G and PDZ-GEF1 work cooperatively in the NGF-induced sustained activation of Rap1. To validate the role of PDZ-GEF1 in PC12 cells, we finally examined whether PDZ-GEF1 is involved in the BDNF-induced neurite outgrowth in rat primary cultured hippocampal neurons. It has been reported that BDNF, but not NGF, enhances axon elongation in hippocampal neurons (; ; ; ). BDNF binds to its receptor, TrkB receptor, in hippocampal neurons (; ). Retrograde transport of BDNF-TrkB and the signaling complex associating with it is involved in the axon elongation and cell survival signal (; ). Hippocampal neurons at 2.0 days in vitro were stimulated by BDNF for indicated periods of time. PDZ-GEF1 was recruited to LBPA-positive late endosomes at 30 min after BDNF stimulation (). PDZ-GEF1 was not recruited to EEA1-positive early endosomes. Hippocampal neurons were cotransfected with an expression vector for GFP as a morphological marker along with the PDZ-GEF1 siRNA expression vector. At 1.5 days in vitro after plating, BDNF-induced recruitment of endogenous PDZ-GEF1 to late endosomes was not observed in the PDZ-GEF1-knockdown neurons (). The length of axon in the PDZ-GEF1-knockdown neurons was less than those of the control neurons (). These results indicate that PDZ-GEF1 is involved in the BDNF-induced axon outgrowth in rat primary cultured hippocampal neurons. We first showed here that NGF induced the sustained activation of Rap1 at late endosomes and neurite outgrowth in cultured PC12 cells. The activation of Rap1 at late endosomes was catalyzed by PDZ-GEF1. Earlier FRET studied showed that Rap1 was activated at peri-nuclear regions and it was practically difficult to distinguish the sites of the activation of Rap1 at the plasma membrane, early endosomes, and late endosomes by this method (; ). We could not directly show the activation of Rap1 at late endosomes but indirectly showed it by immunofluorescence microscopy and immunoisolation analysis: PDZ-GEF1 was recruited to late endosomes in response to NGF. PDZ-GEF1 was not recruited to early endosomes, whereas C3G was recruited to early endosomes, but not to late endosomes. Thus, these two GEFs for Rap1 play roles in different compartments. As Rap1 is first activated by C3G at the plasma membrane and early endosomes after NGF stimulation (; ), activated Rap1 may affect the GEF activity of PDZ-GEF1. We found that the inhibitory effect of the knockdown of C3G on the NGF-induced sustained activation of Rap1 and neurite outgrowth was less effective than that of the knockdown of PDZ-GEF1. The knockdown of PDZ-GEF1 alone did not completely abolish the NGF-induced sustained activation of Rap1 or neurite outgrowth. The double knockdown of C3G and PDZ-GEF1 completely abolished the NGF-induced sustained activation of Rap1 and neurite outgrowth similar to the knockdown of Rap1. Thus, C3G and PDZ-GEF1 seem to work cooperatively in the NGF-induced sustained activation of Rap1 and neurite outgrowth. We found that C3G was recruited to early endosomes, but not to late endosomes. The activation of Rap1 was observed on late endosomes, not on early endosomes, at 30 min after NGF stimulation. Moreover, inhibition of the transport of TrkA receptor form early to late endosomes reduced the activation of Rap1 at 30 min after NGF stimulation. Collectively, we propose the following model for the activation of Rap1 by C3G and PDZ- GEF: Upon NGF stimulation, C3G first activates Rap1 on early endosomes; GTP-Rap1 on early endosomes is transported to late endosomes with TrkA receptor; and Transported GTP-Rap1 activates PDZ-GEF1 and induces the sustained activation of Rap1 on late endosomes, eventually inducing neurite outgrowth. The long-distance retrograde NGF signaling from axon terminals to cell bodies is crucial for neuronal survival and plasticity (; ). It was shown that the internalization and endosomal trafficking of TrkA receptor are important for the long-distance retrograde NGF signaling (; ; ). TrkA receptor continues to transduce a signal after its internalization into endosomes (; ; ). Certain specialized endosomal organelles might represent signaling platforms from which specific pathways emerge (; ). Earlier studies demonstrated that late endosomes are the major endosomal population labeled by endocytosed iodinated NGF in cell bodies of sympathetic neurons (). As the interaction of TrkA receptor and NGF is resistant to decreasing pH values within the endosomal pathway (), it appears likely that NGF is still bound to the extracellular domain of TrkA receptor within late endosomes. Accordingly, immunoelectron microscopic study found phospho-TrkA receptor (pTrkA) immunoreactivity in late endosomes of sciatic-nerves () and biochemical study found pTrkA in late endosomes in PC12 cells (). Rab7, a member of the Rab family small G proteins, which regulates TrkA receptor transport from late endosomes to lysosomes, was shown to regulate the persistence of TrkA receptor at late endosomes and TrkA receptor signaling (). Consistent with these earlier observations, we showed here by use of bafilomycin, an agent which inhibits the vesicular traffic from early endosomes to late endosomes, that the NGF-induced internalization and transport of TrkA receptor to late endosomes was essential for the sustained activation of Rap1 at late endosomes, the activation of ERK, and neurite outgrowth. Collectively with our findings, like TrkA receptor in clathrin-coated vesicles () and early endosomes (), TrkA receptor in late endosomes is competent of signaling. In addition, there are several reports that receptor tyrosine kinases (RTKs) induce signaling from late endosomes. For instance, it is known that late endosomes contain phospho-EGF receptor (pEGFR and its activated downstream signaling components (). The MAPK scaffold p14 localizes to the outer limiting membrane of late endosomes, and this localization is essential for EGFR signaling (). There is also in vivo evidence pointing toward the importance of late endosomes in the regulation of neuronal TGF-β signaling (). Thus, late endosomes appear to be ideally suited for regulating spatial and temporal compartmentalization of signal transduction, beyond its conventional role in cargo degradation. Our report is the first to demonstrate that late endosomes represent functional TrkA receptor signaling platforms. This system might play a role in continuing activation state of NGF signaling during transport of TrkA receptor from axon terminals to cell bodies. We furthermore showed here that PDZ-GEF1 was recruited to late endosomes to form a complex with TrkA receptor, which was internalized and transported there, and that recruited PDZ-GEF1 induced sustained activation of Rap1 and ERK. NGF-bound TrkA receptor activates its own tyrosine kinase and gathers the signaling complex consisting of C3G, CrkL, and FRS2 on the plasma membrane (; ). TrkA receptor and the FRS2-Crk-C3G complex associating with the receptor are internalized into clathrin-coated vesicles and move into early endosomes within 5 min (). Rap1 is activated at early endosomes and activated Rap1 induces the activation of the MAPK cascade through activating B-Raf (). TrkA receptor then passes through early endosomes without entering into the recycling pathway to the plasma membrane and reaches to late endosomes within 30 min. Like the activated TrkA receptor complex at early endosomes, the TrkA receptor complex containing ARMS, S-SCAM, and PDZ-GEF1 at late endosomes was competent in activation of Rap1. While the GEF activity of C3G is activated by its interacting partner Crk (), it remains unknown how the activity of PDZ-GEF1 is regulated at late endosomes. PDZ-GEF1 binds to the PDZ1 domain of S-SCAM through its C-terminal PDZ-binding motif and forms a stable complex (). However, the binding of S-SCAM to PDZ-GEF1 does not affect GEF activity in vitro (). The complex was recruited to ARMS-positive late endosomes through directly binding to ARMS, in a manner dependent on transport of TrkA receptor. The C-terminal PDZ-binding motif of PDZ-GEF1 contributed to direct the subcellular localization of PDZ-GEF1, leading to the augmented activation of Rap1 at late endosomes. Collectively, PDZ-GEF1 is recruited to late endosomes through its C-terminal PDZ-binding motif. PDZ-GEF1 then activates Rap1 at late endosomes and the GEF activity of PDZ-GEF1 is augmented by a positive feedback mechanism, resulting in the sustained activation of Rap1 at late endosomes. It may be noted that when TrkA receptor is transported from early endosomes to late endosomes, the FRS2-Crk-C3G complex associating with TrkA receptor remains at early endosomes, suggesting that this complex is dissociated from the receptor. The mechanism of this dissociation remains unknown and is an issue to be addressed in the future. We finally showed here that the PDZ-GEF1-induced sustained activation of Rap1 at late endosomes was involved in the NGF-induced neurite outgrowth in PC12 cells and the BDNF-induced axon outgrowth in rat hippocampal neurons. What is the target protein(s) of Rap1 in the neurite outgrowth? Many downstream targets of Rap1 have been identified: they include c-Raf, B-Raf, RalGDS, afadin/AF-6, PI3-kinase, and RAPL (; ; ; ). c-Raf and B-Raf are protein kinases connecting Ras to ERK; RalGDS is a regulator of another small G protein Ral; afadin/AF-6 is an actin filament- and nectin-binding protein at adherens junctions; PI3-kinase is a phosphatidylinositol kinase, and RAPL is a small protein which transduces a signal from Rap1 to integrin. Among them, for activating MAP kinase cascade, B-Raf was shown to be the target protein of Rap1 for the NGF-induced neurite outgrowth (). Several transcription factors, which are downstream of the MAP kinase cascade, contribute to neurite outgrowth by gene expression. On the other hand, RAPL and its binding protein Mst1 were recently shown to be involved in transport of integrin LFA-1 and its activation in lymphocytes (). Transport of transmembrane proteins from late endosomes to the plasma membrane is an emerging paradigm and generally accepted (; ; ). Accordingly, activation of Rap1 at late endosomes might be involved in the transport of integrin from late endosomes to the plasma membrane and thereby regulate the activation of MAP kinase cascade indirectly. Moreover, we previously identified RA-RhoGAP as another direct downstream target of Rap1 (). RA-RhoGAP has the RA and GAP domains in addition to the PH and annexin-like repeat domains. It indeed shows a GAP activity specific for Rho and this GAP activity is enhanced by GTP-Rap1. The Rap1-RA-RhoGAP-Rho pathway plays an important role in the neurite outgrowth. Collectively, during the initiation of the neurite outgrowth, a neurite formation signal(s), such as NGF and BDNF, induces the C3G-mediated activation of Rap1 at endocytic vesicles or early endosomes in a growth cone of the nascent neurite. Subsequently, RA-RhoGAP is recruited and activated by Rap1 to cause the inactivation of Rho at the growth cone of the nascent neurite. Repression of the Rho-mediated signaling pathway induces rapid actin depolymerization of the growth cone and thereby extends the incipient neurite further. In addition, to replenish the requirement for neurite outgrowth, such as signaling molecules and cytoskeletal proteins, endocytic vesicles or early endosomes are transported to late endosomes to sustain the activation of Rap1 by PDZ-GEF1. The sustained activation of Rap1 prolongs the activation time of the MAPK cascade and results in the up-regulation of gene expression. Thus, the Rap1-RA-RhoGAP-Rho system and the PDZ-GEF-Rap1-B-Raf-MAPK system could cooperatively regulate neurite outgrowth. In future studies, it will be important to elucidate how these two systems are spatially and temporally activated during the neurite outgrowth. pFLAG-CMV2-PDZ-GEF1 and pFLAG-CMV2-PDZ-GEF1-ΔRA were obtained from Dr. Tohru Kataoka (Kobe University, Kobe, Japan). The RNAi-resistant mutant of FLAG-human PDZ-GEF1 (pERedNLS-FLAG-PDZ-GEF1) was generated by mutagenesis of 5′-GAGAGATTGTATGGTGAA-3′ to 5′-GGAATGTATGGTAA-3′ using the QuikChange site-directed mutagenesis kit (Stratagene). The letters in the codons, which are different from the letters in the RNAi sequence of rat PDZ-GEF1, are underlined. pCMV-HA-PDZ-GEF1 was constructed using standard molecular biology methods. pCIneo-Myc-S-SCAM, pCIneo-Myc-S-SCAM-8, pCIneo-Myc-S-SCAM-10, and pCIneo-Myc-S-SCAM-12 were obtained from Dr. Yutaka Hata (Tokyo Medical and Dental University, Tokyo, Japan). pCIneo-Myc-S-SCAM-ΔPDZ4 (aa 929–1010 deletion) was constructed using standard molecular biology methods. KIAA1250/ARMS cDNA was obtained from Dr. Takahiro Nagase (Kazusa DNA Research Institute, Chiba, Japan). pCMV-FLAG-ARMS, pGEX4T-2-ARMS-Ct (aa 1616–1715), and pGEX4T-2-ARMS-Ct-ΔSIL (aa 1616–1712) were constructed using standard molecular biology methods. pCMV-FLAG-C3G was constructed using standard molecular biology methods. The RNAi-resistant mutant of FLAG-human C3G (pERedNLS-FLAG-C3G) was generated by mutagenesis of 5′-GTCTCATGGAGGTTA -3′ to 5′-GTCTCATGGAGTTA-3′ using the QuikChange site-directed mutagenesis kit (Stratagene). The letters in the codons, which are different from the letters in the RNAi sequence of rat C3G, are underlined. pGEX4T-1-RalGDS-3xRBD was constructed using standard molecular biology methods. The GST-fusion fragment of PDZ-GEF1 (aa 1–250) was produced in , purified, and used as an antigen to raise a pAb in rabbit. The rabbit anti-PDZ-GEF1 pAb was affinity purified by using MBP-PDZ-GEF1 (aa 1–250) immobilized on Amino-link agarose beads (Pierce Chemical Co). A rabbit anti-S-SCAM pAb was obtained from Dr. Yutaka Hata (Tokyo Medical and Dental University, Tokyo, Japan). A mouse anti-LBPA mAb was obtained from Dr. Toshihide Kobayashi (RIKEN, Wako, Japan) and Dr. Jean Gruenberg (University of Geneva, Geneva, Switzerland). A rabbit anti-Rab7 pAb was obtained from Dr. Marino Zerial (Max Plank Institute, Dresden, Germany). A rabbit anti-Rap1 pAb, a mouse anti-pan-Trk mAb (B-39), a rabbit anti-C3G pAb, a mouse anti-cMyc mAb (9E10), and a mouse anti-GST mAb (B-14) were purchased from Santa Cruz Biotechnology, Inc. A mouse anti-phospho-ERK mAb (E10) and a rabbit anti-ERK pAb were purchased from Cell Signaling. A mouse anti-Ras mAb and a rabbit anti-TrkA receptor pAb were purchased from Upstate Biotechnology. A mouse anti-EEA1 mAb and a mouse anti-GM130 mAb were purchased from Transduction Laboratories. A mouse anti-FLAG M2 mAb and a rabbit anti-FLAG pAb were purchased from Sigma-Aldrich. A mouse anti-Kidins220/ARMS mAb was purchased from Abcam. A rabbit anti-Kidins220/ARMS pAb was purchased from ABR. A mouse anti-Tau-1 mAb and a rabbit anti-MAP2 pAb were purchased from Chemicon International. A mouse anti-HA mAb was purchased from Babco. For PDZ-GEF1-knockdown, double-stranded 19-nucleotide RNA duplexes (QIAGEN) for PDZ-GEF1 #1 (5′-GAGAGAUUGUAAUGGUGAA-3′), #2 (5′-GAGUAGAGAGAGUCUUGAA-3′), scramble RNA for #1 (5′-AAUUGUAAUAGGAUGGGAG-3′), or double-stranded 25-nucleotide RNA duplexes (StealthTM RNA-mediated interference; Invitrogen) for PDZ-GEF1 #3 (5′-CGAUCCAGUAUUGUCAGCAAUUCUU-3′) was transfected into PC12 cells using Lipofectamine 2000 reagent. For C3G-knockdown, double-stranded 25-nucleotide RNA duplexes (Stealth RNA-mediated interference; Invitrogen) for C3G #1 (5′-CAGAACGAGAAAUGGAGAUUCUGAA-3′), double-stranded 19-nucleotide RNA duplexes (QIAGEN) for C3G #2 (5′-GAUGCUCAUGGAGGUCUAU-3′), #3 (5′-CCAGACUACAUAGACGGGAAGGUCA-3′), or scramble RNA for #1 (5′-AGAGUUAGCGGCGAAUAUAAAAAGC-3′) was transfected into PC12 cells using Lipofectamine 2000 reagent. For ARMS-knockdown, double-stranded 25-nucleotide RNA duplexes (StealthTM RNA-mediated interference; Invitrogen) for ARMS #1 (5′-CAGGCCGAGUAUAGAGACGCCUAUA-3′), #2 (5′-CGGCUCUCAACAGAAGGGACACUUA-3′), #3 (5′-GGGCUCCAUCCAUUCUACUCUAGAA-3′), or scramble RNA for #1 (5′-CAGGAGCAUAUAGAGCCGCUCGAUA-3′) was transfected into PC12 cells using Lipofectamine 2000 reagent. For Rap1-knockdown, double-stranded 25-nucleotide RNA duplexes (StealthTM RNA-mediated interference; Invitrogen) for Rap1 #1 (5′-CAGAAUUUAGCAAGACAGUGGUGUA-3′), #2 (5′-CAGCAAUGAGGGAUUUGUAUAUGAA-3′), or #3 (5′-UGGGAAAGUCUGCUCUGACAGUUCA-3′) was transfected into PC12 cells using Lipofectamine 2000 reagent. For detection of knockdown efficiency, the whole cell lysates of PC12 cells were subjected to SDS-PAGE, followed by immunoblotting with the anti-PDZ-GEF1 pAb, the anti-C3G pAb, the anti-ARMS pAb, or the anti-actin mAb. For FRET analysis and rescue experiment in neurite outgrowth, the siRNA expression vector, pSUPER-retro was used for expression of shRNA in PC12 cells. The following inserts were used: PDZ-GEF1 gene-specific insert was a 19-nucleotide sequence corresponding to nucleotides 740–758 (5′-GAGAGATTGTAATGGTGAA-3′) of PDZ-GEF1 cDNA, which was separated by a 10-nucleotide noncomplementary spacer (TTCAAGAGA) from the reverse complement of the same 19-nucleotide sequence. C3G gene-specific insert was a 19-nucleotide sequence corresponding to nucleotides 1407–1425 (5′-GATGCTCATGGAGGTCTAT-3′) of C3G cDNA, which was separated by a 10-nucleotide noncomplementary spacer (TTCAAGAGA) from the reverse complement of the same 19-nucleotide sequence. GEF assay was performed as described previously (). Lipid-modified GST-Rap1A was generated in Sf9 cells using the baculovirus expression system as described previously (; ). FLAG-PDZ-GEF1 and FLAG-PDZ-GEF1-ΔRA were purified from transfected HEK 293 cells as described previously (). Lipid-modified GDP-bound form of GST-Rap1A (5 pmol) was incubated at 30°C for indicated periods of time in a reaction mixture (50 μl) containing FLAG-PDZ-GEF1 (0.5 pmol) with or without the GST-Rap1A-CA (5 pmol), 50 mM Tris/HCl at pH 8.0, 12 mM MgCl, 2 mM EDTA, 0.4 mM DTT, 0.06% CHAPS, and 12 μM [S]GTPγS (6 × 10 cpm/pmol). The mixture was applied to a nitrocellulose filter, and the radioactivity retained on the filter was measured. PC12 cells were transfected with an empty pSUPER-retro vector or pSUPER-retro-PDZ-GEF1. After selection with puromycin, cells were further transfected with pRaichu-Rap1. PC12 cells expressing Raichu-Rap1 were starved for 6–12 h with phenol red-free DMEM/F12 medium containing 0.1% bovine serum albumin (BSA), and then treated with 50 ng/ml of NGF. The medium was covered with mineral oil (Sigma-Aldrich) to preclude evaporation. Cells were imaged with an IX71 inverted microscope (Olympus) equipped with a Cool SNAP-HQ cooled CCD camera (Roper Scientific) controlled with MetaMorph software (Universal Imaging), as described previously (). The filters used for the dual-emission imaging studies were obtained from Omega Optical: an XF1071 (440AF21) excitation filter, an XF2034 (455DRLP) dichroic mirror, and two emission filters (XF3075 (480AF30) for CFP and XF3079 (535AF26) for YFP). Cells were illuminated with a 75-W Xenon lamp through a 12% ND filter and viewed through a 60x immersion objective lens. The exposure times for 4 × 4 binning were 400 msec for CFP and YFP images, and 100 msec for differential interference contrast (DIC) images. After background subtraction, YFP/CFP ratio images were created with the MetaMorph software and the images were used to represent FRET efficiency. Staining was performed as follows: PC12 cells were seeded onto poly-- lysine coated coverslips in 24-well plates a day before the experiment. PC12 cells were cultured in DME with serum for 24 h, serum-starved for 16 h, treated with bafilomycin for 2 h, and stimulated by NGF for 30 min. The cells were fixed with 4% paraformaldehyde for 15 min and permeabilized with 0.05% Saponin for 30 min at room temperature. Images were captured using a Carl Zeiss confocal laser scanning microscope using a 63× oil immersion objective lens (model LSM 510-V3.2; Carl Zeiss MicroImaging, Inc.) or a confocal laser scanning microscope (model TE2000; Nikon) using a 60× oil immersion objective lens. The fluorescence area and intensity of each protein in PC12 cells were measured by use of NIH Image software. Collected data were exported as 8-bit TIFF files and processed using Adobe Photoshop 7.0. The pull-down assay was performed as described previously (). PC12 cells were transfected with siRNA for 2 d and stimulated by NGF for indicated periods of time. The PC12 cells were washed with 1 ml of ice-cold PBS and lysed in Buffer A (50 mM Tris/HCl at pH 7.4, 150 mM NaCl, 5 mM MgCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM phenylmethylsulfonyl fluoride, 1 mM sodium vanadate). The samples were centrifuged at 100,000 x for 10 min, and the supernatant was collected as the cell lysates. For the Rap1 pull down assay, the cell lysate (600 μg of protein) was incubated with GST-RalGDS Rap binding domain (RBD) (10 μg of protein) immobilized on glutathione-Sepharose beads (50 μl) for 30 min. After the beads were washed three times with Buffer A, GTP-Rap1 was detected by immunoblotting with an anti-Rap1 pAb. For the Ras pull-down assay, the cell lysate (600 μg of protein) was incubated with GST-c-Raf Ras binding domain (RBD) (30 μg of protein) immobilized on glutathione-Sepharose beads (50 μl) for 30 min. After the beads were washed three times with Buffer A, GTP-Ras was detected by immunoblotting with an anti-Ras mAb. 5% of the total cell lysates of PC12 cells was used for immunoblotting with either the anti-phospho-ERK mAb or the anti-ERK pAb to determine the amount of activated ERK (Phospho-ERK) or the total amount of ERK (Total-ERK), respectively, in each transfected cells. PC12 cells were obtained from Dr. Shinya Kuroda (Tokyo University, Tokyo, Japan) and maintained in DMEM with 5% horse serum and 10% bovine calf serum (Hyclone) in a 10% CO atmosphere. PC12 cells were cultured in DMEM with serum for 48 h and stimulated by NGF for 24 h. Images were captured using a confocal laser scanning microscope using a 60× oil immersion objective lens (model TE2000; Nikon). Collected data were exported as 8-bit TIFF files and processed using Adobe Photoshop software. The assay for neurite outgrowth was performed on 40–60 cells randomly chosen in each group. The number of primary neurites per cell was defined as the number of thin cell processes with a length longer than one cell diameter. The statistical significance of differences between each group was analyzed by the two-tailed test. Rat hippocampal neurons were prepared form embryonic day (E) 18 Sprague-Dawley rats with slight modifications as described (). For cDNA transfection, before plating, the cells were resuspended in an optimized transfection solution for primary rat hippocampal neurons (Amaxa). Each sample (2 × 10 cells in 100 μl) was transfected with 2 μg cDNA as indicated by using Nucleofector electroporation device (Amaxa), according to the optimized protocol for primary rat hippocampal neurons (Amaxa). Immediately after transfection, the cells were cultured in 24-multiwells at a density of 3 × 10 cells/well. After 1.5 d culture, neurons were processed for immunohistochemistry. Images were captured using a confocal laser scanning microscope using 40 and 60× oil immersion objective lens (model TE2000; Nikon). Collected data were exported as 8-bit TIFF files and processed using Adobe Photoshop software. The morphometrical analysis for axons and minor neurites was performed on between 40 and 60 GFP-positive neurons (). The number of axons or minor neurites per cell was defined as the number of Tau-1-positive processes at least twice as long as the other processes or MAP2-positive processes longer than one cell diameter, respectively. The length of individual axons for each neuron was measured by use of the NIH Image software. Axon length encompasses all visible parts of an axon without the length of its branches. The statistical significance of differences between each group was analyzed by the two-tailed test. Immunoisolation of vesicles was performed as described previously (). In brief, PC12 cells (cultured on 4 × 10 cm plates) were starved for 16 h in phenol red-free DMEM/F12 medium containing 0.1% BSA, and either left unstimulated or stimulated by NGF for 30 min. Subsequently, the cells were cracked with a ball-bearing homogenizer in Buffer B. Subsequently, a PNS was prepared by centrifugation at 3,000 rpm for 10 min (Kubota 5800). Dynabeads Protein A (DYNAL Inc.) magnetic beads were coated with the anti-PDZ-GEF1 pAb or the control IgG at a density of 10 μg Ab per 50 μl beads according to manufacturer's instructions. The anti-PDZ-GEF1 pAb-coated beads were added to the PNS and incubated with continuous rotation at 4°C for 2 h. The immunoisolated beads were washed five times with Buffer B using a magnet and transferred to fresh tubes. The beads were finally resuspended in 80 μl of the SDS sample buffer for SDS-PAGE, followed by immunoblotting with the mouse anti-EEA1 mAb, the rabbit anti-Rab7 pAb, the anti-ARMS mAb, and the rabbit anti-PDZ-GEF1. Fig. S1 shows the effect of GTP-Rap1 on the GEF activity of PDZ-GEF1-ΔRA. Fig. S2 shows the effect of overexpression of PDZ-GEF1 on neurite outgrowth. Fig. S3 shows the recruitment of C3G to early endosomes in a NGF- dependent manner. Fig. S4 shows the Inhibition of the transport of TrkA receptor from early endosomes to late endosomes by bafilomycin. Fig. S5 shows the expression levels of S-SCAM and ARMS in S-SCAM-expressing- and ARMS-knockdown PC12 cells, respectively. Online supplemental material is available at .

Myelinated axons are organized into a series of discrete domains that are distinguishable by their molecular composition and physiological function. These domains include the nodes of Ranvier, which are enriched in voltage-gated sodium channels essential for saltatory conduction, the flanking paranodal junctions, and the juxtaparanodes, which are enriched in Shaker type K channels (; ). Each of these domains forms as the result of instructive contact-dependent signals from myelinating glia (i.e., Schwann cells in the peripheral nervous system [PNS] and oligodendrocytes in the central nervous system). Adhesion molecules on the glial cell bind to and recruit a complex of axonal adhesion molecules and cytoskeletal proteins; the latter include ankyrin G at the node and 4.1B at the paranodes and juxtaparanodes. Interactions with these cytoskeletal proteins target and stabilize the localization of additional proteins (i.e., sodium channels at the node and potassium channels at the juxtaparanodes). However, together, these domains (the node, paranodes, and juxtaparanodes) only account for ∼1% of the longitudinal extent of the axon. The remaining and by far the largest domain of the myelinated axon is the internode, the portion of the axon located under the compact myelin sheath. Axons and myelinating glia exhibit an intimate functional relationship in this region, as reflected in the highly regular apposition of their respective plasma membranes, which are separated by 12–13 nm. This separation persists after osmotic changes or in various pathologic states (). Conversely, the periaxonal space as well as attachment of the myelin sheath to the axon is disrupted by the action of proteases (). These results indicate that interactions between the glial and axonal membranes along the internode are actively maintained by cell surface proteins. The molecules that mediate axonal–glial interactions along the internode have remained largely elusive. The myelin-associated glycoprotein (MAG), a member of the Ig superfamily expressed by myelinating Schwann cells and oligodendrocytes, has been specifically localized to this region (). MAG is expressed in the periaxonal glial membrane at initial stages of myelination () and interacts with several axonal components (); at later stages of myelination, it localizes to Schmidt-Lanterman incisures as well (). However, mice deficient in MAG myelinate appropriately and exhibit only modest alterations in the periaxonal space (; ), suggesting that other molecules are likely to mediate axo–glial adhesion along the internode. Recently, a family of adhesion molecules termed the Nectin-like (Necl) proteins were described (). Like the nectins, a family of adherens junction proteins, the Necl proteins contain three extracellular Ig-like domains, a single transmembrane domain, and a short cytoplasmic segment (). Necl proteins notably differ from the nectins in their cytoplasmic sequences, which are linked to the cytoskeleton via a FERM (4.1, ezrin, radixin, moesin)-binding domain in their juxtamembrane region and contain a class II PDZ (PSD-95, DLG, Z01)-binding sequence at their C terminus. The Necl proteins have been implicated in a variety of biological activities, including cell adhesion, regulation of cell growth and synaptic function, and cell polarity (for review see ). They were originally identified as tumor suppressor in lung cancer 1 (TSLC1)–like proteins, as their limited expression in lung cancer cell lines (; ; ) and other tumor types () correlates with abnormal cell proliferation. They have also been shown to promote synapse formation and glutamate receptor clustering, leading to an alternative designation as synaptic cell adhesion molecules (SynCAMs; ). Five Necl proteins have been identified, each independently by several groups, which accounts for the diverse nomenclature. The Necl proteins include Necl-1/TSLL1/SynCAM3/IgSF4B (), Necl-2/TSLC1/SynCAM1/IGSF4/RA175/SgIGSF (; ; ), Necl-3/SynCAM2, Necl-4/TSLL2/SynCAM4, and Necl-5/Tage4 (). As Necl-5 lacks the characteristic FERM- and PDZ-binding motifs of the cytoplasmic domain, its classification as a member of the Necl family is controversial (). Necl-2 is the prototype and most intensively studied member of this family. It is widely and highly expressed, including in the developing nervous system and various epithelial tissues (; ; ). Necl-2 promotes adhesion via homophilic (; ; ) and heterophilic interactions (). Like other Necl proteins (; ), Necl-2 is believed to mediate adhesion as a cis-dimer. Necl-2 has been reported to interact intracellularly with DAL1, a truncated form of the band 4.1B protein (). Other Necl proteins are highly expressed in the nervous system, including Necl-1 (; ) and Necl-4 (). They are inferred to mediate adhesion although their ligands, and their modes of adhesion are less well established. Because of their role in mediating cell adhesion and their linkage to 4.1 proteins, including 4.1B, which has been localized to the paranodal and juxtaparanodal domains (; ; ), we considered the Necl proteins to be candidates for the mediation of axo–glial interactions in one or more of the domains of myelinated fibers. We now report that the Necl proteins are localized to the internode of myelinated fibers in the PNS, that they bind to both axons and Schwann cells via heterophilic interactions, and that Necl-4 is up-regulated with and essential for myelination. These results implicate the Necl proteins as a major new set of cell adhesion molecules that mediate and are required for the axo–glial interactions along the internode of myelinated axons. To characterize their expression in axo–glial interactions and for use in functional studies, we isolated full-length cDNAs for Necl-1–4 by PCR amplification using Necl sequences present in the genomic database. An alignment of the cloned Necl protein sequences is shown in Fig. S1 A (available at ); their schematic organization is shown in . To examine the pattern of Necl protein expression in Schwann cells, neurons, and peripheral nerves, we used a series of antibodies. These included affinity-purified polyclonal antibodies to the C termini of Necl-1 and -2; the latter were previously reported (; ). Analysis demonstrated that these antibodies were of high titer but exhibited some cross-reactivity with the other Necl proteins; in particular, both Necl-2 antibodies cross reacted with Necl-3 (Fig. S2 B, available at ), likely as a result of the extensive identity of the C termini of these two proteins. To generate specific antibodies to the Necl proteins, we constructed recombinant proteins containing the ectodomain of each Necl protein fused to the human Ig Fc domain. These Necl-Fc fusion proteins were individually purified and injected into guinea pigs, resulting in the generation of antisera to Necl-1, -3, and -4. Western blotting (Fig. S2 B) and immunocytochemistry of permanently transfected CHO cell lines expressing full-length Necl proteins (Fig. S2 C) corroborated the specificity of each of these antisera. For Western blot analysis, detection of the Necl proteins was markedly enhanced by treating lysates with peptide -glycosidase F to remove N-linked oligosaccharides (Fig. S2 A); this likely reflects the extensive glycosylation of the Necl protein ectodomain (; ). We first analyzed expression of the Necl proteins by immunofluorescence of Schwann cells and dorsal root ganglion (DRG) neurons (). Necl-1 was not expressed by Schwann cells but was highly expressed at the membranes of DRG neuron cell bodies and neurites. Antibodies to the C terminus of Necl-2 stained Schwann cell membranes and nuclei and robustly stained the membranes of neurites and DRG somas. Necl-3 did not stain Schwann cells or neurites, although low level Necl-3 expression was detected at the membranes of some neuronal somas. The lack of Necl-3 expression on Schwann cells and DRG neurites suggests that their staining by Necl-2 antibodies reflects the bone fide expression of Necl-2 at these sites and is not the result of cross-reactivity with Necl-3. Necl-4 was strongly expressed at the membranes of Schwann cells, including fine filopodial-like extensions of the Schwann cell membrane (unpublished data). No staining of neurites or neuronal cell bodies by Necl-4 could be detected. These results indicate that neurons and Schwann cells express quite different sets of Necl proteins: neurites strongly express Necl-1 and -2, whereas Schwann cells principally express Necl-4 and some Necl-2. Interestingly, Necl-1–3 accumulated at sites of contact between neuronal somas in contrast to Necl-2 and -4, which were more diffusely expressed by Schwann cells. Western blotting of lysates after deglycosylation confirmed that Necl-1 is expressed by DRG neurons but not Schwann cells, whereas Necl-4 is expressed by Schwann cells but not by DRG neurons (). Necl-2 was expressed by Schwann cells and at substantially higher levels by DRG neurons. Each of these proteins (i.e., Necl-1, -2, and -4) accumulated in 21-d-old Schwann cell/neuron cocultures, particularly under myelinating conditions. These results strongly suggest that expression of the Necl proteins is up-regulated with myelination. We could not detect Necl-3 expression by Schwann cells and detected only faint expression in neurons by Western blotting, which is consistent with the immunofluorescence data. These latter results further indicate that the band recognized by the anti–Necl-2 antibodies in Schwann cell and neuronal lysates is indeed Necl-2. We also examined Necl expression by PCR and Northern blot analysis. Northerns confirmed the high level expression of Necl-4 in sciatic nerves (); analysis of developing sciatic nerve demonstrated the expression of Necl- 4 mRNA perinatally that increased with myelination (). PCR analysis confirmed that Necl-1 and Necl-4 mRNA expression was principally restricted to DRG neurons and Schwann cells, respectively (Fig. S3 A, available at ), whereas Necl-2 is expressed by both cell types (Fig. S3 B). Modest differences in the expression of six previously described Necl-2 isoforms, which vary slightly in their ectodomain sequences (), were also observed by PCR analysis of neuron and Schwann cell first-strand templates (Fig. S3 B). To characterize the distribution of the Necl molecules in myelinated fibers, we stained adult mouse teased sciatic nerves. Staining demonstrated striking and robust expression of all three Necl proteins along the internode (i.e., the region under the compact myelin sheath extending into the juxtaparanodal region; ). Necl-1 and -2 are essentially excluded from the paranodes (, insets), as indicated by the limited overlap with the paranodal marker Caspr (). Necl-4 is also expressed along the internode and juxtaparanodes but could be detected in some paranodes, although at reduced levels. None of the Necl proteins were detectable at the node. We were unable to detect any specific staining for Necl-3 in the teased sciatic nerves (unpublished data). These findings identify the Necl proteins as novel components of the internodal and juxtaparanodal domains. Necl-1, -2, and -4 are also expressed at high levels in the Schmidt-Lanterman incisures of myelinating Schwann cells (, arrowheads). Necl-4 was expressed throughout all layers of the incisures, whereas Necl-1 and -2 were more variable and were frequently present only in the outermost incisural layers. The expression of Necl-1 in the incisures, a Schwann cell–specific structure, was unexpected, as Western blots demonstrated no expression of Necl-1 by Schwann cells (). Therefore, we undertook a developmental Northern analysis of Necl-1 in sciatic nerve (Fig. S3, C and D). This confirmed that Necl-1 mRNA is faintly expressed in sciatic nerve but has a delayed onset of expression compared with Necl-4. These results demonstrate that all three of these Necl proteins are expressed in the clefts, indicating that Necl-1 and -2 expression is up-regulated in Schwann cells with myelination. Finally, we compared the localization of Necl-1, which is enriched on axons, to Necl-4, which is expressed by Schwann cells. We stained sciatic nerves with a rabbit polyclonal antibody to the C terminus of Necl-1 and the guinea pig antibody to Necl-4. These proteins exhibited essentially identical expression patterns along the internodal domain (, yellow in the merged image). An example of a node is shown at higher magnification in . In the incisures, these proteins were frequently but not precisely colocalized; illustrates an example showing Necl-1 only in the outer layers in contrast to Necl-4, which is present throughout all layers. Double staining for myelin basic protein (MBP) confirmed that Necl-4 (, red) is localized just internal to the compact myelin sheath (, blue) adjacent to the axon. The distribution of Necl-4 in the periaxonal glial membrane and incisures is similar to that described for MAG (); double staining confirms that these proteins indeed colocalize in these domains (). Collectively, these studies indicate that Necl-1 on the axon is directly apposed by Necl-4 on the periaxonal Schwann cell membrane in the internodal domain and that the Necl proteins are new constituents of the Schmidt-Lanterman clefts. To analyze their potential function in axo–glial interactions, we systematically examined the ability of the Necl proteins to mediate homophilic or heterophilic adhesion, focusing on Necl-1, -2, and -4, which are expressed in the PNS. Interactions mediated by Necl-3, which is not expressed at detectable levels in the PNS, were also examined and are shown in Fig. S4 (available at ). To characterize binding within the Necl protein family, we incubated Necl-Fc fusion proteins with monolayers () or dissociated cell suspensions of transfected CHO cells expressing epitope-tagged Necl proteins (Fig. S4 B). Live staining with an anti-HA antibody verified that the Necl proteins are expressed at the surface of the transfected cells ( and S4) and facilitated the isolation of high level expressing cells via FACS. Necl-Fc protein binding was monitored with a fluorescently labeled anti–human Fc antibody. In the case of the dissociated CHO cells, this binding was quantitated by FACS analysis; results are summarized in . In general, the Necl proteins preferentially mediate heterophilic adhesion. The most robust binding detected was between Necl-1 and -4; substantial binding of Necl-1 to -2 was also observed as previously reported (). We also detected modest homophilic binding by Necl-2 but not by Necl-1 or -4. No binding of Necl-Fc proteins to control CHO cells (i.e., cells transfected with the empty pcDNA3.1 vector) was observed, indicating that Necl-Fc binding to transfected cells was specific. As axons express Necl-1 and Schwann cells principally express Necl-4, these results suggest that Necl-1 on axons binds heterophilically to Necl-4 on Schwann cells. We next characterized binding of the Necl proteins to purified cultures of DRG neurons and Schwann cells using a similar strategy. In general, Necl-2 and -4 bound strongly to DRG neurites, whereas Necl-1 and -2 bound robustly to Schwann cells (). No binding of Necl-1 to neurons and minimal binding of Necl-4 to Schwann cells was observed. These findings indicate that the Necl proteins mediate specific adhesion to neurons and Schwann cells. To test the ability of the Necl proteins to promote direct Schwann cell adhesion, we examined Schwann cell attachment to Necl-Fc fusion proteins spotted onto plastic dishes at increasing concentrations; human IgG served as a control. The density of adherent Schwann cells at each Necl protein site was then measured. Representative images are shown in , and the quantitation of four separate experiments is summarized in . Schwann cells bound at high density to the substrate-adsorbed Necl-1–Fc but not to the Necl-2– or -4–Fc proteins. These results confirm that Necl-1 is able to promote Schwann cell attachment and suggest that it mediates physiologically relevant binding. These results highlight a potential role of the Necl proteins in general and of Necl-4 on Schwann cells in particular in the axo–glial interactions of PNS myelination. To characterize the function of the Necl proteins, we focused on Necl-4, which is the major Necl expressed by Schwann cells and is strikingly up-regulated with myelination in the cocultures (). For the knockdown, we targeted two distinct sequences within the first Ig domain of Necl-4 (Fig. S5 E, available at ) and subcloned the corresponding sequences into the pLL3.7 vector (). We infected purified Schwann cells with the lentiviral short hairpin RNA (shRNA) constructs; Schwann cells infected with the empty pLL3.7 vector or pLL3.7 targeting a nonspecific sequence in luciferase (shLuc) served as controls. Both shRNAs to Necl-4 resulted in an essentially complete knockdown of Necl-4 in Schwann cells (Figs. S5 A and 6 C), whereas Necl-2 levels were unchanged, underscoring the specificity of the knockdown. Schwann cells were then added to established neuron cultures and maintained under myelinating conditions for an additional 10 d. Knockdown of Necl-4 in Schwann cells resulted in a striking inhibition of myelin segment formation compared with control (uninfected) Schwann cells, Schwann cells infected with vector alone, or the shRNA to luciferase (). Quantitation of the number of MBP-positive myelin segments demonstrated that the shRNA constructs to Necl-4 significantly inhibited myelination, corresponding to ∼75% inhibition for shNecl-4 construct #1 (P < 0.001) and essentially complete inhibition with the second (shNecl-4 #2) construct (P < 0.001). In a complementary and independent set of experiments, we also infected premyelinating neuron–Schwann cell cocultures with shRNA construct #1; myelination was then induced by the addition of ascorbate to the culture media. shRNA knockdown again resulted in the extensive inhibition of myelination in the cocultures (quantitated in Fig. S5 D). In contrast, infection of neurons with the Necl-4 shRNA construct before seeding with Schwann cells had no effect on myelination in cocultures (unpublished data). To corroborate the specificity of the shRNA effects on myelination and to exclude nonspecific off-target effects (for review see ), we performed a knockdown rescue experiment. We expressed a modified Necl-4 protein in Schwann cells treated with shRNA (construct #1); this Necl-4 protein contained codon substitutions that rendered it insensitive to shRNA treatment while preserving its amino acid sequence (Fig. S5 E). The modified Necl-4 protein was robustly expressed in shRNA-treated Schwann cells () and, notably, fully rescued myelination (). These results underscore the specificity of the shRNA knockdown and confirm the crucial role of Necl-4 in PNS myelination. Intriguingly, overexpression of the Necl-4 construct may enhance myelination in the cocultures based on increased numbers of myelin segments (). These results suggest that Necl-4 may provide a positive signal for myelination. Schwann cells remained associated with axons despite the knockdown of Necl-4 based on phase microscopy (unpublished data) and immunofluorescence staining (Fig. S5 F). Schwann cell numbers were comparable in the knockdown cultures (Hoechst stain; ). Western blot analysis of the effects of Necl-4 knockdown corroborated the inhibition of myelin protein expression () and also demonstrated substantial effects on Schwann cell transcription factor expression (). Thus, there was dramatic inhibition of Oct-6, a transcription factor expressed by promyelinating Schwann cells (), and nearly complete inhibition of the expression of Krox-20, a transcription factor required for the myelinating phenotype of Schwann cells (). These results indicate that Necl-4 expression is required for progression to the promyelinating stage and subsequent myelination of axons. Finally, we tested the effects of knockdowns of Necl-1 and -4 on the binding of Necl proteins to axons and Schwann cells. Knockdown of Necl-4 effectively ablated the expression of Necl-4 at the Schwann cell membrane (). In the absence of Necl-4, Necl-1 no longer bound to Schwann cells (). In contrast, Necl-2 still bound to Schwann cells in the absence of Necl-4 (), indicating that its binding is independent of Necl-4. In complementary studies, we performed a knockdown of Necl-1 on neurons by lentiviral-mediated shRNA (). This effectively inhibited Necl-1 expression based on Western blotting (Fig. S5 B). In the absence of Necl-1, Necl-4 binding to neurites was substantially reduced (). Together, these results indicate that Necl-1 on the axon and Necl-4 on the Schwann cell are obligate binding partners. h a v e d e m o n s t r a t e d t h a t n e u r o n s a n d S c h w a n n c e l l s e x p r e s s d i s t i n c t m e m b e r s o f t h e N e c l p r o t e i n f a m i l y . W e a l s o s h o w t h a t t h e N e c l p r o t e i n s p r e f e r e n t i a l l y m e d i a t e h e t e r o p h i l i c i n t e r a c t i o n s ( w i t h N e c l - 1 o n t h e a x o n b i n d i n g t o N e c l - 4 o n t h e S c h w a n n c e l l ) , t h a t t h e s e p r o t e i n s d e l i n e a t e t h e i n t e r n o d a l d o m a i n , a n d t h a t N e c l - 4 m e d i a t e s a x o n a l – g l i a l i n t e r a c t i o n s r e q u i r e d f o r m y e l i n a t i o n . T h e s e r e s u l t s h a v e i m p o r t a n t i m p l i c a t i o n s f o r t h e d o m a i n o r g a n i z a t i o n o f a x o n s a n d t h e m e c h a n i s m s t h a t i n i t i a t e S c h w a n n c e l l m y e l i n a t i o n , a s c o n s i d e r e d f u r t h e r b e l o w . To clone the rat Necl-1, sense (ATGGGGGCCCCTTCCGCCCT) and antisense (CTAGATGAAATATTCCTTCTTGTCATCCCCGCC) primers were used to PCR amplify a full-length cDNA from rat DRG neuron first strand (GenBank/EMBL/DDBJ accession no. ). To clone Necl-2, we designed primers based on the mouse cDNA for TSLC1 (). We amplified a full-length cDNA from adult mouse brain with sense (ATGGCGAGTGCTGTGCTGCC) and antisense (CTAGAAGTACTCTTTCTTTTCTTCGGAGTT) primers. This mouse isoform of Necl-2 corresponds to isoform 4 of SynCAM 1 () that lacks the putative mucin-like domain of the predicted rat sequence. The mouse Necl-2 cDNA is available from GenBank/EMBL/DDBJ (accession no. ). Necl-3 was amplified from rat Schwann cell first strand using sense (ATGATTTGGAAACGCAG) and antisense (TTAAATGAAATACTCTTTTTTCTC) primers, and the sequence is available from GenBank/EMBL/DDBJ (accession no. ). A full-length Necl-4 cDNA was amplified from rat Schwann cell first strand using sense (ATGGGCCGGGCCCGGCGCTTCC) and antisense (AATGAAGAATTCTTCTTTCCGTTTGTGTCCATCGCCGC) primers, and the sequence is available from GenBank/EMBL/DDBJ (accession no. ). CHO cells were transfected with full-length constructs for each Necl protein and with the empty pcDNA3.1 vector using the LipofectAMINE 2000 reagent (Invitrogen); permanently transfected cells were selected by maintenance in G418-containing medium (α-MEM, 10% FBS, and 750 μg/ml G418). To facilitate analysis, the sequence encoding the influenza HA epitope was added by PCR to the N terminus of each Necl protein immediately after the signal peptide as predicted by the SignalP program. CHO cells were stained for the HA epitope and selected by FACS (MoFlo; DakoCytomation). Constructs encoding the extracellular domain of each Necl protein fused to the hinge region of the human IgG-Fc were subcloned into the pcDNA 3.1 TOPO vector (Invitrogen) and transiently transfected into HEK 293FT cells (Invitrogen). Transfected cells were maintained in media containing DME, 1% Ultra Low IgG FBS (Invitrogen), 1 mM nonessential amino acids (Invitrogen), and 2 mM -glutamine. Conditioned media was collected after 3 d and alkalinized by 1 M Hepes buffer (Invitrogen) at 10% vol, and the salt concentration was increased by adding 10× Dulbecco's PBS (dPBS) at 25% vol. The conditioned media was then filtered, and Necl-Fc proteins were collected batchwise with protein A–agarose beads (Roche). The beads were washed with 6× dPBS plus 0.2% Triton X-100 (Sigma-Aldrich) followed by 6× dPBS and 1× dPBS washes. The proteins were eluted off the beads with 200 mM glycine buffer at pH 2.8, immediately neutralized with 1 M Hepes, pH 8.5, at 10% vol, and dialyzed overnight against dPBS (1:500 vol/vol) using Slide-A-Lyser dialysis cassettes (Pierce Chemical Co.). Protein concentrations were assessed with the BCA Protein Assay (Pierce Chemical Co.). Biochemical analysis confirmed that each construct encoded proteins of the predicted size of ∼65 kD after deglycosylation. Guinea pig polyclonal antibodies were generated against Necl-1, -3, and -4 by immunization with the corresponding Necl-Fc fusion protein. Antibodies to the human Fc moiety were removed by passing antiserum over an agarose–human IgG (Jackson ImmunoResearch Laboratories) column. Rabbit polyclonal antibodies to Necl-2 have previously been described (; ). A rabbit polyclonal antibody was raised to the C-terminal sequence (AEGGQSGGDDKKEYFI) of Necl-1 coupled to keyhole limpet hemocyanin and injected into rabbits, and the resulting antibodies were affinity purified against the immunizing peptide coupled to Sepharose beads. Other antibodies included mouse monoclonal antibodies to MBP (SMI-94) and neurofilament (SMI-31 and SMI-32; all obtained from Sternberger Monoclonals) and the HA epitope (HA.11; Covance). Polyclonal antibodies specific to Krox-20 and Oct-6 (provided by D. Meijer, Erasmus University, Rotterdam, Netherlands), Caspr (gifts from E. Peles [Weizmann Institute, Rehovot, Israel] and M. Bhat [University of North Carolina, Chapel Hill, NC]), MAG (), and neurofilament (Covance) were also used. Secondary antibodies included donkey anti–mouse (FITC, rhodamine X, Cy5, and amino-methylcoumarin conjugated), donkey anti–guinea pig (FITC and rhodamine X conjugated), donkey anti–rabbit (FITC and rhodamine X conjugated), donkey anti–chicken (Cy5 and amino-methylcoumarin conjugated), and goat anti–human-Fc (FITC and rhodamine X conjugated; Jackson ImmunoResearch Laboratories). For FACS analysis, we also used phycoerythrin (PE)-conjugated donkey anti–mouse (eBioscience). Establishment of primary rat Schwann cell and DRG neuron cultures has been described previously (). In brief, neurons were isolated from embryonic day 16 DRGs by trypsinization and plated on a collagen substrate (Biomedical Technologies Inc.) in standard neuronal medium (neurobasal medium, 2% B27 supplement, 2 mM -glutamine, 0.4% glucose, and 50 ng/ml 2.5S NGF). Nonneuronal cells were removed by feeding the cultures every 2 d alternately with standard neuronal medium supplemented or not supplemented with 5-fluorodeoxyuridine and uridine (both at 10 μM) over a week. Schwann cells prepared from postnatal day 2 sciatic nerves () were expanded in D media (DME, 10% FBS, and 2 mM -glutamine) supplemented with 4 μM forskolin and 5 ng/ml of the EGF domain of rhNRG-1-β1 (over a period of 2–3 wk; R&D Systems). Schwann cells were then maintained in D media for 3 d before use. Myelinating Schwann cell–neuron cocultures were established by seeding purified DRG neuron cultures with 200,000 Schwann cells in C media (MEM, 10% FBS, 2 mM -glutamine, 0.4% glucose, and 50 ng/ml 2.5S NGF). After 3 d, cocultures were supplemented with 50 μg/ml ascorbic acid to initiate basal lamina formation and myelination and were maintained for an additional 10–21 d. DME was obtained from BioWhittaker Bioproducts. α-MEM, MEM, neurobasal media, B27 supplement, -glutamine, trypsin, and G418 were purchased from Life Technologies. Glucose, forskolin, 5-fluorodeoxyuridine, uridine, and ascorbic acid were purchased from Sigma-Aldrich. FBS was obtained from Gemini, and NGF was purchased from Harlan Bioproducts. Sciatic nerves were removed from 10–12-mo-old C57BL mice, fixed with 4% PFA (Electron Microscopy Sciences) for 2 h, and stored in dPBS (Invitrogen) at 4°C until teased. Teased sciatic nerve fibers were mounted on glass slides, dried overnight at room temperature, and stored at −80°C until use for immunofluorescence staining. To further characterize homophilic versus heterophilic Necl binding, we performed a FACS analysis with Necl-Fc constructs. CHO cells expressing Necl proteins were briefly trypsinized for 1 min with 0.125% trypsin and 0.5 mM EDTA at room temperature, collected in ice-cold HBSS medium (Invitrogen) containing 10% FBS to terminate trypsinization, and washed twice in HBSS medium. Cells were diluted in HBSS to a final concentration of 750,000 cells in 200 μl and incubated with either one of the Necl-Fc proteins or whole human IgG at a final concentration of 200 nM; cells were incubated on ice for 30 min. Anti-HA antibody at 1:500 was also added to detect the CHO-Necl cells. After incubation, cells were washed once with ice-cold HBSS and incubated with PE-conjugated donkey anti–mouse (to detect expressing cells) and FITC-conjugated anti–human Fc (to detect Necl-Fc binding; both at 1:100) for 30 min on ice. After two washes in ice-cold HBSS and one wash in Ca/Mg2-free ice-cold dPBS, fluorescence measurements of individual cells were performed using FACS (FACScan; Becton Dickinson). Log fluorescence was collected for FITC (channel FL1-H) and PE (channel FL2-H) and displayed as double-parameter (PE/FITC) graphs. For each CHO-Necl clone, the FL2-H channel was calibrated by labeling cells for the HA tag only (no FITC), whereas the FITC channel (no PE) was calibrated by detecting Necl-Fc, giving the strongest binding as determined initially by binding to cells in culture. Additional controls included omitting anti-HA antibodies and Necl-Fc proteins, with only PE- and FITC-conjugated secondary antibodies added. Analysis was first gated on single cells on the basis of forward and side light scatter based on data acquisition from 10,000 cells. Each CHO-Necl clone was incubated with human IgG as a control. The FITC fluorescence value below which 99% of the events were found was noted and used as the boundary between no binding versus binding (Fig. S4 B, left and right red boxes). Binding of Necl-Fc was determined by counting the number of cells in the binding gate and dividing by the total number of cells. FACS analysis was performed with the FlowJo software package (Tree Star, Inc.). Necl-Fc proteins serially diluted in HBSS were spotted (2 μl/spot) onto 100-mm nontissue culture polystyrene dishes (Fisherbrand; Fisher Scientific) to minimize nonspecific cell attachment to the plastic. Duplicate spots were made for each dilution of Necl-Fc per experiment. Plates were incubated in a 37°C tissue culture incubator for 1 h, washed twice with HBSS, blocked with 1% BSA in HBSS for 1 h at 37°C, washed twice with HBSS, and washed once with D media. Schwann cells were briefly trypsinized, resuspended in D media at a density of 250,000 cells/ml, and 10 ml were added to each plate. The cultures were incubated in a 10% CO and 37°C tissue culture incubator for 90 min. The plates were then gently washed with dPBS twice, and cells were fixed with 4% PFA. Phase-contrast digital images of the spots (10× fields) were captured with a microscope (Eclipse TE2000-U; Nikon), and the number of cells per field was determined using ImageJ software (National Institutes of Health). To generate shRNAs, we used the pLentiLox (pLL3.7) vector in which the U6 promoter drives shRNA expression and GFP is expressed under separate promoter control (provided by L. Van Parijs; ; ). Two 21 nucleotide shRNAs (#1: nt 79–99, GTGCAGACAGAGAATGTGACG; #2: nt 151–171, GGGTCTATAGTCGTCATTCAG) that targeted sequences within the first IgG domain of Necl-4 were designed using Easy siRNA (ProteinLounge), BLOCK-iT RNAi designer (Invitrogen), and siRNA sequence selector (CLONTECH Laboratories, Inc.). The shRNA stem loops for pLL3.7 vector were designed to contain a sense shRNA sequence followed by a short (9 nt) nonspecific loop sequence and an antisense shRNA sequence followed by six thymidines, which serve as a stop signal for RNA polymerase III. The 5′-phosphorylated PAGE-purified oligonucleotides were annealed and subcloned into HpaI–XhoI sites of pLL3.7. The lentiviral vector was transfected into 293FT cells together with packaging plasmids Δ8.9 and pCMV-VSVG (provided by J. Milbrandt, Washington University, St. Louis, MO) using LipofectAMINE 2000 (Invitrogen). As controls, we used the empty pLL3.7 vector or a vector encoding shRNA to a nonspecific (luciferase) sequence. Viral supernatants were collected 72 h after transfection, centrifuged at 3,000 rpm for 15 min, aliquoted for one-time use, and frozen at −80°C. An shRNA targeting a 21-nt sequence (nt 733–753; GTGCAGACAGAGAATGTGACG) in the third IgG domain of Necl-1 was designed using the same approach. Freshly plated Schwann cells (10 cells per 100-mm plate) were incubated for 3 d with viruses at a 2/3 dilution (vol/vol) in D media (DME, 10% FBS, and 2 mM -glutamine) supplemented with forskolin and rhNRG-β1 (EGF domain). Cells were expanded for an additional week and maintained for 3 d in D media before use. Protein knockdowns were confirmed by Western blotting and by immunohistochemistry. RNA was isolated from rat sciatic nerves and Schwann cells by CsCl gradient centrifugation (). Equal amounts (10 μg) of total RNA were electrophoresed in 1% agarose and 2.2 M formaldehyde gels, transferred to nylon membranes (Duralon; Stratagene) in 6× SSC, and UV cross-linked (0.12 J). Blots were prehybridized, hybridized, and washed using standard techniques; the final stringency of the wash was 0.2× SSC at 65°C for 30 min (). cDNAs corresponding to nt 1–981 of rat Necl-1 and nt 1–966 of rat Necl-4 were used as probes. The probes were generated by PCR, isolated by agarose gel electrophoresis, and purified with the QIAquick Gel Extraction kit (QIAGEN). Sequence alignments were performed with the Clustal V program (). The Necl proteins were analyzed for the presence of putative domains using the following programs and databases: SignalP () and TMHHM (), Pfam (), SMART (), and Scansite 2.0 (). Statistical analysis was performed with the Prism software package (GraphPad). Fig. S1 presents amino acid sequences of the Necl proteins. Fig. S2 shows Necl expression CHO cell lines and specificity of anti-Necl antibodies. Fig. S3 shows PCR of Necl-2 isoforms and Northern analysis of Necl-1. Fig. S4 presents the binding of Necl-Fc fusion proteins to Necl-expressing cells. Fig. S5 shows that the knockdown of Necl-4 inhibits myelination in cocultures. Online supplemental material is available at .

The nervous system is a highly integrated network of neurons and glia that work together to generate, propagate, and modulate action potentials. In most neurons, high densities of voltage- gated Na (Nav) channels initiate action potentials at the axon initial segment (AIS; ; ; ), whereas rapid and efficient action potential conduction along axons depends on high densities of Nav channels located at nodes of Ranvier. In cells that fire repetitively, the integration of synaptic inputs into trains of action potentials is determined by the AIS (). However, despite its importance, the molecular mechanisms underlying AIS formation and maintenance are poorly understood. In contrast, the cellular and molecular mechanisms of nodal Nav channel clustering are better characterized. In the peripheral nervous system (PNS), the available data point to a model where channel clustering at nodes is initiated by interactions between gliomedin, secreted by Schwann cells, and the axonal cell adhesion molecule (CAM) neurofascin-186 (NF-186; , ; ). Subsequently, the cytoskeletal and scaffolding proteins ankyrinG (ankG) and βIV spectrin are recruited to nodes. Finally, Nav channels bind to ankG and mediate the currents necessary for action potential conduction (). Thus, the clustering of ion channels at nodes of Ranvier is thought to depend on binding to cytoskeletal and scaffolding proteins that are positioned along axons by extracellular, heterophilic interactions between axonal and glial CAMs (). Intriguingly, the molecular compositions of the AIS and nodes are nearly identical, consistent with the fact that these axonal domains perform similar functions. These similarities suggest that the developmental mechanisms regulating their formation may be conserved (). By analogy with nodes, the CAMs NF-186 and neuron glia–related CAM (NrCAM) may initiate ion channel clustering at the AIS through as-yet-unknown extrinsic mechanisms. One possibility is that the ECM contributes to formation of the AIS. Consistent with this idea, reported that in vitro a specialized brevican-containing matrix surrounds the AIS. An alternative model places ankG as the central intrinsic scaffold to which all other AIS components become tethered. The latter view has strong experimental support, as AIS localization of Nav and KCNQ2/3 Kv channels, NF-186, and βIV spectrin depend on binding to ankG (; ; ; ). Further, mice lacking ankG in their Purkinje neurons fail to cluster ion channels or any other AIS proteins (; ; ). However, loss of βIV spectrin has also been proposed to disrupt the proper assembly of the AIS (). Finally, ablation of Nav channels by RNAi blocked the accumulation of ankG, NF-186, and NrCAM in cultured motorneurons, indicating that ion channels themselves may play previously unappreciated roles in domain formation or stability (). What are the molecular requirements for AIS formation? Is NF-186 necessary for Na channel clustering, as at nodes of Ranvier, or does it play some other unknown function at the AIS? To answer these questions, we eliminated each AIS protein and determined whether its loss affected the localization of other AIS proteins. Our results show that although nodes and the AIS share a common molecular organization, their mechanisms of assembly are unique. In particular, AIS NF-186 is dispensible for ion channel clustering but is required for the assembly of the specialized brevican-based ECM. Nav channels, NF-186, NrCAM, βIV spectrin, and ankG are all highly enriched at the AIS in vivo (; NrCAM, NF-186, and ankG not depicted). We used cultured hippocampal neurons to determine the role these proteins play in AIS assembly. This well-characterized model () has been used to determine the mechanisms regulating protein sorting and localization in central nervous system (CNS) neurons (; ; ; ; ). In the experiments described here, we define the AIS by the presence of high density, restricted clustering of at least one of the following five protein components: NF-186, NrCAM, βIV spectrin, Nav channels, and ankG. Both in vivo and in cultured hippocampal neurons, NrCAM, NF-186, ankG, βIV spectrin, and Nav channels all colocalize within the first 20–40 μm of the axon (). Further, these proteins are excluded from somatodendritic domains defined by microtubule-associated protein 2 (MAP2; , blue) immunoreactivity. Which protein is the first to define the AIS? To answer this question, we immunostained hippocampal neurons at 2, 4, 6, 8, and 10 d in vitro (DIV) using antibodies against Nav channels, ankG, βIV spectrin, NrCAM, and NF (pan-NF). However, similar to , we were unable to clearly identify a single protein that clustered first at the AIS (Fig. S1, available at ), as we observed only small differences in the time course of protein clustering. These results likely reflect differences in antibody affinity rather than provide an accurate picture of the temporal regulation of AIS protein localization. Thus, other methods are necessary to determine the molecular events leading to AIS formation. Neuronal excitability depends not only on the kinds of ion channels and receptors expressed by a cell but also on their specific localization. In axons, ion channels are clustered in high densities at the nodes of Ranvier and the AIS. Analysis of βIV spectrin mutant mice suggested that βIV spectrin could be responsible for the assembly of the AIS and nodes by directing the recruitment of ankG to these domains (). More recent results indicate that βIV spectrin localization depends on ankG binding. Further, βIV spectrin by itself cannot recruit other interacting proteins to the AIS (). Instead, βIV spectrin contributes to maintenance of these domains and the overall plasma membrane structure of the node and the AIS (). Consistent with these ideas, we observed no impairment in ankG, Nav channel, or CAM clustering in hippocampal neurons lacking βIV spectrin at the AIS. Nav channels themselves have also been proposed to be required for AIS assembly (). This result was based on the observation that knockdown of Nav channels in spinal motor neurons (using the same shRNA expression construct used here) blocked the clustering of ankG at the AIS. This result is different from the one we report for hippocampal neurons. We do not have a clear explanation for these differences, although one notable distinction between the two models is that the AIS assembles much more rapidly in motorneurons than hippocampal neurons. Although nodes and the AIS share a common molecular organization, the data presented here show that the initial events leading to their assembly are unique: NF-186 is dispensible for Nav channel clustering at the AIS but not at nodes (). Despite this clear difference, there are also common elements to the mechanism of Nav channel clustering in these domains. For example, we show that ankG is essential for AIS formation. This conclusion is consistent with analyses of mutant mice lacking ankG in their Purkinje neurons (). AnkG is also required for node of Ranvier formation based on its simultaneous interactions between NF-186, Nav channels, and KCNQ2/3 channels (; ; ). Furthermore, one recent report demonstrated that knockdown of ankG in myelinated DRG Schwann cells results in failure to cluster Na channels (). Thus, ankG functioning as a scaffold to facilitate clustering of Nav channels is common to both nodes and AIS. How does ankG become localized to the AIS? The answer to this question remains unknown and will likely depend on a more complete description of the proteins located at the AIS and the developmental mechanisms that underlie axon specification (). It is interesting to note that other ankyrin repeat–containing proteins are also found at the AIS, including IκBα, a component of the NFκB signaling pathway (). However, as ankyrinB is restricted to distal axons rather than the AIS (), it is not likely that ankyrin repeats themselves determine AIS localization. Indeed, structure function analyses of ankG's targeting mechanism indicated that multiple protein domains in ankG are necessary for its restriction to the AIS (). Animals were housed at the University of Connecticut Health Center, and all experiments involving animals were approved by the institutional animal care and use committee in accordance with all National Institutes of Health guidelines for the humane treatment of animals. Brevican brains were generously provided by R. Faessler (Max Planck Institute of Biochemistry, Martinsried, Germany). The mouse monoclonal pan-NF and pan-Nav channel antibodies and the polyclonal Caspr antibodies were previously described (). The polyclonal βIV spectrin antibody was previously described (). The polyclonal ankG antibody was provided by V. Bennett (Duke University, Durham, NC). A mouse monoclonal ankG antibody was purchased from Zymed Laboratories. The rat monoclonal MBP and the rabbit polyclonal anti-Aggrecan antibodies were purchased from Chemicon International. The chicken MAP2 antibody was purchased from EnCor Biotechnology, Inc. The rabbit polyclonal NrCAM antibody was purchased from Abcam. The rabbit polyclonal anti–NF-186 antibody was provided by A. Gow (Wayne State University, Detroit, MI). The anti-Neurocan antibody 1F6 developed by R.U. Margolis and R.K. Margolis was obtained from the Developmental Studies Hybridoma Bank developed under the auspices of the National Institute of Child Health and Human Development by the University of Iowa (Iowa City, IA). The guinea pig anti-brevican antibodies were previously described (). The rabbit and mouse anti-GFP antibodies were purchased from Invitrogen and CLONTECH Laboratories, Inc., respectively. All fluorescent secondary antibodies were purchased from Invitrogen except for the AMCA-conjugated anti-chicken antibody, which was obtained from Accurate Chemical. The shRNA constructs were made as previously described (). The sense sequences for each oligonucleotide are as follows: Nav1.x, 5′-GTTCGACCCTGACGCCACT-3′; ankG, 5′-GCCGTCAGTACCATCTTCT-3′; βIV spectrin, 5′-CACTGGATAGCCGAGAAGG-3′; NF, 5′-TGCCTTCGTCAGCGTATTA-3′; NrCAM, 5′-CCAATAATCCTCCGAAGTG-3′; control, 5′-CTACTGAGAACTAAGAGAG-3′. The ankG-GFP and NrCAM constructs were gifts from V. Bennett. The HA-tagged NF-186 cDNA was provided by S. Lambert (University of Massachusetts Medical School, Worcester, MA). The myc-tagged βIV spectrin construct was a gift from M. Komada (Tokyo Institute of Technology, Yokohama, Japan). pEGFP-N1 was provided by J. Hewett (University of Connecticut Health Center, Farmington, CT). shRNAs used in electroporation experiments were constructed using pENTR 11. The H1 promoter driving shRNA expression and the shRNA sequence were inserted to pENTR 11 via the EcoRI and HindIII sites. The CAG promoter driving expression of membrane-bound gap-EGFP was inserted using the HindIII and XhoI sites. In the brevican-EGFP fusion construct, EGFP was fused to the C terminus of the full-length cDNA of rat brevican. We introduced an EcoRI and a BamHI restriction site into the brevican cDNA and cloned the entire open reading frame into the pEGFP-N1 vector. Optic nerves and brains were dissected rapidly and immediately fixed with 4% PFA at 4°C for 30 min (optic nerves) or 1 h (brains). Optic nerves and brains were then transferred to ice-cold 20% sucrose (wt/vol) in 0.1 M PB until equilibrated. The tissue was sectioned and immunolabeled as previously described (). For cultured neurons, cells to be immunostained using pan-Nav or ankG antibodies were fixed with 1 or 2% PFA, respectively, whereas cells to be immunostained using βIV spectrin, pan-NF, or NrCAM antibodies were fixed with 4% PFA. Cells were permeabilized using 0.3% Triton X-100 (Sigma-Aldrich) in milk. Fluorescence images were collected on an Axiovert 200M (Carl Zeiss MicroImaging, Inc.) fitted with an apotome for optical sectioning, and a digital camera (AxioCam; Carl Zeiss MicroImaging, Inc.). Images were taken using a 63× Plan-APOCHROMAT (1.4 NA) objective. AxioVision (Carl Zeiss MicroImaging, Inc.) acquisition software was used for collection of images. In some cases, stacks of images were acquired and volume reconstructions were generated using AxioVision software. In some images, contrast and brightness were subsequently adjusted using Photoshop (Adobe). No other processing of the images was performed. All figures were assembled using CorelDraw. Primary hippocampal neurons were prepared from E18 rat embryos as described previously (). Neurons were plated at a density of 200 cells/mm. Primary myelinating co-cultures were prepared as described by with minor alterations. In brief, DRG were dissected from E16 Wistar rats and collected in HBSS without calcium and magnesium (Invitrogen). Cells were mechanically dissociated after 15 min of trypsin (0.25% in HBSS; Invitrogen) digestion at 37°C. Trypsinization was stopped by adding 10% FBS (Invitrogen) in Neurobasal medium (Invitrogen). Cells were centrifuged, washed, and resuspended in growth medium (Neurobasal; 2% B27), penicillin/streptomycin, Glutamax, and 100 ng/ml 7S NGF (all from Invitrogen). Cells were plated at a density of 230 cells/mm on glass (Carolina Biological) or Permanox slides (Nunc) coated with 0.4 mg/ml matrigel (BD Biosciences) and10 μg/ml poly--lysine (Invitrogen). Cells were maintained and myelinated with 50 μg/ml ascorbic acid (wt/vol; Sigma-Aldrich) as described by . Primary hippocampal neurons were transfected 3 h after plating using Lipofectamine 2000 (Invitrogen). Each shRNA construct was cotransfected into hippocampal neurons with pEGFP-N1, permitting GFP to serve as a marker for transfection. Cells were transfected using 1 μg of shRNA vector plus 1 μg pEGFP-N1 combined with 3 μg of Lipofectamine 2000 per ml of OptiMem (Invitrogen). Transfections were allowed to proceed for 4 h, after which cells were returned to growth medium. After 10 or 17 DIV, cells were fixed and immunostained as described. All experiments were performed in triplicate with independent dissections. Cells transfected with each shRNA separately were immunolabeled with ankG, pan-NAV, βIV spectrin, pan-NF, or NrCAM antibodies. 25–30 MAP2- and GFP-positive cells were counted per dissection for each shRNA and AIS antibody combination (∼90 transfected cells for each condition). Cells containing any positive AIS clustering of the immunolabeled protein were counted as having properly formed AIS with respect to that protein, even if the immunofluorescence intensity was clearly reduced compared with untransfected neurons. Only GFP-positive cells without any detectable AIS immunoreactivity were counted as negatives. For validation of shRNAs, COS-7 cells were cotransfected as above with a combination of each shRNA and ankG-GFP, myc-βIV spectrin, HA–NF-186, or NrCAM cDNAs. Cells were lysed, and proteins were size fractionated by SDS-PAGE (). Proteins were then electrophoretically transferred to nitrocellulose membranes and immunoblotted for each protein of interest as described previously (). In utero electroporation was performed as described previously (), except that we introduced shRNA plasmids into embryonic rats at E16 rather than mice. We used rats because the pan-NF antibody only recognizes rat NF-186. COS-7 cells were transfected with 2 μg/ml brevican-EGFP cDNA using 3 μg/ml Lipofectamine 2000 (Invitrogen) in OptiMem. Serum-free media (VP-SFM; Invitrogen), supplemented with Glutamax replaced the transfection solution, and the culture supernatant was collected 3 d after transfection. For NF-186 and NrCAM transfections, COS-7 cells were plated on 25-mm glass coverslips (Fisher Scientific) and transfected with either 1 μg/ml of HA-tagged NF-186 or NrCAM cDNAs. 2 d after transfection, cells were incubated in serum-free media for 2 h before treatment. Brevican-EGFP was pre–cross-linked using anti-GFP antibodies for 1 h. Cells were treated with pre–cross-linked brevican-EGFP for 2 h. Cells were fixed with 4% PFA and immunolabeled with either pan-NF or NrCAM antibodies. Hoechst (Sigma-Aldrich) was used to identify nuclei. Fig. S1 shows immunostaining of cultured hippocampal neurons immunostained for Nav channels, NrCAM, ankG, βIV spectrin, and NF-186 as a function of DIV. Fig. S2 shows that control shRNA does not affect AIS clustering of Nav channels, ankG, βIV spectrin, NrCAM, or NF-186 and that the shRNAs are specific and do not affect the expression of other AIS proteins. Fig. S3 shows a myelinated DRG neuron triple labeled with anti-MBP, anti–pan-NF, and anti–βIV spectrin. Fig. S4 shows the effect of shRNA-mediated protein knockdown on the localization of other AIS proteins at 17 DIV. Fig. S5 shows an untransfected neuron and a transfected neuron expressing low levels of HA–NF-186. In both neurons, brevican is restricted to the axon. Online supplemental material is available at .

The ubiquitous and dynamic remodeling of membranes through fusion and fission defines cellular compartmental organization (; ; ). Controlling membrane dynamics in developmental intercellular fusion, intracellular trafficking, and cell invasion by enveloped viruses and parasites may lead to new strategies for quelling diseases. An understanding of membrane remodeling at the physicochemical level that might guide the development of such strategies requires interdisciplinary investigation of protein–lipid interactions. Only proteins have sufficient complexity and information content to organize and regulate membranes, whereas fusion and fission ultimately unite and separate membrane lipids. In this mini-review, we focus on the hypothesis that all fusion is essentially lipidic at its core (; ). The hemifusion–fusion or stalk–pore pathway of membrane fusion was identified and explored first in theoretical work and experiments on artificial protein-free bilayers (; ; ), and then in viral fusion (for review see ), in intracellular fusion (; ; ; ), and, most recently, in developmental cell fusion (). This pathway starts with hemifusion, a stalklike connection between the contacting membrane leaflets where the distal leaflets and the aqueous inner contents remain distinct (). Hemifusion is followed by the opening of an expanding lipidic fusion pore to complete the fusion reaction (). Each of the essential stages of the pathway will be first described for lipid bilayers, and then for biological membranes with an emphasis on the mechanisms by which proteins may drive each stage. An alternative pathway (; ), featuring a proteinaceous gap junction–like fusion pore, is discussed in . Contact of two lipid bilayers is determined by the thickness of a layer of water separating the polar heads of lipids at equilibrium (Luzzati planes; ). For lipid bilayers without surface electric charge, this equilibrium distance is set by the interplay between intermembrane interactions, such as long-range Van der Waals attraction, short-range repulsive interactions referred to as hydration forces (), and an effective repulsion originating from bilayer undulations (). Based on x-ray measurements, the characteristic values of the interbilayer distances for most biologically ubiquitous lipids, such as phosphatidylcholine (PC), are 2–3 nm (). These distances are only a few times larger than the dimensions of the lipid polar groups. Hence, they are comparable to the scale of membrane surface roughness. Initial contact between fusing biological membranes is fundamentally different from that between two protein-free bilayers. First of all, the distance between bilayers of biological membranes is as wide as 10–20 nm, and the contact is almost always mediated by tethering molecules. The contact zone is crowded with membrane-associated proteins, including those involved in membrane binding and fusion (). For some membranes, such as the envelopes of alphaviruses, protein networks are very tight and coat the membrane surface. The intimate contact of fusion requires an opening of protein-depleted patches in the opposed membranes (). This may additionally crowd the proteins outside of these patches. Interactions between membrane proteins and the cytoskeleton can restrict protein mobility along the membrane surface, hindering the displacement of proteins. The subcortical actin meshwork itself can prevent direct contact of protein-free patches (). These obstacles must be removed or considerably weakened to enable an appropriate bilayer contact. If these obstacles cannot be removed, fusion is inhibited. Clinically, inhibitors of this protein-displacement fusion stage can act as potent and broad-range antiviral agents. For instance, multivalent lectins of the innate immunity system block membrane fusion during enveloped virus entry by cross-linking sugar moieties of membrane-surface proteins (). The resulting network of immobilized glycoproteins decreases the access of membrane bilayers to each other, inhibiting fusion. This strategy may be exploited for the discovery of antifusion drugs. In productive fusion, plasma membrane patches partially depleted of proteins can be generated by a local disruption of the cytoskeleton network adjacent to the membrane, resulting in partial shrinkage of the otherwise stretched network to reposition many integral membrane proteins (). A more general way of facilitating protein displacement is to produce a very limited area of tight contact between bilayers of biological membranes. Fusion proteins may bring two bound membranes into very close contact by acting on only one of them. For instance, in fusion mediated by a homotrimeric influenza virus HA, insertion of its functionally essential amphipathic “fusion peptide” (FP) domains () into the viral membrane and subsequent restructuring of the protein generate deformation of the viral bilayer in the vicinity of HA trimer. To minimize the energy of deformation, trimers assemble into ringlike clusters and dimple the viral envelope with protein-depleted top toward the target membrane (). Fusion proteins may also pull bilayers together by acting on both of the membranes (). Viral fusion proteins are anchored in the viral envelope by their transmembrane domains (TMDs) and, under fusion conditions, insert their FP into the target membrane. As a result, the two membrane-inserted domains of the protein are positioned in different membranes. Further conformational changes “zipper” the protein into a hairpin-like shape with TMD and FP at the same side of a rigid structure, thus, bringing the two membranes into close proximity. The bilayer contact of intracellular fusion involves a formation of hairpin structure composed of membrane proteins anchored in opposing membranes (). For both viral and intracellular fusion, the interbilayer distance reached by bridging membranes with hairpin structure can be close to the hairpin thickness constituting several nanometers, and might be further decreased by insertion of the membrane-proximal regions of the hairpin into membranes (). Establishment of a protein-free contact, although a prerequisite, is insufficient for hemifusion, even when the membranes are separated by only 2–3-nm gaps (). Hemifusion is observed only for specific lipid compositions and specific ions in the aqueous bathing solution or upon dehydration of the intramembrane contact (; ). Special conditions that promote hemifusion are characterized by a common property; in the initial state, the membrane monolayers accumulate energy, which is released upon hemifusion. Hemifusion-driving energy will accumulate if the curvature of the contacting membrane monolayers differs from their spontaneous curvature. The spontaneous curvature of a lipid is determined as the preferred curvature of a monolayer formed by this lipid (). The spontaneous curvature characterizes the effective shape of lipid in a monolayer that fully reflects interactions of lipid molecules between themselves, as well as with the bathing solution. A monolayer that tends to bulge spontaneously toward the layer of polar heads is seen to consist of molecules having an effective shape of inverted cones, and its spontaneous curvature is conventionally defined as positive. A lipid monolayer that bulges spontaneously toward the hydrocarbon tails has a negative spontaneous curvature, and is described as consisting of cone-shaped lipid molecules. Based on experimental studies, lyso PC (LPC) has positive spontaneous curvature, whereas cone-shaped phosphatidylethanolamine, oleic acid (OA), diacylglycerol, and probably cholesterol at a moderate membrane concentration, have negative spontaneous curvature. If the curvature of the monolayer in the bilayer deviates from its spontaneous curvature, the monolayer is under elastic stress and, if allowed, would release this stress by bending toward its spontaneous curvature. Theory indicates that negative spontaneous curvature of monolayer favors hemifusion, and that positive spontaneous curvature hinders hemifusion (). These predicted effects were borne out in experiments on the dependence of fusion on bilayer composition (). Hemifusion might also be boosted by distortion of lipid monolayer packing by inclusions such as amphiphilic peptides (), which generate monolayer deformation of a complex character, including bending and tilting of the lipid hydrocarbon chains in respect to the monolayer plane (). Inclusions would promote hemifusion only if the latter releases the elastic stresses. Finally, bilayers hemifuse when brought to distances much smaller than the equilibrium one by external effects. These effects might be produced by adding polyethylene glycol to draw water from the contact zone (), or by a direct dehydration of the membrane contact in a multilamellar lipid sample (). When bilayers are separated by only 1 nm, the accumulated energy of intermembrane hydration repulsion is expected to drive fusion because formation of a fusion stalk and its expansion into a hemifusion diaphragm (HD) partially relaxes the hydration energy (). Therefore, at these interbilayer distances, hemifusion becomes energetically favorable, as is observed experimentally (). Hemifusion of biological membranes is operationally defined as lipid mixing without aqueous content mixing and/or as lipid mixing between contacting leaflets of the membranes in the absence of lipid mixing between distal membrane leaflets. Formation of a single expanding fusion pore identifies the fusion event as a complete fusion (lipid and content mixing), even if there are hundreds of hemifusion sites present. Thus, to detect hemifusion, complete fusion is inhibited by lowering temperature, modifying fusogenic proteins and decreasing their numbers (; ; ). Hemifusion intermediates can be also stabilized by altering lipid composition. The effects of lipids on viral fusion, intracellular organelle fusion, and exocytosis (; ; ) are similar to those previously discussed for protein-free bilayers. For instance, OA in the contacting and distal membrane leaflets promotes hemifusion and, as discussed in Fusion pores in biological membranes, inhibits breaking of the hemifusion structure into a fusion pore (). Thus, adding OA to the fusing biological membranes is expected to and, indeed, facilitates detection of hemifusion. Identification of hemifusion as lipid mixing without content mixing has several limitations, chiefly (a) the masking of hemifusion by protein–membrane interactions that restrict lipid flux in viral fusion (), and (b) the masking of complete fusion events that yield pores too transient or too small to allow detectable content mixing (). Additional complications arise from the dynamics of hemifusion intermediates (; ). Because detection of hemifusion relies upon the integrated lipid flux over time, membranes may have dissociated by the time of assay. In spite of these methodological difficulties, the hemifusion phenotype has been established in many fusion reactions. Although opening of a fusion pore within a hemifusion connection awaits unambiguous demonstration, diverse lines of indirect evidence, including similar lipid dependences of biological fusion and fusion between artificial lipid bilayers, and the ability of dissimilar fusion proteins to mediate hemifusion, suggest the central place of hemifusion in protein-mediated fusion (). To drive hemifusion by generating the elastic stresses of the mismatch between the actual curvature of membrane leaflets and their spontaneous curvature, proteins might change the lipid composition of contacting leaflets of membranes to that with negative spontaneous curvature (). Hence, phospholipases and acyltransferases that initiate enzymatic cascades leading to increased concentrations of such lipids as diacylglycerol and phosphatidylethanolamine may promote hemifusion. Lipid-modifying enzymes have, indeed, been implicated in some intracellular fusion reactions (). Note, the flux of lipids out of their site of synthesis must be slowed down to accumulate a sufficiently large local concentration of fusogenic lipids. Restriction of lipid flow across the fusion site is, indeed, observed at the early stages of viral fusion (; ; ). For the elastic stresses causing hemifusion to be driven by a distortion of bilayer packing, fusion proteins have to interfere with the bilayer structure. For example, fusion mediated by influenza HA critically depends on a specific boomerang-like conformation of the membrane-inserted FP that is hypothesized to produce a bilayer distortion required for hemifusion (). Note, however, that mechanisms of this kind can only work if fusion allows a relaxation of the stresses induced by the distortions. Current models do not account for this crucial step and, therefore, have to be developed further to offer a plausible scenario for fusion mediated by membrane inclusions. Fusion proteins might drive hemifusion by producing bending stresses in bilayers (). For instance, membrane-bulging deformations that bring bilayers of two biological membranes into a very close contact also generate stresses in the protein-depleted patches of bilayers at the top of the bulges (; ). Hemifusion between these bulges or between bulges and flat bilayer of the target membrane relieves the bending stress of the outer leaflets of the bulged membranes. The energy for stressing the bilayer can come from protein restructuring or protein–membrane or protein–protein interactions. Let us first estimate the minimal energy release required from one fusion protein to enable fusion. The energy of the initial fusion intermediate—the fusion stalk—is a few tens of kilocalories per mole, and its characteristic area is ∼100 nm (). Hence, the stress (energy per unit area) needed to drive this fusion stage should be at least ∼0.1 kcal/mol·nm ∼0.7 mJ/m. Assuming that the fusion proteins form a ring around the fusion site, and that the diameter of one fusion protein in the membrane plane is ∼5 nm, the energy required from one fusion protein is only a few kilocalories per mole. Both fusion protein refolding and FP–membrane interactions might readily provide the required energy (). Sufficient energy for hemifusion can be also released by moderately strong protein–protein interactions based on electrostatic and hydrophobic forces or hydrogen bonds, such as the interactions in actin assembly and antibody–antigen binding (dissociation constant in micromolar range). In contrast, interactions mediated by membrane elasticity, such as the aggregation of membrane proteins based on a mismatch between the length of their TMD and membrane thickness, release <1 kcal/mol () and, thus, are too weak to drive hemifusion. Expansion of a stalk into HD requires much more energy than stalk formation. The fusion proteins have to produce ∼100 pN force acting on the HD rim (). Generation of such force requires the energy released per protein to reach the value of a few tens of kilocalories per mole, which may be provided by fusion protein refolding (). Proteins must be sufficiently rigid to effectively transmit the released energy into bilayer stress, i.e., the effective bending rigidity of fusion protein domains or multiprotein structures has to exceed the bending rigidity of a lipid bilayer. For example, zippering of fusion proteins into hairpin conformations will bulge the bilayers toward each other (; ; ) only if the protein domains that connect the hairpins with the bilayer matrix are more difficult to bend than the lipid bilayer. Whereas the bending rigidity of lipid bilayer is ∼12 kcal/mole, rigidities of relevant protein domains are unknown. Molecular dynamic simulations performed for SNAREs () suggest that the protein domain that links the helical bundles formed by these proteins with the bilayer is rigid enough to transfer required mechanical energy from proteins to membranes. The force that the protein machine can apply to the membrane, and, consequently, the strength of the resulting bilayer stress, is also limited by how tightly this machine is membrane anchored. Although TMDs of integral proteins anchor well, FPs are less reliable. The force needed to detach the FP of HA from a bilayer is estimated to be ∼20 pN (), somewhat exceeding the force needed to bend a lipid bilayer into a fusogenic bulge. In addition, all three FPs of the HA trimer may be engaged in membrane attachment. For other fusion proteins, including those with shorter or less hydrophobic membrane-interacting sequences, the protein-generated force is likely delivered to membranes via the concerted action of multiple anchors (; ). Hemifusion might be promoted by protein assemblies at the fusion site that scaffold lipids onto protein surfaces. Perhaps conformational changes in the proteins that form a hypothetical protein scaffold between the membranes raise hydrophobicity of the surface of the scaffold. Hydrocarbon tails of the lipids of the contacting membrane leaflets would cover this scaffold and, thus, merge the membranes (). However, the subsequent stages of fusion would require a radical transformation of the protein properties to release the lipid–scaffold interactions. The energy of such a hemifusion connection would likely be dominated by the lipid–scaffold interactions and, thus, is expected to be rather insensitive to the spontaneous curvature of the lipid monolayers. Thus, the similarity between the effects of nonbilayer lipids on protein-mediated hemifusion and hemifusion of protein-free bilayers indirectly argues against this mechanism. A more likely role for protein scaffolds is to function after stalk formation to increase the radius of the stalk, using electrostatics, which is a weaker force than the hydrophobic effect. This putative protein scaffold can be located outside rather than inside the hemifusion connection (). For instance, the C2b domain of synaptotagmin, one of the key components of the intracellular fusion machinery, may arrange around the fusion site and present an electrostatic surface sufficiently positive to strongly bind negatively charged lipids (). Indeed, synaptotagmin has a large positive charge when its ligand Ca is bound. Further, the quenching of fluorescence upon membrane binding suggests that membrane curves around the globular C2b domain. Hence, the geometry of poststalk stages may be impacted upon by such protein–lipid interactions to promote expansion of the stalk (). Conversely, proteins may act to specifically prevent the widening of a ring of proteins surrounding the stalk. These would act as brakes, or clamps, to the fusion process. This may be the mode of action of complexin, a molecule that binds to SNAREs and is important for exocytosis (). Indeed, in reconstituted systems of complexin plus SNAREs, only hemifusion results (). Effective transition from hemifusion to complete fusion upon reversing complexin inhibition by synaptotagmin and calcium suggests that complexin prevents the SNARE ring from widening the stalk radius. An intermediate fusion stage that is set for rapid fusion completion upon a final triggering event might be important for the fastest fusion reactions, such as neurotransmitter release (; ). To complete fusion, the hemifusion intermediate must transition to a fusion pore. The pore might open directly from a fusion stalk (; ) or within the HD formed upon expansion of the stalk (; ). The dependence of fusion pore opening on the composition of distal membrane leaflets that form HD is consistent with the latter pathway. Because the curvature of the distal lipid monolayer forming the edge of the pore in the HD is opposite to that in a fusion stalk, lipids that inhibit hemifusion (e.g., LPC) are expected and, indeed, promote pore formation (). The elastic energy of the bent lipid monolayer at the edge of a lipidic pore is rather high (∼12 kcal/mol for a 1-nm radius pore in a PC bilayer; ). Thus, until the fusion pore expands beyond the HD and the area of a tight membrane contact, pore development remains very energy intensive. Further expansion likely proceeds spontaneously (). In many cases both HD expansion and the opening and expansion of a fusion pore in a lipid bilayer are driven by lateral tension generated in a membrane monolayer. The effects of tension on hemifusion and fusion have been observed experimentally (; ; ) and confirmed in numerical simulations of the fusion process (). Theory shows that effective formation of a pore in HD requires tension to reach values of at least a few milliNewtons/meter () that is significantly higher than the estimate of the apparent plasma membrane tension for fibroblasts (0.03 mN/m; ), but within the range of tensions described for biological membranes (). Fusion pores in biological membranes resemble those in protein-free bilayers in their electrophysiological characteristics (; ; ) and in their dependence on lipids in the distal membrane leaflets for pore formation (). The contrasting dependence of hemifusion and fusion pore development on the composition of different leaflets of the fusing membranes may explain the promotion of neurotransmitter release by snake venom phospholipase A2 (; ). Phospholipase A2 hydrolysis produces LPC and OA, and whereas OA quickly partitions into the inner leaflet of plasma membrane and promotes hemifusion between this membrane and synaptic vesicle, LPC stays in the outer leaflet of the plasma membrane and promotes fusion pore opening. In viral and intracellular fusion, pores can close and reopen with the final outcome of the process dependent on both the proteins involved and the membrane lipids (; ). The transition from a small flickering pore to a larger expanding pore likely represents the most energy-demanding fusion stage (; ; ; ). For intracellular fusion, the decision between closing a fusion pore or complete fusion is sometimes referred to as “kiss-and-run versus complete fusion.” Indeed, small fusion pores in mast cell exocytosis are stabilized in hyperosmotic solutions that delay the hydration of exocytotic vesicle contents (unpublished data). This suggests that lateral tension developed by swelling of the vesicle helps to expand the fusion pore until the vesicle contents are fully released and it flattens into the plasma membrane, completing fusion. In contrast to these final fusion stages, opening and moderate widening of a fusion pore proceed in flaccid vesicles (), indicating that tension that drives these stages is generated by fusion proteins rather than by the swelling of the entire vesicle. Stalk–pore transition may involve the aforementioned electrostatic attraction of biological membranes to ringlike scaffolds of protein () that drives stalk expansion toward the point of fusion pore formation. Because such an electrostatic switch can operate extremely quickly, it helps to explain the extremely rapid fusion pore opening that characterizes synaptic release in the nervous system. Another mechanism by which proteins might generate lateral tension that drives opening and expansion of a fusion pore is suggested by the fusion coat hypothesis (). Activated fusion proteins interconnect into a membrane coat that bends the membrane out of its initial shape and expands the fusion site (). A requirement for this mechanism is that the bending rigidity of the protein coat greatly exceeds that of a lipid bilayer. Indeed, estimates show that the coat must be 50–100 times more rigid than the lipid bilayer (). Accordingly, the formation of large fusion pores requires the participation of a considerably larger number of activated fusion proteins than that needed for all the previous fusion stages (; ). o l u t i o n h a s h a d m a n y m i l l i o n s o f y e a r s t o d e s i g n m e m b r a n e f u s i o n r e a c t i o n s , a n d w e a r e j u s t s c r a t c h i n g t h e s u r f a c e i n o u r u n d e r s t a n d i n g o f t h e i r c o m p l e x i t y . T h e c o u p l i n g b e t w e e n p r o t e i n s a n d m e m b r a n e s t h a t i s a t t h e h e a r t o f f u s i o n i s l i k e l y t o b e p a r t i c u l a r t o e a c h s y s t e m . H o w e v e r , s i m i l a r e f f e c t s o f m e m b r a n e l i p i d s o n f u s i o n b e t w e e n p r o t e i n - f r e e b i l a y e r s a n d o n b i o l o g i c a l f u s i o n , a l o n g w i t h r e c e n t f i n d i n g s t h a t d i v e r s e f u s i o n p r o t e i n s f o r m h e m i f u s i o n i n t e r m e d i a t e s , s u b s t a n t i a t e t h e h y p o t h e s i s t h a t p r o t e i n s d r i v e m e m b r a n e r e a r r a n g e m e n t t h r o u g h a c o n s e r v e d p a t h w a y d e f i n e d b y t h e p r o p e r t i e s o f l i p i d b i l a y e r s . A c c e p t a n c e o f t h i s p a r a d i g m w i l l h o p e f u l l y a c c e l e r a t e t h e o n g o i n g e x p l o r a t i o n o f t h e s p e c i f i c m e c h a n i s m s b y w h i c h p r o t e i n s c a t a l y z e a n d d i r e c t d i s t i n c t s t a g e s o f t h i s l i p i d i c p a t h w a y .

TLRs are perhaps the best-studied pathogen-recognition receptors in mammals, and they serve essential functions in mediating innate immunity and establishing adaptive immunity (). TLRs specifically recognize a wide array of microbial components, referred to as pathogen-associated molecular patterns (PAMPs), and upon activation, they engage a signaling pathway leading to proinflammatory responses against pathogenic infection. In addition to a well-described role in immunity (), Toll, which is the orthologue of the TLRs, plays crucial roles in establishing the dorsoventral axis polarity during embryogenesis (), in synaptogenesis, and in axon pathfinding (). Such nonimmune functions of this receptor family remain undiscovered in mammals, despite the fact that TLRs are evolutionarily conserved across species (). In the mammalian central nervous system (CNS), TLRs are expressed in microglia and astrocytes and activate inflammatory pathways in response to pathogenic infection, sterile tissue injury, or in neurodegeneration (; ). The expression of certain TLRs has been recently documented in mammalian neurons (; ), but the functional significance in this cell type has yet to be elucidated. In this study, we define the expression and localization of TLR8 in mouse neurons and reveal the dissociable roles for TLR8 in neurite outgrowth and neuronal apoptosis. Western-blot analysis for TLRs within the developing mouse brain revealed a unique expression profile for TLR8. TLR8 expression in brain () was detected by embryonic day 12 (E12), increased in late embryonic and neonatal stages, and then declined drastically after postnatal day 21 (P21), which is when the basic patterns of neurogenesis and axonogenesis are complete. In adult brain, TLR8 expression is low, but detectable (). The remarkable abundance of TLR8 in embryonic brains, and its developmentally regulated expression, was unexpected because mammalian TLRs are thought to be expressed predominantly in pathogen-sensing tissues and to function in immunity. We further examined the expression pattern of TLR8 in the developing mouse nervous system by immunohistochemistry with an anti-TLR8 polyclonal antibody whose specificity we verified by human embryonic kidney cell transfection and antibody absorption (Fig. S1 A, available at ). In early embryos, TLR8 is highly expressed in peripheral sensory and sympathetic ganglia and in postmitotic migrating CNS cells, but not in the periventricular cell proliferation zones ( and Fig. S1 A, c). Whole-mount in situ hybridization with a -specific probe yielded an mRNA distribution signal at E12 that closely approximates the immunostaining pattern (). TLR8 expression in late embryonic brains was sharply restricted to axonal tracts, including the olfactory nerve fiber layer, cortical intermediate zone, internal capsule, anterior commissure, fimbria of hippocampus, optic chiasm, and other major fiber systems ( and Fig. S1 B). Postnatally, TLR8 is diffusely expressed in most regions of the brain and localizes mainly to neuronal somata (). This dynamically changing spatiotemporal expression pattern implies a role for TLR8 in development of the mammalian nervous system. In neurons isolated from E16 mouse neocortex, TLR8 is expressed at a markedly higher level than in macrophages (). TLR8 in cultured cortical neurons localizes to the perinuclear cytoplasm and neurites, including their growth cones (). The myeloid differentiation factor 88 (MyD88), which is an essential adaptor protein for signaling through all TLRs, except for TLR3 (), was also detected in cortical neurons (). TLR4 is not present in cortical neurons (), as we previously reported (), and indicates the purity of our neuronal cultures, as microglia are known to express high levels of TLR4 (). It is noteworthy that TLR7, a TLR family member phylogenetically and structurally related to TLR8 (), is not expressed by neurons (). Mouse TLR8 was previously suggested to be nonfunctional, based on the observation that ligand stimulation of human, but not mouse, TLR8 induces NF-κB activation (). Because the genomic locus is conserved between human and mouse (), and because the amino acid residues within the TIR domain critical to TLR signaling are identical between human and mouse TLR8 (unpublished data), the mechanism underlying such a species-dependent NF-κB activation by TLR8 remains unclear. However, the inability of mouse TLR8 to activate NF-κB does not necessarily infer a lack of function, as TLR8 may function in biological processes that do not require NF-κB activation, or may alternatively operate in a cell type–specific manner. To investigate the function of TLR8 in neurons, we analyzed the morphological response of freshly plated cortical neurons to a highly permeable synthetic compound, resiquimod (R-848). Although it is a dual TLR7 and TLR8 agonist (), R-848 functions only through TLR8 in neurons, as TLR7 is absent (). Primary neurites of neurons stimulated with R-848 for 24 h were significantly shorter in length () and fewer in number () compared with controls. A slight, but substantial, increase in neuronal death after R-848 exposure was observed by nuclear morphology (). Anticleaved caspase3 immunostaining suggested that the neuronal death was mediated by the classical effector caspase pathway (). A dose–response () and time-course () analysis revealed that the effects of R-848 on neurons were concentration-dependent and relatively slow. Corresponding results were obtained with R-848 treatment on more mature neurons replated from a 5-d-in vitro (DIV5) culture (Fig. S2, available at ). As controls, the exposure of neurons to lipopolysaccharide (LPS), which is a potent inducer of neuronal death in mixed CNS cultures through activation of TLR4 on microglia (), and to loxoribine, which is a TLR7-specific agonist (), produced no detectable effect (), suggesting a lack of contaminating CNS immune cells in the culture and a selective role for TLR8 in the R-848–induced neuronal responses. The effects of R-848 did not appear to be mediated through soluble secreted factors because conditioned medium from R-848–stimulated cultures failed to affect morphology of freshly plated neurons (). Thus, R-848 specifically inhibits neurite outgrowth and triggers apoptosis in cultured neurons. We noted that in R-848–stimulated cultures, ∼13% of the cleaved caspase3-positive neurons exhibited neurite lengths comparable to those of untreated neurons, whereas 12% of the cleaved caspase3-negative neurons lacked processes (). This observation, as well as the similar timing for the onset of R-848–induced neurite outgrowth inhibition and apoptosis (), implies that these events are not necessarily sequential, but may result from two parallel processes. Therefore, we next addressed whether the effects of R-848 on neurons could be dissociated. Addition of a pan-caspase (unpublished data) or a caspase3-specific inhibitor (Z-DEVD-FMK) completely inhibited R-848–induced neuronal apoptosis (). Despite this elimination of the apoptotic response, R-848 stimulation still profoundly inhibited neurite outgrowth (), suggesting that R-848–induced inhibition of neurite outgrowth is not a consequence of apoptosis. Furthermore, removal of R-848 from the culture medium immediately before the appearance of the earliest morphological changes in neurons restored neurite outgrowth, but did not prevent apoptosis (). Collectively, these results suggest that R-848–induced neurite outgrowth inhibition and apoptosis likely occur independently of one another. To determine whether TLR8, which is intracellularly localized in neurons (), specifically mediates the observed effects of R-848, we introduced an anti-TLR8 polyclonal antibody with validated specificity (Fig. S1 A) into freshly cultured neurons by using Chariot, which is a protein-transduction reagent previously demonstrated to deliver antibodies efficiently into postmitotic neurons (). As protein delivery bypasses the transcription–translation process associated with conventional transfection techniques, it provides the opportunity to study early and rapidly proceeding cellular events, such as neurite outgrowth. Anti–TLR8-transduced neurons exhibited substantially longer neurite lengths () and reduced apoptosis () in response to R-848 compared with control IgG-transduced neurons. Similar results were obtained with an anti-TLR8 monoclonal antibody (unpublished data). The attenuating effect of anti-TLR8 antibody was abrogated by co-delivery of an inhibitory peptide specific to the antibody (unpublished data). These results show that R-848 inhibits neurite outgrowth and triggers apoptosis through TLR8. The partial blocking effect of anti-TLR8 antibody may be attributed to incomplete inhibition of TLR8, or alternatively, to the implication of other, yet unknown, mechanisms. R-848 was recently shown to activate caspase 1 in macrophages through cyropyrin/Nalp3 in a MAPK/NF-κB/TLR7-independent manner (), and future work will determine whether this cyropyrin pathway mediates R-848 effects on neurons and if it relates to TLR8 signaling. To dissect the mechanism underlying TLR8-activated neuronal responses, we investigated whether TLR8 in neurons signals through its conventional pathway, the MAPK and NF-κB cascades (). A highly sensitive ELISA-based analysis revealed that R-848 did not induce NF-κB () or AP-1 (unpublished data) transactivation in cultured neurons. Furthermore, IκBα (Ser32) phosphorylation and other characteristic hallmarks of TLR signaling, including the phosphorylation of ERK1/2, SAPK/JNK, Akt, and GSK3β, were not detected in neurons (). In contrast, these signaling molecules were readily activated by R-848 in macrophages (). As additional evidence that neuronal TLR8 signaling occurs independently of the canonical TLR–MyD88–NF-κB pathway, MyD88 deficiency did not confer resistance to R-848 effects on the morphology of cultured neurons (unpublished data). Interestingly, two essential components of the canonical TLR pathway, IκBα and interleukin 1 receptor–associated kinase 4 (IRAK4), were markedly down-regulated in cultured neurons after prolonged (9 h) R-848 stimulation (). A transient reduction of IκBα was also observed 5 min after R-848 administration (). The decrease of IκBα likely occurred through a previously described () phosphorylation-independent degradation process because IκBα (Ser32) phosphorylation was not detected in R-848–stimulated neurons (). Prolonged TLR stimulation was shown to cause IRAK4 degradation in macrophages (). Notably, the timing of the down-regulation of IκBα and IRAK4 coincided with the onset of R-848– induced neurite outgrowth inhibition and apoptosis (), suggesting a possible link between these events. IκBα may be important in neuronal TLR8 signaling, in light of its capacity to regulate gene transcriptional activity independent of NF-κB (). It is also interesting to note that a proapoptotic role has been recently suggested for IRAK4 (). R-848–induced down-regulation of IRAK4 may provide a feedback mechanism to prevent excessive neuronal death. Our results suggest that, fundamentally different from in immune cells, TLR8 in neurons functions in a NF-κB–independent manner. Similarly, our work with neuron cultures from -deficient mice suggests that TLR3 stimulation by polyinosinic-polycytidylic acid inhibits neurite extension, but does not activate NF-κB (unpublished data). As NF-κB activation has been implicated in promoting both neurite outgrowth and neuronal survival (), it is conceivable that TLR-signaling in neurons leading to neurite suppression and apoptosis does not involve the NF-κB pathway. Although the physiological relevance of TLR8 in neurodevelopment is yet to be determined, its role may involve processes negatively regulating axonogenesis and neuron number in the developing nervous system, in light of the developmentally regulated expression of TLR8 in axons and neurons, as well as the capability of TLR8 to inhibit neurite outgrowth and induce neuronal apoptosis. Our findings also add important evidence supporting the emerging concept that traditional “immune molecules” may possess distinct functions in neuronal processes (). Swiss-Webster mice were obtained from Taconic Farms. All animal procedures were conducted in accord with the National Institutes of Health and the Harvard Medical School guidelines. R-848 was purchased from GL Synthesis; LPS was purchased from List Biological Laboratories; Z-DEVD-FMK was purchased from Biovision; loxoribine was obtained from Invivogen; rabbit anti-TLR7 polyclonal antibody, rabbit anti-TLR8 polyclonal antibody, and synthetic inhibitory peptide were obtained from Invitrogen; rabbit anti-TLR4 and rabbit anti–NF-κB p65 (C-20) polyclonal antibodies were purchased from Santa Cruz Biotechnology, Inc.; rabbit anti-MyD88 polyclonal, mouse anti-neurofilament (200 kD) monoclonal, and mouse anti–MAP-2 monoclonal antibodies were obtained from CHEMICON International, Inc.; recombinant mouse TNFα, mouse anti–β-actin monoclonal antibody, cytosine arabinoside (Ara-C), and avertin (2,2,2-Tribromoethanol) were purchased from Sigma-Aldrich; mouse anti-neuronal class III β-tubulin (TUJ1) monoclonal antibody was obtained from Covance; rabbit anti-cleaved caspase3 (Asp175) monoclonal (5A1), rabbit anti-IκBα polyclonal, rabbit anti–phospho-IκBα (Ser32) polyclonal, rabbit anti-p44/42 MAPK polyclonal, rabbit anti–phospho-p44/42 (Thr202/Tyr204) polyclonal, rabbit anti–phospho-SAPK/JNK (Thr183/Tyr185) polyclonal, rabbit anti–phospho-Akt (Ser437) polyclonal, and rabbit anti-GSK3β (Ser9) polyclonal antibodies were obtained from Cell Signaling Technology; rabbit anti-IRAK4 polyclonal antibody was purchased from Millipore; goat anti–rabbit/mouse IgG-HRP was obtained from GE Healthcare; and goat anti–rabbit/mouse–FITC/Cy3 was purchased from Jackson ImmunoResearch Laboratories. The nucleotide sequence CATGGATTCTGACGTGCTTTTGTCTGCTGTCCTCTGGAACCAGTGCCA located within the N terminus of mouse (nt 81–128; GeneBank accession no. ) was selected as the probe, of which specificity was assessed by BLAST. The 48-bp oligonucleotide antisense and sense probes were synthesized and labeled with 10 optimally spaced DIG molecules by GeneDetect.com Ltd. Whole-mount in situ hybridization with E12–12.5 embryos was performed following the manufacturer's protocol. The anti–DIG-AP Fab fragment was purchased from Roche. Neocortical neurons from mouse E16–17 embryos were prepared as previously described (). Typically, >97% cells generated from the procedure were neurons, as estimated by neuron-specific βIII-tubulin (TUJ1) staining. Cells were seeded at a density of 4 × 10 cells/well on poly--lysine–coated 12-mm coverslips or at a density of 2.5 × 10 cells/well on 6-well plates, and cultured in neurobasal medium (Invitrogen) supplemented with B-27 (Invitrogen), 0.5 mM -glutamine, and 1% antibiotic/antimycotic solution. Neurons freshly cultured for 4 h or transduced with antibodies (see the following section) were treated with various stimuli and further incubated for the times indicated. When applicable, 20 μM Z-DEVD-FMK or 1% DMSO (vehicle control) was added into the culture 1 h before the application of R-848. For all assays, including ELISA, morphological, and Western blot analysis, every condition studied was performed in triplicate wells. The antibody delivery procedure was performed with Chariot reagents (Active Motif) following the manufacturer's instructions. In brief, 2 μg anti-TLR8 polyclonal antibody or nonimmune IgG was incubated with 2 μl of the Chariot reagent for 30 min at RT. The formulated antibody–Chariot complex was then applied onto neurons (which had been grown for 4 h after isolation) for 4 h of incubation to allow antibody internalization. When needed, 2 μg inhibitory peptide specific to the anti-TLR8 polyclonal antibody was transduced together with the anti-TLR8 antibody. The antibody- or nonimmune IgG–transduced cells were either cultured under normal conditions or subjected to R-848 stimulation. For tissues, adult spleen or mouse brains at designated developmental stages were dissected out under a stereomicroscope from the timed-pregnant or postnatal mice. 3–20 brains (depending on stages) from the same developmental stage were pooled together to eliminate the discrepancy between individuals. For cell cultures, primary cortical neurons and Raw264.7 macrophages (American Type Culture Collection), which were untreated or treated with various stimuli, were collected at the times indicated. The collected tissues or cells were lysed in RIPA buffer (150 mM NaCl, 50 mM Tris, 1% NP-40, 0.25% sodium deoxycholate, and 1 mM EGTA) supplemented with protease inhibitor cocktail tablet (Roche) and phosphatase inhibitors sodium orthovanadate (1 mM) plus NaF (1 mM). The protein concentration was determined using a Bradford-based assay (Bio-Rad Laboratories). The equally loaded protein samples were separated by SDS-PAGE using 10–20% linear Criterion gels (Bio-Rad Laboratories) and then electro-transferred onto a polyvinylidene difluoride membrane (Bio-Rad Laboratories) at 4°C overnight. The membrane was incubated in blocking solution (5% nonfat dry milk and 0.1% Tween-20 in Tris-buffered saline) at RT for 1 h, and then incubated with primary antibodies diluted (anti-TLR8 polyclonal antibody 1:1,000; other primary antibodies 1:1,000–2,000) in the blocking solution at 4°C overnight, followed by a thorough washing procedure and subsequent incubation with HRP-conjugated goat anti–rabbit or goat anti–mouse IgG (1:4,000 dilution) at RT for 1 h. Finally, ECL Plus reagents (GE Healthcare) were applied onto the membrane to detect the antibody-bound bands according to the manufacturer's instruction, and the resultant chemiluminescent signals were visualized with Kodak X-OMAT film (Kodak). Band densitometry was performed using IPLab3.5 software (Scanalytics) for Western-blots from three independent experiments. Nuclear fractions from the stimulated neurons were prepared using the Nuclear Extract kit (Active Motif), and NF-κB assay was performed using the TransAM ELISA kit (Active Motif) according to the manufacture's protocols. In brief, 5 μg nuclear proteins were incubated in a 96-well plate coated with the oligonucleotide containing the NF-κB–binding sequence (5′-GGGACTTTCC-3′). The activated transcription factor specifically bound to the immobilized oligonucleotide was detected using the antibody against p65 and followed by HRP-conjugated secondary antibody detection. The color-developing solution was applied in the sample wells, and the absorbance was quantified at 450 nm by spectrophotometry using a microplate reader (Spectra MAX250; Molecular Devices). For tissues, whole embryos or embryonic brains were dissected out and fixed by immersion in 4% PFA overnight at 4°C. Postnatal and adult mice were perfused transcardially with 4% PFA after anesthetization with avertin, and tissues were subsequently removed and postfixed overnight at 4°C. The collected tissues were then embedded in paraffin and cut into 5-μm-thick sagittal sections, which were deparaffinized using a standard histology protocol immediately before immunohistochemical staining. For cultures, cells grown on coverslips were fixed with either methanol for 10 min at −20°C or 4% PFA for 10 min at RT for immunocytochemistry. In the staining procedure, tissue sections or cell coverslips were permeabilized with 0.5% Triton X-100 (Sigma-Aldrich) for 10 min, and then blocked with the buffer containing 10% normal goat serum (Sigma-Aldrich), 1% (wt/vol) BSA, and 0.2% (vol/vol) Triton X-100 for 2 h at RT, followed by incubation with primary antibodies that were diluted (anti-TLR8 polyclonal and anti-neurofilament 200 kD monoclonal antibodies 1:50 for immunohistochemistry, 1:100 for immunocytochemistry; anti-MAP2 monoclonal antibody 1:50; and anti-βIII-tubulin and anti-cleaved caspase3 monoclonal antibodies 1:100) in the dilution buffer (2% normal goat serum, 1% BSA, and 0.1% Triton X-100) overnight at 4°C. Samples were subsequently incubated with FITC- and/or Cy3-conjugated species-specific secondary antibody/antibodies in the dilution buffer (1:200 dilution) for 1 h at RT. VECTASHIELD Mounting Medium with DAPI (Vector Laboratories) was used to mount the fluorescently labeled samples and to stain cell nuclei. To evaluate apoptosis, images were captured with a 10× objective lens (Plan Fluor; Nikon) in an unbiased manner. As the majority of cells that display condensed nuclei (visualized by DAPI), which is a characteristic morphology of apoptotic cells, were clearly stained by anti-cleaved caspase3 (Asp175) monoclonal (5A1) antibody, the rate of apoptosis was expressed as the percentage of the cleaved caspase3-positive cells relative to the total cells within a given field (300–400 cells/field). The apoptosis rate presented in the figures was obtained as a mean from 12 fields randomly chosen from the triplicate wells of each condition studied ( = 12 fields). To determine neurite outgrowth, images were acquired with a 20× objective lens (Plan Fluor; Nikon) from two randomly chosen fields in each well, triplicate wells of every condition. Neuron-specific βIII-tubulin (TUJ1) and the cleaved caspase3 were stained to visualize the neuronal somata and processes and to identify the apoptotic cells, respectively. Unless indicated, only the neurites of the cleaved caspase3-negative cells were used for measurement to eliminate the impact of the apoptotic cell morphology on the outcome of statistical analysis for neurite parameters. A primary neurite was defined as a process extending from the cell body by at least one cell diameter (∼10 μm). The primary neurites of nonapoptotic cells were individually traced using Spot software (Version 4.6) with the Curve tool for 20–30 cells within a given field for all fields acquired for every condition ( = 100–180 cells/condition). The length of individual neurites was automatically calculated according to the calibrated scales using the same software. The total number of the measured primary neurites was counted. The measurement data were then exported into Excel 2003 (Microsoft) for statistical analysis. The average neurite length and neurite number were obtained by dividing the total neurite length and total number of neurites, respectively, by the total number of the cleaved caspase3-negative cells (including cells bearing no neurites) measured for each condition. The significance of difference for quantitative analysis was assessed by pair-wise comparisons with -test. Data are presented as the mean ± the SEM. Unless indicated, all cell culture experiments were performed with samples from three independent cell preparations. Fig. S1 shows the specificity of the affinity-purified anti-TLR8 polyclonal antibody and the axon-specific expression of TLR8 in the embryonic brain. Fig. S2 shows that R-848 inhibits neurite outgrowth and triggers apoptosis in neurons developed in vitro. There is also a Supplemental materials and methods. Online supplemental material is available at .

The clinical hallmark of Fanconi anemia (FA) is the development of pancytopenia (loss of blood cells) in childhood (; ; ). There is a progressive loss of hematopoietic stem cells by enhanced apoptosis, and it affects all blood lineages (for review see ). Another consistent feature of FA is a high propensity toward both hematological and nonhematological malignancies, including myelodysplastic syndrome, acute myelogenous leukemia, and squamous cell carcinomas. A wide variety of birth defects, such as short stature, skeletal abnormalities, abnormal skin pigmentation, and developmental abnormalities of other organs, are also observed. Markedly reduced life expectancy has been observed in FA patients with death resulting from hematological complications and cancer. Thus, understanding the basis of the bone marrow failure is of critical importance to improve current treatment approaches for patients with FA. The cellular phenotype of FA is characterized by the occurrence of spontaneous chromosomal aberrations and hypersensitivity to DNA cross-linking agents, such as mitomycin C (MMC) and diepoxybutane. There are 12 complementation groups (A–C, D1, D2, E–G, I, J, L, and M), and 11 of the FA genes have been cloned (; ; ). A key event in the FA pathway is the activation of FA subtype D2 protein (FANCD2) by monoubiquitination, which critically depends on the formation of a core complex of at least eight FA proteins (FANCA–C, E–G, L, and M) in the nucleus, in which FANCL is likely to function as the E3 ubiquitin ligase (, ; ). Localization studies suggest that a high molecular weight FA complex is found in both the nucleus and cytoplasm (; ). For example, a microscopic study revealed that FANCA and FANCG are cytoplasmic in G1 and G2-M phase but are predominantly nuclear during S phase (). In addition, the FANCC protein has been found to interact with several cytoplasmic proteins involved in redox metabolism, including GSTP1 and the molecular chaperones GRP94 and HSP70 (; ; ; ). FANCG interacts with CYP2E1, which is associated with the production of reactive oxygen species (ROS) and the bioactivation of carcinogens, possibly implicating FANCG in protection against oxidative damage (, ). These data suggest that the FA proteins may function in more than one cellular compartment. Oxygen sensitivity of FA cells was first reported by , and further studies demonstrated abnormal oxygen metabolism of FA cells, suggesting a defective antioxidant mechanism (; ). In particular, oxidative damage and a senescent phenotype in response to hypoxia followed by reoxygenation is accentuated in FA cells (). In FA fibroblasts, however, the activity of antioxidative enzymes, including catalase, superoxide dismutase, glutathione reductase, and phospholipid hydroperoxide glutathione peroxidase, are normal (; ). Oxidative stress caused multimerization and an increased interaction of FANC proteins (). Antioxidants have been shown to be beneficial for DNA stability and survival of FA cells (; ). Thioredoxin (Trx) is an intracellular antioxidant and regulator of redox-sensitive gene expression. The overexpression of Trx in FA fibroblasts prevents the cytotoxic and DNA-damaging effect of MMC and diepoxybutane (), suggesting a direct association of oxidative stress with the primary genetic defect in FA (; ; for review see ). This hypothesis is strengthened by the finding that and superoxide dismutase double knockout mice exhibit severe defects in hematopoiesis, including histological evidence of bone marrow hypoplasia (). These biochemical and genetic data strongly suggest that FA cells, including subtypes A, C, and G, experience increased oxidative stress, which can alter multiple cellular processes such as apoptosis, senescence, and DNA damage. Although identified evolutionally conserved tetratricopeptide repeat domains in the FANCG protein, little is known about the biochemical function of the FANCG protein. To determine potential functional roles of FANCG protein, we performed a yeast two-hybrid screen. Peroxiredoxin-3 (PRDX3), the major Trx-dependent mitochondrial peroxidase, was identified as interacting with FANCG. Mitochondrial PRDX3 is an important cellular antioxidant that regulates physiological levels of HO, leading to decreased cell growth while protecting cells from the apoptosis-inducing effects of high levels of HO (). PRDX3 depletion resulted in the acceleration of apoptosis, with increased rates of ΔΨm (mitochondrial membrane potential) collapse, cytochrome release, and caspase activation (). PRDX3 contains a mitochondrial localization sequence, is found exclusively in the mitochondrion, comprises 5% of the total mitochondrial matrix, and uses mitochondrial Trx2 as the electron donor for its peroxidase activity (; ). PRDX3 expression is induced by oxidants in the cardiovascular system to maintain oxidant homeostasis (; ). expression is regulated by myc (), and elevated levels of PRDX3 protein were detected in breast cancers () and hepatocellular carcinoma (). - and -deficient mice have severe hematological disorders and hemolytic anemia. knockout mice also develop malignant cancers (; ). In this study, we demonstrate that the FANCG protein interacts physically with PRDX3 and that the proteins colocalize in the mitochondria. In FA cells from patients with subtypes G, C, and A, PRDX3 undergoes cleavage by a calpainlike cysteine protease, and mitochondrial peroxidase activity is diminished. These defects are associated with distorted mitochondrial structures. The overexpression of PRDX3 can restore the resistance of FA cells to HO exposure, and down-regulation increases MMC sensitivity, suggesting that diminished mitochondrial peroxidase activity underlies the sensitivity of FA cells to oxidative stress. To identify interacting partners of the FANCG protein, we initially used both the N-terminal (aa 1–322) and C-terminal (aa 301–622) regions of the FANCG protein as baits for yeast two-hybrid analysis. A human B lymphocyte cDNA library cloned into the pACT vector was used as prey for the two-hybrid screen. The bait and prey were sequentially transformed into yeast, and positive colonies were selected by growth on the minimal synthetic plates containing 3-aminotriazole lacking adenine, tryptophan, and leucine. These clones were further tested for β-galactosidase activities, and positive plasmids were isolated and sequenced. A cDNA encoding aa 1–256 of the PRDX3 protein was identified in the screen using the C-terminal portion of FANCG as bait (). A bait construct containing a specific missense mutation (G546R) identified in an FA-G patient was used to further probe this interaction (). Quantitative β-galactosidase assays demonstrate that wild-type PRDX3 interacts with the FANCG bait, and this activity is diminished when the mutant FANCG bait is used (). To confirm the interaction of PRDX3 and FANCG in mammalian cells, we used a coimmunoprecipitation assay. The full-length wild-type and G546R mutant cDNAs were cloned into the pCDNA3 mammalian expression vector as a fusion with the HA epitope. An unrelated HA-tagged nuclear factor 1 (NF1) construct, pCH-HA-NF1, was used as a negative control (). These expression constructs were transfected individually into COS-1 cells. 48 h after transfection, total cell lysates were precipitated with rabbit C-terminal PRDX3 antibody. The mutant and wild-type FANCG constructs were expressed equivalently when probed with the HA antibody (, input). Immunoprecipitated proteins were analyzed on SDS gel followed by immunoblotting with HA antibody (, top). The HA-FANCG protein is detected consistently with the endogenous PRDX3 protein interacting with the HA-FANCG protein. Similar to our two-hybrid result, there was little to no interaction between the endogenous PRDX3 protein and the HA-FANCG G546R mutant protein, and there was no evidence for interaction with an unrelated HA-NF1–tagged protein. This further demonstrates the specificity of the interaction between FANCG and PRDX3. A reverse experiment was performed with the same lysates but with coimmunoprecipitation using the anti-HA epitope antibody. The immunoprecipitated proteins were visualized by Western blotting performed with PRDX3 antibody (, bottom). This antibody detects endogenous PRDX3 protein in the immunoprecipitate from cells transfected with wild-type but not G546R mutant HA-FANCG or HA-NF1. Therefore, by two-hybrid analysis and coimmunoprecipitation assays, the FANCG and PRDX3 proteins interact. This interaction is diminished in FANCG protein containing the patient-associated G546R mutation. Previous studies have found that PRDX3 is exclusively a mitochondrial protein (; ). Therefore, we wanted to identify whether FANCG protein colocalized with PRDX3 within the mitochondria. The pDsRed2-Mito plasmid was used to identify mitochondria. We transfected the pCDNA3-HA FANCG and pDsRed2-Mito constructs into HeLa cells and used immunofluorescence to detect the localization of transfected proteins. C-terminal PRDX3 antibody was used to detect the endogenous PRDX3 protein. Inspection of asynchronous HeLa cells transfected with the HA-FANCG construct revealed cells with localization either in the nuclear or cytoplasmic compartments as previously reported (). shows a deconvolution analysis of a representative cell with predominantly cytoplasmic staining of HA-FANCG. There was substantial colocalization of HA-FANCG with the mitochondrial marker (). As expected, PRDX3 colocalized with the mitochondrial marker (not depicted; see ). Staining of both the HA-FANCG and endogenous PRDX3 demonstrates colocalization (). These immunofluorescent experiments suggest that both the FANCG and PRDX3 proteins are found within mitochondria. To confirm the localization of FANCG in the mitochondrial compartment, we performed immunoblotting of lysates from fractionated cells. After transfection with either the HA-FANCG wild-type or G546R mutant construct, HeLa cells were first fractionated into a nuclear and cytoplasmic lysate. The cytoplasmic lysate contains organelles, including mitochondria. Mitochondria were then further isolated from this extract. Immunoblots of the fractionated lysates from cells transfected with the wild-type construct are shown in the left panel in . Immunoblotting with appropriate antibodies demonstrates the expected pattern with cytoplasmic/mitochondrial lysates containing both tubulin and cytochrome and the purified mitochondrial lysate containing only cytochrome . Consistent with our immunofluorescence experiments, antibody to the HA epitope detects a single protein band with the expected size of HA-FANCG in both the mitochondrial and nuclear lysates. The pattern of localization of the G546R mutant protein is similar to the wild-type HA-FANCG (, right). Thus, the lack of coimmunoprecipitation of PRDX3 and mutant FANCG is not related to mislocalization of the mutant protein and is most likely caused by a direct impact of the mutation on FANCG–PRDX3 interaction. Because transiently transfected proteins may be localized differently in comparison with the endogenous protein, we also isolated mitochondrial extracts from the FA-G mutant and stably corrected lymphoblasts and also from HeLa cells transiently transfected with HA-FANCG as a control (). An antibody to a FANCG peptide detects the FANCG band in the mitochondrial lysate of HeLa cells and FA-G–corrected cells but not in the FA-G mutant cells (). Thus, both the immunofluorescent and Western blot assays are consistent with a portion of FANCG protein localizing to the mitochondria. To understand the physiological relevance of the interaction between FANCG and PRDX3 in mitochondria, we determined the expression of PRDX3 in a FA-G mutant lymphoblast cell line compared with FA-G–corrected cells. We isolated total protein from both the mutant and corrected FA-G lymphoblasts, and equal amounts of protein were analyzed by Western blotting with C-terminal PRDX3 antibody (). In addition to the expected 25-kD protein, we found an additional smaller molecular mass band (named PRDX3-S) of ∼16 kD in the lysates from FA-G mutant cells (, lanes 3 and 4). This band was not seen in the FANCG-corrected FA-G cells (, lanes 1 and 2). To determine whether the PRDX3-S species is also a mitochondrial protein, we isolated the mitochondria from the mutant and corrected lymphoblasts. Similar to the whole cell lysates, PRDX3-S is found in the mitochondrial extracts of untreated FA-G mutant cells (, lane 4) but not the corrected cells (, lane 1). Previous investigators have reported that FA cells are sensitive to oxidative stress (for review see ). We treated the mutant and corrected FA-G cells with 100 μM and 1 mM of HO for 90 min and isolated mitochondrial lysates for Western blotting. Interestingly, the FA-G–corrected cells exposed to 1 mM HO demonstrate a band of the same mobility as the PRDX3-S band seen in unexposed FA-G mutant cells (, lanes 3 and 4). These results suggest that the PRDX3-S band may be generated by ongoing oxidative stress in the FA-G mutant lymphoblasts. Several proteases are activated by oxidative stress, including members of the cysteine protease family. To determine whether PRDX3-S is generated by the cleavage of full-length PRDX3 with a cysteine protease, we added a cysteine protease inhibitor, ALLN, to the medium 4–5 h before harvesting mitochondria from the FA-G mutant lymphoblasts. Immunoblotting demonstrates that the ALLN treatment blocks generation of the PRDX3-S band (). This experiment suggests that a cysteine residue in PRDX3 is cleaved by a calpainlike cysteine protease active in the FA-G mutant cells. We tested this hypothesis directly in vitro. We produced S-labeled PRDX3 protein using a wheat germ in vitro transcription/translation assay and incubated the protein with recombinant calpain II enzyme. As demonstrated in , the addition of increasing amounts of calpain II (0.25–1.5 μg; lanes 2–5) to the in vitro–translated protein generates a band of the same molecular weight as PRDX3-S. Calpain II cleavage activity requires high calcium (20 mM), and the addition of 20 mM EGTA blocks the production of PRDX3-S by 1.5 μg calpain II (, lane 6). These data are consistent with the endogenous PRDX3-S band seen in Western blots being a cleavage product of full-length PRDX3. The finding of PRDX3-S in Western blots of lysates from isolated mitochondria () suggests that the protease activity is mitochondrial in origin. To test this directly, we incubated in vitro–translated S-labeled PRDX3 protein with mitochondrial extracts isolated from untreated FA-G mutant and FA-G–corrected cells before and after exposure to 1 mM HO. The cleavage product PRDX3-S was generated by incubation with the untreated FA-G mutant mitochondrial extract, but only from FA-G–corrected cells after exposure to oxidative stress (). The addition of in vitro–translated FANCG protein protects PRDX3 from calpain-mediated cleavage (). Collectively, these results suggest that PRDX3 is cleaved by a mitochondrial calpainlike cysteine protease in vivo to produce PRDX3-S. This protease activity is continuously active in FA-G mutant cells and can be induced in FA-G–corrected cells by acute exposure to HO. PRDX3 has been reported as an exclusively mitochondrial protein that reduces the level of HO continuously generated in mitochondria (; ; ). To determine whether the localization of PRDX3 is altered in cells from FA-G patients, we performed immunofluorescence of PRDX3 in FA-G mutant (PD352-F1°) and corrected (PD352-F2) primary fibroblasts and normal human primary fibroblasts. The endogenous PRDX3 was hybridized with C-terminal PRDX3 antibody and was detected by FITC-labeled anti–rabbit IgG secondary antibody. Localization of the FITC-stained PRDX3 with pDsRed2-Mito was studied by deconvolution microscopy. As expected from prior studies (; ), in normal human fibroblasts, there is nearly 100% colocalization of the mitochondrial and PRDX3 signals (). In contrast, in FA-G mutant cells, most of the PRDX3 (FITC signal) does not colocalize with the pDsRed2-Mito RFP marker (). Interestingly, in the FA- G–corrected cells, there is an intermediate pattern, with the amount of colocalization (yellow signal) intermediate between the FA-G mutant and normal fibroblasts (). These immunolocalization experiments suggested that there might be a distortion of mitochondrial structures in FA-G cells, which is only partially restored when the cDNA is reintroduced (FA-G–corrected cells). To study mitochondrial structure further, we performed EM on the same cell lines used for the immunolocalization studies. As shown in , the mitochondrial structures are quite abnormal in FA-G cells. The FA-G mutant cells possess mitochondria with markedly irregular outlines and sizes. There was noticeable branching, irregular angulation, and elongated shapes, including mitochondria with acute and right angle outlines, as well as some with a z shape (, arrowheads). There were also areas with constriction of the mitochondria, with decreased diameters and elongation of the mitochondria with bleblike extensions. The crista and internal structures of the mitochondria from both FA-G mutant and corrected cells did not show alterations. Mitochondria from the FA-G–corrected cells (, top arrowheads) had occasional mitochondria with irregular outlines and less noticeable branching. The majority of the mitochondria from corrected cells had more typical mitochondrial morphology (, bottom arrowhead). Overall, the total number of abnormally shaped mitochondria in FA-G mutant fibroblast cells () is higher compared with the corrected cells (). These immunofluorescent and EM structural experiments suggest that there are defects in both PRDX3 localization and mitochondrial structure in cells from FA-G patients, implicating defective mitochondrial function in FA patients. The lack of complete rescue in the FA-G cells after correction with the cDNA may result from inefficient turnover of the abnormal mitochondria already formed in FA-G mutant cells. Several lines of evidence argue that the proteins encoded by the different FA genes function coordinately. This includes the similar clinical phenotypes of patients from the different subtypes as well as biochemical experiments demonstrating that many FA proteins form a complex required for FANCD2 monoubiquitination. Therefore, we tested whether the expression of PRDX3 protein is altered in FA-C and -A mutants and in corrected lymphoblasts, as these represent the most common FA subtypes. Isolation of mitochondrial protein and immunoblotting with C-terminal anti-PRDX3 antibody is shown in . PRDX3-S was found in FA-A and -C mutant lymphoblasts (, lanes 2 and 4) but not in the corresponding corrected lymphoblasts (, lanes 1 and 2). This result suggests that the loss of any one of these three FA proteins results in the cleavage of PRDX3 in untreated cells. The aforementioned result could be an indirect effect secondary to diminished FANCG protein in these other subtypes or a direct effect of loss of the FANCA or FANCC function. Therefore, we determined endogenous FANCG expression in mitochondrial extracts of FA-A, -C, and -G mutant cells (). As expected, FANCG protein is absent from FA-G mutant cells. FANCG is present in the FA-C mitochondrial extract and is somewhat diminished in the FA-A extract. Previous studies demonstrated diminished FANCG levels in FA-A cells (; ). The presence of FANCG protein in the mitochondria from FA-C cells suggests that PRDX3 cleavage is not simply caused by the loss of FANCG and that multiple FA proteins are required for the maintenance of PRDX3 integrity. This model is consistent with the fact that FA cells from multiple subtypes demonstrate increased sensitivity to oxidative stress (for review see ). PRDX3 reduces the HO level in mitochondria by catalyzing the conversion of HO into HO using Trx reductase (TR) and Trx as electron donors (). An important question is whether the cleavage and mislocalization of PRDX3 observed in the previous experiments is associated with alteration in mitochondrial peroxidase activity. We performed Trx-dependent peroxidase activity assays in mitochondrial lysates from mutant and corrected FA cells. We used recombinant PRDX1 as a positive control for these assays. In , the addition of 0.6 μg of recombinant PRDX1 shows a ninefold higher peroxidase activity than the negative control, which contains all of the reagents except the enzyme. 0.6 μg mitochondrial lysate from the FA-G–corrected cells supplies a similar level of peroxidase activity. The mitochondrial extract from FA-G mutant cells showed almost a sevenfold lower peroxidase activity than the FA-G–corrected cells. We excluded the possibility of an inhibitor of peroxidase activity in the mutant cells by assaying a 1:1 mixture of lysates from FA-G mutant and corrected cells. This activity was comparable with the FA-G–corrected lysates alone. Similar to the finding of the PRDX3-S cleavage product in FA-A and -C cells, mitochondrial extracts from FA-A and -C mutant (but not corrected) lymphoblasts also had decreased mitochondrial peroxidase activity similar to the FA-G mutant cells (). Given that PRDX3 is the major Trx-dependent mitochondrial peroxidase, these data suggest that the PRDX3-mediated peroxidase activity is diminished in all three FA subtype mutant cells compared with corrected cells. Thus, we see both the cleavage of PRDX3 and diminished mitochondrial peroxidase activity in multiple FA subtypes. Our data are consistent with the decreased mitochondrial peroxidase activity of PRDX3 in cells from FA-G, -A, and -C patients as the basis of increased sensitivity to ROS. To test this hypothesis, we determined whether the transient overexpression of PRDX3 in FA-G mutant cells increases resistance to treatment with HO. FA-G mutant and corrected fibroblasts (PD352-F1° and PD352-F2) were transfected with either the pEF6/His6 empty vector or PRDX3 expression construct pEF6-PRDX3 using a protocol that transfects ∼80% of the cells. 48 h after transfection, cells were exposed to varying concentrations of HO for 90 min, and the number of viable cells was determined using the 3-(4,5-dimethyl-thiazol-2yl)-2,5-diphenyl-tetrazolium bromide (MTT) colorimetric assay. As previously reported (), FA-G mutant cells have increased sensitivity to HO compared with FA-G–corrected cells (). Transient transfection of PRDX3 restores the survival of FA-G mutant cells to similar levels as the FA-G–corrected cells transfected with the empty vector. We next determined whether decreasing PRDX3 expression in cells with a normal FA pathway would alter the response to MMC. HeLa cells were transfected with either a control siRNA or PRDX3 siRNA (as previously developed by ) to reduce PRDX3 expression. Cells with diminished PRDX3 expression showed dose-dependent increases in cell death after 24 h of MMC treatment (), which is consistent with a previous study of the impact of Trx levels on MMC sensitivity (). The increase in MMC sensitivity occurred despite the maintenance of FANCD2 monoubiquitination in the PRDX3 knockdown cells (). FANC proteins, including G, A, and C, have been reported to be present both in the nucleus and cytoplasm. However, the methods used in previous studies did not distinguish between cytoplasmic and mitochondrial proteins (; ; ). In this study, we present data based on two-hybrid interaction, coimmunoprecipitation, and immunolocalization that the FANCG protein binds and colocalizes with PRDX3 in mitochondria. The interaction of these two proteins is lost in the patient-associated G546R-FANCG mutant protein. We specifically demonstrated that a portion of FANCG localizes with mitochondria in HeLa and FA-G–corrected cells. These results suggest that the FANCG protein can translocate into mitochondria and interact with PRDX3. The finding of FANCG interacting with PRDX3 is further strengthened by the characterization of defects in the structure and function of PRDX3 in FA cells from subtypes A, C, and G. PRDX3 is one of the major proteins that controls the level of ROS in the mitochondria (), and the level of ROS impacts proliferation and apoptosis. Therefore, we focused our experiments on determining whether cells from FA patients have intrinsic defects in regulation of the PRDX3 protein and enzymatic activity. We find a proteolytic cleavage product, PRDX3-S, in FA cells that is also seen in corrected cells exposed to high levels of HO. Importantly, the PRDX3-S product is found in three different FA subtypes (G, C, and A), suggesting that multiple FA proteins are required to maintain the integrity of PRDX3 in the mitochondria. We propose that the cleavage of PRDX3 in FA mutant cells or normal cells exposed to oxidative stress is caused by a calpainlike cysteine protease based on inhibition by ALLN in vivo and by in vitro cleavage assays using mitochondrial extracts from FA cells and recombinant calpains. Calpains, a family of Ca-activated neutral cysteine proteases, are involved in cell death in a variety of models. For example, p35 (a neuron-specific activator) is cleaved by calpain into p25, which accumulates in the brains of Alzheimer's patients (). Although there are some reports of mitochondrial calpains, they are not well characterized (; ). Oxidative stress can release Ca from the mitochondrial Ca pool (), which may then activate mitochondrial calpains. Our results suggest that this activation occurs constitutively in FA cells. Further research to characterize mitochondrial calpains and to understand their activation by oxidative stress is needed. PRDX3 mediates the level of ROS through its peroxidase activity. The endogenous level of mitochondrial peroxidase activity is substantially reduced in FA-G, -A, and -C mutant cells, suggesting that a basic function of FA proteins is to maintain the integrity and biochemical activity of PRDX3. Consistent with this model, the transient overexpression of PRDX3 restores the resistance of FA-G cells to acute oxidative stress. The loss of peroxidase activity in mitochondria from FA cells may result in ongoing oxidative stress with an increase in ROS and ROS-mediated DNA damage (). Collectively, our data suggest that FA proteins have important roles outside the nucleus in maintaining the structure and peroxidase activity of the mitochondrial PRDX3 protein. This is similar to the finding that the p53 protein translocates to both the nucleus and mitochondria after DNA damage (; ). Others have also suggested the importance of PRDX3 in modulating levels of ROS and apoptosis in non-FA cells (; ). We propose that the loss of PRDX3 activity results in a negative feedback loop (). With decreased peroxidase activity, ROS can build up within mitochondria, triggering apoptosis as well as further activating calpainlike proteases to further cleave PRDX3, which is consistent with the constitutive PRDX3 cleavage activity demonstrated in mitochondrial extracts of FA-G cells. Apart from cleavage, ROS can also directly interfere with PRDX3 activity by overoxidation of the active site cysteine to cysteine sulfinic acid. This form of inactivation may be particularly important for PRDX3 because the rate of regeneration of active PRDX3 by sulfinylation is slower than for PRDX1 and PRDX2 (). This model suggests that cells from FA patients would accumulate mitochondrial lesions from constitutive oxidative stress and may result in the distorted mitochondrial structures seen in EM experiments. mice (). The reduction in PRDX3 expression in non-FA cells results in increased sensitivity to MMC with an intact FANCD2 monoubiquitination pathway. In FA cells, the increased ROS-mediated DNA damage would be coupled with impaired DNA repair processes and could also increase genomic instability and cancer susceptibility. This model is consistent with the recent finding in that the loss of PRDX function increases the mutator phenotype of strains deficient in DNA repair functions (). Thus, minimizing the production of ROS, inhibiting calpainlike protease activity, or using treatments that activate other ROS scavenger pathways may achieve improvements in the FA phenotype. The N-terminal (aa 135–1,093) and C-terminal (aa 1,044–2,000) regions of cDNA were cloned into pAS2-1 bait vector (CLONTECH Laboratories, Inc.). A human B lymphocyte cDNA library fused with the GAL4 activation domain (gift from S. Elledge, Harvard University, Boston, MA) was screened. A total of 800,000 transformants was screened in the yeast strain PJ-69-4A (). Colonies capable of growth on minimal synthetic plates containing 3-aminotriazole and lacking adenine, tryptophan, and leucine were selected. The liquid β-galactosidase activity assay was performed () to quantify the interaction of specific bait-prey combinations. Full-length cDNA was subcloned into the HindIII and NotI sites of pcDNA3 to encode an N-terminal HA epitope-tagged FANCG protein. cDNA was cloned into the BamH1 and Xba1 sites of the pEF6/His6 mammalian expression vector (Invitrogen). pCH-HA-NF1 has been previously described (). The construct pDsRed2-Mito encoding RFP fused to the mitochondrial targeting sequence from subunit VII of human cytochrome oxidase was purchased from BD Biosciences. HeLa and COS-1 cells were grown in DME and 10% FBS (American Type Culture Collection). The mutant and corrected FA-A (HSC72) and -C (HSC536) lymphoblasts were obtained from G. Bagby (Oregon Health Sciences University, Portland, OR) and H. Joenje (VU University Medical Center, Amsterdam, Netherlands), EUFA316 (FA-G corrected and mutant) cells were obtained from A. D'Andrea (Dana Farber Cancer Institute, Boston, MA), and PD352 cells were obtained from the Fanconi Anemia Research Fund. Normal human fibroblasts were obtained from the American Type Culture Collection (cell line GM08398). All FA cells were grown in RPMI and 15% FBS except that the PD352-F fibroblasts were grown in α-MEM and 15% FBS. All cells were grown in a 37°C incubator with 5% CO. Transfections were performed with Effectene (QIAGEN) according to the manufacturer's protocol. Polyclonal rabbit antibodies to PRDX3 were generated against peptides from N-terminal (GEFKELSLDDFKGKY) and C-terminal (SPTASKEYFEKYHQ) regions of mouse PRDX3 (Zymed Laboratories). Rabbit polyclonal antibody against a FANCG peptide was obtained from S.-H. Lee (Indiana University School of Medicine, Indianapolis, IN; ). Sources of commercial monoclonal antibodies were as follows: anti-HA (HA.11) antibody (Covance), cytochrome antibody (BD Biosciences), α-tubulin (Oncogene Research Products), and FANCD2 (Novus). Cells were lysed with RIPA buffer, and ∼500 μg of lysate was incubated with 4–5 μl of antibody (HA) for 30 min on ice. 20 μl of protein A plus/protein G agarose beads (Oncogene Research Products) was added and incubated overnight at 4°C with constant shaking. Beads were washed with either PBS or RIPA three to four times and boiled with the sample buffer for 5 min followed by 10% SDS-PAGE and immunoblotting as described in the previous section. Cells were plated onto poly--lysine–coated coverslips in 60-mm dishes and transfected the next day with the indicated constructs using Effectene. 48 h later, the cells were fixed for 30 min at RT in a 4% PFA-buffered solution and permeabilized with 0.5% Triton X-100 for 10 min at RT. Staining with either the polyclonal PRDX3 C-terminal antibody and/or monoclonal HA antibody at a 1:300 dilution was performed for 1 h at RT in TBS-T containing 5% nonfat dairy milk. The species-specific fluorescein (FITC)- or Texas red–conjugated secondary antibodies (Abcam) at a 1:400 dilution were applied for 1 h at RT followed by counterstaining with DAPI and mounting together with Vectashield mounting medium (Vector Laboratories) for 10–15 min at RT in the dark. The cells were analyzed on a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) equipped with a digital CCD camera (C4742-95-12ERG; Hamamatsu) and the Openlab 3.1.5 and Volocity 2.1d19 software packages (Improvision) using a plan-Apochromat 63× NA 1.40 oil objective (Carl Zeiss MicroImaging, Inc.). Primary fibroblasts of FA-G mutant (PD352-F1°) and corrected (PD352-F2) cells at approximately passage 8 were harvested and washed three to four times with 1× PBS. The cell pellets were fixed in 3% glutaraldehyde and underwent graded dehydration and infiltration with plastic to create plastic tissue blocks. Semithin sections stained with toluidine blue were created to determine the suitability for ultrastructural examination. Thin sections were prepared, placed on copper grids, and stained with uranyl acetate and lead citrate. The section was then examined with transmission EM to assess the ultrastructure of the mitochondria using an electron microscope (model 1210; JEOL). Cells were harvested, washed once with ice-cold PBS, pH 7.4, resuspended, and incubated in a hypotonic buffer (10 mM Tris, 10 mM NaCl, 3 mM MgCl, 1 mM EDTA, and 1 mM EGTA) for 30 min on ice. Cells were lysed by 15 strokes with a tight-fitting Dounce homogenizer, and nuclei and unbroken cells were pelleted by centrifugation for 15 min at 600 and 4°C. The supernatant was centrifuged for 20 min at 12,000 and 4°C to pellet mitochondria. The pellet was resuspended in lysis buffer (100 mM Tris-HCl, pH 7.4, and 0.1% Triton X-100 supplemented with protease and phosphatase inhibitors as described for RIPA buffer) and broken by sonification for 15 s. Membranes were removed by centrifugation at 12,000 for 15 min at 4°C (). Mitochondrial protein was quantified using protein reagent (Bio-Rad Laboratories). NADPH, recombinant human PRDX1, human Trx, and TR from rat liver were purchased from Sigma-Aldrich. The initial rate of NADPH oxidation was monitored spectrophotometrically (Ultraspec2100 Pro; GE Healthcare) at A. The 150-μl reaction mixture contained 50 mM Hepes-NaOH, pH 7.0, 250 μM NADPH, 46 nM TR, 2.2 μM Trx, and 0.6 μg of recombinant human PRDX1 or mitochondrial extract of the indicated cells. The reaction was initiated by the addition of 0.5 mM HO (). S-labeled PRDX3 protein was produced by in vitro transcription-translation using the TNT-coupled wheat germ extract lysate (Promega). 10 μl of labeled PRDX3 protein was incubated with increasing amounts (0.25–1.5 μg) of rat recombinant calpain II (Calbiochem) or mitochondrial lysates and the activation buffer (20 mM CaCl) for 10–15 min at 37°C. Calpain was diluted with assay buffer (50 mM Tris-HCl, pH 8.0). After digestion, the proteins were separated on a 10% SDS polyacrylamide gel and transferred to polyvinylidene difluoride membrane to decrease background. The membranes were exposed to phosphorimager screen (Molecular Dynamics) for quantitative analysis as well as x-ray film (Kodak). Approximately 10 FA-G primary fibroblasts (mutant, PD352F1°, and corrected PD352-F2) were split into 96-well plates. 24 h later, cells were transfected with the indicated construct using Effectene. 48 h after transfection, cells were treated with H0 for 90 min. Cells were washed several times with PBS, and 15 μl MTT (5 mg/ml in PBS) was added to each well (). Plates were incubated for 4 h, 150 μl DMSO was added to each well, and the optical density at 550 nm was determined with a plate reader. Varying numbers of each cell line were plated, and, 24 h later, the MTT assay was performed. These data were used for log-linear regression of cell number versus OD to determine the number of viable cells remaining after treatment with HO. The PRDX3 and control siRNAs (as described in ) were introduced into HeLa cells by transfection with the use of a Nucleofactor instrument (Amaxa Biosystems). 24 h later, cells were treated with the indicated dose of MMC for 24 h. MTT assay was performed as described in the previous paragraph for cell viability assay. For immunoblotting, cells were lysed with RIPA buffer and separated by 8% SDS-PAGE followed by immunoblotting with FANCD2 antibody (Novus Biologicals). Fig. S1 shows the mislocalization of PRDX3 in FA-G mutant fibroblasts. Online supplemental material is available at .

Mitochondria are made up from two different types of noncontiguous membranes, the outer membrane (OM), and the inner membrane. The OM forms an envelope. It presents a barrier only for macromolecules, as it contains pore-forming proteins that allow the free passage of solutes up to a molecular mass of a few thousand Dalton. The inner membrane encloses the matrix space. It is a membrane in the strictest sense, as even small solutes like ions and metabolic substrates cannot pass through it without the help of carrier proteins. Further, it is one of the most protein-rich lipid bilayers in biological systems, with a protein/lipid mass ratio of ∼75:25 (; ). Various multisubunit protein complexes are located in this membrane, fulfilling several fundamental processes. The most abundant complexes are by far those involved in oxidative phosphorylation (OXPHOS). In addition, several other processes, such as protein translocation, metabolite exchange, protein assembly, iron–sulfur biogenesis, and protein degradation, take place in this membrane as well; proteins involved in inheritance of mitochondrial DNA, fusion and fission of mitochondria, and apoptosis are important for functionality of mitochondria. The inner membrane can be subdivided in two morphologically and presumably functionally distinct subdomains. The first domain is the inner boundary membrane (IBM). It is closely apposed to the OM and can be considered as a second envelope structure. It interacts with the OM in many ways. In particular, it forms contact sites with the OM. These have been termed “morphological contact sites.” They appear to represent tight attachments, as they survive when mitochondria are subjected to procedures to separate outer and inner membranes by mechanical means (). The OM and the IBMs also interact in functional terms. During import of proteins, the TOM and TIM complexes engage in close interaction (). ATP is exported from the matrix by the ADP/ATP carrier in conjunction with porin of the OM (). Outer and inner membrane are fusing and dividing in a coordinated manner (). The second subdomain of the inner membrane is the cristae membrane (CM). In most mitochondria, it makes up the majority of the inner membrane surface, in particular in mitochondria of cells with a high energy demand, such as muscle cells. It forms invaginations of the IBM, in which two leaflets of inner membrane are juxtaposed to each other. In most cases, they form extended sheets. In some organisms or tissues, however, cristae have other shapes, such as tubules or fenestrated sheets (; , ; ; ). Other distinct substructures of the inner membrane are the cristae junctions. They connect the IBM with the cristae. In most types of mitochondria, these are narrow ring- or tubulelike structures, so small that they were proposed to form barriers between the intracristal space and the intermembrane space (, ; ). They were further proposed to undergo remodeling during apoptosis (). Cristae junctions have been studied in mitochondria from various organisms by EM and electron tomography (; ; ; ; ; ; ). Despite the rather detailed insights into the composition, structure, function, and dynamics of mitochondria, the molecular basis of the structural and functional diversity of the inner membrane has remained elusive. Little is known about how proteins are distributed between the various subdomains of the inner membrane or the dynamics of their lateral movements. Even less is known about which proteins are responsible for the architecture of the various inner membrane substructures. Biogenesis and maintenance of subdomains are virtually a blank area on the landscape of mitochondrial biology. Attempts have been made to characterize the different parts of the inner membrane by biochemical means (; ; ; ). Fractionation of submitochondrial membrane vesicles obtained by sonication and enzyme activity determinations were used, yet the information from such studies has remained rather limited, mostly because of the inability to correlate the fragments generated with their origin in the intact mitochondria. In this study, we have used quantitative immuno-EM of intact yeast cells to characterize the various parts of the mitochondrial inner membrane. This appears to be the least distorting method to localize proteins in different subdomains of mitochondria. The high resolution of transmission EM allowed us to analyze wild-type yeast mitochondria that were reported to have a mean diameter of only 350 nm (). We have combined this with the analysis of various processes in mitochondria, such as protein import, protein synthesis, and complex assembly. This was done not only to verify the morphological data but also to obtain insights into the mechanisms that determine mobility of proteins in the inner membrane. We show that each protein has a characteristic distribution between the various subcompartments of the inner membrane. We discuss these results in terms of the dynamic architecture of the subdomains, of mitochondrial functions in metabolism, and of mitochondrial biogenesis. To determine the submitochondrial location of mitochondrial proteins, we used immunogold labeling of chemically fixed, cryosectioned yeast cells grown under respiratory conditions. The various steps of the procedure were optimized for obtaining sensitive low-background readout. The amount of antibodies added was kept very low to minimize unspecific labeling. For quantitative analysis, we plotted mitochondrially located gold particles found in hundreds of sections of mitochondria onto a single model representing part of the OM, the IBM, the CM, and the matrix (). For each protein, on average, ∼300 gold particles (174–1,254) were plotted, yielding a graphical representation of the mean distribution of the protein under investigation. We assigned each gold particle to one of the following three predefined zones: OM/IBM, CM, or background. The relative density was calculated after subtracting the background signal and taking into account that the length of the CM is, on average, 1.5-fold longer than that of the IBM under the growth conditions used in these experiments. Examples of raw data for immunogold labeling are presented for Core1, which is a subunit of complex III in the inner membrane, for the OM protein Tom20, and for matrix-targeted GFP (mtGFP; ). Core1 is located predominantly in the CM (), consistent with its known role in OXPHOS, which is reported to take place predominantly in the CM in bovine heart mitochondria (). Tom20 shows a distribution that is consistent with its known location in the OM protein (). Also, the matrix location of the mtGFP is apparent (). Collectively, this type of representation of gold particles accumulated in silico allows the localization of mitochondrial proteins on a subcompartmental level. For further validation of the method used in this study, we analyzed the distribution of two proteins, Tim23 and Cox2, which were suspected to be enriched in the IBM or the CM, respectively. Tim23 is a component of the TIM23 translocase that functions in cooperation with the TOM complex. Thus, it should be located preferentially in the IBM. In three independent experiments using an antibody raised against the N terminus of Tim23 on a single preparation of fixed yeast cells, on average, 59.5 ± 3.2% ( = 180; = 259; = 294; = 733) of gold particles per unit membrane length were found to be in the OM/IBM zone and, correspondingly, 40.5% were found to be in the CM (; Fig. S1 N, available at ). Therefore, Tim23 is enriched in the IBM, but is not exclusively located to this compartment under steady-state growth conditions during respiration. In addition, these data show that our approach is highly reproducible. Furthermore, a large proportion of the signal for the N terminus of Tim23 appears to be located at the outer face of the OM (; Fig. S1, N and O), supporting the reported two membrane–spanning topology of Tim23 with its N terminus crossing the OM (). As a control, we determined the submitochondrial location of a fusion protein in which GFP was fused to the N terminus of Tim23. With this fusion protein we aimed to artificially anchor Tim23 stably and not dynamically to the OM. The fusion protein can fully replace the essential endogenous Tim23 protein, as indicated by its normal growth (unpublished data). First, we determined the submitochondrial location of GFP-Tim23 by biochemical means. Isolated mitochondria under isotonic conditions from the GFP-Tim23–expressing strain were either incubated with proteinase K or not, and then subjected to SDS-PAGE and immunoblotting. The N-terminal GFP part was degraded almost completely by the added protease, similar to the OM protein Tom40, whereas the inner membrane protein Aac2 was not (). The remaining fragment representing Tim23 lacking GFP was detected using an antibody raised against the Tim23 in proteinase K treated mitochondria (unpublished data). Thus, the N-terminal GFP moiety was largely located at the outer face of the OM and the inner membrane protein Tim23 is spanning the OM. Upon quantitative immuno-EM using an antibody raised against GFP 87.4% ( = 214) of gold particles per unit membrane length were found in the OM/IBM zone (). This is highly consistent with the aforementioned biochemical data and shows that GFP-Tim23 is highly enriched in the IBM. It further suggests that, in wild-type, the N terminus of Tim23 is dynamically, rather than stably, spanning the OM. Cox2 is a subunit of cytochrome oxidase (COX, complex IV) and is predicted to be present at the main site of OXPHOS. In three independent experiments using an antibody raised against the C terminus of Cox2 on a single preparation of fixed yeast cells, on average, 59.4 ± 6.8% ( = 220; = 199; = 162; = 581) of all gold particles per unit membrane length were found to be in the CM zone and 40.6% were found in the OM/IBM zone (; Fig. S1 B). It appears that Cox2 is enriched in the CM, but it is not exclusively located in this subcompartment. Further, we checked whether the SD increased when yeast cells were grown and processed in separate experiments. In the three experiments performed, we observed 63.4 ± 3.9% ( = 733; = 226; = 295; = 1,254) of Tim23 in the OM/IBM (Fig. S1 O; Table S1). In a further step, we determined the distribution of seven subunits from a total of four membrane-spanning protein complexes involved in OXPHOS under respiratory growth conditions. All subunits were found to be enriched in CMs (). The highest enrichment observed for any protein in this study was that for Core1 (complex III), showing 67.1% ( = 357) of gold particles per unit membrane length in the CM zone ( and ). Another subunit of the same complex, Core2, showed a very similar distribution of gold particles in the CM (66.6%; = 266; ; Fig. S1 Q). The subunit Cox2 of the cytochrome oxidase (complex IV), and two subunits (Su and Su ) of the FF-ATP synthase (complex V), the subunit Rip (Fe/S-Rieske protein; complex III), and the ADP/ATP carrier were found enriched in the CM to a similar extent (; Fig. S1). Therefore, the CM is the major site of OXPHOS, albeit not the exclusive one. Complex I is not present in and therefore could not be localized. For the FF-ATP synthase, we used two inner membrane subunits (Su and Su ) that are required for dimerization and oligomerization of monomers of FF-ATP synthase and that are present in the dimeric and oligomeric forms of this complex (). The majority of subunits and were found in the CM (55.4%, = 375 and 60.1%, = 327, respectively; ; Fig. S1, vK and L) suggesting that FF-ATP synthase, the oligomeric forms in particular, are predominantly located in the CM. Again, two subunits of the same oligomeric complex showed a similar distribution. We addressed the question of where nuclear-encoded proteins imported from the cytosol and mitochondrial-encoded proteins from the matrix are inserted into the inner membrane. First, we determined the distribution of proteins involved in mitochondrial biogenesis: Tom20, Tom40, Tim50, Tim14, Tim16, Tim23, Tim17, Mia40, Oxa1, and Mrpl36 (; ; ; and Fig. S1, D, E, G, M–P, and R–T). The aforementioned enrichment of the subunit Tim23 of the TIM23 complex was corroborated by sublocalization of Tim17, Tim50, Tim14, and Tim16 (; Fig. S1, M and R–T). These other essential and tightly associated subunits of the same complex, showed a very similar extent of enrichment in the IBM (59.4%, = 217) of gold particles per unit membrane length in the OM/IBM zone for Tim17, 59.5% ( = 733) for Tim23, 66.1 ± 6.5% ( = 240; = 280; = 333; = 853) for Tim50, 66.9% ( = 875) for Tim14, and 64.2% ( = 847) for Tim16. Not surprisingly, both subunits of the TOM complex analyzed, Tom20 and Tom40, are strongly enriched in the OM, reaching 84.9% ( = 213) and 77.2% ( = 179), respectively, of gold particles per unit membrane length in the OM/IBM zone (; ; and Fig. S1 P). An enrichment in the CM was observed for the mitochondrial large ribosomal subunit protein Mrpl36 and for Oxa1, showing 59.4% ( = 196) and 58.9% ( = 174) of gold particles per unit membrane length in the CM zone, respectively (; Fig. S1, E and G). Ribosomes bind to the C terminus of Oxa1 (; ), explaining their similar distribution. This is also consistent with a much earlier report showing that mitochondrial ribosomes in yeast are present in the matrix, near the CM, and aligned at the IBM (). Collectively, these results support the view that mitochondrially translated proteins are synthesized preferentially close to the cristae and are also inserted at this location in the inner membrane. Mia40 is the initial import receptor for small cysteine-containing proteins in the intermembrane space of mitochondria (; ; ). Mia40 was found to be enriched in the IBM, showing 55.5% ( = 214) of gold particles per unit membrane length in this subcompartment (; Fig. S1 D). This is in line with the proposed role of this protein in early trapping of incoming substrate proteins by disulfide bridge formation. A physiological process that renders mitochondria essential for cell viability is iron–sulfur cluster biogenesis. This process involves several matrix-located proteins (for review see ). We determined whether iron–sulfur biogenesis generally occurs near one of the subcompartments of the inner membrane. Two proteins involved in this process were analyzed Isd11 and Nfs1. These two proteins were reported to fractionate with membrane vesicles after sonication of isolated mitochondria at low salt concentrations (; ). Immuno-EM demonstrated both proteins to be closely attached to the inner membrane and enriched in the CM (; Fig. S1, C and F). For Isd11 57.1% ( = 200) and for Nfs1 66.0% ( = 204) of gold particles per unit membrane length are found in the CM zone. This suggests that at least some steps in iron–sulfur biogenesis occur predominantly at the CM. We investigated the submitochondrial location of components involved in protein degradation. Prohibitin 1 (Phb1) and 2 (Phb2) are not proteases themselves, but they were reported to be in a supercomplex with the m-AAA protease Yta10/Yta12 (). We observed for Phb1 58.0% ( = 210) and for Phb2 52.0% ( = 250) of gold particles per unit membrane length in the CM (; Fig. S1, H and I). Phb1 appears more enriched in the CM than Phb2, which almost seems evenly distributed within the inner membrane. No immunogold signal was observed for Phb1 in a strain and for Phb2 in a strain showing the high specificity of each antibody. Because both subunits are in the same complex with the m-AAA protease Yta10/Yta12, we propose that protein degradation mediated by this complex occurs both in the CM and the IBM. Mgm1 is a dynamin-like protein in mitochondria that is required for mitochondrial fusion (; ). It exists in two isoforms, both of which are required for function (, ), and was found to interact with two OM proteins, Fzo1 and Ugo1, which are both essential for mitochondrial fusion themselves (; ). Using an antibody recognizing both isoforms, we found Mgm1 to be enriched in the IBM, showing 61.2% ( = 317) of gold particles per unit membrane length (). This is in line with the proposed function of Mgm1 in mitochondrial fusion, and furthermore, suggests a role in coordinating the fusion of the outer and the inner membranes. Tim23 in yeast was proposed to span the outer and the inner mitochondrial membranes (). This raises the question of whether this topology is static or dynamic. We asked whether Tim23 can redistribute between the two subcompartments depending on the activity of the protein import machinery. We treated a yeast strain overexpressing mtGFP with the translation inhibitor puromycin and followed the distribution of Tim23 by quantitative immuno-EM. When mtGFP was overexpressed, Tim23 was enriched in the OM/IBM zone showing 67.2% ± 3.1 ( = 197; = 366, = 378; = 941) of gold particles in this zone. This enrichment was slightly higher than under steady-state conditions in wild-type cells, without overexpressing a mitochondrial protein (see above). Upon inhibition of protein translation and, consequently, protein import, Tim23 redistributed from the IBM to the CM (; and Fig. S2, A and B, available at ). After 30 min of puromycin treatment, the density of Tim23 was nearly equal for the IBM (52.4 ± 2.6%; = 180; = 248, = 246; = 674) and the CM (47.6 ± 2.6% ; = 180; = 248, = 246; = 674). The distribution of an inner membrane protein between IBM and CM is apparently dynamic, and changes with the physiological state of mitochondria. Furthermore, it suggests that the CM can act as a reservoir for proteins that otherwise mainly fulfill their function in close apposition to the OM. We asked whether membrane association of mitochondrial ribosomes is also dynamic. In the same experiment, we followed Mrpl36, which is a ribosomal protein of the large subunit of mitochondrial ribosomes. After puromycin treatment, a partial detachment of this protein from the inner mitochondrial membrane was observed, accompanied by an increase in the matrix (). This is consistent with a recent study showing that upon addition of puromycin only a fraction of Mrpl36 remained associated with the inner membrane (). Our observation suggests that the entire mitochondrial ribosome, the large ribosomal subunit, or Mrpl36 alone partially detaches from the inner membrane upon block of translation and release of nascent chains. Where in the inner membrane do newly inserted subunits become associated to form assembly intermediates and, eventually, complexes and supracomplexes? To address these questions, we analyzed the submitochondrial distribution of partially assembled complexes in a mutant with a deficiency in the formation of complex III. That mutant lacked the factor Bcs1, which is required for a late assembly step of complex III and for formation of the respiratory chain supracomplex IIIIV (, ). Mutant strains and wild-type control were grown on fermentable carbon source, as deficiency of fully assembled complex III leads to the inability to grow on nonfermentable carbon sources. Quantitative immuno-EM was performed using antibodies against Rip and Core1 (both complex III), and against Cox2 (complex IV). In wild-type cells, the Core1 and Rip proteins were found strongly enriched in the CM, with only 29.9% ( = 292) and 38.3% ( = 214) of gold particles per unit membrane length in the IBM (; Fig. S2, E and G). In the deletion mutant, both the Core1 protein, which is present in the partially assembled complex III, and the nonassembled Rip protein were found less enriched in the CM. 40.3% ( = 237) and 47.2% ( = 202) of Core1 and Rip protein, respectively, per unit membrane length were present in the IBM (; Fig. S2, F and H). Apparently, both subunits accumulated more in the IBM when assembly of complex III was impaired. On the other hand, Cox2 was more enriched in the CM in the deletion strain (65.3%; = 153), as compared with the wild-type strain (54.7%; = 205; and Fig. S2, C and D). In the case of complex III, unassembled or partially assembled subunits accumulate in the IBM, but after full assembly, of complex III they end up mostly in the CM. When the supracomplex of III and IV was not formed, the assembled complex IV accumulated in the CM. This may be explained by the fact that less complex III is present in the CM, and therefore more space is available for complex IV. In summary, we show that assembly intermediates and fully assembled proteins of a respiratory complex distribute differently between the subcompartments. The specific location of single subunits may influence or regulate the assembly of such complexes. Further, these results suggest that assembly of respiratory complexes involves migration in the inner membrane. The inner membrane of mitochondria consists of two morphologically distinct subdomains. One subdomain, the IBM, is tightly apposed to the OM, and the other one forms the CM. These two domains are connected by narrow structures called the cristae junctions. We have addressed the question of whether the morphological differences are reflected in the protein composition of the two domains. For this we have established a procedure for quantitative immuno-EM using yeast cells, an approach that is the least disturbing to the native organization of the membranes. We report on the localization of 20 different proteins in the two subdomains of the inner membrane. We show that the proteins indeed have different characteristic distributions between IBM and CM. For instance, the five subunits of the TIM23 complex investigated were more abundant, relative to unit membrane length, in the IBM than in the CM. This is consistent with the reported two membrane–spanning topology of Tim23 () and earlier reports on the presence of translocation contact sites (; ; ; ). Our data further corroborate earlier studies, using both similar and different approaches, such as biochemical fractionation or immunocytochemistry, regarding the preferential location of complexes involved in OXPHOS within the CM (; ; ). The level of respiratory complex III in our study was roughly twofold higher in the CM than in the IBM. This is similar to the reported 2.2–2.6-fold higher density in CM, as compared with IBM of bovine heart mitochondria (). At least two major possibilities can be envisaged to explain how such diversity is generated and maintained at a molecular level. One is that IBM and CM are two distinct membrane compartments that are separated by the rather narrow cristae junctions, which might serve as strict barriers for lateral diffusion of membrane components, as suggested for ions, metabolites, and small proteins in the intermembrane space (, ; ). A second explanation is that there are no such barriers, or only selective barriers or filters that allow more or less free lateral diffusion, and that diversity is generated in a different way. Possible mechanisms might involve differential topological effects, as the IBM is only facing the OM, whereas the CMs are facing each other. Our results would favor, perhaps surprisingly, the second explanation. First, the distributions observed in our study are, in quantitative terms, not all or none. Rather, all proteins investigated are present in both domains of the inner membrane at different levels. Second, components of the mitochondrial protein import machinery are preferentially found in the IBM and are known to interact with components involved in import in the OM, specifically the TOM complex. Components of the OXPHOS are, not surprisingly, present at higher levels in the CM, which is apparently generated for accommodating these abundant complexes and supracomplexes. Third, our experiments demonstrate a dynamic behavior of certain components in the context of their function. Tim23, which is preferentially present in the IBM under normal conditions, redistributes into the CM when it is no longer engaged in the translocation of polypeptide chains. What could be the mechanisms that are responsible for establishing an uneven but dynamic distribution between IBM and CM? In cases of proteins localized preferentially in the IBM, contacts of a direct or indirect nature would be a way to generate sequestration of proteins. Indeed, Tim23 in yeast has been shown to interact with the OM with a peripheral segment (), thereby generating a bias toward a preferential location in the IBM. As Tim23 interacts tightly with Tim17, the latter component is also maintained in the IBM. By a hierarchy of interactions, whole pathways could be recruited to the IBM. Such interactions have also been established for components of the mitochondrial fusion and fission machinery. For instance, Mgm1 was reported to contact Fzo1 and Ugo1, which are OM proteins involved in the fusion of mitochondria (; ). Further organizing determinants of the IBM are the morphological contact sites that link it tightly to the OM. These types of contacts are observed in electron micrographs of virtually all mitochondria investigated (; ). They appear to play a major role, as they guarantee the presence of permanent linkages and may help to bring together components of both membranes that have to find each other. Unfortunately, the protein components involved have not been identified thus far. For the cristae, a different principle of stabilizing organization can be proposed. The main components of the CM are the complexes of OXPHOS that associate with each other to form supracomplexes of megadalton molecular mass. A specific role has been assigned in various studies to the FF–ATP synthase. This large complex was suggested to form oligomeric supracomplexes that are required for cristae formation (; ; ). In this way, a scaffold would be created to keep not only the FF-ATP synthase but also respiratory complexes in the cristae. Other proteins may join this macromolecular network, retaining themselves in the CM. Such an organization could trap supracomplexes within the CM because of their high molecular weight and prevents them from diffusion to the IBM through narrow cristae junctions. A major conclusion from our study is that both subcompartments are dynamic in the sense that distribution of proteins can change upon changes of the physiological state of the cell. The observed dynamic changes are correlated with the function of the proteins investigated. For instance, Tim23 redistributed from its primary site of action during protein translocation, the IBM, to the CM when protein synthesis and consequently protein import were stopped. In such a situation, the cristae could act as a reservoir for proteins not needed for activity in the IBM. Also the mitochondrial ribosomal protein Mrpl36 moved from its site of action, in this case, from the CM to the matrix, when nascent chains were released from ribosomes. The observed dynamic nature of the subcompartmentalization has particular relevance in the context of the biogenesis of mitochondria. Newly synthesized proteins are imported into the mitochondrial inner membrane from the cytosol and from the matrix. The complexes involved in OXPHOS, which make up the larger part of the inner membrane proteins, are made from ribosomes present both outside and inside mitochondria. Thus, questions of where they are inserted into the membranes and where they meet to assemble to functional complexes arise. Our data show that the insertion of imported proteins does occur preferentially into the IBM, and that the insertion of mitochondrial translation products occurs mostly into the CM. On the other hand, movement of assembly intermediates between the subcompartments was suggested in our experiments. This again suggests that there are not exclusive sites of complex formation. The lateral movement of assembly intermediates may be facilitated by assembly factors described in mitochondria that shield and protect aggregation-prone surfaces. Furthermore, the preferential accumulation of certain subunits of complex III and IV assembly intermediates correlates well with the location of the components responsible for their import. This suggests that the site of insertion of individual subunits also plays a role, possibly a kinetic one, in the pathway of multisubunit complex formation. The structure and function of cristae junctions deserves particular attention in further attempts to understand the organization of the mitochondrial inner membrane. They apparently are not absolute barriers, but they could well represent kinetic or diffusion barriers for individual proteins and mark as specific gates. In the future, it is of the utmost importance to learn whether they influence the traffic of proteins between the domains of the inner membrane. strains used are wild-type D273-10B; wild-type W303 (Mat α, , , , , and ) strain; Tim23/Δtim23 (Mat a/α, his3Δ1/his3Δ1, , , LYS2/lys2Δ0, MET15/, and TIM23/∷Kan; Open Biosystems); Δbcs1 (W303, Mat α, , , , , and ; ); and Oxa (YPH499, Mat α , , , , and ; ). Δtim23 + pRS315-GFP-Tim23 (Mat α, his3Δ1 , LYS2, , ∷Kan). Culturing of yeast was performed according to standard protocols at 30°C on complete liquid media containing 2% lactate or 2% galactose. Mitochondria were prepared according to . The GFP-Tim23 construct was obtained by PCR amplification of the complete open reading frame from genomic yeast DNA using the primers Tim23up (5′-AAAGGATCCATGTCGTGGCTTTTTGGAGA-3′) and Tim23down (5′-CCCAAGCTTTCATTTTTCAAGTAGTCTTTTCTTGACAC-3′). The Tim23-promoter region was amplified by PCR using the primers Prom23up (5′-CCTGAGCTCACTGTGACGTCG-3′) and Prom23down (5′-CCCGGTACCGATTGTGTGTGATCTGTTAAAC-3′) on genomic yeast DNA. To amplify the GFP pYX242 mtGFP () was used as a template. The primers GFPup (5′-CCCCGGTACCATGAGTAAAGGAGAAGAAC-3′) and GFPdown (5′-CCCCGGATCCTGCTCCTGCTGATCCTCCTTTGTATAGTTCATCCATGC-3′). The resulting PCR-fragments were cloned into pRS315 yeast expression vector in three single steps. The correctness was confirmed by DNA sequencing, and the plasmid was transformed into the heterozygous diploid Tim23/Δtim23 strain. The transformed strain was sporulated, and haploid spores were tested for expression of GFP-Tim23 and used for further analysis. Mitochondrial membrane proteins were localized by postembedding immunogold labeling of chemically fixed cryosectioned whole yeast cells. Wild-type (D273-10B) cells were grown on complete liquid media containing 2% lactate at 30°C (if not stated differently) and fixed during early log phase in a mixture of freshly prepared 4% formaldehyde and 0.5% glutaraldehyde in 0.1 M sodium citrate buffer adjusted to growth conditions for temperature and pH. Cells were washed with phosphate buffered saline, incubated with 1% sodium metaperiodate for 1 h, immersed in 25% polyvinyl-pyrrolidone (PVP, K15/MW 10000; Fluka) and 1.6 M sucrose () at 30°C for 2–3 h, mounted on specimen holders, frozen in liquid nitrogen, and sectioned at −115°C with an ultracryotome Ultracut S attached with a FCS unit (Leica). Ultrathin, thawed cryosections were prepared with glass/diamond knives and placed on formvar/carbon-coated copper grids (200 mesh, hexagonal). Labeling with IgG primary antibodies and secondary antibody–gold complexes (10 nm; Dianova) was performed as previously described (). Finally, the sections were stained and stabilized by a 1:1 mixture of 3% tungstosilicic acid hydrate (Fluka) and 2.5% polyvinyl alcohol (MW 10000; Sigma-Aldrich), and analyzed by standard transmission EM. Specific epitope recognition was ensured by initially optimizing the dilution of each antibody used in such a way that mitochondria were still labeled but giving no or negligible cytosolic background (see Table S2, available at , for details). Because labeling occurs only at the surface of the 50–70-nm-thick sections, and because of the low antibody concentration, only a weak signal for endogenously expressed proteins is found. To overcome this disadvantage, we developed a novel tool in which an in silico accumulation of signals onto a single draw-to-scale model representing part of OM, IBM, and CM () was performed. The location of all gold-particles found in mitochondria showing clearly resolvable CMs connected by cristae junctions to the IBM on electron micrographs (22,000× enlargement) were transferred to this model using a microfiche device with 24× enlargement. For each antigen, on average, 250 gold particles were plotted, producing a graphical representation of the average distribution of the protein under investigation. To calibrate the positions of the gold particles, a point of origin and an internal size standard was used. To digitalize the distribution of the gold-particles, Photoshop CS2 (Adobe) and Image J (National Institutes of Health) programs were used. The calibrated x/y-values for each gold particle were used to reconstruct the particle distribution with the program Delta Graph 4.0.1. The result of this procedure is a statistically significant distribution-pattern of the mitochondrial protein investigated. The accuracy of the method is in the range of 10 nm. The quantification and assignment to one of the three predefined zones was done with Excel 2002 (Microsoft). Assignment to the matrix was done for gold particles that are in the background zone, but only those located on the matrix side of the inner membrane. Because mostly integral membrane proteins were analyzed, the background signal for each antibody in the OM/IBM and the CM zone was assumed to be represented by those gold particles found experimentally in the background zone. Therefore, we subtracted the number of gold particles expected to be background of a particular zone from the number of gold particles found in this zone. The relative areas of the OM/IBM, CM, and background zones in our model are 1.23 to 1.0 to 3.48 and were taken into account in this background subtraction. From this the number of gold particles per unit, membrane length of CM or OM/IBM was calculated, taking into account that the length of the CM in these sections was, on average, 1.5-fold longer than that of the IBM under the growth conditions used in these experiments. For galactose-grown cells, the length of the CM was identical to the length of the IBM. Finally, the relative density of gold particles in the CM or the OM/IBM (percentage of gold particles per unit membrane length) can be obtained by dividing the number of gold particles in the CM or the OM/IBM, respectively, by the sum of the densities of gold particles in the CM and the OM/IBM. Therefore, by first subtracting the normalized background determined for each experiment and normalizing for membrane length, we obtain a normalized percentage of gold particles per OM/IBM or CM membrane length. This allows us to compare numerous different proteins that vary markedly in their expression level and/or their reactivity with their corresponding antibody, in respect to a possible enrichment in one or the other subcompartment. Wild-type cells (w303α) transformed with the mtGFP expression plasmid pVT100U-mtGFP () were grown on selective liquid media containing 2% lactate. In the exponential growing phase, 100 μg/ml puromycin was added for various time periods before fixation of the cells. Fig. S1 shows the distribution of gold particles after immuno-EM of mitochondrial proteins. Fig. S2 shows the dynamics of the distribution of mitochondrial proteins. Table S1 shows the quantitative immuno-EM of mitochondrial proteins. Table S2 shows the antibodies and dilutions used.

Chloroplasts are subdivided by three noncontiguous membrane systems into at least six suborganellar compartments that serve to segregate and organize several essential metabolic functions. The chloroplast outer and inner envelope membranes form the organelle boundary and effectively segregate chloroplast and cytoplasmic metabolism by controlling metabolite and ion transport (), whereas the internal thylakoid membrane performs the light harvesting and photophosphorylation reactions of photosynthesis. The vast majority of chloroplast membrane proteins are encoded in the nucleus (; ). As a result, the biogenesis of these membranes relies on the selective targeting and insertion of hundreds of proteins from their site of synthesis in the cytoplasm. Except for the translocon component, Toc75, outer envelope membrane proteins are targeted directly from the cytoplasm to the membrane via targeting signals contained within and adjacent to their transmembrane helices (). The majority of thylakoid membrane proteins contain cleavable N-terminal transit peptides that target them across the envelope. Import is mediated by the same translocon complexes within the outer (Toc) and inner (Tic) envelope membranes that mediate the import of soluble proteins (; ). Upon import, thylakoid membrane proteins are released into the soluble stroma and processed to remove their transit peptides, and intrinsic secondary targeting signals direct them to the thylakoid (). The protein translocons at the thylakoid are homologous to protein export translocons found in prokaryotic cytoplasmic membranes (e.g., Sec, SRP, and TAT pathways), indicating conservation in these targeting systems from the original bacterial endosymbiont (; ). The limited studies on targeting to the chloroplast inner envelope membrane (IM) leave open the question of whether nucleus-encoded proteins insert into the membrane during the import process via a stop-transfer mechanism or target to the membrane after the completion of import by inserting from the stromal side of the membrane. Two classes of nucleus-encoded integral IM proteins are known to exist. The first class contains proteins that lack cleavable transit peptides (; ). Although the targeting determinants for these proteins have not been completely defined, they appear not to use the Toc–Tic translocons for import. The second, larger class consists of IM proteins, is synthesized with cleavable transit peptides, and initially engages the Toc–Tic import machinery. The abundant inner membrane protein, Tic110, is a member of this class (, ). The analysis of deletion mutants and fusion proteins indicates that the N-terminal region of Tic110, including its two transmembrane segments, is required for targeting to the IM (). Interestingly, a fusion protein containing the N-terminal targeting determinants of Tic110 transiently accumulated as a soluble intermediate in the stroma before inserting into the IM (). In a subsequent study, dominant-negative mutants of Tic110 that disrupt Tic complex formation resulted in the accumulation of normal Tic110 in the stroma in vivo (). Consistent with the fusion protein studies, chloroplasts isolated from these mutants transiently accumulated a soluble, mature form of Tic110 in in vitro import assays (). Collectively, the two studies suggested that Tic110 inserts into the membrane from the stroma after import. In contrast to the studies with Tic110, in vitro import studies with several chloroplast polytopic membrane transporters led to the conclusion that IM proteins do not use stromal intermediates en route to the membranes (; ; ). In one case, soluble forms of fusion proteins to the envelope phosphate translocator were observed but were not shown to represent targeting intermediates (). These results suggested that chloroplast IM proteins use a stop-transfer mechanism of targeting that results in direct insertion of the proteins into the inner membrane during import through the Toc–Tic system. In this report, we wished to examine the process of targeting to the chloroplast IM by studying the import and insertion of a simple IM protein. To this end, we investigated the targeting of pre-atTic40, a nucleus-encoded chloroplast inner membrane protein with a single transmembrane helix (; ). We show that the import and membrane insertion of native pre-atTic40 involves a size intermediate that inserts into the inner membrane after import from the cytoplasm through the Toc–Tic machinery. Furthermore, we demonstrate that atTic40 and atTic110 can insert directly and selectively into isolated IM vesicles. These data are consistent with a pathway for the targeting of IM proteins that is independent of their import from the cytoplasm. A schematic diagram of the structure of pre-atTic40 is shown in . The protein contains a 76-amino-acid transit peptide that is removed upon import into chloroplasts. The single transmembrane helix is located within the N-terminal region between amino acids 106 and 127 of the preprotein with a C-terminal ∼35-kD soluble region extending into the stroma (; ). As a first step in examining the targeting of pre-atTic40, we performed a time course of import of in vitro–translated [S]-labeled pre-atTic40 into isolated pea chloroplasts (, lanes 2–4). In addition to pre-atTic40, two other major forms of atTic40 are apparent in the import assays. The polypeptide with the highest mobility was confirmed to be mature atTic40 by comparing its mobility with that of endogenous atTic40 as detected by immunoblotting (unpublished data). In addition to mature atTic40, a major intermediate-sized form of the protein (int-atTic40) with a mobility between pre-atTic40 and atTic40 is observed in the import assays. As a first step in defining the relationship between int-atTic40 and mature atTic40, we examined their suborganellar location. To this end, chloroplasts from each time point in import were treated with thermolysin or trypsin (). Thermolysin has been shown to digest proteins that are exposed at the chloroplast surface, but it does not penetrate the outer membrane (). In contrast, trypsin can penetrate the outer membrane and access the intermembrane space, digesting envelope proteins that are not protected by the inner membrane (, ; ). demonstrates that mature atTic40 accumulates with time and becomes increasingly insensitive to either protease, consistent with targeting and insertion in the IM (). Similarly, the portion of int-atTic40 that is insensitive to protease digestion increases with time, rising from 8–10% at 2 min to 40–45% of the total int-atTic40 at 30 min (). Toc75, a protein deeply imbedded in the outer membrane, is not digested with thermolysin but is nearly completely degraded with trypsin (), confirming the differential accessibility of the two proteases to the envelope compartments. These data suggest that int-atTic40, like atTic40, accumulates inside the inner membrane, where neither thermolysin nor trypsin can access the polypeptide. Both pea and chloroplasts gave similar patterns of import, confirming that the intermediate-sized product was a true import product and not an artifact of the heterologous import system (Fig. S1 A, available at ). Pea chloroplasts were used for all subsequent experiments. To ensure that the lack of atTic40 and int-atTic40 proteolysis was not a result of incomplete digestion, samples from the 5-min time point in import were treated with a range of thermolysin and trypsin concentrations (). Increasing the protease concentrations had no effect on the proportions of protease-insensitive atTic40 or int-atTic40. Furthermore, the addition of Triton X-100 to the samples before protease treatment resulted in complete degradation of int-atTic40 and atTic40 (), indicating that the proteins are not intrinsically protease resistant. These data confirm that both forms are localized inside the inner membrane. Although atTic40 and int-atTic40 accumulate inside the chloroplast with time, a portion of each population is sensitive to both proteases. The highest levels of sensitive forms are observed at early time points in import (). Their sensitivity to both proteases indicates that they are exposed at the surface of the chloroplast and therefore appear to be in the process of import into the organelle. These forms are similar to import intermediates observed with stromal preproteins that are captured within the Toc–Tic machinery (; ; ). To obtain additional evidence that int-atTic40 accessed the stroma, we tested whether the stromal processing peptidase (SPP) that cleaves the transit peptides of soluble proteins (; ) was involved in converting pre-atTic40 to the intermediate form. Analysis of the pre-atTic40 sequence identifies a potential SPP processing site after amino acid 42 of the transit peptide (). Incubation of in vitro–translated pre-atTic40 with a stromal extract results in the conversion of 35% of the precursor to an intermediate-size form coincident in size with int-atTic40 (). The processing by SPP is likely to occur at the predicted cleavage site (residue 42) because the mobility of a truncated form of pre-atTic40 lacking the first 42 residues is identical to int-atTic40 (Fig. S1 B). Furthermore, the truncated form is not cleaved by the stromal extract (), eliminating the possibility of processing at another site. As a control, we also tested cleavage of the precursor to the triose-phosphate translocator (pre-TPT; ). As expected, 50% of pre-TPT was converted to its mature form in the assay (). These data provide additional evidence that int-atTic40 is exposed to the stroma before IM insertion and processing to mature atTic40. To determine if fully imported int-atTic40 and mature atTic40 had reached the IM, we fractionated chloroplasts from 2-, 5-, and 30-min import reactions. The samples were treated with trypsin to remove external or partially imported atTic40, osmotically lysed and separated into membrane and soluble fractions. As expected, the majority of mature atTic40 (∼75%) partitioned with the membrane fraction at the 5-min and 30-min time points (, left). In contrast, 88–94% of int-atTic40 was not associated with the membrane at the three time points (, right). Collectively, the data in and suggest that the protease-insensitive form of int-Tic40 is soluble and located in the chloroplast stroma. To confirm that int-atTic40 is a true intermediate on the pathway of IM targeting, we tested whether int-atTic40 was the precursor to mature, membrane-integrated atTic40. For this experiment (), chloroplasts were incubated with pre- atTic40 in the presence of 0.1 mM ATP to form an early import intermediate that is inserted in the Toc–Tic machinery but has not fully crossed the envelope (pulse; ; ). The chloroplasts were isolated to remove any unbound translation product (, lane 1) and resuspended under import conditions to promote import of the early intermediate. Chloroplasts from each time point during the chase were lysed and separated into soluble and membrane fractions. As expected, bound pre-atTic40 was quantitatively converted to mature atTic40 after a 30-min chase (). Both forms were predominantly associated with the membrane fraction. The abundance of total int-atTic40 peaks at 2 min of chase and then largely disappears by 30 min (). The peak of int-atTic40 follows the decrease in pre-atTic40 and precedes the accumulation of mature atTic40 (), consistent with it representing an intermediate in the conversion of pre- atTic40 to atTic40. Furthermore, int-atTic40 accumulates in the soluble fraction of chloroplasts before conversion to atTic40 (). Treatment of the chloroplasts from each time point in the pulse-chase experiment with thermolysin demonstrates that protease-resistant forms of both int-atTic40 and mature atTic40 accumulate during the chase (). As expected, membrane-integrated atTic40 attains 90% protease resistance after the 30-min chase (). The soluble int-atTic40 is almost completely protease resistant at 7.5 min into the chase (), confirming that the intermediate is passing through the stroma en route to its insertion into the membrane. To test whether int-atTic40 was the direct precursor to atTic40, we performed a variation of the pulse-chase experiment. For this assay, isolated chloroplasts were incubated with pre-atTic40 for 5 min under import conditions to accumulate soluble int-atTic40 (). The chloroplasts were chilled to stop the import reaction and were treated with trypsin to remove pre- atTic40 and int-atTic40 that had not been completely imported into the organelle (). The treated chloroplasts were isolated to remove the protease and incubated again under import conditions for 5 or 60 min (). The soluble int-atTic40 that accumulated after 5 min of import could be quantitatively converted to membrane-integrated mature atTic40 by the additional incubation (). On the basis of these results, we conclude that the targeting of atTic40 involves a soluble intermediate that targets to the IM after import into the stroma. A previous study on the import of pea Tic40 into isolated chloroplasts did not report a size intermediate in IM targeting (). Although we observed a soluble intermediate with characteristics similar to int-atTic40 when we repeated import with pea pre-Tic40 (unpublished data), the size of the intermediate was only slightly larger than the mature protein, indicating that the intermediate processing site is distinct from that in pre-atTic40. The difference in processing between pea and Tic40 raised questions about the role of the intermediate processing. Therefore, we tested whether the C-terminal region of the atTic40 transit peptide plays a role in IM targeting. We generated a series of increasingly larger deletion mutants in the intermediate sequence (residues 43–76; ) and examined their import and insertion properties. All deletions disrupt the putative processing site that gives rise to mature atTic40. Elimination of the four C-terminal residues of the transit peptide does not significantly affect import efficiency or processing to an intermediate size (). However, larger deletions dramatically reduce import efficiency in addition to inhibiting processing to the mature size (). Although import efficiency is reduced, deletions up to Δ58-76 appeared to target and insert into the IM, albeit with somewhat lower efficiencies than pre-atTic40 (). The effects of deletions larger than Δ58-76 on IM targeting could not be determined because of the combination of poor import efficiency and aberrant cleavage (), presumably because of disruption of the SPP processing site. To further ensure that the soluble forms of the deletions were productive intermediates in the targeting pathway, we performed a chase experiment similar to that shown in using the Δ58-76 mutant (). The Δ58-76 soluble intermediate can be chased to the membrane with a 30-min incubation, similar to authentic int-atTic40. Collectively, these results indicate that the intermediate sequence of the transit peptide participates in import but is not essential for membrane insertion. To further explore the determinants for pre-atTic40 targeting, we replaced its transit peptide with that of the small subunit of rubisco (psTP-Tic40). The import efficiency of this construct was comparable to pre-atTic40, but the protein was converted to mature atTic40 directly with no apparent intermediate processing (). Remarkably, the chimera targeted to the IM, albeit with approximately one half the efficiency of pre-atTic40 (). These results are consistent with the conclusion that the int-atTic40 intermediate sequence is not required for membrane integration but might facilitate the process. Finally, we tested whether the transmembrane helix of atTic40 was required for IM targeting. As expected, deletion of the transmembrane region (ΔTM-atTic40) had little effect on import but completely eliminated membrane integration (). To test directly whether membrane insertion is independent of import, we studied the ability of int-atTic40 to integrate into isolated IM vesicles. Inside-out IM vesicles were purified from isolated chloroplasts using previously published procedures (; ). To ensure that the majority of the vesicles were inside out, thereby exposing the inner face of the IM to the external buffer, we tested the protease sensitivity of the IM marker proteins psTic110 and psTic22 (; ; ). The bulk of psTic110 (∼95 kD) extends into the stroma (; ), whereas psTi22 is bound to the outer face of the IM in the intermembrane space (). The majority of psTic110 is degraded by either trypsin or thermolysin, whereas psTic22 is resistant to both proteases, indicating that the isolated vesicles are predominantly inside out (). Disruption of the vesicles with Triton X-100 resulted in complete degradation of psTic110 and psTic22 (). To quantify the proportion of inside- out vesicles in the population, we imported radiolabeled atTic110 into chloroplasts and treated isolated IM vesicles from these chloroplasts with thermolysin (). Greater than 90% of the radiolabeled atTic110 was degraded within 30 min, suggesting that at least 90% of the vesicle population was in the inside-out orientation. As an initial substrate for the vesicle insertion reaction, a soluble stromal extract containing int-atTic40 (, lane 2) was isolated from chloroplasts after a 5-min pre-atTic40 import reaction (, lane 1). This extract was incubated with isolated IM vesicles, and the association of int-atTic40 and mature atTic40 with the vesicles was assayed by differential centrifugation. As evident in , int-atTic40 bound to IM vesicles. Binding was time dependent, with the majority of int-atTic40 (55%) associated with the vesicles during the incubation (, graph). A small but substantial fraction of the int-atTic40 is converted to a form with a mobility identical to mature atTic40, suggesting that a portion of int-atTic40 is processed to atTic40 upon association with vesicles. Extending the incubation to 2.5 h increased the proportion of int-atTic40 that was converted to atTic40 in the presence but not the absence of IM vesicles (). Extraction of the vesicles with alkaline buffer demonstrates that the membrane-associated forms of atTic40 are integrated into the membrane bilayer (). Furthermore, treatment of the vesicles with thermolysin after the insertion reaction results in complete degradation of the atTic40 proteins (), indicating that they correctly inserted with the bulk of their sequences exposed at the inner face of the IM. As a control for vesicle targeting specificity, we tested the binding and insertion of int-atTic40 to canine pancreatic microsomal membranes. Binding to the heterologous membranes was insignificant when compared with control reactions lacking membranes (). As an additional control, we tested the protein dependence of int-atTic40 insertion into IM vesicles by treating the IM vesicles with varying concentrations of thermolysin before the insertion reaction (). The protease treatments dramatically reduced the binding and processing of int-atTic40 in IM vesicles, demonstrating the requirement of proteinaceous components at the IM for insertion. We also examined the energy dependence of the insertion reaction (). The stromal extract containing atTic40 was dialyzed to remove free nucleotides before the insertion reaction. As shown in , int-atTic40 bound and inserted into vesicles with similar efficiencies in the absence of added nucleoside triphosphates or in the presence of ATP or GTP. This observation suggests that direct insertion into the IM does not require nucleotide hydrolysis. Finally, we examined the role of stromal factors in int-atTic40 targeting by testing whether in vitro–translated forms of atTic40 could target directly to vesicles. We compared the targeting of in vitro–translated int-atTic40, atTic40, or the ΔTM-atTic40 mutant to stromal int-atTic40 (). Remarkably, in vitro–translated int-atTic40 and atTic40 both associated with vesicles and inserted into the membranes with efficiencies only slightly lower than that observed for stromal int-atTic40 (, graph). Furthermore, in vitro–translated int-atTic40 was partially processed to mature atTic40. As expected, in vitro–translated mature atTic40 was not processed (). These data confirm that int-atTic40 processing to mature atTic40 occurs upon association with the IM. The ΔTM-atTic40 mutant showed no significant association or insertion into the membrane vesicles (), confirming the selectivity of the insertion assay. Collectively, these results indicate that in vitro–translated int-atTic40 is capable of inserting into isolated IM vesicles, suggesting that specific stromal factors are not required for the targeting reaction. Previous studies on the import of Tic110 suggested that this IM protein also utilizes a soluble targeting intermediate (; ). Therefore, we examined the ability of in vitro–translated atTic110 to insert into IM vesicles. demonstrates that mature atTic110 binds to IM vesicles, and ∼50% of the bound protein is integrated into the membrane. As controls, we tested the binding and insertion of atTic110 lacking one (atTic110-ΔTM1 or atTic110-ΔTM2) or both of its transmembrane segments (atTic110Sol; ). Consistent with previous in vitro import experiments, the binding and insertion of the deletion mutants was dramatically lower than for full-length atTic110. atTic110 showed considerably lower binding and no appreciable insertion into pancreatic microsomal membranes (), demonstrating the membrane specificity of the reaction. Overall, the data with atTic40 and atTic110 demonstrate that these proteins can insert directly into the IM from a soluble state independent of protein import from the cytoplasm. We investigated the mechanism of import and insertion of atTic40 to better understand the process of protein targeting to the chloroplast IM. We demonstrate that the targeting and insertion of this simple single-pass transmembrane protein involves a soluble intermediate that inserts into the IM after it completes import from the cytoplasm. These observations were extended by our demonstration that atTic40 and a second IM protein, atTic110, can insert directly and selectively into isolated IM vesicles. These data are consistent with earlier observations with Tic110 (; ) and indicate that IM targeting of Tic110 and Tic40 involves soluble intermediates. Both preTic40 and preTic110 appear to initially use the general Toc–Tic pathway for import into the organelle (, ; ) and are processed by SPP, generating soluble, stromal intermediates. The intermediates insert into the IM from the stroma side of the membrane. Insertion is dependent on information within the transmembrane helices of the proteins (, , and ) and requires proteinaceous components of the inner membrane (). Our initial studies suggest that insertion occurs independent of nucleoside triphosphate hydrolysis () or the need for stromal factors (). The precise role of the two-step processing of the pre-atTic40 transit peptide in targeting, if any, remains to be determined. Our data indicate that the intermediate region, spanning amino acids 43–76 of the transit peptide, participates in import but is not required for IM insertion (). Furthermore, the intermediate processing site does not appear to be conserved because the pea orthologue of Tic40 appears to be initially processed at a site much closer to the C-terminal end of the transit peptide (; unpublished data). The initial cleavage to generate int-atTic40 during import is mediated by SPP (). The data in , , and indicate that the final processing step requires association with the inner membrane. Therefore, the processing could be performed by a specific envelope protease. Although the protease responsible for processing int-atTic40 to mature atTic40 remains to be identified, two potential processing peptidases have been identified at the envelope (; ). Although our data demonstrate that import and IM targeting are independent processes, one observation suggests that the two reactions can be coupled under some circumstances. The protease protection studies of pre-atTic40 import demonstrate that both int-atTic40 and mature atTic40 are largely exposed to the chloroplast surface at the earliest time points in import and, therefore, have not completed the import process (). Protease-resistant forms corresponding to fully imported int-atTic40 and mature atTic40 accumulate with time in the standard import assay () and in our import pulse-chase experiment (), reaching their maximum levels at later time points. Collectively, these observations suggest that IM targeting can initiate while int-atTic40 is in the process of import. The fact that the intermediate accumulates in the stroma with time suggests that the rate of import of pre-atTic40 from the cytoplasm exceeds the rate of insertion into the inner membrane, resulting in the inability of the insertion reaction to keep pace with import. Our studies provide compelling evidence that the import and membrane insertion of atTic40 and atTic110 are independent processes. However, we cannot rule out the possibility that other IM proteins (e.g., polytopic transporters) use a stop-transfer mechanism. In mitochondria, both processes appear to operate for targeting of proteins to the IM (; ; ; ). Nucleus-encoded mitochondrial IM proteins appear to initially engage the general import machinery of the outer mitochondrial membrane (TOM complex; ). A small number of nucleus-encoded mitochondrial IM proteins are inserted into the membrane from the matrix after they complete import (; ; ). The Oxa1 pathway mediates insertion of several of these proteins (; ; ). This pathway is a member of the Oxa1/Alb3/YidC family of protein export pathways that are conserved from prokaryotes. Oxa1 and a second pathway also mediate the insertion of mitochondria-encoded IM proteins (; ). However, for the majority of mitochondrial IM proteins, complete translocation into the matrix is interrupted by stop-transfer sequences, resulting in the lateral insertion of the proteins into the lipid bilayer via the Tim22 or Tim23 complexes (; ). Additional studies with a wider array of chloroplast IM proteins should demonstrate whether stop-transfer mechanisms also exist in chloroplasts. Although the pathway of Tic40 and Tic110 targeting in chloroplasts and the Oxa1 pathway in mitochondria are analogous at first examination, it is unclear whether the chloroplast proteins use a conservative targeting pathway. Genomic and proteomic studies fail to identify proteins homologous to bacterial export pathways in the chloroplast IM (, ; ; ; ). Furthermore, the conserved protein export pathways, including the Oxa1/YidC homologue, Alb3, are all present at the thylakoid membrane (). Therefore, the relationship of the chloroplast IM targeting translocon to the conserved translocons at the molecular level remains to be determined. Identification of the factors involved in IM targeting in the chloroplast will provide more definitive evidence for the molecular mechanism of this novel membrane protein targeting system. Complete pre-atTic40 (available from GenBank/EMBL/DDBJ under accession no. ) and pre-psTic40 (available from GenBank/DMBL/DDBJ under accession no. ) coding regions were amplified from total seedling cDNA by RT-PCR. The atTic40 cDNA was modified by overlap extension PCR to introduce a silent mutation that eliminated the internal Nco1 restriction site. Both cDNAs were cloned into the Nco1 and BamH1 sites of pET21d. The atTic40 deletion mutants were generated by overlap extension PCR of the atTic40 cDNA to eliminate the specific residues indicated in the figures. The psTP-Tic40 construct (pET21d-pssuTP-matTic40) corresponds to the 57 amino acids of the complete transit peptide of the pea small subunit of rubisco (), fused to mature atTic40. The recombinant int-atTic40 (pET21d-int-atTic40) and mature atTic40 (pET21d-atTic40) constructs for in vitro translation lack the N-terminal 42 and 76 amino acids of pre-atTic40, respectively. The int-atTic40 construct contains an additional Gly codon following the initiation codon to introduce a 5′ Nco1 restriction site. The atTic110ΔTM1 (pET21d-mature atTic110-ΔTM1-His) and atTic110ΔTM2 (pET21d-mature atTic110-ΔTM2-His) constructs contained deletions of amino acids 43–62 and 70–89 of mature atTic110, respectively. The atTic110 and atTic110Sol constructs were previously described (). Import substrates were generated using a coupled in vitro transcription–translation system derived from reticulocyte lysate (Promega). Chloroplast isolation from pea and and the in vitro import experiments were performed as previously described (). For the time course of in vitro import, reactions were stopped at the indicated time points by rapid dilution with three volumes of ice-cold 50 mM Hepes-KOH, pH 7.5, and 330 mM Sorbitol (HS buffer). Protease treatments were performed by dividing the import reactions into three equal parts and diluting each with threefold excess ice-cold HS buffer alone or buffer containing trypsin (50 μg/ml final concentration) or thermolysin (100 μg/ml final concentration). The reactions were incubated on ice for 30 min, and proteolysis was stopped with 1 mM PMSF, 0.05 mg/ml TLCK, 0.1 mg/ml soybean trypsin inhibitor, and 2 μg/ml aprotinin (trypsin inhibitor; ) or 10 mM EDTA (thermolysin inhibitor). The chloroplasts were isolated through 40% Percoll silica gel containing the corresponding protease inhibitor, washed once with HS buffer, and processed for SDS-PAGE analysis. All quantitative analysis of radiolabeled samples was performed with a FLA-5000 phosphorimager and Multi Gauge v. 3.0 software (Fujifilm). Chloroplasts were lysed by suspension in HS buffer to a concentration of 0.5–1 mg chlorophyll/ml and diluted with five volumes of 2 mM EDTA. The lysate was mixed vigorously and incubated on ice for 10 min. The samples were adjusted to 0.2 M NaCl, and the membrane fraction was collected by centrifugation at 18,000 for 30 min at 4°C (). For alkaline extraction, the membrane pellet was resuspended with a small volume of HS buffer and diluted with 20 volumes of 0.2 M NaCO, pH 12. The samples were homogenized with a Teflon homogenizer (Kontes Glass Co.) and incubated at room temperature for 10 min, and the membrane fraction was collected by centrifugation at 100,000 for 15 min. The soluble fractions were removed and concentrated by precipitation in 20% trichloroacetic acid. The stromal extract was prepared as previously described ( ). For the SPP processing assays, 1 μl of in vitro–translation product was incubated with 20 μl stromal extract at 26°C for 90 min. As a control, 1 μl of in vitro–translation product was mock treated with 20 μl of 5 mM Hepes-KOH, pH 8.0. For the pre-atTic40 binding and chase experiments in , isolated chloroplasts were depleted of internal ATP by incubation in the dark at room temperature for 15 min. In vitro–translated pre-atTic40 was depleted of nucleotides by gel filtration. The early import intermediate was generated by incubating chloroplasts with pre-atTic40 in the presence of 100 μM ATP under import conditions (; ). The chloroplasts were reisolated through 40% Percoll silica gel, washed once with ice-cold HS buffer, and resuspended in ice-cold HS buffer containing 50 mM KOAc and 4 mM MgOAc (import buffer) without ATP. A fraction was removed as the 0-min sample, and the remaining chloroplasts were diluted into 10 volumes of import buffer containing 2 mM ATP at 26°C. Equal fractions of the import reaction were removed at 2, 4, 7.5, and 30 min, reisolated, and separated into membrane and soluble fractions. For the experiment in to test int-atTic40 conversion to atTic40, chloroplasts (400 μg chlorophyll) were incubated with pre-atTic40 for 5 min under standard import conditions. The reaction was stopped, and chloroplasts were treated in the absence or presence of trypsin as described in Chloroplast isolation and in vitro import assays. After recovery through Percoll silica gel, the chloroplasts were suspended to 2 ml in prewarmed (26°C) import buffer containing 2 mM ATP. Equal fractions of the reaction mixture were collected at 5 and 60 min, reisolated, and separated into membrane and soluble fractions. The chloroplast IM vesicles were isolated as described previously (). The membranes were recovered by differential centrifugation and stored in HS buffer containing 1 mM dithiothreitol at −80°C. The stromal extract containing int-atTic40 was isolated using chloroplasts from a 5-min pre-atTic40 import reaction as described in SPP assay. The chloroplasts (1–2 mg chlorophyll/ml final concentration) were lysed hypertonically with 10 mM Hepes-KOH, pH 8.0, and 10 mM MgCl () on ice for 10 min. The stromal fraction was separated from the membrane fraction by centrifugation at 40,000 for 30 min, adjusted to 50 mM Hepes-KOH, pH 7.5, 50 mM KOAc, and 4 mM MgCl, and stored at −80°C. For a typical in vitro insertion assay, 15 μl of isolated stromal extract (50 μg protein) or 1 μl in vitro–translated atTic40 or atTic110 protein was incubated with 30 μg inner membrane vesicles under conditions identical to those used for standard chloroplast import assays. The vesicle insertion was performed at 26°C for 1 or 2.5 h. Then, three volumes of ice-cold HS buffer were added and the vesicles were reisolated by spinning at 100,000 for 20 min. The vesicles were washed once with HS buffer and resuspended in 30 μl HS buffer before further treatments, such as carbonate extraction. Fig. S1 shows that the int-atTic40 targeting intermediate is observed during import into both pea and chloroplasts. Online supplemental material is available at .

Quality control of newly synthesized glycoproteins involves recognition of misfolded proteins in the ER, where they are either returned to a productive folding pathway or are targeted for degradation (). Terminally misfolded glycoproteins are transferred to the cytoplasm for proteasomal proteolysis, a process termed dislocation (,). How the cell distinguishes between newly synthesized proteins that have not yet acquired their correct folding state and proteins that are terminally misfolded remains a mystery. In yeast, genetic analysis has shown the involvement of a limited set of proteins that contribute to recognition of misfolded proteins and their subsequent degradation. The secretory protein carboxypeptidase Y (CPY), when engineered to yield a misfolded product, CPY*, has served as a substrate to identify the genetic factors that interfere with its disposal. Der1p was identified as a key player in clearing the yeast ER of misfolded CPY* (; ; ; ). HMG-CoA reductase, which is a transmembrane protein, has similarly served as a reporter substrate, allowing to define HRD1 and HRD3 as essential for its degradation (, ). Hrd1p/Der3p has ubiquitin ligase activity (E3) and forms a complex predominantly with the ubiquitin-conjugating enzymes (E2s) Ubc7p and Ubc1p (), which are themselves recruited by the protein Cue1p () to the site of degradation. Hrd3p is required for regulating the activity and stability of Hrd1p/Der3p (), but the function of Hrd3p in protein degradation remains obscure. Hrd3p has a large luminal domain that contains different sets of repeated regions that might be involved in substrate recognition or form complexes with chaperones. Apart from the Ring-H2 ligase Hrd1p/Der3p, there are additional ER membrane–resident E3s, such as Doa10p (). Depending on the topology of the ER degradation substrates, different proteins are required for their clearance (). Substrates with defects in their cytosolic domain require Doa10p in yeast. Substrates with defects in their luminal portion require the ER lectin Htm1p/Mnl1p, the ubiquitin ligase Hrd1p/Der3p-Hrd3p, Der1p, and proteins involved in ER–Golgi trafficking (). The two pathways merge when leaving the ER; extraction of the ubiquitin-modified substrate occurs with the assistance of Cdc48p/p97 and its cofactors Ufd1p and Npl4p, culminating in delivery to the proteasome and proteolysis of the substrate (, ; , ; ; ). Recent studies analyzed the composition of the protein complexes involved. The Doa10p complex contains Ubc7p, Cue1p, Ubx2p, Cdc48p, and its cofactors Ufd1p and Npl4p. These proteins are mainly cytosolic, supporting Doa10p's role in clearing proteins with defects in their cytosolic domain. In addition to these proteins, the Hrd1p complex consists of Hrd3p, Der1p, the ER lectin Yos9p, and Usa1p (; ). Yos9p has been shown to specifically bind misfolded glycoproteins (; ; ). Ubx2p recruits Cdc48p to the membrane (). Usa1p is thought to link Der1p to the Hrd1p ligase and thereby assist in clearing luminally misfolded proteins from the ER (). In mammalian cells, the dislocation pathway is more complex. Because of the lack of a genetic approach, the dissection of the degradation pathway in mammalian cells relies on the use of substrates, such as mutant versions of the cystic fibrosis chloride conductance channel (; ; ; ), of proteins considered terminally misfolded, such as the mutant null Hong Kong (NHK) version of the secretory glycoprotein α1 antitrypsin () or the truncated and misfolded version of the ER-resident glycoprotein ribophorin (RI), termed RI (; ). Many membrane proteins fail to fold properly for lack of the correct partner subunits, such as the unpaired T cell receptor α chain (TCRα; ,) or the free immunoglobulin μ chain (). Many parallels exist between the yeast and the mammalian glycoprotein quality control systems; however, the players in the mammalian system are more numerous. The mammalian version of HRD1 is also an ER membrane–resident ubiquitin ligase (; ; ), which forms a complex with SEL1L, a mammalian orthologue of yeast Hrd3p (). Additional ubiquitin ligases exist, such as gp78, which has similarity to HRD1 in its Ring finger and interacts with UBC7 via its CUE domain to ubiquitinate TCRα (). There are at least three Der1p homologues in mammals, Derlin-1, -2, and -3, which play roles in the disposal of proteins from the ER (, ; ). Among the better-studied routes of membrane glycoprotein degradation are the pathways used by human cytomegalovirus (HCMV) to destroy the class I major histocompatibility complex (MHC) heavy chains (HCs; ). Class I MHC products serve as a warning system to alert cytotoxic T cells to the presence of virus-derived polypeptides inside the cell. The infected cell, thus, invites attack by the cytotoxic T cell as a means of eradicating the source of the virus (). Large DNA viruses, such as HCMV, are under strong selective pressure to avoid recognition by the immune system. Although widespread amongst the [herpesviridae as a strategy to avoid detection, HCMV in particular has amassed a set of genes whose products evolved to interfere with assembly and intracellular transport of class I MHC products. Amid the HCMV-encoded immunoevasins, two, US2 and US11, which are small membrane glycoproteins that assist in the degradation of class I MHC HCs, stand out. US2 and US11 catalyze superficially similar reactions, characterized by complex formation of US2 or US11 with their target class I MHC HCs, and subsequent extraction of the class I MHC HCs from the ER, a process referred to as dislocation (,). After dislocation, the class I HC is destroyed by the proteasome. In the presence of proteasome inhibitors, a diagnostic intermediate in this pathway is the result of attack by N-glycanase on the newly dislocated class I MHC HCs (,; ; ). This intermediate consists of a fully cytoplasmic, yet intact, class I MHC HC, devoid of its N-linked glycan (). It occurs not only in cells that express US2 or US11, but also in Daudi cells, which are unable to assemble class I MHC products for lack of the light chain β2-microglobulin (; ). Similar deglycosylated intermediates have been reported for a misfolded fragment of RI (, ), but for most other glycoproteins examined, degradation occurs without the obvious production of deglycosylated intermediates. These observations raise the question as to whether the pathways exemplified by US2- and US11-dependent degradation are indeed emblematic of glycoprotein turnover, as we have argued in the past. The furthest advanced is the characterization of the US11 pathway, for which we identified the Derlin-1 protein as an essential participant (; ). Derlin-1, in turn, forms a complex not only with itself and other members of the Derlin family, but also with the mammalian Hrd1p and Hrd3p (SEL1L) homologues, as well as with Cdc48p/p97 (; ). The US2 pathway is impervious to the action of a dominant-negative version of Derlin-1, whereas the US11 pathway is inhibited by it (). Direct involvement in mammalian glycoprotein degradation has now been suggested for components of this ER-resident complex, based on interference with degradation of α1 antitrypsin NHK by means of overexpression and knockdowns of Derlins (). Still, in mammalian cells, much of the relevant data stem from the analysis of reactions that depend on the action of the viral accessories US11 and US2. We analyze the contributions of a mammalian Hrd3p homologue, SEL1L. We show that reduction of the levels of SEL1L by RNA interference results in impairment of US11-mediated dislocation of class I MHC HC molecules. Expression of the same interfering small hairpin RNAs (shRNAs) against SEL1L, however, does not affect the US2 pathway. Expression of SEL1L shRNAs inhibits the degradation of RI, which is a process that occurs independently of the presence of viral accessories. These results allow us to place SEL1L in a more general context of glycoprotein degradation, and suggest that SEL1L might be involved in substrate recognition of misfolded proteins in the ER. SEL1L is part of a mammalian ER multiprotein complex involved in dislocation (; ). SEL1L associates with Derlin-1, -2, p97, and HRD1, as well as with additional proteins that remain to be identified, some or all of which may also play a role in dislocation. Some of these proteins are presumably recruited to the site of dislocation through SEL1L. SEL1L is predicted to be a type I transmembrane protein (Fig. S1, available at ) with 5 N-linked glycans (). Because the bulk of the SEL1L protein is predicted to be in the ER lumen, it is possible that SEL1L first plays a role in substrate recognition and identification of misfolded proteins, and then recruits them to the site of dislocation. We performed coimmunoprecipitations from steady-state radioactive [S]methionine/cysteine–labeled cells using mild lysis conditions (2% digitonin) to preserve ER membrane protein complexes, to analyze whether composition of protein complexes with SEL1L are conserved for the cell lines used in this study. Our earlier observations concern the composition of protein complexes that include the Derlins in the astrocytoma cell line U373 (). We first verified that the types of complexes detected in U373 cells are not cell type–specific, but occur in other cell lines as well (). We were particularly interested in cell lines that would lend themselves to transient transfection experiments (, HeLa cells, lanes 1–4) and in multiple myeloma cells (, MM1.S cells, lanes 5–8), which represent cells with a high secretory capacity, and, presumably, a correspondingly pronounced requirement to clear misfolded proteins from the ER. When we performed immunoprecipitations with antibodies against Derlin-1, -2, and SEL1L, the composition of the complexes obtained was very similar to that reported for U373 cells (). We conclude that the multiprotein complexes, which were initially identified for U373 cells, are representative of the complexes detected in other, unrelated cell types. The pattern of the protein complexes is equivalent in all tested cell lines, including U373 (astrocytoma), HeLa (epithelial), 293T, MM1.S (multiple myeloma), and C6 (Rat glioma) cells. SEL1L is likely to be a crucial component of a ubiquitous ER-dislocation complex. When we analyzed the half-life of SEL1L, we found that it decays with a half-life of ∼180 min (, lanes 1, 3, and 5). SEL1L degradation appears to be proteasome-dependent because it is inhibited by inclusion of the proteasome inhibitor ZLVS (, lanes 7, 9, and 11). SEL1L remains completely susceptible to endoglycosidase H (EndoH) treatment, suggesting that SEL1L is an ER protein and does not traffic through the secretory pathway (, lanes 2, 4, 6, 8, 10, and 12). To address whether SEL1L plays a direct role in the process of dislocation from the ER, we generated small hairpin RNAs (shRNAs) that target SEL1L or a control unrelated protein (GFP) using the pRETRO-SUPER in vivo expression system (). The shRNA plasmids reduced SEL1L protein levels to 30% compared with control cells (). We analyzed the stability of class I MHC HCs in cells expressing US11 and the SEL1L shRNAs. The fate of HC is best analyzed in a pulse-chase experiment in view of the rate of dislocation ( = 2–5 min) in US11 cells (). In cells expressing US11 and shRNAs against GFP (control cells), HC disappears completely when proteasome inhibitors are not included (, , lanes 2 and 3). In cells expressing SEL1L shRNA, the rate of degradation of class I MHC HC is much reduced (, , lanes 8 and 9). In the presence of the proteasome inhibitor ZLVS, the deglycosylated class I MHC HC accumulates as the diagnostic intermediate that characterizes the dislocation reaction. In control shRNA US11 cells, complete conversion to the deglycosylated form of HC is seen after 30 min of chase (lane 6). In SEL1L knockdown cells, >50% of HC remains in its fully glycosylated form. This persistence of HC is attributable to compromised dislocation, as only ∼50% of HC accesses the cytosolically disposed N-glycanase (, 1 [lanes 11 and 12], and c). This impairment in dislocation is not caused by reduced levels of US11 in SEL1L knockdown cells, as US11 is expressed at comparable levels, nor is it caused by aberrant ER insertion and processing of US11, as determined by normal cleavage of US11's signal sequence () in SEL1L knockdown cells (, ). Because SEL1L is a mammalian homologue of yeast Hrd3p, we looked for possible parallels between the two systems that could account for the observed phenotype. Hrd1p, which is a yeast E3 ligase known to be necessary for degradation of some substrates (; ; ; ; ), is regulated by Hrd3p and destabilized in yeast (; ). Therefore, we analyzed HRD1 levels in SEL1L knockdown cells to examine whether inhibition of dislocation is primarily caused by reduced levels of SEL1L or, alternatively, attributable to a reduction in HRD1 levels. We find that HRD1 is stable throughout the chase periods over which HC dislocation occurs (, ). A comparable reduction of SEL1L levels in US2 cells () has no observable effect on HC dislocation. In US2 cells exposed to proteasome inhibitor, we see conversion of almost all class I MHC HCs to the deglycosylated species after 30 min (, , lane 6). In US2 cells, knockdown of SEL1L results in a rate of dislocation equal to that in control cells and yields a pattern noticeably distinct from that seen for the SEL1L knockdown in US11 cells (, b and c). US2 levels are also comparable for control and SEL1L knockdown cells (, ). Unimpaired degradation in US2 cells of class I MHC HCs is consistent with the idea that US11 co-opts a conserved complex to catalyze degradation of class I MHC HC, whereas US2 utilizes a distinct pathway (, ; ). This result is all the more striking because US2 and US11 target the same set of substrates, class I MHC products. Given the homology of SEL1L to Hrd3p, it is likely that SEL1L plays a general role in glycoprotein turnover, similar to what was shown for the Der1p homologues of the Derlin proteins (; ). Thus far, the only substrates that for their dislocation depend on Derlin-1 are class I MHC HCs in US11-expressing cells and US2 itself (). Overexpression and knockdowns of Derlin-2 and -3 have been reported to affect the half-life of α1 antitrypsin NHK (). Therefore, we examined other model substrates of ER dislocation for their susceptibility to inhibition by a knockdown of SEL1L. We analyzed the effect of SEL1L shRNAs on the degradation of a mutant version of RI, RI. The RI mutant lacks the C terminus of RI; unlike the very stable RI protein, the RI fragment has a half-life of ∼90 min (; , ). We chose RI as substrate because its mode of disposal includes an intermediate that is the product of N-glycanase activity (). We transiently transfected RI into HeLa cells, which had been stably transduced with shRNAs against either GFP or SEL1L (). SEL1L knockdown cells showed a reduction of SEL1L levels to 40% of controls. We transfected a GFP expression plasmid into both GFP shRNA and SEL1L shRNA HeLa cells as a transfection control, as well as a control for shRNA-mediated knockdown in the HeLa shRNA GFP control cells. The percentage of cells expressing GFP was equivalent (∼40%), but, as expected, the intensity of the GFP signal was much lower in GFP shRNA cells (unpublished data). Two days after transfection, cells were harvested, pulsed for 20 min with [S]methionine, and chased for 0, 90, and 180 min. Immunoprecipitation with anti-RI antibody retrieves both the truncated and wild-type RI. In control shRNA GFP HeLa cells, ∼90% of truncated RI is degraded after 180 min, whereas in SEL1L knockdown cells, we see a considerable stabilization of RI; ∼40% of RI remains (, b [lanes 3 and 6] and c). The observed stabilization of RI is all the more noteworthy as the achieved stable knockdown of SEL1L in HeLa cells is ≥40% (). Wild-type RI (a stable protein in the ER) remains unaffected, and expression of SEL1L shRNA does not affect overall glycosylation as assessed by class I MHC HC and US11 maturation and glycosylation earlier. SEL1L is essential for dislocation of class I MHC molecules from the ER in cells that express US11. Upon reduction of the levels of the SEL1L protein, fully glycosylated, EndoH-sensitive class I MHC HCs accumulate in the ER. In contrast, the US2 pathway is not perturbed by reduction of SEL1L protein levels. These results underscore the relevance of other observations that place US11 and US2 in distinct pathways, e.g., based on their susceptibility to interference with Derlin-1 function; a Derlin-1 fusion protein impairs degradation of class I MHC molecules via the US11-dependent, but not via the US2-dependent, pathway (). There are several lines of evidence to suggest that US2 and US11 use different principles to accelerate degradation of class I MHC molecules, in addition to the aforementioned difference in Derlin-1 dependency. US2 must exploit features in its relatively short cytoplasmic tail, whereas tailless US11 remains dislocation-competent (; ). The preference of the US2 and US11 pathways for folded and unfolded class I MHC molecules appears to be different (), as is the requirement for elements in the tail of the class I MHC molecules themselves (). Our data now establish that the US11, but not the US2, pathway is sensitive to a SEL1L knockdown. These distinctions are all the more remarkable given the similarities of the substrates targeted for degradation, the identical class I MHC HCs in the same parent cell line. We also report that down-regulation of SEL1L affects the degradation of a RI fragment whose route of degradation is similar to what we have described for class I MHC products targeted by the HCMV immunoevasins (; ). In the presence of proteasome inhibitors, a deglycosylated RI species, the product of N-glycanase digestion, is observed (). Our data, thus, support the notion that human SEL1L is the orthologue of yeast Hrd3p because reduction of SEL1L levels perturbs the degradation of a misfolded substrate, RI. We conclude that SEL1L and its associated partners operate in a pathway that is neither restricted to class I MHC products, nor strictly dependent on the involvement of virus-encoded proteins. The concept of physically extracting proteins from the ER and their delivery to the cytosolic proteasome achieves the compartmentalization required to spare nascent and properly folded ER proteins from premature degradation. The advantages of studying the pathways for the HCMV US2 and US11 proteins are the speed with which degradation occurs and the involvement of well-characterized substrates, the class I MHC products. The available antibodies allow an easy distinction between various folding intermediates, and the occurrence of the tell-tale deglycosylated degradation intermediate is an accurate reporter for its localization and for the dislocation reaction per se (,). We have taken advantage of the fact that US2 and US11 attract very similar substrates in one and the same cell, yet apparently do so by mechanistically different pathways (; ; ). This does raise the question, however, of whether the results obtained for these HCMV immunoevasins and class I MHC substrates can be extrapolated to other substrates, and to pathways that operate independently of viral accessories. To our knowledge, the observation that SEL1L is involved in the degradation of RI, a known dislocation substrate, is the first example for mammalian cells that directly places SEL1L in a pathway of protein degradation. The dominant-negative version of Derlin-1, Derlin-1, had no effect on RI degradation (unpublished data), making the mammalian degradation pathways even more complex. The different degradation pathways that, in yeast, process distinct types of substrates (luminally vs. cytoplasmically exposed proteins) likely operate in mammals, too, but involve more factors that comprise the different complexes. In addition to the multiple Derlin proteins, ubiquitin ligases, and lectins involved in mammalian ER degradation, there is an additional SEL1L-like protein in mammals that bears homology to Hrd3p (Fig. S1). Whether this protein binds to HRD1 and participates in the degradation of misfolded proteins remains to be determined. In yeast, substrates that require Der1p for degradation require Hrd1p/Hrd3p and usually belong to the set of completely luminal substrates. In mammals, depending on the substrate, Derlin-1 and SEL1L can act in concert with each other, as we show for class I MHC HC, but SEL1L can also assist in the degradation of substrates independently of Derlin-1, as is the case for RI. A recent study suggests that, in yeast, Hrd3p and Der1p can recruit substrates independently of each other (). We also examined a substrate that is recognized as misfolded by means of charged residues in its transmembrane domain, TCRα. We observed modest stabilization when SEL1L expression is compromised. The lesser effect on this substrate compared with RI, which is an entirely luminal substrate, further suggests that the nature of the substrate determines recruitment of an otherwise conserved complex. We cannot exclude that different degradation substrates require different levels of SEL1L. Perhaps TCRα degradation usually proceeds in the presence of very few SEL1L molecules. The ER machinery involved in recognition of substrates is likely to be specific for certain substrates, but the details of recognition of misfolded substrates in yeast are no better resolved than in mammalian cells. In yeast, Yos9p, a putative lectin protein, is involved in the degradation of a membrane-bound version of CPY* () and might be involved in identifying and targeting substrates for degradation (; ; ). A recent study suggests that proteins that are misfolded in yeast bind to Hrd3p, which itself binds to Yos9p. Yos9p then ensures that only terminally misfolded proteins are being degraded (). Another avenue to identification of misfolded glycoproteins might be through their high-mannose–containing, N-linked glycan modifications (). Mannose residues are trimmed in the ER by the enzyme α-mannosidase I (; ), generating a Man8GlcNac2 tag implicated in targeting glycoproteins for degradation (). Proteins with this tag bind to calnexin and other lectins, the ER degradation-enhancing mannosidase-like (EDEM) proteins, EDEM1 and EDEM2 (, ; ; ), which then target some glycoproteins for degradation (; ; ) There is a mammalian orthologue of Yos9p, OS-9, whose function is unknown but might be critical in degradation of certain substrates. Similar to EDEM, SEL1L might be involved in substrate recognition and bind to a subset of misfolded proteins. SEL1L is remarkably conserved, and contains several repetitive structural and functional domains (sel1-like repeats of the TPR family) and a type II fibronectin domain in its large luminal part that could bind to chaperones or misfolded proteins. Alternatively, SEL1L might recruit substrate recognition proteins such as EDEM or OS-9 through its type II fibronectin domain at its N terminus. In yeast, Hrd3p's N-terminal domain is essential for its function in ER degradation, and its central region required for interaction with Hrd1p (). SEL1L forms a 1:1 stoichiometric complex with HRD1 in mammalian cells (), but until now, SEL1L's direct involvement in protein degradation has not been shown. Thus, we conclude that SEL1L, based on its associated partners, its direct involvement in glycoprotein degradation, and its structural relationship to yeast Hrd3p, helps select proteins for entry into a degradative pathway. Our data provide the first functional evidence to support the notion that SEL1L is indeed the orthologue of yeast Hrd3p. Anti-HC, anti-US2, anti-US11, anti–Derlin-1, anti–Derlin-2, and anti-SEL1L antibodies have been described previously (; ; ; , ). Anti-RI antibodies and the RI DNA construct were a gift from N.E. Ivessa (University of Vienna, Vienna, Austria) and G. Kreibich (New York University Medical Center, New York, NY), anti-TCRα antibodies were purchased from Sigma-Aldrich, and anti-HRD1 antibodies were a gift from E. Wiertz (Leiden University, Leiden, Netherlands; ). The TCRα expression construct was described previously (). For stable shRNA expression, the pRETRO-SUPER vector was used (). U373-MG astrocytoma/glioblastoma cells, US2, US11, 293T, and HeLa cells were cultured as previously described (; ). These procedures were similar to those described in . Treatment of cells with the proteasome inhibitor ZLVS has been previously described (). Analysis of class I MHC HC stability in US11 and US2 cells, pulse-labeling, cell lysis (SDS and digitonin), and re-immunoprecipitations have been previously described (). NEM (N-ethylmaleimide) was included in digitonin and SDS lysis at a concentration of 2.5 mM. SDS PAGE has been previously described (). 19-nt target sequences in SEL1L were chosen according to the online design software available from the Whitehead Institute webpage (). Sense and antisense strands were annealed to form the shRNA template insert and ligated into the retroviral vector pRETRO SUPER () to generate shRNA construct s2 for stable shRNA expression and for transient transfection (19mer sequences; GGCTATACTGTGGCTAGAA and TTCTAGCCACAGTATAGCC, respectively). As a control, the unrelated construct (GFP shRNA) targeting the enhanced GFP reporter was used (). All constructs were sequence-confirmed and used for virus production in HEK 293T cells (). pRETRO-transduced U373 cells and HeLa cells were selected with 0.375 μg/ml and 1 μg/ml puromycin, respectively, as previously described (). Fig. S1 shows a schematic of the Hrd3p homologues in mammals. Fig. S2 shows SEL1L interacting with RI 293T cells transfected with RI. Online supplemental material is available at .