Datasets:

chemNLP

/

ncbi-text

like 0

DNA damage response mechanisms function to maintain genomic stability in normal cells. Because genomic instability is a characteristic of cancer cells, it is evident that at least some of these damage response pathways become impaired during progression to carcinogenesis. Additionally, patients with defective damage response pathways frequently show cancer predisposition, of which ataxia telangiectasia (A-T) is a well-known example. Significant insight has been gained into the roles of individual damage response pathways. Understanding the efficiency as well as the interplay between them is an important next step (; ). DNA double-strand break (DSB) repair and cell cycle checkpoint arrest represent two pathways to maintain genomic stability (; ; ; ; ; ). A-T mutated (ATM) plays a critical role in regulating cell cycle checkpoint arrest in response to DSBs (; ; ) and regulates a component of DSB repair (; ). The prevailing evidence suggests that in G0/G1, ATM is required for Artemis, a nuclease, to process a subset (∼15%) of radiation-induced DSBs before rejoining. A-T, a disorder caused by mutations in ATM, is associated with pronounced chromosomal instability, cancer susceptibility, and clinical radiosensitivity. This has generally been attributed to ATM's role in cell cycle checkpoint regulation. However, older cytogenetic data (; ) and the recent repair defect described in A-T cells () raises the issue of how ATM's repair and checkpoint functions interplay to maintain chromosome stability. Here, we exploit A-T as a model to define the efficiency and dissect the interplay between DNA repair and cell cycle checkpoint pathways, focusing our attention on two ATM-dependent functions, DSB repair in G2 and G2/M checkpoint arrest. To investigate the contribution of ATM and Artemis to DSB repair in cell cycle phases other than G0, we analyzed asynchronously growing cell cultures to avoid the potential introduction of DSBs during synchronization. In one approach, we used pan-nuclear centromere protein F (CENP-F) staining to identify G2 cells (; ; Fig. S1 A, available at ) and added aphidicolin to prevent S phase cells from progressing into G2 during analysis. Mitotic cells exhibiting distinct centromeric CENP-F staining and condensed chromatin were excluded from analysis. Under these conditions, S phase cells do not progress into G2 (Fig. S1 B) and a considerable proportion of cells irradiated with 1.5 Gy x-rays remain positive for pan-nuclear CENP-F staining for 6–8 h (i.e., they remain in G2), providing sufficient time to detect the ATM/Artemis repair defect, which is measurable at >4 h after irradiation in G0 cells. Enumeration of γH2AX foci in aphidicolin-treated CENP-F–positive primary human fibroblasts after 1.5 Gy x-irradiation demonstrated that ATM- and Artemis-dependent DSB repair operates in G2 (). Aphidicolin treatment did not affect the repair capacity of G2 cells (Fig. S1 C) but caused pronounced H2AX phosphorylation in cells that were CENP-F negative but positive for the S/G2 marker, cyclin A, most likely because of the activation of ATR after replication arrest (Fig. S1 A). Enumeration of γH2AX foci in CENP-F–negative cells that were also negative for the pronounced, aphidicolin-induced γH2AX phosphorylation allowed the analysis of repair in G1 phase cells (). For all cell lines, we observed similar kinetics and magnitude of repair in G1 and G2, which was also similar to that previously observed in G0 cells (). Foci numbers correlated with DNA content being twice as high in G2 compared with G1 (). In analogy to our previous study (), we confirmed that ATM and Artemis operate in the same repair pathway by analyzing the repair defect in Artemis cells treated with the specific ATM small molecule inhibitor KU55933 (). The dual deficiency in Artemis and ATM did not cause an increased repair defect relative to the defect in Artemis cells (). Thus, ATM and Artemis are epistatic in G1 and G2 and function to repair a subfraction of DSBs similar to that observed in confluent cells. Because our results were obtained with nonisogenic human cell lines, we also investigated γH2AX foci formation in matching wild-type (WT), A-T, and Artemis mouse embryonic fibroblasts (MEFs) using procedures similar to those used with human cells and observed identical repair kinetics (Fig. S1 D). To substantiate that γH2AX foci analysis monitors DSB repair, we developed and applied a pulsed-field gel electrophoresis (PFGE) technique to monitor DSB repair specifically in G2 phase cells (). Exponentially growing primary human fibroblasts were pulse-labeled with [methyl-H]thymidine for 1 h and irradiated with 80 Gy 4 h after labeling (when in G2; Fig. S1 E). After 48 and 72 h of repair, cells were harvested and the fraction of radioactivity released (FAR) from the gel plug into the gel was quantified by liquid scintillation counting. The FAR values after repair incubation provide an estimate of the level of unrepaired DSBs and can be compared with FAR values obtained from samples analyzed immediately after irradiation without repair. FACS analysis of parallel samples labeled with BrdU instead of [methyl-H]thymidine showed that labeled cells have progressed to late S/G2 at the time of irradiation (4 h after labeling) and remained in G2 for at least 72 h after irradiation with 80 Gy (Fig. S1 E). We obtained a similar level of unrepaired DSBs in A-T and Artemis cells, which was similar to (or slightly higher than) the level of DSBs induced in cells irradiated with 10 Gy and not incubated for repair (i.e., ∼1/8 of the DSBs induced by 80 Gy remain unrepaired; ). Thus, the magnitude of the G2 repair defect measured by PFGE is similar to the ∼15% repair defect observed by γH2AX foci analysis of G2 or G1 cells () and confluent cells (). The identical repair defect of A-T and Artemis cells in G2 and G1 is perhaps surprising, given that ATM has been reported to be required for homologous recombination. One possible explanation is that Artemis has a role in DSB repair processes other than nonhomologous end joining. Alternatively, our findings could indicate that the majority of ionizing radiation (IR)–induced DSBs are repaired by nonhomologous end joining in G1 and G2. In support of this, we have observed that DNA ligase IV– and Ku80-deficient MEFs have a similar, major DSB repair defect in G1 and G2 (unpublished data). Previously, we presented evidence that Artemis cells show normal G2/M checkpoint activation assessed by counting mitotic cells up to 9 h after IR (). Subsequently, , using phosphoH3 FACS analysis, concluded that cells treated with Artemis siRNA show premature release from the G2/M checkpoint, implicating Artemis in IR-induced checkpoint responses. To examine the maintenance as well as the activation of G2/M arrest, we counted mitotic cells up to 24 h after IR in cells treated with nocodazole to accumulate cells in mitosis. We confirm that Artemis cells, in contrast to A-T cells, show normal G2/M checkpoint induction and, importantly, remain arrested for the same length and possibly greater than WT cells (Fig. S2 A, available at ). We next analyzed the G2/M checkpoint by phosphoH3 FACS analysis and observed checkpoint activation in Artemis but not A-T cells (). WT cells were released from checkpoint arrest 4–6 h after 1.3 Gy and 12 h after 6 Gy x-irradiation. Artemis cells were released slightly later after 1.3 Gy and failed to be released for at least 16 h after 6 Gy ( and Fig. S2 B). Normal checkpoint induction and a prolonged arrest at the G2/M border was also observed in irradiated Artemis MEFs compared with WT MEFs (). We also evaluated the time course for the progression of G2 cells through mitosis into G1 by analyzing BrdU-labeled cells. Exponentially growing fibroblasts were pulse-labeled with BrdU for 1 h and irradiated with 1 Gy 4 h after labeling (when in G2). G2/M checkpoint arrest results in the retention of BrdU-labeled cells in G2. Quantification of the BrdU-labeled G2 cells for up to 12 h after irradiation confirmed that Artemis cells exhibit a prolonged G2/M arrest (). The prolonged arrest of Artemis cells in was less evident in the experiments involving mitotic counting (Fig. S2 A), which may reflect the use of nocodazole in the latter approach, which delays reentry from G2/M arrest. Our observation of a prolonged arrest in Artemis cells is consistent with a role of Artemis in DSB repair in G1 and G2. One explanation for the difference between our results and those of is that their study used human tumor cells for siRNA knock down of Artemis. Such cells frequently behave aberrantly because of abnormal levels of Chk1/Chk2 or cell cycle checkpoint regulation. Having established that Artemis affects ATM's role in G2 DSB repair but not its function in G2 checkpoint control, we wished to evaluate the contribution of these two ATM functions to the prevention of chromosome aberrations in primary human WT, A-T, and Artemis fibroblasts (). We focused on chromosome breaks arising from G2-irradiated cells by adding aphidicolin to prevent S phase cells from progressing into G2 during analysis. Growing cell populations were irradiated with 1 Gy and analyzed for chromosome breaks per mitotic cell at early times (2 and 4 h) after IR, similar to that undertaken in previous studies (). In all cell lines, chromosome breaks decreased with time reflecting DSB repair (). A-T cells show a pronounced elevation of the number of chromosome breaks per mitotic cell (approximately threefold higher than WT cells), whereas Artemis cells exhibit about twofold more breaks per cell than WT cells, consistent with Artemis's repair function in G2 (). Thus, a combined checkpoint and repair defect is more severe than a repair defect alone. We also evaluated the contribution of repair and checkpoint loss to chromosome aberration formation by using the checkpoint inhibitor SB218078. This drug has been described to impact upon Chk1 activity () and abolishes 53BP1 foci formation after hydroxyurea treatment, a Chk1-dependent phenotype (Fig. S3 A, available at ; ). Addition of SB218078 completely abolished the G2/M checkpoint response in primary human fibroblasts as well as in MEFs (Fig. S3, B and C), whereas repair of IR-induced DSBs in G2 remained unaffected (Fig. S3 D). SB218078 had no effect on chromosome aberration formation in A-T cells but increased the level of chromosome breaks per cell in Artemis cells to that of A-T cells (). WT cells treated with SB218078 showed considerably fewer breaks per cell than drug-treated A-T or Artemis cells, which represents the contribution of ATM/Artemis-dependent DSB repair to the prevention of chromosome aberrations in the absence of checkpoints. It is noteworthy that the cells forming chromosome aberrations are those in G2 at the time of irradiation, as the addition of aphidicolin prevented S phase cells from progressing into G2 during analysis. Thus, any role of Chk1 in replication fork stability will not affect chromosome aberration formation. Moreover, SB218078 did not cause chromosome breaks in the absence of IR. Our studies predict that 1 Gy–irradiated G2 phase Artemis cells would harbor 9–12 DSBs that remain unrepaired for prolonged times. The release of Artemis cells from G2/M checkpoint arrest 6–8 h after irradiation suggested that the G2/M checkpoint might be unable to detect 9–12 DSBs. To investigate whether DSB repair is complete at the point of checkpoint release, we evaluated chromosome aberrations in mitotic cells that arise after checkpoint release (i.e., at time points >4 h after IR; ). Because WT and Artemis cells progress from G2 into G1 within 12 h after IR with 1 Gy (see ), we evaluated chromosome breakage up to this time point. Strikingly, the level of chromosome aberrations in WT and Artemis cells at times when the cells that had initiated the checkpoint leave G2 (4–8 h in WT and 6–10 h in Artemis) is approximately one to two breaks per cell (), which is >10-fold above the background number of chromosome breaks. Thus, almost all cells released from the G2 checkpoint exhibit chromosome aberrations in mitosis. This observation represents direct experimental evidence that the human G2/M checkpoint is not maintained until the completion of repair. This prompted us to investigate the time course for the appearance of chromosome aberrations in mitosis. Cells entering mitosis at early times exhibit more chromosome breaks than cells entering at later times (). However, this analysis fails to consider the number of cells reaching mitosis at each time point. Thus, we assessed the number of cells reaching mitosis under the same conditions used for our chromosomal studies (i.e., in the presence of aphidicolin) by using phosphoH3 FACS analysis () and estimated the total number of mitotic chromosome breaks by multiplying the chromosome breaks per cell by the number of mitotic cells (; see Materials and methods for details of this estimation). Considering this novel concept, we examined the kinetics for mitotic chromosome breakage and observed a maximum at times after the G2/M checkpoint has been released (i.e., at 6–8 h in WT and at 8–10 h in Artemis cells). Thus, cells released from the checkpoint (at ≥6 h after IR) as opposed to cells that escape checkpoint arrest at early times (at ≤4 h after IR) represent a major cause of mitotic chromosome breakage (). We also evaluated the number of cells reaching mitosis from the progression of BrdU-labeled G2 cells (obtained from ). An estimation of the kinetics for mitotic chromosome breakage using this analysis provided similar results to that using the phosphoH3 FACS data (Fig. S3 E). Thus, the concept of evaluating chromosome breakage by considering breaks per mitotic cell as well as the number of mitotic cells reveals the striking finding that checkpoint release before the completion of repair represents a major cause for chromosome aberration formation. Remarkably, the total number of breaks in released cells is similar in WT and Artemis cells, although they arise with delayed kinetics in the repair-defective cells. The decrease in breaks at prolonged times after treatment (>10 h) is due to the depletion of irradiated G2 cells; i.e., nearly all cells have left G2. A-T cells display entirely different kinetics. Because of the lack of checkpoint arrest, they display an elevated number of chromosome breaks that decreases with time in part because of DSB repair and the rapid depletion of the G2 population. Our findings establish that all cells released from the G2 checkpoint harbor unrepaired damage, strongly suggesting that the G2/M checkpoint has a threshold. Our observation that Artemis cells remain checkpoint arrested longer than WT cells () but are released with a similar number of γH2AX foci () or mitotic chromosome breaks () supports this notion. However, we sought other procedures to confirm the presence of DSBs in G2 at the time of checkpoint release and to evaluate the sensitivity limit of the G2 checkpoint. As one approach, we performed premature chromosome condensation (PCC) of G2 cells using the phosphatase inhibitor calyculin A (). G2 cells are readily distinguished from mitotic cells and allow the analysis of PCC breaks (). At 4 and 6 h after 1 Gy x-irradiation, the time at which checkpoint release commences in WT and Artemis cells, respectively, we observed three to four PCC breaks per cell consolidating the presence of DSBs at the time of checkpoint release (). Moreover, WT cells at 4 h and Artemis cells at 6 h harbor a similar number of PCC breaks. Interestingly, these studies also provide an additional demonstration of a repair defect in Artemis cells. Previous studies equating PCC breaks with DSBs estimated by PFGE have reported a 1:3–6 relationship (i.e., 3–6 DSBs equate to 1 PCC break; ). Thus, our PCC data suggest a sensitivity level of 10–20 DSBs. We also used γH2AX foci as a further marker to determine whether DSB repair is complete at the time of checkpoint release. We scored the number of foci in CENP-F–positive G2 phase cells at differing times after IR and, in the same population of cells, counted the number of mitotic cells (). We used exponentially growing transformed human fibroblasts, which provide a high mitotic index (MI). Mitotic cells were scored as phosphoH3-positive cells with condensed chromatin. Consistent with these findings, we observed that checkpoint duration increases with dose and that cells are released from the checkpoint with ∼20 foci (). Similar results were obtained with hTert-immortalized fibroblasts (unpublished data). We also analyzed mitotic cells at the 6-h time point and observed foci numbers similar to those of G2 cells, demonstrating that the cells released from the checkpoint do enter mitosis with foci and that there is no selection for cells exiting the checkpoint (Fig. S3 F; ; ). Previously, we and others have observed a 1:1 relationship between γH2AX foci and DSBs (; ; ). Although it is possible that γH2AX foci analysis could overestimate DSBs remaining if repair is completed before the loss of visible foci, this is unlikely to occur in Artemis-deficient cells, where unrepaired DSBs persist for many days in G1 and G2. Thus, our studies analyzing γH2AX foci are consistent with a threshold of 10–20 DSBs. Additionally, our PFGE studies with G2 (), and previously with G0 cells, show that ∼15% of the induced DSBs remain unrepaired in Artemis cells for many days. PFGE studies estimated 30–40 DSBs induced per Gy in G1 (; ). Because G2 Artemis cells irradiated with 1 Gy are completely released from G2 by 12 h, the estimated persisting damage level (15% of 60–80 DSBs induced: 9–12 DSBs) is unable to maintain the checkpoint. In contrast, after 6 Gy, the level of DSBs remaining exceeds the threshold and results in arrest being maintained for at least 16 h. Hence, our PFGE data, which do not rely on γH2AX foci analysis, also indicate that the G2/M checkpoint threshold is >9–12 DSBs. To evaluate whether induction of the G2/M checkpoint has a similar sensitivity limit, we analyzed transformed and immortalized fibroblasts exposed to doses up to 2 Gy at 2 h after irradiation, the earliest time point at which we observed complete arrest in pilot experiments (). Cells irradiated with 0.6 Gy or higher show complete checkpoint arrest. The foci level 2 h after 0.6 Gy is ∼20. Lower levels cause a partial arrest (). Because repair occurs during the 2-h incubation necessary to measure checkpoint induction, our findings are consistent with a level of ∼20 foci being required to activate checkpoint arrest. We also considered it important to examine primary human cells. The low MI of primary cells necessitated FACS analysis to estimate MI, precluding a parallel evaluation of γH2AX foci formation. Our findings were similar to those obtained using transformed/immortalized cells (). Importantly, use of a lower dose, inducing <10 γH2AX foci did not induce any detectable arrest. Based on 30–40 DSBs induced per Gy in G1, 20 DSBs are induced after doses of 0.25–0.33 Gy in G2 cells. This correlates with the mild checkpoint induction observed here after 0.2–0.3 Gy and the absence of checkpoint arrest after 0.1 Gy (). Thus, these findings are consistent with a similar threshold number of DSBs (10–20) both activating and maintaining checkpoint arrest. The existence of a threshold for G2/M checkpoint arrest provides a potential explanation for low-dose hypersensitivity, a phenomenon describing exquisite cellular sensitivity at low radiation doses (). Indeed, a G2/M threshold of ∼20 DSBs would predict the reported survival responses. Cells were grown as described previously (). 10, 15, or 20% FCS was used depending on the cell line. For the FAR assay, cells were labeled with 37 kBq/ml [methyl-H]thymidine (2.81 TBq/mmol; GE Healthcare) for 1 h (electrophoresis was performed according to ). Aphidicolin and nocodazole (Sigma-Aldrich) were added at 3 μg/ml and 100 ng/ml, respectively. Inhibition of Chk1 activity was achieved by addition of 2.5 μM SB218078 (Calbiochem) 30 min before IR. ATMi (KU55933; provided by G. Smith, KuDos Pharmaceuticals Ltd, Cambridge, UK) was added at 10 μM 30 min before IR. X-irradiation was performed at 90 or 120 kV, γ-irradiation using a Cs-source. Dosimetry was performed with ion chambers and considered the increase in dose for cells grown on glass coverslips relative to plastic surfaces. To collect metaphases, 100 ng/ml colcemid (Sigma-Aldrich) was added 2 h before harvesting (1 h for the 2-h time point and 4 h for the 12-h time point). For PCC analysis, cells were treated with 50 ng/ml calyculin A (Calbiochem) for 30 min before harvesting. Chromatid breaks and excess fragments (counted as two chromatid breaks) were scored in 20–100 chromosome spreads from at least three independent experiments per data point. Cells pulse-labeled with 10 μM BrdU (Roche) for 1 h were analyzed according to standard protocols. For phosphoH3 staining, cells were permeabilized with PBS/0.25% Triton X-100 (15 min on ice), incubated in 100 μl α-phosphoH3 antibody (Ser10; 7.5 μg/ml PBS/1% BSA; Upstate Biotechnology) overnight, and treated with the Alexa Fluor 488–conjugated goat α-mouse (MoBiTec) or an FITC-conjugated swine α-rabbit antibody (DakoCytomation) in PBS/1% BSA for 1 h, followed by 50 μg/ml propidium iodide containing 0.5 mg/ml RNase in PBS for 30 min at room temperature. Analysis was performed on a FACScan or FACSCalibur using the CellQuest software (Becton Dickinson). Cells grown on coverslips were fixed in 100% methanol (−20°C) for 30 min, permeabilized in acetone (−20°C) for 1 min, and washed three times for 10 min in PBS/1% FCS. Samples were incubated with primary antibodies (monoclonal or polyclonal α-γH2AX antibody [1:200; Upstate Biotechnology], polyclonal α–CENP-F and α–cyclin A antibody [1:200; Santa Cruz Biotechnology, Inc.], or polyclonal α-phosphoH3 antibody [Ser10; 1:200; Upstate Biotechnology]) in PBS/1% FCS for 1 h at room temperature, washed in PBS/1% FCS three times for 10 min, and incubated with Alexa Fluor 488–, Alexa Fluor 546–, or Alexa Fluor 594–conjugated secondary antibodies (1:500; MoBiTec) for 1 h at room temperature. Cells were washed in PBS four times for 10 min and mounted using Vectashield mounting medium containing 4,6 diamidino-2-phenylindole (Vector Laboratories). In a single experiment, cell counting was performed until at least 40 cells and 40 foci were registered per sample. Each data point represents two to three independent experiments. Error bars represent the SEM between the different experiments. aims to compare the time course for total mitotic chromosome breakage for three different cell lines: A-T, Artemis, and WT. We have measured for all three lines the MI at defined times after irradiation by phosphoH3 FACS analysis under the same conditions used for the chromosomal analysis, i.e., in the presence of aphidicolin (). However, different cell lines can vary considerably in their fraction of G2 phase cells. Moreover, the majority but not all G2-irradiated cells leave G2 within 12 h with slight differences between the three cell lines (). We have considered the first variation (different G2 proportions) by normalizing the phosphoH3 data in such that the sum of the MIs measured up to 12 h after irradiation is the same for all three cell lines and the second variation by multiplying these MIs with the measured proportion of G2-irradiated cells that leave G2 within 12 h. The latter values are derived from . For example, Artemis cells entering mitosis at 8 h after IR exhibit ∼1.5 breaks per mitotic cell (). At this time, the relative MI for Artemis cells is ∼0.35. Thus, we multiplied the value of 1.5 by 1,000 (to normalize it to 1,000 irradiated G2 cells), by 0.75 (because 75% of all irradiated G2 Artemis cells leave G2 within 12 h; ), and finally by 0.35 (because 35% of all cells that leave G2 within 12 h do this at the 8-h time point). This provides a value of ∼400 mitotic breaks for Artemis cells at 8 h (). Fig. S1 provides additional information for the experimental procedures used to measure DSB repair in G2 and shows that A-T and Artemis MEFs exhibit a DSB repair defect in G1 and G2. Fig. S2 provides additional information that Artemis cells show normal G2/M checkpoint induction and a prolonged arrest by counting mitotic cells and by using phosphoH3 FACS analysis. Fig. S3 shows that the Chk1-inhibiting drug SB218078 abolishes the G2/M checkpoint without affecting IR-induced DSB repair in G2 and provides evidence that cells released from G2/M checkpoint arrest exhibit chromosome breaks and γH2AX foci in mitosis. Online supplemental material is available at .

Centromeric chromatin, in which centromere protein A (CENP-A) replaces histone H3, is organized during mitosis to form a surface for kinetochore assembly (; ; ). Studies of neocentromeres, which spontaneously form at low frequency on acentric chromosome fragments, have suggested that CENP-A nucleosomes rather than specific DNA sequences define centromere activity (). The mechanisms that load CENP-A nucleosomes onto chromatin and maintain their restricted distribution are topics of active investigation (). Work in budding and fission yeasts has identified proteins that contribute to CENP-A localization (; ; ), including the general chromatin-remodeling complex CAF-1. However, because CAF-1 inhibition destabilizes CENP-A protein levels, it is unclear whether CAF-1 has a direct role in CENP-A loading (). To identify proteins that are required to localize CENP-A, we took an unbiased functional genomic approach in the nematode has holocentric chromosomes in which kinetochores form along the entire length of each sister chromatid instead of being confined to a localized chromosomal region. Fundamental similarities in the mitotic kinetochores of holocentric and monocentric chromosomes are indicated by both high resolution ultrastructural studies and conservation of the constituent proteins, from CENP-A on the DNA to components of the spindle microtubule interface (for review see ). In both types of chromosome architectures, CENP-A is restricted to a subset of chromatin organized on opposing faces of sister chromatids to geometrically constrain kinetochore assembly and ensure correct attachment to the spindle (; ). These similarities suggest that the mechanism of CENP-A loading is likely to be conserved between holocentric chromosomes of and monocentric chromosomes of vertebrates. In this study, we provide evidence for this assertion by identifying and characterizing a protein family with a Myb-like DNA-binding domain that is specifically required for CENP-A loading in both nematodes and mammalian cells. Identification of this protein class provides a starting point for understanding the mechanism of this critical step in genome inheritance. In , RNAi-mediated protein depletion is highly penetrant and largely independent of intrinsic protein turnover, making it possible to generate oocytes that are >95% depleted of a target protein (). Fertilization of depleted oocytes triggers the first embryonic mitosis, which is highly stereotypical and, therefore, amenable to quantitative live imaging assays (Video 1, available at ). In CeCENP-A–depleted embryos, kinetochores fail to form, resulting in a distinctive kinetochore-null (KNL) phenotype characterized by clustering of chromosomes from each pronucleus, premature spindle pole separation, defective chromosome alignment, and failure of chromosome segregation ( and Video 2; ). To systematically identify the set of proteins whose depletion results in a KNL phenotype, we used a fluorescence microscopy–based assay to rescreen a collection of ∼250 genes implicated in chromosome segregation by a previous comprehensive genome-wide screen. In the initial screen that targeted 98% of the ∼19,000 predicted genes (), embryos individually depleted of each of the ∼1,700 gene products required for embryonic viability were filmed by differential interference contrast (DIC) microscopy. Cluster analysis of the DIC video data revealed a set of ∼250 genes that are generally required for chromosome segregation (). Because the KNL phenotype is not discernable by DIC, we used high resolution fluorescence time-lapse imaging to analyze living embryos coexpressing GFP-histone H2b and either GFP–γ-tubulin or GFP–α-tubulin that were depleted of each of these 250 gene products (unpublished data). This approach, which is expected to uncover all nonredundant gene products whose inhibition results in a KNL phenotype, identified five proteins: CeCENP-A, CeCENP-C, KNL-1, KNL-2, and KNL-3. Three of these, CeCENP-C, KNL-1, and KNL-3, function downstream of CeCENP-A because they require CeCENP-A to localize to kinetochores, and their depletion does not prevent CeCENP-A targeting (; ; ). The remaining KNL protein, KNL-2 (K06A5.4), specifically targets CeCENP-A to chromatin (see ) and represents the only such protein identified by this comprehensive strategy. The depletion of KNL-2 resulted in a defect in mitotic chromosome segregation that was essentially identical to that of CeCENP-A–depleted embryos ( and Video 3, available at ). Premature separation of spindle poles, indicating the absence of kinetochore-microtubule attachments, was quantitatively similar for the two depletions (). In contrast to CeCENP-A depletions (Video 2; ), KNL-2 depletion also resulted in a meiotic chromosome segregation defect that is evident from the aberrant nature of the oocyte pronucleus (on the embryo anterior/left side in Video 3). The meiotic role of KNL-2 in the segregation of holocentric chromosomes is not discussed further here and will be the subject of a separate study. KNL-2 is an ∼103-kD basic protein with a short coiled-coil stretch and a bipartite nuclear localization sequence. Sequence analysis revealed the presence of a divergent version of a DNA-binding domain at its C terminus, which was first defined in the protooncogene Myb (). Myb domains as well as the related SANT domains are present in a large number of proteins implicated in chromatin dynamics, including transcription factors and subunits of chromatin-remodeling enzymes that interact with histones (; ; ). Homologues of KNL-2 were present in nematodes closely related to , but a clear link to Myb domain proteins in other organisms was not evident in initial bioinformatic analysis (see for more on this topic). To determine the subcellular localization of KNL-2, we generated an affinity-purified antibody and a stable strain expressing a GFP fusion with KNL-2 (Video 4, available at ). Both tools revealed that KNL-2 localizes to the centromere/kinetochore region throughout mitosis (). The specificity of antibody localization was established using RNAi-mediated protein depletion (). Because our screening strategy focused on the first embryonic division, it was possible that the function of KNL-2 was limited to the specialized changes in genome architecture that immediately follow fertilization. However, KNL-2 was observed at kinetochores throughout embryogenesis (), and an RNAi-based strategy in which KNL-2 is not inhibited until after early embryogenesis () indicated a continuous requirement for KNL-2 during development (Fig. S1). The depletion of KNL-2 prevented the localization of CeCENP-A to chromatin ( = 29 first-division embryos; ) but did not affect the stability of CeCENP-A protein (). The chromosomal targeting of KNL-2, in turn, required CeCENP-A (), indicating that these two proteins are interdependent for their localization. In contrast to its effect on CeCENP-A localization, the depletion of KNL-2 did not affect the chromosomal targeting of histone H3 () or GFP-histone H2b (), indicating that KNL-2 is specifically required for loading CeCENP-A. The interdependence of KNL-2 and CeCENP-A localization was further confirmed using the live imaging of embryos expressing GFP–KNL-2 or GFP–CeCENP-A (Videos 4–7, available at ). Reciprocal depletion and localization experiments place CeCENP-A at the top of the kinetochore assembly hierarchy (; ; ). The interdependence of KNL-2 and CeCENP-A for chromosomal targeting () predicted that KNL-2, like CeCENP-A, should be upstream of other kinetochore components. To test this prediction, we examined the recruitment of three widely conserved proteins from distinct positions within the substructure of the mitotic kinetochore: CeCENP-C (inner), KNL-1 (medial), and BUB-1 (outer). As expected, all three proteins failed to localize to chromosomes in KNL-2–depleted embryos (). In reciprocal depletions, GFP–KNL-2 was present on chromosomes in embryos depleted of CeCENP-C, KNL-1, or BUB-1 ( and Videos 8 and 9, available at ), placing KNL-2 with CeCENP-A at the top of the kinetochore assembly hierarchy (). CENP-A containing nucleosomes play an important structural role in the condensation of mitotic chromosomes that is independent of their role in directing kinetochore assembly (). To determine whether KNL-2 is also required for chromosome condensation during mitotic entry, we used a recently developed quantitative assay to compare condensation after the depletion of CeCENP-A, KNL-2, or CeCENP-C. The depletion of KNL-2 resulted in a severe condensation defect that was essentially identical to that observed after the depletion of CeCENP-A and distinct from the more subtle effects of CeCENP-C depletion ( and Fig. S2, available at ). Condensation analysis is performed solely on chromatin in the sperm pronucleus, which is formed before introduction of the inhibitory double-stranded RNA (dsRNA) and is consequently free of complications arising from meiotic defects (). This result further confirms that KNL-2 is specifically required for the assembly of CENP-A chromatin. The aforementioned results suggest that KNL-2 physically associates with CeCENP-A nucleosomes to facilitate their loading and function. To test this idea, we isolated embryonic interphase nuclei from a strain expressing GFP–KNL-2, sonicated the chromatin to 500–1,500-bp fragments, and immunoprecipitated the KNL-2 fusion protein with an anti-GFP antibody (). Analysis of the immunoprecipitates revealed that CeCENP-A but not histone H3 was coenriched with KNL-2 by this procedure (). Complementing this biochemical approach, high resolution imaging revealed the partial colocalization of KNL-2 with dispersed CeCENP-A foci in interphase nuclei (Fig. S3, available at ). CeCENP-A and KNL-2 are the only centromere/kinetochore proteins identified in to date that are present in nuclei throughout the cell cycle. Cumulatively, these results indicate that KNL-2 and CeCENP-A are in close physical proximity on chromatin (and may associate directly), where they coordinately maintain centromere structure, initiate kinetochore assembly, and direct chromosome segregation. Our initial efforts to identify KNL-2 homologues outside of nematodes were unsuccessful because the three nematodes with sequenced genomes are closely related rhabditids. Searches of the recently sequenced genome of the parasitic filarial nematode () revealed a more distant nematode KNL-2 homologue (). By aligning the Myb domain sequence from the filarial KNL-2 with the analogous region of the three rhabditid proteins, we generated a profile that allowed us to identify a KNL-2–related subfamily of the Myb/SANT domain–containing protein superfamily with members in all sequenced vertebrates (). The human homologue of KNL-2 was independently discovered in a recent study based on its physical association with Mis18 (). Mis18 was first found in fission yeast, and both fungal Mis18 and its human homologue are implicated in CENP-A loading (; ). No homologue of Mis18 is evident in nematodes, and no KNL-2 homologue is recognizable in fungi. To determine whether the human homologue of KNL-2 (which we refer to as HsKNL2) is required for CENP-A loading, we characterized its localization using an affinity-purified antibody generated against amino acids 661–899 and performed siRNA-mediated depletion. HsKNL2 localizes to the centromere region of chromosomes prominently between late anaphase/telophase and early G1 phase (), which coincides with the time of CENP-A loading in HeLa cells (see Jansen et al. on p. of this issue). The reason for the difference in the localization timing of KNL2 between nematode embryos and mammalian cells in culture is currently unclear. Most importantly, the depletion of HsKNL2 reduced centromeric levels of CENP-A as well as levels of the core kinetochore components hDsn1 (a subunit of the human Mis12 complex; ) and Hec1 (the Ndc80-like subunit of the Ndc80 complex; ; ). In contrast, phosphohistone H3 (S10) staining was unaffected (). HsKNL2 depletion did not affect CENP-A protein levels (not depicted), which is in agreement with the findings in embryos (). HsKNL2-depleted cells exhibited severe chromosome segregation defects, which is consistent with their dramatic kinetochore assembly defect (Fig. S4 and Video 10, available at ). These results establish that involvement of the KNL2 protein family in CENP-A chromatin assembly is conserved between the holocentric chromosomes of nematodes and the monocentric chromosomes of vertebrates. In summary, by using a functional genomic approach in based on the signature CENP-A loss of the function phenotype during the first embryonic division, we identified a conserved Myb domain–containing protein family that is critical for CENP-A loading. The depletion of KNL-2 results in all of the defects expected for a CENP-A–loading factor, and, after extensive functional genomics analysis, KNL-2 is the only protein identified to date in that meets this essential criterion. The fact that KNL2 contains a Myb domain raises the exciting possibility that recognition of short, specific DNA sequences may play a role in CENP-A deposition. Although studies of neocentromeres have suggested that centromere identity is not strictly defined by DNA sequence (; ), it remains possible that short sequence stretches bias centromeric chromatin assembly. Further investigation of how the Myb domain of the KNL2 protein family contributes to CENP-A loading will help elucidate the mechanism by which this widely conserved protein class directs centromeric chromatin formation. RNAi was performed by injection of dsRNA against the target gene as described previously (). Embryos were analyzed from injected adults that were kept for 48 h at 20°C after injection. For KNL-2, oligonucleotides AATTAACCCTCACTAAAGGTCGACTTGGTCGGACAGATT and TAATACGACTCACTATAGGTGCGATATGTGGCGTTATGT were used to create an ∼1,000-bp dsRNA by standard methods; all other dsRNAs were previously described (). siRNA treatment was performed as described previously () using a cocktail of oligonucleotides purchased from Dharmacon. Polyclonal antibodies were generated to amino acids 2–153 of KNL-2 and amino acids 661–899 of HsKNL2 fused to GST and were affinity purified. Immunofluorescence was performed as described previously using directly labeled polyclonal antibodies (). All immunofluorescence images were acquired using a DeltaVision-modified inverted microscope (IX70; Olympus) and Softworx software (Applied Precision) and were deconvolved. Quantification of immunofluorescence was performed in MetaMorph software (Molecular Devices) on images acquired using fixed exposure conditions. HeLa cells were cultured in DME as previously described (). was grown using standard conditions (). All live images were acquired via a spinning disk confocal microscope (CSU10; McBain Instruments) mounted on an inverted microscope (TE2000e; Nikon). Strain TH32 coexpressing GFP-histone H2b and GFP–γ-tubulin was imaged using a 60× 1.4 NA plan Apo objective with 1.5× auxiliary magnification and a cooled CCD camera (Orca ER; Hamamatsu) binning 2 × 2. Live imaging of HeLa cells was performed using similar imaging conditions at 37°C. Strain OD31 expressing GFP–KNL-2 was imaged in the same manner without the 1.5× auxiliary magnification. To construct strain OD31, oligonucleotides CGCTTCCACTAGTGGTGATACGGAAATTGTTCCTC and CGCTTCCACTAGTTTAGTAGATGGATGTGTCTTCTTCA were used to PCR KNL-2 from cDNA (provided by Y. Kohara, National Institute of Genetics, Shizuoka, Japan), and the resulting fragment was digested with SpeI and cloned into pIC26 to create pPM3. pPM3 was biolistically transformed into to generate a stable integrated strain. The HCP-3–GFP strain OD101 expressed HCP-3 with an internal GFP fused between the N-terminal tail and the histone core of HCP-3. The DNA construct for HCP-3–GFP bombardment was made by digesting a GFP PCR product (forward oligonucleotide GCGCGGAGCTCCATGAGTAAAGGAGAAGAACTTTTCAC and reverse oligonucleotide CGCGCGAGCTCGCTTTGTATAGTTCATCCATGCCAT) with Sac-I and inserting at a Sac-I restriction site within the HCP-3 coding region at residue 173. The HCP-3–GFP construct was then inserted at the Bam-HI site of pAZ132 and biolistically transformed into to generate a stable, integrated strain. The YFP–CENP-A clonal cell line was a gift from D. Foltz (Ludwig Institute for Cancer Research, San Diego, CA; ). Nuclei were isolated from adult OD31 . In brief, synchronous liquid cultures of OD31 were grown until 5–10 embryos were present per adult. Embryos were isolated by bleaching, and blastomeres were dissociated by treatment with chitinase (0.5-μg/ml final concentration; 30 min at RT; Sigma-Aldrich). Nuclei were isolated from blastomeres by dounce homogenization in a low ionic strength buffer (0.35 M sucrose, 15 mM Hepes-KOH, pH 7.6, 0.5 mM EGTA, 5 mM MgCl, 10 mM KCl, 0.1 mM EDTA, and protease inhibitor cocktail) with digitonin (0.125% final concentration; Sigma-Aldrich). Nuclei were washed in lysis buffer (50 mM Hepes, pH 7.4, 5 mM EGTA, 1 mM MgCl, 300 mM KCl, 10% glycerol, 0.1% Triton X-100, 0.05% NP-40, and protease inhibitors) and sonicated to create chromatin fragments. The sonicated mixture was clarified, and the supernatant was used for immunoprecipitations as described previously (). Western blots were performed using standard methods. Quantification of mitotic chromosome condensation was performed as described previously () using strain TH32 and the imaging conditions described in the Live imaging and GFP fusions section. Fig. S1 shows that KNL-2 function is not restricted to the first embryonic division of . Fig. S2 shows that KNL-2 makes an equivalent contribution to CeCENP-A in condensation and partially colocalizes with CeCENP-A in interphase nuclei of embryos. Fig. S3 shows that HsKNL2 is required for chromosome segregation in human cells. Video 1 shows chromosome segregation in a control embryo coexpressing GFP-histone H2b and GFP–γ-tubulin. Videos 2 and 3 show chromosome segregation in CeCENP-A– (Video 2) and KNL-2–depleted (Video 3) embryos coexpressing GFP-histone H2b and GFP–γ-tubulin. Video 4 shows GFP–KNL-2 in a control embryo, and Video 5 shows GFP–KNL-2 in a CeCENP-A–depleted embryo. Video 6 shows GFP–CeCENP-A in a control embryo, and Video 7 shows GFP–CeCENP-A in a KNL-2–depleted embryo. Videos 8 and 9 show GFP–KNL-2 in CeCENP-C– (Video 8) and KNL-1–depleted (Video 9) embryos. Video 10 shows DIC and YFP–CENP-A time lapse of control (top) and HsKNL2-depleted (bottom) mitotic HeLa cells. Online supplemental material is available at .

italic xref #text We set out to evaluate the utility of for in vitro cell biological and biochemical investigations. eggs (∼0.6 mm diam) are approximately one fifth the volume of those of (1.2 mm diam). To test whether eggs could be used to prepare functional cellular extracts, we collected, dejellied, and crushed unfertilized eggs, which, like those of , are arrested in metaphase of meiosis II by cytostatic factor (CSF) activity (). Metaphase-arrested egg extracts assembled spindle structures around exogenously added sperm nuclei, entered interphase, and replicated DNA when released from the arrest, and then cycled back into mitosis (). Although yields of extract per frog were 10–20% that of egg extracts effectively recapitulated cell cycle events in vitro. The utility of extracts would be maximized if reagents generated for could be applied. Fluorescence microscopy revealed that antibodies against histone H1 (chromatin component; ), nuclear mitotic apparatus protein (NuMA; spindle pole component; ), and kinesin-like DNA-binding protein (Xkid; ) gave identical staining patterns in extracts, compared with ( and not depicted). Furthermore, the addition of an inhibitory antibody to Xkid resulted in chromosome congression defects in spindle reactions that were very similar to those observed in (unpublished data), suggesting that many reagents will be useful in both species. Because has a pseudotetraploid genome, many genes are present in multiple copies, and without selective pressure, some may be expressed but may not be functional, like one of the Vg1 isoforms (). Whether or not the isoforms are functional, there is frequently more than one. Rae1 (mRNA export factor/spindle regulator; ), RCC1 (guanine exchange factor for Ran; ), and histone H1 were represented by multiple bands on a Western blot, whereas contained a single band for each protein (). This suggests that the pseudotetraploid genotype of contributes to the complexity of the proteome, and the use of could simplify protein analysis. biochemistry is not underpinned by genomic information, making identification of proteins by mass spectrometry difficult. To test whether proteins could be identified using the database, we immunoprecipitated the microtubule-associated developmental regulator nuclear factor 7 (Xnf7; ) from both and extracts (), and then used MALDI mass spectrometry to identify each protein using databases from both species. Xnf7 was identified from both immunoprecipitates using either database (), although the number of peptides identified was higher when queried against the database of the same species. Conceptual trypsin digestion of Xnf7 from both species and comparison of the peptides revealed that although the two proteins are highly conserved (), only 45% of the peptides have identical masses (not depicted). This analysis suggests that although the database will greatly facilitate the identification of proteins by mass spectrometry, it will be more efficient to identify homologous proteins. While validating the use of extracts, we noticed that spindles assembled around sperm nuclei in extracts were considerably smaller (∼30% shorter) than those assembled in extract (). Comparing the fluorescence area of the two types of spindles revealed an approximately threefold difference (unpublished data), indicating a substantially greater microtubule mass in spindles compared with those of . This prompted us to examine the poorly understood phenomenon of spindle scaling further. Spindle sizes were extract dependent, suggesting that they may be defined by diffusible cytoplasmic components. To determine whether these factors reflect a balance of activities present in both extracts, or a dominant activity in one of the extracts, we combined extracts in different proportions and examined spindle length in the mixed reactions. We found that spindle length had a direct and linear dependence on the proportion of to extract (), suggesting equilibrium behavior of cytoplasmic regulatory activities. Previous work has also demonstrated a role for chromatin mass in determining spindle size (). To investigate this phenomenon, we compared spindle length in , , and mixed extracts using sperm from , whose diploid genome (1.7 × 10 bp) is ∼55% that of (3 × 10 bp; ), and found that spindles assembled around chromosomes were ∼10% shorter in all cases (). Therefore, we conclude that although chromatin mass does influence spindle length, soluble cytoplasmic factors are the major determinant in egg extracts. To examine whether spindle size regulation is a static or dynamic process, we added fresh extract to preassembled spindles. To extracts containing spindles that had incorporated X-rhodamine–labeled tubulin, we added three volumes of either or extract containing Alexa Fluor 488 tubulin, and examined spindle length at various time points after mixing. Whereas extract did not affect spindle length over the course of the experiment, addition of extract caused rapid growth of spindles, by ∼5 μm in length within 2 min, and to the size of premixed (75% , 25% ) reactions within 5 min (). Reciprocally, the addition of extract to preformed spindles rapidly shrank them to the size of the premixed controls, whereas addition of extract did not (). The added extract tubulin appeared to incorporate at the plus ends of growing microtubules in the central spindle (). These results demonstrate that soluble cytoplasmic factors dynamically govern spindle length in extracts, in agreement with results obtained in cells (), and indicate that nonmicrotubule spindle matrix elements determining length, if they exist, cannot be purely static structures (). We found that chromatin bead spindles and asters induced by addition of DMSO or RanGTP were smaller in extracts compared with extracts (unpublished data). To determine if differences in global microtubule dynamics could account for the differences in spindle and aster size, we measured the parameters of microtubules nucleated from centrosomes in and extracts (). Overall, although the microtubule growth rate was ∼20% slower in the extracts (14.7 μm/min vs. 18.5 μm/min in ; P < 0.015 in test), catastrophe and rescue frequencies were similar, and the calculated steady-state microtubule lengths were not significantly different (; ). Another mode of microtubule turnover in the spindle is poleward microtubule flux, which is when microtubules coordinately polymerize at their plus ends and depolymerize at their minus ends as antiparallel microtubules slide apart (). We measured these rates using speckle microscopy to mark the spindle microtubule lattice, and found values in both extracts similar to those previously described for (; 1.79 ± 0.33 μm/min for ; 2.25 ± 0. 25 μm/min for ; ). Thus, our results indicate that the coordinated microtubule sliding with balanced plus end polymerization and minus end depolymerization are not significantly different in the two extracts. What is the underlying cause of spindle size difference in the two extracts? One possibility is that we are unable to make precise enough measurements to distinguish potentially causal differences in the dynamics of microtubules in the two extracts. Alternatively, other microtubule dynamic parameters, such as the frequency of nucleation, pausing, or severing, may generate differences in spindle size (). Although different in morphology, centrosomal microtubules in the two extracts grew to similar lengths, whereas microtubule structures induced by taxol, DMSO, and RanGTP were significantly smaller in extracts (unpublished data). An intriguing possibility is that extract factors in the two systems respond differently to microtubule stabilizing/destabilizing agents, including mitotic chromatin. We think that this is possible because we have identified extract factors, such as the microtubule plus-end binding protein Xorbit, whose depletion does not obviously affect centrosomally nucleated microtubules, but strongly destabilizes spindle microtubules (). Extract-dependent changes in the activity of spindle factors or their regulation caused by a different kinase/phosphatase balance or RanGTP signal surrounding chromosomes could locally alter microtubule stability and overall spindle size. The challenge will now be to identify molecules that can account for the observed differences in spindle size in the egg extracts and compare the activities of the orthologous factors between the two species. In conclusion, provides molecular advantages over as a genetically and proteomically tractable system that can be applied to address cell biological questions using in vitro approaches. Although it could be expected that the smaller eggs would have smaller spindles, our results show that this is because of a difference in cytoplasmic factors, rather than the size of the cell itself. By comparing cytoplasmic activities that are intrinsic to and extracts, new insights can be gained into mechanisms regulating cellular morphogenesis. CSF-arrested extracts were prepared essentially as previously described for (; ), with the following exceptions. To induce egg laying, frogs were primed with 10 U of human chorionic gonadotropin (HCG) ∼16 h before a hormone boost of 200 U HCG. Laying commenced ∼4–5 h after the second HCG injection, and eggs were collected into water at 27–28°C (). Frogs were also squeezed ∼6 h after the second HCG injection, and we found no substantial difference in the quality of the laid versus squeezed eggs. Pooled eggs were dejellied using 3% cysteine in water adjusted to pH 7.8 with NaOH. Incubation of frogs at temperatures below ∼23°C inhibited egg laying, and resulted in poor extracts. Once the extract was prepared it was stored at 16°C, as extended storage at 4°C resulted in a loss of CSF arrest. Typical extract yield was 200–400 μl per frog. and extracts were mixed in different proportions and supplemented with either or sperm nuclei prepared by standard procedures () at a concentration of 500 sperm/μl, and X-rhodamine tubulin at 0.125 mg/ml. Cycling reactions were performed, and reactions were diluted into spindle fix (30% glycerol, 1× BRB80, and 0.5% Triton X-100), spun onto coverslips, fixed in –20°C methanol, and mounted in Vectashield according to standard procedures (). Images were collected with a fluorescence microscope (BX51; Olympus) and a cooled charge-coupled device camera (Orca II; Hamamatsu), and spindle lengths were measured using MetaMorph software (Molecular Devices). Spindle area measurements were made using thresholded images in MetaMorph. Mixing experiments were performed three independent times by three different investigators, and the results were averaged. The Xnf7 polyclonal antibody was coupled to protein A dynabeads, as previously described (). For immunoprecipitating Xnf7, 110 μl of either or CSF extract was subjected to three successive 45-min incubations on ice with anti-Xnf7–coated dynabeads. The beads from each round were pooled and washed extensively with XB before eluting for 5 min at room temperature into SDS sample buffer and retrieving the beads on a magnet. Half of the supernatant was subjected to SDS-PAGE, the gel was stained with Gradipure colloidal G-250 Coomassie blue stain (NuSep), and the indicated bands were excised and subjected to mass spectrometry at the University of California Berkeley Mass Spectrometry Facility. CSF reactions containing sperm nuclei and 25 μl extract (with X-rhodamine tubulin) or extract (with Alexa Fluor 488 tubulin) were cycled through interphase and back into metaphase by the addition of 25 μl of the same type of extract (no sperm). Once metaphase spindles had assembled, the extract was split into two tubes, each containing 25 μl of the reaction mixture. As a control, 75 μl of the same type of extract supplemented with the other labeled tubulin was added to one of the tubes, while 75 μl of the opposing extract was added to the other 25-μl reaction. Each of the 100-μl reactions were quickly split into 4 separate tubes, and spindles from each condition were diluted and spun down onto coverslips as described in Spindle size determination in mixed extract for imaging and length measurements. All spindle reactions were incubated at ∼23°C. Centrosomes were prepared from KE37 cells as previously described () and stored at −80°C. CSF extracts were prepared as described in the section Preparation of CSF extracts from , and supplemented with either rhodamine tubulin (Cytoskeleton) or Alexa Fluor 488 tubulin (a gift from T. Whittman, University of California, San Francisco, San Francisco, CA) at 0.2 mg/ml. Centrosome reactions consisted of 8 μl of extract plus labeled tubulin, 1 μl of centrosomes, and 1 μl of Oxyrase. extracts were incubated at ∼20°C, and extracts were incubated at ∼23°C. To image centrosomes, 1 μl of extract was squashed under a 22-mm coverslip and imaged using a 100×/1.3 NA objective. All glassware was base cleaned and stored in 95% ethanol until the time of use. Images were acquired every 3 s for 1–3 min. Microtubule lengths were measured if the microtubule could be followed for at least five successive frames. Microtubule lengths were measured as the distance from the center of the centrosome to the tip of the microtubule. Dynamics measurements were calculated using a custom-made spreadsheet (a gift from R. Tournebize, Institute Pasteur, Paris, France). Calculation of F and F were made by manual inspection of raw growth and shrinkage measurements. MT flux measurements were made using fluorescence speckle microscopy by incubation of cycled spindles with rhodamine-labeled tubulin at a concentration of 1 μg/ml (). 2 μl of extract was gently squashed under a 22 × 22-mm coverslip previously outlined using a PAP pen. Images of speckles were collected every 5 s for 2 min using a 60×/1.4 NA objective. Speckle movements were tracked on kymographs to calculate the rate of flux. Measurements were made from at least five separate spindles from three different extracts for both and .

The budding yeast undergoes polarized growth and provides an excellent model system for the study of cell polarity. Yeast cells are surrounded by a rigid cell wall, which protects the cells from the environment but also physically restrains membrane expansion. For budding, cells not only need to deliver proteins and lipids to the bud membrane but also need to secrete enzymes to remodel the cell wall in order to allow surface expansion. discovered that yeast cells have two distinct classes of exocytic vesicles carrying different sets of cargoes for secretion. One class carries plasma membrane proteins and cell wall modification enzymes such as Bgl2p (hereafter referred to as Bgl2p vesicles), whereas the other class carries proteins such as the periplasmic enzyme invertase (hereafter referred to as invertase vesicles). Several proteins are implicated in the generation of specific vesicles (; ). However, whether the tethering or docking of these vesicles at the plasma membrane is differently regulated is unknown. The exocyst is an octameric protein complex composed of Sec3p, Sec5p, Sec6p, Sec8p, Sec10p, Sec15p, Exo70p, and Exo84p. These proteins are localized to the bud tip, the site of active exocytosis and cell surface expansion, where they tether the post-Golgi secretory vesicles to the plasma membrane before fusion (for reviews see ; ). In the present study, we report that Exo70p functions primarily at early stages of the cell cycle to regulate the secretion of specific vesicles that are critical for polarized cell growth. Exo70p was initially identified through purification of the exocyst complex from yeast lysates (). To study the function of Exo70p, we generated mutant strains. The mutant has a growth defect that is more pronounced at temperatures ≤25°C, whereas is a temperature-sensitive mutant that grows normally at 25°C but fails to grow above 35°C (Fig. S1, available at ). To study secretion in these mutants, we examined the cells that were shifted to their restrictive temperatures by thin-section EM. As shown in , the and cells accumulated post-Golgi vesicles that were 80–100 nm in diameter. These vesicles were preferentially distributed in the daughter cells. Consistent with the EM result, immunofluorescence staining of Sec4p, the Rab protein residing on the post-Golgi vesicles, was polarized to the bud in both mutants (). These results indicate that the polarized delivery of exocytic vesicles is not affected in the mutants, which is consistent with previous studies indicating that the exocyst functions at the vesicle-tethering step after vesicles are transported to the daughter cells (for reviews see ; ). All of the previously identified exocyst mutants were defective in invertase secretion (; ). Therefore, we examined invertase secretion in the mutants. Surprisingly, the and mutants were able to secrete invertase at levels that were close to the wild-type cells (). As controls, all of the other exocyst mutants showed substantial invertase secretion defects at the restrictive temperature (). We next examined the secretion of Bgl2p in these mutants. As shown in , accumulated Bgl2p at both 25 and 34°C, and accumulated Bgl2p at the restrictive temperature of 37°C. The newly accumulated Bgl2p in after shifting to 37°C was comparable with other temperature-sensitive exocyst mutants. The growth defect of is also similar to other exocyst mutants (Fig. S1 B). To better analyze the secretion of these cargoes, we performed pulse-chase experiments. As shown in , the wild-type cells secreted >90% of the newly synthesized Bgl2p within 30 min at 37°C, whereas only secreted ∼40% of Bgl2p. also showed the Bgl2p secretion defect, which was less severe than that in . On the other hand, secreted >80% of the newly synthesized invertase within 30 min at 37°C, whereas only secreted ∼30% of invertase under the same condition (). To further investigate the selectivity of vesicle block in mutants, we separated the vesicles by density gradients as previously described (; ). As shown in , both and accumulated Bgl2p vesicles, whereas only a diminutive amount of invertase vesicles was detected in . Our results confirmed an early prediction (; ) that certain secretory mutants may be specifically defective in one branch of the exocytic pathways. To date, all of the other mutants have been characterized by their block of both classes of vesicles (; ; unpublished data). Thus, the mutants are unique in their selectivity in Bgl2p vesicle block. We do not totally exclude the possibility that new mutant alleles that have more detectable invertase secretion defects can be identified. However, analyses of the two mutant alleles that are different in nature (Fig. S1) strongly suggest that Exo70p primarily mediates the secretion of Bgl2p vesicles. Cargoes in the invertase vesicles travel through endosomal compartments before reaching the plasma membrane. Vps1p is a dynamin implicated in the formation of vesicles that are destined to the endosomes from the TGN (). Pep12p is a SNARE protein involved in the fusion of Golgi or early endosomal vesicles with the late endosomes (). In the or cells, invertase is delivered to the cell surface by routing through the Bgl2p pathway (; ). We found that was synthetic lethal with or (). This result suggests that in the cells that are constitutively defective in Bgl2p vesicle secretion, further disruption of the other exocytic route is detrimental to the cell. We have also analyzed carboxypeptidase Y (CPY) processing and found that this protein is correctly sorted, and there is no kinetic delay for CPY traffic in the mutant (Fig. S2 A, available at ). The normal processing of CPY together with the evident accumulation of post-Golgi secretory vesicles suggest that protein sorting to vacuole and vesicle formation at the TGN is not defective in mutants. Rather, the selective block of Bgl2p vesicle exocytosis is likely imposed at the plasma membrane stage, which suggests that different exocytic pathways can be specifically regulated at the target membrane as well as at the TGN. Using fluorescence microscopy, we examined exocyst localization in the mutants. As shown in , all of the exocyst subunits were polarized, including the mutant protein itself. We also found that the immunoisolated exocyst complex was intact, and the protein was able to associate with other exocyst proteins in the complex (). Similar results were obtained from the cells (Fig. S2 B). Because the localization and composition of the exocyst complex are mostly unaffected in the mutants, the secretion defects are most likely caused by the malfunction of Exo70p itself. During our EM analyses of the cells, we noted that vesicle accumulation was much more prominent in small-budded cells than in large-budded cells. The correlation between bud size and vesicle number suggests that the secretion block in mutants mostly occurs during early stages of budding. In contrast, the other exocyst mutants accumulate secretory vesicles at various stages of the cell cycle (Fig. S3, available at ). Failure in secretion results in an ∼5% increase of cell density, which allows separation of the mutant cells and wild-type cells by density gradients (). In the present study, wild-type and mutant cells were subjected to Percoll density gradients. As shown in , the wild-type cells were distributed to a single low density fraction, and the cells were accumulated at a single high density fraction. However, the cells were distributed in both the low and high density fractions. We found that most of the large-budded cells (93%; = 250) were present in the lighter fraction. In contrast, the denser fraction contained very few large-budded cells; instead, most were small- or medium-budded cells corresponding to the vesicle-accumulating population observed by EM. The cells with lower density were not caused by incomplete phenotypic penetration or spontaneous reversions because cells collected from the two density peaks were equally defective in growth on plates (unpublished data). Besides and , we have also analyzed and other exocyst mutants, including , , and . Although the cells were distributed to both heavy and light fractions, the other exocyst mutants were all distributed to a single high density fraction (, top). Further microscopic analysis of the cells obtained from the heavy fraction demonstrated that most of these cells were in their small- or medium-budded stage, whereas cells obtained from the other mutants were heterogeneous in budding stages (, bottom). This study clearly indicates that mutants block secretion primarily at early stages of the cell cycle. It has been proposed that the two types of secretory vesicles play distinct roles in yeast cells (; ). The invertase vesicles carry proteins such as invertase, acid phosphatase, and possibly the general amino acid permease (), which are secreted under certain physiological conditions. The Bgl2p vesicles contain materials for plasma membrane expansion and cell wall remodeling. We examined whether block of the Bgl2p vesicle pathway in mutants could affect daughter cell growth. Budding of the wild-type and cells was compared at various time points after release from α-factor arrest at G. As shown in , by 60 min, the wild-type cells exhibited much larger bud sizes than the mutant cells (1.5-fold; = 150). Clearly, block of the Bgl2p vesicle alone is sufficient to cause a delay in daughter cell growth. Furthermore, because secretion in the mutants is primarily blocked at early stages of the cell cycle, we speculated that growth defects of the mutant cells would be more prominent during early budding. To test this, we examined the growth of the temperature-sensitive mutant synchronized at different cell cycle stages. We first assessed the growth of cells at the early budding stage by examining bud formation from cells released from G at 37°C. As shown in , the majority of the wild-type cells were capable of generating buds after 60 min. In contrast, most of the cells were either unbudded or at the tiny-budded stage. We then tested whether growth of the cells at later stages of the cell cycle was affected in . Cells released from G were first incubated at 25°C for 90 min to generate large-budded cells and were shifted to 37°C for various periods of time. We found that both the wild-type and cells were able to complete cytokinesis after 30 min of incubation at 37°C, as revealed by the disconnection of the cytoplasm between the mother and daughter cells. However, after cytokinesis, the cells were arrested at the next budding stage, whereas the wild-type cells were able to proceed to new rounds of the cell cycle (). On the other hand, the cells were defective in both budding and later stages of cell growth (). The selective block of Bgl2p vesicles primarily during budding in mutants suggests that the function of Exo70p in exocytosis is spatially and temporally coupled to polarized cell growth. The exocyst complex is under the control of Rho GTPases (). It was previously reported that a Cdc42p mutant, , specifically accumulated Bgl2p vesicles at early stages of the cell cycle (). The phenotypical similarity between and the mutants suggests that Cdc42p and Exo70p function together during polarized cell growth. One possibility is that Cdc42p activates Exo70p during yeast budding. We have examined the interaction between Exo70p and Cdc42p in vitro using recombinant fusion proteins. However, the interaction was very weak, and there was no clear nucleotide dependence in the binding (unpublished data). Nevertheless, Cdc42p may regulate Exo70p function via intermediate molecules. Exo70p functions in concert with the other exocyst components, which mediate the secretion of both the invertase and Bgl2p vesicles. Future studies may likely identify specific mutant alleles in those subunits that also preferentially affect Bgl2p secretion. Overall, characterization of the mutants revealed the intimate coupling between exocytosis and polarized cell growth. Future studies of the mutants will help us understand the nature of the Bgl2p vesicle pathway and how this pathway is linked to polarized cell growth and cell cycle progression. Standard methods were used for yeast media, growth, and genetic manipulations. The yeast strains used in this study are listed in Table S1 (available at ). The mutant was isolated by random mutagenesis. For generation of the mutant, site-directed mutagenesis on the C terminus of Exo70 was performed on a plasmid. The mutant alleles were introduced into , in which the endogenous gene was disrupted by and supplemented with a plasmid. The transformants were replicated onto synthetic complete media (SC) plates containing 5-fluoroorotic acid (5-FOA) to select for the loss of the plasmid. To generate and double mutants, and strains in which the endogenous was disrupted by and supplemented with were transformed with a or plasmid. The transformants were replicated on SC plates containing 5-FOA to select for the loss of the plasmid. Wild-type and cells were grown at 34°C in yeast extract/peptone/glucose (YPD) medium to early log phase and were shifted to 25°C for 3 h. The and cells were grown at 25°C and shifted to 37°C for 2 h. Cells were processed for thin-section transmission EM using a transmission electron microscope (model 1010; JEOL) at 100,000× as previously described (). Fluorescence microscopy of the GFP-tagged exocyst and immunofluorescence observation of Sec4p were performed as previously described (). The images were captured by a fluorescence microscope (DM IRB; Leica) using a 100× objective and a high resolution CCD camera (ORCA-ER; Hamamatsu) and were analyzed by OpenLab 3.1.4 software (Improvision). Invertase assays were performed as described previously (). For and other exocyst mutants, the cells were first shifted to 37°C for 1 h and then were grown in low glucose medium (0.1% glucose) at the same temperature for 1 h. For , cells were started at 34°C and were transferred to 25°C for 1 h before shifting to low glucose medium for 1 h and processing for the assay. Bgl2p assays were performed as described previously (). For and other exocyst mutants, the cells were shifted to 37°C for 2 h. For , cells were started at 34°C and were transferred to 25°C for 2 h before being processed for the Bgl2p assay. To analyze the secretion of newly synthesized Bgl2p, cells were cultured to early log phase at 25°C and were shifted to SC lacking methionine/cysteine at 37°C for 15 min. The cells were then labeled with [S]methionine/cysteine for 10 min and chased for various times at 37°C. Bgl2p from external and internal fractions were immunoprecipitated using anti-Bgl2p polyclonal antibody and were analyzed by SDS-PAGE and autoradiography. Accordingly, to analyze the secretion of newly synthesized invertase, the wild-type, , and cells were cultured to early log phase at 25°C and shifted to 37°C for 15 min. Cells were then shifted to low glucose medium for 10 min to derepress the invertase expression (pulse). The chase was initiated by adding 2% glucose to the culture to repress invertase expression. The secretion of newly synthesized invertase at various chase times was determined by invertase enzymatic assay. Vesicle fractionation was performed as previously described (). Yeast cells (100-ml culture per gradient) were grown in YPD medium to early log phase and shifted to YP plus 0.1% glucose at the restrictive temperatures to induce invertase expression. Cells were spheroplasted, lysed, and subjected to differential centrifugation. The pellets from the 100,000- spin were fractionated on 20–55% Percoll step gradients (4 ml in TLA100.3 tubes [Beckman Coulter]), and 160-μl fractions were collected for invertase and Bgl2p assays. The wild-type and cells were grown at 34°C to early log phase and were shifted to 18°C for 3 h. The cells were washed and resuspended in 0.2 ml of synthetic minimal media. The cells were then layered on a 70% Percoll/synthetic minimal medium mixture and centrifuged for 20 min at 25,000 in TLA100.3 tubes. For and other exocyst mutants, the cells were shifted for 2 h at 37°C before centrifugation. For synchronization of cells, early log-phase cells were arrested at G phase using 5 μM α factor for 2.5 h at 34°C. The cells were then released into fresh YPD at 25°C for various periods of time and were collected for light microscopy. To examine the growth of cells at different cell cycle stages, cells were treated with α factor for 2.5 h at 25°C and either directly released into fresh medium at 37°C or first released at 25°C for 90 min to generate large buds followed by a shift to 37°C for various times. Cells were then collected for light microscopy observation. The completion of cytokinesis was determined by staining with 1 mg/ml DAPI. Table S1 lists the major yeast strains used in this study. Fig. S1 shows the growth properties of the mutants and the mutation sites on the protein. Fig. S2 A shows CPY processing in by pulse-chase assay. Fig. S2 B is the immunoprecipitation experiment using radiolabeled yeast lysate showing that the isolated exocyst complex is mostly intact in the mutant. Fig. S3 displays the EM images of representative , , and cells at their small-budded and large-budded stages. Online supplemental material is available at .

Epithelial cells play fundamental roles in separating compositionally different compartments to regulate homeostasis and maintain physiological functions in multicellular organisms. These functions are established by organized junctional complexes, cytoskeletal architecture, and highly polarized membrane domains (). During epithelial cell polarization, E-cadherin– and nectin-mediated cell–cell contacts induce the formation of primordial “spot-like” adherens junction (AJ) complexes (). Through interaction between actin filaments and components of primordial AJs, these junctions are gradually fused side by side and finally become “belt-like” AJs (; ). In parallel with this event, tight junctions (TJs) are formed at the apical side of AJs. However, how belt-like AJs and TJs are evolved from primordial AJs and sorted during the polarization process of epithelial cells remains mostly to be clarified. The molecular architecture of AJs and TJs has been unraveled rapidly in recent years (; ; ; ; ). Among them, ZO-1 and Par-3–Par-6–aPKC are unique in that they localize at primordial AJs in the initial phase of epithelial polarization (; ), but they eventually localize at TJs and not at belt-like AJs after the maturation of epithelial polarization (; ). Par-3–Par-6–aPKC protein complexes are known to be required for the formation of belt-like AJ and TJ formation in epithelial polarization (); however, our knowledge about the functional roles of ZO-1 in cell polarization is limited. ZO-1/ZO-2/ZO-3 is a membrane-associated guanlyate kinase (MAGUK) protein composed of the following domains: three PDZ (PSD95/Dlg/ZO-1) domains, an SH3 domain, a GK domain, an acidic domain, and an actin binding region (). The PDZ1 domain binds to claudins. ZONAB is localized to TJ plaque by binding to the SH3 domain. The GK domain is the binding site for occludin (). SH3-GK domains are responsible for the binding to α-catenin and afadin (; ). In addition to diverse interactions, the SH3-GK domain is thought to play a role in the dimerization of MAGUK proteins as reported for other MAGUK proteins, especially as shown by PSD-95 () and Dlg/SAP90/SAP102 (), but direct evidence is lacking in the case of ZO-1. The acidic domain has not been well characterized in previous studies. As ZO-1 binds to not only TJ proteins (such as claudins and occludin), but also to AJ proteins (such as α-catenin and afadin), we speculate that ZO-1 may orchestrate the behavior of binding partners during epithelial cell polarization and play a role in sorting belt-like AJs and TJs from primordial AJs. We have previously established an epithelial cell line lacking the expression of all ZO-1/ZO-2/ZO-3 to clarify their function. Using mouse EpH4 epithelial cells in which ZO-3 was not expressed, we established cell lines with a knocked-out ZO-1 gene (ZO-1 cells) with homologous recombination (). As the next step, clones with suppressed ZO-2 expression (1[ko]/2[kd] cells) were obtained from ZO-1 cells by stably expressing short interfering RNAs (). We previously reported that these cells possessed well-polarized cell architecture in terms of the differentiation of apical/basolateral membranes and formation of belt-like AJs but lacked TJs completely in the confluent state. The exogenous expression of N-terminal PDZ1-3 domains of ZO-1 was inefficient to rescue the formation of TJs in 1(ko)/2(kd) cells; however, when N-terminal PDZ1-3 domains of ZO-1 were forcibly recruited to the lateral membrane by adding a myristoylation signal and dimerized using the FKBP system, claudins were polymerized in 1(ko)/2(kd) cells, indicating that dimerization of the PDZ domains of ZO-1 determine whether and where claudins are polymerized in epithelial cells (). In the present study, we carefully observed the formation process of junctional complexes in 1(ko)/2(kd) cells and parental EpH4 cells using the Ca switch assay and examined the roles of ZO-1 in the formation of belt-like AJs and junction-associated linear actin cables besides TJs during epithelial polarization. Our data indicate that ZO-1 plays crucial roles not only in TJ formation, but also in the conversion from “fibroblastic” AJs to belt-like “polarized epithelial” AJs during epithelial polarization. Furthermore, to examine whether ZO-1 itself mediates the formation of both belt-like AJs and TJs, we performed a mutational analysis of ZO-1. We examined AJ formation carefully during epithelial cell polarization in 1(ko)/2(kd) cells and parental EpH4 cells using the Ca switch assay. The cells were cultured in a low Ca medium containing 5 μM Ca overnight under confluent conditions, and their polarization was initiated by transferring to a normal Ca medium. The degree of AJ formation was evaluated by immunofluorescence staining with anti-afadin mAb and anti–E-cadherin mAb. As shown in , parental EpH4 cells began to form belt-like AJs at 2 h and appeared to have mostly completed the process at 4 h after being transferred to normal Ca medium. In clear contrast, in 1(ko)/2(kd) cells, even after a 24-h incubation in normal Ca medium, AJ staining was still punctate. The number of spots of primordial AJs increased in 1(ko)/2(kd) cells along the time course, but each spot of primordial AJs was not fused (). The rearrangement of the actin filaments during junctional maturation was also significantly delayed in 1(ko)/2(kd) cells compared with parental cells. In parental cells, the cortical actin cytoskeleton was aligned in a linear fashion along the cell–cell junction 4 h after the Ca switch. In contrast, actin bundles were not organized at the cell cortex even 24 h after Ca repletion in 1(ko)/2(kd) cells. To confirm the retardation of junction maturation in 1(ko)/2(kd) cells, 1(ko)/2(kd) cells were cocultured with parental EpH4 cells and double stained with antibodies against components of AJs or TJs, phalloidin and anti–ZO-1 antibody, 24 h after replating (). We examined the behavior of integral membrane proteins of AJs (E-cadherin and nectin) and undercoat proteins of AJs (α-catenin and afadin) in 1(ko)/2(kd) cells 24 h after replating. Judging from their staining, belt-like AJs were formed between parental EpH4 cells; however, spot-like AJs were still present in 1(ko)/2(kd) cells, representing a general defect in the assembly of belt-like AJs in 1(ko)/2(kd) cells. Afadin and β-catenin normally colocalized at primordial AJs in 1(ko)/2(kd) cells, indicating that the E-cadherin–catenin and nectin–afadin complexes were associated even in the absence of ZO-1/ZO-2 (Fig. S1 A, available at ). In addition, Par-3 is also normally colocalized with afadin at primordial AJs in 1(ko)/2(kd) cells (Fig. S1 B), and we concluded that molecular assembly of primordial AJs is normal in 1(ko)/2(kd) cells. TJ components (claudin-3, tricellulin, and cingulin) were not present at cell–cell contacts of 1(ko)/2(kd) cells (). These phenotypes of 1(ko)/2(kd) cells, the absence of TJ formation and delayed formation of belt-like AJs and junction-associated linear actin cables, were rescued by recovery of ZO-1 or -2 ( and not depicted). The same results were obtained in the case of F9 cells lacking ZO-1 and -2 by homologous recombination (unpublished data). The discontinuity of AJs in 1(ko)/2(kd) cells was decreased by further culture in normal Ca medium judging from the staining of afadin, Par-3, and E-cadherin (Fig. S2, available at ). Cortical actin cables were normally formed in 1(ko)/2(kd) cells 72 h after Ca switch; however, junction-associated actin cables were not observed (Fig. S2 B). The staining of actin filaments became sharp in more confluent state as shown in our previous paper (). The formation of TJs was not restored by long culture, and we confirmed that ZO-1/ZO-2 is a structurally essential component of TJs. Although 1(ko)/2(kd) cells finally became well-polarized epithelial cells in a confluent state after Ca switch (Fig. S2 A), the retardation of belt-like AJ formation during cell polarization in 1(ko)/2(kd) cells clearly indicated that the loss of ZO-1/ZO-2 affected the initial phase of epithelial cell polarization. The polarity protein complex, which consists of Par-3, Par-6, and aPKCλ/ζ, has been shown to be required for the maturation of belt-like AJs and TJs from primordial AJs in epithelial cells (; ). The newly discovered phenotype of 1(ko)/2(kd) cells, the persistence of primordial AJs, is similar to a previously reported phenotype of the dominant-negative mutant of aPKCλ overexpressing epithelial cells (). During epithelial cell polarization, the polarity protein complex is known to be recruited to primordial AJs, and its activation at primordial AJs triggers belt-like AJ and TJ formation (). In addition to the polarity protein complex, several groups reported that the Rac1-specific guanine nucleotide exchange factor, Tiam-1, acts upstream of Par-3, Par-6, and aPKCλ/ζ during epithelial polarization (; ; ). We first examined whether molecular assembly of the Par-3– Par-6–aPKC complex and Tiam-1 at primordial AJs was changed in 1(ko)/2(kd) cells. There was no obvious difference in the expression levels of Par-3, Par-6, aPKCλ/ζ, and Tiam-1 between parental Eph4 cells and 1(ko)/2(kd) cells (). 24 h after replating, Par-3, Par-6, aPKCλ/ζ, and Tiam-1 localized at primordial AJs in 1(ko)/2(kd) cells, indicating that the localization of them at primordial AJs was not affected by the loss of ZO-1/ZO-2 ( and not depicted). A small G protein, Rac1, is known to be activated upon E-cadherin and nectin mediated cell–cell contact formation (; ). The activation of Rac1 is required for the activation of aPKC and subsequent cell polarization (). We examined whether the activation of Rac1 during cell polarization was altered in 1(ko)/2(kd) cells. We analyzed Rac1 activation in parental EpH4 cells and 1(ko)/2(kd) cells upon a Ca switch. In parental EpH4 cells, Rac1 activation occurred within 10–30 min, whereas Rac1 activity was hardly stimulated in 1(ko)/2(kd) cells, suggesting that delayed cell polarization in 1(ko)/2(kd) cells was due to the impaired activation of Rac1 in primordial AJs (). Indeed, the exogenous expression of dominant-active (DA) Rac1 (but not Cdc42-DA) led to the maturation of belt-like AJs in 1(ko)/2(kd) cells, whereas the polymerization of claudins was not restored by the exogenous expression of Rac1-DA (). These data demonstrated that ZO-1 plays a critical role in the establishment of belt-like AJs through the activation of Rac1. The relationship between the activation of Rac1 and ZO-1 upon cell–cell contact should be clarified in future studies. The aforementioned findings suggested that ZO-1 plays crucial role in the conversion from fibroblastic AJs to belt-like polarized epithelial AJs through activation of Rac1 at primordial AJs in the initial phase of epithelial polarization. On the other hand, 1(ko)/2(kd) cells completely lacked TJs (). Because a previous study suggested that AJ formation was a prerequisite for the assembly of TJs (), we examine whether ZO-1 is directly involved in the establishment of two distinct junctional domains, belt-like AJs and TJs, during epithelial polarization. To determine whether ZO-1 itself might mediate the formation of both belt-like AJs and TJs, we performed a mutational analysis of ZO-1. We tested whether these mutants could rescue belt-like AJ and/or TJ formation by transiently expressing them in 1(ko)/2(kd) cells. We first examined whether the N-terminal half of ZO-1 containing three PDZ domains, an SH3 domain, and a GK domain (ZO-1) restored the formation of linear actin cables and belt-like AJs and TJs in 1(ko)/2(kd) cells. When ZO-1 was expressed in parental EpH4 cells, ZO-1 localized at lateral membranes ( and Fig. S3, available at .). Claudin-3 and occludin were incorporated into the ectopic TJs formed at cell– cell contacts between ZO-1 expressing parental EpH4 cells ( and not depicted). ZO-1 localized at lateral membranes in 1(ko)/2(kd) cells and did not induce reorganization of actin filaments and formation of belt-like AJs, whereas ZO- 1 induced claudin polymerization in 1(ko)/2(kd) cells (). TJs formed at cell–cell contacts between ZO-1 expressing 1(ko)/2(kd) cells were abnormal in that the TJs were discontinuous and expanded along the lateral membrane. Although ZO-1 was an artificial construct, ZO-1 induced TJ formation without belt-like AJ formation, indicating that ZO-1 has a potential to form TJs independently of fully blown belt-like AJ formation in epithelial cells. Because SH3 and GK domains of PSD-95, another member of MAGUK homologues, were reported to interact with each other and dimerize through the intermolecular association, we considered that SH3-GK domains of ZO-1 also function as a dimerization module and contribute to the formation of a MAGUK network and the subsequent polymerization of claudins. We confirmed that ZO-1 (but not ZO-1) formed a self-dimer in vitro (Fig. S3 C). The ability to form a self-dimer of ZO-1 is consistent with the idea that SH3-GK domains of ZO-1 function as a dimerization unit and contribute to the formation of a MAGUK network. Previously, we reported that the N-terminal half of ZO-1 containing three PDZ domains, an SH3 domain, and a GK domain was enough to induce the formation of normal TJs in 1(ko)/2(kd) cells (). The construct used as the N-terminal half of ZO-1 in the previous study was characterized by and encoded 1–862 amino acid residues of ZO-1. More precisely, 805–871 amino acids of ZO-1 comprise an acidic domain, and the construct contained most of an acidic domain. Therefore, as the next step, a longer construct including ZO-1 and an acidic domain (ZO-1) was introduced into parental EpH4 cells and 1(ko)/2(kd) cells. In parental EpH4 cells, exogenously expressed ZO-1 was incorporated into TJs efficiently (). In 1(ko)/2(kd) cells, formation of belt-like AJs and TJs was restored completely by exogenous expression of ZO-1 (). Our knowledge about the function of the acidic domain is limited so far. In the present study, for the first time, we demonstrate that the acidic domain is required for the proper segregation of belt-like AJs and TJs during epithelial polarization. As PDZ domains of ZO-1 directly bind to the C terminus of claudins, we examined whether PDZ domains of ZO-1 are required for proper formation of belt-like AJs. The construct of ZO-1 lacking PDZ1-3 (ZO-1) did not rescue TJ formation but restored the formation of belt-like AJs and linear actin cables, judging from the staining of phalloidin and afadin in 1(ko)/2(kd) cells (). The lines of actin cables induced by ZO-1 in 1(ko)/2(kd) cells were less sharp than those of parental EpH4 cells. This indicated that ZO-1 was insufficient for the formation of junction-associated actin cables. Finally, the construct containing an acidic domain and actin binding region (ZO-1) was not effective for the recovery of both belt-like AJs and linear actin cables, indicating that SH3-GK domains are indispensable for the formation of belt-like AJ formation (). Collectively, these data demonstrate that ZO-1 rescued at least polymerization of claudins independently of AJ formation, whereas ZO-1 rescued only belt-like AJ formation. Thus, the required domains for belt-like AJ and TJ formation differed. The SH3-GK domains of ZO-1 are essential for both belt-like AJ and TJ formation. As ZO-1 rescued both belt-like AJ and TJ formation, future studies have to clarify how the acidic domain regulates the proper segregation of belt-like AJs and TJs during epithelial polarization. Mouse anti–ZO-1 mAb (), rat anti-occludin mAb (), rat anti-tricellulin mAb (), and rat anti-cingulin mAb () were raised and characterized previously. Rat anti–E-cadherin mAb (ECCD2) and rabbit anti-PAR-3/ASIP pAb were provided by M. Takeichi (Center for Developmental Biology, Kobe, Japan) and S. Ohno (Yokohama City University, Yokohama, Japan), respectively. Mouse anti-afadin mAb and rat anti–nectin-2 mAb were provided by Y. Takai (Osaka University, Osaka, Japan). Rabbit anti–α-catenin and mouse anti–α-tubulin (DM1A) were purchased from Sigma-Aldrich. Rabbit anti–claudin-3 pAb and rabbit anti–ZO-2 pAb were purchased from Zymed Laboratories. Rat anti-HA mAb was purchased from Roche Applied Science. Rabbit anti-aPKC pAb and anti-Tiam1 pAb were purchased from Santa Cruz Biotechnology, Inc. Mouse anti-Rac1 mAb was purchased from Upstate Biotechnology. pGEX4T1-CRIB-Pak for pull-down assay of Rac1 has been described previously () and was provided by F. Oceguera-Yanez and S. Narumiya (Kyoto University, Kyoto, Japan). pEF-BOS-myc-Rac-DA and pEF-BOS-myc-Cdc42-DA were provided by Y. Takai (Osaka University, Osaka, Japan) and T. Sasaki (University of Tokushima, Tokushima, Japan). A diagram of the expression constructs of deletion mutants used in this study is shown in . Each fragment was amplified by PCR and subcloned into the vector pCAGGS-NGFP or pCAGGS-NHA. Mouse EpH4 epithelial cells, 1(ko)/2(kd) cells, and MDCK II cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum. EpH4 cells were a gift from E. Reichmann (Institute Suisse de Recherches, Lausanne, Switzerland). Transfection was performed using Lipofectamine Plus Reagent (Invitrogen) according to the manufacturer's instructions. Immunofluorescence microscopy was performed as described previously (). In brief, cells cultured on coverslips were fixed with 3% formalin in PBS for 10 min at RT, treated with 0.2% Triton X-100 in PBS for 5 min, and washed with PBS. Blocking was done by incubating the fixed cells with 5% BSA in PBS for 30 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 for 30 min with the secondary antibody. For actin staining, Alexa Fluor 568 phalloidin (Invitrogen) was added to the secondary antibody. Specimens were observed at RT with a photomicroscope (BX51; Olympus) and with a confocal microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) equipped with a Plan-APOCHROMAT (60/1.40 N.A. oil-immersion objective) with appropriate binning of pixels and exposure time. The images were analyzed with IPLab version 3.9.5 (BD Biosciences) and LSM510 Meta version 3.0 (Carl Zeiss MicroImaging, Inc.). The in vitro binding assay was performed as previously described (; ). Samples were resolved by SDS-PAGE and electrophoretically transferred to a nitrocellulose membrane (Schleicher & Schuell). This membrane was incubated successively with primary antibodies, which were visualized using a blotting detection kit (GE Healthcare). Fig. S1 shows the colocalization of β-catenin, afadin, and Par-3 at primordial AJs in 1(ko)/2(kd) cells. Fig. S2 shows that 1(ko)/2(kd) cells restore the formation of belt-like AJs and cortical actin cables 72 h after the Ca switch. Fig. S3 shows that exogenous expression of ZO-1 induced aberrant TJs also in MDCK II cells and that ZO-1 formed a self-homodimer in vitro. Online supplemental material is available at .

Signaling by the multifunctional cytokine TGFβ1 is regulated by other secreted proteins that influence TGFβ1's biological activity and localization (). Central to these interactions is the fact that TGFβ1 is secreted in a latent form. TGFβ1 latency arises from a noncovalent interaction between TGFβ1 and its propeptide, latency-associated peptide (LAP). In addition to blocking access of TGFβ1 to TGFβ receptors, LAP interacts via disulfide bonds with proteins of the latent TGFβ-binding protein (LTBP) family (). The trimolecular complex of TGFβ1, LAP, and LTBP is referred to as the large latent complex and is thought to be the major secreted form of latent TGFβ1. LTBPs bind to matrix molecules, thereby anchoring latent TGFβ1 in the extracellular space, and influence the release of TGFβ1 from LAP, a process called latent TGFβ1 activation (). Two other isoforms of TGFβ (TGFβ2 and -3) are secreted in similar latent forms. Proposed TGFβ1 activators include thrombospondin-1 (TSP1), proteases such as plasmin and matrix metalloproteinases (MMPs), the integrins αvβ6 and αvβ8 (which recognize an RGD sequence in LAP), and reactive oxygen species (; ; ; ; ; ; ). By degrading LAP or changing its conformation, these activators permit TGFβ1 to engage TGFβ receptors. Because TGFβ1 regulates numerous processes (immune function, cell proliferation, apoptosis, extracellular matrix formation, and vascular development, among others), one can speculate that multiple TGFβ1 activation mechanisms are used to activate TGFβ1 in diverse contexts. However, we lack a comprehensive picture that connects specific biological effects of TGFβ1 with specific TGFβ1 activators. In some cases, deletion of the gene encoding a putative TGFβ1 activator results in a phenotype consistent with a TGFβ1 deficit. For example, mice lacking TSP1 or the integrin β6 subunit ( and mice) develop inflammation, although it is not as severe as that in mice, which develop marked infiltrates of activated T cells in multiple organs and die soon after weaning (; ; ). In addition, mice are protected from TGFβ-dependent fibrotic tissue reactions (). Some embryos lacking αvβ8 () have defective vasculogenesis that appears identical to defects observed in a subset of embryos (; ). However, knockouts of genes encoding other putative TGFβ1 activators, such as plasminogen and various MMPs, do not result in phenotypes clearly related to TGFβ1 deficits. Although ; mice have a more severe inflammatory phenotype than either single-null animal (), these mice do not fully phenocopy mice. To delineate the role of RGD-binding integrins in generation of TGFβ1 activity, we created mice with a selective loss of integrin-mediated TGFβ1 activation. To do this, we made a TGFβ1 gene mutation that encodes an inactive version of LAP's integrin-binding site (RGE instead of RGD) and used embryonic stem (ES) cells containing the mutant gene to generate mice. These mice produce latent TGFβ1 that cannot interact with RGD-binding integrins. A targeting vector was made by inserting two contiguous fragments of DNA, with appropriate mutations in the fragment containing exon 5, into the cloning sites of the pKSNT vector (). One mutation creates a conservative single amino acid change (D246E) within the RGD motif located near the C terminus of LAP (), and the other creates a new restriction site that can be used for identification of the mutant allele. The vector was used to transfect ES cells, and cells with correct targeting of were injected into blastocysts to generate mice carrying the mutation (). These mice were crossed with Cre-deleter mice to remove the -flanked cassette from the targeted gene. mice display several abnormalities (), the most prominent of which is T cell–mediated multiorgan inflammation that leads to death soon after weaning. In addition, a strain-dependent fraction of embryos dies around embryonic day (E) 10 because of failure of yolk sac vasculogenesis and/or hematopoiesis (). Also, mice lack Langerhans cells (LCs), which are dendritic cells residing in the epidermis (). We predicted that mice would have an incomplete version of the phenotype, as occurs in , , and mice (see for a comparison of knockout phenotypes). However, mice display the cardinal features of mice. mice appear normal at birth but are smaller than littermates by 14 d and have markedly reduced survival (). Histologic examination of mice between the ages of 18 and 28 d revealed marked mononuclear cell infiltration of multiple tissues, in particular, the heart, lung, liver, stomach, and pancreas (which were abnormal in >85% of mice examined), and occasionally in the CNS ( and not depicted). In the lung, inflammatory lesions are usually localized around bronchi and larger vessels and sometimes diffusely within alveolar walls; in the liver, lesions are localized within portal canals. We did not note inflammation in the skin or kidney. These findings are similar to those reported for mice. A side-by-side comparison of the inflammatory lesions in and mice reveals them to be indistinguishable (Fig. S1, available at ). Because the β6-integrin subunit, which is expressed predominantly in epithelial cells, is up-regulated by injury and inflammation, we assessed β6 expression in 3-wk-old mice. β6 protein is markedly increased in lung and gastric epithelium () and is occasionally increased in biliary epithelium (not depicted). We did not detect β6 protein expression in the heart. We also compared the developmental consequences of the RGE mutation with the developmental abnormalities observed in mice. We determined the genotype of mice born to parents heterozygous for the mutant allele. Of 378 births, 36.4% were , 50.2% were , and 13.4% were . Thus, there is a substantial difference between the Mendelian ratios 1:2:1 (: :) and the observed ratios of ∼1:1.4:0.4, indicating an embryonic lethality of ∼50% associated with the genotype and ∼25% for the genotype. To determine when lethality occurs, we collected embryos for genotyping and histologic analysis. Of 146 E10.5 embryos, 24.3% were , 50% were , and 25.7% were . Therefore, embryonic death occurs at or after E10.5. Approximately half of the E10.5 yolk sacs had grossly evident anemia and/or absence of normal vasculature (). Histological abnormalities consisted variably of a paucity of vessels, absence of hematopoietic cells, and extensive buckling between the mesodermal and endodermal layers (). Similar findings have been reported for mice and mice expressing dominant-negative TGFβ receptors (; ). We also compared the LC status of mice with that of mice. Normally, peripheral blood monocytes enter the epidermis and differentiate into LCs, but mice completely lack LCs (). The target cell for TGFβ signaling appears to be the LC or a precursor, not epithelial cells, and TGFβ1 production by nonmarrow-derived cells is sufficient for LC production. We compared the presence of LCs in mice and littermate normal controls by staining for LCs in epidermal sheets obtained from the back or ear. LCs are absent from both sites (), aside from very rare small clusters of LCs in skin from ears, typically near the ear edge (not depicted). Because αvβ6 is functional in the epidermis, in that mice develop skin inflammation (), we suspected that this integrin might be the major or sole activator of TGFβ1 involved in LC generation. Therefore, we examined epidermal sheets from and mice (C57BL/6 strain) for the presence of LCs (). LCs are almost completely absent from ear epidermis. In contrast, epidermal sheets from the backs of mice have about half as many LCs as equivalent samples from mice (). Although some mice develop inflammatory skin lesions at sites of tissue trauma, characterized by hair loss and macrophage infiltration, these changes were not present in skin of our mice, and the lack of LCs in mice is independent of the inflammatory phenotype (). We conclude that the TGFβ1 required for LC generation is activated by RGD-binding integrins, but the integrins involved vary by region. In ear epidermis, the αvβ6 integrin is required for LC generation, whereas in back epidermis, αvβ6 appears to contribute but is not absolutely necessary. The phenotypes of and mice are similar to that of mice, which lack CD4+CD25+ regulatory T cells (Tregs). However, we detected no difference in the abundance of Tregs among CD4+ cells isolated from spleens of 8–10-d-old and mice, assessed as the percentage of either CD25+ or Foxp3+ cells (unpublished data). Because the phenotype of mice is highly similar to that of mice in the processes we examined, we assessed in vivo transcription and translation of the mutated gene to confirm that the observed phenotype is not due to impaired gene function. The targeting vector introduced a expression sequence into the intron between exons 5 and 6 of , and such sequences often interfere with expression of the targeted gene. Serum levels of TGFβ1 reflect both tissue production of TGFβ1 and release of TGFβ1 by platelets during clotting and can be measured with a luciferase-based bioassay after heating the serum to activate latent TGFβ1 (see Materials and methods). The assay is performed with and without anti-TGFβ antibody to confirm specificity. Serum levels of TGFβ1 were reduced in mice and undetectable in mice when the sequence was present (). However, after removal of the sequence, serum levels of latent TGFβ1 in mice heterozygous or homozygous for the RGE mutation were equivalent to those in mice (). Levels of latent TGFβ1 in serum-free medium conditioned by lung fibroblasts derived from , , or mice were also equivalent (). To confirm that the genetic manipulations had not altered mRNA sequence, we used RNA from lung fibroblasts to amplify the coding sequence by RT-PCR and found no changes other than the expected mutations introduced in exon 5. gene expression, measured by semiquantitative RT-PCR using RNA isolated from lung, liver, and heart, was not measurably affected by the RGE mutation ( and not depicted). The RGD-to-RGE mutation eliminates integrin-mediated TGFβ1 activation in cell culture experiments but does not appear to affect other LAP functions, such as TGFβ inhibition or interaction with LTBP. The D-to-E mutation is conservative, and the location of the RGD sequence is remote from the TSP1-binding site and an MT1-MMP cleavage site (; ). Purified recombinant LAP with the RGD-to-RGE mutation does not support adhesion by cells expressing αvβ6 or αvβ8 (; ), indicating that these integrins cannot effectively engage the mutated binding site. In contrast, LAP's interaction with TGFβ1 appears unaffected by the RGE mutation. Recombinant LAP binds and inhibits the bioactivity of recombinant TGFβ1, and the dose–response curve for this inhibition is not affected by the RGE mutation (); also, cells transfected with TGFβ1-RGE cDNA secrete normal amounts of latent TGFβ1 that is bioactive after activation by heating (not depicted). We also tested the ability of the RGE mutant form of LAP to form disulfide linkages with LTBP1 and to be incorporated into ECM, because these functions are critical for activation. CHO cells expressing LTBP1 () were transfected with cDNAs encoding wild-type or RGE TGFβ1, and conditioned media were immunoblotted with anti-LAP antibody. Equal amounts of high molecular weight bands consistent with LTBP1–LAP complexes () are seen, along with small amounts of monomeric LAP (∼37 kD). We previously showed that cells expressing αvβ6 do not activate soluble latent TGFβ1-RGE (). In cell culture conditions similar to those in previous reports (, ), the RGD and RGE forms of latent TGFβ1 are equivalently incorporated into cell-derived matrix, but only the RGD form of matrix-bound latent TGFβ1 is activated by αvβ6- or αvβ8-expressing cells (Fig. S2, available at ). In summary, loss of latent TGFβ1 activation by RGD-binding integrins in vivo recapitulates the major features of mice. We conclude that in these biological contexts, essentially all active TGFβ1 is generated in conjunction with integrin–LAP interactions. Issues requiring clarification include the identity of the RGD-binding integrins involved, the role of RGD-binding integrins in activation of TGFβ3, and the role of nonintegrin mechanisms of TGFβ1 activation. Integrins are heterodimers of α and β subunits. Of 24 mammalian integrins, 8 (αvβ1, αvβ3, αvβ5, αvβ6, αvβ8, αIIbβ3, α5β1, and α8β1) bind RGD sequences in their respective ligands. Of these, 2 (αvβ6 and αvβ8) clearly activate TGFβ1 in vitro and in vivo (; ). αvβ5 binds LAP and, when expressed by scleroderma fibroblasts, activates latent TGFβ; however, activation has not been noted in other αvβ5-expressing cells (; ). Three other RGD-binding integrins (αvβ1, αvβ3, and α8β1) bind LAP but have not been shown to activate latent TGFβ1 when expressed in cultured mammalian cells (; ; ). The phenotypes of αv-, β6-, and β8-null mice (; ; ) are somewhat similar to those of mice (), but the phenotypes of other integrin-null mice are not (α5- and β1-null mice die early in embryogenesis, precluding comparison). Nevertheless, it is plausible that these LAP-binding but non–TGFβ-activating integrins promote TGFβ1 activation in vivo, perhaps by concentrating latent TGFβ1 at cell surfaces where it might be activated by another process. TGFβ3-LAP contains an RGD sequence, and latent TGFβ3 is activated by αvβ6 and αvβ8 expressed in cultured cells (; ). The phenotypes of and mice do not overlap, but 10% of mice have cleft palate (), which occurs in all mice (; ; ). Therefore, αvβ8 may be partially responsible for TGFβ3 activation during palate formation. There are numerous reports of physiologic TGFβ1 activation by nonintegrin mechanisms, such as TSP1, proteases, and oxidants. Although our results suggest that RGD-binding integrins play a dominant role in TGFβ1 activation, several caveats exist. First, our results are limited to three phenotypes that are evident within the first few weeks of life, and it is possible that other TGFβ1-mediated processes occur independently of integrin-mediated activation. Second, integrins likely act in parallel or in series with nonintegrin mechanisms to generate TGFβ1 signaling in vivo. For example, αvβ8 and MT1-MMP cooperatively activate TGFβ1 at cell surfaces (), and proteolytic activity releases latent TGFβ1 from the ECM (), thereby potentially enhancing access of latent TGFβ1 to integrins for activation under some circumstances. The phenotype of TSP1-null mice partially overlaps that of mice (), suggesting that TSP1 may cooperate with RGD-binding integrins in TGFβ1 activation. Our findings reveal a critical role for RGD-binding integrins in the generation of TGFβ1 signaling activity in three disparate processes (control of inflammation, vasculogenesis, and LC genesis) and illustrate the utility of a subtle genetic mutation approach to complement gene knockout studies. Further work is needed to establish which RGD-binding integrins are involved in specific TGFβ1 effects and how integrins interact with other molecules involved in TGFβ1 activation. A 1.4-kb mouse TGFb1 cDNA was used to screen a 129/SvEv mouse Lambda FIX II Library. A 13-kb clone containing exons 1–6 was restriction mapped. HindIII digestion generated a 6-kb fragment containing exons 2–5 and a contiguous 5-kb fragment containing most of the intron between exons 5 and 6. These were subcloned into pBluescriptSK. Exon 5 encodes the RGD in TGFβ1-LAP. We used the QuikChange site-directed mutagenesis kit (Stratagene) to mutate the codon encoding D to a codon encoding E and to create an adjacent silent mutation creating a BstUI site (cg/cg). The mutagenesis primers were ggatcagccccaaacgtcgcggcgagctgggcaccatccatgac and gtcatggatggtgcccagctcgccgcgacgtttggggctgatcc. A targeting vector was made using the pKSNT plasmid (a gift from A. Joyner, New York University School of Medicine, New York, NY). The plasmid containing the 5-kb HindIII fragment was digested with BamHI to generate a 3.3-kb BamHI fragment (one BamHI site is derived from the pBluescript cloning site), which was inserted at the BamHI cloning site. The 6-kb HindIII fragment containing the RGD-to-RGE mutation was excised with SalI and XhoI and ligated into the targeting vector at the SalI site. A clone with correct insert orientation (SalI site closest to the cassette destroyed) was identified. The vector was extensively sequenced to confirm that genomic sequences and sites were intact and correctly oriented. The targeting vector was linearized by SalI digestion. 50 μg of DNA was used to transfect 5 × 10 W4 ES cells by electroporation. ES cells were selected with 200 μg/ml G418 and 2 μM gancyclovir. Correctly targeted clones were identified by Southern blot using external probes (exons 1 and 6). With ApaI-digested genomic DNA, an exon 6 probe revealed an 8-kb fragment of the wild-type gene and a 7.4-kb band in the targeted locus (RGE). With EagI-digested genomic DNA, an exon 1 probe revealed a 20-kb fragment of the wild-type gene and an 11-kb band in the targeted locus. Three ES clones with a homologous recombination event were identified, two of which included the RGE mutation. An ES clone was used for injecting blastocysts of C57BL/6J mice. The chimeric mice were bred with C57BL/6J mice to achieve germline transmission. mice were bred to generate homozygous mice. TGFβ1 protein levels were undetectable in these mice because of the presence of . Cre-deleter mice (a gift of A. Joyner) were crossed with mice. Two Cre lines were used, one of Swiss-Webster background and the other C57BL/6J. Removal of was confirmed by PCR showing undetectable product using primers for (gaacaagatggattgcacgc and gaagaactcgtcaagaagggc) and an appropriate-sized fragment using primers flanking the insertion site (gaggagacaagatctctcaga and caatggccctacacacacacagag). Lungs from newborn mice were minced and placed in tissue culture dishes with high-glucose DME containing 10% FCS, glutamine, penicillin, and streptomycin to allow lung fibroblasts to proliferate. Within the first three passages, cells were transfected with an expression plasmid containing cDNA for the SV40 large T antigen. Sera from wild-type and mutant mice were heated to 80°C for 10 min to activate latent TGFβ and then diluted 1:4 in DME. Bioactive TGFβ was measured by adding samples to mink lung epithelial cells stably transfected with a TGFβ-responsive luciferase reporter construct and then assaying cell lysates for luciferase activity (). Conditioned media were obtained by incubating equal numbers of cells in DME containing 0.1% BSA overnight, heat activated as described, diluted with an equal volume of DME, and assayed as described. Samples were tested in the presence and absence of a TGFβ1-specific inhibitory antibody (R&D Systems) or an inhibitory antibody against all three TGFβ isoforms (1D11; R&D Systems). Results are the means of triplicates ± SEM. Recombinant simian wild-type and RGE LAP were produced as described previously (). Active recombinant TGFβ1 was added to DME supplemented with 0.1% BSA. Aliquots of this stock were combined with varying concentrations of recombinant LAP and then added to mink lung epithelial reporter cells and assayed as described. CHO-K7 cells stably expressing LTBP1 (a gift from J. Annes and D. Rifkin, New York University) were transfected with expression vectors containing cDNA encoding different forms of TGFβ using Lipofectamine Plus (Invitrogen) or with empty plasmid as described previously (, ). The next day, cells were used to condition serum-free medium (DME supplemented with 0.1% BSA and glutamine) for 24 h. Immunoblotting with anti-LAP mAb VB3A9 was done as described previously (). The three types of serum-free conditioned media obtained as described above were incubated with 11 nM of purified, stripped TSP1 (a gift from J. Murphy-Ullrich, University of Alabama, Birmingham, AL; ) for 1 h at 37°C and then assayed with mink lung epithelial reporter cells as described, with or without addition of a TGFβ-inhibitory antibody (1D11). Primary cultures of lung fibroblasts (see Primary fibroblasts) were used as the source of RNA. RNA was extracted and reverse transcribed by standard techniques. PCR was performed using primers based on the 5′ and 3′ ends of the TGFβ1 coding sequence (5′ primer, tactgccgcttctgctcccac; and 3′ primer, caggagcgcacaatcatgttg). The resulting PCR product was the expected size and was completely sequenced. Semiquantitative RT-PCR was done with RNA isolated from lungs, liver, and heart of and mice (forward primer, exon 4: cggaatacagggctttcgatt; and reverse primer, exon 6: cttgctgtactgtgtgtccaggc). The PCR program was 94°C for 4 min; 94°C for 45 s, 60°C for 45 s, and 72°C for 90 s for 20, 25, or 29 cycles; and 72°C for 10 min. Amplification of β-actin cDNA (primers used: atctggcaccacaccttctacaatgagctgcg and cgtcatactcctgcttgctgatccacatctgc) was done as a control. 5-μm sections of formalin-fixed tissue were used for detection of the β6-integrin subunit. Endogenous peroxidase activity was quenched with 3% hydrogen peroxide in methanol for 15 min. Antigen retrieval was with Digest-All 3 Pepsin (Zymed Laboratories) for 5–7 min. Blocking was with avidin/biotin block solution (Vector Laboratories) followed by 0.5% casein solution for 15 min. Anti-β6 monoclonal antibody ch2A1 (a gift from S. Violette, Biogen Idec, Cambridge, MA), a humanized version of a mouse anti-β6 IgG mAb, was used at 0.5 μg/ml in 0.1% BSA for 1 h at room temperature. A Vectastain ABC kit (Vector Laboratories) with anti-human secondary antibody was used according to directions, with 3,3-diaminobenzidine/hydrogen peroxide as chromogen (Sigma Fast tablets; Sigma-Aldrich) and hematoxylin counterstaining. Specificity was confirmed by absent staining on lung sections from mice. Ia+ LCs in epidermal sheets were detected as described previously (). For counts, epidermal sheets from four and three wild-type mice were stained (6–12 sheets per mouse; mean 8.6). A digital image (400×) of each stained sheet was captured, and all LCs were counted. The mean cell count for each mouse was the average count of all sheets for that mouse. Statistical significance of the difference between means for wild-type and mice was assessed by two-tailed -test. CD4+ splenocytes were isolated from total splenocytes stained by negative selection with immunomagnetic beads (CD4+ T Cell isolation kit; Miltenyi Biotec). Flow cytometry was performed after labeling with phycoerythrin-conjugated anti-CD4 and FITC-conjugated anti-CD25 mAbs (BD Biosciences) or isotype controls to determine the fraction of CD4+ cells that were CD25+. To determine the fraction of CD4+ cells that were Foxp3+, cells were labeled with FITC-conjugated anti-CD4 antibody, and intracellular staining with a phycoerythrin-conjugated anti-Foxp3 antibody was performed (eBioscience). Primers are CGGAATACAGGGCTTTCGATT and GGTACGGGCATTCTGGATAC. PCR program is 94°C for 4 h and then 30 cycles of 94°C for 30 min, 60°C for 30 min, and 72°C for 50 min, followed by 72°C for 10 h. PCR products are digested with BstUI and electrophoresed on agarose gels. A light microscope (DM LB; Leica) captured images with a 10, 20, or 40× objective lens at room temperature. Permount imaging medium was used. The camera used was a Spot Insight Color (model 3.2.0), and the acquisition software was the Spot program version 4.0.9, both by Diagnostic Instruments. Fig. S1 shows the similarity of inflammatory lesions in and mice. Fig. S2 shows data from transfection experiments demonstrating that the normal and RGE forms of latent TGFβ1 can be incorporated into ECM and that cells expressing αvβ6 or αvβ8 can activate the normal form but not the RGE form of ECM-bound latent TGFβ1. Online supplemental material is available at .

Kinetochores mediate the interaction between chromosomes and spindle microtubules, thereby enabling mitotic chromosome movement, and produce a mitotic checkpoint signal that ensures bipolar attachment of all chromosomes before anaphase onset (; ). Assembly of the kinetochore during mitosis takes place at the centromere, a megabase-sized specialized chromatin region typically formed on arrays of α satellite DNA (; ; ). Despite the prevalence of centromeres at adenine-thymine–rich repetitive α satellite DNA, the DNA sequences themselves appear to play a nonessential role in centromere specification. This is most clearly exemplified by the characterization of human neocentromeres. In these rare but naturally occurring patient cases, a specific centromere has relocated to another site on the chromosome without any apparent DNA rearrangements, concomitant with vacating the original α satellite–containing locus (; ; ). This shows that DNA sequences normally associated with centromeres are neither necessary nor sufficient to promote centromere propagation and that maintenance of centromeres is determined predominantly in an epigenetic manner. Centromere protein A (CENP-A) is a conserved histone H3 variant that replaces canonical H3 specifically at centromeres in all known eukaryotes (; ; ; ) and has been shown to be required for the localization of nearly all other centromere and kinetochore components (; ; ; ; ; ; ). We have recently shown that the loop1 and α2 helix of the CENP-A histone fold domain is responsible for forming a rigid/inaccessible interface with histone H4 and that this region, when transplanted into canonical histone H3, confers centromere targeting (, ) and provides an essential function of CENP-A (). CENP-A chromatin directly recruits a six-component CENP-A nucleosome-associated complex (CENP-A) that forms the foundation for the assembly of other centromere components and the kinetochore during mitosis (). The existence of a CENP-A–directed centromere-specific chromatin structure makes CENP-A a prime candidate for the epigenetic propagation of centromere identity. This directly implies that CENP-A propagation at the centromere is a partially or completely self-directed process. It is, however, unknown how CENP-A is discriminated from canonical histone H3 and how its specific incorporation at centromeric nucleosomes is achieved. Earlier models have suggested that differences in timing of replication of centromeric DNA versus the genome overall may provide a temporal window permissive for CENP-A loading (; ). However, this appears not to be the case, as replication of centromeric DNA is not restricted to a specific time during S phase (; ). Alternatively, CENP-A loading could be separate from assembly of canonical histones altogether by allowing CENP-A loading outside S phase. Indeed, DNA replication is not required for CENP-A assembly and CENP-A mRNA, and protein levels peak only after S phase during late G2 phase, consistent with a disconnect between the timing of CENP-A and H3 assembly (, ). Whether propagation of centromeric chromatin and general chromatin is indeed temporally distinct and how and when CENP-A nucleosomes turn overis not known. This we now test by developing and exploiting a novel, covalent fluorescent pulse-labeling strategy with SNAP tagging. The SNAP tag, a modified variant of the suicide enzyme O-alkylguanine-DNA alkyltransferase, whose normal function is in DNA repair, has been extensively engineered to covalently and irreversibly modify (and inactivate) itself through acceptance of the cell-permeable guanine derivative O-benzylguanine (BG; or fluorescent derivatives thereof). In effect, this allows labeling of SNAP fusion proteins at will in vivo (, , 2006). We applied pulse labeling with this methodology to determining CENP-A turnover specifically at centromeres () as well as quench-chase-pulse labeling to follow the fate of newly synthesized CENP-A (). We established cell lines stably expressing centromere-localized CENP-A–SNAP at near endogenous levels in HeLa cells ( and Fig. S1, available at ). Multiple lines of evidence indicated that CENP-A with the SNAP tag substituted functionally for CENP-A in centromere maintenance. We have previously reported that transgene-encoded CENP-A expression leads to reduction of the endogenous CENP-A pool through competition at the protein level (). Here, a similar reduction in endogenous CENP-A in response to CENP-A–SNAP expression resulted in an unchanged overall CENP-A pool, the majority of which was SNAP tagged (Fig. S1 A; line 23 is used for all pulse-labeling experiments). Because chronic reduction of CENP-A to <50% is a cell-autonomous lethal event (), the SNAP-tagged CENP-A pool in the stable CENP-A–SNAP cell lines not only competed for assembly at centromeres with authentic CENP-A but also provided essential aspects of CENP-A function in centromere maintenance. (Retention of substantial CENP-A function by CENP-A–SNAP [219-aa tag] is in agreement with what has been shown for N-terminally YFP-tagged [240 aa] or C-terminally tandem affinity–tagged [TAP; 172 aa] CENP-A, which, respectively, rescue CENP-A lethality and incorporate into bona fide centromeric nucleosomes that are associated with a six-member complex of centromere components [CENP-A–TAP; ; ].) 15-min pulse labeling with the tetramethylrhodamine (TMR)-conjugated SNAP substrate, TMR-, specifically identified CENP-A–SNAP already assembled into centromeric chromatin (, top). Preincubation of CENP-A–SNAP–expressing cells with the nonfluorescent SNAP substrate (BG-block) led to complete quenching of SNAP and rendered CENP-A undetectable with TMR- (, bottom). To determine turnover of CENP-A at centromeres, cells were synchronized at the G1–S boundary by tandem treatments with thymidine. CENP-A bound to unreplicated centromeres was pulse labeled with TMR- and chased for up to two cell cycles ( and ). Consistent with earlier reports on total CENP-A levels that had indicated slow protein turnover (; ), centromere duplication in the initial round of DNA synthesis produced a 60 ± 14% reduction in intensity of TMR-–labeled CENP-A–SNAP at individual centromeres by the first mitosis and through the subsequent G1 (). After a second cycle of DNA replication, the previously labeled, centromere-bound CENP-A–SNAP was diminished to 25 ± 5% of its initial level, whereas the total number of fluorescent centromeres positive per cell remained unchanged throughout the experiment (). Thus, despite continued synthesis of both SNAP-tagged and endogenous CENP-A, CENP-A already loaded into centromeric chromatin by late G1 is redistributed to, and retained by, daughter centromeres. CENP-A must be replenished at centromeres after DNA replication to complete duplication of new centromeres. To determine the timing of CENP-A incorporation into chromatin of newly replicated centromeres, cells stably expressing CENP-A–SNAP were synchronized at the G1–S boundary by double thymidine block, and both centromere-associated and any free pool of CENP-A–SNAP were quenched with nonfluorescent BG ( and ). The cells were then released into S phase for 6.5 h, nascent CENP-A–SNAP was pulse labeled for 15 min by reaction with TMR-, and incorporation of the fluorescent CENP-A was examined in late S, G2, M, and the subsequent G1. CENP-A synthesized during S phase was diffusely localized in the nucleus ( and Fig. S2, available at ) but did not appear at centromeres during any phase of G2 or M. Rather, only after passage through mitosis and entry into the G1 phase of the next cell cycle did CENP-A assemble at centromeres (). Quantification of the number of cells positive for CENP-A loading confirmed that the initially synchronized cell population did not load substantial levels of CENP-A before ∼11 h after release from thymidine and concomitant with entry into G1 (). Close examination of cells just before and after mitotic entry revealed that the earliest time CENP-A loading could be detected at centromeres was concomitant with nuclear envelope reformation and completion of furrow ingression as indicated by midbody formation in late telophase/early G1 ( and Fig. S3, available at ). No newly made CENP-A was observed at centromeres at any time during mitosis before telophase. The absence of CENP-A at centromeres at these earlier time points cannot be attributed to a pool of newly synthesized CENP-A too small to be detected because the prelabeled pool size is the same for all time points. The pattern of loading restricted to late mitosis/early G1 was not a result of thymidine treatment per se because randomly cycling cells were also dependent on mitotic progression to permit CENP-A loading (Fig. S1). To time the arrival of CENP-A–SNAP at the centromere more accurately, we followed live cells containing a pool of TMR-–labeled but nonassembled CENP-A–SNAP from metaphase through early G1 (). TMR- labeling of live cells resulted in the nonspecific retention near the cell periphery (presumably in internal membranes) of a proportion of the dye, a proportion that is removed during normal fixation and washing conditions. Nevertheless, no TMR- signal could be detected specifically at centromeres in metaphase (). However, assembly of nascent CENP-A–SNAP could be detected as early as ∼50 min after anaphase onset and continued for several hours in early G1 ( and Video 1, available at ). Next, we determined whether loading of CENP-A is unique to the early hours of G1 or whether loading is also permissive at any other point in the cell cycle, including the possibility of a secondary CENP-A loading stage, as has been suggested in fission yeast (). CENP-A–SNAP in mitotic cells arrested with nocodazole treatment was initially quenched with nonfluorescent BG, and a G1 phase cell population was generated by release from nocodazole arrest. The CENP-A–SNAP pool produced during mid-to-late G1 was labeled and monitored for timing of centromeric deposition (). DNA content was assayed by FACS to verify cell cycle position (unpublished data). No assembly of CENP-A–SNAP–TMR- was detectable at centromeres during the subsequent S, G2, or M phases (). However, fluorescent CENP-A from the prior G1 assembled into the new daughter centromeres after exit from this subsequent mitosis (). Thus, despite the presence of a stable noncentromeric CENP-A pool, no loading occurred at any stage of the cell cycle before the following G1 phase (10–18 h after CENP-A synthesis and labeling). Although SNAP-tagged CENP-A faithfully tracks to centromeres and provides an essential function of CENP-A in centromere maintenance, there remained the possibility that the SNAP tag or the cell synchronization methods we have used would interfere with the timing of CENP-A loading. If CENP-A loading is normally restricted to early G1, levels of CENP-A (tagged or endogenous) on individual centromeres should double from early mitosis to when loading is completed, in late G1. On the other hand, if CENP-A loading occurs before mitosis, as previously proposed (), CENP-A levels in mitosis and G1 would be similar. Examination using indirect immunofluorescence to track endogenous CENP-A or direct fluorescence measurement of a cell line stably expressing YFP–CENP-A revealed that in both cases CENP-A levels increased from M to late G1 (∼3.4- and ∼2.5-fold, respectively; ), findings only consistent with CENP-A loading in G1 rather than before mitosis. The discrete, abrupt onset of CENP-A assembly as cells exit from mitosis suggested that passage through mitosis is a prerequisite for CENP-A assembly. Alternatively, entering the G1 cell cycle state may be triggering CENP-A assembly without any mechanistic involvement of mitosis per se. To distinguish these possibilities, the G1 cell cycle phase was disconnected from mitotic passage by combining the SNAP-based CENP-A assembly assay with a classic cell–cell fusion approach (). Heterophasic heterokaryons were generated by fusing G1 cells with G2 cells, each expressing CENP-A–SNAP and each uniquely marked by stable expression of CFP-tagged histone H3 or tubulin, respectively, to mark nuclei or microtubules (). The two differentially marked CENP-A–SNAP cell populations were synchronized by double thymidine treatment. Previously deposited CENP-A–SNAP was quenched, and each population was released for differing lengths of time so as to produce two synchronized populations, one of which was at mitosis/early G1 (H3-CFP cells) and the other at late S/early G2 phase (CFP-tubulin cells). The two populations were mixed, and cell fusion was induced with polyethylene glycol (PEG). Nocodazole was added to prevent any further passage through mitosis. 4 h after fusion, TMR- labeling was used to assay in both nuclei of the heterokaryons for assembly at centromeres of the unloaded, newly synthesized CENP-A–SNAP pool that was present in all nuclei (). After cell–cell fusions, nuclei originating from G2 and G1 cells share the same cytoplasm and, in principle, the same cell cycle state. Control fusions revealed that, as expected, loading of CENP-A in both nuclei occurred exclusively in G1 cell to G1 cell fusions. In contrast, no binucleate heterokaryons derived from fusion of two G2 populations could be found in which both nuclei loaded CENP-A (), and the vast majority (86%) loaded it in neither nucleus. It should be noted that because of the short time cells spend in mitosis (∼1 h) and the inherent spread in synchrony as cells transverse across the cell cycle, an early G1 phase cell population will invariably contain a fraction of cells that are in G2. (In this case, 33% of the H3-CFP G1 population had in fact not yet reached G1 by the time cells were fused.) Therefore, in all fusions, a spread of heterokaryons loading CENP-A–SNAP at centromeres at one, both, or neither of the nuclei is expected. Nevertheless, despite this inherently imperfect synchrony, a striking finding was that in binucleate heterokaryons derived from fusion between cell populations enriched in G1 and G2, most (66%) G1 cell–derived nuclei (H3-CFP marked) recruited CENP-A–SNAP to centromeres to levels indistinguishable from surrounding nonfused G1 cells. In contrast, no heterokaryons were found that had assembled CENP-A in both nuclei, indicating that G2-derived nuclei, although sharing the same cytoplasm with a CENP-A–assembling G1-derived nucleus, did not assemble CENP-A (), despite the presence of fluorescently labeled CENP-A–SNAP. The frequency of heterokaryons loading CENP-A–SNAP in one or neither nucleus corresponded to the frequency of H3-CFP G1 and G2 cells at the time of fusion (, arrows), indicating that in heterokaryons the G1- and G2-derived nuclei are neither inducing nor inhibiting CENP-A assembly in the other nucleus. Therefore, the early G1 cell cycle state that is permissive for CENP-A assembly does not directly dictate the ability to load CENP-A. Rather, passage through mitosis is crucial to allow CENP-A assembly as cells enter G1. Our experiments suggest that mitosis is a key cell cycle determinant in initiating CENP-A loading. To exclude the possibility that proficiency for CENP-A loading is determined by a “timing” mechanism rather than actual mitotic passage and G1 entry, cells were arrested using nocodazole to produce a nascent unloaded pool of CENP-A–SNAP in mitosis. Nocodazole-treated cells never assembled CENP-A–SNAP, even by the time 94% of control cells had reentered G1 and loaded CENP-A–SNAP (), reaffirming the notion that exit from mitosis is required for CENP-A loading. Multiple processes occur during mitosis that might act to trigger new CENP-A nucleosome recruitment. These include chromatin stretching, which occurs during metaphase and has been proposed as a mechanism for the exchange of histone H3 for CENP-A–containing nucleosomes (; ; ). Although the concept of functional reinforcement of centromere location that is part of this model is appealing, no experimental evidence has been generated in support for such a mechanism. Alternatively, DNA decondensation or the presence of other mitotic kinetochore components may be integral to triggering the process of centromere assembly. To test the tension-dependent CENP-A loading model, cells were produced that completed mitosis in the absence of microtubule attachment (and therefore microtubule-mediated chromatin stretching). To do this, cells were depleted of BubR1 with transcription-mediated short hairpin RNA and treated with nocodazole to block microtubule assembly, and CENP-A loading was assessed (). Under these conditions, cells enter mitosis without spindle assembly or kinetochore attachment, but quickly exit without the BubR1-dependent mitotic checkpoint (). Depletion of BubRI alone did not affect the ability of cells to load CENP-A, whereas nocodazole treatment of cells with normal BubR1 levels prevented mitotic exit and any loading (). Nocodazole treatment of cells depleted of BubR1 () or another mitotic checkpoint component Mad2 (not depicted) produced CENP-A loading to levels comparable to that seen in untreated cells, along with normal interphase nuclei with twice the number of resolved centromeres. This was indicative of a successful mitotic exit where sister chromatids had disjoined but failed to segregate and complete cytokinesis because of the absence of microtubules. Conversely, cells that did not load CENP-A were either arrested in mitosis (i.e., not depleted in BubRI) or had not yet entered mitosis (; as indicated by a centromere number consistent with unresolved sister centromeres, as is the case in G2 phase). Thus, passage through mitosis is critical for CENP-A loading, but microtubule attachment or microtubule-generated tension across centromeric chromatin is not. Our effort validates the SNAP tag (, , ), coupled with indirect immunofluorescence or live-cell imaging, as an approach capable of visualizing and tracking intracellular dynamics of protein pools synthesized at different times. SNAP technology also stands out from other cell biological tools to determine protein dynamics, such as FRAP experiments, in that it allows the determination of protein turnover on a much longer time scale and is therefore well suited for proteins with long half-lives. To earlier efforts that had shown CENP-A to be long lived (; ), with the SNAP tag approach we have demonstrated that, once assembled at centromeres, CENP-A does not turn over measurably within the ∼50-h time frame of our experiments. (An added benefit of this outcome is demonstration that the covalent SNAP-BG binding is indeed irreversible.) Moreover, the ability to differentially label SNAP protein pools synthesized at different times allows direct assessment of the fate of nascent proteins, including the turnover rates of proteins at the same cellular location, but assembled at different times. Using the SNAP tag approach, nearly all centromeric CENP-A is shown to remain centromere associated even during centromeric DNA replication, consistent with a role for CENP-A as an epigenetic marker maintaining centromere identity though cell division (; ; , ). More surprisingly, loading of newly synthesized CENP-A occurs in a discrete cell cycle window in early G1. A mitosis intervening between centromere DNA replication and new CENP-A loading is a prerequisite for CENP-A assembly. Although earlier work suggested that CENP-A may load in G2 phase based on an increase in overall CENP-A protein levels at this time (), our direct visualization with the SNAP tag has demonstrated that, despite its continued expression throughout the cell cycle, newly made CENP-A is accumulated in a nuclear, but noncentromeric, form before mitosis. The abrupt onset of CENP-A assembly at centromeres initiating at the end of mitosis firmly supports a model in which loading of CENP-A requires one or more key events during mitosis that may include nuclear envelope breakdown or chromatin decondensation, thereby allowing potential CENP-A assembly factors access to centromeric chromatin. Alternatively, assembly may be dependent on mitotic modification of CENP-A itself, which creates an environment that is permissive for subsequent CENP-A loading. Although passage through mitosis itself is a strict requirement for CENP-A loading, microtubule attachment at kinetochores has no apparent role in CENP-A assembly, in contrast to previous proposals (; ; ). It remains possible, however, that components of the greater centromere/kinetochore affect CENP-A loading or stabilization after loading. Defects in structural centromere proteins have been shown to affect CENP-A levels (; ). It is therefore possible that components of the centromere (which themselves are dependent on CENP-A for their localization) recruit new CENP-A or parts of the loading machinery. This would serve a direct epigenetic feedback between active centromeres and the propagation of new centromeric chromatin. Finally, propagation of CENP-A chromatin may await the availability of an active loading factor or an adaptor molecule at centromeres. The recently identified hMis18α, hMis18β, and M18BP1/hsKNL2 proteins, of which the M18BP1/hsKNL2 Myb domain–containing protein is an evolutionary conserved component, have been proposed to be required for CENP-A loading (; see Maddox et al. on p. of this issue). Strikingly, all these proteins display a pattern of centromere localization coincident with CENP-A assembly (from anaphase through early G1). Thus, recruitment of hMis18α, hMis18β, and M18BP1/hsKNL2 at the centromere during late anaphase could be dictated by a modification of centromeric chromatin coincident with mitotic exit or may represent a component of the CENP-A loading machinery that is itself activated during mitotic exit. The sudden onset of CENP-A assembly exclusively after reentry into G1, but not in mitosis, carries with it two important implications for epigenetic centromere inheritance. First, a requirement for a subsequent mitosis as a prerequisite for loading of CENP-A onto previously replicated centromeric DNA intrinsically couples centromere replication and maturation to cell cycle progression. Second, loading of new CENP-A after mitosis dictates that centromeres and the kinetochores assembled on them proceed through mitosis with only half the complement of CENP-A. During S phase, CENP-A protein is redistributed among sister centromeres, leaving vacant DNA sequences that are not replenished by CENP-A but are most likely occupied by typical histone H3.1–containing nucleosomes, which are available in excess during DNA replication. Indeed, histone H3–containing nucleosomes have been detected on mitotic centromeres interspersed with CENP-A–containing nucleosomes and have been shown to occupy centromeric chromatin when CENP-A levels are depleted (; ). Our work now indicates that the mixed chromatin state generated in S phase does not represent an intermediate state of centromeric chromatin where canonical nucleosomes serve as transient placeholders but that in fact this centromeric chromatin composition is what promotes kinetochore formation during mitosis (). CENP-A–SNAP–3XHA was constructed by inserting a PCR-generated fragment carrying the human CENP-A open reading frame flanked by KpnI and XhoI sites into the corresponding sites of pSS26m (Covalys) in frame with SNAP26m. A triple HA tag was introduced in frame at the SNAP26m C terminus, resulting in an 371-amino-acid open reading frame producing a 41-kD fusion protein (referred to as CENP-A–SNAP throughout this paper). HeLa cells and their derivatives were cultured in DME supplemented with 10% newborn calf serum (from here onward referred to as complete medium). HeLa monoclonal cell lines expressing CENP-A–SNAP, H3-CFP, or CFP–α-tubulin were generated by stable integration via Moloney murine leukemia retroviral delivery essentially as described previously (; CFP–α-tubulin retroviral construct was provided by J. Shah, Harvard Medical School, Boston, MA). Cells stably expressing the CENP-A–SNAP fusion protein were selected by blasticidin S (5 μg/ml; Calbiochem) and were isolated and individually sorted by flow cytometry. The resulting monoclonal lines were expanded and examined by fluorescence microscopy after TMR- labeling and by Western blot to identify lines expressing proper levels of the CENP-A–SNAP fusion protein. Clone 23 (Fig. S1) was used for all experiments in this paper unless stated otherwise. Ratio of CENP-A–SNAP levels to endogenous CENP-A in parental HeLa cells is ∼0.7:1. H3-CFP or CFP–α-tubulin cell lines were isolated by puromycin selection (1 μg/ml; Calbiochem). BubRI or control short hairpin RNA producing pSUPER plasmids and transfection procedures were as described previously (). SNAP tag activity in cells was quenched by addition of 20 μM O-BG (BG-block; Covalys) in complete growth medium for 30 min at 37°C or pulse labeled with 2 μM TMR- (Covalys) in complete growth medium for 15 min at 37°C. After quenching or pulse labeling, cells were washed twice with prewarmed PBS, after which cells were reincubated in complete medium to allow excess compound to diffuse from cells. After 30 min, cells were washed again twice in PBS followed by reincubation in complete medium. Unless stated otherwise in figures or legends, HeLa cells were treated with 2 mM thymidine in complete medium for 17 h, washed twice in PBS, and released in complete medium containing 24 μM deoxycytidine for 9 h followed by addition of thymidine to a final concentration of 2 mM for 17 h, after which cells were released again into complete medium containing 24 μM deoxycytidine and assayed. Nocodazole was used at 100 ng/ml. Cells were grown and SNAP assayed on glass coverslips followed by fixation and processed for immunofluorescence using standard procedures. Cells were not preextracted before fixation. Anti–CENP-A (a gift from K. Yoda, Nagoya University, Nagoya, Japan) was used at a dilution of 1:100, anti–CENP-C (a gift from W. Earnshaw, University of Edinburgh, Edinburgh, UK) sera was used at a dilution of 1:1,000, and anti-Mad1 (a gift from A. Musacchio, European Institute of Oncology, Milan, Italy) tissue culture supernatant was used at a dilution of 1:20. YL1/2 α-tubulin (Serotec) was used at a dilution of 1:2,500. Anti-HA11 (Covance Research Products, Inc.) was used at a dilution of 1:1,000. Donkey secondary antibodies (anti-mouse Cy5- or FITC-conjugated and anti-rabbit FITC-conjugated) were obtained from Jackson Immunoresearch Laboratories. Samples were stained with DAPI before mounting in ProLong (Invitrogen). Double thymidine–arrested H3-CFP and CFP-tubulin–expressing CENP-A–SNAP cells were SNAP-quenched with BG-block followed by release either directly or after 4 h to generate out-of-phase populations. After release of the trailing population, cells were coseeded on 18 × 18 mm uncoated coverslips. 5 h after seeding, coverslips were washed once in prewarmed PBS and incubated cell-side down in a 100 μl PEG-1500 (Roche) for 30 s followed by addition of 500 μl PBS and three subsequent washes in PBS. Coverslips were returned to complete medium containing 100 ng/ml nocodazole to prevent G2 cells from entering G1 and were TMR- labeled and fixed 4 h after PEG fusion. Digital images were captured using a DeltaVision RT system (Applied Precision) controlling an interline charge-coupled device camera (Coolsnap; Roper) mounted on an inverted microscope (IX-70; Olympus). For each sample, images were collected at 1× binning using a 100× oil objective at 0.2 μm z sections spanning the entire nucleus and were subsequently deconvolved, and maximum signals were projected as 2D images using softWoRx (Applied Precision; all images are deconvolved except those shown in Fig. S2). For quantification, images were converted to unscaled TIFF images. Centromere signal intensity was determined using MetaMorph (Molecular Devices) by measuring integrated fluorescence intensity within an 8 × 8 pixel square. Background signal was subtracted from an area within the nucleus not containing centromeres. For live-cell imaging, cells were grown on 22 × 22 mm glass coverslips transfected with YFP–CENP-C () using Effectene (QIAGEN) 48 h before SNAP labeling, after which coverslips were mounted on a slide separated in a double-stick tape chamber in phenol red–free CO independent DME (Invitrogen) containing 0.5 U/ml of the oxygen-scavenging enzyme, Oxyrase (Oxyrase, Inc.), and sealed with a 1:1:1 mixture of vasalin, lanolin, and paraffin. Images were acquired at 2× binning using a 60× oil objective for TMR- and YFP, as well as differential interference contrast at 10-min intervals. For each time point, 5 × 1 μm z sections were acquired for fluorescence images, and a single differential interference contrast image was acquired at the middle z position. Stacks were deconvolved, and maximum intensity was projected using softWoRx and assembled into a paneled video using MetaMorph. Whole cell extracts equivalent to 50,000 cells were separated by SDS-PAGE and transferred to nitrocellulose. Blots were probed by human anti-centromere serum (Antibodies, Inc.) at a dilution of 1:300. Fig. S1 shows the cell cycle–dependent CENP-A assembly of independently established cell lines expressing different levels of CENP-A–SNAP. Fig. S2 shows the diffuse nuclear localization of noncentromere-loaded CENP-A–SNAP in G2 cells. Fig. S3 shows evidence for CENP-A–SNAP loading coincident with cytokinesis and nuclear envelope reformation. Video 1 shows a time lapse of early G1 CENP-A–SNAP assembly at centromeres corresponding to stills shown in E. Online supplemental material is available at .

The cell cycle is orchestrated by the coordinated actions of several kinases whose activity is regulated positively by cyclins () and negatively by cyclin-dependent kinase (cdk) inhibitors (CKIs; ). Entry into the cell cycle from previous quiescence depends on the activation of G1-phase kinases. These chiefly include cdk4 and cdk6 kinases (cdk4/6) activated by D-type cyclins during early to mid-G1 phase and the cdk2 kinase, whose activation at the G1/S boundary depends on cyclins E and A (; ). The single most important substrate of these cdks is the retinoblastoma protein (pRb), whose phosphorylation is a prerequisite for S-phase initiation (). Recent results have shown that cell cycle reentry is facilitated by the activity of the cyclin C–cdk3 complex, which is also a pRb kinase (). CKIs belong to two families known as INK4 and Cip/Kip. The INK4 family comprises four members that are indicated according to their approximate molecular size as p15, p16, p18, and p19. These inhibitors specifically bind cdk4/6, preventing heteroduplex formation with D cyclins. Cip/Kip inhibitors include p21, p27, and p57. These molecules show much wider binding specificity, as they are able to bind essentially all cyclin–cdk complexes and, albeit with lower affinity, free cyclins (). In addition to their inhibitory role on cell cycle kinases, Cip/Kip family molecules facilitate cyclin–cdk complex formation (); the precise balance between these two opposite activities is still debated. The vast majority of the cells that make up a vertebrate's body spend most of their time in different nonproliferating states, which are collectively labeled as G0 phase of the cell cycle. Physiologically nonproliferating cells can be found in at least three distinguishable conditions, including reversible quiescence, replicative senescence, and the postmitotic state that characterizes and defines terminal differentiation. These three states are quite disparate with respect to both their phenomenology and the molecular mechanisms responsible for proliferative arrest. Quiescence is defined as a temporary, reversible absence of proliferation. This state can be induced by a variety of conditions including, among others, growth factor deprivation, contact inhibition, and loss of anchorage (). Quiescence can be usually readily reverted by removing the conditions that determined it. Indeed, the succession of events that follow exit from quiescence has long served as the principal experimental paradigm for cell cycle studies. Quiescence is usually associated with very low levels of cyclins, and it is generally believed that such low levels are its main determinants (; ). By way of example, serum-starved quiescent fibroblasts express very low amounts of cyclins associated with any phase of the cell cycle. Serum refeeding triggers exit from quiescence by bringing about a surge of cyclin D expression followed by cyclins E, A, and B in a coordinated succession (). Unlike cyclins, G1-controlling cdks are not primarily regulated at the protein expression level, thus making their cognate cyclins the limiting factors for cell cycle reentry (). Replicative senescence, which is also called cell aging, is normally a permanent state of proliferation arrest. It was originally described as an intrinsic limit to the number of duplications that cells can undergo in vitro (). More recently, replicative senescence has come to be viewed as the result of telomere shortening, damage at the molecular level, or both (). Indeed, preventing telomere attrition () and/or DNA damage accumulation () often results in cell immortalization. Unlike their reversibly quiescent counterparts, aging cells can express high levels of G1 cyclins, which are nonetheless devoid of associated kinase activity (). Removal of p53 () or pocket (pRb family) proteins (; ) has been shown to weaken replicative senescence or prevent it altogether. In most cases, both pathways must be disrupted to produce full immortalization (). The involvement of CKIs in establishing cell aging has long been recognized. p21 and/or p16 expression has been shown to associate with senescence in a variety of cell types (), whereas the role of p27 is somewhat more restricted (). Ablation of relevant CKIs such as p21 () or p16 () before senescence takes place can delay or prevent its establishment or reduce resistance to immortalization. p21 is a transcriptional target and is one of the most important effectors of the DNA damage response mediated by p53. p16 prevents the formation of cyclin D–cdk4/6 kinase complexes, thus interfering with pRb phosphorylation (). In addition, it has been shown that injection of anti-p53 antibodies into senescent fibroblasts induces cell cycle reentry along with a reduction of p21 expression (). Thus, CKIs belong to regulatory pathways that are essential for the onset of senescent arrest; surprisingly, however, whether or not they are absolutely required for the maintenance of senescence as opposed to its establishment is not known (). Terminal differentiation defines cells that permanently exit the cell cycle in the course of acquiring functional specialization. In extreme cases, terminally differentiated (TD) cells such as keratinocytes and erythrocytes lose their nuclei altogether, thus irreversibly relinquishing their ability to divide. However, most TD cells keep a full replicative apparatus while nonetheless becoming extremely refractory to spontaneous or induced cell cycle reactivation (). The molecular basis of such refractoriness is incompletely understood. It has been shown that pRb is absolutely necessary for the establishment of terminal differentiation in several histotypes, including skeletal muscle myotubes (). However, pRb is not required for maintenance of the postmitotic state (; ). Although several cyclins are down-regulated early upon skeletal muscle differentiation (; ), the up-regulation of several CKIs has been shown to correlate with myotube formation. p18 (), p21 (; ), p27 (; ), and p57 (CDKN2C, CDKN1A, 1B, and 1C, respectively; ) have all been indicated to participate in control of the cell cycle in muscle differentiation, making it difficult to sort their respective functions. Most conclusions concerning the role played by CKIs in muscle differentiation rely on knockout animals (), whose deficiencies may have been compensated for in the course of ontogenesis. Although useful to investigate the significance of CKIs in establishing the postmitotic state, such animals are ill suited to assess their relevance in the maintenance of terminal differentiation. Mice lacking one or more CKIs before the onset of terminal differentiation cannot help distinguish between continuing proliferation during differentiation and cell cycle reentry after the establishment of the postmitotic state. The three aforementioned proliferation arrest states behave differently at the molecular level. For instance, the forced expression of cyclin E–associated kinase activity can induce cell cycle reentry in quiescent cells (; ) but not TD myotubes (). Similarly, the functional suppression of p53 can reactivate senescent fibroblasts () but not TD cells (). The postmitotic state cannot be established in myotubes in the absence of pRb (), whereas this protein is not essential for initiating either quiescence or cell senescence (). Conversely, the removal of pRb reactivates the cell cycle in quiescent and senescent cells () but not in TD myotubes (; Huh et al., 2004). Despite their conspicuous differences, conditions sharing nonproliferation as an endpoint might also share a core mechanism responsible for their common phenotype. In this study, we suppress the expression of various CKIs in cells representative of all three states. Unexpectedly, CKI removal led to widespread cell cycle reentry in all cases even in the absence of exogenous growth factors. Thus, all states of nonproliferation depend critically on the constant expression of cell cycle inhibitors. We conclude that even the deepest, genuinely permanent growth arrest states are the result of a decision that must be continuously reasserted by the cell. We have previously shown that the cell cycle can be reactivated in postmitotic TD skeletal muscle cells, adipocytes, and neurons by restoring physiological levels of cdk4/6 activity. However, such restoration requires gross overexpression of both cyclin D1 and cdk4, suggesting the need to overwhelm preponderant levels of CKIs (). We sought to alleviate the requirement for cyclin D1–cdk4 overexpression by identifying and removing functionally significant CKIs in myotubes, which is a prototypic TD model system. To determine which CKIs are most relevant in our experimental settings, mouse C2C12 myoblasts were induced to terminally differentiate into myotubes that were then coinfected with recombinant adenoviruses carrying cyclin D1 and cdk4. Infection multiplicities were chosen so as to achieve a moderate overexpression of both moieties in order to maximize capture of available inhibitory molecules without causing cell cycle reentry. Immunoprecipitations were performed with antibodies to cdk4 that were equally capable of precipitating free cdk4 and its cyclin D complexes. The coprecipitated molecules were electrophoretically resolved (). Several specifically precipitated molecules were identified by liquid chromatography/tandem mass spectrometry. Most relevant among these were mouse cyclin D3 and, as the only representative of CKIs, p21 (). In an effort to reduce the load of CKIs in myotubes to facilitate cell cycle reactivation, we suppressed p21 expression by RNAi (). Unexpectedly, the sole suppression of p21 brought about a massive reentry of myotubes into the cell cycle, as shown by immunofluorescent detection of incorporated BrdU (). DNA synthesis took place equally well in the presence of 5% FBS and in the absence of exogenous growth factors. Frequent mitotic figures were noted in the reactivated myotubes, although mitoses were nearly always morphologically aberrant. Such mitoses were quickly followed by apoptotic death as detected by TUNEL staining 48 h after RNAi: 40% of the p21-depleted myotubes were TUNEL positive versus 14.3% of the control cells. Both abnormal mitoses and cell death resemble those described in the case of viral oncogene–reactivated muscle cells (). The specificity of p21 RNAi was ascertained by comparing the ability of four different double-stranded RNA oligonucleotides (siRNAs) for p21 to lower p21 protein levels and induce DNA synthesis in C2C12 myotubes (Fig. S1 A, available at ). The role played by the other known CKIs was systematically investigated by interfering with each of them singularly or in selected combinations (). The effectiveness of RNAi in reducing the expression of the different CKIs is shown in Fig. S1 (B–D). Ablation of no other single CKI stimulated BrdU incorporation in myotubes. Interference with p27 by itself also did not trigger DNA synthesis but reproducibly induced a small increase in the percentage of myotubes that were reactivated by interfering with p21 (). Fig. S2 (A–C; available at ) shows the association of three CKIs with cyclin D1, cyclin D3, or cdk4 in both myoblasts and myotubes. Total and cdk4-bound p21 and p27 were absolutely quantitated (Fig. S2 D). The relevance of p21 does not descend from its abundance, at least in comparison with p27. Thus, the pivotal role of p21 in maintenance of the postmitotic state in C2C12 myotubes is dictated by variables other than mere cellular concentration. CKI-RNAi experiments similar to those performed with C2C12 cells were performed with primary myoblasts (mouse satellite cells [MSCs]) freshly explanted from FVB mice. shows that in myotubes derived from these cells, multiple CKIs are involved in maintenance of the postmitotic state. Although in primary myotubes, ablation of p21 alone elicits very little BrdU incorporation, successive suppression of additional CKIs progressively increases the percentage of reactivated cells. However, interference with single CKIs other than p21 elicited no reactivation (negative data not depicted); even the combined suppression of p18, p27, and p57 was completely ineffective unless p21 was also suppressed (, right two histograms). These results suggest that key cdk complexes in MSC myotubes are controlled primarily by p21, whereas other CKIs take charge of suppressing cdk activity upon the loss of p21. Similarly, primary human myotubes incorporated BrdU () and went through aberrant mitoses (not depicted) after the simultaneous ablation of p21, p18, and p27. Altogether, the results described thus far show that postmitotic myotubes can be reactivated by the sole removal of critical CKIs even in the absence of exogenous mitogenic stimuli. In turn, this suggests that fully competent cyclin–cdk complexes are present, although normally inhibited, even in permanently proliferation-arrested cells. Reactivation of the cell cycle in the absence of exogenous growth factors presented an apparent paradox. Growth factors are required for cyclin D1 synthesis and assembly of cyclin D–cdk4/6 complexes (), which should be necessary to ignite a new cell cycle. Because the immunoprecipitation experiments have shown that sizable amounts of endogenous cyclin D3 associate with cdk4 ( A and S2 E), we hypothesized that preformed complexes containing at least cyclin D3, cdk4, and p21 are present in TD muscle cells. Indeed, similar complexes have been described in C2C7 myotubes (). Such preexisting complexes would require neither de novo synthesis nor assembly, thus being growth factor independent. The subtraction of p21 would allow cell cycle–promoting kinase activity to be expressed. This model finds strong support in a published quantitative appraisal of cell cycle–regulatory molecules in C2C12 myotubes showing low combined levels of D-type cyclins with a 3:1 prevalence of cyclin D3 over cyclin D1 (). Western blot analyses shown in Fig. S2 E agree qualitatively with this previous study. Our model made verifiable predictions: impairment of either cyclin D3 or cdk4/6 should abrogate the effects of p21 removal. Indeed, in C2C12 myotubes, suppression of cyclin D3 or infection with a recombinant adenovirus carrying a dominant-negative mutant of Cdk4 (dnK4) abolished BrdU incorporation stimulated by interference with p21 (). To assess the specificity of these results, similar experiments were performed with double p21/cyclin D1 RNAi. Interference with cyclin D1 determined only a modest reduction in myotube DNA synthesis in comparison with p21 RNAi alone (unpublished data). The specificity of dnK4 activity was verified by expressing it 24 h after p21 RNAi. In this case, there was no reduction in BrdU incorporation, as the newly activated cell cycle had already progressed past the point where cdk4/6 activity is required (). As also entailed by our model, interference with p21 should bring about substantial cdk4 kinase activity, which should be abrogated by the functional ablation of either cdk4 or cyclin D3. To test this prediction, C2C12 myotubes were subjected to RNAi for p21 either alone or in combination with cyclin D3 RNAi or infection with dnK4. shows that p21 protein levels were reduced in all p21-interfered cultures compared with myotubes transfected with control siRNA. Cyclin D3 protein levels were also profoundly decreased by the corresponding siRNA. In agreement with the model, cdk4 activity, which is absent in control-transfected TD muscle cells, was readily measured in p21-interfered myotubes at levels at least comparable with those found in proliferating myoblasts (). Such activity became undetectable upon either dnK4 infection or cyclin D3 RNAi. The specificity of cyclin D3 interference was verified by Western blot and cell cycle reentry analyses (Fig. S3, available at ), as already performed for p21. We noticed that p21 protein levels, which were already much reduced by p21 RNAi alone, were further lowered by simultaneous transfection of siRNA for cyclin D3 ( and see ). Because p21 is known to facilitate cyclin–cdk complex assembly (), it was conceivable that exceedingly low levels of p21 rather than cyclin D3 suppression might be responsible for impairing cell cycle reactivation. To exclude this possibility, we interfered with cyclin D3 and cotransfected progressively decreasing amounts of p21 siRNA in C2C12 myotubes (). Double transfection with cyclin D3 and 1.7 nM p21 RNA oligonucleotides yielded p21 levels comparable with those obtained with p21 siRNA alone, yet BrdU incorporation was still entirely suppressed. Thus, the suppression effect can be ascribed entirely to cyclin D3 interference rather than to its modest indirect effect on p21 protein levels. To explore the reasons why cyclin D3 plays a decisive role in the reactivation of C2C12 myotubes upon p21 removal, we measured cyclin D1– and cyclin D3–associated kinase activities in C2C12 myoblasts and in myotubes transfected with p21 or control siRNA. shows that in myoblasts, the kinase activity associated with cyclin D3 is three times as high as that of cyclin D1. In control myotubes, there is virtually no measurable cyclin D–linked kinase activity. p21 RNAi in myotubes brought about a marked decrease in cyclin D1 levels and a much more modest reduction of cyclin D3, which is consistent with the stabilizing role of p21 (). Although in p21-interfered myotubes the activities associated with both cyclin D1 and D3 raise, the latter is eightfold higher than that of cyclin D1. Thus, the vast majority of cyclin D–associated kinase activity in p21-depleted myotubes is carried by cyclin D3, which is the most abundant D-type cyclin in these cells (), explaining its pivotal role in myotube reactivation. Cip/Kip family CKIs have the ability to bind most cyclins and cdks (). As a consequence, removal of p21 or other Kip-type CKIs might derepress G1 kinases other than cdk4/6, most importantly the kinases associated with cyclin E. To investigate this possibility, C2C12 myotubes were transfected with control siRNA or various combinations of siRNAs for p21, p27, and cyclin D3. shows that the corresponding protein levels were effectively reduced. Substantial activity was found in cyclin E immunoprecipitates from myotubes interfered for p21 or p21 and p27. However, such activity was essentially abrogated by cyclin D3 RNAi (), indicating that the cyclin E–associated kinase is triggered as a result of the previous activation of cyclin D3–containing complexes. This result is in keeping with the notion that ordinarily, the activation of cdk2 is preceded by and is partially dependent on previous cdk4/6 firing (). In the present case, the removal of CKIs might be supposed to free cyclin E–cdk2 complexes, but cyclin E–associated kinase activity still depends on previous cdk4/6 activation. shows that upon cyclin D3 RNAi, the association of residual p27 with cyclin E is enhanced in spite of the strong reduction in p27 total levels, possibly as a result of diminished competition from cyclin D3–containing complexes. This finding provides a plausible mechanistic explanation for the lack of cdk2 activation upon simultaneous removal of CKIs and cyclin D3. Altogether, these results indicate that in myotubes, the primary target of CKI-mediated inhibition is not a cyclin E–driven kinase but rather a cyclin D3–cdk4/6 complex. This conclusion is also in agreement with our previous results showing that TD myotubes can be reactivated by physiological levels of cdk4 activity but not by the cdk2 kinase even if hyperactive (). We wished to investigate whether different CKIs could substitute for one another upon the selective suppression of one of them. We selected a well-studied model system: mouse embryo fibroblasts (MEFs) converted into muscle cells by the expression of MyoD (; ). Wild-type and p21 MEFs were infected with a MyoD-carrying adenovirus and were induced to differentiate into myotubes (myotubes from MyoD-converted MEFs [MEF-Mts]). TD myotubes thus obtained were subjected to RNAi for multiple CKIs. Similar to C2C12- and MSC-derived myotubes, wild-type MEF-Mts were reactivated only if the suppressed CKIs included p21. In particular, the suppression of p27 or p57 singularly or in various combinations could not elicit DNA synthesis (negative data not depicted). On the contrary, a meaningful fraction of p21 MEF-Mts were reactivated by the subtraction of either p27 or p57 (). Combinations involving more than one CKI allowed even higher percentages of myotubes to reenter the cell cycle. Thus, as suggested by multiple RNAi experiments (), diverse CKIs can functionally replace p21 in maintenance of the postmitotic state. Prompted by the results obtained with TD myotubes, we asked whether other physiological nonproliferative states also need constant maintenance by CKIs. RNAi experiments were performed with low passage primary mouse (not depicted) and human fibroblasts () that were made quiescent by 48-h serum starvation. Interference with p21 alone or in association with p27 and/or p18 in the presence of 0.1% FBS was up to 82% as efficient in inducing cell cycle reentry as refeeding with 10% serum (). Mitotic figures were much more frequent among the CKI-interfered cells than in those transfected with control siRNA (unpublished data). To assess whether the reactivated cells are capable of substantial proliferation, serum-starved p21-interfered fibroblasts were kept in culture medium containing 0.1% FBS and were counted in the 4 d after RNAi. shows that p21-depleted fibroblasts doubled in four days despite the absence of added growth factors, whereas control cells proliferated negligibly. Cell division was not accompanied by an appreciable increase in cell death: the percentage of TUNEL-positive p21-interfered cells never exceeded that of control fibroblasts (unpublished data). Importantly, cell cycle reentry driven by CKI RNAi was again completely abrogated by either cyclin D3 RNAi or dnK4 expression (). As a specificity control, although dnK4 also abolished serum-mediated cell cycle reentry, cyclin D3 RNAi reduced serum-promoted reactivation only partially, which is consistent with the notion that in human fibroblasts, serum acts mainly via cyclin D1 (). Immunoprecipitation and Western blot analyses show that low levels of all three D cyclins are present in quiescent fibroblasts along with levels of cdk4 and cdk6 as high as those found in proliferating cells (Fig. S4, A and B; available at ). However, only cyclins D1 and D3 are found associated with cdk4 (Fig. S4 C). Although the suppression of cyclin D3 abolishes the p21 RNAi–mediated reactivation of these cells (), we cannot exclude a proliferative role for cyclin D1–containing complexes. A quantitative evaluation of p21, p27, and p16 (Fig. S4, D–F) suggests that, as in the case of myotubes, the absolute expression level of a cell cycle inhibitor is not the primary determinant of its relevance in the maintenance of growth arrest. These results agree with and expand upon a previous study showing that the reduction of p21 levels induces mitotic reactivation of serum-starved fibroblasts (). Our data indicate that in fibroblasts as well as myotubes, cyclin D3–cdk4/6 complexes are present and potentially functional even during quiescence. Furthermore, they show that multiple CKIs are involved in silencing their activity. Finally, we investigated the role of CKIs in the maintenance of proliferative senescence. Primary human embryo kidney (HEK) cells have a limited proliferative life span in vitro (). Young cells are small and elongated, but, after a short term in culture, they become large, flat, irregularly shaped, and express acidic β-galactosidase activity, a marker of cellular senescence (; ). Aging cells become unresponsive to growth factors and cease proliferation (unpublished data). Senescent HEK cultures (2% BrdU incorporation in 24 h) were subjected to RNAi for p21 and/or p16 (). Interference with p21 induced reactivation of the cells, bringing almost half of the population into the mitotic cycle. In these cells, p16 RNAi was almost equally effective, which is consistent with the known role of this inhibitor in establishing the replicative senescence of human cells (; ). Combined interference for the two CKIs demonstrated almost no synergy, suggesting that in this experimental system, unlocking either pathway is sufficient to induce cell cycle reentry. Similar results were obtained with senescent human foreskin fibroblasts. The depletion of p21 or p16 brought the majority of these cells back into the cell cycle (). Although senescent cells are known to express high levels of inactive G1 cyclin–cdk complexes (), the reversal of senescence by direct removal of CKIs has not been shown before. Both HEK cells () and fibroblasts (not depicted) displayed abundant mitotic activity upon RNAi-mediated reactivation, which suggested that they could actually proliferate. To test this possibility, senescent HEK cells were transfected with p21 or p16 siRNA twice at a 48-h distance. Daily counts showed that the number of control-transfected cells did not increase in 3 d, whereas the CKI-interfered populations grew (). In particular, the cells transfected with p21 siRNA increased more than threefold over a 72-h period, showing that interference with p21 brings about not only cell cycle reentry but actual population growth. Cell cycle reactivation in either HEK cells or fibroblasts did not cause any increase in the spontaneous rate of cell death (). Altogether, these data allow us to conclude that senescent as well as TD and quiescent cells rely on the continuous presence of cell cycle inhibitors to maintain their nonproliferative state. In this study, we show that cells in such diverse nonproliferation states as terminal differentiation, quiescence, and replicative senescence can be allowed to reenter the cell cycle with comparable high efficiency simply by removing appropriate CKIs even in the absence of exogenous mitogenic stimuli. To determine which molecules are most important in preventing cdk activation in TD myotubes, we selected an unbiased mass spectrometry–based approach. Such a strategy would be effective in uncovering both known and potentially unrecognized molecular brakes in our system. Immunoprecipitation/mass spectrometry experiments pointed at p21 as the main inhibitor in the C2C12 experimental system. Accordingly, interference with p21 was sufficient to reactivate the cell cycle in C2C12 myotubes. The unexpected discovery that removal of CKIs allows full cell cycle reactivation in notoriously refractory TD cells led us to suspect that more pliable cell types might be reactivated by the same strategy. This hypothesis was fully confirmed by results obtained with quiescent as well as senescent cells. Absolute quantitation of some of the CKIs present in myotubes and in quiescent fibroblasts showed, somewhat surprisingly, that the inhibitor playing the most substantial role in suppressing proliferation was not the most abundantly expressed. In both mouse myotubes and resting human fibroblasts, p27 accumulates at much higher levels than p21 (Figs. S2 D and S4 D). However, the suppression of p21 is much more momentous than that of p27. Likewise, although in quiescent fibroblasts p16 protein levels are comparable with those of p21, the suppression of p16 is essentially inconsequential. There might be trivial explanations for these unexpected findings. For example, variable effectiveness in interfering with the different CKIs might skew the results. However, the large difference in the molar concentrations of p27 versus p21 (60-fold) strongly suggests the existence of partially unrecognized functional differences between the two molecules. Several points concerning specific nonproliferation states are worth discussing. Myotubes are prototypic TD cells that are capable of resisting the mitogenic pressure of growth factors, retroviral oncogenes, and many positive cell cycle regulators that act as powerful proliferative stimuli in non-TD cells (). The discovery of CKIs was quickly followed by reports of the high expression of a variety of them in numerous TD histotypes (; ; ). However, they are generally viewed as an aid in cell cycle exit at the inception of terminal differentiation () and as a safety feature against accidental cell cycle reactivation after the establishment of the postmitotic state. Indeed, there is no previous demonstration that CKIs are essential for maintenance of the postmitotic state in TD cells. As already discussed above (see Introduction), knockout animals cannot address this point, as they already lack CKIs before the onset of terminal differentiation. Thus, the appropriate means to investigate the role of inhibitory molecules in maintenance of the postmitotic state is to acutely ablate them after the establishment of this condition (). Our results are all the more unexpected, as the functional suppression of p21 has been reported to be insufficient to reactivate C2C12 myotubes (). However, this conclusion was reached indirectly by suppressing p21 with an E1A mutant whose other functions are incompletely understood; moreover, the activity of the essential cdk4/6 kinase was not investigated. The present results are consistent with our previous conclusion that the cdk4/6 kinase must be suppressed to conserve the postmitotic state. Conversely, activation of this specific kinase (but not other cdks) is sufficient to drive TD cells into S phase (). Indeed, we show here that CKI removal results in the reactivation of cdk4 and cell cycle reentry. For the first time, however, TD cells have been reactivated not through the forced expression of exogenous molecules but by the ablation of endogenous inhibitors, highlighting the existence of latent kinase complexes, whose presence had previously drawn little attention. Furthermore, in light of potential applications of our findings, it is certainly desirable to achieve cell cycle reactivation with a comparatively mild, reversible manipulation of endogenous molecules. Although the reactivation of quiescent fibroblasts by antisense suppression of p21 has been described previously (), we show here that D cyclins are the main players in such reactivation, which is similar to our finding in the myotubes. By analogy with TD muscle cells, resting fibroblasts possess considerable levels of preformed cyclin D–cdk4 complexes, whose activity is counteracted by specific CKIs. At least two pathways are deeply involved in determining cell senescence: those of p53 and pRb. These two pathways concur to establish senescence proliferation arrest, at least in part, via p21 and p16. Although the roles of these two CKIs in initiating cell senescence are firmly established, their contribution to the maintenance of this state is far less clear (). Our data show that even after its attainment, replicative senescence must be actively maintained by expressing high levels of cell cycle inhibitors. Removal of these CKIs brings about not only cell cycle reactivation but, at least in the short term, true cell proliferation. Whether CKI suppression is sufficient to overcome both the senescence and crisis barriers to immortalization remains to be investigated. The absence of proliferation is usually thought of as a passive state deriving from the unavailability of key components of the cell cycle engine, with particular emphasis on G1 cyclins (; ). In the present study, we show that TD, quiescent, and senescent cells can be reactivated by the sole removal of key CKIs. Thus, both temporary and permanent growth arrest, far from being negatively determined states, entail a relentless commitment by the cell, which must actively maintain such states through the continuous, critical expression of inhibitors. As a corollary, these results imply that even cells in permanent, physiologically irreversible arrested states harbor functionally relevant levels of key cyclins whose unchecked activity is sufficient to reactivate the cell cycle. The biological role of these cyclins in apparently incongruous cellular environments remains to be investigated. An ever-growing number of biotechnology applications requires large quantities of specific cell types to be grown in the shortest possible time, provided that this goal can be achieved without introducing or causing harmful alterations in the cells themselves. An ideal proliferation-inducing/accelerating protocol should leave the cell's genome and differentiative properties intact. A relatively gentle intervention such as reversible CKI removal might fulfill these requirements, although its suitability for these purposes must be thoroughly investigated. We believe that the results summarized in this study have the potential to allow, facilitate, and/or accelerate wound healing, tissue repair, and in vitro culture of fastidious cell types. As CKIs are pivotal inhibitors of proliferation in all of the cases assessed, their suppression could be exploited to induce the proliferation of cell types that cannot be currently cultured. The mouse C2C12 myoblast cell line () was cultured as previously described (). Differentiation was induced by plating the cells in DME containing 2% horse serum and 50 μM 1-β--arabinofuranosylcytosine (). Myoblasts used as a reference were synchronized in mid-G1 phase to approximate myotubes synchronously reactivated by CKI knockdown. Synchronization was obtained by serum starvation (2% horse serum) for 16 h followed by refeeding with 20% FBS for 8 h. Primary MSCs were isolated and cultured as previously described (; ). Primary human satellite cells were cultured in growth medium (PromoCell) and induced to differentiate by plating them onto gelatin-coated dishes in DME containing 5 μg/ml human insulin, 5 μg/ml human holotransferrin, and 50 μM 1-β--arabinofuranosylcytosine for 6 d. Wild-type and p21 MEFs were cultured in DME supplemented with 10% FBS (growth medium). MEF conversion into muscle cells was obtained by infecting them with a recombinant adenovirus carrying MyoD at an infection multiplicity of 400. 4 d later, terminal differentiation was induced following the same protocol used for C2C12 myoblasts. Low passage human primary foreskin fibroblasts (FB1329) were cultured in growth medium; quiescence was induced by switching confluent cultures to 0.1% FBS for 48 h. Synchronization in mid-G1 phase, which was attained for the same reasons described for C2C12 cells, was achieved by serum starving nonconfluent cultures in 0.1% FBS for 48 h followed by refeeding with 20% FBS for 8 h. Fibroblasts became senescent after prolonged culturing (passage >40). Primary HEK cells (a gift from S. Bacchetti, McMaster University, Hamilton, Ontario, Canada) were grown in α-MEM supplemented with 10% FBS and 2 mM -glutamine. The overexpression of human cyclin D1 and cdk4 was obtained by coinfecting C2C12 myotubes with the corresponding recombinant adenoviruses () at infection multiplicities of 15 and 150, respectively. Cultures were collected 24 h after infection and processed for immunoprecipitation and mass spectrometry identification. The expression of dnK4 was obtained by infecting C2C12 myotubes or FB1329 cells with a recombinant adenovirus () at a multiplicity of infection of 800 or 150, respectively. Target cells were transfected with the HiPerfect transfection reagent (QIAGEN) complexed with siRNAs. We optimized the HiPerfect/siRNA ratio to obtain efficient suppression of the target molecule with the smallest possible amounts of both reagents. RNAi for a single molecule in C2C12, MSC, and MEFs converted into myotubes was performed by complexing 6 μl HiPerfect and 5 nM siRNA per 35-mm petri dish. Human satellite cell myotubes, FB1329 fibroblasts, and HEK cells were interfered with 12 μl HiPerfect and 10 nM siRNA per 35-mm petri dish. RNAi for multiple molecules was performed by adding each siRNA at a final 5- or 10-nM concentration, depending on the cell type, to the same volume of HiPerfect used in single transfections. Transfection complexes were kept in the culture medium until the end of the experiment. To identify cdk4-associated molecules, C2C12 myotubes infected with human cyclin D1– and cdk4-carrying adenoviruses were lysed in cdk4 immunoprecipitation buffer containing 20 mM Hepes, pH 7.5, 250 mM NaCl, 5 mM MgCl, 0.5 mM EGTA, 0.25% NP-40, 5 mM ATP, 1 mM DTT, 0.5 mM PMSF, 10 μg/ml aprotinin, 10 μg/ml chymostatin, and 20 μg/ml leupeptin. Cell lysates were briefly sonicated and centrifuged. The small-scale analytical SDS-PAGE shown in was performed by immunoprecipitating cdk4 from 1 mg of whole-cell protein extracts, whereas large-scale mass spectrometry sample preparation was performed as follows. 50 mg of protein extract was subjected to 4 h of immunoprecipitation with a mixture of antibodies to cdk4: mouse mAb-1 and rabbit pAb-5 antibodies to human and mouse proteins, respectively (NeoMarkers), or control IgGs (Bioscience) bound to magnetic beads (Dynal). After extensive washes, the precipitates were detached from the antibodies by incubating the beads in 0.1 M glycine buffer, pH 3, supplemented with 0.1% Triton X-100 for 15 min; the suspension was then neutralized with 0.75 M Tris, pH 8, boiled, and separated on a 4–12% polyacrylamide gel (NuPAGE; Invitrogen). Selected Coomassie-stained bands were removed and processed for mass spectrometry identification. In brief, gel slices were excised, treated essentially as described previously (), and digested with modified sequencing grade trypsin (Promega). Liquid chromatography/mass spectometry/mass spectometry analyses were performed on an ion trap instrument (LCQ-DECA XP; Thermo Electron) equipped with a μ-ESI probe. Peptide mixtures were separated on a C BioBasic column (100 × 0.18 mm; 5 μm; Thermo Electron) at a 2 μl/min flow rate using a 60-min linear gradient from 5 to 60% acetonitrile in 0.1% formic acid. Protein identification was obtained using Bioworks Browser 3.1 software (Thermo Electron) and searching against a human protein database from the National Center for Biotechnology Information. To evaluate cdk4 activity in C2C12 cells, myoblasts and myotubes were lysed in cdk4 immunoprecipitation buffer and treated as described for cdk4 immunoprecipitation. 1.8 mg of protein extract was subjected to 4 h of immunoprecipitation with the anti-cdk4 antibody Ab-5 or control IgG bound to magnetic beads. After extensive washes, immunoprecipitates were resuspended in kinase buffer (), and the kinase assay was performed on a GST-Rb fragment substrate (Santa Cruz Biotechnology, Inc.) as previously described (). Labeled proteins were resolved on a 4–12% polyacrylamide gel and detected by autoradiography. To measure cyclin D1– and cyclin D3–associated kinase activities in C2C12 cells, myoblasts and myotubes were lysed in cdk4 immunoprecipitation buffer and treated as described for cdk4 immunoprecipitation. 350 μg of protein extract was subjected to 3-h immunoprecipitation with anti–cyclin D1 (72-13G; Santa Cruz Biotechnology, Inc.) or –cyclin D3 (C-16; Santa Cruz Biotechnology, Inc.) antibodies or control IgG bound to protein G–Sepharose beads (GE Healthcare). Kinase assay and autoradiography were performed as previously described (). To assess cyclin E–associated kinase activity, C2C12 myotubes were lysed in immunoprecipitation buffer, briefly sonicated, and centrifuged (). 3 mg of protein extracts were immunoprecipitated for 3 h with an anti–cyclin E antibody (M-20; Santa Cruz Biotechnology, Inc.) or control IgG bound to magnetic beads. Immunoprecipitations were performed as previously described (). Labeled proteins were resolved on a 4–12% polyacrylamide gel and detected by autoradiography. To characterize cdk4 and cdk6 complexes in human fibroblasts, proliferating and quiescent FB1329 cells were lysed in cdk4 immunoprecipitation buffer and treated as described for cdk4 immunoprecipitation. 1 mg of whole-cell lysates were subjected to 3 h of immunoprecipitation with anti-cdk4 or -cdk6 antibodies or control IgG bound to protein G–Sepharose beads. Pictures were obtained using a fluoresence microscope (Axioskop 2; Carl Zeiss MicroImaging, Inc.) with a 20× NA 0.50 or a 40× NA 0.75 plan Neofluar objective. Images were digitized with a camera (AxioCam; Carl Zeiss MicroImaging, Inc.) and acquired with AxioVision 3.1 software (Carl Zeiss MicroImaging, Inc.). Images were contrast enhanced by applying the brightness/contrast regulation of Photoshop 7.0 software (Macintosh version) to the whole image. Double immunofluorescence images were obtained by superimposing two single-color pictures with Photoshop (Adobe). In immunofluorescence procedures, mAbs to BrdU (Bu20a clone; DakoCytomation) and myosin heavy chain (MyHC; MF20, ) were used. AlexaFluor488- or 594-conjugated antisera to mouse Igs were obtained from Invitrogen. Nuclei were counterstained with Hoechst 33258. HEK cultures were stained for acidic β-galactosidase activity as described previously (). siGenome SMARTpool reagents (Dharmacon) composed of four siRNAs were used to interfere with the following human and mouse transcripts: CDKN1A (p21), CDKN1B (p27; human only), CDKN1C (p57), CDKN2A (p16), CDKN2B (p15), CDKN2C (p18), CDKN2D (p19), CCND3 (cyclin D3), and CCND1 (cyclin D1). The antisense sequence of the Individual siGenome duplex (Dharmacon) for mouse p27 was 5′-P.UAUCCCGGCAGUGCUUCUCUU-3′. Individual siRNAs for mouse p21 and cyclin D3 (siGENOME SMARTpool Upgrades; Dharmacon) were used in some experiments; the antisense sequences are as follows: p21, #1 (5′-P.CAAAGUUCCACCGUUCUCGUU-3′), #2 (5′-P.UUUCGGCCCUGAGAUGUUCUU-3′), #3 (5′-P.CAACGGCACACUUUGCUCCUU-3′), and #4 (5′-P.GAACAGGUCGGACAUCACCUU-3′); and cyclin D3, #1 (5′-P.AAUCACGGCAGCCAGGUCCUU-3′), #2 (5′-P.UAUAGAUGCAAAGCUUCUCUU-3′), #3 (5′-P. CUCGCAGGCAGUCCACUUCUU-3′), and #4 (5′-P.UACCUAGAAGCUGCAAUUGUU-3′). The siCONTROL Non-Targeting siRNA #2 (Dharmacon) was transfected as a control. Proteins were separated on 4–12% polyacrylamide gels and analyzed by Western blotting with the following mAbs or pAbs: mAbs to cyclin D1 (72-13G; Santa Cruz Biotechnology, Inc.), cyclin D2 (DCS-3; Abcam), p16 and human cdk4 (both Ab-1; NeoMarkers), p27 (Transduction Laboratories), and β-tubulin (Sigma-Aldrich); pAbs to mouse cdk4 (Ab-5; NeoMarkers), p18 (Biosource International), and p16, p21, p57, cyclin D3, cdk2, cdk6, and cyclin E (Santa Cruz Biotechnology, Inc.). Peroxidase-conjugated antisera to mouse and rabbit Igs were purchased from Bio-Rad Laboratories. Western blots were developed using the SuperSignal kit (Pierce Chemical Co.). In immunofluorescence procedures, mAbs to BrdU (Bu20a clone; DakoCytomation) and MyHC (MF20; ) were used. AlexaFluor-conjugated antisera to mouse Igs were obtained from Invitrogen. Nuclei were counterstained with Hoechst 33258. HEK cultures were stained for acidic β-galactosidase activity as described previously (). In semiquantitative RT-PCR procedures, total cellular RNA was extracted from MSC-derived myotubes with TRIzol reagent (Life Technologies). cDNA synthesis and PCRs were performed starting from 1 μg of total cellular RNA using the GeneAmp Gold RNA PCR Reagent kit (Applied Biosystems). Individual CKIs were amplified along with mouse β-actin in the same test tube. PCRs proceeded for 25 or 30 cycles (p15 or p16 and p19, respectively) on a personal machine (Mastercycler; Eppendorf). The following primers were used: p15 forward, 5′-GCCCAATCCAGGTCATGATGATG-3′; p15 reverse, 5′-GATGGGGCTGGGGAGAAAGAAG-3′; p16 forward, 5′-GTCGCAGGTTCTTGGTCACTG-3′; p16 reverse, 5′-CGCACGATGTCTTGATGTCCC-3′; p19 forward, 5′-GTCTGCGTCGGCGACCGGTTG-3′; p19 reverse, 5′-CAGGAGCTAGGAAGCTGACCACG-3′; β-actin forward, 5′-TGTTACCAACTGGGACGACA-3′; and β-actin reverse, 5′-CTTTTCACGGTTGGCCTTAG-3′. For absolute quantitation of CKIs in C2C12 and human fibroblasts, the following recombinant-tagged proteins were used as standards: mouse GST-p21, poly-His-p27, and human GST-p16, -p21, and -p27. A dilution curve of each reference protein was run on an SDS-PAGE gel along with the cell extracts or immunoprecipitates to be quantitated, and the proteins were analyzed by Western blotting. Developed films were computer scanned and subjected to densitometry. The data thus acquired and the unknown samples (lysates or immunoprecipitates) were plotted as exemplified in Fig. S4 F, yielding raw quantity estimates that were corrected to take into account the difference in the molecular weight of natural proteins and recombinant standards. To directly compare the kinase activities associated with cyclin D1 and D3 (), the densitometric measurements of pertinent bands (GST-Rb) were adjusted by subtracting their respective backgrounds (IgG). The adjusted values were then normalized for immunoprecipitation efficiency, which was calculated as follows: OD/OD × μg/μg × 100, where OD = OD of the immunoprecipitated protein (arbitrary units), OD = OD of the protein band in the total lysate (arbitrary units), μg = total lysate proteins loaded (micrograms), and μg = amount of total lysate subjected to immunoprecipitation (micrograms). Fig. S1 shows p21 RNAi specificity and the knockdown efficiencies of other CKIs in mouse myotubes. Fig. S2 shows a characterization of cyclin–cdk–CKI complexes and quantification of total and cdk4-bound p21 and p27 in mouse C2C12 cells. Fig. S3 shows cyclin D3 RNAi specificity. Fig. S4 shows the expression of D-type cyclins, cdk4/6, p21, p27, and p16 and cyclin D–cdk4 complexes in proliferating and quiescent human fibroblasts. Online supplemental material is available at .

Cilia and flagella are highly conserved microtubule-based organelles that perform a wide array of crucial motile and sensory functions in many types of cells (for review see ). is a unicellular green algae that has been used extensively to study ciliary/flagellar function and assembly. Each cell has a pair of equal length flagella whose length is tightly monitored and regulated. When cells are induced to shed their flagella, they regenerate flagella rapidly to the predeflagellation length within 90 min (). After amputation of one of the two flagella, the remaining one shortens and waits for the other one to regrow to the same length; both then grow out to the predeflagellation length. The most striking example of the active regulation of flagellar length occurs when wild-type (WT) cells are mated to mutant cells with abnormally long flagella. Within minutes after cell fusion, the long flagella shorten to the WT length (). These observations demonstrate the existence of a vigorous regulatory mechanism that assesses and enforces flagellar length. Flagella are dynamic structures that undergo continuous assembly and disassembly, mainly at their distal ends (; ). The steady-state length of flagella is likely to be the result of equilibrium between flagellar assembly and disassembly. A wealth of experimental evidence indicates that flagellar assembly and maintenance require intraflagellar transport (IFT), a kinesin/dynein-based transport system that involves at least two protein complexes of >17 polypeptides (; ). IFT particles have been observed to associate with flagellar proteins and preassembled complexes () and to move at defined rates up and down the flagella (; ; ). Recent studies indicate that IFT is involved in the transport of signaling molecules (; ) and in Hedgehog signaling in mouse primary cilia (). The compartmentalization of IFT particles can also be modulated in response to flagellar adhesion during mating in (). Because IFT is essential for flagellar assembly, it is a likely target of regulation for controlling the length of flagella. One model for length control proposes that the length of flagella is governed by intrinsic properties of IFT that determine the extent of flagellar assembly by balancing rates of assembly and disassembly (). Genetic studies demonstrate that flagellar length is regulated by specific protein products (; ; ). There are four genetic loci ( (), , , and ) at which mutations result in abnormally long flagella, often two to three times the normal length (; ; ). The mutant has very long flagella and regrows flagella very slowly after deflagellation. Five mutant alleles of have been identified, and they cause varying degrees of excessive flagellar length and defective flagellar regeneration. Four previously described mutant alleles cause the assembly of long flagella, but they can regenerate flagella normally. Recently, we described two new null mutations at that confer a distinct unequal length flagella phenotype; the two flagella are different in lengths on most mutant cells (). The null mutants also regenerate flagella very slowly and have prominent swellings at the distal ends of their flagella that are filled with IFT-like particles. About a dozen mutants, which are isolated after DNA insertional mutagenesis, have very long flagella but can regrow flagella with WT kinetics after deflagellation. The gene products of three of these genes have been identified. and encode novel proteins of unknown function (; ). encodes a MAPK (), providing the first evidence that protein kinase pathways are involved in flagellar length control. Recent studies also implicate glycogen synthase kinase 3, an aurora kinase, and an NIMA-related kinase in the regulation of flagellar assembly and disassembly (; ; ). Similar mechanisms for regulating the length of cilia/ flagella may exist in other organisms. For example, in , the length of flagella can be shortened or increased by the overexpression or deletion of a MAPK (). In sea urchin blastula, cilia at the apical tuft are two to three times longer than cilia present on the rest of the embryo (). In mammals, motile cilia within the same organ can be different in length depending on their location (), and the length of primary cilia varies with the diameter of bile ducts (). Perturbation of ciliary length has been shown to correlate with human diseases such as primary ciliary dyskinesia (). Recently, it has been shown that mice with a particular form of juvenile cystic kidney disease have kidneys with abnormally long primary cilia (). In this study, we cloned and identified its gene product as a new member of the cyclin-dependent kinase (CDK) family. CDKs have attracted intense research interest because many of them play essential roles in cell cycle progression (). In addition, CDK5 performs multiple important functions in terminally differentiated neuronal cells (). In the present study, we identify a new function for this class of kinases in regulating the size and development of an organelle. Although a close homologue of LF2p has not been identified in other organisms, several CDK-related kinases of unknown functions are highly expressed in the testis, in which germ cells are differentiating into flagellated sperm cells (; ), raising the possibility that these kinases may be the counterparts of LF2p in regulating flagellar length in higher organisms. The current study will inspire a new direction for exploring additional roles of CDKs in nondividing cells. cells have a pair of equal length flagella that never exceed 16 μm (). Cells containing any of the five previously identified mutant alleles assemble extra-long flagella up to 30 μm (). Using DNA insertional mutagenesis, we have identified a new mutant allele () whose flagellar length defect is very similar to that of the two null mutants of (). Although some cells had flagella longer than 16 μm, most of the cells possessed stumpy or unequal length flagella (). The length of both flagella on each cell in a population of cells was measured, and the majority of cells had a very short flagellum (<2 μm) and a longer flagellum (). Moreover, there was an obvious defect in the morphology of the mutant flagella: instead of having a tapered shape, the distal ends of all flagella, whatever their lengths, were swollen (, arrows). Examination of these flagella by thin-section EM revealed the accumulation of IFT-like particles at these distal flagellar bulbs (). Western blot analysis of flagella isolated from WT and cells also revealed the overaccumulation of IFT proteins in the mutant flagella (unpublished data). WT cells can regenerate their flagella rapidly and synchronously after they shed or resorb their flagella, reaching full length by 90 min after flagellar amputation. Several of the alleles with a long flagella phenotype are unable to regrow their flagella after deflagellation, or they regenerate only very slowly (). The mutant showed the most severe defect in flagellar regeneration. All of the cells were unable to grow flagella beyond a short stump of ≤1 μm () even hours after flagellar excision. The tagged mutant enabled us to identify genomic DNA clones that rescued the mutant phenotype upon transformation. A 7.5-kb XhoI–XhoI fragment containing a predicted CDK gene was able to complement the mutant phenotype upon transformation into cells (). By combining EST sequences with RT-PCR and 5′ rapid amplification of cDNA ends (RACE) analysis, we were able to obtain a complete 2,244-bp cDNA sequence for (genomic and cDNA sequences are available from GenBank/EMBL/DDBJ under accession nos. and ). When cloned into a gene expression cassette, the 1.1-kb coding region of this putative cDNA rescued the mutation by transformation. The predicted gene encodes a protein of 354 aa with a calculated mass of 37,546. The N-terminal 300 aa include the 11 subdomains that are characteristic of all serine/threonine kinases (). BLAST searches indicated that LF2p is most homologous to members of the CDK family. The percentages of identity/similarity to rat CDK2, CDK4, CDK5, and CDK7 are 40/58%, 37/53%, 34/53%, and 39/53%, respectively. LF2p is most similar (47% identity and 61% similarity) to rat PNQARLE, a CDK-related kinase of unknown function (). In addition to the overall sequence homology, PNQARLE and LF2p both have a C-terminal extension that is absent in many other CDKs. LF2p contains all of the aa known to be critical for the catalytic activity of CDK kinases, including the glycine-rich loop (GXGXXG) for ATP binding in subdomain I and the invariant lysine (position 41) for phosphotransfer in subdomain II (). In CDK2, phosphorylation of threonine at position 14 and tyrosine at position 15 within the glycine-rich loop inhibits its activity, whereas phosphorylation of threonine at position 160 within subdomain VIII activates the kinase (). LF2p lacks the threonine and tyrosine residues in the glycine-rich loop but retains the second threonine at position 168 (). Another feature of many CDKs is the α helix (subdomain III) containing the PSTAIRE motif required for cyclin binding. Cyclins expressed at various phases of the cell cycle regulate the level and activity of CDKs (). In LF2p, a unique sequence, PDVVVRE, replaces the PSTAIRE motif, indicating that LF2p probably does not interact with cyclins and, therefore, is classified as a CDK-related kinase. Expression of the transcript in WT and various mutant cells was analyzed on RNA blots with a DNA hybridization probe from the predicted coding region (). A 2.4-kb RNA was detected in WT cells (CC-620) as well as in long flagella mutants , , , and . The transcript was not observed in the insertional mutant , strongly suggesting that is a null mutant. Interestingly, the transcript level was substantially reduced in , a null mutant of , and in two double mutants, and , all with unequal length flagella. This observation shows that accumulation of the RNA could be affected by null mutations in or by double mutations of , , or . During flagellar regeneration, the transcript levels for flagellar proteins increase rapidly and transiently (). We examined the transcript level of at various times after flagellar amputation in WT cells. In contrast to flagellar gene transcripts such as , whose level increased during flagellar regeneration, the amount of RNA appeared to decrease early during regeneration (), suggesting that LF2p is unlikely to be a flagellar protein. Attempts to generate antibodies to bacterially expressed LF2 protein or to synthetic peptides were unsuccessful. Therefore, we tagged the cDNA construct with the HA epitope so that LF2p could be detected with a commercially available HA antibody. The HA-tagged construct rescued the mutants as efficiently as the untagged version. When total proteins from several rescued strains were analyzed on Western blots using an HA antibody, a cluster of at least three protein bands migrating at ∼42 kD was detected only in cells rescued with the HA-tagged construct (, top and bottom). The amount of the higher mol wt isoforms decreased, whereas the level of the smallest mol wt isoform increased when cell extracts containing the HA-tagged proteins were incubated at 37°C (, no PPI). However, this size change was inhibited if phosphatase inhibitors were included during the preparation of cell extracts (, +PPI). These results may be explained by the presence of phosphatase activity in cell extracts and because the higher mol wt forms of LF2p are phosphorylated forms of LF2p. The LF2p isoforms were resolved by 2D gel electrophoresis. The two higher mol wt forms of LF2p were focused at a lower pH (), as would be expected if they are phosphorylated forms of LF2p. pI values of the isoforms were estimated by their positions on the first dimension. On three separate gels, estimated pI values between the largest and smallest isoforms were within 0.6 of each other, which can be accounted for by the possible addition of one to three phosphate groups to the protein. Because plays a regulatory role in flagellar assembly, we examined the distribution of LF2p in the cell body and in the flagella. The majority of LF2p was detected in the cell bodies, although a very small amount was consistently observed in flagella (). The presence of LF2p in these flagellar samples could not be explained by the presence of contaminating cell bodies in these samples because very little staining with antibodies to a chloroplast protein, OEE1 (), was observed with these samples (). The amount of LF2p in flagella isolated from 3.6 × 10 cells was about the same as the amount contained in 10 cell bodies, indicating that ≤0.3% of LF2p was distributed in flagella (). Some of the flagellar form of LF2p was detergent insoluble and remained with the axonemal fraction after 1% NP-40 extraction. It is noteworthy that there was no difference in the relative amount of the different LF2p isoforms in flagella versus cell bodies. Previous studies indicated that the protein products of and are predominantly localized inside the cell, probably in protein complexes (; ). When cells expressing HA-LF2p or HA-LF3p were analyzed in parallel by immunofluorescence, similar punctate staining of both proteins was observed inside the cell bodies (). No flagellar staining was detectable in either case. To test whether LF2p may form a complex with the other LF proteins, we studied the sedimentation of HA-tagged LF2p or LF3p from cell extracts on sucrose density gradients. LF2p and LF3p copurified in the same fractions, with the majority of the proteins sedimenting in fractions 9–13, around 11S (, top). Previously, we found that LF1p cosedimented with LF3p in the same sucrose density fractions (). In contrast, when the sedimentation of LF4p was compared with LF2p in another experiment, most of the LF4p sedimented away from LF2p in fractions 6–9 (, bottom). These results suggest that LF2p may be a member of a protein complex involving LF1p and LF3p but not LF4p. We used a GAL4 yeast two-hybrid system to test for direct interactions among LF proteins. A 1.1-kb cDNA of and full-length coding regions of , , and were cloned into yeast vectors, and their interaction was determined by the level of β-galactosidase activity (). LF2p interacted specifically with LF3p and LF1p. Interaction between LF1p and LF3p was also detected. In contrast, LF4p did not interact with any of the three LF proteins. Five previously identified alleles differ in the severity of their flagellar length defects and their ability to regenerate flagella after deflagellation (see description in the following two paragraphs; ). We determined the sequences of four of the mutant alleles () and correlated the lesions with the severity of their mutant phenotype (). In , the 5′ donor site of intron 1 was changed (GT to AT), resulting in an abnormal splicing event at a position 26 bp downstream. The 26-bp addition to the mature RNA caused a shift of the reading frame, thereby introducing incorrect aa and a premature translational stop. Therefore, is expected to make no protein or, at best, a nonfunctional truncated product that includes only two kinase subdomains. Although cells have been reported to have long flagella, our reexamination of two available strains revealed that many of these cells had unequal length flagella that have swollen distal tips similar to the null mutant phenotype of cells (unpublished data). mutant cells also did not regenerate flagella after amputation (). In , a nucleotide change at the 3′ acceptor site of intron 2 changed the consensus AG to TG. As a result, splicing took place 6 bp downstream at the next AG site, removing a glycine and a glutamine located in the junction between subdomains IV and V of the protein. These two residues are not well conserved among different CDKs (). has the weakest mutant phenotype of all alleles. Only 4–30% of cells displayed long flagella (16–20 μm), and flagellar regeneration was only slightly delayed in relative to WT (; unpublished data). In , two nucleotide changes were found: one in the 5′ splice site (GT to AT) and one in the 3′ splice site (AG to AC) of intron 1. As a consequence, intron 1 was retained in the mature RNA, and 39 aa were inserted in subdomain I right after the glycine-rich loop. The insertion in is expected to disrupt the structure of the protein and to affect its activity. The mutant strain has a severe phenotype: >50% of cells have abnormally long flagella (), and many cells did not regenerate flagella until 2–3 h after deflagellation (). In , a nonsense mutation changing the codon TGG to TGA was found. The mutant protein is predicted to contain 233 aa lacking the last kinase subdomain XI and the C-terminal tail. The function of subdomain XI and the C-terminal tail is not known, but the truncated protein appears to retain partial function, as has a moderate flagellar length defect and can regenerate flagella with nearly normal kinetics (). Analysis of mutants suggests that the catalytic kinase function of is important for the regulation of flagellar length. To directly test the role of the kinase activity of LF2p in flagellar length control, we engineered specific mutations in the HA-tagged gene construct that should affect only the catalytic activity but not the overall structure of LF2p. One change converted the invariant lysine at position 41 to an arginine (K41R), a change used routinely to create kinase-dead versions of various kinases (Snyder et al., 1985; ). In a second construct, we converted the second glycine in the glycine-rich loop to a valine (G21V), a change that can greatly diminish the kinase activity of the mutated enzyme (). The WT construct and the two mutant constructs were introduced into the null mutant by transformation to test for the effect of the mutations on the function of the protein. Although the WT construct could rescue the phenotype completely, the mutant constructs could not. Nevertheless, cells expressing the mutant constructs were clearly distinguishable from because many of the transformed cells assembled flagella (). Notably, >50% of cells containing the K41R construct and >20% of cells containing the G21V construct still possessed flagella longer than 16 μm (). In addition, although could not regenerate flagella after deflagellation, cells harboring the mutant gene constructs were able to regrow flagella slowly (). It is noteworthy that cells transformed with the K41R construct displayed a flagellar length and regeneration phenotype almost identical to that of the mutant (), in which an insertion was predicted to disrupt the kinase function of the protein. These results clearly demonstrate the importance of the kinase activity of LF2p for maintaining proper flagellar length. In contrast, the G21V and K41R constructs were able to support flagellar assembly in the null mutant background. One possible explanation for this result is that LF2p supports flagellar growth by a function independent of its kinase activity, such as facilitating the assembly of a protein complex. Alternatively, the mutated constructs may have residual kinase activity that is sufficient to support flagellar assembly but not proper length control. To distinguish these two possibilities, we combined both G21V and K41R mutations in the same construct to further decrease the kinase activity of the protein without affecting the LF2 protein in other ways. When transformed into cells, the double mutant construct was unable to rescue the null mutant. We conclude from these results that the single mutant constructs allow flagellar assembly because they retain partial kinase activity. The protein products from six K41R and G21V transformant lines were examined by Western blot analysis. In all samples, only a single band was observed at the position of the fastest migrating isoform of LF2p (K41R transformants are shown in ). The higher mol wt isoforms, which are the presumed phosphorylated proteins (), were missing, suggesting that LF2p kinase activity may be required directly or indirectly to phosphorylate LF2p. In this study, we show that the gene encodes a novel CDK-related kinase. Analysis of an allelic series of mutants exhibiting different degrees of flagellar length and flagellar growth phenotypes has provided useful insights into the functions of . Three of the mutants that display a long flagella phenotype—, , and —contain mutations that are expected to produce a hypomorphic effect caused by reduced catalytic activity of the protein. The severity of the mutant phenotype in these three mutants appears to correlate with the extent of disruption in the kinase domain of the proteins. On the other hand, when LF2p is completely absent, as in the null mutant , cells exhibit multiple defects: a reduced ability to assemble flagella, as reflected by the high percentage of stumpy flagella cells in a population; an inability to maintain an equality of length between the two flagella of a cell; an inability to regenerate flagella after deflagellation; and an accumulation of IFT-like particles at the distal ends of flagella, where flagellar assembly and disassembly occur. Based on these results, we conclude that LF2p plays multiple roles in flagellar assembly, including support of flagellar growth, balance of length between the two flagella of a cell, and enforcement of normal flagellar length. Nine CDK proteins have been annotated in the genome (). LF2p, which is annotated as CDKI1, appears to have evolved to perform specific functions in flagellar assembly. Although no orthologue of LF2p can be identified in other organisms, it is most similar to the mammalian CDK-related kinase PNQALRE in sequence identity and overall structural organization. PNQALRE was first proposed to be a CDK-activating kinase that phosphorylates other CDKs (). However, a more recent study reported that the CDK-activating activity is caused by the association of PNQALRE with CDK7 (). Interestingly, PNQALRE mRNA is most abundant in testes (), the site of assembly of sperm tail axonemes, raising the possibility that PNQALRE may play a similar role to LF2p in ciliogenesis. Some proteins with conserved kinase domains do not require kinase activity to function (). We used site-directed mutagenesis to test the role of LF2 kinase activity in vivo. A change at the second glycine in the highly conserved ATP-binding domain or a change in the invariant lysine in subdomain II often produces inactive protein kinases. Both mutant proteins fail to maintain flagellar length control, indicating that the kinase activity of LF2p is required to enforce flagellar length. Importantly, these mutations do not completely abolish the function of LF2p because the transformed cells were able to assemble flagella. It is possible that these mutations do not totally eliminate the catalytic activity of LF2p. There is a precedent that even mutating the invariant lysine could leave residual catalytic activity in some kinases (), and there is at least one example of an active kinase that lacks this conserved lysine (). When both G21V and K41R mutations were introduced in a single construct to further reduce the kinase activity of LF2p, the double mutant construct became completely nonfunctional in rescue of the mutant phenotype. Because these mutations should only affect the phosphotransfer reaction but not the structure of the protein, the kinase activity of LF2p is the only activity required to carry out its functions. In addition, it appears that a higher level of kinase activity is required for flagellar length control, whereas a lower level of kinase activity is sufficient to support flagellar growth. We are developing in vitro kinase assays to test the predictions from these experiments. Most CDKs are phosphorylated by other kinases, and a few can phosphorylate themselves (). For LF2p, there are at least three isoforms. The higher mol wt forms may be phosphorylated proteins, as they could be converted to the lowest mol wt form by a phosphatase-like activity in cell extracts and were focused at a lower pH on 2D gels. Interestingly, only the smallest isoform of the protein was detected in strains expressing the low activity K41R and G21V constructs, raising the possibility that LF2p may autophosphorylate or that there is a feedback loop that regulates LF2p phosphorylation. Previous genetic studies indicate that the gene products of , , and may work together to regulate flagellar assembly (; ). Double mutants of and any of the hypomorphic alleles of or produce a synthetic stumpy flagella or unequal length flagella phenotype (; ). Although the length defect of mutants can be complemented rapidly when these cells fuse with WT cells to form temporary dikaryons during mating, all pairwise crosses of , , and alleles fail to restore their flagellar length control in dikaryons (), indicating that some common structure or process is defective in these mutants. The resemblance of null or hypomorphic mutants in and also lends support to the idea that these genes work in similar pathways. During flagellar regeneration, the level of RNA transcript for all three genes did not increase but rather decreased transiently during the first 30 min after deflagellation (; ; this study). In addition, the RNA transcript of was diminished in the -null mutant and in double mutants of and , and a similar reduction in the RNA transcript of has also been observed in the null mutant of and in double mutants (). These almost identical accumulation patterns of transcripts suggest that the expression of these three genes is coordinated. At the protein level, the LF1, LF2, and LF3 proteins were localized by immunofluorescence to similar cytoplasmic foci, and these proteins also cosedimented on sucrose density gradients, suggesting that they may form a protein complex. Using yeast two-hybrid assays, we demonstrated the specific interaction of these proteins in vivo. Based on all of these results, we propose that LF1, LF2, and LF3 proteins work together as a protein complex in the cytoplasm, which we call the length regulatory complex (LRC) because of its role in regulating flagellar length. We hypothesize that LF2p is the catalytic subunit of the LRC, with LF1p and LF3p being the accessory proteins for activation or recruitment. Many CDKs are inactive as a monomer and require binding to cyclins for activation. All cyclins contain a conserved region of 100 aa known as the cyclin box that binds the PSTAIRE helix of CDKs (). Noncyclin binding partners also exist for CDKs. One class of noncyclin regulatory proteins is distinct from cyclins in their primary sequence but have a similar tertiary structure to cyclins. For example, p35/p25 binds and regulates the activity of CDK5, a kinase with important functions in neuronal cells (). Other noncyclin regulatory proteins that bind CDKs have no structural similarity to cyclins: for example, the Ringo/Speedy proteins that interact with CDK1 and CDK2 (), PIF-1B and PIF2 that bind to the N-terminal extension of the CDK-related kinase PFTAIRE (), and MAT1, a RING finger protein that can stabilize and alter the substrate specificity of a CDK7–cyclin H complex (; ). LF1p and LF3p are novel proteins that are distinct from cyclins, and they are likely to be new examples of noncyclin binding partners for CDKs. Previously, our laboratory identified a MAPK, LF4p, as a regulator of flagellar length (). LF4p, unlike LF2p, does not affect flagellar assembly. Null mutants can assemble flagella both during the cell cycle and after deflagellation (). In addition, cell fractionation studies show that LF4p is enriched in flagella (), but very little of LF1, LF2, and LF3 proteins are located in the flagella (; ; this study). We do not know how LF4p localizes inside the cell because antibodies to this protein do not work for immunolocalization, but the cytoplasmic form of the LF4 protein sedimented in lighter fractions in sucrose density gradients than the other LF proteins. Previously, it was shown that mutations can suppress the regeneration phenotype in double mutants, and mutations can suppress the stumpy flagella phenotype of double mutants to produce long flagella (). However, we are doubtful that LF4p works directly downstream from the LRC because the mutation cannot suppress the null or mutant phenotype (unpublished data). Our yeast two-hybrid experiments also did not show any direct interaction of LF4p with LRC proteins. It is more likely that the kinase pathway and LRC pathway work in parallel to regulate flagellar length, with some cross talk between them. CDKs have diverse roles and many phosphorylation targets (). The complex phenotypes of mutants suggest that the LRC may also have multiple targets. One obvious level of regulation is RNA accumulation. We observed a decreased level of transcript accumulation for the and genes in the null mutants or double mutants of , , and (; this study). Many CDKs have been shown to be part of the transcriptional protein complex, to phosphorylate the C-terminal domain of the RNA polymerase large subunit, or to be part of RNA splicing complexes (for review see ). It will be important to determine whether the LRC regulates these cellular functions. Another possible target for the LRC is the IFT machinery. mutants, especially the null mutants, show an overaccumulation of IFT particles at the distal ends of their flagella, the major location for flagellar assembly and disassembly. A similar abnormal accumulation of IFT particles at flagellar tips occurs in mutants defective in specific IFT components (; ; ; ). Moreover, some IFT mutants show unequal length flagella (; ), leading to the hypothesis that LF1, LF2, and LF3 proteins may regulate IFT (). measured the rate of IFT particle movements on various mutant flagella and did not detect any major difference in the kinetics of retrograde or anterograde transport in mutants. Recently, demonstrated that the cargo-carrying capacity of IFT particles could be modified. When flagella were induced to resorb by chemical treatment, a large amount of cargo-free IFT particles entered into the shortening flagella and carried disassembled flagellar components back to the cell. It is possible that the LRC regulates qualitative properties of IFT such as cargo loading, the unloading capacity of IFT particles, or the activation/inactivation switch between the anterograde/retrograde motors that occurs at the distal end of flagella. Because the LRC is localized mainly in the cytoplasm, a satisfactory model must explain how IFT is modified in the cytoplasm to affect its function in flagella. The identity of the LRC as a protein kinase complex suggests that phosphorylation may be the mechanism. Little is known about how the IFT particles are assembled and transported inside the cytoplasm before they enter the flagella. Based on the similarity of domain structures between IFT proteins and the protein components of intracellular coated vesicles, it has been postulated that the IFT machinery may have originated as specialized membrane vesicles (). There are many examples of CDKs or CDK-related kinases that phosphorylate intracellular membrane components (; ; ; ). It is not difficult to envision the use of a specialized protein kinase complex, the LRC, to regulate a specialized transport system for cilia and flagella. WT strain (CC-620), mutant (CC-803 and CC-912), (CC-2288), and (CC-2287) were obtained from the Genetics Center. (cs89) and were laboratory strains that were previously described (). is an insertional mutant tagged with the nitrate reductase () plasmid (). An mutant was crossed with or to create double mutant strains for transformation. DNA from was cloned into a λ-bacteriophage vector using the Lambda Fix II/XhoI partial fill-in vector kit (Stratagene) according to the manufacturer's instructions and was screened with P-labeled pUC119 DNA to identify clones containing genomic sequences flanking the integrated plasmid. A fragment from one of the phage clones was used to identify eight clones from a bacterial artificial chromosome (BAC) library of genomic DNA (available from the Clemson University Genomics Institute). Crude DNA from BAC clones was prepared by alkaline lysis. A fragment from one of the BAC clones was used as a hybridization probe to identify a 20-kb genomic clone, λ2-4, from a WT λ-phage library. To test for the presence of the gene in these clones, BAC or λ-DNA was cotransformed with pArg7.8 () into strains containing or mutations as described previously (). The sequence of a 17-kb region around was determined by sequencing plasmid subclones of λ2-4 using universal sequencing primers and gene-specific primers (Advanced Genetics Analysis Center, University of Minnesota). The complete cDNA sequence was obtained and verified using three different approaches: RT-PCR and 5′ RACE products, two ESTs available from the Genome project, and PCR products amplified from a gametic cDNA library (provided by B. Snell, University of Texas Southwestern Medical Center, Dallas, TX). PCR, 5′ RACE, RNA analysis, and quantitation were performed as described previously (). The genetic lesions in mutants were determined by direct sequencing of genomic PCR and RT-PCR products from the mutant alleles. No sequence change could be found in (CC-2288), and it no longer displayed any mutant phenotype, suggesting that the mutation may have reverted. The coding region of was amplified from a gametic cDNA library with a primer with an NdeI site (5′-CCGCATATGCCGTCGACGCTTCAAGGC-3′) and a primer with an EcoRI site (5′-GACTGAATTCTCACACGAGCGGCAATGACG-3′) using PfuUltra polymerase (Stratagene) and was cloned into pPCR2.1 (Invitrogen). A resulting clone was digested with NdeI–EcoRI, and the 1.1-kb insert was cloned into a psaD promoter/3′ untranslated region cassette () to generate the cDNA clone psaD-LF2. To add the HA epitope to psaD-LF2, the HA cassette was excised from p3XHA () and cloned into a unique PmlI site 16 aa before the translational stop codon. Lysine 41 and glycine 21 were mutagenized to an arginine and a valine, respectively, according to protocols from the QuikChange II XL Site-Directed Mutagenesis kit (Stratagene). All clones were sequenced to ensure that no error was introduced during the amplification or cloning processes. Cells were induced to shed their flagella by pH shock and were grown under bright light with shaking or stirring. Cells were taken at different times, fixed with an equal volume of 2% glutaraldehyde, and examined with differential interference contrast (DIC) microscopy optics using a microscope (Diaplan; Leica) with a 100× NA 1.25 objective (Leitz). Images of cells were captured using a video camera (CCD 72; Dage-MTI, Inc.) and Image 1.59 software (Scion). Flagellar length was measured with Image J version 1.31 software (National Institutes of Health). Thin-section EM of whole cells was performed as described previously using a microscope (1200EXII; JEOL; ). All DIC and EM images were assembled using Photoshop CS2 software (Adobe). To determine the distribution of LF2p in flagella, cells were harvested by centrifugation, resuspended in 10 mM Hepes and 5 mM MgSO4, and deflagellated by pH shock. Cell bodies were collected by low speed centrifugation. The supernatant fraction containing flagella was underlaid with 25% sucrose and centrifuged to remove the remaining cell bodies. Purified flagella were collected by centrifugation at 23,000 . Half of the flagella was further extracted twice with 1% NP-40 in buffer to prepare axonemes. Protein concentration was determined using the Protein Assay kit (Bio-Rad Laboratories). Whole-cell or cell body protein samples were prepared by boiling 5 × 10 cells in 0.5 ml SDS-PAGE buffer (62.5 mM Tris, pH 6.8, 5% 2-mercaptoethanol, 2% SDS, 10% glycerol, and 0.05% bromophenol blue). Flagella were boiled in SDS-PAGE buffer to a final concentration of 2 mg/ml. Soluble proteins were extracted from cells by two to three cycles of freezing/thawing and were analyzed on sucrose density gradients as described previously (). Phosphatase inhibitors (50 mM NaF, 25 mM β-glycerophosphate, and 1 mM sodium orthovanadate) were added to the extraction buffer except when noted otherwise. All protein samples were size fractionated by SDS-PAGE and transferred to polyvinylidene difluoride (PVDF) membranes. Primary antibodies used were a rat HA antibody (3F10; 1:1,200; Roche Biochemicals), a rabbit anti-OEE1 (1:3,000), and a rabbit antibody to LF4p (). Secondary antibodies are an anti–rat HRP at 1:8,000 and an anti–rabbit HRP at 1:24,000 dilutions (Sigma-Aldrich). Blots were analyzed sequentially with different antibodies without stripping. ECL reagents (GE Healthcare) were used for detection. Immunofluorescence was performed on methanol-fixed cells using the HA antibody at a 1:400 dilution and an AlexaFluor488 fluorochrome–conjugated secondary antibody at a 1:500 dilution (Invitrogen) according to procedures described previously (). Images were captured on a CCD camera (CoolCam; Cool Camera Co.) and assembled using Photoshop CS2 (Adobe). The interaction of LF proteins was studied according to procedures detailed in the Yeast Protocols Handbook (CLONTECH Laboratories, Inc.). The full-length coding region of , , and and a partial cDNA encoding aa 1–330 were cloned into plasmid vectors pGBKT7 (GAL4 DNA-binding domain) and pGADT7 (GAL4 activation domain) and tested in the yeast strain SFY526. All LF constructs were also tested alone or in combination with control plasmids containing murine p53 or SV40 large T-antigen to ensure that they did not activate the reporter gene without a specific interacting partner. Expression of the fusion proteins in transformant yeast strains was confirmed by Western analysis using anti-Myc and -HA antibodies. The interaction was quantified by liquid β-galactosidase assay using -nitrophenyl β--galactopyranoside as the substrate and was expressed in Miller units.

Spinal muscular atrophy (SMA) is a genetic disorder associated with recessive loss-of-function mutations in the human () gene (). SMA is a broad-spectrum disorder whose severity is inversely proportional to levels of full-length SMN protein (). The most severe form of SMA is also the most common one, and these patients typically die within the first 2 yr of life (). SMA is characterized by loss of motoneurons from the anterior horn of the spinal cord and progressive muscular atrophy in the limbs and trunk, usually culminating in respiratory failure (). SMN is the central member of a large oligomeric protein complex implicated in a variety of subcellular processes, including pre-mRNA transcription and splicing, RNP biogenesis and transport, neuritogenesis, and axonal pathfinding, as well as in the formation and function of neuromuscular junctions (; ). However, the only SMN function that has been well-documented to date is its role in the biogenesis of Sm-class small nuclear RNPs (snRNPs; ; ; ). Despite the observation that SMA patient-derived mutations lead to defects in Sm-core assembly in vitro (; ; ), a definitive link between snRNP biogenesis and the etiology of the disease has not been established in a model organism. Null mutations in single-copy genes are lethal in every organism studied to date (). In humans and higher primates, there are two genes, and (). is dispensable, but can partially compensate for homozygous loss of (). Patients with additional copies of display milder phenotypes, a finding that has been confirmed using several transgenic mouse models (). Because SMA is caused by reduced expression of SMN, modeling SMA in other genetically tractable organisms has been hampered by the need to create hypomorphic mutations. We describe the generation of a model of SMA. Hypomorphic mutants are characterized by an inability to fly or jump, and they display severe neuromuscular defects. The analysis of this phenotype has led to the surprising discovery that SMN is a sarcomeric protein, implicating a muscle-specific function. (CG16725) is a single-exon gene in (), encoding a 226-aa protein (). The expression profile shows that dSMN is highly expressed during embryogenesis, but that the levels decrease sharply during subsequent developmental stages ( and not depicted). Because SMN is essential for Sm-core RNP assembly in human cells (; ; ), we investigated whether the protein has a similar conserved function. Schneider 2 (S2) cells treated with double-stranded RNA (dsRNA) targeting , but not , were efficiently and specifically depleted of dSMN (). As assayed by two independent methods, dsRNA-treated S2 cells were deficient in assembly of new Sm cores (). Thus, we conclude that SMN's function in snRNP assembly is conserved in invertebrates. A previous study identified two missense mutations ( and ; ) in the conserved Y-G box of the gene, the homozygous inheritance of which results in late-larval lethality (). To identify additional alleles, we searched transposon insertion databases and found one element and two piggyBac transposon insertions in both coding and noncoding regions of (). EY14384 (henceforth referred to as ) is a insertion located 94 bp upstream of the putative transcription start site, whereas f05960 (S) and f01109 (S) are piggyBac insertions within the coding region. Developmental analysis of these mutants demonstrated that homozygotes are completely viable, with no apparent phenotype. The and alleles are late-larval lethals. Genetic complementation studies revealed that the A-D alleles failed to complement each other, and that crossing them over appropriate deletions did not accelerate the lethal phase. Importantly, transgenic expression of a UAS-YFP construct under control of a GAL4 driver completely rescued the larval lethality of the two alleles we tested, producing viable adults. Thus, the A-D alleles are genetic-null mutants. Consistent with these genetic results, Western blotting of lysates from the phenocritical phase (second to third instar) indicated a complete loss of dSMN protein ( and not depicted). However, despite the loss of dSMN, we could detect no appreciable differences in spliceosomal snRNA levels in the mutant larvae ( and Fig. S1, available at ). These findings are similar to those of , which showed that reduced SMN expression and impaired snRNP synthesis caused a slow growth phenotype, but did not affect steady-state snRNP levels in chicken DT40 cells. Thus, despite the observation that dSMN functions in snRNP biogenesis in , the lethality associated with the null mutants is not caused by a systemic depletion of snRNPs. SMA is caused by reduced levels of SMN in mammals; complete loss of function results in early lethality (). To generate a better model for SMA, we screened for neuromuscular phenotypes in adult flies by imprecise excision of the element in . From a total of ∼170 independent excisions, we isolated two lines ( and ) that displayed overt motor dysfunction. and homozygotes (henceforth referred to as E2 and E33 mutants, respectively) each showed marked defects in flying and jumping. The E2 mutants exhibited a 2-d delay in pupation, reflecting an extended larval period, and ∼20% of the E2 pupae died at the pharate adult stage. However, the phenotype of the E2 mutants was incompletely penetrant; ∼45% of E2 animals had flight and jump defects. Moreover, dSMN expression levels in these animals were also variable (unpublished data). In contrast, E33 mutants were completely viable and fertile, and 100% of the animals were incapable of flying or jumping (Videos 1 and 2, available at ). Because the E33 phenotype was fully penetrant, this allele was chosen for further characterization. The indirect flight muscles (IFMs) of the thorax are among the best characterized muscles in the adult animal and are essential for flight (). Because E33 mutants are flightless, we prepared hemithoraces by dissection and analyzed the IFMs of wild-type and mutant animals by light microscopy. The IFMs of the fruitfly are composed of dorsal longitudinal muscles (DLMs) and dorsoventral muscles (DVMs). The mutant IFMs were highly disorganized, even when observed at a gross level (). Although DLM fibers in wild-type flies span the entire anteroposterior length of the dorsal thorax, E33 DLMs often failed to extend the whole length of the thorax, and DVMs were typically unidentifiable (). When salicylate-cleared thoraces were imaged under plane-polarized light, significant muscle degeneration was apparent in the mutant fibers (). Under higher magnification, wild-type IFMs showed the characteristic striations, whereas the mutant muscles were extremely irregular, with numerous bulges and constrictions throughout (). Thus, E33 mutants display severe muscular atrophy, which is one of the hallmark features of SMA. Genetic complementation analyses revealed that the loss-of-function (null) alleles, described in the previous section (), each failed to complement the motor dysfunction phenotype of the E33 allele. Furthermore, the flightlessness of the E33 mutation was rescued by expression of the UAS-YFP transgene under the control of a -GAL4 driver, demonstrating that the observed motor defects in E33 are, indeed, caused by loss of dSMN function. Therefore, we were somewhat surprised when we analyzed developmentally staged lysates by Western blotting of dSMN and found no appreciable differences between adult wild-type and E33 mutants (). Upon further analysis, we found that the amount of dSMN produced in the mutant thoraces was substantially reduced relative to either wild-type or parental lines (). Although the reason for this tissue-specific depletion is currently unknown, the mutants presented us with an opportunity to analyze the neuromuscular defects associated with reduced dSMN expression. During the embryonic–larval transition, motoneurons contact muscle fibers only after the completion of myogenesis (). In contrast, adult motoneurons establish contact with the developing muscle fibers during myogenesis, which is a situation more akin to that of vertebrate development (). Fruitfly DLMs are innervated by remodeled larval motoneurons whose cell bodies lie within the thoracic ganglion and project dorsally into the flight muscles (). Using monoclonal antibody 22C10, which stains neuronal processes (), we analyzed the DLMs of E33 hypomorphs for defects in the organization of their DLM motoneurons. As shown in , both the number and routing of primary motoneuron branches was clearly compromised in E33 flies, as compared with controls. Secondary branches were disorganized () and arborization defects ranged from moderate to severe. Because the mutant muscle fibers were spatially disorganized, it was difficult to assign parameters to the individual motoneurons. Using criteria established by as a guide, we scored flies from each genotype and found that ∼80% of the mutant thoraces showed motoneuron routing defects; a smaller fraction (∼40%) showed defects in secondary branching and arborization. No such defects were observed in the control animals. Thus, we conclude that reduced thoracic expression leads to acute neuromuscular dysfunction in . The development of muscles and motoneurons in the adult fruitfly are spatiotemporally associated. In other words, the synthesis of structural proteins important for one tissue can require the presence of the other tissue. For example, formation of the male-specific muscle (MSM) and the development of its specialized characteristics are absolutely dependent on innervation (). Furthermore, expression of an MSM-specific actin isoform, Act79B, is also dependent on innervation (). Similarly, in vertebrates, denervation is known to result in reduced expression of skeletal muscle actin (). Given that SMA is characterized by motoneuron denervation, it seemed likely that the neuromuscular defects observed in the E33 hypomorphs was also coupled with innervation failure. Because one of the hallmarks of innervation failure is reduced expression of actin, we tested whether the IFM-specific actin isoform Act88F () was expressed in the hypomorphs. RT-PCR was performed on thoracic RNA from wild-type, , and E33 animals, using primers specific for , , and transcripts. Expression of was undetectable in the mutant thoraces, whereas the transcripts were easily detected in the wild-type and strains (). We note that loss of expression was not caused by a general loss of transcription or splicing in the IFMs, as the RT-PCR products for and (which cross intron–exon boundaries) were comparable in all three samples (). Consistent with the idea that splicing is unaffected in the mutants, Northern analysis of thoracic lysates from wild-type or E33 flies showed no significant differences in steady-state levels of either snRNAs () or trimethylguanosine-capped snRNPs (Fig. S1). Staining of wild-type () and mutant () muscles with phalloidin confirmed the general loss of filamentous actin in the hypomorphic IFMs. Most importantly, actin staining and proper myofibril formation were rescued by transgenic expression of YFP-dSMN (). Thus, reduced levels of dSMN protein result in loss of Act88F expression and severe neuromuscular disorganization. In a parallel set of experiments, we were interested in purifying and characterizing the SMN complex. After transient transfection of S2 cells with a Flag-dSMN construct, we performed a pulldown experiment with anti-Flag beads. Associated proteins were eluted from the beads by boiling in sample buffer and subjected to SDS-PAGE. The protein profiles of transfected versus nontransfected cells were compared by Coomassie staining (). Although a more detailed analysis (unpublished data) suggests that there are likely to be many other differences between the two samples, three prominent bands could be easily distinguished. The bands were excised from the gel and analyzed by mass spectrometry. The two smaller proteins were identified as Flag-dSMN and endogenous dSMN ( and not depicted). Because SMN is known to self-oligomerize (), we expected to recover the endogenous dSMN protein. However, the third, ∼100-kD band was identified by 39 distinct peptides in the mass spectrogram as α-actinin (Fig. S2, available at ). Because α-actinin is a protein known to play a major role in cross-linking actin filaments within numerous cell types, including muscle (), we tested whether dSMN interacts with α-actinin in additional biochemical assays. We performed GST-pulldown and coimmunoprecipitation analyses using adult and thoracic lysates. shows that α-actinin was recovered using GST-dSMN, but not with GST alone (control). As shown in , anti-dSMN antibodies weakly, but reproducibly, coprecipitated α-actinin. Control (anti-myc) antibodies and beads coprecipitated only background amounts of α-actinin. Collectively, these studies show that dSMN forms a complex with α-actinin in vivo. Given the aforementioned findings, we analyzed the localization of dSMN in the IFMs of wild-type and YFP-dSMN transgenic flies (). Each IFM myofiber is a single multinucleate cell composed of several myofibrils that contain numerous functional units called sarcomeres. Within each sarcomere, actin (thin) and myosin (thick) filaments interdigitate to form contractile elements. Thick filaments are anchored together within a structure known as the M-line, whereas thin filaments are anchored at the Z-line or Z-disc (). The region between the M-lines is called the I-band. Actin is typically excluded from the M-line, localizing throughout the I-band, and it is often enriched at, or depleted from, the Z-line. In wild-type and YFP-dSMN transgenic myofibers, dSMN was detected not only in muscle cell nuclei but also in the individual myofibrils (; and not depicted). Within the myofibrils, three distinct patterns of sarcomeric localization were observed: I-band, Z-line enriched, and granular (). The same three SMN localization patterns were visible in IFM myofibrils from YFP-dSMN transgenic flies (). The I-band pattern was the predominant one, as confirmed by phalloidin costaining (). However, the dSMN and actin staining patterns did not always correlate, especially in respect to accumulation at the Z-line. That is, when actin was enriched at the Z-line, dSMN did not necessarily show a similar enrichment, and vice versa. It is possible that dSMN localization within the sarcomere is dynamic; future experiments will be required to address this point. Meanwhile, control antibodies and secondary antibodies alone clearly failed to stain the sarcomeres (). Similarly, a 2xYFP transgene construct showed only background staining in the myofibrils (). Importantly, in myofibrils where dSMN was enriched at the Z-line, it colocalized with α-actinin (). We conclude that dSMN localizes to flight muscle sarcomeres. The studies detailed in the previous sections suggested that the flightlessness associated with the allele might be caused by a loss of actin expression in the IFMs. Therefore, we examined the IFMs of flies that contain a null mutation in (), which is called KM88. homozygotes are flightless and fail to form proper IFMs, but are otherwise viable and fertile (). As shown in , KM88 mutants express dSMN, but the protein is delocalized. Similarly, α-actinin, an actin filament cross-linking factor (), was mislocalized in both the and the mutant backgrounds, which is consistent with a complete failure to form thin filaments. Further analysis demonstrated that thick filament formation was largely unperturbed in the hypomorphic IFMs, and that α-actinin staining did not overlap with the myosin filaments (). This scenario is, again, very similar to that of the KM88 mutation, wherein the vast majority of myosin-positive filaments are devoid of α-actinin (unpublished data). Collectively, these findings indicate that dSMN and Act88F are required for proper formation of flight-muscle myofibrils. To determine whether the sarcomeric localization of SMN is a conserved feature among vertebrates, we prepared myofibrils from mouse hindlimb muscles and analyzed the distribution of Smn and α-actinin. As shown in , Smn colocalizes with α-actinin in a Z-line pattern. Two independent anti-SMN antibodies revealed the same pattern, and control antibodies were negative (). Unlike the situation in the fly, mouse Smn localized exclusively to the Z-line; no granular or I-band like patterns were observed. In summary, four lines of evidence demonstrate that SMN is a bona fide sarcomeric protein. First, localization to the sarcomere was not simply caused by cross-reactivity of the antibodies, as transgenic expression of YFP-dSMN showed prominent myofibrillar staining that was completely absent in control IFMs (). Second, control antibodies fail to stain the sarcomeres of both flies and mice ( and ). Third, dSMN staining was lost in the E33 mutants, resulting in a loss of expression and thin filament formation (). Fourth, dSMN interacts with α-actinin in vitro and in vivo (), and reduced dSMN expression leads to complete disorganization of α-actinin in situ (). Thus, we conclude that SMN is a sarcomeric protein. Despite the well-established gene–disease relationship between and SMA, the connection between protein function and molecular etiology has been obscured by a plethora of putative cellular functions attributed to SMN (; ). Previous investigations have shown that the SMN complex is required for assembly and transport of spliceosomal snRNPs (; ; ; ; ). Additional findings point to roles for this complex in neurite outgrowth and pathfinding (; ; Sharma et al., 2005), neuromuscular junction formation (), profilin binding (; Sharma et al., 2005), and axonal transport of β-actin mRNPs (). A common link between each of these additional studies is the actin cytoskeleton. Our finding that reduced dSMN expression leads to motor axon routing and arborization defects, coupled with a loss of Act88F expression in the muscle, is consistent with this actin-related theme. The vast majority of SMA studies continue to be focused on a motoneuron-specific role for SMN; our results do not exclude such a function. However, the idea that SMN might also have a muscle-specific function is not a new one. Cocultures of SMA type I and II muscles with wild-type motoneurons failed to sustain innervation, whereas muscles from control or SMA type III patients maintained stable connections, suggesting a muscle-specific requirement for SMN (; ). Similarly, down-regulation of in mouse C2C12 cells revealed defects in myoblast fusion () and tissue-specific knockouts of in mouse muscle resulted in pronounced dystrophic phenotypes (; ). Also in support of a muscle-specific function is the observation that, despite having comparable levels of SMN, mouse skeletal muscle extracts failed to support efficient Sm-core assembly, whereas extracts from spinal cord were quite active (). Collectively, these studies show that relatively high levels of SMN are required in muscles, although the reason for this requirement was unclear. Our discovery that SMN is a sarcomeric protein required for expression of muscle-specific actin not only provides a plausible role for the protein in muscles, but highlights the potential importance of this tissue in SMA pathophysiology. At least 20 different skeletal muscle diseases are thought to be caused by mutation or mislocalization of sarcomeric proteins (; Hauser et al., 2000; ; ; ). In this regard, it is particularly interesting that SMA patients have been shown to display varying degrees of myofibrillar/sarcomeric (including Z-line) abnormalities (; ). Notably, αB-crystallin was recently reported to form a complex with SMN in HeLa cells (). αB-crystallin is an intermediate filament protein that, in muscle cells, accumulates at the Z-line (). Thus, the SMN complex can interact with at least two distinct Z-line proteins, α-actinin, and αB-crystallin. We have shown that reduced thoracic dSMN levels result in loss of Act88F expression with no apparent defect in either snRNP biogenesis or pre-mRNA splicing. Because expression of muscle-specific actin is known to be dependent on motoneuron innervation (; ), the neuronal defects observed in the hypomorphs are consistent with those expected of an SMA model. Further, the data are consistent with denervation as either the cause or a consequence of muscle degeneration. Notably, a fraction of SMA type III patients display dystrophic phenotypes without evidence of neurogenic abnormalities (; ; ). Although motoneuron loss is generally regarded as a late event in disease progression, one of the main problems in studying SMA, especially the severe forms, is that we are only able to analyze the end-stage of the disease. We currently do not know whether the mutant phenotype observed in the hypomorphs is caused by reduced dSMN expression in the thoracic muscles, the motoneurons, or a combination of tissues. Future work using tissue-specific rescue constructs and a detailed analysis of motoneuron development and myogenesis in the hypomorphic pupae will address these important issues. Regardless of the actual disease trigger, the identification of SMN as a sarcomeric protein underscores the importance of muscle cell function in modulating the severity of SMA. Oregon-R was used as the wild-type allele. Missense alleles and were gifts from M. van den Heuvel (Oxford University, Oxford, UK). Transposon insertion alleles (f05960) and (f01109) were obtained from the Exelixis collection at Harvard Medical School. (EY14384) was a gift from H. Bellen (Baylor College of Medicine, Houston, TX; ). Excision alleles were generated by mobilization of the element using standard protocols. KM88 was a gift from J. Vigoreaux (University of Vermont, Burlington, VT). All stocks were cultured on standard cream-of-wheat agar at room temperature (24 ± 1°C) in half-pint bottles. Genetic complementation analyses were performed using standard methods. 2xEYFP, which was used as control for nonspecific localization, and the () lines were obtained from the Bloomington Stock Center. For rescue experiments, a YFP-dSMN transgene (a gift from J. Gall, Carnegie Institution of Washington, Baltimore, MD) was expressed under the control of in the following homozygous backgrounds: , , and . Embryonic, larval, adult (total and thoracic), and S2 cell lysates were prepared, electrophoresed, and blotted using standard protocols. Anti–rabbit dSMN antibody (a gift from J. Zhou, University of Massachussetts Medical School, Worcester, MA) was affinity-purified and used at a dilution of 1:10,000. Antibodies against SNF (4G3, monoclonal), SmB (Y12, monoclonal), and tubulin (anti–rabbit; Sigma-Aldrich) were used as loading controls. Appropriate HRP-conjugated secondary antibodies were used for detection. and dsRNAs were transcribed in vitro from PCR products flanked with T7 promoters. S2 cells were placed in SF-900 media containing 14 μg/ml of dsRNA. Extracts were generated 3 d after dsRNA treatment using the Ne-Per nuclear/cytoplasmic extraction kit (Pierce Chemical Co.) and dialyzed in reconstitution buffer (20 mM Hepes-KOH, pH 7.9, 50 mM KCl, 5 mM MgCl, and 0.2 mM EDTA) as previously described (). 40 μg of cytoplasmic extract were loaded on a gel for Western blotting to confirm knockdown. For the assembly assay, wild-type U1 snRNA and U1 snRNA containing a deletion of the Sm assembly site were in vitro transcribed from PCR products in the presence of [P]UTP and m7G cap analogue (Promega). 100,000 counts of radiolabeled U1 snRNA were incubated in 100 μg of cytoplasmic extract at 22°C for 40 min in reconstitution buffer. Assembled snRNPs were precleared with protein G beads before immunoprecipitation with monoclonal antibody Y12 in RSB-100 buffer (600 mM NaCl, 20 mM Tris-HCl, pH 7.4, 2.5 mM MgCl, and 0.01% NP-40; ). Immunoprecipitated RNAs were denatured in formamide loading buffer, run on a 6% acrylamide TBE-urea gel, and exposed to a phosphorimager. GST-dSMN was cloned as previously described (). Total fly lysate was homogenized in NET buffer (150 mM NaCl, 50 mM Tris, pH 7.5, and 5 mM EDTA). 500 μg of lysate was passed over recombinant 8 μg GST or GST-dSMN beads overnight in NET buffer. The pulldown products were washed and loaded on a denaturing gel. 15 μg of dsRNA against or (control) were added twice to S2 cell medium over a course of 6 d. After knockdown, cells were transfected with GFP-SmB. Northern blotting of GFP-SmB immunoprecipitate using radiolabeled U1 and U2 snRNA probes was performed following established protocols. Total RNA from adult thoraces was extracted using TRIZOL (Invitrogen). Adult thoraces were pulled apart from the main body, and wings and legs were clipped close to the thorax. Thoraces were immediately transferred to TRIZOL and homogenized. Total RNA was extracted following the manufacturer's instructions. RNA was run on a standard 10% polyacrylamide-urea gel (Invitrogen), transferred to a nylon membrane, and probed with P-labeled PCR products corresponding to the U1, U2, and U6 snRNAs and U3 snoRNA. Total RNA from adult thoraces was prepared as described in the previous section. RT-PCR, with appropriate controls, was performed using SuperScript First-Strand synthesis system (Invitrogen). In brief, oligo-dT–primed first-strand synthesis products were subjected to 20 cycles of PCR using gene-specific primers. Sequences are as follows: Act88F, sense 5′-CCACGCCATTCTGCGTCTGG-3′ and antisense 3′-GCTGCCTTTGAAGAGCTTTCGCG-3′; troponin I, sense 5′-TTGTGAAGGCCAGAAATGGG-3′ and antisense 5′-GACTTCATTTCTGATCAAAT-CCAT; and tropomyosin 2, sense 5′-CACCATGGACGCCATCAAGAAG-3′ and antisense 5′-TTGGTATCGGCATCCTCAGC-3′. The IFMs were dissected from the thorax with the help of fine forceps and needles in a drop of PBS. The entire muscle preparation was fixed in 4% paraformaldehyde, and the fibers were partially teased apart with needles. Immunofluorescence was performed following established protocols. Certain preparations were also stained for filamentous actin by adding 1 μM FITC-conjugated phalloidin (Sigma-Aldrich) 20 min before the secondary antibody incubation was completed. Images were taken using either a TCS SP2 laser scanning confocal microscope or a DM6000 microscope (both Leica), and assembled using Photoshop (Adobe). The Leica Confocal Scanner is interfaced with Leica Confocal Software, and the DM6000 microscope is interfaced with Volocity software. Images were captured at room temperature using a 62× oil immersion objective. Wild-type and transgenic thoracic muscles were dissected in a drop of phosphate-buffered saline (PBS) and fixed in 4% paraformaldehyde. Certain preparations were stained with 1 μM TRITC-phalloidin for 20 min to reveal filamentous actin. Teased myofibrils were washed in PBS and mounted in antifade (50% glycerol and 2.3% 1,4-diazobicyclo-2,2,2-octane). Images were obtained using a TCS SP2 laser scanning confocal microscope or DM6000 microscope, and assembled using Photoshop. Hemithoraces were generated by freezing in liquid nitrogen and dissecting with a razor blade along the central axis of the body. The bisection was slightly offset from the midline, so as to preserve the other half of the thorax intact. Tissues were then processed for immunostaining essentially as previously described (). During fixation, IFMs of the bisected side were flipped over to expose the contralateral hemithorax. The routing of the primary axons and the number of secondary branches was quantitated as described in the text. Hemithoraces were prepared as described in the previous section, and the thoraces were observed and imaged under bright-field optics, with or without counterstaining with safranin for contrast. Analysis of muscles using plane-polarized optics was performed essentially as previously described (). Mouse skeletal myofibrils were prepared by the method of . In brief, hind leg skeletal muscles were depleted of calcium by incubating in an EGTA–Ringer's solution overnight at 4°C. The sample was placed in rigor buffer and homogenized using a glass Dounce tissue grinder. The homogenate was spun (2,000 for 5 min) and washed in repeated cycles until a pure preparation of myofibrils (as monitored by phase-contrast microscopy) was obtained. Purified myofibrils were adsorbed onto a gelatin-coated slide and subjected to immunofluorescence analysis using standard protocols. Note that for dual staining of Smn and α-actinin, mouse monoclonal antibodies targeting a-actinin were incubated with the purified myofibrils, followed by incubation with secondary antibodies conjugated to Alexa Fluor 594. After extensive washes, the preparations were incubated with FITC-conjugated monoclonal anti-SMN antibodies. Fig. S1 shows the analysis of snRNA and snRNP levels in wild-type and mutant animals. Fig. S2 shows the amino acid sequence of α-actinin that was covered by mass spectrometric analysis of the band shown in A. Videos 1 and 2 show the flight behavior of wild-type and E33 mutant adult flies, respectively. The online version of this article is available at .

Programmed cell death (PCD) is essential for the normal development of most, if not all, metazoans. The developmental time at which specific cells or tissues are removed is often specified by the release of systemic or locally acting signaling molecules. During amphibian metamorphosis, for instance, thyroid hormone signals cell death that leads to resorption of the tadpole tail and other larval tissues (; ). During vertebrate limb development, separation of the limb digits requires death of the interdigital regions that is controlled by BMP signaling (). Although much is known about the temporal aspect of regulation in these and other systems, it is less well understood why some cells and tissues, but not others, die in response to widespread signals (). A system that is particularly well suited to address this question is the removal of larval tissues by PCD during insect metamorphosis. In particular, the larval salivary glands of have been extensively used to unravel signaling pathways that control developmental cell death (; ). Death of the larval salivary glands takes place in the early pupa and is triggered by a pulse of the steroid hormone 20-hydroxyecdysone (20E). The salivary glands survive an earlier 20E pulse that leads to the destruction of the larval midgut (). The two consecutive hormone pulses that trigger these stage-specific responses are referred to in this study as the late-larval and the prepupal 20E pulse (). Salivary gland death is foreshadowed by transcriptional activation of the death genes () and (; ). The protein products of both genes kill by interfering with caspase inhibition by the inhibitor of apoptosis protein (IAP) 1 (DIAP1). A critical target of DIAP1 is the apical caspase Dronc, which is required for execution of salivary gland death (for review see ). The mammalian cell death regulators Smac/Diablo and Omi/HtrA2, which are related to and , act in a similar way by antagonizing IAP function (; ; ). Loss of , but not , leads to salivary gland persistence (; ). However, has been shown to synergize with in bringing about salivary gland death (). Induction of both and requires the up-regulation by 20E of transcription factors encoded by , (), and (; , ). In addition, full induction of depends on direct binding of the 20E receptor EcR/Usp to a salivary gland enhancer of the gene (). Proper expression of the early hormone response genes and salivary gland death require the transient expression of the nuclear receptor βFtz-F1 in mid–prepupae (). Thus, βFtz-F1 has the properties of a competence factor for stage-specific hormone signaling (; ). However, βFtz-F1 expression is observed in almost all larval tissues (), leaving the question open of how the tissue specificity of salivary gland death is achieved. Tissue-restricted expression of 20E-regulated genes in the larval salivary glands has been shown to require coregulation by the transcription factor Fork head (Fkh; ; ). is already expressed in the salivary glands during embryogenesis, and is required for the proper development of this organ (; ). Expression of during larval development is restricted to the salivary glands and a small number of other tissues, including the lymph glands and Malpighian tubules (; ; ). The mammalian counterparts of Fkh are the FOXA1, 2, and 3 proteins (also known as HNF3α, β, and γ; ; ), which are members of the larger family of Fkh/HNF or Fox transcription factors (; ; ). Similar to , FOXAs play a role in specifying tissue-specific responses to steroid signaling, suggesting that aspects of FOXA function are evolutionarily conserved (). In mutants, the embryonic salivary glands undergo extensive apoptosis, which is foreshadowed by and expression. Whereas this indicates that the presence of Fkh is required for survival of the embryonic salivary glands, other data suggest that the protein has an independent developmental role in secretory cell invagination (). Thus, is part of a long list of developmental genes that cause ectopic cell death when impaired in their function. It has been estimated that nearly 20% of all genes can cause PCD when mutated (). It is difficult to establish whether these genes normally participate in the control of apoptosis, or whether activation of the default death pathway is an indirect result of aberrant development (). We show that plays a key role in specifying a cell death response to steroid signaling during normal development. Fkh is lost from the larval salivary glands at the onset of metamorphosis, and this loss is required for the subsequent steroid-induced removal of the tissue. Ectopic expression of rescues the salivary glands and premature knockdown of leads to the premature 20E-induced activation of PCD and of the death genes and . Transcription of is down-regulated in a –mediated response to the late-larval 20E pulse, followed by a loss of the Fkh protein during prepupal development. These data indicate that Fkh protects the salivary glands from hormone-induced death until a stage-specific, hormone-induced loss of the protein earmarks the tissue for destruction in response to future hormone exposure. We previously showed that is transcriptionally down-regulated in the salivary glands in response to the late-larval 20E pulse, and that this response is mediated, at least in part, by the early 20E response gene (). However, it was not clear whether this down-regulation is a transient event followed by a resumption of expression, or whether expression remains low or absent in prepupal and pupal salivary glands. To resolve this point, we dissected salivary glands from staged larvae, prepupae, and early pupae. Total RNA extracted from these glands was analyzed for expression by Northern blot hybridization (). Consistent with our earlier results, we found that is expressed in the salivary glands of early and mid-third instar larvae (, lanes P and -18), and that expression is turned off in an apparent response to the late-larval 20E pulse (, lanes -8 and -4). Our Northern analysis did not detect mRNA in early prepupae (, lane 0), and it did not indicate that expression resumes at any time before the salivary glands die at ∼14 h after puparium formation (APF). These data suggested that Fkh protein is lost from the larval salivary glands before the tissue is removed in response to the prepupal 20E pulse. To test this prediction and to determine the temporal profile of Fkh protein expression at the larval–prepupal transition, we stained salivary glands dissected from staged animals with a Fkh antibody (). Strong immunostaining was observed in the cell nuclei at −4 h APF, a time at which mRNA has almost disappeared from the salivary glands (). Fkh protein is still present in considerable amounts in the glands of freshly formed prepupae (0 h APF). 2 h later, the concentration has greatly diminished, and by 4 h APF the protein is reduced to very low levels. Collectively, these data show that is transcriptionally down-regulated in response to the late-larval 20E pulse, and that the protein is still present in the cell nuclei in substantial amounts at the larval–prepupal transition. Subsequently, the protein is lost from the salivary glands during prepupal development. We next asked whether the down-regulation of might be required for the salivary glands to undergo PCD in response to the prepupal 20E pulse. When is ectopically expressed from a heat-inducible transgene in the transformant P[], the gene, which is normally repressed by the late-larval 20E pulse, fails to be down-regulated (). We used the same transgenic line to express ectopically at 10 h APF, shortly before the prepupal 20E pulse signals salivary gland destruction. Pupae of the -expressing line and heat-shocked control pupae () were dissected 20 h APF, which is ∼6 h after the salivary glands are normally destroyed. All -expressing pupae still possessed larval salivary glands at this time ( > 37; penetrance = 100%), whereas no salivary glands could be found 20 h APF in any of the heat-shocked control pupae or in non–heat-shocked P[] pupae. The structure of the cells and cell nuclei of most rescued salivary glands appeared well preserved, and the overall morphology of the glands was very similar to that of salivary glands before the onset of PCD (). In some of the rescued salivary glands, the cells appeared to be more round in shape and contained large vacuole-like structures, suggesting that the cell death program had been initiated in these glands. However, in contrast to dying cells at 14 h APF, which were largely depleted of filamentous actin, these structures were still supported by portions of a well-developed actin cytoskeleton. At 26 h APF, the number of persisting salivary glands was substantially reduced (11% persistence; = 18), which is consistent with the interpretation that some of the glands at 20 h APF had entered the death pathway. We suspected that this was caused by a waning effect of the ectopic Fkh. Therefore, we tested whether survival of the salivary glands could be further prolonged by sustained expression of . After the first heat shock at 10 h APF, we applied a second heat shock at 16 h APF, which resulted in 67% salivary gland persistence at 26 h APF ( = 21). This suggests that activation of the cell death program can indeed be continuously suppressed by maintained expression. Collectively, these data demonstrate that Fkh is sufficient to prevent PCD in the larval salivary glands. Importantly, they suggest that the down-regulation of Fkh before the prepupal 20E pulse provides competence to the salivary glands to respond to this pulse by activation of the death pathway. We wondered whether rescue of the larval salivary glands by was a tissue-specific effect or whether ectopic would block PCD in other tissues as well. To address this question, we ectopically expressed shortly before the larval midgut normally dies. This did not lead to a delay in the initiation or execution of PCD in this tissue (Fig. S1, available at ). Thus, our results support a model in which specifically functions in the larval salivary glands in the developmental control of death. In an attempt to identify genes that might mediate the survival function of Fkh, we analyzed the effect of on the expression of genes that had previously been implicated in the control of salivary gland death. RNA was isolated from the salivary glands of P[] and control animals at different times after heat treatment at 10 h APF. Northern blots of this RNA were first hybridized to detect expression of genes of the 20E-controlled signaling pathway (). The primary hormone response genes , , and which are required for proper salivary gland death (; , ), were all expressed in the presence of ectopic Fkh. However, the amount of RNA and the timing of expression differed from the controls. Expression of all three genes started earlier when Fkh was present, and and were also expressed more strongly. The mRNA level appeared to be somewhat diminished, but did not go down by 16 h APF, as it did in the control (). failed to be down-regulated in 16 h APF salivary glands as well. These data indicate that salivary gland survival in the presence of ectopic Fkh is not likely to be caused by a reduced expression of upstream regulators of steroid-induced death. Next, we tested whether expression of the downstream death activators or was changed by Fkh. We found that mRNA was diminished when compared with the control, but still detectable. More strikingly, expression of , which was strongly expressed in the control glands, appeared to be completely suppressed by the ectopic Fkh (). As is known to be required for proper salivary gland death, this suggested that repression of was at least partially responsible for the suppression of salivary gland death by Fkh. In mutants, the transcription factor Senseless (Sens) is not properly expressed and, similar to mutants, mutants exhibit embryonic salivary gland apoptosis (). We were therefore interested to also determine the expression of in the presence and absence of Fkh. Hybridization with a probe showed that induction of was followed by a brief burst of transcription at 12 h APF (), confirming that is a target gene of Fkh. However, although massive overexpression of from a transgene delayed salivary gland death, it did not affect the transcript levels of or (unpublished data). Thus, uses a different pathway than to protect cells from PCD (unpublished data). This conclusion is consistent with the finding that forced expression of in mutants does not rescue the embryonic cell death phenotype of (). Collectively, our results show that at least part of the effect of Fkh on cell death is mediated by repression of the death genes and . They further suggest a crosstalk between and another survival pathway that acts through . We also note that is epistatic to the cell death regulator (), which is not sufficient to initiate activation of the death pathway as long as is expressed. To provide a broader foundation for our conclusions, we performed a microarray analysis of gene expression at 14 h APF in the presence and absence of ectopic Fkh. Expression of was induced by heat shock in P[] animals, and RNA extracted from the salivary glands of these and heat-shocked control animals was hybridized to Affymetrix Genome Arrays. The microarray analysis confirmed that and were down-regulated in response to Fkh (16- and 2-fold, respectively). In addition, it revealed that another known IAP inhibitor, (), was down-regulated. Overall, the microarray analysis identified 55 genes annotated as functioning in apoptosis whose expression was at least 1.5-fold changed by Fkh ( and S1). was among the four most strongly down-regulated genes, as was , a gene that has been reported to synergize with to promote apoptosis (). Genes encoding the Apaf-1 orthologue Ark and the apical caspase Dronc () were down-regulated by approximately twofold by Fkh. As these proapoptotic proteins are known to be required for the destruction of the larval salivary glands (; ; ; ), down-regulation of the corresponding genes is likely to contribute to the antiapoptotic effect of . The two Bcl-2 family members of , and , were 2- and 2.5-fold up-regulated by . We found that -independent overexpression of either or using the UAS/Gal4 system did not affect salivary gland death (unpublished data). These genes therefore have no, or at least no essential, role in mediating the effect of . Interestingly, among the genes strongly up-regulated by was the PDK1 orthologue. PDK1 is an essential activator of the protein kinase Akt (). As signaling through the PI3K–Akt pathway can protect salivary glands from PCD (), PDK1 is likely to contribute to the survival function of . In summary, the microarray data suggest that Fkh ensures survival by the coordinated repression of IAP antagonists and the regulation of other apoptosis-related genes. In particular, they confirm that the death gene , which is required for salivary gland death (), is a prime candidate for a target of the survival function of . Our results showed that the down-regulation of in response to the late-larval 20E pulse is required for proper induction of and in response to the prepupal 20E pulse and subsequent death of the salivary glands. This prompted us to ask whether down-regulation of is also sufficient to specify these responses to steroid signaling. To address this question, we down-regulated prematurely in early third instar larvae using RNAi. We generated transgenic fly stocks that use a heat-shock promoter to drive expression of a double-stranded (ds) RNA (). Northern analysis confirmed that these lines strongly expressed dsRNA upon heat treatment (unpublished data), followed by a dramatic decline in the amount of Fkh protein (determined by Western analysis; Fig. S2, available at ). To further ascertain that expression of the dsRNA led to an effective knockdown of activity in the salivary glands, we examined the expression of known target genes of by Northern blot hybridization. RNA was isolated from the salivary glands of staged dsRNA-expressing animals and controls. A Northern blot of this RNA was hybridized to detect expression of the and genes (). is a well-characterized target gene of that encodes a glue protein and is strongly expressed in third-larval instar salivary glands until puparium formation (). As expected, large amounts of mRNA were detected in the salivary glands of the heat-shocked control larvae. In striking contrast, mRNA was absent from the salivary glands of the dsRNA-expressing larvae. encodes a bHLH transcription factor that is enriched in larval salivary glands and down-regulated at the larval–prepupal transition (). Recently, it has been shown that is directly activated by , although some expression is still observed in a mutant (). Consistent with these findings, RNA was severely, but not completely, reduced after dsRNA expression (). These data indicate that function in the larval salivary glands is severely compromised by RNAi. Importantly, the RNAi knockdown had no effect on overall development and growth of the larvae, as heat-shocked P[] larvae pupariated at the same time and formed prepupae of similar size as heat-shocked control larvae (unpublished data). Next, we inspected salivary glands that had been treated with dsRNA for signs of PCD. Staining with acridine orange revealed no difference between RNAi and control glands before puparium formation (). This indicated that the premature down-regulation of by itself was not sufficient to cause cell death. However, salivary glands after puparium formation exhibited a progressive loss of tissue integrity and showed strong nuclear acridine orange staining (). The salivary glands of heat-shocked control animals showed no nuclear acridine orange staining, or any signs of tissue disintegration, and died at the normal time in response to the prepupal 20E pulse (). The nuclei of the salivary glands of dsRNA-treated animals were stainable with acridine orange as early as 4 h APF (). At 9 h APF, salivary glands could still be dissected from some, but not all, prepupae. These glands contained acridine orange–positive cell nuclei, but the cell boundaries, which were clearly discernable in the control glands, had disappeared (). Salivary glands could occasionally also be found at later time points. However, the cellular structure and shape of these glands were severely compromised. This suggests that, although the death program had been prematurely activated, the final steps of tissue disintegration and removal could not be performed properly. We overexpressed in early prepupae from a heat-inducible transgene and found that the salivary glands responded in a similar way, showing nuclear acridine orange staining, but persisting for several more hours (unpublished data). Collectively, these results strongly suggested that loss of was sufficient to specify a cell death response to 20E signaling. To further test this conclusion, we asked whether death gene expression was changed after the knockdown of , and, if yes, whether this change occurred in response to the late-larval 20E pulse or earlier. Northern blot hybridization revealed that and were strongly activated at or shortly before puparium formation in salivary glands that had been treated with dsRNA. reached a peak in expression at 4 h APF, whereas was induced earlier, already reaching a very high transcript level at 0 h APF (). This profile is strikingly similar to the temporal profile of and expression observed in response to the late-prepupal 20E pulse. Also at this time, shows maximal expression earlier than , which peaks in expression several hours after the early hormone response genes and are first detected (). This suggests that, similar to the response in late-prepupal glands, induction after premature loss of is a secondary hormone response mediated by and . This hypothesis is supported by the expression profiles of and , which are very similar in late-larval and -prepupal salivary glands (). Hybridization with an probe confirmed that this early response gene is induced at the normal time after knockdown of (). Collectively, the morphological and gene expression data demonstrate that loss of leads to a premature activation of the death program in response to the late-larval 20E pulse. They identify and as two key death regulators whose hormone responsiveness is controlled by Fkh. The response of to 20E signaling at the end of larval development is mediated, at least in part, by the early 20E-inducible gene (). The down-regulation of that is normally observed at this time does not occur in mutants of . The continued expression of in these mutants is sufficient to maintain expression of the gene in prepupal salivary glands. mutants also show defects in salivary gland death, which are even more pronounced in mutants of the subfunction of (; ). Collectively, these observations raised the possibility that a derepression of might be responsible for the persistence of the larval salivary glands in mutants. To test this possibility, we performed a Northern analysis of expression in the salivary glands of late prepupae and early pupae of the mutant (). As expected, the salivary glands of control animals did not show expression of shortly before their destruction, at 12 or 14 h APF. In striking contrast, salivary glands of hemizygous mutant animals exhibited strong expression of mRNA both at these times and also in 16-h pupae. In 16-h control pupae, the disintegration of the salivary glands was too far advanced to obtain RNA for a Northern analysis. These data indicate that is indeed derepressed in the persisting salivary glands of mutants. Hybridization of the Northern blot with a probe confirmed that this death gene was not properly induced in the mutant (). The expression of in provides an explanation for the persistence of larval salivary glands in mutants and identifies as at least one critical target of the function of in programmed cell death. These results lend further support to a model in which is tied into the hormonal signaling pathway through the 20E-controlled gene (). Developmental cell death in invertebrates and vertebrates is often controlled by systemic signals, which provide the trigger for cell and tissue destruction (; ). However, it is not well understood why these signals induce death only at a particular time and only in some cells and tissues, but not in others. The experimental data presented in this study support a model that explains how a specific tissue of the fruit fly is singled out for destruction in response to the steroid hormone 20E (). It explains why the larval salivary glands are destroyed in response to a particular 20E pulse, the prepupal pulse, and why they survive earlier pulses of the same hormone. It thus provides a framework for our understanding of how tissue-specific developmental cell death is precisely timed. In more general terms, our data suggest that a key event in acquiring competence for a cell death response to a systemic signal is the loss of a tissue-specific survival factor. This loss occurs in response to a temporal signal that precedes the death-inducing signal. In the salivary glands, the tissue-specific survival factor is Fkh, and the signal that leads to the loss of Fkh is provided by the late-larval 20E pulse. Our data show that loss of Fkh is required for the death response to the prepupal 20E pulse (). They further suggest that the salivary glands survive all previous hormone pulses because they are protected by the presence of Fkh (). Thus, Fkh not only plays a key role in determining the tissue selectivity of salivary gland death, but also in the proper timing of this event. Elimination of Fkh is mediated by the 20E-induced transcription factor (). We find that is strongly expressed at the normal time of salivary gland death in the mutant of , which demonstrates that is required for the continued repression of beyond the larval–prepupal transition (). The derepression of in is sufficient to explain why the salivary glands do not die in the mutant. Loss of the survival factor renders critical death regulators responsive to the death-inducing signal. In the salivary glands, these death regulators are the IAP antagonists and , which together are required for salivary gland death (). In the absence of Fkh, the two genes are inducible by hormone, as shown by the premature induction of and after RNAi knockdown of (). Importantly, loss of by itself is not sufficient to activate and , or to kill the salivary glands within the ∼36 h between knockdown and the late-larval 20E pulse. Strong activation of and and death only occur in response to the hormonal signal. After elimination of Fkh, the hormone induces expression of and at a level that is sufficient to kill (). This observation supports the conclusion that there are no other repressors present in prepupal salivary glands that are sufficient to prevent hormonal induction of cell death. All that seems to be needed to induce death is the hormone 20E and one or more 20E-induced transcriptional activators. Previous work has shown that and play the role of hormone-induced activators of and in late-prepupal salivary glands (). Both and are required for the induction of , which has the characteristics of a secondary-response gene. Intriguingly, the activation of after premature loss of shows the same secondary-response characteristics, suggesting that the same 20E-induced transcription factors are responsible for the activation of by the late-larval 20E pulse (). Full induction of depends not only on but also on direct binding of the hormone receptor EcR/Usp to the gene (). Thus, has characteristics of both a primary- and secondary-response gene, leading to an earlier induction of the gene in response to the prepupal 20E pulse. Again, premature activation in response to the late-larval pulse shows the same temporal characteristics (). This suggests that and are responsible for the premature activation of and after knockdown of , and that such an activation is normally prevented by the presence of Fkh. Our immunostaining data support this conclusion by showing that Fkh protein is still present in the larval salivary glands at the time when the two genes are active. It only disappears from the tissue 2–4 h APF (). These data explain why and mediate a death response exclusively to the prepupal 20E pulse, despite a very similar induction pattern of the two genes in response to the preceding late-larval pulse. Our results exclude that repression of and is mediated by the target . Repression may thus be mediated by another downstream target of or by direct binding of Fkh to transcriptional control regions of and . In support of the latter possibility, we found that the first intron of contains a cluster of 13 Fkh binding sites. One of these sites exhibits strong binding of Fkh in in vitro DNA-binding assays, whereas the other sites have weak to moderate binding affinity (unpublished data; de Banzie, J., personal communication). Although this region may function as a silencer of expression in vivo, reporter gene assays in transgenic flies did not reveal that it has an enhancer function. We were not able to identify a similar binding site cluster in . Our microarray data identify other apoptosis-related genes that are down- or up-regulated by Fkh. Therefore, it is likely that Fkh protects cells from death by interfering with the cell death program at multiple levels. Regulation of genes such as , , or , is likely to mediate a general function of as a survival factor. This function appears to be required for the survival of the developing salivary glands during embryogenesis (). However, it is not essential for the survival of postembryonic salivary glands, as demonstrated by the failure of the glands to die in the absence of Fkh during prepupal development. Our data confirm this conclusion by showing that the salivary glands fail to undergo PCD within the ∼36 h between the premature knockdown of and the steroid induction of death. They separate a general protective function of Fkh from a specific function that Fkh has in the control of steroid-induced developmental PCD. Tissue-specific developmental cell death controlled by steroid hormone plays an important role not only in insects but also in humans and other vertebrates. Glucocorticoids, for instance, control the development of the immune system by killing specific types of thymocytes (). Many genes regulated by glucocorticoids are coregulated by the vertebrate FOXA counterparts of Fkh (). It will be interesting to see whether FOXAs have evolutionarily conserved functions in glucocorticoid-induced death and in other types of developmental cell death. For construction of the P[] transformation plasmid, a segment of the coding region of (corresponding to amino acid positions 189–435) was amplified by PCR from the plasmid fkh-pET3b. Two copies of the product were then sequentially cloned in a head–head orientation into the transformation vector pCaSpeR-hs-act (), leaving a 130-bp spacer between the copies. Five independent transformant lines were obtained, which all expressed dsRNA upon heat shock; a detailed description of the construction steps can be obtained upon request). element injections were performed by BestGene, Inc. Third instar larvae were staged using the blue gut method as previously described (). For the −4-h time point (clear gut) in , only salivary glands that had completed glue protein secretion into the gland lumen were used. For the −4-h time point in , this additional criterion could not be used because glue production was severely affected in heat-shocked P[] animals. Prepupae and pupae were staged by collecting freshly formed prepupae within 30 min of puparium formation and keeping them on damp filter paper at 25°C for the indicated lengths of time. For the Northern analysis of expression in a mutant background, we used the stock (). allele). The males were separated based on this phenotype, and salivary glands were dissected for RNA extraction. RNA extraction, fractionation by gel electrophoresis, transfer to nylon membranes, and hybridization with radioactive DNA probes were performed as previously described (). Probes were derived by restriction digest from the following plasmids: , 848-bp AflII–BglII fragment from RE03865; , 1-kb EcoRI–HindIII fragment from pOW3Sal; , 1.4-kb EcoRI fragment from pBS-sens (provided by H. Bellen; Baylor College of Medicine, Houston, TX); , 480-bp StuI–PvuII fragment from paaDM527; and , 1.6-kb I fragment from cDNA (provided by C. Thummel; University of Utah, Salt Lake City, UT). , , , and probes were prepared as previously described (). For ectopic expression of , prepupae of the transformant line P[] and control prepupae were collected at 9.5 h APF and incubated for 30 min in a 37°C water bath. The animals were transferred to damp filter paper in a Petri dish and kept at 25°C until the salivary glands were dissected for RNA extraction or microscopic analysis. For RNAi knockdown of , third instar larvae of P[] and were collected within 3 h of the second–third larval instar molt and transferred to fresh yeast paste. Larvae were kept at 25°C and subjected to heat shocks, as described in the previous section, at 12, 26, and 40 h after collection. The salivary glands of larvae expressing dsRNA were defective in glue protein production and showed a reduced size compared with control glands (). Glue proteins make up >30% of the total gland protein in wandering third instar larvae (), suggesting that the reduced organ size was caused by smaller cell size and not by reduced cell number. Staining of the salivary glands with the nuclear dye Hoechst 33342 confirmed this prediction. Impaired glue production was an expected result of the knockdown of , as expression of all glue protein genes, including (), depends on (; ; our microarray data). Acridine orange staining of salivary glands and midguts was performed as previously described (). Independent of the genotype of the animals, the salivary glands of both heat-shocked and non–heat-shocked wandering larvae had a speckled appearance after acridine orange staining (). Microscopic analysis of single nuclei at high magnification revealed that this was caused by staining of the nucleoli, which stain brightly with acridine orange (). The polytene chromosomes did not show staining above background (Fig. S3). Staining of salivary glands with Fkh antibodies, FITC-labeled phalloidin (Alexis Biochemicals), and Hoechst 33342 (AnaSpec, Inc.), was performed using standard procedures. The anti-Fkh antibody was affinity purified and used at a dilution of 1:500 (). Bound Fkh antibodies were detected with a Cy2-conjugated goat anti–guinea pig secondary antibody (1:400; Jackson ImmunoResearch). Images of the fluorescently labeled tissues shown in , , and were taken with an inverted confocal microscope (TCS SP2; Leica), overlaying a z series of 30–50 sections. Tissues were mounted in PBS, and images were acquired with a 10×/0.40 NA ( and ) or 20×/0.70 NA () HC PL APO objective (eyepiece lens 10×/22 HC PLAN; Leica). Bright-field images shown in Fig. S1 were taken with a 3–charge-coupled device digital imaging camera (KY-F75U; JVC) using a microscope (Axioskop 2 Plus; Carl Zeiss MicroImaging, Inc.) and Auto-Montage imaging software (Syncroscopy). The differential interference contrast and fluorescent images shown in Fig. S3 were taken with a camera (AxioCam MRm; Carl Zeiss MicroImaging, Inc.) using an Axioskop 2 Plus microscope and AxioVision 4.1 software. 2.5×/0.075 NA (Fig. S1) and 40×/0.75 NA (Fig. S3) Plan-Neofluar objectives (Carl Zeiss MicroImaging, Inc.) were used for image acquisition (10×/23 eyepiece lens). Images were processed using Photoshop 7.0 (Adobe), with uniform adjustments made to brightness and contrast, and were assembled using Illustrator CS2 (Adobe). All images were taken at room temperature. P[] prepupae and control prepupae were heat shocked at 9.5 h APF for 30 min at 37°C and the salivary glands dissected 4 h later. Samples were prepared in three replicates from the P[] glands and in two replicates from the control glands. Total RNA was isolated using Trizol (Invitrogen) and purified on RNAeasy columns (QIAGEN). Hybridization to Affymetrix Genome Arrays was performed by the microarray facility of the University of Maryland Biotechnology Center. Raw data provided by the Center were normalized, pooled, and compared using dChip (). Analysis was performed using the PM-only model with outlier detection. The datasets were filtered for genes that showed an at least 1.5-fold relative change in their mean expression and an absolute expression change of at least 400. Query for apoptosis-related genes was performed based on annotation using dChip and Microsoft Access. Fig. S1 shows that misexpression of does not affect PCD in the larval midgut. Fig. S2 shows a Western blot analysis of Fkh protein expression after knockdown of the gene by RNAi and after overexpression from P[]. Fig. S3 shows that acridine orange strongly stains the nucleoli of salivary gland nuclei of normal third instar larvae. Table S1 lists all apoptosis-related genes identified by microarray analysis that showed an at least 1.5-fold response to .The online version of this article is available at .

Stroke is a major cause of death and a major factor behind people spending their lives confined to bed, as the consequences of a stroke include loss of functions such as memory, sensory perception, and motor skills. These symptoms are caused by various kinds of ischemia, which drive brain neurons toward death. In most cases with brain ischemia, neuronal death is composed of necrosis and apoptosis, which remove all damaged neurons (; ). Necrosis occurs first in the ischemic core, whereas apoptosis occurs several days later in the region surrounding the core, called the penumbra. Both cell death modes after ischemia are initiated by the rapid loss of cellular ATP, followed by disturbances in cellular signaling mechanisms, including Ca homeostasis (; ). The apoptosis machinery is accelerated after reperfusion, which partially supplies blood flow to produce the ATP required for the execution of apoptosis (; ; ). Many studies have revealed that several compounds that inhibit apoptosis in cells have protective roles against ischemic damage in vivo, although their potencies are limited (; ; ; ). This may be related to the possibility that rapid and expanding necrosis largely contributes to the total loss of brain neurons after ischemia. Thus, rapid treatments are currently the focus of investigations into cures for brain strokes (The National Institute of Neurological Disorders and Stroke rt-PA Stroke Study Group, 1995; ; ). Compared with the machinery of apoptosis, necrosis is a more passive process in which energy failure leads to mitochondrial swelling, accompanied by cristae disruption. These processes then lead to rupture of the plasma membrane with concomitant loss of intracellular proteins and ions. However, little is known about how to develop compounds that inhibit necrosis. We recently demonstrated that cultured cortical neurons die by necrosis under low-density (LD) and starvation stress without serum or any supplements (; ,). Of particular interest are the findings that neuronal death in high-density (HD) cultures is markedly inhibited and that addition of conditioned medium (CM) from HD cultures prevents necrosis in LD cultures (). Here, we report the identification of a CM molecule, prothymosin-α1 (ProTα), that mediates necrosis inhibition and note the clinical potential of this protein to prevent brain strokes. As previously reported (; ,), rat embryonic cortical neurons in serum-free LD (10 cells/cm) cultures rapidly died by necrosis. As early as 6 h, but not at 3 h, after the start of serum-free culture, neurons under LD conditions showed many pores on their surfaces by scanning EM analysis (). At 12 h, the cell surface membranes were largely destroyed and only the nuclei remained. By transmission EM analysis, typical necrotic features, such as membrane destruction, loss of cytoplasmic electron density, and swollen mitochondria with a disrupted cristae structure, were observed at 6 h (,). Necrotic features were also observed by staining with propidium iodide (PI). PI staining was substantially observed after 3 h of LD culture and showed a time course that was parallel to the decrease in survival activity (). Addition of CM derived from 72-h HD (5 × 10 cells/cm) cultures delayed the cell death in LD cultures in a concentration-dependent manner, with the concentration dependency also being parallel to the decrease in survival activity (). When the factor mediating this survival activity was purified from prefractionated extracts, 6.3 μg of an ∼20-kD protein was obtained by molecular weight cutoff ultrafiltration, ion-exchange filtration, and SDS-PAGE from 20 ml of the CM (; and Table S1, available at ). After SDS-PAGE, this 20-kD protein was analyzed by matrix-assisted laser desorption/ionization–time of flight (MALDI-TOF) mass spectrometry (MS), and a search in the nonredundant National Center for Biotechnology Information protein database for matching peptide mass fingerprints revealed 17 peptides that were unique to rat ProTα. Moreover, tandem MS analysis confirmed that the N terminus of purified ProTα was an acetylated serine (129.612 vs. Ser 87.343 m/z; ), in agreement with a previous report (). For biological identification, we performed several experiments using an anti-ProTα IgG, which had already been characterized (Figs. S1 and S2, available at ). When CM factors were applied to anti-ProTα IgG-conjugated beads, the eluates obtained from acid-treated beads exhibited a single band that corresponded to recombinant ProTα on SDS-PAGE and an “acidic blot,” with no substantial signal in the flow-through, whereas the ProTα signal was observed in the flow-through from preimmune IgG-conjugated beads, but not in the control eluates (). After pretreatment of the CM with anti-ProTα IgG-conjugated beads, but not preimmune IgG-conjugated beads, ∼80% of the original CM-induced survival activity was lost (). Thus, it is evident that a large proportion of the survival activity in the CM can be attributed to the action of ProTα. For quantitative analysis, ProTα was extracted from the CM by acid phenol (Fig. S1 and supplemental text), subjected to SDS-PAGE, and directly detected by the highly sensitive blue stain method without a blotting procedure, as transfer of ProTα to a blotting membrane is unstable because of its acidity. ProTα was detected in the CM as early as 1 h after the onset of serum-free culture, and the level was maintained for up to 12 h, whereas the intracellular ProTα level was reduced (). The amount of ProTα in the CM of HD culture (72 h) was determined to be 66 pmol/cm. This release into the CM was observed in serum-free, but not in serum-containing (serum-plus), cultures. Because cortical neurons showed no substantial plasma membrane damage at 12 h after the start of serum-free HD culture in terms of PI staining or transmision EM analysis (,), the ProTα in the CM is likely to have been released from neurons whose membranes have not yet been disrupted. ProTα lacking a signal peptide sequence is probably released in a nonvesicular manner (unpublished data), as seen in the case of FGF-1 (; ). When anti-ProTα IgG was simply added to HD cultures, a concentration-dependent decrease in the survival activity was observed, despite no extra incubation for immunoabsorption (). This finding provides strong evidence that ProTα released under serum-free stress plays a substantial neuroprotective role. ProTα purified to homogeneity exhibited a concentration dependency equivalent to that of the recombinant protein and had a maximum survival activity in LD cultures that was equal to that in HD cultures (). Furthermore, addition of ProTα mutants that lacked the N-terminal region (Δ1–29), including thymosin-α (), or the C-terminal region (Δ102–112), including the nuclear localization signal TKKQKK, retained the original activity of ProTα. As the culture plates were precoated with ProTα in the aforementioned experiments, the site of ProTα action seems to be through unidentified cell surface membrane targets, but not through thymosin-α receptors or nuclear binding sites. In this experiment, the survival activities of ProTα were equivalent when the same amount of protein (25 pmol/cm) was used to precoat culture plates or added directly to cultures (initial medium concentration: 80 nM; ). Recombinant ProTα reversed the rapid decrease in survival activity in cortical neurons caused by the serum-free or permanent ischemia model (). The addition of ProTα abolished all the typical necrotic features, such as disrupted plasma membranes and swollen mitochondria, but no damage to the nucleus, at 6 h in the transmission EM analysis, and instead caused apoptosis, as observed by nuclear fragmentation, at 12 h (). A similar cell death mode switch from necrosis to apoptosis was observed by pretreatment with CM factors (20%) derived from HD cultures, whereas treatment with 1 μg/ml anti-ProTα IgG inhibited the cell death mode switch (). When the cell death mode was evaluated by double staining with necrosis (PI) and apoptosis (annexin V, caspase-3, and TUNEL) markers, 69, 86, and 92% of neurons, respectively, died by necrosis under serum-free stress, whereas only 15, 22, and 5% of neurons, respectively, died by apoptosis (). Addition of ProTα or CM totally switched the cell death mode from necrosis to apoptosis, and the CM-induced changes were abolished by further addition of anti-ProTα IgG. These findings suggest that the cell death mode switch induced by CM factors is largely attributable to the action of ProTα. A pharmacological study using various inhibitors revealed that the survival activity of recombinant ProTα was mediated through activation of PLC and PKC (), consistent with a previous report regarding CM factors (). In the present study, we used 1 μM U73122, a PLC inhibitor, and GF109203X, a pan-type PKC inhibitor. These findings were supported by a biochemical study, in which addition of ProTα significantly increased the PKC activity and the effect was reversed by U73122 (). This survival activity at 12 h was inhibited by Go6976, a PKCα/β inhibitor, but not by HBDDE, a PKCα/γ inhibitor, or Rottlerin, a PKCα/δ/θ inhibitor (). Therefore, the PKCβ isoform is likely to be involved. Significant ProTα-induced survival activity was observed after 12 h of serum-free culture, but not at 24 h (). However, more potent and long-lasting survival activity was observed in the low-oxygen and low-glucose (LOG) ischemia and reperfusion model. It should be noted that no change in the survival activity was observed between 24 and 48 h in the latter condition. The incidence of apoptosis in ProTα-treated samples was markedly lower in the latter reperfusion model (38.4 ± 6.16%; = 4) than in the serum-free model (86.0 ± 8.25%; = 6), suggesting that this difference could be attributed to the action of antiapoptotic serum factors. This view was clearly confirmed when antiapoptotic growth factors and ProTα were added to the serum-free culture (). At 48 h after the start of serum-free culture, the survival activity was as low as 5%, even in the presence of ProTα alone. However, further addition of NGF, brain-derived neurotrophic factor (BDNF), basic FGF, or interleukin-6 rescued the survival activity to >80% of the control level, whereas these factors alone had no effects on the survival. There was no mitochondrial cytochrome (cyto ) release, which induces apoptosis through the formation of the apoptosome with Apaf- 1 and caspase-9, whereas addition of ProTα caused cyto release (). It should be noted that ProTα-induced cyto release was abolished by further addition of BDNF or BIP-V5, which blocks the translocation of Bax to mitochondria (), but not by zVAD-fmk, a pan-type caspase inhibitor. demonstrates the time-dependent changes in cell death status when the culture was performed in the presence of ProTα and BDNF, zVAD-fmk, or BIP-V5. The addition of ProTα alone inhibited necrosis throughout 48 h and increased the number of living cells (or necrosis, apoptosis double negative) more prominently at the early stage (12 h), but not at the later stage (24 or 48 h). On the contrary, the number of cells showing apoptosis time-dependently increased in the presence of ProTα. Further addition of BDNF or BIP-V5 showed complete survival by inhibiting apoptosis throughout 48 h. However, zVAD-fmk caused a marked cell death by necrosis at the later stage, though it showed complete survival at the early stage. It is generally accepted that necrosis is caused by energy failure because of the loss of cellular ATP (; ,; ). The cellular ATP levels of cortical neurons rapidly decreased immediately after the start of serum-free culture (). This decrease was markedly inhibited by the addition of ProTα or CM, and further addition of anti-ProTα IgG abolished the CM effects. As previously reported (), this rapid decrease and its reversal by ProTα seem to be parallel to the activity of glucose transport, as [H]-2-DG uptake was markedly decreased by serum-free treatment and reversed by ProTα (). Similar changes were also observed in the ischemia-reperfusion model of culture (). Addition of ProTα reversed the rapid decrease in the cellular ATP levels of cortical neurons after LOG ischemic stress and reperfusion with serum-containing medium (). We previously revealed that the membrane translocation of the glucose transporters GLUT1 and -4 is largely inhibited in serum-free cultures of cortical neurons, which leads to necrotic cell death (). In the present study, inhibition of GLUT1 and -4 membrane translocation was also observed under LOG stress by immunocytochemistry (). Biochemical evidence was also found when the cell surface proteins were biotinylated before Western blot analysis (). An immunocytochemical study revealed that ProTα activated PKCα, -β, and -β at 10 min (). A knockdown study using antisense oligodeoxynucleotides (AS-ODNs) for these kinases demonstrated that only PKCβ, not PKCα or -β, plays roles in the ProTα-induced GLUT1/4 translocation (, h and i). Further characterization revealed that ProTα induced GLUT1/4 translocation by activation of PLC through pertussis toxin–sensitive Gα, but not Gα, suggesting that a putative Gα-coupled ProTα receptor may be involved in this action. Furthermore, the PKCβ AS- ODN treatment reversed the ProTα-induced necrotic PI staining (). The molecular machineries for apoptosis are relatively better characterized than those for necrosis. In terms of the activation of various caspases, caspase-3 is believed to be the final execution molecule for apoptotic cell death linked to DNA breakdown and nuclear fragmentation (; ). ProTα activated caspase-3 in the serum-free or permanent ischemia model (). Similar activation was also observed for caspase-9, but not for caspase-8 or -12. These findings suggest that ProTα causes apoptosis through a caspase-9–mediated mitochondrial pathway. This view was clearly confirmed by the findings that ProTα increased the expression of proapoptotic Bax and Bim, but slightly decreased the expression of antiapoptotic Bcl-2 and -xL, which regulate mitochondrial apoptotic signaling (). On the other hand, the PKCβ AS-ODN also reversed the ProTα-induced proapoptotic Bax expression in the LOG stress model (). However, it should be noted that the Bax expression was also abolished by treatment with the AS-ODN for PKCβ, but not that for PKCα. To examine the role of Bax and Bim in the ProTα-induced apoptosis, we performed the experiments using siRNA in the LOG ischemic stress and reperfusion model. As shown in , the ProTα treatment markedly up-regulated the Bax expression in all cells. The pretreatment of Bax siRNA 24 h before ProTα treatment caused a complete loss of Bax in 10–18% of total cells, and these Bax-negative cells did not show any apoptotic active caspase-3 or necrotic PI staining. This finding was also confirmed by the quantification of incidence of apoptotic, necrotic, and living cells in experiments without and with Bax siRNA treatment. As mentioned in , the ProTα treatment alone abolished the necrosis, whereas it increased the survival with some apoptosis (, left). A similar cell death ratio was observed in Bax-positive cells, which are unlikely to be transfected with siRNA. However, Bax-negative cells were all alive, or apoptosis and necrosis negative. However, the down-regulation of Bim showed less significant changes in the number of cells showing apoptosis (Fig. S4, available at ). These results strongly suggest that ProTα causes apoptosis through an up-regulation of Bax. ProTα is a highly acidic nuclear protein of the α-thymosin family and is widely distributed throughout the body (; ). It is generally thought to be an oncoprotein that is correlated with cell proliferation by sequestering anticoactivator, a repressor of estrogen receptor activity, in various cells (; ). On the other hand, ProTα has also been reported to act as an extracellular signaling molecule, as observed in the activation of macrophages, natural killer cells, and lymphokine-activated killer cells, and in the production of interleukin-2 and TNF-α (). Here, we isolated ProTα from CM of primary cultures of cortical neurons as a molecule providing protection against neuronal necrosis (). By using a specific antibody, ProTα was proven to be the major CM factor involved in density-dependent survival under conditions of serum-starvation stress. The identity of the target of ProTα in respect to cell death regulation is a very interesting issue. ProTα was reported to inhibit apoptosome formation in NIH3T3 cells (). This observation is in good contrast with the present finding that addition of ProTα to neuronal cultures caused apoptosis, suggesting that ProTα has opposite functions inside and outside of the cell. Furthermore, as a ProTα deletion mutant lacking the nuclear localization signal retained the full survival activity, it is unlikely to be the aforementioned genomic action. The most probable candidate would be a cell surface receptor. Indeed, the presence of a cell surface ProTα receptor has been reported in lymphoid cells (; ), and we confirmed this in cortical neurons by using ProTα–Alexa 488 (Fig. S5, available at ). Further strong evidence to support the presence of a cell surface ProTα receptor is the fact that ProTα-induced membrane transport of GLUT1/4 was mediated through a Gα-coupled receptor, which activated PLC and PKCβ. Because ProTα-induced translocation of PKCβ was observed within 10 min, it is evident that this signaling can be attributed to a direct action through a membrane receptor. The distinctive advantage of ProTα-induced neuroprotection can be attributed to the inhibition of necrosis. Necrosis is characterized by bioenergetics failure and rapid loss of plasma membrane integrity, which may result from decreased glucose transport (), as well as enzymatic destruction of cofactors required for ATP production, increased mitochondrial reactive oxygen species production, and channel-mediated calcium uptake (; ; ). In our serum-free or LOG stress model, most cortical neurons died by necrosis. We found that rapid cell death or necrosis was accompanied by decreased glucose uptake and cellular ATP levels. The glucose transport mechanism is one of the most important sets of molecules for maintaining cell survival. Various species of GLUT have been identified in different cell types (). Because GLUT3, which is most abundant in neurons, is constitutively localized in membranes, its function is unlikely to be regulated by environmental factors. In contrast, it was reported that some survival factors induce translocation of GLUT1 and -4 into plasma membranes through activation of protein kinases, including AKT and PKCs (; ; ). This is consistent with our previous study showing that serum-free stress reduces GLUT1/4 translocation (). Here, we successfully demonstrated that ProTα prevented the stress-induced reduction of GLUT1/4 transport through PKCβ activation. The second important issue is that ProTα switches the cell death mode by causing apoptosis. Because serum-free stress alone did not cause mitochondrial cyto release, this stress by itself is unlikely to induce the machinery for apoptosis as well as that for necrosis. This view is supported by our previous report that addition of pyruvate to serum-free cultures to maintain the cellular ATP levels prevented necrosis but did not induce apoptosis (). Although the possibility still remains that pyruvate has an unidentified mechanism to remove the trigger for apoptosis, it is very likely that apoptosis does not always occur after the prevention of necrosis. This finding strongly supports the view that ProTα induces apoptosis. In the present study, we have demonstrated that ProTα up-regulates proapoptotic Bax and Bim, and down-regulates antiapoptotic Bcl-2 and -xL. Because the treatment with Bax siRNA blocked the ProTα-induced apoptosis, and the treatment with BIP-V5 blocked the ProTα-induced cyto release and apoptosis, it is evident that the up-regulation of Bax plays an important role in ProTα-induced apoptosis. On the other hand, the caspase inhibitor zVAD-fmk blocked the ProTα-induced apoptosis and caused necrosis. This may be explained by the view that the up-regulation of Bax by ProTα causes a cyto depletion from mitochondria, followed by the necrosis induction through a damage of mitochondrial ATP production (; ), as apoptosis is inhibited by zVAD-fmk. ProTα-induced up-regulation of Bax was found to be mediated by PKCβ and -β activation, consistent with previous reports that PKCβ activates the I-κB kinase complex, IKK (; ), leading to NF-κB activation followed by Bax up-regulation. Thus, PKCβ is likely to be an important switch molecule to determine the cell death mode. The lack of contribution of PKCβ1 to the ProTα-induced necrosis inhibition may be related to the deficiency of the membrane-anchoring C-terminal peptide of PKCβ (). By use of acid-phenol extraction and blue staining, the amount of ProTα in the CM was determined to be 66 pmol/cm. The data of revealed that CM amounts are ∼50% of the total (CM + cells) at 6–12 h after the start of serum-free HD culture. As this value in CM (∼30 pmol/cm of ProTα) corresponds to the concentration of ProTα (25 pmol/cm) required to make the conversion of necrosis to apoptosis, this mechanism seems to be physiologically relevant. The possible in vivo roles of ProTα in brain stroke represent the most important issue to be discussed. ProTα inhibits the rapid cell death of neurons after serum-free ischemic stress by inhibiting necrosis. This property seems to be beneficial, as the representative growth factors used in the present study had no effects on the necrosis, although they have potent antiapoptosis activities. Furthermore, ProTα is a unique cell death regulatory molecule in that it converts the intractable cell death necrosis into the controllable apoptosis. Because this apoptosis would be inhibited by growth factors secreted upon ischemic stress, it is expected that ProTα may have neuroprotective roles in brain stroke. As mentioned in , the combined use of ProTα with growth factors, but not caspase inhibitors, may have a potential clinical availability. In conclusion, we have identified the survival factor secreted from cortical cultures as the nuclear protein ProTα. We have also demonstrated that this protein plays an in vivo neuroprotective role in brain ischemic events. Moreover, it has the potential for clinical use against brain strokes. Cell culture medium and FCS were purchased from Invitrogen. The antibodies used in the present study were GULT1 and -4; BDNF; phosphorylated PKCα, -β, -β, -γ, -ɛ, and -δ (all from Santa Cruz Biotechnology, Inc.); activated caspase-3 and phosphorylated PKCζ (Cell Signaling); and cyto (BD Biosciences). The reagents for staining were PI (Sigma-Aldrich), TUNEL, Hoechst 33342 (Invitrogen), and Gelcode blue stain reagent (Pierce Chemical Co.). After several trials, we optimized our procedures for purifying ProTα. Purification was started with 20 ml CM, which had been collected at 72 h after the start of HD culture, as previously reported (). The CM was first subjected to ultrafiltration (Vivaspin 2; Sartorius KK), and the active materials observed in the >5-kD fraction were applied to an ion-exchange membrane spin column (Vivapure Q Mini; Sartorius KK), which had been equilibrated with 20 mM sodium acetate, pH 5.2. The sample was eluted with different concentrations of NaCl (0.2–1 M), and the active fraction was finally separated by SDS-PAGE and stained with Gelcode blue stain. The appropriate band was excised from the gel, washed with 50 mM NHHCO and 50% acetonitrile, and incubated with 100% acetonitrile for 10 min. The gel segment was rehydrated in 50 mM NHHCO and then dehydrated in 100% acetonitrile. The resulting gel plug was incubated overnight with 5 ng/μl trypsin in 25 mM NHHCO. The digested peptide mixture was diluted with the matrix 4-hydroxy-α-cyanocinnamic acid (HCCA) in 1:1 acetonitrile/0.1% TFA (vol/vol), deposited on a target, and dried to allow MALDI-TOF MS analysis (Bruker Daltonik). Purification of recombinant rat ProTα was performed as described previously (). This procedure using acid phenol was also available for simple purification of endogenous ProTα for SDS-PAGE analysis. In the recombinant protein preparation, the genes for ProTα and its deletion mutants (Δ1–29 and Δ102–112) were first amplified from cDNAs derived from rat embryonic brain using specific primers (rat and mouse 5′-primer, 5′-AACATATGTCAGACGCGGCAGTGGA-3′; rat 3′-primer, 5′-AAGGATCCAGTGGAGGGTGAATAGGTCAC-3′; rat Δ1–29 5′-primer, 5′-AAGAATTCGGAAGAGACGCACCTGCC-3′; rat 3′-primer, 5′-GAGTCGACCTAGTCATCCTCATCAGTCTTC-3′; rat Δ102–112 5′-primer, 5′-AAGAATTCATGTCAGACGCGGCAGTG-3′; rat 3′-primer, 5′-GAGTCGACCTACTCAACATCATCATCCTCATC-3′). The PCR products were cloned into pGEM-T Easy and subcloned into pET16b. BL21 (DE3) cells were transformed with pET16b-ProTα. Recombinant rat ProTα and its derivatives were induced by 0.1 mM IPTG, purified (Biophoresis; ATTO), and dialyzed against PBS for later use. Recombinant and endogenous ProTα isolated by the acid-phenol extraction procedure were detected as described previously (). Primary culture of the cerebral cortex from 17 d of embryonic rats was performed according to the previously reported protocol (; ). They were seeded onto 96-well culture dishes, 8-well Lab-Tek chambers (Nunc), and 3.5- and 9.0-cm culture dishes that had been all coated with poly--ornithine (Sigma-Aldrich) and cultured in DME/F-12 medium at 37°C in 5%-CO atmosphere. For ProTα coating, recombinant ProTα was added to culture dishes and incubated for 2 h at 25°C. The dishes were washed twice with PBS for immediate use. Primary cultures of 17-d-old embryonic rat cerebral cortex were prepared as described previously (). After being cultured for 3 d, cortical neurons were washed twice with glucose-free balanced salt solution (BSS; 116 mM NaCl, 5.4 mM KCl, 1.8 mM CaCl, 0.8 mM MgSO, and 1 mM NaHPO, pH 7.3), which had been deaerated using a vacuum. After replacement of the BSS with fresh BSS containing 1 mM glucose, neurons were exposed to hypoxia (<0.4% O, 5% CO, and 95% N) for 2 h at 37°C in a commercially available culture incubator (Nuair). After the ischemic stress, the culture medium was exchanged for fresh DME/F-15 medium (1:1) containing 5% horse serum and 5% FBS, and the neurons were further incubated for the indicated periods in a 5% CO atmosphere (reperfusion). Survival activity was determined by the WST-8 reduction assay throughout the experiments. The modes of cell death were determined by various means, including PI staining, activated caspase-3, GLUT1, GLUT4, TUNEL, ATP measurement, and scanning and transmission EM analyses, as previously reported (; ,). In the GLUT translocation analyses, the cortical neurons were biotinylated (Pierce Chemical Co.), lysed, immunoprecipitated with streptavidin-conjugated beads, and subjected to Western blot analysis. Characterization of the modes of cell death and the immunocytochemistry analysis are described in the supplemental text. Cortical cells on 8-well Lab-Tek chamber slides were fixed with 4% PFA in PBS for 30 min at 25°C, followed by permeabilization with 50 and 100% methanol for 5 min each at 25°C. The cells were then rinsed twice with PBS and preincubated in blocking buffer (2% BSA with 0.1% Tween 20 in PBS) for 1 h at 25°C. Next, the cells were incubated with each primary antibodies in blocking buffer overnight at 4°C, rinsed with PBS, and incubated with FITC-conjugated anti-rabbit IgG (1:200; Santa Cruz Biotechnology, Inc.) or FITC-conjugated anti-goat IgG (1:200; Rockland) for 2 h at 25°C. The immunolabeled cells were mounted with Permafluor (Thermo Scientific). For imaging cells, a laser-scanning confocal microscope imaging system consisting of a microscope (Axiovert 200 M; Carl Zeiss MicroImaging, Inc.) and a scan module (LSM 510 META and LSM 5 PASCAL; Carl Zeiss MicroImaging, Inc.) with image browser software (Carl Zeiss MicroImaging, Inc.) were used at ambient temperature, equipped with 40×/1.3 and 63×/1.4 oil-immersion lens and nonimaging photodetection device (photomultiplier tube; Carl Zeiss MicroImaging, Inc.). The imaging medium used was immersion oil (Immersol 518; Carl Zeiss MicroImaging, Inc.). A dynamic range adjustment was used to optimize the signal for the fluorophores, and images were collected in multitrack mode (Carl Zeiss MicroImaging, Inc.). Any brightness and contrast adjustments were performed in Photoshop (Adobe). SDS-PAGE using 10–15% polyacrylamide gels and immunoblot analyses were performed as described previously (). The primary antibodies were an anti-phosphorylated PKCα antibody, anti-PKCα antibody, rabbit anti-PKCβ antibody, and rabbit anti-PKCβ antibody (1:1,000; Santa Cruz Biotechnology, Inc.). Visualization of immunoreactive bands was performed using an enhanced chemiluminescent substrate (Super Signaling Substrate; Pierce Chemical Co.) for HRP detection. To determine the activation of various PKC isoforms and G protein in the mechanism of GLUT translocation, cultures were grown in the presence of AS-ODNs for PKCα, PKCβ, PKCβ, or Gα. The AS-ODNs were diluted in water to a concentration of 20 μM and added to the cultures at a final volume of 1/50 of the culture medium every 12 h after seeding for 3 d. In parallel, some cultures were treated with the corresponding missense ODNs containing the same bases as the AS-ODNs but in a random order. None of the ODNs resembled any other sequences in the GenBank database. Using Western blot and immunocytochemical analyses, we demonstrated that treatment of cortical neurons in culture with these AS-ODNs, but not the missense ODNs, reduced the levels of the target proteins. The probes had the following sequences: PKCα AS-ODN, 5′-CGGGTAAACGTCAGC-3′ (); PKCβ AS-ODN, 5′-GTTTTAAGCATTTCG-3′; PKCβ AS-ODN, 5′-GTTGGAGGTGTCTCT-3′; PKCβ missense ODN, 5′-ACGAGCCCGAACCACCGT-3′ (); and Gα AS-ODN, 5′-ATGGACTCCAGAGT-3′ (). All the ODNs were purchased from QIAGEN. Bax and Bim siRNA constructs were purchased from Ambion (siRNA ID 49750 and 47149). Gene silencing was attained by transfection of siRNA into cells using Lipofectamine 2000 transfection reagent (Invitrogen) according to the manufacturer's instructions. The gene silencing was verified by detecting protein with immunocytochemical analysis 48 h after the transfection of primary cortical neurons with siRNA. In brief, cells (1 × 10 cells/cm) grown in an 8-well Lab-Tek chamber slide were transiently transfected with 50 nM siRNA using 20 μl/ml Lipofectamine 2000 in a total transfection volume of 0.2 ml DME (Invitrogen). After incubation at 37°C in 5% CO for 6 h, the medium was replaced by fresh serum–containing medium. 2 d after the incubation, treated neurons were used for the characterization of cell death modes, as described. Multiple comparisons of analysis of variance followed by test were used for statistical analysis of the data. The criterion of significance was set at P < 0.05. All the results are expressed as the mean ± SEM. The supplemental text contains additional methodological details on characterization of anti-ProTα IgG used, as well as protocols used for cell survival activity, intracellular ATP levels, [H]-2-DG uptake, PKC kinase assay, and immunostaining protocol. Table S1 shows a summary of the procedures for purifying ProTα from CM. Fig. S1 shows a characterization of anti-ProTα IgG. Fig. S2 shows an evaluation of membrane localization of GLUT1/4 by fluorescence imaging. Fig. S3 shows specific down-regulation of Gα and PKC isoforms by treatment with AS-ODNs. Fig. S4 shows immunostaining of ProTα-induced apoptosis under the Bim siRNA–treated LOG stress condition. Fig. S5 shows ProTα–Alexa 488 binding to cell membranes. Online supplemental material is available at .

Cell polarization is required for T cell processes such as transendothelial migration, proliferation, homotypic interactions, activation in response to antigen presentation, and cytotoxity (; ). Polarized cell migration or chemotaxis in response to chemoattractants stimulates leucocytes to transmigrate through the endothelial barrier to reach secondary lymphoid organs or sites of infection (). During the process of T cell polarization, morphological and functional changes result in a bipolar asymmetric shape of T cells, which is mediated by the recruitment of surface receptors, signaling complexes, and cellular organelles to discrete areas of the plasma membrane (). Polarized T cells are characterized by a leading edge, in which chemokine receptors and activated integrins (such as LFA1) are clustered, and a uropod at the back, in which ICAM-1/3 and CD44 accumulate (). The Ras-like GTPase Rap1 has been implicated in adhesion processes, such as inside-out signaling, integrin-mediated cell–matrix adhesions, and the control of cell polarity (for reviews see ; ). Rap1 and its effector protein RAPL are two key proteins that are required for the establishment of T cell polarity. Indeed, inhibition of Rap1 signaling by the overexpression of a GAP for Rap impairs chemokine-induced T cell polarization and transendothelial migration, as well as the adhesion to ICAM-1 and VCAM-1 (). Expression of the truncated mutant RAPL ΔN, which is unable to bind to Rap1, abrogates V12Rap1, as well as chemokine-induced T cell polarization (), suggesting that RAPL functions downstream of Rap1. However, little is known about the signaling pathways used by Rap1 and chemokines to induce T cell polarization. In various cell types and species, three conserved protein complexes, termed the partitioning defective (Par), Scribble, and Crumbs complexes, have been shown to regulate cell polarity (; ). Of these, the Par polarity complex, consisting of a core of Par3, Par6 (for partition-defective), and atypical PKC (aPKCλ/ι and aPKCζ), controls different aspects of cell polarity. These include polarization of astrocytes, asymmetric cell division in yeast, axon specification and synaptogenesis in neuronal cells, and apical–basal polarity in epithelial cells (for reviews see ; ; ; ). A recent study has shown that various polarity proteins (e.g., Par3, aPKC, Scribble, Dlg, and Crumbs3) are differentially localized throughout polarized T cells (), suggesting that one or more of the polarity complexes may regulate T cell polarization. Whether the Par, Scribble, or Crumbs polarity complexes are indeed functionally required for chemokine-induced T cell polarization is unknown. Rho-like GTPases have been shown to function in the polarization processes of various cells, including T cells (; ; ). In earlier studies, we have identified the T lymphoma invasion and metastasis 1 (Tiam1) gene using retroviral insertional mutagenesis in combination with in vitro selection of invasive T lymphoma variants (). Tiam1 encodes a guanine nucleotide exchange factor (GEF) that specifically activates the Rho-like GTPase Rac (). However, the physiological function of Tiam1 in lymphoid cells is unknown. We have recently shown, along with other studies, that Tiam1 interacts with Par3 of the Par polarity complex, and thereby is a critical component of Par-mediated regulation of neuronal and epithelial (apical–basal) cell polarity (; ; ; ). Moreover, Tiam1 is able to associate with Rap proteins in fibroblasts (), suggesting that Tiam1 may control Rap1-induced T cell polarization. Therefore, we have investigated the potential function of Tiam1 and the Par polarity complex in T cell polarization. We show here that Tiam1 in conjunction with the Par polarity complex is an important regulator of Rap1- and chemokine-induced polarization and chemotaxis of T cells. GTPases of the Rho and Rap family have been implicated in T cell polarization (; ). To identify one of the first requirements for T cell polarization, constitutively active mutants of these GTPases (i.e., V12Cdc42, V12Rac1, and V12Rap1) were expressed in BW5147 T lymphoma. None of the constitutively active Rho GTPases tested were able to induce polarization (). This indicates that, although necessary (; ), activated Rac1 and Cdc42 are not sufficient to induce T cell polarity. In contrast, expression of constitutively active Rap1A (V12Rap1) induced a polarized phenotype in ∼75% of the BW5147 T lymphoma cells, as determined by morphological changes and the localization of CD44 in the uropod (). A wild-type (WT) form of Rap1A (WTRap1) was unable to induce T cell polarization in BW5147 cells (), indicating that the activity of Rap1 was necessary to induce the polarization process. To confirm that V12Rap1 expression induces a fully polarized phenotype, we investigated the localization of additional proteins reported to be restricted either to the leading edge or to the uropod during chemokine-induced T cell polarization. Talin, CXCR4, and LFA-1, were present at the leading edge of V12Rap1-expressing T cells, and were excluded from the uropod, where CD44 and ezrin specifically accumulated (). These results confirm that V12Rap1 expression is sufficient to induce T cell polarization. Interestingly, V12Rap1 was not uniformly distributed at the plasma membrane, but was strongly enriched at the leading edge, where it colocalized with CXCR4 (), suggesting that V12Rap1 locally initiates downstream signaling pathways required for T cell polarization. Because Rap1 is activated by chemokines within seconds (), our results suggest that local Rap1 activation is one of the key events initiating T cell polarization. We therefore used V12Rap1-expressing T lymphoma cells as a model to study the biochemical events leading to T cell polarization. Par3 of the Par polarity complex is asymmetrically localized in the uropod-containing murine T-cell line MD45 (). Therefore, we analyzed if proteins of the Par polarity complex and Cdc42, which is a major activator of the Par polarity complex, also differentially localize in polarized V12Rap1-expressing BW5147 T lymphoma cells. Confocal analysis showed that Par3, PKCζ, and Cdc42 were uniformly distributed in nonpolarized control cells ( and not depicted). In polarized V12Rap1-expressing cells, Cdc42, Par3, and PKCζ were devoid from the uropod where CD44 accumulates, and colocalized at the leading edge where CXCR4 is present (). RhoA was not differentially localized in polarized cells (Fig. S1, available at ). Moreover, V12Rap1 colocalized with the Par polarity proteins at the leading edge (Fig. S2). These data suggest that Par polarity proteins may function in V12Rap1-induced T cell polarization. To test whether Rap1 can activate Cdc42 and the Par polarity complex in T cells, we investigated if exogenous expression of V12Rap1 in BW5147 T lymphoma cells modifies the activation status of Cdc42 and the Par polarity complex. Expression of V12Rap1 in BW5147 cells enhanced Cdc42 activation compared with untransfected control cells (), indicating that Rap1 is able to activate Cdc42. Atypical PKCζ is a key protein in the Par polarity complex, and its activation by phosphorylation controls Par-mediated cellular polarization (). To determine the activation state of the Par polarity complex, we analyzed the phosphorylation status of PKCζ on Thr410 in polarized and nonpolarized BW5147 cells. As shown in , PKCζ phosphorylation was increased in polarized V12Rap1-expressing BW5147 cells when compared with nonpolarized cells, indicating that V12Rap1 activates the Par polarity complex, and thereby PKCζ, in polarized T cells. Because activated PKCζ is localized at the plasma membrane in contrast to nonactivated PKCζ (), we also determined the intracellular localization of PKCζ by a biochemical fractionation method. Consistent with the increased phosphorylation observed upon polarization, we found that PKCζ was enriched in the membrane fraction of polarized V12Rap1-expressing BW5147 T lymphoma cells when compared with nonpolarized cells (). Together, these data indicate that constitutively active V12Rap1 leads to the activation of Cdc42 and the Par polarity complex, as determined by the activation and membrane translocation of PKCζ. To investigate whether the activation of Cdc42 has a functional effect in V12Rap1-induced T cell polarity, we inhibited Cdc42 activity in BW5147 cells. As shown in , expression of dominant-negative N17Cdc42 reduced the number of polarized V12Rap1-expressing cells from 70 to ∼30%. This indicates that Cdc42 activity is required for V12Rap1-induced T cell polarization. To investigate the hierarchical activation of Cdc42 and the Par polarity complex in V12Rap1-expressing cells, we also analyzed the phosphorylation status of PKCζ in the presence of N17Cdc42. We found decreased PKCζ phosphorylation in the nonpolarized cells coexpressing V12Rap1 and N17Cdc42 compared with polarized V12Rap1–expressing cells (). This suggests that Cdc42 activates the Par polarity complex, leading to T cell polarization. Indeed, expression of Par3 shRNA, which reduced the Par3 protein levels to ∼50% (Fig. S3, available at ), impaired V12Rap1-induced T cell polarization when compared with cells expressing control shRNA (). In addition, inhibition of PKCζ downstream signaling by the expression of a kinase-dead mutant of PKCζ (PKCζKD) in BW5147 cells (), or by a myristoylated PKCζ pseudosubstrate (PKCζ inhibitor; ) in primary T lymphocytes (), abrogated V12Rap1-induced polarity in both BW5147 cells and primary T cells. From these data, we conclude that V12Rap1 activates Cdc42, leading to the activation of the Par polarity complex that is required for the establishment of T cell polarity. The Par complex is known to control changes in the actin and microtubule cytoskeleton that are required for cell polarization. The Rho GTPase Rac controls actin cytoskeleton remodeling during various processes, including cell migration and cell polarization (; ). Because T cell polarization is dependent on actin remodeling (), it is likely that Rac has a function in V12Rap1-induced T cell polarization. Indeed, expression of V12Rap1 or PKCζ enhanced Rac activity in BW5147 T lymphoma cells (). Most importantly, V12Rap1-mediated Rac activation could be inhibited by the expression of dominant-negative Cdc42 () or PKCζKD (). These data indicate that Rap1 regulates Rac activity through the Cdc42–Par–PKCζ pathway, and that Rac acts downstream of PKCζ to mediate T cell polarization. Indeed, N17Rac1 expression inhibited T cell polarization in BW5147 cells induced by V12Rap1 (), demonstrating that Rac activity is required for V12Rap1-induced polarity in T lymphoma cells. Therefore, we conclude that Rac is activated downstream of the Par polarity complex and PKCζ to mediate the actin remodeling required for the polarization of T cells induced by V12Rap1. Tiam1 is a GEF that specifically activates Rac (). Tiam1 has recently been shown to act in conjunction with the Par polarity complex in the establishment of neuronal and epithelial cell polarity (; ; ; ). Because Tiam1 is also able to interact with Rap1 in fibroblasts (), we hypothesized that Tiam1 could control V12Rap1-induced Rac activation in T cells. Interestingly, GST pulldown experiments show that endogenous Tiam1 interacts with activated V12Rap1, but not with WTRap1 (), suggesting that the Tiam1–Rap1 interaction is dependent on the activation state of Rap1. Similarly, coimmunoprecipitation experiments show that endogenous Tiam1 interacts with V12Rap1 in polarized T lymphoma cells (). Immunoprecipitation of endogenous Tiam1 revealed that Par3 associates with Tiam1 irrespective of the presence of V12Rap1 (). Interestingly, Tiam1 interacts with PKCζ only in polarized V12Rap1-expressing T cells, suggesting that Tiam1 is associated with the activated Par complex during T cell polarization. Indeed, Tiam1 colocalizes with PKCζ and V12Rap1 in the front of polarized V12Rap1-expressing cells, whereas it is homogenously distributed in nonpolarized T cells (). From these data, we conclude that Tiam1 interacts with V12Rap1 and components of the Par polarity complex, and may thereby have a function in connecting the Par complex and Rac activity at sites where Rap1 is activated and cell polarization is initiated. To investigate whether Tiam1-mediated Rac activation is essential for V12Rap1-induced T cell polarization, we used a dominant-negative mutant of Tiam1, corresponding to the PHnCCex domain of Tiam1 (). This mutant lacks the catalytic DH domain, still contains the Par3 interaction domain, and inhibits Tiam1-mediated Rac activation (). Expression of dominant-negative Tiam1 in BW5147 T lymphoma inhibits both V12Rap1- induced Rac activation () and cell polarization (). These results demonstrate that Tiam1 not only associates with Rap1 and the Par polarity complex but is also essential for the Rac activation required for Rap1-induced T cell polarization. To further substantiate the function of Tiam1 in V12Rap1-induced T cell polarization, we used primary lymphocytes isolated from WT (Tiam1+/+) and Tiam1 knockout (Tiam1−/−) mice (). V12Rap1-IRES-GFP was transduced into lymphocytes of both genotypes, and the degree of cell polarization was analyzed in the GFP-expressing cells. V12Rap1 induced cell polarization in ∼80% of the GFP-positive Tiam1+/+ T lymphocytes, whereas Tiam1-deficient (Tiam1−/−) cells showed only background polarization (∼20%) as found in nontransduced cells (). These data demonstrate that Tiam1 is also required for V12Rap1-induced cell polarization in normal T lymphocytes. The polarized characteristics induced by chemokines and V12Rap1 expression are indistinguishable in terms of morphology and cell surface receptor expression (). Therefore, we investigated whether Rap1 and the Par polarity complex also function in chemokine-induced T cell polarization. First, we analyzed the activation kinetics of Rap1, Cdc42, the Par complex, and Rac1 in primary T lymphocytes upon chemokine stimulation. As shown in , Rap1, Cdc42, PKCζ, and Rac1 are rapidly and transiently activated after stromal cell–derived factor-1 α (SDF1α) stimulation. Similar results were found in chemokine-treated Jurkat cells (unpublished data). Upon SDF1α stimulation of Jurkat cells, WTRap1A is recruited to the leading edge (), suggesting that upon activation it induces polarity by activation of the Par polarity complex at specific sites in T cells. These data are consistent with the intracellular localization of V12Rap1 in polarized BW5147 T lymphoma cells (Fig. S2). To further substantiate that Rap1 activates the Par polarity complex during chemokine stimulation, Rap1 downstream signaling was inhibited by the expression of Rap1-GAP. As shown in , Rap1-GAP inhibited chemokine-induced activation of Cdc42, PKCζ, and Rac in Jurkat T cells. Consistent with these findings, chemokine-induced polarity was strongly impaired in cells expressing Rap1-GAP (). From these results, we conclude that Rap1 activity is required for both chemokine-induced T cell polarization and activation of the Par polarity complex. To confirm that Rap-induced activation of the Par polarity complex is necessary for chemokine-induced T cell polarity, we investigated the function of the Par polarity complex and its downstream effector Tiam1 in T cell polarization. First, we analyzed the function of Tiam1 in Rac activation and T cell polarity induced by the chemokine SDF1α. Interestingly, stimulation with SDF1α induced an increase in Rac activation in Tiam1+/+, but not in Tiam1−/−, lymphocytes (), indicating that Tiam1 is necessary for SDF1α-induced Rac activation. These data are consistent with the requirement of Tiam1 for Rac activation induced by V12Rap1 (). We also determined the SDF1α-induced degree of T cell polarization in Tiam1+/+ and Tiam1−/− T lymphocytes. As shown in , SDF1α-induced polarity was ∼50% reduced in Tiam1−/− T cells compared with Tiam1+/+ T cells. Comparable results were found using the secondary lymphoid tissue chemokine (unpublished data). From these data, we conclude that Tiam1-mediated Rac activation controls, to a large extent, the chemokine-induced polarization of T cells. To functionally test the effect of Tiam1 in chemokine-induced T cell polarity, we analyzed the chemotactic migration capacity of T cells of both genotypes in response to SDF1α using a Boyden chamber assay. Tiam1−/− T cells showed ∼50% reduction in their chemotactic response to different concentrations of SDF1α compared with Tiam1+/+ T cells (). Based on our findings in V12Rap1-induced polarity in T lymphoma cells, we also analyzed the association of Tiam1 and the Par polarity complex in primary T lymphocytes upon chemokine stimulation. Similarly, as found for V12Rap1-induced polarization of T lymphoma cells (), SDF1α stimulation of primary T cells promoted Tiam1–PKCζ association, whereas interaction between Tiam1 and Par3 was independent of SDF1α (). In addition, Tiam1 and PKCζ of the Par complex colocalized at the leading edge of SDF1α-stimulated primary T cells (). Interestingly, Tiam1-deficiency did not inhibit the activation of PKCζ induced by SDF1α (), whereas inhibition of PKCζ signaling in Tiam1+/+ T lymphocytes inhibited SDF1α-induced Rac activation (). These findings confirm our earlier conclusion that Tiam1 activates Rac downstream of PKCζ and the Par polarity complex. Treatment of Tiam1+/+ T cells with PKCζ pseudosubstrate also inhibited SDF1α-induced T cell polarization () and chemotaxis (). Intriguingly, the inhibition of PKCζ signaling did not alter the residual 50% polarization and chemotactic migration capacity of Tiam1−/− T lymphocytes (), indicating that Tiam1 and the Par polarity complex function in the same signaling pathway during chemokine-induced T cell polarization and chemotaxis. Apparently, residual chemokine-induced polarization and chemotaxis, as found in the primary Tiam1−/− T cells, are not dependent on the Par polarity complex. Rap1 and its effector protein RAPL function in chemokine-induced integrin activation (), migration, and polarization of T cells (; ), but how these processes are regulated is unknown. In this study, we show that the Par polarity complex, in conjunction with Tiam1 and Rac, regulates both Rap1-induced T cell polarization and chemokine-induced polarization and migration of primary T cells. We found that Rap1 triggers polarity by activating Cdc42, which in turn activates the Par polarity complex. Activation of PKCζ through the Par polarity complex leads to the activation of Rac via Tiam1 (). Because V12Rap1-induced Rac activation is inhibited by dominant-negative N17Cdc42 and kinase-dead PKCζ, it is unlikely that V12Rap1 is able to activate Rac directly in T cells, as has been reported in fibroblasts (). In lymphocytes, Tiam1 mediates V12Rap1-induced Rac activation as a result of the activation of Cdc42 and the Par polarity complex. Although we show that Cdc42 and Rac are required for T cell polarization, constitutively active V12Cdc42 or V12Rac1 are not sufficient to induce polarization, in contrast to constitutively active V12Rap1. Apparently, specific properties of Rap1 are required to initiate T cell polarity. Similar to Rap1, other proteins that are capable of inducing T cell polarity (i.e., RAPL and Mst1) have been implicated in inside-out signaling of integrins, and are localized in small vesicles in nonstimulated cells (, ). Upon stimulation, these proteins translocate to a specific site at the plasma membrane, specifically at the future leading edge (). Indeed, we found that WTRap1 is present in the leading edge of chemokine-induced polarized T cells upon SDF1α stimulation, where it colocalizes with the proteins of the Par polarity complex (). During axon specification in neuronal cells, Rap1B acts as the primary cue upstream of the Par polarity complex and defines which of the growing neurites becomes the future axon (). The striking similarity by which Rap1 in conjunction with the Par polarity complex determines polarization of T cells (as shown in this study) and axon specification in neuronal cells (; ) suggests that Rap proteins are able to recruit the Par polarity complex in various cellular systems, and thereby controls the initiation of cell polarity. We found that Rap1, through activation of Cdc42, not only localizes but also activates the Par polarity complex and Rac through Tiam1. Because Tiam1 associates with both activated Rap1 and Par3 (), Tiam1 may also function as a scaffold protein that couples activated Rap1 to the Par polarity complex. In fact, in fibroblasts it has been shown that Rap1 promotes cell spreading by binding Tiam1, thereby localizing Rac activity at specific sites in cells (). Moreover, we found that Tiam1 associates with active, but not inactive, Rap1. Based on these data, it is tempting to speculate that upon a polarization signal in lymphoid cells, Rap1 is activated at a specific site of the plasma membrane and recruits Tiam1 and the Par polarity complex to initiate polarity. Rap1 activates Cdc42 through an associated unknown GEF. Activated Cdc42 binds to Par6 (; ), which leads to activation of the Par polarity complex, including PKCζ, and, subsequently, to activation of Rac through Tiam1. Within this scenario, Tiam1 may have two functions; i.e., to connect the Par polarity to Rap1 at the site where T cell polarity is initiated and to activate Rac downstream of the Par polarity complex to achieve the actin remodeling required for T cell polarization. Consistent with this, it has been shown that PKCζ is able to phosphorylate Tiam1 (), which could regulate Tiam1 activity. The fact that we identified Tiam1 as a gene that promotes infiltration of T lymphoma cells into fibroblast monolayers () might be explained by its function in polarity signaling in lymphoid cells. In fact, we have shown that increased Tiam1/Rac signaling promotes the infiltration of T lymphoma cells into fibroblast monolayers in a polarized fashion (). Our data provide new insights into the mechanisms by which chemokine-mediated polarity is established. However, it is likely that additional signaling pathways play a role. In fact, the incomplete inhibition of SDF1α-induced polarization in T cells lacking Tiam1 or PKCζ activity suggests that other polarity complexes contribute to chemokine-induced T cell polarization. Tiam1−/− mice develop, grow, and reproduce normally (; unpublished data). In addition, no obvious defects have been found in mice deficient for the Par polarity protein PKCζ, except for a small delay in the formation of secondary lymphoid organs (; ). Apparently, both in vitro and in vivo, other polarization pathways contribute to the polarization process of T cells and/or can overcome the deficiency of the Par3–PKCζ–Tiam1 pathway. Indeed, proteins of the Scribble and Crumbs polarity complexes have been found asymmetrically distributed in polarized T cells (), suggesting that they also contribute to T cell polarization. Dlg and Scribble control uropod formation in the uropod-containing T-cell line MD45, and regulate asymmetric distribution of proteins in T cells (). In addition, other Rac-specific GEFs have been shown to function in T cell polarization. Deregulation of Vav1-signaling by the expression of a dominant-active or -negative mutant of Vav1 reduces chemokine-induced T cell polarization (). However, T cell polarization is normal in Vav1-deficient T cells compared with Tiam1+/+ cells (), suggesting that other Vav proteins (e.g., Vav 2 or 3) or Tiam1 compensate for the loss of Vav1. As Vav2 has also been shown to associate with Rap1 in fibroblasts (), Vav proteins might also have a potential role in Rap1-induced polarity signaling. DOCK2 is an unconventional GEF for Rac that belongs to the CDM family (CED-5 in , DOCK180 in man, and Myoblast in ; ). In DOCK2-deficient mice, various defects in T lymphocytes have been described, including migration defects in response to chemokines, lymphocytopenia, and atrophy of lymphoid follicles (). Deletion of DOCK2 inhibits chemokine-induced formation of a bipolar shape (; ), but the mechanism by which DOCK2 influences T cell polarization is unknown. Chemokine-induced Rap1 and PKCζ activation is not dependent on DOCK2 (). Moreover, V12Rap1 induces polarization in BW5147 T lymphoma cells that do not express DOCK2 (). These data exclude a function of DOCK2 in Par complex–mediated polarization of T cells. It is conceivable that at least two distinct pathways are required for chemokine-induced T cell polarization. One of these pathways involves DOCK2 through an unknown mechanism and the other pathway involves Rap1 and the Par polarity complex in conjunction with Rac activators such as Tiam1 or possibly Vav1–3. Other polarity complexes may also contribute to Rap1- or chemokine-induced polarization of T cells, as they have also a function in other Par complex-mediated polarization processes such as the establishment of epithelial apical–basal cell polarity (). Collectively, our data implicate Tiam1 in conjunction with the Par polarity complex in Rap1- and chemokine-induced T cell polarization. To achieve T cell polarization, Rap1 is activated by chemokine stimulation leading to activation of Cdc42, and thereby of the Par polarity complex. Tiam1 associates with Rap1 and components of the Par complex and may function to connect the Par polarity to Rap1 at the site where T cell polarity is initiated. Furthermore, Tiam1 is required to activate Rac downstream of the Par complex, presumably to regulate actin remodeling required for T cell polarization. The following antibodies were used in immunoprecipitations, immunoblottings, and immunofluorescent stainings. Antibodies against PKCζ, Tiam1 (C16), Cdc42 (P1), GST, RhoA, and Rap1 were purchased from Santa Cruz Biotechnology, Inc. Antibodies against c-myc (9E10) and Rac1 were purchased from Millipore. Antibodies against Par3 were purchased from Zymed Laboratories. Antibodies against Nf-κB and phospho-PKCζ/λ (Thr 410/403) were purchased from Cell Signaling Technology. Antibodies against talin were purchased from Sigma-Aldrich. Antibodies against ezrin and CD44 were purchased from BD Biosciences. Antibody against GFP was purchased from Roche. Antibody against ICAM3 was purchased from Abcam PVC. Antibody against LFA-1 (M17-4) was provided by E. Roos (The Netherlands Cancer Institute, Amsterdam, Netherlands). All the conjugated secondary antibodies for immunofluorescent staining were purchased from Invitrogen. Poly--lysine was purchased from Sigma- Aldrich. Recombinant SDF1α was purchased from Peprotech. The PKCζ pseudosubstrate inhibitor was purchased from Calbiochem. Myc-tagged V12Rac1, N17Rac1, V12Cdc42, or N17Cdc42 sequences were cloned into the retroviral vector LZRS-IRES-Neo as previously described (). Myc-PKCζ-WT and myc-PKCζ-K281W were subcloned into the Swa1 and Not1 sites of the LZRS-IRES-GFP vector (). Myc-tagged V12Rap1 was subcloned into the XhoI and NotI restriction sites of the retroviral vectors pMX-eGFP and LZRS-IRES-Bsd. Myc-tagged WTRap1 (derived from the University of Missouri- Rolla cDNA Resource Center) was subcloned into the XhoI and NotI restriction sites of the retroviral vector LZRS-IRES-Bsd. GST-WTRap1 and GST-V12Rap1 (derived from UMR cDNA Resource Center) were generated by subcloning WTRap1 and V12Rap1 into the SmaI restriction site of pGEX 6P2. Dominant-negative Tiam1 (PHnCCEx) has been previously described (). pMT2-HA-Rap1-GAP (Rap1-GAP) was provided by J.L. Bos (University Medical Center, Utrecht, Netherlands). The shRNA oligonucleotides targeting Par3 RNA (Par3 shRNA) and luciferase RNA have been designed as previously described (; ). Sequences of the primers are as follow: Par3 shRNA sense primer, 5′-GATCCCCGGCATGGAGACCTTGGAAGTTCAAGAGACTTCCAAGGTCTCCATGCCTTTTTGGAAA-3′; Par3 shRNA antisense primer, 5′-AGCTTTTCCAAAAAGGCATGGAGACCTTGGAAGTCTCTTGAACTTCCAAGGTCTCCATGCCGGG-3′; luciferase shRNA sense primer, 5′-GATCCCCCGTACGCGGAATACTTCGATTCAAGAGATCGAAGTATTCCGCGTACGTTTTTGGAAA-3′; and luciferase shRNA antisense primer, 5′-AGCTTTTCCAAAAACGTACGCGGAATACTTCGATCTCTTGAATCGAAGTATTCCGCGTACGGGG-3′. BW5147 T lymphomas and Jurkat JA16 T cell subclone (provided by J.A. Nunes, Institut National de la Santé et de la Recherche Médicale, Marseille, France; ) were grown in RPMI 1640 medium supplemented with 10% fetal calf serum. Rac-11P cells were cultured in DME supplemented with 10% fetal calf serum. A laminin-5 matrix was obtained by culturing Rac-11P cells to confluency, after which cells were detached with 10 mM EDTA in PBS containing a mix of protease inhibitors (Sigma-Aldrich) at 4°C. Phoenix retrovirus packaging cells () were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum. Single T-cell suspensions were isolated from lymph nodes and spleen of 4–8-wk-old Tiam1+/+ and Tiam1−/− mice (). Negative selection was performed by using a pan T-cell isolation kit (MACS; Miltenyi Biotec), according to the manufacturer's instructions. T cell purity was >95% as determined by flow cytometry. Jurkat cells (10 × 10) were electroporated at 960 μF, 250 V, for 25 ms with 20 μg of plasmid using a gene Pulser Xcell (Bio-Rad Laboratories). Gene expression was assessed after 24 h. BW5147 T lymphoma cells were infected with retrovirus containing supernatants, as previously described (). Cells were selected for 2 wk, unless otherwise specified. For retroviral transduction of primary T lymphocytes, single T-cell suspensions were stimulated with 3 μg/ml CD3ɛ antibody (145-2C11; R&D Systems) and 25 U/ml IL-2 (Peprotech) for 18 h at 37°C. Subsequently, 3 × 10 T lymphocytes were incubated with 1 ml of virus containing supernatant in the presence of 8 μg/ml polybrene (Sigma-Aldrich) and spin-infected for 2 h at 2,000 rpm. After a 5-h incubation, cells were washed and allowed to grow for 48 h. Infection efficiency was between 10 and 30%. Lysates were prepared in standard NP-40 lysis buffer (10% glycerol, 50 mM Tris-HCl, pH 7.4, 1% NP-40, 150 mM NaCl, 20 mM NaF, 2 mM MgCl, 1 mM NaVO, and 1 mg/ml protease inhibitors cocktail [Sigma-Aldrich]) for 10 min at 4°C and centrifuged at 13,000 rpm for 10 min at 4°C. For fractionation experiments, pellets of BW5147 cells were lysed using the ProteoExtract Subcellular Proteome Extraction kit (Calbiochem) according to the manufacturer's instructions. The efficiency of subcellular fractionation was determined by SDS-PAGE and immunoblotting with selected marker proteins. Rac and Cdc42 activity was determined as described previously (), using a biotinylated Rac1–Cdc42 interactive binding motif peptide of PAK1. For this, BW5147 cells were starved for 18 h in IMDM medium with 0.5% BSA and lysed in standard NP-40 buffer. Purified T cells or Jurkat cells (10 × 10 cells) were stimulated as indicated with 500 ng/ml SDF1α, and lysed in standard NP-40 buffer. Rap activity was determined as previously described () using a GST-RalGDS-RBD fusion protein. For this, purified T cells (10 × 10 cells) were stimulated as indicated with 500 ng/ml SDF1α, and lysed in standard NP-40 buffer. The inner and outer face of Transwells (Costar; 5-μm pore size) were coated with 0.5% Ovalbumin (Ova) for 2 h at RT. Purified T cells (10 in 150 μl RPMI and 0.1% Ova) were treated with 2 μM PKCζ inhibitor for 1 h, where indicated, and loaded in an Ova-coated transwell, which was placed into a 24-well plate containing 250 μl RPMI supplemented with 0.1% Ova and various concentrations of SDF1α. After 1 h at 37°C, the cells that migrated into the lower chamber were collected and counted. Data were expressed as the mean ± the SD. Comparisons between groups were analyzed with tests. Data were considered as statistically significant when P ≤ 0.05. Fig. S1 shows the intracellular localization of PKCζ, Par3, and Cdc42 at the leading edge of V12Rap1-expressing BW5147 cells, in comparison with RhoA localization. Fig. S2 shows the colocalization of PKCζ, Par3, and Cdc42 with V12Rap1 at the leading edge of V12Rap1-expressing BW5147 cells. Fig. S3 shows the down-regulation of Par3 expression by shRNA in BW5147 cells. The online version of this article is available at .

The osteoclast is a polykaryon whose capacity to mobilize bone requires the organization of its unique cytoskeleton. Thus, upon mineralized matrix recognition, the osteoclast polarizes its fibrillar actin, eventuating in the formation of an acidified extracellular microenvironment wherein the resorptive machinery of the cell is activated. Failure to undergo this polarization event results in osteoclast dysfunction and, consequently, in varying degrees of osteopetrosis (). made the surprising observation that the principal phenotype of c-Src–deleted mice resides in the skeleton. These animals develop severe osteopetrosis despite an abundance of osteoclasts. Thus, the disorder does not reflect the failure of osteoclast recruitment but defective osteoclast function: specifically, disorganization of the cell's actin cytoskeleton (). In fact, c-Src is the dominant Src family kinase (SFK) expressed in the osteoclast and regulates the cell's cytoskeleton as an adaptor protein as well as through its kinase activity (; ). Syk is another tyrosine kinase that modulates osteoclast function, at least in vitro (; ). The discovery of Syk in this regard was a consequence of the observation that immunoregulatory adaptor molecules such as Dap12 and FcRγ (; ), which bear immunoreceptor tyrosine-based activation motifs (ITAMs), are central to osteoclastogenesis, as mice lacking both proteins are osteopetrotic because of the failure to generate osteoclasts (). Alternatively, whether ITAM proteins impact the resorptive capacity of mature osteoclasts is unknown. The signaling events emanating from these proteins are activated by SFKs, which phosphorylate tyrosine residues within the ITAM (). These ITAM phosphotyrosines bind to and activate Syk (; ). However, the maintenance of ITAM-initiated Syk activity is under the aegis of the associated SFK (). Activated ITAM-bound Syk targets guanine nucleotide exchange factors such as Vavs, leading to induction of the Rho GTPase Rac (). Integrins are αβ-transmembrane heterodimers and the principal cell/matrix recognition molecules. The osteoclast is particularly rich in the αvβ3 integrin, which mediates the capacity of the cell to polarize, spread, and optimally degrade bone. Although less pronounced than that of c-Src–deficient mice, those lacking αvβ3 develop enhanced bone mass as a result of osteoclast dysfunction, again reflecting a failure to undergo cytoskeletal organization (; ). The β3-integrin subunit is also expressed by platelets, wherein it associates with the αIIb chain. Similar to ITAM-bearing proteins, αIIbβ3 recognizes Syk (), which is also essential for integrin-mediated signaling in neutrophils (). Interestingly, Syk binds directly to the β3 subunit of αIIbβ3 via the kinase SH2 domain but, unlike its interaction with ITAMs, does so independently of the phosphorylation of integrin tyrosine residues (). In contrast to its association with phosphorylated ITAMs, recognition of the β3 subunit itself has no immediate impact on Syk activation, which occurs upon integrin/ligand occupancy and is presumably secondary to the induction of an associated SFK (). As a result of αvβ3-induced Syk phosphorylation in the osteoclast, Vav3, which regulates the cell's cytoskeleton, is activated (). Thus, Syk is downstream of immunoregulatory proteins such as Dap12 and FcRγ, which govern osteoclastogenesis, and of integrins, which regulate the cytoskeleton. Because αvβ3 and its associated proteins are therapeutic targets in the context of pathological bone resorption (), we explored the mechanism by which the integrin regulates Syk in the osteoclast. We find that Syk, c-Src, and αvβ3 form a ternary complex in the cell. Syk–c-Src binding requires the terminal three amino acids of the β3 subunit and the kinase and C-terminal SH2 domains of Syk. The consequence of αvβ3-ligand occupancy is the activation of c-Src, which, in an ITAM-dependent manner, phosphorylates Syk, thus organizing the osteoclast cytoskeleton. In keeping with these observations, Syk osteoclasts have a disorganized cytoskeleton leading to subnormal bone resorption in vitro and in vivo. To overcome the perinatal lethality of Syk mice, we generated bone marrow chimeras by transplanting Syk fetal liver cells into lethally irradiated wild-type (WT) recipients (). Syk (nongenotyped Syk or Syk mice, which are phenotypically identical) fetal liver cells, which were injected into similar hosts, served as a control. These animals are designated Syk and Syk chimeras. Donor and recipient cells are distinguished on the basis of surface antigen Ly-5.1 expression. Specifically, recipient mice (B6.SJL) express Ly-5.1, whereas donors (129.sv) do not. Purified osteoclast precursors in the form of bone marrow macrophages (BMMs) that recovered from Syk chimeras consist almost exclusively of donor cells (no Ly-5.1; Fig. S1 A, available at ). Furthermore, Syk chimeric BMMs express normal levels of receptor activator of nuclear factor κB (RANK) and the macrophage colony–stimulating factor (MCSF) receptor c-Fms (Fig. S1 B). Syk is essential for T cell development, and we asked whether the same is true regarding the osteoclast. To this end, we cultured Syk and Syk BMMs with MCSF and RANK ligand (RANKL) for 5 d. Syk-deficient osteoclasts, which were generated in culture, are distinctly abnormal. Both genotypes differentiate into tartrate-resistant acidic phosphatase (TRAP)–expressing multinucleated cells. However, Syk BMMs form sheets of characteristic well-spread osteoclasts, whereas those lacking Syk are small with an irregular crenated appearance, suggesting a cytoskeletal defect (). To ensure that these morphological abnormalities of Syk osteoclasts do not reflect arrested differentiation, we measured a series of markers of osteoclastogenesis in BMMs exposed to MCSF and RANKL with time. The expression of characteristic osteoclastogenic proteins is not delayed in the Syk cells (Fig. S2 A, available at ) nor are intracellular signaling events required for efficient osteoclast differentiation (Fig. S2 B). These pathways include RANKL activation of nuclear factor κB, which is assessed by IκB-α phosphorylation and degradation, as well as JNK, ERK1/2, and p-38 phosphorylation. Furthermore, MCSF-driven ERK1/2 and AKT phosphorylation is normal in Syk osteoclasts (Fig. S2 C). These data establish that Syk deficiency does not impair osteoclast differentiation, and, therefore, we turned to osteoclast function. To this end, we plated BMMs on dentin, added MCSF and RANKL for 6 d, and stained the actin cytoskeleton with FITC-phalloidin. Documenting that the abnormal shape of the Syk polykaryons reflects deranged cytoskeletal organization, they are incapable of forming actin rings (). In keeping with their dysfunctional cytoskeleton, Syk osteoclasts are also unable to degrade mineralized matrix in vitro, as indicated by a complete absence of dentin-resorptive lacunae (). As expected, the failure of Syk-deficient osteoclasts to spread and form actin rings is completely normalized by retroviral reconstitution of the WT tyrosine kinase () as is the cells' resorptive capacities (). In contrast, the kinase-inactive mutant Syk fails to rescue the mutant cell's spreading defect (). Thus, the kinase activity of Syk is central to its capacity to organize the osteoclast cytoskeleton. Rac is a downstream effector of αvβ3-mediated cytoskeletal organization (). To determine whether Rac signals downstream of αvβ3-activated Syk in committed osteoclast precursors, we cultured Syk or Syk BMMs in osteoclastogenic conditions for 3 d. The cells were lifted and maintained in suspension or replated on αvβ3 ligand for 30 min. Densitometric analysis of reveals that αvβ3 occupancy increases Rac activation 1.8-fold in Syk cells, whereas no detectable change (1.1-fold) occurs in the absence of the tyrosine kinase. Having established that Syk deficiency impacts osteoclastic bone resorption in vitro, we asked whether the same obtains in vivo. Similar to the in vitro situation, osteoclast number in Syk chimeric mice 2 mo after the generation of the animals is indistinguishable from that of irradiated controls transplanted with WT liver cells whether in the basal state or after stimulation with the resorptive agonist parathyroid hormone (PTH) (1–34) (). On the other hand, Syk-deficient osteoclasts are small and appear incapable of adhering to the bone surface or forming well-demarcated resorption lacunae after receiving the bone-resorptive hormone (). Not only does Syk deficiency reduce the percentage of bone surface covered by osteoclasts adherent to bone, but, unlike WT, the proportion of bone-apposed cells does not increase in response to PTH (). Confirming an osteoclast defect in these Syk chimeras, PTH(1–34)-enhanced serum levels of the global bone-resorptive marker CTx are blunted (). For reasons unknown, Syk chimeras survive only 1–2 mo after transplantation, preventing the meaningful assessment of skeletal mass with age. On the other hand, 18.5-d Syk embryos have increased bone density as compared with Syk littermates ( and Fig. S3, available at ). Moreover, osteoclasts resident in Syk-deleted embryos have the same crenated appearance as their counterparts in radiation chimeras (). In keeping with increased bone density, the marrow space of Syk mice contains a network of trabeculae that is absent in their Syk counterparts. Alternatively, growth plate morphology of the mutant mice is unremarkable. Osteoblasts may increase skeletal mass by accelerating bone formation or arresting osteoclast formation and function as a result of reduced RANKL or increased osteoprotegrin (OPG) expression. To address this issue in the context of Syk deficiency, we cultured Syk and Syk calvarial osteoblasts in osteogenic conditions for 20 d, after which they were stained with Alizarin red. As seen in (A and B), there is no apparent difference in the capacity of the two genotypically distinct osteoblasts to generate mineralized bone nodules. Next, we treated the same combination of osteoblasts with 1,25-dihydroxyvitamin D or TNF-α, as each is a modulator of RANKL synthesis. Similar to their bone-forming capacity, Syk and Syk osteoblasts express equivalent amounts of RANKL and OPG mRNA in response to both agents (). Finally, and in keeping with the unaltered biological activity of Syk osteoblasts, Syk protein is also not detectable by immunoblotting in their Syk counterparts (). Osteoclasts deficient in Syk, the αvβ3 integrin (), or c-Src () each have striking cytoskeletal defects eventuating in subnormal bone resorption. Furthermore, c-Src () or Syk () in other circumstances associates with the β3-integrin cytoplasmic domain, prompting us to ask whether the three molecules form a complex in osteoclasts. To address this issue, we turned to β3 BMMs transduced with human β3 (hβ3). These transductants express physiological levels of αvβ3 (). Human constructs of the integrin were used because the WT cytoplasmic component is identical to its mouse counterpart, and an antibody that effectively recognizes its extracellular domain is available. The cells were placed in MCSF and RANKL for 5 d to generate mature, adherent osteoclasts. Total cell lysates (TCLs) were then immunoprecipitated with anti–c-Src, anti-Syk antibodies, or irrelevant IgG. Both tyrosine kinases associate with αvβ3 in mature osteoclasts (). To determine whether formation of the kinase–integrin complexes depends on αvβ3 occupancy, we generated mononuclear preosteoclasts by exposing BMMs to MCSF and RANKL for 3 d. In this circumstance, the cells are committed to the osteoclast phenotype as they express TRAP, αvβ3, and c-Src () but, unlike mature osteoclasts, are easily suspended. The cells were maintained in suspension or were adherent to the αvβ3 ligand vitronectin for 30 min. Reciprocal immunoprecipitation and immunoblotting show that Syk associates with the integrin in an adhesion-dependent manner (). In contrast, β3–c-Src recognition is constitutive and is not impacted by integrin occupancy (). The interaction of both Syk and c-Src with αvβ3 in osteoclasts suggests that the two kinases may coimmunoprecipitate. To address this issue, we cultured WT BMMs in MCSF with or without RANKL for 3 or 5 d. Syk immunoprecipitates and TCLs were then immunoblotted for c-Src content. Consistent with the progressive expression of c-Src as cells assume the osteoclast phenotype, the kinase is increasingly abundant in Syk immunoprecipitates derived from MCSF- and RANKL-containing cultures and is absent in those treated with MCSF alone (). Also, in contrast to c-Src, whose expression by macrophages is specific to the osteoclast phenotype, the production of Syk is unaffected by time or RANKL. To determine the functional role of the β3 integrin in c-Src–Syk association, WT or β3 BMMs were cultured with RANKL and MCSF for 5 d, after which we immunoprecipitated c-Src or Syk and reciprocally immunoblotted the other. As shown in , association of the two tyrosine kinases is substantially reduced in the absence of αvβ3, although the expression of neither c-Src nor Syk is altered. Confirming its role in regulating c-Src–Syk recognition, retroviral reconstitution of β3-deficient osteoclasts with the integrin subunit restores coimmunoprecipitation of the kinases (). We next asked whether the integrin's state of occupancy impacts c-Src–Syk association. Committed preosteoclasts were once again generated by incubating WT BMMs in MCSF and RANKL for 3 d. The cells were lifted and either maintained in suspension or plated on vitronectin for 30 min. Syk immunoprecipitates were then immunoblotted for c-Src and Syk content. Adhesion of preosteoclasts to αvβ3 ligand enhances c-Src–Syk coprecipitation in circumstances of equal Syk content (). These data suggest that Syk mediates αvβ3 function in osteoclasts. Thus, we asked whether an essential property of the integrin, namely rapid spreading on αvβ3 ligand, is disturbed in the absence of Syk. To this end, committed preosteoclasts were suspended and replated on vitronectin-coated plates. In keeping with disturbed αvβ3 function, Syk deficiency markedly reduces the capacity of preosteoclasts to rapidly spread on the integrin's ligand( and Fig. S4, available at ). c-Src directly associates with the distal three amino acids of the β3-subunit cytoplasmic domain in vitro and in transformed cells (, ). To determine whether this motif regulates c-Src–Syk interaction in the osteoclast, we substituted these residues with the C-terminal three amino acids of the β1-integrin chain to generate hβ3–hβ1(C3R) (; ). hβ3–hβ1(C3R) or hβ3WT was retrovirally transduced into β3 BMMs, which were then placed in osteoclastogenic conditions. TRAP staining of the cultures after 5 d demonstrates that the mutated construct yields small, poorly spread osteoclasts that are indistinguishable from those generated from naive β3 BMMs, whereas the WT subunit completely rescues the osteoclast phenotype (). This observation and the diminished coprecipitation of c-Src and Syk in the hβ3–hβ1(C3R)-bearing cells () establish that the three terminal β3 amino acids are required for optimal association of the two kinases and their ability to organize the osteoclast cytoskeleton. To further explore this issue, we asked whether the three terminal β3 residues regulate the integrin's interaction with Syk and/or c-Src in the osteoclast. Thus, we transduced β3 BMMs with hβ3WT or hβ3–hβ1(C3R). After 5 d in MCSF and RANKL, the integrin constructs were immunoprecipitated, and the immunoprecipitates were blotted with anti-Syk or β3 antibodies. Unlike transformed cells expressing αIIbβ3 (), Syk binds hβ3–hβ1(C3R) in the osteoclast and, for reasons unknown, does so more effectively than to hβ3WT (). However, this conundrum does not reflect the enhanced affinity of Syk for the β1 integrin in the absence of β3 (, available at ). Syk phosphorylation in the context of αIIbβ3 does not depend on the presence of the two Y residues in the β3 cytoplasmic domain (). On the other hand, αIIbβ3–Syk binding is arrested by the alanine mutation of β3 (). We find that unlike the platelet integrin, in which β3 promotes a bleeding dyscrasia (), the same double mutant does not impact the osteoclast (). In keeping with this observation, Syk association with β3 is unaltered by β3 (). In contrast to Syk, the binding of c-Src to hβ3 is arrested by the hβ3–hβ1(C3R) mutation (). Consistent with this observation, no c-Src–β3 association is evident upon deletion of the entire integrin cytoplasmic domain (hβ3ΔC) or its termination at residue 751 (hβ3Δ752–762). Syk must be phosphorylated to induce its downstream targets and, when bound to the ITAM domain of immune response receptors, is activated via autophosphorylation (). To determine whether such is the case regarding its association with αvβ3 in the osteoclast, we suspended WT preosteoclasts in the presence of the Syk kinase inhibitor piceatannol (Pice; Calbiochem) or a carrier (DMSO). After 20 min, the cells were either maintained in suspension or replated on the αvβ3 ligand vitronectin. 30 min later, Syk immunoprecipitates were analyzed for phosphotyrosine content by immunoblotting. shows that Pice does not detectably impact adhesion-induced Syk phosphorylation. Confirming that it is an effective Syk kinase inhibitor in osteoclasts, Pice arrests integrin-mediated phosphorylation of the Syk downstream target Vav3 (; ). We next transduced WT or kinase-inactive Syk (Syk) into Syk BMMs. The cells were plated on vitronectin, and tyrosine-phosphorylated Syk was measured. Consistent with our Pice-based studies, adherent Syk preosteoclasts bearing kinase-inactive Syk undergo normal adhesion-associated tyrosine phosphorylation (). These data indicate that although Syk kinase activity is necessary for organization of the osteoclast cytoskeleton (), αvβ3-associated Syk is itself phosphorylated by another kinase. Given that αvβ3 induces Syk–c-Src interaction, the latter is a likely Syk kinase. In fact, SFK inhibitors diminish c-Src–Syk association () and adhesion-induced Syk phosphorylation (). More specifically, Syk activation in response to αvβ3 occupancy is markedly reduced in Src preosteoclasts (). Having established that c-Src activates Syk in the osteoclast, we asked the reciprocal question: does αvβ3 occupancy activate c-Src, and, if so, is the event mediated by Syk? We have shown that the absence of Syk attenuates c-Src activity in the context of an entire population of preosteoclasts (). However, these previous experiments included both αvβ3 ligand–adherent and –nonadherent cells, the latter being abundant in the absence of Syk. As our present aim involved the impact of Syk on αvβ3-stimulated c-Src activity, we maintained Syk or Syk preosteoclasts on vitronectin for 30 min, after which nonadherent cells were removed. TCLs of the remaining adherent preosteoclasts, representing the population with occupied αvβ3, were then immunoblotted for activated c-Src. Adhesion to the αvβ3 ligand phosphorylates the activating residue c-Src equally in the presence or absence of Syk (). Thus, although αvβ3-associated c-Src activates Syk, the reciprocal is not true. Syk and c-Src recognize and signal through ITAM-bearing immunoreceptors as well as the αvβ3 integrin (). To determine whether interdependency exists between these two receptor-mediated signaling pathways in osteoclasts, we turned to osteopetrotic mice deleted of two essential osteoclastogenic ITAM immunoreceptors, namely Dap12 and FcRγ (). In keeping with previous studies (; ), we find that terminal osteoclastogenesis, which was induced in our system by 5-d exposure of spleen cells to RANKL and MCSF, is attenuated in Dap12/FcRγ mice (unpublished data). Surprisingly, however, the appearance of TRAP-stained mutant cells after only 3 d in osteoclastogenic medium is similar to WT (). Furthermore, the expression of β3-integrin mRNA () and c-Src () is also indistinguishable in the two genotypes at this earlier time. Next, Dap12/FcRγ day 3 preosteoclasts were lifted and maintained in suspension or replated on αvβ3 ligand for 30 min. Subsequent immunoprecipitation and phosphotyrosine immunoblotting shows that the adhesion-mediated activation of Syk is obviated in the absence of the two ITAM proteins (). Furthermore, integrin occupancy of Dap12/FcRγ preosteoclasts fails to phosphorylate the Syk kinase target Vav3 but continues to activate c-Src, which is upstream of Syk in the αvβ3 signaling cascade (). These data indicate that integrin activation of Syk requires ITAM immunoreceptors. If so, one would expect the disruption of Syk–ITAM association to blunt the tyrosine kinase's capacity to organize the osteoclast cytoskeleton. To address this issue, we turned to Syk, which continues to bind the β3 integrin but not ITAM proteins (; ). We expressed WT Syk or this mutant of the C-terminal SH2 domain in Syk preosteoclasts generated by 3-d exposure to RANKL and MCSF. The cells were suspended and replated on vitronectin for 30 min. In this circumstance, the WT protein but not Syk is phosphorylated (), and Syk and c-Src no longer meaningfully associate (). These collective observations indicate that ITAM proteins mediate αvβ3-stimulated Syk activation. Importantly, WT Syk rescues the Syk osteoclast cytoskeleton, but Syk fails to do so (). Therefore, αvβ3- and ITAM-bearing immunoreceptors function in tandem regarding the impact of c-Src–Syk on the osteoclast. Osteoclasts are characterized by a unique cytoskeleton that mediates the resorptive process. Upon contact with bone, the cell generates two polarized structures enabling it to degrade skeletal tissue. These include a villous organelle unique to the resorbing osteoclast, which is known as the ruffled membrane, and the actin ring or sealing zone, which isolates the resorptive microenvironment from the general extracellular space (). The β3-integrin knockout mouse serves as an important tool for determining the role of the osteoclast's most abundant integrin in its capacity to resorb bone (). Failure to express αvβ3 results in dramatic alterations of the cell's actin cytoskeleton, including failure to form actin rings or normal ruffled membranes in vivo. These abnormalities result in attenuated bone resorption. The mechanisms by which c-Src exerts its effects in osteoclasts are complex. In the first instance, it functions as an adaptor protein independent of its capacity to induce tyrosine phosphorylation (). Alternatively, complete reversal of the functional abnormalities of c-Src–deficient osteoclasts also requires its kinase activity (; ). The similar cytoskeletal phenotypes of c-Src and αvβ3-deficient osteoclasts suggest a commonality of intracellular signaling. c-Src binds directly to the terminal three amino acids of the β3 subunit in the context of the platelet integrin αIIbβ3 (), and we have established that the same residues regulate αvβ3–c-Src association in the osteoclast. However, the means by which c-Src is recruited to αvβ3 in the osteoclast is not fully understood. A current model holds that Pyk2 and c-Cbl mobilize c-Src to the integrin (). In this paradigm, αvβ3 occupancy induces the phosphorylation of Pyk2, which then activates c-Src by occupying its SH2 domain. Although provocative, these data await documentation that Cbl or Pyk2 deletion prompts a meaningful bone phenotype. Furthermore, although c-Src and c-Cbl phosphorylation depends on αvβ3 occupancy, this is not the case regarding Pyk2 (). On the other hand, there is substantial evidence that Syk partners with c-Src, and, therefore, we asked whether these two nonreceptor tyrosine kinases impact the osteoclast in the context of the αvβ3 integrin. Differentiation of Syk-deficient BMMs into osteoclasts is not delayed as determined by the progressive expression of markers of the mature resorptive cell in the presence of RANKL and MCSF. Moreover, activation of many intracellular signals induced by the two osteoclastogenic cytokines is not retarded in the mutant cells. Like those lacking c-Src or αvβ3, however, Syk osteoclasts have a deranged cytoskeleton, as they fail to spread or form actin rings in vitro and are largely unattached to the bone surface in vivo. Failure of a kinase-inactive Syk mutant to rescue Syk osteoclasts establishes that activated Syk is required to organize the cell's cytoskeleton. Confirming the functional significance of these abnormalities, Syk embryos have increased bone mass, and basal and PTH-stimulated bone resorption are attenuated in radiation chimeric mice whose osteoclasts lack the kinase. The morphological similarity of osteoclasts lacking c-Src, αvβ3, or Syk raises the possibility that these molecules work in tandem in the osteoclast. In fact, we find that each associates with the other two in the mature resorptive cell and that c-Src–Syk interaction requires the presence of the liganded integrin. The β3-integrin cytoplasmic domain consists of 47 amino acids, and we previously noted that of six candidate substitutions, only one, the human mutation β3, fails to rescue the dysfunctional β3-deficient osteoclast (). In the present study, we extend these data to include the terminal three residues of the subunit as essential for osteoclast cytoskeletal organization. The fact that the absence of this motif arrests both β3–c-Src and c-Src–Syk association provides compelling evidence that the two tyrosine kinases partner with the integrin to regulate the osteoclast cytoskeleton. On the other hand, neither substitution of the three C-terminal β3 residues nor the absence of c-Src dampens Syk's capacity to bind the integrin. Therefore, in osteoclasts, Syk binds to the β3 subunit independently of c-Src and at a site that is probably distinct from the SFK. In contrast to the platelet (), mutation of the two β3 cytoplasmic domain tyrosine residues to phenylalanine does not impact the osteoclast. This observation may be a result of the fact that although the β3 chain recognizes Syk via the latter's N-terminal SH2 domain, it does so in a phosphotyrosine-independent manner (). This is in keeping with our finding that mutation of the two tyrosine residues in the β3 cytoplasmic domain does not alter its recognition of Syk in osteoclasts. This circumstance permits Syk in association with αvβ3 to continue to activate downstream cytoskeletal-organizing molecules such as Vav3 and Rac and explains the capacity of β3 to completely rescue β3 osteoclasts (). The earliest insights into the mechanisms by which Syk impacts the platelet and the osteoclast relate to its capacity to interact with ITAM-containing proteins (). In the osteoclast, these ITAM proteins include Dap12 and FcRγ (; ). Although the deletion of each enhances bone mass, the elimination of both promotes severe osteopetrosis, an event mediated, at least in part, by Syk dysfunction. In contrast to the means by which its N-terminal SH2 domain associates with β3 integrin, Syk's C-terminal SH2 motif binds ITAM proteins via the classic mechanism of phosphotyrosine recognition (). Given the individual flexibility of each Syk SH2 domain (), these observations suggest that Syk associates with αvβ3 and ITAM proteins in the same complex. Deletion of both Dap12 and FcRγ arrests terminal osteoclastogenesis as a result of the failed expression of NFATc1 (). However, the absence of Syk does not impact the osteoclastogenic transcription factor. This observation is consistent with the fact that osteoclasts lacking the kinase, like those deleted of the integrin, mature normally but fail to organize their cytoskeleton. Thus, although Syk is regulated by both αvβ3 and ITAM proteins, the impact of each on the osteoclast differs. On the other hand, we find that ITAM proteins are not required for early commitment to the osteoclast phenotype, enabling us to explore the role of Dap12 and FcRγ in αvβ3 activation of Syk and its downstream targets. Syk activation in osteoclasts is independent of its own kinase and occurs under the aegis of c-Src. Furthermore, Syk's ability to undergo phosphorylation and associate with c-Src in the bone-resorptive cell depends on an intact ITAM-interactive C-terminal SH2 domain (; ). These data suggest a model in which Syk is first recruited to the activated integrin at a site other than the three terminal amino acids (). Syk, in turn, functions as an adaptor for the integrin and ITAM proteins wherein the kinase interacts with the latter in an SH2 domain–dependent manner. c-Src binds independently of integrin occupancy to the distal β3 cytoplasmic domain, where it associates with and phosphorylates Syk upon αvβ3 occupancy. Importantly, these observations establish for the first time that ITAM proteins not only regulate osteoclast function but the capacity of the mature resorptive cell to resorb bone as well. Thus, our data show that Syk is a nexus of a novel signaling pathway regulating osteoclast function and, like c-Src () and αvβ3 (), is a candidate anti–bone-resorptive therapeutic target. Animals were housed in the animal care unit of the Washington University School of Medicine and were maintained according to guidelines of the Association for Assessment and Accreditation of Laboratory Animal Care. All animal experimentation was approved by the Animal Studies Committee of the Washington University School of Medicine. Syk mice () were maintained on a 129.sv background (which does not carry the Ly-5.1 allele). Mice were genotyped by PCR using 5′-AGAGAAGCCCTGCCCATGGAC-3′ (Syk) and 5′-CCTTGGGAAAAGCGCCTCCCCTACCC-3′ (Syk) as forward primers in combination with the 5′-GTCCAGGTAGACCTCTTTGGGC-3′ reverse primer. The products (86 and 120 bp for Syk and Syk, respectively) were resolved on a 2.5% agarose gel. Syk and littermate Syk fetal liver cells were obtained from embryonic day 15–17 embryos from timed matings of Syk carriers. Bone marrow chimeras were generated by intravenous injection of unfractionated fetal liver cells into 6–8-wk-old lethally irradiated congenic recipient B6.SJL mice (carrying the Ly-5.1 allele). To evaluate marrow engraftment, chimeric marrow cells were incubated for 30 min with FITC-conjugated anti-Ly-5.1 mAb IgG. The cells were washed and subjected to FACS analysis. Chimeras were used 4–8 wk after the bone marrow transplantation. Primary BMMs were prepared as described previously () with slight modification. Marrow was extracted from femora and tibia of 6–8-wk-old mice with α-MEM and cultured in α-MEM containing 10% inactivated FBS, 100 IU/ml penicillin, and 100 μg/ml streptomycin (a-10 medium) with 100 ng/ml MCSF () in petri dishes. In c-Src and Dap12/FcRγ mice, macrophages were cultured from their spleen cells. Cells were incubated at 37°C in 6% CO for 3 d. Cells were washed with PBS and lifted with 1× trypsin/EDTA (Invitrogen) in PBS. A total of 5 × 10 cells were cultured in 200 μl α-MEM containing 10% heat-inactivated FBS with 100 ng/ml GST-RANKL and 26 ng/ml of mouse recombinant MCSF in 96-well tissue culture plates, some of which contained a sterile whale dentin slice. Cells were fixed and stained for TRAP activity after 5 d in culture using a commercial kit (387-A; Sigma-Aldrich). For actin ring staining, cells were fixed in 4% PFA, permeabilized in 0.1% Triton X-100, rinsed in PBS, and immunostained with AlexaFluor488 phalloidin (Invitrogen). To quantitate resorption lacunae, cells were removed from the dentin slices with 2 N NaOH and mechanical agitation. Dentin slides were stained with Coomassie brilliant blue. For preosteoclast generation, 1.2 × 10 BMMs were plated per 10-cm tissue culture dish. After a 3-d culture in 26 ng/ml MCSF and 100 ng/ml GST-RANKL, TRAP-expressing preosteoclasts were lifted with 0.02% EDTA in PBS. WT human Syk, hβ3 integrin subunit, and their mutants in pMX retrovirus vector were transiently transfected into Plat-E packaging cells using FuGENE 6 Transfection Reagent (Roche). The virus was collected 48 h after transfection. BMMs were infected with the virus for 24 h in the presence of 26 ng/ml MCSF and 4 μg/ml polybrene (Sigma-Aldrich). Cells were selected in the presence of MCSF and 1 μg/ml blasticidin (Calbiochem) for 3 d before analysis of osteoclastogenesis. Cultured cells were washed twice with ice-cold PBS and lysed in radioimmunoprecipitation buffer containing 20 mM Tris, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1% Triton X-100, 2.5 mM sodium pyrophosphate, 1 mM β-glycerophosphate, 1 mM NaVO, 1 mM NaF, and 1× protease inhibitor mixture (Roche). After incubation on ice for 10 min, the cell lysates were clarified by centrifugation at 15,000 rpm for 10 min. 40 μg of total lysates were subjected to 8 or 12% SDS-PAGE and transferred onto nitrocellulose membrane. The filters were blocked in 5% milk/TBS and 0.1% Tween 20 for 1 h and incubated with primary antibodies at 4°C overnight followed by probing with secondary antibodies coupled with HRP (Santa Cruz Biotechnology, Inc.). The proteins were visualized using SuperSignal West Dura Extended Duration Substrate (Pierce Chemical Co.). Goat anti–β3-integrin pAb was purchased from Santa Cruz Biotechnology, Inc. Mouse anti-Syk mAb was purchased from Abcam, and antiphosphotyrosine mAb 4G10 was purchased from Upstate Biotechnology. Antiphospho-Vav3 (pY173) antibody was purchased from Biosource International. mAb 327 directed against the c-Src protein and full-length c-Src cDNA were gifts from A. Shaw (Department of Pathology, Washington University School of Medicine, St. Louis, MO). Anti–c-Src mAb was purchased from Cell Signaling. Anti–β1-integrin mAb was a gift from R. Hynes (Massachusetts Institute of Technology, Cambridge, MA). For immunoprecipitation, BMMs were cultured with MCSF and RANKL in culture dishes for 3 d. Preosteoclasts were lifted and plated onto 5-μg/ml vitronectin-precoated dishes or cultured in suspension. Cells were washed in cold PBS and lysed on ice in radioimmunoprecipitation lysis buffer. 800 μg of protein was incubated with 2 μg monoclonal anti–human β3-integrin antibody (7G2; gift from E. Brown, University of California, San Francisco, San Francisco, CA) or polyclonal anti-Syk antibody (N19; Santa Cruz Biotechnology, Inc.) at 4°C overnight with rotation. Protein A/G agarose (Santa Cruz Biotechnology, Inc.) was then added and incubated with rotation for 3 h at 4°C. Immunoprecipitates were washed three times in lysis buffer, and the beads were boiled in 2× SDS sample buffer for 5 min. After centrifugation, proteins were separated by 8 or 10% SDS polyacrylamide gels. Primary calvaria-derived osteoblasts were isolated in collagen-gel culture from 17.5-d Syk or Syk embryos (). 5 × 10 primary osteoblasts were maintained in 48-well plates in α-MEM for 3 d, after which osteoblastogenic medium consisting of α-MEM containing 50 μg/ml ascorbic acid and 2 mM β-glycerophosphate was added. 20 d later, bone nodules were stained with Alizarin red. 1.2 × 10 Syk or Syk BMMs were cultured with 100 ng/ml RANKL and 25 ng/ml MCSF for 3 d. Preosteoclasts were lifted with 0.02% EDTA in PBS and washed twice with α0 media. 1.5 × 10 preosteoclasts were plated on vitronectin (BD Biosciences)-precoated six-well nontissue culture-treated plates for 5 or 10 min. WT or Syk preosteoclasts were lifted and plated onto 5-μg/ml vitronectin-precoated dishes or were cultured in suspension. Cells were washed in cold PBS and lysed with lysis buffer. Levels of Rac1-bound GTP were determined according to the protocol of the Rac1 activation kit (Pierce Chemical Co.). 10 μg synthetic human PTH(1–34) (Bachem California Inc.) in 25 μl of vehicle (1 mM HCl and 0.1% BSA) or vehicle alone were injected four times daily for 4 d into the subcutaneous tissue overlying the calvariae using a Hamilton syringe. After killing by CO narcosis, calvariae were removed intact, soft tissues were gently dissected, and the calvariae were fixed in 10% phosphate-buffered formalin for 24 h and were further processed as described previously (). Osteoclast number was determined using the Bioquant System (BIOQUANT Image Analysis). The mouse TRAP assay kit was purchased from Immunodiagnostic Systems Ltd. Serum TRAcP 5b was determined according to the manufacturer's instructions of the mouse TRAP assay kit (Immunodiagnostic Systems Ltd). All in vivo experiments were approved by the Washington University Animal Studies Committee. For whole-mount skeletal staining, 18-d embryos were eviscerated, fixed in 95% ethanol, and stained with Alcian blue and Alizarin red according to standard protocols (). For histological analyses, embryonic bones were fixed in 10% formalin, and decalcified sections were stained for TRAP activity. All data are expressed as means ± SD, and statistical significance was calculated by test. Images from (A–E), 2 B, 3 B, 4 B, 6 B, and 8 (A and G) were acquired using a microscope (Eclipse E400; Nikon) with plan Fluor lenses at RT. All magnifications are 20× except for , which is 60×. Zimmerson oil (Fisher Scientific) served as the imaging medium in . Fluorochrome FITC was used in . Images from were acquired using a microscope (MZFLIII; Leica) with a plan Apo 1× lens (model 10447157; Leica) at RT. No imaging media or fluorochromes were used. Photographs were taken with a camera (MagnaFire S99802; Optronics) and displayed with MagnaFire software (version 2.1B; Meyer Instruments) in all circumstances except . The images were organized in Photoshop (version 7.0.1; Adobe). was acquired with a digital camera (Powershot A75; Canon) and organized in Photoshop. Fig. S1 shows the generation of Syk chimeras. Fig. S2 shows that Syk deficiency does not delay osteoclast differentiation or arrest intracellular signaling. Fig. S3 shows that bone density is increased in Syk embryos. shows that Syk preosteoclasts do not rapidly spread on αvβ3 ligand. Fig. S5 shows that the deletion of β3 integrin does not enhance Syk–β1 integrin association in osteoclasts. Online supplemental material is available at .

xref sub sc italic #text To examine the effect of LacCer stereoisomers on mechanisms of endocytosis in human skin fibroblasts (HSFs), various fluorescent endocytic markers were used. These probes were previously characterized as relatively specific markers of uptake via caveolae- (anti–β1-integrin [β1-Fab] and fluorescent albumin), clathrin- (transferrin [Tfn]), and RhoA (interleukin-2 receptor β subunit [IL-2R])-regulated mechanisms using dominant-negative proteins and biochemical inhibitors (; ; , ; ). We used 1 mg/ml of fluorescent dextran as a probe for Cdc42-regulated endocytosis, as this concentration of dextran has been shown to selectively label endosomes derived from this pathway (; ). An Fab fragment of an antibody (Ab) against GFP was used to follow the endocytosis of glycosyl-phosphatidylinositol (GPI)–GFP via this pathway. Endocytosis of GPI-GFP was perturbed by the expression of dominant-negative Cdc42 (Fig. S1 A, available at ). In addition, endocytosed GPI-GFP closely colocalized with endocytosed dextran (Fig. S1 B). These results indicate that dextran and GPI-GFP are internalized via the Cdc42-regulated pathway in HSFs as they are in CHO cells (). Cells were coincubated with the endocytic probes and 5 μM L--LacCer or the corresponding natural (D-) stereoisomer for 30 min at 10°C (see Materials and methods). Biochemical analysis indicated that approximately equal amounts of either LacCer stereoisomer became cell associated under these incubation conditions (∼3 pmol/10 cells). The samples were then briefly warmed to 37°C to allow endocytosis to occur, and the amount of internalization was quantified by image analysis. Pretreatment of cells with --LacCer inhibited the uptake of caveolar markers (labeled albumin or anti–β1-integrin Fab fragment) by ∼70% relative to untreated control cells, whereas there was a slight stimulation (∼10–20%) of Cdc42-regulated (dextran and GPI-GFP) and RhoA-regulated (IL-2R) endocytosis; little effect was seen on the clathrin-dependent internalization of Tfn (). In contrast, when β--lactosyl--octanoyl---sphingosine (D--LacCer) was used, internalization via caveolae was increased 60–100% relative to untreated controls as previously reported (). Quantitative analysis of surface caveolae by electron microscopy () after treatment of cells with the LacCer stereoisomers showed that the - isomer increased the number of caveolae present at the PM, which is consistent with the inhibition of internalization by this route, whereas the D- isomer reduced the number of surface-connected caveolae, presumably reflecting an increase in internalization of caveolae-derived vesicles (; ). The results in show that pretreatment of cells with the nonnatural LacCer stereoisomer blocks caveolar endocytosis while not markedly affecting other mechanisms of internalization. The effect of L--LacCer treatment on SV40 infection in monkey CV1 cells was examined next because endocytosis of this virus has been extensively characterized in this cell type and is shown to occur via caveolae (; ; ). Pretreatment of cells with L--LacCer dramatically reduced viral infection as monitored by the expression of the large T antigen after 14 h at 37°C, whereas D--LacCer had no effect relative to untreated control samples (). Because the inhibition of SV40 infection by L--LacCer could result from effects on virus binding as well as endocytosis, we also studied the effect of L--LacCer on SV40 binding. We found that the L--LacCer treatment inhibited SV40 binding by ∼80% (), as detected using a mAb to the SV40 major capsid protein VP1 (). Thus, the effect of the nonnatural LacCer isomer on SV40 infection was probably mainly caused by its inhibition of SV40 binding to the cell surface. We next addressed potential mechanisms by which L--LacCer might selectively inhibit caveolar endocytosis as well as the binding of SV40 to the cell surface. One possibility is that L--LacCer might disrupt PM microdomains, which are local regions of the PM enriched in GSLs and cholesterol that may act as organizing centers for particular proteins (for reviews see ; ). To test this possibility, cells were incubated with AlexaFluor594-CtxB at 10°C to label GM ganglioside at the PM followed by an anti-CtxB IgG. This treatment caused the formation of numerous micrometer-size clusters of CtxB at the PM, which were not present in the absence of anti-CtxB IgG (, left vs. middle). Importantly, when cells were pretreated with L--LacCer for 30 min at 10°C before incubation with the labeled CtxB and cross-linking anti-CtxB IgG, clustering into PM domains was prevented (, right). No inhibition of domain formation was observed using D--LacCer; rather, the D- isomer induced the formation of large domains enriched in GM ganglioside and cholesterol in the absence of the cross-linking IgG (Fig. S2, available at ; ). We then incubated CV1 cells at low temperature with SV40 virus and found that this treatment induced the formation of PM domains enriched in GM ganglioside and cholesterol (; left vs. middle), which is remarkably similar to the clustered microdomains induced by CtxB cross-linking (). Interestingly, the pretreatment of CV1 cells with L--LacCer prevented the SV40 induction of these PM domains (; right). These results provide a potential explanation for the reduction in SV40 binding to CV1 cells because GM ganglioside is a receptor for SV40 (), and clustering of GM may be required for maximal SV40 binding. Together, these experiments demonstrate that the induction of PM domains by various treatments (cross-linking Abs or SV40) is prevented by L--LacCer. This modulation of PM domain organization may disrupt the distribution of potential cargo in microdomains and, thus, inhibit their endocytosis via caveolae. The concept that microdomain clustering is important for caveolar endocytosis is supported by previous observations. First, unlike BODIPY–D--LacCer, which associates with microdomains and is internalized via caveolae, the L- isomer of BODIPY-LacCer does not partition into microdomains at the PM and is not selectively endocytosed via caveolae (). Second, agents that stimulate the clustering of GSL-enriched microdomains at the PM (e.g., exogenous GSLs and β1-integrin cross-linking Abs) also stimulate caveolar endocytosis (, ). A second mechanism by which L--LacCer might inhibit caveolar internalization is by disrupting transmembrane signaling events required for endocytosis. We focused on signaling through β1 integrin because this integrin is internalized via caveolae in HSFs and other cell types (; ) and because an early event after integrin activation is signaling through src, a kinase whose activity is required for caveolar endocytosis (; ; ). We first examined the activation of β1 integrin in HSFs after cross-linking with a stimulatory β1-integrin Ab (β1-stim Ab) using the HUTS-4 Ab, which only binds to β1 integrins in their activated conformation (). Treatment with the stimulatory Ab dramatically increased HUTS binding (), whereas pretreatment with L--LacCer before incubation with β1-stim Ab reduced HUTS-4 binding to levels seen in untreated control cells. In contrast, when D--LacCer was used, the Ab-induced activation of β1 integrin was not inhibited. When L--LacCer was incubated with HSFs in the absence of β1-stim Ab, no increase in HUTS binding was seen (Fig. S3, A and B; available at ). This is in contrast to D--LacCer, which activated β1 integrin in the absence of the β1-stim Ab to a similar extent as when the Ab was used alone (Fig. S3, A and B; ). Because β1 integrins have been shown to activate src kinases and src phosphorylation is required for caveolar endocytosis (; ), we also examined the effect of L--LacCer treatments on this process. Cells were incubated for 30 min at 10°C ± the LacCer stereoisomers and/or the β1-stim Ab followed by a 30-s incubation at 37°C. Cell lysates were then immunoblotted for src and phospho-(Y416) src. Src kinase was activated (phosphorylated at Y416) by treatment with D--LacCer to a level similar to that seen with the stimulating Ab, and this activation could be blocked by the src inhibitor PP2 (). In contrast, little or no src activation was observed upon treatment with L--LacCer. Furthermore, pretreatment of cells with L--LacCer inhibited src phosphorylation induced by the β1-stim Ab, whereas pretreatment with D--LacCer did not substantially affect src phosphorylation in response to β1-stim Ab (). Several features of our studies suggest that the LacCer stereoisomers may regulate caveolar endocytosis via modulation of β1-integrin signaling. First, both D--LacCer and β1 integrin stimulate src activation and caveolar endocytosis to a similar degree (; , ). Second, treatment with L--LacCer inhibits β1 integrin–mediated src signaling as well as caveolar endocytosis ( and ). To further examine the role of β1 integrin in the regulation of caveolar endocytosis in HSFs, we used an siRNA approach to deplete this integrin and studied the uptake of various markers. We first validated the use of three different β1-integrin siRNAs and found that these reduced β1-integrin levels in HeLa cells ∼80% relative to untreated cells (Fig. S3 C). We then used electroporation to transfect HSFs with a β1-integrin siRNA and found that the levels of β1 integrin were reduced by ∼75% in the transfected cells as assessed by immunofluorescence (Fig. S3 D). We then examined the effect of β1-integrin knockdown on the uptake of multiple endocytic markers. Endocytosis of the caveolar markers albumin and BODIPY-LacCer was dramatically reduced in cells transfected with β1-integrin siRNA relative to that in nontransfected cells. In contrast, no effect of β1-integrin siRNA treatment was seen on clathrin (Tfn), Cdc42 (dextran and GPI-GFP), or RhoA (IL-2R) internalization (). Control experiments showed that clathrin heavy chain siRNA treatment resulted in a strong inhibition of Tfn uptake but had little effect on albumin or BODIPY-LacCer internalization (). Finally, we examined the effect of β1-integrin knockdown on src phosphorylation in HSFs treated with the LacCer stereoisomers using immunofluorescence. Although the phosphorylation of src was increased in D--LacCer– treated cells (), this stimulation was reduced by 65–75% in individual cells in which β1 integrin had been depleted (). As expected, no further reduction of phospho-src was observed in cells treated with L--LacCer and depleted of β1 integrin (not depicted) because src phosphorylation was inhibited by the L- stereoisomer (). The inhibition of caveolar endocytosis by β1-integrin knockdown in HSFs suggests a key role for β1 integrin in the regulation of this endocytic process by GSLs. For example, natural GSLs such as D--LacCer that stimulate caveolar endocytosis may do so by increasing β1-integrin clustering, whereas L--LacCer may inhibit endocytosis via caveolae by preventing this clustering. Further studies are needed to determine the applicability of our findings to other cell types, the role of other integrin components (e.g., various αβ heterodimers) in modulating caveolar uptake, and whether the regulation of caveolar endocytosis is restricted to integrins that are internalized via caveolae or also includes integrins that are endocytosed via the clathrin pathway (e.g., αβ5; ). Our data support a general model in which exogenously supplied D--LacCer or other sphingolipids with the natural D- stereochemistry promote the coalescence of PM microdomains, leading to the clustering and activation of transmembrane proteins that can initiate a signaling cascade required for caveolar endocytosis. In HSFs, β1 integrin appears to be a key molecule involved in transducing this signal across the PM bilayer; however, other signaling proteins that partition into lipid microdomains may be similarly affected by exogenous lipids in HSFs or other cell types. Our results also suggest a potential mechanism whereby certain tumor cells that shed gangliosides can alter the properties of nearby cells (; ). Most importantly, the results of the current study document a dominant-negative lipid, L--LacCer, which selectively inhibits caveolar endocytosis of multiple markers by interfering with microdomain clustering and β1-integrin signaling. Of additional interest is our finding that the nonnatural stereoisomer of LacCer dramatically inhibited SV40 infection, most likely by disrupting PM domains that are required for optimal binding of the virus to the cell surface. Finally, we suggest that the disruption of membrane microdomains by L--LacCer may represent a new approach for the treatment of certain diseases and infectious agents that use lipid rafts or raft proteins as targets (). Normal HSFs (Coriell Institute for Medical Research), HeLa, and CV1 cells (American Type Culture Collection) were used as described previously (; ). The SV40 construct (pUCSVH388-2) was obtained from the American Type Culture Collection, transfected into CV1 cells, and the virus was harvested after maximum cytopathic effect. The virus was titered and used at an MOI of 15. D--LacCer was purchased from Avanti Polar Lipids, Inc., and L--LacCer was synthesized by N. Gretskaya (Shemyakin Institute, Moscow, Russia). LacCer isomers were complexed to defatted BSA and incubated with cells at a final concentration of 5 μM (). Fluorescent AlexaFluor594- and -647–labeled CtxB, Tfn, dextran (10 kD), Ab labeling kits, and fluorescent secondary Abs were obtained from Invitrogen. An anti–β1 integrin (IgG1) from BD Biosciences was used as a stimulatory Ab (β1-stim Ab) and for the preparation of an Fab fragment using a kit from Pierce Chemical Co. Abs were fluorescently labeled with an AlexaFluor dye as described previously (). An AlexaFluor594-Fab fragment against GFP was prepared similarly from an anti-GFP Ab (Invitrogen). The HUTS-4 β1-integrin Ab was purchased from Chemicon. A phycoerythrin-labeled IL-2R Ab was obtained from BD Biosciences. and phospho- Abs for Western blotting were obtained from Cell Signaling Technology. Antiphospho-src (Y416) mAb for immunofluorescence studies was purchased from Upstate Biotechnology. CtxB, large T antigen Abs, and tyrosine kinase inhibitor (PP2) were purchased from Calbiochem. The biotinylated derivative (BC-Θ) of perfringolysin O (Θ toxin) was obtained from Y. Ohno-Iwashita (Tokyo Metropolitan Institute, Tokyo, Japan) and used as described previously () using AlexaFluor594-streptavidin (Invitrogen) to visualize its distribution on the cell surface. A mAb (pab597) against SV40 VP1 was provided by L. Norkin (University of Massachusetts, Amherst, MA) and used as described previously (). All other reagents were purchased from Sigma-Aldrich. For endocytosis of BODIPY-LacCer or AlexaFluor-labeled anti–β1-integrin Fab or Tfn, HSFs were preincubated with the fluorescent marker for 30 min at 10°C, washed, and further incubated for 3 or 5 min at 37°C. For AlexaFluor-labeled albumin and dextran, cells were incubated for 3 or 5 min at 37°C without preincubation. Endocytosis of IL-2R was performed as described previously (). For GPI-anchored protein uptake, HSFs were first transfected with GPI-GFP for 48 h and then were incubated with AlexaFluor594-labeled anti–GFP-Fab for 30 min at 10°C, washed, and incubated for 5 min at 37°C. All samples were then either acid stripped or back exchanged to remove cell surface fluorescence before fluorescence microscopy (). CV1 cells were incubated ± LacCer stereoisomers for 30 min at 10°C and coincubated for 1 h at 10°C with SV40 virus (MOI = 15). Samples were then washed and stained for SV40 binding at the PM using anti-SV40 VP1 mAb for 30 min at 10°C . Cells were washed, incubated with AlexaFluor594-labeled anti–mouse secondary Ab, and observed by fluorescence microscopy at 10°C. For the visualization of GM ganglioside microdomains at the PM, HSFs and CV1 cells were incubated for 30 min at 10°C ± L--LacCer and further incubated (for 30 min at 10°C) with fluorescent CtxB. Samples were washed and incubated with anti-CtxB IgG (30 min for HSFs) or SV40 virus (1 h for CV1 cells) at 10°C. Cells were then washed and observed under the fluorescence microscope (IX70; Olympus) at 10°C. For the visualization of cholesterol microdomains, CV1 cells were incubated (for 30 min at 10°C) ± L--LacCer followed by SV40 virus for 1 h at 10°C. Cells were then washed and incubated with biotinylated BC-Θ followed by AlexaFluor594-streptavidin. HSFs were serum starved for 2–3 h and treated with D--LacCer, L--LacCer, or 50 μg/ml PDGF for 30 min at 10°C in HBSS. In one experiment, cells were pretreated with 15 μM PP2 for 30 min at 37°C before incubation with D--LacCer. In another experiment, cells were either untreated or treated with the LacCer isomers for 30 min at 10°C followed by incubation with β1-stim Ab for 30 min at 10°C. All samples were then washed and warmed for 30 s at 37°C before cell lysis. Lysates were then immunoblotted for and phospho- (Y416). Fluorescence microscopy was performed using a fluorescence microscope (IX70; Olympus) equipped with 60× 1.4 NA and 100× 1.35 NA objectives and a CCD camera (C4742; Hamamatsu). For quantitative studies, all photomicrographs in a given experiment were exposed and processed identically for a given fluorophore and were analyzed using the MetaMorph image processing program (version 6.2; Universal Imaging Corp.). Quantitative results are expressed as means ± SDs. Images were prepared for individual figures using Photoshop CS (Adobe). Fig. S1 shows the characterization of endocytosis of GPI-GFP via the Cdc-42–regulated pathway in HSFs. Fig. S2 illustrates microdomains containing cholesterol, GM ganglioside, and β1 integrin in HSFs treated with D--LacCer versus L--LacCer. Fig. S3 shows the effects of D-- and L--LacCer on β1-integrin activation and the characterization of β1-integrin knockdown in HeLa cells and HSFs. Online supplemental material is available at .

Apoptosis signal-regulating kinase (ASK) 1 is a MAP3K family member that activates both the JNK and p38 MAPK signaling cascades and is activated in response to various stimuli, including oxidative stress, endoplasmic reticulum stress, calcium influx, and inflammatory cytokines (; ; ). Expression of ASK1 protein has been reported to be strongly induced surrounding wounds in rat palatal epithelium (). It has also been demonstrated that ASK1 induces keratinocyte differentiation and regulates the innate immunity of the skin (, ). These findings have suggested that ASK1 may play an important role in epithelial wound healing. Mammalian skin is composed of three differentiated epithelial compartments: the interfollicular epidermis, sebaceous glands, and hair follicles (). A bulge within each hair follicle contains stem cells, which in turn proliferate and differentiate into new hair follicles (; ). Wounding of skin has been reported to induce hair growth (). It was recently shown that the pattern of expression of epithelial stem cells in hair follicles around wound areas is similar to that in spontaneous hair cycling (). This suggested that understanding of wound-induced hair regrowth may elucidate the general mechanisms of hair growth. Furthermore, it is known that onset of the developmental program in epithelial stem cells is triggered by environmental signals (). However, the hair regrowth factors and mechanisms by which wounding induces hair regrowth remain to be determined. In this study, we found that ASK1-deficient ( ) mice exhibited marked delay of wounding-induced hair regrowth. mice, and that transplantation of activated macrophages induced hair regrowth in both wild-type (WT) and mice. These findings demonstrate the critical involvement of macrophages in hair growth and the need for regulation of infiltration and activation of macrophages in skin wounds by ASK1 for macrophage-dependent hair regrowth. mice, and wound areas were monitored for up to 20 d. mice ( and not depicted). mice (). Although it has been reported that hair regrowth is induced by wounding, the reason for this has remained unclear (; ). Hair follicles undergo repeated cycles of proliferation and differentiation (anagen stage), apoptosis (catagen stage), and resting (telogen stage) of epithelial cells (). mice at 12 d after wounding (; ). These findings suggested that lack of ASK1 delays wounding-induced anagen initiation. in the normal development of hair follicles of 14.5-d-old embryos or the spontaneous anagen-stage hair follicles of 18-d-old neonatal mice (unpublished data). mice (). These findings suggested that ASK1 deficiency does not affect the hair-regenerating activity of epithelial cells, per se. mice (). Wounding clearly induced activation of ASK1, p38, JNK, and ERK in WT mice. mice, indicating that ASK1 is required for activation of p38 and JNK in wounded skin (). mice. mice because of reduction of activity of p38 and JNK, which are required for the production of various inflammatory cytokines (see next section). mice) were collected, and gene expression in each group was monitored using high-density Affymetrix GeneChips. wounded skin. mice. A total of 138 of the 185 genes (76%) were up-regulated in an ASK1-dependent fashion by wounding (, cluster A). Altogether, 53 of the 138 genes (38%) were classified in the category of “immune-response regulation,” including various cytokines and chemokines using the Gene Ontology Consortium classifications (Table S1, available at ). Together with the results shown in , these findings suggested that ASK1 is required for regulation of postwounding immune responses, such as cytokine and chemokine production in wound areas. mice (Table S2, available at ). To confirm altered expression of macrophage-related genes, relative mRNA expressions of a macrophage-specific marker () and chemotactic factors ( [] and []) were measured by real-time RT-PCR in the same RNA samples used for GeneChip analysis. mice (). mice was significantly reduced compared with that in WT mice (). These findings suggested that ASK1 is required for infiltration of macrophages into wound areas, which may be caused by ASK1-mediated postwounding immune responses, such as the production of macrophage-specific chemotactic factors. Expression of β and in the wound area was also found by microarray and real-time RT-PCR analyses to be increased in an ASK1-dependent manner (Table S2 and ). IL-1β and TNFα are typical macrophage-activating factors, which may be expressed in wounded skin and in activated macrophages. mice compared with WT mice (). These findings suggested that ASK1 is also required for activation of macrophages, which may be mediated by macrophage-activating factors, such as IL-1β and TNFα, in the wound area. mice may be caused by reduction of infiltration of activated macrophages in wounds. To examine whether macrophages promote hair growth, we performed intracutaneous transplantation of bone marrow–derived macrophages (BMDMs) into the dorsal skin of WT mice. Hair growth was strongly induced by transplantation of BMDMs derived from WT mice, but by neither mouse embryonic fibroblasts (MEFs) nor dendritic cells (DCs; and Fig. S1 c, available at ). Staining for Ki67 (a cell proliferation marker) in the area of transplantation of BMDMs revealed that proliferation of epidermal basal cells and hair follicle cells, including bulge stem cells, was accelerated by macrophages, but not by PBS control, only in the injected area ( and Fig. S1, a and b). These findings clearly indicated that macrophages possess the ability to induce hair growth. mice, as well as those derived from WT mice (), indicating that ASK1 is not required for the macrophage function by itself to induce hair growth. skin wounds. To confirm that infiltration of macrophages is required for hair regrowth, we examined whether hair growth depends on the number of transplanted macrophages. wounded skin and the delay of hair regrowth. mice (). As shown in , ASK1 was also required for activation of macrophages, which may be induced by macrophage-activating factors in the wound area. These findings suggested that infiltration of macrophages into the wound area is not sufficient for induction of hair growth, and that activation of macrophages is also required for the induction of hair growth. mice may not be sufficiently activated, which is due in part to reduced production of macrophage-activating factors. mice. IL-1β–treated BMDMs induced hair in WT mice more effectively than did nontreated BMDMs. mice, as well as in WT mice ( and Fig. S1 d). We also performed intracutaneous transplantation of macrophages of a different type, which were differentiated by monocyte colony-stimulating factor (M-CSF–BMDMs). Robust hair growth was induced by transplantation of M-CSF–BMDMs in WT mice (, top). mice than in WT mice, but IL-1β–stimulated M-CSF–BMDMs induced hair growth in mice that was almost equivalent to that in WT mice (, bottom). These findings indicated that activated macrophages possess the ability to induce hair growth, and that regulation of both recruitment and activation of macrophages by ASK1 is required for macrophage-dependent hair regrowth. In this study, we found that macrophages play a crucial role in wounding-induced hair regrowth, and that regulation of infiltration and activation of macrophages in wound areas by ASK1 is required for macrophage-dependent hair regrowth (). BMDMs (). mice ( and not depicted). These findings suggested that ASK1 is selectively required for infiltration and subsequent activation of macrophages in wound areas (, left and middle), but not for the induction by macrophages of hair growth (, right). It is likely that infiltration of macrophages and activation of them are induced by macrophage-specific chemotactic factors and macrophage-activating factors, including MCP-1, MIP-1α, TNFα, and IL-1β, which are produced around the wound area in an ASK1-dependent fashion (). Although it has been reported that hair growth is induced by wounding, the reasons for this have remained unclear (; ). Synchronized hair follicle cycling in mice has been reported to be related to the number and activation of perifollicular macrophages (). This study is the first to provide direct evidence that infiltration and activation of macrophages are critically involved in postwounding hair regrowth, in which ASK1 plays an essential role. We are currently identifying hair growth–promoting factors produced by activated BMDMs, which appear to be thermosensitive proteinaceous factors (unpublished data). It has recently been found that macrophage-stimulating factor (MSP), which was originally identified as a chemotactic factor for macrophages, could induce hair growth, and that RON, the receptor for MSP, is strongly expressed in hair follicles (). We therefore examined whether ASK1-dependent up-regulation of MSP and/or RON is responsible for macrophage-induced hair growth. in levels of expression of and after wounding (Fig. S2, available at ), suggesting that MSP and RON may not be responsible for ASK1-dependent hair growth. () and WT mice were derived from heterozygote crosses of mice and constantly housed in a specific pathogen-free facility with a 12-h light/dark schedule and constant temperature. All experiments were performed using 8-wk-old female mice whose dorsal skin hair follicles were all in telogen stage. mice used in this study have been backcrossed on the C57BL/6J strain for 12 generations. All experiments were in accordance with protocols approved by the Animal Research Committee of the Graduate School of Pharmaceutical Sciences (University of Tokyo, Tokyo, Japan). Before injury, mice were anaesthetized and the dorsal hair was shaved. Two equidistant 5-mm full-thickness incisional wounds were punched in the middle of the dorsum as previously described (). Each wound region was digitally photographed (DSC-D700; Sony) at the indicated time points. Total RNA extraction was performed using the Isogen Reagent (Nippon Gene Co. Ltd.). These RNA extracts were used for oligonucleotide microarray analysis and RT-PCR analysis. The levels of expression of over 45,102 transcripts and variants were analyzed by oligonucleotide microarray (GeneChip Mouse Genome 430 2.0 Arrays; Affymetrix). Sequence clusters were created from the UniGene database (Build 107, June 2002). Analysis was performed essentially as previously described (). mice/WT mice) and ≤2 for the ratio (untreated skin of mice/WT mice). Classification of functions of genes was performed by Gene Ontology bioinformatics analysis using information from annotation files (2004.4 version). Single-stranded cDNA was synthesized with oligo (dT) primers from total RNA using SuperScript III reverse transcriptase (Invitrogen). The abundance of transcripts in cDNA samples was measured by real-time PCR with specific primers. Real-time RT-PCR amplifications of cDNA were performed using the Quantitation kit on the ABI Prism 7000 Sequence Detection System (Applied Biosystems). Skin extracts from each mouse were resolved by SDS-PAGE and analyzed by immunoblotting as previously described (). For morphological analysis, 5-μm sections of paraffin-embedded wounded skin were stained with hematoxylin and eosin. 7-μm sections of fresh-frozen wounded skin were immunostained with antibodies to F4/80 (A3-1; Serotec), CD11b (M1/70; Bioscience), and MHC class II (ER-TR3; BMA) at 4°C. 8-μm sections of fresh-frozen transplanted skin were immunostained with antibodies to Ki67 (TEC-3; DAKO Cytomation) at 4°C. Each staining was performed according to the manufacturer's instructions. For image acquisition, a microscope (DM4000B; Leica) equipped with PL FLUOTAR objective lenses (5×/0.15 NA, 20×/0.50 NA, and 40×/0.75 NA) and a digital camera (DC300FX; Leica) were used. Images were processed with IM50 image manager (Leica) and Photoshop CS software (Adobe). A crude population of BMDMs was generated in vitro from mouse bone marrow, as previously described (), but with some modifications. Bone marrow cells were cultured in RPMI 1640 medium containing 10% heat-inactivated FBS, 100 units/ml of penicillin G, 10 ng/ml of recombinant mouse granulocyte–monocyte colony-stimulating factor (Strathmann Biotech GmbH), and 5 ng/ml of recombinant mouse IL-4 (Genzyme/Techne) or 10 ng/ml of recombinant human M-CSF (PeproTech EC). The medium was replaced on days 2 and 4, and the nonadherent granulocytes were rinsed away. On day 7 of culture, the firmly adherent cells were obtained as a macrophage population characterized by certain features of morphology and phenotype, such as expression of high levels of F4/80 but low or undetectable levels of MHC class II and CD86 molecules (as determined by flow cytometry), and loosely adherent cells were obtained as DCs. We generated two types of BMDMs, which were cultured with granulocyte–monocyte colony-stimulating factor and IL-4 (BMDMs) or with M-CSF (M-CSF–BMDMs), and used as BMDMs in subsequent experiments. For activation of cells, the BMDMs, M-CSF–BMDMs, and MEFs obtained as described in this section were stimulated with media containing 10 ng/ml of recombinant human IL-1β (Roche) for 24 h. Cells were then washed five times with PBS and subjected to transplantation as described in the next section. mice were resuspended in 50 μl PBS, and cells or vehicle alone were injected intracutaneously in the dorsal skin, as previously described (), but with some modifications. The number of transplanted BMDMs was similar to the number of macrophages that had infiltrated into the wound area (∼10 cells), as estimated by counting macrophages in sections of wounded skin stained immunohistochemically with an antibody to the macrophage-specific marker F4/80. Fig. S1 (a and b) shows that the number of Ki67-positive cells increased only in the injection site of BMDMs, but not of those PBS control. Fig. S1 c shows that transplantation of DCs did not induce hair growth in WT mice. Fig. mice. Fig. S2 shows levels of expression of and in wounded skin. Table S1 shows Gene Ontology Consortium classifications. Table S2 shows the results of GeneChip analysis.

During mitosis, a parental cell divides to generate two new cells, each with genome content identical to that of the parental cell. In contrast, during meiosis, a single parental cell generates four gametes, each of which has half the number of chromosomes in the parental cell. The dramatically different pattern of meiotic and mitotic chromosome segregation is achieved in part by the temporal and spatial regulation of sister chromatid cohesion (for review see ). Both mitotic and meiotic cells replicate their chromosomes to generate sister chromatids that are held in juxtaposition by cohesion around the centromeres and along the length of the arms. In mitosis, the paired sister chromatids become attached to the mitotic spindle, all cohesion is dissolved, and sister chromatids segregate from each other. In contrast, during meiosis I (MI), homologues become linked as a consequence of reciprocal exchange and the presence of sister chromatid cohesion on the arms. After the linked homologues become attached to the MI spindle, cohesion on the arms is dissolved, allowing homologues to segregate from each other, but sister chromatids remain paired and segregate together to the same pole because cohesion at the centromeres and pericentric regions is protected from dissolution. MI is followed by MII, in which paired sister chromatids attach to the spindle, the remaining centromere and pericentric cohesion is dissolved, and sister chromatids segregate from one another. Therefore, the temporal and spatial regulation of sister chromatid cohesion plays a key role in generating the meiotic pattern of chromosome transmission, distinct from that of mitosis. Several studies have provided substantial insights into the molecular basis for the meiotic regulation of sister chromatid cohesion (for reviews see ; ). Cohesion is mediated largely by cohesin, a multisubunit protein complex conserved from yeast to human (; ). Cohesin is loaded onto the chromosomes at S-phase to generate cohesion between sister chromatids. In order for homologues to dissociate from each other in MI, cohesin is removed from the chromosome arms by two different mechanisms: one involves condensin and Polo kinase and acts before anaphase I (); the other requires separase and Polo kinase and acts at the onset of anaphase I (; ). The separase pathway is able to remove all cohesin and, therefore, to dissolve all sister chromatid cohesion, but the separase pathway and possibly the condensin pathway are blocked from removing cohesin at the centromeric and pericentric regions by a mechanism involving a conserved protein, called MEI-S332/Sgo1 (; ). The mechanism of meiotic cohesion protection by the MEI-S332/Sgo1 pathway is complex. MEI-S332/Sgo1 localizes to the centromeres and pericentric regions (; ; ; ). The pericentric localization depends on cohesin and Bub1, a component of the spindle assembly checkpoint (; ). Subsequently, Sgo1 recruits to the centromere phosphatase PP2A by binding its B regulatory subunit, called Rts1 (; ). The centromeric localization of PP2A is thought to enhance the ability to dephosphorylate centromeric cohesin and thus shields it from Polo kinase–dependent removal by the separase. As expected, mutations in Bub1, MEI-S332/Sgo1, and Rts1 cause precocious separation of sister chromatids during MI. These studies show that Bub1, MEI-S332/Sgo1, and PP2A are key components for the retention of centromere and pericentric cohesin. A recent study of cohesin protection in has implicated two new players, the Aurora B kinase and its binding partner inner centromere protein (INCENP; ). In vitro MEI-S332 is bound by INCENP and phosphorylated by Aurora B. In mitosis, MEI-S332 localization to the centromeric region is compromised by a mutation in INCENP with reduced activity or by a mutation in MEI-S332 that reduces its Aurora B–dependent phosphorylation as defined by in vitro studies. These observations have led to a model in which Aurora B–INCENP complex protects cohesion by phosphorylating MEI-S332 and thereby increasing MEI-S332 ability to bind to centromeres. The authors suggest that INCENP plays a similar role in meiosis because a partially defective allele of INCENP causes partial precocious separation of sister chromatids in meiosis I and partially compromises MEI-S332 localization. These important studies pose several questions. Is the role of Aurora B in protecting centromere cohesion conserved among eukaryotes? Given the partial defects, how important is Aurora B relative to other components that are known to be essential to protect centromere cohesion? Does Aurora B protect sister chromatid cohesion through cohesin localization? Does Aurora B help recruit MEI-S332 to the centromere in meiosis, as it apparently does in mitosis, or might it have a different or additional function? Previously, we discovered an essential function of Ipl1, the founding member of Aurora B, in modulating meiotic chromosome transmission in the budding yeast (). In this work, we show that one function of Ipl1 is to ensure the protection of centromeric cohesin during MI, indicating that this function of Aurora B kinase is conserved between yeast and flies. The role of Ipl1 in protection of meiotic centromere cohesion is as critical as MEI-S332/Sgo1, Bub1, and Rts1. Ipl1 is only marginally required for Sgo1 localization to the centromeres. Rather, Ipl1 is critical to maintaining the PP2A subunit Rts1 at centromeres after but not before the onset of anaphase I. The continued centromeric localization of Rts1/PP2A presumably ensures that centromeric cohesion is protected from separase until MII. Previously, we and others have shown that the Aurora B kinase is essential for meiosis in a diversity of organisms (; ; ). To dissect Aurora B function during meiosis of budding yeast, we generated a meiosis-specific null allele of () by replacing the endogenous promoter with a mitosis-specific promoter from (). The expression of is preserved in mitosis, and cells have no detectable mitotic mutant phenotype as judged by cell cycle progression and cell viability (unpublished data), but because no new Ipl1 is made from this promoter in meiosis and preexisting mitotic Ipl1 is degraded at the end of mitosis, cells with this allele have no detectable Ipl1 during meiosis (Fig. S1, available at ). cells initiate meiotic nuclear divisions, albeit with a delay (Fig. S1). These cells produce tetrads with unequal nuclei, and <1% spore viability (unpublished data). These observations suggest that Ipl1 has an essential function in meiotic chromosome transmission. To determine the role of Ipl1 in meiotic chromosome segregation, we examined cells for changes in several different aspects of chromosome structure. No detectable changes were evident in chromosome compaction or assembly of the synaptonemal complex (unpublished data). To monitor cohesion in a region near the centromere, we used the GFP chromosome-marking system. Tandem arrays of Tet operators were inserted into one homologue of chromosome V at the locus, which is ∼35 Kb away from the centromere (). If cohesion is preserved at the centromeres and in pericentric regions (henceforth referred to collectively as centromeric regions) throughout MI, a single GFP spot should be observed (, A and B, top). Failure to maintain sister chromatid cohesion leads to the appearance of two GFP spots (, A and B, middle and bottom). Before onset of anaphase I, a very small percentage of both wild-type and cells exhibited two GFP spots, indicating that Ipl1 function is not required for sister chromatid cohesion before anaphase I. We then examined telophase I cells for precocious separation of sister chromatids. Only 8% of wild-type cells exhibited two GFP spots (), indicating that the vast majority of sister chromatids maintain cohesion at the centromeres as expected. In contrast, 38% of cells (27% of type I and 11% of type II; ) show precocious separation of sister centromeres, about five times as many as in wild-type cells (). The severity of precocious chromatid separation in cells is similar to that in meiotic null alleles of () and (; ). These data suggest that Ipl1, like Sgo1 and Bub1, plays a major role in the protection of sister chromatid cohesion during MI of budding yeast. If the sole function of Ipl1 were to protect centromeric cohesion during MI, then the removal of Ipl1 would lead to segregation of both separated sister chromatids to only one of the two MI products (, A and B, middle), but in , about half of the cells with two GFP spots (the sister chromatids) segregate to opposite poles and thus to different MI products (, bottom). Furthermore when we used the same GFP system to mark both homologues of chromosome V, we observed that, in 52% of cells in MI, both homologues segregated to the same spindle pole, compared with only 3% in wild-type cells (unpublished data). These observations suggest that Ipl1, in addition to protecting centromeric cohesion, is required to ensure the co-oriented attachment of sister kinetochores to the MI spindle (, top). Because sister chromatid cohesion is mediated by cohesin, we addressed whether Ipl1 is required for the maintenance of cohesin in the centromeric region during MI. We followed the localization of different cohesin subunits in wild-type and cells by indirect immunofluorescence on spread nuclei. Similar results were obtained with cohesin subunits Smc1, Scc3, and Rec8. We show the localization of the representative cohesin subunit Rec8 tagged with a 3xHA epitope (). In prophase I of wild-type and cells, cohesin is able to associate with the chromosomes along their entire length (). By telophase I, 89% of wild-type cells lose cohesin staining on the chromosome arms but retain staining around the spindle poles, where the centromeres are positioned (). The remaining 11% show no cohesin staining. In contrast, by telophase I, 78% of cells lose all cohesin staining, a sevenfold difference (). We also examined cohesin localization in cells lacking Ipl1 that were arrested in metaphase I or in telophase I (Fig. S2, available at ). This analysis confirmed that the absence of Ipl1 eliminates cohesin localization to centromeres in telophase I but not to centromeres or arms in metaphase I. Finally, the severity of cohesin loss in cells is similar to that of and ( and see the following paragraphs), suggesting that, like Sgo1 and Bub1, Ipl1 plays a major role in the protection of meiotic cohesin. Furthermore, similarity of their mutant phenotypes implies that these proteins act in the same pathway to protect cohesin. To begin to investigate how Ipl1 protects centromeric cohesin, we examined the localization of Ipl1 through meiosis. We incorporated 3xHA into the C terminus of Ipl1 and visualized the localization of the tagged protein by indirect immunofluorescence in spread nuclei. Ipl1 localizes to prophase chromosomes, as intense dispersed foci superimposed on a diffuse weak general staining (, top). At metaphase and telophase I, Ipl1 is more concentrated around the spindle poles (). This pattern of staining is reminiscent of centromeric proteins (see the following paragraphs). At late anaphase I, a substantial portion of Ipl1 is localized along the microtubule spindle (). The pattern of dynamic localization of Ipl1 during yeast meiosis is very similar to that of Aurora B kinase and INCENP observed in mitosis in many organisms, which led to their designation as passenger proteins (; ). The dispersed foci of Ipl1 in prophase I and spindle pole body–proximal foci in metaphase and telophase I cells are similar to the localization pattern previously reported for Sgo1. Sgo1 immunostaining was subsequently demonstrated to correspond to its binding to centromeric regions by chromatin immunoprecipitation (ChIP; ), so we tested for colocalization of Ipl1 with the centromeric Sgo1 foci by performing indirect immunofluorescence to localize both Ipl1 and Sgo1 in yeast meiotic cells. At prophase I, when homologues are paired, Sgo1 forms ∼16 foci, corresponding to 16 paired yeast centromeres (). The Ipl1 foci colocalize with those of Sgo1 (). The colocalization of Ipl1 and Sgo1 at the centromere is also evident during telophase I, when yeast centromeres are clustered around the spindle poles (). Collectively, both Ipl1 and Sgo1 are concentrated at the yeast centromeres, supporting the idea that these proteins act together to protect centromeric cohesin and sister chromatid cohesion. The centromeric localization of Sgo1 depends on the spindle checkpoint protein Bub1 (; ). We therefore determined whether Bub1 is required for the localization of Ipl1 to yeast meiotic centromeres. We created a conditional allele that depletes Bub1 protein specifically during meiosis (). This mutant allele fails to protect centromeric cohesin during meiosis I (), and as expected, sister chromatids separate precociously (). As shown in , neither Ipl1 nor Sgo1 is concentrated at the centromeres in cells. They are dispersed throughout the chromosomes (, bottom). We used ChIP to confirm the failure of Sgo1 to associate with centromeric DNA in the absence of Bub1 (). Unfortunately, Ipl1 has not been detectable by ChIP even in wild-type cells. Bub1 is therefore required for the proper localization of both Ipl1 and Sgo1 to centromeric regions, lending further support to the idea that Ipl1 and Sgo1 act together in the same pathway to protect centromeric cohesin. We asked next whether the centromeric localization of Ipl1 depends on Sgo1. We examined Ipl1 localization in cells harboring a conditional allele (), which depletes Sgo1 during meiosis (not depicted; a similar allele was reported previously by and ). In wild-type cells, Ipl1 forms discrete foci that colocalize in prophase I with centromeric protein Sgo1 () and in telophase I with Sgo1 and spindle poles ( and ). In contrast, in >90% of cells, Ipl1 either fail to form detectable foci () or form a few foci that do not colocalize with Ctf19, a centromere protein (not depicted). The centromeric association of Ipl1 therefore depends on Sgo1, suggesting that Ipl1 acts concomitant with or downstream of Sgo1 in protection of centromeric cohesin. To address whether Ipl1 acts downstream of or concomitant with Sgo1, we asked whether Ipl1 function is needed for the centromeric association of Sgo1 or Rts1. Rts1 is a regulatory subunit of PP2A phosphatase. The binding of Rts1 by Sgo1 recruits PP2A to the centromeric region, where its phosphatase activity is used to protect cohesion (; ). In prophase I of wild-type cells, Sgo1 and Rts1 are concentrated at the centromeres (, , and ). In prophase I of cells, Sgo1 is able to associate with the centromeric DNA, as shown both by ChIP () and by indirect immunofluorescence of spread nuclei (). Similarly, Rts1 is also able to associate with centromeric regions in the Ipl1 mutant (). These results show that Ipl1 function is not necessary to establish localization of Sgo1 and Rts1 to the centromeric regions, but the presence of Sgo1 and Rts1 at centromeric regions in prophase I is clearly not sufficient to protect centromeric cohesin in anaphase I, given the cohesin defect in the mutant. We then examined Sgo1 and Rts1 localization in telophase I cells by indirect immunofluorescence of spread meiotic nuclei. In telophase I of wild type, 94% of cells show robust staining of Sgo1 and Rts1 at centromeric regions. In telophase I of , a similar fraction of cells show Sgo1 staining of centromeric regions, although the intensity is slightly reduced in a subset of cells (). Therefore, Ipl1 function appears to play a minor role in maintaining Sgo1 at centromeric regions in telophase I. In contrast, 66% of cells have no Rts1 staining in centromeric regions in telophase I (). In the remaining 34%, the level of Rts1 staining appears reduced relative to wild type. This cytological observation is confirmed by a biochemical analysis of Rts1 association with the centromeric DNA using ChIP (). These data, collectively, show that Ipl1 function acts downstream of Sgo1 localization by playing a critical role in the maintenance of Rts1 in centromeric regions in telophase I cells. During MI, sister chromatids segregate together because sister chromatid cohesion is protected at centromeric regions, and sister kinetochores are modulated to co-orient, attaching to microtubules from the same poles. Here, we report that Ipl1, the Aurora B kinase in budding yeast, is required for the protection of centromeric cohesin during meiosis I. In , mutations in INCENP and potential Aurora B phosphorylation sites of MEI-S332 have also implicated INCENP and Aurora B in the protection of centromeric cohesion in MI (). Given the evolutionary distance between these organisms, the role of Aurora B in the protection of meiotic cohesin is probably conserved in all eukaryotes. Interestingly, in budding yeast, Ipl1 also plays a critical role for the co-oriented spindle attachment of sister kinetochores (). Therefore, Aurora B controls both co-orientation of sister kinetochores and protection of centromere cohesion, the two modifications that differentiate chromosome behavior in meiosis I from mitosis. Having these two key features of MI chromosomes under the control of the same kinase may help to ensure they occur in a coordinated fashion. Our analyses of Ipl1 in budding yeast suggest that it protects centromeric cohesion by controlling the centromeric localization of Rts1, the regulatory subunit of PP2A, and a key component of the MEI-S332/Sgo1 pathway (; ). We propose that the spindle-assembly checkpoint protein Bub1 recruits Sgo1 to the centromeres. Sgo1 recruits the Aurora B complex and Rts1/PP2A. Aurora B ensures that Rts1 remains stably bound to centromeric regions in telophase I, thereby ensuring that Rts1/PP2A is properly positioned to protect centromeric cohesion. This model is driven by the following observations. From our phenotypic analyses, Ipl1 is as important as Sgo1, Bub1, and Rts1 in the protection of meiotic cohesion. Ipl1, Bub1, Sgo1, and Rts1 proteins are required to maintain the binding of cohesin to centromeric regions during MI. Sgo1 and Ipl1 are enriched in centromeric regions in a Bub1-dependent manner. Ipl1 depends on Sgo1 to associate with centromeric regions. Finally, we show that Ipl1 is required for the efficient maintenance of Rts1 but not Sgo1 in the centromeric regions. The role of Ipl1 in Rts1 localization is sufficient to explain its role in protecting cohesion, given the essential role of Rts1 in protecting centromeric cohesion (; ). In budding yeast, we observed a minor defect in Sgo1 localization in mutant cells, suggesting that Aurora B may also protect centromeric cohesion by facilitating the localization of MEI-S332/Sgo1 to centromeric regions, as has been proposed from a study in (). Although this study suggests that phosphorylation of MEI-S332 by Aurora B is important for MEI-S332 centromere localization, our study suggests that Aurora B may also modulate MEI-S332/Sgo1 localization through its effect on Rts1. The dissociation of Rts1 from centromeric regions in Aurora B–defective cells (this study) may partially destabilize Sgo1 localization. In support of this possibility, Rts1 binds directly to Sgo1 family members in several organisms (; ; ), and in human cells, Sgo1 localization depends on Rts1/PP2A family members (). Importantly, if MEI-S332/Sgo1 recruitment were the only function of Aurora B in the protection of centromeric cohesin/cohesion, then chromosomes in Aurora B mutant cells that manage to recruit MEI-S332/Sgo1 should protect centromeric cohesion, but this is clearly not the case in mutants of budding yeast, because chromosomes with Sgo1 (the vast majority) fail to protect cohesin (this study). In this light, it will be interesting to assess whether Aurora B in also plays a role in Rts1 recruitment as well as stable binding of MEI-S332. The regulation of Rts1 by Ipl1 provides several important insights into the protection of centromeric cohesion. In recent analyses, mislocalization of Rts1 to different regions of chromosomes gives rise to ectopic cohesion in MI (; ). This result suggests that the localization of Rts1 and presumably PP2A to chromosomes is sufficient to protect cohesion, but in this study we showed that, even though Rts1 localization occurs properly in cells before anaphase I, failure of cohesion ensues. Therefore, Rts1 localization to centromeric regions before anaphase I is not sufficient to protect cohesion. Either this localized Rts1/PP2A is inactive and must be activated by Ipl1 or Rts1/PP2A is constitutively active but must be maintained at the centromeric region until telophase I and possibly until metaphase II to protect centromeric cohesion. Ipl1 may regulate Rts1 function/localization by phosphorylating Rts1 or Sgo1. In , MEI-S332 is a target of Aurora B in vitro, and mutations in phosphorylation sites do lead to cohesion defects. In budding yeast, no changes in mobility of either Rts1 or Sgo1 have been detected in the absence of Ipl1, although at least Rts1 is a phosphoprotein (unpublished data). Determining the target of Ipl1 in cohesion protection will therefore require more sensitive studies of Rts1 and Sgo1 or the identification of additional potential targets. The regulation of cohesion by PP2A is one of the most recently discovered of a long list of processes influenced by this phosphatase. PP2A has been implicated in cell proliferation, gene expression, insulin regulation, and hyperosmotic stress, to name a few. In many cases, the specific function of the PP2A appears to result from association of the core catalytic subunits with a different B regulatory factor, like Rts1, but very few studies have examined the regulation of these regulatory factors. Understanding the regulation of a B subunit (Rts1) by Aurora B kinase (Ipl1) may provide a new paradigm for understanding how PP2A is regulated and how it in turn regulates so many other processes. Yeast strains used in this study are SK1 derivatives and are listed in Table S1 (available at ). The promoter of was used to generate meiotic conditional alleles of , , and as described previously (). The allele was obtained from A. Amon (Massachusetts Institute of Technology, Cambridge, MA; ). A PCR-based method was used to generate gene deletions and C terminus protein tagging (). Synchronous yeast cultures were induced for meiosis as described previously (). A single colony was inoculated in 5 ml YEPD overnight. This culture was diluted into YEPA medium to reach an OD (λ = 600) of 0.1. When the YEPA culture reached an OD of 1.4, yeast cells were harvested by centrifugation and washed once in prewarmed water. Cells were resuspended in the same volume of 2% KOAC to induce meiosis. All yeast cultures were inoculated at 30°C. Yeast nuclei spread and immunofluorescence were performed as described previously (). Anti-HA antibodies (12CA5 and 3F10; Roche) were used to detect HA epitope–tagged proteins (0.5 μg/ml). An anti-MYC antibody (9E10; Roche) was used to detect MYC epitope–tagged proteins. The α-tubulin antibody (YOL1/34; Serotec) was used to localize the microtubule spindle (1:500 dilution). A polyclonal antibody against yeast γ-tubulin (1:10,000 dilution; a gift from J. Kilmartin, Medical Research Council, Cambridge, UK) was used to localize the spindle pole bodies. A GFP antibody (ab290; 1:5,000; Abcam) was used to detect TetI-GFP in spread nuclei. Secondary antibodies (goat anti-mouse, anti-rat, and anti-rabbit; Jackson ImmunoResearch Laboratories) were used at a dilution of 1:500. Fluorescence images were acquired with a microscope (Axioplan 2 [Carl Zeiss MicroImaging, Inc.]; 100× objective lens [NA = 1.40] or 63× objective lens [NA = 1.40]) equipped with a charge-coupled device camera (Quantix; Photometrics). Displayed images were processed with IP-Lab for contrast adjustment and pseudocoloring. Yeast protein extraction and Western blot were performed as previously described (). Yeast synchronous cultures were induced for meiosis, and aliquots were withdrawn at 1-h intervals. Yeast total proteins were separated by SDS-PAGE and followed by immunoblotting with an ECL kit (Pierce Chemical Co.). ChIP was performed as previously described (). Table S1 shows yeast strains used. Fig. S1 shows depletion of Ipl1 protein during meiosis. Fig. S2 shows that Ipl1 protects centromeric cohesin at anaphase I. Online supplemental material is available at .

Attachment of chromosomes to spindle microtubules (MTs) is performed by kinetochores, which are large proteinaceous structures that assemble at the centromeric regions of each sister chromatid. According to the topology of its components, the kinetochore may be subdivided to two domains: the outer kinetochore (OKt) and the inner centromeric region (ICR). The OKt consists of several electron-dense zones, and it contains proteins that are involved in MT capture and regulation of the spindle assembly checkpoint (; ; ). The ICR is positioned between sister centromeres, and contains protein complexes that are involved in regulation of sister chromatid cohesion and modulation of MT attachment. These proteins include chromosomal passenger complex (CPC), mitotic centromere-associated kinesin (MCAK), and Shugoshin (Sgo; ; ). The CPC consists of Aurora B, INCENP, Survivin, and Dasra/Borealin (). Aurora B phosphorylates and inhibits the MT depolymerase MCAK, thus, controlling the polymerization/depolymerization state of tubulin filaments to achieve correct end-on attachment of MTs to the kinetochore (; ). The inner centromere protein (INCENP) subunit of the CPC can directly bind MTs, and a recent study suggests that the CPC can function as a bridge between the centromere and kinetochore MTs (). The CPC itself localizes along chromosomes in prophase, and then concentrates at the ICR in prometaphase and metaphase (). The mechanism that regulates CPC relocalization is unknown. Sgo plays critical roles in both cohesion and MT dynamics during metazoan mitosis. Sgo functions as an adaptor protein for phosphatase PP2A, recruiting it to the ICR. PP2A dephosphorylates SA2 subunit of the cohesin complex, preventing the latter from Plk1-dependent release during the G2–M transition, thus, maintaining centromeric cohesion until anaphase (; ). Localization of Sgo has been reported to depend on Bub1 activity (). Bub1 was first isolated in a screen for budding yeast mutants that were sensitive to benomyl, which is an inhibitor of MT polymerization (). It was later characterized as a protein kinase that is involved in spindle checkpoint response in yeast and in vertebrates (; ). Bub1 is not only involved in control of the checkpoint (), but also regulates the loading of spindle checkpoint proteins to kinetochores. Bub1 is recruited to centromeres early in prophase and promotes binding of Plx1, BubR1, Mad1, Mad2, Cenp-E, and Cenp-F to the OKt (; ; ). Interestingly, recruitment of these proteins does not require Bub1 kinase activity, suggesting that Bub1 plays a structural role in organization of the OKt. However, yeast Bub1 has additional functions in chromosome segregation that are independent of its ability to recruit the OKt components (; ). Recent studies suggested that Bub1 kinase may play a role in localization of Sgo in the ICR, thus providing a possible link between Bub1 kinase activity and chromosome segregation (). We decided to analyze whether Bub1's function in ICR assembly is restricted to Sgo targeting. We show that Bub1 kinase works as a master organizer of the ICR in both egg extracts and mammalian cells. Bub1 controls both stability and correct positioning of the CPC to the ICR in a kinase-dependent manner. Moreover, we find that soluble Bub1 kinase mediates binding of Sgo to mitotic chromatin, whereas CPC directs relocalization of chromatin-bound Sgo specifically to the ICR. Together, our results indicate that Bub1's dual role in Sgo and CPC targeting to the ICR represents a novel and important new paradigm for its action at multiple levels of kinetochore assembly. To assess a precise role of Bub1 in kinetochore formation, we immunodepleted Bub1 from meiotically arrested (cytostatic factor [CSF]) egg extracts (). Quantitative Western blotting showed that immunodepletion removed Bub1 to undetectable levels ( A and S1 A, available at ). As reported earlier (; ), we found that a sperm chromatin directly assembled into condensed chromosomes within Bub1-depleted CSF extracts was essentially devoid of BubR1 and the Dynein–Dynactin complex (Fig. S1 B and not depicted). To observe the effect of Bub1 depletion upon replicated chromatin (), we added sperm nuclei to mock- or Bub1-depleted CSF-arrested extracts that had been driven into interphase through the addition of 0.06 mM CaCl. After completion of DNA replication, mitosis was induced with a fresh aliquot of corresponding CSF extract. Under these circumstances, although Bub1 depletion caused a substantial reduction in the kinetochore recruitment of BubR1, Mad2, Bub3, and Dynein–Dynactin complex in the chromatin-bound fraction, small amounts of these proteins were generally visible on kinetochores ( A and S1 C). This residual population may reflect a difference in the organization of the unreplicated and replicated sperm chromatin assembled in extract (; ). Interestingly, depletion of Bub1 by RNAi in somatic cells similarly inhibited, but did not eliminate, recruitment of Mad2, CENP-E, and BubR1 to kinetochores (; ). These findings support previous conclusions that Bub1 has an important role in kinetochore formation, and suggest that this function is modulated by the status of the mitotic chromatin. It has been reported that Bub1 also regulates the recruitment of some (), but not all (; ), proteins associated with the ICR of the kinetochore. Therefore, we were curious to also examine the loading of proteins associated with the ICR. Surprisingly, we observed that Bub1 depletion reduced the amount of Aurora B and other CPC components (Survivin and Dasra A) into the chromatin fraction (), although no changes in the concentration of any CPC constituents were observed in total extracts. We examined the localization of the residual CPC bound to chromatin in the absence of Bub1. Although the amounts of BubR1 and p150 were reduced after Bub1 depletion (), they were still conspicuous and properly positioned at kinetochores according to their colocalization with the centromeric protein CENP-A (Fig. S1 C and not depicted), thereby allowing the use of BubR1 and p150 as kinetochore markers in Bub1-depleted extracts. In mock-depleted extract, Aurora B localized precisely on the ICR and partially colocalized with BubR1. Remarkably, Aurora B staining was no longer juxtaposed to BubR1 in Bub1-depleted extracts, suggesting that it was not properly targeted to the ICR (). Immunofluorescent analysis of Survivin and Dasra A localization in Bub1-depleted extracts similarly showed that these CPC components were displaced from kinetochore markers (). Notably, depletion of Bub1-related checkpoint kinase, BubR1, caused no changes in the Aurora B staining pattern (Fig. S2, B and C, available at ), showing that CPC localization is specifically regulated by Bub1. Because Aurora B binds to chromosome arms during prophase (), we wished to test whether CPC mislocalization in Bub1-depleted extracts was a result of prophase-like arrest. To do this, we assayed whether Cohesin was released normally from chromosomes in extracts lacking Bub1. Cohesin complexes largely dissociate from chromosome arms at prophase/prometaphase, but a small portion is retained at centromeres (; ). To monitor Cohesin dynamics, we isolated chromatin at three different stages: at interphase (at the end of DNA replication), at nuclear envelope breakdown (NEB), and at metaphase (30 min after induction of mitosis), and probed it for the presence of Scc1, which is a component of the cohesin complex. As expected, the levels of chromatin-bound Scc1 gradually decreased throughout progression from interphase to metaphase in mock-depleted extracts (). The dynamics of dissociation and the levels of Scc1 bound to metaphase chromosomes were indistinguishable in mock- and Bub1-depleted extr acts, indicating that depletion of Bub1 does not affect prophase–prometaphase transition. Collectively, our results suggest Bub1 depletion disrupts metaphase recruitment of the CPC to the ICR in egg extracts, but that this disruption does not reflect a defect in cell cycle progression. To test whether Bub1 plays a comparable role in other systems, we depleted Bub1 from HeLa cells by RNAi and analyzed distribution of Aurora B in control (Lamin A/C RNAi) and Bub1-depleted cells. 24–48 h after transfection of siRNA duplexes, the cells were incubated with nocodazole for 1 h, immediately followed by fixation and staining. In control mitotic cells, Aurora B consistently localized to the ICR, as expected (, top). In contrast, cells treated with Bub1 siRNA showed mislocalization of Aurora B (, middle and bottom), similar to the displacement that we observed in egg extracts lacking Bub1. Analysis of individual chromosomes indicated that Aurora B localized along the chromosome arms in the absence of Bub1, and did not display considerable colocalization with centromeric antigens (CREST; ). Also consistent with our observation that CPC recruitment is quantitatively reduced in Bub1-depleted egg extracts, the intensity of Aurora B staining was also reduced in Bub1-depleted cells (). Similar results were obtained using HeLa cells stably expressing Survivin-GFP; 24 h after Bub1 siRNA transfection, most prometaphase cells showed dispersed distribution of Survivin along chromosomes arms, whereas cells treated with Lamin A/C siRNA localized Survivin-GFP preferentially to the ICR (). Thus, it appears that the role of Bub1 in CPC recruitment to the ICR may be a general feature of metazoan systems. We reasoned that the decreased association of the CPC to chromatin in egg extracts or in cells lacking Bub1 might be linked to some properties of the complex that were altered in the absence of Bub1. Notably, depletion of Bub1 from egg extracts did not affect phosphorylation of histone H3, which is a well-known substrate of Aurora B (), arguing that mislocalized Aurora B was not inactivated as a kinase (). To test whether CPC stability was compromised, we produced recombinant xAurora B fused with a zz tag (Aurora B-zz) by translation of its mRNA in Aurora B (CPC)-depleted egg extracts. Recombinant protein was added to control or Bub1-depleted CSF extracts at concentration approximately equal to that of endogenous Aurora B, and CPC complexes that formed on Aurora B-zz were purified by IgG–Sepharose beads. Extracts lacking Bub1 did not promote efficient binding of CPC constituents, such as INCENP, Survivin, and Dasra A, to Aurora B-zz beads (). CPC stability is regulated by Aurora B kinase activity (). To understand whether impaired CPC formation is mediated by Bub1-dependent modulation of Aurora B itself, we performed the same kind of assay, but using a kinase-dead version of Aurora B (Aurora B-zz) as bait. Both Survivin and INCENP bound to Aurora B-zz less efficiently than to Aurora B-zz, as expected. However, the absence of Bub1 further exacerbated CPC formation so that Survivin and INCENP became barely detectable on Aurora B-zz beads (). These data suggest that Bub1 controls CPC stability in a manner that is independent of Aurora B activity. It is formally possible that in the absence of Bub1, CPC becomes more stable; this is an alternative explanation for why exogenous Aurora B-zz accumulated less CPC components (). However, because it is known that down-regulation of the single CPC component compromises residual CPC recruitment to the chromatin (; ), the phenomena that resembles our observations in Bub1-depleted extracts ( and ), we believe that Bub1 stabilizes the CPC complex. Because Bub1 controls stability of the CPC, we reasoned that Bub1 might phosphorylate one or several of its subunits. To address this issue, we performed a kinase assay using recombinant Bub1 and CPC purified from Bub1-depleted extracts by immunoprecipitation (, left) or by Aurora B-zz pulldown (, right). The INCENP subunit of the CPC is phosphorylated by Aurora B in the absence of Bub1 (; ). Strikingly, however, addition of Bub1 enhanced INCENP phosphorylation levels (). Only the INCENP subunit of the CPC appears to be phosphorylated by Bub1, as we could not detect any [P]-containing band corresponding to Dasra, Survivin, or Aurora B (unpublished data). Moreover, Bub1 appears to phosphorylate INCENP directly because similar assays that also included core histones as substrates showed that the level of histone H3 phosphorylation was not affected by the presence of Bub1, arguing that Bub1 does not stimulate Aurora B activity (). Our results suggest that Bub1 controls stability of the CPC by phosphorylating its INCENP subunit. Bub1 recruits several proteins onto kinetochores through protein–protein interactions that are independent of its kinase activity (). To determine whether Bub1 kinase activity is similarly dispensable for CPC localization, we expressed wild-type Bub1 and a kinase-dead mutant (Bub1; ). Recombinant Bub1 and Bub1 were added to Bub1-depleted extracts. Consistent with previous observations, we found that Bub1 kinase activity was not required for restoration of BubR1, Mad2, or p150 recruitment to kinetochores ( and not depicted; ). Moreover, exogenous wild-type Bub1 quantitatively restored recruitment of Aurora B, Dasra A, and Survivin to chromatin, as well as their localization to the ICR (, A and B; and not depicted). In striking contrast to its ability to restore localization of the OKt components, however, Bub1 failed to rescue mislocalization of CPC components caused by Bub1 depletion (). These observations indicate that Bub1 kinase activity is absolutely necessary for the regulation of CPC localization during prometa- and metaphase. Collectively, our results demonstrate that Bub1 controls localization of the chromosome passenger complex in the ICR during mitosis in vertebrates in a kinase-dependent manner. We examined two additional ICR components, MCAK and Sgo, to assess whether ICR structure was generally disrupted in the absence of Bub1, or whether this effect was limited to CPC. Remarkably, Bub1 depletion by RNAi caused a substantial reduction in the amount of MCAK associated with kinetochores (). Because localization of MCAK to the ICR depends on Aurora B activity (; ; ), it is highly possible that this defect in its recruitment is a secondary consequence of Aurora B displacement. In this scenario, it is notable that Aurora B kinase activity alone appears to be insufficient for MCAK targeting, suggesting that CPC localization and/or interactions among CPC members may also be critical for its full biological function. Bub1 is essential for kinetochore localization of Sgo in both yeast and mammals (; ). It has also been shown that Bub1 kinase activity is required in fission yeast for the centromeric localization of spSgo1 and spSgo2 (). Consistent with these studies, removal of Bub1 from egg extract prevented binding of Sgo (xSgo) to mitotic chromatin (), although total xSgo levels within depleted extract remained unchanged. We further sought to determine whether Sgo targeting required Bub1 kinase activity, as the CPC does, or is kinase-independent, as is the case for OKt components. Remarkably, the ability of xSgo to bind mitotic chromatin in Bub1-depleted extracts could be rescued by addition of wild-type Bub1, but not Bub1 ( B). In combination with our earlier finding on the CPC (), these data strongly argue that Bub1 kinase activity is critical for general organization of a functional ICR. Because it is the kinase activity of Bub1 that is generally required for targeting of ICR components, we wondered whether Bub1 itself must localize at kinetochores to perform its function. Depletion of Aurora B from egg extracts has been reported to prevent Bub1 binding to chromosomes (). We confirmed this observation and examined Bub1 after Aurora B depletion ( and not depicted). We found that neither its total levels nor its kinase activity were substantially affected. In the absence of Aurora B, xSgo still bound to chromatin, albeit at slightly reduced levels ( A and S2 B). As expected, simultaneous depletion of Aurora B and Bub1 resulted in loss of xSgo from mitotic chromatin, suggesting that Aurora B depletion does not bypass the requirement for Bub1 in recruiting xSgo to chromatin (). These data demonstrate that Aurora B and other CPC components are not required for recruitment of xSgo to chromatin. Additionally, they indicate that although Bub1 kinase activity is essential for xSgo recruitment to mitotic chromatin, its own association to chromosomes is dispensable. We examined whether xSgo loaded onto chromatin in the absence of CPC was correctly localized. In contrast to the well- defined kinetochore staining of xSgo in control extracts, xSgo was diffusely distributed throughout chromosomes assembled in CPC-depleted extracts (). Although Bub1 was thus sufficient to establish xSgo chromatin binding, the CPC appears to be essential for restriction of xSgo to the ICR. To determine whether the capacity of the CPC to restrict xSgo to the ICR might involve direct interactions, we made reciprocal immunoprecipitation using Aurora B and xSgo antibodies. We found that xSgo could be coprecipitated using anti-Aurora B antibody and Aurora B could be coprecipitated with xSgo (), clearly demonstrating that the CPC and xSgo interact with each other. Notably, xSgo did not appear to be a stochiometric component of the CPC on Coomassie-stained gels (). In addition, we could co-deplete neither xSgo from egg extracts through Aurora B depletion nor Aurora B through xSgo ( and Fig. S3 A, available at ). We had observed that addition of recombinant Bub1 to Bub1-depleted egg extracts, either at the start of reaction () or at the induction of mitosis (not depicted), fully rescued CPC and xSgo localization. Finally, we wished to determine the execution point of Bub1's role in ICR assembly, and, specifically, whether Bub1's kinase activity is essential during the interval of mitotic chromosome assembly. To answer this question, we first assembled fully condensed mitotic chromosomes in Bub1-depleted extracts, and then added recombinant Bub1 or Bub1 (). Consistent with our previous results, Aurora B was displaced from the ICR and xSgo was unable to bind to the mitotic chromatin in Bub1-depleted extracts. Moreover, even allowing a prolonged interval for mitotic chromatin assembly did not cause accumulation of Aurora B at the ICR and loading of xSgo onto chromatin in the extracts lacking Bub1 (). Addition of exogenous wild-type Bub1, but not Bub1, to egg extracts 30 min after mitotic induction also rescued proper localization of Aurora B and xSgo in the ICR (). These findings clearly demonstrate that Bub1 can promote the formation of the ICR in preformed mitotic chromosomes. Furthermore, they suggest that the assembly of the ICR is independent of many other aspects of chromosome condensation. We have shown that Bub1 plays a central role in ICR formation, acting at multiple points in this assembly pathway. First, Bub1 controls CPC localization to the ICR. In the absence of Bub1, the CPC can bind to chromosome arms, albeit with reduced efficiency, but it does not become associated to the ICR ( and ). Although the activity of Aurora B as a histone H3 kinase was not lost under these circumstances, the stability of the CPC was markedly altered (). Second, as in earlier studies, we found that Bub1 mediates xSgo recruitment to the ICR ( and ). In addition, we found that Bub1 acts primarily by promoting xSgo binding to mitotic chromatin (); Bub1 can accomplish this function even when it is not being stably associated to mitotic chromosomes or kinetochores. Third, in contrast to chromatin binding of xSgo, Bub1 by itself is insufficient to direct xSgo to the ICR in the absence of the CPC (). Together, these findings suggest that Bub1 regulates localization of ICR components through mechanisms that are both CPC-dependent and -independent. Remarkably, we find that Bub1's kinase activity is essential for all of its roles in ICR assembly. It is also notable that Bub1 kinase can accomplish its essential roles in ICR formation in a manner that is not coupled to chromosome condensation or the OKt formation because it was able to fully restore ICR assembly on completely condensed replicated chromosomes (). Our data suggest that Bub1 plays an indispensable role in localizing the CPC to the ICR. In egg extracts, Bub1 depletion completely prevented CPC recruitment to the ICR in a manner that could be fully rescued with wild-type Bub1, but not with kinase-dead mutant Bub1 (). We similarly observed mistargeting of the CPC to chromosome arms in prometaphase-arrested HeLa cells that had been depleted of Bub1 through RNAi (), arguing that Bub1's role in controlling CPC distribution may be a general feature of metazoan systems. Notably, our results do not agree with those of earlier studies, which concluded that the CPC could localize to centromeres in a Bub1-independent manner (; ). There are two possible sources of this discrepancy. First, our finding that soluble Bub1 can promote xSgo localization suggests that it does not need to achieve a high level of kinase activity on kinetochores to execute this function. If a limited level of soluble (or kinetochore-associated) Bub1 activity is able to promote CPC recruitment, then partial RNAi-mediated depletion of Bub1 should not cause redistribution of the CPC. Indeed, we were also able to find cells that had substantially reduced levels of Bub1 on kinetochores after Bub1 siRNA, but which contained several chromosomes with the proper localization of Aurora B in the ICR, as might be expected in this case (unpublished data). Second, even after depletion of Bub1 to immeasurable levels, we continue to observe loading of the CPC throughout chromosome arms in both egg extracts and HeLa cells ( and ), implying that recruitment of the CPC to prophase chromosomes is independent of Bub1. Because costaining with centromere markers was not provided in the earlier studies, it is conceivable that arm-associated foci of CPC staining might have been incorrectly attributed to ICR- associated populations. Our results strongly suggest that formation of the OKt and the ICR differs by their sensitivity to Bub1's activity. Recruitment of such OKt components as Plx1, BubR1, Mad1, Mad2, Cenp-E, and Cenp-F depends on Bub1 itself, but not on its kinase activity (; ; ). One notable exception is Mps1, whose localization to the OKt is controlled by Bub1's kinase function (). On the other hand, localization of all of the ICR elements tested (CPC and xSgo) absolutely requires Bub1's kinase activity (– ). Together, our data indicate that initiation of OKt assembly relies on physical interaction of their elements with Bub1, but that the formation of the functional ICR requires only kinase activity of Bub1. Our result that Sgo localization requires both Bub1 and Aurora B is consistent with data obtained in other model systems (; ; ; ). However, our findings address several key issues that were not predicted. First, we show that Bub1 mediates binding of xSgo to the mitotic chromatin, by itself, not to the kinetochore. Second, we show that soluble Bub1 kinase can promote binding of xSgo to mitotic chromatin, whereas Aurora B (CPC) directs chromatin-bound xSgo to the ICR. It is also worth mentioning that localization of both Aurora B (CPC) and Bub1 to centromeres depend on each other (, , and ; ). Based on these observations, we would like to propose a scheme for Bub1-mediated events in kinetochore assembly. During prophase, Aurora B kinase initiates kinetochore formation, probably by phosphorylation of centromeric proteins like CENP-A (). This results in recruitment of Bub1 onto kinetochores and assembly of the spindle checkpoint components at the OKt. In return, Bub1 kinase, while soluble or kinetochore bound, controls formation of the ICR by two pathways. First, it promotes relocalization of the CPC from chromosome arms to the ICR. It is feasible that Bub1 controls not only stability of the CPC but also its association with yet unidentified component that is essential for CPC targeting to the ICR (). This idea is supported by the observation that a mixture of recombinant CPC components (Aurora B, Dasra A, INCENP, and Survivin) does not rescue CPC depletion in egg extracts (), implying that the in vivo CPC is built up of more than these four constituents. Second, Bub1 may phosphorylate Sgo or its mitotic chromatin binding sites to promote its recruitment. This chromatin-bound xSgo requires the CPC to further direct its localization at the ICR; then Bub1 itself or Bub1-mediated accumulation of the CPC targets MCAK into the ICR. In summary, the ICR is a dynamic structure whose assembly is independent from many other aspects of chromosome condensation and is controlled by Bub1 kinase through a web of interactions. Bub1 promotes binding of xSgo to chromatin and mediates relocalization of CPC from chromosome arms. A cDNA encoding the Bub1 kinase-dead mutant and Aurora B kinase-dead mutant was generated by PCR. Wild-type and kinase-dead mutant of Bub1 were cloned into modified pGEM transcription vector that contains 5′ and 3′ UTR regions of β-globin; wild-type and kinase-dead mutant of Aurora B, both fused with zz-tag, were cloned into similarly modified SP6-based vector (both vectors were provided by Y.-B. Shi, National Institutes of Health, Bethesda, MD). RNA transcripts were produced using mMessage mMACHINE T7 or SP6 transcription kit correspondingly (Ambion). Production of proteins in egg extract was performed as previously described (). His-tagged wild-type xBub1 was expressed in High Five cells (baculovirus was provided by J. Maller, University of Colorado, Denver, CO) and purified as described previously (). Antibodies against the following proteins were used: Sgo, RCC1, PIASy, and topoisomerase II have been described previously (; , ; ); Dynein IC (clone 74.1; Abcam); p150; Aurora B (Beckman Dickinson); Survivin (R&D Systems); phosphorylated histone H3 (Millipore); Cenp-A (either a gift from A. Straight [Stanford University, Stanford, CA] or raised in rabbits against peptide MRPGSTPPSRRKSRPPRRVS-C); Bub3 and Mad2 (a gift from R.H. Chen, Institute of Molecular Biology, Taipei, Taiwan); Dasra A (a gift from H. Funabiki, The Rockefeller University, New York, NY); INCENP (a gift from P.T. Stukenberg, University of Virginia, Charlottesville, VA); and human Bub1 (either a gift from S.S. Taylor [University of Manchester, Manchester, UK] or purchased from Sigma-Aldrich, clone 14H5). CREST sera were a gift from I. Ouspensky (National Institutes of Health, Bethesda, MD). Polyclonal anti– Bub1 (aa 274–467), anti– Aurora B, anti– RanGAP1, anti-Scc1 (EPYSDIIATPGPRFH), anti– BubR1 (aa 189–359), anti-hMCAK (C-IQKQKRRSVNSKIPA), and anti-xSgo (aa 1–663) were raised in rabbits or chickens and affinity purified. All secondary antibodies conjugated with Alexa Fluor 488, 568, or 647 were obtained from Invitrogen. sperm nuclei and low-speed extracts of eggs arrested by CSF were prepared as previously described (Kornbluth, 2001). For immunoprecipitation, protein A–conjugated Sepharose beads coupled to corresponding antibodies were prepared. CSF-arrested extract diluted fivefold with CSF-XB buffer (5 mM Hepes-KOH, pH 7.7, 100 mM KCl, 2 mM MgCl, 10 μm CaCl, and 5 mM EGTA) supplemented with 20 mM β-glycerophosphate and 10 μg/ml each of leupeptin, pepstatin, and chymostatin was incubated with antibody-coated beads for 1.5 h at 4°C. After incubation, beads were washed four times with CSF-XB buffer supplemented with 20 mM β-glycerophosphate and 0.5% Triton X-100, and the precipitates were eluted from beads by addition of 0.1 M glycine, pH 2.3. For immunodepletion, protein A–conjugated magnetic beads (Dynal) were incubated overnight with indicated antibodies or rabbit IgG (Vector Laboratories) at 4°C and then covalently coupled using Dimethyl pimelimidate 2 HCl (Pierce Chemical Co.) according to the manufacturer's protocol. Beads were blocked with 10% gelatin hydrolysate (Sigma-Aldrich) in CSF-XB buffer for 20 min, washed with CSF-XB buffer and incubated with extracts for 1 h at 23°C or at 4°C. Beads were removed by magnetic separation, and supernatants were used for the experiments. Interphase was induced by addition of CaCl at a final concentration 0.06 mM to CSF-arrested egg extracts. Sperm chromatin was added at concentration 1,000–3,000 nuclei/μl. After DNA replication, 2/3 vol of correspondent CSF-arrested extract was added to induce mitosis. Nocodazole (Sigma-Aldrich) at a final concentration 20 μg/ml was added along with CSF extracts, where indicated. For pulldown assay, extracts supplemented with either Aurora B-zz or Aurora B-zz were diluted five times with CSF-XB buffer (containing 20 mM β-glycerophosphate and 10 μg/ml LPC) and incubated with IgG–Sepharose beads for 2 h at 4°C. After incubation, beads were washed four times with CSF-XB buffer supplemented with 20 mM β-glycerophosphate and 0.5% Triton X-100, and the precipitates were eluted from IgG–Sepharose beads by addition of 0.1 M glycine, pH 2.3. For in vitro kinase assays, the CPC was precipitated from Bub1-depleted egg extracts using antibody against Aurora B, or purified on IgG–Sepharose through affinity to exogenously added Aurora B-zz. Control beads and beads containing precipitates were incubated with addition of either baculovirus-expressed xBub1 or buffer alone (0.8× CSF-XB buffer containing 20 mM β-glycerophosphate, 1 mM DTT, 5 mM MgCl, 1 mM ATP, and 1 μCi of γ[P]ATP). To analyze activity of Aurora B toward exogenous substrates, core histones (a gift from R. Kamakaka, University of California, Santa Cruz, Santa Cruz, CA) were added to the kinase assay reaction at a final concentration of 70 μg/ml. After 35-min incubation at 23°C, the reactions were stopped by the addition of SDS sample buffer. Protein samples were separated by SDS-PAGE and phosphate incorporation was determined by PhosphorImager (GE Healthcare). 100-μl aliquots of each reaction were diluted fivefold with 0.8× CSF-XB buffer containing 20 mM β-glycerophosphate, 5% glycerol, and 0.5% Triton X-100, and were incubated for 1 min at RT. The samples were then layered onto a 35% glycerol-containing CSF-XB cushion and centrifuged at 10,000 for 5 min at 4°C. The pellets were resuspended in the same buffer, and the centrifugation was repeated. 60 μl of SDS-PAGE sample buffer was added to the resulting pellet, and the samples were heated to 100°C for 5 min, followed by vortexing. For purification of interphase chromatin, 100-μl aliquots of extract were diluted with 0.8× CSF-XB buffer containing 20 mM β-glycerophosphate and 5% glycerol and incubated for 1 min at RT, followed by centrifugation through the cushion at 10,000 for 5 min at 4°C. HeLa and HeLa cells were cultured in DME containing 10% FBS (BioWest) at 37°C. EGFP-Survivin plasmid was provided by S. Dimitrov (Institut Albert Bonniot, La Tronche Cedex, France). HeLa cells stably expressing EGFP-Survivin were made using Effectene (QIAGEN) according to the manufacturer's protocol. siRNA duplexes designed to repress Lamin A/C (Dharmacon) or Bub1 (corresponding to nt 273–295 of the Bub1 coding region; QIAGEN; ), were transfected using Oligofectamine (Invitrogen) according to the manufacturer's instructions. Cells were analyzed 24–48 h after transfection. Cells on coverslips were washed with PBS containing 1 mM MgCl and immediately fixed with 4% PFA. After fixation, cells were washed in TBS-T and permeabilized with 0.2% Triton X-100. Chromatin or cells were blocked with 3% BSA for 30 min, and then stained with corresponded primary antibody for 40 min at RT, followed by the staining with secondary antibodies for 40 min. DNA was counterstained by 4 μg/ml Hoechst 33342 (Sigma-Aldrich). Samples were mounted in medium (Vectashield; Vector Laboratories) and sealed. Specimens were observed by a fluorescent microscope (Axioskop; Carl Zeiss MicroImaging, Inc.) with Iris 100×/1.4 NA objective (Carl Zeiss MicroImaging, Inc.). Images were taken with a charge-coupled device camera (Orca II; Hamamatsu) operated by Openlab software (Improvision). 0.9-μm-wide slices with 0.1 μm distance were taken. Flattened stacks of images were taken for the same exposure and processed in the same manner. For , images were acquired with a LSM 510 Meta system (Carl Zeiss MicroImaging, Inc.). The scale bar is 20 μm throughout, unless otherwise specified. Fig. S1 shows that Bub1 is undetectable in Bub1-depleted egg extracts, and that BubR1 does or does not localize to kinetochores in such extracts, depending on the status of mitotic chromatin. Fig. S2 shows that depletion of BubR1 does not affect formation of the ICR and that depletion of Aurora B co-depletes CPC components. Fig. S3 shows that depletion of Sgo does not alter localization of the CPC in the ICR. The online version of this article is available at .

Alternative splicing is a regulatory mechanism that allows genes to encode for multiple protein isoforms that often play different biological roles (; ; ). It arises from the optional use of alternative splice sites within a pre-mRNA. In mammals, the signals that define the beginning and the end of an intron are ill defined, and the authentic splice sites can be identified only with the help of additional cis-acting elements, named “splicing enhancers” and “splicing silencers.” Usually, any given region of a pre-mRNA contains, in addition to various potential exon–intron boundaries, several splicing enhancers and silencers that antagonize each other (). This enormous body of information is decoded by two families of RNA-binding proteins: the serine/arginine (SR) proteins and the heterogenous nuclear RNPs (hnRNPs; ; ). The SR proteins consist of one or two RNA-binding domains and a domain rich in SR dipeptides. They bind to splicing enhancers and usually activate splicing at nearby splice sites. The hnRNPs bind mostly, but not always, to splicing silencers and therefore inhibit splicing at nearby splice sites. Thus, the fate of a pre-mRNA region is usually decided by the antagonism between hnRNP and SR proteins (for review see ). The best-described example is the antagonism between the SR protein ASF/SF2 and hnRNP A1. The ratio of ASF/SF2 to hnRNP A1 determines the usage of splice sites, because ASF/SF2 recruits the U1 small nuclear RNP to the pre-mRNA, a crucial step in establishing a 5′ splice site, whereas hnRNP A1 counteracts U1 binding (). However, the limited number of hnRNP and SR proteins, as well as their limited binding specificity for the pre-mRNAs, cannot faithfully determine splicing in so many pre-mRNAs (). Additional proteins are likely required to confer specificity to the regulation of alternative splicing in live cells and to decipher the signaling pathways triggered by environmental cues. A well-suited candidate to integrate signal transduction pathways and RNA metabolism is Sam68. This RNA-binding protein belongs to the signal transduction and activation of RNA (STAR) metabolism family (; ), and it was first identified as a target of the tyrosine kinase Src (; ). STAR proteins are characterized by a KH (hnRNP K homology) domain embedded in a highly conserved region called GSG (GRP33/Sam68/GLD1) domain, which is required for homodimerization and sequence-specific RNA binding (; ). Sam68 has been implicated in the regulation of cell cycle progression and apoptosis (; ). The subcellular localization and the affinity of Sam68 for RNA are regulated by posttranslational modifications, like phosphorylation and methylation (, ; ; ). The localization of Sam68 is predominantly nuclear, suggesting a function in pre-mRNA processing. A role for Sam68 in alternative splicing was demonstrated by its ability to induce inclusion of the variable exon v5 in the CD44 mRNA (). The activation of Ras in response to phorbol ester stimulation triggered the phosphorylation of Sam68 by Erk1/2 and Sam68-dependent v5 inclusion in mouse T-lymphoma cells. More recently, it has been observed that Sam68 cooperates with the splicing activator SRm160 in the regulation of CD44 alternative splicing (). In addition, Sam68 was found to associate with Brm, a component of the SWI/SNF chromatin remodeling complex (), suggesting that it may participate in splicing events during the initial stages of CD44 pre-mRNA synthesis. Aside from CD44 alternative splicing, no information is available on other pre-mRNAs that could be regulated by this protein. Given the role of Sam68 in the regulation of apoptosis, it would be important to identify cellular targets involved in such a process. Alternative splicing plays a crucial role in the control of apoptosis. Several pre-mRNAs for cell death factors are alternatively spliced, yielding isoforms with opposing functions during programmed cell death (). A clear example is the Bcl-x transcript, which is alternatively spliced to produce the antiapoptotic Bcl-x(L) or the proapoptotic Bcl-x(s) (). The choice of alternative splicing of Bcl-x pre-mRNA reflects the sensitivity of cells toward agents that induce apoptosis. For example, cancer cells often up-regulate the antiapoptotic Bcl-x(L) isoform, and this event is associated with increased risk of metastasis, reduced sensitivity to chemotherapeutic treatments, and poor prognosis (; ). On the other hand, treatment of cancer cells expressing high levels of Bcl-x(L) with antisense oligonucleotides complementary to the Bcl-x(L) splice site favor the expression of Bcl-x(s) and sensitize cells to undergo apoptosis (). These observations strongly indicate that manipulation of Bcl-x alternative splicing may have important applications in cancer treatment. However, despite the crucial importance of this process, little information is available on splicing regulators controlling splice site selection of Bcl-x pre-mRNA. Recent reports have shown that hnRNP F/H induce Bcl-x(s) expression (), whereas SAP155 is required for selection of the Bcl-x(L) splice site (). Moreover, the ratio between the two isoforms can be shifted toward Bcl-x(s) by an increase in ceramide (). Hence, cells can quickly switch to an apoptotic program as a consequence of altered growth conditions through a change in splice site selection of the Bcl-x pre-mRNA. In the present work, we have investigated the role of the RNA-binding protein Sam68 in the regulation of apoptosis. We have identified the first cellular targets of Sam68 involved in apoptosis and provide evidence that Sam68 favors the selection of the Bcl-x upstream 5′ splice site and the production of the proapoptotic Bcl-x(s) isoform. Sam68 is a RNA-binding protein whose intracellular levels regulate cell cycle progression and apoptosis (; ). We set out to investigate whether Sam68 plays a direct role on the posttranscriptional regulation of mRNAs encoding known apoptotic regulators. Endogenous Sam68 was immunoprecipitated with an anti-Sam68 antibody from HEK293 cell extracts () prepared under conditions that preserve RNA and RNPs (). RNA was extracted from the immunoprecipitates and analyzed by RT-PCR for the presence of genes involved in the regulation of apoptosis. We found that Sam68 specifically binds to the endogenous Bak, Bax, Bcl-x, and Bim mRNAs, but not to Bcl2 mRNA. Control immunoprecipitations with preimmune IgGs gave no enrichment in any of the mRNAs analyzed (). To test whether Sam68 can recognize these mRNAs on its own, we purified GST-Sam68 from bacteria and total RNA from HEK293 cells and performed GST pull-down experiments. RT-PCR analysis showed that Sam68 can directly bind to these mRNAs in a cell-free assay (). To date, the only cellular target identified for the splicing activity of Sam68 is CD44 (). Because Sam68 binds to the mRNA for Bcl-x, an apoptotic regulator known to undergo alternative splicing (), posttranscriptional regulation of this mRNA by Sam68 was investigated further. Alternative splicing is usually governed by a delicate balance of activators and inhibitors of splice site selection that have redundant functions in the cell (; ). Hence, we investigated whether changes in the intracellular concentration of Sam68 modulate the alternative splicing of Bcl-x in HEK293 cells. We found that down-regulation of Sam68 by RNAi changed the ratio of alternatively spliced Bcl-x isoforms, with an increase in the antiapoptotic Bcl-x(L) and a decrease in the proapoptotic Bcl-x(s) transcript (approximately twofold; , right). A decrease in Bcl-x(s) was observed also at the protein level after depletion of Sam68 (, left). As a second approach, we overexpressed GFP or GFP-Sam68 constructs in HEK293 cells. GFP-positive cells were sorted, and cellular extracts were prepared for RT-PCR and Western blot analyses. Up-regulation of Sam68 caused an increase in the Bcl-x(s)/Bcl-x(L) ratio at the mRNA level (4.5-fold; ). A similar increase in ratio was obtained by transfection of HEK293 cells with a Bcl-x(L) antisense oligonucleotide (3.8-fold; ). Up-regulation of Sam68 also increased Bcl-x(s) protein (). These results demonstrate that fluctuations in the intracellular concentration of Sam68 affect the alternative splicing of endogenous Bcl-x mRNA. To investigate the regulation of Bcl-x splicing by Sam68 in more detail, we used a Bcl-x minigene that spans the whole alternatively spliced region from exon 1 to 3, with a shortened intron 2 (; ). Cotransfection of GFP-Sam68 with this reporter strongly enhanced the formation of the short variant of the Bcl-x transcript with respect to transfection of GFP alone (). The selection of the upstream 5′ splice site in exon 2 was stimulated in a dose-dependent manner when the Bcl-x minigene was cotransfected with increasing amounts of GFP-Sam68 (). Hence, the Bcl-x minigene recapitulates the effects of Sam68 on the endogenous Bcl-x mRNA. Phosphorylation of Sam68 by the MAPK Erk1/2 induces inclusion of exon v5 in the CD44 mRNA (). To determine the effects of Erk1/2 signaling on Sam68-mediated splicing of Bcl-x, HEK293 cells were transfected with GFP or GFP-Sam68 and treated with selective inhibitors of the MAPK pathway. We observed that neither U0126, which blocks activation of Erk1/2 by MEK, nor JNK inhibitor 1, which blocks activation of the JNK family of MAPKs that are also involved in alternative splicing (), substantially affected Sam68-mediated selection of the 5′ splice site in the Bcl-x minigene (Fig. S1 B, available at ). Under these conditions, the localization of GFP-Sam68 was also not affected (Fig. S1 A). To determine whether constitutive activation of Erk1/2 signaling affected Sam68-mediated Bcl-x alternative splicing, HEK293 cells were cotransfected with MEKK1 and suboptimal concentration of GFP-Sam68 (Fig. S1 D). Under these conditions, Erk1/2 was activated by MEKK1 (Fig. S1 C, left). Interestingly, activation of Erk1/2 signaling caused a partial redistribution of GFP-Sam68 in the cytoplasm (Fig. S1 A). Analysis of immunoprecipitated GFP or GFP-Sam68 by Western blot with phosphospecific antibodies indicated that MEKK1 increased phosphorylation of Sam68 at serines, whereas threonine phosphorylation was already detected under basal conditions and only mildly affected by MEKK1 (Fig. S1 C, right). Sam68-mediated induction of Bcl-x(s) alternative splicing was only marginally affected in the presence of activated Erk1/2 (2.59 vs. 2.04 in the x(s)/x(L) ratio; Fig. S1 D, left). To rule out the possibility that the basal threonine phosphorylation of Sam68 was sufficient to modulate its activity, a similar experiment was performed using wild-type myc-Sam68 or -Sam68m1, a mutated version lacking the eight Erk1/2 phosphorylation sites of Sam68 (). As illustrated in Fig. S1 D, we observed that myc-Sam68m1 was as efficient as the wild-type protein to favor the Bcl-x(s) 5′ splice site selection. These results indicate that, different from what was observed with CD44, Erk1/2 signaling does not affect Sam68-mediated Bcl-x alternative splicing. Sam68 was originally identified as a substrate of Src in mitosis (; ), and tyrosine phosphorylation of Sam68 by the Src-like kinase Fyn decreases its affinity for synthetic homopolymeric RNA in vitro (). Hence, we asked whether tyrosine phosphorylation affected the splicing activity of Sam68 toward a cellular target. Coexpression of wild-type Fyn counteracted the ability of GFP-Sam68 to favor the Bcl-x(s) 5′ splice site selection, resulting in an isoform ratio similar to the GFP control (). Remarkably, the constitutively active mutant Fyn (), which induces a stronger tyrosine phosphorylation of Sam68 (), completely reverted the effect and induced predominant selection of the Bcl-x(L) 5′ splice site in a Sam68-dependent manner (, bar graph). Both Fyn and Fyn caused the relocalization of Sam68 from a diffuse distribution in the nucleoplasm to discrete subnuclear foci (). No effect of Fyn on Bcl-x splicing were observed when Sam68 was not overexpressed. Because the endogenous Sam68 was not strongly tyrosine phosphorylated under these conditions (20 h after transfection), we performed splicing assays after prolonged Fyn expression. Remarkably, tyrosine phosphorylation of the endogenous Sam68 was strongly increased 48 h after transfection of Fyn (). In these cells, the effects of Fyn on Bcl-x minigene splicing were similar to those obtained with overexpressed Sam68 (), indicating that tyrosine phosphorylation of endogenous Sam68 affects Bcl-x splicing. Moreover, a similar regulation of Bcl-x isoform ratio was also observed with endogenous mRNAs (Fig. S2, available at ). Next, we asked whether tyrosine phosphorylation by Fyn also affected the binding of Sam68 to the endogenous Bcl-x mRNA. Bcl-x mRNA was coimmunoprecipitated with GFP-Sam68 but not with GFP (). Coexpression of Fyn abolished binding of Bcl-x mRNA to GFP-Sam68 (), indicating that tyrosine phosphorylation reduces the affinity of Sam68 for this cellular target in live cells. A specific mutation (V276F) in the GSG domain of the Sam68 homologue GLD-1 () affects germ cell apoptosis after DNA damage in (). Because this valine residue (V229) is conserved in human Sam68, we generated the homologous mutant allele (GFP-Sam68) and tested its activity toward Bcl-x splicing. First, the RNA-binding activity was tested in vitro by pull-down assays with synthetic homopolymeric RNA. Both wild-type and mutant Sam68 bound to polyU-Sepharose beads with approximately the same affinity (see ). By contrast, Sam68 was defective in binding to polyA–Sepharose beads (see ), indicating that this mutation affects the RNA-binding specificity of the protein. To test the effect on splicing, wild-type and mutant GFP-Sam68 were coexpressed with the Bcl-x minigene. Remarkably, Sam68 was completely unable to favor the selection of the Bcl-x(s) 5′ splice site. Rather, its expression enhanced the selection of the Bcl-x(L) 5′ splice site (). Moreover, unlike the wild-type protein, Sam68 was unable to favor the expression of the endogenous Bcl-x(s) mRNA (). Densitometric analysis () and real-time PCR quantification () of RNA coimmunoprecipitation experiments showed that Sam68 was partially defective in binding to endogenous Bcl-x mRNA. Interestingly, we also observed that Sam68 localized to discrete nuclear foci and was not diffused in the nucleoplasm like wild-type Sam68 (). Confocal microscopy showed that these foci were different from the speckles where splicing factors like SC35 and ASF/SF2 accumulate (Fig. S3, available at ). On the other hand, the localization of Sam68 was similar to that observed when wild-type Sam68 was coexpressed with activated Fyn. Because in both conditions the Bcl-x(L) 5′ splice site was favored, these results suggest an inverse correlation between splice site modulation by Sam68 and its localization to subnuclear foci. Importantly, Sam68, in which two arginine residues in the nuclear localization signal () were substituted (R436A/R442A), did not affect splicing of Bcl-x (compared with GFP alone), indicating that nuclear localization of Sam68 is required to modulate splice site selection of Bcl-x pre-mRNA (). Next, the correlation between Sam68 activity toward Bcl-x splicing and induction of apoptosis was investigated. Similar to what is reported in NIH3T3 cells (), up-regulation of Sam68 induced apoptosis in HEK293 cells, as determined by annexin V binding (), nuclear fragmentation (), and cleavage of caspase 3 (). Induction of apoptosis was similar to that elicited by transfection of the Bcl-x(L) antisense oligonucleotide (), which affects the Bcl-x(s)/Bcl-x(L) ratio similarly to Sam68 (). On the other hand, Sam68 and Sam68, which did not induce Bcl-x(s), were unable to trigger apoptosis in the same experimental setting. These results indicate that modulation of Bcl-x alternative splicing by Sam68 strongly correlates with the ability of the protein to elicit an apoptotic response. Our results demonstrate that Sam68 affects the choice of Bcl-x splice sites in live cells. Normally, the Bcl-x(L) isoform is expressed predominantly. Sam68 could either antagonize or favor the recruitment of SR proteins in the vicinity of the Bcl-x splice sites. To test these hypotheses, we attempted to shift the balance between Sam68 and SR proteins. Our previous results already showed that silencing Sam68 by RNAi increased Bcl-x(L) splicing (). Next, we checked whether up-regulation of ASF/SF2, a prototypical SR protein that is essential for cell survival (), exerted the same effect. Indeed, coexpression of ASF/SF2 induces virtually exclusive usage of the Bcl-x(L) splice site in the reporter minigene. When GFP-Sam68 is also expressed, ASF/SF2 prevails but the Bcl-x(L) is no longer exclusively selected. Up-regulation of ASF/SF2 did not affect the diffuse localization of GFP-Sam68 in HEK293 cells (). This result suggests that Sam68 does not function by recruiting SR proteins to the Bcl-x(s) splice site. If this is true, forced recruitment of a RS domain by Sam68 should not improve its splicing activity. To confirm this, we artificially fused the RS domain of ASF/SF2 to GFP-Sam68. The resulting fusion protein retains its RNA-binding activity, as measured by polyA and polyU pull-down assays (), indicating that its RNA-binding motif is not aberrantly folded. Nevertheless, the in vivo splicing assay indicated that the RS domain neutralized, rather than increased, the splice site modulating activity of Sam68 (). Interestingly, GFP-Sam68-RS was only partially diffused in the nucleoplasm, whereas a portion of the protein accumulated in foci similar to those observed with GFP-Sam68 () but different from SC35 and ASF/SF2 speckles (Fig. S3). Next, we investigated the effect of the coexpression of Sam68 and ASF/SF2 on apoptosis. Remarkably, we found again a close correlation between Bcl-x splicing and apoptosis. ASF/SF2 up-regulation almost completely suppressed the number of annexin V–positive cells () and nuclear fragmentation (). In line with their effect on Bcl-x splicing, expression of GFP-Sam68-RS or coexpression of ASF/SF2 with Sam68 strongly reduced the number of apoptotic cells. Alternative splicing is regulated by the concerted action of several splicing regulators (). To investigate whether Sam68 interacted with other splicing factors in HEK293 cells, the endogenous protein was immunoprecipitated, and the presence of associated factors was tested by Western blot analysis. As shown in , endogenous Sam68 did not interact with SAP155 and hnRNP F/H, two splicing regulators that were recently reported to affect Bcl-x alternative splicing (; ), or with ASF/SF2. By contrast, we found that Sam68 specifically interacted with hnRNP A1, a splicing regulator known to antagonize the function of ASF/SF2 (). A similar interaction with hnRNP A1 was also observed with transfected Sam68 (). Confocal microscopy showed that GFP-Sam68 and hnRNP A1 were both diffused in the nucleoplasm and that they partially colocalized in some nuclear foci (, arrows). On the other hand, neither colocalization () nor interaction () was observed between ASF/SF2 and Sam68. The association between Sam68 and hnRNP A1 was only slightly affected by RNase treatment (). To confirm a protein–protein interaction between them, we performed pull-down assays. Bacterially purified GST fusion proteins containing different regions of Sam68 were incubated in vitro with HEK293 nuclear extracts. As shown in , hnRNP A1 strongly bound to the C-terminal region of Sam68 (276–443), which does not contain the RNA-binding domain. The minimal region required for binding was mapped to the last 93 amino acids of Sam68 (351–443). On the other hand, the N-terminal region of Sam68 (1–277), containing the whole GSG domain, weakly bound to hnRNP A1, suggesting that weak interaction through a common RNA could also occur. No binding to GST alone was observed. These results indicate that Sam68 and hnRNP A1 can form a protein–protein interaction. To test whether association between Sam68 and hnRNP A1 was inhibited by conditions in which Bcl-x(L) splicing was favored, we tested the coimmunoprecipitation in the presence of Fyn or with GFP-Sam68 mutant proteins. Remarkably, we found that neither GFP-Sam68 nor GFP-Sam68-RS was able to associate () or colocalize (Fig. S3) with hnRNP A1 and that this interaction was strongly reduced when wild-type Sam68 was phosphorylated on tyrosine by coexpression of Fyn (). To determine whether hnRNP A1 expression influences the Sam68 effects, we depleted hnRNP A1 by RNAi. HEK293 was transfected with hnRNP A1 siRNAs or scrambled controls 24 h before transfection with the Bcl-x minigene and GFP or GFP-Sam68. We observed that depletion of hnRNP A1 attenuated Sam68-induced Bcl-x(s) splicing (). Remarkably, annexin V staining indicated that depletion of hnRNP A1 also reduced the number of apoptotic cells elicited by Sam68 transfection (). These results suggest that hnRNP A1 and Sam68 cooperate to modulate the alternative splicing of Bcl-x and apoptosis in live cells (). The RNA-binding protein Sam68 has recently been proposed to regulate apoptosis (; ). However, no direct evidence on the mRNAs regulated by Sam68 during induction of apoptosis or on the mechanisms involved in this process has been provided. The results presented herein indicate that regulation of apoptosis by Sam68 involves changes in alternative splicing of its cellular target, Bcl-x. We provide evidence that Sam68 binds to endogenous Bcl-x mRNA and that fluctuations in the intracellular levels of this RNA-binding protein affect the ratio between the antiapoptotic Bcl-x(L) and the proapoptotic Bcl-x(s) mRNA. Finally, by using multiple experimental approaches, we show that posttranslational modifications and point mutations affect both the splicing activity of Sam68 toward Bcl-x and induction of apoptosis in live cells. Our work provides the first cellular target of Sam68 involved in apoptosis and suggests a mechanism of action for this RNA-binding protein in response to an altered intracellular environment. The cellular targets of Sam68 potentially involved in apoptosis were identified by coimmunoprecipitation experiments. Using this approach, we have recently identified additional mRNAs that are targets of Sam68 in mouse male germ cells (). The specificity of the binding was assessed by a parallel immunoprecipitation with preimmune IgGs and by confirmation of the targets identified in pull-down assays using purified proteins and RNAs. More important, we show that fluctuations in Sam68 intracellular levels achieved by RNAi or transient transfection profoundly affect alternative splicing of one of these targets: Bcl-x. An increase of Sam68 shifts the balance toward the proapoptotic Bcl-x(s) isoform, and there is a correlation between Sam68-induced Bcl-x(s) splicing and apoptosis. However, as Sam68 binds the mRNAs for other regulators of apoptosis, it is possible that its effects on cell death also involve the regulation of additional targets. STAR proteins are thought to link signal transduction pathways to mRNA processing (). Indeed, it was recently shown that ser/thr phosphorylation of Sam68 by Erk1/2 affected exon v5 inclusion in the CD44 mRNA (). Surprisingly, we found that Erk1/2 signaling does not strongly affect Sam68-mediated alternative splicing of Bcl-x, suggesting that posttranslational modifications of this RNA-binding protein may differentially regulate its mRNA targets. By contrast, we found that tyrosine phosphorylation of Sam68 by the Src-like kinase Fyn reverts the ratio of Bcl-x(s)/Bcl-x(L) induced by Sam68 and promote the expression of the antiapoptotic Bcl-x(L) mRNA. The same effect was also achieved by creating the V229F substitution or by fusing an RS domain to Sam68. Intriguingly, in all these cases, the localization of Sam68 changed from being diffuse in the nucleoplasm to accumulating in discrete subnuclear foci that are different from the speckles where SC35 and ASF/SF2 accumulate. These observations suggest that Sam68 is in a dynamic equilibrium in the nucleus and that posttranslational modifications affect both the localization and the activity of the protein, resulting in alternative processing of its pre-mRNA targets. The activity of Src-like kinases is often up-regulated in cancer cells and contributes to cell proliferation, survival, and invasiveness (). Src activity is up-regulated in human prostate carcinomas at advanced stages, and it correlates with tyrosine phosphorylation of Sam68 (). A similar observation was made in breast cancer cells (), suggesting that it is a common mechanism of control of this protein in neoplastic cells. Our results provide a possible explanation for these observations. Tyrosine phosphorylation of Sam68 in cancer cells may protect them from apoptosis by altering the Bcl-x(s)/Bcl-x(L) ratio in favor of Bcl-x(L) (). In line with this hypothesis, it has been shown that Bcl-x(L) levels are increased in more aggressive prostate cancer cells and that their treatment with synthetic oligonucleotides that promote the Bcl-x(s) splice site selection in the pre-mRNA triggers apoptosis (). It is possible that Sam68 participates in the mechanisms that render prostate cancer cells more resistant to apoptosis and that treatments affecting tyrosine phosphorylation of Sam68 can influence survival of cancer cells expressing high levels of Bcl-x(L). The experiments presented herein demonstrate for the first time that Sam68 can interact with hnRNP A1. Interestingly, we found that tyrosine phosphorylation by Fyn, or mutations like V229F and fusion to a RS domain, disrupted this complex and affected Bcl-x(s) splicing. Moreover, depletion of hnRNP A1 by RNAi strongly attenuated Sam68-induced Bcl-x(s) splicing and apoptosis, suggesting that the interaction between Sam68 and hnRNP A1 is functionally relevant. The interaction between Sam68 and hnRNP A1 is only in part mediated by RNA, suggesting that these proteins could bind cooperatively to pre-mRNAs. Thus, hnRNP A1 might be recruited to certain pre-mRNA regions by the presence of Sam68 in the vicinity. Although future mechanistic experiments are required to define this functional interaction, we hypothesize that Sam68 may affect Bcl-x splicing by attracting hnRNP A1, which is known to compete with ASF/SF2 and to cause switches in 5′ splice sites (). Generally, there are rather few proteins that influence splice site selection, compared with the number of regulated alternative splice events. It has been postulated that the choice of splice sites is governed by combinations of splice factors, rather than individual proteins (; ), but practical examples of interactions between splice factors remain few. Thus, the interaction of Sam68, a tightly regulated RNA-binding protein, with an abundant factor that has a function in splice site selection, may well become a paradigm in regulated alternative splicing. In conclusion, the experiments presented herein demonstrate that Sam68 affects the alternative splicing of Bcl-x pre-mRNA and suggest that perturbations of intracellular signaling pathways affecting its tyrosine phosphorylation status can finely tune the splicing activity of Sam68 and predispose the cell to survive or to undergo programmed cell death. The Bcl-x minigene has been described previously (). The cDNA of human Sam68 was subcloned from pcDNA3-Sam68 () into EcoRI–SalI restriction site of pEGFP. Site-directed mutations were inserted by PCR using oligonucleotides containing the mutated residue (Fig. S4, available at ). Wild-type and mutated Sam68 cDNAs were also subcloned into pCDNA3-myc eukaryotic expression vector. The human ASF/SF2 cDNA (available from GenBank/EMBL/DDBJ under accession no. ) was amplified by RT-PCR using Proofstart polymerase (Pfu; Stratagene) and HEK293 RNA. The cDNA was subcloned into SalI–XbaI restriction sites of p3XFLAG (Sigma-Aldrich). GFP-Sam68-RS was generated by fusing the RS domain of ASF/SF2 (aa 194–248) to the C terminus of Sam68. The cDNA encoding the RS domain of ASF/SF2 was amplified by PCR and subcloned into SalI–BamHI restriction sites of pEGFP in fusion with Sam68 upstream of the TAA codon. All cDNAs used in the experiments were sequenced by Cycle Sequencing (BMR Genomics). Expression vector for Sam68m1, containing the phosphorylation mutated sites, was provided by H. Konig (Institut für Toxikologie und Gnetik, Karlsruhe, Germany). HEK293 cells were maintained in DME (Invitrogen) supplemented with 10% FBS (BioWhittaker Cambrex Bioscience), penicillin, and streptomycin. For transfections, HEK293 cells were plated in 35-mm dishes 1 d before and transfected with 1 μg of DNA (Bcl-x minigene, pEGFP-Sam68, pEGFP-Sam68, pEGFP-Sam68, pEGFP-Sam68RS, pFLAG-Asf/SF2, pCMV5-Fyn, pCMV5-Fyn, pCDNA3-MEKK1, pCDNA3-myc-Sam68, pCDNA3-myc-Sam68) using lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. At 24 h after transfections, cells were collected for RNA or biochemical analysis. For RNAi, cells at ∼50/60% confluence were transfected with siRNAs (MWG Biotech) using Oligofectamine and Opti-MEM medium (Invitrogen). Transfections were performed for 2 or 3 consecutive days. Sequences for Sam68 and hnRNP A1 siRNAs are listed in Fig. S4. Transfection of antisense Bcl-x(L) oligonucleotide (Calbiochem) was performed as for RNAi using Oligofectamine. For protein extraction, HEK293 cells were resuspended in lysis buffer (100 mM NaCl, 10 mM MgCl, 30 mM Tris-HCl, pH 7.5, 1 mM dithiothreitol, 10 mM β-glycerophosphate, 0.5 mM NaVO, and protease inhibitor cocktail [Sigma-Aldrich]) supplemented with 0.5% Triton X-100. The extracts were centrifuged for 10 min at 12,000 at 4°C, and the supernatants were collected and used for Western blot or immunoprecipitation experiments. HEK293 cells were homogenized in lysis buffer (100 mM NaCl, 10 mM MgCl, 30 mM Tris-HCl, 1 mM DTT, protease inhibitor cocktail, and 40 U/ml RNase OUT [Invitrogen]) supplemented with 0.5% Triton X-100. Soluble extracts were separated by centrifugation at 10,000 for 10 min, and they were precleared for 2 h on protein A–Sepharose beads (Sigma-Aldrich) in the presence of 2 μg rabbit IgGs, 0.05%BSA, and 0.1 μg/ml yeast tRNA. After centrifugation for 1 min at 1,000 , supernatants were incubated with 2 μg anti-Sam68 (Santa Cruz Biotechnology, Inc.), anti-GFP (Roche), or rabbit IgGs for 3 h at 4°C under constant rotation. Beads were washed three times with lysis buffer, and an aliquot was eluted in SDS sample buffer for Western blot analysis. The remaining beads were incubated with lysis buffer in the presence of (RNase-free) DNase (Roche) for 15 min at 37°C and washed three times with lysis buffer before incubation with 50 μg proteinase K (Roche) for an additional 15 min at 37°C. Coprecipitated RNA was then extracted by standard procedure and used for RT-PCR using BclX-1 and rtBclX-2 primers (Fig. S4). For the coimmunoprecipitation experiment with hnRNP A1, nuclear extracts were prepared by resuspending cells in ipotonic buffer (10 mM Tris/HCl, pH 7.4, 10 mM NaCl, 2.5 mM MgCl, 1 mM DTT, protease inhibitor cocktail, 30 U/ml RNase OUT, 10 mM β-glycerophosphate, and 0.5 mM NaVO). After incubation on ice for 7 min, samples were centrifuged at 700 for 7 min. Pelleted nuclei were resuspended in ipotonic buffer supplemented with 90 mM NaCl and 0.5% Triton X-100, sonicated, and stratified on 30% sucrose. After a centrifugation at 5,000 for 15 min, nuclear extracts were precleared and immunoprecipitated as described. Where indicated, nuclear extracts were treated with 100 μg/ml RNase A (Sigma-Aldrich). For protein–protein interactions, GST and GST-Sam68 were purified from as previously described () and incubated for 1 h with nuclear extracts. For RNA–protein interactions, 2 μg of purified GST proteins were equilibrated for 1 h in 50 mM Tris-HCl, pH 7.4, 100 mM KCl, 2 mM MgCl, 1 mM EDTA, 1 mM DTT, 40 U/ml RNase OUT, and 0.2% Nonidet-P40 supplemented with 0.05% BSA and 0.1 μg/ml yeast tRNA. Purified total RNA from HEK293 was added to the beads and incubated at 4°C under constant rotation. Beads were washed and RNA was extracted as described in the previous paragraph. 1 μg of RNA from HEK293 transfected cells or all of the coimmunoprecipitated RNA was used for RT-PCR using M-MLV reverse transcriptase (Invitrogen) according to manufacturer's instructions. 10% of the reverse-transcription reaction was used as template together with the following primers: endogenous Bcl-x, Bcl-X-1 (forward) and Bcl-X-2 (reverse); real-time PCR, rt-BclX-1 (forward) and rt-BclX-2 (reverse); Bcl-x minigene, mg-BclX-1 (forward) and mg-BclX-2 (reverse). All primer sequences are listed in Fig. S4. Real-time PCR was performed using the iQ Sybr-green Supermix (Bio-Rad Laboratories) according to manufacturer's instructions. Cell extracts or immunoprecipitated proteins were diluted in SDS sample buffer and boiled for 5 min. Proteins were separated on 10% SDS-PAGE gels and transferred to Hybond-P membranes (GE Healthcare) as previously described (). The following primary antibodies (1:1,000 dilution) were used (overnight at 4°C): rabbit anti-Sam68, rabbit anti-Erk2, rabbit anti-Fyn, and mouse anti-phosphotyrosine (PY20; Santa Cruz Biotechnology, Inc.); mouse anti-ASF/SF2 (US Biological); mouse anti–hnRNP A1, mouse anti-tubulin, and rabbit anti-actin (Sigma-Aldrich); mouse anti–hnRNP H/F (Abcam); rabbit anti–phospho-ERKs (Cell Signaling); rabbit anti-GFP (Roche); rabbit anti-phosphoserine and anti-phosphothreonine (Stressgen); and rabbit anti–Bcl-x (BD Biosciences). Secondary anti-mouse or anti-rabbit IgGs conjugated to horseradish peroxidase (GE Healthcare) were incubated with the membranes for 1 h at room temperature at a 1:10,000 dilution in PBS containing 0.1% Tween 20. Immunostained bands were detected by chemiluminescent method (Santa Cruz Biotechnology, Inc.). Transfected cells grown on 35-mm plates were harvested and processed for annexin V staining or for Western blot analysis. For annexin V staining, cells were washed in PBS and stained with the annexin V–PE (BD Biosciences) according to the manufacturer's instructions. GFP-positive cells were then analyzed with a FACSCalibur Flow Cytometer (Becton Dickinson). Transfected HEK293 cells were fixed in 4% paraformaldehyde and washed three times with PBS. Cells were permeabilized with 0.1% Triton X-100 for 7 min and incubated for 1 h in 0.5% BSA. Cells were washed three times with PBS and incubated for 2 h at room temperature with antibodies against cleaved casp3 (1:400; Sigma-Aldrich), SC35 (1:200; Sigma-Aldrich), Asf /SF2 (1:100), or hnRNP A1 (1:400), followed by 1 h of incubation with cy3-conjugated anti-mouse IgGs (Chemicon). After washes, slides were mounted with MOWIOL reagent (Calbiochem) and analyzed by confocal microscopy using an inverted microscope (Carl Zeiss MicroImaging, Inc.). Images in , , , and Fig. S1 A were taken from an inverted microscope (IX70; Olympus) using an LCA ch 20×/0.40 objective. Images in were taken from a microscope (Axioskop; Carl Zeiss MicroImaging, Inc.) using a Pan-Neofluar 40×/0.75 objective. Images were acquired at room temperature using a RT-slider camera (Diagnostic Instruments) and the IAS2000 software (Biosistem82; Delta Sistemi). The confocal images in and Fig. S3 were taken from a confocal microscope (LSM510; Carl Zeiss MicroImaging, Inc.) using a Plan-Neofluar 40×/1.3 oil differential interference contrast objective and the LSM510 software (Carl Zeiss MicroImaging, Inc.). Images were acquired as TIFF files, and Photoshop and Illustrator (Adobe) were used for composing the panels. Fig. S1 shows that the ERK1/2 pathway does not regulate Sam68-mediated alternative splicing of Bcl-x. Fig. S2 shows the effects of overexpressed Fyn on the alternative splicing of endogenous Bcl-x transcripts. Fig. S3 shows the localization of Sam68 mutants and splicing regulators in HEK293 cells. Fig. S4 shows a list of the oligonucleotides used for the PCR reactions in the article. Online supplemental material is available at .

The ER is a three-dimensional membranous network that extends throughout the cell, and is organized into discrete domains: the nuclear envelope, the smooth ER, and the rough ER. The ER participates in a variety of cellular functions, such as protein and lipid synthesis, the regulation of intracellular calcium levels, degradation of glycogen, and detoxification reactions (). The fine structure of the ER in any given cell may be organized depending on which of these functions predominates. For example, during myogenesis, the ER differentiates into the sarcoplasmic reticulum, which is an ER specialized in Ca regulation (). Proteins and lipids synthesized in the ER are transported to their destinations by the secretory pathway, and different cells use the secretory pathway to varying degrees. In addition, the amount of exocytic vesicular traffic may vary in a cell during development. oogenesis is a good model system to study how cells prepare for the varying requirements of vesicular transport activity. The oocyte is interconnected with 15 nurse cells by ring canals (; ), and cytoplasm transported from nurse cells to the oocyte constitutes a major contribution to oocyte growth (). Another source for oocyte growth is yolk taken up during vitellogenesis (stages 8–10), when the oocyte becomes highly endocytic. During vitellogenesis, the rate of oocyte growth overtakes the rate of nurse cell growth such that, ultimately, the oocyte volume equals the total volume of the 15 nurse cells. In addition, a high density of microvilli forms on the oocyte surface during these stages (). Therefore, a large amount of membrane must be added to the oocyte plasma membrane during vitellogenesis. However, mechanisms for increasing exocytic membrane trafficking is poorly understood. In this study, we identified a novel gene, , which is required for oocyte and bristle growth. The gene encodes a novel conserved ER membrane protein. We show that Jagunal is required for reorganizing the ER through ER clustering in the oocyte during vitellogenesis. The failure to reorganize the ER in mutant oocytes results in reduced vesicular traffic and slowed cell growth. We propose that Jagunal is involved in reorganizing the ER in cells that must increase exocytic membrane traffic during development. After the completion of vitellogenesis, the oocyte doubles its volume within 30 min to reach its maximum size (). The final growth of the oocyte depends on rapid cytoplasm transport from nurse cells. Mutations that affect the ability of nurse cells to transport cytoplasm result in the production of small eggs (); this phenotype has been referred to as “dumpless.” To identify essential genes involved in oocyte growth, we examined a collection of lethal mutations on chromosome 3R that were made () and preselected for a small egg phenotype in Ruth Lehmann's laboratory at New York University (New York, NY). We found a mutation that severely affects oocyte growth and named it (), which means small egg in Korean. Germline clones (GLCs) of ) showed a fully penetrant dumpless phenotype with severe defects in oocyte integrity. Four additional alleles were isolated from a noncomplementation screen, three of which were lethal and one () that was semilethal (). Homozygous mutant , , and animals died during the first and second instar larval stages. GLCs of and showed a dumpless phenotype. mutant oocytes appeared normal until stage 9. During stage 10, the anterior region of the oocyte began to detach from nurse cells and follicle cells (, arrow). We examined the phenotype in live egg chambers using a protein trap line (G413) that produces a GFP fusion with Basigin (Bsg) and labels plasma membranes (; ). In time-lapse images, Bsg-GFP allowed us to view both rapid oocyte growth caused by intercellular cytoplasm transport from nurse cells, and ooplasm streaming in the oocyte during stage 11 (Video 1, available at ). As observed in fixed egg chambers, live mutant oocytes expressing Bsg-GFP detached from other cells at the anterior region (, arrow). In severely affected mutant egg chambers, the oocyte detached from nurse cells completely and the cytoplasm of nurse cells leaked into the space between nurse cells and the oocyte (, arrow; and Video 2). In more mildly affected egg chambers, the oocyte was connected to nurse cells, and intercellular cytoplasm transport proceeded to some extent during stage 11 (Video 3). However, the oocyte failed to grow to its maximum size. GLC phenotypes of were less severe in younger flies, likely because of the perdurance of wild-type gene products expressed before the formation of mutant clones. Therefore, we focused on the phenotypes in egg chambers from females at least 7 d old. To determine whether the growth defect caused by mutations is oocyte specific, we examined the growth of mutant bristles by making somatic cell clones. mutant bristles were thinner and shorter than wild-type bristles. This defect was observed in several bristle types: wing anterior margin bristles (), microchaetes (), and macrochaetes (not depicted). mutant bristles also exhibited a defect in their surface structure. During the development of bristles, the plasma membrane forms bulges between the 8–12 actin bundles associated with the plasma membrane, and the secretion of cuticle produces a ridged structure (). Wild-type bristles contain several parallel ridges with deep valleys between them, whereas mutant bristles have a weak ridge structure with shallow valleys (, inset). The mutation mapped to the chromosome interval 83C1- 83D4 as delineated by the proximal breakpoint of the noncomplementing deficiency and the proximal breakpoint of the complementing deficiency . We mapped the mutation further by single-nucleotide polymorphism mapping to within a 70-kb region containing 20 genes. By sequencing open reading frames of genes in the 70-kb interval, we identified mutation sites in all alleles within a predicted gene, (). The viability and fertility defects associated with all alleles were rescued by ubiquitous expression of a cDNA, confirming that is . () encodes a protein of 197 amino acids with predicted homologues in the genomes of human, mouse, zebrafish, mosquito, and (). Jagunal shows 31% identity and 51% similarity to the human homologue. The first 40 amino acids of Jagunal are highly conserved, and a semilethal allele, , is a missense mutation at a conserved Asp residue in the N-terminal region (, asterisk), suggesting that the N-terminal region has an important function. The protein has four predicted transmembrane domains and a putative dilysine motif at the C-terminal end (, boxed area). The dilysine motif is known to be an ER-retention motif for ER membrane proteins (). Jagunal homologues in human and zebrafish match the consensus of the motif (KKXX), whereas arginine replaces lysine in Jagunal (RKXX). However, the C-terminal sequence is likely to act as an ER retention motif; some substitutions of lysine by arginine are permitted in the dilysine motif (), and interference with this sequence in Jagunal alters its localization (see the following section). To examine the subcellular localization of Jagunal, we made an antibody against Jagunal. Unfortunately, the antibody did not detect endogenous Jagunal proteins in ovaries by Western blotting or immunofluorescence. However, overexpressed Jagunal was detected (), and because ectopic expression of Jagunal rescued the lethality of mutations, the localization of ectopically expressed Jagunal likely reflects that of endogenous Jagunal. In Western blot analysis, overexpressed Jagunal migrated slightly faster than the expected molecular weight of 23 kD on a denaturing gel (, arrow); several background bands were also detected. In immunofluorescence, a strong signal was detected in cells overexpressing Jagunal (). Jagunal colocalized with an ER marker, EYFP-ER (EYFP fused to the KDEL ER retention sequence; ; ). Moreover, the distribution of Jagunal showed a characteristic ER pattern during early stages, where it was enriched at the nuclear envelope (). Jagunal began to be enriched in the oocyte during stages 2–3, where it was uniformly distributed in the cytoplasm (). During stages 9 and 10, Jagunal became concentrated in the oocyte subcortex adjacent to follicle cells (). Double labeling revealed that Jagunal was on the cytoplasmic side of cortical actin filaments in the oocyte (, arrow). Jagn-GFP and -Venus fusion proteins provided evidence for a functional ER retention motif. We made Jagn-GFP and -Venus transgenes by using a genomic DNA construct with GFP fused in-frame to the C-terminal end of Jagunal and with Venus fused in-frame between the first and second transmembrane domains, respectively (). Because the expression level of these proteins in ovaries was low, we used a GFP antibody to detect them. Jagn-Venus showed an ER distribution and was enriched in the oocyte subcortex (), whereas Jagn-GFP localized at the plasma membrane, including ring canals and the oocyte plasma membrane (, arrow and arrowhead), in addition to the apparent ER localization. The aberrant localization of Jagn-GFP at the plasma membrane was also found in follicle cells during stage 10, when vesicular traffic occurs actively (). In contrast, Jagn-Venus did not localize at the plasma membrane of follicle cells (). The aberrant localization of Jagn-GFP at the plasma membrane likely results from loss of ER retention activity because GFP blocks the dilysine motif that must occupy the extreme C-terminal position to be functional (). In addition, the Jagn-Venus transgene, but not the Jagn-GFP transgene, rescued the lethality of alleles, further supporting the role of the dilysine motif in controlling the localization of Jagunal. The fact that Jagunal is an ER membrane protein suggested that defects caused by mutations might be caused by a defect in the ER. Therefore, we examined the distribution of several ER proteins in wild-type and mutant oocytes. In wild-type egg chambers, EYFP-ER, which is an ER luminal protein, was evenly distributed in the oocyte during stage 8, which is the earliest vitellogenic stage (). However, beginning at stage 9, EYFP-ER became concentrated in the oocyte subcortical region (, arrow), and the enrichment continued through stage 10 (, arrow), which is similar to distribution of Jagunal (). However, the enrichment of EYFP-ER in the subcortex was greatly reduced or eliminated in GLCs (). We examined two other ER luminal proteins, protein disulfide isomerase (PDI)–GFP () and Boca. Boca is an ER chaperone and is required for correct folding of Yolkless (). Similar to EYFP-ER, these proteins also became concentrated in the oocyte subcortical region during stages 9 and 10 in wild-type egg chambers (, G and I, arrows). The enrichment of Boca and PDI-GFP was greatly reduced in mutant oocytes; instead, the proteins were uniformly distributed in ooplasm (). We also examined two ER membrane proteins, Sec61-α-GFP and Rtnl1 (reticulon-like)-GFP. Sec61-α is a component of the translocon that translocates proteins across ER membranes (), and reticulons are ubiquitous ER membrane proteins proposed to stabilize highly curved ER membrane tubules (). Similar to ER luminal proteins, GFP fluorescence showed that these ER membrane proteins became concentrated in the oocyte subcortical region in wild-type egg chambers (, K and M, arrows), and the enrichment was again greatly reduced in mutant oocytes ( liter and N). The concentration of all examined ER proteins in the oocyte subcortex suggests that the subcortical enrichment of ER proteins is not caused by uneven distribution of some ER proteins. Instead, the ER itself is likely to be concentrated in the subcortex. These results suggest that the ER is reorganized to produce a subcortical enrichment of the ER in the oocyte during vitellogenesis, and that Jagunal is required for this reorganization. During stages 9 and 10, ER proteins were concentrated into clusters in the subcortex (′, arrow) in wild-type egg chambers. However, ER clusters were not found in mutant oocyte subcortex (′). To further characterize ER organization in the oocyte during vitellogenesis, we examined ER organization in the ultrastructural level. In wild-type oocytes, the ER membranes were dispersed in the cytoplasm during stage 8 (, arrows). However, beginning at stage 9, many ER clusters were found in the subcortical region (, arrow), corresponding to ER clusters observed by confocal microscopy (′, arrow). Although some ER clusters were found far away from the oocyte cortex, the overall ER density was much lower deep inside the ooplasm (″) compared with the subcortical region (′), which is, again, consistent with the subcortical enrichment seen by confocal microscopy. The ER clusters continued to exist through stage 10 (, arrow). However, ER clusters were not found in stage 12 oocytes (, arrows), suggesting that the clusters disassembled after completion of vitellogenesis. In addition to ER clustering, ER morphology changed during vitellogenesis. The ER luminal space increased during stages 9 and 10 (), compared with stages 8 and 12 (). The swollen ER morphology might indicate high ER activity during stages 9 and 10. The formation of ER clusters was severely impaired in mutant oocytes. ER clusters were found in only 17% (3/18) of stage 9 and 10 mutant oocytes; instead, the ER remained dispersed in the cytoplasm in most mutant oocytes (, E and F, arrows). Furthermore, the diameter of the ER lumen did not expand in mutant oocytes compared with stage 9 and 10 wild-type oocytes (). These data show that Jagunal is required for the formation of ER clusters and ER morphology in the oocyte during vitellogenesis. The dramatic reorganization of the oocyte ER during stages 9 and 10 likely reflects a major increase in oocyte ER function needed for vitellogenesis and membrane expansion. To determine whether mutations affect the organization of components downstream of ER in the exocytic membrane traffic pathway, we examined exocyst and Golgi complex localization. The exocyst complex is involved in targeting secretory vesicles to the appropriate exocytic sites on the plasma membrane (), and two components of the exocyst complex, Sec5 and Sec8, localize at the oocyte plasma membrane (, A and A′; ; ). In contrast to Sec5, Sec8 does not localize at the interface of the oocyte and nurse cells (). Both Sec5 and Sec8 localized normally at the oocyte plasma membrane in GLCs (, B and B′). We used antibodies to Lava lamp, which is a golgin protein, to examine the distribution of Golgi complexes (; ). In wild-type egg chambers, Golgi complexes were enriched in the oocyte and distributed evenly throughout the ooplasm during stage 8 (), as previously reported (). However, we found that many Golgi complexes began to accumulate in the subcortical region during stage 9 (, arrowhead). The enrichment of Golgi complexes in the subcortical region became more evident during stage 10, with more Golgi near lateral plasma membranes compared with the posterior region (, arrowhead). Because enrichment of Golgi in the subcortical region was not previously reported (), we examined other Golgi markers, EYFP-Golgi () and dCOG5-GFP, which is a subunit of the COG complex (). These two proteins were also enriched in the oocyte subcortex during stages 9 and 10 (unpublished data). However, unlike ER enrichment, the distribution of Golgi was unaffected in mutant oocytes (). Although Jagunal is conserved in metazoans, yeast do not have a Jagunal homologue, indicating that Jagunal is not an essential component of secretory pathway. Indeed, GLCs of did not show any evident defect in previtellogenic stages, as do GLCs of and (; ), suggesting that membrane trafficking is not blocked in mutant cells. Instead, Jagunal might be involved in increasing vesicular traffic to the oocyte surface during vitellogenesis by reorganizing the ER. To determine whether membrane traffic to the oocyte surface is reduced in mutant oocytes during vitellogenesis, we examined the localization of Yolkless in mutant oocytes. Yolkless is the yolk receptor, and its presence at the cell cortex coincides with vitellogenesis. Yolkless is distributed uniformly in the oocyte during previtellogenic stages and begins to be transported to the oocyte surface with the onset of vitellogenesis during stage 8 (; ). Yolkless is transported to the oocyte plasma membrane by exocytosis, and it accumulates in the cortex as a result of endocytic recycling (). As oogenesis proceeds, the intensity of Yolkless at the cortex increases (); by stage 10, Yolkless is almost exclusively in the cortex (). In GLCs, the level of Yolkless in the oocyte lateral cortex was reduced (). The amount of Yolkless staining at the oocyte lateral cortex varied from near normal () to absent (). Posterior enrichment of Yolkless in mutant oocytes was most evident during stage 9 (, arrow), when 72% (39/54) of mutant oocytes showed a nonuniform cortical Yolkless distribution. Although we occasionally saw wild-type oocytes with a very weak posterior enrichment of Yolkless, the posterior enrichment of Yolkless was much stronger in mutant oocytes. The behavior of Yolkless suggests that mutations affect exocytic vesicular traffic to the oocyte lateral membrane. In addition, posterior enrichment of Yolkless in mutant oocytes suggests the existence of a Jagunal-independent transport pathway polarized to the posterior pole. For example, the posterior enrichment of Yolkless may be caused, in part, by posterior-polarized endocytic recycling requiring Rab11 (). To determine whether endocytosis of yolk is affected in mutant oocytes, we examined yolk granules in electron micrographs of stage 10 wild-type and mutant oocytes. Yolk granules were reduced in size and abundance in mutant oocytes (). We measured the area of yolk granules in several egg chambers, and found that the overall area occupied by them was reduced by ∼70% in mutant oocytes (from 29 to 9.3%; ). The amount of yolk in oocytes varied widely, from 0 to 20% of oocyte area, but did not overlap wild-type amounts of yolk. To further examine endocytosis in mutant oocytes, we quantitated the number of coated pits and vesicles in the plasma membrane and cortex of stage 10 oocytes. Compared with wild-type egg chambers (), the number of coated pits and vesicles was reduced by ∼60% in mutant oocytes (from 6.3 to 2.4 vesicles/μm; and G). Again, the amount of reduction varied, but numbers of coated pits and vesicles in oocytes did not overlap wild type. These data show that yolk endocytosis and overall endocytic activity at the oocyte surface are reduced to similar extents in mutant oocytes. Considering that Jagunal is an ER membrane protein and mutations affect ER organization during vitellogenesis, the decreased endocytosis in mutant oocytes is likely to be caused by reduced transport of factors required for endocytosis (including Yolkless and membrane) to the oocyte surface. Indeed, the surface area of stage 10 oocytes () was reduced compared with wild type (). Mutant oocytes ranged from a slight reduction in the number of microvilli () to nearly absent microvilli (). Interestingly, we observed that microvilli were nearly absent on the surface of an oocyte that was detached from neighboring follicle cells (, arrowheads). The lack of sufficient membrane reserves in the form of microvilli could explain why the mutant oocytes fail to expand during stage 11. #text All crosses and culturing of were performed using standard procedures (). Canton-S and flies were used as wild-type controls. The first allele () described in this study was from a subset of lethal mutations on chromosome 3R () selected by R. Lehmann (New York University, New York, NY) because germline clones produced small eggs. We made four more alleles of with an ethylmethane sulphonate noncomplementation-lethal screen. EYFP-ER and -Golgi flies () were obtained from the Bloomington Stock Center (Indiana University, Bloomington, IN). The dCOG5-GFP (GFP-Fws) fly stock was obtained from M. Fuller (Stanford University, Palo Alto, CA; ). Bsg-GFP (G00413), PDI-GFP (ZCL1503), Rtnl1-GFP (ZCL1569), and Sec61α-GFP (ZCL0488) flies were generated by protein trap screening (). The driver () or germline triple driver (a combination of [], [], and []) were used to induce ubiquitous or germline-specific expression, respectively, of pUASp-Jagunal in the ovary. Flies used for clonal analysis were obtained from the Bloomington Stock Center. A cDNA clone of (1.1 kb) was obtained by RT-PCR and subcloned into pUASp, creating . To make the Jagn-GFP and -Venus constructs, a genomic DNA (6 kb) was used. GFP was inserted at the C terminus, and Venus was inserted between the first and second transmembrane domains. The Jagn-GFP and -Venus constructs were subcloned into CaSpeR, creating and . All alleles were induced on a chromosome (). For clones marked by loss of (), second–third instar larvae of the genotype +/ were heat shocked for 1–2 h in a 38°C water bath on two consecutive days. Ovaries were dissected from female flies that were fed yeast paste for ∼7 d to dilute out wild-type gene products. flies, and their progeny were heat shocked as first–second instar larvae for 1–2 h in a 38°C water bath. To overexpress Jagunal in a mosaic manner, flies were crossed to ; > > flies (), and their progeny were heat shocked as second–third instar larvae for 1–2 h in a 38°C water bath. Anti-Jagunal antibody was raised against the C-terminal 17 amino acids. Purified peptide antigen, CYNLVKAWKARTATRKFQ, was obtained from the Keck Biotechnology Resource Laboratory (Yale University, New Haven, CT). The antigen was injected into a rabbit and boosted four times according to a standard protocol (Cocalico Biologicals, Inc.). Antibodies were purified with a Sulfolink kit (Pierce Chemical Co.) using the original peptide. Ovaries were dissected in IMADS () and fixed in 6% formaldehyde saturated with heptane as previously described (). To visualize F-actin, egg chambers were incubated with 1 U of rhodamine-conjugated phalloidin in PBT. The following antibodies were used: rabbit anti-Jagunal at 1:200 (this study), mouse monoclonal anti–Hts-RC at 1:10 (), rabbit anti–Lava lamp at 1:200 (from John Sisson, University of Texas, Austin, TX; ), rat anti-Yolkless at 1:200 (from Christopher Schonbaum, University of Chicago, Chicago, IL; ), mouse monoclonal anti-Sec5 at 1:200 (from Thomas Schwarz, Harvard University, Boston, MA; ), guinea pig anti-Sec8 at 1:1,000 (from Ulrich Tepass, University of Toronto, Toronto, Ontario, Canada; ), guinea pig anti-Boca at 1:500 (from Richard Mann, Columbia University, New York, NY; ), and rabbit anti-GFP at 1:500 (Torrey Pines Biolabs, Inc.). Anti-GFP antibodies were used to stain EYFP-ER, Jagn-GFP, and Jagn-Venus. Fluorescence micrographs were obtained at room temperature using an inverted microscope (Axiovert 100M; Carl Zeiss MicroImaging, Inc.) equipped with a laser scanning confocal imaging system (LSM510; Carl Zeiss MicroImaging, Inc.) and a 40×/1.2 NA water-immersion objective lens (C-Apochromat; Carl Zeiss MicroImaging, Inc.; Center for Cell and Molecular Imaging, Yale University School of Medicine, New Haven, CT). Images were processed using Photoshop (Adobe) and assembled in Illustrator (Adobe). Dissected ovaries were homogenized in SDS loading buffer and boiled for 5 min. Equal amounts of ovary extract were loaded and separated on a 12% polyacrylamide gel. Proteins were transferred to nitrocellulose, blocked with 5% milk, and probed with primary antibodies overnight at 4°C. Rabbit anti-Jagunal antibody was used at 1:2,000. Flies were put through an EtOH dehydration series (25, 50, 75, and 100%), followed by treatment with hexamethyldisilazane. The treated flies were dried in a low vacuum overnight. Flies were mounted on stubs, rotary shadowed, and viewed on an ISI SS-40 scanning electron microscope (Department of Molecular, Cellular and Developmental Biology EM Facility, Yale University). Images were recorded onto Polaroid 53 film and scanned at 300 dpi. Ovaries were dissected in IMADS (), fixed in 2% glutaraldehyde in 0.1 M sodium cacodylate buffer, pH 7.4, for 1 h. After washing with 0.1 M sodium cacodylate buffer overnight, samples were postfixed in 1% osmium tetroxide in 0.1 M sodium cacodylate buffer for 1 h. After making sections, they were stained with 2% uranyl acetate for 5 min and 1% lead citrate for 2.5 min and examined in a Tecnai 12 Biotwin electron microscope (Center for Cell and Molecular Imaging, Yale University School of Medicine, New Haven, CT). Digital images were collected with a Morada charge-coupled device camera using iTEM software (Olympus). To quantitate the percentage of yolk area in electron micrographs of oocytes, regions of ooplasm were selected and the areas of yolk granules within the selected regions was measured using ImageJ v. 1.36b (National Institutes of Health). Coated pits and vesicles were counted in selected cortical region of oocytes. The linear length of the selected oocyte surface (excluding microvillar surface) was measured using ImageJ. Ovaries expressing Bsg-GFP were dissected in mineral oil and ovarioles were spread well on the coverslip. Immediately after dissection, GFP fluorescence was visualized on a LSM 510 confocal microscope with a 20× objective. Images were collected every 20 s for 30–60 min. The images were converted to a movie using Graphic Converter (Lemkesoft). Movies are 240 times faster than real time. Wild-type egg chambers (Video 1) and GLCs (Videos 2 and 3) are presented. The online version of this article is available at .

The actin cytoskeleton is essential for numerous cellular and developmental processes involving membrane dynamics. These include endocytosis, cell migration, cytokinesis, and various morphogenetic processes. The polymerization, depolymerization, and 3D organization of actin filaments in cells are governed by vast number of actin-binding proteins. Most actin-binding proteins are composed of multiple domains, performing also regulatory and signaling functions. Among the plethora of actin-binding proteins are the actin filament nucleating Arp2/3 complex and its activators Wiskott-Aldrich syndrome protein (WASP) and WASP family verprolin homologous proteins (WAVEs), which promote formation of membrane protrusions downstream of the Rho-family GTPases (). Although most actin-dependent processes involve reshaping of cellular membranes, the direct effects of actin-binding proteins on the organization of membranes has not been reported. One central group of proteins functioning at the interface between signaling and the actin cytoskeleton are insulin receptor substrate (IRS) p53, missing-in-metastasis (MIM), and their homologues. These proteins share similar domain organization to each other, possessing a recently identified IRSp53/MIM domain (IMD) at their N terminus. In addition to the IMD, MIM and some IRSp53 isoforms possess a C-terminal WH2 domain that binds actin monomers with high affinity (; ). Although the exact functions of IRSp53 and MIM are not defined, both proteins are linked to the Arp2/3-mediated actin filament assembly and formation of plasma membrane protrusions. IRSp53 interacts with the small GTPases Cdc42 and Rac through its N-terminal region and with WAVE2 through its central SH3 domain (). IRSp53 regulates the Arp2/3-modulating activity of the WAVE2 complex and is involved in lamellipodia and filopodia formation in motile cells (; ). MIM was originally identified as a putative tumor suppressor because it is expressed in nonmetastatic, but absent from metastatic, bladder cancer cells (; ). MIM is a sonic hedgehog (Shh) responsive gene and is strongly expressed during development in muscles and postmitotic neurons and in adult mice in kidneys, liver, and Purkinje cells of the cerebellum (; ). MIM enhances Arp2/3-mediated actin polymerization through interactions with cortactin but inhibits WASP-mediated actin polymerization (). In cells, MIM and IRSp53 localize to the plasma membrane and are involved in the formation of membrane protrusions. The filopodia/membrane ruffle–inducing activity of MIM and IRSp53 resides in the N-terminal IMD, as indicated by a drastic induction of filopodia when this domain is ectopically expressed in mammalian cells (; ). Previous studies demonstrated that recombinant IMDs bind and bundle actin filaments and interact with the small GTPase Rac, providing a plausible explanation for their filopodia-forming activity (; ). The crystal structures of the IMDs from IRSp53 and MIM revealed an α-helical coiled-coil domain that self-associates into a “zeppelin-shaped” dimer (; ). Surprisingly, the closest structural homologues of IMD are the lipid-binding BAR (Bin/amphiphysin/Rvs) domains. BAR domain proteins (e.g., amphipysin, endophilin, and Rvs161/167) and related F-BAR domain proteins (e.g., toca and syndapin) induce tubular invaginations from the plasma membrane during the formation of an endocytic vesicle (). In vitro, BAR domains evaginate liposomes into narrow tubules. They interact with negatively charged lipids through patches of positively charged residues at the concave face. Membrane deformation is driven by the intrinsic curvature of the rigid “banana-shaped” BAR domain dimer (; ). In addition, an amphipathic N-terminal helix found in a subset of BAR domains (N-BARs) penetrates the membrane and potentiates the membrane-tubulating activity (; ). In contrast to BAR domains, IMDs are involved in the formation of membrane protrusions, rather than membrane invaginations, and have not been reported to deform membranes. Here, we provide evidence that IMDs deform PIP-rich membranes into tubular structures. Unlike previously characterized membrane-tubulating domains, the IMD appears to bind to the inner surface of the membrane tubule and therefore promote the formation of plasma membrane protrusions rather than invaginations. Identification of the PIP-binding interface of the IMD provided a molecular explanation for this membrane-tubulating activity and revealed how this novel function is linked to filopodia formation in cells. We also show that the filopodia-forming activity of MIM is independent of F-actin bundling and GTPase binding of the IMD. IMDs share remote structural homology with BAR domains, which bind and deform lipid membranes in vivo and in vitro (). Despite the structural similarity, IMDs were proposed to promote filopodia formation through their F-actin–bundling activity and thus form a functionally distinct group within the BAR domain family (; ; ). To examine whether IMDs display a BAR domain–like membrane-binding activity, we studied the interaction of the IMD of MIM with various lipids by native gel electrophoresis. We found that it interacts with micelles containing PIP and PIP (). The polar head group of PIP, IP, or other phospholipids did not shift the motility of MIM/IMD, suggesting the lack of a high-affinity interaction. The MIM/IMD–PIP interaction was examined further in a more physiological context with a high-speed cosedimentation assay using synthetic lipid vesicles. The IMD of MIM cosedimented with vesicles containing 30% PIP, whereas only weak cosedimentation was detected with vesicles without PIP (). A more detailed analysis revealed that already 5–10% of PIP considerably increased the affinity of MIM/IMD to vesicles (Fig. S1, available at ). This PIP density is similar to the one reported for other PIP-interacting proteins, such as N-WASP, which requires 10–15% PIP density for activation (). It is also important to note that, based on native gel electrophoresis () and cosedimentation assays (Fig. S1), the IMD of MIM binds PIP with considerably lower affinity than PIP. Sequence database searches revealed two alternative splice variants of mouse MIM's IMD. These differ from each other by a four-amino-acid insertion encoded by exon 7 (Fig. S2, A and B, available at ). These residues are located at distal ends of the dimeric IMD (Fig. S2 C). The two alternatively spliced forms of the IMD of MIM are referred as MIM/IMD-L (longer splice variant) and MIM/IMD-S (variant lacking the four residues coded by exon 7). Importantly, both MIM/IMD-L and -S, as well as the IMD of IRSp53, interacted strongly with PIP-containing vesicles, suggesting that PIP binding is a function common among all IMDs ( and Fig. S3). To investigate the possible effects of IMDs on the structure of PIP-rich membranes, we performed an EM analysis of MIM/IMD-L–PIP vesicle complexes. Surprisingly, these experiments revealed a strong membrane deforming/tubulating activity. Synthetic vesicles without MIM/IMD-L were heterogeneous in size, mostly spherical or curved in shape, and evenly distributed throughout the sample. In the presence of MIM/IMD-L, these vesicles formed clusters with complex tubular structures (). Similar structures were also predominant in samples containing the IMD of IRSp53 (unpublished data). The structural organization of the IMD-induced membrane tubules was examined in more detail by using electron tomography. Because of the complex nature of structures induced by 22 μM MIM/IMD-L (), 1.1 μM protein was used in this experiment. The tomography analysis of a selected section is displayed in and Video 1 (available at ). MIM/IMD-L–induced tubules were of regular width (measured diameter of 78 nm; SD = 7 nm; = 65; the narrowest tubule was 60 nm and the widest 93 nm) independent of the protein concentration (unpublished data). Amount of tubulation corresponded with the amount of protein, and similar tubulation was observed in assays performed either with unilamellar or multilamellar vesicles (). In contrast to the long tubular extensions induced by BAR/F-BAR domains, electron tomography revealed that in intact vesicular structures MIM/IMD-induced tubules are typically invaginating toward the interior of the vesicle ( and Video 1). This observation suggests that IMDs tubulate membranes in a direction opposite that of BAR/F-BAR domains. It is also important to note that the amount of membrane tubulation induced by MIM/IMD correlated with the PIP density of the vesicles. High amounts of membrane tubules were observed in experiments performed with 10–30% PIP density. MIM/IMD also induced membrane tubulation at 3–5% PIP, but this was less efficient than higher PIP density (Fig. S1 C). Only very weak tubulation was detected with vesicles without PIP (unpublished data). We also performed a similar assay by using two other PIP-binding proteins, α-actinin and heterodimeric capping protein. These proteins did not induce membrane tubulation under similar conditions, confirming that this activity is not a general feature of PIP-bnding proteins (unpublished data). To reveal the mechanisms of actin and phospholipid interactions of the IMD, the PIP- and actin-binding sites of MIM/IMD-L were mapped by systematic mutagenesis. 19 clusters of mutations were individually created on the surface of MIM/IMD-L using the 3D structure of the domain () as a guide. Each cluster of mutations contained one to three charged amino acids or surface-exposed hydrophobic residues substituted with alanines. The mutant clusters are evenly distributed over the surface of the IMD (, B and C; and Fig. S3 A). The PIP binding was examined by high-speed cosedimentation assay with synthetic lipid vesicles containing 30% PIP Eight mutants displayed a substantially reduced affinity for PIP ( and Fig. S3 B). Five of these (Mut10, -11, -12, -15, and -16) were clusters of charged residues that map close to the distal ends of the dimeric IMD, and two (Mut14 and -17) contained substitutions of surface-exposed hydrophobic residues. These data show that the PIP-binding site of IMD consists of relatively large positively charged regions that are located at each end of the dimeric domain. This is further supported by the observation that neutralizing the negative charge within this region (Mut13 [D143A]) increased the affinity of IMD for PIP. In addition, surface-exposed leucines located in these regions contribute to PIP binding. Actin filament cosedimentation assay performed with 1 μM IMD and 0–10 μM F-actin revealed that only four clusters of mutations (Mut11, -12, -15, and -16) displayed considerably lower affinity for F-actin as compared with wild-type IMD. Combining these mutations further reduced F-actin binding, demonstrating the importance of these residues for actin interactions ( and Fig. S3 C). These residues map close to the ends of the dimeric IMD and overlap with the PIP-binding site, revealing that the positively charged region located in helix-2 forms the main actin- and PIP-binding site of the IMD (). As with the observed increase in PIP binding, neutralization of the negative charge within this region (Mut13 [D143A]) also increased the affinity of IMD to F-actin. However, in contrast to PIP binding, none of the mutants (or double mutants) resulted in a complete lack of F-actin binding, showing that IMDs interact with F-actin mostly through nonspecific electrostatic interactions. To confirm the integrity of the mutants, we performed a urea denaturation assay for each double mutant used in the study. The mutants unfolded at the same urea concentration (4.5 M) as wild-type MIM/IMD, indicating that the mutations did not affect folding of the protein (unpublished data). Expression of IMDs in cultured mammalian cells induces a dramatic formation of filopodia (; ; ). However, the dynamics of these structures (i.e., whether they are true membrane protrusions or the result of cell retraction) have not been reported. Live-cell analysis of U2OS cells expressing MIM/IMD-L revealed that IMD-induced filopodia are highly dynamic and extend with a rate of up to 1 μm/min ( and Video 2, available at ). These protrusions are dependent on an intact actin cytoskeleton because treatment of the cells with latrunculin A decreased the dynamics of IMD-induced filopodia (Video 3). Quantification of the data revealed that ∼40% of the filopodia in untreated cells displayed elongation during the 60-s detection period, whereas after 6, 12, or 18 min of latrunculin A treatment, <5% of the filopodia displayed elongation. To examine the in vivo roles of PIP interactions of IMD, we assayed filopodia formation in U2OS cells expressing mutant IMDs. One of these (Mut12+15) displays severe defects in F-actin binding and a nearly complete lack of PIP binding () and loss of lipid vesicle tubulation activity (unpublished data). The other mutant (Mut14+17) interacts with F-actin with a similar affinity to wild-type IMD but shows a moderate defect in PIP binding ( and Fig. S3, B and C). The mutant defective in actin and PIP binding failed to induce detectable filopodia in U2OS cells. Interestingly, Mut14+17, which displays defects only in PIP binding, induced significantly less filopodia than wild-type IMD (). This demonstrates that interaction with PIP is critical for filopodia formation by IMDs. The experiments described above showed that IMDs bind and deform PIP-rich membranes in vitro and that the PIP binding appears critical for the filopodia-inducing activity in vivo. 3D analysis, derived from seven confocal planes of filopodia from cells expressing GFP-tagged MIM/IMD-L, revealed that the IMD does not localize to the F-actin bundle but instead localizes to the plasma membrane surrounding the F-actin bundle. The divergent localization was most obvious at the base of the filopodium, where the GFP–MIM/IMD-L signal followed the plasma membrane rather than the F-actin bundle that extended into the cytoplasm (, right). In , MIM/IMD that does not colocalize with F-actin is shown in green. Because it surrounds the interior of filopodia, the colocalization (yellow) and F-actin (red) signals are not visible in thin filopodia when all channels are merged. A more detailed analysis revealed that F-actin and MIM/IMD colocalized only at a thin region within the filopodia (, middle). The channel correlation in colocalized volume, taking into account all the pixels in the 3D projection, was 0.6713 (calculated by Pearson's correlation coefficient, where 1 represents perfect colocalization, 0 is no colocalization, and −1 is perfect inverse colocalization). Together, these data demonstrate that MIM/IMD does not localize to the F-actin bundle but is instead found at the plasma membrane surrounding the bundle. Colocalization channel analysis further confirmed only a very thin interface where both proteins are present. To investigate the possible cooperation between lipid- and actin-binding activities of IMDs, we examined samples containing PIP-rich vesicles, F-actin, and IMD by light microscopy. Alexa 568–labeled F-actin, MIM/IMD-L, and fluorescein phosphatidylethanolamine (PE)–labeled vesicles with or without PIP were mixed, applied on polyornithine-coated glass slides, and imaged by light microscopy. In the presence of MIM/IMD-L and PIP, clear actin dots were found to localize at positions corresponding to vesicles (). Importantly, vesicles without PIP showed significantly weaker colocalization with actin, as was also found for PIP vesicles in an experiment using Mut15 (). The intensity of actin staining on vesicles was quantified, confirming the importance of PIP binding of IMD in the formation of these denser F-actin dots (). Similar results were also obtained with the IMD of IRSp53 (unpublished data). The results demonstrate that IMDs are capable of associating actin filaments with PIP vesicles, at least in vitro. However, it is important to note that the actin and PIP-binding sites overlap on the surface of the IMD (). This suggests that a fraction of IMDs associates in this assay with membranes only through one pole, leaving the other binding site accessible to actin. Alternatively, the membrane-bound IMD may interact with actin through partially different interface than the soluble domain. IMDs were previously reported to induce filopodia formation through their actin filament–bundling activity (; ; ). However, our data showed that IMDs bind and deform membranes and do not localize to actin bundles in filopodia, as would be expected for an F-actin cross-linking protein. We thus performed low-speed sedimentation assays to compare F-actin–bundling activities of IMDs. Surprisingly, at physiological ionic conditions, MIM/IMD-L, MIM/IMD-S, and IRSp53/IMD displayed almost undetectable F-actin–bundling activity (), even though these proteins bound F-actin with affinities similar to the one previously reported for IMDs (Fig. S3 C). Under identical conditions, 1.25 μM α-actinin efficiently bundled F-actin, suggesting that at physiological salt concentration, IMDs display only very weak actin filament–bundling activity (). Because some previous F-actin bundling assays with IMDs were performed at low ionic strength () and IMDs appear to have a tendency to aggregate at low salt (), we performed a low-speed F-actin sedimentation assay at various salt concentrations. These data showed that the F-actin–bundling activity of IMDs (i.e., the amount of actin in the pellet fraction) increased sharply at low ionic strength (). We next used dynamic light scattering (DLS) to reveal whether the increased F-actin–bundling activity at subphysiological salt concentrations resulted from aggregation of MIM/IMD-L. DLS measures the fluctuation of scattered light, giving the distribution of hydrodynamic radii, ( ). Thus, DLS not only provides the information about the size (hydrodynamic radius; ) but also about the polydispersity of the sample (). DLS performed at different KCl concentrations revealed a peak corresponding to the apparent size of IMD dimer and peaks of larger radius corresponding to protein aggregates. At 100 mM KCl, the amount of aggregates was small, keeping in mind that the intensity of the peaks is proportional to the square of molecular mass and thus strongly enhances the signal from large particles. of >50 nm increased dramatically, indicative of protein aggregation (). Similar results were also obtained from a high-speed sedimentation assay, which demonstrated that MIM/IMD-L precipitated at low salt (unpublished data). Together, these data suggest that the previously reported F-actin–bundling activity of IMDs resulted from protein aggregation at nonphysiological ionic conditions. IMDs display only very weak F-actin bundling at physiological conditions that is unlikely to contribute to filopodia formation in vivo. Furthermore, our analysis revealed that MIM/IMD-L does not bind G-actin or affect the nucleotide exchange or kinetics of actin polymerization (unpublished data). We next compared the binding of MIM/IMD-L and -S to recombinant Rac by GST pull-down and surface plasmon resonance (SPR) assays. Surprisingly, these assays revealed a clear difference between MIM/IMD splice variant binding to Rac. In the GST pull-down assay, MIM/IMD-S interacted with dominant-active (V12) and -inactive (N17) Rac, as described previously (), whereas no detectable binding of MIM/IMD-L to Rac was seen in this assay (). In the SPR assay, a concentration-dependent interaction of MIM/IMD to Rac was studied. Consistent with the pull-down assays, the obtained equilibrium binding level values showed that Rac interacted with MIM/IMD-L with a binding level of only 10–15% of that of MIM/IMD-S (). A high background was obtained from the His-GST control channel, which makes the observed affinity of the MIM/IMD-L variant difficult to interpret, further proposing a very weak Rac interaction for this variant. Furthermore, our pull-down and SPR assays demonstrated also that neither variant of MIM/IMD bound other small GTPases, such as RhoA and/or Cdc42, with detectable affinity (unpublished data). Although only MIM/IMD-S binds Rac with detectable affinity, overexpression of either MIM/IMD-S or -L induced filopodia formation in cells (). Quantification of filopodia number from transfected cells revealed no apparent differences between cells expressing MIM/IMD-S and -L (). This provided evidence that, in contrast to PIP binding (), Rac binding is not necessary for filopodia formation by IMDs. IRSp53 and MIM are relatively large, multidomain proteins that regulate cytoskeletal dynamics during motile and morphogenetic processes. Expression of full-length proteins or their N-terminal IMDs, which bind Rho family GTPases and bundle actin filaments, results in dramatic filopodia/microspike formation in cultured mammalian cells (; ). In this study, we determined the mechanism by which IMDs induce membrane protrusions in cells. We show that IMDs display only very weak F-actin–bundling activity at physiological ionic conditions that is unlikely to contribute to filopodia formation. The previously reported actin-bundling activity appears to result from protein aggregation at low salt conditions. Interaction with small GTPase Rac is not required for IMD-induced filopodia formation. IMDs bind and deform PIP-rich lipid membranes in vitro and localize to the interface of plasma membrane and F-actin bundles in filopodia. IMDs interact with membranes through a similar interface to the structurally related BAR domains. However, IMDs and BAR domains generate an opposite membrane curvature because of the different geometries of their membrane-binding interfaces. Interaction with PIP-rich membranes is necessary for the filopodia-inducing activity of IMDs. Together, these findings show that IMD is a new functional member of the membrane-deforming BAR domain family. However, in contrast to previously characterized membrane-deforming domains, IMDs induce formation of membrane protrusions rather than invaginations. Here, we show that, in contrast to previous studies (; ; ), the IMD does not function as an F-actin–bundling motif at physiological ionic conditions. Although the IMD does not bundle F-actin, it binds actin filaments with a moderate affinity. Previous charge-reversal mutagenesis studies revealed that a cluster of four positively charged residues at the ends of IMD dimer plays a central role in F-actin binding and bundling (; ). Our systematic mutagenesis analysis revealed that the F-actin–binding site of the IMD of MIM is considerably larger and covers an ∼4-nm-long positively charged region along helix-2. Importantly, none of the mutations in this study resulted in a complete lack of F-actin binding. Even mutants in which five lysines/arginines were replaced by alanines bound F-actin with detectable affinity. This suggests that IMDs do not display a specific actin-binding site, but instead interact with F-actin through unspecific electrostatic interactions. In contrast, neutralizing specific residues at the actin-binding sites of other proteins such as actin-depolymerizing factor/cofilin or capping protein results in dramatic defects in actin binding (; ). The affinity of IMDs to F-actin is relatively low compared with most other F-actin–binding proteins (; ), suggesting a nonspecific interaction with actin. In filopodia, the IMD of MIM did not localize to F-actin bundles, as would be expected for an actin-bundling protein. At physiological ionic strength, this domain displayed an extremely weak actin-bundling activity compared with the well-characterized cross-linking protein, α-actinin (). Efficient bundling was, on the other hand, induced by subphysiological ionic conditions, where IMDs also form aggregates. It is important to note that although the IMD dimer contains two F-actin–binding sites, usually indicative of cross-linking activity, it appears to bind F-actin with relatively low affinity. Therefore, simultaneous interaction of both sites with F-actin, required to induce filament cross-linking, is likely to be a rare event. Alternatively, the actin filament may interact with these sites in an orientation that sterically prevents the binding of another filament. Together, these data suggest that IMDs do not bundle F-actin in cells. MIM binds and activates Rac through its IMD. The Rac-binding site is located at the ends of the IMD dimer and overlaps with the actin-binding site (). Here, we show that only the shorter splice variant of MIM's IMD interacted with Rac, whereas the longer splice variant containing a four-amino-acid insertion in the loop between helix-2 and -3 did not bind Rac or other Rho-family GTPases with a detectable affinity. In the 3D structure, the insertion in MIM/IMD-L is located at the ends of IMD dimer, providing further support that this region plays a central role in Rac binding. Importantly, both MIM/IMD-L and -S, which displayed dramatic differences in Rac binding, induced similar filopodia formation when expressed in cells. These data, together with previous studies showing that filopodia formation by IMDs was not perturbed by the coexpression of dominant-negative Cdc42 or Rac1 (), provided evidence that Rac binding is not a critical function of IMDs during the formation of membrane protrusions. However, the differences between the two MIM splice variants in Rac binding suggest another, more specific role for this interaction in the context of the full-length proteins. Our biochemical and EM analyses revealed that IMDs bind phosphatidylinositides and deform membranes into tubular structures. The observed PIP binding is in line with recent data showing that the N-terminal region of IRSp53 (containing the IMD) binds phospholipids in vitro (). IMDs did not bind IP, the polar headgroup of PIP, with detectable affinity. This result also implicates that the fatty acid chains of PIP are important for the interaction or that IMD interacts only with membranes where the inositol headgroups are aligned. The latter alternative is supported by our mutagenesis analysis, revealing that the PIP-binding site is composed of relatively large positively charged surfaces at each end of the dimer. Although the peculiar symmetry of IMDs results in an ∼140° rotation of the dimer ends to each other, the two positively charged PIP-binding patches are facing the same direction when the domain is aligned with amphiphysin BAR domain (). This model shows that IMDs bind lipids through an interface similar to that found in BAR domains. It is also important to note that membrane, Rac, and actin binding seem to take place roughly through the same area of the molecule, suggesting that these interactions compete with each other. However, of these activities, only PIP binding is necessary for the filopodia formation, suggesting that the other (competitive) interactions may have a regulatory role. The membrane-tubulation activity of IMDs is also supported by a recent study, which demonstrated that the IMD of IRSp53 can deform PIP-rich membranes in vitro (). However, this activity was reported to be Rac dependent, whereas our data demonstrates that IMD-induced membrane tubulation is independent from Rac both in vitro and in vivo. Furthermore, in contrast to our study, the IMD of MIM was not capable in deforming membranes in the study by . These differences may arise from the fact that these authors used PIP in their EM studies, whereas PIP was used in our study. Our results showed that MIM/IMD binds PIP with substantially higher affinity than PIP both at native gel electrophoresis and cosedimentation assays. Because PIP is also >25-fold more abundant at the plasma membrane than PIP (), PIP is most likely the physiological binding partner of IMDs during membrane deformation/filopodia formation in cells. It is also important to note that the membrane-binding interface of an IMD determined in our work is much more extensive than the one identified by . This may arise either from the different lipids used in these assays or from the more extensive mutagenesis approach that was applied in our study. Importantly, several lines of evidence suggest that IMDs generate a membrane curvature opposite to that of BAR domains. Our electron tomography analysis revealed that the tubular structures induced by MIM/IMD-L often penetrate toward the interior of the vesicular structure. In contrast, BAR and F-BAR domains deform liposomes into morphologically different, separate narrow tubes (; ; ). The lipid-binding interfaces of BAR and IMDs display opposite curvatures (). Interestingly, the curvature of the PIP-binding interface measured from a space-filling model of the 18-nm-long IMD dimer suggests that MIM/IMD would induce a formation of a membrane tubule with a diameter of ∼95 nm. This corresponds well with the diameter of MIM/IMD-L–induced membrane tubules measured from electron micrographs (∼80 nm). When expressed in cells, IMDs induce filopodia, as compared with plasma membrane invaginations induced by BAR domains. Thus, in contrast to the previously characterized lipid-deforming domains (BAR, F-BAR, and ENTH/ANTH), IMDs appear to generate an opposite membrane curvature. A hypothetical model for how IMDs induce filopodia formation through their membrane-tubulating activity is presented in (B and C). In addition to membrane deformation, IMDs may also cross-link F-actin to the plasma membrane. However, further studies are required to elucidate the possible biological significance of this activity. Actin dynamics are closely linked to membrane-deformation processes such as endocytosis (). Our data provide first evidence that, in addition to N-WASP–induced endocytic processes (), plasma membrane protrusions are generated through interplay between actin polymerization machinery and direct membrane deformation. IRSp53 has been implicated in lamellipodia formation via WAVE/SCAR-2 complex and shown to activate WAVE2, an Arp2/3 activator (; ; ). Localization of IRSp53 to the plasma membrane is dependent on its N-terminal IMD (; ; ; ). MIM localizes to filopodia, lamellipodia, and cell–cell junctions (unpublished data), suggesting a role in actin-dependent morphogenetic processes at the plasma membrane. Similarly, several BAR domain proteins also interact with the regulators of actin dynamics. For example, yeast Rvs167/161 complex is intimately involved in actin assembly during endocytosis and mammalian tuba binds Arp2/3 activator N-WASP (; ). Similarly, F-BAR proteins toca-1 and syndapin interact with N-WASP (; ). In conclusion, our study places the IMD in a new functional subbranch of the membrane-deforming BAR domain family. However, because of the unique geometry of its PIP-binding site, an IMD induces membrane protrusions rather than invaginations. In the future, it will be important to reveal the detailed mechanism by which IMDs interact with membranes. Because IMDs have a tendency to form multimers at low salt conditions, it will be interesting to determine the possible role of the IMD oligomerization or cooperativity during membrane deformation. Elucidating the mechanism by which the interplay between different functions of IRSp53 and MIM (membrane deformation, actin monomer binding, and WAVE2 interactions) contribute to cell migration and morphogenesis will also provide important challenges for future research. MIM/IMD-L was cloned into the SpeI–HindIII sites of pHAT1 vector (). The site-directed mutagenesis was performed as in . GFP fusions of MIM/IMDs were constructed into the XhoI–BamHI sites of pEGFP-N1 (CLONTECH Laboratories, Inc.). Alleles encoding mutant GTPases were cloned into BamHI–HindIII (Rac) or EcoRI–HindIII (RhoA, Cdc42) sites of pGEX-2T (GE Healthcare; ). The human IRSp53 construct was provided by H. Nakagawa (Kyushu Institute of Technology, Kyushu, Japan), and IRSp53/IMD cDNA was cloned into SpeI–NsiI sites of pHAT1. All IMD constructs were expressed as His tag fusion proteins, enriched with Ni-NTA Superflow beads (Sigma-Aldrich), and purified with Q-Sepharose high-performance anion- exchange column (GE Healthcare). Small GTPases Rac, RhoA, and Cdc42 were expressed as GST fusion proteins, enriched with glutathione-agarose beads (Sigma-Aldrich), and purified by Superdex-75 HiLoad gel filtration column (GE Healthcare). Human skeletal muscle α-actinin 2 (provided by J. Ylänne, University of Jyväskylä, Finland) was expressed and purified from pET8c-6HTEV plasmid (). Rabbit muscle actin was prepared from acetone powder as described by . PIP (-a-phosphatidylinositol-4,5-bisphosphate; porcine brain triammonium salt), PIP (1,2-dioctanoyl--glycero-3-[phosphoinositol-3,4,5-trisphosphate], tetra-ammonium salt), phosphatidylcholine (PC; brain), phosphatidylserine (PS; brain), and fluorescein PE (18:1 PE/CF 1,2-dioleoyl--glycero-3-phosphoethanolamine-N-[carboxyfluorescein]) were purchased from Avanti Polar Lipids, Inc. PC and PS were dissolved in 9:1 chloroform/methanol, PIP was dissolved in 1:2 chloroform/methanol, PIP was dissolved in 65:35:8 chloroform/methanol/water, and fluorescein PE was supplied in chloroform. The following mixtures were prepared: 5% fluorescein PE, 20% PS, 45–75% PC, and 0–30% PIP. Samples were vacuum dried under N and hydrated for a minimum of 4 h in 0.2 mM Hepes-KOH, pH 7.5, and 100 mM NaCl to a total lipid concentration of 1 mM. Finally, the vesicles were vortexed thoroughly for 2 min to allow formation of large multilamellar vesicles. Vortexing was repeated before each experiment. When preparing unilamellar vesicles, hydrated lipids were subjected to extrusion through a 1-μm filter according to manufacturer's instructions (Mini-Extruder; Avanti Polar Lipids, Inc.). Vesicles (167 μM total lipid and 3, 5, or 30% PIP) were mixed with 0, 1.1 (for tomography), or 22 μM IMD dimer in F-buffer (5 mM Tris-HCl, pH 7.5, 0.2 mM ATP, 0.2 mM DTT, 0.2 mM CaCl, 2 mM MgCl and 100 mM KCl). Reactions were incubated for 30 min and fixed by adding 2.1% glutaraldehyde and 0.1 M Hepes for 30 min at RT. Samples were sedimented (17,000 at 4°C) and postfixed by osmication (1% OsO, 15 mg/ml K[Fe{CN}], and 0.1 M Na-cacodylate buffer, pH 7.4), for 1 h, followed by en bloc staining with uranyl acetate (1% uranyl acetate and 0.3 M sucrose) for 1 h at 4°C, dehydration, and Epon embedding. Thin (60 nm) or semi-thick (120 nm) sections were prepared and stained with uranyl acetate and lead citrate for visualization with a transmission electron microscope (FEI Tecnai 12; FEI Corp.) operated at 120 kV. Images were recorded using a charge-coupled device camera (Erlangshen ES500W; Gatan Corp.). For 3D electron tomography, a 250-nm-thick section was prepared as described above and imaged with an electron microscope (Tecnai 20 FEG; FEI Corp.) operating at 200 kV. Images were recorded with a 1k × 1k change-coupled device camera (Multiscan 794; Gatan Corp.) at a magnification of 11,500× (1.63 nm/pixel). For collection of tilt series, the specimen was tilted ±70° at 1° intervals around two orthogonal axes (). The alignment and reconstruction of tilt series was done with the IMOD program package () using 10-nm colloidal gold particles as fiducial markers. The tomographic reconstruction was visualized and modeled with Amira software using volume rendering (TGS, Inc.). Actin filament cosedimentation assays were performed as described previously (; ). For lipid-binding assays, 20 μl of buffer (20 mM Hepes-KOH, pH 7.5, and 100 mM NaCl) or buffer with vesicles of 167 μM total lipid concentration was added to 25 μl of F-buffer. IMD constructs (5 μl) were diluted in desired concentrations in F-buffer. In high-speed (F-actin or vesicle binding) assays, samples were sedimented by centrifugation at 360,000 for 30 min at RT. In low-speed F-actin bundling assays, samples were sedimented at 17,000 for 30 min at 4°C. 12% SDS-polyacrylamide gels were scanned with a calibrated imaging densitometer (GS-710; Bio-Rad Laboratories, Inc.) and quantified with Quantity One software. Immunofluorescence and transfection of human osteosarcoma (U2OS) cells with GFP-MIM/IMD constructs were performed as described by . F-actin was visualized with Alexa 568 phalloidin (dilution 1:400; Invitrogen). To block actin polymerization, cells were treated with 0.2 μg/ml latrunculin A (Invitrogen). For analysis of the association of actin and vesicles in vitro samples containing 2.5 μM MIM/IMD-L dimer, 1.67 μM lipid vesicles (0 or 30% PIP), and 1 μM F-actin (50% Alexa Fluor 568–labeled actin) were prepared for light microscopy in modified F-buffer (10 mM imidazole, pH 7.5, 0.2 mM CaCl, 0.2 mM ATP, 1 mM DTT, 100 mM KCl, 2 mM MgCl, and 1 mM EGTA). 2-μl samples from reactions were applied on polyornithine-coated glass slides. Images were acquired through a charge-coupled device camera (DP70; Olympus) on a microscope (AX70 Provis; Olympus). For the image acquirement, the AnalySIS software (Olympus) and PlanApo 60×/1.40 (oil) objective (Olympus) was used. For in vitro samples, images were acquired with fixed exposure times for quantification with TINA software. The confocal image stacks were deconvoluted with AutoQuant AutoDeblur 3D Blind Deconvolution (AutoQuant Imaging, Inc.), and the 3D reconstructions and the colocalization analysis were made with Bitplane Imaris (Bitplane Inc.). The time-lapse images were acquired with an inverted microscope (IX70; Olympus) equipped with a Polychrome IV monochromator (TILL Photonics) with the appropriate filters, heated sample environment, CO control, and 40×/1.35 (oil) objective. Total internal reflectance fluorescence (TIRF) was performed using 60×/1.45 (oil) TIRF objective and 488 nm laser. Samples (0.5 mg/ml of IMD in F-buffer with desired KCl concentrations) were prepared, incubated for 1 h at 20°C, and monitored with a batch DLS instrument (Precision Detectors) equipped with deconvolution software for correlation function analysis (). GST pull-down assays were performed as in . Beads containing 5 μM GST or GST-Rac (V12 and N17) were incubated with 2.5 μM MIM/IMD dimer in 10 mM Tris, pH 7.5, and 100 mM NaCl for 10 min at RT, washed three times with 500 μl of reaction buffer, and analyzed on 12% SDS-PAGE. Interaction of Rac with MIM/IMD was studied with SPR on nitrilotriacetic acid sensor chip on a Biacore 2000 (Biacore AB) with a flow rate of 20 μl/min according to the manufacturer's instructions. The assay was performed in 10 mM Hepes-KOH, pH 7.2, and 150 mM NaCl with an injection contact time of 5 min and a dissociation time of 10 min. The recombinant His-tagged MIM/IMD splice variants and His-tagged GST proteins were immobilized by nickel chelation on nitrilotriacetic acid sensor surface to saturating levels (in the range of 9,000 resonance units) to facilitate detection of putatively low-affinity interactions. Binding of GST-Rac to MIM/IMD splice variants was studied under the system conditions between 20 and 300 nM Rac concentrations. To obtain binding levels at steady state, the control sensorgram, coated with His-GST, was subtracted from each MIM/IMD sensorgram. The steady-state binding levels of sensorgrams were evaluated by Langmuir 1:1 binding model modified to mass transfer effect in Biacore Evaluation Software 3.1. The obtained equilibrium binding level responses are given as resonance units. The coordinates for 3D structure of the MIM/IMD () were obtained from R. Dominguez (University of Pennsylvania, Philadelphia, PA). Protein concentrations were determined with a diode array spectrophotometer (8452A; Hewlett Packard) by using the calculated extinction coefficients for mouse MIM/IMD-L and -S (ɛ = 15,220 Mcm), IRSp53/IMD (ɛ = 19,770 Mcm), and actin (ɛ = 26,600 Mcm). The native PAGE analysis for detecting the lipid interaction of MIM/IMD was performed as previously described (). Fluorescence-monitored urea denaturation assays were performed as described by Lappalainen et al.. Fig. S1 shows that MIM/IMD prefers PIP over PIP, the binding affinity correlates with phosphoinositide density of vesicles, and MIM/IMD is capable of tubulating membranes with low PIP density. Fig. S2 illustrates two alternative splice variants of MIM/IMD. Fig. S3 shows the PIP and F-actin–binding experiments on MIM/IMD-L mutants. Video 1 shows electron tomography of MIM/IMD-induced membrane tubules. In Video 2, the dynamics of MIM/IMD-induced filopodia are shown, and Video 3 demonstrates that the dynamics of MIM/IMD-induced filopodia is abolished by latrunculin A treatment. Online supplemental material is available at .

Transmission of force from skeletal muscle myofibrils to the ECM is thought to be mediated largely by intermediate filaments (IFs). Several IF proteins are expressed in muscle, including vimentin, nestin, synemin, syncoilin, lamins, cytokeratins, and desmin, the major muscle-specific IF protein (for review see ). The desmin IF network forms a 3D scaffold surrounding Z-disks, extends from one Z-disk to the next, and finally connects the contractile apparatus to the plasma membrane at the level of Z-disks but also to organelles such as mitochondria and the nucleus (for review see ). The dystrophin–glycoprotein complex (DGC) has been implicated in mediating the IF-ECM link through syncoilin and synemin, which interact with desmin and bind to the DGC protein α-dystrobrevin (; ; ). The DGC is a large protein complex consisting of integral membrane proteins (α- and β-dystroglycan [βDG], α-, β-, γ-, and δ-sarcoglycan, and sarcospan), the >425-kD large actin-binding protein dystrophin, and dystrophin-associated proteins such as the syntrophins and α-dystrobrevin. Components of the DGC are part of the costameric protein network that, among other proteins, also includes integrins, vinculin, talin, α-actinin, and caveolin-3. Costameres are subsarcolemmal protein assemblies that circumferentially align in register with the Z-disks of peripheral myofibrils (for reviews see ; ); some authors include elements located above M-lines and in longitudinal lines in this term (). Muscular dystrophies (MDs) are a group of clinically and genetically heterogeneous diseases characterized by progressive muscle wasting. Lack of dystrophin leads to the most common form, Duchenne MD (DMD), but MD can also result from mutations in genes whose products are not known to associate with the DGC (). Most patients with plectin defects, who mainly suffer from various subtypes of the skin blistering disease epidermolysis bullosa (), have also been diagnosed with MD, and muscle phenotypes have been observed in plectin-deficient mice (). The cytolinker protein plectin is prominently expressed in striated muscle cells and has been visualized at Z-disks, the sarcolemma, and at mitochondria (; ; ; ), but the molecular mechanisms involved in plectin-related muscle disease/defects are unknown. Plectin is a large ( > 500,000) protein consisting of N- and C-terminal globular domains separated by an ∼200-nm–long rod. The N-terminal domain contains a multifunctional actin-binding domain (ABD; ) that is capable of also interacting with integrin β4 (; ) and vimentin () and also contains binding sites for nesprin-3 () and the nonreceptor tyrosine kinase Fer (). The C-terminal domain contains binding sites for IFs (Nikolic et al., 1996), the γ1 subunit of AMP kinase (), and the PKC scaffolding protein RACK1 (). Several different plectin isoforms, which are generated by tissue and cell type–dependent alternative splicing of transcripts from a single gene with >40 exons, form the basis for its broad versatility (; ). Isoforms with eight alternative N termini have been identified, and specific functions have been linked to distinct isoforms. Plectin 1a anchors keratin IFs to hemidesmosomes in basal keratinocytes (), and a specific role in fibroblast and T cell migration has been demonstrated for plectin 1 (). In skeletal muscle, four isoforms (plectins 1, 1b, 1d, and 1f) are expressed at considerable levels. In this study, we address the following issues: Where on the subcellular level are these plectin isoforms localized in muscle fibers? What are their muscle-specific (novel) binding partners? Are they differentially regulated during differentiation? What role do they play in dystrophic muscle, such as that of mice? Plectins 1d, 1f, 1b, and 1, the isoforms most abundantly expressed in skeletal muscle, show relative mRNA ratios of >10:4:3:1, respectively (). To obtain data about their expression and localization in skeletal muscle on the protein level, we isolated the quadriceps, a typical fast-twitch muscle composed of mainly type 2 fibers, from 10-wk-old mice and processed it for immunolabeling. Anti–pan-plectin antiserum revealed strong subsarcolemmal and moderate sarcoplasmic staining in cross sections of small diameter fibers and only faint sarcoplasmic and sarcolemmal staining in larger diameter fibers (). On longitudinal sections, Z-disks were stained in all fibers, but the signal was much stronger in small diameter fibers, where additionally the plasma membrane was stained (). These fibers, which showed strong autofluorescence at 488 nm (, F and H; insets), were positive for myosin heavy chain (MyHC)–2A (; also see E, G, and I), whereas those with larger diameters were MyHC-2B positive (). Therefore, it appears that in quadriceps, fast 2A fibers express plectin at higher levels than type 2B fibers, as has previously been reported for type 2 compared with slow type 1 fibers (). Double immunolabeling of plectin 1f and MyHC-2A on longitudinal sections revealed this plectin isoform to be located at Z-disks in 2A fibers but to be hardly expressed in 2B fibers (, D and E; and not depicted). On cross sections, 2A fibers showed moderate sarcoplasmic plectin 1f–specific staining as well as irregular and weak staining of the membrane (). Staining of longitudinal sections using a plectin 1–specific antiserum revealed this isoform to be much less abundant, if at all present, at Z-disks. However, a strong signal came from sarcolemma-associated structures, primarily in 2A fibers (). On cross sections, plectin 1–specific signals were detected as irregularly distributed accumulations at the sarcolemma of 2A but not 2B fibers (). As we were unsuccessful in generating isoform-specific antibodies directed against plectins 1b and 1d, we ectopically expressed and visualized GFP fusions of all four full-length plectin isoforms (1, 1b, 1d, and 1f) in myotubes (). Plectin 1 was expressed in a diffuse dotty pattern throughout the cytosol (; see virtual cross sections in insets 1 and 2) and was concentrated in the vicinity of nuclei. Immunolabeling with antibodies specific for sarcomeric α-actinin, a marker for Z-disks, revealed that areas positive for α-actinin were completely devoid of plectin 1 (, a and b; see areas marked by identically positioned arrowheads). Plectin 1b was distributed throughout the sarcoplasm in a pattern somewhat more patchy but similar to that observed for plectin 1, also mostly excluding areas that were positive for α-actinin (; a, b, and cross sections). Plectin 1d was located exclusively at structures identified as Z-disks (; arrowheads in a and b indicate the same exemplary positions). Contrary to expectations based on the immunostaining of tissue sections, plectin 1f was found not to be associated with Z-disks (). However, the observed sarcolemma association of this isoform was impressively confirmed (, virtual cross sections 1–4; a and b show individual confocal sections as indicated in panel 1). Immunolabeling of in vitro–differentiated C2C12 cells with plectin 1– and 1f–specific antibodies revealed the same localization of the native isoforms (unpublished data). To further investigate the sarcolemma association of plectin, extensor digitorum longus (EDL) muscle was teased into single fibers, which were immunolabeled for plectin and βDG, a costameric membrane marker (). Both proteins colocalized in costameric structures. Whereas βDG staining resembled a gridlike pattern (Z-disks and longitudinal lines), pan-plectin serum revealed prominent Z-disk and perinuclear localization and only a rare association of plectin with longitudinal lines. Virtual cross and longitudinal sections (, insets 1 and 2) through confocal stacks showed plectin at the sarcolemma but also extending into the fibers in regular intervals. Analysis with isoform-specific antibodies showed distinct staining patterns for plectins 1 and 1f. Whereas plectin 1 was found in the perinuclear area, at longitudinal lines, and in a dotty pattern at Z-disks (), plectin 1f was strongly expressed at Z-disks and tightly encircled nuclei (). Intriguingly, when the staining patterns for both isoforms are merged, they match that observed with pan-specific plectin antibodies, suggesting that plectins 1 and 1f are the major sarcolemma-associated isoforms. Costaining of plectin 1f with dystrophin revealed a partial colocalization of both proteins (). At the sarcolemma, plectin 1f was concentrated at Z-disks and extended into the fiber, whereas dystrophin was limited to the sarcolemma but was also found between Z-disks (, inset 1). To define the role of plectin in the process of differentiation from myoblasts to myotubes, we profiled the expression of plectin and other skeletal muscle proteins (). After 96 h of differentiation, myoblast cultures had formed myotubes that started to twitch (unpublished data). During differentiation, plectin isoform 1 expression peaked at 8–16 h, which is similar to that of utrophin. Plectin 1f was expressed only later, starting between 24 and 48 h, and reached plateau levels after 72 h. Interestingly, dystrophin showed a very similar expression profile, whereas βDG was detectable earlier (16 h), and caveolin-3 was not detected before 48 h. Integrin α7B was already expressed in myoblasts and showed peak levels after ∼48 h of differentiation. The spatially and temporally coordinated expression of plectin 1f and dystrophin suggested a possible direct interaction of both proteins. In immunoprecipitation (IP) experiments using IP lysates (see Materials and methods) from wild-type muscle tissue, plectin coprecipitated with dystrophin and vice versa (, lanes 5 and 7). From IP lysates, which were used as negative controls, plectin was not precipitated using dystrophin antibodies (, lanes 6 and 8). Interestingly, although plectin was expressed at higher levels in compared with wild-type muscles (, lanes 1 and 2), less of it was immunoprecipitated from (, lanes 5 and 6; also see ), indicating a shift of plectin into an insoluble pool. Utrophin and plectin were coprecipitated from rat fibroblast lysates (unpublished data). To identify interacting subdomains of the proteins, we immobilized His-tagged fragments of utrophin, including its N-terminal ABD, the C terminus, and the entire WW-ZZ domain as well as its three subdomains WW, EF, and ZZ on nitrocellulose membranes and overlaid them with various plectin samples (Fig. S1 A, available at ; summarized in ). Using either purified full-length plectin or plectin-rich cell lysates, we found plectin bound to the ABD and WW-ZZ domain of utrophin but not to its C-terminal part. A recombinant plectin ABD showed similar specificity, although its binding to the utrophin ABD was very weak compared with that of full-length plectin. When overlaid onto WW-ZZ subdomains, positive signals were obtained for the EF and ZZ domains, whereas a fragment encoded by exons 9–12 of plectin did not show binding to any of the utrophin fragments used. Confirming these results, a Eu-labeled version of the plectin ABD specifically bound to WW-ZZ domains of utrophin and dystrophin when overlaid onto microtiter plate–immobilized proteins (Fig. S1 B). Furthermore, the WW-ZZ domain of dystrophin competed with that of utrophin for plectin ABD binding as did actin (with considerably higher efficiency; Fig. S1 C). This suggested that simultaneous binding of actin and WW-ZZ domains to the plectin ABD was unlikely to occur. To define the relation of plectin with costameric membrane complexes in muscles lacking dystrophin, we characterized plectin localization in skeletal muscle fibers at different stages of MD in mice by immunolabeling cross sections of quadriceps from 2-, 4-, and 14-wk-old animals with plectin-specific antibodies (). Compared with normal muscle, no differences were observed at the (prenecrotic) age of 2 wk (not depicted) and in unaffected areas of quadriceps from 4-wk-old (peak necrotic) mice (, A and B and D and E). Plectin 1f was found in the sarcoplasm and irregularly at the sarcolemma, whereas plectin 1 was localized only in subsarcolemmal accumulations. Dystrophic areas were clearly distinguished by the presence of high numbers of small-diameter fibers with centralized nuclei and loose connective tissue, expressing very high levels of plectins 1 and 1f (). After 14 wk, most fibers had already passed through one round of degeneration/regeneration, and, as was expected from findings in DMD muscles (), we observed increased plectin staining at the sarcolemma of regenerated fibers when using a pan-plectin antibody (unpublished data). Isoform-specific antibodies revealed that this increased sarcolemmal staining was caused by the up-regulation of plectin 1f (, H and I vs. G). This was especially evident in 2B fibers, which are identified as large-diameter fibers lacking autofluorescence (; insets in G–I show autofluorescence). For plectin 1, on the other hand, we could not detect notable differences between and control samples (). Thus, during the regeneration of muscles, plectins 1 and 1f were both up-regulated in regenerating myotubes, but only plectin 1f associated with the sarcolemma and stayed there at high levels after regeneration was complete. To obtain quantitative estimates of plectin up-regulation in mice compared with other sarcolemma-associated proteins, we prepared KCl-washed microsomes from skeletal muscle of 8–10-wk-old and control mice (; ). Compared with total muscle lysates (), microsome fractions from control muscle were enriched in DGC components (dystrophin, utrophin, and βDG) and the membrane markers caveolin-3 and integrin α7B; in addition, actin and plectins 1 and 1f but not tubulin were detected in microsomes (). Comparing corresponding control and samples, we found increased levels of plectin (∼170%) and utrophin (∼140%) in total muscle lysates, whereas those of actin were similar (). Total plectin was two- to threefold more abundant in versus control microsome fractions, with relative levels of plectin 1 and plectin 1f of ∼300% and ∼150%, respectively. Interestingly, the levels of utrophin in the sarcolemmal fraction were only ∼50% of those found in the wild type, suggesting a weaker membrane association of utrophin in muscle. With our lysis protocol (see Materials and methods), the levels of βDG were found at ∼50% compared with the wild type, although βDG levels as high as ∼100% of wild type have been reported when samples were treated with cholate detergent (). Caveolin-3 appeared slightly increased, and no notable difference was observed in the case of actin. Interestingly, the levels of integrin α7B were approximately fourfold increased (). Thus, these biochemical data were in agreement with the observed up-regulation of plectin in muscle and the sarcolemma association of isoform 1f observed in the immunolabeling of tissue sections. To assess whether utrophin was substituting for dystrophin as a linker protein between βDG and plectin in dystrophin-deficient muscle, we stained cross sections of gastrocnemius for βDG and utrophin (). The antiserum to βDG gave a strong signal in control samples of all ages (2, 4, and 14 wk; ) and reduced but still clearly positive signals in the corresponding samples (). Actively regenerating fibers, which are marked by small diameters and centralized nuclei, showed a much stronger staining at the age of 4 wk that was almost similar in intensity to the signal in control fibers (). Utrophin was strongly expressed at the sarcolemmas of 2-wk-old control animals, whereas only limited staining was observed in corresponding muscle samples (, compare B with D). At later developmental stages of normal muscle, utrophin was detectable at myotendinous (, arrowheads) and neuromuscular junctions (not depicted), but no general sarcolemmal staining was observed (). In the case of 4- and 14-wk-old muscle, utrophin was present exclusively in regenerating small-diameter fibers (, H and L; asterisks) and at myotendinous junctions (, arrowheads). Faint sarcolemmal utrophin-specific staining that was not visible in wild-type muscle was observed in muscle (). The low levels of utrophin and the positive identification of βDG at the sarcolemma of muscle prompted us to costain teased wild-type and muscle fibers for plectin and βDG (). Both mAbs as well as an antiserum to βDG revealed that the gridlike staining pattern typical for costameres was lost in fibers, and, instead, βDG was found exclusively above Z-disks together with plectin 1f (, compare the insets of N and P, which are magnified in Q and R). We also examined microsome fractions and sections prepared from muscles of /utr mice and found a similar situation as in muscles (Fig. S2, available at ). The redistribution of βDG to sites above Z-disks where plectin 1f was concentrated suggested that plectin could directly interact with the cytoplasmic domain of βDG. Co-IP of both proteins from lysates of C2C12 myoblasts, mouse keratinocytes, and the human colon adenocarcinoma cell line CaCo-2 using anti-βDG antibodies was successful (). Using no or irrelevant antibodies, neither plectin nor βDG was detectable in the corresponding precipitates (unpublished data). Plectin and βDG could also be coprecipitated from lysates of skeletal muscle from mice (). Immuno-EM confirmed the close association of both proteins at the sarcolemma of muscle fibers (). When the plasma membrane was lost as a result of Triton X-100 extraction, βDG remained anchored to subsarcolemmal filamentous structures (, white arrowheads), which were also positive for plectin (). βDG and plectin labeling was most prominent at subsarcolemmal regions overlying Z-disks. Additionally, the plectin label was concentrated at the periphery of Z-disks (). To map the βDG-binding sites on plectin, a panel of His-tagged plectin fragments representing different structural domains () were blotted onto nitrocellulose and overlaid with the cytoplasmic domain of βDG. The WW-ZZ domain of dystrophin, which binds to the C terminus of βDG (), and the utrophin ABD or BSA were used as positive and negative controls, respectively. Using βDG-specific antibodies for detection, we found an interaction of βDG with two nonoverlapping plectin fragments (). One was encoded by plectin exons 12–24, representing part of the plakin domain located C terminally of the ABD within the plectin N-terminal globular domain, and the other corresponded to the C terminus of plectin, starting within repeat domain 4. No other protein tested showed binding except for the dystrophin WW-ZZ domain. To narrow down the region of βDG involved in binding to plectin, a fragment (βDGcytΔDBS) corresponding to roughly 70% of the βDG cytoplasmic domain (lacking the C-terminal region containing the dystrophin/utrophin-binding motif) was overlaid onto the same panel of recombinant proteins (). It bound to both plectin fragments identified before but bound much weaker to the C-terminal plectin fragment (Ple R-C), suggesting that additional C-terminal βDG sequences were needed for efficient binding to this fragment. As expected, the truncated βDG fragment failed to bind to the dystrophin WW-ZZ domain. Binding of βDG to both plectin fragments was efficiently blocked by the dystrophin WW-ZZ domain (). When dystrophin was added to the overlay solutions at an equimolar ratio, βDG–plectin binding was only slightly reduced, but increasing the molar ratios to 1:5 or 1:10 in favor of dystrophin led to strongly reduced binding and no binding, respectively. Unexpectedly, in this competition experiment, positive signals were observed in two additional lanes (Ple E1–12 and Utr ABD; ), suggesting that the dystrophin WW-ZZ domain mediated the indirect binding of βDG to the ABDs of utrophin and plectin. To confirm the simultaneous binding of WW-ZZ domains to plectin and βDG, we performed an overlay assay in which the utrophin WW-ZZ domain was immobilized on microtiter plates and incubated with constant amounts of Eu-labeled βDG in the presence of increasing concentrations of labeled or unlabeled versions of the plectin ABD (). In the first case, the amounts of labeled protein bound remained unchanged (, gray bars), whereas signals were additive (, black bars) in the latter, clearly demonstrating that plectin and βDG bound to independent binding sites within the WW-ZZ domain. Muscle formation is an incremental process in which differentiating myoblasts fuse and form primary and finally secondary myotubes. As this process involves massive rearrangements of the cytoskeleton, it was not unexpected to find that plectin isoforms were differentially expressed during its course. Of special interest was the striking similarity of the temporal expression patterns of plectin 1f and dystrophin, which suggested a role for this particular plectin isoform in the formation and maturation of costameres. This is supported by the observed exclusive membrane association of a recombinant version of this isoform in differentiated myotubes but not at the myoblast stage, where dystrophin is absent. A similar role of plectin had been suggested previously by , ), who concluded from their myoblast differentiation experiments that the association of plectin with Z-disks is a prerequisite for formation of the intermyofibrillar desmin cytoskeleton and, furthermore, that plectin is a component of primary longitudinal adhesion structures, which are precursors of costameres that form mature costameres only after being subjected to contractile forces. This would also explain the apparent discrepancy in plectin 1f localization in the tissue and in teased fibers (sarcolemma and Z-disks) versus transfected myotubes (sarcolemma only), as the latter represent a less mature stage. Interestingly, using mAb 121 to plectin, identified a membrane-associated plectin variant that is up-regulated during human myotube differentiation. Our results would suggest that this variant is plectin 1f. However, it is unexplainable how a mAb with an epitope in plectin's rod domain could specifically detect one rod-containing isoform (1f) over others. Immunostaining of muscle tissue revealed that plectin expression levels in individual fibers varied and were dependent on the fiber type. In cross sections of normal striated human muscle, a moderate to intense cytoplasmic and sarcolemmal staining of plectin has been reported in type 1 (slow twitch) fibers, whereas only faint staining of the sarcolemma was observed in type 2 (fast twitch) fibers (). In the present study, we show using an antiserum to plectin not discriminating among isoforms that in quadriceps (a typical fast muscle composed of mainly type 2 fibers), plectin clearly was localized at Z-disks in both type 2A and 2B fibers, with a stronger signal in 2A fibers. This corresponds well with the intense staining obtained with plectin 1f–specific antibodies in this fiber type. Neither with anti–plectin 1f nor anti–plectin 1 antibodies did we detect substantial Z-disk staining in 2B fibers. Based on this observation, one other plectin isoform expressed in skeletal muscle, plectin 1b or 1d, must be associated with Z-disks in type 2B fibers. From our overexpression experiments, we conclude that this isoform is plectin 1d, as it was found localized exclusively at Z-disks. At this time, the molecular mechanism of this targeting is unknown. Using co-IP and in vitro binding assays, we demonstrate direct interactions of plectin via multiple interfaces with components of the DGC, including direct binding of its plakin domain to the cytoplasmic domain of βDG; direct binding of its C-terminal portion to βDG; binding of its ABD to the WW-ZZ domains of dystrophin and utrophin; and binding of its ABD to the ABD of utrophin ( and , schematics). Whether the ABD of plectin would also interact with that of dystrophin remains an open question considering that the functionalities of the ABDs of dystrophin and utrophin differ (), whereas their WW-ZZ domains are highly conserved (). Plectin and dystrophin had previously been coimmunoprecipitated from muscle lysates, but their interaction was assumed to be indirect via actin (). In dystrophin-lacking mice, one may thus expect to find the reduced sarcolemma association of plectin, but our immunofluorescence and tissue fractionation experiments revealed that plectin was instead enriched at the sarcolemma of muscle. It was widely believed that dystrophin deficiency in skeletal muscles of mice and DMD patients leads to the reduced expression and sarcolemmal association of dystrophin-associated proteins, including a strong reduction or even absence of βDG immunoreactivity () despite its mRNA levels being similar to those in normal samples (; ). However, in a recent study, challenged this view when they demonstrated that in muscle samples treated with 2% cholate, βDG was detectable at levels comparable with those of wild-type samples. The authors proposed that βDG was targeted to the plasma membrane normally in dystrophin- deficient muscles but remained inaccessible to antibodies and, when tissues were lysed, became part of an SDS-insoluble pool. Using our protocols, we also found considerable levels of βDG in skeletal muscle microsome fractions (∼50% of wild type even without cholate treatment), and we were able to immunodetect βDG with variable intensities throughout muscle regeneration. Thus, our results support the hypothesis of , and the direct interaction of plectin with βDG provides an explanation for the observed increase in sarcolemmal plectin in muscle (). In the absence of dystrophin, more plectin can bind to βDG, causing at the same time the redistribution and accumulation of βDG above Z-disks, where plectin is normally localized. Matching our βDG immunostaining results, have observed a corresponding redistribution of αDG in teased fibers. It has been suggested that utrophin could substitute for dystrophin in dystrophic muscles, but its intimate association with βDG may be limited to the time of regeneration only (). The ∼50% reduction of utrophin observed in microsome preparations from muscle of ∼10-wk-old mice despite the overall higher expression of utrophin (∼150% of wild-type levels) would support this notion. Furthermore, this observation also suggests that plectin binds to βDG with a higher affinity than utrophin. Remaining sarcolemmal utrophin staining that was observed in transgenic mice overexpressing Dp71, a short splice variant of dystrophin lacking rod and N-terminal domains (; ), was explained by to possibly be caused by the binding of utrophin to subsarcolemmal actin via its ABD. Our finding that βDG immunostaining signals observed in tissue from 4-wk-old animals was almost as strong as in normal muscle would also fit the proposed model (), as the higher expression of utrophin during the phase of peak necrosis/regeneration may displace plectin from βDG and, thus, restore accessibility of the epitopes masked by plectin. Similarly, the overexpression of Dp71 (or other dystrophin or utrophin deletion variants) in restored normal DGC components, whereas a full phenotypic rescue was achieved only by proteins with functional ABDs and intact C-terminal βDG-binding domains (). The DGC has been considered to be responsible for connecting the subsarcolemmal actin cytoskeleton to the ECM, and disruption of this link causes a dystrophic phenotype. However, in recent years, it has been established that the contractile actions of a muscle fiber are mechanically integrated by desmin IFs, which are responsible for linking individual myofibrils laterally with each other and to the sarcolemma at the level of the Z-disks. Previous studies have implicated the DGC as the transmembrane complex linking the IF network with the ECM (for reviews see ; ; ). The necessary link would be created by an α-dystrobrevin–synemin/syncoilin–desmin bridge. Synemin may also directly interact with vinculin, providing an alternative anchorage of desmin IFs to costameres (). Plectin directly interacts with desmin via its C-terminal IF-binding domain () and also with multiple components of the DGC, including its transmembrane core protein βDG. Thus, we propose that plectin acts as a direct linker between the DGC and the desmin IF network. There is precedence for such a function of plectin in basal keratinocytes, where the protein directly links the keratin IF network to the cytoplasmic domain of the β subunit of the laminin receptor integrin α6β4 (). It could be that synemin plays the essential role in establishing the direct linkages between heteropolymeric IFs and the myofibrillar Z-disk and costameric regions, and plectin might only provide additional structural support at these sites. However, this would be in conflict with observations in differentiating human skeletal muscle cultures, where plectin was already localized in a cross-striated pattern, whereas desmin was still found in longitudinal filaments (). Recently, it was shown that besides type III and IV IFs, the cytokeratins K8 and K19 are also expressed in striated muscle and localize to Z-disks and M-lines and that K19 directly interacts with the dystrophin ABD (). Whereas plectin has been shown to directly bind to keratins 5, 14, and 18 (; ), it is unknown whether it can also interact with the muscle-specific keratins and possibly plays a role in their anchorage as well. Based on our observations, we propose the following model for plectin's association with the DGC (). Because binding of plectin to βDG via its C-terminal binding site was abolished in the absence of the C-terminal part of βDG's cytoplasmic domain (harboring the PPxY motif required for interaction with the WW domain of dystrophin and utrophin; see for a discussion of the WW domain and its interactions) and binding of plectin to βDG was efficiently blocked by dystrophin, only a portion of the βDG cytoplasmic tails would normally be available for binding to plectin when dystrophin is present. However, plectin could remain associated with the DGC via binding of its ABD to dystrophin (utrophin). In such a constellation, the plectin C-terminal domain, separated from the N terminus by the ∼200-nm–long rod domain, would be exposed and available for binding to muscle IFs (desmin and potentially also cytokeratins 8/19). In the absence of dystrophin, however, the available binding sites on βDG are occupied by plectin, leading to the increased sarcolemmal plectin 1f signal observed in /utr, and DMD muscle fibers and consequently to an increased insolubility of βDG and masking of its epitopes. Binding via the C terminus would also leave the plectin ABD available for interaction with filamentous actin, possibly taking over some of the responsibilities of dystrophin. Finally, based on the cellular targeting of overexpressed recombinant full-length versions of plectin isoforms, we propose plectin isoform 1d to be involved in the anchorage of desmin IFs to Z-disks (, bottom). Recently, several proteins involved in signaling such as nonreceptor tyrosine kinase Fer, the PKC scaffolding protein RACK1, and the key enzyme involved in energy homeostasis, AMP kinase, have been identified as novel interaction partners of plectin and led to the proposal that plectin acts as a scaffolding platform for signaling proteins in addition to serving as a cytoskeletal linker. Having established plectin as a component of the DGC with multiple binding interfaces to its key components, it will now be a challenge to define its role in DGC-mediated signaling. An important consequence implied by our model could be that misguided signaling mediated by the accumulation of plectin scaffolds at the sarcolemma of and DMD dystrophic muscle contributes to the disease phenotype. Full-length (mouse) plectin isoform cDNA constructs (including respective 5′ untranslated regions) encoding proteins with C-terminal GFP have been described previously (). For bacterial expression of plectin fragments, the corresponding cDNAs were PCR amplified by using primers with EcoRI tails and were cloned into pJD1, a modified pET-15b (Novagen) that was obtained by replacing the EcoRI–BamHI fragment of pBN120 with that from pAD29 (); expressed proteins contained an N-terminal His tag and a C-terminal tag. The following plectin fragments were used: Ple E1–12 (M–R), Ple E9–12 (E–V), Ple E12–24 (M–E), Ple Rod (E–Q), Ple R–R (A–K), and Ple R–C (L–A), which were all from a rat (), as well as Ple ABD (D–N) from a mouse (NM_201389). To express N-terminally His-tagged versions of human utrophin (X69086) fragments, Utr ABD (S–D), Utr WW-ZZ (A–M), Utr WW (A–K), Utr EF (I–S), Utr ZZ (N–M), and Utr C-terminal (M–M) EcoRI-flanked cDNAs were cloned into pBN120 (). The WW-ZZ domain (A–M) of human dystrophin (X14298) was also expressed from pBN120. Fragments βDGcyt (L–P) and βDGcytΔDBD (L–D) of human βDG (NM_004393) were expressed from pLJ1 (a pET32a [Novagen]-derived plasmid in which the sequence between the NcoI and XhoI restriction sites has been replaced by 5′-AATTCCTGGTGCCACGCGGTTCT-3′) as proteins with N-terminal Trx-His-S tags and C-terminal His tags. The cDNA fragments encoding Ple ABD and Dys WW-ZZ were also inserted into the EcoRI site of pMal-c2 (New England Biolabs, Inc.) to generate fusion proteins with N-terminal maltose-binding protein. For immunoblotting (IB), IP, immunofluorescence microscopy (IFM), and immuno-EM, the following antibodies were used: mAbs 5B3 (IB) and 7A8 (EM; ) to plectin; antisera #9 (IB), #46 (IFM), and #123 (IFM) to plectin (); anti–plectin isoform 1 antiserum (IB and IFM; ); anti–plectin isoform 1f antiserum (IB and IFM), which was prepared and affinity purified as described previously () using amino acids M–K of plectin 1f () as immunogen; mAb EA-53 (Sigma-Aldrich) to sarcomeric α-actinin (IFM); mAb AC-40 (Sigma-Aldrich) to actin (IB); mAb B-5-1-2 (Sigma-Aldrich) to tubulin (IB); mAb 43DAG1/8D5 (IB, IFM, and IP; Novocastra) and rabbit antisera #1709 and #1710 (Tyr 895-P; IB, IFM, IP, and EM; ) to βDG; anti-utrophin antiserum RAB5 (IB, IP, and IFM; ); mAb DY4/6D3 (Novocastra) to dystrophin (IB); mAb (clone 26) to caveolin-3 (IB and IFM; BD Biosciences); anti–integrin α7 antiserum (IB; provided by U. Mayer, University of East Anglia, Norwich, UK; ); mAbs to MyHC-2A (SC-71) and -2B (BF-F3; IFM; hybridomas were obtained from the German Resource Center for Biological Material; ); and mAb to myc epitope tag (1-9E10.2; IB; American Type Culture Collection). As secondary antibodies, we used goat anti–rabbit IgG AlexaFluor488 (Invitrogen), goat anti–mouse IgG Texas red (Jackson ImmunoResearch Laboratories), and donkey anti–rabbit Cy5 (Jackson ImmunoResearch Laboratories) for IFM and used goat anti–rabbit and goat anti–mouse IgGs conjugated to AP or HRP (Jackson ImmunoResearch Laboratories) for IB. Thin sections (3–5 μm for longitudinal and 8–10 μm for cross sections) were prepared from skeletal muscle (quadriceps and gastrocnemius) dissected from C57BL/10 control and mice (Institut für Labortierkunde, Medical University of Vienna) and frozen in liquid nitrogen–cooled isopentane. Sections were placed on slides, fixed with acetone for 10 min, and incubated for 1 h in 5% goat serum in PBS to block nonspecific binding of antibodies. Samples were incubated with primary and secondary antibodies diluted in PBS for 1 h each. Signal specificity was controlled by the omission of primary antibodies or by using normal mouse or rabbit serum in their place. To prepare teased fibers, mice were anesthetized with isoflurane and perfused with 2% PFA in PBS. EDL was dissected and incubated with the same fixing solution for 10 min. Using fine forceps, the muscle was teased into single fibers, which were then adhered onto chrome-alaun/gelatin-coated slides. Slides were blocked with PBS containing 0.1% BSA and 0.1% Triton X-100 for 1 h, incubated for 2 h with primary antibodies (diluted in blocking solution), washed with PBS for 30 min, incubated with secondary antibodies (diluted in PBS) for 1.5 h, and washed again with PBS. Finally, samples were briefly rinsed with water and mounted in Mowiol. Immortalized (p53 negative) mouse myoblasts () were cultivated on collagen-coated (5 mg/ml in PBS overnight; Sigma-Aldrich) tissue culture dishes in F-10 (Invitrogen) medium containing 20% FCS, 2.5 ng/ml human basic FGF (Promega), 100 U/ml penicillin, and 100 μg/ml streptomycin. Myoblasts were transfected using FuGENE6 (Roche), and differentiation was initiated after 12–16 h by switching the medium to DME containing 5% horse serum, 100 U/ml penicillin, and 100 μg/ml streptomycin. After 4–6 d, myotubes were fixed with chilled (−20°C) methanol and processed for immunolabeling and subsequent laser-scanning confocal microscopy. For IB analysis of protein expression profiles during differentiation, myoblast cultures were differentiated, and cells were lysed directly in reducing SDS sample buffer at different time points. Immunolabeled tissue and cell samples mounted in Mowiol were viewed in a fluorescence microscope (Axiophot; Carl Zeiss MicroImaging, Inc.) using plan Neofluar 40× NA 1.2 (for tissue sections; Carl Zeiss MicroImaging, Inc.) and plan Apochromat 63× NA 1.3 (for cells and teased fibers; Carl Zeiss MicroImaging, Inc.) objectives. Confocal images were recorded using the LSM510 module (Carl Zeiss MicroImaging, Inc.) and the LSM510 software package (version 3.2 SP2; Carl Zeiss MicroImaging, Inc.). Images were processed using LSM Image Browser (generation of projections of confocal stacks; gamma/contrast adjustments; version 3.2; Carl Zeiss MicroImaging, Inc.) and Photoshop CS2 (cropping and splitting of color channels; Adobe) and were mounted/labeled using Illustrator CS2 (Adobe). For preembedding immuno-EM, perfusion-fixed (2% PFA in PBS) adult rat EDL was dissected and teased into small fiber bundles before shock freezing in liquid nitrogen–cooled isopentane. Samples were thawed in PBS, treated with 0.2% Triton X-100 in PBS for 1 h, and blocked for 1 h in 0.1% Triton X-100, 0.1% BSA, and 1:25 normal goat serum in PBS (blocking solution). For primary immunolabeling, teased fibers were incubated overnight at 4°C in a mixture of mAb 7A8 to plectin and antiserum #1710 to βDG (diluted in blocking solution without serum). After washing for 1 h (0.2% BSA in PBS), samples were incubated overnight at 4°C in a mixture of gold-conjugated goat anti–mouse (5 nm) and goat anti–rabbit (10 nm) secondary antibodies (British Biocell International). Gold-labeled fibers were washed in PBS, postfixed in 2.5% glutaraldehyde for 30 min, and washed in double-distilled water before silver enhancement for 1 h (R-GENT SE-EM kit; Aurion). Samples were immersed in 0.5% OsO in PBS for 15 min, dehydrated, and embedded in epoxy resin (agar 100; Agar Scientific Ltd.). Thin sections were cut with an ultramicrotome (Ultracut S; Leica), mounted on copper grids, counterstained with uranyl acetate and lead citrate, and examined at 80 kV in an electron microscope (JEM-1210; JEOL). Digital images were acquired and processed using a camera (Morada; Olympus) and the analySIS software package (Olympus). Hind leg muscles were dissected from C57BL/10 control and mice, snap frozen in liquid nitrogen, and ground in a mortar. Muscles were homogenized in solution A (20 mM NaPO, 20 mM Na-PO, pH 7.4, 0.303 M sucrose, 0.5 mM EDTA, 1 mM MgCl, 2 mM PMSF, and Complete mini protease inhibitor cocktail [Roche]) using a Dounce homogenizer (∼10 strokes). Part of the crude homogenate (total muscle lysate) was mixed with an equal volume of SDS sample buffer (0.4 M Tris, pH 6.8, 0.5 M DTT, 10% SDS, 50% glycerol, and 0.1% bromophenol blue) for further analysis; the rest was centrifuged for 15 min at 20,000 , and the pellet was rehomogenized. Combined supernatants were filtered through six layers of cheesecloth and centrifuged for 15 min at 25,000 . The pellet was discarded. To the supernatant, solid KCl was added to a final concentration of 0.6 M. After centrifugation for 35 min at 200,000 , the pellet was resuspended in solution B (20 mM Tris-maleate, pH 7.4, 0.303 M sucrose, 0.6 M KCl, and the same protease inhibitors as in solution A). After incubating for 1 h, KCl-washed microsomes were pelleted for 35 min at 200,000 and resuspended in solution B without KCl. 5 g of muscle yielded 0.5 ml of microsome suspension. All steps were performed at 4°C on ice. For subsequent IB analysis, microsome suspensions were mixed with 5 vol SDS sample buffer. Proteins were separated using standard 5 or 15% SDS-PAGE. Note the considerably higher concentration of SDS (∼60 mg/ml) in our samples compared with standard conditions (∼20 mg/ml). Under these conditions, immunoblot analysis of microsome fractions generally gave much better results, which are likely caused by the enhanced solubilization of membrane-associated protein complexes in these lipid-rich fractions. For IB, proteins were transferred to nitrocellulose membranes, and membranes were blocked with 5% nonfat dried milk in PBS–0.05% Tween 20 and incubated with primary and AP-conjugated secondary antibodies. For quantitation, stained membranes were scanned, and bands were evaluated using the ImageQuant 5.1 software package (Molecular Dynamics). Normalization factors based on total protein content or tubulin signals were applied to the values measured. Mouse myoblasts (), mouse keratinocytes (), and CaCo-2 (HTB-37; American Type Culture Collection) were cultured as recommended or described previously. IP with antisera to plectin and βDG was performed essentially as described previously (). In brief, clarified cell extracts in RIPA buffer were incubated for 2 h at 4°C with antibodies or, for control experiments, without antibodies or with host sera. Immunocomplexes were collected by centrifugation after a further 1-h incubation with protein A– or G–Sepharose beads (GE Healthcare) and extensive washing with RIPA buffer and were subsequently analyzed by IB. For IP from muscle tissue, total muscle lysates were prepared as for microsome preparations with the addition of 0.5% Triton X-100 to solution A. Part of the lysates was mixed with SDS sample buffer (total lysates), and the rest was incubated for 3 h with protein A–Sepharose beads and centrifuged (IP lysates) before incubation with antibodies overnight. Immunocomplexes were captured by protein A–Sepharose beads and eluted with SDS sample buffer. Recombinant protein fragments were expressed in BL21(DE3) and purified as described previously (). Protein fragments were transferred to nitrocellulose membranes after 10% SDS-PAGE. Membranes were blocked with 5% BSA in TBS containing 0.5% Tween 20 and overlaid with 10 μg/ml of proteins in 20 mM Hepes, pH 7.5, 150 mM NaCl, 2 mM MgCl, 1 mM DTT, and 5% BSA and incubated overnight at 4°C with agitation. Bound proteins were detected by IB using protein- or epitope tag–specific primary and HRP-conjugated secondary antibodies or (in the case of His- and S-tagged fragments) by using the India HIS detection system (Pierce Chemical Co.) and HRP-conjugated S protein (Novagen), respectively. The experimental details of this binding assay have been described previously (). In brief, proteins immobilized on microtiter plates were overlaid with Eu-labeled proteins in solution at different concentrations. After washing, the amounts of proteins bound were determined by measuring Eu fluorescence in comparison with a standard. Fig. S1 shows the binding data (blot overlay assays) summarized in the table in B as well as additional microtiter plate–binding data. Fig. S2 shows immunofluorescence images of tissue sections from wild-type and /utr mice coimmunolabeled with antibodies to plectin and βDG as well as immunoblots of wild-type and /utr muscle lysates using primary antibodies to plectin, dystrophin, utrophin, and βDG. Online supplemental material is available at .

Congenital muscular dystrophies (CMDs) represent a clinically and molecularly heterogeneous group of autosomal recessive neuromuscular disorders with a typical early onset of symptoms. Estimates in Italy suggest an incidence rate of 4.65 × 10 (). Thus, after Duchenne muscular dystrophy (DMD), CMDs represent the second most frequent neuromuscular disorder. Laminin-α2–deficient CMD, classified as MDC1A, accounts for ∼30–40% of all CMD patients. MDC1A is a severe progressive muscle-wasting disease that leads to death in early childhood (; ; ). It shows a rather homogenous clinical picture, with severe neonatal hypotonia associated with joint contracture and inability to stand or walk. Moreover, MDC1A is accompanied by a peripheral neuropathy that is caused by demyelination in the peripheral and central nervous system. However, no mental retardation is observed in most patients. Laminins are cruciform-like molecules formed by α, β, and γ chains (). There are 5 α, 3 β, and 3 γ chains described so far that give rise to 15 isoforms (). The central role of laminins can be explained by their dual function in organizing a structured basement membrane through interaction with other basement membrane proteins and connecting basement membranes to adjacent cells via cell surface receptors. Inactivation of different laminin chains in mice causes distinct phenotypes (for review see ). The laminin-α2 chain assembles to laminin-211 (LM-211; α2, β1, and γ1) and LM-221. LM-211 is the main isoform in the basement membrane of muscle and peripheral nerve, whereas laminin-221 is restricted to neuromuscular junctions (). In the basement membrane, LM-211 and -221 bind to other laminins, to nidogen (which in turn binds to collagen IV and perlecan), and to agrin (). The self-polymerization activity of LM-211 is thought to be particularly important for the formation of a proper muscle basement membrane. The main receptors for laminin-α2 in adult muscle are dystroglycan and α7β1 integrin (, green). Dystroglycan is cleaved into the peripheral α-dystroglycan and the transmembranous β-dystroglycan. In the membrane, dystroglycan associates with the sarcoglycans and sarcospan and intracellularly binds to dystrophin, which in turn links the complex to the f-actin cytoskeleton. The complex between LM-211, dystroglycan, sarcoglycans, and dystrophin, which is called the dystrophin–glycoprotein complex (DGC), has been shown to be of utmost importance for the maintenance of muscle integrity, as mutations in these components cause different types of muscular dystrophies (for review see ). Similarly, mice or humans that are deficient of α7 integrin display a mild muscular dystrophy (; ), and muscle-specific inactivation of β1 integrins has a major impact on muscle development (). Thus, the evidence is strong that both receptor systems contribute to the linking of basement membrane to the f-actin cytoskeleton, and it is likely that the two systems act synergistically. MDC1A is among the most severe muscle dystrophies, which may be based on the observation that the absence of laminin-α2 leaves both receptor systems unoccupied by its ligand. As a compensatory mechanism, muscle fibers of MDC1A patients and laminin-α2–deficient mice increase synthesis of laminin-α4 (; ; ; ). However, LM-411 is truncated at the N-terminal end, which prevents its self-polymerization, and it also does not bind to α-dystroglycan or α7β1 integrin with high affinity (; ). There is also evidence that muscle fiber membranes of MDC1A patients, and mice models thereof, contain significantly lower levels of α7β1 integrin () and α-dystroglycan (; ). In addition, the ability of muscle to regenerate is greatly impaired (; ). These deficiencies lead to the dystrophic phenotype characterized by high levels of creatine kinase (CK) in the blood, large variation in fiber size, successive replacement of muscle by fibrous tissue, and infiltration of adipose tissue. Good models for the disease are / mice generated by homologous recombination (). Like human patients, / mice have an early onset and severe dystrophic phenotype, which is often lethal between 6 and 16 wk. They grow at a slow rate, the histology of muscles is very similar to that of human patients, and they have a prominent peripheral neuropathy based on defective myelination of the peripheral nerve. There is no curative treatment for MDC1A. However, a “replacement therapy” using a miniaturized form of the basement membrane component agrin (mini-agrin) was shown to markedly lower muscle degeneration and mortality in / mice (). This is caused by both increasing the tolerance to mechanical load and improving the regenerative capability of the muscle (). These studies left several questions unanswered that were addressed in the current study. First, an efficacious treatment also needs to work after the onset of the disease. Second, a requisite to envisage pharmacological treatment options that aim at increasing synthesis of endogenous agrin is to show that full-length agrin can also have a beneficial effect. Finally, although it is highly suggestive that the beneficial effect of mini-agrin is based on the linking of the up-regulated LM-411 with α-dystroglycan (), other mechanisms (e.g., via integrins) may also contribute. In an attempt to answer these open questions, we prepared a panel of constructs to generate different transgenic mouse lines (). First, we used the tet-off system (Fig. S1 A, available at ) to generate / mice in which expression of mini-agrin can be temporally controlled (). Second, we generated transgenic / mice that overexpress chick full-length muscle agrin (c-FLag) in muscle (). Third, we generated / mice that overexpress a fusion construct in which we replaced the α-dystroglycan binding region of chick mini-agrin (c-mag) with that of mouse perlecan (AgPerl; ). Domain V of perlecan (also called endoreppelin; ) binds to α-dystroglycan (), but not to integrins that are expressed in muscle. In this study, we show that mini-agrin can slow down the progression of MDC1A at any stage of the disease, full-length agrin is capable of improving muscle function, and the fusion construct between agrin and perlecan also counteracts the disease. In summary, our results are conceptual proof that linkage of laminin isoforms with α-dystroglycan is a means to treat MDC1A also at progressed stages of the disease. #text xref italic #text The m-mag cDNA was obtained by two independent RT-PCRs on mRNA isolated from mouse skeletal muscle. The 0.75-kb cDNA encoding the 25-kD N-terminal agrin and the 2.2-kb cDNA encoding the 95-kD C-terminal half were ligated, and a 5× myc-tag (0.25 kb) was added to the 3′ end to yield the m-mag–myc (m-mag) construct. The 3.2-kb m-mag construct was sequenced and subcloned downstream of the uni-directional pTRE2 tet-responsive promoter (tetO7-CMV; BD Biosciences; pTRE2, 3.8 kb). A PacI site was inserted into the pTRE2 vector to allow linearization of the construct as a 4.9-kb PacI–AseI fragment for injection into mouse oocytes. All transgenic mouse lines in which the cDNA was stably inserted into the genome were mated with transgenic mice expressing the tetracycline-dependent transcription activator (tTA) under the control of a 3.3-kb fragment of the human muscle CK (MCK) promoter (Fig. S1). MCK-tTA mice were obtained from N. Raben (National Institutes of Health, Bethesda, MD; ) and were shown to drive expression of transgenes in skeletal and heart muscle (; ). The AgPerl fusion protein was created by fusing the cDNA encoding the 0.75-kb 5′ region of chick agrin () with a cDNA coding for domain V of mouse perlecan (full-length cDNA encoding mouse perlecan was a gift from T. Sasaki, Max-Planck-Institut for Biochemistry, Martinsried, Germany). Both the 6.2-kb c-FLag and the 3.1-kb AgPerl were subcloned downstream of the 1.3-kb MCK promoter (Fig. S1, B–D). Constructs were linearized and injected into male pronuclei. C-mag transgenic mice (c-mag) were created as previously described (). / mice () containing a LacZ insertion in the gene served as the mouse model for MDC1A. Genotyping of heterozygous and homozygous / mice was done as previously described (). M-mag mice were genotyped by primers designed to amplify a 683-bp-long fragment, including the linker region of the N- and C-terminal parts of the m-mag construct (5′-GCGGATCACTTTGCGGAACC-3′ and 5′-TCGAACCTGAACTGTACATGACC-3′). Both c-FLag and c-mag mice were genotyped with primers amplifying a 591-bp-long fragment coding for the C-terminal part of agrin (5′-ACCTGGATAAGCGTTTTGTT-3′ and 5′-CTTCTGTTTTGATGCTCAGC-3′). Genotyping of AgPerl transgenic mice was performed on the chick agrin portion of the construct (5′-GTCCCTTGCTGATGACCTTGA-3′, 5′-ACCCAGCCCCTCAGTACATGT-3′). To distinguish hemi- from homozygous MCK-tTA mice, we performed quantitative TaqMan PCR (TaqMan PCR core reagent kit; Applied Biosystems) on the genomic DNA. The following primers were used: -transgene, 5′-GCCTACATTGTATTGGCATGTAAAA-3′, 5′-CAAAAGTGAGTATGGTGCCTATCTAACA-3′, and Probe 5′-FAM-CTTTGCTCGACGCCTTAGCCATTGAG-TAMRA 3′. For normalization of copy number, the following probes for β-actin were used: 5′-CCACTGCCGCATCCTCTT-3′, 5′-GCTCGTTGCCAATAGTGATGAC-3′, and Probe 5′-FAM-CCCTGGAGAAGAGCTATGAGCTGCCTG-TAMRA-3′. For temporal regulation of m-mag expression under the tetracycline-regulated tet-off expression system (), 5 μg doxycycline (doxycycline hydrochloride; Sigma-Aldrich) per milliliter of drinking water (enriched by 4% sucrose) was administered in dimmed bottles. For repression after transgene expression, 50 μg of doxycycline per milliliter of drinking water was applied. The cDNAs encoding m-mag or AgPerl were subcloned into the pCEP-Pu vector () and transfected into HEK 293 EBNA cells. Conditioned medium was collected, and the relative amount of the protein was determined by dot blot assays. Such supernatants were directly used for experiments. 96-well plates were coated with either chick α-dystroglycan enriched from skeletal muscle as previously described () or with laminin-111 (0.5 μg/well), which was a gift from J. Engel (Biozentrum, University of Basel, Switzerland). Proteins were coated in 50 mM sodium carbonate buffer, pH 9.6, and incubated overnight at 4°C. After blocking with PBS containing 0.05% Tween-20, 1 mM CaCl, 1 mM MgCl, and 3% BSA (blocking buffer), wells were incubated with a dilution series (1:6) of supernatant containing m-mag (pure supernatant as the starting concentration) or of purified c-mag (50 nM as the starting concentration). The wells were washed with blocking buffer. Bound protein was detected with polyclonal antibodies raised against the C-terminal, 95-kD part of chick or mouse agrin. Alternatively, the monoclonal antibody 9E10 () directed against the myc-tag was used. For detection, appropriate horse radish peroxidase–conjugated antibodies, followed by McEvans solution, ABTS, and HO, were used. The absorbance was measured on an ELISA reader at 405 nm after 15 min. Lysates enriched for α-dystroglycan were obtained from chick or mouse skeletal muscles, as previously described (). Proteins were separated on a 3–15% SDS gel and blotted to nitrocellulose membrane. Blots were blocked for 2 h with PBS containing 0.05% Tween-20, 1 mM CaCl, 1 mM MgCl, and 5% dry milk powder (blocking buffer). Supernatants containing recombinant proteins were added and incubated overnight at 4°C. After several washes with blocking buffer, bound m-mag was detected with the anti-myc antibody 9E10, whereas detection of AgPerl was done using a polyclonal antiserum raised against the N-terminal part of agrin. For detection, appropriate horse radish peroxidase–conjugated antibodies were used, and immunoreactivity was visualized by the ECL detection method (Pierce Chemical Co.). Tissues were homogenized in protein extraction buffer (75 mM Tris-HCl, pH 6.8, 3.8% SDS, 4 M urea, 20% glycerol, and 5% β-mercaptoethanol). Equal amounts of protein were separated on a 3–12% SDS–PAGE and immunoblotted. Protein signals were normalized to β-actin (Santa Cruz Biotechnology; sc-8432) or β-tubulin (BD Bioscience). Northern blot assays were performed on total RNA extracted from skeletal muscles using Northern Max Kit (Ambion). Signals were normalized to corresponding β-actin signals. Quantitative TaqMan PCR was performed on the m-mag transgene (5′-TGTGCCAATGTGACCGCTA-3′, 5′-GCTGAAACCCTTGCCAGAA-3′, and Probe 5′-FAM-CCCCCAAAGTCCTGTGATTCCC -TAMRA 3′) and was normalized to β-actin (5′-CCACTGCCGCATCCTCTT-3′, 5′-GCTCGTTGCCAATAGTGATGAC-3′, and Probe 5′-FAM-CCCTGGAGAAGAGCTATGAGCTGCCTG-TAMRA-3′). Locomotive behavior was determined as previously described (). In brief, mice were placed into a new cage and motor activity (walking, digging, and standing upright) was measured for 10 min. Grip strength was evaluated by placing the animals onto a vertical grid and measuring the time until they fell down. The cut-off time was 3 min. Blood for CK assays was collected from the tail vein. 2 μl of serum was applied using the CK CK-NAC Liqui-UV kit (Rolf Greiner Biochemica). In all tests, at least three animals of each genotype were analyzed, and values were normalized to values obtained from WT animals. Muscles were immersed in 7% gum tragacanth (Sigma-Aldrich) and rapidly frozen in liquid nitrogen–cooled isopentane (–150°C). 12-μm-thick cross sections or longitudinal sections were cut and collected on SuperFrost Plus slides (Menzel-Glaser). In the case of longitudinal sections, the slides were pretreated with 3% aqueous EDTA. General histology was performed using HE (Merck). Masson's Trichrome staining () was used to visualize collagenous tissue. Membrane-bound and extracellular epitopes were visualized with Alexa Fluor 488–conjugated WGA (Invitrogen). Polyclonal rabbit anti–mouse laminin-α5 (Ab 405) and monoclonal rat anti–mouse laminin-α1 (Ab 198; ) were a gift from L. Sorokin (Lund University, Lund, Sweden). Polyclonal sheep anti–mouse α-dystroglycan was a gift from S. Kröger (University of Mainz, Mainz, Germany). The remaining antibodies were produced in-house or obtained as follows: monoclonal mouse anti–rat dMyHC (Novocastra), monoclonal rat anti–mouse laminin-γ1 chain (CHEMICON International, Inc.), polyclonal rabbit anti–chick (produced in-house; ), and anti–mouse agrin (produced in-house). Mouse monoclonal anti-myc antibody (9E10) was produced and purified from hybridoma cell line 9E10 and was biotinylated (D-Biotinoyl-E-aminocaproic acid-N-hydroxysuccinimidester; Roche). Depending on the source of the primary antibody, appropriate Cy3-conjugated (Jackson ImmunoResearch Laboratories) Alexa Fluor 488–conjugated secondary antibodies (Invitrogen) or TRITC-labeled streptavidin were used for visualization. DAPI was used to stain nuclei. The muscle fiber size was quantified using the minimum distance of parallel tangents at opposing particle borders (minimal “Feret's diameter”), as previously described (). Pictures of WGA-stained cross sections were collected using a fluorescence microscope (DM5000B; Leica), a digital camera (F-View; Soft Imaging System), and analySIS software (Soft Imaging System). Measurement of minimal Feret's diameter of notexin-treated muscle was done on cross sections stained for laminin-γ1 and dMyHC. Normalization of the number of fibers in each fiber Feret class of 5 μm was based on the total number of muscle fibers in each picture. Fibrosis was quantified by measuring the fibrotic area of WGA-stained muscle cross section and normalizing it to the entire area of the cross section. For quantification of immunostainings of m-mag, c-mag, c-FLag, laminin-α5, or α-dystroglycan, images were collected and analyzed by a confocal microscope (TCS-8P; Leica) and appropriate software. InSpeck Microscope Image Intensity Calibration kit (Invitrogen) was used to determine the linear range of the laser. Specific intensity was calculated for each image as the signal intensity of the muscle circumference minus that of an adjacent, nonstained region (). Five different pictures were taken using the same parameters on each section, and four different sections were used for each individual mouse. In all quantification experiments, at least three mice of each genotype were analyzed. Transgenic c-mag and c-Flag were detected by the polyclonal rabbit anti–chick agrin (). For comparison of the transgenic m-mag and the endogenous agrin, an antiserum recognizing the 95-kD, C-terminal half of mouse agrin was used. Chick and mouse agrin immunostainings were quantified separately, as described in the previous section. Under the premise that c- and m-mag ameliorate the disease phenotype to the same extent, the relative expression levels of both were set to 100%. This clearly shows that levels of endogenous agrin expressed in kidney (expression level, 113% of m-mag) are sufficient to at least produce the ameliorating effect of c-Flag (expression level, 81% of c-mag). Tibialis anterior of 5-wk-old mice was injured by injection of 15–20 μl notexin (50 μg/ml; Sigma-Aldrich), as previously described (). Mice were killed 6, 14, or 28 d after injection, and muscles were isolated and processed as described in Quantification of immunostainings. Fibrosis in triceps brachii muscles was measured by assaying for the exclusive collagen-specific modified aminoacid hydroxyproline (; ). Tendons were carefully removed before muscles were vacuum-speed dried and sent to Analytical Research Services (Bern, Switzerland) for amino acid analysis. There, each muscle was hydrolyzed under vacuum in 50 μl of 6 N HCl for 22 h at 115°C. Hydrolysates were evaporated to dryness and resuspended in 0.1% trifluoroacetic acid. Aliquots were diluted 1:100 for determination of amino acids by a routine method (), including derivatization with phenylisothiocyanate, followed by HPLC, identifying, and quantifying the collagen-related amino acid hydroxyproline. Relative hydroxyproline amount was assessed in reference to the total amount of amino acids. To compare the different genotypes, p-values were calculated using the unpaired two-sample -tests, assuming equal variances. Fig. S1 represents the regulation of expression by the inducible tetracycline-regulated tet-off expression system () and the breeding scheme to obtain / mice with a tight spatial and temporal regulation of mini-agrin expression. Fig. S2 shows the binding of the transgenic m-mag and the fusion protein AgPerl to laminin and α-dystroglycan in both solid-phase and overlay binding assays. In Fig. S3, immunohistochemical staining of cross sections visualizes the regulation of different agrin-binding proteins, including laminin-α5, α-dystroglycan, and laminin-α1 in /m-mag mice. The online version of this article is available at .

Hereditary spastic paraplegia (HSP) is a genetic disease characterized by neurodegeneration of the long axons of spinal neurons of the dorsal columns and cortical spinal tract, with clinical symptoms resulting from this pathology. HSP is a heterogeneous disorder both clinically and genetically with dominant, recessive, and X-linked forms linked to >20 loci (). One of the most common forms of autosomal dominant HSP is caused by mutations in Spastin (; ). We and others have shown that Spastin is a microtubule (MT)-severing enzyme (; ), meaning that Spastin makes internal breaks in MTs. The mechanisms of enzyme assembly, tubulin recognition or engagement, and severing are unknown. Many disease-associated missense mutations in Spastin impair both ATPase and MT-severing activities, suggesting that HSP and axon loss paradoxically result from failure of MT severing. Spastin is a member of the AAA ATPase (ATPases associated with various cellular activities) family (for review see ). The defining feature of these enzymes is the ∼250-amino-acid module known as the AAA ATPase domain. In the majority of cases examined, AAA proteins assemble into ring-shaped hexamers, sometimes in an ATP-dependent manner. In part, this is because the ATP binding pockets are constituted from residues from adjacent monomers in the oligomer. Many AAA proteins disassemble protein complexes without destroying their substrates. For instance, NSF apparently unwinds the coiled coils that define the SNARE complex, resulting in SNARE complex disassembly so that the individual SNARE components can be reused for successive rounds of membrane fusion (). Particularly well understood AAA proteins are the so-called translocators. These enzymes, whose members include the Clp family, are often involved in protein quality control. The hexameric forms of these proteins engage target proteins by recognizing C-terminal peptides () and pulling or translocating these polypeptides through the central pore of the hexamer (; ). Successive cycles of ATP hydrolysis lead to unfolding of target proteins and complete and directional translocation of the substrate through the pore often into an associated proteolytic machine (; ). Therefore, action of the translocating type AAAs results in denaturation or destruction of the native target protein. Structural studies of several AAA proteins, particularly translocators, revealed the existence of loops that project into the pore cavity in the hexamer. In a hexamer, each monomer contributes two loops. Mutational analysis of many AAA proteins has shown that pore loop integrity is critical for function. Pore loop 1 lies near the surface of the pore and contains a conserved YVG motif also found in Spastin (; ; ; ). Mutation of these residues often results in retained ATPase activity and enzyme assembly but loss of function. In the P4 phage packaging motor, the equivalent of pore loop 1 has been shown to alternate between “up” and “down” conformations throughout the ATPase cycle (). With ClpA and ClpB (; ) or p97/VCP (), cross-linking studies have shown that substrate proteins are in direct contact with pore loop 1. These data suggest that pore loop 1 may engage target proteins and be responsible for translocation. Pore loop 2 lies deeper inside the pore. In translocators, pore loop 2 has not been cross-linked to target proteins, but mutations in this loop block function and sometimes lead to an inability to bind target proteins (; ). Although pore loops are present in AAAs that are not known to translocate target proteins, the contribution of these loops to the function of these enzymes is not as well investigated. Here, we show that Spastin assembles into a hexamer in the presence of ATP, that hexameric but not unassembled Spastin recognizes the extreme C-terminal amino acids of its target protein, tubulin, and that this interaction is required for MT severing. Mutational analysis of the pore loops, including analysis of a disease-associated mutation in pore loop 2, shows that both pore loops are critical for MT severing and probably are directly involved in recognizing the extreme C terminus of tubulin. Despite the fact that the tubulin C-terminal tail is present in tubulin dimer, Spastin does not bind well to tubulin dimer compared with MT polymer. In addition to the pore interaction with tubulin, we define an N-terminal MT binding domain (MTBD) in Spastin that mediates the attachment of Spastin to the MT. This interaction is also required for severing. We have previously shown that recombinant Spastin severs MTs (). To simplify the task of addressing the mechanism by which Spastin severs MTs and to produce a more soluble recombinant protein that expresses better in , we first sought to define which domains of Spastin are dispensable for MT severing. Like many other AAA proteins, Spastin contains a C-terminal AAA cassette (roughly aa 343–616) and an N-terminal region (aa 1–342). Homology to other AAA proteins is only within the AAA domain. The N-terminal region consists of four subdomains (see for schematic): the Atlastin binding domain (), the MIT domain, alternatively spliced exon 4 (), and as we show in , an MTBD that is both necessary for MT severing and sufficient for MT association. To qualitatively assay MT severing, we exploited the fact that Spastin overexpression results in nearly complete loss of MTs (; ). Expression of the Δ1–227 deletion mutant resulted in loss of MTs, suggesting that the first 227 amino acids of Spastin are not required for severing activity (). Deletion of the first 279 amino acids of Spastin did not abolish severing (, Δ1–279 Spastin). However, Spastin lacking the first 328 amino acids did not sever MTs in transfected cells (, Δ1–328). In aggregate, the deletion experiments suggest that a region between residues 279 and 328 is important for MT severing. Unexpectedly, we found that recombinant Spastin lacking this region (aa 279–328) could still sediment with purified MTs in vitro (unpublished data). Therefore, we deleted a few more amino acids and produced recombinant Spastin proteins lacking either aa 1–227 (Δ1–227 Spastin) or both aa 1–227 and aa 270–328 (called ΔMTBD Spastin; ). Using a previously published sedimentation assay for MT destruction () where MTs are recovered in the pellet and released tubulin in the supernatant after centrifugation, we found that Δ1–227 Spastin could associate with MTs in the absence of ATP and caused MT disassembly in the presence of ATP (). In contrast, ΔMTBD Spastin neither bound nor severed MTs, demonstrating that aa 270–328 are necessary for MT severing in vitro. Consistent with this in vitro data, overexpression of ΔMTBD Spastin in cells did not result in MT loss (). We next examined whether the MTBD was sufficient to confer MT association both in cells and in vitro. In transfection experiments, 1–328 stop Spastin but not 1–279 stop Spastin decorated MTs, suggesting that amino acids 280–328 are important for MT binding (). We next wanted to examine whether this domain was sufficient for direct MT interaction. We found that we could use the heterobifunctional chemical cross-linker ethyl-3-(dimethylaminopropyl)-carbodiimide (EDC) to detect a direct interaction between Spastin and tubulin/MTs (Fig. S1, available at ). This cross-link did not require ATP. A GST fusion protein containing only aa 227–279 of Spastin was not able to cross-link to MTs (). However, a fusion protein containing aa 227–328 of Spastin was able to cross-link to MTs, suggesting that this MTBD is sufficient for direct interaction with MTs. Despite the fact that GST-227-328 Spastin could be cross-linked to MTs, it did not sediment with MTs (unpublished data). This may suggest a low-affinity interaction. Together, these data suggest that an N-terminal MTBD encompassing aa 270–328 in Spastin is also important for MT severing. This domain also appears to be sufficient for MT association in a nucleotide-independent manner. We made internal deletions within this region to refine it further (Fig. S2, available at ) and found that both halves of this domain seem to mediate MT association in cells but one half appears more critical for MT severing than the other. Consistent with our results, others have shown that deletion of the entire N-terminal region preceding the AAA domain in Spastin resulted in an enzyme incapable of MT severing (). They did not address whether this N-terminal domain bound to MTs directly. Our data demonstrating that Spastin lacking the first 270 amino acids can still associate with MTs and sever in vivo seem inconsistent with the claim by others of a MTBD in the first 269 amino acids of Spastin (). Those experiments did not use purified proteins, and it is possible that the observed MT association in crude cell lysates was indirect, perhaps mediated by endogenous Spastin in the cells. Our deletion results suggested that Δ1–227 Spastin severs MTs. Compared with recombinant full-length Spastin, this deletion mutant expressed much better, was more soluble, and was less prone to aggregate. Therefore, we characterized this version of recombinant Spastin and, unless otherwise indicated, all further experiments and effects of point mutations were examined in this “backbone.” We performed ATPase assays for Δ1–227 Spastin and calculated the K and V for the enzyme (). Because these values were previously determined for full-length Spastin using a Lineweaver-Burke plot (), we recalculated K and V using the nonlinear regression method; these numbers are also shown in . Δ1–227 Spastin severed MTs in a videomicroscopy-based assay that we had previously used to study Spastin (). We conclude that the first 227 amino acids of Spastin are not required for MT severing. Although Δ1–227 Spastin severs MTs and has similar ATPase kinetics compared with full-length protein, severing as assessed by the videomicroscopic assay is not amenable to kinetic analysis in our experience. Therefore, we cannot ascertain whether the truncated enzyme severs as well as full-length protein. Limited proteolysis of MTs with subtilisin, which cleaves off small peptides from the extreme C terminus of α- and β-tubulin (), renders them resistant to severing by Spastin (). Therefore, we hypothesized that Spastin might engage the free C terminus of tubulin during the severing reaction. Examination of MT structure shows that the last ∼100 amino acids of α- and β-tubulin project from the surface of the MT primarily as two helices (, ). The last several amino acids of α- and β-tubulin are unstructured and are not resolved in crystal structures. We prepared recombinant GST fusion proteins representing the C-terminal 104 amino acids of α-tubulin. Because of posttranslation detyrosination of α-tubulin, we prepared tails ending either in Tyr or Glu and predicted that Spastin should bind directly to these tails. Accordingly, we immobilized the GST–tubulin tails on glutathione–Sepharose and found that wild-type (WT) Spastin did not bind to the bead-immobilized tubulin tails (). Given that Spastin can associate with MTs in vitro in the absence of ATP (), this result was surprising. Because WT Spastin could not bind to the tubulin tail and because AAA proteins harboring a canonical Walker B motif point mutation (E442Q in Spastin) often remain kinetically trapped on their cellular target proteins (; ), we tested whether E442Q Spastin could bind the tubulin tail. Indeed, E442Q Spastin bound to the tail only in the presence of ATP. E442Q Spastin bound equally well to both Glu and Tyr tails. E442Q Spastin did not bind well to truncated GST–tubulin tails lacking the last 22 amino acids (), suggesting that the most C-terminal region of these peptides are important for Spastin binding. This ATP-dependent binding appears to represent a distinct interaction with tubulin as compared with the ATP-independent interaction between tubulin and the N-terminal MTBD discussed earlier (). In other AAA proteins, the E to Q Walker B mutation allows for ATP binding but not hydrolysis, trapping the oligomeric form of the enzyme. This suggests that oligomeric Spastin is binding to the tubulin tail. To specifically test whether Spastin can assemble into a homooligomer and whether the E442Q mutation traps the oligomer as a stable complex, we analyzed Spastin by gel filtration. E442Q Spastin, which is predicted to bind but not hydrolyze ATP, assembled into a homooligomer in an ATP-dependent manner (, compare top two panels). In contrast to the relatively symmetrical elution profile of oligomeric Spastin shown in , in some experiments, the peak was less symmetrical and left shifted. Neither ATP nor ATPγS trapped WT Spastin as a stable oligomer. We presume that WT Spastin can form a hexamer but that even with ATPγS, the hexamer is not as stable as that formed by the E442Q mutant. To more accurately characterize the Spastin homooligomer, we collected the high molecular mass oligomer fraction of E442Q Spastin obtained by gel filtration and analyzed the protein complex by sedimentation velocity in the presence of ATP. shows that the normalized sedimentation coefficient distributions obtained over a broad concentration range are nearly superimposable, with a major peak at 7.6 ± 0.05 Svedbergs (uncorrected). Thus, Spastin forms a stable, nonequilibrating oligomer over the concentration range examined. As described in Materials and methods, we also calculated the translational diffusion constant from the sedimentation velocity data. We then used the Svedberg equation to calculate the molecular mass of Spastin. The sedimentation data fit to a mean mass of 250 ± 10 kD, which agrees well with a predicted hexamer mass of 258 kD (Δ1–227 Spastin has a mass of 43 kD). We conclude that Spastin assembles into a hexamer in the presence of ATP. The binding data in suggest that the hexameric form of Spastin binds to the tubulin tail. In light of the assembly data, it is likely that the WT Spastin hexamer dissociates upon ATP hydrolysis, whereas the E442Q hydrolysis-deficient mutant remains trapped as a stable hexamer. We imagine that WT Spastin does form a hexamer and binds transiently but that we cannot detect binding on account of hexamer instability. Why might hexameric but not monomeric Spastin bind to the tubulin tails? Assembly of AAA proteins results in the formation of a central pore (). Because pore loops in some AAA proteins have been shown to directly contact substrate proteins, we hypothesize that the observed binding to the tubulin tails is mediated by pore residues in Spastin and that mutations in these pore loops should abrogate the tubulin tail–Spastin interaction and MT-severing activity (). We introduced several pore loop mutations into E442Q Spastin and asked whether these pore mutations could suppress the ATP-dependent binding to the tubulin tails (). Residues 415–417 in Spastin constitute the YVG motif in pore loop 1 (). Many groups have found that the aromatic residue in pore loop 1 is crucial for function (; ; ). E442Q/Y415A Spastin did not bind to the GST–α-tubulin tails (). Because the tubulin tail is extremely acidic, we predicted that basic amino acids in pore loop 2 might be important for tail binding. Compared with some other AAA proteins, pore loop 2 in Spastin lacks a conserved acidic residue and harbors an extra basic residue (). Therefore, we engineered missense mutations in pore loop 2: two that alter charged amino acids (R451G and A457E) and one representing a disease-associated mutation (C448Y; ). shows that C448Y/E442Q Spastin but not R451G/E442Q or A457E/E442Q Spastin bound to the GST–α-tubulin tail. The Arg residue that we mutated is not conserved in all Spastins (e.g., Spastin). However, in dSpastin, there is a Lys residue in the loop at a more C-terminal position, and the “charge balance” in the loop is therefore the same as human Spastin. Tubulin is a heterodimer, and we therefore examined the binding of Spastin to β3-tubulin tails. We chose to examine this neuron-specific β-tubulin isoform both because HSP is a neuronal disorder and because, although, like α-tubulin, it has an acidic tail, the sequence, particularly the last four amino acids, is quite distinct from α-tubulin (). Similar to the results for the α-tubulin tail, we found that E442Q Spastin bound to GST–β3-tubulin tail in an ATP-dependent manner (). As with the α-tubulin tail, introduction of the Y415A, R451G, or A457E pore loop mutations into E442Q Spastin abolished binding. In contrast to the result obtained with α-tubulin tail, the C448Y HSP-associated mutation abrogated binding to the β3-tubulin tail. In summary, mutations in both pore loops 1 and 2 can affect tail binding. Importantly, compared with many disease and engineered mutations in Spastin that we previously examined (), the pore loop mutants all retained some ATPase activity (). Also, we found that R451G/E442Q and A457E/E442Q Spastin could still oligomerize in the presence of ATP (). The ATPase measurements and gel filtration experiments suggest that pore loop mutations do not prevent pore assembly and that the point mutations in pore loops more or less specifically prevent tubulin tail binding without globally disrupting Spastin. We conclude that the tubulin tail likely binds in the pore in hexameric Spastin in a loop-dependent manner. If the tubulin tail binds inside the central pore of the Spastin hexamer, the C-terminal Tyr residue in the tubulin tail should be relatively resistant to proteolysis by carboxypeptidase A (CpA), which specifically removes this Tyr from α-tubulin (). We observed partial protease protection, consistent with this notion (Fig. S3, available at ). If the pore–tubulin tail interaction is important for MT severing, the pore mutations should impair Spastin's MT-severing activity, even though the enzymes can hydrolyze ATP (). Indeed, when cells were transfected with plasmids encoding YFP-Spastin harboring pore mutations, severing did not occur (). Instead, the mutant Spastin proteins decorated MTs. Similarly, using the previously described sedimentation assay, recombinant Spastin proteins harboring pore loop mutations (Y451A, C448Y, and R451G) did not sever MTs in the presence of ATP (), as nearly all of the tubulin remained in the pellet as polymer. Similar results were obtained for the A457E mutation (not depicted). Note that a small fraction of each protein sedimented into the pellet only in the presence of MTs. Because we showed that pore loops engage the tubulin tail and that mutations that prevent this interaction also prevent severing, we further predicted that C-terminal peptides derived from tubulin should act as competitive inhibitors of Spastin-mediated MT severing. We incubated the GST–tubulin tail fusion protein with Spastin and taxol-stabilized MTs in either the presence or absence of ATP, as indicated in . We used the previously described MT sedimentation assay to monitor MT disassembly. In the absence of GST–tubulin tail, Spastin effectively severed MTs. In contrast, when GST–tubulin tail was added in 10-fold molar excess compared with tubulin, severing was substantially blocked. Truncated GST–tubulin tail lacking the C-terminal 22 amino acids did not act as a competitive inhibitor of Spastin. Although the last 22 amino acids are necessary for severing, a fusion protein consisting only of the 22 amino acids of tubulin fused to GST did not act as a competitive inhibitor (unpublished data). This may indicate that the rest of the tubulin tail is important or that when expressed alone, the last 22 amino acids of the tail are not properly ordered. The experiment in was performed with tubulin tail ending in Tyr, and similar results were obtained with peptides terminating in Glu (not depicted). Because the tubulin tails are present on nonpolymerized tubulin heterodimer, we might have predicted that Spastin, particularly the hexameric form, would interact with nonpolymerized tubulin heterodimer. To test this idea, GST-E442Q Spastin was immobilized on glutathione agarose, and tubulin was added in the presence of ATP (). In contrast to the tail binding experiments (), where about one third of the added Spastin could be captured on tubulin tail beads, only negligible amounts of tubulin heterodimer were captured on GST-Spastin beads. This suggested that Spastin only interacts poorly, if at all, with tubulin dimer. Consistent with this notion, soluble tubulin dimer does not cofractionate with hexameric E442Q Spastin by gel filtration analysis (unpublished data). These binding studies suggest that hexameric Spastin does not bind tubulin well despite the presence of the tail. This may indicate that the tails are perhaps more accessible to Spastin when tubulin is polymerized in a MT or that Spastin plays a role in exposing the tail to the pore loops. Also, as mentioned earlier, we can detect interaction of Spastin with tubulin using the chemical cross-linker EDC (Fig. S1). We compared the efficiency of Spastin cross-linking obtained using MTs to that obtained using nonpolymerized tubulin and found that cross-linking to polymer is much more efficient than to nonassembled tubulin, again suggesting that Spastin interacts better with MTs (, compare cross-links in lanes 4 and 8). Our deletion analysis revealed the existence of an N-terminal MTBD in Spastin (aa 270–328) that is required for MT-severing activity in vivo and in vitro. Additionally, this domain may be sufficient for MT association. Analysis of this region of Spastin did not reveal any homology to any known protein or functional motif. Although Katanin has an MTBD preceding its AAA domain, there is no homology between the N terminus of Spastin and Katanin. In fact, the only region of homology between these MT-severing proteins is in the AAA ATPase domain. Our deletion analysis of the N-terminal region of Spastin also revealed that the first three subdomains in the N terminus of Spastin encompassing the first 227 amino acids are not required for MT severing. Previously, we () and others () found that Atlastin binds to a region in the first 116 amino acids of Spastin. Atlastin, an integral membrane GTPase, is encoded by another autosomal dominant HSP gene. We hypothesize that Spastin-interacting proteins, such as Atlastin, binding to the first 227 amino acids of Spastin might serve to localize Spastin to sites of activity (). Because we mapped an MTBD between aa 270 and 328, it is unlikely that known Spastin binding proteins interacting with regions between aa 1 and 227 would interfere with MT severing. This is consistent with the observation that Atlastin binding does not appear to regulate MT severing by Spastin (). We demonstrated that Spastin can assemble into a hexamer during its ATPase cycle. This is significant because Vps4, another AAA protein closely related to Spastin, has been proposed to assemble into larger oligomers such as 12 mers rather than hexamers (; ). Although slowly hydrolyzable ATP analogues allow for trapping of some WT AAA ATPases as oligomers, this is not the case for Spastin (). Similarly, Katanin () and Vps4 cannot be trapped as stable oligomers with ATP. Instead, the demonstration of homooligomeric forms of these enzymes also necessitated the use of the Walker B E to Q mutation. Together with Fidgetin, Spastin, Katanin, and Vps4 constitute the meiotic subfamily of AAA ATPases (), and in this respect, this subgroup appears distinct from other AAA proteins. Subtilisin treatment of MTs, which removes small peptides from the C terminus of α- and β-tubulin, renders MTs resistant to severing by both Spastin and Katanin (; ), the other known AAA ATPase that severs MTs. This prompted us to investigate the role of the tubulin C terminus in MT severing. Our results show that pore loops in hexameric Spastin recognize the extreme C terminus of its target protein (tubulin) and that this interaction is critical for MT severing. Whether Katanin recognizes tubulin in a similar manner is not known. Several microtubule-associated proteins and motor proteins also bind to the tubulin tail; this raises the possibility that certain microtubule-associated proteins may protect MTs from severing as proposed (; ). Though the role of pore amino acids has been well studied in translocating AAA proteins involved in protein quality control, their role in other AAA proteins is not as well understood. A pore loop 1 mutation in Vps4 abolishes retroviral budding in a cell-based assay (), just as a similar mutation in Spastin abolished severing. Our analysis of pore loop 2 in Spastin reveals that it plays a critical role in severing and in tubulin C-terminal tail recognition, akin to the role of pore loop 2 in ssrA tag recognition (). The C448Y disease-associated mutation uncouples ATPase from severing activities ( and ) and binds to the α-tubulin tail (). Interestingly, it cannot bind to the β3-tubulin tail (). Thus, even in non-Clp type AAA ATPases, pore loops are critical for function and for target protein recognition. Even though pore loops are present in unassembled Spastin, hexamerization is important for stable, robust tail binding. This could mean that there is simultaneous engagement of the tail by more than one loop or that the loops adopt different structures in the hexameric Spastin. In transfected cells, pore loop mutations in Spastin, whether they abolish tail binding or not, result in Spastin proteins that decorate MTs. Therefore, it is probable that MT association with these mutants occurs via the N-terminal MTBD and that the enzyme assembles and perhaps hydrolyzes ATP. However, we imagine that severing fails to occur either because the tubulin tail cannot engage the pore loop (YA, RG, or AE for α- and β-tubulin and these plus C448Y for the β-tubulin tail) or the normal response of the AAA ring to the engagement is blocked because of the mutated pore loops (CY). In either case, the protein apparently does not release from or sever the MTs. Spastin lacking the N-terminal MTBD domain does not sever or decorate MTs despite having intact pore loops. This suggests that although tubulin tail–loop interaction is necessary for severing, this additional Spastin tubulin interaction must also be necessary for MT severing. Although Δ270–328 Spastin/E442Q can hexamerize and hydrolyze ATP, it could not bind to the GST–α-tubulin tails (Fig. S4, available at ). The reason for this is not clear but may suggest that the N-terminal MTBD–MT interaction is important in positioning the hexamer such that the tubulin tail is pore accessible or that this MTBD actually plays a role in exposing that tail to the pore. We do not know where the N-terminal MTBD contacts the MT. We have seen that purified recombinant Spastin can bind MTs. This includes WT enzyme as well as mutants that are predicted not to bind ATP (e.g., K388A Spastin) or hydrolyze ATP (e.g., E442Q; ). Thus far, only deletion of the N-MTBD prevents MT association in vitro and in vivo. In contrast, when mutant versions of Spastin are overexpressed in cells, two distinct patterns are observed. Most mutants analyzed in this and past work () do not sever MTs. However, some mutant proteins decorate MTs (e.g., E442Q and pore loop mutants), whereas others do not even though they bind in vitro (e.g., K388A and R499C; ). We imagine that decoration of MTs in cells reflects more than mere binding and may reflect kinetic trapping of hexameric Spastin on MTs. Because E442Q forms an ATP-dependent oligomer, it is possible that this mutant protein assembles into a hexamer on MTs but, because it cannot hydrolyze the bound ATP, remains kinetically trapped on the MT, resulting in decoration. Interestingly, the pore loop mutant Spastins decorate MTs even though they can hydrolyze ATP, albeit not quite as well as WT enzyme. This may imply that, although necessary for release, completion of an ATPase cycle is not sufficient to result in release from the MT. shows a schematic of the two distinct interactions Spastin makes with MTs. We propose that Spastin exerts force on the MT in part by engagement of the tubulin tail by Spastin's pore. Severing could be achieved by pulling the tail into the pore. In this model, the N-MTBD would merely anchor the hexamer and severing would be analogous to untying a shoelace by pulling on a slipknot. Alternatively, the importance of the tail binding could be that relative motions occur between the two sites where Spastin is anchored to an MT. Soluble tubulin has a curved conformation but has a straight conformation in the MT polymer (; ; ). Other proteins, such as stathmin, cause MT catastrophe by binding to and sequestering curved tubulin dimer (), and depolymerizing kinesins may engage tubulin tails to peel off dimers at MT ends by flipping them into the curved conformation (). Spastin could sever MTs by flipping a tubulin heterodimer in the middle of a polymer into the curved conformation, making it incompatible with polymer packing. Given that the tubulin tails are present on nonpolymerized dimer, one might have expected hexameric Spastin to bind to dimer. The fact that it does not () may mean that the tails are not accessible in the dimer but may become exposed or properly oriented in the MT polymer either by virtue of some polymer-induced conformational change in this region of tubulin or by virtue of Spastin binding to the polymer. As mentioned earlier, Spastin could conceivably be restricted to MT polymer if it indeed interacts preferentially with tubulin in a straight conformation. As Spastin apparently does not interact well with soluble dimer, any ATP-driven change in tubulin conformation occurring during severing may result in Spastin release from the severed tubulin. Expression and purification of GST-Spastin was as previously described () with the following modifications. Deletions and point mutations were made using the QuikChange kit (Stratagene). Δ1–227 Spastin expressed better and was more soluble than full-length Spastin, resulting in better yields. GST-tagged protein was expressed from the pGEX-6P-3 vector (GE Healthcare). Protein was expressed using BL21-CodonPlus (DE3)-RIPL (Stratagene). Cultures were grown in LB supplemented with carbenicillin, chloramphenicol, and streptomycin to an OD of 1.0 and induced overnight at 30°C by adding IPTG to 0.5 mM. Cells were lysed in 1 ml of lysis buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 5 mM MgCl, 5 mM DTT, 1 mM ATP, and EDTA-free protease inhibitor tablet [Roche]) per 20 ml culture using a high-pressure homogenizer (Emulsiflex C-5; Avestin). Triton X-100 was added to 1% to the lysate at this stage. Detergent was not kept in the buffer during purification. Lysates were cleared by centrifugation (Ti60 rotor [Beckman Coulter]; 45,000 rpm, 45 min, 4°C) and pumped over a 1- or 5-ml glutathione–Sepharose column (GE Healthcare). The column was washed with 10 column volumes of lysis buffer with 500 mM NaCl and 1 mM DTT (no ATP) followed by 10 column volumes of lysis buffer with 150 mM NaCl and 1 mM DTT (no ATP). Protein was eluted using a gradient of 1–50 mM glutathione in lysis buffer with 1 mM DTT and no ATP. For cleavage of the GST moiety, protein was desalted (PD10 column; GE Healthcare) into enzyme buffer (50 mM Hepes, pH 7.4, 150 mM NaCl, 5 mM MgCl, and 1 mM DTT) and cut overnight at 4°C with PreScission Protease (GE Healthcare). Protease and GST were removed using glutathione–Sepharose beads (GE Healthcare). ATPase assays of Spastin activity by the malachite green method are as described previously (). K and V were calculated by nonlinear regression analysis using the Prism 4 software package. Gel filtration chromatography was performed using a Superdex 200 column (bed volume = 24 ml) attached to an FPLC (GE Healthcare) system that was run at 0.4 ml/min. Typically, 50–100 μg Spastin in a volume of 100 μl was centrifuged (TL100 rotor [Beckman Coulter]; 50,000 rpm, 10 min, 4°C) immediately before analysis. 0.4-ml fractions were collected. The column was calibrated with the size standards indicted in the figures. Typically, the buffer composition was 50 mM Hepes, pH 7.3, 100 mM NaCl, 5 mM MgCl, and 1 mM DTT. Where indicated in the figures, the column buffer and the sample contained 1 mM ATP or 0.1 mM ATPγS. Sometimes Spastin was concentrated before chromatography using Amicon concentrators blocked with 5% Tween-20. Sedimentation velocity analysis was performed with an Optima XL-I (Beckman Coulter) in aluminum-filled epon double-sector cells at 20°C and 50,000 rpm. Interference scans were collected at 45-s intervals. Four dilutions of E442Q Spastin, ranging from 0.94 to 0.08 mg/ml, were prepared in 50 mM Hepes, 150 mM NaCl, 5 mM MgCl, 1 mM DTT, and 1 mM ATP. The sedimentation of macromolecules in a centrifugal field is governed by the Lamm equation:where is the concentration, is time, is the radial distance, is the translation diffusion constant, is the sedimentation coefficient, and ω is the rotation velocity. The sedimentation coefficient and translation diffusion constant were obtained by analysis sedimentation velocity data. These parameters were then used to obtain the molecular weight using the Svedberg equation:where is the molecular weight, is the gas constant, is the temperature, is the partial specific volume of the macromolecule, and ρ is the solvent density. We used the program DcDt+ to analyze the association state of Spastin. This program computes the apparent sedimentation coefficient distribution function g(s*), where the peak of the function is related to the sedimentation coefficient and the width of the function is related to the diffusion constant. These parameters are obtained by nonlinear least squares fits to the g(s*) functions, and the molecular masses are then computed using the Svedberg equation. The data were analyzed using the time-derivative method () to obtain normalized * profiles with the DcDt+ analysis software (, ). The buffer density and the protein mass and partial specific volume were calculated using the SEDNTERP program (). Videomicroscopy, cell culture, transfections, and immunofluorescence were performed as described previously (). Cos-7 cells were transfected with Polyfect (QIAGEN) for 36–48 h before being fixed in methanol. Tubulin was stained with YL1/2 antibody. In general, Spastin was visualized as a YFP fusion, and tubulin was visualized using a Cy5-labeled secondary antibody. All images were obtained with a microscope (Axiovert 200; Carl Zeiss MicroImaging, Inc.) using a 63× 1.6 NA plan-apochromat oil-immersion lens. Images were acquired with a CE camera (Roper Scientific) using MetaMorph software (Universal Imaging Corp.) and processed with Photoshop (Adobe). Site-directed mutagenesis of the YFP-Spastin plasmid was achieved using the QuikChange kit. The centrifugation or spin-down assay for MT severing is essentially as published previously (). For the tubulin tail competition experiment, 25-μl assays contained 0.5 μg Spastin, 1.0 μg taxol-stabilized MTs, and 8.6 μg GST–tubulin tail peptide. The GST–tubulin tail construct encodes GST fused to the C-terminal 104 amino acids of murine α-tubulin. Recombinant protein production was essentially as described for GST-Spastin. Although this protein has Tyr as the ultimate amino acid, we also produced a similar peptide corresponding to detyrosinated tubulin that ends in Glu. These plasmids were the gift of G. Gundersen (Columbia University, New York, NY). 8.6 μg of Spastin was added to the aforementioned GST–tubulin tail peptides (6.0 μg) immobilized on glutathione–Sepharose in a total volume of 100 μl (50 mM Hepes, pH 7.3, 150 mM NaCl, 5 mM MgCl, and 1 mM DTT) with or without 1 mM ATP, as indicated in the figures. After 30 min at 25°C, beads were washed three times with buffer supplemented with 0.1% Triton X-100 and either lacking or containing ATP. Next, SDS-PAGE sample buffer was added, and proteins were visualized by Coomassie staining after SDS-PAGE. EDC (Pierce Chemical Co.) was added to mixtures of proteins to 1–5 mM final from a 50 mM stock. Reaction buffer was 50 mM Hepes, pH 7.4, 150 mM NaCl, and 5 mM Mgl. After 15 min at room temperature, reactions were quenched by the addition of SDS-PAGE sample buffer. Samples were analyzed by Coomassie blue staining after SDS-PAGE. Fig. S1 shows EDC cross-linking between Spastin and MTs. Fig. S2 shows a detailed deletion analysis of the N-terminal MTBD. Fig. S3 shows that the C-terminal tail of tubulin is partially protease protected by the Spastin hexamer. Fig. S4 shows the biochemical characterization of Spastin lacking the MTBD. Online supplemental material is available at .

Mature resting B lymphocytes capture antigen (Ag) via their specific B cell receptor (BCR), which corresponds to a surface Ig coupled to a signaling module formed by the Igα/Igβ dimer (; ). In addition to Ag internalization, BCR stimulation triggers a complex cascade of signaling events that ultimately leads to the activation of B lymphocytes, which can then initiate the development of germinal centers. To complete germinal center formation, activated lymphocytes must process and present internalized Ag onto major histocompatibility complex (MHC) class II molecules to primed CD4 T cells, a process referred to as T-B cooperation (; ). It was recently shown that upon immunization, Ag-specific B lymphocytes are among the first lymphoid organ cells to acquire Ag and express the corresponding surface MHC–peptide complexes, highlighting the capacity of B cells to efficiently process and present BCR-internalized Ag onto MHC class II molecules in vivo (; ). MHC class II molecules assemble shortly after synthesis in the ER with a type II transmembrane protein, the invariant chain (Ii), which prevents their premature association with endogenous peptides (). In addition, Ii contains in its cytoplasmic tail the targeting signals that deliver MHC class II molecules into endocytic compartments for them to be loaded with antigenic peptides (; ; ). Such peptides result from the degradation of internalized Ag by endocytic proteases, which must also cleave Ii to free MHC II molecules for loading, a reaction catalyzed by the chaperone H2-DM (; ; ). Therefore, successful Ag presentation relies on its efficient targeting into endocytic compartments competent for processing (i.e., wherein it concentrates together with MHC class II, proteases, and H2-DM molecules). This corresponds to an essential function of the BCR: Ag captured through the BCR undergoes accelerated transport to endosomes and enhanced presentation efficiency as compared with Ag taken up by fluid-phase endocytosis (; ). Translocation of BCR–Ag complexes to lipid rafts as well as an intact actin cytoskeleton have been proposed to be essential for accelerated transport to endosomes (; ; ). In addition, this process is accompanied by substantial modifications in the endocytic pathway of B cells, as highlighted by studies using various mouse lymphoma cell lines (; ; ). In particular, we and others have shown that intracellular MHC class II molecules and BCR-internalized Ag converge into nonterminal LAMP-1–containing lysosomal compartments that display a multivesicular morphology and wherein Ag processing occurs, a process that depends on MHC class II–associated Ii (; ; ). The molecular mechanisms involved in the biogenesis of multivesicular endosomes have been documented, in particular by highlighting the importance of ubiquitylation in targeting membrane proteins to multivesicular endosome luminal vesicles (). The key role of Ag-triggered BCR ubiquitylation in directing Ag trafficking toward multivesicular lysosomes enriched for MHC class II was recently reported (). In addition, differential ubiquitylation of MHC class II β chain was shown to regulate its surface expression in immature versus mature dendritic cells (DCs; ). However, little information is available on the nature of the motor proteins that connect the vesicles carrying MHC class II molecules to the cytoskeleton, thereby helping their sorting to lysosome-like multivesicular compartments. This could involve microtubule- and/or actin-dependent forces, which are both known to control in concert the intracellular location and trafficking of organelles. We aimed to understand how BCR engagement in primary lymphocytes coordinates the transport of Ag- and MHC class II–containing vesicles for them to converge and ensure efficient Ag processing. In this study, we identify the actin- based motor protein myosin II as playing an essential role in this process. Myosin II is activated upon BCR engagement and becomes physically associated with MHC class II–Ii complexes. Myosin II inhibition or depletion impairs the concentration of MHC class II molecules together with BCR–Ag complexes into lysosomes devoted to Ag processing. Accordingly, cells lacking myosin II activity do not efficiently form MHC class II–peptide complexes from BCR-internalized Ag. Thus, myosin II regulates MHC class II trafficking and Ag processing in B lymphocytes. BCR stimulation of mouse B lymphoma cell lines triggers the appearance of MHC class II–containing compartments wherein BCR-internalized Ag is processed (; ). To explore the mechanisms underlying this process in primary B lymphocytes, we used MHC II–GFP knockin mice (). B cells purified from the spleen of MHC II–GFP mice were or were not BCR stimulated with a polyvalent BCR ligand (hereafter referred to as Ag) and analyzed by confocal microscopy. Resting B cells exhibited the peripheral distribution of MHC II–GFP molecules (), which was reminiscent of both surface and intracellular localization (in both lysosomes and the ER; see the next paragraph; not depicted). In sharp contrast, BCR engagement induced the redistribution of MHC II–GFP-containing vesicles that clustered near the cell center (). A substantial change in the nuclear shape of BCR-activated lymphocytes was also observed: whereas the nucleus of resting cells was round and filled most of the intracellular space, the vast majority of stimulated B cells displayed an important nuclear invagination where the MHC-GFP cluster sits (). Kinetics of MHC II–GFP clustering showed a peak at 60 min upon BCR triggering for ∼70% of the cells (). A similar observation was recently made using MHC class II–GFP knockin mice expressing a specific BCR and stimulated with a polyvalent Ag (). MHC class II/Ag clusters displayed the features of lysosomal-like Ag processing compartments: they stained positive for BCR–Ag complexes, LAMP-1, and H2-DM (; and Fig. S1 A, available at ) but contained reduced amounts of full-length Ii (Fig. S1 A). Accordingly, BCR stimulation induced the rapid cleavage of MHC II–bound Iip31 as indicated by the appearance of the Ii proteolytic intermediary fragments Iip25 and Iip10 (Fig. S1 B). No substantial change in the steady state levels of surface MHC class II was observed at any time period between 5 and 60 min upon BCR engagement (Fig. S1 C). Although this result does not exclude the possibility that MHC class II molecules are transported to lysosomes by trafficking through the cell surface, it indicates that the MHC class II/Ag lysosomal cluster does not result from massive BCR-induced endocytosis of surface MHC class II molecules. Ultrastructural experiments showed that MHC class II/Ag clusters corresponded to a network of vesicles and tubules, which included variable amounts of internal membranes enriched for MHC class II, LAMP-1, and H2-DM molecules ( and not depicted). No such compartment was observed in resting B cells (unpublished data). Therefore, we conclude that similar to mouse B lymphoma cells (; ), BCR-stimulated primary lymphocytes rapidly reorganize their endocytic route, accumulating lysosomal compartments that cluster near the center of the cell and to which MHC class II molecules and BCR–Ag complexes converge for processing. We next used time-lapse video microscopy to analyze the dynamics of BCR-induced clustering of MHC class II–GFP vesicles. Purified spleen MHC II–GFP B cells were or were not BCR stimulated and analyzed by 3D deconvolution time-lapse fluorescence microscopy. MHC II–GFP vesicles showed random movements in resting B cells and did not experience any substantial change in morphology during the duration of image acquisition (Video 2; available at ). In contrast, Ag-activated B lymphocytes exhibited major morphological changes during time-lapse (see Video 1 for general field and Video 3 for single cell). Strikingly, the plasma membrane of the vast majority of B cells underwent successive contraction events after BCR engagement ( and Videos 1 and 3). We consistently observed that cell contraction was coupled to MHC II–GFP clustering: MHC II–GFP clusters transiently associated with a contracted membrane portion and then moved toward the center of the cell (, arrowheads). Importantly, the centripetal movement of MHC II–GFP clusters occurred concomitantly to contraction arrest followed by cell spreading ( and Video 3). We conclude that the dynamics of MHC class II–containing vesicle clustering are coupled to BCR-induced cell contractility. Cell contractility is controlled by the actomyosin network and relies, in part, on phosphorylation of the regulatory myosin II light chain (MLC). MLC phosphorylation allows the activation of myosin II, which then initiates contraction by moving on actin microfilaments (). To investigate whether myosin II was activated upon Ag stimulation, we analyzed the levels of phosphorylated MLC in resting and activated spleen B lymphocytes. Immunoblot experiments performed with an antibody (Ab) that specifically recognizes phosphorylated MLC showed that BCR stimulation considerably increased MLC phosphorylation (). MLC phosphorylation showed kinetics that was consistent with its potential implication in the BCR-induced clustering of MHC class II vesicles, reaching maximal levels at 60 min upon BCR engagement (threefold as compared with nonactivated B cells). Analysis of the subcellular distribution of MLC showed that although resting cells exhibited homogenous peripheral MLC distribution, a 30-min BCR stimulation induced the concentration of MLC in contracted cell portions, as contractile rings were often appreciated (). Both MHC II–GFP and Ii were enriched in such contraction sites together with MLC molecules (, bottom). This finding strengthens the time-lapse results showing that before moving toward the center of the cell, clusters of MHC class II vesicles transiently associate with contractile membrane portions. Thus, BCR stimulation of resting mature lymphocytes triggers myosin II activation as indicated by MLC phosphorylation and redistribution to contracted membrane portions, where it colocalizes with MHC class II vesicle clusters. To demonstrate that myosin II–driven contraction is indeed functionally required for BCR-induced clustering of MHC class II– and Ag-containing vesicles, we first used pharmacological inhibitors. The inhibitors Y27632 and ML-7 target RhoA effector kinase (Rho-associated kinase) and MLC kinase, respectively (). Rho-associated kinase inhibits MLC-phosphatase and, thereby, indirectly promotes the levels of phosphorylated MLC, a task performed by MLC kinase (). Blebbistatin is a highly specific inhibitor of the ATPase activity of myosin II that has been used to demonstrate its involvement in cell division and migration (; ; ; ). Time-lapse experiments showed that when spleen B cells were pretreated for 45 min with these inhibitors before Ag stimulation, they underwent neither contraction nor cell spreading (Videos 4 and 5, available at ). Prevention of myosin II activation or inhibition of its activity strongly impaired the clustering of MHC class II– and Ag-containing lysosomes at the center of BCR-stimulated spleen B cells (). Only a few clusters were observed in drug-treated activated cells, and they seemed to contain reduced amounts of MHC class II, LAMP-1, and Ag molecules. In addition, these clusters remained dispersed at the cell periphery rather than concentrated at the center of myosin II–inhibited cells (). Quantitative analysis showed that although ∼75% of control Ag-stimulated cells displayed MHC II– GFP/LAMP-1/Ag central clusters, only ∼15% did when myosin II activation was hampered ( = 200; ). Impaired clustering did not result from a defect in Ag-induced BCR capping or internalization, which remained unaffected in the presence of myosin II inhibitors, excluding a general paralysis induced by drug treatment ( and Fig. S2, available at ). No effect of myosin II inhibition on MHC class II surface levels was observed by cytofluorometry (unpublished data). Thus, MHC class II– and Ag-containing vesicles do not converge and cluster together at the cell center when myosin II activity is compromised. Having shown that myosin II activity is required for the proper positioning of MHC II/Ag compartments, we next addressed by immunogold cryoelectron microscopy whether it also affects the maturation of these vesicles. The network of tubular and vesicular lysosomes wherein LAMP-1, H2-DM, Ag, and MHC class II molecules accumulate was barely observed in cells lacking myosin II activity (). Instead, drug-treated BCR-stimulated lymphocytes showed vacuolar compartments that labeled for LAMP-1, Ag, or MHC class II molecules at their limiting membrane (; inset). Consistent with immunofluorescence experiments, these vacuolar compartments were dispersed in the cytoplasm of drug-treated cells rather than clustered together, leading to the reduced concentration of MHC class II molecules and Ag in their lumen as compared with control BCR-stimulated lymphocytes (see for quantifications). Thus, the lack of myosin II activity affects the maturation of lysosomes to which MHC class II molecules and BCR–Ag complexes are targeted. The impact of myosin II inhibition on the maturation of MHC II/Ag lysosomes was further investigated by biochemical means using magnetic nanoparticles (NPs) coupled to anti-BCR Abs. As shown by , these NPs undergo proper BCR-mediated uptake and targeting to MHC class II–containing lysosomes (Fig. S3, available at ), allowing us to semipurify the endocytic compartments in which they accumulate (). We found that both the amounts of MHC class II and Ii were strongly decreased in endosomes purified from BCR-stimulated cells whose myosin II activity was compromised (). In contrast, levels of the early endosomal marker Rab5 and of lysosomal mature cathepsin D found in the magnetic fraction were not affected by the lack of myosin II activity (). In conclusion, both our morphological and biochemical results indicate that myosin II is required for the proper trafficking and maturation of MHC class II– and Ag-containing vesicles: it promotes their clustering in activated B lymphocytes and, thereby, allows them to concentrate together into a multivesicular and tubular lysosomal network. We next assessed whether myosin II activity is required for the presentation of BCR-internalized Ag using both B lymphoma cells and primary B lymphocytes. A20/DNP cells were pretreated with Y27632 and/or ML-7, further incubated with ovalbumin (OVA)-DNP (DNP-OVA), fixed, and cultured together with OVA-responding T cell hybridoma (). Pretreatment of A20/DNP cells with drugs that prevented myosin II activation or activity inhibited OVA presentation to CD4 T cells in a dose-dependent manner (). Equivalent results were obtained when targeting OVA protein to the BCR of spleen B cells by chemically cross-linking it to anti-IgM F(ab′) fragments (). No presentation of noncoupled OVA was detected under such experimental conditions, showing that the effect of the drugs indeed concerned the presentation of BCR-internalized Ag (). Importantly, the presentation of exogenously added OVA peptide was not affected by any drug treatment, demonstrating that the inhibition of Ag presentation resulted from impaired processing (). Pulse-chase experiments performed on Ag-stimulated spleen B cells not treated or treated with the drugs showed that synthesis, maturation, and processing of MHC II molecules were not compromised upon myosin II inhibition: equivalent amounts of SDS αβ–peptide complexes were immunoprecipitated from control and drug-treated cells. This indicates that the basal interaction of neosynthesized MHC II molecules with endogenous peptides does not require myosin II activity, whereas the processing of BCR-internalized Ag strictly relies on it (Fig. S4, available at ). Thus, the activity of the actin-associated motor myosin II is strictly required for BCR-driven Ag presentation. Is myosin II necessary for the formation of MHC II–peptide complexes from BCR-uptaken Ag, or is it required to export these complexes to the cell surface for interaction with T cells? To address this question, we took advantage of the mAb 2C44 that specifically recognizes the complexes formed between I-A molecules and the 156–173 peptide from Ag LACK but does not bind to any of the free components (). Recombinant LACK protein was targeted to BCR uptake by coupling it to the aforementioned NP together with anti-BCR Abs. The appearance of I-A–LACK complexes was analyzed by immunofluorescence using 2C44 at 2 and 4 h upon Ag internalization. At both time points, I-A–LACK complexes were mainly detected in lysosome clusters that labeled for LAMP-1 (). No complex formation was observed when using NPs only coupled to LACK or to anti-BCR Abs ( and not depicted). This assay was used to assess the requirement of myosin II activity on the formation of MHC II–peptide complexes. Blebbistatin treatment strongly reduced the percentage of 2C44 cells, indicating that myosin II activity is indeed necessary for the formation of I-A–LACK complexes (see for quantifications). Moreover, the few 2C44 blebbistatin-treated cells displayed staining in peripherally distributed lysosomes rather than in central clusters (). Impairment of I-A–LACK complex formation in myosin II–inhibited cells was further confirmed biochemically by showing that they contained reduced 2C44-reactive material after immunoprecipitation as compared with mock-treated lymphocytes (). To introduce genetic evidence in the involvement of myosin II in the formation of MHC II–peptide complexes, we performed depletion experiments using siRNA. RT-PCR analysis had shown that MyH9 was a major myosin II form expressed in mouse B lymphoma cells (unpublished data). Three myosin IIA (MyH9)–specific siRNAs were electroporated into B lymphocytes either individually or as a pool. MyH9 depletion was obtained with the three individual siRNAs and even more efficiently when using the siRNA pool (). The percentage of 2C44 cells was considerably reduced by MyH9 depletion (see for quantifications), demonstrating that the absence of myosin II severely compromised the formation of I-A–LACK complexes from BCR-internalized LACK. As expected, MyH9 depletion reduced the frequency of central lysosome clusters (). In addition, certain MyH9-depleted cells exhibited peripheral LAMP-1 patches similar to the ones observed in stimulated primary B cells whose myosin II activity was compromised (, arrowheads). Therefore, we conclude that myosin II is essential for the efficient formation of MHC II–peptide complexes from BCR-internalized Ag. Having established that myosin II activity regulates the trafficking of MHC class II molecules upon BCR engagement, we next investigated whether these proteins physically associate. The trafficking of MHC class II molecules mainly relies on the cytosolic portion of its chaperone-associated molecule, Ii, which is essential for the processing and presentation of BCR-internalized Ag (). Thus, we raised the hypothesis that Ii may physically link MHC class II–containing vesicles to myosin II through its cytosolic tail. In support of this hypothesis, B cells stimulated for short time periods showed a strong Ii staining in the contracted membrane portions and the contractile rings to which MLC and MHC class II molecules were recruited () as well as Ii/MLC colabeling in membranes from ER, Golgi, and endocytic vesicles when analyzed by immunogold cryoelectron microscopy (Fig. S5, available at ). Moreover, MLC was pulled down in the aforementioned B lymphoma cell line by using anti-Ii Abs in coimmunoprecipitation experiments (). MHC class II molecules were also found in anti-Ii immunoprecipitates. Strikingly, MLC–Ii complexes were retrieved from BCR-stimulated cells only. The interaction was found to be transient, and its kinetics was compatible with the one observed for MHC II vesicle clustering, reaching maximal levels at 60 min and then decaying. In agreement with this observation, the amounts of Ii-associated actin increased with similar kinetics as well (). MLC was also retrieved when using anti-MHC II Abs for the immunoprecipitation (). Importantly, the inhibition of myosin II activity with blebbistatin reduced its ability to associate with MHC class II–Ii complexes (). These results show that MHC II–Ii complexes and myosin II dynamically associate upon BCR engagement and further indicate that BCR-triggered myosin II activity is necessary for this association. Are both Ii and MHC class II required for the association to myosin II? Because insufficient protein amounts are generated from mouse spleen B cells, we used DCs purified from either Ii- or MHC II (I-Aβ)–deficient mice to address this question. Indeed, a 30-min lipopolysaccharide (LPS) stimulation of DCs induces the dynamic association between MHC class II–Ii and myosin II, as observed in BCR-activated cells (unpublished data). We found that the interaction between MHC class II and MLC was completely abolished in Ii-deficient cells (). This equally applies to myosin II heavy chain (MyH9) as observed when immunoprecipitating Ii and analyzing the coimmunoprecipitated material by immunoblotting or by an unbiased detection method (Coomassie staining followed by mass spectrometry analysis). Indeed, MyH9 is the major protein retrieved in this experiment, and its levels are comparable with those of immunoprecipitated Ii (). Strikingly, Ii retained the capacity to associate to MLC in I-Aβ knockout cells, although less efficiently (). No difference in the expression levels of MLC or myosin II heavy chain was observed in total extracts from I-Aβ– or Ii-deficient cells (). Together, these results show that Ii is essential for MHC II–Ii–myosin II interaction and further suggests that this association links MHC class II–containing vesicles to the actin network. In support of this conclusion, we found that Ii-deficient BCR-stimulated lymphocytes exhibited the peripheral distribution of MHC II–, LAMP-1–, and Ag-containing vesicles instead of central clusters, which is similar to cells whose myosin II activity was compromised (). Two nonexclusive hypotheses can be raised to explain the lack of MHC II–myosin II association in Ii-deficient cells: it could result from MHC II mislocalization, which accumulates in the ER when Ii is missing (; unpublished data), or the requirement for Ii cytosolic tail to form a complex with myosin II. The fact that Ii retains its ability to associate to myosin II in the absence of MHC class II molecules argues in favor of the second hypothesis. To further analyze the involvement of Ii cytosolic tail in the association with myosin II, we took advantage of the specific inhibitor of cathepsin S, LHVS, which prevents the removal of Ii cytosolic tail from MHC II–Ii complexes that have reached LAMP-1 compartments (; ). LHVS-treated cells displayed modified kinetics of MHC II–myosin II association: although 60 min upon BCR activation, the retrieval of myosin II with anti–MHC II Abs was diminished as compared with untreated cells, the association was still maintained 120 min upon BCR engagement (). This result shows that the presence of the cytosolic tail of Ii on MHC II–Ii complexes in lysosomes regulates their ability to associate with myosin II. We conclude that the lack of association of MHC II to MLC in Ii- deficient cells likely reflects the direct involvement of Ii in the formation of this protein complex rather than from the mislocalization of MHC class II molecules. Our data further indicate that MHC II–Ii–myosin II association can be modified by inhibiting myosin II activity with blebbistatin as well as by altering the processing of MHC II–Ii complexes with LHVS, providing strong evidence in favor of a direct role for myosin II in regulating MHC class II trafficking. #text I-Aβ–GFP knockin (referred to as MHC II–GFP), I-Aβ knockout, and Ii knockout mice were previously described (; ; ). The mouse lymphoma cells A20/DNP expressing surface IgM anti-DNP (), IIA1.6 (FcγR-defective variant of A20), and the DO54.8 T cell hybridoma were cultured as reported previously (). DCs were differentiated from mouse bone marrows cultured during 12 d in a granulocyte/macrophage colony-stimulating factor–containing medium as described previously (). The following Abs and inhibitors were used: rat anti–mouse LAMP-1 (BD Biosciences), rabbit anti-Igα (), rabbit anti-DM (provided by L. Karlsson, Johnson & Johnson Pharmaceutical Research and Development, San Diego, CA), rabbit anti–full-length Ii (JV11; ), rat anti-Ii N terminal (In-1; BD Biosciences), rat anti–I-A (M5114; ), rabbit anti–I-Aβ (), rabbit anti-Rab5 (provided by J.P. Gorvel, Universite de la Mediterranee Parc Scientifique de Luminy, Marseille, France), goat anti–cathepsin D (Santa Cruz Biotechnology, Inc.), mouse anti-RhoA (provided by P. Chavrier, Centre National de la Recherche Scientifique/Institut Curie, Paris, France), rat anti–α-tubulin (Serotec), rabbit anti-MLC and rabbit antiphospho-MLC (Thr18/Ser19; both were obtained from Cell Signaling Technology), rabbit anti–myosin IIA (Abcam), mouse antiactin (MP Biomedicals), Cychrome-conjugated anti-B220 (BD Biosciences), phycoerythrin-conjugated anti-IgM (BD Biosciences) and phycoerythrin-conjugated anti–I-A (BD Biosciences), ML-7 and Y27632 (Calbiochem), blebbistatin (Tocris), and LHVS (provided by H. Ploegh, Whitehead Institute, Cambridge, MA). A single-cell suspension was generated by the mechanical disruption of spleens from 8–12-wk-old mice, and resting mature IgM/IgD B cells were purified by negative selection (Miltenyi Biotec). Cell purity was 80–90% as assessed by flow cytometry using anti-IgM, IgD, and B220 Abs. For activation, B cells (10 cells/ml for biochemical assay and 5 × 10 B cells for immunofluorescence) were stimulated using multivalent BCR ligands: 10 μg/ml F(ab′) goat anti–mouse IgM (Cappel) for primary spleen B cells or 10 μg/ml F(ab′) goat anti–mouse IgG (MP Biomedicals) for IIA1.6 mouse lymphoma cells plus 20 μg/ml F(ab′) donkey anti–goat (Jackson ImmunoResearch Laboratories). For BCR activation with NP (8-nm diameter; FeO; provided by J. Roger, Université Paris 6, Centre National de la Recherche Scientifique, Paris, France), 10 μg/ml F(ab′) goat anti–mouse IgG or IgM was mixed with 10 μg/ml of recombinant LACK protein and a 3.2 vol of NP. B cells plated on poly--lysine–coated glass coverslips were fixed in 4% PFA for 20 min at RT, and PFA was quenched in PBS plus 1 mM glycine for 10 min. Fixed cells were incubated with Abs in PBS plus 0.2% BSA and 0.05% saponin. For detection of I-A–LACK complexes, biotinylated 2C44 and streptavidin-546 (Tyramide Signal Amplification kit; Invitrogen) were used. Immunofluorescence images were acquired on a confocal microscope (LSM Axiovert 720; Carl Zeiss MicroImaging, Inc.) with a 63× 1.4 NA oil immersion objective (Carl Zeiss MicroImaging, Inc.). For video microscopy, B cells attached on poly--lysine–coated slides were incubated in a Ludin chamber at 37°C in the absence or presence of BCR multivalent ligands. Fluorescence 3D + time images were acquired every 2–5 min during ∼3 h on an inverted fast 4D deconvolution microscope (DMIRB2; Leica) using a PL APO HC 1.4 NA oil immersion objective (Leica). It was equipped with a cooled interline CCD detector (CoolSNAP HQ; Roper) with a pixel size of 6.45 × 6.45 μm, 12 bits of dynamics, and a read-out speed of 20 MHz. Z positioning was accomplished by a piezoelectric driver (10-nm precision and 40-nm repetitiveness; LVDT; Physik Instrument) mounted beneath the objective lens. Illumination was provided by a fast wavelength switcher (DG-4; Sutter Instrument Co.). We used a 2 × 2 binning and a z distance between planes of 0.3 μm, giving a voxel size of 129 × 129 × 300 nm, which is compatible with the deconvolution process. Images were deconvolved with the MetaMorph (Universal Imaging Corp.) point-spread function–based iterative constrained modified gold algorithm. Films were reconstructed using MetaMorph 6.2 software. 1–3 × 10 of purified spleen B cells not pretreated or pretreated with myosin II inhibitors were stimulated with multivalent BCR ligands (see Preparation and stimulation of B cells section) and processed for immunoelectron microscopy as previously described (). OVA was coupled to DNP (). Monovalent F(ab) anti–mouse IgM was prepared by reducing 4 mg F(ab′) goat anti–mouse IgM with 1 mg Mesna (Pierce Chemical Co.) and was incubated with 8 mg maleimide-activated OVA (Pierce Chemical Co.). Fractions were purified on a Sephadex 75 column (GE Healthcare) and purity tested by immunoblotting. Spleen B cells from MHC II–GFP mice (I-A haplotype) or A20 anti-DNP cells (I-A haplotype) not pretreated or pretreated with myosin II inhibitors were used for stimulation of the B097.10 (I-A) or DO54.8 T cell hybridomas (I-A) as previously described (). For immunoprecipitations, IIA1.6 cells were not pretreated or pretreated with inhibitors for 1 h at 37°C followed by activation with multivalent BCR ligands. Alternatively, day 12 bone marrow–derived DCs treated with LPS during 30 min were used. Cells were lysed (100 mM Tris, 300 mM NaCl, 0.5% NP-40, and 5% glycerol plus protease cocktail inhibitors), and 900 μg of cell lysates was precleared with rabbit and mouse nonimmune sera and/or protein G–coated Sepharose beads. Ii and MHC class II molecules were immunoprecipitated using the In-1 rat mAb or M5114 rat mAb, respectively. Samples were washed, resuspended in reducing laemmli buffer, boiled, and loaded onto a 12% SDS-PAGE gel. Proteins were transferred to polyvinylidene difluoride membranes (Immobilon-P; Millipore), and membranes were incubated with the appropriate Abs and revealed with ECL (GE Healthcare). Alternatively, gels were directly stained with colloidal Coomassie blue (Bio-Rad Laboratories). Pulse-chase experiments were performed as previously described (). In brief, spleen B cells (20 × 10 cells for each time point) were starved in methionine/cysteine-free RPMI medium for 45 min at 37°C, pulsed with 0.5 mCi [S]methionine for 30 min at 37°C, further chased for different time points, and lysed, and cell extracts were immunoprecipitated using anti–MHC II Ab. 50 × 10 primary B cells were activated with magnetic NP coupled to anti-BCR Abs for 1 h at 37°C, washed with PBS, and resuspended in cold homogenization buffer (3 mM imidazol, 8% sucrose, 1 mM DTT, 1 mM EDTA, and protease inhibitors). Cells were broken using a cell cracker (), nuclei and intact cells were removed by centrifugation at 750 for 10 min, and postnuclear supernatants were collected. NP intracellular compartments were semipurified by incubating postnuclear supernatants on a magnet O/N at 4°C. Nonmagnetic and magnetic fractions were recovered. Equal protein quantities of the magnetic fraction were analyzed by SDS-PAGE. 4 × 10 IIA1.6 cells were electroporated using nucleofactor R T16 (Amaxa) in the presence of 20 nM siRNAs, SMARTpool siRNA, or siGENOME ON-TARGETplus (#1, 5′-GGGCUUAUCUACACCUAUUUU-3′; #2, 5′-AYAAGAACCUGCCCAUCUAUU-3′; #3, 5′-GCAGACAAGUACCUCUAUGUU-3′) specific for the gene or negative control duplex siRNA (5′-UUCUCCGAACGUGUCACGUTT-3′; QIAGEN). Cells were cultured for 96 h at 37°C and analyzed by immunoblotting or immunofluorescence. Videos 1 and 3 show that B cells undergo cell contraction and MHC II–GFP clustering upon BCR stimulation. Video 2 shows the behavior of nonstimulated purified B lymphocytes from MHC II–GFP mice. Videos 4 and 5 show that treatment of BCR-stimulated cells with inhibitors of MLC phosphorylation prevents cell contraction and MHC II–GFP clustering. Fig. S1 A shows that full-length Ii is weakly detected in central lysosome clusters, whereas H2-DM is enriched in these compartments. Fig. S1 B shows that the proteolysis of Ii is induced upon BCR stimulation. shows that there is no variation in the steady state levels of MHC class II surface expression at various times after BCR stimulation. Fig. S2 shows that BCR capping is not modified in B cells whose myosin II activity is inhibited. Fig. S3 shows a characterization of the method used to semipurify the endosomes to which BCR–Ag complexes are targeted. Fig. S4 shows that the formation of MHC II–peptide complexes from endogenous proteins is not affected by the inhibitors of myosin II activity. Fig. S5 shows that Ii and MLC colocalize on ER, Golgi, and endosomal membranes. Online supplemental material is available at .

Rap1 has been implicated in the control of cell adhesion in a variety of cell types (; ; ; ). In mammalian cells, Rap1 controls cell spreading by mediating the functions of integrins, by binding to and localizing the Rac exchange factors (guanine nucleotide exchange factors [GEFs]) VAV1 and TIAM1 to the sites of cell spreading, and by regulating cadherin-mediated cell–cell contacts (; ; ; ; ; ). Rap1 is thought to control cell adhesion and the leading edge function of moving cells by interacting with and regulating adaptor proteins that help control the cytoskeleton, although there is no direct evidence of a regulatory role for Rap1. RIAM and lamellipodin are related members of the MRL family of adaptor proteins containing a Ras association (RA) Rap1-GTP–binding domain, a PIP-binding pleckstrin homology (PH) domain, and a polyproline domain that interacts with Ena–vasodilator-stimulated phosphoprotein and profilin. RIAM1 induces cell spreading and lamellipodia formation and localizes to the sites of membrane protrusion (), whereas lamellipodin controls leading edge formation and filamentous actin (F-actin) polymerization in PDGF-treated fibroblasts (). Other Rap1 effectors include the following: affidin, which is associated with cell junctions and binds ZO-1, nectins, and profilin and, thus, may provide a link between Rap1 and the regulation of F-actin filaments (); RapL, which binds LFA-1 (lymphocyte function–associated antigen 1) and regulates LFA-1 localization in a Rap1-GTP–dependent manner (); ARAP3, which contains an Arf6 GTPase-activating protein, a RhoA GTPase-activating protein domain, and several PH domains and affects PDGF-induced lamellipodia formation (); and PDZ-GEF, a Rap1-GTP–binding Rap1-GEF (; ). In , Rap1 has been linked to cytoskeletal regulation, phagocytosis, and the response to osmotic stress (, ; ). The overexpression of wild-type Rap1 leads to cell spreading and reduces the contraction of cells in response to treatment with azide. The reduction of Rap1 expression suppresses osmotic stress–induced cyclic guanosine monophosphate (cGMP) production and impairs cell viability, whereas the expression of constitutively active Rap1 enhances cGMP production (). Osmotic stress causes the activation of Rap1 with kinetics paralleling those of cGMP production. Rap1 was implicated in phagocytosis by an analysis of the putative Rap1 effector Phg2, a kinase with an RA domain that preferentially binds Rap1-GTP but not Ras-GTP (). This -null strain exhibits defects in particle binding during phagocytosis, cell substratum adhesion, and motility. The mechanisms by which Rap1 controls cytoskeletal reorganizations in are not understood. To investigate this, we analyzed the role of Rap1 in cell adhesion and chemotaxis and found that Rap1 controls myosin II assembly and disassembly. We demonstrate that Rap1 is activated at the cell's cortex and preferentially at the leading edge of chemotaxing cells, the site at which Phg2 localizes. Cells expressing constitutively active Rap1 are unable to effectively regulate myosin II assembly and disassembly. We determined that Phg2 is required for myosin II phosphorylation in an in vitro assay. Our findings suggest a model in which localized recruitment to and activation of Phg2 at the leading edge by Rap1-GTP of chemotaxing cells are required for proper myosin II phosphorylation and disassembly at the newly formed pseudopod. We suggest that this leading edge activation of Rap1 regulates cell adhesion and, during chemotaxis, helps establish cell polarity by locally modulating myosin II assembly and disassembly. has a single Ras subfamily member that is homologous to human Rap1 (). We investigated whether Rap1 plays a role in chemotaxis by examining kinetics of the chemoattractant (cAMP)-mediated activation of endogenous Rap1 and myc-tagged wild-type, constitutively active (Rap1), and dominant-negative Rap1 (Rap1) expressed in cells. We used a pull-down assay using the human RalGDS Rap1-GTP–binding domain, which shows high but not exclusive specificity for the GTP-bound form of Rap1 and only poorly binds Ras-GTP in crude extracts in this assay (; see Materials and methods; ; ). shows that endogenous Rap1 was rapidly activated in response to stimulation, with a peak at 5–10 s. Myc-Rap1 (Rap1) expressed in wild-type cells displayed a similar activation profile (). There was a high level of myc-Rap1 in the GTP-bound form in unstimulated cells, and the level did not change on stimulation. Myc-Rap1, which was expressed at approximately the same level as myc-Rap1 and myc-Rap1, did not exhibit binding before or after stimulation. The relative (normalized) level of endogenous Rap1-GTP (level of Rap1-GTP compared with total Rap1 as determined by Western blotting) was slightly reduced in cells expressing Rap1. We investigated Rap1 localization using GFP-Rap1 and myc-Rap1. GFP-Rap1 exhibited the same activation kinetics as endogenous Rap1 or myc-Rap1, indicating that it was biologically active (unpublished data). GFP-Rap1 and myc-Rap1 localized predominantly to intracellular membranes, including the ER and endosomal membranes, which is similar to observations in mammalian cells (; ; ; ). In addition, a portion of GFP-Rap1 was present on the plasma membrane. This is more readily seen after removal of the soluble cytoplasmic fraction (). Total GFP-Rap1 showed a similar localization in chemotaxing cells and is absent from the domain immediately posterior to the leading edge, presumably because the endosomal membrane fraction is excluded by the pseudopodial F-actin cortex (). The localization of total Rap1 is not detectably altered on global (uniform) chemoattractant stimulation (). GFP-Rap1 exhibited a similar subcellular localization (unpublished data). We created a YFP-RalGDS reporter to examine the spatial-temporal activation of Rap1 (). We observed a low level of Rap1-GTP (localized YFP-RapBD) at the plasma membrane before chemoattractant stimulation, which is consistent with a basal level of Rap1-GTP in these cells. In response to chemoattractant stimulation, Rap1-GTP levels at the cell cortex rapidly increase within 3 s of stimulation and peak at ∼6–8 s (). These Rap1 kinetics are slower than those of Ras (; ). In chemotaxing cells, Rap1-GTP is found predominantly at the leading edge and extends weakly along the sides of cells (). This domain of Rap1-GTP is broader than that observed for Ras-GTP (). No localization of the YFP-RapBD reporter is observed on vesicles and other intracellular membranes where the majority of Rap1 is found, suggesting that Rap1-GTP activation is restricted to the cortex. Cells overexpressing wild-type Rap1 (Rap1 cells) are more flattened and spread than wild-type cells as previously described (; , ). This phenotype is substantially enhanced in cells overexpressing Rap1 and reduced in cells overexpressing Rap1 (), suggesting that Rap1 may control cell-substratum attachment. This conclusion was confirmed using cell-substratum attachment assays, which measure the fraction of cells that detach from a membrane during agitation. Rap1-overexpressing cells exhibit a small increase in cell attachment (i.e., a decrease in cell detachment; and Fig. S1, available at ). The level of cell attachment is considerably greater for Rap1 cells. We were unable to obtain a Rap1 knockout, suggesting that Rap1 is essential. To examine the phenotype of cells exhibiting reduced Rap1-GTP levels, we expressed Rap1 in wild-type cells. Although expressing Rap1 has little effect on endogenous Rap1 activation (), Rap1 cells exhibit reduced attachment. Unconventional myosin VII and talin A form a complex and are required for adhesion of cells to the substratum (; ; ). We examined whether the overexpression of Rap1 could bypass the –null cell defects. Overexpressing Rap1 in –null cells results in cells with a strong substratum attachment similar to that of wild-type cells overexpressing Rap1. This suggests that either Rap1 lies downstream of myosin VII–talin A or Rap1 and myosin VII–talin A function in distinct pathways to control attachment and that Rap1 can promote attachment independently of talin A. Wild-type cells become polarized and chemotax to a micropipette emitting the chemoattractant cAMP with a high speed and chemotaxis index, an indicator of the directionality of movement (). Rap1 cells move at <50% of the speed of wild-type cells and have a substantially decreased chemotaxis index resulting from the production of lateral pseudopodia (and an increased number of turns) and the decreased formation of a prominent leading edge in the direction of the chemoattractant gradient ( and Fig. S2, available at ). Rap1 cells have a milder phenotype; Rap1 has a speed and directionality similar to that of wild-type cells, although with a reduced number of turns and lateral pseudopodia (). Myc-Rap1 and myc-Rap1 cells experience delayed multicellular development as a result of a delay in aggregation and morphogenesis. The aggregates are also smaller than those of wild-type cells (Fig. S3). In contrast, Rap1 cells develop more rapidly than wild-type cells. Wild-type cells rapidly round up and contract after treatment with azide (NaN) as described previously (; ; ). As shown in , cells overexpressing Rap1 do not change shape after NaN treatment, and, in agreement with a previous study, NaN-mediated contraction does not occur in cells lacking myosin II ( cells; ), suggesting an essential role for myosin II in this response. (A and B) indicates that ( null) cells exhibit a flattened, spread shape and a high degree of substratum attachment compared with wild-type cells. cells is further increased by overexpressing Rap1. These results suggest that Rap1 functions through myosin II and other pathways to control cell adhesion. Myosin II assembly in is regulated by the differential phosphorylation of three Thr residues (Thr, Thr, and Thr) in the myosin II tail by a family of myosin II heavy chain kinases (MHCKs; ; ; ; ). Unphosphorylated myosin II self-assembles, whereas myosin II is disassembled in vivo by phosphorylation by the MHCKs. Substitution of Thr, Thr, and Thr with alanines (3XAla) leads to overassembled myosin II, whereas Asp substitutions (3XAsp) cause constitutively unassembled myosin II. cells expressing myosin II behave similarly to cells: they exhibited high attachment, are spread, and do not change shape upon NaN treatment (). cells expressing myosin II are more rounded before NaN treatment and do not round further upon NaN treatment (), presumably because they are already contracted. These cells also exhibit less adhesion (). cells and cells expressing Rap1, we investigated the effect of Rap1 on the reorganization of the cytoskeleton in response to chemoattractant stimulation. In wild-type cells, chemoattractant stimulation results in a rapid, transient phosphorylation of myosin II, causing a small but reproducible drop in assembled myosin II in the Triton X-100 pellet after ∼5–10 s (). This response is followed by an ∼2.0–2.5-fold increase in cortical myosin II that peaks at ∼40 s (; ; ). Myosin II assembly is not visibly affected by overexpressing wild-type Rap1. However, the overexpression of Rap1 causes a major alteration in chemoattractant-stimulated changes in assembled myosin II. After the small decrease in cortical myosin II levels, there is only a slight and gradual rise in assembled myosin II that reaches a first plateau of ∼20% of the level of myosin II in wild-type cells at 40 s, the time when assembled myosin II levels peak in wild-type cells (). There is a secondary, slow rise in assembled myosin II after 60 s, reaching a level of ∼50–60% of that of assembled myosin II in wild-type cells at 120 s. The response in wild-type cells returns to near basal levels by 80 s. cells, which complements cell defects, to examine the changes in cortically localized myosin II in resting cells and in response to chemoattractant stimulation (; ). Unstimulated cells exhibit a basal level of GFP–myosin II at the cell cortex (). Upon chemoattractant stimulation, after a lag of a few seconds, there is a decrease in GFP– myosin II at the cortex with a minimum at ∼10 s followed by an increase to above basal levels with a broad peak centered at 25–30 s and a subsequent decrease to basal levels. Rap1 cells have a similar response. As with the assay that quantifies myosin II in the Triton X-100–insoluble fraction, Rap1 cells exhibit an abnormal GFP–myosin II response (). At 5 s after stimulation, the level of cortical GFP–myosin II decreases substantially. Although a minimum is reached at ∼10 s in wild-type cells, the level of cortical GFP–myosin II continues to decrease in Rap1 cells with a low point at 15–20 s. Levels increase only slightly through 45 s. These data are consistent with a delayed and weak chemoattractant-mediated cortical accumulation of myosin II in cells overexpressing Rap1. Our findings suggest that Rap1 directly or indirectly negatively regulates myosin II assembly at the cell cortex. We used GFP–myosin II to study the spatial localization of myosin II in unstimulated vegetative cells and in developed cells chemotaxing to a micropipette. Wild-type cells have some cortically localized GFP–myosin II, and some cells have small patches of myosin II (). In contrast, vegetative Rap1 cells are highly polarized, with one pole exhibiting a broad domain of GFP–myosin II concentrated along the cortex. The other pole has a ruffled front similar to what is sometimes observed at the leading edge of chemotaxing cells. We found that although wild-type cells have a very low level of myosin II (endogenous myosin II or GFP–myosin II) along the bottom of cells (), Rap1 cells have a highly polarized myosin II localization along one pole of the cell. In chemotaxing wild-type cells, myosin II localizes to the posterior and along the lateral sides toward the cell's posterior as reported previously (). In Rap1 cells, much of the myosin II is found in a highly localized region at the very posterior of cells. These findings suggest that Rap1 may be involved in spatially restricting myosin II away from the leading edge in polarized cells. Wild-type cells exhibit a biphasic F-actin polymerization profile with a sharp peak at 5 s and a second, lower, and broader peak at ∼45–60 s that has been linked to pseudopod extension (; ). Overexpressing Rap1 results in an ∼40% increase in the basal level of F-actin. The kinetics of the response are similar to those of wild-type cells, with the level of F-actin proportionally higher in Rap1 cells as a result of the higher basal level of F-actin in unstimulated cells. The overexpression of wild-type Rap1 has little effect on the F-actin profile. The Phg2 Ser/Thr kinase was identified in a genetic screen for genes required for phagocytosis () and was identified independently by us in a yeast two-hybrid screen for proteins that interact with the PH domain–containing protein PhdB. Phg2 has an N-terminal PIP-binding domain, a Rap1-GTP–binding RA domain, and a Ser/Thr kinase domain and colocalizes with F-actin protrusions in vegetative cells (; ). We confirmed that the Phg2 RA domain preferentially binds Rap1-GTP over Ras-GTP (; unpublished data). We examined the subcellular localization of Phg2 using a GFP fusion (see Materials and methods). indicates that GFP-Phg2, which complements the -null phenotypes (not depicted), exhibits a low level of cortical localization in unstimulated cells. Upon chemoattractant stimulation, GFP-Phg2 rapidly and transiently translocates to the cell cortex. The level of cortical Phg2 peaks at ∼10 s. Phg2 localizes to the leading edge of polarized chemotaxing cells, with the localization extending weakly toward the posterior (). Thus, Phg2 shows a temporal and spatial localization similar to that of Rap1-GTP. This contrasts with Ras-GTP and PH domain proteins (e.g., CRAC and Akt/PKB) that are highly restricted to the leading edge and are absent from the lateral sides of chemotaxing cells (). To determine whether the RA domain is required for the leading edge localization of Phg2 in chemotaxing cells, we created GFP-Phg2ΔRA, a Phg2 construct lacking the RA domain. In chemotaxing cells, the localization of GFP-Phg2ΔRA was similar to that of GFP-Phg2, indicating that the RA domain is not required for Phg2 localization (). We also studied the possible mechanisms by which Phg2 is recruited to the cortex. The localization is unaffected in cells treated with LY294002 and in cells lacking PI3K1 and PI3K2 (-null cells) nor is localization affected by pretreatment with the actin inhibitor latrunculin A (unpublished data). These results suggest that Phg2 localization is not controlled by the PI3K or F-actin pathways that mediate the leading edge localization of several signaling components involved in the formation of a stable leading edge (). To investigate the function of Phg2, we created a -null strain in our wild-type strain KAx-3 (see Materials and methods; unpublished data). -null cells in either HL5 axenic growth medium or phosphate buffer are very flat and spread compared with wild-type cells (and are not multinucleated) and, like Rap1 cells, often have large lamellipodia-like structures on one side of the cell (Rap1 cells are shown as a comparison; ). The cells often have numerous and more prominent filipodia-like structures compared with wild-type cells, which are not observed in Rap1 cells. The phenotypes were complemented by expressing GFP-Phg2 but not GFP-Phg2ΔRA. Collectively with the localization of GFP-Phg2ΔRA, these data suggest that the RA domain and presumably Rap1-GTP are required for Phg2 function but not its localization. We created a second -null strain in which the Bsr selection cassette was inserted in the opposite orientation. Both -null strains had similar phenotypes that were complemented by expressing tagged wild-type Phg2 (unpublished data). Previous studies determined that -null cells exhibit defects in particle binding, cell attachment, and basal cell motility when assayed in HL5 axenic growth medium (). We confirmed a decrease in the rate of phagocytosis in our -null strain (unpublished data). Our -null strain exhibited a very strong increase in cell attachment compared with our parental wild-type cells ( A and S1). These phenotypes were complemented by the expression of tagged Phg2 in the -null strain. There are conflicting data concerning the cell adhesion of -null cells. found that their -null cells in a DH1 background exhibited ∼90% less adhesion compared with wild-type cells as assayed in a flow chamber. Using a plate-shaking assay, showed that the same strain had a mildly reduced adhesion. To confirm our data and to directly compare the two strains, we measured cell adhesion in the plate-shaking assay of (Fig. S1). These findings confirmed both our results and those of and demonstrated an intrinsic difference between the two -null strains and parental backgrounds. Our -null strain phenotypes are fully complemented by GFP-Phg2, indicating that the phenotypes result from the disruption of Phg2. Chemotaxing -null ( ) cells exhibit strong chemotaxis phenotypes. The cells move slower than wild-type cells, have a broader leading edge than wild-type cells, and produce numerous dominant lateral pseudopodia and more pronounced smaller projections from the leading edge (). These phenotypes are complemented by expressing GFP- or GST-Phg2 in -null cells. As shown in and Fig. S4 (available at ), F-actin levels were ∼1.5-fold higher in unstimulated -null cells compared with those in wild-type cells. Upon chemoattractant stimulation, -null cells showed only a small first peak of F-actin polymerization relative to wild-type cells, which had a 2.2-fold increase. A small second peak was observed at ∼40 s. This overall reduced level of F-actin polymerization might account for the reduced speed of -null cell chemotaxis. Unstimulated -null cells had ∼30% more myosin II in the Triton X-100 pellet compared with wild-type cells (). Upon stimulation, -null cells exhibited a small decrease in myosin II levels at ∼5 s and a subsequent peak at ∼40 s, which were changes similar to those in wild-type cells. To study the regulation of myosin II associated with the cortex in -null cells, we examined the spatial-temporal localization of GFP–myosin II in vivo. Myosin II does not delocalize from the cortex in response to chemoattractant stimulation (unpublished data). As myosin II dissociation depends on myosin II phosphorylation (), this observation suggested that Phg2 might regulate myosin II phosphorylation. demonstrated that the expression of Phg2 in led to the in vivo phosphorylation of bacterial proteins on Ser/Thr residues, indicating that Phg2 has Ser/Thr kinase activity. However, Phg2 purified from bacteria did not exhibit kinase activity under the conditions assayed. To examine Phg2 kinase activity, we expressed GFP- and GST-tagged Phg2 in -null cells and found that both proteins complemented the Phg2 chemotaxis and attachment defects. Triton X-100 lysates of cells before and at various times after chemoattractant stimulation were fractionated into soluble (10K supernatant) and insoluble (cortical) fractions, which were assayed for potential Phg2 activity. For the soluble fraction, we purified Phg2 (GFP-Phg2 by immunoprecipitation and GST-Phg2 on glutathione–Sepharose beads) and assayed for kinase activity using myelin basic protein or histone 2B as substrates. No activity was observed (unpublished data). We assayed the insoluble/cortical fraction similarly and detected no activity on exogenous substrates (unpublished data). We studied Phg2 activity indirectly in the cortical fraction by assaying kinase activity against endogenous substrates by incubating the Triton X-100–insoluble pellet with Mg–γ- [P]ATP and comparing samples from wild-type and -null cells. We observed a single [P]O-labeled band (of ∼250 kD) in samples from wild-type cells that was not observed in samples from -null cells (; the remainder of the gel is not depicted). The band was barely detectable in samples from unstimulated cells, strong in samples from cells stimulated for 10 and 20 s, and barely detectable in samples from later time points. As the band is not observed in -null cells, we suggest that Phg2 is required for phosphorylation of the ∼250-kD protein in this assay. The kinase activity is restored in -null cells expressing wild-type Phg2 but not GFP-Phg2ΔRA, suggesting that the RA domain is required for Phg2 activity and that Phg2 activity may be activated by Rap1-GTP. Our assay does not distinguish between Phg2 directly phosphorylating the ∼250-kD protein or being required for the activity of another kinase that directly phosphorylates myosin II. Coomassie blue–stained gels showed that the ∼250-kD phosphorylated protein comigrated with the myosin II band. cells and found that the band was absent (). cells. The phosphorylated band was then present, and its mobility corresponded to GFP–myosin II (), indicating that the phosphorylated protein is myosin II. This activity was unaffected by pretreatment with LY294002 or latrunculin A, which is consistent with our findings that Phg2 cortical localization is unaffected by the treatment of these drugs and indicating that these pathways were not dependent on the PI3K pathway or F-actin polymerization (). cells expressing myosin II or myosin II to examine whether Thr, Thr, or Thr might be the sites of myosin II phosphorylation. illustrates that neither of these substituted myosin II proteins are phosphorylated. Although these findings are consistent with the sites of Phg2-dependent myosin II phosphorylation being Thr, Thr, or Thr, we cannot exclude the possibility that Thr, Thr, or Thr are not the sites of phosphorylation and that the actual sites of phosphorylation are unavailable in myosin II and myosin II in our assay for an unknown reason. We also investigated myosin II phosphorylation in cells lacking three of the four MHCKs (/-/-–null cells), which exhibit an extensive myosin II overassembly (). indicates that myosin II phosphorylation is highly reduced in this strain, suggesting that the myosin II phosphorylation pathway requires MHCK. The overexpression of MHCK-A in -null cells suppresses the lack of myosin II phosphorylation in our assay (). The kinase activity is found in unstimulated as well as stimulated cells, with the activity increasing after stimulation. We find that as previously reported for wild-type cells (), MHCK-A localizes to the leading edge of chemotaxing -null cells, suggesting that the loss of myosin II kinase activity in -null cells is not caused by the inability to recruit MHCK-A (unpublished data). Phg2 has a Rap1-GTP–binding RA domain and, therefore, is expected to lie downstream from Rap1. Cells expressing Rap1G12V have an elevated level of myosin II phosphorylation that is also extended, which is consistent with increased and extended activation mediated by the activated form of Rap1 (). Rap1 cells are very flat and adherent to the substratum, which is consistent with findings in mammalian cells, indicating the involvement of Rap1 in controlling cell shape and attachment. These cells chemotax slowly and produce numerous lateral pseudopodia, probably in part because they cannot regulate their attachment to the substratum and restrict the activation of pseudopod formation to the side of the cell closest to the chemoattractant source. Rap1 cells have a modest reduction in attachment. The Rap1 phenotypes are weak, presumably because Rap1 does not effectively inhibit the Rap1 GEFs, as there is little change in the activation of endogenous Rap1 in these cells. Our studies suggest that one mechanism by which Rap1 controls cell spreading and attachment in acts at least partially through the regulation of myosin II assembly. Rap1 cells exhibit weak chemoattractant-mediated myosin II assembly, which is consistent with an inability to control cell contraction. In response to NaN treatment, wild-type but not Rap1 cells contract, which is consistent with previous studies (; ). cells expressing myosin II. These strains exhibit a spread, flat cell shape and a high substratum attachment. In Rap1 cells, GFP–myosin II dissociates from the cell cortex in response to chemoattractant stimulation but undergoes only a slight and greatly delayed reassociation with the cortex. Rap1 cells exhibit an abnormal and deeply reduced chemoattractant-mediated myosin II assembly with slow kinetics. These findings suggest that Rap1 negatively controls myosin II assembly. MHCKs are the major kinases that phosphorylate myosin II on Thr, Thr, and Thr and result in myosin II disassembly (; ). MHCK-A is recruited to the leading edge by binding to F-actin, which activates the kinase (), thereby disassembling myosin II at the site of leading edge protrusion. Our findings that Rap1 is activated at the leading edge and that the expression of Rap1 causes extended myosin II disassembly suggest that Rap1 may be a key upstream spatial regulator of myosin II assembly. We suggest that Rap1 may prevent myosin II assembly at the leading edge and may indirectly (passively) promote its assembly at the posterior of cells. Myosin II assembly and phosphorylation are regulated by cGMP in (, ; ). Cells lacking the cGMP-binding protein GbpC exhibit a spatial-temporal profile of GFP–myosin II cortical disassembly/assembly similar to that of Rap1 cells. The profile of myosin II in the Triton X-100–insoluble pellet is similar to that of cells lacking both guanylyl cyclases (, ). Thus, cells expressing Rap1 exhibit phenotypes similar to those lacking cGMP or the cGMP effector GbpC, suggesting that Rap1 may be a negative regulator of the cGMP pathway. Just before submission of this manuscript, a study by demonstrated that GpbD is a Rap1 GEF and that its overexpression leads to increased cell adhesion and chemotaxis defects similar to those of the Rap1 cell presented here. They present evidence that Rap1 lies upstream of Phg2. Our model by which Rap1 may control anterior and posterior functions may be analogous to the proposed mechanism by which anterior and posterior functions are differentially regulated by Gαi and Gα12/13 in neutrophils (). Gαi promotes the leading edge function and inhibits posterior function (activation of RhoA and assembly of myosin II), whereas Gα12/13 controls posterior functions by mediating RhoA activation. The loss of Gαi function increases RhoA activity, whereas the inhibition of RhoA function increases leading edge function, indicating that the regulation of anterior and posterior functions is balanced. In , anterior and posterior functions have different response kinetics and are spatially restricted in polarized cells (; ). Chemotaxing Rap1 cells often have a flat, extended posterior region that does not lift from the substratum, suggesting that Rap1 plays a part in controlling the posterior function of moving cells. In addition, the localization of myosin II is altered in unstimulated Rap1 compared with wild-type cells. Although myosin II has a general localization along the cortex except at sites of pseudopod extension with an enrichment at the posterior in wild-type cells (), resting Rap1 cells have a polarized cytoskeletal organization with myosin II present in a single, extended domain along approximately one third of the cortex opposite a domain that has the appearance of a projecting lamellipod. In chemotaxing Rap1 cells, myosin II is found almost exclusively in a small cap/domain at the posterior of cells, as though myosin II sites are restricted from other areas of the cell through the constitutive activation of Rap1. This lack of visible myosin II along the cell's lateral sides may contribute to a possible loss of myosin II–mediated cortical tension and the presence of lateral pseudopodia. We demonstrated that Phg2 is a key regulator of the cell's cytoskeleton and cell attachment. -null cells are highly adhesive, which is a finding that contrasts that of . We discovered that under two different assay conditions, their -null strain has increased adhesion in phosphate buffer, suggesting that differences may be caused by assay conditions. As -null cells exhibit high adhesion, we suggest that Phg2 negatively regulates cell adhesion. Phg2 is required for proper chemotaxis. Chemotaxing -null cells move slowly and have a broad leading edge with numerous small protrusions. The broad leading edge structure (more of a lamellipod than a pseudopod) and protrusions are presumably related to the broad lamellipod and filopodia-like protrusions exhibited by vegetative -null cells. These cells have reduced random motility (; unpublished data). Phg2 leading edge localization does not require the RA domain, but the ability to complement the null phenotypes does, indicating that the genetic function of Phg2 is dependent on Phg2's ability to bind Rap1-GTP. These findings demonstrate that Phg2 and, thus, its putative regulator Rap1 play key roles in controlling leading edge formation and in suppressing these lateral and adventitious protrusions. -null cells have a deeply reduced F-actin response, which, along with the high attachment of these cells, may account for their reduced chemotactic speed. A further understanding of the possible mechanism by which Phg2 controls the cytoskeleton was obtained by examining the physiological and biochemical phenotypes of -null cells. -null cells show an elevated basal level of myosin II and F-actin in vegetative and developmentally competent cells, which are normal kinetics of chemoattractant-mediated myosin II assembly, but show a substantially reduced F-actin polymerization response. Motile amoeboid cells (e.g., neutrophils and ) spatially restrict F-actin polymerization and leading edge protrusion from myosin II assembly to the opposite poles of polarized cells (; ; ). Myosin II thick filaments are partially responsible for cortical tension or rigidity. To protrude a pseudopod, cells must disassemble the myosin II thick filaments at that site. Our imaging data suggest that Phg2 is required for the disassembly of myosin II thick filaments; in the absence of Phg2, cells accumulate elevated levels of myosin II in the Triton X-100–insoluble fraction. We revealed that Phg2 is required for in vitro myosin II phosphorylation in a cortical fraction in response to chemoattractant stimulation. As this phosphorylation is thought to be responsible for myosin II disassembly from the cortex (), our combined data suggest that Phg2 is required for efficient cortical myosin II phosphorylation and disassembly. We suggest that the defect in delocalizing myosin II from the cortex that we observe in -null cells may be responsible for the restricted chemoattractant-mediated F-actin response. In Rap1 cells, myosin II disassembly is extended, and there is minimal reaccumulation over a 2-min time frame. The initial responses observed after a global stimulation recapitulate the changes that occur at the leading edge of chemotaxing cells. Like other leading edge responses, Rap1 is activated rapidly at the cortex, and Phg2 translocates to the cortex in response to global stimulation. Rap1 is also activated at the leading edge, and Phg2 accumulates at this site. We suggest that Phg2 is a new component of the rapidly activated leading edge regulatory network that, in this case, controls myosin II phosphorylation and disassembly at this site. , - , or cells and constitutive activity in MHCK-A cells suggest that this network may regulate MHCK-A (or other MHCKs), which localizes to the leading edge. The higher basal level of myosin II in -null cells is consistent with this hypothesis. Phg2 may play other roles, as -null cells have many cellular protrusions and extended lamellipodia in resting and chemotaxing cells, phenotypes that are not observed in cells lacking myosin II. -null cells exhibit a normal myosin II response when one examines total myosin II that associates with the Triton X-100–insoluble fraction and are able to contract upon NaN treatment. As such, we expect that -null cells can assemble and disassemble myosin II in other domains of the cell. In conclusion, Rap1 is a major regulator of the cytoskeleton in resting cells and cells stimulated by chemoattractant or responding to osmotic stress, and it functions, at least in part, to control myosin II through Phg2. Therefore, Rap1 may be part of the signaling cascade by which leading edge and posterior functions of the cell interact to control cell polarization and coordinate directional movement. We obtained latrunculin A, LY294002, aprotinin, and leupeptin from Sigma-Aldrich, anti-myc antibodies from Santa Cruz Biotechnology, Inc., glutathione–Sepharose beads from GE Healthcare, and γ-[P]ATP from MP Biomedicals. Anti-Rap1 antibodies have been described previously (). 13-mm nitrocellulose filters (0.22 μm) used for the adhesion assay were obtained from Millipore. The full coding sequence of the cDNA was generated by RT-PCR and was subcloned into the EcoRI–XhoI site of pSP72. An EcoRI restriction site in the 5′ end region of the gene was removed by substitution of cytosine at nucleotide 18 with thymine without changing the coding of amino acids. For GFP- or myc-Rap1 fusion proteins, the EcoRI–XhoI fragments of pSP72 were subcloned into the expression vector EXP-4(+) containing either a GFP or myc fragment. The mutants Rap1 and Rap1 were generated using the QuikChange Site-Directed Mutagenesis kit (Stratagene). Because the gene product contains two additional N-terminal amino acids compared with the human and gene products, amino acids were numbered according to the consensus alignment of Ras proteins to facilitate comparison (). GST-RalGDS–Ras-binding domain (RBD) was a gift from J.L. Bos (University Medical Center Utrecht, Utrecht, Netherlands). For RBD-RalGDS-YFP fusion protein expression, the RBD of RalGDS was amplified by PCR using GST-RB-RalGDS as a template, was ligated to a YFP fragment, and was cloned into the expression vector EXP-4(+). All clones were confirmed by DNA sequencing. A knockout construct was made by inserting the blasticidin resistance cassette into a BamHI site created at nucleotide 595 of the cDNA and was used for a gene replacement in the KAx-3 parental strains. Randomly selected clones were screened for a gene disruption by PCR, which was then confirmed by Southern blot analysis and RT-PCR. These studies confirmed that the knockout strains did not express Phg2 transcripts 5′ or 3′ to the site of insertion. A full-length sequence was obtained by PCR using a series of primers and was cloned into pBluescript, sequenced, and subcloned into EXP-4(+) plasmids for GFP or GST fusion protein expression. To create a GFP-Phg2 construct lacking the RA domain (GFP-Phg2ΔRA), two pairs of primers were used to amplify the sequence 5′ and 3′ to that encoding the RA domain, creating an XbaI site at the junction site. This created an in-frame deletion of the RA domain. This construct was then subcloned into GFP-fused pEXP4 plasmid. The GFP-Phg2ΔRA coding region was confirmed by sequencing. The -null mutant strain in the KAx-3 background and MhcA-GFP plasmids were provided by J.A. Spudich (Stanford University, Palo Alto, CA). The – and –null cells were obtained from M.A. Titus (University of Minnesota, Minneapolis, MN), and DH1-10 and -null cells in the DH1-10 background were obtained from P. Cosson (Universite de Geneve, Geneva, Switzerland). The myosin heavy chain phosphorylation mutant constructs 3XAsp and 3XAla were gifts from T.T. Egelhoff (Case Western Reserve University, Cleveland, OH). The Rap1-GTP–binding domain of mammalian RalGDS was expressed in as a GST fusion protein as described previously (). The purified GST-RalGDS-RBD was used for the detection of activated Rap1. Log-phase vegetative cells were washed twice and resuspended at a density of 5 × 10 cells/ml in Na/K phosphate buffer and pulsed for 6 h with 30 nM cAMP every 6 min. The cells were collected and resuspended at a density of 2 × 10 cells/ml and were treated with 1 mM caffeine for 30 min. The 300-μl aliquots were stimulated with 15 μM cAMP and lysed by mixing with an equal volume of 2× lysis buffer (100 mM Tris, pH 7.5, 300 mM NaCl, 50 mM MgCl, 20% glycerol, 1% NP-40, 2 mM DTT, 2 mM vandate, aprotinin, and 5 μg/ml leupeptin) at the indicated times. The lysates were centrifuged for 10 min, and the supernatants were incubated with 10 μg GST-RBD on glutathione–Sepharose beads at 4°C for 30 min in the presence of 1 mg/ml BSA. The beads were washed three times and subjected to SDS-PAGE and Western blot analysis with either an anti-myc mAb or an anti-Rap1 pAb. For control of the input amount of total Rap1 proteins, 40 μl of the cells were taken right after lysis of the cells and analyzed. Analyses of chemotaxis toward cAMP were performed as described previously (; ). The images of chemotaxing cells were taken at time-lapse intervals of 6 s for 30 min using an inverted microscope (TE300; Nikon) with a plan Fluor ELWD 40× NA 0.6 lens and a camera (CoolSnap HQ; Roper Scientific). The frames were captured using MetaMorph software (Molecular Devices), and the data were analyzed using DIAS software (Soll Technologies; ). Cells for analysis were randomly chosen with the requirement that they move for at least 15 min without touching another cell and were examined. Differential interference contrast images were hand traced (10 frames/min for 15–20 min). At least five randomly chosen cells were analyzed from each of at least three independent experiments performed on separate days. The quantitation of membrane or cortical localization of GFP fusion proteins was performed as described previously () with slight modification. Fluorescence images were obtained using a confocal microscope (DM IRE2; Leica) with HCX plan Apo NA 1.40 100× or 63× objective lenses (oil CS; Leica) and a camera (EM-CCD or ORCA-ER; Hamamatsu Photonics). Images were captured using SimplePCI software (Compix Inc., Imaging Systems) and were analyzed using ImageJ software (National Institutes of Health). Differential interference contrast images were taken with the same equipment but using an ORCA-285 camera (Hamamatsu Photonics). Adhesion assays were performed as described previously () with minor modifications. Log-phase growing cells on the plates were washed with Na/K phosphate buffer and resuspended at a density of 2 × 10 cells/ml. The amount of 4 × 10 cells in 200 μl were plated onto 13-mm circular nitrocellulose filters (Millipore). After 30 min, unattached cells were removed by dipping filters into Na/K phosphate buffer. Filters were transferred to microcentrifuge tubes filled with 800 μl Na/K phosphate buffer and were vortexed for 1 min with a mixer (model 5432; Eppendorf) simultaneously. 150 μl of the detached cells from the filters were plated onto a 30-mm Petri plate with a hole covered by a 0.17-mm glass coverslip, and an additional coverslip was placed on top. The cells were photographed and counted (detached cell number). To determine the total cell number, 200 μl of the cells was transferred into microcentrifuge tubes filled with 600 μl Na/K phosphate buffer and counted. Cell adhesion is presented as a percentage of detached cells compared with total cells. This experiment was repeated three times or more, each time with four filters for each strain. 5 × 10 cells/ml in Na/K phosphate buffer were pulsed and stimulated with cAMP as described in the PKB/Akt activity assay (). After stimulation with cAMP, the cells were lysed by mixing with an equal volume of 2× lysis buffer. The cytoskeleton fractions were isolated by centrifugation of the lysates at 4°C for 10 min. The pellets were washed once with 1× lysis buffer and then with kinase buffer (25 mM MOPS, pH 7.4, 25 mM β-glycerophosphate, 20 mM MgCl, and 1 mM DTT). The pellets were resuspended and incubated in 50 μl of kinase buffer containing 10 μCi γ-[P]ATP and 5 μM ATP for 20 min at 22°C. The reaction was stopped by the addition of 25 μl of 4× sample buffer followed by SDS-PAGE. The gels were stained with Coomassie blue, dried on a Hoefer slab gel drying apparatus, and exposed to film. F-actin polymerization and myosin II assembly were assayed as described previously (). Fluid-phase endosomes were labeled using 70,000- TRITC-dextran (Sigma-Aldrich). Log-phase cells expressing GFP-Rap1 proteins were incubated with 0.75 mg/ml TRITC-dextran for 1 h and were washed twice with Na/K phosphate buffer. The cells were observed using a confocal microscope (DM IRE2; Leica). Sodium azide treatment of cells was performed as described previously (). Vegetative cells were washed twice with MES buffer (20 mM MES, pH 6.8, 2 mM MgSO, and 0.2 mM CaCl) and oxygenated by bubbling for 20 min. The cells were allowed to adhere to glass coverslips for 10 min and were inverted onto a plate with a hole covered by a glass coverslip holding 150 μl MES buffer with or without 25 μM NaN. After 30 min of incubation, the cells were photographed. Fig. S1 shows that the -null strain created in the KAx-3 background exhibits strong attachment both in the plate-shaking method described in Kortholt et al. and in the cell attachment assay described in Materials and methods. Fig. S2 shows an increased number of turns and more frequent extrusion of lateral pseudopodia in cells expressing Rap1. Fig. S3 shows that the multicellular development of Rap1 and Rap1 cells is delayed. In contrast, Rap1 cells develop more rapidly than wild-type cells. Fig. S4 shows basal levels of F-actin and assembled myosin II in vegetative cells or in aggregation-competent cells. Online supplemental material is available at .

The phosphoinositide 3-kinase (PI3K) signaling pathway mediates a multitude of cellular responses after extracellular stimulation by peptide growth factors and hormones (). Deregulation of this pathway is associated with human diseases such as cancer and diabetes (). PI3K and its product, phosphatidylinositol 3,4,5-trisphosphate (PtdInsP), are key signaling molecules in cell motility, particularly in chemotaxis, which is a process involved in a wide range of cellular responses, including morphogenesis, wound healing, immune response, angiogenesis, and metastasis of tumor cells (; ; ). Upon chemoattractant stimulation, a PtdInsP gradient is created and maintained at the leading edge of cells with amoeboid motility, such as leukocytes and (). This process involves both localized accumulation and activation of PI3Ks, which generate PtdInsP/PtdInsP, and the phosphatase PTEN, which removes them (; ). PtdInsP serves as a docking site for a subclass of PH domain–containing proteins that are recruited at the leading edge. However, it is not clear which PI3K downstream effectors lead to activation of the actin polymerization machinery required for cell migration. Akt is one of the candidate molecules through which PI3K regulates chemotaxis, but the role of Akt in the control of cell polarity and chemotaxis has been established only in , where Akt phosphorylates PAKa, regulating its subcellular localization and myosin II assembly (). Unfortunately, PAK1, the mammalian homologue of PAKa, lacks the Akt phosphorylation site, suggesting the existence of a different signaling pathway. One important downstream effector of PI3K is the 3-phosphoinositide–dependent protein kinase-1 (PDK1; ). PDK1 phosphorylates and activates a group of related protein kinases belonging to the AGC family (). These include isoforms of Akt (), p70 ribosomal S6 kinase (S6K; ), p90 ribosomal S6 kinase (RSK; ), PKC (), and serum- and glucocorticoid-induced protein kinase (SGK; ). Evidence has indicated that PDK1 is constitutively active, and that regulation involves the conversion of substrates to forms that can be phosphorylated by PDK1 in agonist-treated cells (; ). For example, the phosphorylation of Akt by PDK1 is regulated by the conformational change induced by engagement of the PH domain to the membrane by PtdInsP/PtdInsP, which relieves autoinhibition of the active site, allowing PDK1 to access T308 on the activation loop (; ). PDK1 also contains a PH domain that binds PtdInsP with high affinity and has a crucial role in activation of Akt (; ). Other PDK1 substrates, such as S6K and SGK, which lack a PH domain and are phosphorylated by PDK1 at the same ratio in the presence or absence of PtdInsP, interact with a pocket in the kinase domain of PDK1, called the PIF-binding pocket (). The prior phosphorylation of S6K and SGK at their hydrophobic motif promotes their interaction with the PIF-binding pocket of PDK1 and their T-loop phosphorylation (). The key role that PDK1 plays in activating certain AGC kinase members was substantiated by the finding that mouse embryonic stem (ES) cells lacking PDK1 fail to activate Akt, S6K, and RSK in response to stimuli that trigger the activation of these enzymes in wild-type ES cells (). Unexpectedly, although Akt and RSK have been reported to play important roles in regulating survival and proliferation, ES cells lacking PDK1 were viable (). Nevertheless, PDK1 is required for normal embryo development, as mice embryos lacking PDK1 die at day E9.5 displaying multiple abnormalities, including lack of somites, forebrain, and neural crest–derived tissues (). PDK1 hypomorphic mice, in which a general and extensive reduction of PDK1 expression was obtained by intron insertion of a neomycin resistance gene, were viable and fertile, but were 40–50% smaller than control animals, and their organ and cell sizes were also proportionately reduced. Interestingly, activation of Akt and S6K1 by insulin was normal in the PDK1 hypomorphic mice, showing that regulation of cell size by PDK1 is independent of insulin's ability to activate Akt and S6K (Lawlor et al., 2002). Moreover, PDK1 knock-in mouse embryos, in which the PH domain was disrupted, die at embryonic day (E) 11.5 (). In these knock-in cells, Akt was not activated by IGF1, whereas RSK was normally activated, indicating that PtdInsP binding to PDK1 is required for Akt, but not RSK activation. However, the cause of death in PDK1 mice appears to be a lack of functional circulation. Because the inability to form a functional circulatory system might result from the inability of ECs to migrate, our initial approach in understanding the role of PDK1 in cell migration was to study the vessel formation in embryoid bodies (EBs) derived from ES cells lacking PDK1 and the motility of ECs differentiated from them. Moreover, to gain further insights into the role of PDK1, we modulated PDK1 activity in human ECs using retroviral vectors expressing PDK1 mutants. Our results suggest that both PDK1 and Akt are involved in EC motility. The starting point for our investigation was an examination of the role of PDK1 in blood vessel formation. We used a well-established model of early vascular plexus formation—the ES differentiation into EBs (; ). PDK1 mice die at E9.5 displaying multiple abnormalities, including the lack of a circulatory system (Lawlor et al., 2002). ES cells, in which both copies of the PDK1 gene have been disrupted, are viable and proliferate at the same rate as PDK1 (). We generated EBs from PDK1 and PDK1 ES cells. ES cells were cultured for 5 d in suspension to form EBs, and then plated on tissue culture dishes. Gross examination of EBs revealed differences in size and morphology starting from day 3 of differentiation on a gelatin-coated dish (). EBs from ES PDK1 cells exhibited reduced cell size and spreading, which were probably caused by defective cell motility and adhesion, although we could not exclude proliferation defects. In PDK1 EBs, CD31-positive ECs aggregated in dense clusters that started to form a vascular-like network after 3 d (). On day 7, the PDK1 ECs organized into tubular structures that became more evident, numerous, and branched after 10 d (). When PDK1 EBs were analyzed 3 d after plating, only CD31-positive cell clusters were found without any signs of vessel formation (). After 7 d, an immature network of ECs began to form in some areas of the EBs (). However, this network was unable to differentiate in vessel-like structures, and after 10 d it regressed, appearing as clusters of ECs with few branches (). To exclude that the observed differences might be a consequence of a reduced number of ECs, we determined whether lack of PDK1 modified the number of CD31- and Flk1-positive cells in EBs. After 3 d in culture, anti-CD31 staining confirmed the presence of ECs in both PDK1 and PDK1 EBs (). The percentage of CD31-positive cells decreased after 7 and 10 d in culture, with no considerable differences between PDK1 and PDK1 EBs (). Similar results were obtained by staining with anti-Flk1, suggesting that PDK1 is not essential for EC differentiation from ES cells (). Directional migration is a key event in angiogenic remodelling during vascular morphogenesis (). The inability of PDK1 EBs to form a vascular network, even though ES PDK1 cells can differentiate into ECs, suggests that this phenotype could result from defective cell migration. To verify this hypothesis, we performed migration assay with cells isolated from PDK1 and PDK1 EBs and stimulated with VEGF-A. CD31-positive ECs from PDK1 EBs weakly migrated in response to VEGF-A gradient compared with those from PDK1 (). We also observed that, in the absence of VEGF-A, ECs derived from PDK1 EBs randomly migrate, whereas ECs from PDK1 EBs appear completely incapable of migration (). Taking advantage of previously made ES knock-in cells in which either the PH domain (PH/PH) or PIF pocket (155E/155E) was disrupted, we studied the vascular phenotype of EBs derived from these ES cells (; ). EBs generated from PDK1 ES cells display normal vasculature development, similar to that of PDK1 EBs (). In contrast, PDK1 EBs failed to develop well-defined, cordlike structures, lacking the elaborate organization displayed by PDK1 (). However, compared with PDK1 EBs, the defective vascular structures of PDK1 EBs appeared less severe and did not regress after 10 d (compare and ). The quantification of the total length of vessel-like structures at day 7 of differentiation pointed out the impaired vascular development of PDK1 and PDK1 EBs compared with PDK1 and PDK1 EBs (). Migration of CD31-positive ECs derived from these EBs was analyzed. As shown in , ECs derived from PDK1 EBs migrated in response to VEGF-A as efficiently as wild-type cells, whereas ECs from PDK1 EBs displayed migration defects similar to those of PDK1 cells. These results indicate that PDK1 regulates vascular formation and EC migration in a PH domain–dependent way. To further assess the role of PDK1 in cell migration, we studied the chemotactic response of human ECs where PDK1 was overexpressed by retroviral transduction (EC-PDK1). Contrary to ECs transduced with vector alone (EC-vector), either EC-PDK1 or ECs expressing a membrane-tagged form of PDK1 (EC- PDK1caax) migrated more efficiently in a gradient of VEGF-A (). Few cells migrated in the absence of chemoattractant, and the overexpression of PDK1 did not increase the number of migrating cells; in some batches of EC-PDK1caax, we observed a relatively higher number of migrating cells in basal conditions compared with EC-vector, but not a statistically significant amount. To distinguish between chemotaxis toward VEGF-A gradient and random motility induced by VEGF-A, we added the same concentration of VEGF-A in both the upper and lower compartments of the Boyden chamber. The results clearly showed that PDK1 and PDK1caax were not able to significantly enhance random migration (). This result was supported by time-lapse videomicroscopy experiments, in which ECs were homogenously stimulated with VEGF-A. shows that neither PDK1 nor PDK1caax were able to increase random motility compared with EC-vector. We considered whether the PDK1-enhanced migration might reflect increased activation of PDK1 by VEGF-A. The regulatory mechanisms controlling PDK1 activity are poorly understood. PDK1 has been reported to be constitutively active in resting cells and autophosphorylated at S241 (). According to these previous data, we observed that S241 was basally phosphorylated in ECs, and that stimulation with VEGF-A did not modify the phosphorylation level (, top). The high level of S241 phosphorylation of EC-PDK1 compared with EC-vector depends on the high expression level of exogenous proteins, as demonstrated by the amount of myc-tagged protein (, bottom). We then examined the phosphorylation of Akt, which is the main substrate of PDK1. Upon stimulation with VEGF-A, T308 of Akt was very poorly phosphorylated in EC-vector (, first row, and Fig. S1 A, available at ). In contrast, we found that in EC-PDK1 and EC-PDK1caax the level of phosphorylation was already detectable without stimulation and increased upon VEGF-A stimulation (, first row, and Fig. S1 A). Interestingly, S473 of Akt, which is not a substrate of PDK1, showed similar levels of phosphorylation in unstimulated EC-vector, EC-PDK1, and EC-PDK1caax; in the presence of VEGF-A, the level of phosphorylation increased in all different cell types at the same rate (, second row, and Fig. S1 B). To evaluate the kinase activity of Akt we looked at the phosphorylation of two substrates: GSK3β and FKHR/AFX. As shown in , the phosphorylation of GSK3β and FKHR/AFX in PDK1 and PDK1caax ECs increased in either basal condition and upon VEGF-A stimulation, and it partially correlated with the phosphorylation of T308, but not with that of S473 of Akt. Collectively, these results indicate that Akt phosphorylation level at T308 correlates with the EC's ability to migrate, whereas the role of S473 seems to be marginal. It was important to be sure that the increased migration did not result from effects on proliferation or survival. A high level of Akt activity was reported to confer EC survival in the absence of attachment (). Moreover, in some cell lines, PDK1 regulates cell proliferation and survival, although ES PDK1 cells do not display any growth defects (; ). To determine the effects of PDK1 on these biological processes, we studied the growth and the survival rate of ECs expressing the wild-type and the membrane-tagged PDK1 mutant. PDK1 expression caused slight increase of VEGF-A–induced cell proliferation only after 96 h of stimulation (Fig. S2 A, available at ). As for proliferation, cell viability in the absence of serum and growth factors was not dramatically increased by PDK1 and PDK1caax expression (Fig. S2 B). PDK1 contains a C-terminal PH domain, a centrally located catalytic domain, and a 50-aa N-terminal region that binds a Ral-GEF (). Additionally, recent studies indicate that the ability of PDK1 to phosphorylate S6K, SGK, and RSK is dependent on a docking site, the PIF pocket, which is located on the small lobe of the PDK1 kinase domain (). To determine which region was required for PDK1-induced migration, PDK1 mutants of these regions of the protein were produced and tested for their ability to modulate VEGF-A– induced EC migration (Fig. S3, available at ). ECs expressing PDK1 kinase-dead (PDK1-KD), PDK1 lacking the PH domain (PDK1-ΔPH) or the Ral-interacting domain (PDK1-Δ50), and PDK1 with a mutation on the PIF pocket (PDK1-L155E) were assayed for migration properties in a Boyden chamber in the absence or presence of VEGF-A as chemoattractant. The PDK1-KD mutant lost the ability to enhance EC migration (), and, in some experiments, it even exhibited a reduced motility compared with EC-vector (not depicted). As expected, the kinase-dead mutant was unable to enhance the phosphorylation of T308 of endogenous Akt (, fourth row). ECs transduced with PDK1-Δ50 and PDK1-L155E were assayed in a similar manner and still showed an increased migration when stimulated by VEGF-A (). In contrast, PDK1-ΔPH did not function in enhancing EC migration (). When tested for the ability to phosphorylate Akt, it showed levels of T308 phosphorylation comparable with that of EC-vector (, fourth row). As we have previously shown for PDK1 and PDK1caax, the expression of all PDK1 mutants did not change the level of S473 Akt phosphorylation in both unstimulated and VEGF-A–stimulated conditions (, fifth row). Because it has been demonstrated that the PH domain of PDK1 binds with high affinity to PtdInsP and this interaction enhances its ability to activate Akt (), we investigated the potential involvement of PI3K, the enzyme generating PtdinsP, in the mechanism of PDK1-induced migration. Treatment with the inhibitor of PI3K, LY294002, reduced both basal and VEGF-A–stimulated EC migration (). The inhibition of PI3K activity completely abolished the effect of PDK1 and PDK1caax expression on EC migration, suggesting that production of PtdInsP is necessary for PDK1 activation and the subsequent promigratory effect. As PtdInsP is required for both PDK1 and Akt activity, the inhibitory effect of LY294002 was not attributable exclusively to one of them. To clarify this point, we treated ECs expressing Akt-myr, which is a membrane-targeting mutant form of Akt that is PtdInsP independent, with LY294002, and we observed the phosphorylation level of T308 of Akt. In unstimulated conditions, T308 is highly phosphorylated, whereas a little increase was observed after VEGF-A stimulation (). Unexpectedly, the inhibition of PI3K activity with LY294002 did not decrease the phosphorylation level of T308 of Akt, but exclusively inhibited the phosphorylation increase stimulated by VEGF-A (). These results () clearly show that during VEGF-A–stimulated migration, PI3K activation is required for both PDK1 and Akt activity. To assess the role of Akt in the PDK1-induced migration, we performed experiments with ECs infected with lentiviral vectors carrying shRNA sequences designed to silence the expression of Akt. The best performing sequence among those tested caused a decrease in Akt expression of ∼70% (). The reduction of Akt expression paralleled the reduced ability to migrate in response to VEGF-A in both EC-vector and EC-PDK1 (). PDK1 has been shown to move to the plasma membrane in response to platelet-derived growth factor and insulin (; ). However, other studies have not supported these reports (). To determine the subcellular localization of PDK1 in ECs responding to VEGF-A, we stained EC with anti-myc and -PDK1 Abs. In unstimulated cells, PDK1 was localized in the cytoplasm and perinuclear region (). Once the cells were stimulated with VEGF-A, a PDK1 fraction translocated to the plasma membrane (). As expected, EC-PDK1caax showed a strongly marked plasma membrane localization in both stimulated and unstimulated cells (). Deletion of the PH domain of PDK1 resulted in a loss of its ability to localize to the plasma membrane upon VEGF-A stimulation (, F and G, for quantification). The association of PDK1 with the plasma membrane in stimulated cells and its involvement in the process of directional cell migration prompted us to test whether PDK1 localized at the leading edge of migrating cells. After wounding a confluent monolayer of EC-PDK1, we detected PDK1 in the lamellipodia at the leading edge of migrating cells in the direction of the wound (). In contrast, PDK1-ΔPH failed to move to the leading edge, whereas PDK1caax was localized all around the plasma membrane (). To test whether PDK1 phosphorylated Akt in the lamellipodia at the leading edge, we stained a wounded monolayer of PDK1-transfected murine embryonic fibroblasts (MEFs) with anti-PDK1 and anti-pT308Akt antibodies. The level of phosphorylation on T308 of Akt increased at the wound edge and colocalized with PDK1 on large regions of the plasma membrane (). In this experiment, MEFs were used instead of ECs to avoid the interference of GFP of infected ECs. The MEF behavior was similar to that of ECs in migration assays (unpublished data) and in translocation of PDK1 to plasma membrane in response to PDGF (Fig. S4, available at ). These findings suggested that both PDK1 and Akt move to the leading lamellipodia, where Akt is phosphorylated by PDK1. It has been demonstrated in different cell types that motility and chemotaxis rely on the activation of PI3K, and chemotactic factors elicit intracellular PtdInsP gradients in the plasma membrane. Therefore, the localization of PDK1 at the leading edge could be a consequence of an intracellular gradient of PtdInsP, mediated by the binding of PtdInsP to the PH domain of PDK1. To further investigate the importance of PDK1 localization at the leading edge in cell motility, we compared behaviors of ECs expressing membrane-targeted mutants of PDK1 (PDK1caax), Akt (Akt-myr), and PI3KCA (p110caax) in chemotaxis experiments. The expression of the membrane-targeted catalytic subunit of PI3K did not increase EC migration, but rather, in some experiments, slightly inhibited chemotaxis (). Similar results were obtained by , who observed chemotaxis defects after the expression of membrane-targeted PI3KA in PI3K1/2-null cells. In contrast, when we transduced EC with membrane-targeted Akt, a strong increase in migrating cells both in the absence and presence of VEGF-A was observed (). This increase in motility is characterized by the presence of multiple pseudopodia (unpublished data). These findings raise the question of whether restricted activation of PDK1 and Akt to the leading edge is critical for the chemotaxis process or not. We considered the possibility that PDK1 and Akt interact with each other in a manner that depends on PtdInsP In this model, the correct membrane localization of both occurs in response to lipid. To address this issue, we transduced ECs with PDK1caax and Akt-myr together, expecting that PDK1–Akt complex might form in higher concentrations, but that neither component could be localized properly. These cells showed a reduced VEGF-A–induced migration compared with EC-PDK1caax or EC-Akt-myr alone (). However, ECs expressing Akt-myr transduced with wild-type PDK1 did not change their ability to migrate in comparison with nontransduced cells, demonstrating that the inhibitory effect was not caused by the double infection (). Finally, we analyzed by immunofluorescence the localization of PDK1 and pT308Akt in a wounded monolayer of double-infected EC. We detected both PDK1 and pT308Akt on the lamellipodia at the leading edge of migrating EC-PDK1 (, B and C, respectively). In ECs carrying both PDK1 and Akt-myr, a gradient of phosphorylated Akt along the cells was still present (), in accordance with the localization of PDK1 at the leading edge (). On the other hand, when PDK1 was forced to the whole plasma membrane by infecting the cells with PDK1caax, phosphorylated Akt was mainly localized at the front of migrating cells (). When both PDK1 and Akt were constitutively linked to the membrane, the staining of pT308Akt was evident along the entire surface of the cells (). These data, together with chemotaxis results (), suggest that proper membrane localization of both PDK1 and Akt are required to correctly instruct the chemotaxis process. Cells with altered PI3K or PTEN activity can usually migrate, but exhibit a significantly reduced ability to move directionally toward a chemoattractant gradient (). However, the mechanism behind how PtdInsP accumulation is followed by the formation of leading edge is still obscure. Although it has been described as a positive feedback loop between PtdInsP and Rac GTPase, resulting in enhanced formation of membrane protrusions at the leading edge, relatively little is known about how PI3K downstream effectors regulate cell migration in response to external stimuli (; ). Moreover, there is not yet a consensus about the importance of localized PI3K signaling during migration of mesenchymal cells (e.g., fibroblasts and EC) that do not adopt an amoeboid movement, and especially not when tyrosine kinase receptor signaling is involved (; ). In this study, we found that PDK1, a PtdInsP-binding protein, plays an important role in the regulation of cell migration stimulated by VEGF-A, a ligand of the tyrosine kinase VEGF receptor 1 and 2. We observed that cells lacking PDK1 exhibited a reduced motility, and completely lost their ability to migrate in response to a chemoattractant in vitro. The role of PDK1 in cell migration is controlled by its PH domain because knock-in cells with a mutated PH domain, and therefore unable to bind PtdInsP, migrate at the same reduced rate as knock-out cells. Thus, it is plausible that PDK1 may follow the internal gradient of PtdInsP generated by VEGF-A– dependent PI3K activation, and translocates to the plasma membrane in the direction of the stimulation. We can, indeed, observe that in response to VEGF-A, PDK1 moves to the plasma membrane at the leading edge of migrating cells and phosphorylates Akt on T308. The essential role of PDK1 in cell migration is supported by the findings that PDK1 knock-out EBs exhibit evident developmental defects that can be ascribed to defective EC motility. The vascular phenotype displayed by PDK1 EBs is obvious and cannot be accredited to lack of EC differentiation. In PDK1 EBs, the number of differentiated EC is similar to that of PDK1 EBs, but the ECs are not able to form capillary-like structures. Given that cell migration is a critical event in the angiogenic remodelling during vascular morphogenesis (), we suggest that the vascular phenotype observed in the PDK1 EBs could be caused by reduced cell motility. This hypothesis is supported by the evidence that PDK1 mouse embryos die at E9.5, displaying multiple abnormalities, including lack of somites, dorsal root ganglia, forebrain, and a circulatory system (Lawlor et al., 2002). The lack of dorsal root ganglia, together with the absence of branchial arches, strongly suggests a defective migration of neural crest cells. Moreover, ES cells lacking PDK1, which failed to activate Akt and RSK, are viable, despite the reported important role of Akt and RSK in regulating survival and proliferation. Although it has been reported that PDK1 plays an important role in cell proliferation and survival in some cell lines, at least for ES cells and ECs, PDK1 is not intrinsically required for survival and proliferation (; ). Collectively, our results and the phenotype of PDK1 knock-out mice (Lawlor et al., 2002) indicate that PDK1 is necessary for EC migration in vitro and in vivo, and potentially regulates the migration process of other cell types and tissues. Expression of wild-type and mutant PDK1 was used to provide insight into how migration might be regulated. The initial finding that overexpression of PDK1 promotes cell migration supports the central role of PDK1 in this process. Interestingly, overexpression of PDK1 exclusively increased cell motility in the presence of a chemoattractant, such as VEGF-A, but did not modify the basal cell motility. The ability of PDK1 to increase EC migration is kinase-dependent, as demonstrated using the kinase-dead mutant. Some PDK1 domains that are important in other functions, such as the RalGEF-interacting N terminus and the PIF pocket, are not involved in this process. The PH domain is clearly critical. Both the ECs expressing PDK1 lacking a PH domain and the EB-PDK1–derived cells make this point. However, inactivation of the PH domain in EBs gave rise to a vascular phenotype less severe than that caused by complete deletion of PDK1. In contrast, the motility of EC derived from EB-PDK1 was completely defective and comparable to that of EB-PDK1. These results suggest that other PDK1 domains may be involved in the vascular network formation, regulating different biological processes rather directional motility. Consistent with these observations, PDK1-induced migration was blocked by PI3K inhibitor. Upon activation of VEGF receptor, activation of PI3K promotes the membrane localization of PDK1 and Akt, resulting in an increase of cell migration. Whether PtdInsP is also required for PDK1 activation in EC is unclear. A previous report () and our experiments indicate that PDK1 is constitutively active, suggesting that the involvement of PI3K in PDK1 activation could be preferentially linked to PtdInsP-induced conformational changes of Akt that enable PDK1 to phosphorylate this kinase (). In addition to this mechanism, we demonstrated that PDK1 activity can be positively regulated by VEGF-A in cells expressing the membrane-targeted mutant of Akt, which does not require conformational changes to be phosphorylated. The increased phosphorylation of Akt-T308 in PDK1-overexpressing cells paralleled the chemotaxis increase, whereas no phosphorylation change of S473 was observed. This suggests that T308 phosphorylation determines the activation state of Akt, whereas S473 is dispensable. However, two substrates of Akt, GSK3β and FKHR, were more phosphorylated in PDK1-overexpressing cells, but their phosphorylation level did not completely correlate with the EC ability to migrate. A possible explanation could be that high levels of T308 phosphorylation modify the substrate specificity of Akt, stimulating its activity on other substrates. A similar event has recently been described for S473. Cells lacking the kinase for S473 retained the Akt activity, but the ability to phosphorylate some substrates was dramatically reduced (). The evidence that PDK1 overexpression does not modify the basal cell motility suggests that this enzyme is mainly involved in the directional movement. Knowledge on chemotaxis mechanisms is essentially derived from studies on leukocytes and amoeba, which are able to move rapidly toward a variety of chemoattractants. These studies have demonstrated that PH domain–containing proteins specific for PtdInsP accumulate at the leading edge of migrating cells and that PI3K and PTEN associate with the membrane at the front and back, respectively, of chemotaxing cells (; ). As with the aforementioned cell type, motility of cells with nonamoeboid movement, such as fibroblasts and EC, rely on the activation of PI3K, and PDGF gradients elicit intracellular PtdInsP gradients in the plasma membrane (; ). However, there are indications that, in these cells, the PtdInsP-mediated spatial gradient–sensing mechanism differs (; ). Our observations that PDK1 overexpression regulates EC chemotaxis, together with PDK1 localization at the leading edge of migrating cells, demonstrate that the PI3K signaling pathway regulates EC directional motility in accordance with the previously proposed models in leukocytes and . Moreover, the expression of membrane-targeted PI3K in ECs did not increase, rather than decrease, the chemotaxis. Similar results have been described by in cells, in which they observed a chemotaxis deficiency in PI3K1/2 cells expressing membrane-targeted PI3K. However, these models, which are focused on PI3K local accumulation, are not able to explain why EC transduced with PDK1 membrane-targeted mutant exhibit a great increase in chemotaxing cells compared with wild-type EC. We suggest a model in which the PH domain of PDK1 and Akt contributes to their localization; as long as either PDK1 or Akt is localized in a lipid-specific way, a signaling gradient results (). When one of them is overexpressed and forced to the plasma membrane, the result is the local increase of Akt activation and chemotaxis. When both are overexpressed and forced to the membrane, the gradient of PtdInsP is no longer needed for localization and activation that would result in directional migration (). The enrichment of PDK1 and Akt at the leading edge contributes to a local increase of Akt phosphorylation at T308 that could be responsible for the stimulation of directional migration. This model is confirmed by immunofluorescence staining of Akt pT308 showing that only when both PDK1 and Akt are membrane-anchored is Akt local activation lost. In spite of this, the high level of phosphorylation of Akt-myr observed in unstimulated condition or in presence of PI3K inhibitor is difficult to explain. In these conditions, PDK1 mainly localizes to the cytosolic and perinuclear regions; thus, it cannot efficiently phosphorylate the T308 of Akt-myr. A possible explanation is that the presence of the Myr tag on Akt leads to effects that are not solely related to its constitutive membrane association. In conclusion, we demonstrate that PDK1 is required for migration of ECs in vivo and in vitro. Moreover, the increase of chemotaxing cells, obtained in PDK1-overexpressing ECs, indicates that one of the mechanisms by which PDK1 controls the motility is the regulation of the directional migration. We also suggest that PDK1-mediated Akt activation at the leading edge is responsible for this effect. To our knowledge, this is the first genetic evidence that the PI3K signaling pathway controls the motility response mediated by tyrosine kinase receptors and the directional motility process in ECs. Human ECs were isolated from umbilical cord vein, characterized, and grown as previously described (). MEF and Phoenix cells were grown in DME (Cambrex) supplemented with 10% FCS, 2 mM -Glutamine (Cambrex), and antibiotics. MEFs were transiently transfected with Lipofectamine Plus reagent (Invitrogen) according to the manufacturer's instructions. All ES cells (PDK1, PDK1, PH/PH, and 155E/155E) were provided by D.R. Alessi (University of Dundee, Dundee, UK) and maintained in an undifferentiated state by culture on a feeder layer of MEFs pretreated with 10 μg/ml mitomycin C (Sigma-Aldrich), in high-glucose DME (Invitrogen), 15% FBS (HyClone), 0.1 mM nonessential amino acids (Invitrogen), 1 mM sodium pyruvate (Invitrogen), 0.1 mM β-mercaptoethanol (Sigma-Aldrich), -Glutamine, antibiotics, and 1,000 U/ml LIF (CHEMICON International, Inc.). The cDNAs of wild-type and mutant PDK1 (with the exception of L155E) and the cDNA of the membrane-targeted catalytic subunit of PI3K (p110caax) were previously described (). The L155E mutant of PDK1 was made by site-directed mutagenesis. The cDNA of membrane-targeted Akt1 (Akt1-myr) was provided by W. Sessa (Yale University, New Haven, CT; ). All cDNAs (with the exception of HA-tagged Akt1-myr) were myc-tagged by polymerase chain reaction and subcloned under the control of LTR promoter into EcoRI–EcoRI sites of the retroviral vector Pinco, which also contains GFP cDNA as tracking marker (). cDNAs of myc-tagged PDK1 and PDK1caax were also subcloned into EcoRI–EcoRI sites of Pinco modified with dsRed cDNA instead of GFP for double-infection experiments. The amphotropic cell line Phoenix was transfected with retroviral vectors, and the retroviral supernatants obtained were collected, filtered (0.45 μm; Millipore), and supplemented with 4 μg/ml of polybrene (Sigma-Aldrich). Medium of ECs were replaced with the appropriate retroviral supernatants, and cells were incubated at 37°C with 5% CO for 5 h. 72 h after infection, cells were analyzed for GFP expression by microscopy and for specific transgene expression by Western blot. Short hairpin RNAs (shRNA) against human Akt1 (4 sequences) were designed according to the TRC shRNA guidelines (), and subcloned into the MluI–ClaI sites of the pLVTHM vector (), which was provided by D. Trono (University of Geneva, Geneva, Switzerland). Efficacy of the constructs was tested through transduction into Sup-M2-TS cells and Western blot analysis of total cell lysates with α-Akt1–specific antibody (Cell Signaling Technology) after 4 d. The sequence for the sense oligonucleotides for the most effective knockdown constructs is: 5′-GGACTACCTGCACTCGGAGAA-3′ (based on positions 1,207–1,238 of human AKT1). Self-inactivating retroviral and lentiviral particles were produced as previously described (). Aliquots of virus, plus 8 μg/ml of polybrene, were used to infect exponentially growing cells (10/ml). Fresh medium was supplemented at 24 h after the infection. The infectivity was determined (after 72 h) by FACS analysis of GFP-positive cells. Chemotaxis assays with human ECs were performed in a Boyden chamber, as previously described (). In brief, PVP-free polycarbonate filters (8 μm pore size; Neuroprobe) were coated with 1% gelatin for 2 h at 37°C. 10 ng/ml VEGF-A (R&D Systems) dissolved in serum-free medium was seeded in the lower compartment of the chamber; cells were serum starved overnight, and then suspended in serum-free medium at a concentration of 2.5 × 10 cells/ml, and 50 μl of the suspension was added to the upper compartment. For experiments with EBs, at day 3 of differentiation, they were disaggregated with PBS and 2 mM EDTA for 5 min at 37°C and trypsin for 1 min at 37°C. Cells were seeded on the upper side of a 24-well, 8-μm pore HTS FluoroBlok insert (BD BioSciences; 1 × 10 cells/well) that was coated on the lower side with 20 μg/ml fibronectin and incubated in EB medium supplemented with 2% FCS, rather than 20% FCS; the lower compartment was filled with EB medium and 2% FCS with or without 30 ng/ml VEGF-A. After 5 h of incubation at 37°C with 5% CO, the upper surface of the filters was scraped with a rubber policeman, and the filters were fixed and stained with Diff-Quick (Dade Behring) or rat α-CD31, as described in Indirect immunofluorescence. Four random fields of each sample in the lower surface of the filters were counted at 10× magnification. Starved ECs were plated onto gelatin-coated 24-well plates and allowed to adhere in serum-free medium for 1 h at 37°C. 100 ng/ml VEGF-A was or was not added to the medium, and ECs were observed with an inverted microscope equipped with a thermostatic and CO-controlled chamber (AS MDW workstation; Leica). Fluorescent video images of ECs were recorded at 10-min intervals for 6 h with a charge-coupled device camera (Orca HiRes; Hamamatsu Photonics) and analyzed using DIAS image processing software (Solltech). Speed parameter (in micrometers/minute) of 60 cells from three different experiments were calculated and plotted. Confluent cells were serum deprived for 2 h, pretreated or not pretreated with 50 μm LY294002 for 45 min, and stimulated or not stimulated with 30 ng/ml VEGF-A for 10 min. Cells were transferred on ice, washed three times with cold PBS containing 1 mM Na orthovanadate, and lysed in RIPA-modified buffer containing 20 mM Tris, pH 7.2, 150 mM NaCl, 1% Triton X-100, 0.5% Na desossicolate, 0.1% SDS, 5 mM EDTA, and protease and phosphatase inhibitors (50 μg/ml pepstatin, 50 μg/ml leupeptin, 10 μg/ml aprotinin, 1 mM PMSF, 100 μM ZnCl, 1 mM Na orthovanadate, and 10 mM NaF). After centrifugation (15 min at 10,000 ), supernatants were precleared by incubation for 1 h with protein G–Sepharose (GE Healthcare). Samples (700 μg of proteins) were incubated with rat α-HA (Roche) for 2 h to isolate overexpressed Akt-myr, and immune complexes were recovered on protein G–Sepharose. Beads were washed four times with lysis buffer and detected by immunoblot. Proteins were separated by SDS-PAGE electrophoresis, transferred to polyvinylidene difluoride (PVDF) membrane (Millipore), incubated with rabbit α-pT308Akt and, after stripping, rabbit α-Akt (Cell Signaling Technology), and visualized by ECL system (GE Healthcare). For lysates, cells were serum deprived for 2 h and stimulated or not with 30 ng/ml VEGF-A for 10 min. Total proteins were extracted in Laemmli buffer (62.5 mM Tris-HCl, pH 6.8, 2% SDS, and 10% glycerol) and quantified, and equal amounts of each sample were resolved by SDS-PAGE and transferred to PVDF membrane. After blocking with TBS/0.1% Tween 20/5% BSA, membranes were incubated with primary antibody overnight at 4°C. The following primary antibodies were used: rabbit α-pS241PDK1, α-pS473Akt, α-pT308Akt, α-Akt, α-pFKHR, α-pGSK3β (all from Cell Signaling Technology), mAbs α-myc, and α/β tubulin (both Santa Cruz Biotechnology). Immunoreactive proteins were identified with secondary antibody coupled to HRP antibody and visualized by ECL. For in vitro differentiation of ES cells, the same ES medium was used, except that LIF was omitted and the FBS concentration was 20% (EB medium). The hanging drop procedure was followed (). ES cells grown to confluence on the feeder layer of MEFs were harvested with trypsin and centrifuged. The pellet was resuspended in ES medium, and ES plus MEF cell suspension was seeded on a tissue culture dish for 30 min at 37°C. During this time, MEFs attached, whereas ES cells remained in suspension. This step was repeated twice. The ES cell suspension was centrifuged, and the pellet was resuspended in EB medium for counting. 30-μl drops, containing 400 cells, were placed on the undersurface of the lids of Petri dishes and incubated at 37°C. After 2 d, the cell aggregates contained in the drops were collected and cultured in suspension in bacterial Petri dishes for 3 d. The aggregates were then transferred onto gelatin-coated regular culture dishes, where they spread and differentiated; this was day 0 of differentiation. For fluorescence immunostaining, at day 0 EBs were plated onto gelatin-coated glass coverslips in 24-well plates; after 3, 7, or 10 d of differentiation, they were fixed with paraformaldehyde 3.7% in PBS for 20 min at room temperature, washed three times with PBS, and stained as described in Indirect immunofluorescence. The quantification of total length of vessel-like structures stained by α-CD31 was performed with the imaging software winRHIZO Pro (Regent Instruments, Inc.), as described by . EBs at different differentiation stages were rinsed twice and then disaggregated with PBS 2 mM EDTA for 5 min at 37°C and trypsin for 1 min at 37°C. 2 × 10 cells were then incubated on ice with 5 μg/ml of the following primary antibody for 30 min: rat α-Flk1 (Becton Dickinson), rat α-CD31, and control rat IgG. After three washes with PBS 1% BSA, cells were incubated on ice with 2.5 μg/ml R-phycoerythrin–conjugated α-rat antibody (Southern Biotechnology Associates) for 20 min. After final washes with PBS, samples were fixed with PBS, 1% BSA, and 2% PAF and analyzed using FACScan (Becton Dickinson). ECs or MEFs were seeded at high density on gelatin-coated glass coverslips; after 12 h of adhesion, monolayer cells were wounded by dragging a plastic pipette tip across the cell surface and 50 ng/ml VEGF-A or PDGF was added to serum-free medium. After 6 h, cells were washed with PBS, fixed with PAF 3.7% for 10 min at room temperature, and analyzed by indirect immunofluorescence, as described in the following section. The protocol described was followed both for cells (ECs and MEFs) and EBs. For immunofluorescence staining, ECs were plated onto gelatin-coated glass coverslips in 24-well plates. After 12 h of adhesion in complete medium, they were serum deprived for 2 h and then stimulated or not stimulated with 50 ng/ml VEGF-A. Medium with growth factors was removed, and cells were fixed with 3.7% PAF for 10 min at room temperature. After fixation, cells or EBs were rinsed three times with PBS, and then quenched with 50 mM NHCl for 20 min at room temperature, washed twice with PBS, and permeabilized with PBS 0.5% Triton X-100 for 5 min at room temperature. After two washes with PBS, coverslips were blocked with PBS 0.3% Triton X-100, 1% donkey, 1% or goat serum for 1 h at room temperature, and incubated with primary antibodies overnight at 4°C in a humidified chamber. For EC staining of EBs, rat α-CD31 (1:100; BD Biosciences) was used, and for ECs and MEFs, mAb α-PDK1 (1:80; BD Biosciences), mAb α-myc (1:40; Santa Cruz Biotechnology) and rabbit monoclonal α-pT308Akt (1:80; Cell Signaling Technology) were used. After three washes with PBS, coverslips were incubated for 1 h at 37°C in a humidified chamber with fluorescent secondary antibodies; donkey α-rat Cy2 (1:200; Jackson ImmunoResearch Laboratories), donkey α-mouse Alexa Fluor 488 or goat α-mouse Alexa Fluor 405 (1:400; Invitrogen), donkey α-rabbit Alexa Fluor 555, or goat α-rabbit Alexa Fluor 405 (1:400). Coverslips were then rinsed three times with PBS, mounted, and analyzed using an inverted fluorescence microscope (DM IRB; Leica) equipped with 63×/1.30 HCX Plan-Apochromat (Carl Zeiss MicroImaging, Inc.) glycerin-immersion and 4×/0.10 C Plan objectives or a confocal laser-scanning microscope (TCS SP2 with DM IRE2; Leica) equipped with 63×/1.40 HCX Plan-Apochromat oil-immersion objective. Confocal images are the maximum projections of a z section of ∼1.50 μm. The images were arranged and labeled using Photoshop software (Adobe). Fig. S1 shows the quantification of Western blots presented in D. Fig. S2 shows that PDK1 overexpression didn't affect the proliferation rate and survival capacity of ECs. Fig. S3 shows the PDK1 mutants used in the overexpression experiments. Fig. S4 shows the localization of PDK1 in MEF stimulated or not stimulated with PDGF. Online supplemental materials are available at .

Under various pathological conditions, including infection, malignancy, and autoimmune diseases, tissues are incessantly exposed to reactive oxygen species (ROS) produced by infiltrating inflammatory cells. We previously disclosed that superoxide stimulated cell motility in many types of malignant cells, leading to invasion and metastasis (; ; , ; ). This notion is compatible with earlier findings that superoxide generated by red blood cells from sickle cell anemia patients stimulated transendothelial migration of human myeloid leukemia, HL-60 cells, and human monocytes () and that leukotriene B induced transmigration of human neutrophils across epithelium, which was inhibited by intracellular ROS scavenger -acetylcystein (NAC) or NADPH oxidase inhibitor diphenylene iodonium (DPI; ). Furthermore, respiratory burst (superoxide generation) and prominent motility tracking to pathogens have been known to be pivotal features of inflammatory cells, in particular, monocytes/macrophages and neutrophils, and various chemokines that stimulate inflammatory cells to generate superoxide (; ) have been shown to simultaneously evoke chemotaxis (), suggesting a close relationship between superoxide production and motility in these cells. Cell motility results from remodeling of acto-myosin system, which is regulated by Rho family of small GTP-binding proteins (). There is diversity in extracellular stimulants for cell migration, such as lysophophatidic acid (), platelet-derived growth factor (), and hepatocyte growth factor (), and many different intracellular signaling molecules that correspond to each stimulant are implicated in the activation of RhoGTPases. However, despite the fact that superoxide and chemokines are considerably important as stimulants of motility, not only from the view of tumor biology but also from the view of innate immunity, no detailed exploration on the motility relevant to these stimulants has been performed to date. In the present study, we first reveal that, in human squamous carcinoma SASH1 cells, superoxide activates PKCζ, which phosphorylates RhoGDI-1, in turn liberating RhoGTPases from RhoGDI-1, leading to their activation. Then, using human peripheral monocytes (hPMs) and murine macrophage-like cell line J774.1, we examined whether the superoxide extracellularly generated by hypoxanthine/xanthine oxidase (HPX/XOD), or which they themselves produced upon treatment with a chemotactic peptide, -formyl-methionyl-leucyl-phenylalanine (fMLP), stimulated their motility, and found that both extracellularly generated and self-produced superoxide augmented their motility. Furthermore, we confirmed that the PKCζ–RhoGDI-1 phosphorylation–RhoGTPases activation signaling pathway was also involved in the motility of hPMs and J774.1 stimulated with superoxide or fMLP. To examine the effect of superoxide on cell motility for SASH1 cells, we performed a phagokinetic track assay (). The cells without any treatment showed some movement (intrinsic motility), which was not affected by the addition of Cu-Zn superoxide dismutase (SOD). When the cells were treated with HPX/XOD, the motility was enhanced as compared with that of nontreated cells. The enhancement of cell motility by superoxide stimulation was canceled with the addition of Cu-Zn SOD, indicating that superoxide by itself, not other types of ROS, is the factor that enhances the cell motility. Treatment of the cells with NAC resulted in almost complete suppression of intrinsic motility to the basal level, suggesting that endogenous ROS is affecting their motility. This effect of endogenous ROS on motility was not seen in Cu-Zn SOD transfectant (clone 1), suggesting that the endogenous ROS responsible for the cell motility is mainly superoxide. Incidentally, the suppressive effects observed in the other Cu-Zn SOD clones, including clone 2, were essentially the same as those seen in clone 1 (unpublished data). We also performed migration a assay. Cells treated with HPX/XOD showed higher motility than nontreated cells, and it was blocked by the addition of Cu-Zn SOD (Fig. S1 a, available at ). To confirm that the motility of SASH1 cells is related to their intracellular ROS level, we analyzed dihydrorhodamine (DHR) 123 staining of SASH1 cells by FACS (). The nontreated SASH1 cells showed some staining intensity (endogenous ROS), which was not suppressed by extracellularly added Cu-Zn SOD. This staining intensity was clearly enhanced by treatment with superoxide. NAC pretreatment or Cu-Zn SOD overexpression suppressed the intensity to the basal levels and maintained the basal intensity even after superoxide treatment. Incidentally, extracellular superoxide was not detectable with this cell line measured by cytochrome method (unpublished data), indicating that the cells were not releasing superoxide extracellularly. These results suggest the relevance of superoxide to motility of SASH1 cells. To investigate the mechanism of enhancement of cell motility by superoxide, we first examined the morphological changes in SASH1 cells (). The parental cells had a rather round shape with small lamellipodia. In the parental cell stimulated with superoxide, there was increment of F-actin and a more discrete formation of lamellipodia and filopodia than in that without stimulation. Cu-Zn SOD transduced clone 1 showed less actin staining with spindle-shaped morphology (dendrite-like formation). Similar morphological features were observed with NAC-treated cells. This spindle-shaped morphology was unchanged even after superoxide stimulation. When we analyzed the activity of Rho family of small GTPases (Rho, Rac, and Cdc42) by pull-down assay, some activated forms of these proteins were identified in parental cells, and the superoxide stimulation apparently enhanced these proteins (). This activation of Rho family GTPases was markedly inhibited by overexpression of Cu-Zn SOD (). The results of the pull-down assay were verified by analyzing the proteins relocalizing to the plasma membrane (Fig. S2, available at ). RhoA, Rac1, and Cdc42 each showed an apparent membrane translocation followed by spontaneous detachment from the membrane in a relatively short time after superoxide treatment; however, the detachment of Rac1was somewhat retarded compared with the other two GTPases. Transfectants of dominant-negative (DN) Cdc42 (DNCdc42) and Rac1 (DNRac1), exhibited impaired motility similar to that of C3-treated cells treated with or without superoxide (). When the morphology of SASH1 cells was examined, treatment with C3 resulted in a slight reduction of F-actin intensity () compared with that of nontreated cells () and showed new dendrite-like formations and multiple nuclei in a single cell caused by inhibition of cytoplasmic division. In these cells, superoxide treatment did not increase F-actin intensity, but apparently induced lamellipodia or filopodia formation (). DNRac1 transfectant was not substantially different from the parental cells without the stimulation (), whereas superoxide treatment of the cells induced F-actin increment and filopodia formation, although lamellipodia formation was not observed (). Transduction of DNCdc42 caused loss of cell polarity with relatively concentrated F-actin staining in the center of the cell (). The morphological characters of DNCdc42 became more apparent by treatment with superoxide (). These results are compatible with the previous notion that F-actin is regulated by Rho; that activation of Rac1 is associated with lamellipodia formation (), although it does not associate much with F-actin or filopodia formation; and that Cdc42 regulates cell polarity and filopodia (). As evidence that shows the relationship between PKC and RhoGTPases is accumulating (; ; ; ; ; ; ; ), we examined the possibility that PKC is a molecule that links superoxide with Rho family GTPases. The PKC inhibitor, calphostin C, dose-dependently inhibited cell motility () as well as the activity of Rho family GTPases () in superoxide-treated or nontreated SASH1 cells (). At 500 nM calphostin C, cell motility was suppressed to basal level and almost complete inhibition of the RhoGTPase activity was attained. The results suggested that PKC may be an upstream molecule of RhoGTPases and that it regulates cell motility stimulated by superoxide. To determine which isozyme of PKC is responsible for the Rho family activation, we studied PKC isozymes that were expressed in SASH1 cells by immunoblotting and seven PKC isozymes, PKCα, -β, -γ, -δ, -ɛ, -μ, and -ζ, were found to be expressed (, lane 1). As members of the Rho family of small GTPases form complexes with RhoGDI-1 in cytosol in their resting state and the first step of Rho family activation is their liberation from RhoGDI-1 (; ; ; ; ), we next identified the PKC isozyme that interacted with RhoGDI-1–RhoGTPases by pull-down assay using the anti–RhoGDI-1 antibody. As shown in , among seven isozymes expressed in SASH1 cells, only PKCζ was detected in the immunoprecipitates, and the association was enhanced by superoxide treatment. Furthermore, this coprecipitation of PKCζ with RhoGDI-1 was clearly inhibited in NAC-pretreated () or Cu-Zn SOD transduced cells (), indicating dependency of PKCζ activation on superoxide stimulation. We then examined whether PKCζ kinase activity is stimulated in superoxide-treated cells. As shown in , phosphorylated PKCζ was increased in superoxide- treated cells as compared with nontreated cells revealed by in vitro phosphorylation assay. The PKCζ activity was suppressed by calphostin C in a dose-dependent manner as described previously (; ; ; ). Further, PKCζ activation was confirmed by its translocation to the plasma membrane (). The PKCζ molecule consists of two functional domains, reportedly, a regulatory N-terminal domain and a catalytic C-terminal domain (). We investigated which domain interacts with RhoGDI-1 in superoxide-treated SASH1 cells by pull-down assay. As shown in Fig. S3 a (available at ), RhoGDI-1 did associate with the C-terminal catalytic domain but not with the N-terminal regulatory domain. The fact that PKC is a member of serine-threonine kinases, and RhoGDI-1's interaction with the catalytic domain of PKCζ, prompted us to examine the possibility of phosphorylation of RhoGDI-1 by PKCζ. Immunoprecipitates with anti–RhoGDI-1 antibody were found to contain threonine but not serine-phosphorylated RhoGDI-1 in both superoxide-treated and nontreated cells, though in the latter the threonine-phosphorylated RhoGDI-1 increased more than in the former (). To further elucidate whether PKCζ directly phosphorylate RhoGDI-1 upon superoxide stimulation or, rather, activates a putative kinase, which in turn phosphorylates RhoGDI-1, RhoGDI-1 was purified from SASH1 cell lysate by immunoprecipitation (IP) and incubated with recombinant active PKCζ. As shown in , the intensity of the band representing [P]RhoGDI-1 in the PKCζ-treated preparation was apparently increased compared with that in the nontreated preparation, indicating the direct action of PKCζ on RhoGDI-1. On the other hand, when recombinant RhoGDI-1 was simply incubated with recombinant active PKCζ, phosphorylation of RhoGDI-1 was undetectable (unpublished data). Therefore, it is highly conceivable that some cofactors that coprecipitated with RhoGDI from the cell lysate were required for RhoGDI-1 phosphorylation by PKCζ. The small RhoGTPase family proteins are potential cofactors. Upon superoxide stimulation, PKCζ interacts with RhoGDI-1/small RhoGTPase family protein complex; binding of PKCζ to the complex stimulates the release of the small RhoGTPases from RhoGDI (). Incidentally, weak but nonnegligible phosphorylation of RhoGDI-1 observed in the PKCζ-nontreated preparation (, left) may be ascribed to the activity of endogenous PKCζ that has been coprecipitated with RhoGDI-1. We then extended our investigation to examine the kinetics of complex formation of RhoGDI-1 and Rho family GTPases or PKCζ by IP using anti–RhoGDI-1 antibody. An immunoreactive band of PKCζ increased in intensity from 30 s until 2 min after superoxide treatment (). However, in the same time course, RhoGDI-1 showed no appreciable changes, whereas the bands of all Rho family GTPases showed gradual decrement, indicating the dissociation of RhoGTPases from the complex and their eventual activation upon initiation of superoxide stimulation. These dissociation and association phenomena were not obvious in DNPKCζ transduced cells (). To further validate the implication of PKCζ in morphological changes, Rho family activation, and cell motility induced by superoxide, SASH1 cells were transfected with DNPKCζ plasmid to suppress PKCζ activity. We first confirmed that the transfectants expressed the DNPKCζ protein (unpublished data) and that DNPKCζ was truly active to inhibit autophosphorylation of PKCζ in vitro (). This DNPKCζ transfectant showed morphological features that were similar to those of the parental cells (), and these features were not apparently affected by superoxide treatment. Moreover, activation of Rho family proteins with superoxide stimulation was not seen in the DNPKCζ transfectants (). Furthermore, intrinsic cell motility or cell motility induced by superoxide was also impaired in the transfectant (). In the transfectant of DNPKCζ, the associations of RhoA, Rac1, and Cdc42 with RhoGDI were not affected () and showed no increase of threonine phosphorylation of RhoGDI even after HPX/XOD treatment (). These results indicate that the activation of Rho family, the changes of cell morphology, and stimulation of cell motility relevant to superoxide were evoked by PKCζ. These observations led us to speculate that superoxide may be actively involved in the motility of inflammatory cells as well, most likely via an autocrine mechanism, as they are known to exhibit prominent motility and generate superoxide by themselves. We validated this speculation by using macrophage-like cell line J774.1. As shown in , certain amount of superoxide was detected in the medium of the cells and their motility was significantly suppressed by addition of Cu-Zn SOD to the medium (), indicating autocrine stimulation of cell motility by superoxide released from the cells themselves (automotility). When the cells were treated with NAC, their motility (both auto and intrinsic) was almost completely suppressed (), indicating that not only automotility but also intrinsic motility is mediated by ROS inside the cells. Enhancement of the motility was clearly observed with the cells treated with superoxide. This enhanced motility was blocked by the addition of Cu-Zn SOD to the level of the cells treated with Cu-Zn SOD alone and to the basal level by NAC, suggesting that the effect of extracellularly added superoxide is also operating intracellularly as ROS. These results were compatible with the morphological findings (). When nontreated cells are compared with NAC-treated cells, which resemble Cu-Zn SOD transfected SASH1 cells, F-actin increment, small but apparent formation of filopodia, and little lamellipodia were seen, supporting the speculation of autoactivation of cells by self-produced superoxide. The F-actin intensity, filopodia, and lamellipodia formation seen in nontreated J774.1 cells were further enhanced in superoxide-treated cells. To confirm the involvement of RhoGTPases in the superoxide-relevant motility of J774.1 cells, we conducted a pull-down assay for these enzymes. As shown in , superoxide activated RhoA, Rac1, and Cdc42 in a time-dependent manner. Then, to evaluate whether the activation of these GTPases is associated with cell motility, we intended to examine the motility of the cells treated with specific inhibitors for RhoA, Rac1, and Cdc42. However, transduction of DN RhoGTPases was not feasible in this cell line; therefore, only C3 treatment was performed. As shown in , treatment with C3 at a concentration of 5 μg/ml caused suppression of motility of both superoxide-treated and nontreated J774.1 cells to the basal levels, suggesting the dependency of the motility on Rho. We next studied whether there is PKCζ involvement in this signaling pathway. Among PKCα, -β, -γ, -δ, -ɛ, -μ, and -ζ, only PKCζ associated with RhoGDI-1 examined by pull-down assay using RhoGDI-1 antibody (unpublished data). To confirm the role of PKCζ in cell motility and RhoGTPase activation, J774.1 cells were treated with calphostin C. Both cell motility () and RhoGTPase activation () of superoxide-treated and nontreated cells were substantially suppressed to basal levels by 500 nM calphostin C. We used myristoylated pseudosubstrate peptide of PKCζ (myr-PKCζp) to inhibit the enzyme activity. Motility of J774.1 cells with or without superoxide stimulation was impaired by myr-PKCζp compared with control peptide–treated cells (). The phenomena of association of PKCζ and dissociation of GTPases from the RhoGDI-1–GTPases complex upon superoxide stimulation found in SASH1 cells were also confirmed in this cell line (), and dissociation and association were inhibited by calphostin C (not depicted). As shown in Fig. S4 (available at ), Rac2 was also activated by superoxide in J774.1 cells. We further extended our study by examining human monocytes collected from the peripheral blood of a healthy volunteer. As shown in Fig. S5 a (available at ), addition of Cu-Zn SOD significantly lowered the motility of hPMs, indicating autocrine activation of motility by superoxide, which they generated (automotility). Treatment with NAC almost completely suppressed their motility, a finding consistent with that observed with J774.1. Augmentation of motility by superoxide treatment was also observed with hPMs (Fig. S5 a), as was the case with SASH1 and J774.1, and the motility was positively related with superoxide production (Fig. S5 b). Morphologically, NAC-treated hPMs showed spindle form with dendrite formation, whereas nontreated cells exhibited a rather round shape with slightly increased F-actin staining and formation of some small filopodia (Fig. S5 c). Such F-actin staining or filopodia formation of hPMs was further strengthened by treatment with superoxide (Fig. S5 c). As shown in Fig. S5 d, the activation of RhoGTPases by superoxide was also confirmed. Moreover, movement of hPMs treated with or without superoxide was strongly suppressed by C3 to the basal levels (Fig. S5 e). To investigate whether PKCζ is also involved in the motility of hPMs, we pulled down the isozyme with anti–RhoGDI-1 antibody in the cells treated with or without superoxide. Coprecipitation of PKCζ with RhoGDI-1 was clearly observed in cells treated with superoxide and seen with less intensity in those without superoxide (). RhoGDI-1 was apparently phosphorylated, as revealed by immunoblotting using anti-phosphothreonine antibody in both superoxide-treated and nontreated hPMs, with much higher intensity in the former than the latter (). As shown in , PKCζ kinase activity was increased by superoxide stimulation. To examine the involvement of PKCζ in the motility of hPMs, the cells were treated with myr-PKCζp. As shown in , motility of both superoxide-stimulated and nonstimulated cells were almost completely suppressed to basal level by myr-PKCζp. The activation of RhoGTPases in hPMs was also clearly suppressed by treatment with myr-PKCζp (). Association of PKCζ with RhoGDI-1 upon superoxide stimulation and dissociation of RhoGTPases from RhoGDI-1 were confirmed with hPMs as well (). The effect of myr-PKCζp on the morphology of hPMs was further examined (). hPMs had a round shape with small filopodia and mild F-actin staining. Filopodia formation and F-actin staining were enhanced, and lamellipodia formation became evident by treatment with superoxide. In myr-PKCζp–treated cells, filopodia and lamellipodia formation were hardly seen and intensity of F-actin was rather weak compared with nontreated cells. The association of PKCζ with and dissociation of RhoGTPases from RhoGDI-1 were impaired in myr-PKCζ–treated cells (unpublished data). To investigate the role of superoxide in the motility of monocytes/macrophages that are stimulated by chemokine, we examined the superoxide generation ( and Fig. S5 f) and motility ( and Fig. S5 g) of J774.1 cells () and hPMs (Fig. S5, f and g) treated with fMLP. fMLP treatment substantially stimulated superoxide production by these cells ( and Fig. S5 f). The motility of these cells treated with fMLP was significantly increased, and this increment was negated by the addition of Cu-Zn SOD, NAC, or DPI ( and Fig. S5 g), indicating that the motility of these cells is prompted by superoxide generated by themselves through |activated NADPH oxidase in an autocrine manner. When fMLP-treated cells were treated with myr-PKCζp, the motility of the cells was totally suppressed to basal levels. Essentially the same results were obtained by migration assay (Fig. S1 b). Thus, it was suggested that PKCζ is directly activated by fMLP, which utilizes the same PKCζ–RhoGDI-1–GTPases signal pathway as superoxide. In various pathophysiological situations, such as inflammation, malignancy, arterial sclerosis, tissue damage, etc., cells in the lesion are incessantly exposed to ROS. In the present study, all the cells examined, which included human tongue cancer cell line SASH1, the murine macrophage-like cell line J774.1, and hPMs, became motile after superoxide treatment, suggesting a ubiquitous role of superoxide as a stimulator of cell motility. Furthermore, the finding that the addition of Cu-Zn SOD to hPMs or J774.1 cells suppressed their motility, suggests that the automotility of those cells is largely dependent on the superoxide, which they themselves generate, i.e., an autocrine stimulation of the motility by superoxide. Regarding the motility of cancer cells (SASH1), the autocrine mechanism was not operating, as self-produced superoxide was not detectable. However, as they respond to exogenous superoxide (HPX/XOD), it is suggested that cancer cells acquire invasiveness and metastatic ability as a consequence of enhanced motility stimulated by the superoxide generated by infiltrating inflammatory cells in a paracrine manner in vivo. The observation that NAC treatment of hPMs and J774.1 or Cu-Zn SOD transfection to SASH1 cells almost completely nullified the motility of these cells implies the essentiality of intracellular superoxide as a generator of cell motility at the basal level (intrinsic motility). This recognition of the essentiality of intracellular superoxide for cell motility is consistent with our previous observation that an inverse relationship existed between intracellular Cu-Zn SOD activity and invasiveness of tumor cells. We used a tetracycline regulation system for transient expression of Cu-Zn SOD, as it was previously reported that forced long-term expression of SOD gene brought about growth suppression of the transfectants (). With respect to the mechanism whereby extracellular superoxide increases intracellular superoxide, penetration through cell membrane may be a possibility (; ; ; ), although the proof remains to be established. To substantiate the idea that superoxide induces motility, we investigated the underlying molecular mechanism. The first issue we addressed was whether, and to what extent, the Rho family of small GTP-binding proteins was involved in the signal transduction of superoxide-induced cell movement, as these proteins are well accepted as common mediators of the cell signals for various stimuli. RhoA, Rac1, and Cdc42 were each activated by superoxide stimulation in all three cells examined. The motility of SASH1 cells was markedly suppressed by C3, DNRac1, and DNCdc42, with C3 limiting motility to basal levels in hPMs and J774.1 cells. Together, these results indicate that ROS induces cell motility via Rho family GTPase activation. Furthermore, each single GTPase may be indispensable for the cell movement induced by superoxide. Next, we searched for the molecule that may link the superoxide signal with Rho family GTPases. It has been reported that stimulators of cell motility, such as lysophophatidic acid and insulin, activate both Rho family GTPases and PKC simultaneously (; ). Translocation of RhoA to membrane and RhoA-dependent phospholipase D activation induced by GTPγS are reportedly blocked by calphostin C in MDCK cells (). Activation of RhoA by phosphorylating RhoGDI-1 by PKCα has been observed in human umbilical venular endothelial cells stimulated with thrombin (). In addition to these reports, previous evidence indicating activation of PKCs by oxidative stress (; ) led us to hypothesize that PKC is the linking molecule. This hypothesis was supported by the fact that calphostin C suppressed motility and activation of RhoGTPases in all three types of cells. With regard to the mechanism for activation of PKC by superoxide, the involvement of PI3-kinase in linking these molecules is conceivable because PI3-kinase has been known to activate PKCζ via PDK-1 phosphorylation in vitro (). However, the fact that the inhibitor of PI3-kinase, LY294002, did not inhibit superoxide-induced RhoA or Cdc42 activation in our cells (unpublished data) negated this possibility. Thus, it seems likely that superoxide activates PKC either through activation of kinases other than PI3-kinase or through direct activation of PKC by superoxide, possibly by causing conformational change (). Elucidation of these putative mechanisms remains a future task. Given that PKC is involved in signaling from superoxide to all three RhoGTPases, it is highly plausible that PKC interacts with RhoGDI-1 to activate RhoGTPases, because inactive RhoA, Rac1, and Cdc42 are commonly bound to RhoGDI-1. Therefore, using SASH1, J774.1 cells, and hPMs, we examined the direct interaction between PKCs and RhoGDI-1 by IP and found that only one specific isozyme of PKCs, PKCζ, was coprecipitable with RhoGDI-1. This observation, taken collectively with the facts that treatment with DNPKCζ or inhibitory peptide of PKCζ caused suppression of motility and the activation of the GTPases, strongly supports the notion that this particular PKC isozyme is a signal transducer of superoxide. It has been shown that peroxide activates various PKC isozymes expressed in COS cells (). We also found that superoxide treatment of SASH1 brought about activation of all types of PKCs expressed in the cell, as determined by their translocation to the plasma membrane. In this context, the specific binding of PKCζ to RhoGDI-1 can be rationalized by the speculation that this isozyme has higher binding affinity than other PKC isotypes to the substrate (RhoGDI-1), and not by the specific activation of PKCζ by superoxide. It is generally accepted that for activation of RhoGTPases, their release from the RhoGDI-1 molecule is required. It was recently reported that PKCα, activated by thrombin, can phosphorylate RhoGDI-1, catalyzing the release of bound GTPases (). Our present results indicate that superoxide stimulates PKCζ, which in turn leads to RhoGDI-1 phosphorylation at threonine sites to liberate RhoGTPases. These results are compatible with the aforementioned study, except for the stimulant (superoxide) and signal transducer (PKCζ). As inflammatory cells are usually attracted and activated by chemokines in the lesion, we further studied the role of superoxide in motility of J774.1 and hPMs induced by fMLP, a chemotactic peptide that is known to interact with 7-transmembrane G-coupled formyl-peptide receptors (). Our findings that fMLP treatment of these cells stimulated superoxide generation, which was suppressed by DPI, a specific inhibitor of NADPH oxidase, confirmed the notion that chemokines activate PKCζ to stimulate NADPH oxidase at plasma membrane, in turn generating superoxide (; ). On the basis of these findings, we extended our investigation to elucidate the relevance of fMLP-induced motility to superoxide generated by the cells treated with fMLP. The results that Cu-Zn SOD or DPI treatment suppressed the motility of J774.1 and hPMs support the concept of autocrine activation of chemokine-induced motility via superoxide. It was speculated that NAC should scavenge out intracellular superoxide and thereby abrogate superoxide-induced cell motility. However, when the fMLP-stimulated J774.1 cells or hPMs were pretreated with NAC, a certain magnitude of motility unexpectedly remained. The results therefore imply that fMLP interaction with chemokine receptor stimulates multiple pathways to activate cell motility. Furthermore, the fact that myr-PKCζp suppressed the motility of J774.1 cells and hPMs to basal level suggests the possibility that the other signal (as well as superoxide-relevant one) is also mediated by PKCζ. This notion, activation of PKCζ through interaction of chemokine with its receptor, is veritably compatible with the recent report that atypical PKCζ regulates stromal cell–derived factor-1 mediated migration of human CD34+ progenitor cells (). Hence, it may be reasonable to deduce that fMLP binds its receptors to activate PKCζ to generate superoxide, which in turn stimulates the motility in an autocrine manner via the PKCζ–RhoGDI-1–RhoGTPase pathway. On the other hand, PKCζ activated by fMLP simultaneously induces motility via the common PKCζ–RhoGDI-1–RhoGTPase pathway. Accordingly, the motility induced by chemokine is considered to be at least partly dependent on superoxide. In conclusion, we disclosed that superoxide plays a pivotal role in the motility of SASH1, J774.1, and hPMs through a novel signaling pathway of PKCζ–RhoGDI-1–RhoGTPases. Thus, these results suggest a new approach for manipulation of inflammation as well as tumor cell invasion by targeting this novel signaling pathway. SASH1, a poorly differentiated human squamous cell carcinoma line was cultured in DME/F-12 (Invitrogen), and murine macrophage-like cell line J774.1 was cultured in RPMI (Invitrogen) supplemented with 10% FCS in humidified 5% CO at 37°C. To attain the quiescent state, cells were cultured in the medium without FCS for 20 h. CD14+ monocytes were isolated from peripheral blood mononuclear cells of healthy volunteers obtained by the standard Ficoll-Paque method and separated by negative magnetic depletion using hapten-conjugated CD3, CD7, CD19, CD45RA, CD56, and anti-IgE antibodies (MACS; Miltenyi Biotec) and a magnetic cell separator (MACS) according to the manufacturer's instructions. Serum-starved cultures on 100-mm dishes were treated with 4 μg/ml HPX (Sigma-Aldrich) and 7 × 10 U/ml XOD (Sigma-Aldrich) for various incubation periods noted in each experiment. For NAC treatment, NAC (Sigma-Aldrich) was dissolved in the medium, pH adjusted to 7.4, and the cells were treated for 1 h. A uniform carpet of gold particles was prepared as previously described (). In brief, sterilized coverslips were coated with 1% bovine serum albumin and were placed in 35-mm tissue culture dishes. Gold colloidal solution (184 μM HAuCl-4H0 and 11.6 mM NaCO) was boiled shortly to make gold colloidal particles, and the coverslips were coated with them. 2 × 10 serum-starved cells with various pretreatments described in each experiment were seeded, attached to the coverslips, treated with or without superoxide, and 2 h later fixed by 0.1% formaldehyde. The areas cleared of gold particles were examined under microscope and quantified by image processing (KS400; Carl Zeiss MicroImaging, Inc.). Intracellular ROS level was evaluated using DHR 123 (Invitrogen) as a fluorescent probe. After cells were treated with various conditions, as indicated, 1 μl of 43.3 mM DHR was added to the cells suspended with 1 ml HBSS for 20 min, cells were washed, the generation of rhodamine 123 was monitored on a FACScan (Becton Dickinson) with excitation at 488 nm, and the emitted fluorescence was collected at 525 nm. The cells were fixed in 3% paraformaldehyde for 20 min, permeabilized with 0.2% Triton X-100 for 5 min, and stained with rhodamine-labeled phalloidin (Invitrogen) for 40 min to visualize F-actin. Fluorescence images were obtained by a confocal laser-scanning microscope system with 40× objective lens (LSM5 PASCAL; Carl Zeiss MicroImaging, Inc.). Preparation of recombinant C3 exoenzyme and ADP-ribosylation assay were performed as described previously (). The C3-expressing plasmid pET3a C3 was provided by S. Narumiya (Kyoto University Faculty of Medicine, Kyoto, Japan). Cells were lysed in lysis buffer (20 mM Tris/HCl, pH 7.5, 10 mM MgCl, 1 mM pefabloc, 5 mM leupeptin, and 5 mM pepstatin) containing 1% NP-40 and were incubated with protein G–agarose beads (GE Healthcare) for 3 h to block the nonspecific binding. After centrifugation, the supernatant was incubated with anti–RhoGDI-1 antibody (Santa Cruz Biotechnology, Inc.) for 3 h, followed by the addition of protein G–agarose beads, and agitated overnight. The beads were collected, washed, and further analyzed by immunoblotting. For immunoblotting, each aliquot of the sample was eluted by RhoGDI-1 peptide, mixed in SDS sample buffer, heated, separated on SDS-PAGE, and electroblotted. Antibodies against RhoA (26C4), RhoGDI-1, and PKCζ were purchased from Santa Cruz Biotechnology, Inc. Antibodies against Rac1 and PKCα, -β, -γ, -δ, -ɛ, -θ, and -μ were purchased from BD Biosciences. Antibodies against Cdc42 were purchased from Upstate Biotechnology. Antibodies against phosphothreonine and phosphoserine were purchased from Zymed Laboratories and QIAGEN. Activity of Rho family small GTP-binding proteins was assayed by pull-down assay using each assay kit (Upstate Biotechnology) according to the manufacturer's instruction. PKCζ activity from 3 × 10 cells was assayed by assessing its phosphorylation state as described previously (). In brief, PKCζ in cell lysates was immunoprecipitated with the anti-PKCζ antibody, washed, mixed in assay solution (35 mM Tris, pH 7.5, 15 mM MgCl, 1 mM MnCl, 0.5 mM EGTA, 0.1 mM CaCl, 1 mM sodium orthovanadate, and 100 μM γ-[P]ATP), and incubated at 30°C for 10 min. After reactions were stopped by the addition of gel loading buffer, the samples were boiled and analyzed by SDS-PAGE followed by autoradiography. Phosphorylation of RhoGDI-1 in vitro was performed using RhoGDI-1 immunoprecipitated from SASH1 cell lysate with RhoGDI-1 antibody. Confluent cells grown in 100-mm dishes were washed with ice-cold PBS and lysed in IP buffer containing 50 mM Tris, pH 7.4, 150 mM NaCl, 0.25 mM EDTA, pH 8.0, 1% deoxycholic acid, 1% Triton X-100, 5 mM NaF, 1 mM sodium orthovanadate, 1 mM PMSF, 5 μg/ml leupeptin, 5 μg/ml aprotinin, and 1 μg/ml pepstatin A. The cells were collected and then cleared by centrifugation at 4°C at 14,000 for 10 min. The lysate was incubated with anti–RhoGDI-1 antibody for 1 h followed by the addition of protein G–Sepharose beads overnight at 4°C. The beads were collected, washed twice with ice-cold lysate buffer, washed three times with PBS, and washed once with kinase buffer (8 mM MOPS, pH 7.4, and 0.2 mM EGTA). The immunoprecipitated RhoGDI-1 was incubated with 5 ng of recombinant active PKCζ in kinase assay buffer for 10 min at 30°C followed by the addition of magnesium/ATP cocktail (Upstate Biotechnology): 4 mM MgCl and 25 μM ATP in 1 mM MOPS, pH 7.2, 1 mM β-glycerol phosphate, 0.2 mM EGTA, 50 μM sodium orthovanadate, 50 μM dithiothreitol, and 10 μCi γ-[P]ATP (stock 1 mCi/100 μl: 3,000 Ci/mmol; PerkinElmer). The reaction was stopped by the addition of Laemmli sample buffer. The samples were electrophoresed on a 10% SDS-polyacrylamide gel, transferred to nitrocellulose membrane, and exposed to x-ray film. The blots were then subjected to Western blotting with anti–RhoGDI-1 antibody to verify that an equal amount of the protein loaded in each lane. Cells were washed three times with ice-cold PBS and scraped in lysis buffer (20 mM Tris/HCl, pH 7.5, 10 mM MgCl, 1 mM pefabloc, 5 mM leupeptin, and 5 mM pepstatin). The cells were lysed by 18 passes through a 26-gauge needle on ice. Trypan blue staining of the lysate indicated >95% disruption of plasma membrane. The subcellular fractionation was performed by the method as described by . The lysate was first centrifuged at 500 for 10 min to prepare low-speed pellet, and the supernatant was recentrifuged at 120,000 for 45 min to pellet the remainder of the particulate fraction (high-speed pellet). The low-speed pellet was further purified by sucrose gradient centrifugation to obtain plasma membrane fraction. Protein concentrations were determined using BCA protein assay (Pierce Chemical Co.) according to the manufacturer's directions. Alkaline phosphodiesterase I, cytochrome oxidase, and lactate dehydrogenase were used as marker enzymes for plasma membrane, mitochondria, and cytosol, respectively (). DNA and RNA were used as markers for nucleus and endoplasmic reticulum, respectively, and were detected on agarose gels by ethidium bromide staining, with or without RNase treatment. After subfractionation of the cells, plasma membrane fractions were subjected for immunoblotting to measure PKC or RhoGTPase activity. The pTA-Hyg and pET2a vectors were provided by J. Yokota (National Cancer Center Research Institute, Tokyo, Japan). cDNA for the human Cu-Zn SOD expression vector was generated by PCR using the forward primer 5′-TTTCCGTTGCAGTCCTCGGAA-3′ and the reverse primer 5′-CCTCAGACTACATCCAAGGGA-3′ with the human cDNA library as the template. The PCR fragment was ligated into the pET2a vector to generate pET2a Cu-Zn SOD vector, which expresses Cu-Zn SOD when tetracycline is omitted from the medium. SASH1 cells were first transduced with the pTA-Hyg vector, selected by hygromycin B, transfected with the pET2a Cu-Zn SOD vector, and selected by G418. cDNA of N17Rac1 and N17Cdc42, which are DN cDNAs of Rac1 and Cdc42, were provided by Y. Takai (Osaka University, Osaka, Japan). N17Rac1 was first ligated into the pSVneo-Myc/Sra vector and then cDNA coding for Rac1N17 tagged with Myc was digested and ligated into the pIRES/neo vector, resulting in pIRES-Rac1N17-Myc/neo. N17Cdc42 was first ligated into the pFlag-CMV vector and then cDNA coding for N17Cdc42 tagged with Flag was digested and ligated into the pIRES/hyg vector, resulting in pIRES-Cdc42N17-Flag/hyg. cDNA coding for the N- or C-terminal of human PKCζ was generated by PCR with the human cDNA library as the template using the First strand cDNA Synthesis kit (Boeringer). To produce plasmid pRx-bsr-PKCζN that contains the N-terminal regulatory domain of human PKCζ (nucleotides 1–738), we used 5′-CCGGAATTCACCCAAGATGGAAGGGAGCGGCGGC-3′ as the forward primer and 5′-CGCGGATCCTCATAGGTCAAAGTCCTGCAGCCCAAGC-3′ as the reverse primer. PCR products were separated, sequenced, digested with EcoRI and NotI, and ligated into a pcDNA3.1/His (Invitrogen), and the fragment containing His6 was excised by HindIII and NotI. It was inserted into a retroviral vector, pRx-bsr, supplied by H. Hamada (Sapporo Medical University, Sapporo, Japan). We used pRx-bsr-PKCζN regulatory domain expressing vector as DN (DNPKCζ), as this domain reportedly showed autoinhibitory activity (). Phenol red–free medium was used in the assay. 10 cells were stimulated for 1 min with HPX/XOD, and the cells were further incubated for 1 h in the presence of 50 μM cytochrome . The medium was removed and placed on ice, and the absorbance at 550 nm was immediately read by spectrophotometer with H0 as a blank. Superoxide-specific reduction of cytochrome was expressed as the difference in absorbance between cells incubated with or without SOD using an extinction coefficient of 21 mM cm. Myr-PKCζp, SIYRRGARRWRKLYRAN (positions 113–129, which is pseudosubstrate region of human PKCζ), and myr-control peptide, RLRYRNKRIWRSAYAGR (), were custom made by Sawaday Technology Co. They were solubilized immediately before use at 1 mM concentration in PBS, pH 7.2, and heated at 40°C to achieve complete solubility. Statistical analysis of data was performed by standard techniques with the aid of the Stat View computer program for the Macintosh (Abacus Concepts). The means of values were provided with associated SEM. Statistical significance was defined as P < 0.01. Fig. S1 shows a migration assay of SASH1 cells and J774.1 cells. Fig. S2 shows translocation of Rho family small GTPases to plasma membrane in SASH1 cells treated with superoxide. Fig. S3 shows the interaction of PKCζ catalytic domain with RhoGDI and that a SASH1 cell lysate without PKCζ did not phosphorylate recombinant RhoGDI-1. Fig. S4 shows activation of GTP-Rac2 by superoxide in J774.1 cells. Fig. S5 shows that superoxide stimulates RhoGTPases activity and motility of hPMs. Online supplemental material is available at .

Elastic fibers play a crucial role in the structural integrity and function of various organs (). The loss of elasticity leads to many age-related phenotypes, such as wrinkled skin, emphysema, and arteriosclerosis (; ). Therapeutic intervention for these age-related changes has not been realized because the mechanisms of the development and maintenance of elastic fibers remain unclear. Elastic fibers comprise two distinct components: an amorphous core of cross-linked elastin surrounded by a peripheral mantle of microfibrils. In the process of elastic fiber assembly, elastin precursors (tropoelastin) are deposited on microfibrils, aligned in an orderly way, and cross-linked by lysyl oxidase (LOX) enzymes to form mature elastin, which confers the elastic properties to elastic fibers (). However, the molecular mechanism of this multistep process remains largely unknown (). Tropoelastin, which is a major constituent of the elastic fiber, can interact with other tropoelastin molecules and self-aggregate at physiological temperatures. Thus, tropoelastins undergo a phase transition, which is termed coacervation, from soluble monomers to insoluble aggregates (). In elastic fiber assembly, coacervation is considered to play a critical role through concentrating and aligning tropoelastin before cross-linking. However, coacervation of tropoelastins alone is not sufficient to explain the assembly process and the variable forms of elastic fibers in different tissues. For fibrillar deposition of tropoelastin, microfibrils, which are mainly composed of fibrillin-1 and -2, are assumed to act as scaffolds, and to direct the morphogenesis of elastic fibers. In fact, mice deficient in both fibrillin-1 and -2 were recently reported to exhibit impaired elastogenesis (), indicating that fibrillins are required for the initial assembly of elastic fibers. The final cross-linking step is initiated by oxidative deamination of tropoelastins, which is catalyzed by LOXs (). LOXs are copper-containing monoamine oxidases secreted by fibrogenic cells, such as fibroblasts and smooth muscle cells. Five LOX family members have been identified so far: LOX, LOX-like 1 (LOXL1), 2, 3, and 4 (). LOX-deficient mice were reported to die perinatally because of insufficient cross-linking of elastin and collagens (; ). LOXL1-deficient mice were reported to live to adulthood, but to have degenerated elastic fibers, suggesting that LOXL1 is also important for the development and/or maintenance of elastic fibers (). Yet, little is known about how LOXs are recruited to catalyze tropoelastins on microfibrils. Thus, the elastic fiber assembly involves recruitment and organized deposition of various molecules, and seems to be tightly regulated and finely tuned to achieve the appropriate elastic properties in each organ. However, whether there are any specific molecules organizing the assembly of elastic fiber components remains unknown. Recently, we and others showed that the elastic fiber assembly process requires fibulin-5 (also known as developing arteries and neural crest EGF-like [DANCE] or embryonic vascular EGF-like repeat–containing protein [EVEC]) by thoroughly investigating fibulin-5–deficient mice (; ). These mice recapitulate human aging phenotypes, such as loose skin, emphysematous lungs, and stiff arteries because of disorganized elastic fibers. Fibulin-5 has an Arg-Gly-Asp motif at the N-terminal domain and interacts with cell surface integrins, such as αvβ5, αvβ3, and α9β1 (, ). Fibulin-5 also has tandem arrays of calcium-binding EGF-like domains and a characteristic C-terminal fibulin domain, and these structures enable it to interact with several elastic fiber components, such as elastin (; ), EMILIN (), and LOXL1 (). Thus, fibulin-5 is a good candidate for an organizer of elastogenesis that scaffolds elastic fiber components in the vicinity of cell surfaces. However, the specific role of fibulin-5 in elastogenesis, whether fibulin-5 can promote elastogenesis, and whether there is a correlation of aging-associated alterations of fibulin-5 with aging phenotypes, remains unknown. We report that recombinant fibulin-5 protein potently induces elastic fiber assembly, even in serum-free cell culture, without changing the expression of elastic fiber components, suggesting that fibulin-5 serves as an organizer molecule for elastogenesis. Fibulin-5 deposits on microfibrils, promotes aggregation of tropoelastin molecules through coacervation, and also interacts with LOXL1, 2, and 4, which are enzymes that cross-link elastin. We propose a model in which fibulin-5 tethers LOXL enzymes to microfibrils, thus facilitating aggregation and cross-linking of elastin on microfibrils. Intriguingly, a much lower level of full-length fibulin-5, and a higher level of a truncated form of fibulin-5 were detected in aged mouse loose skin than in young mouse skin because of proteolytic cleavage of the N-terminal domain. The cleavage of fibulin-5 abrogates fibulin-5–microfibril interaction, leading to loss of the elastic fiber–organizing activity of fibulin-5. To elucidate the molecular role of fibulin-5 in elastogenesis, we established an in vitro system to evaluate elastic fiber assembly, using human skin fibroblast cultures. In the presence of fetal bovine serum, skin fibroblasts developed an abundant meshwork of elastic fibers, which were stained with antielastin antibody (), as previously reported (). In the absence of serum, they did not develop elastic fibers (). Investigation of elastogenic factors has been hampered by the serum dependency of elastogenesis in cell culture, as serum contains various proteins of known and unknown functions (). Surprisingly, we found that skin fibroblasts developed an abundant meshwork of elastic fibers, even in serum-free media, when they were cultured with purified recombinant fibulin-5 protein (). Staining with anti-FLAG antibody revealed that added fibulin-5 protein colocalized with elastin. Recombinant fibulin-5 protein induced elastic fiber assembly in a dose-dependent manner in serum-free medium (). We scarcely detected endogenous fibulin-5 in serum-free culture, whereas we detected abundant deposition of fibulin-5 in serum-containing culture (Fig. S1, available at ). Addition of fibulin-5 protein did not affect the mRNA expression of elastic fiber components (e.g., elastin and fibrillins; ), indicating that a sufficient amount of elastic fiber components was produced even in serum-free culture, but that they could not be organized into elastic fibers without the addition of fibulin-5 protein. These data suggest that fibulin-5 is not a signaling molecule that changes gene expression, but an organizer molecule that promotes assembly of elastic fiber components. Tropoelastin, which is the precursor of elastin, can interact with other tropoelastin molecules and self-aggregate at physiological temperatures through a phase transition termed coacervation (; ). Coacervation is considered to be an important step before cross-linking of tropoelastin molecules for elastic fiber assembly (; ). Because fibulin-5 can directly interact with tropoelastin (), we examined whether fibulin-5 can affect the coacervation of tropoelastin. Purified recombinant tropoelastin protein coacervated at 35–40°C, causing the solution to become turbid (). In the presence of recombinant fibulin-5 protein, tropoelastin started to coacervate at lower temperatures (), suggesting that fibulin-5 promotes coacervation of tropoelastin. Fibulin-5–deficient mice exhibit aging-related characteristics, including loose skin, emphysema, and stiff arteries, because of disorganized elastic fiber development (; ). Because tissue elastic properties are gradually reduced with aging (; ), we examined the age-related changes of elastic matrices by observing skin tissues of young and old mice with transmission electron microscopy. As shown in , elastic fibers were disorganized and fragmented in old mice, whereas they were organized into thick fibrous structures in young mice (, middle and top). The deteriorated elastic fibers in old mice closely resemble the disorganized and fragmented elastic fibers found even in young fibulin-5–deficient mice (, middle and bottom). Next, we investigated whether the expression of fibulin-5 is altered during aging, as we inferred from the data in the previous sections that suggested fibulin-5 is not only necessary for elastogenesis but also able to promote elastic fiber organization. Using skin tissues of young and old mice, we performed Western blotting with anti–fibulin-5 antibody (BSYN2473). As shown in , we detected two specific bands in wild-type mice, while we did not detect any band in fibulin-5–deficient mice. Notably, the intensity of the upper bands was considerably decreased in the old mice compared with the young mice (, compare lanes 1–3 with lanes 4–6). On the other hand, the intensity of the bottom bands increased remarkably in the old mice (, compare lanes 1–3 with lanes 4–6). From these results, we hypothesized that the switching of fibulin-5 subtypes may affect the elastogenic activity of fibulin-5 and be correlated with the deterioration of elastic matrices during aging. Because only one major transcript of fibulin-5 was detected by Northern blot analysis () and no mRNA or EST that would code for a protein ∼10 kD smaller or larger than the major form of fibulin-5 protein was found in public databases, the two subtypes of fibulin-5 protein we observed were not thought to be splice variants, but were rather considered to be indicative of posttranslational modification. To investigate this modification, we transfected an expression vector containing fibulin-5 cDNA with a signal peptide and a FLAG tag at the N terminus into 293T cells. The culture medium of these cells was collected and subjected to Western blotting. As shown in , we detected 2 specific bands of 45 and 55 kD with an anti–fibulin-5 antibody (BSYN2473; , left) that recognizes amino acids 76–98 (, red mark). On the other hand, we detected only one 55-kD band with anti-FLAG antibody (, right). These results suggest that fibulin-5 may be cleaved at a more N-terminal position than the recognition site of anti–fibulin-5 antibody (). We deduced that the lower 45-kD band is the cleaved form of fibulin-5, while the upper 55-kD band is the full-length fibulin-5. The cleaved N-terminal fragment of fibulin-5 appears to be easily degraded because we did not detect a 10-kD band corresponding to this fragment with anti-FLAG antibody. To identify the cleavage site of fibulin-5, we stably transfected the C-terminal–histidine-tagged expression vector encoding fibulin-5 cDNA into 293T cells. We purified recombinant fibulin-5 protein from the conditioned medium of these cells with immobilized metal affinity chromatography, and examined the purified protein by SDS-PAGE. As expected, we detected two bands stained with Coomassie blue, and the lower 45-kD band was assumed to be the cleaved form of fibulin-5. We sequenced the N-terminal end of the lower band using Edman degradation (, arrow), and determined that fibulin-5 is cleaved after the arginine at position 77 (, top and middle). The upper 55-kD band was also confirmed by protein sequencing to be the full-length fibulin-5, as the N-terminal sequence coincided with the predicted cleavage site after the signal peptide. As many proteases are arginine-specific, we examined whether the arginine at position 77 is necessary for the cleavage of fibulin-5. We constructed an expression vector encoding R77A mutant fibulin-5, which was mutated from arginine to alanine at position 77 (, bottom), and transfected this R77A mutant fibulin-5 vector into 293T cells, followed by Western blotting. As shown in , we detected only the upper 55-kD band, whereas the lower 45-kD band disappeared (lane 2), indicating that the R77A mutant fibulin-5 is resistant to cleavage. These results indicate that fibulin-5 is specifically cleaved after the arginine at position 77 (, compare lanes 1 and 2), although we cannot rule out the possibility that fibulin-5 might not be cleaved in vivo at the same site as in vitro cell culture. Next, we examined the nature of the protease that cleaves fibulin-5. We transiently transfected the expression vector encoding C-terminal FLAG-tagged fibulin-5 cDNA into 293T cells, and subsequently added different types of protease inhibitors to the culture medium, which was subjected to Western blotting. Whereas a cysteine protease inhibitor, E64, did not inhibit the cleavage of fibulin-5 at all, a serine protease inhibitor, aprotinin, completely inhibited the cleavage (, compare lanes 3 and 4). These results indicate that fibulin-5 is cleaved by a serine protease in vitro. We have not yet succeeded in identifying the serine protease that cleaves fibulin-5. To assess the functional consequence of fibulin-5 cleavage, we studied the interaction of fibulin-5 and microfibrils. Microfibrils, which are mainly composed of fibrillin-1 and -2, are considered to serve as scaffolds for tropoelastin deposition and subsequent elastic fiber assembly. Fibroblast cell cultures developed fine meshworks of fibrillin-1 microfibrils, even under serum-free conditions, at days 4–7 of culture (); this is much earlier than elastin deposition, which requires >10 d of culture in serum-containing medium. When we added recombinant fibulin-5 protein to the culture, fibulin-5 colocalized with fibrillin-1 microfibrils. However, when we added recombinant truncated fibulin-5 protein to the culture, the added truncated fibulin-5 did not deposit on fibrillin-1 microfibrils. This was not because the truncated form of fibulin-5 is susceptible to degradation, as Western blot analysis of the culture medium showed that truncated form of fibulin-5 was as stable as the full-length fibulin-5 in the medium (). These data indicate that full-length fibulin-5 can deposit on microfibrils, whereas the truncated form of fibulin-5 cannot. Next, we investigated the consequences of the functional change caused by the cleavage of fibulin-5 in elastic fiber assembly, using the in vitro elastogenesis assay shown in with recombinant fibulin-5 and the truncated form of fibulin-5 proteins. Whereas wild-type fibulin-5 potently induced elastic fiber development, the truncated form of fibulin-5 did not induce elastic fiber development at all (). Therefore, we concluded that the cleavage of fibulin-5 causes inactivation of the elastogenic activity of fibulin-5. The assembly of functional elastic fibers requires not only fibrillar deposition and coacervation of tropoelastin, but also cross-linking of tropoelastin. The aforementioned data suggest that deposition of fibulin-5 onto microfibrils accelerates fibrillar tropoelastin deposition, but it was not clear if the tropoelastin molecules are cross-linked to form mature functional elastic fibers. To examine this, we added β-aminopropionitrile (BAPN), an irreversible inhibitor of LOXs (), to serum-free skin fibroblast culture containing recombinant fibulin-5 protein. It is known that LOXs, which are cross-linking enzymes for elastin monomers, are necessary for elastic fiber maturation (; ; ). As shown in , elastins showed only a dotlike deposition when elastin cross-linking was largely inhibited by BAPN, whereas fibrillar deposition of fibulin-5 on microfibrils was not affected by BAPN. This punctuate distribution of elastin is displayed by tropoelastin coacervates before cross-linking (). These data indicate that elastic fibers organized by the addition of fibulin-5 protein are cross-linked and mature fibers, and that not only deposition and aggregation but also cross-linking of elastin is promoted by prealigned fibulin-5 on microfibrils. We next examined whether cleaved fibulin-5 is merely inactive or works against elastogenesis in a dominant-negative manner. For this purpose, we quantified the amount of elastic fibers induced by full-length fibulin-5 with or without cleaved fibulin-5. We metabolically labeled newly synthesized elastin with [H]valine (), and measured the incorporation of [H]valine in the NaOH-insoluble fraction of these cells, which reflects the amount of mature elastic fibers (). As shown in , we detected substantial, dose-dependent incorporation of [H]valine upon adding recombinant fibulin-5 to the medium (, compare lanes 1–4). On the other hand, cleaved fibulin-5 showed neither an additive nor a dominant-negative effect on fibulin-5–induced elastic fiber development (, compare lanes 1–7). These results indicate that fibulin-5 simply loses its elastogenic activity as a result of cleavage of the N-terminal domain. To develop mature elastic fibers, tropoelastin molecules need to be highly cross-linked by LOXs. Five LOX family members have been identified so far, LOX, LOXL1, 2, 3, and 4 (). Mice lacking LOX and mice lacking LOXL1 were recently reported to show profound fragmentation of elastic fibers (; ; ). Moreover, LOXL1 has been reported to interact with fibulin-5 (). To investigate whether fibulin-5 interacts with other LOX family members, we performed in vitro binding assays using Myc-tagged LOX and LOXLs, and the set of FLAG-tagged fibulin-5 deletion mutant constructs shown in . As shown in , we detected the specific interaction of LOXL1, 2, and 4 proteins with fibulin-5 protein (top, lanes 1, 7, and 13). The interaction of fibulin-5 and these LOXL proteins was eliminated or considerably diminished by the C-terminal deletion of fibulin-5 (, top, lanes 5, 11, and 17). These results indicate that LOXL1, 2, and 4 proteins mainly interact with the C-terminal domain of fibulin-5. We also detected substantial, but weak, interaction of fibulin-5 with LOX and LOXL3 in in vitro binding assays (unpublished data). However, we could not determine the specificity of these interactions because all of the fibulin-5 deletion mutants tested seemed to weakly interact with LOX or LOXL3. Intriguingly, the fibulin-5–LOXL1 interaction was markedly diminished by N-terminal deletion of fibulin-5 (, top left, lanes 2 and 3). Thus, fibulin-5 does not interact with LOXL1 after cleavage of fibulin-5, and this defect might contribute to the loss of elastogenic activity. We have identified an elastic fiber–organizing activity of fibulin-5/DANCE protein that is abrogated by proteolytic cleavage of the N-terminal domain in cell culture and in vivo. Our proposed molecular mechanism of elastic fiber organization by fibulin-5/DANCE protein is illustrated in . We call this the “Line DANCE model.” According to this model, full-length fibulin-5 protein associates with microfibrils. Direct interaction of fibulin-5 and fibrillin-1 () may be important for this association. As fibulin-5 serves as an integrin ligand through its N-terminal domain (, ), integrin–fibulin-5 interaction, in addition to integrin–fibrillin interaction, may help the assembly to occur in the proximity of the cell surface. After fibulin-5 interacts with microfibrils, tropoelastins accumulate on fibulin-5 proteins and coacervate. LOXL enzymes tethered on the C-terminal domain of fibulin-5 promote cross-linking of the aggregated tropoelastins, to form mature elastic fibers. In contrast, as illustrated in , the truncated form of fibulin-5 cannot associate with microfibrils or integrins, and therefore is unable to promote the organization of the elastic fiber assembly, although the truncated form of fibulin-5 itself has the ability to bind tropoelastin and enhance coacervation of tropoelastin as well (unpublished data). In this study, we first succeeded in inducing the development of elastic fibers in serum-free skin fibroblast culture by adding purified fibulin-5 protein. Tropoelastin and other elastic fiber components were not organized into elastic fibers without addition of fibulin-5 protein, suggesting that fibulin-5 is not only a necessary component of elastic fibers (; ) but also an organizer that can induce elastic fiber assembly. Skin fibroblasts are known to develop abundant elastic fibers in serum-containing media. However, serum contains various known and unknown factors, such as cytokines and growth factors, and also causes increases of total cell number and cell density, making it difficult to identify specific molecules that are important for elastogenesis. Indeed, we detected a significant amount of fibulin-5 in serum using immunoprecipitation and ELISA (unpublished data), which might contribute to the elastogenic activity of the serum to some extent. We detected abundant deposition of fibulin-5 that may be derived from fibroblasts or from serum in serum-containing culture of skin fibroblasts, whereas endogenous fibulin-5 was scarcely detected in serum-free culture (Fig. S1). This indicates that the dose of fibulin-5 is crucial for elastic fiber organization in this serum-free cell culture, and that the endogenous expression of fibulin-5 alone is insufficient. Our serum-free elastogenesis system, in combination with gene knockdown by RNAi, would be a useful tool to identify other factors that regulate elastic fiber assembly, as unknown factors in the serum need not be taken into account. It has been reported that fibulin-5 can strongly interact with tropoelastin (). We found that fibulin-5 not only binds with tropoelastin, but also promotes coacervation of tropoelastins, and thus plays a more positive role in the alignment and deposition of tropoelastin. Fibrillin-1 has also recently been reported to promote coacervation (). Because fibulin-5 interacts with both fibrillin-1 and tropoelastins (; ), a fibulin-5–fibrillin-1 complex might cooperatively promote coacervation. Coacervated tropoelastin molecules need to be cross-linked by LOX family enzymes for the development of mature elastic fibers. Our data show that fibulin-5 also promotes cross-linking of tropoelastin, which may be mediated by recruiting LOXL1, 2, and 4 in the vicinity of coacervated tropoelastin (). We found that a much higher level of the truncated form of fibulin-5 was present in aged mouse skin than in young mouse skin (). There are several possible interpretations of this result: (a) fibulin-5 might be cleaved because of more protease activity in aged tissues; (b) truncated fibulin-5 might be stable and accumulate during life; or (c) truncated fibulin-5 might be more easily extracted from aged elastic fibers than from young elastic fibers. In any case, the truncated form of fibulin-5 cannot contribute to new elastogenesis, and therefore the decrease in full-length fibulin-5 may directly reflect a decrease in the elastogenic activity of aged tissues. In this study, we could not exclude the possibility that the truncated form of fibulin-5 might play some roles in elastogenesis in vivo. Although truncated fibulin-5 cannot contribute to new elastogenesis as it cannot interact with microfibrils, fibulin-5 protein cleaved after deposition onto microfibrils might take part in subsequent processes of elastogenesis, as binding with tropoelastin and facilitating coacervation of tropoelastin are not compromised by the N-terminal domain cleavage (unpublished data). To clarify the physiological consequences of the cleavage of fibulin-5, further studies will be needed using gene-targeted mice in which a cleavage-resistant mutant form of fibulin-5 is knocked-in. So far, elastic matrices have been considered to be designed to maintain elastic function for a lifetime (). Elastin is reported to have an exceptionally long half-life, and to have a turnover rate which approaches our lifespan (). On the other hand, elastase activity is provoked by exposure to sunlight and by smoking, causing degradation of the elastic fibers of the skin and lung, respectively (; ). This implies that elastogenic activity needs to be maintained even in the adult tissues to regenerate elastic fibers. Indeed, we detected a significant amount of fibulin-5 in the serum of aged mice (unpublished data), and we and others reported that the expression of fibulin-5 is up-regulated in some pathological conditions, such as in atherosclerotic lesions (; ; ), lung injury by elastase (), and pulmonary hypertension (). The potent elastogenic activity of fibulin-5 might provide future therapeutic applications for aging-associated diseases, such as emphysema and arteriosclerosis, or cosmetic applications for wrinkled skin. Application of fibulin-5 might be extended to tissue engineering, to make elastic skin or vessels by using cell cultures. Further studies are needed to examine whether fibulin-5 might improve age-related changes of elastic matrices in vivo. In summary, we demonstrated the elastic fiber organizing activity of fibulin-5 that is abrogated by age-associated proteolytic cleavage. We demonstrated that the addition of recombinant fibulin-5 protein induces elastic fiber development, and that a decrease in full-length fibulin-5 caused by proteolytic cleavage may contribute to the loss of the elastogenic potential in aged elastic tissues. Our findings might provide a molecular pathway to future therapeutic applications of the elastic fiber organizing activity of fibulin-5. 293T cells and human skin fibroblasts (HSFs) were maintained in DME (Sigma-Aldrich) supplemented with 2 mM glutamine, 10% penicillin/streptomycin, and 10% FBS at 37°C in 5% CO. HSFs were provided by M. Naito (Kyoto University, Kyoto, Japan). Human full-length fibulin-5 cDNA was cloned as previously described (). pEF6/V5 (Invitrogen) was modified by incorporation of a C-terminal FLAG-tag or Myc-tag (pEF6/FLAG, pEF6/Myc). The fibulin-5 ΔN1- (Δnt 247–399), ΔN2- (Δnt 247–504), ΔM- (Δnt 505–1,110), and ΔC- (Δnt 1, 114–1,512) fibulin-5 cDNAs were subcloned into pEF6/FLAG. Human fibulin-5 cDNA sequences are numbered according to GenBank accession no. . Human full-length LOXL1, 2, 3, and 4 cDNAs were obtained from IMAGE (available from GenBank under accession no. , , , and , respectively), and subcloned into pEF6/Myc. All constructs were confirmed by sequencing (ABI Prism 3100). pEF6/V5 (Invitrogen) was modified by incorporation of a C-terminal histidine and FLAG-tag. Human wild-type and each mutant fibulin-5 cDNA were subcloned into pEF6/FLAG-His. The respective stable blasticidin-resistant 293T cell clones were then obtained by transfection with these expression vectors. Recombinant proteins were purified from the serum-free conditioned medium of stable lines using Ni-NTA metal affinity resin (QIAGEN), and desalted with a Hi-Trap desalting column (GE Healthcare). Protein concentrations were determined with Coomassie Plus-200 Protein Assay Reagent (Pierce Chemical Co.). Human tropoelastin cDNA was cloned from human skin fibroblasts by RT-PCR. Of the four splice variants obtained, the longest cDNA, which codes one of the splicing variants of the elastin gene (ELNc; available from GenBank under accession no. ), was subcloned into pTrcHis vector (Invitrogen) and expressed in bacteria, followed by protein purification with Ni-NTA column (GE Healthcare), as previously described (). HSFs were subconfluently plated on microscope cover glasses (Fisher Scientific). 3 d after plating, culture media were changed to serum-free DME/F12 (Sigma-Aldrich), and purified recombinant fibulin-5 proteins were added, and the incubation was continued for another 10 d. The cells were fixed with 100% methanol, and blocked with 2% BSA. The primary antibodies used were anti-FLAG M2 monoclonal (1/100; Sigma-Aldrich), anti–human elastin polyclonal (1/100; EPC, PR533), anti–human fibulin-5 monoclonal (1/100; 10A), and anti–human fibrillin-1 polyclonal (1/100; EPC, PR217) antibodies. The secondary antibodies used were Alexa Fluor 488, 546, or 647 anti–rabbit or –mouse IgG (1/100; Invitrogen). After staining with DAPI, stained cells were mounted with a Prolong Gold Antifade Kit (Invitrogen). Fluorescence images were sequentially collected using a confocal microscope featuring 405-, 488-, 543-, and 633-nm laser lines with an open pinhole setting operated by the built-in software; either a microscope (LSM510 META; Carl Zeiss MicroImaging, Inc.) with a C-Apo 40×/1.2 NA lens ( and ) or a microscope (FV-1000; Olympus) with a UPlanSApo 40×/0.9 NA lens ( and S1). 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 CS2 (Adobe), and imported into Illustrator CS2 (Adobe) for assembly. Anti–human fibulin-5 monoclonal antibody (10A) was raised by Iwaki Co., Ltd. by immunization of fibulin-5–deficient mice with purified recombinant human fibulin-5 protein. Coacervation of tropoelastin with or without fibulin-5 protein was assessed using the method described (). Soluble tropoelastin and fibulin-5 were prepared in PBS and mixed on ice. Each reaction mixture was transferred to an 8-channel quartz cuvette for light scattering measurements at 440 nm. The temperature of a spectrophotometer (UV-1650; Shimazu) fitted with a heating/cooling block was regulated by a circulating water bath. All channels were simultaneously cooled and heated in the heating/cooling block by the circulating water. 293T cells were transfected using Lipofectamine PLUS (Invitrogen). The mixtures of conditioned media were subjected to immunoprecipitation with anti-FLAG M2 affinity gel (Sigma-Aldrich). After washing, immune complexes were resolved by SDS-PAGE, transferred to PVDF membranes (Bio-Rad Laboratories), and reacted with HRP-conjugated anti-Myc monoclonal (9E10; 1/500; Santa Cruz Biotechnology) or anti-FLAG M2 monoclonal (1/1,000; Sigma-Aldrich) antibody. Signals were detected using Western Lightning Chemiluminescence Reagent PLUS (Perkin Elmer). Proteins were extensively extracted from skin tissues with 8 M urea solution. After dialysis against PBS, the protein concentration was measured with Coomassie Plus-200 Protein Assay Reagent. Protein samples of the same amount were subjected to SDS-PAGE (Invitrogen), followed by Western blotting with anti–fibulin-5 antibody (1/400, BSYN2473). HRP-conjugated anti–rabbit antibody (1/2,000; Santa Cruz Biotechnology) was used as a secondary antibody. For transmission electron microscopy, skin tissues were fixed with 4% paraformaldehyde, 2.5% glutaraldehyde, 0.1% tannic acid in 0.1 M cacodylate buffer. Sections were then stained with uranyl acetate. HSFs were subconfluently plated on 60-mm dishes in quadruplicates. 3 d after plating, 20 μCi [H]valine was added to each dish, together with purified recombinant fibulin-5 protein or cleaved fibulin-5 protein. The cultures were then incubated at 37°C in 5% CO for 10 d. The cells were harvested in 0.1 M acetic acid on ice. After centrifugation, the pellets were boiled in 0.1 N NaOH for 1 h. Subsequently, the NaOH-insoluble pellets were boiled with 5.7 N HCl for 1 h, mixed with scintillation fluid, and measured for radioactivity with a scintillation counter (Beckman Coulter). Total RNAs were extracted using RNeasy Plus Mini Kit (QIAGEN) and transcribed to cDNA with random hexamers using the SuperScript III First-Strand Synthesis System (Invitrogen). For quantitative PCR, the reaction was performed with a QuantiTect SYBR Green PCR kit (QIAGEN), and the products were analyzed with the Mx3000P QPCR System (Stratagene). Primers used for quantitative PCR are provided in Table S1 (available at ). Fig. S1 shows that serum induces fibulin-5 deposition on microfibrils. Table S1 shows primers used for quantitative PCR. The online version of this article is available at .

xref #text Range and standard deviation (SD) are used for descriptive error bars because they show how the data are spread (). Range error bars encompass the lowest and highest values. SD is calculated by the formulawhere refers to the individual data points, is the mean, and Σ (sigma) means add to find the sum, for all the data points. SD is, roughly, the average or typical difference between the data points and their mean, . About two thirds of the data points will lie within the region of mean ± 1 SD, and ∼95% of the data points will be within 2 SD of the mean. As you increase the size of your sample, or repeat the experiment more times, the mean of your results () will tend to get closer and closer to the true mean, or the mean of the whole population, μ. We can use as our best estimate of the unknown μ. Similarly, as you repeat an experiment more and more times, the SD of your results will tend to more and more closely approximate the true standard deviation (σ) that you would get if the experiment was performed an infinite number of times, or on the whole population. However, the SD of the experimental results will approximate to σ, whether is large or small. Like , SD does not change systematically as changes, and we can use SD as our best estimate of the unknown σ, whatever the value of . In experimental biology it is more common to be interested in comparing samples from two groups, to see if they are different. For example, you might be comparing wild-type mice with mutant mice, or drug with placebo, or experimental results with controls. To make inferences from the data (i.e., to make a judgment whether the groups are significantly different, or whether the differences might just be due to random fluctuation or chance), a different type of error bar can be used. These are standard error (SE) bars and confidence intervals (CIs). The mean of the data, , with SE or CI error bars, gives an indication of the region where you can expect the mean of the whole possible set of results, or the whole population, μ, to lie (). The interval defines the values that are most plausible for μ. Because error bars can be descriptive or inferential, and could be any of the bars listed in or even something else, they are meaningless, or misleading, if the figure legend does not state what kind they are. This leads to the first rule. when showing error bars, always describe in the figure legends what they are. y o u c a r r y o u t a s t a t i s t i c a l s i g n i f i c a n c e t e s t , t h e r e s u l t i s a P v a l u e , w h e r e P i s t h e p r o b a b i l i t y t h a t , i f t h e r e r e a l l y i s n o d i f f e r e n c e , y o u w o u l d g e t , b y c h a n c e , a d i f f e r e n c e a s l a r g e a s t h e o n e y o u o b s e r v e d , o r e v e n l a r g e r . O t h e r t h i n g s ( e . g . a m p l e s i z e , v a r i a t i o n ) b e i n g e q u a l , a l a r g e r d i f f e r e n c e i n r e s u l t s g i v e s a l o w e r P v a l u e , w h i c h m a k e s y o u s u s p e c t t h e r e i s a t r u e d i f f e r e n c e . B y c o n v e n t i o n , i f P < 0 . 0 5 y o u s a y t h e r e s u l t i s s t a t i s t i c a l l y s i g n i f i c a n t , a n d i f P < 0 . 0 1 y o u s a y t h e r e s u l t i s h i g h l y s i g n i f i c a n t a n d y o u c a n b e m o r e c o n f i d e n t y o u h a v e f o u n d a t r u e e f f e c t . A s a l w a y s w i t h s t a t i s t i c a l i n f e r e n c e , y o u m a y b e w r o n g ! P e r h a p s t h e r e r e a l l y i s n o e f f e c t , a n d y o u h a d t h e b a d l u c k t o g e t o n e o f t h e 5 % ( i f P < 0 . 0 5 ) o r 1 % ( i f P < 0 . 0 1 ) o f s e t s o f r e s u l t s t h a t s u g g e s t s a d i f f e r e n c e w h e r e t h e r e i s n o n e . O f c o u r s e , e v e n i f r e s u l t s a r e s t a t i s t i c a l l y h i g h l y s i g n i f i c a n t , i t d o e s n o t m e a n t h e y a r e n e c e s s a r i l y b i o l o g i c a l l y i m p o r t a n t . I t i s a l s o e s s e n t i a l t o n o t e t h a t i f P > 0 . 0 5 , a n d y o u t h e r e f o r e c a n n o t c o n c l u d e t h e r e i s a s t a t i s t i c a l l y s i g n i f i c a n t e f f e c t , y o u m a y n o t c o n c l u d e t h a t t h e e f f e c t i s z e r o . T h e r e m a y b e a r e a l e f f e c t , b u t i t i s s m a l l , o r y o u m a y n o t h a v e r e p e a t e d y o u r e x p e r i m e n t o f t e n e n o u g h t o r e v e a l i t . I t i s a c o m m o n a n d s e r i o u s e r r o r t o c o n c l u d e “ n o e f f e c t e x i s t s ” j u s t b e c a u s e P i s g r e a t e r t h a n 0 . 0 5 . I f y o u m e a s u r e d t h e h e i g h t s o f t h r e e m a l e a n d t h r e e f e m a l e B i d d e l o n i a n b a s k e t b a l l p l a y e r s , a n d d i d n o t s e e a s i g n i f i c a n t d i f f e r e n c e , y o u c o u l d n o t c o n c l u d e t h a t s e x h a s n o r e l a t i o n s h i p w i t h h e i g h t , a s a l a r g e r s a m p l e s i z e m i g h t r e v e a l o n e . A b i g a d v a n t a g e o f i n f e r e n t i a l e r r o r b a r s i s t h a t t h e i r l e n g t h g i v e s a g r a p h i c s i g n a l o f h o w m u c h u n c e r t a i n t y t h e r e i s i n t h e d a t a : T h e t r u e v a l u e o f t h e m e a n μ w e a r e e s t i m a t i n g c o u l d p l a u s i b l y b e a n y w h e r e i n t h e 9 5 % C I . W i d e i n f e r e n t i a l b a r s i n d i c a t e l a r g e e r r o r ; s h o r t i n f e r e n t i a l b a r s i n d i c a t e h i g h p r e c i s i o n . Science typically copes with the wide variation that occurs in nature by measuring a number () of independently sampled individuals, independently conducted experiments, or independent observations. the value of (i.e., the sample size, or the number of independently performed experiments) must be stated in the figure legend. It is essential that (the number of independent results) is carefully distinguished from the number of replicates, which refers to repetition of measurement on one individual in a single condition, or multiple measurements of the same or identical samples. Consider trying to determine whether deletion of a gene in mice affects tail length. We could choose one mutant mouse and one wild type, and perform 20 replicate measurements of each of their tails. We could calculate the means, SDs, and SEs of the replicate measurements, but these would not permit us to answer the central question of whether gene deletion affects tail length, because would equal 1 for each genotype, no matter how often each tail was measured. To address the question successfully we must distinguish the possible effect of gene deletion from natural animal-to-animal variation, and to do this we need to measure the tail lengths of a number of mice, including several mutants and several wild types, with > 1 for each type. Similarly, a number of replicate cell cultures can be made by pipetting the same volume of cells from the same stock culture into adjacent wells of a tissue culture plate, and subsequently treating them identically. Although it would be possible to assay the plate and determine the means and errors of the replicate wells, the errors would reflect the accuracy of pipetting, not the reproduciblity of the differences between the experimental cells and the control cells. For replicates, = 1, and it is therefore inappropriate to show error bars or statistics. If an experiment involves triplicate cultures, and is repeated four independent times, then = 4, not 3 or 12. The variation within each set of triplicates is related to the fidelity with which the replicates were created, and is irrelevant to the hypothesis being tested. To identify the appropriate value for , think of what entire population is being sampled, or what the entire set of experiments would be if all possible ones of that type were performed. Conclusions can be drawn only about that population, so make sure it is appropriate to the question the research is intended to answer. In the example of replicate cultures from the one stock of cells, the population being sampled is the stock cell culture. For to be greater than 1, the experiment would have to be performed using separate stock cultures, or separate cell clones of the same type. Again, consider the population you wish to make inferences about—it is unlikely to be just a single stock culture. Whenever you see a figure with very small error bars (such as ), you should ask yourself whether the very small variation implied by the error bars is due to analysis of replicates rather than independent samples. If so, the bars are useless for making the inference you are considering. Sometimes a figure shows only the data for a representative experiment, implying that several other similar experiments were also conducted. If a representative experiment is shown, then = 1, and no error bars or P values should be shown. Instead, the means and errors of all the independent experiments should be given, where is the number of experiments performed. error bars and statistics should only be shown for independently repeated experiments, and never for replicates. If a “representative” experiment is shown, it should not have error bars or P values, because in such an experiment, = 1 ( shows what not to do). italic #text Suppose three experiments gave measurements of 28.7, 38.7, and 52.6, which are the data points in the = 3 case at the left in . The mean of the data is = 40.0, and the SD = 12.0, which is the length of each arm of the SD bars. (in this case 40.0) is the best estimate of the true mean μ that we would like to know. But how accurate an estimate is it? This can be shown by inferential error bars such as standard error (SE, sometimes referred to as the standard error of the mean, SEM) or a confidence interval (CI). SE is defined as SE = SD/√. In , the large dots mark the means of the same three samples as in . For the = 3 case, SE = 12.0/√3 = 6.93, and this is the length of each arm of the SE bars shown. The SE varies inversely with the square root of , so the more often an experiment is repeated, or the more samples are measured, the smaller the SE becomes (). This allows more and more accurate estimates of the true mean, μ, by the mean of the experimental results, . We illustrate and give rules for = 3 not because we recommend using such a small , but because researchers currently often use such small values and it is necessary to be able to interpret their papers. It is highly desirable to use larger , to achieve narrower inferential error bars and more precise estimates of true population values. illustrates what happens if, hypothetically, 20 different labs performed the same experiments, with = 10 in each case. The 95% CI error bars are approximately ± 2xSE, and they vary in position because of course varies from lab to lab, and they also vary in width because SE varies. Such error bars capture the true mean μ on ∼95% of occasions—in , the results from 18 out of the 20 labs happen to include μ. The trouble is in real life we don't know μ, and we never know if our error bar interval is in the 95% majority and includes μ, or by bad luck is one of the 5% of cases that just misses μ. The error bars in are only approximately ± 2xSE. They are in fact 95% CIs, which are designed by statisticians so in the long run exactly 95% will capture μ. is a critical value from tables of the statistic. This critical value varies with . For = 10 or more it is ∼2, but for small it increases, and for = 3 it is ∼4. Therefore ± 2xSE intervals are quite good approximations to 95% CIs when is 10 or more, but not for small . CIs can be thought of as SE bars that have been adjusted by a factor () so they can be interpreted the same way, regardless of . This relation means you can easily swap in your mind's eye between SE bars and 95% CIs. If a figure shows SE bars you can mentally double them in width, to get approximate 95% CIs, as long as is 10 or more. However, if = 3, you need to multiply the SE bars by 4. 95% CIs capture μ on 95% of occasions, so you can be 95% confident your interval includes μ. SE bars can be doubled in width to get the approximate 95% CI, provided is 10 or more. If = 3, SE bars must be multiplied by 4 to get the approximate 95% CI. Determining CIs requires slightly more calculating by the authors of a paper, but for people reading it, CIs make things easier to understand, as they mean the same thing regardless of . For this reason, in medicine, CIs have been recommended for more than 20 years, and are required by many journals (). illustrates the relation between SD, SE, and 95% CI. The data points are shown as dots to emphasize the different values of (from 3 to 30). The leftmost error bars show SD, the same in each case. The middle error bars show 95% CIs, and the bars on the right show SE bars—both these types of bars vary greatly with , and are especially wide for small . ; the values are shown at the bottom of the figure. Note also that, whatever error bars are shown, it can be helpful to the reader to show the individual data points, especially for small , as in and , and rule 4. When comparing two sets of results, e.g., from knock-out mice and wild-type mice, you can compare the SE bars or the 95% CIs on the two means (). The smaller the overlap of bars, or the larger the gap between bars, the smaller the P value and the stronger the evidence for a true difference. As well as noting whether the figure shows SE bars or 95% CIs, it is vital to note , because the rules giving approximate P are different for = 3 and for ≥ 10. illustrates the rules for SE bars. The panels on the right show what is needed when ≥ 10: a gap equal to SE indicates P ≈ 0.05 and a gap of 2SE indicates P ≈ 0.01. To assess the gap, use the average SE for the two groups, meaning the average of one arm of the group C bars and one arm of the E bars. However, if = 3 (the number beloved of joke tellers, Snark hunters (), and experimental biologists), the P value has to be estimated differently. In this case, P ≈ 0.05 if double the SE bars just touch, meaning a gap of 2 SE. when = 3, and double the SE bars don't overlap, P < 0.05, and if double the SE bars just touch, P is close to 0.05 (, leftmost panel). If is 10 or more, a gap of SE indicates P ≈ 0.05 and a gap of 2 SE indicates P ≈ 0.01 (, right panels). Rule 5 states how SE bars relate to 95% CIs. Combining that relation with rule 6 for SE bars gives the rules for 95% CIs, which are illustrated in . When ≥ 10 (right panels), overlap of half of one arm indicates P ≈ 0.05, and just touching means P ≈ 0.01. To assess overlap, use the average of one arm of the group C interval and one arm of the E interval. If = 3 (left panels), P ≈ 0.05 when two arms entirely overlap so each mean is about lined up with the end of the other CI. If the overlap is 0.5, P ≈ 0.01. with 95% CIs and = 3, overlap of one full arm indicates P ≈ 0.05, and overlap of half an arm indicates P ≈ 0.01 (, left panels). The rules illustrated in and apply when the means are independent. If two measurements are correlated, as for example with tests at different times on the same group of animals, or kinetic measurements of the same cultures or reactions, the CIs (or SEs) do not give the information needed to assess the significance of the differences between means of the same group at different times because they are not sensitive to correlations within the group. Consider the example in , in which groups of independent experimental and control cell cultures are each measured at four times. Error bars can only be used to compare the experimental to control groups at any one time point. Whether the error bars are 95% CIs or SE bars, they can only be used to assess between group differences (e.g., E1 vs. C1, E3 vs. C3), and may not be used to assess within group differences, such as E1 vs. E2. Assessing a within group difference, for example E1 vs. E2, requires an analysis that takes account of the within group correlation, for example a Wilcoxon or paired t analysis. A graphical approach would require finding the E1 vs. E2 difference for each culture (or animal) in the group, then graphing the single mean of those differences, with error bars that are the SE or 95% CI calculated from those differences. If that 95% CI does not include 0, there is a statistically significant difference (P < 0.05) between E1 and E2. in the case of repeated measurements on the same group (e.g., of animals, individuals, cultures, or reactions), CIs or SE bars are irrelevant to comparisons within the same group (). #text

Centrosomes are critical microtubule (MT) nucleators and organizers in animal cells (). Centrioles form the centrosome core and are surrounded by pericentriolar material (PCM) containing MT nucleating factors like γ-tubulin (γtub; ). Centrosomes play key roles in many processes, including organizing mitotic spindle poles (). In animal cells, centrosome duplication occurs by a conserved cycle (). It begins with centriole disengagement in late mitosis (), followed by procentriole assembly along the wall of each centriole in S phase. By G2, cells contain two mother/daughter centriole pairs that remain in proximity until mitosis. Both centriole pairs form functional centrosomes, maturing synchronously before mitotic entry, by recruiting PCM and acting as MT organizing centers (MTOCs; in contrast, there is a 10-min delay in activating the second yeast MTOC; ). The centrosomes then move to opposite sides of the nucleus to organize spindle poles and asters that position the spindle with respect to cortical cues. The essential role of centrosomes in animal cells was called into question by the fact that flies lacking functional centrosomes, or lacking centrioles entirely, live to adulthood (; ). However, not all is well: these animals have defects in divisions of larval neural stem/progenitor cells, the central brain neuroblasts (NBs). Adult tissue stem cells play key roles in tissue maintenance/repair (). In each division, the daughters differ in fate: one retains stem cell character and the other differentiates. central brain NBs are a superb model for asymmetric divisions of postembryonic tissue stem cells (; ). Both embryonic and larval NBs are polarized cells exhibiting strict division patterns crucial for their roles as stem cells. Unlike the precise relationship between the embryonic NB division axis and adjacent epithelium, larval central brain NBs () do not appear to divide with specific orientations relative to the brain as a whole (). However, each NB creates a simpler microenvironment (): the NB and its ganglion mother cell (GMC) daughters. NBs divide asymmetrically, and the NB daughter retains stem cell character, whereas the GMC daughter goes on to differentiate. NBs divide according to strict local rules; each GMC is born adjacent to the previous GMC ( and Video 1, available at ; ), creating a GMC cap on one side of the NB (). Although differential fate allocation is critical in stem cells, we have much to learn about how a stereotyped division axis is established. NBs must coordinate cortical and spindle polarity so that neural determinants are packaged into the differentiating daughter (). Mutations affecting polarity or astral MT cortical interactions result in asymmetric division defects (). The importance of a properly aligned spindle is also suggested by spindle alignment defects in the absence of centrioles (14% symmetric divisions; ) or in mutants that lack PCM ( [] or []) and have few or no astral MTs (; ). Thus, proper interactions between the spindle, astral MTs, and cortical polarity cues help maintain a constant division axis. Previous analyses revealed that NB spindles form at prophase already roughly aligned with the ultimate division axis () but did not define how the initial axis forms. Here, we address how this model stem cell maintains a persistent division axis. male germline stem cells also have a persistent division axis. It was proposed that one centrosome is cortically anchored by MT–adherens junction interactions (). To test whether a similar mechanism exists in NBs, we analyzed the centrosome cycle using 4D or 5D spinning disk confocal microscopy on brains prepared with no physical distortion (Fig. S1 A, available at ), maintaining NB shape to replicate normal mitosis. By prophase, NBs contain two MTOCs that are almost fully separated and aligned along the NB/GMC axis (), but analysis of fixed NBs revealed a single MTOC positioned opposite the GMCs before mitotic entry (). We thus examined MTOC behavior throughout the cell cycle as an initial approach to test the hypothesis that fixing the position of one MTOC through successive divisions helps ensure persistent spindle orientation. We analyzed live NBs expressing GFP-G147, an MT-associated protein (), revealing a striking temporal difference in MTOC activity. During interphase, a single detectable MTOC persists opposite the previous division site; we refer to this as the dominant MTOC. As NBs approach mitosis, this MTOC increases activity (matures; empirically judged by size and MT fluorescence intensity), forming an MT basket around the nucleus (, 0:00, arrows; and Videos 2 and 3, available at ). We refer to this as preprophase; this stage is also seen in fixed samples stained for tubulin (Fig. S1 B). Soon after, sometime before the dominant MTOC fully matures, something striking happens: a second MTOC appears distant from the first (, 1:45, arrowheads; and Video 2). We refer to this as the second MTOC and this stage as prophase onset. The second MTOC increases activity, maturing ∼10 min before nuclear envelope breakdown (NEB; , 1:45–13:24). Using 4D imaging, we excluded the possibility that the second MTOC was present earlier in another focal plane. To further assess this, we imaged forming spindles end on ( and Video 3). It is clear that the second MTOC did not emerge from the dominant MTOC (, top) or travel around the nucleus (, middle). Instead, the second MTOC appeared roughly opposite the dominant MTOC (, bottom, arrowheads; 132 ± 38° from the dominant one, using the centroid of the nucleus as a fixed reference; = 30; , prophase). MTOC separation began immediately, and by NEB, they were 146 ± 20° apart ( = 18, ; this is slightly less than seen by [171°], likely because of different measurement methods). Thus, NBs form two distinct MTOCs: an MTOC persisting from the previous division and another only activated at mitotic entry. This distant activation of the second MTOC raised questions about the centrosome cycle. One possibility is that NBs have two MT nucleating centrosomes, but only one can retain MTs and act as an MTOC during interphase, whereas the second acquires MT retention ability during mitotic entry, explaining the second MTOC's sudden appearance. There is precedent for this: mouse L929 cells have two γtub-bearing centrosomes that can nucleate MTs, but only one contains Ninein and can retain MTs to form an MTOC (). To test this hypothesis in NBs, we used EB1-GFP. This binds growing MT plus ends and reliably identifies MT nucleation sites (; ). Only one nucleation site was present in interphase and preprophase (, arrows; 0:00; z series not depicted), and a new nucleation site appeared distant from the first (, arrowheads), consistent with spatially and temporally distinct second MTOC activation. Thus, NBs regulate MT nucleation and not just MT retention. To examine how the new nucleation center forms, we imaged centrosomes using a PCM protein, GFP-Cnn (). NBs contain a single detectable centrosome during interphase (, 0:51–2:36). When NBs reenter mitosis, a second Cnn-positive centrosome appears distant from the first (, 2:36, blue arrow), mimicking activation of the second MTOC. To verify that these occur simultaneously, we imaged NBs expressing mCherry-Tubulin (chTub) and GFP-Cnn ( and Fig. S2 B, available at ). This revealed perfect temporal and spatial correlation between the appearance of the second centrosome and activation of the second MTOC (, arrowheads). We never saw physical separation of two centrosomes/MTOCs ( > 60). To our knowledge, this is the first example of asynchronous and physically distant centrosome maturation, suggesting that NBs use a novel centrosome cycle. Higher temporal/spatial resolution imaging revealed that two GFP-Cnn spots separate during mitotic exit (, inset; and Video 5, available at ). One GFP-Cnn spot persists as the NB interphase centrosome, forming the dominant MTOC, whereas the other spot disappears. The persistent Cnn spot (centrosome) remains relatively stationary in interphase (see The dominant centrosome predicts spindle alignment), consistent with the hypothesis that coarse spindle alignment begins in interphase by anchoring the dominant centrosome (, 0:00–2:48, red arrows). We also examined centrosome fate in the two daughters (new NB and GMC). They differ dramatically in PCM retention, in contrast to mammalian cells, where both daughters' centrosomes retain PCM. The GMC centrosome sheds all PCM (, 0:27–0:51, green arrowheads; centrioles remain [see Asymmetric centrosome regulation]; GMCs regain PCM when reentering mitosis, ). The new NB centrosome (that becomes the dominant MTOC) retains PCM throughout interphase (, 0:27–0:51, red arrows) and further accumulates PCM during the next mitosis (, 1:39–2:36, red arrows and Fig. S2 A). The complete shedding of PCM in GMCs appears to be the normal behavior of interphase centrosomes in most fly cells (; Rogers, G., personal communication), whereas in syncytial early embryos, both daughters retain PCM foci through the cell cycle. In contrast to both cell types, the NB daughters exhibit differential PCM retention. We examined centrioles live to test this hypothesis, using the centriole marker GFP-PACT () and Histone-GFP (). Mother/daughter centrioles disengaged in late telophase (, 0:24; and Fig. S1 C), as in mammalian cells and fly embryos (; ). Thus, two NB centrioles are present throughout interphase despite the presence of only one MTOC. The two centrioles then exhibit different behaviors. One remains fairly stationary (, arrows), whereas the second moves to roughly the other side of the nucleus (arrowheads). Disengagement perfectly correlates with separation of Cnn spots (), suggesting that the stationary centriole retains PCM to form the dominant MTOC and the mobile centriole completely sheds PCM. To test this, we imaged NBs expressing chTub and GFP-PACT ( and Video 6, available at ). The stationary centriole retained MTs (, arrows), whereas the mobile centriole did not (arrowheads). Upon reentering mitosis, the mobile centriole regained nucleation activity, forming the second MTOC. This suggests that full separation of the MTOCs that organize the spindle is biphasic. It begins in interphase, when one centriole retains PCM, remains stationary, and forms the dominant MTOC, whereas the second centriole sheds PCM and becomes mobile. Movement of the second centriole away from the dominant MTOC in interphase accounts for ∼70% (132/180°; ) of the separation needed to form a spindle. Mechanisms of transporting the mobile centriole remain to be identified, but it is nonrandom, as in 26/30 NBs, the second MTOC emerged ≥90° from the dominant MTOC (). After the second MTOC is activated, the two separate the last 30%, most likely via MT sliding forces. This might explain defects in mutants, where MTOCs are only separated by 124° at NEB (). Perhaps interphase centriole movement is normal, but MT-based MTOC separation is defective. These data suggest that NBs differentially regulate the activity of their two centrioles within the same cytoplasm. Interestingly, a similar observation was made in clam eggs, which have three centrosomes just after fertilization. The sperm centrosome is functionally inactivated, whereas female centrosomes organize the meiotic spindle (). We next examined NB centrosome regulation. In preprophase, one centriole (marked by anti-DPLP; Fig. S1 C) formed the dominant MTOC (, arrows), whereas the second centriole had no associated MTs and was randomly positioned (, arrowheads), confirming our live-cell data. We thus examined whether γtub is recruited asymmetrically. Fixed preprophase NBs had two centrioles; only that opposite the GMCs accumulated γtub (; neither GMC centriole carried γtub [yellow arrowheads], consistent with complete Cnn loss in interphase GMCs). Further, both γtub and Cnn are absent from the NB centriole nearest the GMCs in interphase/preprophase (Fig. S2 C). Polo kinase promotes centrosome maturation by promoting γtub recruitment during mitotic entry (). Differences in Polo localization/activity might underlie differences in timing of NB centrosome maturation. We examined NBs expressing Polo-GFP and the centriole marker mCherry-DSAS-6 (). Only the centriole pair that forms the dominant centrosome was Polo-GFP positive during preprophase (, arrows). Polo-GFP accumulated on the mobile centriole pair as the NB entered mitosis (, arrowheads; and Video 7, available at ), increased on both centriole pairs through prophase, and moved on to kinetochores (). When we imaged Polo-GFP in cells exiting mitosis, we could see it retained at low levels on the dominant centrosome ( and Fig. S2, D–G). In the future, it will be interesting to examine the localization of Aurora A, another centrosome regulator. Unlike the distal appendages of mammalian mother centrioles, fly mother and daughter centrioles have no known ultrastructural () or molecular differences. Our data suggest that differences exist. It is unlikely that this differential regulation is a result of location, as both centrioles are initially adjacent after disengagement. The differences may be due to centriole age or procentriole maturation state. The NB spindle is largely aligned by NEB (). Based on our data, we tested the hypothesis that the dominant centrosome helps define one spindle pole before prophase. We calculated the angle between the dominant centrosome/MTOC axis (, top) and the anaphase axis (bottom), using the nuclear centroid as a fixed reference. This revealed two phases in defining the future spindle axis. Through prophase onset, the dominant centrosome remains fairly stationary roughly opposite the GMCs (, coarse alignment), agreeing with fixed images (), whereas the second centriole moves to a distal position (to within 46 ± 33° [ = 25] of the anaphase axis; , prophase). This is consistent with our hypothesis. The dominant centrosome may be immobilized by aster–cortex interactions or by absence of an active displacement mechanism. In the second phase, alignment is refined in prophase and prometaphase (the angle between the NB centrosome and anaphase axes decreases from 31 ± 29° to 15 ± 12°; = 15), as shown by . To further test whether anchoring the dominant centrosome helps roughly align the spindle, we imaged mutant NBs live. They lack functional centrosomes (; Fig. S3 A, available at ) and astral MTs. Mutant NBs lack a dominant interphase centrosome, allowing us to assess its role in spindle orientation and asymmetric cell division. Live imaging revealed robust chromatin-mediated MT nucleation and spindle assembly producing fairly normal spindles ( and Video 9). Spindle poles emerge from a disorganized MT array near the chromosomes that focuses as the spindle lengthened. Spindles do not rotate during formation, always forming along the initial pole separation axis, but do rotate during metaphase (23 ± 15°; = 11), suggesting that rotation can occur without astral MTs or that mutants have a reduced astral array sufficient for rotation (). Surprisingly, consecutive divisions in mutants usually produce adjacent or near-adjacent daughters ( = 5/5; and Video 10), as in wild type (). In a few cases, however, spindles form parallel to the GMC cap and, presumably, the polarity axis (2/13; ∼15%); these NBs divide symmetrically (). This suggests that the second phase of spindle alignment can occur without a dominant centrosome and can rescue misalignment, as long as it is not too extreme, but occasional atypical symmetric divisions occur. This results in defective brain anatomy, with ectopic paired, smaller NBs, presumably progeny of symmetric divisions (Fig. S3 B). Our data reveal two new aspects of asymmetric division in this stem cell model. First, cells can adjust the canonical centrosome cycle to allow novel cell behaviors, as was observed during clam meiosis (). Central brain NBs also alter this cycle: rather than both centrosomes maturing in synchrony and proximity (), the two centriole pairs are differentially regulated, maturing asynchronously and distant from one another (). One retains MT nucleating activity throughout the cell cycle, forming the dominant MTOC during interphase, whereas the second is initially inactive, only forming a functional centrosome and nucleating MTs at mitotic entry. One speculative possibility is that these are mother and daughter centrioles and that one is preferentially retained in the stem cell, a hypothesis that will now be tested. It is also of interest to ask whether other stem cells use this mechanism. Second, our data suggest that this novel centrosome cycle helps ensure high-fidelity spindle positioning and thus asymmetric division (). We propose a model in which NB mitotic spindles are aligned in two phases to ensure that GMC daughters are born next to the previous GMC. Rough alignment is achieved by confining the dominant MTOC to a relatively fixed position from the previous division and moving the second centriole to the other side of the cell. As spindles form, a second process refines this initial alignment. In mutants, without centrosomes, the first mechanism is inactive, but the second mechanism can align the spindle unless initial alignment is wildly off axis (). In mutants, centriole separation must occur normally, as prophase MTOCs are nearly fully separated, but alignment of spindle poles to cortical polarity cues is defective (). The normal two-step process is a robust mechanism ensuring successful asymmetric divisions and reproducible brain anatomy. flies were the wild-type controls for all immunostained samples. For live-cell imaging, we used the following strains: (), (), (GFP-tagged MT-associated protein; ), (a gift from S. Rogers [University of North Carolina at Chapel Hill, Chapel Hill, NC] and B. Eaton [University of Texas at San Antonio, San Antonio, TX]), (), and (). We generated transgenic flies of the genotype and by using a standard P-element transformation (). mCherry-α-tubulin (human tubulin) was PCR amplified from an unknown expression vector (a gift from A. Straight, Stanford University, Stanford, CA) and cloned into the pUASg vector. mCherry-SAS-6 (generated by G. Rogers, University of North Carolina at Chapel Hill) is expressed under its endogenous promoter and was cloned into the pCaSpeR4 vector. All UAS promoters were driven by Gal4-1407 (Bloomington Drosophila Stock Center). For fixed samples of mutants, we identified homozygous larvae by selecting against the marker on the Balancer (). For live imaging of MTs in the background, we generated recombinants of the genotype . Crawling third instar larvae were dissected in Schneider's Drosophila Medium (Invitrogen) with 10% FCS. The entire brain was explanted and placed anterior side down (ventral nerve cord upward) in our imaging chamber (Fig. S1). Brains were allowed to settle in the center of a pool of media in a glass-bottomed dish (MatTek). The media was surrounded by Halocarbon oil 700, which supported a glass coverslip used to seal the chamber. Samples were imaged using a Yokogawa spinning disk confocal (PerkinElmer) mounted on a microscope (Eclipse TE300; Nikon). It is equipped with an interline cooled charge-coupled device camera (ORCA-ER; Hamamatsu), a z-focus motor (Prior Scientific), an excitation and an emission wheel controlled by the Lambda 10-2 controller (Sutter Instrument) and emission filters from Semrock. Objectives used were 100× 1.4 NA, 60× 1.4 NA, and 40× oil 1.3 NA. 4D and 5D (x, y, z, time, wavelength) video sequences were collected using the multidimensional acquisition add-on in MetaMorph (Molecular Devices). Brains of and flies were fixed in 9% formaldehyde or 4% paraformaldehyde in PTA (PBS + 0.1% Tween 20 + 0.2 g/l sodium azide) for 15 min, blocked in 1% normal goat serum for 3 h, and stained in a microcentrifuge tube in primary antibody and 1% normal goat serum in PTA overnight at 4°C. Brains were washed and incubated in secondary antibodies for 2 h at room temperature. The following antibodies were used: E7 mouse anti–α-tubulin (1:250; Developmental Studies Hybridoma Bank), rabbit anti-DPLP (1:1,000; ), mouse GTU-88 anti–γ-tubulin (1:500; Sigma-Aldrich), and rabbit anti-GFP (1:750; ab290 [Abcam]). Secondary antibodies were Alexa 488 and 546 (Invitrogen) and were used at a final concentration of 1:500. For each selected time point, the (x, y, z) coordinates of the centrosome was recorded. We also recorded the coordinates for the point of origin at each time point, which we designated as center of the nucleus from interphase to NEB, the center of the chromosomal mass at initial metaphase, and half the distance between the slightly separated sister chromatids at anaphase onset. At each time point, the origin was normalized to and the centrosome coordinates were adjusted accordingly. This method eliminated the effects of x, y, z stage/microscope drift. The following equation was used to measure the angle between the two defined vectors (x1, y1, z1) and (x2, y2, z2): Dot Product = (x1 × x2) + (y1 × y2) + (z1 × z2) = L1 × L2 × cos(Θ), where L, length of vector, equals the square root of (x + y + z) and Θ is the angle between the two vectors. Note that all the measured angles are in relation to the anaphase-onset vector, which was always designated as the (x2, y2, z2) vector. For the interphase time points, we used GFP-Cnn (, blue), because the interphase centrosome could not always be identified with high confidence using an MT marker. We used GFP-G147 to stage cells at prophase (appearance of second MTOC), NEB (flood of fluorescence into the nucleus), initial metaphase (judged by spindle shape), and anaphase onset (kinetochore MT shortening; , pink). Video 1 shows wild-type NBs expressing actin-GFP through two rounds of mitosis. Video 2 presents a side view of wild-type GFP-G147–expressing NBs during mitotic entry. Video 3 gives an end-on view of wild-type GFP-G147–expressing NBs during mitotic entry. Video 4 shows wild-type NBs expressing GFP-Cnn through an entire cell cycle. Video 5 shows wild-type NB expressing chTub and GFP-Cnn, showing PCM splitting during mitotic exit. Video 6 presents wild-type NB expressing chTub and GFP-PACT. Video 7 shows wild-type NB expressing mCherry-SAS-6 and Polo-GFP during mitotic entry. Video 8 shows wild-type GFP-G147–expressing NBs during mitotic entry and through the end of telophase. Video 9 shows mutant NBs during spindle assembly and through mitosis. Video 10 shows mutant NBs through two rounds of mitosis. Fig. S1 provides a sample preparation and MT distribution in NBs. Fig. S2 shows CNN and Polo behavior throughout the cell cycle. Fig. S3 shows that mutant brains contain supernumerary central brain NBs. Online supplemental material is available at .

Many internal organs are lined by a monolayer of polarized epithelia with separate apical (AP) and basolateral (BL) surfaces that are defined by distinct protein and lipid compositions and are separated by tight junctions (). The AP surface serves as a barrier to the outside world and is specialized for the exchange of materials with the lumen. The BL surface is adapted for interaction with other cells and for exchange with the bloodstream. Among its many roles, this epithelial barrier is one of the most fundamental components of the innate immune system, protecting organisms from the vast environmental microbiota; indeed, >90% of infectious agents enter through mucosal epithelia. Although an effective defense mechanism against most microbes, pathogenic bacteria have evolved or acquired strategies to circumvent the mucosal barrier (). For example, some professional pathogens such as and inject into the host cell toxins that subvert host signal transduction pathways and manipulate the host cell cytoskeleton in ways that allow entry through the AP surface of the mucosal barrier (). In contrast, for opportunistic pathogens, of which is a prime example, the mucosal barrier represents a formidable challenge to bacteria-mediated damage or entry. However, in the setting of injured or poorly polarized epithelium, can initiate colonization and unleash its arsenal of potent virulence factors (). Indeed, this gram-negative pathogen is a leading cause of nosocomial infections in hospitalized patients and accounts for its predilection to cause ventilator-associated pneumonia, skin infections in burn patients or at the site of surgical incisions, and catheter-related infections, among others (). is also a cause of chronic lung infections and ultimately death in patients with cystic fibrosis (). Although usually considered an extracellular pathogen, ∼50% of all isolates can be measurably internalized into nonphagocytic cells in vivo as well as in vitro (). A seemingly simple but very important question is from what surface optimally enters the host cells. In tissue culture models, is observed to preferentially bind to and enter the cells at the edge and BL surfaces at the site of mechanical wounding, corresponding to injured and poorly polarized cells (). Consistent with this, we have found that several strains of enter more efficiently into incompletely polarized cells (; unpublished data). Phosphatidylinositol 3,4,5-trisphosphate (PIP3) has recently emerged as both a key determinant of epithelial polarity and of pathogen interaction with host cells (; ; ). In MDCK cells, a well-studied model of polarized epithelium, PIP3 is stably localized at the BL membrane and is excluded from the AP plasma membrane (; ). The mechanism by which a gradient of this freely diffusible lipid is maintained has not been fully elucidated, but it most likely involves localization of the lipid phosphatase PTEN (phosphatase and tensin homologue) to the tight junction (). Phosphatidylinositol 3-kinase (PI3K) induces scattering and tubulogenesis in epithelial cells through a novel pathway (). We have recently shown that PIP3 plays a key role in determining the composition and identity of the BL surface (). Insertion of exogenous PIP3 into the AP surface results in the rapid transformation of regions of the AP surface into a membrane with the composition of the BL surface by redirecting BL transcytosis. Conversely, reduction in the synthesis of PIP3 by the inhibition of PI3K causes a decrease in BL surface area. Together, these results suggest that PIP3 is necessary and sufficient for the specification of BL membrane. Interestingly, PIP3 is also involved in morphogenesis of the AP surface of photoreceptor cells in (). In a previous study, we have discovered that the PI3K pathway is necessary and sufficient for the internalization of into epithelial cells (). In the present study, we demonstrate that this pathogen subverts PI3K to alter membrane polarity. We show that in response to binding to the AP surface, many BL proteins as well as PI3K and actin are rapidly redistributed to the regions of bacterial binding at the AP surface without disrupting the integrity of the tight junctions. Concurrently, proteins that are normally resident at the AP surface are removed from the regions of bacterial attachment. We demonstrate that BL constituents of the protrusions originate from the BL membrane and are redirected to the AP membrane by a dynamin- and PI3K-dependent transport mechanism likely involving transcytosis. By transforming an AP surface into one with BL characteristics, the bacteria create a local microenvironment that facilitates colonization and entry into the mucosal barrier. In polarized MDCK cells, the pleckstrin homology (PH) domain of Akt fused to GFP (PH-Akt-GFP), a protein probe for PIP3, localizes exclusively at the BL surface in an LY294002 (LY)-dependent manner, whereas a point mutant unable to bind 3′ phosphoinositides is cytoplasmic (; unpublished data). We have previously shown that upon AP infection, PH-Akt-GFP relocalized to the site of AP bacterial binding (). This process as well as subsequent bacterial entry was blocked by the PI3K inhibitor LY, suggesting a key role for PI3K in these events (). These findings suggested that the bacteria induce the relocalization of lipids normally found only at the BL surface (; ). As the PH domain of Akt can also bind to phosphatidylinositol 3,4,-bisphosphate, we confirmed that PIP3 was recruited to the site of AP bacterial binding using a more specific probe, the PH domain of Grp1 fused to GFP (, ). Polarized confluent MDCK cells grown on transwell filters were transiently transfected with PH-Grp1-GFP and infected with (MOI of 50) for 30 min and fixed. Similar to our results with PH-Akt-GFP, this PIP3 probe showed a BL distribution and localized to the bacterial binding site (Fig. S1, available at ). We further investigated the spatial relationship of the bacteria to the cell surface by scanning electron microscopy. Extensive membrane protrusions from the AP surface were observed at the site of the attachment of bacterial aggregates (). Most of the bacterial binding occurred at or near cell–cell junctions (, right). As preferentially binds to BL surfaces when they are accessible, we considered the possibility that the bacteria were binding to BL surfaces of areas of multilayered cells. MDCK cells stably transfected with GFP-PH-Akt were cocultivated with PKH26 red–labeled for 30 min. In general, PIP3-containing protrusions were not associated with multilayered cells (). By 2 h, a time at which bacterial internalization is complete, protrusions were no longer visible (). We investigated whether PI3K, the enzyme that generates PIP3, is recruited to the AP protrusions. MDCK cell monolayers were infected with bacteria for 30 min, fixed, stained with an antibody to PI3K, and examined by confocal microscopy. In uninfected cells, cytoplasmic PI3K was preferentially found in the BL regions of the cells (unpublished data). Within 30 min after the addition of bacteria, PI3K accumulated at the AP surface underneath the site of bacterial attachment (). Membrane protrusion and generation of PIP3 are events often associated with actin rearrangement (), and entry is inhibited by cytochalasin D (). Phalloidin staining revealed filamentous actin in the bacteria-induced protrusions (). Protrusion formation was completely blocked by latrunculin B, indicating that their formation requires filamentous actin (unpublished data). We examined the membrane constituents of the bacteria-induced protrusions using antibodies to AP and BL proteins. Surprisingly, gp135/podocalyxin, a component of AP membranes, was absent from the PIP3-rich protrusions (). Confocal XZ sections revealed the acquisition of the BL markers p58, β-catenin (), and β1-integrin (not depicted). These remarkable changes in membrane composition are evident in the 3D reconstructions, which clearly show the bacteria aggregates associated with protrusions of the AP membrane that contain BL proteins and exclude AP proteins. Interestingly, the PIP3 probe GFP-PH-Akt and staining of the BL markers did not completely overlap, which may reflect the relative rapid degradation of PIP3 at the AP membrane by lipid phosphatases (). Together, these experiments reveal that binding of to the AP surface of the epithelial barrier is able to radically alter the composition of the membrane, transforming it from an AP surface to one with BL constituents. This transformation is accompanied by the recruitment of PI3K, the generation of PIP3 at the AP surface, and the recruitment of actin into membrane protrusions. AP proteins are lost from protrusions, which instead become enriched in BL proteins. These observations rule out the possibility that BL proteins are simply intermixed with AP proteins in these protrusions. Proteins normally destined for the BL surface could be redirected to the AP membrane from intracellular stores or from the BL membrane. To distinguish between these two possibilities, we used a pulse-chase experiment to investigate whether proteins from the BL surface were present in the AP protrusion. BL surface molecules were selectively biotinylated at 4°C by the addition of sulfo--hydroxysulfosuccinimide-biotin to the BL medium. This compound reacts with free amino groups on the surface but does not cross the tight junctions (). was added to the AP surface, and biotinylated BL molecules were detected by fluorescent streptavidin. As seen in (left), biotinylated BL proteins were detectable in the AP protrusions. In contrast, AP staining was not detected in adjacent or uninfected cells (unpublished data). This experiment suggests that BL constituents of the AP protrusions originate from the BL membrane either by disrupting tight junctions or by transcytosis. To test whether the integrity of the tight junction was affected, we assayed permeability to FITC-inulin, a low molecular weight molecule that is not able to diffuse through functional tight junctions (). In control experiments, the monolayer was exposed to 10 mM EDTA for 10 min, a treatment known to disrupt AP junctions. Diffusion of FITC-inulin from the AP to the BL compartment was observed within 20 min and continued to increase for up to 60 min. In contrast, AP addition of bacteria did not increase the diffusion of FITC-inulin compared with uninfected cells. Instead, less FITC-inulin diffusion was observed, suggesting that bacteria may increase the stability of tight junctions (). We confirmed these findings by testing whether in the absence of cell permeabilization the AP addition of antibodies to an extracellular domain of the BL protein β1-integrin was able to access to the BL membrane. As seen in , AP addition of the β1-integrin antibody only stained the AP surface at the site of the bacteria-associated protrusion. No BL staining was observed, confirming that did not disrupt the gate function of the adjacent tight junction. Together, these experiments strongly suggest that the –induced rerouting of BL markers does not occur by disruption of tight junctions. We tested whether a dominant-negative (DN) mutant of dynamin II, which has previously been shown to inhibit BL endocytosis in MDCK cells (), blocked the relocalization of BL proteins to the bacteria-induced AP protrusion. Infection of MDCK cells with an adenovirus construct expressing the DN dynamin II (K44A) under the control of a tetracycline-sensitive repressor prevented the accumulation of BL proteins at sites where bound to the AP surface (). Compared with nontransfected cells in the same sample ( = 13), relocalization of BL proteins to the site of AP binding was inhibited in cells expressing DN dynamin II by 56% ( = 14; P < 0.01 by chi-squared analysis). LY has been shown to inhibit BL to AP transcytosis of ricin or dimeric IgA (; ). Consistent with our observed role for transcytosis, LY also inhibited the accumulation of biotinylated BL proteins at the site of AP bacterial binding (). Compared with cells treated with DMSO alone ( = 12), LY prevented the relocalization of biotinylated BL proteins by 54% ( = 13; P < 0.01 by chi-squared analysis). We have recently shown that PIP3 is necessary and sufficient to specify BL membrane and that the exogenous addition of PIP3 transforms AP membrane in BL membrane by redirecting transcytosis (). We have also reported that activation of the PI3K and Akt pathways are necessary and sufficient for internalization (). The present study demonstrates that in the context of polarized MDCK cells, , an important human pathogen, subverts this pathway to create a local microenvironment that facilitates colonization and entry into the mucosal barrier. Similarly, binding of to the AP surface of the epithelial barrier is able to radically alter the composition of the membrane, transforming it from an AP surface to one with BL constituents. This transformation is accompanied by the recruitment of PI3K, the generation of PIP3, and the recruitment of actin into membrane protrusions. It is also possible that has additional ways to increase AP PIP3 such as by stimulating PIP3 diffusion and/or by decreasing the activity or levels of lipid phosphatases such as PTEN. In any case, by increasing AP PIP3, the bacteria appear to stimulate BL to AP transcytosis. For some molecules, Src kinase family–mediated phosphorylation inhibits BL recycling and allows cryptic AP sorting signals to predominate (). Interestingly, entry correlates with the activation of Src family kinases (). We propose that as vesicles fuse with the AP membrane and release their cargo, AP constituents are pushed aside. In addition to promoting bacterial pathogenesis, the rerouting of BL molecules may also benefit the host by enhancing the innate immune response. We found that the bacteria-induced protrusions were commonly found at cell–cell junctions. The close proximity to BL constituents at tight junctions may contribute to the rapid kinetics of the transformation of membrane polarity. This may also reflect the location of the host cell receptor. Although the cystic fibrosis membrane receptor and asialoganglioside GM1 have been suggested as receptors for entry (; ), in the context of MDCK cells, at least, we consider them to be unlikely candidates and are actively investigating other candidate receptors. We imagine that this subversion of host cell polarity is not absolute. Rather, in wounded or disrupted epithelium, the cells at the edge of the wound are poorly polarized, and is able to efficiently bind and coopt PI3K and generate PIP3 at the surface. As the mucosal barrier becomes increasingly differentiated and polarized, there is less binding (unpublished data), and the rerouting of BL markers to the AP surface is not as efficient, correlating with the decreased susceptibility of the intact epithelium to –mediated invasion or damage. MDCK cells (clone II, T23, or cells stably transfected with GFP-PH-Akt; ) were grown for 24 h on transwell filters as previously described (). For transient transfections, MDCK cells were transfected with LipofectAMINE 2000 (Invitrogen) according to the manufacturer's instructions. The mammalian expression vector pEGFP containing the PH domain of Grp1 was a gift from M. Birnbaum (University of Pennsylvania, Philadelphia, PA). pEGFP-AH, which contains the PH domain of Akt, and the mutant AHR25C (both constructs encode amino acids 1–147 of Akt), in which the lipid-binding site is mutated (), were gifts from J. Downward (London Research Institute, London, UK). Antibodies were obtained from the following sources: PI3K antibodies were purchased from Transduction Laboratories; HA antibody was purchased from Santa Cruz Biotechnology, Inc.; β1-integrin was obtained from the American Type Culture Collection; p58 antibody was a gift from K. Matlin (University of Cincinnati, Cincinnati, OH); anti–ZO-1 (R40.76) was a gift from B. Stevenson (University of Edmonton, Alberta, Canada); and gp135 antibody was a gift from G. Ojakian (State University of New York Downstate Medical Center, Brooklyn, NY). Secondary antibodies used were AlexaFluor594- or 647-conjugated antibodies obtained from Invitrogen. Actin filaments were stained with AlexaFluor350-phalloidin (Invitrogen). Streptavidin and AlexaFluor594 were obtained from Invitrogen. LY and latrunculin B were purchased from Calbiochem. Immunofluorescent staining was performed as previously described (). Samples were examined with a confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.) using a plan Apochromat 63× oil objective (Carl Zeiss MicroImaging, Inc.). Images were acquired using Meta 510 software (Carl Zeiss MicroImaging, Inc.), collected as TIFF files, and further analyzed with Photoshop 7.0 software (Adobe). The strain PAK (obtained from J. Mattick, University of Queensland, Brisbane, Australia) was routinely grown shaking overnight in Luria-Bertani broth at 37°C. These stationary phase bacteria were labeled with the Red Fluorescent Cell Linker Mini kit (PKH26; Sigma-Aldrich), added at an MOI of 250 to MDCK cells plated at a density of 7 × 10 cells on 12-mM transwells (Costar), and grown for 24 h. After 30 min of infection at 37°C, cells were washed three times with PBS, fixed as previously described (), and analyzed by confocal microscopy. Similar results were seen at lower MOIs, but fewer bacterial aggregates bound to cells were observed. Filter-grown confluent polarized MDCK cells grown as described in Cell preparation and culture were infected with with an MOI of 50. After 30 min of infection at 37°C, cells were washed three times with PBS and fixed with 2% cacodylate gluteraldehyde for 30 min and 2% aqueous osmium tetroxide for 30 min as primary and secondary fixatives, respectively. Samples were then rinsed with distilled water, dehydrated to 100% ethanol, and incubated with hexamethyldisilazone (Sigma-Aldrich) until dry. Dried samples were mounted on aluminum holders, sputter coated with gold palladium (20-nM coating), and examined with a scanning electron microscope (model 5410; JEOL). MDCK cells were grown as described in Cell preparation and culture, and BL membrane proteins were specifically biotinylated with sulfo--hydroxysulfosuccinimide-biotin at 4°C for 30 min as described previously (). The biotin solution was removed, and cells were washed with PBS and incubated at 37°C for 5 min, at which time bacteria were added to the AP surface for 30 min. Transcytosis of BL proteins to the AP surface was detected by adding streptavidin and AlexaFluor594 conjugate to the AP and BL media for 30 min at 37°C, and cells were analyzed by confocal microscopy as described in Immunofluorescence staining and image analysis. In some experiments, cells were preincubated with LY (100 μM in DMSO) for 1 h. LY was also present during biotinylation and infection. Construction of recombinant adenovirus and infection were performed as previously described (). MDCK T23 cells () were grown for 3 d on 12-mm transwells. 18 h before the assay, cells were washed five times with PBS lacking CaCl and infected for 2 h with 0.14 μl of virus stock per transwell (∼40 plaque-forming units/cell) in PBS lacking CaCl. The recombinant adenovirus carries a HA-tagged version of a DN dynamin II mutant (K44A). The monolayers were washed three times with MEM media supplemented with 5% FBS, further incubated in the same medium containing a low concentration (1 ng/ml) of doxycycline for 16 h, and titrated to allow sufficient expression of the DN dynamin II protein without obviously disrupting the integrity of the monolayer. MDCK cells were grown as described in Cell preparation and culture and infected with for 30 min or exposed to 10 mM EDTA for 10 min. After washing with PBS, 100 μg/ml of prewarmed PBS + FITC-inulin (Sigma-Aldrich) was added to the AP chamber, and prewarmed PBS was added to the BL chamber. Cells were incubated at 37°C, and 100-μl samples were collected from the BL chamber every 30 min. Fluorescence was quantified using a fluorescence plate reader (excitation/emission wavelength at 480/530 nm; 70% gain; Cytofluor). Chi-squared analysis was performed. P < 0.05 was considered statistically significant. Fig. S1 shows that –induced AP protrusions contain Grp1-PH-GFP, a probe for PIP3. Online supplemental material is available at .

The eukaryotic oligosaccharyltransferase (OST) transfers preassembled oligosaccharides onto asparagine residues as nascent polypeptides are translocated across the rough ER membrane (for review see ). The consensus site for N-linked glycosylation in eukaryotic organisms is conserved and corresponds to the simple tripeptide sequence N-X-T/S, where X can be any residue except proline (). The oligosaccharide donor assembled by most eukaryotes for N-linked glycosylation is the dolichol pyrophosphate–linked oligosaccharide GlcManGlcNAc-PP-Dol (abbreviated here as GMGN-PP-Dol). Synthesis of the dolichol-linked oligosaccharide (OS-PP-Dol) donor occurs by the stepwise addition of monosaccharide residues onto the dolichol-pyrophosphate carrier by a family of ER-localized membrane bound glycosyltransferases (asparagine-linked glycosylation [ALG] gene products; for review see ). ManGlcNAc-PP-Dol (MGN-PP-Dol) is assembled on the cytoplasmic face of the ER membrane, with UDP-GlcNAc and GDP-Man serving as monosaccharide donors. Man-P-Dol and Glc-P-Dol are the donors for the luminally oriented glycosyltransferases that add four mannose and three glucose residues to OS-PP-Dol assembly intermediates within the ER lumen. Depletion of the yeast Rft1 protein causes severe hypoglycosylation of proteins and accumulation of ManGlcNAc-PP-Dol () even though Alg3p, not Rft1p, is the mannosyltransferase that adds the sixth mannose residue. Rft1p has been proposed to flip cytosolically oriented MGN-PP-Dol across the ER membrane (). Certain kinetoplastids ( and ) and the ciliate assemble OS-PP-Dol compounds that lack the glucose residues (MGN-PP-Dol by ) and/or the mannose residues (GMGN-PP-Dol by and MGN-PP-Dol by ) that are transferred by the luminally oriented ALG gene products (; ). Searches of fully sequenced genomes using yeast ALG proteins as query sequences has revealed considerable diversity in OS-PP-Dol biosynthesis amongst unicellular organisms (). Biochemical studies have confirmed bioinformatic predictions that synthesizes GN-PP-Dol, and synthesize MGN-PP-Dol, and the pathogenic fungi synthesizes MGN-PP-Dol (; ). The diversity of eukaryotic OS- PP-Dol donors was proposed to have occurred by secondary loss of ALG genes during the evolution of current eukaryotes from a last common ancestor with a complete ALG pathway (). In fungi and vertebrate organisms, the OST is an oligomer composed of seven to eight nonidentical subunits (for review see ). Of the eight OST subunits (Stt3p, Ost1p, Ost2p, Ost3p or Ost6p, Ost4p, Ost5p, Wbp1p, and Swp1p), five are encoded by essential yeast genes (, , , , and ). With the exception of STT3, which contains the enzyme active site (; ; ), relatively little is known about the roles of the essential or nonessential subunits (for review see ). Vertebrate, plant, and insect genomes encode two forms of the catalytic subunit that are designated as STT3A and -B (; ). The canine STT3 homologues are assembled with a shared set of noncatalytic subunits (ribophorin I [Ost1 homologue], ribophorin II [Swp1], OST48 [Wbp1], DAD1 [Ost2], and TUSC3 or IAP [Ost3 or -6] and OST4) to generate OST isoforms with kinetically distinct properties (). Protein and DNA sequence database searches of fully sequenced eukaryotic genomes using the yeast and human OST subunits as query sequences suggest that the OST in protist organisms has a simpler subunit composition (; ). The genomes of and the kinetoplastids and encode several different STT3 proteins (), yet lack genes encoding the noncatalytic subunits. Four-subunit complexes, consisting of STT3, OST1, OST2, and WBP1, are predicted for and . A six-subunit complex (STT3, OST1, OST2, OST3, OST4, and WBP1) is predicted for . The genome encodes readily identifiable homologues of all OST subunits with the exception of Ost5p (). The absence of glucose residues on OS-PP-Dol compounds assembled by most protists and is of particular interest because the terminal glucose residue on GMGN-PP-Dol is a critical substrate recognition determinant for the OST. OS-PP-Dol assembly intermediates that lack the terminal glucose residue are transferred less rapidly by the vertebrate and yeast OST (; ; ; ; ), thereby minimizing synthesis of glycoproteins with aberrant oligosaccharide structures. Glycosylation of proteins with an oligosaccharide assembly intermediate could interfere with glycoprotein quality-control pathways in the ER as well as subsequent oligosaccharide-processing reactions in the Golgi complex (for review see ). Cellular defects in GMGN-PP-Dol biosynthesis cause a family of inherited diseases (congenital disorders of glycosylation [CDG-I]) due to hypoglycosylation of nascent glycoproteins by the OST in cells that accumulate an assembly intermediate or are unable to maintain a normal concentration of fully assembled GMGN-PP-Dol (for review see ). Preferential utilization of GMGN-PP-Dol by the yeast and vertebrate OST occurs by allosteric interactions between a regulatory OS-PP-Dol binding site and the active site subunit, as well as by oligosaccharide structure–mediated alterations in tripeptide acceptor binding affinity (; ). Kinetic analysis of the purified canine OST isoforms has suggested that the regulatory OS-PP-Dol binding site is not located on STT3A or -B, but is instead associated with one or more of the noncatalytic subunits (). Does the OST from organisms that synthesize nonglucosylated OS-PP-Dols transfer the in vivo donor in preference to OS-PP-Dol assembly intermediates or GMGN-PP-Dol? Previous studies indicate that the OST transfers glucosylated (GMGN-PP-Dol) and large nonglucosylated (MGN-PP-Dol) donors at similar rates in vitro (), suggesting that the OST is nonselective. Can biochemical analysis of the OST from primitive eukaryotes reveal properties of the higher eukaryotic OST that arose as additional subunits were added to the STT3 catalytic core? Here, we report a comparison of the OST from , , , , and , with emphasis placed upon an analysis of donor substrate selection. Our results support the hypothesis that terminal mannose residues on the OS-PP-Dol are important for donor substrate recognition by the OST in organisms that assemble nonglucosylated OS-PP-Dol compounds. Cooperative OS-PP-Dol binding, a feature of the yeast and canine OST complex that facilitates exquisite GMGN-PP-Dol selection, is not a property of the predicted one- and four-subunit protist OST complexes. Is preferential utilization of the in vivo oligosaccharide donor an OST property that is restricted to eukaryotes that assemble triglucosylated OS-PP-Dols? To address this question, the OST from selected protists and was assayed using a synthetic tripeptide acceptor and a heterogeneous bovine OS-PP-Dol library that consists of donors that range in size between MGN-PP-Dol and GMGN-PP-Dol. Enzyme concentrations were adjusted to ensure that a maximum of 3% of the total donor substrate was converted into glycopeptides. Radiolabeled glycopeptide products that were captured with an immobilized lectin (ConA Sepharose) were subsequently eluted and resolved by high-pressure liquid chromatography (HPLC) according to the number of saccharide residues (). As expected, GMGN-NYT was the most abundant product when the purified OST was assayed (). In contrast, GMGN-NYT was less abundant in the glycopeptide products () and barely detectable in glycopeptide products derived from assays of the (), (), or (not depicted) OST. The composition of the OS-PP-Dol donor library was determined as described previously () by incubating an excess of the purified yeast OST with a low quantity of the donor substrate (OST endpoint assay; Fig. S1, available at ). A normalized initial transfer rate (glycosylated tripeptide [OS-NYT]/ OS-PP-Dol; ) was calculated for the eight most abundant donors by dividing the glycopeptide product composition by the composition of the OS-PP-Dol donor substrate library. A normalized initial transfer rate of 1 (, dashed lines) indicates nonselective utilization of a donor substrate relative to the total donor pool. The and OST transfer the mannosylated donors (MGN-PP-Dol) threefold more rapidly than GMGN-PP-Dol (). Although the OST utilizes compounds ranging in size between MGN-PP-Dol and GMGN-PP-Dol at rates similar to those reported previously (), OS-PP-Dol donors with fewer mannose residues (MGN-PP-Dol) were transferred less rapidly (). The OST showed preferential utilization of the in vivo donor (MGN-PP-Dol) relative to assembly intermediates (MGN-PP-Dol) and the glucosylated donor (). Careful inspection of the glycopeptide elution profiles () revealed that several of the smaller glycopeptide peaks (e.g., MGN-NYT) have prominent shoulders, suggesting oligosaccharide structural heterogeneity. The structural diversity of the OS-PP-Dol donor substrate library is thought to arise by exposure of GMGN-PP-Dol to cellular glucosidases and mannosidases during isolation (). Mannosidase degradation of MGN-PP-Dol (, compound a) could yield seven MGN-PP-Dol isomers (, compounds c–i) that differ from biosynthetic MGN-PP-Dol (, compound b). Biosynthetic MGN-PP-Dol (, compound b) can be readily distinguished from these other isomers by digestion with α-1,2 mannosidase, as it is the only isomer that has two α-1,2–linked mannose residues (, red circles). MGN glycopeptides produced in an OST endpoint assay were purified by preparative HPLC (, left) and digested to completion with α-1,2 mannosidase. Resolution of the digestion products by HPLC (, right) yielded three peaks (M3, M4, and M5) that are derived from MGN-PP-NYT isomers that contain 2, 1, or 0 α-1,2–linked mannose residues. Quantification (, black bars) of two independent experiments revealed that 22% of the MGN glycopeptides were derived from biosynthetic MGN-PP-Dol (, compound b), 63% from compounds c–h, and 15% from compound i. If the OST from , , , or selects biosynthetic MGN-PP-Dol (, compound b) in preference to other MGN-PP-Dol isomers, the MGN glycopeptides synthesized in the presence of excess donor substrate should be enriched in glycopeptides that contain two α-1,2–linked mannose residues. To ensure that our glycopeptide product analysis provided a reliable measure of the relative initial transfer rate, the OST assays were terminated when <10% of the total MGN-PP-Dol was converted into glycopeptides. Typical HPLC profiles of the α-1,2 mannosidase digestion products of the MGN glycopeptides are shown in , and the results from assays of all four organisms are quantified in . We observed a very similar distribution of MGN isomers for the donor substrate pool and the initial glycopeptide products (, compare black and white bars), thereby indicating that the OST does not discriminate between MGN-PP-Dol isomers. The MGN glycopeptides synthesized by the and OST also resembled the MGN-PP-Dol donor pool; hence, the OST from these organisms does not select biosynthetic MGN-PP-Dol in preference to other MGN-PP-Dol isomers (). MGN glycopeptides synthesized by the OST were twofold deficient in biosynthetic MGN-NYT and enriched in one or more MGN-NYT isomers that have one α-1,2–linked mannose residue (). This result, taken together with a reduced transfer rate for MGN-PP-Dol relative to MGN-PP-Dol () by the OST suggests that a terminal α-1,2–linked mannose residue on the B or C antennae of MGN-PP-Dol serves as a positive determinant for substrate selection by the OST. Donor substrate competition experiments were conducted using purified biosynthetic MGN-PP-Dol (, isomer b), MGN-PP-Dol (, compound a), and GMGN-PP-Dol. The OST will synthesize GMGN-NYT when GMGN-PP-Dol is the sole donor substrate (, profile a). The absence of the MGN-NYT product indicates that the endogenous donor substrate is not abundant in the assay mix relative to the exogenous donor substrate. Analogous results were obtained using detergent extracts prepared from , , and (unpublished data). When the MGN-PP-Dol/ GMGN-PP-Dol ratio is 2.5:1, the yeast OST primarily synthesized GMGN-NYT, unlike the OST that synthesized MGN-NYT (, profiles b and c). Quantification of this competition experiment, as well as additional assays containing 1.5 μM MGN-PP-Dol and variable concentrations of GMGN-PP-Dol, showed that donor substrate selection by the (, squares) and (, circles) OST occurs across a wide range of donor substrate ratios (). Additional donor substrate competition experiments were conducted using 1:1 mixtures of the three purified oligosaccharide donors (). The OST selects GMGN-PP-Dol in preference to both nonglucosylated donors () but does not discriminate between MGN-PP-Dol and MGN-PP-Dol (). The OST from and selects both nonglucosylated donors in preference to GMGN-PP-Dol () but does not discriminate between MGN-PP-Dol and MGN-PP-Dol (). The OST does not discriminate between MGN-PP-Dol and GMGN-PP-Dol (), but both donors are selected in preference to MGN-PP-Dol (). In contrast, both the glucosylated donor and biosynthetic MGN-PP-Dol are nonoptimal donors for the OST relative to the in vivo donor (). The observation that GMGN-PP-Dol but not MGN-PP-Dol is a nonoptimal donor for the and OST suggests that the A antennae of MGN-PP-Dol may be recognized by the OST in these organisms. To test this hypothesis, an additional competition experiment was performed using a mixture of purified MGN-PP-Dol and an enriched sample of GMGN-PP-Dol. The GMGN-PP-Dol preparation contains GMGN-PP-Dol as a minor component. Glycopeptide products synthesized by the and OST were resolved by HPLC (, top). The initial transfer rates of ∼1 for the OST serves as an important control for the observed lower transfer rates of GMGN-PP-Dol and GMGN-PP-Dol relative to MGN-PP-Dol by the OST. Each additional glucose residue on the A branch of the oligosaccharide reduces the normalized initial transfer rate by the OST. Enzyme kinetic experiments suggest that selection of the fully assembled OS-PP-Dol by the yeast or mammalian OST occurs by allosteric communication between a regulatory OS-PP-Dol binding site and the donor substrate binding site on STT3, in addition to oligosaccharide structure–dependent alterations in tripeptide substrate binding affinity (; ). Nonlinear Lineweaver-Burk plots for the OS-PP-Dol substrate are diagnostic of the cooperative OS-PP-Dol binding kinetics of the yeast and mammalian OST (). Donor substrate saturation experiments for the (), (), and enzymes () were conducted using a constant concentration of tripeptide acceptor and increasing concentrations of purified OS-PP-Dols. The linear Lineweaver-Burk plots yielded K values in the submicromolar range for the in vivo donor substrate. The experimental data for the OST was replotted as an Eadie-Hofstee plot (). The linear Eadie-Hofstee plot for the OST is inconsistent with cooperative OS-PP-Dol binding kinetics. In contrast, the OST binds the same donor substrate (MGN-PP-Dol) in a cooperative manner, as revealed by a nonlinear Eadie- Hofstee plot (, inset). Additional donor substrate saturation experiments using the nonoptimal donors (MGN-PP-Dol for OST and GMGN-PP-Dol for OST) did not reveal differences in the apparent K that could account for the lower transfer rates of the nonoptimal donor substrate (unpublished data). Donor substrate selection by the protist OST does not involve a regulatory OS-PP-Dol binding site, nor is it explained by a reduced affinity for the nonoptimal oligosaccharide donor. Reduced transfer rates for nonoptimal donors by the yeast and mammalian OST is in part explained by a reduced binding affinity for the tripeptide acceptor in the presence of an OS-PP-Dol assembly intermediate (; ; ; ). The () and () OST were assayed in the presence of a constant concentration of the optimal and nonoptimal oligosaccharide donors and increasing concentrations of the tripeptide acceptor. The linear Lineweaver-Burk plots for the tripeptide acceptors were indicative of a single acceptor tripeptide binding site, as observed for the yeast and mammalian OST (; ). The nonoptimal donor substrate (GMGN-PP-Dol for and MGN-PP-Dol for ) reduces the binding affinity of the OST for the tripeptide acceptor. In both cases, the threefold decrease in acceptor tripeptide binding affinity is responsible for the reduction in the normalized transfer rate when the acceptor tripeptide is not saturating. The apparent V is not influenced by the structure of the OS-PP-Dol donor, as revealed by a shared I/V intercept, when the oligosaccharide donors are present in fourfold excess relative to the apparent K for the donor substrate (). GN-PP-Dol is the smallest oligosaccharide donor that is an effective substrate for the yeast OST (; ; ). The 2′ -acetyl modification on the first saccharide is critical for catalysis, whereas the 2′ -acetyl modification on the second residue is important for substrate recognition (). Efficient -glycosylation by the OST from higher eukaryotes is also dependent on the terminal glucose residue on the A antennae of the oligosaccharide (; ). As the OS-PP-Dol donors synthesized by many protists and the fungi lack glucose residues, one might predict that the OST from these organisms would only recognize the GlcNAc core of the donor substrate. However, donor substrate competition experiments demonstrate that the in vivo oligosaccharide donor for , , , and is a preferred substrate relative to certain larger and/or smaller OS-PP-Dol compounds. To our knowledge, this is the first evidence that oligosaccharide donor substrate selection is not restricted to organisms that synthesize the triglucosylated oligosaccharide donor. In all four cases, preferential utilization of the in vivo donor is less stringent than that observed for the or mammalian OST both in terms of the size range of compounds that are optimal in vitro substrates and the fold selection of the in vivo donor substrate relative to nonoptimal donors. The predicted one-subunit OST from utilizes larger OS-PP-Dol compounds, including the in vivo donor MGN-PP-Dol in preference to MGN-PP-Dol. The latter compound is one of four lumenal OS-PP-Dol assembly intermediates that could compete in vivo with MGN-PP-Dol as a donor substrate. The observed two- to threefold more rapid in vitro transfer of MGN-PP-Dol than MGN-PP-Dol appears to be sufficient to ensure that small assembly intermediates are rarely used in vivo, in part because MGN-PP-Dol is more abundant in the ER than the luminally oriented (MGN-PP-Dol) assembly intermediates (). The presence of a terminal α-1,2–linked mannose residue on the B or C antennae appears to be important for preferential utilization of MGN-PP-Dol by the OST, as revealed by the relative transfer rates of MGN-PP-Dol isomer classes () and by the reduced utilization of MGN-PP-Dol relative to MGN-PP-Dol. In vivo transfer of an assembly intermediate may be deleterious, as protein-linked high-mannose oligosaccharides that lack the terminal mannose residue on the B antennae (M8B isomer) or C antennae (M8C isomer) are less efficiently glucosylated by the UDP-glucose glycoprotein glucosyltransferase (UGGT; ). UGGT, which was first detected in , serves as the folding sensor for the glycoprotein quality-control pathway in the ER (). The predicted four-subunit OSTs from and () transfer the in vivo donor (MGN-PP-Dol) at the same rate as other OS-PP-Dol compounds that lack glucose residues (MGN-PP-Dol), including MGN-PP-Dol isomers that lack one or more mannose residues on the A antennae. Because synthesis of the MGN-PP-Dol donor is completed on the cytoplasmic face of the rough ER, the and OST do not need to discriminate between luminally oriented MGN-PP-Dol and cytoplasmically oriented OS-PP-Dol assembly intermediates. Consequently, MGN-NYT is the major glycopeptide synthesized in vitro when an acceptor tripeptide is incubated with intact or membranes () despite the lack of a mechanism to discriminate against underassembled oligosaccharide donors. We propose that the STT3 active-site subunit of the OST has evolved to have a catalytic site that is optimal for the in vivo oligosaccharide. For , , and , the proposed loss of genes that encode the ALG glucosyltransferases (ALG6, -8, and -10; ) has apparently been accompanied by compensatory alterations in the STT3 structure that are optimal for an oligosaccharide donor with an A antennae that lacks all three glucose residues. The predicted seven-subunit OST transfers the larger mannosylated OS-PP-Dol donors (MGN-PP-Dol) more rapidly than smaller assembly intermediates or GMGN-PP-Dol. Utilization of the fully assembled in vivo donor in preference to luminally exposed OS-PP-Dol assembly intermediates may be a shared property of the OST in organisms that synthesize donors larger than MGN-PP-Dol. The relatively modest (∼1.5-fold) preference for MGN-PP-Dol relative to biosynthetic MGN-PP-Dol leads to selective synthesis of MGN-NYT when the acceptor tripeptide is incubated with intact membranes (). Kinetic analysis of the and OST revealed that oligosaccharide structure–mediated modulation of acceptor substrate binding affinity is a conserved property of the eukaryotic OST that can be ascribed to the STT3 active site. The threefold reduction in acceptor substrate binding affinity readily accounts for the reduced transfer of nonoptimal donors when the acceptor tripeptide is present at subsaturating levels. Future studies will address the order of substrate binding to the one- and four-subunit OSTs that are predicted for and . One objective of these experiments will be to determine whether the subunit composition of protist complexes matches the bioinformatic predictions. Cooperative OS-PP-Dol binding by the OST is not explained by dimerization of heterooctamers, as coimmunoprecipitation experiments using yeast strains that express STT3-HA and STT3-HisFLAG from chromosomal loci did not reveal higher order OST oligomers (). Potential explanations for the discrepancy between a recent report describing dimeric assembly of the yeast OST complex () and our previous conclusions are being explored. Cooperative OS-PP-Dol binding is not explained by separate but interacting binding sites for the chitobiose core of GMGN-PP-Dol and the terminal glucose residue, because cooperative binding by the yeast or canine OST is not dependent on the presence of glucose residues on the oligosaccharide donor, as confirmed here using MGN-PP-Dol as a donor substrate. Instead, our results indicate that cooperative donor substrate binding is diagnostic of a regulatory OS-PP-Dol binding site that is primarily responsible for the highly selective utilization of the GMGN-PP-Dol donor (; ). Based on a kinetic analysis of canine OST isoforms, we proposed that the regulatory OS-PP-Dol binding site is not located on the catalytic subunit (STT3A or -B), but is instead provided by one or more of the shared noncatalytic subunits. Support for this hypothesis has now been provided by recent experiments showing that a STT3 can assemble with the noncatalytic yeast OST subunits and, upon doing so, mediate selective utilization of GMGN-PP-Dol as the donor substrate both in vitro and in vivo (). One objective of this study was to determine whether protist OSTs use a regulatory OS-PP-Dol binding site to select the in vivo oligosaccharide donor. Unlike the and OST, the predicted one-subunit OST from (STT3) and the predicted four-subunit OSTs from and (STT3-OST1-OST2-WBP1) do not bind OS-PP-Dol in a cooperative manner; hence, the OST from these organisms lacks the regulatory OS-PP-Dol binding site. The simplest interpretation of this observation is that the regulatory OS-PP-Dol binding arose as additional subunits were acquired during evolution of the eukaryotic OST. The IAP and TUSC3 (N33) proteins dissociate from the canine OST during purification, so these OST3/OST6 family members are not candidates for the regulatory OS-PP-Dol binding site. OST4 and -5 can be discounted based on structural considerations because neither of these polypeptides has more than a few residues exposed to the lumen of the ER (). Therefore, cooperative OS-PP-Dol binding by the yeast or vertebrate OST correlates with the presence of a Swp1p/ribophorin II subunit in the OST complex. Extensive biochemical and genetic evidence supports direct interactions between Wbp1, Swp1p, and Ost2p (; ), as well as between their respective mammalian homologues, OST48, ribophorin II, and DAD1 (; ). We hypothesize that the regulatory OS-PP-Dol binding site is located on the Swp1p–Wbp1p–Ost2p subcomplex. Interestingly, OS-PP-Dol protects a critical cysteine residue in Wbp1p from modification by a cysteine-directed protein modification reagent (). A role for the Swp1p–Wbp1p–Ost2p subcomplex as the regulatory OS-PP-Dol binding site might help explain why expression of each of these subunits is essential for viability of (, ; ). With the exception of , there is a strong correlation between organisms that assemble a glucosylated oligosaccharide donor (either GMGN-PP-Dol or GMGN-PP-Dol) and organisms that express or are predicted to express a Swp1p/ribophorin II homologue (; ). Trophozoites of strain HM1:IMSS were grown axenically (in the absence of bacteria or other cells) in TYI medium supplemented with 10% heat-inactivated adult bovine serum at 37°C. Axenic cultures of strain G3 were maintained in TYM medium supplemented with 10% heat-inactivated horse serum at 37°C. Axenic cultures of epimastigotes (strain Y) were grown in the LIT medium supplemented with hemin and 10% heat-inactivated fetal calf serum at 25°C. strain B3501, maintained on YPD plates, was grown in YPD broth for 20 h at 30°C. Whole cells were collected by centrifugation and resuspended in 10 mM Hepes, pH 7.4, 25 mM NaCl, 10 mM MgCl, and 1× protease inhibitor cocktail (PIC; as defined by ). , , or cells were homogenized using 50 strokes of a Teflon-glass homogenizer. The cell suspension was mixed with an equal volume of glass beads and vortexed extensively (200 5-s bursts). Total membrane fractions were collected by a 30-min centrifugation of the cell homogenate at 267,000 using a rotor (TLA 100.4; Beckman Coulter). The membrane pellets were solubilized in 1.5% digitonin, 20 mM Tris-Cl, pH 7.5, 500 mM NaCl, 1 mM MgCl, 1 mM MnCl, 1 mM DTT, and 1× PIC at a membrane concentration of 2 eq/μl (1 eq/μl = 50 A in 1% SDS). The detergent extracts were clarified by a 5-min centrifugation at 66,600 using the rotor. The OST was purified from an epitope-tagged (6xHisFLAG-OST1) yeast strain as described previously (). Detergent extracts of the , , , and membranes were diluted fourfold with 20 mM Tris-Cl, pH 7.4, 1 mM MgCl, 1 mM MnCl, 1 mM DTT, and 1× PIC. 5-μl aliquots of the 4×-diluted soluble extracts were assayed for OST activity in a total volume of 100 μl as described previously (), using Nα-Ac-Asn-[I]Tyr-Thr-NH as the acceptor substrate and either structurally homogeneous OS-PP-Dol compounds or a previously described heterogeneous bovine pancreas OS-PP-Dol pool () as the donor substrate. OST assays were supplemented with 1.4 mM deoxynojirimycin, 1.4 mM mannojirimycin, and 1.4 mM swainsonine to inhibit glucosidases and mannosidases. Glycopeptide products from OST assays were isolated with ConA Sepharose and quantified by gamma counting. Structurally homogeneous GMGN-PP-Dol, MGN-PP-Dol, MGN-PP-Dol, and an enriched GMGN-PP-Dol preparation were purified as described previously () from porcine pancreas (GMGN-PP-Dol and GMGN-PP-Dol), an Δ yeast strain (MGN-PP-Dol), or an Δ yeast strain (MGN-PP-Dol). The concentration and composition of OS-PP-Dol samples was determined from the yield and oligosaccharide distribution of radiolabeled glycopeptides obtained in the OST endpoint assay (). In brief, 12–15 pmol of OS-PP-Dol was incubated with 60 fmol of purified yeast OST for 24–48 h under OST assay conditions to quantitatively convert the donor substrate into glycopeptides. The previously isolated heterogeneous bovine pancreas OS-PP-Dol pool consists of a mixture of biosynthetic OS-PP-Dol assembly intermediates and OS-PP-Dol degradation products that were produced by exposure of the OS-PP-Dol to endogenous mannosidases and glycosidases during isolation (). As shown in Fig. S1 B, the bovine OS-PP-Dol pool has the following oligosaccharide composition: 4.7% MGN, 14% MGN, 19% MGN, 23% MGN,17% MGN, 11% MGN, 3.7% MGN, 1% GMGN, 1.5% GMGN, and 5.1% GMGN. Because of low abundance in the donor pool, initial transfer rates are not reported for GMGN-PP-Dol and GMGN-PP-Dol. Assays designed to analyze the donor substrate preference of the OST using the bovine OS-PP-Dol library contained 1.2 μM OS-PP-Dol and were terminated before 3% of the total donor was consumed. Donor substrate competition experiments using purified donors were terminated before 10% of the substrate was consumed. Glycopeptide products from the competition experiments were eluted from the ConA beads and resolved according to oligosaccharide size by HPLC as described previously (; ), except that the HPLC buffer A was acetonitrile/water/acetic acid/triethlyamine (73.6:23:2.4:1), whereas HPLC buffer B was water/acetic acid/triethlyamine (91:3:6). Glycopeptides prepared using the yeast OST and purified OS-PP-Dol compounds served as HPLC elution standards. OST assays to prepare MGN glycopeptides using the heterogeneous OS-PP-Dol library were designed to ensure that <10% of the total MGN-PP-Dol was converted to MGN-NYT. HPLC fractions corresponding to MGN-NYT were dried and resuspended in 50 μl of 1× reaction buffer supplied by the manufacturer (Prozyme) and incubated for 18 h at 37°C with 0.33 mU α-1,2 mannosidase. The Savant-dried glycopeptide digestion products were dissolved in 500 μl HPLC buffer A and resolved by HPLC as described in the preceeding paragraph. The kinetic parameters for the tripeptide acceptor and oligosaccharide donor for the , , and enzymes were determined by a nonlinear least-squares fit of the kinetic data to the Michaelis-Menten equation and by linear least-squares fits of Lineweaver-Burk plots or Eadie-Hofstee plots. The kinetic parameters for the dolichol-oligosaccharide donor for the enzyme were obtained using a nonlinear least-squares fit of the kinetic data to equations for a substrate activated enzyme as described previously (). Kaleidagraph 3.5 (Synergy Software) was used for curve fitting. Fig. S1 shows the oligosaccharide composition analysis of the OS-PP-Dol library used for the experiments in and Online supplemental material is available at .

The movement of molecules between the cytoplasm and the nucleoplasm occurs through nuclear pore complexes (NPCs), which form channels across an impermeable nuclear envelope (NE) membrane. The NPCs are elaborate, symmetrical structures consisting of repetitive subunits composed of proteins termed nucleoporins, or nups. Passage of most macromolecules through NPCs involves the recognition of a nuclear transport signal by soluble transport receptors called karyopherins, or kaps, which use their affinity for nups to traverse the NPC with their cargos in tow (for reviews see ; ). There are 14 members of the karyopherin family in yeast. Importins move cargos into the nucleus, whereas exportins move them out, and each kap binds a different, but sometimes overlapping, set of cargos. This redundancy may explain why only five kaps are essential for viability in yeast. Three of these are the importins: Kap121p and the heterodimeric complex of Kap95p–Kap60p. The latter complex binds to what is referred to as a classical NLS (cNLS), which has been identified in a large number of cargo molecules. The number of known cargos imported by Kap121p is more limited but of significant interest, as this kap is required for normal progression through mitosis. Moreover, Kap121p-mediated transport is specifically regulated during mitosis (). In addition to their function in cargo import, both Kap121p and the Kap95p–Kap60p complex control the association of specific proteins with the NPC (; ; ). One example is the yeast protein Ulp1p. Both Kap121p and the Kap95p–Kap60p complex bind to Ulp1p and concentrate this protein at the NPC (). However, the functional significance of the association of Ulp1p with the NPC remains unclear. Ulp1p is one of two isopeptidases in yeast that specifically target the small ubiquitin-like modifier, SUMO. One of the functions of Ulp1p is to cleave the SUMO precursor (encoded by the gene) to its mature form. Processed SUMO can then be covalently linked to its target proteins in a sequential process that involves an E1 activating enzyme, an E2 conjugating enzyme, and one of several E3 ligases (for review see ). Importantly, sumoylation is a reversible process, as Ulp1p and another isopeptidase, Ulp2p, are capable of desumoylating proteins (, ). The regulation of the sumoylation state of a protein can profoundly affect its function, often by dictating its localization, interactions, and stability. The number of proteins identified as being modified by the addition of SUMO is rapidly growing (; ; Y. ). Many of the targets are nuclear proteins involved in DNA replication and repair, chromatin remodeling, and transcriptional control (; ). However, members of a group of proteins that form the septin ring at the bud neck of budding yeast are also major sumoylation targets (). The septin ring appears before bud formation in G1 phase of the cell cycle. As the bud grows, the ring extends through the bud neck, forming an hourglass-shaped structure that appears as a double ring, which later divides during cytokinesis. The ring persists in the mother and daughter until disassembling during G1 (for reviews see ; ). At least three septins, Cdc3p, Cdc11p, and Shs1p, become sumoylated specifically during mitosis; the modification occurs before anaphase and is removed at cytokinesis (; ). The function of this modification has not been studied to the point that it is fully understood. One suggestion is that it plays a role in the disassembly of the septin ring (). The sumoylation of the septins during M phase requires the E3 ligase Siz1p (; ). Siz1p is located in the nucleus during interphase. At a point before anaphase, and coincident with septin sumoylation, Siz1p is phosphorylated. This is accompanied by an egress of Siz1p from the nucleus and its accumulation at the septin ring (), where a visible accumulation of this protein remains until again concentrating in the nucleus after cytokinesis. These changes in the localization of Siz1p suggest an active role for the nuclear transport machinery in controlling septin sumoylation by dictating the movement of Siz1p. Septin desumoylation occurs at cytokinesis () and requires Ulp1p (; see Results); however, how its activity is regulated is less clear. Ulp1p is localized to the nucleoplasmic face of the NPC (X. ) but presumably must gain access to the cytoplasmic septins to promote desumoylation during cytokinesis. Interestingly, the regions of Ulp1p that function as kap binding sites () are also required for normal sumoylation of numerous targets (), suggesting that, like Siz1p, kaps may regulate the activity of Ulp1p through controlling its localization in the cell. To further understand the role of the nuclear transport machinery in cell cycle progression, we have focused on the role of kap-mediated movement of Siz1p and Ulp1p in controlling the cell cycle oscillation of septin sumoylation. The kaps Kap95p, Kap121p, and Kap142p/Msn5p play specific roles in regulating septin sumoylation. Kap95p imports Siz1p into the nucleus, and Kap142p/Msn5p exports Siz1p to the cytoplasm at M phase. Ulp1p localization and activity are regulated by Kap121p and the Kap95p–Kap60p complex. We propose that Ulp1p is transiently released from the NPC during mitosis, facilitating the desumoylation of the septins during cytokinesis. This event is controlled by the association of Ulp1p with Kap121p and changes in the interactions of this kap with the NPC. Defects in any of these kap interactions lead to abnormalities in the cycle of septin sumoylation and desumoylation. Siz1p is concentrated in the nucleus throughout most of the cell cycle until mitosis, when nuclear amounts decrease and the protein accumulates at the septin ring. Here it resides until reentering the nucleus after mitosis (; ; , wild type [WT]). We investigated the nuclear import and export pathways that control Siz1p localization. The nuclear localization of Siz1-GFP, a chimeric protein that functions similar to WT Siz1p (), was examined in strains containing temperature-sensitive mutations in either of two essential kaps, Kap121p () and Kap95p (). In each mutant, transport of cargos recognized by each kap is strongly inhibited at the nonpermissive temperature for growth (37°C; ; ). We observed that the localization pattern of Siz1-GFP was unaltered in cells at both 23 and 37°C (). In contrast, the nuclear concentration of Siz1-GFP was not observed in the mutant at either temperature and was only restored after introducing a plasmid-born WT copy of . In the mutant, Siz1-GFP was diffusely distributed throughout the cell with some generally small-budded cells exhibiting a nuclear rim signal. The significance of this localization is unclear. The association of Siz1-GFP with the bud neck in large-budded cells, in either asynchronous or nocodazole-arrested cultures, was not affected by the mutation (). We conclude from these results that Kap95p is required for the import of Siz1p into the nucleus but not for its targeting to the septin ring. The nuclear pool of Siz1p has been suggested to be the source of Siz1p that is recruited to the septin ring during M phase (). However, as shown with the mutant, the nuclear localization of Siz1p is not a prerequisite for its septin ring association (). To further define the dynamics of Siz1p recruitment to the septin ring and the role of nuclear export machinery, we examined whether the export kap Kap142p/Msn5p was required for this event. Kap142p/Msn5p was targeted for analysis, as it has been shown to export cargos whose transport is induced by phosphorylation (; ; ), and phosphorylation of Siz1p occurs concomitantly with its recruitment to septin rings (). WT and cells expressing Siz1-GFP were synchronized in M phase, and the subcellular localization of Siz1-GFP was examined by confocal microscopy. As shown in , deletion of inhibited the export of Siz1-GFP and its concentration at the bud neck. We examined the effect of inhibiting nuclear export of Siz1p on the sumoylation of septin ring components. A GFP-Sumo fusion protein was used to monitor the subcellular localization of SUMO-modified proteins by fluorescence microscopy (; ). As shown in , the GFP-Sumo signal concentrated in the nucleus and a diffuse cytoplasmic signal was also detected. In addition, the septin ring was strongly labeled in large-budded (M phase) WT cells (; ). Western blotting confirmed that, in addition to the free form, GFP-Sumo was readily incorporated into cellular proteins, including septins (unpublished data). However, no septin signal was detected in the mutant, suggesting that export of Siz1p from the nucleus is required for septin sumoylation. This conclusion was further supported by the results of experiments examining the sumoylation state of a septin, Cdc3-HA, containing a C-terminal HA tag. Consistent with previous reports, various sumoylated species of Cdc3-HA could be detected by Western blotting in cell lysates derived from WT cultures arrested in M phase with nocodazole (; ). However, the slower migrating sumoylated Cdc3-HA species were greatly reduced in lysates from cells, further supporting the conclusion that Kap142p/Msn5p-mediated export of Siz1p is required for normal septin sumoylation. Previously published data suggest that Ulp1p is required for septin desumoylation at cytokinesis (). Consistent with this, we observed that a strain (YRW122) containing a mutation (; ) in the catalytic domain of Ulp1p exhibited abnormal septin sumoylation. Large-budded cells expressing exhibited GFP-Sumo–labeled septins in both the mother and the bud. This was rarely observed in WT cells, where sumoylation was confined to the mother side of the septin ring (; and ). Moreover, septin sumoylation persisted beyond separation of the ring and cytokinesis in cells expressing the mutation (). This phenotype was not observed in WT cells (; ). Ulp1p has been localized to the nucleoplasmic side of the NPC (X. ), where it interacts with Kap121p and the Kap95–Kap60p complex (). Because has a closed mitosis, this localization raised the question as to how Ulp1p could desumoylate the cytoplasmic septins. We hypothesized that the association of Ulp1p with the kaps and the NPC may control its accessibility to the septin ring. We therefore examined the effects of the and alleles on the NPC association of Ulp1p and the sumoylation state of the septins at various points in the cell cycle. As shown in , the mutation appeared to have little effect on the NPC association of Ulp1p at either 23 or 37°C. In contrast, shifting the –containing strain to the nonpermissive temperature caused a decrease in the level of NPC-associated Ulp1p and a corresponding increase in its cytoplasmic levels. Quantification of these signals revealed a change in the nuclear/cytoplasmic ratio from ∼5 at 23°C to ∼1.5 at 37°C (unpublished data). This observation is consistent with data obtained with other mutant alleles (). Septin sumoylation was also examined in the and strains expressing or . Cells were synchronized in G1 phase at 23°C. The cultures were then split, and cells were placed in fresh medium containing α-factor (to maintain G1 arrest), hydroxyurea (to arrest in S phase), or nocodazole (for M phase arrest). After incubating at 37°C, Cdc3-HA sumoylation was evaluated by Western blot analysis. Consistent with previous results (), in a WT strain, sumoylated forms of Cdc3-HA were detected predominantly in nocodazole-arrested cells (). In contrast, M phase accumulation of sumoylated Cdc3-HA was completely inhibited in a mutant at 37°C, but was normal at 23°C (unpublished data) or upon complementation of the mutant with WT . Moreover, GFP-Sumo failed to accumulate at septin rings in M phase cells at 37°C, whereas nuclear levels were unaffected (). The strain appeared more similar to WT cells both in terms of Cdc3-HA sumoylation and GFP-Sumo labeling of the septins. However, GFP-labeled septin rings were detectable in both mother and daughter cells (). cells also exhibited increased amounts of sumoylated Cdc3-HA in α-factor–arrested cultures (); however, the relevance of this observation was difficult to assess, as these cells exhibited a slight decrease in the efficiency of arrest (88% of vs. 92% of WT cells). To assess whether changes in septin sumoylation observed in the kap mutants were linked to Ulp1p, we tested the effects of deletion mutations that remove the various kap binding domains of Ulp1p on the sumoylation state of Cdc3-HA and correlated this with their subcellular localization. Previous data from suggest that the Kap121p and Kap95p–Kap60p binding sites in Ulp1p are positioned within residues 1–150 and 150–340, respectively, and are outside of the region containing the catalytic domain (residues 403–621). In addition, Ulp1p contains a putative nuclear export signal (NES) in a region between residues 340 and 403. Deletion of the Kap121p binding domain of Ulp1p (Ulp1) led to a partial mislocalization of Ulp1p and a localization pattern similar to that observed for Ulp1p in the mutant (). Concomitant with the mislocalization of Ulp1p from the NPC was a failure to accumulate sumoylated Cdc3-HA at M phase (induced by nocodazole arrest; ). These data support the contention that Kap121p-mediated localization of Ulp1p at the NPC is important for efficient septin sumoylation. Deletion of the NES region (Ulp1) did not affect Cdc3-HA sumoylation or alter Ulp1p association with the nuclear periphery (). We also examined mutations lacking the Kap95p–Kap60p binding site (Ulp1). Ulp1-GFP exhibited no obvious changes in NE association compared with WT Ulp1p. Surprisingly, however, disruption of the Kap95p–Kap60p binding site led to the accumulation of Ulp1p-GFP at the septin rings in large-budded cells (, arrows). Likely as a consequence of this abnormal localization, septin sumoylation was completely inhibited in cells containing Ulp1 (). These data suggest that Kap95p–Kap60p negatively regulates Ulp1p binding to the septin ring. To further understand how Kap121p and Kap95p–Kap60p regulate the localization of Ulp1p, a mutant that lacked both Kap121p and Kap95p–Kap60p binding sites (Ulp1) was examined. This mutant failed to bind to either the septin ring or the NE (; unpublished data). Thus, Kap121p binding to Ulp1p, in addition to its role in targeting Ulp1p to the NPC, appeared to be required for the association of Ulp1 with the septin ring. We therefore directly examined the effect of the mutation on the binding of Ulp1 to the septin ring. In these cells, the Ulp1 mutant failed to accumulate at the NE or septin rings at 37°C and was dispersed throughout the cell (), supporting a role for Kap121p in directing the association of Ulp1p with the septin ring. Our data are consistent with a model in which the interaction of the Kap–Ulp1p complexes with the NPC control Ulp1p recruitment from the NPC to the septin rings. To further develop this model, we performed several experiments that focused on the nature of Ulp1p's interactions with the kaps and its relevance to septin sumoylation. For this analysis, plasmid-born gene fusions encoding Ulp1-GFP (containing the Kap121p binding domain) or Ulp1-GFP (containing the Kap95p–Kap60p binding domain) were introduced into cells (). Variation in plasmid numbers between cells within the culture allowed us to evaluate the localization of the fusion proteins at differing cellular levels. At low levels of expression, both Ulp1-GFP and Ulp1-GFP were visible at the NE. These observations support the conclusions of that both the Kap121p and the Kap95p–Kap60p binding regions of Ulp1p were capable of independently mediating its binding to the NPC, albeit with reduced efficiency relative to the WT protein. However, when levels of Ulp1-GFP were elevated, the chimera began to accumulate in the nucleus, suggesting that the NE binding sites were limited and excess protein was imported into the nucleoplasm. In contrast, Ulp1-GFP was generally observed evenly distributed along the NE in all cells. A punctate pattern was only occasionally detected in low-expressing cells (unpublished data). A similar uniform NE distribution was also observed upon overproduction of full-length Ulp1p (X. ). The localization patterns of Ulp1-GFP and Ulp1-GFP were dependent on functional Kap121p and Kap95p, respectively (unpublished data). These results suggest that Kap121p and Kap95p mediate the association of Ulp1p with distinct sites at the NPC and perhaps locations on the inner nuclear membrane (INM). The basis for the binding of Ulp1p to the NPC was further evaluated by examining the energy dependence of this interaction. Certain NPC-associated proteins, such as Mad1p, Mad2p, the Mlp proteins, and the nups themselves, including those that exhibit mobility, such as Nup2p, remain bound to the NE in the presence of metabolic poisons such as deoxyglucose (; ; ; unpublished data). Moreover, Kap60p and Kap95p concentrate at the NPCs after treatment of cells with deoxyglucose (). In contrast, Kap121p, although still visible in the cytoplasm, shows a marked accumulation in the nucleus under these same conditions (). The localization of Ulp1-GFP, however, was distinct from that observed for its interacting kaps. Within 20 min of treatment of cells with deoxyglucose and sodium azide, the NE concentration of Ulp1-GFP was largely lost and the protein was distributed throughout the cell (). These effects were rapidly reversed after removal of the poison. The apparent mobility of U1p1p led us to conclude that, at some point during mitosis, Ulp1p gains access to the cytoplasm to desumoylate the septins. To further test this idea, experiments were performed to determine whether a mutation that alters its mobility affects the sumoylation state of septins. Strains were isolated that express a mutant form of Ulp1p in which the kap binding domains were removed and the catalytic domain was fused to the nucleoporin Nup60p. This fusion protein is predicted to localize to the nucleoplasmic face of the NPC in a region where WT Ulp1p has been reported to reside (X. ). Consistent with this, Ulp1C-GFP-Nup60 localized to the NPC and could rescue the growth of the deletion mutant (; ; unpublished data). However, unlike the WT protein, Ulp1C-GFP-Nup60 does not appear to have access to the cytoplasm. As show in , treatment of cells with deoxyglucose and sodium azide did not induce the release of Ulp1C-GFP-Nup60 from the NE; moreover, it did not induce Cdc3-HA desumoylation in M phase–arrested cultures (). In contrast, these metabolic poisons induced septin desumoylation in WT cells where Ulp1p can access the cytoplasm. The predicted requirement of Ulp1p release from the NPC for septin desumoylation is also supported by data showing that in Ulp1C-GFP-Nup60–containing cells, septin sumoylation persisted beyond separation of the ring and cytokinesis similar to that observed in the mutant (). This GFP-Sumo ring signal is later lost as cells progress through G1 or in α-factor–arrested cells (unpublished data). Elevated levels of sumoylated Cdc3-HA were, however, detected throughout the cell cycle Ulp1C-GFP-Nup60 cells (). The likelihood that Ulp1p is released from the NPC combined with the observation that recruitment of Ulp1 to the septin ring was dependent on Kap121p linked this kap to the targeting step. This prompted us to examine whether mitosis-specific molecular rearrangements in the NPC that alter the binding of Kap121p to the NPC and inhibit its import pathway () contribute to releasing Ulp1p into the cytoplasm. In cells arrested in M phase with nocodazole, Kap121p becomes associated with Nup53p (), and we wondered if this change would alter the binding of Ulp1p at the NPC. Using a strain producing tandem affinity purification (TAP)–tagged Ulp1p, we tested whether Ulp1-TAP was recruited with Kap121p to Nup53p. Ulp1-TAP was purified from cells arrested in G1 (with α-factor) or M (with nocodazole) phase. Ulp1-TAP derived from both cultures was associated with Kap121p, Kap95p, and Kap60p. However, an enrichment of Nup53p was visible in the nocodazole-arrested culture, suggesting that both Kap121p and Ulp1p are recruited to Nup53p (). Consistent with this observation, we were able to reconstitute interactions between recombinant versions of these proteins (). Surprisingly, full-length Ulp1p bound Nup53p and Kap121p independently. However, GST-Ulp bound Kap121p, but not Nup53p. Nonetheless, a preformed complex of GST-Ulp–Kap121p bound to Nup53p, showing that these proteins could form a trimeric complex through Kap121p. The results described above suggest that the Kap121p–Ulp1p complex is recruited to Nup53p during mitosis. We therefore investigated the relevance of this interaction to septin sumoylation using specific mutations in Nup53p that fail to bind Kap121p (nup53). We speculated that if Nup53p plays a role in regulating the accessibility of Ulp1p to the septins, then the kinetics of septin desumoylation might be altered in this strain. To evaluate this, cells were released from nocodazole arrest and Cdc3-HA sumoylation was monitored over time. As shown in , we reproducibly observed that desumoylation of the Cdc3-HA was delayed in the strain relative to the WT strain. This effect was not due to a delay in progression out of mitosis, as both strains showed similar kinetics of Clb2p decay (). In addition, Siz1p release from the septin ring occurred at similar time after release from nocodazole arrest in both the WT and the strain (unpublished data). These results suggest that changes in the NPC association of Kap121p and Ulp1p contribute to timing of septin desumoylation. Several components of the septin ring are major cytoplasmic sumoylation targets. Septin sumoylation is cell cycle regulated and reversible, with sumoylated septins accumulating before anaphase and persisting until desumoylation after mitosis (). Here, we have shown that the nuclear transport machinery regulates septin sumoylation by controlling access of two key SUMO-modifying enzymes, Siz1p and Ulp1p, to the septins. Siz1p is localized to the nucleus during interphase by Kap95p, which would preclude the spurious sumoylation of cytoplasmic targets. Such a role for Kap95p could explain the elevated levels of sumoylated septins seen in the strain (). At the onset of anaphase, Siz1p is phosphorylated (), and it is exported from the nucleus by Kap142p/Msn5p. Phosphorylation may act as a signal for Siz1p export from the nucleus, as Kap142p/Msn5p has previously been shown to recognize NESs containing phosphorylated amino acid residues (; ; ). Kap142p/Msn5p is strictly required for the accumulation of Siz1p at the septin ring, suggesting that export mediated by this kap is required to generate a cytoplasmic pool necessary for the accumulation of Siz1p at the septin ring. Septin desumoylation is mediated by Ulp1p (; ), whose activity is controlled by kaps. Our data suggest a complex situation in which Kap121p and the Kap95p–Kap60p complex direct Ulp1p to distinct locations, which controls the accessibility of this isopeptidase to septin substrates. Mutations that disrupt the interaction of Ulp1p with Kap121p or the Kap95p–Kap60p complex cause an increase in the cytoplasmic levels of Ulp1p and a loss of septin sumoylation ( and ). These results are consistent with the hypothesis that these kaps function as negative regulators of Ulp1p activity against cytoplasmic targets. One possibility is that kap binding directly suppresses the isopeptidase activity of Ulp1p. However, this is unlikely, as not all Ulp1p substrates are similarly affected by Ulp1p mutants that inhibit kap binding (; see below). Instead, we envisage kaps negatively affecting the cytoplasmic activity of Ulp1p by sequestrating it at the NPC. Once Ulp1p is delivered to the NPC, it associates with the nucleoplasmic face of this structure, where it interacts, directly or indirectly, with the Mlp proteins (X. ). Here, Ulp1p would be sequestered away from cytoplasmic targets. Similarly, the cytoplasmic activity of SENP2, the human counterpart of Ulp1p, is suppressed by interaction with the NPC (). Conversely, the concentration of Ulp1p at the nuclear face of the NPC is likely to stimulate its activity directed against nuclear targets, including those in transit through the NPC. Thus, we predict that mutations in Ulp1p that decrease the efficiency of its targeting to the NPC would lead to higher levels of nuclear protein–SUMO conjugates. Consistent with this idea, have shown that deletion mutants of Ulp1p that reduce its NPC association (and binding to the kaps; ; ) lead to an overall increase in cellular levels of SUMO-modified proteins, the vast majority of which are contained within the nucleus (). These observations suggest that kaps control the access of Ulp1p to targets in both the cytoplasm and the nucleus. Although Ulp1p binds both Kap121p and the Kap95p–Kap60p complex, it remains unclear whether it binds simultaneously to these kaps or forms two complexes. The functional significance of binding multiple kaps is unclear, but it seems unlikely that the kaps play an entirely redundant role. For example, when independently examined, the Kap121p (Ulp1) and the Kap95p–Kap60p (Ulp1) binding domains of Ulp1p exhibited distinctly different localization patterns. The Kap121p binding domain is generally visible in a uniform perinuclear pattern, whereas the Kap95p–Kap60p domain displayed a punctuate NE and intranuclear localization pattern. The uniform distribution and the capacity of the NE to accommodate overproduced Ulp1 or WT Ulp1p (X. ) is most likely explained by direct binding to the INM. Several observations led us to this conclusion. First, overproduced Ulp1 and Ulp1p are restricted to the NE and are not detected in the peripheral ER, as would be expected if they were associated with the outer nuclear membrane. Second, Ulp1p is visible in association with INM-derived intranuclear membranes that are induced by the overproduction of the nucleoporin Nup53p (; Fig. S1, available at ). Interestingly, both Nup53p and Ulp1p may follow pathways similar to the INM. In both cases, Kap121p is required for NE accumulation of Ulp1 (unpublished data). These observations raise the intriguing possibility that Ulp1p can access the INM through a process that is controlled by Kap121p. This would place Ulp1p in a position to regulate the sumoylation state of INM-associated proteins, including chromatin components. We have also shown that the steady-state association of Ulp1p with the NE is energy dependent. Energy-dependent association is not a feature of any of the nups or currently known NPC-interacting proteins. Interestingly, the dispersed localization of Ulp1p induced by 2-deoxyglucose/sodium azide is also distinct from that observed for the bulk of Kap121p or Kap95p and Kap60p, which either accumulate in the nucleus or at the NPC, respectively. This suggests that the Ulp1p–kap complexes are specifically dissociated in the absence of energy or that they are functionally distinct from the bulk of the kaps. The former possibility seems remote, as Ulp1–kap complexes can be isolated from cell lysates and reconstituted in vitro in the absence of energy (; ). A more likely scenario is that kaps bound to Ulp1p are not part of the cycling pool of kaps involved in the nuclear transport cycle. Several observations support this model. First, the interaction of Ulp1p with Kap121p and the Kap95p–Kap60p complex may prevent these kaps from binding RanGTP (). have shown that Ulp1p is not dissociated from Kap121p or the Kap60p–Kap95p complex by treatment with RanGTP. Moreover, we have shown that overexpression of Ulp1 inhibits Kap121p-mediated transport and induces a nuclear accumulation of Kap121p (Fig. S2, available at ). Both of these phenotypes can be explained by Ulp1 inhibiting the binding of RanGTP to Kap121p, as similar phenotypes are also observed upon depletion of cellular levels of RanGTP (). On the basis of these observations, including the karyopherin and energy-dependent NPC association of Ulp1p, we suggest that Ulp1p dynamically interacts with the NPC. Although its primary function likely occurs at the NPC, we envisage that Ulp1p can also migrate to other locations in the cell, where it desumoylates specific targets, including those in the INM and the septin rings. In the latter case, this movement appears to be cell cycle regulated, occurring during mitosis. The hypothesis that Ulp1p can leave the NPC can explain previous observations that Ulp1p is required to desumoylate cytoplasmic proteins. For example, in addition to Ulp1p being required for the desumoylation of septins (), the Ulp1 is required to prevent the accumulation of SUMO modifications on the cytoplasmic protein glutamyl-prolyl-tRNA synthetase (). Moreover, human SENP2 has also been shown to shuttle between the nucleus and the cytoplasm (). Finally, the movement of Ulp1p from the NPC appears to be required for normal septin desumoylation, as this process is inhibited by anchoring of the catalytic domain of Ulp1p on the nucleoplasmic face of the NPC by tethering it to Nup60p (). Although both Kap121p and the Kap95p–Kap60p complex function in sequestering Ulp1p at the NPC, we also propose that they play a role in regulating the access of Ulp1p to targets in the cytoplasm, specifically, the septins. These events are likely controlled by the interactions of the Ulp1p–kap complexes with the NPC. Importantly, we show that, when unable to bind to the Kap95p–Kap60p complex, Ulp1p (Ulp1) is detected at the septin ring in large-budded cells. This targeting event is dependent on Kap121p, as mutations that abolish Ulp1p–Kap121p binding (Ulp1 [] and Ulp1 [not depicted]) and the mutation () prevent the association of Ulp1p with the septin ring. Our interpretation is that, although both Kap121p and the Kap95p–Kap60p complex function in concentrating Ulp1p at the NPC during interphase, Kap121p plays a distinct role in targeting Ulp1p to septin rings during mitosis. A key step in this process would likely involve cell cycle–specific changes in the binding of the Ulp1p–kap complex to the NPC that facilitate release and recruitment of Ulp1p to binding sites at the septin ring. Consistent with this idea, the positioning of Ulp1p and Kap121p within the NPC is altered during mitosis, with both being recruited to sites where they interact with Nup53p (). It has previously been shown that the association of Kap121p with Nup53p inhibits its translocation through the NPC and import of its cargos, leading to their accumulation in the cytoplasm (). These, or as-yet-undefined, changes in the NPC could contribute to increased cytoplasmic levels of Ulp1p and facilitate its association with the septin ring. We speculate that, at points during mitosis, the opposing activities of Kap121p-dependent septin association and Kap95p–Kap60p–mediated NPC targeting allow Ulp1p to cycle between the NPC and the septin ring. The concerted targeting functions of the kaps would presumably tightly regulate the levels of Ulp1p that can associate with septin rings. Thus, binding of Ulp1p to the septin rings is likely to be transient. This, coupled with low cellular levels of Ulp1p, would explain our inability to detect cytoplasmic Ulp1p, except under conditions where its interactions with kaps are altered. Interestingly, targeting of Ulp1p to the septin ring may occur before completion of mitosis, when levels of sumoylated septins drop. We observed that Ulp1 is detected at septin rings in large-budded cells before nuclear division (), coinciding with time points when Siz1p is also associated with the septin ring. Moreover, a catalytic domain mutant of Ulp1p showed a more predominant septin sumoylation pattern during M phase and a delay in septin desumoylation after cytokinesis (). This would imply that steady-state levels of sumoylated septins are a product of the relative activities of both Ulp1p and Siz1p, with the balance being shifted to the desumoylated state by the release of Siz1p from the septin ring and its reimport into the nucleus at the end of mitosis. Strains used in this study are presented in Table S1 (available at ). The gene encoding Cdc3-HA was integrated within the strains indicated in Table S1 using a PCR product derived from a previously described yeast strain (), provided by E. Johnson (Thomas Jefferson University, Philadelphia, PA). Yeast strains were grown at 30°C unless otherwise indicated in YPD or synthetic media (SM) supplemented with appropriate nutrients and 2% glucose (). Yeast transformations were performed as described by . All of the following plasmids contain inserts synthesized using Expand High Fidelity PCR system (Roche Diagnostics). pRS316 ULP1-GFP, pRS316 ULP1-GFP, pRS316 ULP1-GFP, pRS316 ULP1-GFP, pRS316 ULP1-GFP, and pRS316 Ulp1-GFP were constructed as follows. The promoter region of (−300 to +3, where +1 corresponds to the A of the start codon) was amplified with Xho1 and EcoR1 linkers and inserted into the corresponding sites of pRS316 (; ). Next, the GFP ORF () was amplified with Sac1 and Not1 linkers and cloned into the corresponding sites of pRS316. The following regions of the ORF were then inserted into the resulting plasmid at an EcoR1 site after the promoter and in frame with the 5′ end of the ORF: +4 to +1863 (encoding Ulp1p), +4 to +450 (Ulp1), +448 to +1020 (Ulp), +451 to +1863 (Ulp1), +4 to +448 linked to +1021 to +1863 (Ulp1), and +4 to +1021 linked to +1210 to +1863 (Ulp1). Note that the latter two inserts contain a BssH II linker between the two ULP1 fragments. pRS316 ULP1-GFP, pRS316 ULP1-GFP, pRS316 ULP1-GFP, and pRS316 ULP1 -GFP were introduced into a -null mutant by plasmid shuffle. Each plasmid complemented the lethal phenotype of a -null mutant. pGULP1 and pGULP1 were constructed by cloning nucleotides 1–450 and 1–1863 of the ORF plus stop codons into the BamH1 site of pGEX-6P-1 (GE Healthcare). pYULP1 was constructed by cloning nucleotides 1–450 of the ORF followed by a stop codon into BamH1 and EcoR1 sites of pYEX-BX (CLONTECH Laboratories, Inc.). pcNLS-GFP (), pGNUP53, pGKAP121 (), pPHO4-NLS-GFP, pKAP121-GFP, and pYNUP53 (; ) were previously described. pULP1C-GFP-NUP60 and pSMT3-GFP () were provided by E.C. Hurt (University of Heidelberg, Heidelberg, Germany). Cell pellets used for fluorescence microscopy were suspended in 20 μl of an appropriate SM, and 2 μl of this suspension was spotted onto a slide. Images of GFP fusion proteins were acquired at room temperature as 0.7-μm optical sections using a Plan-Apochromat 63×/1.4 NA oil differential interference contrast objective on an microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) equipped with a confocal scanning system (LSM 510 META; Carl Zeiss MicroImaging, Inc.). All images were acquired using LSM 510 software and viewed using LSM Image Browser. Images were subsequently incorporated into figures using Draw 10 (Corel) and Photoshop (Adobe). Strains containing the endogenous ORF tagged after the last codon with the HA epitope were made as follows. Genomic DNA was isolated from the strain EJY301 containing the gene (). The gene was amplified by PCR, and the product was transformed into the indicated strains. WT, , , and YRW110 cells expressing were synchronized in G1 phase with 7.5 μg/ml α-factor (Sigma-Aldrich) at 23°C as described previously (). α-Factor was removed by washing cells in YPD. Cells were then resuspended in fresh YPD media containing either 100 mM hydroxyurea (Sigma-Aldrich) or 17 μg/ml nocodazole (Sigma-Aldrich) and shifted to 37°C for 2.5 h. Cell cycle arrest was monitored by examining cell morphology and by Western blotting (see Western blot analysis) to detect levels of the mitotic cyclin Clb2p. Cdc3-HA was detected by Western blotting of whole-cell lysates prepared as described in the following section. Levels of Cdc3-HA sumoylation in strains that were not temperature sensitive were determined as described, except cells were grown at 30°C. Cdc3-HA desumoylation was also examined after release from nocodazole arrest in DF5 and cells. Cells were arrested with 17 μg/ml nocodazole for 2.5 h. After this incubation period, cells were washed with YPD to remove nocodazole and resuspended in YPD. At the indicated times, cells were harvested and washed once with water, and whole-cell lysates were prepared for Western blot analysis. Whole-cell lysates were prepared from cell cultures as follows. Cells derived from 2-ml cultures were harvested by centrifugation, washed with water, and sonicated for 20–30 s in 35 μl SDS-PAGE sample buffer. Samples were then incubated at 75°C for 10 min, and proteins were separated by SDS-PAGE. Proteins were then transferred to nitrocellulose membranes, and membranes were blocked with 5% skim milk and 0.1% Tween 20 in PBS. HA moieties were detected using a monoclonal anti-HA antibodies (3F10) conjugated to HRP (Roche Diagnostics) and the ECL system (GE Healthcare). Specific rabbit polyclonal antibodies were used to detect Clb2p (Santa Cruz Biotechnology, Inc.), GFP, Gsp1p (), Nup53p, and Kap121p (). Binding of primary antibodies was detected using HRP-conjugated donkey anti-rabbit secondary antibodies and ECL. Cells synthesizing Ulp1-TAP () were grown in YPD (0.3 OD/ml) and arrested in G1 or M phase as described previously (). After harvesting, cells were lysed and Ulp1-TAP was isolated using IgG-Sepharose (GE Healthcare) chromatography as previously described (; ). After isolation, bead-bound complexes were washed extensively with lysis buffer containing 25 mM MgCl and eluted using a step gradient of MgCl (200, 500, and 1,000 mM) followed by a 0.5-M acetic acid, pH 3.4, elution. Proteins in each eluate were TCA precipitated, separated by SDS-PAGE, and transferred to nitrocellulose for Western blotting. cells transformed with pGKAP121, pGULP1, pGULP1, or pGNUP53 were grown to midlog phase and induced with 1 mM IPTG for 4 h. Cells were lysed, and GST fusions were purified on glutathione–Sepharose beads according to the manufacturer's instructions (GE Healthcare) using 150 mM NaCl, 1 mM MgCl, 0.1% Tween 20, and 50 mM Tris, pH 7.5, as a lysis and wash buffer. Purified Kap121p and Nup53p were recovered by treatment of the bead-bound fusions with thrombin and Precision Protease (GE Healthcare), respectively. To evaluate binding to Ulp1p, 10 μg Kap121p or Nup53p in 40 μl of lysis buffer were incubated for 45 min at 4°C with 10 μl glutathione–Sepharose beads preloaded with ∼10 μg GST-Ulp1p or ∼5 μg GST-Ulp1. Unbound fractions were collected, and beads were eluted with SDS-PAGE sample buffer. Alternatively, the GST-Ulp1–Kap121p complex was further incubated with ∼10 μg of recombinant Nup53p for 1 h at 4°C, and bound and unbound fractions were collected. Proteins in the various fractions were separated by SDS-PAGE and detected with Bio-Safe Coomassie (Bio-Rad Laboratories). Treatment of cell cultures with metabolic poisons was performed essentially as described previously (). Cell cultures were grown in YPD or SM in the presence or absence of 10 μg/ml nocodazole for 2.5 h. Cultures were then centrifuged, and cell pellets were washed twice with media lacking glucose. Pellets were resuspended at room temperature in medium lacking glucose and supplemented with 100 mM 2-deoxyglucose (Sigma-Aldrich) and 17 mM sodium azide (ICN Biomedicals) and, where indicated, 10 μg/ml nocodazole (Sigma-Aldrich). Localization of GFP-tagged proteins was detected at the indicated time points. For recovery, cells were washed twice with PBS to remove metabolic poisons, resuspended in media containing 2% glucose, and incubated at 30°C for the indicated times. For the induction of promoter–controlled expression of and , strains harboring the pY ULP1 or pY NUP53 plasmids were grown to midlogarithmic phase in SM, and expression was induced for 6 h by the addition of copper sulfate to a final concentration of 0.3 mM. Fig. S1 shows that Ulp1-GFP is associated with the INMs in Nup53p-overproducing cells. Fig. S2 shows that overproduction of the Kap121p binding domain of Ulp1p inhibits the import of Kap121p cargoes and alters Kap121p localization. Table S1 contains a list of the strains used in this study. Online supplemental material is available at .

POSH (plenty of SH3s) was initially identified as a Rac-binding protein and an activator of the JNK and nuclear factor κB signaling pathways (). Subsequently, POSH was shown to activate JNK signaling by acting as a scaffold for mixed lineage kinases (), a function negatively regulated by the protein kinase Akt2 (). As a regulator of JNK, POSH is mainly implicated in the activation of apoptosis and differentiation of neuronal cells (; ; ). Apoptotic stimuli increase the expression of POSH, mixed lineage kinases, JNK, and Siah1, and the latter is a POSH-interacting E3 ligase and a known activator of the JNK pathway (). Conversely, siRNA-mediated silencing of POSH confers neuroprotection (). In contrast to its proapoptotic function in mammalian neurons, the neuronal-specific expression of POSH extends the longevity of adult fruit flies (; ). POSH contains an N-terminal RING finger domain, which is a hallmark of many ubiquitin (Ub) E3 ligases. The E3 Ub ligase family of proteins consists of hundreds of structurally diverse enzymes that determine the specificity of Ub conjugation through specific recognition of substrates and the recruitment of cognate E2 Ub-conjugating enzymes (). Indeed, recent studies implicate the Ub ligase function of POSH in the production of infectious HIV-1 () in degradation of the early endosome resident sorting factor Hrs (hepatocyte growth factor–regulated tyrosine kinase substrate; ) and in control of the immune system via degradation of the JNK activator TAK-1 (). We now report the identification of Herp (homocysteine-inducible ER protein) as a novel ubiquitination substrate and regulator of POSH. Herp, which contains a Ub-like domain, is an ER stress–inducible protein critical for cell survival under stress. The underlying mechanism by which Herp exerts its protective function has been obscure, although it likely involves the control of calcium homeostasis during ER stress (). We previously showed that POSH is a TGN-associated protein despite lacking a detectable transmembrane domain (). In the present study, we show that POSH associates with the TGN membrane through an association with Herp. Within minutes of the perturbation of intracellular calcium by the calcium-perturbing agent thapsigargin (Tpg) or induction of ER stress by the glycosylation inhibitor tunicamycin (Tm), Herp is redistributed from the TGN to the ER. The increased ER expression of Herp occurs long before the ER stress–induced enhancement of Herp expression and is dependent on the POSH-mediated conjugation of lysine-63–linked poly-Ub chains to Herp. Thus, we provide evidence that POSH regulates calcium homeostasis by increasing the levels of Herp in the ER. Herp was identified as a POSH-interacting protein through a yeast two-hybrid screen of a HeLa cDNA expression library. The POSH construct used as bait in the screen lacked the RING domain. To confirm the interaction between POSH and Herp, detergent extracts from cells transiently coexpressing epitope-tagged POSH and Herp were subjected to immunoprecipitation followed by Western blot analysis. POSH and Herp were both coprecipitated by antibodies to the complementary protein (). Similar coimmunoprecipitation was observed in vitro after the incubation of bacterially expressed maltose-binding protein (MBP) POSH fusion protein (POSH) and truncated Herp (tHerp; amino acids 1–272), which lacked the putative membrane-associated domain (). Therefore, the interaction between POSH and Herp is direct. POSH is associated exclusively with the TGN membrane (), whereas Herp was reported as an ER resident protein (). If the interaction between POSH and Herp is physiologically relevant, it would require that both proteins be present in the same intracellular compartment. To resolve this issue, we used immunofluorescence microscopy to determine the intracellular localization of endogenous Herp. The results indicate a polar distribution of endogenous Herp around the nucleus and colocalization with the TGN marker TGN46 (). To further confirm the TGN localization of Herp, we determined the intracellular distribution of Herp in cells expressing a fusion between autocrine motility factor receptor (AMFR) and GFP. The E3 ligase gp78/AMFR is an integral ER membrane protein (). Consequently, AMFR-GFP is detected in a characteristic ER reticular network structure throughout the cell, from which Herp, which retains its polar distribution, is mostly excluded (, d). A similar distribution of Herp is observed in cells stained with POSH and Herp antibodies. Herp colocalizes with POSH in a typical TGN appearance, and both proteins are mostly excluded from the ER based on costaining with an antibody against the ER resident chaperone calnexin (). The discrepancy between our findings and previous studies (; ) showing mainly ER localization for Herp is likely an outcome of different immunostaining protocols. The staining protocol used throughout this study excludes detergents, as we had noticed that Herp loses its TGN localization when 0.05% Tween 20 is used in the immunofluorescence staining procedure (Fig. S1, available at ). Related to this is the unresolved issue of the membrane topology of Herp. A previous study indicated that the majority of Herp faces the cytoplasm and predicted a short transmembrane segment (; ). However, no physical evidence has been provided for membrane integration. The fact that the subcellular localization of Herp is sensitive to mild detergents suggests either peripheral membrane association or dynamic insertion of the protein into the TGN membrane. Nevertheless, we cannot exclude the possibility that Herp is also expressed at low levels in the ER. POSH is a soluble protein associated with the TGN membrane. Because Herp expresses a hydrophobic C-terminal region implicated in membrane binding (), we tested whether Herp mediates POSH binding to the TGN. To this end, we determined the subcellular localization of POSH in cells treated with siRNA to knock down the expression of Herp (). Immunofluorescence analysis demonstrates that silencing of Herp expression causes a redistribution of POSH throughout the cell. Therefore, POSH association with the TGN membrane is indeed Herp dependent. The association of Herp with POSH prompted us to test whether Herp is a POSH ubiquitination substrate. POSH ubiquitinates itself in the presence of the E2 ligases UbcH5 () or Ubc13/Uev1A (). Therefore, we tested whether Herp is ubiquitinated by POSH in the presence of either E2 enzyme (). When tHerp-Flag was incubated in an in vitro ubiquitination reaction in the presence of Ubc13/Uev1a and POSH, high mol wt tHerp–poly-Ub conjugates were efficiently synthesized based on anti-Flag immunoblotting. Interestingly, very low mol wt Herp-Ub conjugates were also generated in the absence of POSH. In a ubiquitination reaction using UbcH5, Herp was also ubiquitinated, but only low mol wt species corresponding to one to approximately six conjugated Ub molecules per Herp protein were generated. The heteromeric Ubc13/Uev1a E2 is exclusively involved in the generation of lysine K63–linked poly-Ub chains (; ). Rather than target proteins for proteasomal degradation, K63-linked polyubiquitination serves regulatory functions mainly through the promotion of protein–protein interactions (). Activation of the inhibitor of nuclear factor κB kinase requires the formation of K63-linked poly-Ub chains on the TNF receptor–associated factor (TRAF) 2 or 6 adaptor proteins. The TRAFs are also RING proteins that catalyze the formation of K63-linked Ub chains on themselves and on other proteins. This activity is promoted by distinct activators and coincides with activator-induced TRAF oligomerization (; ). By analogy, the association of POSH with Herp and the fact that POSH also mediates the formation of K63-linked poly-Ub chains prompted us to test whether Herp also functions as a POSH activator in addition to constituting a POSH substrate. To this end, we performed an in vitro POSH self-ubiquitination assay in the absence or presence of Herp and subsequently detected POSH ubiquitination by Western blot analysis with a POSH mAb. High mol wt POSH-Ub adducts are substantially stimulated upon the addition of increasing concentrations of tHerp (). Activation of POSH by Herp is specific to Ubc13/Uev1a () and requires the Herp Ubl domain (). We next tested whether POSH activation also requires its oligomerization, as occurs with the activation of TRAF. First, we investigated whether Herp influences the oligomeric state of POSH by determining POSH sedimentation in glycerol density gradients in the presence or absence of Herp. The results indicate that when POSH is incubated alone, it migrates as a low mol wt species in fractions 2–4 (, top). tHerp migrates in two distinct positions in the gradient (a minor low mol wt peak and a major high mol wt peak), indicating that the bulk of tHerp is in an oligomeric state (, middle). Notably, when POSH is incubated together with Herp before centrifugation, a portion of POSH subsequently appears as a broad high mol wt peak in fractions 7–10, and this peak coincides with that of tHerp (, bottom). It is noteworthy that the peak of oligomeric Herp moves further toward the bottom of the gradient in the presence of POSH. These observations strongly suggest that POSH and Herp form a heterooligomeric complex. Because Herp both activates POSH self-ubiquitination and induces POSH oligomerization, we asked whether these two Herp activities were related. The Ubl domain of Herp is essential for POSH activation, so we tested whether Herp was also defective for inducing POSH oligomerization. Indeed, unlike tHerp, tHerp failed to increase POSH sedimentation in a glycerol density gradient (, top). To confirm that oligomerization activates POSH, we determined the self-ubiquitination activity of POSH protein isolated from the gradient fractions (, bottom). Aliquots from each fraction were incubated with E1, Ub-activating enzyme, Ubc13/Uev1a, Ub, and ATP in an in vitro ubiquitination assay. In this reaction, Herp-Ub as well as POSH-Ub conjugates are synthesized. Therefore, ubiquitinated POSH was determined only after the removal of Herp-Ub conjugates by immunoprecipitation. The results indicated that after incubation of POSH with tHerp, two activity peaks were detected. The first at fractions 2 and 3 corresponded to low mol wt POSH, and the second at fractions 6–9 coincided with oligomeric POSH. The activity in the oligomerized POSH fractions was considerably higher than that in the monomeric POSH fractions despite the smaller amount of POSH in the high mol wt fractions, indicating that oligomerization substantially increases POSH-specific activity. As expected, only a single activity peak coinciding with monomeric POSH appeared in the presence of tHerp. Together, these results indicate that the oligomerization of POSH is closely linked to Herp-induced POSH activation. The failure of tHerp to activate POSH does not result from an inability to interact with POSH. A Herp pull-down experiment indicated that POSH was present in Herp as well as in Herp immune complexes (). This result is consistent with the decreased size of Herp oligomers as determined by glycerol density centrifugation () and suggests that POSH oligomerizes through interaction with high mol wt Herp oligomers, the formation of which requires the Ubl domain. Herp is an ER stress–induced protein whose reported functions include the enhancement of ER-associated degradation (ERAD) via association with the E3 ligase HRD1/synoviolin () and maintenance of low cytosolic calcium during ER stress (). Both of these functions suggest that the Herp ER stress–protective functions are performed at the ER membrane. Because we find Herp associated primarily with the TGN, we postulated that Herp is deployed to the ER upon the induction of ER stress. To test this hypothesis, we determined the subcellular localization of Herp in resting and ER-stressed cells by immunofluorescence microscopy. ER stress was induced either by perturbation of calcium by the calcium ATPase inhibitor Tpg, which causes the rapid accumulation of calcium from the ER lumen to the cytosol, or by the protein glycosylation inhibitor Tm, which causes the accumulation of misfolded proteins in the ER. Immunostaining of endogenous Herp indicates that after a 1-h incubation with Tpg, Herp distribution changes dramatically from a focal to a diffuse appearance (, top). Confocal microscopy further reveals that after Tpg treatment, Herp staining generally overlaps with the ER marker calnexin (, bottom). Herp stress-induced redeployment is also observed after Tpg treatment of PC3 and SW480 prostate and colon cancer cell lines, respectively, indicating that this is a general response in multiple cell types (Fig. S3, available at ). Upon ER stress, the expression of Herp is strongly elevated as a result of the induction of gene transcription (; ; ). Thus, we tested the possibility that Herp redistribution to the ER is caused by its strong overexpression. To resolve this issue, we compared the kinetics of stress-induced Herp redeployment with the induction of Herp expression. Western blot analysis of endogenous Herp demonstrates distinct kinetics for Herp redistribution and induction of Herp expression. Induction of expression occurs between 3 and 6 h after Tpg addition (, top). In contrast, live cell imaging of Herp-RFP indicates that deployment to the ER is initiated within 15 min of the addition of Tpg and is complete within 45 min. Similarly, Herp relocation is complete within 2 h, whereas Herp protein accumulation occurs only between 3 and 6 h after the initiation of Tm treatment (Fig. S2). Thus, Herp stress-induced redistribution clearly occurs on a much faster timescale than Herp protein induction, so increased Herp protein levels during ER stress cannot account for its rapid relocalization. We also tested the possibility that stress-induced ER deployment of Herp results from the selective stabilization of a rapidly turning over (unstable) Herp population in the ER. Thus, we determined the effect of the protein synthesis inhibitor cycloheximide (CHX) on Herp redistribution. If there was a rapidly degraded pool of ER-localized Herp that was stabilized by Tpg treatment, the appearance of Herp in the ER would require continuous protein synthesis coupled with the inhibition of protein degradation. If the stabilization of Herp in the ER was the mechanism of Herp redistribution, CHX, by preventing new protein synthesis, would also prevent Herp ER appearance. Consequently, cells were incubated either with solvent or CHX alone or with Tpg for 1 h followed by visualization of Herp by immunofluorescence microscopy (, top). The result indicated that Tpg-induced Herp redistribution was indistinguishable in the presence and absence of CHX, establishing that new protein synthesis was not involved. In a parallel experiment, CHX completely abolished the accumulation of unstable Ub-GFP fusion protein () in the presence of the proteasome inhibitor MG132, confirming the potency of the CHX inhibition (, bottom). From these results, we conclude that immediately after the induction of ER stress, Herp relocalizes from the TGN to the ER. The fact that Herp is a POSH ubiquitination substrate prompted us to investigate whether POSH-mediated ubiquitination is involved in the stress-induced Herp redistribution. To this end, we tested whether redistribution is inhibited by expression of the RING finger mutant POSH, which is inactive as a Ub ligase and was previously shown to function in a dominant-negative fashion (). Cells were transfected with Herp-RFP together with either native or the dominant-negative POSH. Tpg was subsequently added to each culture, and Herp-RFP was immediately visualized. The live fluorescence analysis revealed that although Tpg stimulated the redistribution of Herp when wild-type POSH was overexpressed, Herp was refractory to Tpg treatment and retained its polar distribution in the presence of the dominant-negative POSH mutant (). A previous study showed that Herp is rapidly degraded in a proteasome-dependent fashion (). In the present study, we show that POSH promotes the conjugation of K63- but not K48-linked Herp–poly-Ub chains, suggesting that it is not directly involved in the degradation of Herp. Therefore, to further analyze the functional role of POSH and of K63 polyubiquitination, we compared the effect of Tpg on the redistribution of Herp and Herp, a mutant in which all of the lysine residues within the Ubl domain had been substituted with arginines. As shown in , when cells were incubated with DMSO, Herp staining appeared as small speckles discrete from calnexin. As expected, upon Tpg treatment, wild-type Herp was redistributed and colocalized with calnexin, indicating ER localization (, left), whereas Herp remained immobile (, right). The inability of Herp to redistribute in response to stress can result from either an inability to activate POSH or because it cannot be ubiquitinated as a result of the loss of the ubiquitination acceptor sites (or both). To distinguish between these possibilities, we compared the capacity of wild-type and mutant Herp to serve as substrates of POSH and as POSH activators in vitro. The comparison indicated that the lysine mutations completely abolished the Ubc13/Uev1a-dependent ubiquitination, whereas the UbcH5-dependent ubiquitination of both Herp proteins was essentially similar (). Both wild-type and mutant Herp demonstrate a similar ability to activate POSH self-ubiquitination (), indicating that Herp can functionally interact with POSH. The correlation between the abolishment of K63-linked polyubiquitination in vitro and the failure to relocalize Herp in vivo suggests that POSH-mediated K63-linked polyubiquitination of the Ubl domain of Herp is essential for Herp relocalization to the ER. Herp ubiquitination in the presence of UbcH5c indicates that lysine residues downstream of the Ubl domain can also serve as ubiquitination sites. Herp ubiquitination at downstream lysine residues is consistent with the results of , who showed that truncation of the Ubl domain inhibits the proteasomal degradation but does not eliminate ubiquitination. Thus, the functional significance of Herp ubiquitination downstream of the Ubl domain remains obscure. The model by which POSH-mediated K63-linked polyubiquitination is essential for calcium-dependent Herp relocalization predicts that Herp is ubiquitinated in vivo upon the induction of ER stress. To test this prediction directly, we monitored the effects of Tpg and POSH on the kinetics of Herp ubiquitination in a stable POSH knockdown cell line that expresses substantially reduced POSH levels (). To this end, H310 cells were cotransfected with Herp-Flag and HA-tagged Ub, a Ub derivative in which all lysine residues but lysine-63 are mutated to arginines, in the presence or absence of POSH overexpression. The cells were subsequently treated with Tpg for various time periods, after which Herp ubiquitination was analyzed by Western blot analysis of isolated Herp- Flag immune complexes with anti-HA. The results (, time 0) show that the overexpression of POSH markedly increased the ubiquitination of Herp in H310 cells, indicating that POSH is a rate-limiting factor for Herp K63-linked polyubiquitination in vivo. The level of Herp ubiquitination is further stimulated after the addition of Tpg: a considerable increase in Herp polyubiquitination is observed at 10 min and further intensifies at 30 min after Tpg addition. A slight decrease in Herp ubiquitination is subsequently observed after 90 min. Collectively, these results indicate that the ubiquitination of Herp is POSH dependent and is regulated by calcium. Furthermore, the kinetics of Herp ubiquitination in vivo is coincident with the kinetics of Tpg-induced Herp redistribution () and, thus, supports a mechanism whereby POSH-mediated Herp ubiquitination is activated by calcium and regulates Herp ER localization. Herp restricts calcium flow in neuronal cells subjected to agents that transiently increase intracellular calcium such as Tpg and bradykinin (). Maintenance of low cytosolic calcium likely requires the regulation of ER calcium channels by Herp. As POSH-mediated ubiquitination is obligatory for the recruitment of Herp to the ER, we predicted that inhibition of POSH activity would accelerate calcium release from the ER in the presence of Tpg. To test this hypothesis, we determined the initial rates of Tpg-induced calcium release in mock-transfected cells and in cells expressing either wild-type or dominant-negative POSH. Because during the experiment, cells were kept in a calcium-free medium, the observed elevation in intracellular calcium reflected the influx of ER calcium. The results () demonstrate that in the control culture, calcium release was initiated ∼2 min after the addition of Tpg and reached a maximum after 4 min. Consistent with the hypothesis that POSH function is rate limiting for Herp relocalization, the initiation of calcium release was considerably delayed in cultures overexpressing native POSH and started only 6 min after Tpg addition. In contrast, in cultures expressing POSH, the rate of calcium release was dramatically accelerated: it was initiated instantaneously, increased rapidly, and reached a twofold higher maximal value within 2 min. The initial rate of calcium release was also dramatically accelerated in Tpg-treated H310 cells relative to control cells (), further indicating the essential role of POSH in the regulation of cytosolic calcium. Together, these results establish a crucial role for POSH-mediated ubiquitination in the maintenance of calcium homeostasis through the regulation of Herp. Herp is an essential factor for ER stress resistance. A Herp knockout study in mice indicates that in Herp-null cells, ER stress signaling and ERAD are reduced, whereas ER stress–induced cell death is increased (). The molecular basis for Herp function during stress has recently started to be addressed. Herp function was directly linked to the ERAD machinery and to the stabilization of calcium homeostasis, functions that both require ER localization. Nevertheless, all studies describing the role of Herp during stress are based on data obtained from cells that were subjected to prolonged stress periods. In this study, we report a regulatory mechanism that recruits Herp to the ER immediately after the induction of ER stress. Although we cannot exclude ER localization, we provide evidence that in resting cells, Herp is primarily associated with the TGN () and that only after the induction of stress does it appear mainly in the ER ( and ). Stress-induced ER expression of Herp occurs within minutes and is independent of stress-induced Herp protein induction (), suggesting that Herp protein relocalization is involved. The mechanism of Herp relocalization is unclear. Because Herp is a membrane-embedded protein, retrograde transport likely involves release from a TGN membrane retention factor followed by vesicular traffic, possibly mediated via COPI (coatomer complex I) vesicles (; ). Retention of Herp in the ER may be achieved through binding to an ER membrane protein. Recent data demonstrating a direct interaction between the Ubl domain of Herp and the integral ER membrane E3 ligase Hrd1 suggests that Hrd1 may function as a Herp ER retention factor (). Whether Herp interacts exclusively with Hrd1 or with additional integral membrane ERAD E3 ligases such as gp78/AMFR remains to be explored. We establish a critical function for POSH-mediated ubiquitination in stress-induced Herp translocation by showing the inhibitory effect conferred by the expression of dominant- negative POSH (). The role of POSH-mediated Herp K63 polyubiquitination is further strengthened by the observed correlation between the exclusive requirement for lysine residues within the Ubl domain for stress-induced translocation and for Ubc13/Uev1a-dependent Herp ubiquitination in vitro (). As would be anticipated if POSH-mediated Herp ubiquitination drives translocation, we find that Tpg-activated ubiquitination slightly precedes ER mobilization: Herp polyubiquitination is observed as early as 10 min, whereas Herp relocalization is distinguishable only 15 min after the addition of Tpg ( and ). Regulation of signal-induced protein translocation by K63-linked polyubiquitination was initially demonstrated by , who showed an absolute requirement for TRAF6-mediated polyubiquitination for nuclear translocation of the neurotrophin receptor–interacting factor (NRIF) upon interaction of neurotrophin with its receptor. These investigators further demonstrated a correlation between the inhibition of TRAF6-induced NRIF ubiquitination and resistance to neurotrophin-induced apoptosis, thereby establishing a physiological role for TRAF6-mediated NRIF ubiquitination. Similarly, in our study, we establish the physiological role for POSH-mediated Herp ubiquitination by showing that the expression of dominant-negative POSH inhibits ER stress–induced Herp relocalization to the ER () and stimulates free calcium release (). The latter effect of POSH is consistent with the established function of Herp in restricting stress-induced calcium flow, a function that requires ER localization. Silencing of POSH expression by RNA interference also stimulates calcium release in the presence of Tpg () and, thus, further confirmed the critical function of POSH in restricting cytosolic calcium upon the induction of ER stress. Comparing the kinetics of Herp ER mobilization and restriction of free calcium upon Tpg treatment presents a major discrepancy: although the relocalization of Herp is observed after 15 min (), Tpg-induced calcium release is much faster and occurs within a few minutes (). How does Herp control calcium release from the ER if calcium release precedes translocation? This apparent paradox can be resolved if one considers the rapid action of Tpg, which elevates intracellular calcium within minutes (). Therefore, the initial calcium concentration upon the addition of Tpg is most likely determined by the basal levels of Herp at the ER membrane rather than by de novo activation of Herp translocation. Constitutive POSH activity may release small amounts of Herp to the ER. Herp is short lived (; ) and, thus, does not accumulate in the ER. However, the basal levels might be sufficient to attenuate initial stress-induced calcium leakage. According to this model, substantial mobilization of Herp to the ER and further restriction of free calcium flow is only facilitated when POSH is activated and allows the recruitment of additional Herp. The result of the in vivo Herp ubiquitination experiment that demonstrated basal POSH-dependent Herp ubiquitination that was further enhanced by Tpg () is consistent with this model. Because POSH activity is required for Herp ER localization, the relative levels of Herp at the ER membrane under basal conditions is determined by the relative POSH activity: it may be completely inhibited by the overexpression of POSH, partially inhibited by RNA interference (), and obviously elevated by the overexpression of native POSH. Consequently, the Tpg-induced calcium burst is instantaneous in POSH-overexpressing cells, accelerated in POSH-depleted cells, and markedly attenuated in the wild-type POSH- overexpressing cells (). Based on the results presented in this study, we propose the following model for the regulation of Herp ER mobilization (): POSH constitutively polyubiquitinates Herp, releasing small amounts of Herp to the ER. Elevation of intracellular calcium activates the ubiquitination activity of the POSH–Herp complex at the TGN. As a result, Herp polyubiquitination is enhanced, resulting in the activation of Herp mobilization to the ER membrane. At the ER membrane, Herp performs ER stress–resistance functions, one of which is the restriction of further calcium release, probably through an interaction with a calcium channel. A candidate for a Herp-regulated calcium channel is presenilin, with which Herp interacts (, ). In agreement with this hypothesis, presenilins were recently shown to form calcium leak channels mediating the bulk of the passive ER to cytosol calcium flow (). The assumption that the elevation of cytosolic calcium directly triggers the activation of Herp relocalization is based on the findings that the relocalization of Herp is rapidly induced by treatment of cells with Tpg and is independent of new protein synthesis. These findings also exclude a role for the ER stress response–induced transcriptional program. Independence of the ER stress response is further supported by our kinetic analysis, indicating that relocalization of Herp occurs well before any elevation in Herp protein level as part of the ER stress response (; ). In addition, the observed mobilization of exogenous Herp () cannot result from ER stress response transcriptional up-regulation, as it is under the control of a cytomegalovirus promoter, which does not express an ER stress response element. It has been reported that in yeast, accumulation of unfolded proteins in the ER stimulates calcium influx through the plasma membrane (). Similarly, unfolded proteins may activate calcium influx in mammalian cells, which would explain our observation that treatment with Tm, which causes unfolded protein accumulation, also rapidly induces Herp redistribution (Fig. S2). The proposed model for the regulation of Herp traffic requires a mechanism that couples the transient increase in cytosolic calcium with the activation of POSH. One possibility is that Herp is activated by calcium and, in turn, activates POSH. This model is supported by our findings that Herp is a POSH activator. Nevertheless, we have thus far not found evidence that calcium stimulates the association of Herp with POSH or further stimulates POSH activity in the presence of Herp in vitro. This is not unexpected, and, because neither POSH nor Herp express a known calcium-sensing domain, calcium regulation likely involves a calcium-sensing molecule. Evidence for a possible calcium sensor was recently provided by the findings that in , POSH interacts directly in a calcium-dependent fashion with the calcium-binding protein ALG2 (). Whether ALG2 is the molecule that confers calcium regulation on POSH or whether it is mediated by another molecule that activates POSH (either directly or through Herp) such as a TGN resident calcium channel is currently under investigation. In conclusion, we provide evidence for a novel spatial regulation of Herp by a calcium-activated Ub-mediated mechanism. Therefore, we propose that recruitment of Herp to the ER through POSH-mediated ubiquitination plays an essential role in the subsequent resistance of cells to ER stress. pCMV-Herp-FLAG was constructed by adding a C-terminal FLAG tag to IMAGE clone 5575914 (IMAGE Consortium; obtained from the UK Human Genome Mapping Project Resource Center). pCMV-POSH-V5 was described previously (). The siRNA duplex used to knock down Herp (5′-AAGGGAAGUUCUUCGGAACCUdTdT-3′ and 5′-AGGUUCCGAAGAACUUCCCdTdT-3′) was synthesized by Dharmacon. The H310 and H314 cell lines that stably express POSH-specific and control small hairpin RNA (shRNA) are matching clones of the previously described H153 and H187 cell lines, respectively (). Anti-Herp 25B was produced in rabbits immunized with affinity-purified Herp (tHerp). PT1 mouse anti-POSH was produced in mice immunized with affinity-purified POSH, and rabbit anti-POSH was previously described (). Both protein antigens were produced in bacteria as GST fusions and were used for immunization after removal of the GST portion. The following primary antibodies were used for immunofluorescence: anticalnexin (Santa Cruz Biotechnology, Inc.), anti-V5 (Invitrogen), anti-Flag (Sigma-Aldrich), and sheep anti-TGN46 (Serotec). Secondary Cy-conjugated antibodies were purchased from Jackson ImmunoResearch Laboratories. POSH was produced in bacteria as an MBP fusion protein (molecular mass of the fusion protein is ∼150 kD; ). tHerp and tHerp were produced by PCR amplification of Herp codons 1–272 (tHerp) or codons 85–272 (tHerp) followed by a FLAG tag cloned into pGEX-6P-2 (GE Healthcare). tHerp was constructed by two-step PCR mutagenesis mutating all of the lysine residues within the Ubl domain (positions 15, 38, 61, 75, and 78) to arginines. Proteins were expressed in BL21 by IPTG induction. Proteins were purified by glutathione-agarose affinity chromatography followed by removal of GST by PreScission protease (Ge Healthcare) and Q-Sepharose chromatography. A yeast two-hybrid screen was performed using Matchmaker system 3 (CLONTECH Laboratories, Inc.). Bait plasmid was constructed by cloning hPOSH (amino acids 53–888) into pGBK-T7. Yeast AH109 cells containing pGBK-hPOSH were mated with Y187 yeast cells containing a pretransformed HeLa cDNA library (CLONTECH Laboratories, Inc.). Colonies were selected on defined media lacking tryptophan, leucine, and histidine and containing 2 mM 3-aminotriazol. Colonies that grew on the selective media were tested for β-galactosidase activity, and positive clones were rescued from yeast that was sequenced and reintroduced into Y187 to confirm interaction with bait plasmid. One of the clones was identified as a novel splice variant of Herp containing the first 250 amino acids of Herp and a unique C-terminal region (GenBank/EMBL/DDBJ accession no. ). Fluorescence microscopy was performed as previously described (). In brief, the imaging medium in all cases was PBS. Images were taken in a confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.) with 40× NA 1.6 objective lenses using the LSM 510 acquisition software (Carl Zeiss MicroImaging, Inc.). Fluorochromes used in this study and their respective emission wavelengths are as follows: Cy2 and Fluo-4, 488 nm; Cy3, 588 nm; and Cy5, 633 nm. For colocalization experiments, 10 optical horizontal sections with intervals of 1 μm were taken through each preparation (z stack). A single median section is shown in (C and D), 4 A, and 5 B. Cells were lysed in buffer A containing 50 mM Hepes-NaOH, pH 7.5, 150 mM NaCl, 10% glycerol, 1% Triton X-100, 1 mM EDTA, 1 mM EGTA, 1.5 mM MgCl, 0.5 mM DTT, and protease inhibitor cocktail (1:100; Sigma-Aldrich). Immunoprecipitation was performed using protein A–Sepharose beads coated with the antibodies indicated in the figures. The detergent extracts were incubated at 4°C for 2 h. After precipitation, the beads were washed with buffer A containing 0.1% Triton X-100 and without MgCl. Bound proteins were eluted with SDS sample buffer and subjected to Western blot analysis by standard procedures. Immunoreactive proteins were visualized by ECL. 50 nM of purified recombinant E1, 0.3 μM UbcH5c or Ubc13/Uev1, 13 nM of bacterially expressed POSH, and Herp (where indicated) were incubated in a final volume of 20 μl containing 40 mM Tris-HCl, pH 7.5, 1 mM DTT, 2 mM ATP, 5 mM MgCl, 5 × 10% (vol/vol) Tween 20, and 5 μg Ub. After incubation for 30 min at 30°C, reactions were resolved by 7.5% SDS-PAGE and subjected to immunoblot analysis with PT1 and anti-Flag for the detection of POSH and Herp ubiquitination, respectively. 7 pmol POSH and 10 pmol tHerp were incubated either alone or together in a final volume of 50 μl containing 40 mM Tris-HCl, pH 7.5, 0.5 mM DTT, and 5 × 10% (vol/vol) Tween 20. After incubation for 30 min at 30°C, the mixture was applied to a 10–50% glycerol gradient (2 ml) in the same buffer. Gradients were centrifuged at 250,000 for 3 h at 4°C in a 55Ti rotor (Beckman Coulter). Fractions of 200 μl were collected, and aliquots were analyzed by immunoblotting or POSH ubiquitination assays. For measurement of POSH self-ubiquitination activity in the glycerol density gradient fractions (), we used the homogeneous time-resolved fluorescence method (). In solution, energy transfer between a donor fluorophore coupled to Ub and an acceptor fluorophore coupled to anti-POSH occurs only if Ub is conjugated to POSH. Consequently, aliquots of the glycerol gradient fractions were incubated with 3 nM E1 and 20 nM Ubc13/Uev1a and a mixture of 90 nM Ub and 10 nM Eu-cryptate-Ub (Cisbio) in an in vitro ubiquitination reaction as described in Ub conjugation assays. After incubation at 30°C for 30 min, Herp was immunoprecipitated with anti-Flag (Herp) in radioimmunoprecipitation buffer containing 1% SDS for 2 h at 4°C. The mixtures were then briefly centrifuged, and a 10-μl aliquot from the supernatant was removed and further incubated for 2 h at room temperature in a 96-well plate with XL665-PT1 (anti-POSH; Cisbio). Emission at 665 nm was subsequently determined in a homogeneous time-resolved fluorescence reader (RUBYstar; BMG Labtech) and expressed as δ F (function of the 665/620-nm emission ratio). Complete removal of Herp-Ub conjugates was confirmed in a parallel assay that determined the lack of energy transfer between Eu-cryptate-Ub and XL665-conjugated anti-Flag. Cells were grown on glass-bottom microwell dishes coated with poly- -lysine (MatTek). Intracellular calcium was measured by determining emission at 533 nm using the Fluo-4 Calcium Assay kit (Invitrogen) according to the manufacturer's instructions. The cell culture was visualized in a confocal microscope (LSM 510; Carl Zeiss MicroImaging, Inc.), and an image was taken at individual time points (Fig. S3). Calcium concentration was subsequently determined as the intensity of Fluo-4 emission by the ImageJ program (National Institutes of Health; ) using the mean gray value measurement option and is expressed as arbitrary emission units. Fig. S1 shows the localization of Herp in cells immunostained in the presence of a mild detergent. Fig. S2 shows the time course of Herp protein induction and Herp redistribution in the presence of Tm. Fig. S3 shows that calcium induced the redistribution of Herp in PC3 and SW480 cell lines. Fig. S4 shows calcium imaging of the time course of calcium release in cells expressing POSH and dominant-negative POSH. Online supplemental material is available at .

The androgen receptor (AR) is a ligand-dependent transcription factor of the steroid receptor (SR) subfamily of nuclear receptors. ARs regulate expression of genes involved in development and maintenance of the male phenotype and play a role in the growth of prostate cancer. Like all SRs, AR is composed of a central DNA binding domain (DBD), a C-terminal ligand binding domain (LBD), and an N-terminal transactivation domain (NTD; ). In the absence of androgens, ARs are mainly located in the cytoplasm. Upon ligand binding, ARs rapidly translocate to the nucleus, where they bind to androgen response elements (AREs) in the promoters/enhancers of target genes and recruit transcriptional coregulators (; ; ). Many coregulators, like the p160 family, bind via LxxLL motifs to a hydrophobic cleft in the LBD of SRs formed by ligand-induced repositioning of the C-terminal α-helix. The AR differs from the other SRs in that its LBD preferentially interacts with cofactors containing FxxLF rather than LxxLL motifs (; ). In addition, an extra level of regulation of AR function is provided by an FQNLF motif in its NTD, which is able to interact with the liganded C-terminal LBD (N/C interaction; ; ). A well-recognized function of N/C interaction is stabilization of ligand binding (; ). In addition, it has been hypothesized that N/C interactions might block unfavorable protein–protein interactions. Confocal microscopy of GFP-tagged proteins, as well as quantitative assays such as FRAP and fluorescence resonance energy transfer (FRET), have been instrumental in the investigation of the behavior of SRs in living cells (; ; ; ; , ; ; ; ; ). Like many other nuclear factors interacting with DNA, SRs, including the AR, were shown to be highly mobile in the living cell nucleus and dynamically interact with specific binding sites (; ; , ; ; ). We have previously shown, using FRAP analysis based on computer modeling, that agonist-bound ARs are largely mobile in the nucleus and only transiently bind to immobile elements in the nucleus. This transient immobilization was most likely due to DNA binding, as several non–DNA binding mutants were freely mobile and did not show a detectable immobile fraction (, ). In addition, a recent elegant study using ARs double tagged at the N and C termini with the FRET couple CFP and YFP, respectively, has revealed that N/C interactions are initiated promptly after the addition of hormone, before transport to the nucleus (). However, questions regarding the spatiotemporal organization of AR in the nuclei of live cells remain unanswered: when, where, and in what order do interactions with coregulators and N/C interaction take place once an AR has entered the nucleus? Does proper regulation of AR function require compartmentalization of such interactions? In this study, we applied innovative combined FRAP and FRET methodology, and ratio imaging, using CFP and YFP tagging of wild-type ARs and AR mutants, to investigate the spatiotemporal regulation of AR N/C interactions and AR coregulator interactions in living cells. We tagged the fluorescent proteins YFP and CFP to the N and C termini of wild-type AR (YFP-AR-CFP) and to two mutant ARs: an N/C interaction–deficient mutant in which the N-terminal FQNLF motif is changed into an AQNAA motif (AR[F23,27A/L26A]), and the non–DNA binding mutant carrying a point mutation in the DBD, leading to the inability of this mutant to bind to androgen-regulated promoters (AR[A573D]; ). Western blot analysis showed that the expressed fusion proteins were all of the expected size (). In addition, several lines of evidence show that the double tag does not abolish AR function: the wild-type YFP-AR-CFP was able to induce expression of a luciferase reporter gene driven by an androgen-regulated promoter (at ∼35% of the activity of the untagged AR), whereas the DBD mutant YFP-AR(A573D)-CFP was not (). Importantly, although the transcription activation of double-tagged ARs was lower than that of untagged ARs, the presence of the F23,27A/L26A mutations reduced the activity of both double-tagged and untagged AR to the same extent (∼60% reduction), showing that the transcriptional activity of double-tagged ARs is sufficient to investigate its behavior (). Furthermore, the fusion proteins were mainly cytoplasmic in the absence of androgens and, after the addition of the agonistic ligand R1881, translocated to the nucleus at normal rate (; unpublished data). In the nucleus, the typical punctate nuclear distribution patterns were observed for the double-tagged wild-type AR and the double-tagged AR(F23,27A/L26A) mutant, whereas the inactive non–DNA binding mutant YFP-AR(A573D)-CFP displayed the typical homogeneous distribution pattern described previously (; ). In summary, these data show that double tagging the AR and AR mutants did not interfere with their native behavior. We then investigated whether the double-tagged YFP-AR-CFP provided a bonafide tool to study N/C interaction by FRET. The FRET readout system applied was based on photobleaching of the acceptor and measuring the subsequential increase of the donor (abFRET; ; ; ; Fig. S2 A, available at ). In the presence of R1881, cells with a low expression (Fig. S1) of either the wild-type YFP-AR-CFP or the non–DNA binding mutant YFP-AR(A573D)-CFP showed a considerable increase in CFP fluorescence after acceptor bleaching, whereas only a small increase was observed in the N/C interaction–deficient mutant YFP-AR(F23,27A/L26A)-CFP (). In addition, abFRET was not observed in the absence of agonistic ligand (Fig. S2 B). These data indicate that the measured abFRET represents interaction of the FQNLF motif in the AR NTD with the ligand induced groove in the LBD. This was further corroborated by in vitro spectroscopy showing that FRET was strongly reduced by the addition of FQNLF peptide motifs, which compete with the AR N terminus for interaction with the C-terminal LBD, in lysates of cells expressing YFP-AR-CFP (). This reduction in FRET signal was not observed when, instead of FQNLF motifs, noncompeting LQNLL peptide motifs were added to the lysates (), confirming that the observed FRET is due to N/C interaction. Finally, extending previous data (), confocal time-lapse microscopy of living cells stably expressing YFP-AR-CFP showed that the YFP/CFP ratio considerably increased immediately after the addition of hormone, followed by efficient translocation to the nucleus (). In contrast, the N/C interaction–deficient mutant YFP-AR(F23,27A/L26A)-CFP showed only a small increase in YFP/CFP ratio (Fig. S3). Based on these data, it can be concluded that the FRET measured in the double-tagged YFP-AR-CFP represents N/C interaction. We developed a method based on simultaneous measurement of FRAP and FRET to study the mobility of interacting molecules. In this method, FRET-donor (CFP) and FRET-acceptor (YFP) fluorescence are simultaneously measured at regular time intervals after irreversibly photobleaching the acceptor in a defined subregion of the nucleus. Donor fluorescence increase after acceptor photobleaching and subsequent decrease because of diffusion (donor-FRAP) reflects the mobility of only the interacting molecules (). In contrast, acceptor fluorescence redistribution after acceptor bleaching (acceptor-FRAP) reveals the mobility of the total pool of both interacting and noninteracting molecules, similar to a conventional FRAP experiment (; ). Importantly, comparison of donor-FRAP and acceptor-FRAP curves allows us to distinguish the mobility (and immobilization) of the subpopulations of interacting and noninteracting proteins. First, the method was validated in Hep3B cells expressing either a CFP-YFP fusion protein or separate CFPs and YFPs (). In brief, a narrow strip spanning the nucleus was scanned at 458 nm excitation with short intervals (100 ms) at low laser power (YFP is sufficiently excited at this wavelength; Fig. S4 A, available at ). Fluorescence intensities of the donor (CFP) and acceptor (YFP) were recorded simultaneously. After 40 scans, a high-intensity, 100-ms bleach pulse at 514 nm was applied to specifically photobleach YFPs inside the strip (CFP was not bleached by the bleach pulse; Fig. S4 B). Subsequently, scanning of the bleached strip was continued at 458 nm at low laser intensity. Acceptor (YFP) fluorescence in the strip was considerably reduced after bleaching and recovered at a velocity expected () for molecules the size of the fusion proteins (). In parallel, donor fluorescence in the bleached strip increased immediately after acceptor bleaching and decreased at a similar rate compared with the increase of YFP fluorescence (). The observed CFP increase and subsequent decrease was not due to an artifact of YFP or CFP fluorescent properties, as cotransfected separate YFPs and CFPs, as well as ARs tagged with YFP or CFP only, did not show a donor-FRAP signal ( and Fig. S4). We performed simultaneous FRAP and FRET experiments to investigate the AR N/C interaction. As a control experiment, we tested an AR tagged at the N terminus with the CFP-YFP fusion protein. FRET will occur in these fusion proteins independent of the N/C interaction, as CFP and YFP are always in proximity. Donor-FRAP and acceptor-FRAP of CFP-YFP-AR both showed the same redistribution kinetics (), which are slower than that of the CFP-YFP fusion alone () because of transient binding to DNA of wild-type ARs (, ; ). In sharp contrast, donor-FRAP of the two-sided double-tagged YFP-AR-CFP (representing solely the mobility of N/C-interacting ARs) was considerably faster than the corresponding acceptor-FRAP (representing the mobility of the total AR pool; ). The difference between donor-FRAP and acceptor-FRAP was not observed for the double- tagged non–DNA binding AR mutant (YFP-AR[A573D]-CFP; ). Moreover, the YFP-AR-CFP donor-FRAP curve () showed fast kinetics similar to both donor-FRAP and acceptor-FRAP curves of the non–DNA binding AR mutant (). These data strongly suggest that N/C interactions of the wild-type AR occur mainly in the mobile pool and are abolished when ARs are transiently immobilized in a DNA binding–dependent fashion. To further explore the observation that N/C interaction is reduced when ARs are transiently immobilized, we determined the spatial distribution of N/C-interacting and non–N/C-interacting ARs by high-resolution confocal ratio imaging of YFP-AR-CFP. Because YFP and CFP are present in the same quantity in cells expressing YFP-AR-CFP protein, these can be analyzed by straightforward ratio imaging. In brief, ratio images of cells expressing YFP-AR-CFP, CFP-YFP-AR, and the non–DNA binding YFP-AR(A537D)-CFP were obtained by calculating for each pixel the ratio between the YFP and CFP emission intensity. Subsequently, the nuclei were divided into three areas based on the mean fluorescence intensity of the entire nuclear area and corresponding standard deviation. In YFP-AR-CFP images, pixels with intensities higher than the mean plus two times the standard deviation (4.1% of total area; , red bars) coincided largely with the area that is usually referred to as a speckled or focal pattern, whereas pixels with lower intensities coincided largely with the region outside the speckled pattern (, A and B; for image analysis, see Materials and methods). The mean YFP/CFP ratio in each region was then calculated and expressed relative to the mean ratio in corresponding regions in CFP-YFP-AR with a similar intensity (see Materials and methods). Cells expressing CFP-YFP-AR provide an ideal control to correct for potential imaging artifacts because the ratio should be independent of AR folding and absolute fluorescence intensity. The wild-type YFP-AR-CFP showed a significantly reduced YFP/CFP ratio in the speckles compared with the region outside the speckles (; P = 0.0002; see Materials and methods), whereas no correlation is found for the non–DNA binding YFP-AR(A537D)-CFP, which showed a homogeneous distribution (). Apparently, the concentration of non–N/C-interacting ARs is highest inside speckles. The results described in the previous paragraphs, suggesting that N/C interactions are abolished when AR is immobilized because of DNA binding and that N/C interactions are decreased inside speckles, prompted us to investigate whether the AR speckled pattern is correlated to the distribution of sites of active transcription. Previously, it was shown, using 5-bromo-uridine-5′-triphosphate (BrUTP) incorporation in nascent RNA and immunofluorescence (; ), that progesterone receptor (), glucocorticoid receptors (), and several other transcription factors (BRG1, TFIIH, Oct1, and E2F-1; ) do not show a complete, but rather a partial, overlap with active sites of transcription (nascent RNA). Using the same approach (see Materials and methods), we were able to detect sites of transcription in Hep3B cells stably expressing GFP-AR at physiological levels (). Newly incorporated BrUTP was detected by immunofluorescence using Cy3, which is excited at 543 nm excitation, and GFP-AR was detected by 488 nm excitation. 60 dual channel images were recorded at a configuration at which no cross talk occurred ().Interestingly, visual analysis showed only a partial overlap between the AR speckles and sites of active transcription (, C [right] and D [closed vs. open arrows]). We quantified this observation by image analysis in which AR speckles and areas of active transcription were identified based on the mean fluorescence intensity of the entire nuclear area and corresponding standard deviation. Similar to the ratio imaging analysis (), where we used the same procedure to identify AR speckles (see the previous paragraph), pixels in the GFP-AR image with intensities higher than the mean plus two times the standard deviation coincided largely with AR speckles (, left and right). In the Cy3-labeled BrUTP image, pixels with intensities higher than the mean plus two times the standard deviation were defined to be hot spots of transcription (, middle and right). The centers of, on average, 110 AR speckles and 130 hot spots of transcription per nucleus were then determined. Subsequently, the distances between each AR speckle and the closest hot spot of transcription were determined and compared with a randomly distributed set consisting of an equal number of spots with the same size distribution as the measured hot spots of transcription, taking care that the random spots were not in the nucleoli or outside the nucleus. The number of AR speckles at relatively short distance (<350 nm; , first five columns) to the nearest BrUTP spot was significantly higher compared with what is expected on the basis of a random distribution (43 ± 5 measured vs. 24 ± 2 random spots; P = 0.00025; ). Moreover, the largest relative difference between measured and random was highest at the closest detectable distance. In addition, the number of AR speckles that showed overlap with the nearest hot spot of transcription was significantly higher than expected when there would be no correlation between AR and nascent RNA distributions (P = 5.0 × 10; ). The strongly reduced N/C interaction in the transient immobile AR fraction led us to hypothesize that AR coregulators containing FxxLF motifs may gain access more easily to this fraction, as no competition with the N-terminal AR FQNLF motif is expected to occur. We tested this hypothesis using YFP-tagged fragments of the cofactor ARA54, containing an FNRLF motif. ARA54 and ARA54 fragments containing the FNRLF motif were previously shown to display a strong interaction with the AR LBD (; ; ). In agreement with this hypothesis, abFRET between the single-tagged wild-type AR-CFP and YFP-ARA54 fragments was significantly higher (P = 0.003; see Materials and methods) than that of the non–DNA binding mutant AR(A573D)-CFP with YFP-ARA54, suggesting that interactions between AR and ARA54 fragments are significantly enhanced when ARs are bound to DNA (, left). To further test the hypothesis that AR N/C interactions are responsible for blocking coregulator interactions, we performed the same experiment using the N/C interaction–deficient mutant AR(F23,27A/L26A)-CFP. In contrast to wild-type AR, no difference in FRET with the ARA54 fragments was observed for the mutant and its non–DNA binding variant AR(F23,27A/L26A/A573D)-CFP. Moreover, FRET was higher than that of the N/C interaction–proficient wild-type ARs (P = 0.042) and much higher than the N/C interaction–proficient non–DNA binding mutant (, left). No FRET was found between any of the AR mutants and free YFP (, right). These data are in agreement with a model in which YFP-ARA54 fragments bind preferentially to ARs lacking N/C interaction, i.e., either N/C interaction–deficient AR(F23,27A/L26A) mutants or wild-type ARs transiently immobilized as a result of DNA binding. To investigate this more extensively, we repeated the simultaneous FRAP and FRET measurements in living Hep3B cells expressing YFP-AR-CFP, now in the presence of cotransfected YFP-ARA54 fragments. The addition of YFP-ARA54 fragments considerably reduced the kinetics of the donor-FRAP curve compared with YFP-AR-CFP in absence of YFP-ARA54 fragments (). This is explained by the fact that in this experimental setup not only the N/C-interacting mobile ARs, but also the non–N/C-interacting immobile ARs, show FRET, now between AR C-terminal domain and the YFP-ARA54 fragments (which binds to the C-terminal domain instead of the YFP-tagged N-terminal domain of immobile YFP-AR-CFP). This indicates that the ARA54 fragments preferably interact with the C terminus of the AR when it is transiently immobilized because of DNA binding when the N-terminal FQNLF motif does not compete for interaction with the C-terminal domain. In summary, the abFRET data () show that ARA54 fragments interact more frequently with wild-type AR than with the non–DNA binding mutant. The simultaneous FRAP and FRET analysis () suggests that this is because ARA54 fragments gain access more easily to the C-terminal LBD of the wild-type ARs when there is no, or less, competition with the NTD. This occurs either when wild-type ARs are transiently immobilized in a DNA binding–dependent manner () or when the N/C interaction is disrupted (). Activity of SRs is not only regulated by ligand binding but also by interacting cofactors. The best-described binding site for SR coregulators is the hydrophobic cleft in the LBD to which LxxLL motifs can bind. The AR LBD is unique in its preference for the interaction with cofactors carrying FxxLF motifs rather than LxxLL motifs (; ). The AR itself also contains an FQNLF motif in the N-terminal domain, enabling interaction with the LBD (N/C interaction; ; ). The potential competition between the AR N-terminal FQNLF motif and similar motifs in cofactors for interaction with the LBD raises questions regarding the role of the N/C interaction in orchestrating cofactor interactions. To study AR N/C interactions in living cells, we tagged the AR at the N and C termini with YFP and CFP, respectively, or with CFP alone, and applied FRET and simultaneous FRET and FRAP experiments. In addition, to investigate cofactor interactions, we tagged ARA54 fragments containing an FNRLF motif with YFP. The presence of the tags had no effect on AR localization and hormone-induced nuclear translocation () and only limited effect on the transactivation function of the AR (). Acceptor photobleaching FRET assays on living cells and in vitro competition experiments using FxxLF- and LxxLL-peptide motifs demonstrated that FRET represents N/C interaction (). Previously, using FRAP assays, we and others have shown that the mobility of ARs is reduced compared with the mobility of non–DNA binding AR(A573D) mutants (), as well as antagonist-bound ARs (). In addition, the observed hormone-induced slow down of AR mobility was always accompanied by the formation of a speckled distribution pattern in the nucleus, suggesting that ARs transiently immobilize in speckles. We have now shown using combined FRET and FRAP analysis that, surprisingly, the mobility of the pool of N/C-interacting ARs is not reduced in the presence of hormone and that, consequently, the pool of non–N/C-interacting ARs is responsible for the observed overall slow down of AR mobility. This suggests that the N/C interaction is largely lost when ARs are transiently immobilized, most likely because of DNA binding (). This was confirmed by high-resolution ratio imaging showing that FRET is reduced inside speckles (). The loss of N/C interaction in immobilized ARs suggests that the C-terminal hydrophobic groove, to which FxxLF motifs can bind, is optimally accessible for coregulators when the ARs are bound to DNA. Our acceptor bleaching FRET experiments on YFP-tagged FNRLF fragments of the AR cofactor ARA54 and AR-CFP provide evidence that strongly supports this view. First, the experiments indicate that ARA54 fragments interact more frequently with the wild-type AR than with the non–DNA-binding AR mutants (A573D), whereas the non–N/C- interacting mutants of DNA binding and non–DNA binding ARs do not show this difference and interact more frequently than any of the N/C interaction–proficient ARs (). Moreover, when YFP-tagged ARA54 fragments are coexpressed with YFP-AR-CFP in a simultaneous FRET and FRAP assay, the mobility of the N/C-interacting pool is reduced (). This indicates that on top of the mobile N/C-interacting ARs, the immobile double-tagged ARs now show FRET because of their interaction with the YFP-tagged ARA54 fragments. The observed loss of N/C interaction in immobile ARs and frequent interactions of cofactor fragments with immobile ARs are in line with a scenario in which the AR itself dynamically regulates the time and place of interactions with coregulators by blocking the groove using its N-terminal FQNLF motif when not associated to DNA and allowing access of coregulators only after DNA binding (). Because our data suggest that DNA binding occurs in speckles, the question arose whether these speckles also represent sites of active transcription. To investigate this, we performed BrUTP incorporation experiments on Hep3B cells stably expressing AR-GFP. Interestingly, visual as well as statistical analysis showed that although speckles are closer to sites of active transcription than expected on the basis of a random distribution, AR and transcription hot spots only partially overlap (), suggesting that DNA binding of the AR does not always result in the formation of productive transcription complexes. Several lines of previous evidence are in agreement with these observations. First, it has been shown that progesterone receptor (), glucocorticoid receptors (), and several other transcription factors (BRG1, TFIIH, Oct1, and E2F-1; ) showed only a partial correlation with active sites of transcription. Second, recent data on estrogen receptors using chromatin immunoprecipitation on chip assays suggested that SRs have many more binding sites (∼3,600) in the genome than expected on the basis of the estimated number of estrogen receptor–regulated genes, which probably is in the order of hundreds rather than thousands (). Third, it has been shown by chromatin immunoprecipitation that DNA binding of the estrogen receptor occurs in a cyclic pattern and that an initial cycle of binding only prepares promoters for transcription but does not result in a productive transcription complex (for review see ). However, these nonproductive cycles were observed in cells shortly after application of the hormone. It remains questionable whether after longer exposure to hormone, as used in our experiments, promoters would be “shut down” and reactivated. If not all immobile ARs are involved in active transcription, the question of what happens in speckles remains. It has frequently been suggested that many transcription factors, and other nuclear factors involved in DNA metabolism, bind transiently to DNA also at nonspecific sites, thereby scanning the DNA (; ). Possibly the majority of immobile ARs observed in our experiments are involved in such scanning activity. The interaction with cofactors may then play a role in identifying specific binding sites when encountered during scanning. In addition, it is not excluded that part of the speckles represents some sort of storage site. However, as non–DNA-binding mutants do not form speckles and move freely through the nucleus, such a model suggests that the DBD is also involved in storage. In conclusion, we have used a novel combination of FRAP and FRET to investigate interactions of the AR in living cells and provided evidence that AR N/C interactions are involved in the spatiotemporal regulation of interactions with coregulators. The FRET/FRAP assay provides a novel tool to separately investigate the dynamics of interacting and noninteracting molecules. This opens up a multitude of possibilities to investigate the molecular mechanisms underlying not only the regulation of gene transcription but also that of other DNA transacting systems, such as DNA repair and replication. The cDNA construct encoding N-terminally YFP-tagged AR was generated by replacing EGFP in pGFP-(GA)-AR () by EYFP-C1 (CLONTECH Laboratories, Inc.). The C-terminally CFP-tagged AR was generated by replacing EGFP by ECFP-N3 (CLONTECH Laboratories, Inc.) in pAR-(GA)-EGFP in which two AR fragments from pcDNA-AR0mcs (lacking the AR stop codon; ) and pAR0 (), respectively, were sequentially inserted in EGFP-N3 (CLONTECH Laboratories, Inc.) followed by the introduction of a spacer sequence coding for a (Gly-Ala) stretch. The construct coding for double-tagged AR (pYFP-[GA]-AR-[GA]-CFP) was generated by combining a fragment of N-terminally YFP-tagged AR pYFP-(GA)-AR with a fragment of C-terminally CFP-tagged AR pAR-(GA)-CFP. The F23,27A/L26A variants were generated by QuikChange (Stratagene) mutagenesis using primers 5′-ACCTACCGAGGAGCTGCACAGAATGCTGCACAGAGCGTGCGCGAA-3′ and 5′-TTCGCGCACGCTCTGTGCAGCATTCTGTGCAGCTCCTCGGTAGGT-3′. To generate the A573D variants, the AR DBDs of pYFP-AR-CFP and pAR-(GA)-CFP were replaced by a pGFP-AR(A573D) () fragment containing the AR DBD (A573D) mutation. EYFP in pYFP-(GA)-AR was replaced by an ECFP-EYFP fusion to obtain pCFP-YFP-(GA)6-AR. The YFP-tagged ARA54 peptide construct was obtained by annealing the primers 5′-GATCGACCCTGGTTCACCATGTTTTAACCGGCTGTTTTATGCTGTGGATGTTG-3′ and 5′-AATTCAACATCCACAGCATAAAACAGCCGGTTAAAACATGGTGAACCAGGGTC-3′ containing the FNRLF motif and inserting the fragment in pEYFP-C2 (CLONTECH Laboratories, Inc.). Structures of novel constructs were verified by appropriate restriction digestions and by sequencing. Sizes of expressed proteins were verified by Western blotting. pCYFP encoding the ECFP-EYFP fusion was provided by C. Gazin (Hôpital Saint-Louis, Paris, France). The (ARE)-TATA Luc reporter was a gift from G. Jenster (Josephine Nefkens Institute, Rotterdam, Netherlands). 2 d before microscopic analyses, Hep3B cells were grown on glass coverslips in 6-well plates in α-MEM (Cambrex) supplemented with 5% FBS (HyClone), 2 mM -glutamine, 100 U/ml penicillin, and 100 μg/ml streptomycin. At least 4 h before transfection, the medium was substituted by medium containing 5% dextran charcoal stripped FBS. Transfections were performed with 1 μg/well AR or CFP-YFP expression constructs or 0.5 μg/well empty vector in FuGENE6 (Roche) transfection medium. In the indicated experiments, YFP-tagged ARA54 peptide expression constructs (0.5 μg/well) were added. 4 h after transfection, the medium was replaced by medium with 5% dextran charcoal stripped FBS with or without 100 nM R1881. Hep3B cells stably expressing AR constructs were subjected to the same medium-replacement schedule. For the AR transactivation experiments, Hep3B cells were cultured in 24-well plates on α-MEM supplemented with 5% dextran charcoal stripped FBS in the presence or absence of 100 nM R1881 and transfected using 50 ng AR expression construct and 100 ng (ARE)TATA Luc reporter. 24 h after transfection, cells were lysed and luciferase activity was measured in a luminometer (Fluoroscan Ascent FL; Labsystems Oy). Light emission was recorded during 5 s, after a delay of 2 s. Hep3B cells were cultured and transfected in 6-well plates. 24 h after transfection, cells were washed twice in ice-cold PBS and lysed in 200 μl Laemmli sample buffer (50 mM Tris-HCl, pH 6.8, 10% glycerol, 2% SDS, 10 mM DTT, and 0.001% Bromophenol blue). After boiling for 5 min, a 5-μl sample was separated on a 10% SDS-polyacrylamide gel and blotted to Nitrocellulose Transfer Membrane (Protran; Schleicher and Schuell). Blots were incubated with anti-AR (1:2,000; mouse monoclonal F34.4.1) or anti–β-actin (1:10,000; mouse monoclonal anti–β-actin [Sigma-Aldrich]) and subsequently incubated with HRP-conjugated goat anti–mouse antibody (DakoCytomation). Proteins were visualized using Super Signal West Pico Luminol solution (Pierce Chemical Co.), followed by exposure to x-ray film. Interactions between either the N- and C-terminal domain of the YFP-AR-CFP or between AR-CFP and YFP-ARA54 were assessed using acceptor photobleaching. For this, YFP and CFP images were collected sequentially before photobleaching of the acceptor. CFP was excited at 458 nm at moderate laser power, and emission was detected using a 470–500 nm bandpass emission filter. YFP was excited at 514 nm at moderate laser power, and emission was detected using a 560-nm longpass emission filter. After image collection, YFP in the nucleus was bleached by scanning a nuclear region of ∼100 μm 25 times at 514 nm at high laser power, covering the largest part of the nucleus. After photobleaching, a second YFP and CFP image pair was collected. Apparent FRET efficiency was estimated (correcting for the amount of YFP bleached) using the equation abFRET = ([ − ] × )/([ − ] × ), where and are the mean prebleach fluorescence intensities of CFP and YFP, respectively, in the area to be bleached (after background subtraction), and and are the mean postbleach fluorescence intensities of CFP and YFP, respectively, in the bleached area. The apparent FRET efficiency was finally expressed relative to control measurements in cells expressing either free CFP and YFP () or the CFP-YFP fusion protein (): apparent FRET efficiency = ( − )/( − ). For statistical analysis, the abFRET datasets were tested for normality using the Kolmogorov-Smirnov test, and datasets were compared using the one-tailed test. Spectroscopic analysis of crude cell lysates of cells expressing YFP-AR-CFP was performed on a fluorescence spectrophotometer (F-4500; Hitachi) by recording spectra at 425 nm excitation. The apparent FRET efficiency was calculated as the ratio of the emission intensities at 525 and 475 nm. Background fluorescence of lysates of cells not expressing YFP-AR-CFP prepared in the same way was negligible. Spectra were recorded of lysates in the absence and presence of 300 μM of synthesized peptides containing an FQNLF or LQNLL motif, respectively. To study the mobility of interacting proteins, a narrow strip spanning the nucleus was scanned at 458 nm excitation with short intervals (100 ms) at low laser power (YFP is sufficiently excited at this wavelength; Fig. S4 A). Fluorescence intensities of the donor (CFP) and acceptor (YFP) were recorded simultaneously using 470–500-nm bandpass and 560-nm longpass filters, respectively. After 40 scans, a high-intensity, 100-ms bleach pulse at 514 nm was applied to specifically photobleach YFPs inside the strip (CFP was not bleached by the bleach pulse; Fig. S4 B). Subsequently, scanning of the bleached strip was continued at 458 nm at low laser intensity. = ( − )/( − ) or to compare donor-FRAP and acceptor-FRAP curves by calculating = ( − )/( − ), where , , and are the fluorescent intensities before, immediately after, the bleach and after complete recovery, respectively, and is the background intensity. Because YFP and CFP are present in exactly the same quantity in cells expressing YFP-AR-CFP, ratio imaging can be applied to study the spatial distribution of ARs with and without N/C interaction. Local differences in YFP/CFP ratio within the nucleus of cells expressing YFP-AR-CFP will only be observed if the ratio between N/C-interacting ARs, showing a relatively high YFP/CFP ratio, and non–N/C-interacting ARs, showing a relatively low YFP/CFP ratio, are different. For high-resolution YFP/CFP ratio imaging, YFP and CFP were imaged simultaneously using a moderate excitation at 458 nm and a 470–500-nm bandpass emission filter for CFP and a 560-nm longpass emission filter for YFP. To reduce noise, eight times line averaging was used. Images were analyzed using the KS-400 image analysis package (Carl Zeiss MicroImaging, Inc.). and are the intensities of the YFP and CFP emission, respectively, and is the background intensity. and was calculated for each pixel as = ( + )/2. of each nucleus (termed μ in ) and the standard deviation, σ, were then calculated after (manual) selection of the nuclear area and exclusion of the nucleoli (). < μ + σ, μ + σ < < μ + 2σ and > μ + 2σ were then first calculated for CFP-YFP-AR expressing cells. Because these molecules emit at a fixed YFP/CFP ratio irrespective of their conformation or local concentration, any difference in ratio in the three selected areas is due to imaging artifacts. Indeed, CFP/YFP ratio increased in CFP-YFP-AR expressing cells with low intensity and decreased in cells with high intensities probably because of the nonlinearity of the detectors. Therefore, data obtained from each cell expressing YFP-AR-CFP and the non–DNA-binding mutant YFP-AR(A573D)-CFP were expressed relative to the mean ratio measured in corresponding areas in seven cells expressing CFP-YFP-AR with similar expression level. For statistical analysis, the YFP/CFP ratio imaging datasets were tested for normality using the Kolmogorov-Smirnov test, and datasets were compared using the test. Nascent RNA was detected by BrUTP incorporation in permeabilized living Hep3B cells stably expressing GFP-AR () according to . Cells were grown overnight on coverslips in medium containing 5% dextran charcoal stripped FBS in the presence of 100 nM R1881. The procedure of BrUTP incorporation has been previously described (). Cells were permeabilized in glycerol buffer (20 mM Tris HCl, 0.5 mM MgCl, 0.5 mM EGTA, 25% glycerol, and 1 mM PMSF) supplemented with 0.05% Triton X-100 and 10 U/ml RNAsin for 3 min. To allow BrUTP incorporation, permeabilized cells were incubated for 30 min at RT in synthesis buffer (100 nM Tris HCl, 5 nM MgCl, 0.5 mM EGTA, 200 mM KCl, 50% glycerol, 0.05 mM SAM, 20 U/ml RNAsin, and 0.5 mM PMSF) supplemented with 0.5 mM ATP, CTP, GTP, and BrUTP (or UTP as control; Sigma-Aldrich). Next, cells were fixed in 2% formaldehyde in PBS, incubated in 0.5% Triton X-100/PBS for 5 min and in 100 nM glycin/PBS for 10 min, each step followed by two PBS washes. After blocking with PBG (0.05% gelatin and 0.5% BSA in PBS), incorporated BrUTP was immunolabeled overnight with a rat anti-BrdU mAb (Seralab) diluted 1:500 in PBS at 4°C. After four washes with PBG, cells were incubated for 90 min at RT with biotin-conjugated sheep anti–rat IgG (Jackson ImmunoResearch Laboratories) 1:200 in PBS followed by four washes with PBG. The biotinylated antibody was then visualized with Cy3-conjugated streptavidin (Jackson ImmunoResearch Laboratories) 1:250 in PBS for 30 min at RT. After extensive washing with PBG and PBS, cells were embedded in Vectashield containing DAPI. Fig. S1 shows YFP-AR-CFP expression analysis of cells used in the acceptor photobleaching FRET experiments and in the simultaneous FRAP and FRET measurements. Fig. S2 presents the validation of FRET measurements by acceptor photobleaching (abFRET) and shows the hormone dependency of FRET measured in cells expressing YFP-AR-CFP. Fig. S3 shows the minimal YFP/CFP ratio change after the addition of R1881 in cells expressing YFP-AR(F23,27A/L26A)-CFP variant. Fig. S4 presents the control experiments for donor-FRAP and acceptor-FRAP on cells expressing YFP-AR and AR-CFP. Online supplemental material is available at .

We previously reported that the transplantation of muscle-derived stem cells (MDSCs) into diseased muscle results in a large number of regenerated myofibers (; ). These studies were initiated to understand the broad heterogeneity in phenotype and performance that has been reported for MDSCs and other muscle stem cell populations (; ; ; ; ; ; ; ). In our laboratory, we also have observed a high degree of variability in the ability of different populations of MDSCs to regenerate skeletal muscle. While investigating MDSC population heterogeneity and characteristics that define efficient in vivo skeletal muscle regeneration, we have found that cell sex, a rarely considered variable, has a considerable effect on in vivo outcome. Compared with the transplantation of male MDSCs (M-MDSCs), the transplantation of female MDSCs (F-MDSCs) leads to substantially more regeneration of the diseased skeletal muscle of mice, which model Duchenne muscular dystrophy. More than 2,000 yr ago, Aristotle speculated that sexual dimorphism existed at the earliest stage of embryonic growth; he believed that male embryos became animated 40 d after conception, whereas female animation required 90 d (Aristotle, 350 BC). Recent studies support this notion (; Xu et al., 1992; ; ;). Male embryos created by in vitro fertilization grow faster before implantation than female embryos, and these findings support a genetic cellular difference between the sexes that exists before the induction of hormonal stimulation (; ; ; ). Sex-related differences in the expression of genes during early embryo development have also been observed: females exhibit higher mRNA levels of and , which are two genes involved in the detoxification of reactive oxygen species (ROS; ; ). Male and female embryonic neurons (isolated from rats before gonad differentiation or hormonal stimulation) also display different cellular responses. The female cells are more sensitive to apoptosis-inducing agents, whereas male neurons are more sensitive to ischemia and nitrosative stress, and they cannot maintain the proper level of glutathione, which regulates ROS levels (). Few studies have investigated whether sex-related differences affect tissue or organ regeneration by progenitor cells. demonstrated that the regrowth of cartilage, skin, and hair follicles in an ear pinna wound occurred faster and more completely in female mice as compared with male mice. Another study has shown that female mice modeling living donor liver transplantation with ischemia and reperfusion exhibit more efficient liver regeneration than their male counterparts (). Similarly, male rats show more tissue growth than female rats after nephroctomy (). found that blocking antiapoptotic activity in male animals resulted in sex-related differences in the animals' survival after endotoxic stress; the males recovered, but there was no advantage for females. Finally, in a study of human hematopoietic stem cell transplantations, superior survival was observed with maternally donated recipients as compared with recipients of paternal transplantations (). Several theories have been designed to unify cell and tissue aging with the overall aging of organisms and, thereby, explain females' longer life spans. Telomeres, whose length is believed to act as the mitotic clock of cells, are shorter in older males than in older females, which suggests that male cells undergo more rounds of division than female cells (; ). This finding supports a possible link between growth rate and aging (). has proposed that cells that have finite replicative lifespans must undergo more rounds of division to build the larger bodies of male organisms. Other research has shown that estrogen stimulates telomerase, which slows the rate of telomere attrition (for review see ). This finding has ties with the free radical theory of cell aging, according to which ROS causes aging by damaging DNA, lipids, and proteins. Estrogens also up-regulate pathways that induce the expression of antioxidants (such as glutathione peroxidase) that reduce damage by ROS (for example, the MAPK pathway; ). Together, these findings suggest that sex-related differences in the health of the stem cell compartment could partially explain different rates of aging and disease. Stem and progenitor cells are believed to persist throughout life and contribute to the repair and maintenance of tissue. Therefore, investigations of sex-related differences shown by stem cells, as presented in this study, could lead to an improved understanding of sex-related differences in aging and disease. To determine whether any of the standard markers for MDSC characterization are predictive of high in vivo muscle regeneration, we examined 25 populations of MDSCs in terms of five variables: in vivo muscle regeneration efficiency, expression of CD34, expression of Sca-1, expression of desmin, and cell sex. Analysis of these variables revealed a large degree of heterogeneity in the MDSC populations. The distribution of the populations' regeneration indexes (RIs) and CD34, Sca-1, and desmin expression are shown in . The RI is a measure of how efficiently stem cells participate in skeletal muscle regeneration (; ), whereas CD34 and Sca-1 are stem cell markers, and desmin is a myogenic marker. We performed a Pearson product moment correlation analysis to identify significant relationships between any two variables and, in particular, to determine whether any of the markers correlate with high in vivo muscle regeneration. shows the correlation scatter matrix comprising scatter plots for all possible combinations of variables. Only cell sex correlated with the RI (P = 0.070); M-MDSC populations had a significantly lower RI on average than F-MDSC populations (P = 0.035; test). We also detected a significant correlation between cell sex and Sca-1 expression (P = 0.014): the mean expression of Sca-1 for all M-MDSC populations was significantly higher (M-MDSCs, 73 ± 16%; F-MDSCs, 52 ± 22%; mean ± SD; P = 0.001; test). We next separated the correlation matrices on the basis of sex; shows the scatter plots for M-MDSCs (10 populations are shown in each plot), and shows the scatter plots for F-MDSCs (15 populations are shown in each plot). The scatter plot of the RI versus Sca-1 for M-MDSCs indicates that a higher level of Sca-1 expression by M-MDSCs positively correlated with a higher RI (P = 0.021; ). This demonstrates that the significantly higher levels of Sca-1 expression by M-MDSCs (as compared with F-MDSCs) are not directly related to their low RIs (as compared with the RIs of F-MDSCs). The scatter plots of the F-MDSC populations did not reveal such a relationship (). Significant correlations between variables other than RIs are also indicated in , and correlation coefficients are listed in Table S1 (available at ). We focused our subsequent analysis on the sex-related differences exhibited by MDSCs because cell sex was the only variable that correlated with the RI. We used identical methods to isolate M- and F-MDSCs, and both exhibited stem cell characteristics in vitro ( and Fig. S1, available at ). Morphological comparison of the M- and F-MDSC populations revealed no significant differences in cell size or cell shape (Fig. S1, A and B). Analysis of the cells' multilineage differentiation potential showed that with appropriate stimulation in vitro, both male and female cells expressed markers of myogenic (), osteogenic (Fig. S1 C), and adipogenic lineages (Fig. S1 C), although they did so at different rates and to different extents. On average, there was no significant difference in the mean amount of fast myosin heavy chain (MyHC) in M- versus F-MDSCs (). Our analysis of short-term kinetics showed that M-MDSC and F-MDSC populations had similar population doubling (PD) times (PDTs; 13 ± 1.4 h and 14.2 ± 3.1 h, respectively), cell cycle times (12 ± 1.3 h and 12 ± 1.3 h, respectively), and mitotic fractions (0.91 ± 0.10 and 0.85 ± 0.08, respectively; P = 0.28; test) over a 3-d period (mean ± SD; ). However, after moderate expansion over 14 d or extended expansion over 3 mo (>150 PDs), M-MDSC populations exhibited higher growth rates than F-MDSC populations (M-MDSCs: PDT = ∼13 h and mitotic fraction = 0.87–0.97; F-MDSCs: PDT = ∼15 h and mitotic fraction = 0.81–0.87; and not depicted). After extended expansion, there was no significant difference in telomere length (Fig. S1 D) or telomerase activity (Fig. S1 E) as measured by flow cytometry. In addition, we observed a normal modal chromosome number for M- and F-MDSCs that were expanded to 90 PDs (unpublished data); this is consistent with our previous study of F-MDSCs (). After transplanting M- and F-MDSCs into the skeletal muscle of dystrophic mice (i.e., mice that lack dystrophin at the sarcolemma of muscle fibers), we investigated the cells' abilities to regenerate dystrophin-expressing myofibers. Sex-matched experiments involving several cell populations obtained from various isolations revealed that the implantation of F-MDSC populations led to significantly better skeletal muscle regeneration in vivo, as determined by calculating each population's RI (the ratio of dystrophin fibers per 10 donor cells; ). Female cell populations regenerated significantly more dystrophin-positive myofibers ( = 15 F-MDSC populations; two to six muscles per population; RI = 230 ± 52; mean ± SEM) as compared with male populations ( = 10 M-MDSC populations; two to six muscles per population; RI = 95 ± 20; P = 0.035). Representative images of dystrophin-expressing myofibers within the muscle tissue are shown in . Although the RIs of both M- and F-MDSC populations varied, no M-MDSCs exhibited high regeneration efficiency (i.e., RI > 203). In vitro immunostaining of myotubes after MDSC differentiation demonstrated that both M- and F-MDSCs are able to express dystrophin in myotubes containing a similar number of nuclei (). We performed both sex-matched and sex-mismatched transplantations to determine whether the sex of the host tissue might play a role in the cells' differing regeneration abilities. We used two F- and two M-MDSC populations that had similar expression profiles for CD34, Sca-1, and desmin and exhibited similar short-term proliferation characteristics. Regardless of host sex, M-MDSC transplantation resulted in a low RI (160 ± 75 in female hosts and 105 ± 25 in male hosts; ). Two-way analysis of variance (ANOVA) confirmed these findings and revealed both a significant difference as a result of host sex (P = 0.048) and a significant effect as a result of cell sex (P < 0.001). These results are consistent with those observed in previous studies using sorted populations of M-MDSCs (; ,) or using male hosts (). M-MDSCs had a significantly lower RI than F-MDSCs (P < 0.05), and there was a significantly lower RI with the male host as compared with the female host (P < 0.01). The difference caused by host sex suggests that the female microenvironment might be a more receptive environment for skeletal muscle regeneration by MDSCs and might play a role in RI variability. Hormonal differences between male and female hosts may influence the regeneration process; previous studies have described such differences in the responses of progenitor and stem cells to hormones (; ). We first tested the possibility that the low RI of M-MDSCs implanted in female hosts (as compared with female to female transplantations) might be caused by the immune rejection of the M-MDSCs as a result of H-Y antigenicity. We examined the immune response at the site of cell delivery by measuring the total cross-sectional area of CD4 expression after sex-matched and sex-crossed transplantations. We observed a larger area containing CD4-positive cells after sex-crossed transplantations (). We next investigated the effect of host sex in the absence of an immune response by performing sex-crossed and sex-matched transplantations in immune-compromised animals. M-MDSC to female /severe combined immunodeficiency (SCID) mouse transplantations again resulted in a low level of skeletal muscle regeneration (RI = 216 ± 39), which is similar to what we observed in the female host. We observed no significant difference as a result of host sex (P = 0.235), yet the significant difference as a result of cell sex remained (P = 0.018; two-way ANOVA; ). To determine whether estrogen stimulation improves the in vivo regeneration efficiency of MDSCs, we stimulated M- and F-MDSCs with two doses of 17-β-estradiol (10 or 100 nM; physiologic range of 10–100 nM; ). We confirmed the expression of estrogen receptors in M- and F-MDSCs by immunocytochemistry (Fig. S2 A, available at ) and by microarray analysis (not depicted). After 2 wk in vitro with estradiol, M-MDSCs displayed lower proliferation rates than unstimulated M-MDSCs (although the decrease was not significant). Consistent with other results, comparison of the proliferation rates of the unstimulated cells showed that F- MDSCs proliferated more slowly than unstimulated M-MDSCs at all time points after 7 d (9 d, P = 0.070; 12 and 14 d, P < 0.05; test; Fig. S2 B). We compared the in vivo regeneration efficiency of stimulated and unstimulated M- and F-MDSCs. 2 wk after transplantation, F-MDSCs cultured in control conditions of 0 nM estradiol had a higher RI than similarly cultured M-MDSCs (557 ± 229 vs. 115 ± 36; P = 0.06). Stimulation with estradiol had no significant effect on the RIs of M-MDSCs; however a trend toward a decreased RI was detected in F-MDSCs stimulated with 10 or 100 nM estradiol (P = 0.09 and P = 0.07; = 3; test; Fig. S2 C). Based on cell adhesion characteristics, the preplate technique or variations of this technique have been used to isolate cells from adult skeletal muscle tissue (; ). These cells have demonstrated multilineage differentiation potential ( ; ; ; ). To determine whether M-MDSCs are a rare subpopulation with particular adhesion or isolation characteristics, we attempted to isolate potent M- MDSCs from three other methods: other preplates or other subfractions of myogenic progenitors, a defined FACS subfraction expressing the stem cell marker CD34, and clonal populations. The preplate technique used to isolate MDSCs from a skeletal muscle biopsy involves serial preplating of low-adherence cells found within the primary medium supernatant to separate these cells from cells adhered to the collagenated surface of the flask (). Usually, myogenic precursors with characteristics of myoblasts or satellite cells are found within the early preplates (pp3–6), whereas cells isolated from the later preplates (≥pp6) have stem cell characteristics. We compared the in vivo regeneration efficiency (RI) of M-MDSCs with that of male myoblasts obtained from pp3–6 but identified no significant differences (). To determine the effect of the expression of CD34 on regeneration efficiency, we used FACS to purify M- and F-MDSCs and obtain CD34 fractions. The expression of Sca-1 and CD34 by unsorted M- and F-MDSCs was comparable (), and, in both populations, <15% of the cells were desmin positive. Our comparison of the RI of CD34-positive F-MDSCs (RI = 417 ± 56) with that of similarly sorted M-MDSCs (RI = 266 ± 61) revealed that the latter RI was significantly lower (P = 0.053; ). To test whether there are fewer cells within the M-MDSC populations that are capable of contributing to muscle regeneration, we performed in vitro and in vivo clonal analysis of myogenic behavior. We used FACS to single-cell sort and obtain clones to determine the percentage of clones with myogenic potential in the M- and F-MDSC populations. Because some populations already had a high level of expression of the myogenic marker desmin, we analyzed the clones from these populations separately. At 2 wk after single-cell cloning, there was no significant difference in the percentage of clones among male or female cells derived from MDSC populations (with high or low desmin expression), which yielded myogenic colonies as defined by myotube formation (P = 0.776; one-way ANOVA; ). When we compared all three factors (cell sex, desmin expression, and week of analysis), we detected a significant effect (P = 0.066, P = 0.003, and P = 0.756, respectively) as a result of the desmin level and week of myogenic analysis but no effect as a result of the cell sex (three-way ANOVA). We transplanted eight representative clonal populations of MDSCs (four male and four female clones) into animals to determine whether a clonal population could be identified within the M-MDSC populations that demonstrated a high RI. We confirmed that the overall trend we observe in (using 25 populations) is consistent with results obtained from clonal populations (). Although we did observe an F-MDSC clone that exhibited a lower RI as compared with its parent population (F3 vs. C parent population), we did not observe any male clone that exhibited a higher RI as compared with the parent population. We next performed microarray gene ontology analysis of M- and F-MDSCs to explore global gene expression differences between the populations. We identified the expression of more general cell stress–related genes, including oxidative stress and antiapoptotic genes, in F-MDSCs than in M-MDSCs () using both gene ontology based on the NetAffx annotation database and GeneSifter. Of 45 general cellular stress–related genes, 29% of the genes were significantly increased in females, 7% were significantly higher in males (P < 0.05), and 64% showed no significant difference. There were also sexual dimorphisms in genes that were involved in responses to hypoxia and oxidative stress between M- and F-MDSCs. Of 99 apoptosis-related genes, 23% of the genes were significantly increased in females, 4% were higher in males, and 72% showed no significant difference. We confirmed the sex-related differences in the expression levels of several of these genes using RT-PCR (). We also compared microarray data of F-MDSCs that yielded a low level of regeneration with F-MDSCs that yielded a high level of regeneration. We found two stress-related genes ( and ) that showed significant trends in a direction opposite to what we expected. For example, SOD1 was significantly higher in F-MDSCs as compared with M-MDSCs; however, when we explored the data of F-MDSCs with low RIs versus those with high RIs, we observed that SOD1 was higher in F-MDSCs that had only a low level of regeneration (P < 0.05). Similarly, TRP53Inp1 was significantly higher in the low-regenerating M-MDSCs, but, when we examined the low- versus high-regenerating F-MDSCs, we observed significantly higher levels of TRP53Inp1 in the female populations with the best in vivo regeneration (). Although these trends are interesting, this analysis did not provide any clear indicators for genes of importance. In particular, we found that the antiapoptotic factor was twofold higher in F-MDSCs than in M-MDSCs by microarray analysis (, arrow) and Western blotting (not depicted). To test whether the overexpression of in M-MDSCs could provide a gain of function in terms of in vivo skeletal muscle regeneration, we transfected the M-MDSCs with a plasmid. Western blot analysis and quantification showed higher levels of Bcl2 in M-MDSCs that were transfected to overexpress as compared with control M- or F-MDSCs (unpublished data). However, we did not observe a change in the RI of -engineered M-MDSCs as compared with M-MDSC controls (). We transplanted M- and F-MDSCs, which were transduced with a -encoding retroviral vector, into the gastrocnemius muscles of sex-matched mice, and we harvested at several time points between 16 h and 14 d. We quantified the number of -expressing nuclei that were detected at the transplantation site. At 16, 24, 48, and 72 h, there were significantly more nuclei detected in muscles of M-MDSC transplantations as compared with F-MDSC transplantations (P < 0.05; ). In separate experiments, we digested the skeletal muscle 24 and 48 h after the transplantation of M- and F-MDSCs, and we quantified the amount of gene by quantitative RT-PCR. Similar to the histological analysis, we found more in muscles transplanted with M-MDSCs as compared with F-MDSC transplantations, although this was not statistically significant (P > 0.05; ). In support of the finding of more M-MDSCs after transplantation, we detected higher levels of reduced glutathione, an antioxidant peptide whose levels are reported to be maximal in mitotic cells (). We used flow cytometry and monochlorobimane labeling to detect glutathione levels and found significantly more glutathione in M-MDSCs as compared with F-MDSCs (P < 0.01; ). We also examined the amount of donor cell fusion or differentiation at the engraftment site by quantifying the percentage of nuclei that were located within muscle fibers. We examined the transplantation sites at 1, 2, and 5 d after cell injection. At 5 d after transplantation, we observed a trend toward significantly more donor cell fusion or differentiation with M-MDSCs as compared with F-MDSC transplantation to muscle (P = 0.068; ). To evaluate the effect of low oxygen conditions and oxidative stress on M- and F-MDSCs, we examined MDSC viability, phenotype, and myogenic differentiation under these conditions. We hypothesize that after MDSC transplantation, the transplanted cells will experience conditions of low physiologic oxygen environment and oxidative stress. Previous studies have demonstrated that inflammation occurs at the local microenvironment after cell transplantation (; ; ; ) and that oxidative stress occurs after an inflammation response and in the presence of low oxygen (; ). After exposure to physiologic oxygen for 24 h (2.5% O), we observed a trend toward reduced proliferation in both M-MDSCs and F-MDSCs as compared with their controls (P = 0.18 and P = 0.079, respectively; ). There was no significant difference in the amount of cell viability or cell death in M-MDSCs as compared with F-MDSCs under conditions of low oxygen (P = 0.652; ). After exposure to oxidative stress (100 μM HO), we observed a significant decrease in cell numbers of both M-MDSCs and F-MDSCs (P = 0.039 and P = 0.001, respectively; ). However, when we compared the response of M- with F-MDSCs in terms of cell proliferation or total cell numbers after stress, we observed no difference between the populations; there was no difference in the percent change in cell viability for M- vs. F-MDSCs after oxidative stress (41 vs. 55% decrease; P = 0.325; ). We examined the expression of CD34, Sca-1, and desmin in the populations after exposure to cell stress. We did not observe a change in CD34 or Sca-1 expression in M- or F-MDSCs that were exposed to 2.5% O as compared with cells cultured in atmospheric O. We did observe a trend in increased desmin expression in M-MDSCs that had initially low levels of desmin (P = 0.065; ). There was no similar change in desmin expression in a comparable population of F-MDSCs. After treatment with hydrogen peroxide, we observed trends toward an increase in CD34 and Sca-1 in all populations; however, after oxidative stress, there was no significant difference in the phenotype of M-MDSCs as compared with F-MDSCs (). Next, we tested the ability of the cells to undergo myogenic differentiation after exposure to low oxygen or hydrogen peroxide. We plated M- and F-MDSCs at high density. After 4 d in culture, we transferred cells to either low oxygen or applied 100 μM H0 growth media. Control MDSCs were maintained at atmospheric oxygen with no H0. After 24 and 48 h, we performed immunochemistry for fast MyHC expression. Under control conditions, there was no significant difference in the percentage of nuclei that colocalized with MyHC in M- or F-MDSC populations at 24 or 48 h (27–45% of the nuclei colocalized with MyHC). We observed a trend toward increased myogenic differentiation in M-MDSCs as compared with F-MDSCs after 48 h in low oxygen (). The sex-related difference in the differentiation response was also observed after exposure to hydrogen peroxide for 24 (P = 0.060) and 48 h (). In comparison with F-MDSCs, we observed an increase in myogenic differentiation with M-MDSCs as compared with their untreated controls (). We show here that the sex of MDSCs influences their ability to promote skeletal muscle regeneration. These studies were initiated to understand the broad heterogeneity that has been reported for MDSCs and other muscle stem cell populations (; ; ; ; ; ; ; ). The use of different isolation techniques, culture techniques, and tools for stem cell characterization complicates cross comparisons among cell populations isolated in different research laboratories. This study identifies sex-related differences as a factor in MDSC variability in skeletal muscle regeneration. The M- and F-MDSC populations isolated by the preplate technique shared stem cell characteristics; however, extensive in vivo screening showed that only 2/10 male populations had an in vivo RI near 200. In comparison, 60% of the 15 female populations had an RI higher than the mean RI of M-MDSCs (RI = 95), and 40% of the F-MDSCs had an RI higher than the maximal male RI (RI = 203). After transplantation into the skeletal muscle of dystrophic mice, F-MDSCs transplanted into hosts of either sex consistently regenerated more dystrophin-positive myofibers than did M-MDSCs transplanted into hosts of either sex. A modified preplate technique was used to isolate a low-adhering fraction of MDSCs from skeletal muscle biopsies obtained from 3-wk-old normal C57BL mice (; ). Six isolations were performed using anatomically sexed animals, and cell sex was later confirmed by FISH analysis. All populations were negative for CD45 expression (<0.1% of cells). Both M- and F-MDSCs were cultured in normal growth medium (GM): DME (supplemented with 10% FBS; Invitrogen), 10% horse serum, 1% penicillin/streptomycin, and 0.5% chick embryo extract (Accurate Chemical). To examine multipotency, MDSCs were induced to undergo myogenic, osteogenic, and adipogenic differentiation. To promote myogenic differentiation, MDSCs were plated at high confluence (1,500 cells/cm) in GM for 3 d, transferred to low serum–containing medium (2% horse serum/FBS in DME) for 4 d, and examined for myotube formation by MyHC expression (1:250; Sigma-Aldrich), biotinylated IgG (1:250; Vector Laboratories), and streptavidin-Cy3 (1:500) or dystrophin immunocytochemistry. Osteogenic differentiation was induced by stimulation with 10 ng/ml of bone morphogenetic protein 4 for 10 d and was evaluated by AP and von Kossa staining (). Adipogenic differentiation was induced after cells reached confluency by adding adipogenic medium to the cells (dexamethasone, insulin, indomethacin, and 3-isobutyl-1-methyl-xanthinine; Cambrex) as previously described (). Lipid vacuole formation was assessed by Oil Red O staining. To examine short-term growth kinetics, the division time, PDT, and mitotic or dividing fraction were determined from time-lapsed microscopy as previously described (; ). To examine long-term proliferation potential, M- and F-MDSCs were plated in 25-cm collagen-coated flasks, and routine cell passaging was performed every 2–3 d (). At each passage, cells were replated to a density of 225 cells/cm. The number of PDs for each subculturing was calculated as the log (N/N). This process was repeated for >150 PDs. Karyotyping was performed as previously described () for the two M-MDSC populations (A and D) and two F-MDSC populations (B and C). Telomerase activity was determined using the TeloTAGGG Telomerase PCR ELISA PLUS kit (Roche) according to the manufacturer's protocol. 2 × 10 cells were lysed in 200 μl of lysis reagent, and 10 μl of the lysate was used for the telomere repeat amplification protocol reaction. The experiment was performed in triplicate, and relative telomerase activity was recorded and plotted. For the flow cytometry–based measurement of telomere length, telomeres in M- and F-MDSCs were detected with the PNA Kit for Flow Cytometry (DakoCytomation) according to the manufacturer's protocol. 2 × 10 male or female cells were divided into four 1.5-ml tubes. The cellular DNA was denatured for 10 min at 82°C and hybridized with the telomere PNA probe (FITC) at RT overnight. The cells were then analyzed by flow cytometry, and the relative telomere length was calculated. The use of animals and the surgical procedures performed in this study were approved by the Institutional Animal Care and Use Committee of the Children's Hospital of Pittsburgh (University of Pittsburgh Medical Center). mice (C57BL/10ScSn- were obtained from The Jackson Laboratory or bred at the institution's animal facility. /SCID mice were bred by crossing C57BL/10ScSn- and C57BL/6J-/SzJ mice. For all experiments involving the transplantation of MDSCs, satellite cells, or CD34-sorted populations, 1–2 × 10 cells were transplanted into the gastrocnemius muscles of male or female mice as indicated and harvested after 2 wk. For sex-mismatched experiments, two M-MDSCs (populations A and D) were transplanted into female hosts, and two F-MDSCs (populations B and C) were transplanted into male hosts. The populations had similar phenotypes: populations A and C (CD34 [60–80% positive], Sca-1 [70–100% positive], and desmin [>95% positive]), populations B and D (CD34 [60–80% positive] and Sca-1 [40–80% positive], and desmin [<15% positive]). For CD34-sorting experiments, populations A and C were used. A previously described technique (; ) was used to measure the cells' RIs. The RI is the ratio of the number of dystrophin-positive fibers per 10 donor cells (RI is scored at the cross section of maximal engraftment). The large majority of the >150 muscle transplantations of this study were quantitated for the RI in a blind manner by more than one investigator. The RIs or areas containing CD4-positive cells were compared by two-way ANOVA or a test to assess the effects of cell sex and host sex. For transplantations of MDSCs, cells were first labeled with retrovirus encoding for the gene as previously described (). We performed X-gal staining to visualize β-gal activity and quantified the total number of these -positive nuclei and the number that were located within muscle fibers. We used FACS to obtain single-cell clones (FACSAria and visual confirmation) of M- and F-MDSCs (populations A–D) on 96-well plates. Clones were cultured in 50 μl GM. At 1 wk after sorting, we identified viable colonies and scored these clones as myogenic or not myogenic based on the presence of distinct multinucleated myotubes. At 2 wk after cloning, we again determined the cumulative number of myogenic colonies based on the presence of distinct myotubes. For transplantation experiments using clonal populations, we similarly obtained single-cell clones from M- and F-MDSCs (populations A–D). After culture expansion, we selected eight representative populations and performed transplantations as described above (Cell transplantation to skeletal muscle) to determine the RI of each clone. M- and F-MDSCs (populations A–D) were cultured in GM without phenol red and supplemented with 0, 10, or 100 nM 17-β-estradiol (Sigma-Aldrich). Cells were replated in fresh medium every 2–3 d for 14 d. After in vitro stimulation, 10 cells per muscle were transplanted into the gastrocnemius of female /SCID mice ( = 3 muscles per group). 2 wk after transplantation, the muscles were harvested, sectioned, and immunostained for dystrophin, and the RI was analyzed as described above (Cell transplantation to skeletal muscle). RNA was isolated from five M- and five F-MDSC populations using an RNeasy kit (QIAGEN) and analyzed on a GeneChip Mouse Genome 430A 2.0 array (Affymetrix, Inc.). Fold increase was determined using previously established techniques (). Gene ontology was determined using the NetAffx (Affymetrix, Inc.) annotation database (; ). All results were confirmed using GeneSifter (VizX Labs), a certified GeneChip-compatible web-based analysis tool. The resulting data on fold increase for each population were analyzed using a test to determine significant differences (P < 0.05) between the M- and F-MDSC populations. Several genes of interest from the microarray were confirmed with RT-PCR analysis. Total RNA was extracted from 5 × 10 cells using the Nucleospin RNA kit (CLONTECH Laboratories, Inc.). cDNA was synthesized with SuperScript II reverse transcriptase (Invitrogen) according to the manufacturer's instructions. PCR was performed with Taq polymerase (Invitrogen) according to the manufacturer's instructions for 25, 28, and 30 cycles at 58°C annealing temperature, and PCR products were separated by electrophoresis on 1% agarose gels. The primers used are listed in Table S2 (available at ). For quantitative PCR, we used 50 ng cDNA. cDNA and β-gal primers (forward, ACAGTACGCGTAG; reverse, CCATCAATCCGGTAGGTTTTCCGG) were added to SYBR green PCR master mix (Applied Biosystems) according to the manufacturer's instructions. All data were normalized to 18S, which was used as the internal control. M- and F-MDSCs were plated at 2,500 cells/9.6 cm in GM. The following day, the medium was replaced with 500 μl DME (without serum or antibiotics) containing 4.0 μg plasmid, which also encoded for the neomycin resistance gene, and 5 μl LipofectAMINE 2000 reagent (Invitrogen). MDSCs were incubated at 37°C for 6 h in this medium, and then medium was removed and replaced with normal GM. After 5 d in culture, we began a selection process. For the next 2 wk, MDSCs were cultured in GM containing 1.5 mg/ml G418 sulfate (Cellgro). Controls were also transfected with GFP plasmid to confirm transfection. Cells were maintained in GM and transplanted to male /SCID mice as described above (Cell transplantation to skeletal muscle). Western blot analysis was performed using standard molecular biological techniques. Cells were lysed in Laemmli sample buffer (Bio-Rad Laboratories) and resolved on 4–20% precast gradient gels. Mouse was detected with anti- (1:1,000; BD Biosciences) with goat anti–rabbit HRP (1:2,500; BD Biosciences) and imaged using SuperSignal (Pierce Chemical Co.). β-actin levels were detected with anti–β-actin (1:5,000; Sigma-Aldrich) and goat anti–mouse HRP (1:5,000; Pierce Chemical Co.). sub #text W e e x a m i n e d C D 3 4 , S c a - 1 , a n d d e s m i n e x p r e s s i o n a s d e s c r i b e d a b o v e ( M D S C i s o l a t i o n a n d i n v i t r o c h a r a c t e r i z a t i o n ) . The ability of MDSCs to form myotubes under low oxygen conditions or after exposure to hydrogen peroxide was determined by fast MyHC staining. M- and F-MDSCs (populations A–D) were plated at a density of 1,000 cells/cm on collagen-coated plates with normal GM. For low oxygen stress, after 4 d of growth in atmospheric conditions, growth media was refreshed, and cells were transferred to a 2.5% oxygen incubator. Control flasks were maintained in atmospheric O incubators. For oxidative stress, after 4 d of growth in atmospheric conditions with normal media, growth media was refreshed (control), or MDSCs received media with 100 μM HO. After 24 and 48 h of exposure to low oxygen or incubation with HO, MDSC cultures were fixed and immunostained for MyHC as described above (In vitro characterization). Reduced glutathione levels were determined for M- and F-MDSC populations by flow cytometry. Cells were plated at 1,600 cells/cm on collagen-coated flasks with normal GM. After 24 h, cells were incubated in 5 μM monochlorobimane (Invitrogen) in normal growth media for 20 min at 37°C. The cells were then washed twice with PBS and harvested in 0.25% trypsin-EDTA. Intracellular glutathion levels were determined with a FACSAria machine (monochlorobimane excitation of 380 nm and emission of 461 nm). Fig. S1 shows M- and F-MDSC similarities in cell morphology comparisons, multilineage marker expression, telomerase activity, and telomere length. Fig. S2 shows the in vitro and in vivo effects of estrogen stimulation on M- and F-MDSCs. Table S1 presents the correlation coefficients for relations between variables. Table S2 presents the primers for RT-PCR. Online supplemental material is available at .

The mammary gland is a highly dynamic organ that undergoes a series of changes from intrauterine life to senescence. In humans, growth in adulthood commences at puberty where the parenchymal cells branch from a few blunt ending primary and secondary ducts into an elaborate tree with multiple terminal ducts and lobules. With each menstrual cycle, breast proliferation fluctuates, and, in the luteal phase, the growth fraction can become as high as 35% (; ). Accordingly, during pregnancy, there is both a 10-fold increase in the number of alveoli per lobule as well as de novo formation of lobules by lateral budding from existing terminal ductules, leaving behind a small amount of connective tissue space (for review see ). These cellular dynamics led to postulate the existence of a population of precursor cells in the adult human breast that are capable of giving rise to new lobules. From studies mainly in other species, it is known that adult stem cells are generally focal in distribution and not necessarily colocalized with the bulk of transiently amplifying cells (for review see ). In mice, the location of immature mammary gland stem cells was narrowed down to the peripheral cap cells of the terminal end buds (for review see ). In humans, however, in which end buds are not prominent structures (), the identification of a candidate stem cell zone has had to rely on a detailed and microscopically directed sampling of well-defined segments of the organ followed by functional assays. By combining microdissection with colony-forming ability, candidate stem cells in the human hair follicle were prospectively identified in the bulge region more than a decade before the bulge was unequivocally proven to be the epithelial stem cell niche in the skin (; ; ). The proximal ducts of the prostate were also shown to harbor stem cells by this method (). Recently, a side population exhibiting Hoechst dye efflux properties was isolated from the human mammary gland (for review see ), and self-renewing cells were enriched for by the use of nonadherent mammosphere cultures (). Whereas in mice, the ultimate evidence for the existence of mammary stem cells is the clonal repopulating ability and greater morphogenic capacity within the cleared fat pad (; ), such experiments cannot be performed in humans. Fortunately, surrogate assays conducted with mice and human cells demonstrate that putative mammary stem cells in 3D laminin-rich ECM (lrECM) gels function as they do in vivo in terms of several morphogenetic criteria across the species (; ; ; ). However, the existence of a hierarchy as well as a correspondence between stem cell markers and activity with specific regions of the gland has not been described. In this study, we examine in situ whether the resting human mammary gland exhibits stem cell markers, which could identify a stem cell zone, as has been done for other tissues (; ; ; ; ). We took advantage of Ki-67 and laminin-2/4 as biomarkers for cellular turnover and differentiation (; ; ; ; ; ; ). Further dissection of regions of interest and microcollection allowed us to interrogate functional stem cell properties in culture. These included high clonal proliferative capacity on tissue culture plastic, self-renewal in suspension, and mammary morphogenesis in 3D lrECM (; ; ; ; ; ; ; ; ). Multipotency was determined based on fluorescence imaging in situ and in clonal primary cultures using two lineage markers, keratins K14 and K19, in which K14 marks the myoepithelium and K19 is hypothesized to be a neutral switch keratin that permits the changeover of one type of cytoskeleton to the other (). Importantly, double-positive K14 and K19 transitional cells are known also to codistribute with the stem cell zone in the prostate (). To compare mouse and human markers, we isolated mammary stem cell activity on the basis of surface markers CD49f and EpCAM (). The cells thus isolated were further characterized using additional putative stem and progenitor markers. Finally, we took advantage of human papilloma virus (HPV) 16 E6/E7 to bypass cellular senescence (; ) and to interrupt the normal differentiation of stem and progenitor cells to generate cell lines that maintain their phenotypes in culture. The absence of senescence allowed us to follow the developmental hierarchy of the progenitor cells in the breast and made it possible to isolate stable cell lines representing the different stem and progenitor stages, as done previously for other cell types (; ; ; ; ; ). Our results have defined a human breast epithelial stem cell zone in vivo and a progenitor hierarchy both inside and outside this zone. Additionally, we have developed several stem and progenitor cell lines that will aid our understanding of the possible role of these cells in breast cancer. To test whether a stem cell zone could be defined in the human breast, we stained for several postulated surrogate stem cell markers, including stage-specific embryonal antigen-4 (SSEA-4), keratin K6a, keratin K15, keratin K5, Bcl-2, and chondroitin sulfate (; ; ; ; ; ). Three of the markers examined, SSEA-4 (), K15 () and K6a (see ), localized focally to discrete clusters of cells in ducts, including terminal ducts, and were essentially absent from lobules. Specifically, although 73–78% of ducts stained with K6a, SSEA-4, and K15, the frequency of lobules with stained cells ranged from 6 to 13% ( = 19). Keratin K5, Bcl-2, and chondroitin sulfate were more widely distributed but still were more prominent in ducts than lobules (Fig. S1 A, available at ). The frequency of SSEA-4–positive cells was further determined by FACS analysis of trypsinized uncultured organoids. Single cells identified as epithelial and myoepithelial by dual staining with epithelial antigen EpCAM and β4-integrin in the SSEA-4 gate comprised 0.5–0.7% of the total population, which also included the stromal cells (Fig. S1 B). Importantly, SSEA-4 cells were highly enriched for keratin K6a- and K15-positive cells (Fig. S1 B). Because quiescence is a general property of stem cells in their niche, we stained ducts and lobules with Ki-67. The level of staining in lobules varied markedly between biopsies, as 6/12 had moderate to strong staining (up to 50% of cells per lobule), and the other six were negative. However, ducts showed a low but constant level of staining: a mean of 2.8% (range of 1.3–4.7%) of cells stained in 10 biopsies (). The Ki-67 staining pattern was confirmed by staining for minichromosome maintenance protein 2 (; unpublished data). Furthermore, lobules rather than ducts stained specifically for the α-2 chain of laminin-2/4, a basement membrane component surrounding proliferating epithelial cells (; ; ; ; ). RT-PCR for the different chains of laminin using purified populations of myoepithelial cells and fibroblasts revealed that whereas myoepithelial cells contributed α-1, α-3, and α-5 chains of laminin in addition to α-2 chains of laminin-2/4, fibroblasts expressed α-2 chains of laminin-2/4 only (Fig. S1 C). These data show that a candidate stem cell zone resides in ducts that are enriched in cells identified as being SSEA-4/K5/ K6a/K15/Bcl-2 cells, which are generally quiescent and are surrounded by chondroitin sulfate. The more frequently proliferating progenitors are found outside this region and are often surrounded by laminin-2/4. We assessed the proliferative and morphogenic capacity of primary cells on tissue culture plastic, in mammosphere suspension cultures, and in a 3D lrECM assay. The growth of the cells derived from ducts or lobules were compared on tissue culture plastic. Only ductal-derived cells formed colonies that were considerably larger than 100 cells/colony (60 ± 2% of colonies had >100 cells; = 3 × 50 colonies) compared with lobular-derived cells (0% of colonies had >100 cells), indicating that as expected for stem cell activity in culture, the duct-derived cells have a higher proliferative potential than those derived from lobules (). Similar patterns were obtained from two additional biopsies when placed in culture. To further test for self-renewal, ducts were compared with lobules using the mammosphere assay (). Whereas ducts gave rise to relatively large mammospheres (>70 μm), those derived from lobules were small and irregular (<70 μm; ). Primary mammospheres were then trypsinized and replated to derive secondary mammospheres. In our hands, the frequency of primary and secondary mammosphere formation was ∼3/1,000 cells, a figure comparable with that described originally by . We assessed the presence of prospective multipotent cells in mammospheres by multicolor imaging based on the combined staining for myoepithelial keratin K14 and the luminal or switch keratin K19 (). The K14/19 phenotype was most often observed in duct-derived mammospheres (26/48 colonies; 54%) as opposed to lineage-restricted lobule-derived mammospheres (5/48 colonies; 10%; ). Similarly, mammospheres trypsinized and plated at clonal density from ducts and lobules gave rise to K19/K14 double-stained cells in 56% of those derived from ducts ( = 92 colonies) and only 6% of those derived from lobules ( = 92). The morphogenic potential of mammospheres was assessed in a 3D lrECM assay in which 2/49 duct-derived second passage mammospheres developed into terminal duct lobular unit (TDLU)–like structures. The formation of TDLUs was not observed from lobule-derived mammospheres (0/45); these only gave rise to spherical structures (). A similar pattern was observed with or without a feeder layer of primary epithelial cells. Essentially, this behavior was recorded in mammospheres derived from three randomly collected biopsies from reduction mammaplasty. Finally, immunostaining of plated mammospheres for SSEA-4 from a representative biopsy further revealed that 24% of duct-derived colonies contained positive cells compared with 10% of those derived from lobules. Thus, ducts are enriched for a subpopulation of epithelial cells that, in culture, exhibit a highly proliferative, self-renewing, and morphogenic capacity indicative of stem cell activity. When tissue sections were double stained for keratins K19 and K14, four different populations could be distinguished: K19/K14 (+/+), K19/K14 (+/−), K19/K14 (−/−), and K19/K14 (−/+). The +/+ cells were found as scattered single cells or small groups of cells in ducts ( and inset). The +/− cells were present both in ducts (not depicted) and in lobules (). The −/− cells were rare, but, when present, they appeared generally throughout an entire TDLU (), indicating their clonal origin. The −/+ cells were observed in the myoepithelial cell layer as expected (). These profiles reflect the unique subsets of breast epithelial cells that reside in the mammary gland, as was subsequently revealed in cloned primary cultures from crude collagenase digests (). Furthermore, this pattern was confirmed in microcollected ducts and lobules. Thus, whereas lobules gave rise almost exclusively to colonies of +/−, −/+, and a few −/− cells, only ducts also gave rise to colonies of +/+ cells (). The predominance of +/+ cells in cultures of ducts was observed in three experiments from two independent biopsies. Similar data were obtained after recloning. Thus, whereas cells from lobules either did not clone out or formed small abortive +/− clones, the ducts formed large +/+ clones (9/10 clones in an experiment from one biopsy; Fig. S2, available at ). We conclude that ducts house a subpopulation of epithelial cells that exhibit four important attributes of epithelial stem cells: they express stem cell markers in situ; they are slow cycling in vivo; they exhibit a high proliferative, self-renewal, and morphogenic capacity in culture; and they are bipotent. To find the in situ equivalent of ductal/lobular cells isolated by FACS analysis, we defined a set of anatomical markers. Specifically, within the luminal lineage, we used keratin K6a for ductal cells and BCA-225 for lobular cells; within the myoepithelial lineage, we used keratin K17 for ductal cells and WT1 against lobular cells (). Before FACS analysis, we first removed the substantial stromal component of the human breast by depleting endothelial, fibroblastic, lymphocytic, and monocytic lineages using a CD31/1B10/CD34/CD45 immunomagnetic column, allowing a flow through of lineage-negative (Lin) epithelial cells (). Primary cells were then sorted based on their expression of CD49f (α6-integrin) and EpCAM (). Four subpopulations, which were identified by gates I–IV, were sorted and then analyzed by immunofluorescence for the expression of anatomical selective markers ( and Fig. S3 C; see also Materials and methods). Similar FACS profiles were revealed in cells from six independent biopsies. Subpopulations isolated in gates I and II represented the luminal epithelial lineage, and gates III and IV contained the myoepithelial lineage (). A further subdivision into lobular cells (gates I and III) and ductal cells (gates II and IV) could be established by the anatomical markers (). As expected, the K19/K14 cells cosorted with ductal LinCD49fEpCAM cells within the luminal epithelial lineage (gate II; ). In addition, triple color staining showed that SSEA-4 expression was observed in the +/+, LinCD49fEpCAM population (Fig. S3, available at ). Functional assays for stem cell activity demonstrated that colony-forming ability was significantly restricted to the LinCD49fEpCAM population (gate II; ), also suggesting the isolation of multipotent progenitor activity. Furthermore, single-cell sorting into 96-well dishes based on SSEA-4 staining within a CD49f/EpCAM context revealed that the SSEA-4 subpopulation had a cloning efficiency of three colonies per 96-well dish as opposed to zero from the SSEA-4 gate, which suggests that the SSEA-4 subpopulation represents the majority of colony-forming activity. However, our experimental design so far does not allow us to account for the considerable variation in cloning efficiency between donor samples, so we cannot conclude definitively that the SSEA-4 subpopulation is significantly different from the total LinCD49fEpCAM population. Nevertheless, multicolor imaging revealed that all three clones were +/+ (Fig. S3). Most importantly, when cells from gates I–IV were cultured in 3D lrECM at clonal density, only the LinCD49fEpCAM population (gate II) formed budding (TDLU-like) structures (gate II; ). Thus, whereas the cells isolated from other gates morphologically consisted mainly of small spheres (acinus-like) or large, solid spheres as previously described (; ) and were lineage restricted, those isolated from gate II were TDLU-like budding structures and formed +/+, +/−, and −/+ as revealed by staining the whole mounts of gels (). Collectively, analysis of primary breast tissue reveals that all detectable stem cell–like or stemlike activities are restricted to duct-derived cells. We reasoned that if the mammary epithelial cell types identified above by K14 and K19 expression were connected in a hierarchy, only one cell type should give rise to the other lineage-restricted progenitors. To examine this we used HPV16 E6/E7 () to bypass senescence so cells could be followed for long periods of time. After transducing cells only once with retroviruses harboring E6/E7, we successfully recovered several sublines and clones of the four cell types that could be distinguished by K19/K14 expression as described in . We designated these cell lines +/+, +/−, −/−, and −/+ (). HPV16 E6/E7 conferred extended life span or immortality to all four subtypes, although the −/+ myoepithelial cells senesced after passage 30. +/+ cells' unique ability to give rise to the other three subtypes () and their ability to form TDLU-like structures in 3D lrECM cultures () confirmed that +/+ cells exhibited stemlike activity. Quantification of morphogenesis using the 3D lrECM assay revealed 54 ± 2.5% ( = 3) of TDLU-like structures in cultures of +/+ cells, whereas +/−, −/−, and −/+ gave rise to cells of their own subtypes only () and did not form TDLU in lrECM (). Relative to the other E6/E7 cell types, the multipotent +/+ cells were unique in their combined expression of SSEA-4, keratin K15, Oct-4 (), and Musashi-1 () as well as by their capacity to differentiate into more restricted luminal-like and myoepithelial-like cells in the presence of serum. The latter were characterized by a comprehensive panel of lineage markers (Fig. S4, A–C; and Table S1, available at ). These data provide evidence that a stem cell zone in ducts marked by K19/K14 can give rise to K19/K14, K19/K14, and K19/K14 lineage-restricted progenitors. We asked two questions: whether the integration site of the viral genes affected the stability or the expression of the cells and whether lineage-restricted cells represented the end of their differentiation repertoire or whether they could specialize further. We performed a mass transduction of lineage-restricted flow-sorted mammary epithelial cells based on their expression of surface markers. As in other hierarchical tissues, specialization of the lineages in the human breast is defined by the acquisition of certain differentiation markers that are characteristic of luminal epithelial and myoepithelial cells. RT-PCR of five lineage-specific markers revealed that this general pattern carried through to all of the isolated cell lines (). Thus, whereas the stemlike cells (+/+) had a low expression of three luminal and two myoepithelial markers, lineage-restricted progenitors (+/−, −/−, and −/+) displayed a strong expression of these markers in a mutually exclusive manner (). Furthermore, staining of the lineage-restricted progenitors for surface markers revealed that they stained with either the luminal marker (MUC1) or the myoepithelial marker (Thy1). As such, we could use these markers in a large-scale cell sorting to examine the robustness of lineage maintenance after E6/E7 immortalization (acquired self-renewal). Primary cultures were sorted into luminal epithelial (+/− and −/−) and myoepithelial (−/+) lineages based on staining with MUC1 and Thy1 (; before E6/E7, crude). The profile of primary cultures was identical to that of uncultured cells (unpublished data). Further plating and cultivation in a second passage of sorted cells did not shift this phenotype appreciably (; before E6/E7 transduction, sorted). Sorted cells were transduced with HPV16 E6/E7 and selected in the presence of G418. More than 70 different clones emerged during selection, and these were pooled and reanalyzed by the same criteria. The general patterns of MUC1/Thy1 and MUC1/Thy1 were sustained, albeit with a slight drifting in the population as a whole in spite of the different proviral insertion sites and the large number of pooled clones (; after E6/E7, pooled clones). This was confirmed directly by restriction site (RS) PCR and sequence analysis of chromosomal DNA in the unpooled +/− (MUC1/Thy1) and −/+ (MUC1/Thy1) clones to identify individual integrated proviruses (, clonal analysis). These data demonstrate that lineage phenotype is essentially sustained irrespective of the retroviral integration site in these lineage-restricted pooled clones. The fact that retroviral transduction also immortalizes the lineage-restricted cells suggested the possibility that these cells may be derived from the lineage-restricted progenitor cells. To show that this is the case, we searched for lineage-restricted markers in cells on the ductal-lobular junction. The search yielded new intralineage combinations of markers, demonstrating clearly that the three lineage-restricted cell lines were able to differentiate into additional cellular phenotypes within their respective lineages. Accordingly, +/− luminal progenitors can give rise to all combinations of the descendant cells expressing the luminal markers E29 and BCA-225 (). Similarly, −/− and −/+ cells give rise to descendants expressing combinations of either luminal markers K8 and CDw75 or myoepithelial markers WT1 and K17, respectively (). In all three cases, the cellular phenotypes were similar to their counterparts in situ (). These findings support the hypothesis that the lineage-restricted cells are progenitors rather than the ultimate end of their respective differentiation lineages. The data presented in this study provide the first evidence for the existence of a stem cell zone in mammary ducts marked by K19/K14, which can give rise to K19/K14, K19/K14, and K19/K14 lineage-restricted progenitors. The existence of stem and progenitor cells in the human mammary gland has been widely postulated; however, until now, neither the location nor the candidate cellular entities have been definitively identified and functionally characterized. Previously, we had described two cell lines that were derived from the human breast using magnetic sorting and HPV16 E6/E7 immortalization (). One of the immortalized clones, MUC1/ESA, was able to generate itself as well as luminal epithelial and myoepithelial cells; it further expressed keratin K19 and formed TDLU-like structures inside a 3D lrECM. The other, MUC1/ESA, was lineage restricted, keratin K19 negative, and formed acinus-like spheres within the 3D lrECM. Although this was an important advance, several essential questions remained: did these virally transduced cells have identical counterparts in vivo, or were these a result of immortalization? If counterparts existed, where were they located in situ? Were there additional progenitor cells we had not been able to immortalize? Could we develop a procedure whereby one could isolate the untransduced counterparts reproducibly? Were there differences between ducts and lobules, and, intriguingly, was there a hierarchy? In this study, we have answered all of the aforementioned questions: we have demonstrated the existence of four distinct human mammary epithelial cell types in situ, two of which most certainly are precursors to the two immortalized cell lines we had isolated previously using E6/E7. The cells are distributed in a stem cell zone in ducts and outside this zone in lobules. The size of the stem cell zone varies somewhat with the marker used, the most restricted and presumably most specific being the presence of keratin K15 and SSEA-4. Ducts and lobules were dissected, collected under the microscope, cloned, and characterized. We believe it is critical to meticulously dissect lobules distal to the intralobular terminal duct. Failure to do so could explain previously published data on the reported absence of a difference in growth patterns between ducts and lobules (). Permanent cell lines were established, cloned, and characterized further with respect to the profiles defined in situ and in primary cultures. The procedures we have developed, while painstaking and time consuming by necessity because they are for normal human tissues, are nevertheless robust. We show that these four cell types are hierarchically connected such that only one cell type can give rise to all others, which themselves are lineage-restricted progenitors. An important implication of these findings is that for the first time, a stem cell zone containing one or more mammary stem cells was identified in the human breast. Mouse mammary stem cells have recently been isolated prospectively based on various FACS strategies (; ). However, so far, FACS profiles have not translated into the cell of origin or location in situ. In this study, by the use of anatomical markers, we demonstrate that the previously described gates for bipotent progenitors, colony-forming cells, and mammary repopulating units in mice (; ) enrich specifically for cells of ductal origin in the human breast. Older studies had suggested that the mammary ducts of rodents may harbor such a stem cell niche. For instance, in the mammary gland from virgin mouse, candidate stem cells (stained with antibodies JB6 and JsE3) were found in ducts rather than in alveoli, and it was postulated that these served to regenerate ductal epithelium as well as forming new alveolar buds (). This postulate was born out recently in an experiment with rudimentary ducts from postgestational mice transplanted to cleared fat pads of TGF-β1 transgenic mice, which retained the capacity to reactivate lobular structures at late pregnancy (). Apparently, the mouse mammary gland ductal niche responds specifically to the MMTV–c-myc transgene by amplification of the stem cell compartment (). This implied strongly that in mice, an entire TDLU at any time represents the progeny from a single early ductal progenitor. Indeed, seminal studies of X chromosome inactivation in the human breast had demonstrated the presence of contiguous patches of normal mammary epithelium suggestive of being derived from single stem cells (). These patches were explained to be the result of some developmentally important long-term primitive stem cells, which were presumed to be estrogen receptor negative, as opposed to shorter term estrogen receptor–positive stem cells important for adult tissue homeostasis (). The hierarchy we describe here is essentially compatible with the aforementioned findings. However, it is important to recognize that the present identification of a stem cell zone in ducts does not necessarily exclude the existence of stem cells at other locations, although despite an extensive search, we have not observed such additional candidates as of yet. The possibility of one stemlike cell population in an organ, which precedes all of the other stemlike cells during development, appears to be a general phenomenon because such multiple cell-type niches were also previously described for the hair follicle (). The division of labor into that of the maintenance of tissue homeostasis and the more elaborate development or regenerative remodelling between stem cell compartments may also be a general phenomenon because in the skin, homeostasis is maintained exclusively by the epidermal proliferative unit, but wound repair depends on bulge cells from hair follicles (). In analogy to hair follicles, we would propose that mammary tissue homeostasis is maintained by proliferative units in the TDLUs, whereas more elaborate structures, including the formation of new TDLUs from ductal alveolar buds, depend on the recruitment of cells from ducts. Until recently, very few markers were assigned directly to a candidate mammary stem cell pool in humans or, indeed, even in rodents, but scattered single cells within the ducts of adult mice were shown to be positive for keratin K6a (), a marker originally found in putative stem cells in the terminal end bud (). In the present study, we found that the ductal stem cell zone was characterized by the accumulation of K19/K14 cells. If these traits were even indirectly related to stem cell activity, we would expect them to fluctuate with the size of the stem cell compartment. One of the most important pathways responsible for mammary stem cell activity is canonical Wnt signaling (). Interestingly, Wnt1- or β-catenin–induced mammary hyperplasia and tumorigenesis in mice correlate with the accumulation of K6-positive cells even though K6 in itself does not appear to be essential for mammary gland development, at least in embryonic knockouts (). It has been shown that caveolin-1 deficiency in mice, which conveys mammary hyperplasia and tumorigenesis along with an increased stem cell activity, is characterized by the accumulation of keratin K6-positive cells (). However, the mammary glands of mice that are unable to signal through the Notch pathway most closely mimic the profile of the human mammary stem cell zone (; ). Under these conditions, there is a remarkable ductal accumulation of cells expressing luminal keratins coordinately with keratins K14 and K6 (). A similar stem cell–related profile has been recorded in the prostate (; ). The appearance of +/+ cells in the suprabasal position has been interpreted as an accumulation of intermediate immature luminal cells (). Thus, it cannot be excluded that the stem cell zone described here contains bona fide ductal basal cells, which, under the current available culture conditions, are not colony forming. In mass cultures derived from reduction mammoplasty, we and others had shown previously that a subpopulation of the primary breast luminal epithelial compartment was capable of giving rise to both luminal epithelial and myoepithelial cells (; ; ; ). Furthermore, in culture and occasionally in suprabasal cells in the epithelial layer in situ, single cells could exhibit dual staining for these two cell types (). Cells in this position were shown later to be K19 positive and could be immortalized occasionally in mass culture with HPV E6/E7 (). The discovery in the present study of a ductal stem cell zone in the human breast opens the possibility of the existence of a hierarchy among differentiated progeny at other locations. In the mammary gland, the ultimate level of differentiation is reached during lactation. Although we have not addressed the consequence of lactation on ductal differentiation, a previous study strongly suggested that ducts are indeed protected from hormone-induced differentiation (). This is in agreement with the presumed function of candidate stem cells in mature ducts of humans and rodents as a source of lateral branching during the formation of milk-producing lobules (; ). Functionally, one of the hallmarks of stem cells is the ability to self-renew (for reviews see ; ). A culture assay for self-renewal in human breast epithelial cells was successfully adopted from the field of neurobiology combined with the use of 3D laminin-rich gels (). Thus, cells prevented from attachment grew in suspension to form the so-called mammospheres () reminiscent of the originally described neurospheres. Self-renewing stem cells were characterized by the ability to form new multipotent mammospheres even after passaging (). The other widely used culture assay of self-renewal in epithelial tissues is growth after plating at clonal density (for reviews see ; ). We used both assays in the present study and showed that both mammospheres and multipotent clones could self-renew, although we do not yet know how the fraction of the self-renewing clones compare between the assays. Our findings are pertinent to two poorly understood aspects of breast cancer evolution. For example, we know that breast cancers comprise at least two well-defined subtypes with distinct molecular profiles reminiscent of the luminal and basal lineages (). This has led to a resurgence of speculation that breast cancers may arise in the different compartments within a stem cell hierarchy (; ; ; ; ). The hierarchy described here may guide the design of experiments to test the possible role of these progenitors in the origin of breast cancer subtypes, a possibility that is under investigation. Normal breast biopsies ( = 54) were obtained from patients undergoing reduction mammoplasty for cosmetic reasons. The use of human material has been reviewed by the Regional Scientific Ethical Committees for Copenhagen and Frederiksberg and approved with reference to (KF) 263995. Normal breast tissue was prepared as previously described (). Upon collagenase treatment, epithelial organoids were either cultured as crude preparations or first manually separated under an inverted phase-contrast microscope (TMS-F; Nikon) into ducts and lobules before explantation (microcollected). Microcollected ducts and lobules were plated in collagen-coated (8 μg/cm; Vitrogen100; Cohesion) T-25 flasks (Nunc) in the presence of chemically defined medium (CDM3; ) and were allowed to spread for 8 d. Cells were trypsinized and cloned by limited dilution in the presence of serum-supplemented growth medium consisting of Ham's F12 medium (Invitrogen) supplemented with 2 mM glutamine, 50 mg/ml gentamycin (Biological Industries), 5% FCS (PAA Laboratories), 5 μg/ml insulin (Roche), 1 μg/ml hydrocortisone (Sigma-Aldrich), 0.1 μg/ml cholera toxin (Sigma-Aldrich), and 10 ng/ml EGF (PeproTech) at a density of 60 cells/cm (referred to below as F12 medium). After 7 d, some of the cultures were fixed in methanol and stained with hematoxylin. For secondary cloning experiments, cells were first plated at a density of 400 cells/cm and cultured for 12 d and then were subcloned at a density of 50 cells/cm. Cultures were fixed and stained after 10 d, and the number of colonies was quantified. K19/K14 profiles were assessed by fluorescence in parallel cultures and in cloned primary cultures seeded at a density of 350 cells/cm. The K19/K14 profiles of primary cultures derived from microcollected ducts or lobules were assessed at day 7 at densities of 800 cells/cm and in cultures of secondary clones at a density of 60 cells/cm at day 10. For the nonadherent mammosphere assay, large ducts, terminal ducts (identified by connecting alveoli), and lobules were isolated and trypsinized for 10–15 min at 37°C on an orbital shaker to obtain a single cell suspension. Nonadherent mammosphere cultures were prepared as previously described (). In brief, cells were plated at a concentration of 5,000–20,000 cells/ml. The cultures were monitored for up to 12 d for the appearance of mammospheres. After 8 d, cultures were photographed, and structures derived from ducts (large and terminal) and lobules, respectively, were quantified and separated into two categories: >70 μm and <70 μm ( = 3 × 200 structures). For analysis of keratin expression, duct- and lobule-derived mammospheres were either smeared onto a glass slide and stained or trypsinized at day 9, plated at clonal density (200 cells/cm), and propagated for 5 d in F12 medium before immunocytochemical staining. A total of 92 colonies from each segment were quantified using a fluorescence microscope (Dialux 20; Leitz) equipped with a 10× objective. Mammosphere populations derived from ducts and lobules were assessed for morphogenic potential by inoculation for 3 wk of each population in 300 μl lrECM (Matrigel; Becton Dickinson). Some cultures were conditioned by a feeder layer of primary human breast epithelial cells separated from the top gel by 200 μl of cell-free gel. The number of mammosphere-derived budding structures was assessed by phase-contrast microscopy. To passage mammosphere cultures, mammospheres were collected, trypsinized for 10 min at 37°C while shaking, and single cells were plated at a concentration of 1,000 cells/ml (). Secondary mammospheres were sampled and plated in monolayer culture and analyzed for SSEA-4 expression by immunocytochemical staining (see supplemental Materials and methods, available at ). For RT-PCR analysis of laminins (see supplemental Materials and methods), myoepithelial cells and fibroblasts were purified and cultured as previously described before RNA extraction (; ). Cells representing K19/K14, K19/K14, K19/K14, and K19/K14 profiles were recovered in primary culture and were HPV16 E6/E7 transduced using a previously described protocol (). Fig. S4 D shows the culture history of each of the representative cell lines. For 3D culture, single cell suspensions of 10 cells were inoculated in lrECM as previously described (; ) and were observed for a period of 12 d. For analysis of the robustness of transduced lineages, uncultured epithelial organoids or normal breast epithelial cells grown in primary monolayer cultures in chemically defined medium (CDM3; ) were trypsinized and filtered through a 20-μm cell strainer (Miltenyi Biotec) and resuspended in Hepes buffer supplemented with 0.5% BSA (bovine fraction V; Sigma-Aldrich) and 2 mM EDTA (Merck), pH 7.5. The suspended cells were incubated for 30 min at 4°C in the presence of monoclonal primary antibodies recognizing MUC1 (CD227; 1:100) and Thy1 (CD90; 1:50). Control samples were incubated without primary antibody. Upon incubation, the cells were washed twice in Hepes/BSA/EDTA buffer and incubated for 15 min at 4°C with secondary isotype-specific fluorescent antibodies, AlexaFluor488 goat anti–mouse IgG2b, and AlexaFluor633 goat anti–mouse IgG1 antibodies (1:500; Invitrogen). After incubation, the cells were washed twice in Hepes/BSA/EDTA buffer and resuspended in a volume of 1 ml of buffer. Propidium iodide (Invitrogen) was added at a concentration of 1 μg/ml, and the cells were analyzed and sorted using a flow cytometer (FACSAria; BD Biosciences). The sorted populations were subsequently plated in collagen-coated T-25 culture flasks in chemically defined media (CDM6; ) for MUC1/Thy1 cells and in CDM4 for MUC1/Thy1 cells. The sorted cell populations were transduced as previously described (). Untransduced sorted cell cultures were run in parallel as controls. Secondary cultures were analyzed for the expression of MUC1 and Thy1 by FACS analysis as described above. To isolate putative stem cells within the lineage-negative epithelial cell population, breast organoids ( = 3 biopsy samples) were trypsinized for ∼10 min at 37°C under rotation. The solution was robustly agitated a few times during trypsination. Trypsination was stopped with FCS, and the cells were filtered through a 30-μm filter followed by filtration through a 10-μm filter to obtain a single-cell suspension. Cells were incubated with a cocktail of antibodies against stromal cells, which included CD31 (JC70A; 1:50), CD34 (QBEnd/10; 1:50), CD45 (Bra-55; 1:250), and fibroblast surface protein (1B10; 1:50). Cells were incubated at 4°C for 30 min. After incubation, cells were washed twice in Hepes/BSA/EDTA buffer and incubated for 15 min at 4°C with goat anti–mouse IgG and rat anti–mouse IgM microbeads (Miltenyi Biotec). Cells were then washed again twice and further applied to a column for immunomagnetic cell sorting (MACS; Miltenyi Biotec). The flowthrough from this column was collected and incubated with CD49f (GoH3; 1:500) and EpCAM (VU1D9; 1:100). A control solution was also prepared in which the primary antibodies were excluded. The cells were incubated at 4°C for 30 min followed by two washes and were further incubated with the secondary fluorescent antibodies AlexaFluor488 rabbit anti–rat IgG and AlexaFluor633 goat anti–mouse IgG1 (1:500; Invitrogen). Some degree of cross-reaction between CD49f and AlexaFluor633 was observed. For comparison with a directly conjugated EpCAM antibody, please see Fig. S3 C. The four FACS-sorted populations in gates I–IV were analyzed for luminal and myoepithelial markers as described below and were tested for self-renewal by limited dilution cloning in collagen-coated six-well plates (70 cells/cm) in the presence of F12 medium. Morphogenic potential was analyzed in 24 wells in the presence of CDM3 by inoculation in lrECM at a density of 2 × 10 cells in 300 μl lrECM placed on top on a 200-μl cell-free gel solidified on top of a feeder layer of primary breast epithelial cells. The cultures were observed for 3 wk and assessed in triplicate for morphogenesis by phase-contrast microscopy in 50–80-μm spherical acinus-like structures, budding structures, and spherical colonies (>100 μm). Monolayer cultures and cells derived directly from collagenase digested tissue, cryostat sections of biopsies, mammospheres, or cell lines cultured in 3D lrECM were prepared and stained by immunoperoxidase or immunofluorescence as previously described (; ; ). For antibodies and further details, also see . The four CD49f/EpCAM cell populations were smeared onto glass slides and stained for BCA-255, K6a, K17, and WT1 by immunoperoxidase and for K19/K14 by fluorescence staining with isotype-specific AlexaFlour488 goat anti–mouse IgG3 and AlexaFlour568 goat anti–mouse IgG2a and were quantified ( = 3 × 100 cells; in triplicate from a representative biopsy). As controls, sorted cells were smeared without further staining to check for residual background fluorescence, or cells were stained after having switched the secondary antibodies to use different colors than were used previously. For staining of whole mounts of 3D cultures of cloned FACS-sorted cells, gels were fixed in methanol/acetone (1:1) for 30 min at −20°C followed by a 2-h incubation with blocking buffer and were incubated with 400 μl of antibody solution (K14 [1:25] and K19 [1:50]) overnight at 4°C, washed for 3–4 h, and incubated with secondary antibodies at 4°C overnight. Immunofluorescence stainings were evaluated using a laser-scanning microscope (LSM 510; Carl Zeiss MicroImaging, Inc.). RNA isolation and PCR reactions were performed as previously described (,). Specifically, primers included EpCAM forward (AGTGTACTTCAGTTGGTGCACAAA) and EpCAM reverse (AGTGTCCTTGTCTGTTCTTCTGAC; T 56°C for 29 cycles); estrogen receptor forward (CCCTACTGCATCAGATCCAAGG) and estrogen receptor reverse (CTGCAGGAAAGGCGACAGC; T 60°C for 40 cycles); and tenascin forward (TCCTGCTGACTGTCACAATC) and tenascin reverse (TGCTCACATACACATTTGCC; T 60°C for 30 cycles). To determine the integration sites of the HPV16 E6/E7–expressing vector, an RS-PCR assay was used (; ) The HPV-specific primers were modified to locate to the E6/E7 ORF, and additional RS oligonucleotides were added to the assay. In brief, genomic DNA was extracted from each cell line using the DNeasy tissue kit (QIAGEN). Previously used HPV-specific primers () were supplemented with the HPV765-24D/HPV790-25D primer sets (). The RS oligonucleotides contain a T7 phage promoter, 10 random nucleotides, and a specific RS recognition sequence. Amplification of vector-genome hybrid sequence was used by a seminested PCR reaction. The PCR reactions were performed using the Expand High Fidelity PCR System (Roche) in a 20-μl volume using 1 U polymerase enzyme, 1× PCR buffer, 200 μM deoxynucleotide triphosphate, 2 pmol HPV primer, and 20 pmol RS oligonucleotide primer. 1 μl from the first round of PCR was used as a template for the second round of amplification. Cycling conditions for the first round of PCR were 94°C for 2 min followed by 30 cycles of 94°C for 30 s, annealing for 30 s at 65°C descending 0.5°C each cycle, and extension at 72°C for 2 min. This was followed by 15 more cycles at a fixed T of 55°C. The conditions for the second round of PCR were similar, and products were run on agarose gels, bands of interest were cut out, and specific products were extracted using a QIAquick Gel Extraction kit (QIAGEN). The presence of an E6- or E7-specific sequence in the selected products was verified by PCR reamplification using internal E6/E7 primers. Sequencing using a genetic analyzer (ABI PRISM 310; Applied Biosystems) was performed as previously described (). Fig. S1 shows a more detailed characterization of the ductal stem cell zone. Fig. S2 shows that clonal colonies from ducts, in contrast to lobule-derived colonies, are multipotent. Fig. S3 shows that SSEA-4 cells cosort with the LinCD49fEpCAM population and that these cells give rise to multipotent clonal colonies. Fig. S4 shows that the stemlike cells (+/+) express surrogate stem cell markers and are bipotent. A flow chart describing the isolation of one stemlike cell and three lineage-restricted progenitor cells is also shown. Table S1 shows that bypass of senescence maintains the differentiation repertoire of target cells with respect to a panel of markers for differentiated luminal and myoepithelial cells. Supplemental materials and methods provides further details about cell lines and clones, FACS analysis and cloning, immunohistochemistry and cytochemistry, and RNA isolation and RT-PCR. Online supplemental material is available at .

In polarized epithelial cells, the plasma membrane is divided into functionally and morphologically distinct apical and basolateral domains that have different protein and lipid compositions and are separated by tight junctions. The polarized distribution of proteins is achieved by the sorting of newly synthesized proteins at the TGN and/or the recycling endosome into separate carriers destined for the apical or basolateral domain (). In addition, recycling proteins are sorted in the endosomal compartment after endocytosis from the plasma membrane. Sorting of basolateral transmembrane proteins is guided by peptide motifs present in their cytoplasmic tails, such as dileucine motifs or tyrosine-containing sequences, including YxxΦ and other less well-characterized signals (; ). Some, if not all, of these motifs bind to the clathrin adaptor protein complexes (AP)-1 and/or -4, which mediate incorporation of the cargo protein into basolateral carriers. Sorting to the apical domain is less well defined and may involve several pathways; depending on the specific membrane protein, it may require glycosylation of its extracellular domain, lipid raft association, or the presence of cytoplasmic peptide sequences (; ). Polarized epithelial cells express a specific variant of AP-1, called AP-1B, which is important for basolateral targeting of several transmembrane proteins, including vesicular stomatitis virus glycoprotein G (VSV-G) and the low-density lipoprotein receptor (LDLR; ; ; ; ). Both of these proteins contain tyrosine-dependent sorting motifs (, ; ). In contrast, sorting of the Fc γ receptor isoform BII (FcγRIIB), which contains a dileucine motif, appears to be AP-1B independent (; ; ; ). Other components involved in basolateral transport are the actin-regulatory GTPase cdc42, the exocyst complex, and the GTPase Rab8. Functional defects caused by overexpression of cdc42 mutants highlight the importance of the actin cytoskeleton for transport of proteins from the TGN to the basolateral domain (; ). The exocyst is a protein complex of eight subunits that mediates the tethering of secretory vesicles to docking sites on the plasma membrane. In MDCK cells, this exocyst complex is required for delivery of membrane proteins to the basolateral domain, but not the apical domain (). In yeast, the function of the exocyst complex is regulated by the Rab GTPase Sec4 (). The mammalian homologue of Sec4 is Rab8, which is a key regulator of exocytic membrane traffic from the Golgi complex to the plasma membrane (; ). In polarized MDCK cells, Rab8 specifically regulates AP-1B–dependent transport to the basolateral domain (). In these cells, Rab8 and AP-1B can be found in recycling endosomes, which is an important sorting station for certain basolateral proteins that are en route to the plasma membrane (). In nonpolarized cells, Rab8 has been shown to bind to optineurin, a conserved 67-kD protein containing multiple leucine zipper domains and a putative zinc finger, which also binds to myosin VI (; ). Myosin VI is a unique, actin-based motor protein that moves along actin filaments in the opposite direction of all the other classes of myosins characterized thus far (). This multifunctional motor protein is involved in both endocytic and exocytic membrane-trafficking pathways (, ; ; ). Functional studies on nonpolarized cells derived from the myosin VI knock-out mouse (Snell's waltzer) have shown that myosin VI is required for efficient secretion of alkaline phosphatase and for the maintenance of Golgi morphology (). In normal nonpolarized cells, myosin VI is linked to the Golgi complex and to the secretory pathway by optineurin because knockdown of optineurin by siRNA results in the loss of myosin VI from the Golgi complex and impaired delivery of VSV-G to the plasma membrane (). Optineurin appears to be a linker protein between myosin VI and Rab8, as all three proteins colocalize in the perinuclear region at/around the Golgi complex and on vesicles underneath the plasma membrane. Overexpression of constitutively active Rab8Q67L recruits myosin VI onto Rab8-positive tubules and vesicles around the Golgi complex. Four splice variants of myosin VI are expressed in mammalian cells in a tissue-specific manner; isoforms containing a large insert (LI) in the tail are found in polarized epithelial cells with well-developed apical microvilli, whereas isoforms with a small insert (SI) or no insert (NI) in the tail are expressed in polarized and nonpolarized cells (). Because myosin VI is involved in the secretory pathway in nonpolarized cells and is linked via optineurin to Rab8, which regulates basolateral transport in polarized epithelial cells, we investigated the role of myosin VI in polarized biosynthetic membrane traffic in MDCK cells. We observed that the specific isoform of myosin VI with no inserts in the tail domain is required for transport of newly synthesized basolateral membrane proteins to the correct surface domain. Overexpression of the tail domain of this myosin VI isoform selectively inhibits delivery of tyrosine motif–containing cargo to the basolateral surface, but not that of basolateral membrane proteins with a dileucine motif or of membrane proteins with different apical sorting motifs. Thus, only the specific delivery of AP-1B–dependent basolateral cargo is inhibited. Site-directed mutagenesis indicates that a functional complex containing myosin VI, optineurin, and Rab8 mediates sorting along this transport pathway. Furthermore, we show that myosin VI and optineurin colocalize with AP-1, and also with Rab8 and the transferrin receptor (TfR) in clathrin-coated endosomal structures, which are therefore most likely to be recycling endosomes. Our results demonstrate that myosin VI is directly involved in the AP-1B–dependent sorting of proteins to the basolateral plasma membrane in polarized epithelial cells. This is the first motor protein to be identified in this pathway. In mammalian cells, four splice variants of myosin VI are expressed, containing either a LI (21–31 aa), a SI (9 aa), both inserts (SI+LI), or NI in the C-terminal tail domain (). The tissue distribution of these isoforms indicates that the isoform containing the LI is predominantly expressed in tissues with polarized epithelial cells containing apical microvilli. To determine which of the myosin VI isoforms are expressed in nonpolarized and polarized MDCK cells, we performed RT-PCR on MDCK cells grown on normal tissue culture plasticware (nonpolarized cells) and on MDCK cells grown on porous filters (polarized cells), respectively. The RT-PCR results show that nonpolarized MDCK cells contain the myosin VI isoform with NI, whereas the polarized cells express three different isoforms containing either NI, the LI, or the SI+LI (). Sequencing the RT-PCR products revealed that MDCK cells express a myosin VI with a longer 31-aa version of the LI compared with the LI expressed in human brain (21 aa) myosin VI. The sequences of the SI, however, were conserved between the myosin VI isoforms in MDCK cells and human brain. As a first step to elucidate the function of myosin VI in MDCK cells, we studied the subcellular localization of endogenous myosin VI by indirect immunofluorescence using a polyclonal antibody to the tail domain of myosin VI (). Endogenous myosin VI can be found throughout the cell at the apical and basolateral domains, but is slightly enriched in the apical terminal web region. To determine whether the three myosin VI isoforms expressed in polarized MDCK cells display a specific intracellular localization, we made stable MDCK cell lines expressing the corresponding human isoforms with an N-terminal GFP tag. For functional studies, we also raised a complete set of stable MDCK cell lines expressing the four different tail domain isoforms tagged with GFP, which can act as dominant-negative inhibitors of myosin VI function (). At least four individual clones were raised for each isoform, and the highest expressing cells of each clone were enriched by FACS sorting. Immunoblotting results show that the amount of expressed GFP-tagged myosin VI (, top band) or tail (, bottom band) roughly equals the expression level of endogenous myosin VI. The full-length myosin VI isoforms containing NI, the SI, or the SI+LI were found in both the apical and the basolateral regions of MDCK cells (), with a very similar localization to that of endogenous myosin VI. In contrast, the full-length myosin VI LI isoform was almost exclusively concentrated at the apical domain, where it showed good colocalization with actin. Thus, this isoform appears to be present in the apical microvilli, which is a region very rich in actin. All the other myosin VI isoforms showed less colocalization with actin at the apical domain, indicating that these isoforms are not recruited into microvilli (). To assess the role of myosin VI in the sorting of newly synthesized proteins from the Golgi complex to the cell surface in polarized cells, we studied polarized transport in MDCK cells stably expressing the different dominant-negative myosin VI tail isoforms. We used an adenovirus delivery system to express basolateral or apical reporter proteins in these cells. First, we examined the biosynthetic delivery of VSV-G, which is a basolateral membrane protein that is sorted by its cytoplasmic tyrosine motif in an AP-1B–dependent manner. This protein has been used as a reporter molecule for basolateral sorting in numerous studies (; , ). We monitored the surface appearance of VSV-G using pulse-chase radioactive labeling and domain-selective biotinylation (). In these experiments, the MDCK cells were grown on filters for 4 d to full polarization, and then pulse labeled with [S]methionine for 20 min before chasing for the indicated times. At each time point, cells were biotinylated at the apical and basolateral surfaces to measure the amount of reporter protein that had reached each plasma membrane domain. In control wild-type MDCK cells, cells radiolabeled with VSV-G appeared at the cell surface as early as 30 min into chase, which is consistent with previous studies. At all time points (throughout the entire chase period), 90% of the biotinylated protein was detected on the basolateral surface (). In MDCK cells expressing the dominant-negative tail domain containing NI, however, >80% of the biotinylated VSV-G was found on the apical surface throughout the chase (). After 30 min of chase, a very small amount (<20%) of the VSV-G appeared at the basolateral surface and disappeared over time. This suggests that the VSV-G may be indirectly missorted using a transcytosis step via the basolateral domain. However, after 30 min, the bulk of the labeled VSV-G is clearly at the apical domain, suggesting direct delivery to the apical surface. Furthermore, recent studies have confirmed that, in MDCK cells, newly synthesized membrane proteins are transported directly to the apical or basolateral surface, i.e., not along a transcytotic route (; ). The fraction of VSV-G reaching the cell surface was ∼35% of the total newly synthesized protein in wild-type cells and ∼18% in NI tail–expressing cells (Fig. S1, available at ). This reduction of ∼50% in surface delivery of VSV-G was only observed in cells expressing the NI tail and not in cells expressing the other tail isoforms. The dominant-negative effect of the NI tail on the rate of exocytosis is in agreement with our earlier results with Snell's waltzer fibroblasts, which show a 40% reduction in the rate of secreted alkaline phosphatase secretion (). Furthermore, in RNAi experiments, reduced expression levels of myosin VI and optineurin lead to a reduced level of delivery of VSV-G to the cell surface in HeLa cells (; unpublished data). In MDCK cells expressing the SI, the LI, or the SI+LI tail, VSV-G was targeted correctly because >80% of the total cell surface biotinylated protein was detected on the basolateral plasma membrane (). These results suggest that myosin VI is involved in the basolateral sorting of VSV-G, and that it is specifically the NI isoform that carries out this function. To establish whether myosin VI is also required for sorting of proteins destined for the apical domain, we examined the polarized delivery of newly synthesized influenza HA. In control wild-type MDCK cells, >90% of the HA reaching the cell surface was found on the apical plasma membrane after 1 and 2 h of chase (). Consistent with previous studies, HA is transported directly from the TGN to the apical surface without passing through the basolateral surface (; ). In MDCK cells expressing the four different myosin VI tail isoforms after 2 h of chase, >90% of the cell surface HA was also found at the apical plasma membrane (); i.e., no mistargeting to the basolateral domain was observed. These observations suggest that none of the myosin VI isoforms are involved in HA sorting to the apical domain in MDCK cells. The detailed mechanisms involved in the apical sorting pathways are still poorly understood. However, there are at least three different groups of apical sorting determinants that reside either in the extracellular, transmembrane, or cytoplasmic domains of the cargo proteins, and they probably use different pathways. The sorting signal of HA is located in its transmembrane domain, which also mediates lipid raft association (; ). To confirm that myosin VI was not required for sorting of apical proteins with different sorting signals, we also examined the trafficking of endolyn and an endolyn–megalin chimera. The sialomucin endolyn contains N-glycan–dependent apical targeting information in its ectodomain. This protein has a lysosomal steady-state localization in many cell types that is caused by the interaction of its cytoplasmic YxxΦ motif with AP-3; however, this motif binds only weakly to other AP complexes and is recessive to the apical sorting signal for polarized surface delivery (, ). Megalin is sorted apically by a peptide motif present in its cytoplasmic tail (; ). Therefore, we fused this megalin tail to the ecto- and transmembrane domain of endolyn and used this chimera as the third apical reporter. The results obtained with these two additional marker proteins in MDCK cells expressing any of the different tail isoforms, were very similar to those observed for HA (unpublished data). In addition, the endogenous apical protein gp135 shows a normal immunofluorescence distribution under all conditions in these MDCK cells (), which further confirms that myosin VI is not involved in the sorting of proteins to the apical domain of polarized MDCK cells. To establish whether the myosin VI NI isoform affects only the sorting of specific basolateral cargo, we compared the cell surface targeting of AP-1B–dependent and -independent reporter molecules by indirect immunofluorescence. VSV-G and the LDL receptor both contain a tyrosine motif and use the AP-1B–dependent basolateral sorting pathway (; ). To monitor LDL receptor uptake we used a CD8-LDL receptor chimera containing the extracellular and transmembrane domains of CD8 fused to the LDL receptor tail (). Basolateral targeting of the immunoglobulin Fc receptor (FcγRIIB), however, depends on a dileucine motif, which directs the receptor into an AP-1B–independent pathway to the basolateral domain (; ). In the first experiment, we used VSV-G, which we have shown in the cell surface biotinylation assay to be missorted to the apical domain in cells expressing the dominant-negative myosin VI NI tail. The VSV-G used is the temperature-sensitive mutant (tsO45 VSV-G; ), which accumulates in the ER at 40°C, the nonpermissive temperature. Lowering the temperature to 31°C allows the protein to fold and to be transported to the cell surfaces via the Golgi complex. Therefore, after infection with the virus encoding tsO45 VSV-G, the cells were first grown at 40°C to accumulate the protein in the ER before shifting to the permissive temperature in the presence of cycloheximide to initiate synchronous transport of VSV-G through the Golgi to the cell surface. VSV-G present on the cell surface of nonpermeabilized cells was recognized by an antibody against the luminal domain of VSV-G (). VSV-G was mainly present on the basolateral surface in wild-type MDCK cells, as shown in . In contrast, expression of the myosin VI tail isoform with NI resulted in the apical appearance of VSV-G (). Expression of myosin VI tail isoforms with the SI, the LI, or the SI+LI exhibited a subcellular localization similar to wild-type MDCK cells (unpublished data). Thus, these VSV-G immunofluorescent localization data are in good agreement with the results of the domain-selective biotinylation (). As observed with VSV-G, the CD8-LDL receptor tail chimera was missorted to the apical surface in cells expressing the myosin VI dominant-negative NI tail (), whereas it was sorted correctly to the basolateral surface in wild-type MDCK cells () and in cells expressing myosin VI tails with the SI, the LI, or the SI+LI (not depicted). To address the question of whether apical missorting of the CD8-LDL receptor involves transcytosis, we expressed an endocytosis-defective LDL receptor construct with a mutation in the endocytosis signal (NPXY–NPXA) in myosin VI NI tail–expressing cells (). Like the wild-type CD8-LDL receptor, the mutant receptor is delivered to the apical domain (). This result clearly indicates that in myosin VI NI tail–expressing cells AP-1–dependent cargo is missorted directly to the apical domain and does not involve endocytosis and transcytosis via the basolateral domain. In striking contrast to these observations, the FcγRIIB receptor (containing a dileucine sorting motif) was targeted correctly to the basolateral surface in cells expressing the myosin VI NI tail () and in cells expressing the other myosin VI tails (not depicted). These results suggest that the myosin VI NI isoform is specifically responsible for the sorting of AP-1B– dependent cargos to the basolateral plasma membrane in polarized MDCK cells. Further support for these results is provided by the observation that the steady-state distribution of endogenous E-cadherin, whose basolateral sorting also depends on a dileucine motif, is unchanged in cells expressing the myosin VI NI tail () or the other myosin VI tail isoforms (not depicted). Recently, several components involved in AP-1B–dependent basolateral protein sorting, such as the small GTPase Rab8, have been identified. Expression of a constitutively active Rab8 missorts VSV-G and LDL receptors, but not FcγRIIB, to the apical surface in MDCK cells, and overexpression of Rab8 causes disruption of AP-1 localization (). As Rab8 interacts with the myosin VI binding partner, optineurin (; ), we next asked if optineurin might link myosin VI to Rab8 in the AP-1B–dependent sorting pathway. Because we previously identified the optineurin binding site (RRL; aa 1,107–1,109) in the C-terminal globular tail domain of myosin VI, we mutated this binding motif to three alanines (RRL–AAA) in the NI tail isoform and generated stable cell lines expressing this mutant NI tail. We also established stable cell lines expressing the constitutively active form of Rab8 (Rab8-Q67L). Expression of Rab8-Q67L and the wild-type myosin VI NI tail resulted in indistinguishable phenotypes, with VSV-G being missorted to the apical surface (; ). In cells overexpressing the myosin VI NI tail, no change in AP-1 localization was observed (Fig. S2, available at ; ). However, in cells expressing the myosin VI NI tail with the mutated optineurin binding site, VSV-G was correctly transported to the basolateral surface (). These results, together with previous data, strongly suggest that myosin VI, optineurin, and Rab8 operate in the same basolateral sorting pathway and, most likely, form a functional complex. It is still a matter of debate whether AP-1B–dependent sorting occurs at the TGN or in a post-TGN compartment, such as recycling endosomes. Therefore, we determined the localization of myosin VI and optineurin at the EM level in polarized MDCK cells. In nonpolarized fibroblasts, myosin VI has previously been shown, at the ultrastructural level, to be associated with vesicles at the trans side of the Golgi complex and with clathrin-coated structures at the plasma membrane (; ). In polarized MDCK cells, we observed that myosin VI and optineurin colocalize with AP-1 on clathrin-coated endosomes and clathrin-coated vesicles surrounding these endosomes (). To further characterize these endosomes, we raised a stable MDCK cell line expressing GFP-Rab8. We performed double-labeling experiments on cryosections with antibodies to GFP to label Rab8 and with antibodies to myosin VI to visualize the endogenous protein. Finally, we expressed the human TfR in MDCK cells and performed an antibody uptake experiment from the basolateral domain with a monoclonal antibody to the human TfR to localize the endocytosed receptor. Endogenous myosin VI can be found in recycling endosomes colocalizing with the TfR and with Rab8. Collectively, these results show that myosin VI and optineurin are present in Rab8- and TfR-positive recycling endosomes, where they colocalize with AP-1 and may function in the sorting of cargo to the basolateral domain. Compelling evidence is presented that myosin VI is an essential component of the basolateral sorting machinery in the AP-1B–dependent pathway in polarized epithelial cells. Although three different splice variants of myosin VI are expressed in polarized MDCK cells, only the isoform with NI in the tail domain is required for correct sorting of specific membrane proteins to the basolateral domain. Interestingly, none of the myosin VI isoforms is required for the sorting of proteins to the apical domain. Expression of the dominant-negative myosin VI NI tail mutant selectively results in missorting of plasma membrane proteins containing tyrosine-based sorting motifs that interact with the clathrin adaptor complex AP-1B. These results add myosin VI to the growing list of components, such as Rab8 and cdc42, which are essential for the AP-1B–dependent basolateral sorting pathway. The Rho family GTPase cdc42 is a key regulator of the actin cytoskeleton, and the effects caused by overexpression of constitutively active or inactive cdc42 mutants highlights the importance of actin polymerization/depolymerization for basolateral sorting (; ; ). Although the exact role of actin filaments in vesicle transport to the basolateral domain is not known, it is likely that one function is to provide tracks for myosin VI–based movement to exert pulling, tethering, or transport forces on exocytic membranes and vesicles. Using the myosin toxin BDM, which is now believed to solely inhibit myosin II activity (), it has been shown that the release of basolateral transport vesicles from the Golgi complex in MDCK cells is prevented (). Thus, inhibition of myosin II activity, which stabilizes the actin filament network surrounding the Golgi complex, appears to lead indirectly to inhibition of basolateral transport. In contrast, the role of myosin VI in basolateral sorting appears to be directly related to the formation of a functional sorting complex. Interestingly, our previous work in nonpolarized cells demonstrated that optineurin links myosin VI to Rab8, and all three proteins are present in the perinuclear region at/around the Golgi complex and on vesicles close to the plasma membrane (). In nonpolarized cells, overexpressing a constitutively active Rab8 mutant causes the production of long, tubular membrane structures emerging from the perinuclear region. Myosin VI is specifically recruited to these structures, indicating that GTP-bound Rab8 may serve as a membrane anchor for myosin VI (). It is believed that one function of Rab proteins is to serve as molecular tags on the cytoplasmic surface of membrane compartments to mediate recruitment of cytosolic components such as, for example, actin-based and microtubule-based motor proteins (). Therefore, Rab8 may recruit myosin VI to the AP-1B sorting compartment via optineurin. Further support for this idea is provided by the observation that mutation of the optineurin binding site on myosin VI ablates the negative inhibitory effect of the myosin VI tail domain NI isoform on basolateral sorting. These results suggest that a functional complex between myosin VI, optineurin, and Rab8 plays a crucial role in the AP-1B–dependent sorting pathway. But at which step along the AP-1B–dependent sorting pathway does myosin VI function? In our previous studies in nonpolarized cells, we localized myosin VI in vesicles at the trans side of the Golgi complex that were, on average, 200–500 nm away from TGN38, which is a marker protein for the trans-Golgi network (). Interestingly, this distribution of myosin VI on vesicles near the TGN is very similar to the reported localization of the AP-1B–positive compartment (∼500 nm away from Golgi cisternae; , ). In this study, we show that in polarized MDCK cells myosin VI and optineurin colocalize with AP-1 on clathrin-coated endosomal structures, which are most likely recycling endosomes based on the colocalization of myosin VI with TfR and Rab8 in similar structures. Recently, it has been demonstrated that recycling endosomes play an important role in basolateral sorting and contain components of the AP-1B–dependent pathway, such as Rab8 and exocyst subunits involved in the basolateral sorting pathway (). At the recycling endosome, myosin VI could initially be involved in segregating/clustering AP-1B–dependent basolateral cargo into clathrin-coated subdomains. Subsequently, myosin VI could facilitate the formation of a clathrin-coated transport vesicle by pulling membrane away from the endosome, and therefore supporting vesicle scission. Finally, it could transport the cargo vesicle formed over short distances toward the basolateral domain. The orientation of actin filaments around endosomes is currently unknown; however, studies from isolated phagosomes indicate () that the plus-ends of actin filaments are directed toward the surface of phagosomes. Therefore, the minus end directionality of myosin VI favors the pulling of membrane away from the endosomal surface and the movement of vesicles away from the endosome. In nonpolarized cells, we have shown () that myosin VI, optineurin, and Rab8 colocalize in vesicles underneath the plasma membrane. These results suggest that myosin VI, in conjunction with Rab8, might be required for the delivery of basolateral cargo to the plasma membrane, where it might cooperate with the exocyst complex in the tethering of vesicles before fusion. In yeast, the Rab8 homologue Sec4p targets vesicles to sites of exocytosis at the cell surface (). In mammalian cells, the exocyst complex is only required for basolateral, but not for apical, delivery of cargo (). Interestingly, like myosin VI, subunits of the exocyst complex can be found both at the plasma membrane and in the perinuclear area, probably in the recycling endosome (). So how are basolateral cargoes missorted to the apical domain when myosin VI function is blocked? If only myosin VI–driven transport of AP-1B–specific basolateral carriers along actin filaments was inhibited, it would lead to an accumulation of secretory vesicles in the perinuclear area or around recycling endosomes, but would not result in apical mistargeting. In addition, loss of myosin VI function in exocyst-mediated tethering of AP-1B–specific vesicles to the basolateral domain would also not necessarily result in apical missorting of basolateral proteins; instead, one would expect a nonpolarized distribution of these vesicles. Thus, it appears that myosin VI is involved in cargo sorting or vesicle formation at the donor compartment, and when this function is inhibited, it results in the incorporation of basolateral membrane proteins into apical carriers and their delivery to the apical domain. Why is only the NI isoform of myosin VI involved in basolateral sorting, when both the LI and the SI+LI myosin VI isoforms are also expressed in polarized MDCK cells? Our PCR results indicate that the NI and LI are the major isoforms expressed in polarized MDCK cells, and that the SI+LI is only expressed to a lesser extent. The data showing the localization of the different myosin VI isoforms () indicate that the isoform containing the LI is specifically targeted to the apical domain and is not present in the perinuclear area where basolateral sorting, either in the TGN or the recycling endosome, occurs. The LI enhances targeting of myosin VI to clathrin-coated structures via Dab2 (); therefore, this isoform is believed to play a role in clathrin-mediated endocytosis at the apical domain of polarized epithelial cells (). Thus, differential intracellular targeting appears to be an important mechanism for regulating the intracellular functions of the different myosin VI splice variants expressed together in a single cell type. It is not clear what “advantage” the absence of the SI and LI inserts in the tail confers on myosin VI in respect to basolateral protein sorting. The SI and LI inserts do not contain any obvious structural motifs or apparent binding or regulatory sites. However, these inserts are located in or next to the C-terminal “cargo-binding” tail domain, which is responsible for the cellular targeting of myosin VI, and thus may have an effect on the conformation of the tail domain, and possibly on the whole molecule. In accordance with this hypothesis, recent studies have shown that the cargo-binding C-terminal tail domain of myosin Va interacts with the motor “head” domain regulating myosin Va motor activity (; ; ). Thus, structural, functional, and single-molecule imaging studies are underway to determine the importance of the inserts (or their absence) on the regulation and function of myosin VI. In conclusion, we have demonstrated that in polarized epithelial cells, myosin VI is important for sorting and basolateral delivery of newly synthesized membrane proteins that use the AP-1B–dependent pathway. However, the exact cellular locations, the mechanistic details, and whether, for example, the myosin VI NI isoform operates as a nonprocessive monomer or processive dimer, its precise interacting components, and the regulatory signals used are just a few of the unknowns that need to be determined. The affinity-purified rabbit polyclonal antibodies to the whole tail (α-MVI), globular tail of myosin VI (α-MVI-GT), human full-length optineurin (; : ), and mouse monoclonal antibody to rat endolyn (501; ) have been previously described. The mouse monoclonal antibody to human CD8 was obtained from M. Seaman (Cambridge Institute for Medical Research [CIMR], Cambridge, UK); the mouse monoclonal anti-HA Fc125 tissue culture supernatant was obtained from O. Weisz (University of Pittsburgh, Pittsburgh, PA); the mouse monoclonal antibody to the luminal domain of VSV-G was obtained from R. Pepperkok and J. Simpson (European Molecular Biology Laboratory, Heidelberg, Germany); and the rabbit polyclonal antibody to AP-1 was obtained from M. Robinson (CIMR, Cambridge, UK). The following commercial antibodies were used: a mouse monoclonal antibody to GFP (Abcam), a rabbit polyclonal antibody to GFP (Abcam), rabbit anti–mouse (DakoCytomation), protein A gold (Cell Microscopy Centre, Utrecht, Netherlands), a mouse monoclonal antibody to human CD32/FcγRII (R&D Systems), a mouse monoclonal antibody to E-cadherin (Developmental Studies Hybridoma Bank, Iowa City, Iowa), a mouse monoclonal antibody to VSV-G (Sigma-Aldrich), and a mouse monoclonal antibody to human TfR (5E9C11; American Type Culture Collection). RT-PCR analysis was used to assess the expression of myosin VI isoforms in nonpolarized MDCK cells grown on plastic and in polarized MDCK cells grown on transwell filters, as described in . The cDNAs of endolyn (), CD8-LDLR (a gift from M. Seaman), HA (a gift from O. Weisz), FcγRIIB (a gift from A. Floto, CIMR, Cambridge, UK), VSV-G tsO45 (a gift from R. Duden, University of London, London, UK), and human TfR were amplified by PCR and inserted into the pShuttle-CMV vector (Stratagene) for preparing recombinant adenoviruses. Recombinant adenoviruses were produced using the AdEasy Adenoviral vector system (Stratagene) and used to infect filter-grown MDCK cells. MDCK cells grown for 3 d on filters were washed twice with PBS++ (PBS supplemented with 0.9 mM CaCl, 0.52 mM MgCl, and 0.16 mM MgSO) before adding 10 plaque-forming units/cell of purified recombinant viruses diluted in PBS++ (50 μl for 6.5 mm filters; 200 μl for 12 mm filters) to the apical compartment, while the basolateral compartment was incubated with only PBS++ (600 μl for 6.5-mm filters; 1.5 ml for 12-mm filters). After 1.5 h at 37°C, the virus-containing medium was replaced with normal growth medium, and the cells were incubated for an additional 18 h before performing the biochemical assays. MDCK II cells were cultured at 37°C under 5% CO in D-MEM (Invitrogen) containing 10% (vol/vol) fetal calf serum (HyClone), 100 U/ml penicillin, and 0.1 mg/ml streptomycin (Invitrogen). The cells were plated at 10 cells/cm on 0.4 μm polycarbonate membrane Transwell filters (Corning Costar) and grown for 4 d with a daily medium change. Cell polarization was monitored by measuring the transepithelial electrical resistance using a Mitchell-ERS epithelial voltameter (Millipore). Myosin VI, the tail domains, or full-length Rab8 tagged with GFP at the N-terminus were cloned into the ΔpMEP4 vector, which controls protein expression by the inducible metallothionein IIA promotor. Protein expression was induced for 24 h with 100 μM ZnCl. To generate stable MDCK cell lines, transfected cells were selected with 200 μg/ml Hygromycin (Roche). Single clones were isolated, and expression of the various myosin VI isoforms was checked by immunofluorescence and immunoblotting with an antibody to the myosin VI tail. The four best individual clones for each construct were grown in selective medium and the population of highly expressing cells enriched by FACS sorting. Indirect immunofluorescence staining was performed as described in . For TfR antibody uptake experiments, MDCK cells were infected with virus containing human TfR overnight, serum starved for 1 h, and incubated for 30 min in the basolateral compartment with the monoclonal antibody to TfR, before fixation and processing for EM. For cell surface staining of VSV-G, infected MDCK cells were incubated at 39.5°C overnight to accumulate VSV-G in the ER, followed by 31°C for 2 h in the presence of 100 μg/ml cycloheximide (Sigma-Aldrich). MDCK cells expressing either the CD8-LDLR chimera or the FcγRIIB were incubated at 37°C overnight, followed by a 20°C block for 3 h to accumulate proteins in the Golgi complex in the presence of cycloheximide before a temperature shift to 37°C enables protein exit from the Golgi complex to the cell surface. To visualize proteins on the cell surface the cells were fixed and before permeabilization incubated with antibodies recognizing the luminal domain of VSV-G, CD8, or the FcγRIIB. Cells were visualized in a Radiance Plus confocal scanning microscope with a 63×/NA 1.4 objective at a resolution of 1,024 × 1,024 pixels (Bio-Rad Laboratories). Photoshop software (Adobe) was used for image processing. Polarized MDCK II cells were fixed in a mixture of 2% paraformaldehyde and 0.2% glutaraldehyde in PBS for 2 h. The cells were washed with PBS containing 0.02 M glycine to quench aldehydes, scraped off the filter, centrifuged, and embedded in 12% gelatin in PBS. Small blocks of embedded cells were incubated overnight with 2.3 M sucrose at 4°C, mounted on aluminum pins, and frozen in liquid nitrogen. Ultrathin 70-nm cryosections were cut at −120°C and picked up with a mixture of 1% methylcellulose in 1.15 M sucrose. The sections were processed for labeling using previously described protocols (). For S labeling, MDCK cells grown on transwell filters were starved for 30 min in Met-/Cys-free MEM (Sigma-Aldrich) containing 10% dialyzed FCS (Sigma-Aldrich), radiolabeled for 20 min with 0.1 mCi Redivue PRO-MIX (GE Healthcare), and then chased for different times in D-MEM (Invitrogen) supplemented with 5× Met and Cys and 10% FCS. The side of the Transwell to be biotinylated was incubated twice for 20 min with 1.5 mg/ml Sulfo-NHS-Biotin derivatives in 1 ml PBS. The reaction was quenched by addition of 100 mM glycine in PBS. The solubilization and immunoprecipitation procedures are described in . Samples were separated by 10% SDS PAGE, and the dried gel was exposed to a phosphorimaging screen for 24 h. The screen was scanned with a Cyclone phosphorimager (Perkin Elmer), and the intensity of the bands was quantitated using OptiQuant software. The percentage of a reporter protein reaching the cell surface relative to the total cell surface biotinylated protein was calculated by dividing the biotinylated protein on that cell surface by the total biotinylated protein. Values are averages from four experiments using two different stable clones and are presented with standard deviations. Fig. S1 shows graphical representations of the amount of VSV-G at the cell surface. Fig. S2 shows that overexpression of GFP-myosin NI tail does not disrupt AP-1 localization in MDCK cells. The online version of this article is available at .

The endocytic system fulfills several vital roles for the cell, including internalization and degradation of macromolecules, down-regulation of activated signaling receptors, and spatial regulation of intracellular signaling. Another important function of the endocytic system is to regulate the protein composition of the plasma membrane through remodeling. For instance, the number of plasma membrane–localized transporters and receptors present at any one time arises from multiple sorting decisions made within the endosomal system (). Indeed, given that constitutive or bulk endocytosis is constantly occurring, the steady-state localization of several plasma membrane proteins is maintained by endocytic recycling, a process in which internalized proteins are returned to the cell surface (). However, little is known regarding the cis-acting sorting signals and trans-acting protein machinery that mediate recycling of plasma membrane proteins. After internalization, recycling integral plasma membrane proteins are sorted into transport carriers that bud from a tubular early endosomal compartment. These vesicles are then targeted either directly to the plasma membrane or to an intermediate destination (i.e., the endocytic recycling compartment or the Golgi apparatus) before cargo is delivered back to the plasma membrane. In yeast, the best-characterized plasma membrane recycling pathways involve transit of cargo through the Golgi for resecretion. For example, endocytic recycling of the exocytic v-SNARE Snc1p involves endosome-to-Golgi transport, followed by export from the Golgi to the plasma membrane (; ). Therefore, although the pathways that various cargo proteins take in returning back to the plasma membrane may be complex and involve multiple organelles, it is clear that sorting of cargo at endosomes is a common and key event in determining whether a particular protein will be recycled or targeted for degradation. Considerable insight has been gained into the sorting, export, and retrieval of cargo proteins from endosomes. An evolutionarily conserved protein complex, known as retromer, has received much attention for its role in this process (). In yeast, retromer is composed of five subunits that form two subcomplexes: a complex thought to perform cargo selection, composed of Vps26p, Vps29p, and Vps35p, and a complex that mediates membrane association, composed of Vps5p and Vps17p (; , ; ). Retrieval of the vacuolar hydrolase receptor Vps10p and the late Golgi proteins Kex2p and Ste13p from the prevacuolar compartment back to the Golgi has been shown to be dependent on a functional retromer complex (; ). In mammals, retromer is composed of hVps26, hVps29, hVps35 (), and the Bin/Amphiphysin/Rvs (BAR) domain–containing proteins Snx1 and Snx2 (). Retromer also functions in several trafficking pathways, including transcytosis of the polymeric immunoglobulin receptor () and retrieval of cargo, such as the cation-independent mannose-6-phosphate receptor from late endosomes back to the Golgi (; ). Interestingly, recent studies in have also implicated retromer function in establishing long-range Wnt signaling gradients, expanding the role of retromer to include developmental patterning (; ). Thus, evidence indicates that retromer participates in endocytic sorting of multiple distinct cargos, but it is not yet known how such diverse cargo is selected by retromer. The retromer components Vps5p and Vps17p (as well as the mammalian homologues Snx1 and Snx2) are members of the sorting nexin family of proteins, or SNXs, which has been broadly implicated in sorting within the endosomal system (; ; ). This protein family is characterized by the presence of a SNX-PX (phox homology) domain that binds preferentially to phosphatidylinositol-3-phosphate (PtdInsP), which is a lipid enriched on endosomal membranes (). Currently, 28 mammalian and 10 yeast sorting nexin proteins have been identified (; ), and most of the data regarding the functions of these proteins in yeast points to their role in endocytic sorting. Snx4p, Snx41p, and Snx42p, for instance, form a complex that retrieves Snc1p from early endosomes (). Yeast cells that lack Grd19p (the yeast homologue of human Snx3) or cells that lack retromer mislocalize Ste13p and Kex2p to the vacuole (; ). In addition, Vps5p, Vps17p, and Mvp1p (another sorting nexin), have been shown to be essential for retrieval of Vps10p from late endosomes (; ). Thus, there is evidence supporting the notion that these sorting factors cooperate to fulfill their functions in protein retrieval and recycling. In , sorting decisions within the endosomal system have been shown to regulate the abundance, stability, and localization of various nutrient transporters. These sorting decisions are generally based on cues regarding the availability of particular nutrients. The general amino acid permease Gap1p, for instance, is highly expressed at the plasma membrane when cells are grown in media containing a poor nitrogen source, but is rapidly internalized when high concentrations of amino acids are present (). Similarly, the uracil permease Fur4p is down-regulated when cells are exposed to toxic concentrations of uracil (). We have used the yeast high-affinity iron transporter (composed of the proteins Fet3p and Ftr1p) as a model to investigate endocytic sorting events in response to a limiting concentration of extracellular iron. We demonstrate that the plasma membrane localization of Fet3p–Ftr1p in the absence of iron is maintained by endocytic recycling and that the sorting nexin Grd19p mediates this recycling via interactions with both Ftr1p and the retromer complex. The high-affinity reductive iron transporter in is composed of two proteins, Fet3p and Ftr1p (; ), whose biosynthesis and localization are regulated by the concentration of extracellular iron. When iron is limiting in the growth medium, the protein complex is highly expressed and is localized to the plasma membrane. However, when the extracellular concentration of iron is high, transcription ceases and the protein complex undergoes regulated endocytosis and delivery to the lysosome-like vacuole, where it is degraded (). When cells are grown in media containing intermediate concentrations of iron (10–100 μM), transcription of and is barely detectable, yet the abundance of the Fet3 and Ftr1 proteins remains high and the proteins are localized to the plasma membrane (). As the abundance of several other nutrient transporters is regulated by protein sorting within the endosomal–lysosomal system (; ), we hypothesized that the localization and stability of Fet3p and Ftr1p might be due, at least in part, to recycling of endocytosed Fet3p–Ftr1p. According to this hypothesis, Fet3p–Ftr1p are internalized constitutively and are recycled back to the plasma membrane under low-iron conditions; if they were not recycled, they would be delivered to the vacuole and degraded. To test our hypothesis, a GFP tag was integrated upstream of the stop codon of the locus in a collection of yeast deletion mutants with known defects in protein trafficking within the endocytic system (Table S1, available at ), and this collection was visually screened to identify mutants in which Ftr1p-GFP was partially localized to the vacuole when cells were grown in iron-deficient medium. This screen identified nine genes required to restrict Ftr1p-GFP localization to the plasma membrane under these conditions (). Five of these genes, , , , , and , encode subunits of retromer, which is a protein complex that has been implicated in the trafficking of cargo proteins from endosomes to the Golgi (, ). Another gene, /, encodes a member of the sorting nexin family of proteins, and it has also been implicated in trafficking from endosomes to the Golgi (). The other three genes, , , and , encode components of the Ypt6 Golgi Rab GTPase module that functions at the Golgi in endosome-to-Golgi trafficking pathways (Fig. S1 A; ; ). In all of these mutants, Fet3p-GFP was also mislocalized to the vacuole (Fig. S1 B). We have focused here on investigating the roles of Grd19p and retromer in endocytic recycling. In principle, Grd19p and retromer could be functioning in one of two ways to localize Ftr1p to the plasma membrane under iron-limiting conditions. They could act in the biosynthetic pathway to deliver Fet3p–Ftr1p to the plasma membrane, or they could function after Fet3p–Ftr1p has been delivered to the plasma membrane to maintain them there. These models can be distinguished experimentally because in the latter model endocytosis would be required for delivery of Ftr1p-GFP to the vacuole in the mutants. To test this, double mutants were constructed in which the endocytosis-defective allele was combined with Δ or a retromer gene deletion, Δ, and Ftr1p-GFP localization was determined. In both cases, no vacuolar GFP signal was observed with the double mutants (), indicating that the mutation is epistatic to Δ and Δ. As an independent test of these observations, we used quantitative immunoblotting of cell extracts to determine steady-state levels of Ftr1p-GFP in all of these strains, in addition to the double mutant strains Δ Δ and Δ Δ. In all the retromer and mutant strains, the amount of Ftr1p-GFP was significantly reduced compared with a wild-type control extract, and the mutation substantially restored levels of detected Ftr1p-GFP. These results indicate that, in these mutants, Ftr1p-GFP must first be delivered to the plasma membrane to be subsequently delivered to the vacuole and implicate retromer and Grd19p in a recycling pathway that maintains Fet3p–Ftr1p on the plasma membrane. These results further suggest that retromer and Grd19p are likely functioning together and do not act independently or in a sequential manner, as the Grd19p-retromer double mutants have reduced steady-state levels of Ftr1p-GFP that are essentially identical to the single mutants. Because endocytic recycling signals are not well characterized, we sought to design a systematic strategy to identify those features of Fet3p and Ftr1p that are responsible for directing endocytic recycling. We designed a novel assay based on the trafficking itinerary of the vacuolar protein sorting receptor, Vps10p, which cycles between endosomes and the Golgi to mediate the sorting of cargo proteins to the vacuole (; ). After delivery of Vps10p–cargo complexes to the late endosome, retrieval of Vps10p back to the Golgi requires retromer, which recognizes a retrieval signal in the C-terminal cytoplasmic tail of Vps10p (). Removal of the cytoplasmic tail results in vacuolar delivery and degradation of Vps10p, and, as a result, cargos normally destined for the vacuole, such as carboxypeptidase Y (CPY) are instead secreted (; ). Because retromer is required for the stability of both Vps10p and Fet3p–Ftr1p, we reasoned that Fet3p–Ftr1p contains at least one endosome-to-Golgi sorting signal. This served as the basis of an assay aimed at identifying Fet3p–Ftr1p sequences, which can restore function to a Vps10p mutant lacking the cytoplasmic tail. We focused our analysis on Ftr1p because experiments, which are described in the following sections, indicated that Fet3p does not contain any post-Golgi sorting signals. Ftr1p contains four cytoplasmic portions; three of these are loops connecting membrane-spanning segments and the fourth is a C-terminal tail (). The third loop of Ftr1p, connecting the fifth and sixth membrane-spanning segments, was not investigated because it is predicted to consist of, at most, four amino acids. Fusion proteins in which the cytoplasmic tail of Vps10p was replaced with sequences encoding the remaining cytoplasmic regions of Ftr1p were constructed and integrated at the endogenous locus. All constructs also contained 13 copies of the Myc epitope tag (13xMyc) at the C terminus. As controls, full-length Vps10p was tagged with a 13xMyc tag at its C terminus, and a deletion mutant that removed the entire cytoplasmic domain of Vps10p was constructed by integrating a 13xMyc epitope tag after Arg1421 (). Two different methods were used to analyze the functionality of the Vps10p-Ftr1p chimeric proteins relative to the control constructs. Vps10p mutants that are not recycled, such as Vps10pΔtail, are delivered to the vacuole and degraded, resulting in low steady-state levels of the protein (; ). Therefore, we compared the steady-state levels of the Vps10p-Ftr1p fusion proteins to full-length Vps10p and Vps10pΔtail in a wild-type strain and a Δ strain that lacks the predominant vacuolar protease, and thus is severely compromised in vacuolar degradation (). In wild-type cells, the amount of the Vps10p-Ftr1p(C-tail) construct was nearly as high as native Vps10p, whereas the amounts of Vps10p-Ftr1p(loop1) and Vps10p-Ftr1p(loop2) proteins were similar to the level of Vps10pΔtail. The low amounts of these proteins were rescued in the Δ strain, indicating that the low levels were caused by proteolysis in the vacuole. As a second test, we quantified the secretion of CPY from these strains using a CPY colony blot assay (; ). The results show that CPY secretion by the Vps10p-Ftr1p(C-tail) strain was rescued by 33% compared with the Vps10pΔtail strain. No significant rescue of CPY sorting was observed in the Vps10p-Ftr1p(loop1) and Vps10p-Ftr1p(loop2) strains. It is unclear why the Vps10p-Ftr1p(C-tail) chimeric protein is not more effective at rescuing CPY sorting; however, it is likely that the Ftr1p sequence lacks TGN sorting signals that are present in Vps10p. Nevertheless, the corroborative results of these two assays suggest that the C-terminal tail of Ftr1p contains at least one signal that can partially substitute for the Vps10p endosome–Golgi retrieval signal. To further map the recycling signals within the cytoplasmic tail of Ftr1p, a series of truncation mutants was systematically constructed by integrating GFP at various points within the locus and analyzing the localization of these truncated proteins under iron-limiting conditions. Truncation after residue 318 (Ftr1pΔ318) resulted in localization to the lumen of the vacuole in addition to localization at the plasma membrane. However, a mutant truncated after residue 328 (Ftr1pΔ328) was localized solely to the plasma membrane, implicating amino acids 319–328 as important for endocytic recycling. To confirm a requirement for this region, an internal deletion mutant lacking residues 319–328 (Ftr1pΔ319-328) was constructed and localized to the plasma membrane and vacuole lumen, although missorting of this internal deletion mutant was not as severe as the Ftr1pΔ318 mutant (), suggesting that other portions of the Ftr1p C-tail may also play a role in recycling. Importantly, none of the truncated proteins accumulated in the ER, indicating that their folding and association with Fet3p was not grossly affected by removal of these sequences. To interpret these results from the perspective of endocytic sorting signals, it was important to distinguish if the Ftr1pΔ318-GFP–truncated protein was being targeted to the vacuole via endocytosis or the biosynthetic route. We again made use of the strain to determine the localization of the Ftr1pΔ318-GFP construct under iron-starvation conditions. In the strain, Ftr1pΔ318-GFP was localized exclusively to the plasma membrane (), and these results were confirmed by immunoblotting of cell extracts. Less Ftr1p was detected when the cytoplasmic tail of the protein was truncated, and levels of the truncated protein were stabilized in the strain, as well as in a Δ strain (). The results indicate that this Ftr1p truncation mutant is trafficked properly to the plasma membrane, but it is not maintained there. Together, these results demonstrate that aa 319–328 of Ftr1p are important for endocytic recycling. The delivery of newly synthesized Fet3p and Ftr1p to the plasma membrane depends on their coexpression, and iron-induced down-regulation of Fet3p parallels Ftr1p, indicating that these two proteins traffic together throughout the endomembrane system (; ; ). To determine if the aberrant trafficking of Ftr1p truncation mutants also impacts the trafficking of Fet3p, we examined the localization of Fet3p-GFP in the context of the Ftr1p-tail truncation. We integrated a GFP tag on Fet3p in strains expressing full-length Ftr1p or Ftr1pΔ318 and determined the localization of Fet3p-GFP under iron-limiting conditions by fluorescence microscopy. Fet3p-GFP was restricted to the plasma membrane when coexpressed with full-length Ftr1p, but it was also localized to the vacuole lumen (and a very minor fraction in the ER) when coexpressed with Ftr1pΔ318 (). In a strain with a deletion of most of the Fet3p cytoplasmic tail (deletion of the last 45 amino acids), Ftr1p-GFP was localized to the ER and the plasma membrane (), as expected from a previous study (), suggesting that the Fet3p cytoplasmic tail is required for efficient ER export, but does not influence post-ER trafficking. However, in a strain with deletions of both the Fet3p and Ftr1p C-terminal tails (C-tails), both proteins were localized to the ER, plasma membrane, and the vacuole lumen (), indicating that once exported from the ER, the complex was sorted to the vacuole in the absence of the Ftr1p recycling signal. Collectively, the results indicate that a region of the C-terminal cytoplasmic tail of Ftr1p (aa 319–328; GHLPFTKNLQ) is necessary for endocytic recycling of both Fet3p and Ftr1p under iron-depleted conditions. We next sought to address whether any of the endosomal sorting factors identified in the initial screen recognize the Ftr1p recycling signal. Previous work has shown that epitope-tagged Grd19p present in cell extract, but not the retromer subunit Vps5p, can be captured by a GST-Ste13p fusion protein containing the entire Ste13p cytoplasmic domain (). We first tested binding of purified recombinant Grd19p to a panel of GST fusion proteins containing each of the cytoplasmic regions of Fet3p and Ftr1p, as well as several smaller portions of the Ftr1p C-tail (). Grd19p bound the Ftr1p C-tail, but not any of the other cytoplasmic portions of Ftr1p or Fet3p. Remarkably, Grd19p also bound to a GST fusion protein consisting of only aa 319–328 of Ftr1p, but did not bind to the Ftr1p C-tail deletion mutant lacking these residues, narrowing the site of Grd19p interaction to aa 319–328 of Ftr1p. A recent study demonstrated that the C-terminal tails of Fet3p and Ftr1p physically associate via a region that includes the Grd19p binding site of Ftr1p (), and we therefore tested if binding of Grd19p to the Ftr1p cytoplasmic tail in vitro is influenced by coincubation with the Fet3p cytoplasmic tail, revealing that it was not (unpublished data). Finally, we also tested if the retromer complex binds to the recycling signal in the Fet3p–Ftr1p complex using cell lysate prepared from a strain expressing Vps29p- myc; however, no Vps29p was captured by any of the GST fusion proteins (unpublished data). Although this does not rule out that retromer does not bind Ftr1p, all of the binding experiments indicate that Grd19p specifically recognizes the Fet3p–Ftr1p endocytic recycling signal. To confirm an interaction between Ftr1p and Grd19p in vivo, we attempted to coimmunoprecipitate endogenous epitope-tagged Ftr1p and Grd19p from cell extract; however, we were unable to detect a binding interaction using this approach, almost certainly because of the observation that, at steady-state, only a very small proportion of Ftr1p-GFP resides within endosomal compartments (). To circumvent this issue, we overexpressed the indicated GST-Ftr1p C-tail fusion proteins or GST alone in a yeast strain expressing an endogenous epitope-tagged mutant form of Grd19p, Grd19p-R81A-myc (). Arginine81 lies within the highly conserved PtdInsP-binding pocket in the PX domain of Grd19p, and mutation of this residue to alanine results in mislocalization of the protein to the cytosol (Fig. S2 A, available at ). Cells grown in synthetic media were converted to spheroplasts, the bifunctional membrane-permeable chemical cross-linker dithiobis succinimidyl proprionate (DSP) was added before lysis, and the GST fusion proteins were captured on glutathione–Sepharose beads. The purified material was then probed with antibodies to Grd19p-R81A-myc. The results demonstrate that Grd19p-R81A was only coprecipitated from cells expressing the GST-Ftr1p C-tail (315–404) fusion protein, but not from cells expressing the GST-Ftr1p C-tail (329–404) construct, which lacks the putative endocytic recycling signal. These results indicate that Grd19p can bind the Ftr1p C-tail in cells and further confirms that this interaction requires aa 319–328 of Ftr1p. The most straightforward model to explain the dual requirement for Grd19p and retromer in Fet3p–Ftr1p sorting is that they function together at a common sorting step, and this predicts that some portion of Grd19p and retromer colocalize on endosomes. To test this, we constructed strains that expressed functional, C-terminally tagged Grd19p-GFP and Vps17p-RFP, and then imaged the entire volume of these cells by spinning disc confocal microscopy. Three-dimensional reconstructions and two-dimensional maximum projections were generated from the image z stacks ( and Video 1). The majority of labeled organelles contained both Grd19p-GFP and Vps17p-RFP, although some organelles were labeled by only one of the tagged proteins. The three-dimensional representations allowed us to characterize the morphology of the endosomes labeled by Grd19p and Vps17p, many of which had a distinct tubular shape. Retromer proteins have been localized to the tubular domains of early sorting endosomes in cultured mammalian cells (; ; ; ; J.G. ), and this is thought to reflect their roles in a geometric sorting mechanism that segregates integral membrane proteins from lumenal content (). Furthermore, human Snx3 is also enriched on the tubular domains of endosomes (). Our results clearly indicate that a substantial proportion of Grd19p and the retromer subunit Vps17p colocalize on tubular endosomes, suggesting that they may function together, although the exclusive presence of each protein on some endosomes indicates that they function autonomously as well. We also tested the idea that retromer might be required for association of Grd19p with endosomal membranes, as many phosphoinositide-binding modules require other interactions for stable association with intracellular membranes (). Grd19p consists of a PtdInsP-binding PX domain with a short N-terminal extension (). PtdInsP-binding is critical for endosomal targeting and function, as the R81A point mutant resulted in missorting of Ftr1p-GFP to the vacuole (Fig. S2 B). We examined localization of Grd19p-GFP in Δ cells and found that it still localized to endosomes, although the compartments decorated by Grd19p appeared smaller and more numerous (Fig. S2 C). We conclude that retromer does not influence recruitment of Grd19p to endosomes from the cytosol, but it may influence the cellular distribution of Grd19p-positive endosomes, and these results further indicate that Grd19p and retromer are targeted to endosomes independently. In the course of these experiments, we also determined the relative localization of Vps17p to another sorting nexin, Snx4p, which functions in a retromer-independent early endosome-to-Golgi pathway () and is not required for Fet3p–Ftr1p recycling. Although yeast retromer is considered to function solely in late endosome-to-Golgi trafficking pathways, a substantial degree of colocalization between Snx4p-GFP and Vps17p-RFP was observed (). These observations suggest that the localization of retromer in the yeast endocytic system is more widespread than currently appreciated. In addition, we visualized fluorescently labeled Grd19p and Snx4p within the same cell (). Grd19p-GFP– and Snx4p-RFP-labeled organelles exhibited little to no colocalization, indicating that Grd19p and Snx4p are localized almost exclusively to distinct subsets of endosomes. This is consistent with published studies reporting that these two sorting nexins function in retrieval pathways from different populations of endosomes (). Based on the results so far, we hypothesized that Grd19p and retromer may physically cooperate, with Grd19p linking cargo recognition to retromer-dependent export from endosomes. To test this, we purified epitope-tagged Vps29p retromer subunit from a cell lysate and determined if Grd19p copurified with it. Strains were constructed that expressed 3xHA epitope-tagged Vps29p and Grd19p tagged with a 13xMyc epitope, and as a control, Vps29p-3xHA and another retromer subunit, Vps17p, tagged with a 13xMyc epitope. Cells were converted to spheroplasts and lysed, and Vps29p was immunopurified under native conditions, and the precipitates were probed with antibodies to Grd19p-myc. Under these conditions, no Grd19p-myc copurified with Vps29p-HA, but Vps17p-myc did (unpublished data). However, when DSP was added to intact spheroplasts before lysing the cells, both Grd19p and Vps17p copurified with Vps29p (). As a control for this experiment, we used a strain that simultaneously expressed Vps29p-HA and Snx4p-myc and found that Snx4p did not copurify with retromer (), despite the fact that substantial amounts of Snx4p and retromer colocalized on endosomal membranes. Furthermore, cross-linking depended on colocalization of Grd19p and retromer on endosomes as the Grd19p-R81A mutant, which is localized to the cytosol (Fig. S2 A), did not cross-link to retromer (). These results indicate that Grd19p and retromer interact in vivo on PtdInsP-containing endosomes. Our results suggest that the sorting nexin Grd19/Snx3p functions as a cargo-specific accessory component of the retromer complex, which is required for endocytic recycling of the Fet3p–Ftr1p iron transporter. In support of this model, we report that in null mutants of and each of the five retromer subunits, Fet3p–Ftr1p is missorted to the vacuole when cells are grown under conditions that favor recycling, that Grd19p directly binds a sequence in Ftr1p required for endocytic recycling, that Grd19p-GFP and Vps17p-RFP colocalize substantially on tubular endosomes, and that Grd19p can be chemically cross-linked to the retromer complex in vivo. Although the full range of functions and the specific mechanisms by which retromer operates in membrane trafficking have not been elucidated, it is clear that it is a general endosomal sorting factor required for the proper sorting and export of a diverse set of cargo molecules from endosomes. The work presented in this study is relevant for understanding how cargo is identified by retromer, and the results suggest that that the repertoire of retromer-dependent cargos is extended by its interaction with Grd19p. We speculate that Grd19p functions with retromer in a manner analogous to vesicle coat protein adapters that link cargo selection to coat protein recruitment. Recycling of Fet3p–Ftr1p is likely to be initiated through recognition of Ftr1p by Grd19p because the PX domain of Grd19p binds PtdInsP with the highest affinity of all the PX domains encoded in the yeast genome (). In contrast, the PX domains of the retromer subunits Vps5p and Vps17p bind PtdInsP with at least 100-fold lower affinity than Grd19p (). Preliminary analysis of the binding interaction between Grd19p and the Ftr1p recycling signal by surface plasmon resonance indicates that the affinity is relatively weak ( between 10 and 100 μM; unpublished data), which is consistent with PtdInsP binding providing the driving force for recruitment of Grd19p to endosomes. Because the abundances of Grd19p and Vps17p appear to be similar (), Grd19p is expected to preferentially accumulate on early endosomes containing relatively low amounts of PtdInsP. Retromer will subsequently load onto endosomes as they accrue higher levels of PtdInsP during maturation, facilitated by interactions with other factors, such as Grd19p and cargo molecules. In this manner, Grd19p could serve as both a coincidence sensor that detects the presence of Fet3p–Ftr1p on PtdInsP-containing endosomes and as an adaptor that recruits retromer to cargo to initiate export from the endosome. Grd19p and retromer appear to be sufficient for export from the endosome because we have not observed any recycling defects in other known cargo-sorting factors, including the clathrin adaptor AP-1 complex, other sorting nexins, or the recently identified GSE–EGO complex involved in endosome-to-plasma membrane sorting of the general amino acid permease Gap1p (; Table S1). Once exported from the endosome, Fet3p–Ftr1p is probably delivered to the Golgi for resecretion because deletion of the Golgi Rab GTPase Ypt6p and its regulators also results in defective recycling, and because the Ftr1p C-terminal tail can direct Vps10p from the endosome back to the Golgi. In yeast, five proteins have been identified that are sorted via the Grd19p–retromer pathway. These include native proteins that cycle between the Golgi and endosomes, Ste13p, Kex2p, and Pep12p (; ), and Fet3p–Ftr1p, which uses the Grd19p–retromer pathway to be sorted back to the plasma membrane. The Golgi retrieval signals in Ste13p and Kex2p contain key aromatic residues, although their relevance to Grd19p-mediated sorting is not clear because Grd19p (supplied in a cell extract) still bound well to the cytoplasmic tail of Ste13p, even when the aromatic residue-based retrieval signal had been deleted (). Consistent with this, mutation of the single aromatic residue within the Ftr1p recycling signal (Phenylalanine323 to Alanine) did not affect recycling (unpublished data), so a more systematic analysis of this signal is required to identify its key features. It is also interesting that vacuolar targeting of the Ftr1p mutant lacking the Grd19p binding site (Ftr1pΔ319-328) was not so robust, perhaps implying that other sorting determinants are present within the C-terminal tail of Ftr1p. Inasmuch as the available data suggest that the sequences of the signals which confer Grd19p- and retromer-dependent trafficking are diverse, Grd19p probably recognizes structural features of cargo proteins rather than a strict linear amino acid sequence. Importantly, our results showing that Grd19p directly recognizes Ftr1p and Ste13p and links them to retromer establish for the first time how a sorting nexin (other than Snx1) and retromer cooperate to recognize cargo. Although direct interactions between cargo and any subunit of retromer have not yet been confirmed using purified proteins, the current view posits that cargo is recognized by Vps35p, although other sorting nexins, including human Snx1, also have the capacity to bind cytoplasmic regions of some endocytic cargo proteins (). Moreover, the recent discovery that Vps26 has an arrestin-like structure raises the possibility that Vps26p, like arrestins, may also interact with cargo (). Another possible function of retromer is suggested by the crystal structure of Vps29, which has a protein fold resembling that of phosphoesterases (; ), and recent studies have shown that Vps29p exhibits protein phosphatase activity (). Regardless of the specific functions of the individual retromer subunits, it is clear that multiple cargo recognition mechanisms must contribute to the general function of retromer in endosomal sorting because Grd19p is not required for all retromer-dependent trafficking. In mammalian cells, the early endosomal system is comprised of vacuolar domains connected to an extensive network of tubules, which are enriched in integral membrane cargo proteins that are subsequently sorted to a variety of different organelles (). On the basis of the large surface area/volume ratio of tubes compared with spherical structures, it has been proposed that the packaging of integral membrane proteins into tubes is a highly efficient geometry-based mechanism for segregating membrane and lumenal components (; ; ). Human retromer appears to be involved in this sorting mechanism through its cargo-binding activities and the ability of the BAR domain–containing Snx1 subunit to sense regions of high membrane curvature (). The role of this geometric sorting mechanism in yeast is not known due, in large part, to the difficulty in visualizing cargo within domains of yeast endosomes by light microscopy. However, visualization of yeast endosomes by electron microscopy has provided clear evidence that a subset of them is indeed tubular in shape (; ). These results support the notion that retromer-mediated sorting in yeast and mammalian cells involves the same mechanisms, and they open the door to evaluating the roles of various retromer proteins and auxiliary factors, such as Grd19p in the biogenesis of these organelles. We expect that the interaction between Grd19p and retromer in yeast holds true for the human orthologues, as this interaction could explain the observation that overexpression of human Snx3 in cultured cells leads to a huge expansion of tubular early endosomal compartments through enhanced recruitment of the BAR domain–containing Snx1 component of retromer (). The results presented in this work demonstrate that recognition of recycling protein cargo by retromer can be initiated by a sorting nexin (Grd19p) that functions as an adaptor to link cargo to the cellular recycling machinery. With the identification of the endocytic recycling machinery and insight into how it mediates recycling of Fet3p–Ftr1p, the opportunity now exists to explore how, in response to changes in extracellular iron concentration, the iron transporter is channeled into either recycling or degradative pathways. Unless otherwise indicated, all yeast strains were constructed by integration using recombination of gene-targeted, PCR-generated DNAs using the method of to ensure expression from the native loci. The strain background expressing the allele is SEY6210 (a , , , , , and and the strain background for yeast strains expressing Ftr1p-Vps10p chimeric proteins is BHY10 (SEY6210 ∷pBHY11 [CPY-Inv ]). All other yeast strains were constructed in the BY4742 background (α , , , and . For the initial screen, a GFP tag was integrated in the ORF immediately preceding the stop codon in yeast deletion strains from the EUROSCARF deletion collection. To induce iron starvation, cells were grown overnight to OD ≈ 1.0 in synthetic media containing 50 μM of the iron chelator bathophenanthrolinedisulfonic acid. For all other experiments, cells were grown in either yeast extract/peptone/dextrose, or yeast nitrogen base supplemented with the appropriate nutrients as necessary. Construction of strains coexpressing GFP- and RFP-tagged proteins was accomplished by generating PCR DNA to integrate a RFP tag at the or locus using pKT359 (provided by Kurt Thorn, University of California, San Francisco, San Francisco, CA) as a template (pFA6a-link-tdTomato-HIS3MX6; ) into strains already expressing integrated copies of Grd19p-GFP or Snx4p-GFP (integrated with the cassette using the method of ). Strains expressing Vps10p-Ftr1p chimeric proteins were constructed using recombination of PCR-generated DNAs. All chimeric genes were expressed from the endogenous promoter by integrating sequences after codon 1,421 (immediately downstream of the membrane-spanning segment) of the ORF. The segments were chosen based on a published topology analysis (). To construct the fusion gene, the region encoding Ftr1p aa 315–404, followed by a 13xMyc epitope tag and the selectable marker, was amplified from genomic DNA of strain TSY30 (), and the PCR DNA was used to transform a wild-type strain (BHY10). The chimeras containing portions of that encode internal loops were constructed by cotransformation with two PCR DNAs; one encoding the sequence and a second containing a 13xMyc epitope tag followed by the selectable marker. The DNAs encoding the fusion site were amplified such that the segment was flanked by 40 bases of the locus, and on the other end it was flanked by 40 bases of the Myc epitope tag. The second DNA encoded a 13xMyc epitope tag, followed by the selectable marker, followed by 51 bases of sequence that matched immediately downstream of the stop codon. Recombination between the two PCR DNAs within the Myc epitope tag sequences generated -13xMyc- DNAs flanked by sequences, and recombination of these at the locus generated the chimeric genes. The internal portions encoded aa 30–48 (loop1) and 109–153 (loop 2). For all constructs, His+ transformants were screened by immunoblotting using the 9E10 antibody to identify those which expressed myc-tagged proteins of the correct size. PCR of genomic DNA from these strains using primers upstream of the integration site and within the Myc epitope tag were then used to amplify the junctions and all of the sequences. These PCR DNAs were then sequenced to confirm the constructs. Control strains to compare the functionality of the Vps10-Ftr1p fusion proteins included a Vps10p truncation mutant in which a 13xMyc-epitope tag was integrated after codon 1,421 of the ORF, and full-length tagged on its C terminus with the 13xMyc epitope. Both control strains were constructed by transformation of PCR-generated DNAs using pFA6a-13Myc-HIS3MX6 as the template (). Strains expressing mutant and tagged integrated copies of Grd19p were constructed as follows. The ORF including promoter and terminator regions was amplified by PCR as a BamHI–SalI fragment and cloned into vector pRS416. Site-directed mutagenesis (QuikChange Site-Directed Mutagenesis kit; Stratagene) was used to generate the R81A mutation in this plasmid, and this mutation was confirmed by sequencing. A Δ strain was then cotransformed with the BamHI–SalI fragment released from this plasmid along with or PCR DNA, with flanking sequences targeted to the ORF (). Transformants were screened for growth on plates that lacked histidine, for G418 sensitivity, and for the presence of the GFP tag by microscopy or the 13Xmyc tag by immunoblotting using the 9E10 antibody. Enzymes used in DNA manipulations were purchased from New England Biolabs or Promega. Standard molecular biological and microbiological techniques were used throughout. Primary mouse monoclonal antibodies used in these studies included: 9E10 anti-myc (1:10,000; University of Pennsylvania Cell Center), anti-HA (1:1,000; Covance), anti-PGK (1:10,000; Invitrogen), anti-T7 (1:10,000; Novagen), anti-CPY (1:1,000; Invitrogen), and anti-GFP (1:2,000; Covance). Secondary sheep anti–mouse HRP-conjugated antibodies (GE Healthcare) were used at 1:5,000. This assay is based on the method of , with the following exceptions. Filters were incubated with AP-conjugated goat anti–mouse secondary antibodies (Jackson ImmunoResearch Laboratories) for 2 h at room temperature before development by enhanced chemifluorescence (ECF) and analysis on a Molecular Dynamics Storm 860 PhosphorImager (GE Healthcare). The use of ECF instead of ECL was essential here for obtaining linear signals. The amount of secreted CPY was quantified using ImageQuant software version 5.2 (GE Healthcare) and plotted in graphical form using Excel (Microsoft). GST fusion proteins were constructed by amplifying the indicated sequences by PCR as BamHI–SalI fragments using wild-type genomic DNA as a template. The amplified products were cloned in-frame into the corresponding restriction sites of vector pGEX-KG-KAN (Novagen) for expression in bacteria or vector pEG(KG) () for expression in yeast. All constructs were sequenced to ensure that no mutations were present. For bacterial expression, plasmids were transformed into BL21 (DE3; Novagen). Expression was induced by the addition of 1 mM IPTG at 37°C for 3 h. Cell pellets were resuspended in lysis buffer (PBS [1 mM KHPO, 10 mM NaHPO, 137 mM NaCl, and 2.7 mM KCl, pH 7.4] + 1× Complete Mini protease inhibitor cocktail [Roche] + 1 mM AEBSF [Calbiochem]). Cell pellets were lysed either by sonication or by using a French Press pressure cell, followed by centrifugation (20,000 for 20 min) to produce clarified lysates. For expression in yeast, plasmids were transformed into strain TSY103 (). Expression was induced by overnight growth in synthetic media lacking uracil and containing 4% galactose as the sole carbon source according to a published protocol (). For the production of His-Grd19p, the ORF was amplified by PCR as a BamHI–SalI fragment using wild-type genomic DNA as template. The amplified product was cloned in-frame into the corresponding sites of vector pET28a (Novagen), and the resulting plasmid was sequenced. The plasmid was transformed into BL21 (DE3; Novagen), and expression was induced by the addition of 1 mM IPTG at 37°C for 4 h. The cell pellet was resuspended in binding buffer ([0.5 M NaCl and 20 mM sodium phosphate, pH 7.4] + 1× Complete Mini protease inhibitor cocktail) and lysed using a French Press pressure cell. His-Grd19p was then purified according to a published protocol (). GST fusion proteins were affinity purified on glutathione–Sepharose 4B (GE Healthcare) by rocking for 1 h at room temperature in 0.5 ml PBS. Beads were washed three times with 1 ml PBS, and then resuspended in 300 μl of binding buffer (1X PBS, 1 mM DTT, 5 mM MgCl, and 0.1% Triton X-100). 10 μg of purified His-Grd19p or 250 μg (total protein) of yeast extract expressing Vps29p-myc was added to the beads. Binding reactions were incubated for 2 h at 4°C with rocking. Beads were then gently spun down and washed three times with 300 μl of binding buffer. Proteins bound to the beads were eluted by the addition of 2× sample buffer and were separated by 10% SDS-PAGE followed by Western blotting with anti-T7 or anti-myc (9E10) antibodies. The in vivo cross-linking procedure is based on a published method (). In brief, 50 OD of the indicated strains were grown to log phase (OD ≈ 1.0) in synthetic media, harvested, washed once with water, and converted to spheroplasts. Spheroplasts were resuspended in 1 ml cross-linking buffer (25 mM potassium phosphate, 0.2 M sorbitol, and 1× EDTA-free protease inhibitor cocktail, pH 7.4), and 2–5 mM dithiobis succinimidyl proprionate (DSP; Pierce Chemical Co.) prepared freshly in DMSO was added. Cross-linking reactions were incubated for 30 min at 4°C. 25 mM Tris, pH 7.5, was added for 15 min to quench the reaction, and then 1% Triton X-100 was added to solubilize the material. Tubes were spun at 13,000 for 10 min to remove insoluble debris, and the cross-linked cell lysate was incubated with 25 μl packed volume glutathione–Sepharose beads (GE Healthcare) or anti-HA affinity matrix (Roche) and allowed to rock overnight at 4°C to capture Vps29p-HA or for 2 h to capture the GST fusion proteins. Beads were pelleted and washed five times with 1 ml wash buffer (1× PBS and 1% Triton X-100). Proteins bound to the beads were eluted, cross-links were cleaved with 2× sample buffer containing 20 mM DTT, and eluted proteins were separated by 10% SDS-PAGE, followed by Western blotting with anti-HA and/or anti-myc (9E10) antibodies. Table S1 displays the yeast deletion mutants screened for mislocalization of Ftr1p-GFP when cells were grown in iron-deficient medium. Fig. S1 shows the localization of Fet3p-GFP in retromer and -null mutants, as well as the localization of Ftr1p-GFP and Fet3p-GFP in cells deleted for components of the Ypt6p GTPase module. Fig. S2 shows the cytosolic localization of Grd19p-R81A-GFP, the localization of Ftr1p-GFP in cells expressing wild-type Grd19p vs. Grd19p-R81A, and the localization of Grd19p-GFP in Δ cells. The video displays a three-dimensional reconstruction of live cells coexpressing Grd19p-GFP and Vps17p-RFP. The online version of this article is available at .

Rapid receptor phosphorylation in response to agonist stimulation is a posttranslational modification adopted by nearly all G protein–coupled receptors (GPCRs; ). This event is generally accepted to be mediated by the GPCR kinase (GRK) family in a process that results in the recruitment of arrestin adaptor proteins to the receptor and the concomitant uncoupling of the receptor from its cognate G protein (). In addition, GRK phosphorylation can promote receptor activation of G protein–independent pathways such as the MAPK cascade (). This universal adaptive paradigm belies the complex nature of GPCR phosphorylation and regulation. There are >340 nonolfactory GPCR subtypes in the mammalian genome () showing widespread tissue distribution and influencing nearly every biological process from sensory perception to cell growth and differentiation (). Many of these receptors are phosphorylated at multiple serine, threonine (; ; ), and occasionally tyrosine residues (). This multisite phosphorylation has been reported in some instances to be hierarchical and mediated by more than one protein kinase (; ; ). Most enlightening have been studies on GRK knockout animals that have suggested that the same receptor subtype expressed in different tissues may be phosphorylated by a different complement of receptor kinases (). It is also the case that many receptor subtypes are found in more than one tissue type () and mediate very specialized tissue-specific responses. For example, the M-muscarinic receptor regulates membrane excitability in neurons (), contraction and cell growth in smooth muscle cells (), and secretary vesicle priming and fusion in salivary acinar cells (; ). It would appear intuitive that receptors expressed in different cell types, controlling specific cellular responses, would be regulated in a manner specific to that cell type. Hence, a tantalizing alternative model of GPCR regulation is that phosphorylation is a flexible process of receptor modification where tissue-specific differences in phosphorylation would underlie defined physiological functions. In this paradigm, differential deployment of receptor kinases in a tissue-selective manner would result in differential phosphorylation that would facilitate the specific physiological role of that receptor in a particular cell type. Our work on the G-coupled M-muscarinic receptor has demonstrated that this receptor subtype can be phosphorylated in an agonist-dependent manner by casein kinase 1α (CK1α), and this process regulates the coupling of the receptor to the extracellular-regulated kinase (ERK) 1/2 pathway (, ; ). These studies established that agonist-dependent GPCR phosphorylation could be mediated by protein kinases other than the GRKs (). In the current study, we extend our investigation of the CKs in GPCR phosphorylation and provide evidence that protein kinase CK2 can also phosphorylate the M-muscarinic receptor. Furthermore, we show that the M-muscarinic receptor is differentially phosphorylated in different cell types and that the action of specific receptor kinases can determine the signaling outcome of receptor phosphorylation. To investigate the role of the CK2 in M-muscarinic receptor phosphorylation, we raised siRNAs against the catalytic α and α′ subunits of CK2. The effectiveness of the siRNAs was established by cotransfection of the duplexes with plasmids expressing HA-tagged α or α′ subunits. In these experiments, we estimated the transfection efficiency of fluorescently labeled siRNAs to be ∼90% (unpublished data). The siRNAs designated CK2α-4 and CK2α′-1p effectively inhibited expression of the α and α′ subunits, respectively (). Furthermore, these siRNAs were active against the endogenously expressed kinase where the levels of the CK2α subunit fell by >85% with no subsequent change in the levels of CK1α, GRK2, GRK3, or GRK6 (). This corresponded to a fall in CK2 enzymatic activity of 68% compared with control (). When used in phosphorylation experiments where the M-muscarinic receptor was immunoprecipitated from CHO-M3 cells labeled with [P]-orthophosphate, the CK2 siRNA duplexes reduced agonist-mediated M-muscarinic receptor phosphorylation by ∼72% compared with scrambled siRNA controls (). These results were confirmed by raising further siRNA duplexes to different regions of CK2α and -α′ (Figs. S1 and S2, available at ). The third intracellular loop of the M-muscarinic receptor is a serine-rich region containing several consensus sites for CK1α, CK2, and the GRKs, many of which are shared between these acidotropic kinases (). The ability of these kinases to phosphorylate the third intracellular loop is illustrated in , where GRK6 was found to phosphorylate the GST-fusion protein containing the third intracellular loop of the human M-muscarinic receptor (R-T, called here GST-H3iloop) with the highest efficiency followed by CK1α and then GRK2 and CK2 (). None of the kinases phosphorylated GST alone (unpublished data). To confirm of a role for CK2 in the phosphorylation of the M-muscarinic receptor, the CK2-specific pharmacological inhibitors 4,5,6,7-tetrabromo-1H-benzotriazole (TBB) and 2-dimethylamino-4,5,6,7-tetrabromo-1H-benzimidazole (DMAT; ; ) were used. The selectively of these inhibitors was confirmed in assays using the fusion protein GST-H3iloop as a substrate for CK1, CK2, GRK2, and GRK6 (). TBB at a concentration of 1 μM was very potent against CK2, reducing the level of GST-H3iloop by ∼75% but had only a small effect on CK1, GRK2, and GRK6 (). DMAT also strongly inhibited CK2 with no significant effect on CK1 or GRK2 but was able to inhibit GRK6 by ∼45% (). These in vitro experiments were consistent with the reported selectivity of the CK2 inhibitors and established that they can discriminate between the putative receptor kinases responsible for M-muscarinic receptor phosphorylation. The CK2 inhibitors were then used on intact CHO-M3 cells to determine their effects on M-muscarinic receptor phosphorylation. Both TBB and DMAT substantially reduced agonist-mediated phosphorylation of the M-muscarinic receptor by 76 and 58%, respectively (). These data are consistent with the aforementioned siRNA experiments and demonstrate a role for CK2 in the phosphorylation of M-muscarinic receptors. To further investigate the ability of CK2 to directly phosphorylate the M-muscarinic receptor, membranes prepared from CHO-M3 cells that express recombinant human receptor were reconstituted with purified CK2. We noted previously that these membranes maintain the ability to phosphorylate the M-muscarinic receptor (, ); hence, to reduce the endogenous kinase activity, the membranes were washed with 200 mM NaCl. Purified CK2 added to these membranes resulted in an increase in receptor phosphorylation, although this was not agonist regulated (). To test whether CK2 had a role in the phosphorylation of the M-muscarinic receptor in a native cell type, we investigated mouse CG neuronal cultures where this receptor subtype is endogenously expressed (). To facilitate these studies, we raised an antibody against the region S-L in the third intracellular loop of the mouse receptor that specifically immunoprecipitated the mouse M-muscarinic receptor when tested against all five muscarinic receptor subtypes (). This antibody was subsequently used in phosphorylation experiments where 7-d-old mouse CG neurons were metabolically labeled with [P]-orthophosphate and stimulated with methacholine before being solubilized and the receptor immunoprecipitated. In CG neurons from wild-type mice, the M-muscarinic receptor appeared as a ∼95-kD phosphoprotein that increased in the level of phosphorylation after agonist stimulation in a manner that was inhibited by the muscarinic receptor antagonist atropine (). In CG neurons obtained from transgenic mice where the M-muscarinic receptor gene had been knocked out (), this ∼95-kD receptor band was absent (). To determine a role of CK2 in this phosphorylation event, the effect of the inhibitors TBB and DMAT on agonist-mediated receptor phosphorylation in CG neurons was investigated. Both TBB and DMAT reduced agonist-mediated phosphorylation from control levels of 2.47- ± 0.89-fold and 2.97- ± 0.06-fold over basal to 0.88- ± 0.37-fold and 0.98- ± 0.11-fold over basal, respectively (). Chymotryptic phosphopeptide maps were prepared from the receptor immunoprecipitated from CHO-M3 cells under conditions where CK2 had been inhibited by siRNA knockdown (using CK2α-4/CK2α′-1p) or by pharmacological inhibition (with TBB). In these experiments (as before), quantification of receptor numbers by radioligand binding ensured that the same number of receptors had been immunoprecipitated from each sample. These maps revealed that CK2 siRNA and TBB inhibited the phosphorylation of the same subset of phosphopeptides (, arrows and asterisks) while minimally affecting the phosphorylation of other phosphopeptides (). The sites of CK2-mediated receptor phosphorylation were investigated by two-dimensional chymotryptic phosphopeptide mapping. We established previously that the M-muscarinic receptor is phosphorylated on serines in the third intracellular loop of the receptor (). Here, we compared the chymotryptic phosphopeptide map of the receptor phosphorylated in a cellular context by endogenous kinases in CHO-M3 cells with the map of a bacterial fusion protein containing the third intracellular loop of the M-muscarinic receptor (GST-H3iloop) phosphorylated in vitro by CK2. We know from preliminary studies that this fusion protein is phosphorylated by CK2 in the muscarinic receptor portion only. The phosphopeptide map obtained from the phosphorylated M-muscarinic receptor, immunoprecipitated from [P]-orthophosphate–labeled CHO-M3 cells, was complex, with at least 19 distinct phosphopeptides identified (, right). In contrast, the phosphopeptide map from the in vitro phosphorylated GST-H3iloop demonstrated just five major phosphopeptides (, left). Four of these phosphopeptides migrated to very similar positions to phosphopeptides in the in vivo map, suggesting that they may represent the same phosphopeptides (, asterisks). It is also of interest that the peptides that are shown to decrease in the level of phosphorylation in CK2 siRNA–treated cells (, arrows) closely correlated with peptides seen to be phosphorylated in vitro by CK2 (). Edman degradation of peptide 1 in determined that it was phosphorylated in position 6 in both the in vitro sample and in vivo sample (). Similarly, for peptide 2, the major phosphoacceptor site was at position 15 in the in vitro and in vivo sample (). Thus, the fact that peptides 1 and 2 run in very similar positions in the phosphopeptide maps from the in vivo and in vitro samples and that these peptides are phosphorylated at the same position (residue) indicated that they are the same phosphopeptide phosphorylated in vitro by CK2 and in vivo by endogenous receptor kinases. The occurrence of a phosphorylated serine at position 15 in spot 2 corresponded with a predicted chymotryptic peptide where the third intracellular loop serine 351 (S) was the 15th serine (). This serine is in the motif SASSDEED, which is a classical CK2 consensus site (S-x-x-D/E/pS; ). Generation of a bacterial fusion protein 3iloop construct where the SASSDEED motif was mutated to AAAADEED resulted a fusion protein that was phosphorylated predominantly at just two sites (, marked A and B) compared with the multisite phosphorylation seen in the wide-type fusion protein. In contrast, the sites phosphorylated by CK1 were not changed in the AAAADEED mutant (). These data indicate that the SASSDEED motif was not only a phosphoacceptor site for CK2 but that phosphorylation at this motif promoted further subsequent “hierarchal” phosphorylation events, a feature that is typical of CK2-mediated phosphorylation. Similar analysis on spot 1, where residue 6 is phosphorylated, indicated that the potential phosphoacceptor sites could be contained in one of two predicted chymotryptic peptides, which start with the following sequences: VHPTGSSRS and ELQQQSMKRS. In the first of these sequences, the serine in position 6 fits a consensus CK2 site only if position 9 is primed by phosphorylation. In the second peptide, the serine in position 6 does not fit precisely into a consensus CK2 site, although priming by phosphorylation at position 10 might be sufficient. However, there is no evidence from the Edman degradation data presented here that priming at these sites occurs. Nevertheless, CK2 might be mediating phosphorylation at these sites (even in the absence of priming), a possibility that is currently being tested using mutants lacking these sites. As illustrated in , the predicted sites at which the GRKs and CK2 might phosphorylate the receptor show some overlap. To investigate whether phosphoacceptor sites between CK2 and the GRKs were in fact shared, we generated chymotryptic phosphopeptide maps of GST-H3iloop phosphorylated with GRK2 and GRK6 and compared these to maps generated after the phosphorylation with CK2. The phosphorylation of the third intracellular loop protein by GRK2 and GRK6 appeared to occur at very similar sites, as determined by the fact that the phosphopeptides migrated to similar positions and were phosphorylated on the same residues (i.e., spot A in GRK2 and GRK6 migrate to the same position and are both phosphorylated on residue 12; ). In contrast, CK2 phosphorylated the receptor on very different sites, as determined by the different migration of phosphopeptides. In the single peptide that appeared to run in a position similar to that of the GRKs (i.e., spot 1 runs similarly to spot A; ), Edman degradation demonstrated that this peptide was phosphorylated at a different residue (i.e., residue 6) compared with the residue phosphorylated by the GRKs (i.e., residue 12). These data indicate that CK2 phosphorylated the third intracellular loop at sites different from the GRKs. Internalization of GPCRs has long been associated with receptor phosphorylation (). We tested this link for the M-muscarinic receptor by removing putative phosphoacceptor sites through mutation of 16 serines in the third intracellular loop (). This mutant receptor, termed mutant-6, was expressed normally at the cell surface, as determined by radioligand binding (unpublished data) and by biotinylation of cell surface proteins followed by immunoprecipitation of the biotinylated M-muscarinic receptor (). Furthermore, these receptor-biotinylation experiments demonstrated that immunoprecipitation of mutant-6 was as efficient as that of the wild-type receptor and that the efficiency of immunoprecipitation was not altered by ligand stimulation (). In subsequent phosphorylation experiments, the ability of the mutant-6 receptor to be phosphorylated in response to agonist was demonstrated to have been almost completely removed (). The loss of receptor phosphorylation in mutant-6 correlated with a loss in the ability of the receptor to be internalized after prolonged treatment with agonist. As shown in , stimulation of CHO-M3 cells expressing the wild-type M- muscarinic receptor resulted in a decrease in cell surface receptor expression, as measured by radioligand binding using the nonmembrane permeable antagonist [H]--methylscopolamine (NMS). In contrast, stimulation of cells expressing mutant-6 did not result in any significant change in the cell surface expression of the receptor (). This was confirmed using immunohistochemistry, where staining for the wild-type receptor showed a plasma membrane localization before agonist treatment but a characteristic endosomal localization after treatment with agonist for 30 min (). This was not the case for cells expressing mutant-6, where the receptor remains at the cell surface after exposure to agonist (). Despite the fact that these data support a role for receptor phosphorylation in the internalization of the M-muscarinic receptor, inhibition of CK2-mediated receptor phosphorylation by siRNAs directed against CK2 () or pharmacological inhibition using TBB () did not significantly affect receptor internalization. It appears, therefore, that receptor phosphorylation mediated by CK2 is not involved in receptor internalization, but rather other kinases (presumably of the GRK family; ) are responsible for this process. We tested the role of CK2-mediated receptor phosphorylation in the coupling of the M-muscarinic receptor to the ERK1/2 and Jun-kinase pathways. shows that siRNA-directed agonist CK2 had no effect on the ability of the receptor to activate ERK1/2. In contrast, the magnitude of the Jun-kinase response was significantly increased after inhibition of CK2-mediated receptor phosphorylation, with the maximal response being approximately twofold greater in cells transfected with siRNA against CK2 compared with control transfected cells (). Similarly, the onset of the Jun-kinase response is faster in cells transfected with CK2 siRNA, where a significant response was evident at 10 min of agonist stimulation compared with 15 min in cells transfected with control scrambled siRNA (). Importantly, the control Jun-kinase response to sorbitol was not affected by CK2 inhibition (unpublished data). Our study indicates that CK2 is among several protein kinases involved in the phosphorylation of the M-muscarinic, each able to mediate a different signaling outcome (i.e., internalization/Jun-kinase). A natural extension of these findings would be to hypothesize that cell type–specific receptor phosphorylation could contribute to tissue-specific signaling of the receptor. A first step in testing this hypothesis would be to determine if the receptor is differentially phosphorylated in different cell types. Hence, we compared the tryptic phosphopeptide maps of agonist-stimulated mouse M-muscarinic receptors derived from transfected CHO cells with that of maps derived from receptors expressed in mouse CG neurons. As can be seen in , the receptor expressed in CHO cells was phosphorylated both basally and in response to agonist on many more sites than the receptor expressed in CG neurons. By comparing the migration of the spots, it can be seen that several of these phosphopeptides are common in the receptor maps derived from the two cell types (, marked with numbers), indicating that certain receptor phosphoacceptor sites are conserved. However, some of these common sites are differentially regulated by agonist. Hence, the phosphopeptide marked A is agonist regulated in CG neurons but is constitutively phosphorylated in CHO cells. Furthermore, the peptide marked B is agonist regulated only in CHO cells (). Although some of the phosphopeptides are common to the two cell types, others are cell type specific. Five examples of phosphopeptides identified only in CHO-derived receptors are marked with open arrowheads in (left), and an example of a phosphopeptide specific to CG neurons is marked with a closed arrowhead (right). Agonist occupation results in the multisite phosphorylation of GPCRs by receptor kinases in a manner often described as homologous phosphorylation. There is now a large body of evidence to support the role of the GRK family in this process (). However, it is clear from studies using phosphoacceptor site mutants (), peptide mapping (), mass spectrometry (), phosphospecific antibodies (; ), and transgenic animals () that the process of receptor phosphorylation is complex, suggesting the possibility of the involvement of protein kinases in addition to the GRKs (). By use of CK2 inhibitors and siRNA against the α and α′ catalytic subunits of CK2, we demonstrate that CK2 contributes to the phosphorylation of the M-muscarinic receptor in both a heterologous expression system and in mouse neurons. Furthermore, by using phosphopeptide maps, we show that in vitro phosphorylation of the third intracellular loop of the M-muscarinic receptor by CK2 results in the phosphorylation of sites that are also phosphorylated by endogenous kinases in vivo. Importantly, these in vivo sites are seen to decrease in phosphorylation after CK2 siRNA treatment. Finally, we show that purified CK2 can increase the phosphorylation state of the intact M-muscarinic receptor in membranes prepared from CHO-M3 cells. Thus, CK2 can be added to GRK2, GRK6, and CK1α as a protein kinase that can phosphorylate the M-muscarinic receptor in an agonist-dependent manner (; ; ). What might be the significance of multikinase receptor phosphorylation? Despite CK1α and CK2 sharing the same nomenclature, they are structurally distinct protein kinases, a fact highlighted in classification, where CK2 is classified as a CMGC kinase and CK1α as a member of the CK1 family. This compares with the GRKs that are classified as ACG kinases. It appears, therefore, that the M-muscarinic receptor can be phosphorylated in an agonist-dependent manner by protein kinases from very different families with distinct structural features, mechanisms of regulation and subcellular localization. This diversity would allow for a very flexible process of receptor regulation, where not only can different sites on the receptor be phosphorylated by different protein kinases but also the differential mechanisms of activation and regulation of protein kinase activity/localization could influence receptor phosphorylation and signaling. The fact that more than one structurally distinct protein kinase family has a role in M-muscarinic receptor phosphorylation is reflected in the numerous phosphoacceptor sites determined from the proteolytic phosphopeptide maps conducted in this study. Furthermore, by comparing the in vitro CK2-mediated phosphopeptides with in vivo phosphopeptide maps, and by the analysis of phosphopeptide maps from cells where CK2 was inhibited using either siRNA or pharmacological inhibitors, we demonstrate that only a subset of the phosphoacceptor sites on the M-muscarinic receptor are phosphorylated by CK2. This data points to the fact that the distinct receptor kinases for the M-muscarinic receptor are able to phosphorylate defined sites on the receptor. This is supported by comparisons of the phosphopeptide maps after the phosphorylation of the third intracellular loop by GRK2, GRK6, and CK2, which demonstrated that CK2 can indeed phosphorylate sites different from those phosphorylated by the GRKs. In the case of CK2, we determined that one of these sites was the SASSDEED motif in the third intracellular loop. Phosphorylation at this motif promotes further CK2-mediated phosphorylation in a process akin to hierarchal phosphorylation, which is a common feature of this protein kinase. The question that arises from these observations is whether phosphorylation by different receptor kinases can result in different signaling outcomes. We addressed this here by focusing on three signaling processes well known to be regulated by receptor phosphorylation, namely, receptor internalization, activation of the ERK1/2, and activation of the Jun-kinase pathway (). Our study showed that inhibition of CK2-mediated phosphorylation does not affect receptor internalization. This is the case despite the fact that M-muscarinic receptor internalization is a phosphorylation-dependent process. Thus, receptor internalization must be driven by a non–CK2-dependent phosphorylation. Our unpublished data shows that CK1α inhibition similarly does not affect receptor internalization. It appears most likely that M-muscarinic receptors are internalized in a GRK-dependent manner, as has previously been reported for this receptor subtype (). Receptor coupling to the ERK1/2 pathway is similarly not affected by inhibition of CK2-mediated receptor phosphorylation. We have shown previously that this signaling response is likely to be regulated by CK1α (). In contrast, we show here that CK2 activity is important in coupling the receptor to the Jun-kinase pathway. Inhibition of CK2 via siRNA substantially increases both the magnitude and time course of the Jun-kinase response to muscarinic receptor stimulation but has no affect on the receptor-independent activation of the Jun-kinase pathway mediated by sorbitol. Because we show that CK2 is able to mediate receptor phosphorylation, our data point to the possibility that CK2-mediated receptor phosphorylation can regulate the coupling of the M-muscarinic receptor to the Jun-kinase pathway and none of the other phosphorylation-dependent signaling pathways. This supports the notion that site-specific phosphorylation mediated by a single receptor kinase can regulate a defined receptor signaling process. Our studies do not, however, completely rule out the possibility that CK2 has an indirect role on receptor coupling to the Jun-kinase pathway that is independent of receptor phosphorylation. A logical extension of this finding is that GPCR phosphorylation might be used as a flexible adaptive process where a defined complement of protein kinases would be recruited to phosphorylate specific sites in a process that would allow for tissue-specific signaling. Hence, in the case of the M-muscarinic receptor, such a process would contribute to the very different physiological responses mediated by this receptor when expressed in smooth muscle cells compared with the same receptor subtype expressed in salivary acinar cells and neurons. In the current study, we tested whether the M-muscarinic receptor was able to be phosphorylated in a cell type–specific fashion by comparing the tryptic phosphopeptide maps obtained from the mouse M-muscarinic receptor immunoprecipitated from CHO cells and mouse CG neurons. We refer to the pattern obtained from these maps as phosphorylation signatures. By comparing the receptor phosphorylation signatures obtained in CHO cells and CG neurons, it was clear that some elements of receptor phosphorylation were the same between the two cell types, whereas others were cell type specific. The common features of these maps may underlie common regulatory processes, such as receptor internalization, whereas those phosphorylation events that are unique to a given cell type might be involved in cell type–specific signaling. It was clear from these studies that at least between these two cell types the phosphorylation signatures of the M-muscarinic receptors were different, indicating that differential cell type–specific phosphorylation indeed occurs. Our continuing studies aimed at defining the role of these cell type–specific phosphorylation events in physiologically relevant tissues will test further whether this differential phosphorylation pattern is related to cell type–specific functional responses. Mouse CG neurons were cultured as described previously (). In brief, cerebella from 7–8-d-old BALB/c or transgenic pups were mechanically and enzymatically (trypsin) dissociated and plated at 0.25 × 10 cells/cm on 100 μg/ml poly--lysine–coated 6- or 12-well plates (Nunc). CG neurons were then incubated in Eagle's basal medium supplemented with 20 mM KCl, penicillin/streptomycin, 10% fetal calf serum, and 10 μM cytosine arabinoside (added 48 h after plating) in a humidified atmosphere with 5% CO at 37°C for 7–8 d. The phosphorylated M-muscarinic receptor was immunoprecipitated from 50 μCi/ml [P]-orthophosphate–labeled cells as previously described (). Equivalent amounts of immunoprecipitated receptor in each sample were ensured by parallel radiolabeled ligand binding experiments using the antagonist [H]-NMS (). The immunoprecipitated receptor was resolved on 8% SDS-PAGE and visualized by autoradiography or using a phosphorimager (STORM; GE Healthcare). In the case of CG cells, the procedure was the same, but CSS-25 incubation buffer (120 mM NaCl, 1.8 mM CaCl, 15 mM glucose, 25 mM KCl, and 25 mM Hepes, pH 7.4) containing 100 μCi/ml [P]-orthophosphate was used. The receptor was immunoprecipitated from cell lysates obtained by pooling 2 wells of a 6-well plate. In these experiments and others involving the mouse receptor, we used an in-house anti–mouse M-muscarinic receptor antibody (see Generation of mouse M-muscarinic receptor antibody). For phosphopeptide mapping, CHO-M3 cells and CG neurons were labeled with 100 or 200 μCi/ml [P]-orthophosphate, respectively. Stimulation and immunoprecipitation were performed as described. The immunoprecipitated samples of one entire 6-well plate of CHO cells or 2 plates of CG neurons were pooled and resolved by SDS-PAGE. The gel was then electroblotted onto nitrocellulose membrane, and the phosphorylated receptor was visualized by autoradiography. The area of the membrane containing the receptor was cut out, blocked for 30 min at 37°C with 0.5% polyvinylpyrrolidone-K 30 (Sigma-Aldrich) containing 0.6% acetic acid, and washed several times with water. The M-muscarinic receptor contained on the membrane was digested for 20 h at 37°C in ambic solution (50 mM NHHCO and 0.5 mM CaCl) containing 10 μg/ml of either trypsin (Promega) or chymotrypsin (Sigma-Aldrich). The supernatant was then removed, and the membrane slice washed once with water. The wash and supernatant were combined dried. Tryptic/chymotryptic peptides were then resuspended in 10 μl pH 1.9 buffer (88% formic acid/acetic acid/water, 25:78:897 vol/vol) and spotted onto cellulose-coated chromatography (TLC) plates (20 × 20 cm; Merck). The peptides were then separated in two dimensions. The first dimension was electrophoresis at 2,000 V for 30 min in pH 1.9 buffer using a Hunter HTLE-7002 system (CBS Scientific). The second dimension was ascending chromatography in isobutyric buffer (isobutyric acid/n-butanol/pyridine/acetic acid/water, 1,250:38:96:58:558 vol/vol). The resolved phosphopeptides were then visualized using a STORM phosphorimager. The human M-muscarinic receptor third intercellular loop (R-T) cloned in-frame with GST (GST-H3iloop) was expressed and purified from bacteria as previously described (). 35 μg of fusion protein bound to glutathione–Sepharose beads was incubated with 100 units of CK1 or CK2 (New England Biolabs, Inc.) or 100 ng of purified GRK2 (provided by R. Lefkowitz, Duke University Medical Center, Durham, NC) or GRK6 (provided by J. Tesmer, University of Michigan, Ann Arbor, MI) in kinase assay buffer (10 mM MgCl, 20 mM β-glycerophosphate, and 20 mM Hepes, pH 7.4) with 50 μM ATP and 10 μCi γ-[P]ATP in a total volume of 100 μl. The reaction was continued for 30 min at 37°C. To determine the inhibition potency of TBB and DMAT (Calbiochem), protein kinases were preincubated for 10 min in the same buffer without ATP and with 1 μM of each inhibitor. Reactions were stopped by the addition of 1 ml of ice-cold buffer, and the GST-H3iloop:glutathione–Sepharose beads were pelleted and resuspended in 2× SDS-PAGE sample buffer before being resolved by SDS-PAGE. Quantification of the phosphorylation status of GST-H3iloop was determined by phosphorimager analysis. Alternatively, the phosphorylated GST-H3iloop was blotted onto nitrocellulose membrane and used in phosphopeptide mapping as described. Crude membranes were prepared from CHO-M3 and CHO cells and stored at −80°C as described previously (). Before use, membranes were washed for 15 min with kinase buffer (25 mM Tris-HCl, pH 7.4, 20 mM β-glycerophosphate, 200 mM NaCl, 10 mM MgCl, and 100 μg/ml BSA). Membranes were then pelleted in a microfuge (3 min at 21,000 ) and resuspended in kinase buffer before use in the in vitro phosphorylation reaction, which consisted of membranes (200 μg protein) and kinase buffer containing γ-[P]-ATP (2 μM; 1–4 disintegrations per minute/fmol) in the presence or absence of CK2 (200 or 500 units per reaction) in a reaction volume of 200 μl. Reactions were incubated at 37°C for 10 min. Reactions were stopped by pelting membranes in a microfuge (21,000 for 1 min), and solubilization of the membrane pellet and immunoprecipitation of the M-muscarinic receptor were performed as described previously (). Expression of cell surface muscarinic receptors after exposure of cells to methacholine for the indicated times was determined as described previously () except that saturating concentrations (0.5 nM) of the muscarinic antagonist [H]-NMS was used in incubations with whole cells for 60 min at 37°C. Washes were then conducted at 4°C (). To determine subcellular localization of the M-muscarinic receptor using immunohistochemistry, cells were fixed with 4% paraformaldehyde and permeabilized with Triton X-100 and the receptor was stained using a 1:500 dilution of an in-house anti-human M-muscarinic receptor antibody (). Cells plated onto 6-well dishes were serum starved for 1 h and stimulated with 0.1 mM methacholine for the times indicated. The cells were then lysed, and cellular ERK1/2 was immunoprecipitated and assayed in an in vitro assay using the EGFr peptide substrate as described previously (). Alternatively, Jun-kinase was isolated from the cell lysate using a GST–c-Jun fusion protein followed by the in vitro phosphorylation of GST–c-Jun as previously described (). Cells cultured on 6-well plates were incubated in buffer containing 1 mM biotin (Pierce Chemical Co.) for 30 min at 37°C. The cells were then lysed, and the M-muscarinic receptor was immunoprecipitated as described for phosphorylation experiments. Equivalent amounts of receptor were ensured by analysis of receptor levels using [H]-NMS radiolabeling. Samples were then separated by SDS-PAGE, electroblotted to nitrocellulose membranes, and, after blocking with TBST plus 5% milk powder for 1 h, incubated with 50 ng/ml streptavidin conjugated to horseradish peroxidase (Pierce Chemical Co.) for 30 min. Biotinylated proteins were then detected with chemiluminescence reagent (ECL plus; GE Healthcare). The region S-L of the mouse M-muscarinic receptor third intracellular loop was produced as an N-terminal tagged GST/receptor fusion protein, which was used to inoculate New Zealand white rabbits (Harlan Sera-Labs). The resulting antisera was tested in immunoprecipitation studies and shown to specifically react with the mouse M-muscarinic receptor (). 48 h after siRNA transfection, CHO cells cultured in 12-well dishes were harvested in lysis buffer (25 mM Tris-HCl, pH 7.4, 20 mM β-glycerophosphate, and 200 mM NaCl) containing 0.5% NP-40 substitute (Fluka). The kinase assay (using 10 μl of cell lysate) was performed in lysis buffer containing 1 μM ATP, 10 mM MgCl, 100 μg/ml BSA, 50 μM CK2 peptide substrate RRREEETEEE (Promega), and 1 μCi γ-[P]ATP (total volume 50 μl). After a 15-min incubation at 37°C, half of the mixture was spotted on phosphocellulose P81 filter (Whatman) and washed four times with 0.5% HPO. The filters were then transferred to vials containing scintillation fluid (Safefluor) and counted. The phosphorylation assays were performed in either the presence or absence of 1 μM of the CK2 inhibitor TBB. CK2 activity was then determined as the TBB-sensitive component of the peptide phosphorylation, which was ∼20% of the total peptide phosphorylation. Autoradiography densitometric analysis was performed using ImageQuant and AlphaEase FC softwares. SD of at least three determinations is presented, and significance was determined using a one-way analysis of variance (ANOVA). Fig. S1 shows the sequences on CK2α and -α′ subunits that were targeted by siRNAs. These siRNAs were then tested for the ability to reduce expression of CK2 and effects on M-muscarinic receptor phosphorylation (Fig. S2). This data demonstrated that siRNAs distinct from those used in were also able to reduce CK2 expression, and this correlated with a reduction in receptor phosphorylation. Online supplemental material is available at .

Epithelial cell morphogenesis is a key step in the formation and development of multicellular organisms. Morphogenetic processes involving cell shape change or cell displacement inside an epithelial sheet depend on the flexibility of the cohesiveness between epithelial cells, which is ensured by several adhesion systems, such as the adherens junctions (AJs). Thus, elucidating the regulation of the expression of molecules involved in cellular adhesion is crucial to increasing our knowledge of epithelial cell morphogenesis. The somatic follicular cells of ovarian follicles provide a simple system for studying epithelial morphogenesis. Initially, these cells have a cuboidal shape and form a monolayer around each 16-cell germline cyst. Throughout oogenesis, they progressively differentiate into distinct populations that adopt a squamous or columnar shape and undergo various morphogenetic movements. Thus, studying cell morphogenesis of the follicular epithelium should further our understanding of how cell shape change and cell displacement proceed in an epithelial sheet. When a follicle leaves the germarium (stage 1), it consists of a germ cyst of 16 cells (one oocyte and 15 nurse cells [NCs]) surrounded by a monolayer of follicular cells (; ). Three different subpopulations of somatic cells can be identified: the polar cells, the stalk cells, and the follicular cells (). Two pairs of polar cells are present at each extremity of the follicle, whereas the interfollicular cells are organized into a stalk of four to six cells that intercalate between adjacent follicles. These two populations play a critical role in follicle formation (; Lopez-Schier and St Johnston, 2001; ). The differentiation of the follicular epithelium starts with the specification of the terminal follicular cells versus the main body follicular cells (MBFCs; ; ). At the anterior pole, the terminal cells give rise to the border cells, the stretched cells (StCs), and the centripetal cells (). These three populations cannot be recognized before stage 9 or 10, when specific markers or genes are expressed and several morphogenetic features become apparent (; ; ). Of relevance to the present study are the morphogenetic processes that concern the MBFCs and StCs: at stage 9, the MBFCs displace toward the oocyte and take on a columnar shape when they make contact with it, whereas the StCs flatten around the NCs to maintain a continuous epithelium. This flattening starts from the most anterior cells and extends progressively to the cells located more posteriorly. This process can be monitored by the expression of the MA33 and () enhancer trap lines and of the Eyes absent protein (Eya; see Materials and methods). At the end of the process of StC flattening, most of the nuclei of the 50 or so StCs are positioned in the interstitial gaps between the NCs (; ). The rearrangement of MBFCs into columnar cells, their displacement toward the oocyte, and the flattening of StCs are morphogenetic processes that are poorly understood, and the genes involved are almost entirely unknown. The Notch signaling pathway is required at several stages of oogenesis. First, the Notch receptor is required for polar cell differentiation in the germarium. This entails the germline production of the ligand Delta (Dl) and the specific expression of the Notch modulator Fringe (Fng) in the polar cell precursors (; ; ). Second, the Notch signaling pathway is required to induce a transition from the mitotic cycle to the endocycle in follicular cells at stage 6 of oogenesis. This role is also dependent on the germline production of Dl but is independent (). Third, stage 10 follicles mutant for display defects in the differentiation of the border, stretched, and centripetal cells as well as abnormal migration (; ; ; ). Although the first two requirements of have been analyzed in detail, the third is poorly defined because it has only been demonstrated using a thermosensitive allele. Thus, the precise role of Notch in differentiation of the border, stretched, and centripetal cells and in their morphogenesis remains an unresolved question. In this study, I have taken a different approach to address this question by analyzing the role of during StC flattening and MBFC displacement, as its expression pattern from stages 7–10A suggest that it could be involved. Indeed, during these stages, mRNA is present in all of the follicular cells except the outer border cells (; ). Somatic mutant clones for an -null allele () have been generated, and analysis of the phenotypes induced by the mutant clones leads to the conclusion that the Notch pathway functions in an -dependent manner to control AJ remodeling between the StCs and that this activity is essential for proper StC flattening and for the timing of MBFC displacement. Because is required for the differentiation of polar cells, which, in turn, promote terminal cell differentiation (; ; ; ), only follicles with somatic clones that do not include the anterior and posterior polar cells were analyzed. Only the role of in MBFC displacement and StC flattening will be analyzed in detail here. In the phenotypic description, references to the MBFC include the cells destined to become centripetal cells because the two cell types are indistinguishable until stage 10 (). When clones encompass some StCs and some MBFCs, an excess of cells is observed over the NCs in follicles at stage 10A or older (75%; = 60). As described in the next two paragraphs, these supernumerary cells are either MBFCs or StCs. In stage 10A follicles, many of the supernumerary cells appear in clusters (). These clusters present three traits. First, they display a cell density that is more characteristic of the MBFC population than of the StC population. Second, they are always immediately adjacent to the MBFC population. Third, most of their component cells do not express MA33 or Eya except the ones that are at the anterior border of the cluster (), indicating that the majority of these cells do not have a stretched fate. Together, these data suggest that most of the supernumerary cells within these clusters are MBFCs that did not fully complete their displacement. In agreement with this, these clusters are never observed over the NCs at later stages, indicating that they do eventually surround the oocyte (). Thus, these observations show that is required in the StCs and MBFCs to allow the posterior displacement of the MBFCs to be fully completed by the end of stage 9. However, because this displacement does still occur, other mechanisms must act in parallel to to control it. In stage 10B follicles, supernumerary cells, although no longer arranged in clusters, are still observed over the NCs when a large clone encompasses some MBFCs and some StCs (80%; = 25). These supernumerary cells express every StC marker such as MA33 (), Eya (see ), , and the Dpp pathway readout , the latter also being expressed in the StCs during their flattening (see Materials and methods). Thus, all of the supernumerary cells observed over the NCs in stage 10B follicles have an StC fate. As a consequence, in such follicles, the number of StCs increases from ∼50 to between 70 and 110 cells (). These extra StCs could either have arisen from extra divisions of the cells or from MBFCs that did not get displaced and instead adopted a stretched fate. If these were abnormally proliferating follicular cells, a much higher density of cells would be expected in the part of the mutant clone overlying the NCs, as has been shown for mutant alleles of genes required to stop follicular cell mitotic division (). Moreover, no expression of cyclin B and phosphohistone-3 has been detected in mutant clones after stage 6 of oogenesis (; unpublished data). Thus, these cells are most likely MBFCs that did not reach the oocyte and differentiated as StCs. In conclusion, the supernumerary cells observed above the NCs at stages 10A and 10B derive from an abnormal displacement of MBFCs. The penetrance and expressivity of these phenotypes vary based on the type and number of cells that are mutant. The penetrance is higher when the clones encompass both MBFCs and StCs than when they encompass only one of those cell types, showing that is required in both the StCs and MBFCs. Nevertheless, when the clones encompass just MBFCs, only clones in the anterior part of the MBFC population induce delayed displacement (), whereas follicles with a large lateral and/or posterior clone do not (not depicted). This indicates that the requirement of is different within the MBFC population, with a higher requirement in the anterior cells in contact with StCs. On the other hand, the displacement of MBFCs is more delayed in regions where most of the StCs are mutant than in regions where most of the StCs are wild type (WT), indicating that the StCs play an essential role in MBFC displacement (). Finally, MBFC delays are rarely (6%; = 30) observed in small clones (<10 cells; unpublished data), suggesting that the presence of WT cells around the mutant cells prevents it. Equally, some WT MBFCs can be delayed when they are in the immediate proximity of a large clone (; see ). This suggests that the process of displacement for one MBFC depends on the ability of its neighboring cells to be displaced such that the displacement of the MBFC occurs in a coordinated manner all around the follicle and not in a cell-autonomous manner. In conclusion, is required in the StC and in the anterior MBFC to direct StC number and to control the timing of MBFC displacement. It has been proposed that displacement of the MBFC toward the oocyte is the result of cell shape changes of the most posterior cells from cuboidal to columnar, with this change in form forcing their neighboring anterior cells to move posteriorly (; ). Because no MBFC displacement delay is observed in clones encompassing only posterior cells and because no defect in cell shape change has been detected in posterior cells, is not required to generate the force created by the posterior cells (hereafter referred to as the pulling force; unpublished data). In WT follicles, the MBFC displacement probably stops when the StCs are unable to expand any further without compromising the integrity of the epithelium, implying that this displacement also depends on the ability of the cells to respond to the pulling force (for example, by flattening). It is likely, then, that is involved in this response because a delayed MBFC displacement is observed in most follicles with an clone encompassing StCs and anterior MBFCs. In fact, defective StC flattening and delayed MBFC displacement could arise from a physical inability to undergo morphogenetic processes or from an abnormal StC differentiation (). Accordingly, an analysis of the expression of all known StC fate markers (the Eya protein and the , , and enhancer traps) has been undertaken in stage 9 follicles containing an anterior clone on one side of the follicle. This provides an internal reference such that marker expression in mutant cells can be compared with WT cells located at the same anterior-posterior (A/P) position. No noticeable difference in the expression of any of the markers was observed between WT and mutant StCs (), indicating that differentiation of the mutant StCs is not delayed in comparison with the WT cells. Moreover, although the role of the Dpp pathway in StC fate determination is unknown, the correct expression of the enhancer trap in StCs rules out the hypothesis that this pathway is not active in these cells (). In conclusion, no alteration of StC identity is observed in mutant follicles. Because the delay in MBFC displacement in clones is not caused by a delay in StC differentiation, it leaves open two possibilities: either mutant StCs are unable to flatten properly and/or mutant MBFCs cannot move posteriorly properly. One possible underlying cause that could explain both possibilities is a physical inability of the cells to change their shapes and/or to be displaced. Epithelial cell shape change and movement depends on AJ remodeling. A detailed analysis of AJ remodeling was first performed in WT stage 9 follicles because this is as of yet uncharacterized. Because the StCs undergo the greatest cell shape changes within a follicle, it is reasonable to expect that any mutation that causes defects in cell shape changes by modifying the properties of AJs may manifest more dramatically in the StC than in any other cell type. Therefore, to simplify the study of the role of , I have chosen to focus on the AJs and their dynamics between the StCs by monitoring the expression of epithelial cadherin (ECad; ) and the β-catenin protein Armadillo (Fig. S1, available at ). The following observations have been obtained from the analysis of a mean of 15 follicles of each stage. Throughout stage 9, a colocalization between the two proteins is observed. In early stage 9 of oogenesis, a strong ECad expression is detected in all follicular cells except the posterior cells. This expression displays the hexagonal shape of the follicular cells, indicating that AJs are present all around each cell close to their apical side (). As a result of the geometry of the cells, some AJs mediate contact between two cells, whereas others mediate contact between three (). The orientations of the two-cell junctions are essentially either perpendicular or parallel to the A/P axis. During the process of flattening (midstage 9), no ECad expression is detected at the three-cell junctions that mediate contact between the flattened cells and the flattening cells (). In contrast, ECad is still present at the three-cell junctions that mediate contact between the flattening cells and the unflattened cells. Additionally, only a few spots of ECad are observed in the perpendicularly oriented two-cell junctions between the flattened cells and the flattening cells, whereas a strong ECad expression is still observed in those between the flattening cells and the unflattened cells (). Distinct patterns are thus observed depending on the degree of flattening of the cells. In late stage 9, most of the AJs are no longer visible except for the ones that are parallel to the A/P axis (). At stage 10A, these junctions are barely visible, but ECad starts to be accumulated very strongly in the part of the StC located in the intersticial gaps between NCs. ECad is also detected weakly in the part of the StC overlying the NCs (). At stage 10B, the AJs parallel to the A/P axis are no longer detected, but the ECad expression in the part of the StCs overlying the NCs increases, which permits detection of the shape of the StCs. This reveals that most of the StCs are more elongated along the A/P axis than along the dorsal-ventral axis (). Identical results for StC shape were also obtained by assaying α-tubulin expression (). Next, I compared ECad expression patterns in mutant StCs to control WT StCs located at the same A/P position within the follicle. At late stage 9, the three-cell junctions between WT cells are disassembled, whereas they are still intact in regions where most of the StCs are mutant (77%; = 18; , arrows). The StCs immediately posterior to these latter mutant StCs, whether themselves WT or mutant (, long arrow and arrowhead), display ECad staining on all of their edges, indicating that the process of AJ remodeling has not yet started in these cells. Finally, an abnormally high number of StCs as well as a delay in MBFC displacement are observed in the mutant region (77%; = 18). Similar results are obtained by monitoring the expression of Armadillo (75%; = 12; Fig. S1). Together, these observations show that is required for the AJ dynamics to occur properly during StC flattening and, thus, that the MBFC displacement delays and the abnormally high number of StCs observed in mutant follicles derive from a physical inability of the StCs to flatten. ECad expression patterns still differ between mutant and WT regions in a stage 10A follicle (Fig. S2, available at ), but, in late stage 10B follicles, the ECad expression pattern is almost identical between mutant and WT regions, indicating that AJ remodeling is delayed but not blocked in clones (unpublished data). In these follicles, when the majority of the StCs are mutant, there are twice as many StCs as in a WT follicle (). Because the overall size of the NC compartment in follicles with many mutant StCs is not substantially different than in WT, each StC likely covers less surface area than a WT (compare with ). Indeed, the mean surface area of an StC in a WT follicle is 1,526 ± 301 μm ( = 30), whereas the mean surface area of an StC in a follicle with a large clone is 804 ± 261 μm ( = 83; P < 0.0001). This shows that StCs in mutant follicles flatten less than StCs in WT follicles. In addition, no accumulation of ECad is observed in the part of the StCs that are usually squeezed into the interstitial gaps between the NCs, presumably because the StCs are less flattened and their nuclei are no longer forced into these gaps. In contrast, a higher level of ECad is detected in the part of the StCs overlying the NCs (compare with ). This brighter ECad signal could derive from a higher expression level but more likely reflects the fact that because the cells are smaller, their apical surface is reduced, and, thus, ECad molecules are more concentrated. Staining for α-tubulin yields identical results (). In conclusion, is required for proper flattening of the StCs, which, in turn, controls the minimal number of cells that will adopt this fate. To address whether this role of is linked to an activity of the Notch pathway, I conducted a similar analysis with null alleles of () and (). This analysis shows that in stage 9 follicles with anterior clones doubly mutant for and , most of the three-cell junctions are still present in the mutant region but are disassembled in the WT region (91%; = 23; ). Furthermore, an increased density of StCs and a delayed MBFC displacement are observed in the mutant region (). No defects have been observed for single clones ( = 8), whereas single clones yielded similar defects with a weaker expressivity to the double mutant clones (87%; = 15; unpublished data). In addition, as described for follicles, the appearance of StC markers occurs properly in the mutant StC (unpublished data). Together, these data demonstrate that although the Notch pathway appears not to affect StC identity, it does have an -dependent role in remodeling their AJs, with playing an essential role in the soma and having an overlapping function. A detailed analysis of the expression pattern of Notch and Dl during stage 9 reveals that these two proteins are strongly expressed in the flattened and flattening StCs. In particular, a strong accumulation of Notch is detected at the disassembling three-cell junctions, suggesting that this expression corresponds to the role of Notch described in this study (). Similarly, analyses of the expression patterns of some reporters for Notch activation show that they are more strongly expressed in the flattened and flattening StCs than in the MBFCs (Fig. S3, available at ). Thus, the expression patterns of these markers indicate that the Notch pathway is transcriptionally active in the StCs at stage 9. As a direct involvement of the canonical Notch pathway cannot be analyzed (because of its requirement at earlier steps of oogenesis), I tested the effects of overexpressing Hairless, a repressor of Notch transcriptional activity, using an construct. In all stage 9 follicles overexpressing Hairless ( = 28), abnormal AJ remodeling is observed. In contrast to WT follicles, no AJ disassembly is detected between StCs in early midstage 9 follicles overexpressing Hairless (compare with ). Rather, this disassembly commences only in mid- to late stage 9 follicles after several rows of StCs have differentiated ( and Fig. S3 E). This indicates that the transcriptional activity of Notch plays a crucial role in controlling the dynamics of AJ disassembly between StCs at stage 9 of oogenesis. Follicles expressing a constitutively activated form of Notch (Nact) were then analyzed. Surprisingly, the overexpression of Nact in some StCs leads to an autonomous, abnormal flattening of StCs and a delay in AJ remodeling (72%; = 20; Fig. S3). The shape of Nact-expressing cells remains cuboidal longer than the cells undergoing flattening that are located at the same position along the A/P axis (Fig. S3 F). In parallel, AJ remodeling is delayed in the Nact-expressing region in comparison with the WT region (Fig. S3 G). This indicates that some Notch target genes are involved in AJ remodeling. One could suggest that the loss or gain of activity of the Notch pathway affects the expression of different target genes or that the timing of Notch activation has to be controlled very precisely to allow proper AJ remodeling. The chronology of AJ disassembly is essential for StC flattening and the timing of MBFC displacement. This chronology likely depends on local effectors at the cell junctions undergoing disruption. The myosin II heavy chain and regulatory light chain, which are encoded by the and () genes, respectively, are two effectors that have been shown to remodel cell junctions in a polarized manner (). To determine whether these two proteins are involved in junction disassembly during StC flattening, their expression patterns were determined in stage 9 follicles. A detailed analysis during stage 9 shows that Zipper is more strongly accumulated in some specific spots, which correspond to disassembling AJs. Indeed, depending on the degree of the flattening of StCs, Zipper can be strongly accumulated along a lateral line that separates the flattened cells from the ones that are flattening (). Some parts of this line of expression correspond to the edges of the StCs that are oriented perpendicularly to the A/P axis. Anterior to this line, Zipper is accumulated in StCs both in the vicinity of their nuclei as well as along those edges that are parallel to the A/P axis. In some follicles, this line is not detected, but, instead, a strong accumulation is detected at the disassembling three-cell junctions that are between the row of flattening StCs and the row of unflattened cells (). A similar pattern of expression is observed with an Sqh-GFP fusion protein (). Together, these observations suggest that these are good candidate proteins for control of the chronology of AJ remodeling during StC flattening. To confirm this, Zipper expression has been investigated in double-mutant follicles. In such follicles, when the lateral line of Zipper accumulation is observed in WT regions, it appears to be interrupted in regions containing mutant cells (). Moreover, at the A/P position where Zipper is accumulated at the disassembling three-cell junctions in WT regions, no accumulation is detected in mutant regions (). Such accumulations are instead detected at disassembling three-cell junctions located more anteriorly. Thus, the delay in AJ disassembly observed in cells is coincident with the delay in the appearance of Zipper accumulation at disassembling junctions. A similar delay is observed in cells (unpublished data). Thus, the defects in StC flattening and MBFC displacement delay may be caused by a lack of polarization of Zipper during cell junction disassembly. As an impaired Notch pathway results in abnormal delay in AJ remodeling and the modification of Zipper expression pattern during StC flattening, it suggests a role for , , and during this process. To test this, somatic mutant clones for an -null allele () and for an hypomorphic allele () were induced. Stage 9 follicles with mutant clones display a higher density of flattening StCs in the mutant region than in WT ( = 12; ). At stage 10, a delay in MBFC displacement is also detected in the mutant region ( = 10; ). Thus, as observed in follicles with and / mutant clones, the ability of the StCs to flatten and the timing of MBFC displacement are impaired in mutant follicles. One might expect that the absence of would result in the loss of AJs and that StC flattening would occur faster and/or without any constraint. In light of this, the data from the clones could suggest that the cells are still adhesive. As another adhesion molecule, Notch cadherin (NCad), can also be a component of AJs, its expression was investigated in such clones. In WT follicles, NCad is strongly expressed in all follicular cells from stages 1 to 6 (). Its expression decreases during stages 7 and 8 to become almost undetectable at stage 9, with a weak expression in the follicular cells that surround the oocyte and in the StCs (). In stage 7–9 follicles with mutant clones, a strong NCad expression is detected in the mutant cells. This overexpression is cell autonomous in stage 7 follicles ( = 11) but is restricted to the anterior part of the clones in stage 8 ( = 9) and stage 9 follicles (75%; = 20; ; and Fig. S4, available at ). This overexpression of NCad is no longer detected at stage 10A ( = 25; unpublished data). Thus, the absence of ECad is compensated for by the up-regulation of NCad from stages 7 to 10 and leads to delayed StC flattening and MBFC displacement, indicating that the levels of ECad and NCad expression during mid-oogenesis are interrelated and that the regulation of ECad expression is important for StC flattening. The requirement of myosin II cannot be tested directly by using null alleles of or , as mutant cells will be unable to divide. Thus, a hypomorphic allele of was used, and only small clones can be analyzed. In stage 9 follicles, the mutant StCs undergoing flattening present an abnormal pattern of ECad. The AJs are not remodeled according to the pattern described for WT cells (). Elongation of the mutant cells does not appear to occur mainly along the A/P axis but is more randomly oriented. In addition, the AJs parallel to the A/P axis between cells that have already flattened are no longer visible in the mutant region when compared with the WT region. This indicates that activity of the regulatory light chain of myosin II is required to control the dynamic of AJ remodeling and proper StC flattening. r e e i m p o r t a n t m o r p h o g e n e t i c e v e n t s o c c u r i n t h e f o l l i c u l a r t i s s u e s a t s t a g e 9 o f o o g e n e s i s : b o r d e r c e l l m i g r a t i o n , S t C f l a t t e n i n g , a n d M B F C d i s p l a c e m e n t . M y r e s u l t s h a v e a d v a n c e d t h e u n d e r s t a n d i n g o f S t C f l a t t e n i n g a n d M B F C d i s p l a c e m e n t i n t w o c r u c i a l w a y s . F i r s t , t h e y d e m o n s t r a t e t h a t t h e r a t e o f M B F C d i s p l a c e m e n t d e p e n d s o n p r o p e r S t C f l a t t e n i n g , w h i c h s u p p o r t s t h e i d e a t h a t t h e s e p r o c e s s e s o c c u r i n a c o o r d i n a t e d m a n n e r . S e c o n d , t h e y i d e n t i f y t h e N o t c h p a t h w a y a s p l a y i n g a n e s s e n t i a l r o l e i n S t C f l a t t e n i n g . I n a d d i t i o n t o t h e s e n e w i n s i g h t s i n t o f o l l i c u l a r c e l l m o r p h o g e n e s i s , m y r e s u l t s a l s o d e m o n s t r a t e t h a t t h e N o t c h p a t h w a y a c t s d u r i n g t h e s e m o r p h o g e n e t i c e v e n t s b y c o n t r o l l i n g t h e d y n a m i c o f A J d i s a s s e m b l y a n d n o t c e l l i d e n t i t y . , , , , and are null alleles (; ; ; ; ), and is a hypomorph allele (). Canton-S was used as WT, and the reporter lines used were ; P(, SqhGFP)40 (), (), (), (), (), (an enhancer trap in the β gene; gift of S. Bray, University of Cambridge, Cambridge, UK), and (). , , , , and clones were generated by Flipase-mediated mitotic recombination on FRT101, FRT42D, FRT80-3, or FRT82B chromosomes and was marked using or transgenes (). Ectopic expression of activated Notch was performed by generating Flip-out Gal4 clones in animals carrying UAS-ΔN34a () and () transgenes. Flipase expression was induced by heat shocking 2-d-old females at 38°C for 1 h to generate mutant clones and at 32.5°C for 30 min to generate Flip-out clones. Ectopic expression of () was performed by heat shocking (6 h before dissecting) 3–5-d-old females at 36.5°C for 1 h. Immunofluorescent staining of follicles was performed as described previously () using goat anti–β-galactosidase (1:1,000; Biogenesis), rabbit anti-Myc (1:100; Santa Cruz Biotechnology, Inc.), mouse anti-GFP (1:500; Sigma-Aldrich), rabbit anti-Zipper (1:1,000; ), mouse anti-Eya (1:500; Developmental Studies Hybridoma Bank [DSHB]), mouse anti–α-tubulin (1:1,000; clone DM1A; Sigma-Aldrich), rat anti-ECad (1:200; DSHB), mouse anti-Armadillo (1:50; DSHB), and rat anti-NCad (1:20; DSHB) with the following modification: ovaries from females were dissected directly into fixative 5–7 d after Flipase induction (36 h for mutant clones). Follicles were staged according to and . MA33 is expressed only in the StCs from stage 9 onwards (; ; ; ). is also specifically expressed in the StCs at stage 9, as it turns on in some centripetal cells only at stage 10B (; ). Eya expression is specific at stage 9 to border and StC fates (). In WT stage 9 follicles, the expression of these markers occurs progressively from the anterior part of the follicle and progresses posteriorly row by row as a wave. Accordingly, a gradient of expression for all of these markers is detected in the anterior part of a stage 9 follicle, with the strongest expression in the already flattened StCs, a weaker expression in the flattening StCs, and the weakest expression in cells that are about to undergo flattening. Because Eya is expressed in both border and StCs at stage 9, any conclusions drawn from an Eya staining were always confirmed with MA33, , or to avoid ambiguity concerning the fate of the cells. was used as an StC fate marker because its expression pattern is similar to that of during stage 9. Surface areas of StCs were determined with LSM510 Meta software (Carl Zeiss MicroImaging, Inc.). Follicles were flattened under a coverslip such that the StCs in immediate contact with either the coverslip or the slide (i.e., those on the top or bottom of the follicle) could be viewed within a single focal plane. Only these StCs were used for surface area calculations to avoid inaccuracies caused by the natural curvature of the follicle, which was not taken into account. Confocal images were obtained using a microscope (LSM510 Meta; Carl Zeiss MicroImaging, Inc.) with 40× NA 1.3 plan-Neofluar and 63× NA 1.4 plan-Apochromat objectives. All imaging was performed at RT. Figures were processed using the LSM510 Meta software, Photoshop 7 (Adobe), and Freehand 10 (Macromedia). In all panels, unless otherwise stated, the focus is on the top plane, and a projection (z stack) of all of the z sections in which the StCs are visible is presented. Fig. S1 shows that the Armadillo expression pattern mirrors AJ remodeling and shows the expression of Armadillo in WT follicles and in mutant follicles. Fig. S2 shows that AJ remodeling in mutant follicles is delayed but not blocked and shows the expression of ECad in stage 10A mutant follicles. Fig. S3 shows that StC flattening and AJ remodeling required transcriptional activity of the Notch pathway and shows the expression of , , and β reporters in stage 9 follicles as well as the delay in AJ remodeling observed in stage 9 follicles overexpressing and in stage 9 follicles overexpressing Nact. Fig. S4 shows that controls the down-regulation of NCad and shows the expression of NCad in mutant clones at stages 7 and 8. Online supplemental material is available at .

The mammalian skin functions as a barrier to many forms of environmental stress, therefore wounds of the skin need to be repaired efficiently (; ). Wounding of skin can damage both epidermis and dermis; thus, wound healing requires reepithelialization of the epidermis and the formation of new dermal structures, called granulation tissue. During reestablishment of the epithelial barrier, keratinocytes from outside the wound migrate over the injured dermis and the granulation tissue. At the wound edges, these keratinocytes form the so-called hyperproliferative epithelium (HE), which strongly proliferates and migrates to replenish the wounded area with new cells. Cells from the HE displace the fibrin clot over time (; ; ; ). The HE (, HE) is characterized by the expression of keratins 6 and 16, which are also present in the hair follicle, but not in the uninjured epidermis (; ; ). Impairment of wound healing, e.g., in diabetes, can result in the development of chronic wounds (; ; ). Various signaling systems coordinate the wound healing process, as demonstrated by the analysis of growth factors, their receptors, and downstream signaling components (; ). For instance, genetic evidence obtained in mice indicates that signaling of the EGF receptor and the keratinocyte growth factor (KGF/FGF7) receptor are important for reepithelialization of wounds (; ; ). Furthermore, down-regulation of the TGFβ receptor in keratinocytes reduces the rate of reepithelialization (). Smad3 is a downstream component of TGFβ signaling; in contrast, Smad3 mutant mice show an increased rate of reepithelialization and reduced monocyte infiltration during wound healing (). c-Jun and STAT3 participate in the signaling of growth factors, interleukins, and integrins; conditional mutation of c-Jun and STAT3 in the epidermis delays wound closure (; ). Cell culture models have been used to simulate wound closure. In such experiments, monolayers of epithelial cells are scratch wounded, the migration of the cells is traced, and the molecular mechanisms that control migration are studied (; ). Movement of cells into scratch wounds requires modulation of cell adhesion and changes in the cytoskeleton, e.g., membrane protrusion and generation of new sites of substrate adhesion at the front, as well as actin disassembly and cell detachment at the rear. Small GTPases and protein kinases play essential roles in actin dynamics and cell migration processes, and closure of scratch wounds in cultured cells depends on Rho and Rac as well as c-Jun N-terminal kinase (; ; ; ). Wound closure in vitro is not only achieved by activities restricted to cells in the front row, but also involves cells further away from the wound edge (). Thus, dispersion and migration of single cells at the wound edge is observed, which is accompanied by movement of back row cells that maintain their cell–cell contacts and migrate as coherent cell sheets. The cellular and molecular mechanisms during the closure of scratch wounds in vitro resemble those involved in the migration of epithelial cells during the healing of skin wounds. Hepatocyte growth factor/scatter factor (HGF/SF) and its receptor c-Met promote proliferation of epithelial cells in culture, and can dissociate and scatter MDCK cells, which is accompanied by an increase in their motility (; ; ; ; ; for review see ). Mutation of the HGF/SF and c-Met genes in mice demonstrated that this signaling system is important during vertebrate embryogenesis (; ; ). During development, HGF/SF and c-Met control cell survival and proliferation of hepatocytes, as well as the formation of the placenta. Furthermore, HGF/SF and c-Met are essential regulators of cell motility. HGF/SF signals release cells from dermomyotome, and they subsequently migrate to targets where they form skeletal muscle (; ; ). HGF/SF and c-Met have also been implicated in various physiological and pathophysiological processes in the adult. For instance, during liver regeneration, HGF/SF levels in the blood stream raise, and conditional mutagenesis in mice has shown that the c-Met receptor is essential during liver regeneration and repair (; ; ; ). Furthermore, up-regulated HGF/SF and c-Met expression was observed after injury of other tissues, for instance the lung, kidney, heart, and skin (; ; ; ; ). Thus, up-regulated HGF/SF and c-Met expression might be part of a general response to tissue damage, and it is interesting to note that cytokines such as interleukin 1 or 6 activate HGF/SF transcription (). Moreover, application of exogenous HGF/SF to skin wounds promotes the formation of granulation tissue, angiogenesis, and reepithelialization, whereas neutralization of HGF/SF by the application of antibodies delays these processes (; ; ). We report that both HGF/SF and c-Met are up-regulated in the HE during wound repair in mice, suggesting that HGF/SF and c-Met signal may act in an autocrine manner to promote wound healing. We generated conditional mutant mice, in which c-Met was inactivated in the epidermis by the use of a keratin 14 (K14) promoter-driven cre recombinase. This resulted in the mutation of c-Met in ∼95% of the epidermal cells. The HE of wounds was collected by laser capture microdissection from mutant and control mice, and Southern blotting was performed to analyze the contribution of c-Met mutant keratinocytes to the newly formed epithelium. Remarkably, we found that c-Met mutant keratinocytes were completely unable to reepithelialize the wounds. Instead, residual keratinocytes that escaped recombination (5%, c-Met–positive cells) closed the wounds, but the wound healing process was delayed. These results demonstrate that the c-Met signaling system is essential for skin wound healing. Apparently, no other signaling system is able to compensate for a lack of c-Met in this process. We performed full-thickness dorsal skin wounding in mice in such a manner that the epidermis and the underlying dermis were damaged (). We then analyzed HGF/SF and c-Met expression during the wound healing process by in situ hybridization of these sections, i.e., 1–10 d after the injury. Before wounding, HGF/SF was expressed in hair follicles, but not in the epidermis (). After wounding, HGF/SF was initially expressed in the dermis adjacent to the clot (, only the left halves of the wounds are shown; shows the scheme of entire wound; ). 3 d after the injury, HGF/SF was strongly up-regulated in the HE at the edges of wounds (; control is shown in D). At this time point, the new epithelium is already formed and visible. The receptor tyrosine kinase c-Met is expressed in the epidermis and hair follicles of normal skin (), but it is also strongly expressed in the HE during wound repair (, E and F; control is shown in G). We also performed immunofluorescence staining of phosphorylated c-Met on normal skin and could show that activated c-Met is present in both epidermis and hair follicles, including hair bulge stem cells (Fig. S1, available at ). Our data suggest that during wound healing, HGF/SF and c-Met may signal in an autocrine manner in the HE. We examined the function of the c-Met signaling system in the skin using conditional mutagenesis. For this, we crossed K14-cre mice, which express cre recombinase in the epidermis starting on embryonic day (E) 15 (), with c-Met mutant mice, to generate animals with a K14-cre; c-Met genotype. In such animals, one allele of c-Met corresponded to the conventional null mutation, c-Met (), and the other to a “floxed” allele, c-Met (). After cre-mediated recombination, the exon encoding the essential ATP-binding site of c-Met is removed in c-Met, and a functional null allele, which we denote as c-Met is generated (structures of nonrecombined and recombined alleles are shown in ). K14-cre; c-Met mice will henceforth be called conditional c-Met mutant mice. Efficient recombination of c-Met was observed in the epidermis of conditional c-Met mutant mice. Southern blot analyses demonstrated that 95% of the cells in the epidermis had already recombined the c-Met allele at E17.5 (, c-Met). A similar proportion of cells containing the recombined allele were observed in the epidermis of young and adult animals, e.g., at postnatal day (P) 8 and at 12 wk (). In other epithelial tissues, such as pancreas, lung, and liver, mutation of c-Met was not observed, i.e., the nonrecombined c-Met allele was observed (). Recombination introduced by K14-cre in the skin was also assessed histologically, using the Z/AP reporter mice (); in such mice, yellow NBT/BCIP staining measures alkaline phosphatase activity, which is detectable in recombined epidermal cells, whereas blue X-Gal staining indicates β-galactosidase activity, which is observed in nonrecombined cells. The histological analysis of skin sections of these mice demonstrated that the vast majority of the cells in the epidermis had undergone recombination, and only small groups of nonrecombined cells were detectable (; the enlarged picture in shows a group of blue nonrecombined cells, which are marked by arrow). Recombination was also observed in hair follicle cells, particularly in those cells that locate in the bulge region (, F and G, yellow cells); recombination was not apparent in arrector pili muscle cells (, F and G, blue cells). We first examined the appearance of skin and hair in conditional c-Met mutant mice by histology at birth and afterward. No gross morphological changes in the epidermis were detected compared with control mice. For instance, the thickness of the epidermis was comparable, and it did not display any pathological alterations as assessed by immunohistology using antibodies directed against keratin 10 and loricrin ( and not depicted; ; ). Similarly, we did not observe changes in hair cycle progression when control and conditional mutant mice were compared. For instance, the first and second anagen phases occurred at P5 and P30, respectively (). Catagen and telogen occurred at P18 and P20 (). We have now kept conditional c-Met mutant mice for nearly 2 yr, and we have not observed unusual hair loss or other changes in the appearance of the skin. We introduced full-thickness wounds into the dorsal skin () of control and conditional c-Met mutant mice at 12 wk of age, and we analyzed wound closure by histology 3–15 d after the injury. Wound healing did occur in conditional c-Met mutant mice, but was delayed and required about twice as much time as in the control mice (; see quantification in ). For instance, 5 d after the injury, 50% wound closure was observed in control mice; in conditional c-Met mutant mice, this required 9 d (). Histological examinations showed that the HE was thinner, and its formation was delayed in the conditional c-Met mutant mice, as assessed by hematoxylin/ eosin () and by Masson trichrome staining; the latter stains the epithelium in red (). Immunohistological analysis using keratin 6 antibodies also demonstrated a reduction in the thickness of the HE in the conditional c-Met mutant mice (). Keratin 6 is expressed in activated keratinocytes of the HE and in hair follicles (; ; ). Quantification showed that the formation of the HE was delayed in the conditional c-Met mutant mice; compared with control mice, an 80% reduction of the area of the HE was observed 3 d after injury, a 65% reduction 5 d after injury, and only a 25% reduction 7 d after injury (). In contrast, the proportion of proliferating keratinocytes in the HE was increased after injury in the conditional c-Met mutants, as assessed by phospho-histone 3 and BrdU staining (). We did not observe accumulations of proliferate-positive cells at particular sites in the HE, and, in particular, we did not observe an accumulation of proliferating cells close to the remnants of the hair follicle. We also did not observe any difference in the number of apoptotic cells in the skin of control and conditional c-Met mutant mice, as assessed by TUNEL staining (unpublished data). Thus, wound healing occurred in the skin of c-Met conditional mutant mice, but reepithelialization of wounds was delayed and required about twice as much time as in control mice. To assess if the c-Met mutant cells (95% of the keratinocytes in the epidermis) could contribute to the newly formed epithelium of the wounds, we collected hyperproliferative epithelia of many control and mutant wounds by laser capture microdissection (see Materials and methods) and performed Southern blotting. An example of a section of a wound before and after microdissection is shown in . Remarkably, Southern blot analysis demonstrated the absence of c-Met mutant cells (i.e., of the c-Met allele) in the microdissected hyperproliferative epithelia of the mutant mice at day 5 (). Instead, all cells in the mutant HE contained the nonrecombined c-Met allele, despite the fact that this cell population constituted only 5% in the skin before injury. At day 3, a 1:1 mixture of c-Met and c-Met cells was seen. Immunofluorescence and immunohistochemistry staining with anti–phospho-c-Met antibodies revealed that the majority of cells in the HE at day 5 contained the activated and phosphorylated c-Met receptor, both in control and mutant skin (; compare D and F with E and G). We have observed that staining of activated c-Met is more pronounced in the upper layers of the HE, but staining is also visible in lower layers (, arrows). We also generated wounds in the skin of K14-cre; Z/AP reporter mice, and could show that K14-cre recombination occurred in the wound epithelium (). Immunohistological analyses confirmed that K14 was strongly expressed in the HE (). We conclude from these data that only nonrecombined c-Met keratinocytes, i.e., cells that express a functional c-Met, participate in the formation of the HE. In the skin of conditional c-Met mutant mice, the few remaining cells that escaped recombination can apparently compensate and generate the entire hyperproliferative epithelia. These data demonstrate that c-Met plays crucial functions during wound closure. Primary keratinocytes were isolated from the skin of newborn control and conditional c-Met mutant mice (), and monolayers in culture were scratch wounded (; ). Analysis of K14 by immunohistology indicated that the isolated cells, indeed, correspond to epithelial cells (Fig. S2, available at ). We found that in the presence of HGF/SF, primary keratinocytes from control mice closed the wounds within 48 h, as did cells from conditional c-Met mutant mice in the presence of the EGF receptor ligand TGFα (). However, in the presence of HGF/SF, keratinocytes from the skin of conditional c-Met mutant mice did not close the scratch wounds within 48 h, but required 96 h (). Strong proliferate response toward HGF/SF was observed in control cells close to the wound edges at 24 h, but such a response was not observed in the mutant keratinocytes (, quantification is shown in E). We then examined if, while in culture, the few remaining nonrecombined keratinocytes from the skin of conditional c-Met mutant mice preferentially contribute to wound closure. Primary keratinocytes were stained with phospho-c-Met antibody at different stages of scratch wound closure. In the presence of HGF/SF, control cells showed phospho-c-Met staining in a cobblestone manner (). Such a staining was associated with the membrane of virtually all cells that were close to the wounds at 24 h, and also with cells in the healed area at 48 and 96 h (). In contrast, only few keratinocytes derived from the skin of conditional c-Met mutant mice were initially phospho-c-Met positive (, left, arrow). We observed an enrichment of phospho-c-Met–positive cells at the wound edges after 48 h, and after 96 h a large proportion of cells that had closed the scratch wound contained phospho-c-Met (, arrows). Thus, residual c-Met–positive cells preferentially participate in wound closure in vitro and in vivo. We also examined if primary keratinocytes derived from the skin of control and conditional c-Met mutant mice exhibited different properties at the wound edges. The keratinocytes were characterized by immunohistological analysis using antibodies directed against proteins that are important for directed cell migration, such as vinculin, paxillin, and VASP (; , ; ). Keratinocytes derived from control skin in the presence of HGF/SF showed increased numbers of focal adhesions at the wound edges, and these pointed directly toward the wounds (, left, arrows). Actin stress fibers, which were stained by phalloidin, were also oriented preferentially toward the wounds (). Control cells at the wound edges displayed a preferential location of RhoA staining that is located at the rear of the cells (, left, arrows; ; ). These control cells at the edges of the wound also reoriented their microtubules, which was demonstrated by γ-tubulin staining (, left; ). In contrast, keratinocytes derived from the skin of conditional c-Met mutant mice displayed punctuated, cytoplasmatic, and perinuclear staining of RhoA, which is not located at the rear of the cell (, right, arrowheads). Keratinocytes from conditional c-Met mutant mice displayed only few focal contacts and stress fibers, and these were not oriented toward the wounds (, right). We also used the isolated primary keratinocytes from the skin of control and conditional c-Met mutant mice to study signal transduction by molecules that are crucial for cell proliferation and cell migration (; ; ; ; ). Erk1/2, Akt, Gab1, and PAK1/2 were phosphorylated, and thus activated, in control cells in the presence of HGF/SF and TGFα (, left). In contrast, phosphorylation of these molecules in keratinocytes derived from conditional mutant mice was pronounced in the presence of TGFα, but not of HGF/SF (, left; for quantification see ). These data demonstrate that HGF/SF and c-Met signaling is important for the activation of molecules that control proliferation and migration of primary keratinocytes in cell culture. Activation of this signaling system results in reorganization of adhesion and cytoskeleton complexes, such as focal adhesions and stress fibers, which allows cells to move into the scratch wounds. s h o w t h a t c - M e t a n d H G F / S F e x p r e s s i o n i s i n d u c e d i n t h e H E o f s k i n w o u n d s , i n d i c a t i n g t h a t r e c e p t o r a n d l i g a n d m a y a c t i n a n a u t o c r i n e m a n n e r d u r i n g w o u n d h e a l i n g . W e a l s o r e p o r t t h a t c - M e t i s c r u c i a l f o r r e e p i t h e l i a l i z a t i o n d u r i n g w o u n d c l o s u r e i n t h e s k i n a n d i n c e l l c u l t u r e ; c - M e t m u t a n t k e r a t i n o c y t e s c a n n o t c o n t r i b u t e t o t h e g e n e r a t i o n o f t h e H E i n v i v o , a n d c o n t r i b u t e i n e f f i c i e n t l y t o t h e c l o s u r e o f s c r a t c h w o u n d s i n v i t r o . A n a l y s i s o f c u l t u r e d k e r a t i n o c y t e s d u r i n g t h e c l o s u r e o f s c r a t c h w o u n d s i n d i c a t e s t h a t t h e p r i m a r y d e f i c i t o f t h e m u t a n t c e l l s i s t h e i r i n a b i l i t y t o p r o l i f e r a t e a n d t o m i g r a t e i n t o t h e w o u n d e d a r e a . To mutate c-Met in the skin, the K14-cre mouse line was used, which expresses cre after E15 of embryogenesis (). 2–3-mo-old mice were anesthetized, and two full-thickness 4-mm-diam excisional wounds were made on both sides of the dorsal midline, as previously described (). Mice were kept separately in cages to prevent fighting, and no self-induced trauma was observed in control or mutant mice. Littermates of the same sex were used for the analysis of wound closure. Wound closure was determined as the percentage of the distance covered by the epidermis between the wound edges. For histological analysis, dissected wounds were fixed overnight in 4% formaldehyde in PBS, followed by dehydratation through a graded ethanol series and embedding in paraffin. Sections from the middle of the wound were stained with hematoxylin/eosin, and immunofluorescence was performed as previously described (). Histological sections were used to determine the area of HE by using the measure function of the Adobe Photoshop program. BrdU incorporation was assessed by immunohistochemistry 2 h after injection of 75 μg BrdU/g of body weight. The proliferation index represents the percentage of BrdU-positive cells within the HE. We used antibodies directed against keratin 6 (Covance), phospho-histone H3 (Millipore), PCNA (Oncogene Science), BrdU (Sigma-Aldrich), CD34 (Biozo), phospho-Met (Sigma-Aldrich), and K14 (Covance). Sections were incubated overnight with primary antibodies at 4°C, followed by fluorescent-conjugated secondary antibodies (Jackson ImmunoResearch Laboratories). TUNEL staining was performed using an in situ cell-death detection kit (Roche). Sections from the middle of the wound were also stained using the Masson trichrome procedure (Sigma-Aldrich). Stained tissues were analyzed on a confocal scanning laser microscope (LSM; Carl Zeiss MicroImaging, Inc.). In situ hybridization of paraffin sections was performed using digoxygenin-labeled RNA probes (Roche; ). The antisense transcripts of mouse cDNAs were as follows: a 1.4-kb HGF/SF fragment that encompasses the 3′ coding sequence, a 0.7-kb HGF/SF fragment that encompasses the 5′ coding sequence, and a 3.7-kb c-Met fragment. Western blot analysis was performed using antibodies specific to Erk1/2, phospho Erk1/2, Akt, phospho-Akt, phospho-Gab1, and phospho-PAK1/2 (Cell Signaling Technology). For laser capture microdissection, 46 wound areas were excised, embedded in Tissue-Tek OCT compound and snap-frozen in liquid nitrogen. Frozen sections (8 μm) were prepared and fixed in 70% EtOH for 60 s and stained shortly with hematoxylin/eosin. Laser capture microdissection was performed using an Arcturus PixCell II apparatus. Southern blot analysis of hyperproliferative epithelia from 400 microdissected sections was performed according to standard procedures. Primary murine keratinocytes were isolated from newborn mice, as described previously (), and cultured on collagen IV–coated plates in defined keratinocyte serum-free medium (Invitrogen). The confluent monolayers of primary keratinocytes were scratch wounded by a pipette tip and further cultured in the presence of growth factors: 10 U/ml HGF/SF () and 20 ng/ml TGFα (Sigma- Aldrich). For immunostaining, cells were washed with PBS and fixed in 4% formaldehyde in PBS for 15 min at room temperature. The following antibodies were used: antivinculin (Sigma-Aldrich); antipaxillin (BD Biosciences); anti–phospho-c-Met, anti-RhoA, anti–γ-tubulin (Santa Cruz Biotechnology); and anti-VASP (Cell Signaling Technology). Additional reagents used were TRITC/FITC phalloidin (Sigma-Aldrich) and YO-PRO (Invitrogen). Fig. S1 shows the expression of c-Met in bulge stem cells. Fig. S2 shows isolated keratinocytes from control and conditional c-Met mutant mice. The online version of this article is available at .

In recent years, considerable progress has been made in understanding the role of growth factors and cytokines in wound healing. These factors are released from injured vessels and coagulated platelets and trigger an inflammatory response that initiates the deposition of a provisional extracellular matrix. In parallel, mesenchymal and endothelial precursor cells invade the wound to form the granulation tissue and to contract the wounded area. Contraction of the granulation tissue by mesodermal fibroblasts is of high importance for sealing the wound, as it helps to bring the wound margins together. To be able to contract efficiently, mesenchymal cells of the granulation tissue differentiate into myofibroblasts characterized by a well-developed cytoskeleton. Myofibroblasts express α-smooth muscle actin (α-SMA), which is incorporated into actin stress fibers and enables them to develop much higher mechanical forces (). Hence, induction of α-SMA expression in fibroblasts is a critical step in wound healing. Besides growth factors and cytokines, bioactive lipids have been identified as important signal molecules, modulating inflammatory responses, cell growth, and tissue formation. However, the role of lipid-induced signaling and its contribution to wound healing is still poorly understood. We previously showed that sphingosine-1-phosphate (S1P) triggers a signal transduction cascade mediating nuclear translocation of the LIM-only protein Fhl2 in response to activation of the RhoA GTPase (, ). We and others further identified the LIM-only protein Fhl2 as interacting with transcription factors, including androgen receptor (), serum response factor (SRF; ; ), Jun, and Fos (), as well as with integrin receptors (; ) and signal transducers such as β-catenin (). Fhl2 associates with integrins at focal adhesion sites and is translocated into the nucleus upon stimulation by serum or S1P to modulate transcriptional activity of numerous target genes. Because significant amounts of S1P and lysophosphatidic acid are released from platelets during tissue repair (), we investigated the role of Fhl2 signaling in mesenchymal cells during wound healing. Previous studies of Fhl2 in prostate cancer revealed its expression in the prostate cancer cells proper, but also in myofibroblast-like cells within the stroma (, ). To confirm expression and regulation of Fhl2 in mesenchymal cells, we isolated primary embryonic mouse fibroblasts and visualized Fhl2 expression and nuclear translocation in response to serum exposure. When fibroblasts were starved in 0.5% FCS, and then exposed to 10% FCS for 24 and 48 h, respectively, we observed both significant up-regulation of Fhl2 mRNA and nuclear translocation of the respective protein (Fig. S1, available at ), as previously described (). More importantly, we observed strong up-regulation of Fhl2 in α-SMA–positive mesenchymal cells of wounded skin (). A series of five different tissue specimens obtained 5–14 d after wounding were analyzed by Fhl2 immunostaining. In all cases, we observed very strong signals present in the cytoplasm and the nucleus of myofibroblast-like cells of the granulation tissue, but not in differentiated fibroblasts from normal skin (). The myofibroblasts are characterized by α-SMA and SM22 immunoreactivity and, importantly, double-immunostainings of tissue sections for both Fhl2 (, red stain) and α-SMA or SM22 (, brown stain) indicated that the most abundant site of Fhl2 expression in the granulation tissue are indeed myofibroblasts. In contrast, keratinocytes did not reveal any Fhl2 immunoreactivity. These results, along with data obtained from human skin biopsies in vivo and from serum stimulation of cell lines, indicated that Fhl2 is up-regulated in myofibroblasts during wound healing and that it shuttles into the nucleus in response to exposure to bioactive lipids present in blood, as previously described (; ; ). We applied punch biopsy wounds to skin and cutaneous muscle of wild-type () and -deficient ( ) mice and conclusively found significant up-regulation of both Fhl2 mRNA and protein expression in mice during skin regeneration, with a maximum at 5 d after wounding (). mice lack mRNA and protein expression (). genetic background ( ). mice revealed severely impaired wound healing because only 10% of skin wounds were closed after 5 d, compared with 40% in mice (). mice. transgenic mice that express intermediate Fhl2 mRNA and protein levels in a genetic background, displayed a nearly wild-type phenotype, with 30 and 90% wound closure at days 5 and 12, respectively, demonstrating rescue of the wound closure phenotype of mice. The same transgene expressed in a background, however, did not influence wound healing, indicating that the high levels of Fhl2 expression in mice are both necessary and sufficient for efficient wound healing. At each time point, 38–42 lesions were evaluated by measuring wound closure macroscopically, as well as by histological and immunochemical staining of skin sections. Collectively, our data indicate that the efficiency of wound closure correlates with the amount of Fhl2 mRNA and protein expression in wounds. Fibroblasts play a key role in the formation of mechanical forces that lead to wound contraction, which is required to bring the wound margins together. Therefore, we were interested in correlating the Fhl2 function with extracellular matrix remodeling and contraction. We analyzed the fibroblast-mediated contraction of type I collagen gels as an in vitro model of tissue remodeling. mice displayed a severe defect in collagen contraction, with a half-maximal contraction time of >60 h, compared with 10.4 h in cells (). cells that we were unable to measure exactly the half-maximal contraction time within the observation interval. Because bioactive lipids stimulate fibroblast-mediated collagen contraction (), we analyzed collagen contraction in the presence of S1P. This resulted in a decreased half-maximal contraction time of 8.6 h for fibroblasts (). fibroblasts. fibroblasts by transfection of an appropriate expression plasmid fully restored the capability to contract collagen (), demonstrating that Fhl2 is an essential component in this tissue remodeling assay. During wound healing, mesenchymal cells differentiate upon stimulation by inflammatory cytokines into myofibroblasts, which produce high amounts of α-SMA and are involved in wound contraction in vivo (). The early phase of wound healing is triggered by serum components released from injured blood vessels and degranulating platelets (), which activate the transcription factor SRF (). SRF and Fhl2 interact physically () and bind to the promoter of the SRF-responsive α-SMA gene. Degranulating platelets release large amounts of bioactive lipids, including S1P and lysophosphatidic acid, into wounds that, in turn, trigger nuclear translocation of Fhl2 (). Therefore, we addressed the question of whether Fhl2 may function as a transcriptional cofactor of SRF in activating α-SMA expression during wound healing. For these assays, we tested cells of different origin (epithelial human embryonic kidney [HEK] 293 cells, mesenchymal stem cells, and fibroblasts), which are devoid of endogenous Fhl2 expression or with a Fhl2 knockout genotype, respectively. These cells were cotransfected with SRF- and Fhl2-expression constructs, together with a reporter plasmid carrying an α-SMA promoter-driven luciferase gene. Although expression of Fhl2 in HEK293 alone did not change reporter activity, we observed between two- and threefold activation in the mesenchymal cells (). SRF mediated an approximately fourfold increase of reporter expression in all cell lines. Importantly, expression of both SRF and Fhl2 resulted in approximately two- or threefold higher reporter activity than expression of SRF alone (), indicating coactivation of SRF-mediated transcriptional activity in all three cell lines. and mice. As expected, immunohistochemical staining revealed strong expression of α-SMA in myofibroblasts of the granulation tissue below the wound surface at day 5 in mice, but only very weak signals in knockout animals (). mice (relative units, 1.25 ± 0.6 at day 5 and 1.4 ± 1.0 at day 12, respectively) than in mice (relative units, 2.6 ± 0.75 at day 5 and 2.0 ± 0.6 at day 12, respectively). The difference in α-SMA staining intensity was that it was statistically significant at day 5 (P < 0.001) and was still significant at day 12 (P < 0.1). Importantly, immunostainings of the transgenic rescue mouse strain did not reveal any difference in α-SMA reactivity compared with mice. mice. Cutaneous wound healing is inevitably associated with migration of mesenchymal precursor cells and their subsequent differentiation into myofibroblasts. Within the first days after wounding, mesenchymal cells invade the wound to replace the clot and to form a granulation tissue. mice (). The morphology of different Fhl2 clones was quite similar, but differed from that of Fhl2 clones. cells showed a more epithelial-like form and had a less polar shape, often with many short actin stress fibers running in different directions. cells had a more fibroblast-like form, with many filopodial and lamellipodial structures. They displayed a well-organized actin cytoskeleton with long microfilament cables running across the whole cell body (), and Fhl2 was localized at focal adhesion structures, as well as along the actin filaments. cells ( and Videos 1 and 2, available at ). cells showed much less activity in the formation of filopodia or lamellipodia and, consequently, needed almost twice as much time to close a cell-free cleft in comparison with cells. ( and Videos 1 and 2). Importantly, ectopic expression of a myc-tagged Fhl2 protein (Fig. cells ( and Video 3). The ectopic expression of Fhl2 not only rescued the motility phenotype but also reverted the cell shape and actin cytoskeleton organization to that of stem cells (). Impaired cell motility was independent of the substrate on which the cells migrated (fibronectin, laminin-1, or no substrate) and of the cell origin. On uncoated dishes, cell movement was slower, with 10.8 ± 1.4 μm/h for , 5.8 ± 0.9 μm/h for , and 9.6 ± 0.9 μm/h for rescued mesenchymal stem cells. On fibronectin-coated dishes, the migration velocity was 17.1 ± 0.7 μm/h for , 8.8 ± 0.4 μm/h for , and 12.0 ± 1.5 μm/h for -rescued cells, respectively. cells was not caused by changes of the integrin pattern on their surface, as , , and rescued cells all expressed equal amounts of integrin β1–containing receptors (Fig. S2 B). or rescued cells attached equally well to proteins of the extracellular matrix, suggesting that different adhesion properties are not responsible for the reduced migratory capacity. cells remarkably resemble the phenotype of FAK-deficient cells (). In addition, it is known that FAK has to be activated for cell migration (). After adhesion to extracellular matrix molecules, FAK is autophosphorylated at tyrosine Y397, recruiting Src, which in turn phosphorylates FAK at additional Y residues, including Y861, which serves as the binding site for p130Cas. Interaction of p130Cas and FAK leads to recruitment of multiple other proteins, finally resulting in the formation of lamellipodia and cell migration (; ). -rescued mesenchymal cell lines (Fig S3, available at ). cells. We previously showed that Fhl2 directly binds to integrins () and FAK (), and that it is localized at focal adhesion sites (). cells showed that Fhl2 coimmunoprecipitated with FAK and p130Cas, but not with Src, when cells were plated on fibronectin-coated dishes (). cells, whereas the amounts of FAK and Src were not altered (). cells. Next, quantitative real-time PCR experiments were performed to study whether the changes in p130Cas expression levels result from differences in mRNA expression. Amplification curves for p130Cas and, as a reference gene, cyclophilin, were obtained with template cDNA from Fhl2 and Fhl2 mesenchymal stem cells. Each curve shown in the Fig. S4 A (available at ) represents the mean of three replicates from a single cDNA sample. The amplification of p130Cas cDNA was delayed in Fhl2 cells compared with cells, indicating a lower p130Cas mRNA amount. The difference between the average Ct-value of p130Cas and cyclophilin (ΔCt) was calculated for both cell lines. These values were compared (ΔΔCt), and the relative amount of p130Cas mRNA was calculated and diagrammed (Fig. S4 B). In summary, our data clearly indicate that Fhl2 knockout cells express roughly twofold lower p130Cas mRNA levels. Recruitment of p130Cas subsequently leads to activation of Rac and cell migration (; ). mesenchymal stem cells would be able to rescue the defect in cell migration. Knockout stem cells were infected with retroviruses expressing either p130Cas along with GFP or GFP alone as a control. The p130Cas and GFP genes were connected by an internal ribosomal entry-site sequence. Evaluation of GFP-labeled cells indicated that the infection efficiency was 94.1 and 95.2%, respectively (, left). Consistently, Western blots indicated robust expression of p130Cas in the knockout cells (, right). cells, but did not reach the velocity of cells. These results were obtained independently of whether cells migrated on noncoated or on fibronectin-coated surfaces (). cells, but not entirely to the level of cells. Finally, we asked whether changes in expression of p130Cas resulted in different levels of Rac activation. cells stably expressing p130Cas or the empty vector were assayed for Rac activity. The cells were serum-starved overnight, trypsinized, and plated for 15, 30, or 60 min, respectively, on cell culture dishes precoated with 20 mg/ml fibronectin. For precipitation of GTP-loaded Rac, cells were lysed in Triton X-100 lysis buffer, and 400 mg protein were rotated with GST-PAK3–coated glutathione beads. cells was observed. cells reconstituted with p130Cas restore their capability to activate Rac in response to attachment to fibronectin. Previous studies, mainly based on cell lines in vitro, established Fhl2 as a serum-responsible signal transducer shuttling in response to SP1 and lysophosphatidic acid from the cell membrane into the nucleus, where it functions as a nuclear coactivator of transcription factors. However, only few transcriptional targets, including Fhl2 itself, were described, and the function of Fhl2 signaling in vivo is much less explored. Although a function of Fhl2 in promoting differentiation of myoblasts was suggested (), Fhl2 knockout mice developed only a mild phenotype with bone formation defects and an increased sensitivity in respect to a hypertrophic response to β-adrenergic stimulation in the heart (; ; ; ). Data presented in our study indicate that Fhl2 further mediates nonredundant signaling during wound healing. mice clearly revealed delayed wound healing, reduced migration of mesenchymal precursor cells, delayed activation of α-SMA, and impaired wound contraction. The Fhl2 protein is activated in dermal fibroblasts after release of bioactive lipids in wounded tissue and, indeed, we show that Fhl2 regulates the expression of α-SMA by coactivation of SRF, and thereby the contractility of the granulation tissue. Therefore, it seems that nuclear shuttling and transcriptional coactivation of Fhl2 developed as a signaling pathway mediating rapid adaptation of cells and tissues in response to pathological stress conditions. Our data further indicate that Fhl2 signaling is cell-type specific and different from its function in cardiac muscle cells, where it negatively regulates expression of SMA (). In addition, Fhl2 interacts with proteins of focal adhesion structures at the membrane or cytosolic level, and we provide first evidence that because of this interaction Fhl2 regulates cell motility and contractility. Contraction of the granulation tissue facilitates wound closure by bringing the wound margins together. Efficient contraction of myofibroblasts requires a well-developed cytoskeleton, which is established by expression of α-SMA and its incorporation into actin stress fibers (). Hence, expression of α-SMA by skin fibroblasts is a critical step in wound healing. Interestingly, our data for the first time provide a mechanistic link between release of the bioactive lipids S1P and lysophosphatidic acid from platelets during clotting and wound healing and the contractile activity of the granulation tissue. These substances trigger, in a Rho-dependent manner, nuclear shuttling of Fhl2 () where it acts as a coactivator of α-SMA transcription. Consistent with these data, we further demonstrated that in the absence of Fhl2, the contractile forces of fibroblasts are dramatically reduced and that this defect can be rescued by expression of exogenous Fhl2 protein. It is well known that FAK plays a key role in cell migration. It is activated upon integrin engagement and recruits several cytosolic proteins that drive cell migration. cells. Our data, however, also indicate that the mechanism by which Fhl2 regulates cell migration is more complex and cannot be reduced just to the level of p130Cas protein, as its overexpression did not restore migration velocity of mesenchymal cells to the full level of cells. Thus, it appears that Fhl2 activation in mesenchymal cells after wounding regulates different effector functions of activated FAK. A separate study of our group provided evidence that Fhl2 is also involved in organization of focal adhesion structures and in regulation of matrix assembly (unpublished data). mice display a cutaneous wound-healing phenotype that can be rescued by ectopic expression of Fhl2. cells, which lead to severe defects in collagen contraction and migration. Thus, lipid-triggered Fhl2 signaling is mechanistically involved in regulating wound healing and may represent a new therapeutic target. mice were provided by R. Bassel-Duby (University of Texas Southwestern Medical Center, Dallas, TX) and published previously (). For the generation of transgenic mice, the human Fhl2 cDNA was coupled with a 1.4-kb SM22α promoter () and animals were obtained according to published procedures (). Genotyping was done by PCR analysis from tail genomic DNA using the primer pairs 5′-GACTGCTCCAACTTGGTGTCTTTC-3′ and 5′-TCCCGCAGGATGTACTTCTTGC-3′ in 35 amplification cycles (95°C for 30 s, 54°C for 30 s, and 72°C for 30 s). All animals were maintained in a pure C57BL/6 background, and subpairs were used for the wounding experiments. 48 6-wk-old mice (18 , 18 , and 12 transgenic mice) were used. 2–4 0.6-cm punch wounds, including the skin and cutaneous muscle, were cut into each mouse and left to heal by secondary intention, essentially as previously described (). At days 0, 5, and 12, wounds were dissected and paraffin-embedded for histology or snap-frozen in liquid nitrogen for RNA and protein extraction. All experiments were performed in compliance with animal welfare regulations (Permission No. 50.203.2-BN12, 12/02 by the Regierungspräsidium, Cologne, Germany). Mouse embryonal fibroblasts were obtained by standard procedures and maintained in DME (Invitrogen) supplemented with 100 U/ml penicillin, 10 μg/ml streptomycin, and 10% FCS (Invitrogen). Transient transfection of fibroblasts was done using the Amaxa system (Amaxa) with transfection efficiency >50% measured by GFP expression (). Collagen contraction was performed as previously described (; ). In brief, 250 μl of fibroblast suspension (10 cells/ml) were added to 3 ml collagen type I solution (3 mg/ml) and placed into a 30-mm Petri dish (Greiner). Contraction of the developing collagen sponge was determined by measuring the diameter every 1 h. mice as previously published (). The expanded cells had a doubling time of ∼35 h and were positive for CD34, c-kit, sca1, Thy1, and CD13, and negative for CD45, CD10, and CD31 marker as determined by PCR. According to these markers and to their potency to differentiate into osteogenic, chondrogenic, and adipogenic lineages, we identified them as mesenchymal cell lineages. The cells were maintained in a mixture of DMEM and MCDB-201 medium supplemented with 2% FCS, 10 ng/ml EGF (Sigma-Aldrich), 10 ng/ml PDGF (R&D Systems), 1,000 U/ml of mouse LIF (CHEMICON International, Inc.), 1× insulin–transferrin–selenium mixture (Sigma-Aldrich), and 10M dexamethasone (Sigma-Aldrich). rescue cells were obtained by infection of cells with retroviruses containing a myc-tagged human Fhl2, as we previously described (). Total cellular RNA was extracted from harvested cells or homogenized wound specimens by lysis in guanidinium isothiocyanate. 10 μg was separated by electrophoresis in a 1.2% agarose/formaldehyde gel, transferred to a nylon membrane (Hybond N; GE Healthcare), and probed with radiolabeled cDNA. Soluble protein lysates were extracted from cells or homogenized wound specimens in 150 mM NaCl, 10 mM Tris, pH 7.2, 0.1% SDS, 1% Triton X-100, and 1% deoxycholate and 5 mM EDTA and centrifuged at 13,000 for 20 min at 4°C. 15 μg of protein lysates were denatured at 90°C for 10 min, run on 12% SDS-PAGE gels, and electroblotted to a PVDF membrane (Roti-PVDF; Roth GmbH) using standard protocols. After blocking in 5% nonfat dry milk/PBST for 2 h, the membranes were incubated for 1 h with a monoclonal anti-Fhl2 antibody (dilution 1:2,000), washed, incubated with horseradish peroxidase–conjugated secondary antibody (dilution 1:1,000; DakoCytomation), and developed using ECL chemiluminescence (GE Healthcare). As a control, blots were probed with a primary anti–β-Actin antibody (dilution 1:5,000; DakoCytomation). Images were captured on film, digitized, and if needed, minor linear adjustments in contrast were made using Photoshop software (Adobe). Total RNA was extracted with the RNeasy kit (QIAGEN) from two independent samples of Fhl2 and Fhl2 stem cells, respectively. Reverse transcription of RNA (1.5 μg) was performed with oligo(dT) primers and RevertAid H Minus M-MuLV reverse transcriptase (Fermentas MBI). For PCR amplification of cDNA, specific primers (MWG) were used to detect differences in the expression levels of p130Cas; primers for murine p130Cas were chosen according to . Primers for the reference gene cyclophilin were as follows: 5′-CCACCGTGTTCTTCGACAT-3′ (upstream) and 5′-CAGTGCTCAGAGCTCGAAAG-3′ (downstream). The PCR reactions were done in triplicate for each cDNA after the Stratagene protocol with 2× Brilliant SYBR Green QPCR Master Mix (Stratagene), with preheating at 95°C for 10 min; 40 cycles of 95°C for 30 s, 60°C for 1 min, and 72°C for 30 s; and 95°C for 1 min, 60°C for 30 s, and 95°C for 30 s. MxPro Software (Stratagene) was used for analysis. 4-μm tissue slides were cut from formalin-fixed and paraffin-embedded wound specimens and used for staining with hematoxylin and eosin or by immunohistochemistry. Indirect immunohistochemistry was done by the avidine-biotin method, as previously described (). Primary antibodies were anti–human α-SMA (1:25 dilution; DakoCytomation), anti-SM22 (1:100 dilution; DakoCytomation), anti–cytokeratin-5 (1:100 dilution; DakoCytomation), and anti–collagen type I (1:100 dilution; ICN Biochemicals). Slides were incubated with a secondary goat anti–mouse serum (dilution 1:200; DakoCytomation), reacted with the ABC kit (Vector Laboratories), and peroxidase activity was visualized with 3-amino-9-ethylcarbazole (Sigma-Aldrich). Double immunostaining with a second alkaline phosphatase–labeled antibody (DakoCytomation) was done as previously described (). Pictures were taken by using a light microscope DM LB2 (Leica) and the analysis system software Diskus (Hilgers). For immunofluorescence staining, 5 × 10 fibroblasts were seeded in chamber slides (Nunc), grown to 75% confluency, and incubated for 48 h in medium containing 10 or 0.5% FCS. Indirect immunofluorescence staining was done as previously described (), using rabbit anti-Fhl2 antibody (1:300), anti-Fhl2 mAb clone F4B2 (), or anti-myc mAb derived from clone 9E10 (American Type Culture Collection). Cell images were taken using an Axiovert 2000 ApoTome microscope with an AxioCam digital camera and AxioVision software (Carl Zeiss MicroImaging, Inc.). Transfections of 293 cells and luciferase assays were performed as previously described (). 500 ng of the reporter plasmid pSM8pGL3 were cotransfected with expression plasmids coding for SRF (2.5 ng) and Fhl2 (5 ng pCMX-Fhl2) as indicated. fibroblasts were performed with Lipofectamine (Invitrogen), and transfections of stem cells were performed with Fugene 6 (Roche) as recommended by the manufacturers. Relative light units were normalized to protein concentration using the Bradford dye assay (Bio-Rad Laboratories). For construction of pSM8pGL3, the α-SMA promoter and the first intron (SMP8; a gift from E.P. Smith, University of Cincinnati College of Medicine, Cincinnati, OH) were cloned in pGL3 (Promega). SMP8 contains −1,074 bp of the 5′- flanking region, 63 bp of 5′-UT, and the 2.5-kb first intron of the α-SMA. Cell migration studies were performed essentially as previously described (). In brief, 5 × 10 cells in 0.8 ml of DMEM with 10 ng/ml EGF and PDGF were plated onto 48-well plates, which were precoated with fibronectin, laminin-1, or nothing and blocked with 1% BSA. To produce a cell-free “window,” 1-mm-thick steel plates were inserted into wells before seeding the cells and were removed again after the cells had been attached to the bottom. This method has the advantage over the frequently used “scratch window” assay in that the substrate in the window is not destroyed. The migration was monitored by inverted microscopy at the times indicated. For videos, the scratch assay was used. 10 cells were suspended in FACS-PBS (PBS containing 2% FCS and 0.02% NaN). Cells were incubated with integrin anti-β1 Abs (clone 9EG7; BD Biosciences) for 20 min on ice, washed twice with FACS-PBS, and incubated with Cy2-conjugated secondary antibodies (DakoCytomation) for additional 15 min. After washing the cells, measurements were performed with a FACSCalibur flow cytometer (BD Biosciences). For all statistical analyses, the Cochran-Armitage trend test was used and a P-value <0.05 was considered statistically significant. To quantify the α-SMA immunohistochemical staining results, the following scoring system was applied: no staining, 0; weak staining, 1; moderate staining, 2; maximal staining, 3. 100 cells of each sample were evaluated. Fig. S1 shows that Fhl2 mRNA is serum-inducible in embryonic mouse fibroblasts and that Fhl2 translocates into the nucleus and along the actin cytoskeleton in response to FCS. Fig. S2 shows the migration activity of , , and rescued mesenchymal stem cells (Videos 1–3 are time-lapse movies correlating to Fig. S2). Fig. S3 shows that the absence of Fhl2 does not influence FAK autophosphorylation after adhesion of stem cells to fibronectin. Fig. S4 shows results from real-time qRT-PCR, indicating that Fhl2 cells express reduced levels of p130Cas mRNA. Online supplemental material is available at .

Lipoxygenases (LOXs) represent a widely distributed family of nonheme, nonsulfur, iron-containing dioxygenases that catalyze the regioselective and stereoselective dioxygenation of fatty acid substrates containing one or more (Z,Z)-1,4-pentadiene moieties (). Within the mammalian LOX family, a distinct subclass of epidermis-type LOX has been characterized that are preferentially expressed in skin and few other epithelial tissues (). They include the human 15-LOX-2 and its mouse orthologue 8-LOX, 12R-LOX, and eLOX-3. Their genes map close together within a gene cluster on human chromosome 17p13.1 that was found highly conserved within a syntenic region at the central region of mouse chromosome 11 (). Although exhibiting a rather heterogeneous regio- and stereospecificity, the epidermis-type LOX are phylogenetically closely related, sharing ∼50% amino acid identity. Their differentiation-dependent expression pattern in epithelial tissues suggests a common physiological role in the regulation of proliferation and differentiation of epithelial cells, especially keratinocytes. The epidermal 12R-LOX and eLOX-3 differ from all other mammalian LOX in their unique structural and enzymatic features (; ; ). Both proteins contain an extra domain located at the surface of the catalytic subunit. 12R-LOX represents the only mammalian LOX that forms products with R-chirality, and, unlike all other LOX, eLOX-3 does not exhibit dioxygenase activity, but functions as a hydroperoxide isomerase (). Both enzymes act in sequence to convert arachidonic acid via 12R-hydroperoxyeicosatetraenoic acid (12R-HPETE) to the corresponding hepoxilin-like epoxyalcohol, 8R-hydroxy-11R,12R-epoxyeicosatrienoic acid. This sequence has been hypothesized to be part of a novel LOX pathway in skin that plays an important role in terminal differentiation (; ). Recent genetic studies have identified mutations in the coding regions of 12R-LOX and eLOX-3 genes in patients with autosomal recessive congenital ichthyosis (ARCI), linking for the first time mutations of a LOX gene to the development of a disease (; ). ARCI is a clinically and genetically heterogeneous group of skin disorders that is associated with hyperkeratosis and impaired skin barrier functions (). We and others recently showed that the point mutations found in the LOX genes of the ARCI patients completely eliminated the catalytic activity of the LOX enzymes, indicating that mutational inactivation of either 12R-LOX or eLOX-3 is causally linked to the ARCI phenotype (; ). To investigate the physiological role of 12R-LOX and to analyze the molecular mechanisms that underlie the ichthyosiform skin phenotype, we developed mice with targeted inactivation of the 12R-LOX gene. Examination of the resulting phenotype has revealed a crucial role of 12R-LOX in the development of epidermal barrier function, demonstrating for the first time an indispensable function of a LOX isoform for postnatal survival of mice. For targeting the gene, we used the Cre-loxP system. A targeting vector was constructed by placing a resistance cassette flanked by loxP sites into intron 7 of . A third loxP site was inserted downstream of exon 8 (). This exon encodes a highly conserved region containing two of the iron-binding histidines that are absolutely required for LOX catalytic activity (; ). Thus, Cre-mediated excision of this region yields a nonfunctional protein. The targeting construct was electroporated into E14 embryonic stem (ES) cells, followed by G418 selection. Two ES cell clones had correctly recombined alleles and were used to generate germline chimeras. Heterozygous floxed mice ( ) were mated with CMV-Cre deleter mice () to validate our construct and to generate mice harboring a disrupted allele. Correct recombination and complete excision of the resistance cassette and exon 8 were confirmed by PCR analysis and Southern blot analysis, yielding the expected BamHI fragments (). mice were intercrossed to generate homozygous mutant mice. mice. RT-PCR and Western blot analysis demonstrated the expression of 12R-LOX RNA and protein in skin isolated from neonatal mice. The mutated RNA lacking exon 8 was expressed in heterozygous and homozygous newborn mice, and no wild-type RNA was detected in skin from homozygous mutant mice (). mice. A predicted truncated 12R-LOX was not detectable, indicating translational suppression or instability of the mutated protein (). The lack of 12R-LOX protein expression in the skin of homozygous mutant mice was confirmed by immunofluorescence. Using a 12R-LOX–specific mAb, a prominent and specific staining throughout the plasma membranes of keratinocytes in the stratum granulosum was seen in wild-type mice, but was absent in the homozygous mutant mice (). mice were phenotypically indistinguishable from wild-type mice and reproduced normally. mice were hard to distinguish from wild-type and heterozygous littermates upon macroscopic inspection. However, their skin soon began to develop a red, shiny appearance and became somewhat sticky to the touch (). The neonates did not feed and became progressively dehydrated. All homozygous mutant mice died within 3–5 h after birth. The macroscopic appearance and the early neonate death of the mice suggested a perturbed water barrier. mice lost ∼30% of their body weight within 3 h, whereas their heterozygous and wild-type littermates maintained their weight (). Transepidermal water loss (TEWL) of homozygous mutant mice was increased by a factor of ∼8 compared with wild-type and heterozygous littermates (). Thus, the lethal phenotype of the 12R-LOX–deficient mice most likely resulted from water loss as a result of impaired epidermal barrier function. Barrier formation that starts around E16 in a patterned fashion () was measured with a whole-mount skin toluidine blue penetration assay. In wild-type animals, the staining pattern reflects the decrease of skin permeability from embryonic day (E) 16.5 to E17.5, when the barrier development proceeds in a dorsal to ventral pattern, up to E18.5, which is when skin has become completely impermeable. Skin of homozygous mutant mice, in contrast, remained permeable, as indicated by intense staining (). We then assessed the permeability of the newborn epidermal barrier from outside using the fluorescent dye Lucifer yellow. In skin of 12R-LOX–deficient mice, the dye was found to penetrate throughout the stratum corneum, whereas it was retained in the very top layers in the skin of wild-type mice (). These findings clearly indicate that both the inside-out and the outside-in water barrier function were severely affected in the epidermis of 12R-LOX–deficient mice. We also assessed the barrier function of tight junctions by injecting newborn mice subcutaneously with biotin and measuring its diffusion into the epidermis. Prevention of diffusion was observed in the upper granular cells of the skin of homozygous mutant mice, as well as in wild-type mice, indicating that 12R-LOX deficiency did not affect the barrier function of tight junctions (unpublished data). mice did not exhibit overt abnormalities in the stratified organization of keratinocytes at the level of hematoxylin and eosin–stained paraffin section images (). No substantial differences were observed in the basal, spinous, or granular layers. The stratum corneum, however, appeared to be more tightly packed compared with that of control skin. To unveil defects in differentiation, we analyzed the level of expression and distribution of terminal differentiation markers. Western blot analysis revealed that the levels of keratin 5 and 10, loricrin, and involucrin were not altered in the -null versus control mice (). However, a complete loss of filaggrin monomer in knockout epidermis was associated with enhanced levels of proteolytically derived intermediates, indicating that proteolytic processing of profilaggrin was impaired. On the other hand, immunofluorescence showed a comparable distribution of the filaggrin expression in both genotypes. Similarly, other markers for epidermal differentiation, such as repetin and desmosomal proteins (claudin and occludin), revealed staining profiles that did not differ between the 12R-LOX–deficient mice and their wild-type counterparts (unpublished data). Immunofluorescence analysis indicated absence of the hyperproliferative keratin 6. Accordingly, the proliferation index as determined by Ki67 staining was also unchanged (45.4 ± 3.2% vs. 46.4 ± 0.6% in control and knockout mice, respectively). This indicates that basal cell proliferation was unaffected. The structure of the hair follicles also seemed to be normal. Histological analysis of methylene blue–stained semithin sections revealed obvious structural anomalies in the skin of the homozygous mutant mice (). Numerous vacuole-like structures were observed in the upper granular layers of the epidermis underlying the stratum corneum. The vesicles were variable in size and irregularly distributed. Ultrathin section EM confirmed these findings at higher resolution (). mice, the cells of the granular layer regularly contained vesicular structures of variable sizes (, asterisks). Sometimes they seemed to have fused. Most vesicles were electron-lucent with some smaller vesicles inside (). At higher magnifications, we could observe lamellar structures adhering to the surrounding membrane (). Normal looking lamellar bodies were also present (). Using ruthenium tetroxide after fixation, the typical lipid lamellae within lamellar bodies (), as well as the stacks of lipid lamellae representing extruded content of lamellar bodies in the transition zone between the last granular cell layer and the first cornified cell (), could be visualized. No difference in respect to the organization of the stacks or number of lamellae could be determined so far. We prepared CEs from control and 12R-LOX newborn epidermis to assess morphology and resistance to mechanical stress. Microscopic examination revealed no obvious structural abnormalities of the CEs isolated from mutant mice. Upon ultrasound treatment, however, the percentage of intact corneocytes from homozygous mutant mice decreased significantly (P > 0.01) faster with time, compared with CEs from wild-type and heterozygous littermates, indicating an increased fragility of the mutant CEs (). The structural anomalies in the skin of 12R-LOX–null mice indicated that defects in the lipid metabolism may be associated with the observed phenotype. It is well known that lipid constituents, such as free fatty acids, cholesterol, and ceramides, play an important role in epidermal barrier function. On the extracellular surface of the CE there is a covalently bound layer of very long chain ω-hydroxyceramides and ω-hydroxy-fatty acids, called the lipid envelope. The exact function of the lipid envelope still remains unclear, but there is evidence that interactions of protein-bound lipids with free intercellular lipids contribute to the patterned organization of the lamellae seen in EM (). It has been shown that alterations in lipid composition of free or protein-bound lipids impair barrier function of the skin and lead to an increased TEWL (; ). We thus determined the levels of these lipids in the skin of wild-type and mutant mice. The levels of total fatty acids, cholesterol, and total free ceramides were not substantially different between control and 12R-LOX–deficient mice (not depicted). In the free ceramide fraction, we found mainly ceramide EOS, NS, and NP, as well as two ceramide AS species (). These results are in accordance with previously published results (, ). However, in the protein-bound fraction, five different species could be detected on HPTLC (B1–5). The exact identities of these species still have to be elucidated, but B4 is possibly ceramide OS. Significant differences were found in the subfractions of ester-bound lipids between wild-type and knockout mice. Whereas species B1 was found significantly increased, three other species (B2, B4, and B5) were almost completely absent in knockout epidermis (). We report the successful targeted disruption of in mice, which reflects features of ARCI and shows that 12R-LOX has a crucial role in the establishment of the epidermal barrier. The epidermis is a self-renewing stratified epithelium that serves as a protective barrier against mechanical, chemical, and biological insults; it is also a water-impermeable barrier that prevents excessive loss of body fluids. This function is critical for the survival of all terrestrial vertebrates and is established during late embryonic development. Identification of the molecular nature of the barrier is still under investigation. There is consent, however, that specialized structures in the stratum corneum, the CE, and extracellular lipid lamellae, as well as tight junctions in the granular layers, play essential roles in the development of the skin barrier function (; ). The stratum corneum is formed from granular cells during terminal differentiation as keratinocytes ascend from the proliferative cell type in the basal layer through the spinous and granular layers to end up as flat, dead corneocytes within the cornified layer. The CE is assembled underneath the plasma membrane by sequential incorporation and transglutaminase-mediated cross-linking of precursor proteins, followed by the covalent attachment of extracellular lipids. At the granular layer–stratum corneum interface, the lamellar bodies that are thought to be elements of the tubulovesicular TGN fuse with the cell membrane and extrude their contents to form a multilamellar lipid complex that fills most of the intercellular space. Pathological abnormalities in the stratum corneum, with the subsequent breakdown of epidermal barrier function, are observed in various skin diseases, which are referred to as ichthyoses. The loss of barrier function can be caused by several defects in the molecular mechanisms involved in the proper assembly of the CE or the intercorneocyte lipids. Several defective genes have been identified in ichthyosiform skin disorders so far, including genes coding for CE components (keratins, loricrin, and filaggrin) and proteins involved in the assembly and protein turnover (transglutaminase 1 and LEKTI), and, most frequently, genes coding for enzymes involved in lipid metabolism (e.g., fatty aldehyde dehydrogenase, steroid sulfatase, glucocerebrosidase, ATP-binding cassette transporter, (for review see ). Recent studies from our group and others have linked inactivating mutations in the genes of 12R-LOX and eLOX-3 to the development of ARCI (; ). 12R-LOX is a member of the LOX multigene family, exhibiting, among mammals, a unique R-stereospecificity of oxygen insertion. The enzyme is found almost exclusively in skin. In mouse epidermis, a predominant mRNA expression was observed in the differentiated keratinocytes (; ). By immunofluorescence analyses, we could now localize the 12R-LOX protein at the surface of the keratinocytes in the stratum granulosum, indicating a function in late epidermal differentiation. Interestingly, an almost identical expression pattern was observed for eLOX-3 (not depicted), suggesting a colocalization of both LOX in the plasma membranes of the stratum granulosum. The implication of 12R-LOX and eLOX-3 in ARCI has brought forth the concept that both enzymes function in the same metabolic pathway to convert arachidonic acid via 12R-HPETE to hepoxilin- and trioxilin-like metabolites that are critically involved in keratinocyte differentiation (; ; ; ). As shown for enzymes of the leukotriene synthesis, which form multimeric complexes in the nuclear membrane (), a coordinated membrane organization of these two enzymes, probably together with other enzymes and/or accessory proteins, may be a prerequisite for full enzyme activity and the proper generation of the bioactive lipid products of the 12R-LOX–eLOX-3 pathway in skin. Indeed, under cell-free conditions, the recombinant human 12R-LOX exhibits only low catalytic activity converting arachidonic acid to 12R-HPETE, whereas the mouse enzyme does not metabolize free arachidonic acid, but only esterified substrates, including arachidonic acid and linoleic acid methyl esters (). This has raised the question as to the nature of the endogenous substrate and the functional homology of the mouse and human enzyme (; , ). The results of this paper demonstrate an essential role of 12R-LOX in the development of epidermal barrier function in mice, documenting a functional homology of the mouse and human enzyme in skin. 12R-LOX–deficient mice exhibited the most severe phenotype regarding water barrier dysfunction reported so far. All knockout mice died within 3–5 h after birth as a result of severe dehydration. The mice lost ∼10% of their weight per hour. Other knockout mouse models with epidermal barrier defects, including mice deficient in KLF4 (), claudin (), E-cadherin (), LEKTI (), CAP1 (), and FATP4 (), exhibited substantially less weight loss, resulting in a longer life span of the transgenic mice. Analyses of dye permeability and of TEWL clearly demonstrated a severely defective inward and outward epidermal barrier function in 12R-LOX–deficient mice, while the barrier function of tight junctions was unaffected. The knockout mice failed to develop a functional epidermal barrier, which was acquired in wild-type mice around E17. At this time point, expression of 12R-LOX, which starts in embryonic skin at E15.5, was shown to reach high levels that persists at later embryonic stages and in newborn skin (). Defective skin barrier function typically results in compensatory mechanisms involving epidermal hyperproliferation, hyperkeratosis, and/or parakeratosis, which are observed frequently in ichthyosiform human skin and various mouse models (). 12R-LOX–deficient mice did not display such an obvious cutaneous phenotype, which may not develop because of the early neonatal lethality. In fact, markers of keratinocyte proliferation and terminal differentiation appeared to be unaffected, with the exception of filaggrin. This late terminal differentiation marker is the result of a complex proteolytic processing of profilaggrin by several enzymes, including protein phosphatases, proteases, and protease inhibitors (). Only mature filaggrin aggregates keratin filaments to form macrofibrils that crisscross the cornified cells of the stratum corneum, and it is an integral part of CE that contributes to its structural integrity. mice may be caused by the reduced profilaggrin processing. Furthermore, filaggrin monomers are degraded and provide free amino acids that, together with derivatives of amino acids and specific salts, constitute the natural moisturizing factor that is involved in the hydration of the stratum corneum (). Thus, reduction of filaggrin monomers might explain the more densely packed stratum corneum, which could contribute to the impairement of the barrier function in the mutant mice. In fact, disturbance of the proteolytic profilaggrin processing by gene inactivation has been shown to be associated with impairment of barrier function in several mouse models (; , ; ; ). In humans, lack of proteolytically processed filaggrin monomers caused by loss-of-function mutations have been shown to underlie ichthyosis vulgaris and discussed to be a major predisposing factor for atopic dermatitis (; ). An important component of the epidermal barrier is the arrangement of intra- and extracellular lipid accumulation in the stratum granulosum and stratum corneum, in particular the processing of intracellular lipids and the process of their extrusion into the intercellular space. Ultrastructural analysis revealed structural anomalies in the upper granular layers of the skin of 12R-LOX knockout mice that may reflect defects in the lipid metabolism associated with the observed phenotype. The features of the abnormalities are reminiscent of characteristic alterations found in a subgroup of ichthyosis congenital patients (). They include electron-lucent vesicles of variable size with lamellar structures reminiscent of the content of lamellar bodies adhering to the surrounding membrane. The appearance of these structures suggests that they may originate from defects in the assembly and/or extrusion of lamellar bodies, probably caused by aberrant lipid processing. Electron microscopic examination with ruthenium tetroxide postfixation to preserve lipid structure did not reveal major disturbances of the intercellular lipid lamellae. However, we presently cannot exclude more subtle local disturbances of these structures. mice. mice. Ceramides covalently attached to involucrin and other CE peptides are major constituents of the cornified lipid envelope that surrounds the corneocyte and have been discussed to be critical components of the barrier function (; ). The identity of the lipid species altered in the 12R-LOX–deficient epidermis remains to be elucidated. It also remains to be established weather 12R-LOX is directly involved in the enzymatic lipid processing or in the generation of lipid metabolites that are involved in the regulation of lipid metabolism. It is of interest to note that the epoxyalcohol metabolites produced by the 12R-LOX–eLOX-3 pathway are able to transactivate peroxisome proliferator-activated receptors (PPARs; ). Recent studies provide evidence for a role of PPARs in the regulation of terminal keratinocyte differentiation, including lipid synthesis and processing (; ). Moreover, it was recently reported that PPAR activators are able to accelerate permeability barrier recovery after acute barrier disruption (). In summary, this study indicates that the 12R-LOX–eLOX-3 pathway plays a key role in the process of epidermal barrier acquisition by affecting lipid metabolism, as well as protein processing. Mouse genomic DNA for the targeting vector was cloned from a 129/ola mouse PAC library (clone RPCI-21 K13407Q2; obtained from P.J. de Jong and K. Osoegawa, Roswell Park Cancer Institute, Buffalo, NY). By using a 7.4-kb BamHI fragment and a 4.4-kb PCR-generated fragment containing exon 7–12, a homology region was cloned into pBluescript KS+ (Stratagene) that spanned from 226 nt upstream of exon 3 to 1,081 nt downstream of exon 11 (). A selection cassette flanked by two loxP sites was inserted 230 bp upstream of exon 8, replacing a 222-bp AccI fragment; a third loxP site was inserted 84 bp downstream of exon 8, deleting a BamHI site (). The linearized targeting vector was electroporated into ES cells. Screening was done by PCR at both the 5′- and 3′-end using primer pairs consisting of one vector-specific primer and one endogenous -specific primer located outside of the targeting region. The primers used were as follows: 5′-end screening, 5′-CCGTCGACCTCGACCAGCCTGTCTACA-3′ and 5′-CTGAGGCCAGAAGATCACAAGTTCAAG-3′; 3′-end screening, 5′-GAGGCATCCGGGGATCATAACTTCGTATA-3′ and 5′-CAGGTATAGTTCGCAAGCAGGTGG-3′. gene targeting was further verified by Southern blot hybridization of BamHI-digested genomic DNA using 5′- and 3′-end probes flanking the recombination arms (). Two homologous recombinants were identified out of 300 analyzed. ES cells were injected into C57BL/6 blastocysts to generate chimeric mice, which were mated with C57BL/6 females. F1 heterozygotes were then crossed to 129S6. Mice were bred at the central animal facility of the German Cancer Research Center. All animals were kept under an artificial day/night rhythm and were fed standard food pellets (Altromin), with sterile water available ad libitum. The animal experiments were performed in accordance with the guidelines of the Arbeitsmeinschaft der Tierschutzbeauftragten in Baden-Württemberg (Officials for Animal Welfare) and were approved by the Regierrungspräsidum Karlsruhe, Germany. To convert the targeted allele into a mutant allele structurally lacking the essential exon 8 of the gene, F3 heterozygotes were crossed with CMV-Cre transgenic mice exhibiting ubiquitous Cre expression (). Complete excision of the resistance cassette and exon 8 in offspring mice was confirmed by Southern blot and PCR analyses (). Heterozygous mutant mice ) were bred with 129S6, and their heterozygous offspring were intercrossed to obtain homozygous mutant mice ( ). Genotyping was performed by PCR using 100 ng DNA isolated from tail and organs as a template and the primers ol1231 (5′-ACCCTCCCCTGCTGCTGTTGC-3′) and ol709 (5′-AGAGACCTCCCTTGTTGAGAAG-3′) to distinguish the 459-bp mutant allele band from the 1,075-bp wild-type band (). Total RNA was isolated from epidermis as previously described () and mRNA was reverse transcribed with MuLV reverse transcriptase using the SuperScript II first strand synthesis system (Invitrogen). The resulting cDNA was subjected to PCR with the primer pairs ol230 (5′-CTGTGCCCCGATGTGCTTGCTG-3′) and ol709 (3′-AGAGACCTCCCTTGTTGAGAAG-5′) for 12R-LOX. As a control, β-actin cDNA was amplified as a housekeeping gene. Rabbit polyclonal antipeptide antibodies against 12R-LOX and eLOX-3 have been previously described (). Mouse anti–12R-LOX mAb were raised using GST-12R-LOX fusion protein as immunogen. Other primary antibodies used were goat anti-actin (Santa Cruz Biotechnology), rabbit anti–Claudin-1, rabbit anti-occludin (both from Invitrogen), mouse anti-filaggrin (Monosan), rabbit anti-keratin 5, rabbit anti-keratin 10, rabbit anti-involucrin, and rabbit anti-Ki67 (all from CRP). Secondary antibodies used were Alexa Fluor 488 goat anti–mouse IgG (Invitrogen), CY3 anti–rabbit IgG (BD Biosciences), and anti–goat antibodies (Santa Cruz Biotechnology). Trunk epidermis of newborn mice was separated mechanically from the dermis after incubation for 30 s at 56°C in PBS. Epidermal proteins were extracted as described elsewhere (; ), and Western blot analysis was performed as previously described (). For light microscopic observation, samples were fixed for 24 h with 4% formalin in PBS, dehydrated in 70% ethanol, and embedded in paraffin. 5-μm sections were mounted on slides, dewaxed, rehydrated, and stained with hematoxylin and eosin. For methylene blue staining, skin was fixed with 1% glutaraldehyde for 24 h and embedded in epon. 1-μm sections were stained with methylene blue. For immunofluorescence microscopy, cryosections (3-μm thick) were fixed in acetone for 10 min at −20°C and permeabilized with 0.05% Triton X-100 in PBS and flushed with PBS. The slides were blocked in 1% BSA in PBS for 1 h and incubated with the primary antibody for 1 h. After washing three times (10 min each) in PBS, samples were incubated with a fluorescent dye coupled with antibody and Hoechst 33258 diluted in blocking buffer for 30 min and washed three times. Sections were embedded in mounting medium (DakoCytomation) and examined by light microscopy using a photomicroscope (Axioplan 2) with a 25× Plan-Neofluar objective (both Carl Zeiss MicroImaging, Inc.). Images were acquired with a high sensibility digital black/white AxioCam (Carl Zeiss MicroImaging, Inc.). CEs were purified and sonicated at 4°C for various time points in a bath sonicator, as previously described (). To determine the rate of fluid loss, newborns were separated from their mother and kept at 37°C. The body weight was monitored every 30 min, until time of death of homozygous mutant mice. The rate of TEWL from the skin of newborn mice was determined by using a Tewameter (Courage + Khazaka). For penetration assays, backs of newborn mice were immersed in 1 mM Lucifer yellow in PBS at 37°C. After 1 h incubation, mice were killed and the skin was dissected out. Frozen sections were counterstained with propidium iodide and penetration of the dye was assessed by immunofluorescence microscopy. The developmental stage of mouse embryos was determined based on the assumption that fertilization occurred in the middle of the dark cycle the day before plugs were identified. The embryos were subjected to methanol dehydration and subsequent rehydration, as previously described (), washed in PBS for 1 min, and stained for 30 min in 0.1% toluidine blue O/PBS. After destaining in PBS for 15 min, the embryos were photographed. All specimens were fixed for at least 2 h at room temperature in 3% glutaraldehyde solution in 0.1 M cacodylate buffer, pH 7.4, cut into pieces of ∼1 mm, washed in buffer, postfixed for 1 h at 4°C in 1% osmium tetroxide or in 0.5% ruthenium tetroxide, rinsed in water, dehydrated through graded ethanol solutions, transferred into propylene oxide, and embedded in epoxy resin (Glycidether 100; Merck). Ultrathin sections were treated with uranyl acetate and lead citrate and examined with an electron microscope (EM 400; Philips). r a m i d e A S w a s p u r c h a s e d f r o m S i g m a - A l d r i c h . C e r a m i d e N S w a s p r o v i d e d b y S e d e r m a , a n d c e r a m i d e s E O S , E O P , N P , a n d A P w e r e p r o v i d e d b y D e g u s s a . Lipid extraction followed a previously described protocol () with slight modifications. In brief, epidermis homogenate was extracted twice, first overnight with chloroform (methanol 1:2 at room temperature) and second with 2 ml chloroform (methanol 2:1 for 1 h at room temperature). The organic layers of both extraction steps were combined, the solvent was removed using a Christ Speed-Vac Alpha RVC/Alpha 2–4 (Christ), and the residue was redissolved in 100 μl methanol/chloroform at a 1:1 ratio. For recovery of protein-bound lipids, the pretreated pellet was incubated with 1 ml of 1 N NaOH in methanol at a ratio of 19:1, followed by extraction with 2 ml of chloroform for 1 h at 37°C. The organic layer was removed and washed with 3 ml of PBS–buffer; after phase separation, the solvent of the organic layer was evaporated in the Speed-Vac and the residue was redissolved in 100 μl methanol/chloroform at a ratio of 1:1. Lipid extracts were stored at −20°C until use. sub #text

My great uncle had a very old pharmacy with a little lab in the back. When I was ten or eleven, I had my own lab in the basement doing chemistry experiments. I was born in an area where we had trouble with pollution—we had a lot of problems with SO in the air—so I tried to make SO in the lab. I gassed some plants and looked how they tried to survive. I liked the combination that he developed between genetics—finding new components—and understanding a mechanism. In 1999, it was not like now where there is a big ubiquitin paper every week. I had the feeling that it was not completely solved yet—there were a lot of things to do. It turned out that ubiquitin was a very fruitful area, not only for learning a lot of different techniques and a lot of different approaches but also, in the end, how many different processes it controls. I really saw this when I looked for a postdoc position. I was more interested in questions of proliferation and differentiation, but their regulation brought me back very quickly to the ubiquitin. You need this if you want to extract only one subunit out of a complex. Cdc48 can get the subunit out and then attract proteins that deubiquitinate it or can channel it to the proteasome. It's a really crucial, central activity. It shows the power of ubiquitin as a cellular modification. I didn't like the cell cycle too much because it seemed to be so complex with so many different groups competing in a relatively small area. I was more interested in the kinds of decisions that cells have to make during differentiation. These are very fundamental decisions and there are a lot of different layers of regulation. Various people had demonstrated anaphase-promoting complex (APC) activity in postmitotic brain cells, but nobody knew the pathway that the APC was regulating. My idea was to try to get an entry point into neuronal differentiation by finding the pathway regulated by the APC. I started with a degradation screen to find the important players. But my work turned out to go in a different direction. When we tested human extracts with known substrates, the APC degraded some substrates like securin but we could never get other substrates like cyclin A to be degraded. We decided that before going on and screening this whole library, we actually should understand why the APC fails to degrade a couple of well-known substrates. That very quickly brought us back to the regulation of the APC, which we realized wasn't understood as well as people thought at the time. I kind of ended up back in the ubiquitin field as more a matter of chance than what I really intended to do when I looked for a postdoc. At the end of my postdoc, I turned back to this kind of question. We now have a system where the APC can degrade substrates very, very efficiently. I am now in the process of rerunning the screen and getting new substrates. With the new conditions, in vitro expression cloning is a very powerful technique. We have more active extracts and much better negative controls, which allows a much more straightforward identification of substrates. We are using in vivo sensors to measure the localization of degradation. We can test different mutations of the target sites, and couple this with an RNAi screen to identify regulators, seeing if we get more or less degradation or different localization of degradation. We want to do RNAi to identify ubiquitin ligases that control the entry to differentiation. I'm also studying how deubiquitination enzymes regulate the APC. I think the deubiquitination enzymes are probably just as interesting as ubiquitin ligases. There is huge potential. Although tremendous progress has been made, a lot of fundamental questions in the field still haven't been answered. We have only begun to understand how important the chain length is, and how chains of different topologies are made and recognized. On both sides—the mechanistic questions and the substrates—there is a lot of potential. Ubiquitination is not a simple or trivial reaction—it involves at least three different enzymatic activities. There is a lot of enzyme-specific regulation, so it has been hard to get paradigms for the whole field. You always have two sides to it: the mechanism aspect on one side and the process that is controlled on the other side. Even though a lot of people are interested in the process that is controlled on the other side, substrate identification has been extremely difficult. It took 15 years from the identification of Rad6 and the fact that ubiquitination was important for DNA repair to find probably the most important substrate in this process which turned out to be PCNA. It's not trivial. One of my very strong interests is the mechanism of the ubiquitination, and how this mechanism is used to achieve a certain type of regulation. An example of this is how the APC orchestrates progression through mitosis using a mechanism akin to kinetic proofreading. The fun about ubiquitin—this is something I really like about it—is that it can communicate a lot of information. You have monoubiquitination versus multiubiquitination, there are different chains with different topologies, and all of them have different binding proteins. The whole system is reversible so you can add and remove. Really it's something that can contain a lot of information. We have probably around 1,000 ubiquitin ligases in the genome, we have 100 different deubiquitinating enzymes in the genome, and there are so many proteins that recognize ubiquitinated sites—still there are novel ubiquitin-binding domains that are discovered on a regular basis. It's not a surprise it can control so many processes.

There has been a lot of debate not only about how DCSG sorting occurs but also about exactly where in the cell this triage takes place. All cells have the capacity to rapidly secrete proteins after their transit through the constitutive secretory pathway. A great deal of evidence supports the view that in the appropriate cell type, DCSG sorting signals can redirect proteins from the constitutive secretory pathway to DCSGs, thus confirming that it is not a default secretory pathway. Some groups have proposed that DCSG sorting occurs through the action of a sorting “receptor” present in the TGN that latches onto granule-destined proteins at sites where nascent granules will bud (; ). This has been referred to as the “sorting by entry” model. On the other hand, convincing evidence has been presented that in cells that generate DCSGs, all of the contents of the TGN are initially encapsulated into the nascent granules (). This “sorting by retention” model proposes that those proteins destined to be secreted constitutively are progressively extruded in low-density vesicles as the granule matures, ultimately leaving only the correct cargo protein in the mature DCSG. The first truth is that, regardless of the site at which sorting occurs, a mechanism has to exist that establishes and then maintains the segregation of DCSG cargo proteins from those that are constitutively secreted. Thus, it is a reasonable postulate that some mechanism exists to anchor the appropriate cargo proteins to the DCSG as it forms or matures. A second truth is that the sorting of proteins to DCSGs is a prerequisite for certain posttranslational processing steps in hormone and protease activation. For example, the conversion of proinsulin to active insulin, the conversion of proopiomelanocortin (POMC) to its many peptides, including ACTH, and the proteolytic activation of prorenin to renin all occur only after the precursors are encapsulated in the nascent secretory granules (; ; ). This makes sense for the organism because it ensures that the secretion of the active hormones or proteases is under appropriate physiological control. However, for granule-restricted activation to occur, it is necessary that both the protein precursors and the appropriate processing enzymes end up in the same DCSG. In the case of proinsulin, this means that the proprotein convertases PC1/3 and PC2, as well as carboxypeptidase E (CPE), all of which are required for generation of active insulin, have to be cotargeted with proinsulin in the budding granules. Thus, a second postulate is that a mechanism exists to ensure efficient cotargeting of protein precursors and their processing enzymes in the same organelle. DCSGs also share, by definition, the distinguishing trait of a core that appears dark or dense in electron micrographs. However, in spite of this common appearance, there may be important functional and mechanistic differences in DCSGs. For example, the gonadotropes of the pituitary store luteinizing hormone and follicle-stimulating hormone in separate DCSGs, and their release is independently controlled (for review see ). Likewise, there are two types of DCSG in chromaffin cells containing either epinephrine or norepinephrine, and these are morphologically distinct (). The signals for targeting proteins to DCSGs also show tissue-specific variations; removal of 90 amino acids from the C terminus of the granin chromogranin A (CgA) prevents its sorting to DCSGs in pituitary GH4 cells, but has no effect on DCSG sorting in sympathoadrenal PC12 cells (). Likewise, POMC is efficiently stored in DCSGs when transfected into cultured pituitary cells, but not in sympathetic neurons (). Thus, a third truth is that not all DCSGs are alike, and it is a reasonable postulate that DCSGs can be assembled, even within the same cell, through more than one mechanism. Could some of these truths explain the difficulties in reaching consensus on the protein signals necessary for DCSG targeting? There has been no shortage in the variety of DCSG sorting mechanisms proposed in the last 20 yr; these include protein domains that interact with or that traverse membranes and that may or may not interact with additional proteins on the cytoplasmic side of the DCSG, proteins proposed to be a “master switch” for granule formation, universal granule cargo receptors, protein domains that mediate aggregation in the late TGN, certain paired basic protease cleavage sites or α helices in secretory proteins, disulfide-constrained loops, acidifying proton pumps, and other mechanisms. As a result, investigators have become progressively entrenched in defending their favorite mechanisms and commonly use the descriptors “controversial” and “difficult to repeat” to describe the work of others in their publications. Nevertheless, it is possible to accommodate most of these findings in a model that subdivides targeting function into three components (): membrane associated (or traversing) tethers, tether-associated cargo, and aggregation. Peptidyl-α-amidating monooxygenase (PAM), phogrin, and muclin are all type 1 membrane-spanning proteins that are targeted to DCSGs (; ; ). In the case of phogrin, the granule sorting domain is located in the cytoplasmic tail of the protein, and although the exact nature of the signal is still debated, it appears that this domain can bind the clathrin adaptor proteins AP-1 and -2 in vitro (; ). Such interactions might provide a means of communication between the granule cargo proteins and the membrane domains or cytoplasmic proteins that will define the budding DCSGs. The membrane-binding domains of the granule-resident protein CPE (CPE; ) and the prohormone convertases PC1/3 (; ) and PC2 () are also key for their targeting to DSCGs, and there is agreement that the granule sorting is mediated by short α helical domains. An α helical domain has also been shown to be important for targeting prosomatostatin (), CgA (), and VGF () to DCSGs. Recent results suggest that α helices with the ability to direct granule sorting in secretory proteins share the characteristic of charge segregation from a hydrophobic patch, which is consistent with a shallow membrane interaction (). This first group of DCSG proteins could therefore be said to be tethered to the membranes of the TGN or the maturing granule. A second group of granule sorting domains may act by binding cargo proteins to these granule-tethered proteins. For example, CPE has been proposed to interact with several granule cargo proteins including proenkephalin, proinsulin, POMC (), brain-derived neurotropic factor (proBDNF; ), and secretogranin III (SgIII; ) to promote their retention in secretory granules, even though some of these are not enzymatic substrates of CPE. No common mechanism for interaction of these cargo proteins with CPE has yet emerged, although undefined residues within an N-terminal disulfide-constrained hydrophobic loop in POMC () and acidic residues in proBDNF () have been reported, and both seem to be important for sorting the respective proteins to DCSGs (; ). Paired basic amino acids have also been reported to direct DCSG sorting in some proteins, including proneurotensin (), prorenin (), prothyrotropin-releasing hormone (), and progastrin (), and to increase the sorting efficiency of proinsulin (). In the analyses performed to date, it appears that these paired basic amino acids must constitute a cleavage site for one of the granule-resident prohormone convertases (PC1/3 or PC2) to function as a granule sorting domain because changing the cleavage site to one recognized by furin (another member of the family that cleaves its substrates in the early secretory pathway) causes the proteins to be secreted through the default constitutive pathway (). These results raise the possibility that certain DCSG-targeted proteases can act as sorting chaperones for their substrates, in addition to being processing proteases. Muclin has also been suggested to act as a granule cargo receptor in pancreatic cells through its binding of sulfate groups on O-linked glycosylated proteins (). Atrial natriuretic factor (ANF) has also been shown to be tightly bound to the membranes of atrial myocyte secretory granules through its interaction with PAM (), although it is not a substrate of PAM. Thus, a variety of interactions with “tethers” may serve to target proteins to secretory granules. Notably, if this mechanism is correct, it would, in some cases, provide a means to ensure that processing enzymes and their substrates end up in the same DCSGs. A third category of granule-targeting mechanisms involves formation of high molecular weight protein complexes or aggregates. Indeed, many granule-targeted cargo proteins have the ability to multimerize or aggregate, leading, in most cases, to the formation of electron-dense cores. A direct correlation between the ability to aggregate in vitro and to be sorted to secretory granules in transfected cells has been reported for rat pro-ANF () and CgA (). Because granins are acidic proteins that cluster in the slightly acidic environment present in DCSGs (for review see ) it has been suggested that aggregation may serve to prevent their extrusion from the maturing granule. Indeed, showed that treatment of PC12 cells with bafilomycin A1, which is a specific inhibitor of vacuolar H-ATPase, resulted in a decrease in regulated secretion of CgA with a concomitant decrease in visible DCSGs, suggesting that regulated secretion of CgA and dense core formation are linked in DCSGs. also showed that silencing CgA expression in PC12 cells results in a loss of visible DCSGs, leading the authors to the striking conclusion that CgA is not only a component of the dense core, but that it is also a “master regulator” of DCSG biogenesis. Further work by suggests that the role of CgA in this process may not be that simple; they found no correlation between DSCG content and CgA expression in isolated clonal lines of PC12 cells, suggesting that other proteins could be contributing to DCSG appearance. In fact, expression of several other DCSG cargo proteins, including provasopressin, prooxytocin, POMC, secretogranin II, and chromogranin B, is sufficient to induce aggregate-containing cytoplasmic vesicles, even in cells with no regulated secretory pathway (), although these probably do not display all of the functional characteristics of DCSGs (). Regulating the formation of the aggregate may also be physiologically important. recently reported that a polypyrimidine-binding protein (PTB), which is up-regulated under conditions of high insulin demand, stabilizes messenger RNAs of many of these same DCSG cargo proteins in insulin-producing cells and leads to increased granule formation. CgA has also been reported to induce the expression of PN-1, which is a serine protease inhibitor that slows the turnover of several DCSG cargo proteins () that could provide an additional mechanism for increasing DCSG aggregate formation. Because CgA binds to another granin partner, SgIII (), which, in turn, can associate with cholesterol () and CPE (), aggregation may synergize with protein–protein and protein–membrane interactions to improve the retention of cargo proteins in the maturing granule and their regulated secretion. In spite of the compelling arguments presented for these various DCSG sorting mechanisms, their translation to the whole animal has been anything but simple. One example of this difficulty is the proposed role of CgA as a master regulator of granule formation. Although down-regulation of CgA expression was reported to result in the loss of detectable DCSGs in cultured PC12 cells (), CgA gene inactivation in mice leads to either a “reduction” () or no discernable effect () on DCSG formation in the CgA-rich adrenal chromaffin cells in two independent studies. In spite of the differences in the effects on DCSG morphology, both groups report a similar and dramatic effect on catecholamine secretion in the CgA-deficient mice, proving that CgA deficiency is not entirely without consequence. How can these apparent differences in the requirement for CgA be explained? One obvious possibility is that in vivo, other DCSG cargo proteins can complement the function of CgA in the formation of the dense core, but cannot compensate for its absence in catecholamine storage and secretion. In support of this possibility, the group that saw no effect of CgA inactivation on DCSG formation reported an up-regulation of CgB and SgII in the adrenal glands of the engineered mice (). Thus, although CgA may affect DCSG formation in some cultured cells, this particular function can obviously be replaced in vivo. Nevertheless, although experiments to date have not identified a master regulator of DCSG formation, the concept may not be entirely wrong in specific cell types; ANF inactivation in mice leads to a complete loss of visible DCSGs in the cardiac atrium (), and inactivation of the renin gene leads to a complete disappearance of DCSGs in the juxtaglomerular cells of the kidney (). It's important to note, however, that regulated secretion can occur in the absence of a dense core as it does in many neurosecretory vesicles. In the case of the ANF and renin-deficient mice, it will be intriguing to determine if the remaining cargo proteins are still packaged in such vesicles in the absence of the aggregating partner. A similar conundrum exists with CPE as a sorting receptor for a variety of DCSG cargo proteins. originally proposed CPE as the regulated secretory pathway sorting receptor because they observed endocrine disorders in the Cpe fat/fat mouse that harbors a mutation in the CPE gene. Proinsulin and POMC are among the several proteins that were shown to bind to CPE and that were proposed to enter DCSGs by this association (). However, recent results demonstrate that both proinsulin and POMC are correctly targeted to DCSGs in CPE fat/fat mice (; ). What are we missing in this picture? Although there may be many reasons why it has been hard to derive a consensus for the mechanisms and components of the DCSG sorting machinery, the most intuitive is that we are the victims of our own scientific reductionism, i.e., that in our search for a simple canonical sorting mechanism we have developed a grossly oversimplified view of the way in which proteins enter DCSGs. Nearly 100% of the proinsulin produced in pancreatic β enters DCSGs (), whereas only about 25% of the prorenin in the secretory pathway of kidney juxtaglomerular cells is sorted to DCSGs (). What can explain these differences? Proinsulin contains numerous potential DCSG sorting domains, such as a binding domain for CPE (), two paired basic amino acid protease cleavage sites (), and the ability to hexamerize and subsequently aggregate (), whereas prorenin only contains a single DCSG sorting domain: a paired basic amino acid protease cleavage site (). In the case of prorenin, changing even a single one of these basic amino acids completely eliminates DCSG targeting in tissue culture cells (). In contrast, neither the mutation of the protease cleavage sites () nor the hexamerization domain () of proinsulin appears to affect its DCSG sorting. Combined with the finding that proinsulin is still efficiently sorted to DCSG in CPE-deficient mice (), it has been tempting to dismiss the function of these putative sorting signals. However, another possible explanation is that, with its many DCSG sorting signals, proinsulin might be able to compensate for the loss of any single sorting domain. There is, in fact, some evidence to support the view that DCSG sorting signals can synergize; duplicating the disulfide-constrained loop DCSG sorting signal normally found at the N terminus of CgB results in a greater sorting efficiency to DCSG than the native protein (). Furthermore, combination of α helical and paired basic amino acid sorting domains on either the same protein or on two proteins capable of dimerizing led to a dramatic increase in DCSG sorting over proteins containing either individual domain (). Thus, diverse sorting signals may be able to functionally complement each other even through protein–protein interactions. Complementarity in cellular sorting machineries may also occur. also reported that pituitaries of the Cpe fat/fat mouse contain elevated levels of both SgIII and CgA that might compensate for the loss of CPE in targeting POMC to DCSGs. All of these cases are consistent with the existence of multiple sorting mechanisms, each of which can contribute to the overall efficiency of protein sorting or retention in DCSGs. Cell types and the nature and/or the number of the sorting domains contained in the cargo protein would ultimately determine the extent to which each mechanism is active. Multimerization and aggregation could add synergy between mechanisms used by other DCSG cargo proteins in the aggregate. With such a model, it's also easy to imagine how changing conditions within the cell could alter DCSG sorting efficiency of a protein, which is a potential control point that has important implications for hormone secretion but has received little attention to date. #text

xref #text The study of the role of the monoubiquitination of Pex5p observed in wild-type cells () has been hampered by the polyubiquitination of Pex5p, which accumulates at the peroxisomal membrane in cells lacking components required for the late steps in the import pathway (; ). To study Pex5p monoubiquitination, we followed two strategies to avoid polyubiquitination. For both strategies, we assumed the polyubiquitination site to reside within the N-terminal region of the protein because this region of Pex5p from human and rat is sufficient to carry out its docking to, as well as its consecutive dislocation from, the peroxisomal membrane to the cytosol (). First, we fused three epitopes to the N terminus of Pex5p, which for other proteins has been shown to prevent polyubiquitination. Second, we substituted arginine for the first conserved lysine residue (lysine 18) of Pex5p by site-directed mutagenesis. The corresponding lysine residue is required for polyubiquitination of Pex5p from () and of Pex20p from , which is the putative functional counterpart in the peroxisomal PTS2-dependent protein import pathway (). However, in our case, the single-mutant protein (Pex5pK18R) was still polyubiquitinated (unpublished data). Thus, we considered that an adjacent lysine might substitute for the loss, as is the case for many other proteins destined for degradation (), and accordingly, we also replaced lysine 24 of Pex5p with arginine. Both the -tagged Pex5p and Pex5pK18/24R restored the growth defect of a deletion strain on medium with oleate as the sole carbon source (not depicted) and imported GFP-PTS1 properly into peroxisomes (). Moreover, Pex5p and Pex5pK18/24R were normally bound and released from the peroxisomal membrane (). Thus, both variants behaved like the wild-type protein, thereby demonstrating that neither the tag nor the introduced mutations interfered with the physiological role of Pex5p in peroxisomal protein import. or strains, no polyubiquitinated forms of the -tagged or mutated Pex5p could be detected in the cell lysates (). Thus, the exchange of the lysine residues deleted the target residues for ubiquitination. Similarily, the N-terminal tagging prevented polyubiquitination, possibly by interfering with polyubiquitin-specific factors such as Ubc4p/Ubc5p or the corresponding E3 enzyme, or by masking the target lysine residues for polyubiquitin chain formation. The fact that cells harboring Pex5p or Pex5pK18/24R did not exhibit a growth defect on oleic acid medium and import GFP-PTS1 indicates that polyubiquitination is not a prerequisite for functional peroxisomal protein import in . These data are in agreement with the idea that polyubiquitination is part of a quality control system that primes membrane-accumulated Pex5p for proteasomal degradation (; ; ). A similar system (receptor accumulation and degradation in absence of recycling [RADAR]) has also been described for the quality control of membrane-associated Pex20p (). Despite the lack of polyubiquitination, Pex5p and Pex5pK18/24R were still normally monoubiquitinated (). Preparation of membrane pellets in the presence of -ethylmaleimide (NEM) to inhibit deubiquitinating enzymes results in the appearance of a more slowly migrating form of Pex5p, which has been shown to represent monoubiquitinated Pex5p (). The slower migrating form did shift to a higher molecular weight upon expression of Ub, demonstrating its ubiquitin nature. No change in molecular weight was observed when mutated ubiquitin (UbK48R) was expressed, which prevents the formation of polyubiquitin chains (), confirming the monoubiquitination of the proteins. Thus, in respect to monoubiquitination, the myc-tagged and the K18R/K24R-double-mutated Pex5p behave like the endogenous protein. It has been reported that monoubiquitination of Pex5p is independent of , , , and , and even takes place in a double-deletion strain (). However, the observations on monoubiquitination were hampered in mutants lacking UBC10 (Pex4p) and other late peroxins like Pex1p and Pex6p because of the presence of polyubiquitinated Pex5p. We took advantage of the fact that Pex5p is not polyubiquitinated, but is still susceptible to monoubiquitination, enabling us to investigate which of the known UBCs is responsible for the monoubiquitination event. We isolated Pex5p via immunoprecipitation from chosen deletion strains, leaving out the lethal deletion of , as well as of the SUMO-conjugating enzyme Ubc9p and the Nedd8/Rub1-conjugating enzyme Ubc12p. In all of the UBC deletion strains tested (, , , , , , and ), formation of mono-Ub-Pex5p was visible, except in the case of Pex4p (Ubc10p) deletion (). These data clearly demonstrate the dependence of Pex5p monoubiquitination on the presence of Pex4p. To determine whether not only the presence but also the catalytic activity of Pex4p is essential for Pex5p monoubiquitination, we expressed an inactive Pex4p mutant protein, which carries a C–S point mutation at position 115. This amino acid residue is essential for the activity of ubiquitin-conjugating enzymes (). Although Pex4p(C115S) can be expressed to nearly wild-type levels and is properly targeted to peroxisomes (), monoubiquitination of the -tagged or point-mutated PTS1 receptor was completely abolished ( and Fig. S1 A, available at ). In addition to its catalytic activity, the peroxisomal localization of Pex4p also proved to be essential for Pex5p monoubiquitination. Pex5p or Pex5pK18/24R was expressed in a background, which lacks the peroxisomal membrane anchor for Pex4p (). As shown in , monoubiquitination of Pex5p was not observed in the strain. The fact that the presence of an active ubiquitin-conjugating enzyme Pex4p attached to the peroxisomal membrane via Pex22p is indispensable for the formation of monoubiquitinated Pex5p indicates that the PTS1 receptor is a physiological substrate of Pex4p. Next, we addressed the question of whether Pex4p is required up- or downstream of the AAA complex that is responsible for Pex5p release from the peroxisomal membrane (; ). Previous findings demonstrated that the receptor docking at the peroxisomal membrane and transfer to the RING-finger peroxins is prerequisite for monoubiquitination (). Another attempt to elucidate the order of events was made by Collins and co-workers in (). They took advantage of a specific instability of Pex5p in mutant strains lacking components of the AAA and Pex4p–Pex22p complex. Based on the finding that the Pex5p level in a / strain was reduced to the level of the single-mutant strain, it was concluded that Pex4p acts downstream of the AAA peroxins (). In cells, such a Pex5p instability is not observed, but the protein becomes polyubiquitinated and accumulates at the peroxisomal membrane. Thus, the observed Pex5p instability in other yeasts is likely to be a consequence of polyubiquitination and subsequent proteasomal degradation. In this case, the Pex5p polyubiquitination seems to be part of a quality control system that is not directly related to the import process. Instead of this pathological situation, we now took advantage of the physiological monoubiquitination that is also present in wild-type cells to study the epistasis. The analysis revealed that both Pex5p and Pex5pK18/24R, which are monoubiquitinated under wild-type conditions, but not in -affected cells, are still monoubiquitinated in a strain. double-deletion strain ( and Fig. S1 B). This result demonstrates that Pex4p-dependent monoubiquitination occurs independently of the presence of the AAA peroxins. Thus, monoubiquitination of Pex5p takes place before the protein is released from the peroxisomal membrane in an AAA peroxin– and ATP-dependent manner. An explanation for the different conclusion drawn by Collins and co-workers is provided by the different nature of the Pex5p fraction analyzed. Collins and co-workers used the instability of Pex5p as an indicator, and thus, most probably analyzed the Pex5p form designated for proteasomal degradation (), a process for which Pex4p has been demonstrated to be dispensable (). Monoubiquitination requires Pex4p, and thus, is likely to represent an important step in the peroxisomal protein import process. Thus, the epistasis on the basis of monoubiquitination is expected to reflect the sequence of events in the Pex5p receptor cycle. To investigate whether the Pex4p dependence of the monoubiquitination of Pex5p is indeed defined by a direct ubiquitination reaction between both proteins, we established a cell-free ubiquitination assay with recombinant GST-Pex4p. The functionality of the protein was tested by an in vitro autoubiquitination reaction in the presence of recombinant E1 and ubiquitin (). As the higher molecular weight forms of GST-Pex4p were resistant to β-mercaptoethanol, they likely represent conjugates of ubiquitin and Pex4p that are linked via a peptide bond to a lysine residue of the E2 enzyme. Ubiquitin-conjugating enzymes bind ubiquitin via a thioester bond to their catalytically relevant cysteine residue within the UBC- fold before they pass it to the lysine of a target protein. The formation of the Pex4p-ubiquitin thioester was monitored by omitting reducing agents. Under these conditions, slower migration species indicated the presence of the thioester linkage of ubiquitin. (). Alternatively, the conjugation of His-Ub to GST-Pex4p resulted in the appearance of slower-migrating GST-Pex4p (unpublished data). + Pex5pK18/24R cells, which are known to harbor the required E3 activity. To assay the ubiquitination, the samples were incubated with E1 and Pex4p alone or with Pex4p that has been charged with Ub or His-Ub. Pex5pK18/24R was found to be unmodified in the samples with uncharged Pex4p, while it was ubiquitinated, when Pex4p was preloaded with Ub (). Modified Pex5p species with higher molecular weights were observed when Pex4p had been charged with His-Ub. This form was also specifically recognized by the Penta-His antibody. As no additional His-Ub was added to the reaction, the His-Ub acquired by Pex5p had to originate from the Pex4p-His-Ub conjugate, demonstrating a direct ubiquitination reaction. These results identified Pex5p as a molecular target for Pex4p-dependent monoubiquitination. We then asked whether ubiquitination of Pex5p is a prerequisite for the release of Pex5p from the peroxisomal membrane. Previously, we demonstrated that Pex5p can only be exported from membranes derived from a strain when incubated with either cytosol containing the AAA peroxins or the isolated AAA complex (). To test for the ubiquitin requirement, we aimed to delete all possible ubiquitination sites of Pex5p. For the prevention of polyubiquitination, this was achieved by deletion of K18/24R. However, deletion of any of the 15 lysines within the N-terminal half of Pex5p did not abolish the monoubiquitination of the receptor, indicating that the absence of one ubiquitination site could be overcome by using another. Therefore, we took advantage of the fact that Pex5p is not monoubiquitinated in cells. The lack of Pex5p monoubiquitination in and the prevention of polyubiquitination of Pex5pK18/24R enabled us to separately investigate the contribution of mono- and polyubiquitination to Pex5p release from the peroxisomal membrane. shows that Pex5pK18/24R is still exported from the peroxisomal membrane in an AAA peroxin and ATP-dependent manner. cytosol, a fraction of endogenously encoded Pex5p was released from the membrane. This liberation of Pex5p from cells lacking Pex4p still required the presence of ATP and the activity of the AAA peroxins. However, when Pex5pK18/24R was subjected to the export assay in a Pex4p-deficient system, release of the receptor from the membrane was completely blocked. Thus, the simultaneous loss of both polyubiquitination and monoubiquitination of the receptor prevented release of the receptor, demonstrating that Pex5p ubiquitination is required for its release from the membrane. No release of the Pex5pK18/24R mutant protein was also observed when the wild-type Pex4p was replaced by Pex4p (C115S), demonstrating that the catalytic activity of Pex4p is required for release of the receptor from the membrane (). When recombinant GST-Pex4p was added to the Pex4p-deficient export system, as outlined in , the Pex5pK18/24R was released from the membrane, thereby unequivocally demonstrating the functional role of Pex4p in Pex5p export. When Pex22p-deficient membranes were subjected to the assay, no Pex5pK18/24R was released from the membrane, demonstrating that anchoring of the recombinant GST-Pex4p to Pex22p is required for its function in Pex5p release. Our findings show that mono- or polyubiquitination are both sufficient to prepare Pex5p for the AAA-dependent release to the cytosol. The fate of the released mono- or polyubiquitinated Pex5p is proposed to be different, as outlined in the model depicted in . After its release, the polyubiquitinated Pex5p is directed to proteasomal degradation as part of a quality control system. In support of this assumption, prevention of polyubiquitination did not significantly interfere with Pex5p function, or with its AAA peroxin–dependent release from the peroxisomal membrane. The released monoubiquitinated Pex5p is supposed to be deubiquitinated and made available for further rounds of matrix protein import. Interestingly, in our experimental design, only part of the total membrane-bound Pex5p seems to be monoubiquitinated, and the released Pex5p no longer contained the ubiquitin moiety ( and ). This is supposed to reflect the situation that Pex5p is only transiently monoubiquitinated and that the ubiquitin is released during the export step. Future research will also reveal whether all subunits of the supposedly homooligomeric Pex5p may require ubiquitination for its release from the membrane or whether ubiquitination of only one subunit of the Pex5p might be sufficient for its release. Our data provide a plausible explanation for the PTS1 import defect of a deletion strain and for the previously observed accumulation of Pex5p in cells lacking Pex4p (; ). As cells also exhibit a PTS2 import defect, one could assume a similar role of Pex4p in the cycle of the PTS2 receptor or the auxiliary proteins Pex18p/Pex21p. An essential role for polyubiquitination in recycling of Pex5p and Pex18p/Pex21p orthologue Pex20p in and , respectively, has also been reported based on an enhanced degradation of these receptors upon overexpression of Ub(K48R) (; ). The authors suppose that the overexpression of Ub(K48R) might interfere with a constitutive degradation of a thus far unidentified target factor, which certainly is not Pex5p. We have demonstrated that ubiquitination of the Pex5p is a prerequisite for its dislocation from the peroxisomal membrane by the AAA peroxins. Ubiquitination of Pex5p is expected to facilitate the recruitment of the AAA machinery. The functional role of Pex4p/Ubc10p, which is the only ubiquitin-conjugating enzyme known to be involved in the biogenesis of an organelle, has been a mystery for nearly 13 yr. We demonstrate that Pex5p is a molecular target for monoubiquitination by Pex4p and show a direct role for the protein in the membrane release of Pex5p at the end of the import cascade. It has been demonstrated that the ATP-consuming step in this process is not the binding and import of Pex5p, but the AAA peroxin–dependent export of the receptor (; ). As ubiquitination is essential for the recycling of the PTS1 receptor, we have expanded the energy requirement of the peroxisomal import pathway by a second ATP-dependent step, i.e., receptor monoubiquitination. (this study), as well as , , , , , , and strains (). Deletion strains were generated by the “short flanking homology” method, as previously described (). Yeast media have been described in another work (). Pex5p was expressed from a low-copy vector under the control of its own promotor (pHP17-PEX5). Point mutations in were introduced using overlap extension PCR, leading to the PEX5 K–R mutant collection used in this study. Plasmids expressing Pex5p (pRS6 ; this study), Pex4p (pEMBL-PAS2; ), GFP-PTS1 (), ubiquitin (YEp96; ), and ubiquitin (K48R; YEp110; ), as well as Ubiquitin (YEp105; ), were used. GST-Pex4p was expressed in BL21(DE3). Cells were harvested, diluted in PBS buffer (137 mM NaCl, 2.7 mM KCl, 4.3 mM NaHPO4, and 1.4 mM KHPO, pH 7.3), containing protease inhibitors (8 μM antipain, 0.3 μM aprotinin, 1 μM bestatin, 10 μM chymostatin, 5 μM leupeptin, 1.5 μM pepstatin, 1 mM benzamidine, and 1 mM PMSF [Boehringer]) and broken using a French press. The 100,000 supernatant containing the soluble GST-Pex4p was loaded on a glutathione–Sepharose 4B (Pharmacia). After intense washing with PBS buffer, GST-Pex4p was either eluted from the column with 10 mM glutathione or cleaved from the fusion tag with thrombin. Recombinant Pex4p or GST-Pex4p was charged with recombinant ubiquitin by an in vitro autoubiquitination assay. The reaction mixture contained 0.1 μg E1, 3 μg Pex4p, 5 μg ubiquitin, 2 mM ATP, 2 mM CaCl, and 25 mM Tris/HCl, pH 7.6 (Ub buffer I). The reaction proceeded for 1.5 h at 30°C and was quenched by addition of SDS-gel sample buffer. The formation of the Pex4p–ubiquitin thioester forms was monitored by using SDS-gel sample buffer without β-mercaptoethanol. strain, as previously described (). The Pex5pK18/24R samples were dissolved in Ub buffer II (1 mM ZnSO, 1 mM DTT, 20 mM NEM, 2 mM ATP, 2 mM CaCl, and 25 mM Tris/HCl, pH 7.6). 0.1 μg E1 and 5 μg Ub- or His-Ub–charged or uncharged Pex4p were added to the reaction mixture. The reaction proceeded for 20 min at 37°C. After precipitation with trichloroacetic acid, the pellets were washed twice with 80% acetone and dissolved in SDS-gel sample buffer. Membrane sedimentation, in vitro import, and export assays were performed according to . If required, 0.1 μg E1, 5 μg ubiquitin, and 5 μg recombinant GST-Pex4p were added to the in vitro export reaction. Protein complexes were isolated by coimmunoprecipitation as described by . TCA lysates of cellular fractions were prepared as described by . Immunoreactive complexes were visualized using anti–rabbit or –mouse IgG-coupled horseradish peroxidase in combination with the ECL system (GE Healthcare). Polyclonal rabbit antibodies were raised against Pex5p (), Pex13p (), and Fructose-1,6-bisphosphatase (). Monoclonal mouse antibodies were raised against the C- epitope (), GST (Sigma-Aldrich), and Penta-Histidin (QIAGEN). The direct fluorescence of GFP was recorded at room temperature in distilled water with an Axiophot microscope (Carl Zeiss MicroImaging, Inc.) and a 100×/1.4 NA oil immersion objective. Both fluorescence and optical photographs were taken by using the connected hardware in combination with the Spot RT software version 3.1 (Diagnostics Instruments). Adjustments of contrast and brightness were performed with Photoshop software version 7.0 (Adobe), and characteristic cells were cut out and copied to Micromedia Freehand software version 10.0. Fig. S1 shows that monoubiquitination of Pex5p depends on the presence, activity and peroxisomal localization of the ubiquitin- conjugating enzyme Pex4p and occurs before the AAA peroxin Pex1p. The online version of this article is available at .

Myosins are actin-based motors that play essential roles in a variety of cellular processes, including cytokinesis, cellular trafficking, phagocytosis, maintenance of cell shape, and muscle contraction (). Myosin-based movement results from a precise cycle of the myosin head binding and releasing ATP and actin. During this process, the myosin head transitions through multiple folding conformations. Molecular chaperones appear necessary for de novo folding and structural maintenance of the myosin head. Expression of the myosin motor domain in bacteria results in misfolding (). In vertebrate systems, the chaperonin containing TCP-1 (CCT), as well as molecular chaperones Hsp90 and Hsc70, are necessary but not sufficient in the folding of striated muscle myosin (; ). Evidence from a variety of experimental systems indicates that myosins use specialized chaperones during their activity, folding, and assembly. Mutations in UNC-45/Cro1p/She4p(Dim1p) domain (UCS) proteins lead to phenotypes related to defects in myosin folding and assembly (). Decreased UCS domain protein function in fungal mutants produces myosins defective in actin:ATP transduction (; ). In , null alleles lead to embryonic arrest of body wall muscle development (), and temperature-sensitive mutations lead to a paralyzed or uncoordinated phenotype at the restrictive temperature with marked disorganization of myofibrils (; ). In vitro, UNC-45 exerts chaperone activity on the myosin head and acts as a cochaperone that specifically binds Hsp90 (). UNC-45 has recently been shown to be a substrate of an E3/E4-multiubiquitination complex containing CHN-1 (the homologue of CHIP) and UFD-2 (). –null worms are viable and morphologically normal. However, UNC-45 overexpression leads to an uncoordinated phenotype in these worms, suggesting that increased levels of UNC-45 may cause muscle defects. Previous studies have shown that unassembled components of myofibrils are degraded through the ubiquitin/proteasome system (UPS) in muscle wasting conditions, including cancer and starvation (; ; ; ). Our results suggest that enhanced levels of UNC-45 may promote nonnative myosin conformations, rendering them susceptible to degradation by the UPS. Loss-of-function (Lof) temperature-sensitive UNC-45 mutations result in severely paralyzed worms, with pronounced disorganization of the sarcomere (; ). To investigate the consequences of increased UNC-45 levels in a wild-type background, we used the transgenic line, [P∷ ], expressing UNC-45 under control of the strong muscle–specific promoter. In contrast to lines overexpressing UNC-45 from an extrachromosomal array, this line stably transmits the transgene from generation to generation. worms overexpress UNC-45 at ∼10-fold greater amounts than that of the previously studied integrated line [P∷ ] (; ). Although not as pronounced as in Lof alleles, the increased concentrations of UNC-45 in also lead to abnormal thick filament assembly and a concomitant defect in movement. We measured body bend rates and found that worms were 36% slower than wild type (). However, this is not as severe as the temperature-sensitive mutants grown at the restrictive temperature, which is the result of diminished, rather than augmented, UNC-45 function. Transgenic UNC-45 was able to rescue the motility defects of at the restrictive temperature, indicating that it is functional (). young adult (Ya) wild-type muscle cells are ∼120–150 μm in length with a mean of seven to nine myosin thick filament–containing A bands per cell (; ). Similarly staged worms demonstrated cells 108–130 μm in length that contained only five to six A bands (). Thus, UNC-45 overexpression in worms results in a decreased number of myosin-containing thick filaments. This phenotype arises specifically from increased levels of UNC-45. Transgenic overexpression of the unrelated protein GFP under control of the promoter does not result in altered muscle cells (Fig. S1 A, available at ). Furthermore, integration of the UNC-45 transgene by itself is not responsible for the muscle phenotype, as the expression of UNC-45 as an extrachromosomal array also decreased the number of myosin-containing A bands in cells in which it was expressed (Fig. S1 B). Moreover, in rare instances in which individual muscle cells in worms expressed undetectable levels of UNC-45, a wild-type phenotype was observed (Fig. S1 C). We next investigated whether the overall myosin content was decreased in these worms. Four distinct myosin heavy chain (MHC) isoforms are expressed in as major components of the sarcomere (Miller et al., 1983; ). MHC A and B are present in the body wall muscles, whereas MHC C and D are located in the minor pharyngeal muscles (). To assess myosin levels, individual worms were analyzed for MHC A, B, and D. In contrast to MHC D, which was not affected because UNC-45 is specifically expressed in body wall muscle, both body wall MHC A and B were decreased in to ∼70% of wild type (). Worms homozygous for the allele grown at the restrictive temperature resulted in decreased accumulation of all myosin isoforms examined (). This is expected, as endogenous UNC-45 is involved in pharyngeal as well as body wall muscle myosin assembly (). Previous results showing that MHC A content was not substantially affected in mutant worms may be explained by the fact that the protein paramyosin was used for normalization. Paramyosin is also a major component of thick filaments, and worms at the restrictive temperature have a 45% mean decrease in number of thick filaments compared with worms at the permissive temperature (). Paramyosin levels are also decreased in worms, indicating a loss of total thick filaments (Fig. S2, available at ). Because our transgenic line carries multiple copies of the promoter, the possibility existed that the observed muscle phenotype was caused by its excessive binding of transcription factors. However, real-time RT-PCR verified that mRNA levels of both and , which encode MHC A and B, respectively, were not significantly different between N2 wild-type and worms () with GAPDH mRNA as a control. These results indicated that the decrease in body wall myosin is not a result of diminished transcription. To examine a role for the UPS in the degradation of endogenous myosin in , we first tested whether we could detect ubiquitinated myosin species in whole worm lysates. We performed pull downs from wild-type worms with antiubiquitin and antimyosin antibodies and blotted them subsequently with a myosin-specific mAb (). The ubiquitin mAb was able to pull down full-length myosin as well as species of slower mobility, consistent with polyubiquitinated myosin species. Several smaller myosin-immunoreactive bands were also detected in both pull downs, which may represent ubiquitinated proteolytic fragments of myosin, consistent with previous studies (; ; ; ). Sarcomeric myosin can thus be ubiquitinated in . Elevated ubiquitination would lead to accelerated degradation, which in fact was observed in lysates of and worms at 25°C when compared with N2 and worms at 15°C (Fig. S3, available at ). Proteasome-dependent proteolytic degradation requires ATP and is inhibited by MG132 (). We thus tested whether degradation of endogenous myosin exhibited these properties. We incubated separate reactions containing crude wild-type worm extracts with no additions, ATP supplementation, MG132 supplementation, and a combination of both, and examined the levels of full-length myosin and its degradation products by immunoblotting (). The unsupplemented reaction exhibited constant levels of full-length myosin, whereas the reaction supplemented with ATP showed a clear reduction of full-length myosin over time. MG132 by itself had no effects on full-length myosin levels, but the observed ATP-dependent degradation was effectively blocked by addition of MG132. The proteasome is therefore capable of degrading myosin in . We next tested whether proteasome inhibition could restore the diminished myosin levels observed in the worms that overexpress UNC-45. RNAi against the proteasomal subunit RPT-2 has been previously used to inhibit proteasome function in living worms (). Importantly, at the concentration used (see Materials and methods), RPT-2 RNAi did not result in complete inhibition of the proteasome, which allowed the nematodes to survive. When RPT-2 RNAi was fed to worms, we observed clear restoration of body wall myosin (). Furthermore, RPT-2 RNAi also significantly improved the mobility of nematodes (). Because UNC-45 itself is subject to degradation by the UPS, the previous observation that Lof mutants are rescued by RPT-2 RNAi () can be explained by increased levels of UNC-45, myosin, or both. Our current results show that endogenous myosin can be ubiquitinated in and that proteasome inhibition can prevent its degradation, in both a wild-type background as well as that resulting from increased UNC-45 function. We have found that UNC-45 overexpression results in diminished myosin accumulation and assembly because of its increased degradation via the UPS. Lof mutants also show similar but enhanced defects. These results allow us to propose an idealized model in which precise levels of UNC-45 are critical for supporting adequate myosin folding and assembly into thick filaments (). As a myosin chaperone (), UNC-45 binds to newly synthesized myosin motor domains that have not yet attained their full native structure. UNC-45 assists in folding of the motor domain so that it becomes competent for assembly. We hypothesize that myosin produced in the context of UNC-45 Lof mutations is less capable of attaining its native structure and, as a result, is degraded by the UPS, thus explaining its drastically reduced levels () and aberrant assembly into thick filaments (). Myosin synthesized in the context of excess UNC-45 may also be prevented from reaching the assembled state because of mass action and is then susceptible to degradation via the UPS. Because both endogenous and transgenic UNC-45 can associate with assembled myosin in vivo (; unpublished data), excess UNC-45 may also be capable of shifting the myosin equilibrium from the assembled state into a degradation-susceptible unassembled state. The interplay between UNC-45, myosin, and the UPS may be of relevance to similar concentration-dependent phenomena between chaperones and client proteins capable of polymerization, such as those reported for Hsp104 and yeast prions (). Vertebrates contain two distinct genes encoding different UNC-45 isoforms. General cell UNC-45 is expressed in multiple tissues and appears necessary for various cytoskeletal functions, whereas striated muscle (SM) UNC-45 is specifically expressed in heart and skeletal muscle and may be necessary for sarcomere organization (). SM UNC-45 may function during sarcomere assembly in a mechanism similar to the one described here for UNC-45. Alterations in human SM UNC-45 function or concentrations may be significant in hypertrophy, dilation, and failure of the heart, as well as in skeletal muscle wasting in a variety of human disorders. Nematode strains were grown under standard conditions (). was created as previously described (). Genomic integration of the P∷ extrachromosomal array was performed via gamma irradiation (). The resulting worms were outcrossed five times. Strain BC10095 was obtained from D. Baillie (Simon Fraser University, Burnaby, Canada). In , , and Fig. S2, Ya worms were hand-picked, placed in SDS sample buffer, and heated at 95°C for 10 min. Amounts corresponding to a single worm were loaded per lane. In Fig. S3, equal volumes of L4 Ya stage worms were lysed in 50 mM NaPO, pH 7.4, 200 mM NaCl, 5 mM MgCl, 1% Triton X-100, 5 mM DTT, and 5 mM ATP. Samples were resolved by SDS-PAGE and transferred to Hybond nitrocellulose (GE Healthcare). Blots were reacted with mAb 5-6 (anti-MHC A), 28.2 (anti-MHC B), 5-17 (anti-MHC D), 5-23 (anti-paramyosin; Miller et al., 1983), or anti-HDA-1 (Santa Cruz Biotechnology, Inc.) at 2, 5, 5.25, 5, and 2 μg/ml, respectively, for 1 h, followed by HRP-conjugated anti–mouse IgG or anti–rabbit IgG at 1:10,000 dilutions. Densitometry was performed using AlphaEaseFC software (Alpha Innotech). Immunofluorescence microscopy on whole mounts of Ya worms was performed as described previously (). Worms were reacted with rhodamine-conjugated mAb 5-6 (anti-MHC A), FITC-conjugated mAb 5-8 (anti-MHC B), or Cy3-conjugated M2 (anti-FLAG; Sigma-Aldrich). Images were visualized using a microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) with a 40×/0.75 plan-NEOFLUAR objective and acquired using a digital camera (AxioCam MRc5; Carl Zeiss MicroImaging, Inc.) and AxioVision 3.0 software (Carl Zeiss MicroImaging, Inc.) at room temperature. Fluorescence microscopy of strain BC10095 (P∷GFP) was visualized as described. Polarized light microscopy was performed as described previously (), except that image acquisition was performed using the system described above. Total RNA was isolated from synchronized N2 and worms using an RNeasy mini kit (QIAGEN). Real-time RT-PCR was used to determine gene expression using the QuantiTech SYBR Green RT-PCR kit (QIAGEN) according to the manufacturer's suggestions. Triplicate C values were analyzed in Excel (Microsoft) using the comparative C(ΔΔC) method. Individual Ya worms were placed in M9 buffer, and body bends were counted over 15 s (). RNAi was performed by feeding nematodes the HT115 bacteria containing RPT2 double-stranded RNA in plasmid pPD129.36 () as described previously () except that RNA transcripts were induced with 10 nM IPTG to ensure survival. N2 worms were equilibrated and lysed in 20 mM Tris-HCl, pH 8.0, 200 mM NaCl, 5 mM MgCl, and 5 mM DTT. The homogenate was centrifuged at 5,000 for 10 min at 4°C, and the supernatant was divided into four aliquots to which no supplementation, 2 mM ATP, and/or 100 μM proteasome inhibitor MG132 were added. Samples were kept at room temperature, and aliquots were removed at each time point, mixed with SDS-PAGE sample buffer, and immediately heated at 95°C for 10 min. Samples were resolved by SDS-PAGE followed by Western blots using 5 μg/ml mAb 28.2. N2 worms were lysed and incubated with 100 μg mAb 28.2 or 200 μg anti-ubiquitin mAb (Santa Cruz Biotechnology, Inc.) bound to protein G Sepharose (GE Healthcare) for 2 h at 4°C. Pull downs were washed five times with lysis buffer, mixed with SDS sample buffer, and heated at 95°C for 10 min. Samples were separated by SDS-PAGE followed by Western blotting using 5 μg/ml mAb 28.2. Fivefold less volume of the MHC B pull down compared with the ubiquitin pull down was loaded to avoid saturation. Fig. S1 shows that the phenotype arises specifically from UNC-45 overexpression. Fig. S2 shows that paramyosin levels are also decreased in worms. Fig. S3 shows that myosin accumulation is also decreased in whole worm lysates from and worms at the restrictive temperature. Online supplemental material is available at .

Glycophosphatidylinositol (GPI) anchorage is a common feature of surface proteins that leads to membrane raft localization. The observation of different functional GPI anchors (; ) as well as the fact that different rafts show markedly different lipid and protein profiles (; ; ) implies the existence of a heterogeneous set of anchors and matching rafts. Anchor addition is determined by the GPI anchor signal sequence, which consists of a set of small amino acids at the site of anchor addition (the ω site) followed by a hydrophilic spacer and ending in a hydrophobic stretch (). Cleavage of this signal sequence occurs in the ER before the addition of an anchor with conserved central components () but with variable peripheral moieties (). Carcinoembryonic antigen (CEA) is a GPI-anchored protein, whereas the closely related CEACAM1 (CEA-related cell adhesion molecule 1 [CC1]) contains a transmembrane (TM) domain. In vitro, both proteins mediate intercellular adhesion (; ), but CEA, not CC1, blocks cellular differentiation () and inhibits the apoptotic process of anoikis (; ). Exchanging the membrane anchors of these proteins results in a TM version of CEA that does not show CEA-like activity and a GPI-anchored CC1-like protein that now exhibits CEA-like properties, demonstrating the importance of the membrane anchor (). Replacing the GPI anchor signal sequence of neural cell adhesion molecule (NCAM) for that of CEA results in a functionally CEA-like protein (NCB), demonstrating the existence of functionally specific anchors whose addition is determined by a particular signal sequence (). The sole role of the external domains is to mediate self-binding and, thereby, concentration-dependent clustering (; ), as external domain mutations that disrupt the self-binding of CEA abrogate CEA function, whereas irrelevant self-binding external domains (such as that of NCAM in NCB) suffice for function. Because the amino acid sequence of various GPI anchor signal sequences radically affects protein function, we hypothesized that a signal existed within the GPI anchor signal sequence specifying the addition of a particular functional GPI anchor. Chimeras were generated by exchanging fragments of the CEA and NCAM GPI anchor signal sequences and were tested for CEA-like biological properties. We identify a specificity signal consisting of five amino acids that is necessary and sufficient in conjunction with an upstream proline for addition of the CEA-specific GPI anchor. Although the primary sequences of the GPI anchor signal sequences of CEA and CEACAM6 are very similar, mirroring their identical tumorigenic functions, that of NCAM differs greatly (). The signal sequence from CEA is capable on its own of specifying the addition of the CEA anchor, so chimeras were generated reducing (in five–amino acid increments) the CEA-derived sequence in NCB to localize the sequence responsible for this effect (). These chimeras were tested for biological activity in the CHO-derived LR-73 and the rat myoblast L6 cell lines, although NCBΔ20 was not expressed in L6 transfectants (). The sensitivity of these proteins to phosphatidylinositol PLC (PIPLC) and their insolubility in cold Triton X-100, with most of each protein present in the insoluble fraction, confirmed GPI anchorage (; ). LR transfectants showed strong intercellular adhesive ability (), which is indicative of the retention of the self-binding activity of their external domains (; ), although NCBΔ20 adhered somewhat less, likely because of its lower expression level. CEA and NCB but not NCAM alter the activity of integrin α5β1 (; ) and block differentiation (; ), which are characteristics used to determine which chimeras retained the functional activities conferred by the CEA GPI anchor. LR transfectants were tested for binding to the major α5β1 ligand fibronectin (Fn; ), with NCB expression significantly increasing binding compared with NCAM (P < 0.01). Replacing the first five CEA-derived amino acids of NCB with the equivalent NCAM residues had no effect on CEA-like function (P < 0.005 vs. NCAM), but replacing 10 residues resulted in a protein (NCBΔ10) that no longer altered binding. In L6 myoblasts, NCBΔ5 had a similar effect on binding to Fn compared with NCB (P < 0.005; ), whereas NCBΔ10 transfectants lost this ability completely. NCB expression completely blocks L6 morphological differentiation, whereas NCAM transfectants differentiate readily to form large multinucleated myotubes (). NCBΔ5 expression also blocked differentiation, whereas NCBΔ10-expressing cells fused substantially, which is comparable with the parental and NCAM controls (all fused between 75 and 80%; ). NCAM and NCBΔ10 transfectants up-regulated myosin, a biochemical differentiation marker, but NCB and NCBΔ5 transfectants did not (). Thus, addition of the CEA-specific GPI anchor is determined by the residues that differ from NCBΔ5 to NCBΔ10 (GLSAG), as their substitution with the corresponding NCAM amino acids (SASYT) abrogated the CEA-like biological function. Five amino acid stretches in the CEA GPI anchor signal sequence were next replaced with the corresponding NCAM residues (). CC1-CEA (1C) chimeras were used because previous attempts to attach the NCAM anchor to CEA failed (), whereas the CC1-NCAM chimera (1N) is partially processed (), as seen previously in certain CC1 mutants (). These proteins were expressed at high levels with the exception of 1N () and were GPI anchored () and mediated strong intercellular adhesion (). In LR cells, the expression of 1C caused a significant increase (P < 0.001) in cellular binding to Fn compared with parental cells (). Replacing the first five CEA amino acids (1CN5) or amino acids 11–15 (1CN15) downstream of the ω site resulted in proteins that were still active (P < 0.001); however, replacing amino acids 6–10 (1CN10) completely abrogated the increased binding. In L6 cells, 1C, 1CN5, and 1CN15 expression significantly increased binding to Fn (P < 0.001; ), whereas 1CN10 transfectants bound the same as parental cells. Although the expression of 1CN10 was lower than 1C in L6 cells, 1CN5 and 1CN15 still showed increased binding despite having expression levels similar to 1CN10 (). The expression of 1C, 1CN5, and 1CN15 but not 1CN10 also strongly inhibited L6 morphological () and biochemical () differentiation. Replacing amino acid stretches shorter than five residues in this region had no effect on the binding of LR transfectants to Fn (Fig. S1 D, available at ). Therefore, these results confirm the importance of the residues GLSAG in determining the addition of the CEA anchor. It was next examined whether this sequence was sufficient to confer CEA-like biological properties. CEA amino acids were inserted into NCAM at positions 1–5, 6–10, or 11–15 downstream of the ω site, with NC10 containing the GLSAG sequence (), resulting in GPI-anchored proteins (). Although the proteins mediated intercellular adhesion (), chimera expression did not result in increased Fn binding in either LR or L6 cells (). Differentiation of L6 cells was not blocked because morphological () and biochemical () differentiation was observed in these transfectants. Thus, inserting the sequence GLSAG into NCAM was insufficient to give CEA-like biological activities, suggesting a requirement for further CEA residues. Therefore, larger amounts of the CEA-derived sequence were inserted into NCAM to determine the minimum sequence sufficient for specifying the addition of the CEA GPI anchor (). All chimeras contained the GLSAG sequence, with variable amounts of upstream and/or downstream CEA sequence, and the resulting proteins were GPI anchored () and mediated intercellular adhesion (). When examined for the effects on LR binding to Fn, adding five upstream CEA amino acids (NΔ110C) but not five downstream amino acids (NΔ615C) resulted in increased binding (P < 0.0001; ). Simply adding one CEA residue on each side of GLSAG (PGLSAGA; NC7) also produced increased cellular binding, suggesting, along with the NΔ110C result, that the upstream proline was required for CEA anchor addition. This was examined directly by generating constructs containing only the upstream proline (NC6P) or the downstream alanine (NC6A; ). In LR cells, NC6P expression increased binding to Fn, whereas NC6A had no effect, confirming the importance of the proline (P < 0.0001; ). These results were recapitulated in L6 cells, where only transfectants of NΔ110C, NC7, and NC6P showed a significant difference in binding to Fn compared with the parental cell line (P < 0.0001; ). All chimeras containing the sequence PGLSAG blocked differentiation, whereas those lacking the proline fused similarly to NCAM transfectants (). Thus, inserting the sequence PGLSAG into the GPI anchor signal sequence of NCAM is sufficient to generate a protein with CEA-like biological properties, demonstrating a requirement for the presence of the proline. In CEA and all of these NCAM chimeras, this proline is a part of a G(X)XP sequence (). This consensus sequence has been shown to result in a kink in TM helices () and can be suggested to serve a similar function in this GPI anchor signal sequence. The resulting altered structure may be important to determine the addition of a certain functional anchor. However, it should be noted that the lack of the proline can be overcome if a sufficient downstream CEA sequence is included (, NCBΔ5). The sequence GLSAG was also randomly scrambled to give sequences of ASGGL (denoted NCB-K) and SGLGA (NCB-S; ). It was hypothesized that a complete loss of biological function would be observed if there was a requirement for a particular amino acid sequence or a particular amino acid at a given position, whereas at least partial function should be retained if the signal resulted from a general characteristic of this stretch. These proteins were GPI anchored () and promoted intercellular adhesion (). LR transfectants of both chimeras significantly increased binding to Fn compared with the NCAM cell line (P < 0.001; ). However, both scrambled transfectants bound less than NCB, particularly NCB-S (P < 0.002). L6 transfectants showed altered binding compared with the parental cell line (P < 0.003 for NCB and P < 0.05 for NCB-K and NCB-S; ), although transfectants of NCB-K and NCB-S again bound significantly (P < 0.05) less compared with the NCB transfectants. L6 (NCB-K) and L6 (NCB-S) cells differentiated substantially less than the parental or NCAM controls but did not show completely blocked morphological fusion (). Myosin up-regulation was seen in the NCB-K and NCB-S cell lines, which is contrary to NCB, although at lower levels than the NCAM transfectant (). Thus, scrambling this region results in an incomplete loss of function, indicating that the primary source of the signal is the overall region's characteristics, although this signal is maximized by the sequence PGLSAG. The signal for the addition of a GPI anchor consists of a set of small amino acids followed by a spacer and a hydrophobic region (; ). Work on the bovine liver 5′-nucleotidase has demonstrated the requirement for particular lengths of both the spacer and the hydrophobic domain for proper processing (, ). Differences in the efficiency of GPI anchor addition for various signal sequences suggest that these stretches are not processed identically (). However, these previous studies have been concerned with the efficiency of anchor addition; this study is the first to demonstrate that the specificity of anchor addition is the result of a second signal within this sequence. We have previously demonstrated that the CEA GPI anchor signal sequence determines function and localization to a specific membrane raft despite being cleaved in the ER (; ). This study was designed to establish the residues that are critical for this specification, with the demonstration that the amino acids PGLSAG in the hydrophilic region of the GPI signal sequence are necessary and sufficient for this effect. The identified sequence from CEA is fairly well conserved in CEACAM6 (PVLSAV) and CEACAM7 (PDLSAG; ), which are proteins that show similar biological effects to CEA (; unpublished data), suggesting that it may have a similar role in determining the function of these proteins. Because GPI anchors from five other proteins did not show any of these effects when bound to the CC1 external domain (unpublished data), this amino acid set is quite specific. This work has identified a novel signal within the GPI anchor signal sequence of CEA, which determines protein functionality. Studies using various GPI-anchored protein comparisons such as CEA and NCAM (), Thy-1 and the prion protein (), and folate receptor and placental AP () have suggested that different GPI-anchored proteins exist in different microdomains on the cell surface. This distribution is important because switching the anchor and the subsequent distribution of NCAM to that of CEA is sufficient to radically alter its function (; ). This occurs through association with particular rafts and their specific signaling elements, which determine the downstream tumorigenic effects of CEA (). Examining the GPI anchor signal sequences of various other proteins should further elucidate the specificity and importance of this signal in determining specific anchor addition and, ultimately, protein function. It will also be important to characterize how this region is capable of determining the addition of a particular anchor. It is possible that this stretch of amino acids interacts directly with the GPI anchor precursor before the transamidation binding reaction, and only combinations that match structurally proceed enzymatically. Alternatively, the transamidase complex, which is composed of five different subunits, may play a direct role in this effect. One component, Gaa1p, has been suggested to recognize the signal sequence (), whereas another, Gpi8p, functions as the enzymatic subunit (). Either of these or one of the three complex components (PIG-S, PIG-T, and PIG-U) with currently unknown function could serve to bring together specific signal sequences and GPI anchors. This study demonstrates that a specific six–amino acid stretch of the GPI anchor signal sequence determines the addition of a particular functional anchor, which, in turn, can determine the ultimate function of the protein. All chimeras were generated by PCR overlap extension. The ω site of NCAM remains unknown and was assigned to be A736 on the basis of sequence alignment with chicken NCAM, in which other potential anchor addition sites were not conserved between humans and chickens and, as such, are unlikely to serve as the ω site (). Constructs were generated using NCAM, NCB, or C1-C cDNA (note that the C1-C used in this study did not contain the I to F point mutation described in the original study [] and was called 1C to allow for differentiation between the two proteins) and the primers indicated in Table S1 (available at ). Initial PCR reactions involved separate extensions using the CC1 or NCAM sense primer with the corresponding antisense chimera primer and the sense chimera primer with the antisense CEA or NCAM primer. These fragments were joined by overlap PCR using the CC1 or NCAM sense primer and the CEA or NCAM antisense primer. The resulting NCAM-like fragments were inserted into the EcoRI sites of NCAM (at positions 1,616 and 2,794) in the p91023b expression vector. The C1-C chimeras replaced the corresponding sequence in C1-C using the internal CC1 BamHI digestion site (at position 971) and the BamHI digestion site located in the polylinker region of the vector pEGFP-C2 (used as a cloning vector; BD Biosciences) and were subcloned into p91023b using flanking EcoRI sites. CHO-derived LR-73 fibroblasts and rat L6 myoblasts were cultured as previously described (). Cell surface protein expression was determined by FACS analysis using the mouse mAbs J22, which recognizes CEA and CC1 (), and 123C3 (Santa Cruz Biotechnology, Inc.), which recognizes the NCAM external domain. Transfection and sorting for high expression was performed as previously described (). L6 myoblasts were seeded at 10 cells/cm in 60-mm dishes in medium containing 10% FBS on day 0. 3 d later, the culture medium was changed to 2% horse serum. Myogenic differentiation was assayed 5 d after changing media either by staining with hematoxylin (Sigma-Aldrich) to assess fusion into multinuclear myotubes by light microscopy () or by lysing cells and performing Western blots for myosin expression using mouse mAb 47A (). Photomicrographs of representative fields of stained cells were obtained at room temperature using a microscope (Eclipse E800; Nikon) and a 10× NA 0.30 Ph1 ∞/0.17 objective. Images were acquired with a digital camera (DXM1200; Nikon) and ACT-1 image acquisition software (Nikon). The fusion index was determined by counting the number of nuclei present in fused myotubes (taken as cells with three or more nuclei) and comparing this to the total number of nuclei in the field. Assays were performed essentially as described previously (). For Triton X-100 solubility, cells were collected with PBSCE and resuspended in cold 1% Triton X-100 with protease inhibitors. Cells were syringed through a 27-gauge needle, incubated on ice for 15 min, and centrifuged at 13,000 for 15 min. The supernatant fraction was removed, and the pellet was resuspended in the same volume as the supernatant. Partitioning between pellet and soluble fractions was assessed by immunoblotting; integrin α5 should be found in the supernatant (soluble) fraction and was used as a lysis control, with detection by a rabbit polyclonal anti-α5 (H-104; Santa Cruz Biotechnology, Inc.). For PIPLC sensitivity, monolayer cultures were incubated with 0.1 U bacterial PIPLC (Sigma-Aldrich) in a 1:1 solution of DME/PBS containing 0.2% BSA for 45 min at 37°C. Treated and control untreated cultures were then washed with PBS, rendered single-cell suspensions by light (0.063%) trypsin treatment, and were processed for FACS analysis. Percent sensitivity was determined as the percent decrease in mean fluorescence value (relative units) in the treated sample compared with the untreated control. Adhesion assays were performed as previously described (). Cells were removed from culture flasks by light trypsin treatment (for CC1 external domain chimeras, which are insensitive to trypsin) or PBS citrate + 4 mM EDTA (PBSCE; for NCAM external domain chimeras, which are cleaved by trypsin) and resuspended at a concentration of 10 cells/ml in α-MEM containing 0.8% FBS and 10 μg/ml DNase I (Roche). Single-cell suspensions were obtained by syringing through a 27-gauge needle and were allowed to aggegrate at 37°C with stirring at 100 rpm using a magnetic stirring bar (Spinbar Micro Stir Bar; VWR International). Aliquots were removed at the indicated times, and the percentage of single cells was determined by a hemocytometer (Bright-Line; VWR International). Assays were performed essentially as previously described (). Cells were resuspended at a concentration of 4 × 10 cells/ml for LR or 2 × 10 cells/ml for L6. 100 μl of this suspension was added to Fn-coated plates (Chemicon international) and incubated for 1 h at 37°C. Adherent cells were stained with crystal violet, and the optical density was determined with a plate reader (PowerWave; Bio-Tek Instruments) at 570 nm. Note that in certain cases for L6 cells, depending on the particular experiment, integrin activation resulted in decreased cellular binding to Fn, which is likely the result of the previously described integrin activation-dependent formation of a cocoon of polymerized Fn around the cells (). However, the relative difference between parental cells and activated transfectant cells remained in the inverse sense, so data is presented for ease of interpretation as an increase in all cases. Statistical significance was determined using the -test (). Cellular lysates were resolved by SDS-PAGE and transferred to a 0.45-μm polyvinylidene difluoride membrane (Millipore). Antibody binding was detected using the ECL Plus chemiluminescent reagent (GE Healthcare). Table S1 contains the nucleotide sequences of the primers used to generate the chimeras. Fig. S1 demonstrates that replacing less than five CEA amino acids in the region 6–10 is insufficient to cause a complete loss of biological function. Online supplemental material is available at .

DNA double-strand breaks (DSBs) are especially genotoxic DNA lesions because they potentially lead to chromosomal breakage, fragmentation, and translocation. DSBs are commonly caused by exogeneous agents, such as ionizing radiation (IR) or mutagenic chemicals, but are also caused by radicals that emerge during normal cellular metabolism. In addition, DSBs are generated during V(D)J recombination, which is an essential process in the development of functional B and T lymphocytes. It is therefore of vital importance that each cell is equipped with enzymatic machineries that mediate DSB repair. At least two distinct pathways have evolved that mediate the repair of DSBs: homologous recombination (HR) and nonhomologous end-joining (NHEJ; ; ; ; ). NHEJ is considered to be the prevailing pathway during the G0 and G1 phases of the cell cycle in mammalian cells because this repair pathway does not require the presence of an intact DNA template. NHEJ involves juxtaposition of DNA ends by an enzymatic machinery and subsequent ligation. When DNA termini are incompatible or damaged, processing is necessary before ligation can proceed. Two protein complexes make up the catalytic core of the NHEJ process: the DNA-dependent protein kinase holoenzyme (DNA-PK) and the DNA ligase IV–XRCC4 complex (; ). Ligase IV–XRCC4 mediates ligation of the juxtaposed DNA ends in the final NHEJ step. The DNA-PK holoenzyme consists of the Ku70/80 heterodimer and a 470-kD catalytic subunit (DNA-PK) with serine/threonine protein kinase activity. The formation of a kinase-competent DNA-PK complex by Ku70/80 and DNA-PK requires simultaneous binding of these enzymes to a DNA terminus (; ). Because Ku70/80 has much higher affinity for DNA ends than DNA-PK, this heterodimer most likely binds to DNA termini first and subsequently attracts DNA-PK toward the DSB. Many targets for the DNA-PK kinase have been found in vitro, but the biological relevance of these observations is unclear in most cases. It is, however, well established that DNA-PK has the ability to autophosphorylate itself at a cluster of 6 phosphorylation sites between the Thr2609 and Thr2647 amino acid residues (), as well as at an additional site outside this cluster, the Ser2056 residue (). This activity possibly leads to alteration of the protein's affinity for DNA and to inactivation of its kinase activity. Such phosphorylation-induced alterations are important during DSB repair in vivo because mutations in the phosphorylation cluster cause severely increased radiation sensitivity and decreased DNA repair (; ; ). Several studies have shown that the presence of DNA-PK at DNA ends interferes with efficient ligation, most likely caused by the large dimensions of the protein molecule (; ; ; ). This inhibition of ligation can be relieved by DNA-PK autophosphorylation, indicating that autophosphorylation induces a conformational change in the DNA-PK molecule that liberates DNA ends (; ; ; ). These findings gave rise to the current notion that DNA-PK functions as a “gatekeeper,” which protects DNA ends from premature processing and ligation, until the two DNA ends are properly juxtaposed and DNA-PK autophosphorylation can take place (). The spatiotemporal events of this process, however, remain largely unknown. Not much is known about the in vivo dynamics of enzyme–DNA complex formation upon the onset of a DSB event. Several authors have reported the use of microscope-coupled lasers to introduce DNA damage in specified regions of the cell nucleus to address questions concerning the sequential recruitment of different DSB repair enzymes to those damage sites (; Bekker-Jensen et al., 2006). A recent study examined the in vivo recruitment of NHEJ core enzymes to DSBs after the introduction of DNA damage by the use of a pulsed, near-infrared laser (), and it showed that the binding of Ku70/80 to DNA ends is a highly dynamic equilibrium of association and dissociation. We present a study in which we introduce DSBs in a specific region of the nucleus of living cells by using a pulsed nitrogen laser that was coupled to the epifluorescence path of a fluorescence microscope. This setup enables us to examine the recruitment of YFP-tagged DNA-PK to a DSB site in vivo and to follow its behavior throughout the repair process. We monitor the behavior of wild-type (WT) DNA-PK and two mutated forms of DNA-PK that are either impaired in kinase activity or in the ability to be phosphorylated at 7 phosphorylation sites (the Thr2609-Thr2647 cluster and the Ser2056 residue). These mutations are known to interfere with efficient DSB repair (; ; ; ; ). We demonstrate that both WT and mutant DNA-PK proteins readily accumulate at the DSB site in a Ku-dependent manner, indicating that neither the kinase activity nor the clustered phosphorylation of DNA-PK is important for DNA-PK recruitment to DSBs. Impairment of either one of these functions, however, does result in deficient DSB repair. Our data show that DNA-PK remains present at unrepaired DSBs for at least 2 h. By using photobleaching techniques, we demonstrate that DNA-PK is not present at DNA ends as a rigid complex, but that there is a dynamic exchange between DNA-bound and free protein. Impairment of phosphorylation/autophosphorylation causes this dynamic exchange to take place at a much lower rate, suggesting that unphosphorylated DNA-PK forms a more stable complex with DNA ends than phosphorylated DNA-PK. Our findings suggest a model for DNA-PK– mediated end-joining in which autophosphorylation is required to destabilize the protein–DNA complex. This destabilization results in accessibility of DNA ends, which, in turn, facilitates ligation. Such a model could comprehensibly explain the radiation-sensitive phenotype of DNA-PK autophosphorylation mutants. To study the spatial and temporal dynamics of DNA-PK recruitment to DSBs, we generated a DNA-PK–deficient V3 cell line that stably expressed YFP-tagged DNA-PK. The expression of tagged DNA-PK (verified by Western blot analysis; Fig. S1 A, available at ) lead to a considerable decrease in the radiation sensitivity of the V3 cells (Fig. S1 B), indicating that the YFP-tag did not interfere with the function of DNA-PK during DSB repair. We introduced a small area of DNA damage in the nuclei of these cells by using a microirradiation system that utilizes a 365-nm pulsed nitrogen laser. Directly after microirradiation, YFP-DNA-PK started to accumulate at the damaged region (). We verified that the laser treatment generated DSBs by TUNEL labeling and by staining with an antibody that recognized γH2AX. DNA-PK clearly colocalized with TUNEL and γH2AX in the microirradiated area (), demonstrating that DNA-PK localizes in vivo at laser-induced DSB sites. To estimate the number of DSBs that was produced during microirradiation, we compared the DNA-PK accumulation that was caused by our laser-system with the DNA-PK accumulation that was caused by an already calibrated microirradiation system (). In brief, we treated YFP-DNA-PK–expressing cells with either our 365-nm laser system or with the 800-nm multiphoton laser system used by , which is known to introduce between 1,000 and 1,500 DSBs. We then measured and compared the fluorescence intensity of the accumulated YFP-DNA-PK at the microirradiated sites, which is correlated in a linear fashion to the number of induced DSBs. After correction for the differences in accumulation area size, we found that the fluorescence intensity of the accumulated DNA-PK was 2.5-fold higher after microirradiation with the 365-nm laser than after irradiation with the 800-nm laser. From this, we infer that ∼3,000 DSBs are produced at the beam-focus of our pulsed 365-nm laser (see Materials and methods for details). To study the kinetics of DNA-PK accumulation after microirradiation with the 365-nm laser, we performed time-lapse imaging. Recruitment of YFP-DNA-PK to DSB sites was observed as early as 2 s after microirradiation, and the intensity of the YFP signal rapidly increased during the first 20 s (). To verify whether the kinetics of DNA-PK accumulation at laser-induced DSB sites are comparable to those at IR-induced DSB sites, we exposed our YFP-DNA-PK cells to charged uranium particles and observed immediate DNA-PK accumulation. The accumulation pattern matched the pattern of the uranium particles and colocalized with γH2AX (). The fast DNA-PK accumulation at the uranium-induced DSB sites resembled the DNA-PK accumulation at laser-generated DSB sites (), suggesting that the response of DNA-PK to both types of induced damage is comparable. In addition, we investigated the phosphorylation status of accumulated DNA-PK in the microirradiated area. We have previously shown that IR induces phosphorylation/autophosphorylation of the Ser2056 and Thr2609 amino acid residues of DNA-PK (; ). Immunostaining with phosphospecific antibodies revealed that both the Ser-2056 and Thr-2609 residues were phosphorylated in the microirradiated region (), indicating that the laser-induced DNA lesions are capable of inducing a repair response that involves DNA-PK phosphorylation/autophosphorylation. Many in vitro studies have shown that the Ku70/80 heterodimer directs DNA-PK to DNA ends and stabilizes the DNA–DNA-PK interaction (). To investigate whether the recruitment of DNA-PK to DSB sites is dependent on the Ku protein in vivo, we generated a Ku80-deficient Xrs6 cell line that stably expressed YFP-DNA-PK. After microirradiation of the nuclei of these cells, we did not observe any accumulation of DNA-PK, although γH2AX staining confirmed that DSBs were introduced in the irradiated area (). Continued time-lapse imaging for a period of 2 h after microirradiation did not reveal any accumulation of DNA-PK (not depicted). In addition, we complemented our YFP-DNA-PK–expressing Xrs6 cell line with human Ku80, which restored the expression level of Ku80 protein and decreased the radiosensitivity of these cells to the same level as that of the parental AA8 cell line (Fig. S2, A and B, available at ). After microirradiation of these complemented cells, we observed immediate accumulation of DNA-PK (). The kinetics of this accumulation were very similar to those observed in V3 cells (). These data demonstrate that Ku80 is essential for the recruitment of DNA-PK to DSB sites in vivo. To examine whether the phosphorylation of DNA-PK also takes place in a Ku-dependent manner in our system, we immunoprecipitated DNA-PK from Ku-deficient, YFP-DNA-PK–expressing Xrs6 cells and from YFP-DNA-PK– expressing V3 cells after IR treatment with gamma radiation. Subsequently, we performed a Western blot analysis, using phosphospecific antibodies against the Ser-2056 residue of DNA-PK. Ser-2056 phosphorylation clearly took place after IR treatment in the V3 cells, but the Ku-deficient cell line did not show any detectable Ser-2056 phosphorylation (Fig. S2 C). From these data and the previously reported notion that the phosphorylation of Thr2609 does not take place in Ku-deficient cells (), we conclude that phosphorylation/autophosphorylation of DNA-PK is a Ku-dependent process. To analyze the kinetics of Ku80 accumulation at DSB sites, we generated YFP-Ku80–expressing Xrs6 cells and subjected those to microirradiation. Accumulation of Ku80 at DSB sites was observed within 2 s after microirradiation () and the kinetics of Ku80 accumulation were identical to those of DNA-PK accumulation (). These data suggest that Ku80 and DNA-PK show similar spatial and temporal behavior at DSB sites immediately after the damage induction. It has previously been shown that the kinase activity of DNA-PK is essential during NHEJ-mediated DSB repair in mammalian cells (; ). Because mutation of DNA-PK phosphorylation sites is also known to interfere with DSB repair, clustered phosphorylation of DNA-PK is likely to play an important role during NHEJ (; ; ; ). To test whether kinase activity and clustered phosphorylation of DNA-PK are important for its recruitment to DSB sites, we generated two V3 cell lines that stably expressed YFP-tagged mutant forms of DNA-PK In one mutant form (KD), we impaired the kinase ability (). In the other mutant (7A), we substituted all of the 7 major phosphorylation sites (Thr2609-Thr2647 and Ser2056) with alanine residues, impairing phosphorylation of these sites (). Cellular expression of both mutant proteins was verified by Western blot analysis (Fig. S1 A). Upon microirradiation, we observed accumulation of both the KD and the 7A variant of DNA-PK colocalizing with γH2AX (). Accumulated KD protein proved to be phosphorylated at the Thr2609 and Thr2647 amino acid residues, but not at the Ser2056 residue (). This is consistent with our previous finding that Ser2056 is an autophosphorylation site and that phosphorylation of this residue therefore requires the DNA-PK kinase activity (). The Thr2609 and Thr2647 residues, which were previously reported to be autophosphorylation sites (; ), apparently do not require the DNA-PK kinase to be phosphorylated. It has recently been shown that phosphorylation of these two residues can also be mediated by ATM (). As expected, the 7A mutant did not display phosphorylation of Thr2609, Thr2647, or Ser2056 (). Time-lapse imaging showed that the accumulation kinetics of the 7A and KD proteins are indistinguishable from those observed with WT DNA-PK (). In both mutant cells, the intensity of the fluorescent signal in the microirradiated area increased rapidly during the first 20 s and reached a level that is identical to that found in WT cells. In addition, we demonstrated that pretreatment of WT cells with 30 μM of the phosphoinositide 3 kinase inhibitor wortmannin did not alter the accumulation kinetics of DNA-PK (). Collectively, these data show that the kinase activity and clustered phosphorylation of the DNA-PK protein are dispensable for its initial localization at DSB sites. After establishing the accumulation kinetics of both mutants at DSB sites, we performed time-lapse imaging experiments for a period of 2 h after laser irradiation. This allowed us to measure any decrease in the fluorescence intensity of the accumulation area in time. Interestingly, we observed significant differences between the behavior of WT and mutant proteins at the DSB site (). The fluorescence intensity of the accumulation area in WT cells decreased to 20% of the maximum level in a 2-h period, whereas the intensity of the accumulation area in KD and 7A cells only dropped to 80% of the maximum level in the same period of time. These results demonstrate that impairment of either DNA-PK kinase activity or clustered phosphorylation results in a maintained overall presence of DNA-PK at the DSB sites. The slow decrease of the fluorescence intensity of the accumulation area in mutant cells most likely reflects the inability of these cells to quickly repair the introduced DSBs because it is known that 7A and KD mutants display deficient DSB repair (; ; ; ; ). To verify this, we generated an XRCC4-deficient (XR1) cell line that expressed YFP-tagged WT DNA-PK. Because XR1 cells are unable to efficiently repair DSBs (), we expected the presence of DNA-PK to be maintained at the DSB site. Indeed, the fluorescence intensity of the accumulation area decreased only slightly in a 2-h period and reached a level that was comparable to that of KD and 7A (). These results show that the maintained presence of DNA-PK at DSB sites is correlated to deficient DSB repair, independent of the cause of the deficient DSB repair. We subsequently examined the behavior of DNA-PK mutants in which we impaired phosphorylation of a single site within the phosphorylation cluster. Because phosphorylation of both Thr2609 and Ser2056 is thought to be important for DSB repair via the NHEJ pathway (; ; ; ), we generated V3 cell lines that stably expressed YFP-tagged mutant forms of DNA-PK that could not be phosphorylated at those sites (). Cellular expression of both mutant forms of DNA-PK was verified by Western blot analysis (Fig. S3, available at ). We observed rapid accumulation of both phospho mutants at laser-induced DSB sites colocalizing with γH2AX (). As expected, the accumulation kinetics of S2056A and T2609 were identical to those of WT DNA-PK (). Although the S2056A protein disappeared slower from the DSB site than WT protein, neither mutant showed a behavior comparable to that of 7A (). It is very likely that the complete cluster of phosphorylation sites has to be substituted to observe a long, sustained presence of DNA-PK at the DSB sites. We subsequently examined the stability of the protein–DNA complexes that are formed by WT DNA-PK and the DNA-PK mutants at DSB sites by photobleaching the region of DNA-PK accumulation. Photobleaching took place 10 min after DSB induction, thus allowing for maximum accumulation of DNA-PK at the DSB site. We then monitored FRAP at regular intervals for a period of 10 min. As shown in , we observed recovery of fluorescence after photobleaching the area of DNA damage in WT DNA-PK cells, demonstrating that DNA-PK is not attached to DSBs in a rigid complex, but that there is a dynamic exchange between DNA-bound and free DNA-PK. In addition, we found that recovery of fluorescence after photobleaching occurred at a much faster rate in WT cells than in KD and 7A mutant cells (). In WT cells, fluorescence recovery reached a maximum of ∼60% of the prebleach intensity after 300 s. In contrast, fluorescence recovery in both mutants remained incomplete at that time. The 7A mutant recovered ∼30% of the prebleach intensity after 300 s, whereas the KD mutant displayed a remarkably low recovery of 10% at that time. These results clearly show that the dynamic exchange of DNA-bound protein with free protein is much faster in WT cells than in 7A or KD cells, and therefore this dynamic exchange must be influenced by both the kinase activity and the clustered phosphorylation of DNA-PK. We subsequently studied the FRAP dynamics of the WT, 7A, and KD proteins in an area of the cell nucleus where no DNA damage was present. The dynamics of WT, 7A, and KD proved to be identical under these conditions, showing that the 7A and KD mutants do not display an intrinsic tendency to recover slower than WT DNA-PK in the absence of DNA damage (). Maximum recovery of both WT and mutant DNA-PK in the undamaged area was reached in <10 s, which is 30 times faster than the recovery of WT DNA-PK at DSB sites (300 s). The much prolonged recovery rate of DNA-PK at DSB sites thus reflects the binding of DNA-PK to the DNA ends. Henceforth, the difference in recovery rates between WT, KD, and 7A at the DSB site () must reflect a difference in the stability of the DNA–protein species. To effectively study the response of the DNA-PK holoenzyme to the onset of DSBs in vivo, we needed a system that could quickly and reliably induce DSBs in a small region of the cell nucleus. We chose to use a 365-nm laser system, which is known for its ability to generate DNA damage in a controlled fashion (, ). The type and amount of DNA damage that is introduced by this laser can be controlled by adjusting the laser output intensity and the pulse frequency (). A previous study has demonstrated the recruitment of DSB repair proteins (NBS1, BRCA1, RAD52, and WRN) toward regions of the cell nucleus that were treated with this laser system, indicating that the system is capable of generating DSBs (). In addition, we show that the laser system generates DNA ends and that microirradiated areas colocalize with γH2AX, thereby demonstrating the formation of DSBs. It has recently been suggested that visible accumulation of Ku70 and DNA-PK at DNA damage sites is highly dependent on the output intensity of the laser that creates the DNA damage (Bekker-Jensen et al., 2006). In this study, a relatively mild laser treatment of BrdU-sensitized cells did not suffice to induce visible accumulation of both NHEJ enzymes, whereas treatment of nonsensitized cells with higher output intensity did result in detectable accumulation of DNA-PK and Ku70 at the damage site. The authors of this study theorize that the use of such high laser output intensity leads to a widespread chromosomal damage that saturates the cellular capacity to repair the lesions. Henceforth, they infer that high-powered lasers cause too much collateral damage to produce meaningful results (Bekker-Jensen et al., 2006). Our results, however, suggest that repair of the lesions that are generated with our laser system does take place efficiently. This is supported by the fact that accumulated WT DNA-PK disappears much faster from the DSB site in repair-proficient cells than in repair-deficient cells (). A recent study, in which cells were exposed to a comparable number of DSBs, showed conclusively that these cells were still able to complete cell division (). In addition, we would like to stipulate that the recruitment of DNA-PK in our system fully depends on the presence of Ku, ruling out the possibility of artificial, aspecific aggregation. It is generally assumed that the Ku70/80 heterodimer is the first NHEJ factor to bind to a broken DNA strand and that DNA-PK is subsequently attracted to the DSB sites (Chen et al., 1996; ). Recent studies have demonstrated that conserved C-terminal sequence motifs in Ku80 have the ability to interact with DNA-PK (). However, these in vitro studies do not form direct evidence for Ku80's ability to recruit DNA-PK to DSB sites. Some experiments have even suggested that DNA-PK is able to bind to DNA ends in the absence of Ku70/80 (; ). We directly demonstrate that DNA-PK does not visibly accumulate at induced DSB sites in Ku80-deficient cells. Complementing these cells with Ku80 restores the ability of DNA-PK to accumulate. Collectively, our findings indicate that recruitment of DNA-PK to DSB sites is mediated by Ku80 in vivo. In addition, we show that not only the recruitment but also the phosphorylation of (the autophosphorylation Ser-2056 residue of) DNA-PK is Ku dependent. This provides further evidence for the notion that the assembly of a kinase-competent DNA–PK complex relies on the presence of Ku80. The experiments presented in this paper demonstrate that kinase- and phosphorylation-impaired DNA-PK mutants readily accumulate at DSB sites, but that their presence at the DSB site is maintained for a significantly longer period of time than observed for WT DNA-PK. This most likely reflects the inability of these mutant cells to repair DSBs. Apparently, the presence of DNA-PK is maintained at unrepaired DNA breaks in general because even WT DNA-PK hardly disappears from the DSB site in repair-deficient XR1 cells. Clearly, in the latter case, this is not caused by any mutation in the DNA-PK molecule. Hence, we must conclude that whatever the cause of the deficient repair may be, DNA-PK will be present at the remaining DNA ends. But what causes diminished repair in the 7A and KD mutants? Our photobleaching experiments show that the presence of DNA-PK at DNA ends is not rigid, but that it is, in fact, a dynamic equilibrium between dissociating DNA-bound molecules and associating free molecules. These findings are in agreement with recent observations of the in vivo behavior of Ku80 at DSB sites (). The 7A and KD mutations cause a dramatic decrease in the exchange rate of bound and free molecules. In other words, these mutant DNA-PK molecules are more stably bound to DNA ends than WT molecules. Obviously, there is still exchange between bound and free protein, but at a much lower rate than observed for WT DNA-PK. Because inhibition of either one of the two requirements for DNA-PK autophosphorylation (either an active kinase domain or the ability to be phosphorylated) changes this exchange rate, it is reasonable to assume that autophosphorylation changes the stability of the DNA-PK–DNA complex. Biochemical experiments have demonstrated that unphosphorylated DNA-PK blocks DNA ends, and thereby inhibits efficient ligation (; ). This ligation block can be relieved by DNA-PK autophosphorylation, which induces a conformational change in the DNA–protein complex that liberates the DNA ends. Combined with our in vivo data on exchange rates, these data suggest a comprehensive model for the role of DNA-PK autophosphorylation during NHEJ (). In WT cells, unphosphorylated DNA-PK binds to the ends of a broken DNA strand. The dynamic dissociation/reassociation equilibrium of this unphosphorylated molecule will most likely be comparable to that of the 7A or KD mutants, which is a very slow rate of exchange. Hence, the DNA end will usually be occupied by a DNA-PK molecule. This effectively protects the DNA ends from (premature) processing, degradation by nucleases, or undesirable ligation. Subsequently, when the DNA ends are correctly juxtaposed (), the DNA-PK molecules will autophosphorylate each other. This induces a change in the dynamic association/dissociation rate that, in effect, liberates the DNA ends. It is not completely clear how we should envision this conformational change, but our FRAP data suggests that the exchange of DNA-bound DNA-PK with free DNA-PK is (relatively) rapid in a setting that enables DNA-PK phosphorylation/autophosphorylation. The window of time between dissociation of a bound molecule and reassociation of a new molecule may not differ between the unphosphorylated state and the phosphorylated state, but an exchange event will take place more frequently in the phosphorylated state. Hence, the DNA ends are more exposed to other repair factors in the latter case. In view of the aforementioned model, it is interesting to note that the kinase-impaired mutant of DNA-PK is phosphorylated at the Thr2609 and Thr2647 residues () and still displays an exchange rate that is lower than that of the 7A mutant, in which none of those residues are phosphorylated (). This discrepancy can be explained by taking into account that DNA-PK has several autophosphorylation sites that are functional in the 7A mutant, but that will not be phosphorylated in KD cells. For example, we did not mutate the Ser3205 residue that has previously been identified as an autophosphorylation site (). This site can still be phosphorylated in the 7A mutant, but not in the KD mutant. It is reasonable to assume that other DNA-PK–mediated processes are important for NHEJ as well. As a consequence, impairment of the DNA-PK kinase activity or clustered phosphorylation may influence the NHEJ process in other ways than merely by influencing the stability of the DNA-PK–DNA complex. Recently, it has been shown that autophosphorylation is important for the activation of the processing factor Artemis (Goodarzi et al., 2006). Inhibition of Artemis activation could lead to incorrect or absent processing of the DNA ends, which in turn may affect ligation efficiency. In addition, the DNA-PK kinase activity has been demonstrated to play a role in the phosphorylation of XRCC4. Alternatively, DNA-PK (like ATM) may be required for downstream signaling after the initial detection of DNA damage. All of these possibilities require further investigation. In conclusion, we demonstrate that DNA-PK phosphorylation/autophosphorylation facilitates NHEJ by destabilizing the DNA–DNA-PK complex, which, in turn, enables efficient ligation. This is a comprehensive explanation for the reduced repair efficiency that is observed in DNA-PK autophosphorylation mutants, although it is possible that other factors contribute to this phenotype. CHO cell lines AA8 (WT), V3 cells (defective in DNA-PKcs expression; ), and Xrs6 cells (defective in Ku80 expression; ) were maintained at 37°C in α-minimum Eagle's medium with 10% FCS, 100 units/ml penicillin, and 100 μg/ml streptomycin (HyClone). Stable cell lines expressing YFP-tagged DNA-PK or YFP-tagged Ku80 were maintained with 400 μg/ml of G418. Fluorescent immunostaining was performed as previously described (). In brief, the irradiated cells were fixed with cold methanol for 20 min on ice and permeabilized for 10 min in PBS containing 0.5% Triton X-100. After blocking with PBS containing 5% BSA for 20 min, the cells were subjected to incubation with the primary antibodies. Anti-pS2056 and -pT2647 polyclonal antibodies and anti-pT2609 and -γH2AX monoclonal antibodies were generated as described previously (; ; ). Anti-γH2AX polyclonal antibody was purchased from Cell Signaling Technology. Secondary antibodies (anti–mouse or–rabbit conjugated with Alexa Fluor 488/350 or rhodamine) were purchased from Invitrogen. DNA ends in irradiated cells were labeled with TAMRA-dCTP (Roche) according to the manufacturer's specifications. DSBs were introduced in the nuclei of cultured cells by microirradiation with a pulsed nitrogen laser (Spectra-Physics; 365 nm, 10 Hz pulse) as previously described (, ). The laser system was directly coupled (Micropoint Ablation Laser System; Photonic Instruments, Inc.) to the epifluorescence path of the microscope (Axiovert 200M [Carl Zeiss MicroImaging, Inc.] for immunostaining imaging or time-lapse imaging and confocal LSM 510 Meta [Carl Zeiss MicroImaging, Inc.] for FRAP analysis) and focused through a Plan-Apochromat 63×/NA 1.40 oil immersion objective (Carl Zeiss MicroImaging, Inc.). The output of the laser power was set at 60% of the maximum, which is the minimal dose required to induce detectable accumulation of YFP-tagged DNA-PK in living cells. Immunostaining and time-lapse images were taken with an AxioCam HRm (Carl Zeiss MicroImaging, Inc.). During microirradiation, imaging, or analysis, the cells were maintained at 37°C in 35-mm glass-bottom culture dishes (MatTek Cultureware). The growth medium was replaced by CO-independent medium (Invitrogen) before analysis. To estimate the number of DSBs that were induced by microirradiation, we compared the DNA-PK accumulation that was caused by our 365-nm laser system with the DNA-PK accumulation that was caused by a previously calibrated system (). used an 800-nm multiphoton laser to induce DNA damage. They estimated the biological impact of their microirradiation system by comparing the fraction of Ku80 that was immobilized after microirradiation with the fraction of Ku80 that was immobilized after exposure to gamma irradiation. From these data, they calculated their system to induce between 1,000 and 1,500 DSBs (for a given laser power output, target surface, and exposure duration). We microirradiated several YFP-DNA-PK–expressing V3 cells with either our 365-nm laser system or with the microirradiation system used by ; both systems are based on LSM 510 confocal microscopes, set to provide comparable background and fluorescence intensity levels). Using the same microirradiation system that was used by ; facility of the Erasmus Medical Center, Rotterdam, Netherlands) and following the same protocol, a small disk within the nucleus was irradiated with a pulsed 800-nm laser. Allowing for accumulation of YFP-DNA-PK, the average fluorescence in the disc was measured using the LSM 510 software and compared with the same quantity measured in nuclei that were exposed to our 365-nm laser. After subtraction of the average nuclear fluorescence intensity (undamaged area), we found the average fluorescence intensity at the 365-nm microirradiated region to be ∼7 times higher than at the 800-nm microirradiated region (223 arbitrary units for the 800-nm laser versus 1,576 arbitrary units for the 365-nm laser). We subsequently measured the dimensions of the accumulation areas and found those to be 1.7 μm for the 365-nm spot and 4.9 μm for the 800-nm spot (measuring slices of identical thickness in the z direction), which is a 2.8-fold difference. Taking into account these differences in the accumulation areas produced by the laser-systems, we estimated that the total fluorescence intensity of the accumulated YFP-DNA-PK was ∼2.5-fold (7/2.8) higher after microirradiation with the 365-nm laser than after irradiation with the 800-nm laser. From this, we infer that our microirradiation protocol induces between 2,500 and 3,700 DSBs (2.5 × 1,000 and 2.5 × 1,500, respectively) at the beam-focus of our pulsed 365-nm laser. V3 cells, expressing YFP-tagged DNA-PK, were irradiated with uranium particles with an initial energy of 11.4 MeV/nucleon (energy on the target cells, 3.8 MeV/nucleon) at the UNILAC facility at GSI, as previously described (). Cells were cultured in a Petri dish on 40-μm-thick polycarbonate sheets. Duration of irradiation was <5 ms per sample. Kinetic measurements of the recruitment of YFP-DNA-PK at sites of ion traversals were performed using a beamline microscope () equipped with a 7190–53 EBCCD-camera (Hamamatsu). Uranium ion tracks were visualized by etching polycarbonate, as described previously (). During microirradiation and time-lapse imaging, cells were maintained in CO-independent medium (Invitrogen) at 37°C. Images were made with an Axiovert 200M microscope, equipped with an AxioCam HRm, using a Plan-Apochromat 63×/NA 1.40 oil immersion objective. Time-lapse image acquisition was started before laser microirradiation to obtain an image of the unirradiated cell. DSBs were introduced in cell nuclei by microirradiation with our pulsed nitrogen laser (see Laser microirradiation and imaging) at the time of the third image. Exposure time was set at 400 ms, allowing 4 laser pulses to hit a defined region of a single nucleus. All images before and after irradiation were captured with a 400-ms exposure time. Signal intensities of accumulated YFP at the microirradiated site were converted into a numerical value by the use of the Carl Zeiss Axiovision software version 4.5. To compensate for nonspecific fluorescence bleaching during the repeated image acquisition, the fluorescence intensity of an undamaged region was subtracted from the fluorescence intensity of the accumulation spot for each cell at each time point. Relative fluorescence (RF) was calculated by the following formula: RF = (I − I)/(I − I), where I is the fluorescence intensity of the microirradiated region before irradiation, and I represents the maximum fluorescence signal in the microirradiated region. FRAP measurements were performed by photobleaching the entire DNA- PK accumulation spot that formed after introduction of DSBs and by measuring the subsequent recovery of fluorescence signal in that region. Imaging and FRAP measurements were performed on an LSM 510 Meta confocal microscope using a Plan-Apochromat 63×/NA 1.40 oil immersion objective. Images were acquired by scanning with the 514-nm line of the argon laser of the LSM 510 Meta microscope using the standard image acquisition function of the LSM 510 software version 4.0SP1. The tube current of the argon laser was set at 6.1 A, and the laser intensity was set at 0.3%. The pinhole was set at 2.51 Airy units, corresponding to an optical slice of 1.9 μm. DSBs were introduced in the nuclei of cultured cells by microirradiation with the pulsed nitrogen laser (see Laser microirradiation and imaging). Cells were cultured in 35-mm glass-bottom dishes. During measurements, cells were maintained in a CO-independent medium (Invitrogen) at 37°C. FRAP measurements were initiated 10 min after the introduction of DSBs, when maximum accumulation of DNA-PK was observed. A photobleaching pulse was given at 100% laser intensity (514 nm line of an argon laser, tube current set at 6.1 A). The dimensions of the circular FRAP area matched those of the accumulation spot (∼1.7 μm) and were kept constant during all measurements. An image of the nucleus was taken before photobleaching, immediately after photobleaching and at 30-s time intervals for 600 s. In every image, we measured the average fluorescence of the photobleached accumulation spot (F), from which the background fluorescence intensity was subtracted. Subsequently, F was normalized to the prebleach value: F = F (t) / F. Fig. S1 shows the cellular expression of WT, 7A, and KD DNA-PK, as verified by Western blot analysis. In addition, it demonstrates that complementing V3 cells with YFP-tagged DNA-PK reduces the radiation-sensitivity of the V3 cells to the level of AA8 cells. Fig. S2 shows the cellular expression of Ku80 in Ku80-complemented Xrs6 cells, as verified by Western blot analysis and in vivo immunostaining. It also demonstrates that complementing Xrs6 cells with Ku80 reduces the radiation-sensitivity of the Xrs6 cells to the level of AA8 cells. Finally, it demonstrates by Western blot analysis that phosphorylation of the Ser-2056 residue of DNA-PK is a Ku80-dependent process. Fig. S3 shows the cellular expression of WT, S2056A, and T2609A DNA-PK, as verified by Western blot analysis. The online version of this article is available at .

The defining event of mitosis occurs during anaphase, when identical sister chromatids disjoin and separate toward opposite poles of the microtubule (MT)-based spindle. Anaphase chromatid-to-pole motion (anaphase A) is tightly linked to depolymerization of the opposite ends of chromosome-associated MTs. Chromosomes actively depolymerize MTs at their plus ends, thereby “chewing” their way poleward along MT tracks—a type of motility termed Pacman. At the same time, chromosome-associated MTs serve as traction fibers, which are drawn poleward via persistent depolymerization at their minus ends. This process, termed poleward flux because of the resulting poleward flow (flux) of MTs, reels in attached chromatids to spindle poles (for review see ). Pacman and flux have been observed to occur simultaneously in diverse cell types and, in sum, account for the entire velocity of poleward chromosome motility (; ; , ; ; ). Thus, we refer to the general translocation mechanism underlying anaphase A as Pacman-flux. A large and growing set of proteins has been identified that bind to and modulate the functions of spindle MTs () and thus could be incorporated into the Pacman-flux machinery that drives anaphase A. Among the most mysterious of the spindle binding proteins identified to date are MT severing enzymes (; ). Proteins with the capacity to sever MTs have been found associated with spindles in a wide variety of systems, yet their specific mitotic functions remain largely unknown (; ; ; ). In particular, three conserved and closely related members of the AAA protein superfamily have been identified that may function by severing mitotic spindle MTs (). The best characterized of these is Katanin, a heterodimer consisting of a 60-kD AAA catalytic subunit (p60) and an 80-kD targeting and regulatory subunit (p80; ; ). Katanin has been found to target to centrosomes in diverse cell types (; ; ), leading to the proposal that it contributes to MT minus-end depolymerization and flux (; ; ). However, Katanin's role in flux or chromosome motility has not been previously demonstrated. In addition, a Katanin homologue in targets to meiotic chromosomes. Although this protein does not participate in mitosis, but instead is required for normal meiotic spindle assembly and dynamics (, ; ), it is conceivable that a similarly positioned MT severing protein in mitotic cells could contribute to Pacman-based chromosome motility. A second AAA family member, Spastin, has also been found to sever MTs in cells and in vitro (; ). Spastin has been studied primarily for its role in neuronal development and function. Mutations in the Spastin gene are the major cause of hereditary spastic paraplegia, a disorder caused by the degeneration of subsets of neurons and hallmarked by the progressive weakening of lower extremities (). Loss-of-function mutations of the Spastin homologue in also cause behavioral abnormalities and perturb neuromuscular junctions and axonal MT arrays (). Mitotic functions for the protein are unknown, but Spastin has been found to localize to centrosomes and spindle poles in vertebrate cells (; ), raising the possibility that it, too, could contribute to chromosome motility via the generation of poleward flux. A third AAA protein family member, Fidgetin, groups closely with Spastin and Katanin by phylogenetic analysis () and thus may sever MTs as well, though this has not been demonstrated experimentally. Phenotypic analyses of Fidgetin mutant mice indicate important developmental functions for this protein. Mutants display a head-shaking or “fidget” phenotype stemming from defects in auditory development. They also develop small eyes, which are a manifestation of a cell cycle delay, suggesting a potential mitotic role for this protein (). Fidgetin displays both cytoplasmic and nuclear localizations during interphase (), but its mitotic localization and function have not been reported. For this study, a series of live-cell analyses were performed to explore whether and how orthologues of Katanin, Spastin, and Fidgetin contribute to mitotic spindle and chromosome dynamics. Our findings reveal that all three are incorporated into the anaphase Pacman-flux machinery used to separate chromosomes. Surprisingly, the functions of these proteins are segregated so that Spastin and Fidgetin stimulate MT minus-end depolymerization and flux, whereas Katanin stimulates plus-end depolymerization and Pacman-based anaphase A. The genome encodes single orthologues of Spastin (Dm-Spastin; CG5977) and Fidgetin (Dm-Fidgetin; CG3326) and three potential Katanin p60s. The putative Katanin p60 orthologues include the protein product of CG10229 (most similar to human Katanin p60 and referred to here as Dm-Kat60) and the more divergent protein products of CG1193 and CG10793 (; ). This study reports on the mitotic functions of Dm-Spastin, Dm-Fidgetin, and Dm-Kat60 in S2 cells (we found that the CG1193 protein had no impact on mitosis, and analyses of CG10793 were not performed). S2 cells were used for these studies because of their ready susceptibility to targeted protein knockdown by double-stranded (ds) RNAi and their amenability to live-cell visualization of mitotic spindle and chromosome dynamics (; ; ). Only Dm-Spastin has been directly implicated as an MT severing enzyme in cells. Overexpression of Dm-Spastin in S2 cells was found to cause a substantial loss of interphase MTs, which is typical of increased MT severing activity (; ). Using this same assay, we found that Dm-Fidgetin and Dm- Kat60 similarly disrupted MT arrays when overexpressed in S2 cells, consistent with the hypothesis that all three proteins function as MT severing enzymes in cells (). #text S2 cells stably transfected with an EGFP–α-tubulin construct (under control of the copper-inducible pMT promoter or the constitutive pAc5.1 promoter [Invitrogen]) or with an EB1–EGFP construct (controlled by the pMT promoter) were a gift from R. Vale (University of California, San Francisco, San Francisco, CA). Cells were cultured according to published methods (). Sequences to be used for RNAi were selected by alignment of mRNAs to identify 500–600-bp regions for each protein that displayed minimal homology with other proteins in the FlyBase database. Selected sequences are as follows: Dm-Kat60 (CG10229), nucleotides 1885–2448 (in 3′ UTR of NM_080258); Dm-Spastin (CG5977), 2831–3389 (in 3′ UTR of NM_170115); Dm-Fidgetin (CG3326), 150–671 (in CDS of NM_134919); Asp (CG6875), 4207–4872 (in CDS of NM_079764). DNA templates for RNA synthesis were obtained by PCR of ESTs (Drosophila Genomics Research Center) or S2 cell cDNA using the primers listed in the following paragraph. dsRNA was generated using commercial transcription kits (Megascript T7 [Ambion] or Ribomax T7 [Promega]) according to the manufacturers' instructions. For RNAi, S2 cells were treated on day 0, 2, and 4 by incubating for 1 h in 1 ml serum-free Schneider cell medium (Invitrogen) with 20 μg dsRNA, followed by addition of 1 ml Schneider medium containing 20% heat-inactivated FBS. Cells were replated and analyzed on day 5. GST and maltose binding protein fusions of the N-terminal regions of each protein displayed in Fig. S1 were bacterially expressed and purified using glutathione–Sepharose or amylose resins. Polyclonal antibodies were generated against the GST fusions (ProteinTech). Antibodies were affinity purified from sera using their respective maltose binding protein fusion proteins coupled to Affigel resin (Bio-Rad Laboratories). In addition, affinity-purified antibodies were preabsorbed with resin-bound fusions of N-terminal regions of the other AAA proteins to eliminate cross-reactivity. After RNAi, S2 cells were plated on concanavalin A–coated coverslips to stimulate cell spreading for microscopy (). Cells were fixed in −20°C methanol for 20 min and blocked with 5% normal goat serum in PBS containing 0.1% Triton X-100. Primary antibodies (against the Dm-AAA proteins or α-tubulin [DM1a; Sigma-Aldrich], γ-tubulin [GTU-88; Sigma-Aldrich], phospho-histone H3 [Upstate Biotechnology], or centromere marker Cid [a gift from G. Karpen, University of California, Berkeley, Berkeley, CA]) were applied at 1–20 μg ml final concentrations in blocking buffer. Fluorescent secondary antibodies (Jackson ImmunoResearch Laboratories) were used at 7.5 μg ml. DNA was stained with 1 μg ml propidium iodide, 5 μM Draq5, or 0.3 μg ml Hoechst 33258. Specimens were imaged using an Ultraview spinning-disk confocal system (PerkinElmer) mounted on an inverted microscope (Eclipse TE 300; Nikon) with a 100×, 1.4 NA objective and captured with a digital camera (Orca ER; Hamamatsu). Most images are displayed as maximum intensity projections of the captured z stacks. For experiments requiring MT disruption, cells were treated with 30 μM colchicine for 16 h just before fixation. Two similar approaches were used to assay severing activity. For , S2 cells (constitutively expressing mRFP–α-tubulin) were transiently transfected with a copper-inducible gene encoding full-length Dm-AAA protein fused to EGFP. Transfected cells were induced with 500 μM CuSO for 8–12 h and then imaged with the system described. Images of live cells were captured digitally with identical system settings (exposure time, gain, etc.). Using ImageJ (NIH), fluorescence intensities for both EGFP and mRFP were measured from entire cells expressing Dm-AAA-EGFP and neighboring control cells not visibly expressing EGFP. For , wild-type S2 cells were transiently transfected with plasmids encoding EGFP or full-length Fidgetin-EGFP. After induction, cells were fixed (4% paraformaldehyde, 0.14% glutaraldehyde, 1 μM taxol, 0.1% Triton X-100, 1 mM MgCl, 1 mM EGTA, and 80 mM Pipes, pH 6.8; 15 min, 24°C) and immunostained with DM1a to visualize MTs, and digital micrographs were captured as described. Polymer fluorescence intensity was calculated by subtracting the mean cytosolic fluorescence intensity of a cell (calculated from measurements made in several cytoplasmic regions devoid of MTs) from its total mean fluorescence intensity (measured from the entire cell). To measure α-tubulin turnover at MT ends, rectangular regions at the pole (MT minus ends) and spindle equator (MT plus ends) of RNAi-treated S2 cells stably expressing EGFP–α-tubulin were photobleached using the confocal system described in the previous section, and time-lapse videos of the bleached cells were immediately recorded. To measure γ-tubulin turnover at centrosomes, the two large fluorescent spots of S2 cells stably expressing γ-tubulin–EGFP were photobleached and their subsequent recoveries recorded. The half-time for fluorescence recovery ( ) of each bleached region was measured from the plots of fluorescence recovery (corrected for postbleach fluorescence loss because of imaging). values were not included in calculating the mean for a treatment. Photobleaching was not observed to adversely affect cells; for example, photobleached cells were sometimes observed to proceed to anaphase. Datasets were saved as stacks of TIFF files, and time-lapse series were saved as AVI videos. Datasets were processed and analyzed with MetaMorph or ImageJ as described. When fluorescence intensities were to be quantitated (), the digital images were recorded with identical settings of microscope and Ultraview software. The statistical differences between treatments were analyzed using either a one-way nonparametric analysis of variance (Kruskal-Wallis) for multiple group comparisons or a nonparametric test (Mann-Whitney) for two group comparisons (SigmaStat, Systat Software, or GraphPad Prism; GraphPad Software). Measurement means were taken to be statistically different if P < 0.05. Supplemental figures show production of mono-specific antibodies against the AAA proteins (Fig. S1), verification of target AAA protein knockdown after RNAi (Fig. S2), immunolocalization of AAA proteins after colchicine treatment (Fig. S3), analysis of spindle phenotypes after AAA RNAi (Fig. S4), and representative fluorescence recovery curves of photobleached α-tubulin or γ-tubulin of RNAi-treated spindles (Fig. S5). Videos 1–4 are recordings of live EGFP–α-tubulin expressing, anaphase S2 cells after RNAi with control, Dm-Katanin, Dm-Spastin, or Dm-Fidgetin dsRNA, respectively. Videos 5–7 are recordings of live EGFP–α-tubulin expressing, anaphase S2 cells after RNAi to knock down control, Dm-Katanin, or Dm-Spastin, respectively. Online supplemental material is available at .

Cilia and flagella are elaborate organelles that play important roles in generating extracellular fluid currents in eukaryotes (; ; ). For example, left-right asymmetry in mammals is established by the nodal flow generated by the rotary movement of cilia (). The beating motion of cilia/flagella is driven by ciliary dyneins, which produce sliding forces between outer doublet microtubules (MTs) using ATP as an energy source (). Elucidation of the mechanisms underlying the mechanochemical energy conversion by the dyneins is of great importance. Since its discovery four decades ago (), the outer dynein arm (ODA) has been extensively studied by biochemical and structural methods. The ODA molecule is composed of three heavy chains and several smaller proteins (; ; ), and its total mass is ∼2 MD. Isolated ODAs have three heads connected to a common base through thin stems (; ; ; ). The head domain contains six tandemly linked AAA (ATPases associated with diverse cellular activities) modules, which form a ringlike structure (; ). Quick-freeze deep-etch (QFDE) EM studies of axonemes revealed that ODA has an elliptical head, which binds to the A tubule through two spherical feet (P foot and D foot), and a slender stalk that binds to the B tubule in an ATP-dependent manner (, ; ; ; ). It has been hypothesized that the N-terminal stem and/or the stalk domains serve as lever arms that amplify ATP hydrolysis–dependent conformational changes within the head domain (; ; ; ; ; ), but these hypotheses are based on studies of isolated dyneins in the absence of MTs. To elucidate the mechanism by which dynein generates force under physiological conditions, it is necessary to investigate the structural changes of MT-bound dyneins. Recently, cryoelectron tomography revealed the 3D architecture of the axoneme and showed how ODAs and other components are arranged in situ (); however, the tomograph only shows the structure in the rigor state. In addition, other components of the axoneme, such as the inner dynein arms, interdoublet links, and radial spokes, may affect the structural change of the ODA molecule. To elucidate mechanisms of power generation by ODA, it is therefore necessary to observe the nucleotide-dependent 3D conformational changes of ODA in the absence of other axonemal components. In this study, we established a method for reconstructing the 3D structure of in vitro–reconstituted ODA–MT complexes and visualized their nucleotide-dependent conformational changes by cryo-EM and image analysis. By comparing the relaxed and rigor states of the ODA–MT complex, we observed the conformational changes that occur in the head domains. These results suggest that our system has a high potential to reveal the mechanism of dynein power generation. The dynein–MT complex is formed by copolymerizing tubulin in the presence of axonemal dyneins from (). SDS-PAGE analysis of the complex suggests that it contains the heavy, intermediate, and light chains as well as the docking complex (; ). Because two MTs are cross-bridged by ODA, we refer to the complex as the ODA–cross-bridged MT (ODA–CB-MT) complex. QFDE-EM observation of the ODA–CB-MT complex revealed that, like axonemes, ODA has a longitudinal 24-nm periodicity (; ; ). The globular ODA molecules align along a MT () and extend stalks to the other MT (, arrows). When we observed the ODA–CB-MT complex by negative staining and cryo-EM, we found that two MTs twist over each other (; arrowheads). This twisting provides many views of the complex, allowing 3D reconstruction. The computed diffraction pattern of the cryo-EM image again shows a uniform 24-nm periodicity (). The distance between the two MTs of the ODA–CB-MT complex is 24 nm, which is the same as that observed by cryo-electron tomography of the axoneme (). The moiré pattern of the MTs shows that both of the MTs in the complex have the same polarity (). To reconstruct the 3D structure of the ODA–CB-MT complex, we first determined the helical arrangement of the complex from its cryo-EM images. In the layer line–filtered image (), the two rows of ODA molecules are staggered. This arrangement was also quantitatively confirmed by comparing the near- and far-side layer line at 24 nm. In principle, the phase difference between the two sides allows the determination of whether the left- and right-side ODAs is in phase or out of phase (180°; ). Judging from the phase difference at the amplitude peak (, arrowhead), ODAs are staggered as shown in . Therefore, we concluded that the complex has a twofold screw symmetry along its longitudinal axis. Using this helical arrangement, we reconstructed the 3D structure of the ODA–CB-MT complex ( and Video 1, available at ). Because the long twisting of the two MTs does not always follow the ideal helical symmetry, we combined single-particle analysis and helical image analysis to correct the distortions. We divided the images into 48-nm-long segments and calculated the rotation on the longitudinal axis from the distance between the two MTs. Translational alignment of the segments along the longitudinal axis was based on the 24-nm periodicity. Therefore, no assumption was made regarding the ODA structure. After 10–15 cycles of refinements, the 3D structures converged, and the effective resolution was determined to be 2.7 nm for the wild-type complex in the rigor state from the Fourier shell correlation (; ). Understanding the domain organization of ODA is essential for interpreting its structural changes. The end-on view of the ODA–CB-MT complex shows a pair of MTs cross-bridged by two rows of ODA molecules (). The 3D structure of the ODA–CB-MT complex in the rigor state is roughly divided into four domains: the α, β, and γ head domains and a base complex (). The stalk domain, which binds to MTs in a nucleotide-dependent manner (), should be located between the head domains and the B tubule. Although the stalk is visible in the quick-freeze replica (), it did not appear in the 3D reconstruction, possibly because of its flexibility and its thin structure. An ODA molecule binds to the A tubule through a thicker end, which we refer to as a base complex (; faint green densities). The base complex may contain two intermediate chains, which have WD repeat domains (; green bodies; ; ; ). To determine which side of the ODA–CB-MT complex harbors the ATP-sensitive MT-binding sites of ODA (), the complex was treated with 20 μM ATP and observed by cryo-EM (). Because ATP treatment disrupts the ATP-sensitive binding of the ODAs, in the presence of ATP, the cross-bridged MTs separate, whereas the ATP-insensitive binding remains intact (). also reported that ATP depolymerizes the MTs. To prevent this depolymerization, the MTs were stabilized by the addition of taxol. The cryo-EM image of this hemicomplex showed that the 24-nm periodicity of the ODAs was unaffected by ATP treatment. We compared the averaged cryo-EM image of the hemicomplex with different projections from the 3D reconstruction of the ODA–CB-MT complex. We found that the characteristic triangular shape of the ODA density is also seen in the projection of the ODA bound to the MT by its base complex side (). The similarity in the 2D projection was also supported by the high cross-correlation value between the cryo-EM image of the hemicomplex and the projection of ODA. These results demonstrate that the base complex side of ODA–MT binding is ATP insensitive, which is consistent with the findings by . To examine the motor activity of the ODAs in the ODA– CB-MT complex, we performed a motility assay using the ATP-treated ODA–MT complex. We attached the AlexaFluor543-labeled ODA–MT complex onto a glass coverslip (, red) and applied AlexaFluor488-labeled plain MTs (, green). MT translocation was initiated by the addition of 100 μM ATP. All of the sliding MTs translocated along the ODA–MT complex (, arrowheads; and Video 3, available at ). The sliding velocity by the ODA–MT complex was 4.0 ± 2.6 μm/s ( = 20), which is similar to that by the αβ particle of ODA (). This result demonstrates that the row of ODAs in the ODA– CB-MT complex has motor activity. The three heavy chains of ODA have different structures and functions (; ; , ; ). To assign α, β, and γ heavy chains, we compared the 3D structures between the wild type and the mutant (). The lack of an α heavy chain in appeared as a loss of the outermost density in the top view of the ODA–CB-MT complex (, arrowhead). Together with the thin-section EM study by , this allowed assignment of the α, β, and γ heavy chains from the outermost to innermost positions (). If viewed from the side, the planes of the head domain rings appear parallel to the longitudinal axis of the MT ( and ), which is consistent with recent observations by . The top view () and cross section () of our reconstruction also revealed that the planes of the head domain rings are not exactly parallel but rather tangential to the circumference of the B tubule ( and ). In addition, the three head domains are offset from each other longitudinally ( and ). The disk-shaped objects in the densities of the head domains () summarize the aforementioned interpretations. The positions and orientations of the disks were determined from the center of gravity and the long axis of the densities, respectively (). We correlated the surface structures observed by QFDE-EM with our 3D structure (; ). The elliptical head corresponds to the α head domain and a part of the β head domain. The boundary between the α and β head domains appears as a medial cleft within the head by QFDE-EM (see and in ). The D foot corresponds to the γ head domain and part of the β head domain, and the P foot corresponds to the base complex. The interdomain connections (; arrowheads) indicate that the three heavy chains are functionally distinct. The α head domain is connected to the base complex (, red arrowhead). The β head domain has two densities extending to the base complex: one density extends toward the base complex within the same ODA molecule (; orange arrowhead), and the other extends toward the adjacent base complex (; and ; arrowheads). ODAs are connected even in the absence of MTs (), suggesting that the inter-ODA connection is stable. The γ head domain is connected to the base complex (, yellow arrowhead) and also directly binds to the MT (, arrowhead). These observations suggest functional differentiation among the three head domains (see Discussion). To observe the ODA–CB-MT complex in the relaxed state, we analyzed it in the presence of ATP and vanadate ( and Video 2, available at ). It is known that the addition of ATP to vanadate-treated axonemes causes them to straighten and reduces their elastic bending resistance, which is referred to as the relaxation of axonemes (Sale and Gibbons, 1979; ). Treatment with ATP and vanadate also causes large structural changes of ODA (; ; ). To prepare the ODA–CB- MT complex in the relaxed state, we treated the complex in the rigor state with vanadate and ATP. The complex retained the cross-bridging, and the radius of the complex in the relaxed state did not change. Therefore, we applied the same image analysis methods as for the rigor state. The conformational changes are different among the three head domains ( and ). The density of the α head domain is unclear (), probably because of high heterogeneity. The center of gravity of the β head domain shifted 3.7 nm toward the B tubule (), and its head plane inclined 44° inward () compared with its conformation in the rigor state. In contrast, the γ head domain remained in almost the same position as in the rigor state (). As a result of the conformational change, the distance between the β head domain and the B tubule became shorter (; and ), from 14 nm in the rigor state to 10 nm in the relaxed state. Considering the reported length of the stalk domain (∼15 nm; , ; ), the change in the distance may affect the orientation of the stalk domain. Our reconstruction and the QFDE-EM image in the relaxed state are similar except for the D foot (). The D foot in the QFDE-EM image is apart from the P foot of the adjacent ODA. In our reconstruction, the D foot, which corresponds to the γ head domain and part of the β head domain, is close to the next P foot, which corresponds to the base complex. The difference in ATP concentration (1 mM for the QFDE-EM image and 2 μM for our reconstruction) may be the reason for the different appearance. This study establishes that the ODA–CB-MT complex is suitable for reconstructing the 3D structure of dynein from cryo-EM images. Our reconstruction revealed the detailed 3D arrangement of the ODA complex and the conformational changes between ODAs in two functionally critical states without the influence of other axonemal proteins. The elaborate interdomain linkers and nucleotide-induced conformational changes of the ODA indicate that the three head domains have different functions. The appearance of the α head domain both in the rigor and relaxed states suggests that it is more flexible than the other two head domains, which may be related to a proposed regulatory function (; ; ). The β heavy chain may be important for synchronized movement of ODAs along the axoneme by communication through the inter-ODA connections. Given that the swimming velocity of the α heavy chain–missing mutant is slightly slower than that of the wild type (), the nucleotide-induced movement of the β head domain presumably generates major force for the sliding. The γ head domain, which showed little displacement, may serve as a stable anchor to the A tubule. This displacement of the β head is a new observation, and it is distinct from the rotation of the head proposed by . Because their hypothesis is based on the negative-staining EM of isolated dynein in the absence of MTs, the movement of dynein relative to the MT cannot be observed; however, it is possible that both the rotation and the displacement occur at the same time and that they amplify the sliding speed. Because the β head domain displaced vertically in the direction of sliding, there should be a mechanism for converting the direction of the movement. The stalk domain is the most probable candidate for mediating this conversion because it bridges the head domain and the B tubule. One possible explanation for how the stalk domain converts the movement of the head domain is as follows: if the stalk length is constant, the angle of the stalk is a function of distance between the head domain and the MT. For example, if the stalk length is 15 nm, the 3.7-nm movement of the head domain results in a 40° rotation of the stalk, which can produce a 7-nm sliding movement at the stalk head. To investigate these possibilities, the orientation of the head domain and stalk must be determined by specific labeling (for example, by using antibodies). In the 3D reconstructions and in the aforementioned discussions, we assumed that most of the ODAs were in the relaxed state. This assumption is supported by our EM observations as well as previous biochemical data (described in this paragraph). When ODA and tubulin are copolymerized, there is more than one type of ODA–MT complex (). For structural analysis, we used the ODA–CB-MT complex, but it also contains ODAs complexed with single MTs as shown in . In this ODA–single MT complex, ODAs are bound to MTs by their stalks. All of these ODAs seem to dissociate from the MTs upon treatment with ATP and vanadate because we did not observe single MTs decorated by ODA via its stalk in the same grids, in which we collected cryo-EM images of the ODA–CB-MT complex (unpublished data). This demonstrates that stalk-side binding of most ODAs became weak after the ATP-vanadate treatment, which agrees with the finding of . It is also consistent with previous biochemical data: according to , 2 μM is the lowest concentration of ATP that can completely dissociate ODA from MTs within 1 s. Furthermore, according to kinetic studies of dynein, the rate constants of ATP binding, ATP hydrolysis, phosphate release, and vanadate binding are all faster than 8/s in our condition (; ; ; ). As we incubated the ODA–CB-MT complex with ATP-vanadate for 10–15 s, <1% of ODA should have remained in the rigor state. Although the occurrence of the cross-bridging even in the relaxed ODA–CB-MT complex is perplexing, we think that the cross-bridging of the two MTs is maintained by the sum of many weak ODA–MT interactions. The arrangement of ODAs in the ODA–CB-MT complex also raises the concern that the sliding directions induced by the two rows of ODA are antagonistic and should cause distortion of the complex. As we show in this study, cross-bridging of the ODA–CB-MT complex was retained in the presence of ATP and vanadate probably because ODA was in the presliding state and did not exert force in this nucleotide state. It seems that phosphate release from ODA in the ADP + Pi state is required to separate the cross-bridging, as shown in . These results are consistent with other biochemical data showing that ATP and vanadate inhibit the MT sliding (Sale and Gibbons, 1979; ; ) and that force generation is suggested to require the release of phosphate (; Johnson et al., 1984; ). Therefore, it is reasonable to assume that our relaxed structure represents the presliding state. In a recent study, cryo-electron tomography was applied to study axoneme structures, including ODA (). Although their overall architecture of ODA in the rigor state is quite similar to our reconstruction, there are minor differences in the assignment of the molecular boundary and of the inter-ODA linker to a particular heavy chain. did not show the position of the base complex relative to the head domains in the tomograph of the axoneme tomograph. In the present study, we assigned the base complex to the minus end side of the head domains (, green spherical objects). This assignment is based on the globular shape of ODA and the molecular boundary between adjacent molecules ( and and Video 1) and is consistent with the QFDE-EM study (; ). Future high-resolution studies on isolated ODA complexed with MTs may be needed to clarify the domain assignments. also interpreted that ODA molecules are connected along the MT by an outer outer dynein linker, which extends diagonally from the α head domain to the neighboring γ head domain (see in ). In our map, the inter-ODA linker, which is functionally equivalent to outer outer dynein, seems to connect the β head domain and the neighboring base complex. A similar inter-ODA linker was also observed by QFDE-EM (see in ). The difference between the interpretations of and our study could result from the loss of some axonemal proteins during salt extraction. For example, a cape structure of ODA in the axoneme (see in ) appears to be missing in the reconstituted ODA–MT complex (; see in ). This cape structure looks similar to the outer outer dynein linker observed by in that it extends diagonally from the α head domain. We anticipate that our new 3D reconstruction method will serve as a starting point for the more detailed analysis of the dynein power generation mechanism. The future combination of ODA mutants, antibody labeling, and our reconstruction method will reveal the complicated subunit organization and regulation within the ODA. Given the complexity of the axoneme, both bottom-up (e.g., this study) and top-down (studies of whole axonemes) approaches are essential. The 137c strain and the mutant strain, which lacks the α heavy chain and the 16-kD light chain, were provided by R. Kamiya (University of Tokyo, Tokyo, Japan) and S.M. King (University of Connecticut Health Center, Farmington, CT), respectively. Axonemes were obtained from wild type (137c) and the mutant by the dibucaine method (; ). The axoneme extract was desalted using a Sephadex G-25 gel filtration column equilibrated in PEM buffer (100 mM Pipes-NaOH, pH 6.8, 1 mM MgCl, and 1 mM EGTA). The eluent was concentrated to 2–3 mg/ml using Centricon YM-100 (Millipore). Bovine tubulin was purified as described previously (). MTs were polymerized in PEM buffer with 1 mM GTP in the presence of 1.3 mg/ml axonemal extract at 30°C for 1 h. The ODA–CB-MT complexes were purified by centrifugation through a 50% (wt/vol) sucrose cushion in PEM buffer at 35,000 for 30 min at 30°C. ODA–CB-MT was labeled with AlexaFluor543 (Invitrogen), and the molar ratio of unlabeled to labeled tubulin was 20:1. 100 μg/ml of the ODA–CB-MT complex in HMDE buffer (10 mM Hepes-NaOH, pH 7.2, 5 mM MgCl, 1 mM DTT, 1 mM EGTA, and 1 mM PMSF) containing 10 μM paclitaxel was treated with 20 μM ATP and attached onto the glass-bottom microwell dish (MatTek). Next, the glass surface was blocked with HMDE buffer containing 1 mg/ml casein. Plain MTs labeled with 10 μg/ml AlexaFluor488 in HMDE buffer containing 0.1% NP-40 were sheared with a 28-gauge needle and added to the dish. MT translocation was initiated by adding 100 μM ATP (final concentration). The movements of the MTs were observed using a total internal reflection fluorescence microscope (IX71; Olympus) equipped with plan Apo 100× NA 1.45 total internal reflection fluorescence microscopy oil immersion lens (Nikon). The images were projected onto a CCD camera (Cascade II; Photometrics) and were contrast enhanced with RS Image software (Roper Scientific). Experiments were performed at room temperature. For QFDE-EM, the MT pellet was resuspended in PEM buffer without sucrose and centrifuged again. Quick-freeze replicas of the MTs were prepared as described previously (). For negative-staining EM, samples were fixed in 0.5% glutaraldehyde and stained with 5% uranyl acetate. To prepare the ODA–CB-MT complex in the relaxed state, we treated the complex in the rigor state with 10 μM vanadate from the sucrose cushion step and added 2 μM ATP 10–15 s before plunge freezing. This concentration of ATP is shown to be enough for relaxing the axoneme (). ODA–CB-MT complexes that have MTs with 14 protofilaments/three-start helix were selected (), and their polarity was determined based on the moire pattern in the image (). Selected micrographs were digitized with a scanner (LeafScan 45; Scitex) at a pixel size of 10 μm, which corresponds to 2.5 Å in the specimen. Each filament was straightened by fitting the longitudinal axis to a cubic spline curve. The straightened filaments were divided into 48-nm-long segments. For the initial angle assignment for each segment, we assumed that the distance between the two MTs is constant and used a simple relationship between the diameter of the complex and the rotation angle around the axis: φ = cos(d/d), where d is the distance between the centers of the two MTs, and d is the maximum of the distance determined from two representative ODA–CB-MT complexes. The numbers of ODA molecules and filaments used for each of the reconstructions were as follows: 3,737 molecules from 31 filaments for the rigor state, 3,872 molecules from 38 filaments for the relaxed state, and 1,350 molecules from 11 filaments for the mutant. The effective resolution was determined to be 2.7 nm, 2.7 nm, and 3.5 nm for the rigor state, the relaxed state, and the mutant, respectively, from the Fourier shell correlation between two independent datasets using a 0.3 cut off. Image analysis was performed using Ruby-Helix scripts () and Frealign (). Surface renderings were performed with UCSF Chimera (), Pymol (Delano Scientific), or the AVS software package (Advanced Visual Systems). Videos 1 and 2 show 3D reconstructions of the ODA–CB-MT complex in the rigor state (Video 1) and relaxed state (Video 2) rotating on the long axis of the complex. Video 3 shows sliding of the MT by the ODA–MT complex. Online supplemental material is available at .

The extracellular signal–regulated kinase (ERK) 5 is a member of the MAPKs (; ). The N-terminal kinase domain of ERK5 is highly homologous to the prototypical MAPKs ERK1/2. However, ERK5 is selectively activated by its upstream kinase MEK5 () and ERK5-null mice are embryonic lethal (), suggesting that ERK5 has unique biological functions that cannot be compensated for by ERK1/2. Stimulation of ERK5 regulates neuronal survival, muscle cell differentiation, and cellular proliferation and transformation (; ; ; ; ). Hyperactivity of the ERK5 pathway is associated with highly aggressive forms of breast and prostate cancers (; ). Mechanisms for ERK5 regulation of proliferation are not well understood. However, several targets of ERK5 have been identified, including nuclear factor κB (NFκB), c-myc, and cyclin D, all of which are potential regulators of proliferation (). NFκB is of particular interest because it regulates the G1–S transition of the cell cycle and is required for oncogenic transformation by Ras (; ). Mutations that result in constitutive activation of NFκB are common in epithelial tumors, tumor cell lines, and lymphoid malignancies. These NFκB mutations cause increased proliferation rates and metastatic capacity (). Interestingly, ERK5-mediated activation of NFκB promotes cellular transformation in NIH 3T3 cells (). To elucidate mechanisms for ERK5 regulation of cellular proliferation, we investigated the importance of ERK5 in cell cycle progression. We examined ERK5 activity at different stages of the cell cycle and discovered that ERK5 is activated in a cell cycle–dependent manner, with maximal activation at G2–M. Evidence is presented supporting a critical role for ERK5 in the G2–M progression. Our data also identify NFκB-mediated transcription as a key downstream mechanism by which ERK5 regulates the G2–M phase transition. Although ERK5 plays a pivotal role in growth factor–induced cellular proliferation (), its role in cell cycle regulation is unclear. To determine whether ERK5 is activated at specific stages of the cell cycle, Western analysis using an anti-ERK5 antibody was performed on HeLa cells arrested at different stages of the cell cycle. The cell cycle stage for each treatment was confirmed by flow cytometry analysis (FACS; and not depicted). Cell cycle arrest at M phase was further confirmed by Western analysis and immunostaining using an antibody that recognizes phosphorylated mitosis-specific marker proteins (p-MPM-2; ). Treatment of cells with the microtubule destabilizing agent nocodazole, which arrests cells at the start of the M phase, caused a reduced electrophoretic mobility (phosphorylation shift) of ERK5 (), indicative of ERK5 activation (). In contrast to M phase arrest, there was very little activation of ERK5 in asynchronized cells or when cells were arrested at G1, S, or the G1–S boundary of the cell cycle (). Nocodazole activation of ERK5 was apparently due to the arrest of cells at M phase. Arrest of HeLa cells at M phase with taxol also caused ERK5 activation (unpublished data). In addition, we treated primary cultured cortical neurons from newborn rats with nocodazole. In contrast to HeLa cells, nocodazole did not induce ERK5 phosphorylation in these postmitotic cortical neurons, although these cells express abundant ERK5 that is readily activated by brain-derived neurotrophic factor (). Moreover, we collected mitotic HeLa cells by shaking them off the culture dish 9 h after thymidine release (mitotic; ). ERK5 was phosphorylated in this highly enriched mitotic cell population but not in nonmitotic control cells. Finally, ERK5 activation in mitotic and in nocodazole-arrested M phase HeLa cells was confirmed by a direct kinase assay (). MEK5, the only known kinase that phosphorylates and activates ERK5 (), was also activated in mitotic and nocodazole-treated HeLa cells (). We also measured ERK5 activation as S phase–synchronized HeLa cells progressed to M phase. When HeLa cells were released from a single-thymidine block that synchronizes cells at early S phase, ERK5 phosphorylation was detectable 6 h after thymidine release. At this time, there was a small increase in the number of cells harboring 4n DNA but very few mitotic cells (). Mitotic cells were identified by both positive immunostaining of p-MPM-2 and the condensed nuclear morphology typical of mitotic cells (). ERK5 phosphorylation peaked 8 h after release from a single-thymidine block, when the majority of cells (>84%) had 4n DNA content, but only 5% of the cells were mitotic. Similar observations were made when HeLa cells were synchronized with a double-thymidine block (). ERK5 phosphorylation was readily detectable 9 h after thymidine release, a time when the majority of cells (>86%) were in G2–M phases harboring 4n DNA, whereas only 17% of the cells were p-MPM-2 mitotic cells. We conclude that ERK5 is activated at both G2 and M phases. The temporal profile of ERK5 activation suggests that it might play a role in G2–M progression. To test this hypothesis, we expressed various constructs that either activate or inhibit ERK5 signaling and evaluated their effects on mitosis. Expression of constitutive-active (ca) MEK5 with wild-type (wt) ERK5 in HeLa cells, which activates ERK5 in transfected cells, increased the number of cells that are positive for the mitotic markers p-MPM-2 and p-histone H3 (). These cells also displayed the typical mitotic nuclear morphology (unpublished data). This suggests that ectopic activation of ERK5 signaling is sufficient to increase mitotic index in an asynchronous population. In a separate set of experiments, HeLa cells were transfected as in and treated 24 h after transfection with thymidine to synchronize cells at S phase. The mitotic index in the transfected cell population was determined 12 h after thymidine release. Constitutive activation of ERK5 significantly increased the mitotic index (), suggesting that more of these cells progressed from S phase into the mitotic phase. In contrast, transient expression of dominant-negative (dn) ERK5, which blocks ERK5 signaling, reduced the mitotic index. A similar experiment was performed using siRNA to suppress ERK5 expression and inhibit ERK5 signaling (). Transfection of the ERK5 siRNA greatly reduced ERK5 expression 48 h later (, top). When HeLa cells were synchronized by thymidine treatment 24 h after transfection, transfection of ERK5 siRNA significantly reduced mitotic index (, bottom). In the experiments described in , HeLa cells were synchronized at early S phase with thymidine 1 d after transfection, well before a considerable amount of transfected ERK5 was expressed (unpublished data) or siRNA reduction of endogenous ERK5 protein occurred. Furthermore, ERK5 activation at G1 and S phases was almost undetectable in HeLa cells. Therefore, data in (B and C) suggest that ERK5 signaling may regulate mitotic entry and/or progression. To strengthen this conclusion, we designed an experiment to block ERK5 activity from G2 to M phases onward to avoid any potential interference with the G1 and S phases. We took advantage of the fact that adenoviral-mediated gene expression, monitored by GFP expression and Western analysis, occurs 6 h after viral infection (). HeLa cells were synchronized in early S phase by a single-thymidine treatment. 2 h after release from thymidine, cells were infected with an adenovirus that expresses both GFP and dnMEK5 (). This allowed expression of dnMEK5 starting from G2 and continuing onward. The dnMEK5 used in this experiment inhibits the signaling of ERK5 but not other related MAPKs (; ; ). Expression of the adenoviral dnMEK5 significantly reduced mitotic index in adenovirus-infected cells 12 h after thymidine release (). Together, data in suggest that ERK5 regulates G2–M progression. Activation of ERK5 regulates several downstream targets, including NFκB (), a transcription factor implicated in cellular proliferation. To determine whether ERK5 activation at G2–M leads to NFκB activation, we transfected HEK293 cells with an NFκB-luciferase reporter to monitor NFκB activity. Cells were also cotransfected with dnERK5 to block endogenous ERK5 signaling. We used HEK293 cells to measure NFκB activities because the basal NFκB activity in HeLa cells is very high (unpublished data). As in HeLa cells, ERK5 phosphorylation was mainly detected in nocodazole-arrested M phase HEK293 cells (unpublished data). The NFκB reporter gene was activated in HEK293 cells after nocodazole treatment or when cells were released from a single-thymidine block for 12 h (). Importantly, expression of dnERK5 significantly reduced NFκB-luciferase activity under both conditions. Although the NFκB reporter was only activated 150% 12 h after thymidine release, this stimulation was statistically significant and consistent with the mitotic index of 22% in thymidine-released cells compared with 68% in nocodazole-treated cells. In addition, NFκB-luciferase reporter activities were increased in mitotic cells compared with nonmitotic control cells after the mitotic shake off (). Similar experiments were performed using an NFκB DNA binding ELISA assay. As a positive control, HEK293 cells were transfected with caMEK5 + wtERK5 to activate ERK5. NFκB DNA binding was activated in cells expressing caMEK5 with ERK5 and in cells treated with nocodazole for 12 h (). Expression of dnERK5 significantly inhibited NFκB DNA binding activities afforded by caMEK5 expression or nocodazole treatment. Finally, nocodazole treatment also increased the DNA binding activity of NFκB measured by a conventional gel shift assay, and the expression of dnMEK5 prevented this increase (). These data suggest that stimulation of ERK5 at G2–M causes activation of NFκB. In resting cells, the p50/p65 heterodimer of NFκB is normally sequestered in the cytoplasm by the inhibitor protein, IκB (). Upon stimulation, IκB is phosphorylated on serine-32 and serine-36 residues. This targets IκB for proteasome-mediated degradation, thereby releasing NFκB and allowing its nuclear translocation and DNA binding (). Data in demonstrated that ectopic ERK5 activation or M phase arrest by nocodazole treatment increases NFκB binding to DNA, suggesting that NFκB activation may be due to IκB degradation. Indeed, IκB protein levels were reduced in nocodazole-treated HeLa cells, and its degradation was partially blocked by dnERK5 (). This suggests that ERK5 regulates the degradation of IκB during G2–M phases of the cell cycle. Although ERK5 activity is required for NFκB activation, ERK5 does not directly bind to or modulate either IκB or NFκB (unpublished data). This suggests that another kinase may be acting as a mediator between ERK5 and IκB. The most studied IκB kinases are the IKK family of enzymes (). We monitored IKK activation at G2 and M phases by Western analysis using an antibody that recognizes phosphorylated and activated IKK (p-IKK). IKK was phosphorylated 3–6 h after release from double-thymidine block (), a time when the majority of cells were in S phase and ERK5 was not activated. IKK phosphorylation in S phase is consistent with a role for NFκB in G1–S control. However, IKK was not phosphorylated 9–12 h after thymidine release, when the majority of cells were in G2–M and ERK5 was maximally activated. Furthermore, although ERK5 was activated in a cell population enriched with mitotic cells (), IKK phosphorylation was not detectable in this same preparation (). These data indicate that IKK does not mediate ERK5 activation of NFκB at G2–M phases of the cell cycle. RSK1 and RSK2 are downstream targets of the ERK1/2 pathway that can directly phosphorylate IκB, thereby targeting IκB for degradation (; ). It has been reported that ERK5 activates RSK2 in NIH 3T3 cells and in neurons (; ). To determine whether RSK1 or RSK2 is activated by ERK5 in our system, HEK293 cells were transfected with caMEK1, which is sufficient to activate endogenous ERK1/2, or with caMEK5 + wtERK5 to activate ERK5 signaling. Constitutive activation of the ERK1/2 pathways increased the activity of both RSK1 and RSK2, whereas constitutive activation of ERK5 only activated RSK2 (). These data suggest that RSK2, but not RSK1, may mediate ERK5 activation of NFκB. To determine whether RSK2 is activated at G2 and M phases, the kinase activity of endogenous RSK2 was measured in HeLa cells synchronized by double-thymidine block. Like ERK5, RSK2 was activated 9 h after thymidine release, when the majority of cells were at G2–M (). RSK2 was also activated in the cell population enriched with mitotic cells (unpublished data). Furthermore, RSK2 colocalized with ERK5 in nocodazole-treated HeLa cells (). In control cells, both RSK2 and ERK5 were localized in the cytosol. Nocodazole treatment for 6 h induced nuclear translocation of both proteins in some cells. Moreover, immunoprecipitation of endogenous RSK2 isolated from nocodazole-treated HeLa cells pulled down both phospho-ERK5 and nonphospho-ERK5 protein (), suggesting a physical interaction between ERK5 and RSK2. This is consistent with a recent report documenting an interaction between ERK5 and RSK2 (). The kinase activity of endogenous RSK2 toward IκB was directly measured by an in vitro kinase assay. RSK2 phosphorylation of IκB protein increased 2- or 1.6-fold in HeLa cells expressing caMEK5 + ERK5 or in nocodazole-treated HeLa cells, respectively (). Expression of dnERK5 reduced basal RSK2 kinase activity toward IκB by 60%. Furthermore, expression of dnRSK2 abrogated NFκB activation induced by nocodazole or by expression of caMEK5 + ERK5 (). To test the functional consequence of RSK2 activation by ERK5 in G2–M regulation, HeLa cells were transfected and treated as in . A dnRSK2 was transfected to inhibit RSK2 signaling. As shown in , ectopic activation of ERK5 increased the mitotic index 12 h after thymidine release. This increase was blocked by coexpression of dnRSK2 (). Together, these data suggest that RSK2 is activated downstream from ERK5 at G2–M phases of the cell cycle and that ERK5 activates NFκB via RSK2-dependent IκB phosphorylation and degradation. The specific NFκB inhibitors SN50 and helenalin were used to determine whether NFκB-dependent transcription is necessary for G2–M progression. As a positive control, we confirmed that SN50 and helenalin block NFκB-stimulated transcription after TNFα treatment (unpublished data). SN50 and helenalin were added to HeLa cells 8 h after release from a single-thymidine block () or 6 h after a double-thymidine block () to interfere with the G2–M, but not G1–S, function of NFκB. Both SN50 and helenalin greatly decreased the mitotic index 12 h after release from a single-thymidine block (). We determined whether there is a critical time when blocking NFκB signaling inhibits mitosis. SN50 was added 8 or 10 h after release from a single-thymidine block, times corresponding to the beginning and midpeak of appearance of mitotic cells, respectively. Interestingly, addition of SN50 8 h, but not 10 h, after thymidine release reduced the mitotic index (). Similar results were obtained when actinomycin D was used to block general transcription (unpublished data). These data suggest that NFκB-mediated transcription regulates G2–M progression. The kinetics for the effect of helenalin on mitosis were also examined. HeLa cells were synchronized with a double-thymidine block and treated with helenalin 6 h after thymidine release to inhibit NFκB from G2 phase onward. Helenalin did not affect the rate of appearance of cells harboring 4n DNA 9 h after thymidine release (), suggesting that DNA synthesis and S phase completion were unaffected. However, there was a delay in the reappearance of 2n DNA–containing cells (G0–G1) and in the disappearance of 4n DNA–containing cells (G2–M) 9–14 h after thymidine release, suggesting that cells were staying at G2–M longer. This could have resulted from a delay in mitotic entry, meaning cells spent more time in G2, or a delay in mitotic exit in which cells stayed in M phase longer. Because the majority of the cells still harbored 4n DNA and were in G2 or M phases of the cell cycle 10 h after release from a single-thymidine block (), the fact that treatment with SN50 at this time did not affect the mitotic index () favors the former explanation. To more clearly distinguish between these two possibilities, taxol or nocodazole was included to block mitotic exit and trap mitotic cells so any defects in mitotic entry can be detected unambiguously. Helenalin treatment in the presence of taxol or nocodazole delayed accumulation of mitotic cells (), suggesting that NFκB activity is required for mitotic entry. In addition to the pharmacological inhibitors, we used an adenovirus encoding a nondegradable IκB super repressor (SR) mutant protein to specifically block NFκB activation (). When infected 2 h after release from a single-thymidine block, the adenoviral IκB SR was abundantly expressed 6 h later, when cells enter mitosis (). Expression of IκB SR from G2 phase onward delayed the appearance of mitotic cells after HeLa cells were released from a double-thymidine block (). This reduction in mitotic index persisted when mitotic exit was blocked by taxol. Similar to treatment with helenalin, expression of IκB SR did not affect the rate of initial appearance of cells harboring 4n DNA between 6 and 9 h (), suggesting that DNA synthesis and S phase completion were unaffected. However, the disappearance of 4n DNA–containing cells and reappearance of 2n DNA–containing cells slowed down. Together, data in demonstrate that blocking NFκB signaling delays mitotic entry. To establish a direct link between ERK5 and NFκB at G2–M progression, HeLa cells were transfected with caMEK5 + wtERK5. 1 d later, cells were synchronized at S phase by a single-thymidine block. 2 h after thymidine release, cells were infected with the IκB SR adenovirus to selectively block NFκB signaling at G2–M. Expression of this IκB SR at G2–M greatly reduced the mitotic index (). Similar results were obtained when SN50 or helenalin was used to inhibit NFκB signaling at G2–M (). This supports the hypothesis that ERK5 activation of NFκB is critical for G2–M progression during the cell cycle. To identify target genes of the ERK5–NFκB pathway that regulate G2–M transition, quantitative RT-PCR was performed to investigate the effect of blocking NFκB on the transcription of several genes known to be critical for mitotic entry. These include cyclin B1 and B2 (), polo-like kinase-1 (Plk-1; ), and protein phosphatase cdc25B (). HeLa cells were synchronized by a double-thymidine block and then infected with control or IκB SR adenovirus at the time of thymidine release to inhibit NFκB from G2 phase onward. The transcripts of cyclin B1, cyclin B2, Plk-1, and cdc25B were increased 9 h after thymidine release (). Up-regulation of these transcripts was inhibited by IκB SR. The incomplete suppression of transcription suggests that other factors may also contribute to the transcriptional regulation of these genes. We also transfected cells with a cyclin B1-CAT reporter construct to monitor transcription initiated from the cyclin B1 promoter. Nocodazole treatment stimulated cyclin B1-CAT activity, which is inhibited by coexpression of dnERK5 (). Collectively, these data suggest that the ERK5–NFκB signaling pathway regulates expression of several genes key to the control of G2–M transition. We used human artery smooth muscle cells (hSMCs) and human foreskin fibroblast (HFF) cells to investigate whether the ERK5–NFκB signal transduction system is also required for G2–M progression in nonimmortalized, nontransformed, primary human cells. ERK5 was activated in M phase–arrested hSMCs and HFF cells, an activation manifested as a phosphorylation shift of ERK5 (p-ERK5) in nocodazole-treated cells (M; ). Significantly, expression of adenoviral dnMEK5 or IκB SR reduced the mitotic index in hSMC (). Inhibition of NFκB activity by SN50 or helenalin significantly reduced mitotic index in HFF cells (). These results implicate a role for the ERK5–NFκB signal pathway in G2–M progression during the cell cycle of primary human cells. Although regulation of the G1 phase has been extensively investigated, much less is known about regulation of the G2–M phase transition. Besides a requirement for cyclin B and cdc2 activation, few other genes or signaling molecules have been identified for the control of the G2–M transition. Here, we discovered that ERK5 is activated at G2–M and is critical for the G2–M transition and timely mitotic entry. This function requires ERK5 activation of NFκB through RSK2. Furthermore, NFκB regulates the expression of several genes essential for mitosis, including cyclin B1, cyclin B2, Plk-1, and cdc25B. We also observed the activation and requirement of the ERK5–NFκB pathway for the G2–M progression in primary cultured human cells. These data suggest a novel function for the ERK5–NFκB pathway in regulation of the G2–M transition in both primary and transformed cells. To investigate the functional significance of ERK5 activation at G2–M, we performed several types of experiments to examine the effects of blocking or stimulating ERK5 activity on mitosis. Constitutive activation of the ERK5 pathway was sufficient to increase the mitotic index in an asynchronous cell population, whereas blocking ERK5 activity with dnERK5 or siRNA reduced mitotic index in S phase–synchronized HeLa cells. Most important, expression of adenoviral dnMEK5 from G2 onward, which eliminates potential interference of ERK5 signaling at G1 or S phase of the cell cycle, reduced mitotic index. These data are the first report of a causal relationship between ERK5 activation and G2–M progression. Although ERK5 had been implicated in cellular proliferation, the underlying mechanisms were not defined. It has been reported that ERK5 regulates the G1–S transition of the cell cycle; however, evidence supporting this hypothesis is controversial. For example, some reports showed that dnMEK5 inhibition of ERK5 in NIH 3T3 cells blocks EGF stimulation of thymidine incorporation () and ERK5 may activate cyclin D1 expression (). In contrast, although MEK5 knockout mice have a phenotype similar to the ERK5-null mice, mouse embryonic fibroblast cells derived from MEK5 mice do not exhibit a defect in G1–S (). Our data demonstrate that ERK5 phosphorylation is barely detectable in HeLa cells arrested at G1 or S phase when HeLa cells are released from a thymidine block. This is consistent with the report by arguing against a role for ERK5 in G1–S regulation. Thus, ERK5 may regulate cellular proliferation primarily through its action at the G2–M phases of the cell cycle. What are the downstream targets that mediate ERK5 regulation of the G2–M transition? Our data showed that NFκB is activated when cells were arrested at the start of M phase by nocodazole treatment, 12 h after thymidine release, or in mitotic shake-off cells. Furthermore, NFκB activation in G2–M requires ERK5 activity. Inhibition of NFκB starting from G2 phase onward, achieved by treatment with helenalin or SN50, or by expression of adenoviral IκB SR, reduced the mitotic index. The delay in the rate of appearance of mitotic cells was also observed when mitotic exit was blocked by taxol or nocodazole, suggesting a pivotal role for NFκB in timely mitotic entry. In addition, stimulation of mitosis in cells transfected with caMEK5 + ERK5 was suppressed by inhibiting NFκB from G2 onward, demonstrating a direct link between ERK5 and NFκB at G2–M. These data suggest a novel function for the ERK5–NFκB pathway in the regulation of G2–M transition. The transition from G2 to M phase requires activation of the cyclin B–cyclin-dependent kinase 1 (CDK1, also known as cdc2) complex (). The cyclic activation of this complex at G2–M is stimulated by dephosphorylation of CDK1 and increased expression of cyclin B. This process is regulated by several G2–M kinases and phosphatases, including cdc25 and Plk (; ). The activity and expression of cdc25 and Plk are also regulated in a cell cycle–dependent manner. Although it is clear that many of these G2–M regulators are controlled transcriptionally in a cell cycle–specific manner, mechanisms underlying their transcriptional regulation are not well defined. Recently, it was reported that the forkhead family of transcription factors, including FOXM1 and FoxO, play a critical role in the expression of genes important for mitotic entry and exit (; ). In this study, we discovered that blocking NFκB signaling at G2–M inhibits transcription of cyclin B1, cyclin B2, Plk-1, and cdc25, suggesting an important role for NFκB in the transcriptional regulation of these genes. These findings provide new insights concerning transcriptional mechanisms governing G2–M progression of the cell cycle. Although NFκB has been implicated in cell proliferation and many forms of human cancer, NFκB activity has only been implicated in regulation of the G1–S phases of the cell cycle through induction of the G1 cyclin, cyclin D (; ). Our study identifies a new function for NFκB in the regulation of G2–M transition. It has been reported that forced activation of ERK5 stimulates NFκB reporter gene expression; however, the mechanism by which ERK5 activates NFκB was not elucidated (). NFκB-mediated transcription can be stimulated by increased DNA binding or by an NFκB phosphorylation that enhances the recruitment of transcriptional coactivators (). However, we found that ERK5 does not directly stimulate Gal-4–NFκB–mediated transcription (unpublished data). Instead, IκB protein was degraded and NFκB DNA binding activity increased in M phase–arrested cells; both processes required ERK5 activity. These data support the hypothesis that NFκB activation at G2–M is mediated by ERK5 stimulation of IκB degradation and subsequent NFκB binding to DNA. Previous studies showed that NFκB activation by nocodazole requires IKK in some cells and that this promotes cell survival after mitotic cell cycle arrest (). Surprisingly, IKK was not active at G2–M, when HeLa cells were released from a thymidine block, or in a cell population enriched for mitotic cells. Thus, although IKK may counteract nocodazole-induced apoptosis by acting as an IκB kinase in some cells, it does not mediate ERK5 activation of NFκB during G2–M progression of the cell cycle. RSK1 and RSK2 both phosphorylate IκB (; ). However, ERK5 stimulated the activity of RSK2, but not RSK1. Furthermore, RSK2 was activated at G2–M and physically associated with ERK5. Constitutive activation of ERK5 promoted RSK2 phosphorylation of IκB, whereas blocking ERK5 reduced the kinase activity of RSK2 toward IκB. Moreover, RSK2 activity was required for NFκB activation induced by ERK5 or M phase arrest. Expression of dnRSK2 blocked mitotic entry of HeLa cells induced by ectopic ERK5 activation. These results suggest that RSK2 is necessary for ERK5-induced degradation of IκB at G2–M. However, RSK2 phosphorylates IκB on S32, but not S36, which is necessary but not sufficient for IκB degradation (; ). Thus, it is likely that, in addition to RSK2, ERK5-mediated phosphorylation of IκB may require another, unidentified kinase that mediates ERK5 phosphorylation of IκB at S36. We were unable to achieve complete inhibition of mitotic entry by blocking ERK5, RSK2, or NFκB, possibly because none of the experimental approaches completely blocked the signaling events, or there may be additional pathways involved. Furthermore, because inhibition of NFκB delays but does not completely halt mitotic progression (), it is possible that the ERK5–RSK2–NFκB signaling pathway plays a regulatory rather than obligatory role in the G2–M cell cycle control as exemplified by FoxM1 (). In summary, we propose that ERK5 is activated at G2–M and that it stimulates downstream kinases, including RSK2 kinase (). RSK2 phosphorylates IκB, targeting it for degradation, thereby releasing NFκB from IκB. Subsequently, NFκB translocates to the nucleus and activates expression of target genes that are required for the G2–M progression. Our studies identify an unexpected and novel role for the ERK5–NFκB pathway in G2–M regulation. HEK293 and HeLa cells were purchased from American Type Culture Collection. HEK293 cells were used for reporter gene assays and NFκB DNA binding assays because the basal NFκB activity is very high in HeLa cells. However, for flow cytometry and immunocytochemistry, HeLa cells were routinely used because HEK293 cells grow in clusters and clumps, making it difficult to analyze by these assays. Primary hSMCs were provided by B. Askari and K. Bornfeldt (University of Washington, Seattle, WA; ). HFF cells were provided by A. Minella and B. Clurman (Fred Hutchinson Cancer Research Center, Seattle, WA; ). A commercial siRNA to human ERK5 and control nonsilencing RNA (NS) were obtained from Ambion. The following plasmids were obtained from J.D. Lee (The Scripps Research Institute, La Jolla, CA): wt and dnERK5 and HA-tagged caMEK5 (). The flag-tagged dnRSK2 construct was a gift from S. Impey (Oregon Health Sciences University, Portland, OR). The cyclin B1-CAT (pB1-CAT) reporter plasmid was provided by G. Piaggio and C. Gaetano (CRS-IRE, Rome, Italy; ). The dnMEK5 adenovirus was a gift from R. Segal (Dana Farber Cancer Institute, Boston, MA). The NFκB-luciferase reporter and IκB SR adenovirus have been described previously (). The polyclonal antibody to ERK5 was generated in our laboratory (). The following antibodies were obtained commercially: anti-RSK2, anti–phospho-histone H3 (Ser10), and anti–phospho-MPM-2 (Upstate Biotechnology), anti–phospho-IκB Ser32 or Ser36 (Calbiochem), anti–phospho- IKK Ser180/181 and anti-IκB (Cell Signaling Technologies), anti-MEK5 antibody (Santa Cruz Biotechnology, Inc.), and anti-ERK1/2 antibody (Promega). Cell-permeable NFκB inhibitor SN50 and control peptides, helenalin, thymidine, nocodazole, taxol, and aphidicolin were purchased from Calbiochem. Hoechst 33342 was obtained from Sigma-Aldrich. Myelin basic protein (MBP) was purchased from Upstate Biotechnology. HeLa or HEK293 cells (10 cells/60-mm dish) were transiently transfected with plasmid DNA by Fugene 6 (Roche). siRNA was transiently transfected using Oligofectamine (Invitrogen). HEK293, HeLa, and HFF cells were arrested in different cell cycle phases as follows: serum withdrawal for 14 h followed by addition of serum for 1 h (G1), 1 μg/ml aphidicolin (G1–S; 12 h), 2 mM thymidine (S; 14–16 h), or 0.5 μg/ml nocodazole (M; 12 h). Cell cycle arrest was confirmed by flow cytometry analysis. Primary hSMCs were arrested using similar drug treatments for 14 h. Initially, cells were synchronized at early S phase by a single-thymidine block, in which cells were treated with 2 mM thymidine for 15 h, released from the block by washing cells in PBS three times, and grown in complete medium for the indicated times up to 12 h. In later experiments, a double-thymidine block was used to obtain better synchrony. Cells were treated with 2 mM thymidine for 16 h and released for 8 h, followed by an additional 16-h thymidine block and release. The cells were then washed three times with PBS and released into complete medium. To examine the effect of adenoviral gene expression or drug treatment on mitosis, NFκB inhibitors and adenoviruses were added at the indicated times after final thymidine release. HEK293 cells were transfected using Fugene 6 by the manufacturer's instructions. In brief, 0.15 ×10 cells were plated onto each well of a 24-well plate coated with poly--lysine (Sigma-Aldrich). After 24 h, cells were transfected with the appropriate reporter genes (0.36 μg/well NFκB-luciferase plus 0.05 μg/well EF1α-lacZ plasmid DNA). Where indicated, cells were cotransfected with 0.16 μg/well dnERK5. Cells were treated 48 h after transfection, and luciferase and β-galactosidase activities were measured. For the cyclin B1-CAT reporter assay, HEK293 cells were cotransfected with a plasmid encoding dnERK5, the reporter plasmid pB1-CAT, and an EF1α-lacZ plasmid as an internal control. Cell lysates were harvested 48 h after transfection, and CAT activity was assayed in whole-cell extracts as described previously (). The reporter gene luciferase or CAT activities were normalized to β-galactosidase activity and expressed as the fold induction relative to control. Nuclear fractions were prepared from cell lysates of HEK293 cells using the NucBuster Protein Extraction kit (Novagen) per the manufacturer's instructions. Nuclear NFκB DNA binding activity (the p50–p65 complex) was then quantitated by the NFκB Transcription Factor Assay ELISA kit (Chemicon International) per the manufacturer's instructions. The DNA binding activity of NFκB was also measured by an electrophoretic mobility shift assay (gel shift) as described previously (). Except where specified, mitotic cells were scored by nuclear condensation in conjunction with positive staining for p-MPM-2, a mitotic marker. In , p-histone H3 was used as another mitotic marker in addition to p-MPM-2. Alternatively, HeLa cells in G1 (2n DNA) or G2–M (4n DNA) were quantitated by flow cytometry analysis (FACS). In brief, cells were trypsinized, resuspended in ice-cold 70% ethanol, and fixed at 4°C for 30 min. Cells were then resuspended in propidium iodide staining buffer containing 200 μg/ml RNase and 2% FBS in 1× PBS and incubated at 4°C for 3 h. Cells were then resuspended in PBS/2% FBS with an 18-gauge needle and analyzed by FACScan analyzer (Becton Dickinson) equipped with a 518-nm laser. Cell lysates were prepared from HeLa cells as described previously (). Equal amounts of protein lysates (350 μg) were used for each kinase assay. The kinase activities for ERK5 or MEK5 were measured using immunoprecipitation-coupled in vitro kinase assays using recombinant GST-MEF2C or MBP as substrates, respectively (; ). To measure endogenous RSK2 activity, cell lysates were immunoprecipitated with a polyclonal anti-RSK2 antibody. In brief, 5 μg anti-RSK2 antibody was incubated with 50 μl protein A–Sepharose beads overnight at 4°C. The beads were washed once with lysis buffer and then incubated with cell lysates for 3 h at 4°C. RSK2 kinase activity in the immune precipitates was then quantitated by evaluating the incorporation of [P]ATP into either 0.1 mM CREBtide (Sigma-Aldrich; ) or 1 μg of a recombinant human IκB protein (BIOMOL Research Laboratories, Inc.). Total RNA was isolated using RNeasy Miniprep kit (QIAGEN). To remove trace genomic DNA, DNA-free (Ambion) treatment was performed on samples. Total RNA was quantitated on the Mx4000 Multiplex QPCR System (Stratagene) using the RiboGreen RNA Quantitation kit (Invitrogen). Quantitative RT-PCR was performed in a single reaction on an Mx4000 Multiplex QPCR System using 20 ng of total RNA. The RT-PCR reactions were performed in triplicates in a 20-μl reaction using SYBR Green PCR Master Mix (Applied Biosystems; 10 μl 2× Master Mix, 400 nM each primer, 5 U MultiScribe RT, and 8 U RNase inhibitor). The cycling conditions were 48°C for 30 min and 95°C for 10 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. After each assay, a dissociation curve was run to confirm specificity of all PCR amplicons. Resulting C values were converted to nanograms, normalized to total RNA, and expressed as the mean of triplicate samples ± SD. Pooled sample total RNA was used for standard curves as 1:2 serial dilutions. The standard curves showed reaction efficiencies as follows: hcdc25B, 102.6% and R = 0.999; hPLK1, 102.5% and R = 0.998; hCCNB1, 100.6% and R = 0.998; and hCCNB2, 101.3% and R = 0.998. PCR primers were designed using Primer Express 2.0 software (Applied Biosystems) and synthesized by Operon. The sequences were as follows: hcdc25B, forward, CCCTATGGACCCCCACATG, and reverse, ATGGCAAACTGCTCGTTTCG; hPlk1, forward, GAGCGTGACGGCACTGAGT, and reverse, AAGGAGGGTGATCTTCTTCATCAA; hcyclinB1, forward, GAAATGTACCCTCCAGAAATTGGT, and reverse, CCATCTGTCTGATTTGGTGCTTAG; hcyclinB2, forward, TGTCAACAAACAACTGAAACCTACTG, and reverse, CCTCAGGTGTGGGAGAAGGA. For , images were collected with an inverted fluorescence microscope (Leitz DMIRB; Leica) using a 40× objective lens (Leitz; Leica). MagnaFire digital microscope camera and MagnaFire software (Optronics, Inc.) were used for system control and image processing. For and , images were taken by a Marianas imaging system (Intelligent Imaging Innovations, Inc.) incorporating a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with an X,Y motorized stage, shuttered 175 W xenon lamp coupled with a liquid light guide, a digital camera (CoolSNAP HQ; Roper Scientific), and 40× objective lens (Axiovert; Carl Zeiss MicroImaging, Inc.). Slidebook software package was used for system control and image processing. Results were from three or more independent experiments. Data are presented as means ± SEM for all except in , in which error bars are SD. We used two-tailed test assuming equal variance for statistical analysis of the data. *, P < 0.05; **, P < 0.01; ***, P < 0.001.

Toll-like receptors (TLRs) are involved in the recognition and processing of a variety of signals delivered by viral and microbial products (; ). TLRs sense the presence of molecules that are broadly conserved across microbial taxa. TLR activation initiates the innate immune response by inducing the expression of antimicrobial genes and inflammatory cytokines. Activation of TLRs also enhances adaptive immunity through activation of dendritic cells. TLR-mediated recognition of microbial components by dendritic cells induces the expression of costimulatory molecules, such as CD80/CD86, and the secretion of inflammatory cytokines, and it is responsible for the rearrangement of trafficking pathways of class II major histocompatability complex (MHC) products (; ). There are 10 and 12 TLR paralogues in humans and mice, respectively. Both species have TLR1–9. Mice lack TLR10, but have TLR11–13, which humans lack. Each TLR appears to sense the presence of distinct microbial components (; ). For example, TLR4 recognizes lipopolysaccharides (LPSs), which are components of the Gram-negative bacterial outer membrane, whereas double-stranded RNA, single-stranded RNA, and unmethylated bacterial DNA (CpG) engage TLR3, 7, and 9, respectively (; ; ; ; ; ). Mutations affecting TLR-mediated cellular responses have been instrumental in delineating the components of the relevant signal transduction cascades (). These include spontaneous mutations, targeted gene disruptions, and chemically induced mutations, the best characterized of which include the deficiencies in MyD88 and TRIF adaptor molecules (; ) and in the kinases that act downstream of them (; ; ). In a forward genetic screen using mutagenesis with -ethyl--nitrosourea, identified “triple D” (3d) mice that showed defects in TLR3, 7, and 9 signaling, as well as in class I and II MHC-restricted antigen presentation. The mutation was identified as a single histidine-to-arginine substitution (H412R) in the polytopic membrane protein UNC93B. Mice carrying this mutation are highly susceptible to infection with mouse cytomegalovirus, and . Notably, the histidine residue affected in the UNC93B mutation is invariant for all vertebrate orthologues. UNC93B deficiency has also been linked to the etiology of herpes simplex virus-1 encephalitis in human patients (). Similar to what was observed in 3d mice, cells from patients with a mutation in the gene show impaired cytokine production upon stimulation of TLR3, 7, 8, and 9 and are highly susceptible to various viral infections. The molecular mechanisms that underlie the immunological defects in the UNC93B mutant mice and in human patients are not known. Moreover, no specific function has been assigned to UNC93B. In , , which is the founding member and sole homologue of mammalian , encodes a regulatory subunit of a two-pore potassium channel complex and plays a role in coordinated muscle contraction (; ; ). Multiple paralogues of UNC93 exist in mammals, two of which are UNC93A (32% amino acid identity to s UNC93) and UNC93B (20% amino acid identity to UNC93). UNC93A and UNC93B are highly conserved between human and mouse (71% amino acid identity for UNC93A and 90% amino acid identity for UNC93B). Human UNC93A and UNC93B are predicted multispanning transmembrane proteins and the GFP fusion protein of human UNC93A localizes to the plasma membrane (; ). Mammalian UNC93B contains a domain of unknown function (DUF895) between residues 124 and 189, and human UNC93B shows a weak homology to the bacterial ABC-2 type transporter signature between residues 319 and 523 (; ). However, no functional role has been established for such domains. We analyzed the biosynthesis and maturation of murine UNC93B and discovered that wild-type, but not mutant, UNC93B physically interacts with TLR3, 7, 9, and 13, as assessed by mass spectrometry (MS) and biochemical approaches. Using genetic and immunochemical tools, we further confirmed the interaction between endogenous UNC93B and TLRs in primary dendritic cells and splenocytes of wild-type mice, but not UNC93B mutant mice. The essential role for UNC93B in TLR signaling is thus explained by direct interactions with its client TLRs. The murine gene comprises 11 exons and gives rise to a protein of 598 amino acids. Topology prediction programs suggest that UNC93B spans the membrane 12 times, and the 3d mutation (H412R) is located within transmembrane domain 9 (). UNC93B has two putative N-linked glycosylation sites (consensus NxS/T), NHT and NKT (). We raised polyclonal rabbit antibodies against several peptide sequences of the N- and C-terminal portions of UNC93B. The antibodies showed reactivity with both wild-type and mutant UNC93B, as assessed by immunoblotting and immunoprecipitation on cell extracts prepared from a variety of sources (unpublished data). Polar residues in transmembrane domains are involved in helix–helix interactions within a multitransmembrane domain–containing protein to aid helix packing or participate in protein–protein interactions with a neighboring membrane protein (). Missense mutations involving the loss or gain of an arginine residue in a predicted transmembrane domain are often associated with protein misfolding and malfunction, as found in many human diseases (). Therefore, we assessed the effects of the H412R mutation on expression, maturation, and stability of the UNC93B protein. We generated an epitope-tagged version of UNC93B in which the Flag-Tobacco Etch virus (TEV)-HA tag was attached to the C terminus of either the wild-type (UNC93B-HA WT) or mutant (UNC93B-HA H412R) UNC93B protein (). We then introduced epitope-tagged wild-type and mutant UNC93B into the macrophage cell line RAW 264.7 and conducted pulse-chase experiments. Wild-type and mutant UNC93B-HA showed similar half-lives (∼4 h), but exhibited distinct migration patterns in SDS-PAGE (). Wild-type UNC93B migrates as heterodisperse material upon SDS-PAGE, whereas the UNC93B mutant form is dominated by a well-defined distinct polypeptide in addition to more diffuse material (). To rule out the possibility that the distinct polypeptide seen for mutant UNC93B represents a protein that associates with mutant UNC93B rather than the mutant protein itself, we performed immunoprecipitation experiments with the polyclonal antiserum directed against the C-terminal segment of UNC93B (anti–UNC-C), followed by reimmunoprecipitation experiments with an UNC93B antibody raised against the N terminus (anti–UNC-N), anti-HA, or anti-Flag antibodies. We recovered epitope-tagged, as well as endogenous, UNC93B from RAW macrophages with the anti–UNC-C antiserum (). Reimmunoprecipitation with anti–UNC-N antiserum after mild denaturation of the initial immunoprecipitation samples recovered endogenous, as well as epitope-tagged, UNC93B, whereas anti-HA and -Flag monoclonal antibodies recovered only epitope-tagged UNC93B proteins, as expected (). The distinctly migrating polypeptide of mutant UNC93B was also recovered by reimmunoprecipitation (), confirming that it derives from the UNC93B protein and not from a separate UNC93B-associated polypeptide. To examine the maturation of endogenous wild-type and mutant UNC93B proteins, we performed pulse-chase analysis of bone marrow–derived dendritic cells (BM-DCs) from wild-type (C57BL/6) and UNC93B mutant (3d) mice. We confirmed the 3d phenotype of the UNC93B mutant mice through measurement of TNF production in BM-DCs after stimulation with TLR agonists. As reported, cells from 3d mice showed a complete lack of response to TLR7 and 9 agonists, whereas the response to the TLR4 agonist LPS was not affected (Fig. S1, available at ). The cells were first labeled with [S]methionine/cysteine for 30 min. After chase periods of 0, 1, and 4 h, endogenous UNC93B was recovered by immunoprecipitation with the anti–UNC-C antiserum. The immunoprecipitates were subjected to glycosidase digestion. In agreement with the proposed intracellular location of UNC93B as an ER-localized protein (), we found that both wild-type and mutant UNC93B proteins recovered during the entire chase period retained full sensitivity to Endoglycosidase H (Endo H; ). Immunofluorescence analysis of wild-type and mutant UNC93B-HA with an anti- HA monoclonal antibody in paraformaldehyde-fixed RAW cells resulted in a reticular membrane staining pattern that closely overlaps with staining pattern for the ER membrane protein calnexin, confirming ER localization of both wild-type and mutant UNC93B (unpublished data). Notably, the endogenous UNC93B mutant protein recovered from 3d mice showed the same discrete polypeptide, in addition to the presence of diffuse material, whereas wild-type UNC93B only showed the heterodispersed band (). As observed for epitope-tagged UNC93B, endogenous UNC93B is a stable protein, and wild-type and mutant UNC93B do not show any difference in stability. These results further show that tagging UNC93B at its C terminus does not necessarily influence its maturation or stability. The 3d mice not only show a defect in TLR signaling via TLR3, 7, and 9 but are also compromised in their ability to engage in cross-presentation via class I MHC molecules and in class II MHC-restricted antigen presentation (). Even though surface levels of MHC products at steady state may not be affected by the 3d mutation, this leaves open the possibility of alterations in their trafficking. We performed pulse-chase analysis for class I and II MHC products on BM- DCs obtained from wild-type and 3d mice. Maturation of class II MHC molecules was examined by assessing the levels of SDS-stable, peptide-loaded αβ dimers, as well as the kinetics of SDS-stable dimer formation. The results from wild-type and 3d mice were indistinguishable (). We did not observe any difference in synthesis and maturation of class I MHC molecules either (unpublished data). We conclude that, at least at this level of analysis, defects in MHC-restricted antigen presentation in 3d mice are unlikely to result from aberrant trafficking of MHC products. To identify interacting proteins of UNC93B, we conducted pulse-chase experiments in RAW macrophages stably expressing epitope-tagged wild-type or mutant UNC93B-HA (). Cells were lysed after 0, 90, or 180 min of chase in mild detergent (1% digitonin). We used digestion with either Endo H or Peptide:-glycosidase F (PNGase F) to monitor the glycosylation status of UNC93B-HA. As observed for endogenous UNC93B, both wild-type and mutant UNC93B-HA retained Endo H and PNGase F sensitivity throughout the chase period (, middle and right). We noted the presence of a polypeptide of lesser autoradiography intensity, and of a size (∼130 kD) inconsistent with that predicted for UNC93B itself. This polypeptide, too, retained full Endo H sensitivity. The presence of this additional polypeptide was observed only for wild-type, but not for mutant, UNC93B, even though the mutant protein was expressed at levels comparable to that of wild-type UNC93B, as assessed by both immunoblotting (not depicted) and immunoprecipitation on extracts of RAW macrophages stably expressing the relevant constructs ( and ). The ability to recruit the ∼130-kD polypeptide, thus correlates with the functional properties of UNC93B. We also observed that an ∼130-kD, Endo H–sensitive polypeptide was coimmunoprecipitated with wild-type, but not with mutant, UNC93B in A20 B cells ( and not depicted). In A20 B cells, we detected an additional polypeptide of ∼150 kD that had characteristics similar to that of the ∼130-kD polypeptide, in that it was also Endo H–sensitive and was coimmunoprecipitated only with wild-type UNC93B ( and not depicted). To identify UNC93B-associated polypeptides, we prepared large-scale cell cultures of RAW macrophages stably expressing either wild-type or mutant UNC93B-HA and conducted a preparative immunoprecipitation of the tagged UNC93B proteins by retrieval via the HA epitope tag (). Bound materials were released by digestion with TEV protease. After resolution of eluted polypeptides by SDS-PAGE and visualization by silver staining, we excised polypeptides unique to wild-type UNC93B-bound material, as well as polypeptides common to both wild-type and mutant UNC93B samples, and determined their identity by MS ( and Tables S1 and S2, available at ). As expected, peptides for UNC93B were identified from both wild-type and mutant UNC93B samples with good sequence coverage, at 25.9% (12 peptides) and 11.7% (6 peptides), respectively ( and Table S1). Among the polypeptides coimmunoprecipitated only with wild-type UNC93B, we identified TLR3, 7, 9, and 13 with extensive sequence coverage over the entire length of the proteins (TLR3: 22.0%, 17 peptides; TLR7: 33.5%, 35 peptides; TLR9: 19.7%, 20 peptides; TLR13: 24.9%, 26 peptides; and Table S2). The same experimental approach was adapted to identify UNC93B-interacting proteins in A20 B cells. Again, we identified multiple peptides corresponding to TLR7 and 9 from the sample for wild-type UNC93B, but none from the sample for mutant UNC93B (unpublished data). We did not recover any peptides of TLR3, 7, 9, or 13 with mutant UNC93B, or any peptide that corresponded to TLR4 or other TLRs with either wild-type or mutant UNC93B in RAW macrophages or A20 B cells. We also carefully examined the proteomics data for the presence of TLR adaptors or signaling molecules implicated in TLR signal transduction pathways. Peptides derived from MyD88, TRIF, or IRAK were not detected, even when MS datasets were interrogated specifically for their presence. To confirm the interaction between wild-type UNC93B and TLRs, and the failure of mutant UNC93B to associate with TLRs, we generated myc-tagged TLR3, 4, and 9 fusion constructs and coexpressed them together with wild-type and mutant UNC93B-HA in HEK 293-T cells. Cells were lysed under mild conditions and subjected to immunoprecipitation with an anti-myc antibody. The presence of UNC93B in the myc immunoprecipitates was detected by immunoblot analysis with an anti-HA antibody (, top). Wild-type UNC93B interacts with TLR3 and 9, but not with TLR4, confirming our observations made by large-scale coimmunoprecipitation and MS (). Mutant UNC93B failed to interact with TLR3, 4, and 9, mirroring our earlier results. Total lysates were analyzed for expression levels of wild-type and mutant UNC93B by immunoblotting with an anti-HA antibody to confirm comparable expression levels (, bottom). All TLRs consist of an extracellular domain with a series of leucine-rich repeats, a transmembrane domain, and a cytosolic domain, which contains the conserved Toll-interleukin 1 receptor (TIR) domain. To address which region of the TLRs mediates binding to the UNC93B protein, we generated myc-tagged versions of chimeric TLRs (schematically depicted in ). Because TLR4 failed to bind to UNC93B, we exchanged the transmembrane regions of TLR3 and 9 with the transmembrane region of TLR4 (TLR3-4-3 and TLR9-4-9), and the transmembrane region of TLR4 was swapped for the transmembrane regions of TLR3 or 9 (TLR4-3-4 and TLR4–9-4). These myc-tagged TLR chimeras and myc-tagged wild-type TLR3, 4, and 9 were coexpressed with wild-type UNC93B-HA. Cells were then metabolically labeled with [S]methionine/cysteine and lysed under mild conditions. From these lysates, we performed immunoprecipitations with an antibody to either the myc or the HA epitope. The immunoprecipitation with the anti-myc antibody shows the expression levels of the wild-type and chimeric TLR proteins (, middle). By immunoprecipitating UNC93B via the HA tag and subsequent reimmunoprecipitation with TLR-specific antibodies, we recovered TLR3 and 9, but not TLR4 (, top), confirming our earlier observations ( and ). The chimeric TLRs containing the transmembrane domain of TLR4 (TLR3-4-3 and TLR9-4-9) failed to bind to UNC93B, whereas the TLR4 chimeras containing either the transmembrane region of TLR3 or 9 (TLR4-3-4 or TLR4-9-4) were readily recovered with UNC93B (, top). Recovery of UNC93B-HA is shown in (bottom). These results establish that TLR3 and 9 interact with the wild-type UNC93B protein via their respective transmembrane regions. To extend our observations from transduced cells to primary cells, we next analyzed the interactions of UNC93B with TLRs in BM-DCs obtained from wild-type (C57BL/6) and 3d mutant mice. Immunoprecipitation was performed with rabbit anti–UNC-C serum. We recovered considerable quantities of endogenous UNC93B from both wild-type and mutant dendritic cells (), confirming the earlier observation that the 3d mutation does not compromise stability or steady-state levels of the UNC93B protein. As seen for RAW cells transfected with epitope-tagged UNC93B, we observed the presence of several coimmunoprecipitating polypeptides in the size range of TLRs in wild-type, but not in mutant, UNC93B immunoprecipitates (, left, and ). The identity of one of the interacting proteins was revealed by denaturation of the primary anti–UNC-C immunoprecipitates, followed by reimmunoprecipitation with an anti-TLR7 antibody. In addition, we used dendritic cells obtained from TLR7-deficient mice as further evidence for the specificity of the anti-TLR7 antibody used. Wild-type UNC93B immunoprecipitates contain TLR7, as seen from the reimmunoprecipitation experiment with the anti-TLR7 antibody (, middle). As expected, no TLR7 is found in association with wild-type UNC93B obtained from TLR7-deficient dendritic cells. In dendritic cells obtained from 3d mice, we did not observe coimmunoprecipitation of TLR7 and UNC93B. In addition, the additional polypeptides within the size range of TLRs that were coimmunoprecipitated with wild-type UNC93B are absent from the 3d samples, whereas they are still present in the sample from TLR7-deficient mice. Similar results were obtained when we analyzed interactions between UNC93B and TLRs in splenocytes from wild-type, 3d, TLR7, and TLR9 mice. Again, TLR7 and TLR9 were coimmunoprecipitated only with wild-type, but not with mutant, UNC93B (Fig. S2, available at ). The inability of the mutant UNC93B protein to interact with TLRs might be caused by reduced expression of TLRs in cells from 3d mice. The inability of the 3d mice to signal via TLRs could therefore be the consequence of destabilization of TLRs, if UNC93B would serve a chaperone function. However, this is clearly not the case because we recovered equivalent amounts of TLR7 by direct immunoprecipitation from both wild-type and the 3d dendritic cell lysates using the TLR7 antibody (). In addition, we confirmed the interaction between UNC93B and TLR7 by first immunoprecipitating TLR7 and, subsequently, reimmunoprecipitating UNC93B from the denatured TLR7 immunoprecipitates. Wild-type, but not mutant, UNC93B was recovered by immunoprecipitation of TLR7 (). To determine whether activation of TLRs with their respective agonists regulates the interaction with UNC93B, we stimulated A20 B cells that stably express wild-type or mutant UNC93B-HA with TLR7 and 9 agonists; imiquimod or gardiquimod for TLR7 and CpG DNA for TLR9. After metabolically labeling cells with [S]methionine/cysteine, cell lysates were subjected to immunoprecipitation with an anti-HA antibody to retrieve UNC93B. No substantial changes in the levels of coimmunoprecipitated TLR polypeptides were observed after TLR activation when compared with unstimulated cells, nor did we see additional interacting proteins for agonist-exposed cells (). Similarly, stimulation with the TLR agonists did not alter the interaction between endogenous UNC93B and TLRs in splenocytes and dendritic cells from wild-type mice (unpublished data). TLRs are involved in the perception and processing of signals delivered by viral and microbial products and coordinate both innate and adaptive immunity. Mutations in UNC93B cause signaling defects for multiple TLRs and raise the question of the mechanism of its involvement in TLR signaling. By expression of epitope-tagged UNC93B and production of polyclonal antibodies that recognize endogenous UNC93B, we characterized UNC93B as a glycosylated ER-resident protein. The mutant UNC93B (H412R) protein exhibits distinct migration patterns in SDS-PAGE compared with wild-type UNC93B. Its behavior suggests a role for His412 in the intramolecular organization of UNC93B, as denaturation of polytopic membrane proteins using SDS without boiling is likely to preserve at least some features of secondary structure. Nonetheless, the 3d mutation (H412R) in UNC93B does not compromise biosynthesis and maturation of the mutant protein, and both wild-type and mutant proteins are equally stable. To study the role of UNC93B in TLR signaling, we identified UNC93B binding proteins using a large-scale preparative immunoprecipitation in conjunction with MS. We found that wild-type, but not mutant, UNC93B interacts with TLR3, 7, 9, and 13. We also found that TLR4 does not interact with either wild-type or mutant UNC93B. The number of peptides and the extent of sequence coverage of TLRs identified by MS suggest that these TLRs are well-represented among proteins that bind to UNC93B. The interaction between wild-type UNC93B and TLRs (TLR3, 7, and 9) was confirmed by additional biochemical analyses in cell lines and primary cells. The interactions between wild-type UNC93B and TLRs (TLR3, 7, 9, and 13) and the absence of interaction between UNC93B and TLR4 correlate well with the phenotype of the 3d mice, which show specific defects in signaling via TLR3, 7, and 9, but not TLR4 (). In addition, we demonstrated the expression of TLR13 in a macrophage cell line. TLR13 was identified by homology search of TIR domain-containing proteins in mouse ESTs, but no function has been assigned to it (). It will be interesting to see whether signaling via this less well-characterized TLR is also affected in 3d mice, as we predict it will be. Many studies demonstrate that TLR3, 7, and 9 are localized intracellularly (; ; ; ; ). TLR9 resides in the ER before stimulation and reaches lysosomes only upon activation (; ). The ER localization of TLR9 is consistent with our data showing that ER-localized UNC93B physically interacts with TLR9. UNC93B retains full Endo H sensitivity, which is also consistent with previous observations that TLR9 remains sensitive to Endo H digestion even after activation (; ). The transmembrane domains of TLR7 and 9 and the cytosolic linker region between the transmembrane domain and TIR domain of TLR3 determine the intracellular localization for these TLRs (; ; ; ). By using chimeric TLRs, we show that the transmembrane domains of TLR3 and 9 are responsible for binding to UNC93B. When the transmembrane domain of TLR4 is substituted for the transmembrane domain of either TLR3 or 9, the chimeric TLR4 proteins acquire the ability to interact with UNC93B. In contrast, TLR3 and 9 chimeras equipped with the TLR4 transmembrane domain no longer bind to UNC93B. The strength of interaction between UNC93B with its client TLRs is robust, as judged from the quantities of TLRs detected by direct immunoprecipitation and that were found in association with UNC93B. This result suggests that a considerable fraction, if not the majority of TLR3, 7, and 9, may be associated with UNC93B. The results of pulse-chase analyses are likewise consistent with stable association of UNC93B with TLRs. These observations raise the question as to the function of UNC93B. How does UNC93B participate in TLR signaling? One possibility is that UNC93B acts as an ER chaperone for TLR3, 7, 9, and 13. However, given the observation that the transmembrane segment of a TLR is sufficient to dictate interactions with UNC93B, and that interaction between UNC93B and TLRs is not transient, but is maintained for a prolonged period of time, we consider a classical chaperone function for UNC93B as less likely. Chaperone–client interactions are usually transient, and for glycoproteins they involve mostly the luminal/extracellular domains. Moreover, the expression level of TLR7 is not altered in BM-DCs from 3d mice compared with wild-type mice, arguing against the possibility that UNC93B would act by stabilization of its client TLRs. Instead, by analogy with the function of another ER-resident protein gp96, which is essential for TLR4/MD-2 complex assembly (), UNC93B may be involved in the assembly of intracellular TLRs with thus far unidentified proteins that are essential for either ligand recognition or signal transduction. Alternatively, UNC93B may play a role in retaining the client TLRs in the ER until they are ready to traffic to endosomes. TLR9 resides in the ER until activated (; ). Although there is no consensus on exactly where TLR3 localizes, it seems that a large proportion of TLR3 colocalizes with TLR9 at steady-state (; ). The previously demonstrated essential role of the transmembrane domain of TLR9 for ER localization and identification of the TLR9 transmembrane domain as the UNC93B-binding determinant in this study supports the role of UNC93B in ER retention of TLRs. This suggestion is at variance with the report by , which reports no discernible changes in TLR9 localization in 3d mice. MD-2 directly interacts with TLR4 and plays an important role in the recognition of LPS (). In addition, MD-2 was suggested to contribute to the surface localization of TLR4 (). In MD-2–deficient mouse embryonic fibroblasts, TLR4 is retained in the Golgi apparatus. Although MD-2–bound TLR4 is directed to the cell surface constitutively, UNC93B may play a role in the transport of its client TLRs to endosomes in a stimulation-dependent manner. Trafficking of TLR9 from ER to lysosomal compartments after uptake of CpG in dendritic cells may, thus, depend on the interaction between UNC93B and TLR9. Because both UNC93B and TLRs retain full Endo H sensitivity; even after TLR stimulation they may travel to endosomes via an unconventional anterograde route that does not involve the Golgi apparatus, should the UNC93B–TLR complex, indeed, traffic together to endosomes. High-resolution microscopy studies on localization of TLRs after proper stimulation may provide new insights into the possible role of UNC93B in TLR trafficking. Upon activation, many receptors recruit adaptor molecules onto which various downstream signaling molecules assemble for efficient and coordinated signal transduction. For some receptors, a scaffolding protein that constitutively binds the receptor provides a platform to which signaling molecules are recruited. UNC93B may serve as such a scaffold for TLR signaling or be an integral part of a signaling unit. The exposed loops and N and C termini of UNC93B could serve to organize and orient the other components of the TLR signaling complex. The stably maintained interactions between UNC93B and TLRs, even after activation of TLRs, support this hypothesis. Our initial large-scale immunoprecipitation of UNC93B, and the subsequent MS analysis, did not identify any prominent signaling molecules of the TLR signaling pathways. Furthermore, experiments designed to directly test the association of UNC93B with MyD88 did not yield any evidence for such interactions (unpublished data). However, it will still be interesting to see what additional proteins bind to UNC93B after stimulating cells with TLR agonists. The TLR signaling defects in cells from 3d mice are mimicked by pretreating wild-type cells with chloroquine or bafilomycin, which are agents that inhibit endosome acidification, but the 3d mutation does not affect the pH of different intracellular organelles (). Therefore, UNC93B does not seem to be directly involved in endosome acidification. Human UNC93B protein contains a region of weak homology to the bacterial ABC-2–type transporters (). In addition, homology searches of the conserved domains (rpsblast; National Center for Biotechnology Information BLAST) identified a second domain (residues 101 and 185 of UNC93B) with similarity to a bacterial transporter (MelB, E-value of 0.03). Even though sequence homology is weak in both cases, these observations raise the possibility that UNC93B could have a yet to be identified transporter function. The ectodomain of the microbial nucleotide-sensing TLRs (TLR3, 7, and 9) faces the lumen of the ER and endosomes. It is likely that viral DNAs and RNAs need to gain access to such an environment for signaling through their cognate TLRs. Therefore, UNC93B may participate in such processes and position the TLRs for efficient recognition of their agonists. The exact identities of intracellular compartments where the nucleotide-sensing TLRs receive stimulating signals from microbial DNAs and RNAs are still an open question. Studies using chemical inhibitors that prevent endosome acidification have proposed the endosomes as the location where intracellular TLR9 initiates signaling and potentially recognizes ligands (; ). However, upon addition of CpG DNA, even in MyD88-deficient dendritic cells, TLR9 travels normally from the ER to endosomes, where internalized CpG accumulates (). If translocation of TLR9 to endosomes is indeed a signal-mediated event, these data imply that either TLR9 receives a stimulatory signal of CpG DNA before reaching the endosomes, perhaps in the ER, or there is an additional molecule that senses CpG and triggers translocation of TLR9 in a MyD88-independent manner. The possibility that TLR9 may sense CpG in the ER or ER-derived structures deserves consideration. Pathways for retrograde transport of microbial products to the ER include delivery of the intact pathogen itself, as exemplified by SV40 and polyoma virus (). Bacterial toxins, such as cholera toxin and shiga toxin, likewise travel from the cell surface to the ER, from which they are discharged into the cytoplasm to intoxicate the cells exposed to the toxin (). In addition, it has been claimed that exogenously added soluble proteins can access the ER lumen in dendritic cells (). Therefore, the delivery of TLR ligands to the ER itself is certainly a possibility. Potentially, UNC93B could play a role in the perception of ER-delivered microbial nucleotides in concert with the intracellular TLRs. The reported phenotype of 3d mice includes defects in cross-presentation and MHC class II–mediated antigen presentation. Despite the defects in presentation of exogenous antigens, the subcellular distribution of MHC class I and II proteins was not affected (). Consistent with this observation, biosynthesis, maturation, and assembly of MHC class I and II molecules were identical in BM-DCs from wild-type and 3d mice. Although it remains unclear how UNC93B participates in antigen presentation, it is worth noting that a series of recent, controversial, observations suggests the involvement of the ER or ER-derived intracellular compartments in cross-presentation of exogenous antigens (, ; ). The role of UNC93B in cross-presentation deserves to be further explored in this context. In summary, we demonstrate that wild-type, but not mutant, UNC93B (H412R) physically interacts with TLR3, 7, 9, and 13. The established interaction between UNC93B and TLRs sheds new light on the 3d mutation and its TLR signaling-defective phenotype. The prominent ER localization of UNC93B, and of the TLRs to which it binds, raises the intriguing possibility that the ER itself may serve as a compartment from which TLR signaling is initiated. Murine RAW 264.7 macrophages (TIB-71; American Type Culture Collection [ATCC]) and human embryonic kidney (HEK) cells 293-T (CRL-11268; ATCC) were maintained in DME containing 10% heat-inactivated fetal calf serum (HIFS) and penicillin/streptomycin. Murine A20 B cells (TIB-208; ATCC) were maintained in RPMI 1640 medium supplemented with 10% IFS and penicillin/streptomycin. 293-T cells were transfected with FuGene-6 (Roche) according to the manufacturer's instructions. C57BL/6 wild-type mice were purchased from Taconic. The TLR7 () and TLR9 () mice were obtained from A. Marshak-Rothstein (Boston University, Boston, MA). All animals were maintained under specific pathogen-free conditions, and experiments were performed in accordance with institutional, state, and federal guidelines. Antibodies against mouse UNC93B were generated against the N-terminal (anti–UNC-N; aa 1–59) and the C-terminal (anti–UNC-C; aa 515–598) region. Three peptides were chosen for each region with an Antigen Profiler (Open Biosystems) and synthesized with a cysteine residue added to the N terminus by the Massachussetts Institute of Technology Center for Cancer Research Biopolymers Laboratory. N-terminal UNC93B peptides were as follows: 1N, C-DRHGVPDGPEAPLDE; 2N, C-PDGPEAPLDELVGAY; and 3N, C-GAYPNYNEEEEERRYYRRK. C-terminal UNC93B peptides were as follows: 1C, C-LQQGLVPRQPRIPKP; 2C, C-RYLEEDNSDESDMEG; and 3C, CPYEQALGGDGPEEQ. Peptides were analyzed by HPLC and MS for purity and coupled individually to keyhole-limpet hemocyanin (Pierce Chemical Co.) via the cysteine residue using sulfosuccinimidyl 4-[-maleimidomethyl]-cyclohexane-1-carboxylate (sulfo-SMCC; Pierce Chemical Co.) according to the manufacturer's recommendations. After coupling, the three peptides (1–3N or 1–3C) were mixed, and antisera were raised in rabbits by an outside vendor (Covance). The anti–UNC-C polyclonal serum was affinity purified using the three peptides of the C-terminal region (1C, 2C, and 3C) that are coupled to the resin using the Sulfo Link kit (Pierce Chemical Co.). The affinity-purified anti–UNC-C antibody was used for immunoprecipitations and immunoblotting. The TLR7 antibody and the TLR4 antibody (M16) were obtained from Imgenex and Santa Cruz Biotechnology, respectively. Rabbit polyclonal TLR3 and 9 antibodies are raised against peptide sequences in the C-terminal domain of the respective protein. The Flag antibody (M2; mouse monoclonal) was purchased from Sigma-Aldrich. The anti-HA affinity matrix (3F10, rat monoclonal) was purchased from Roche. The anti-HA 12CA5 antibody (mouse monoclonal) was produced in our laboratory from hybridoma cells. The anti-myc (9E10, mouse monoclonal) antibody was purchased from Invitrogen. The MHC class II antibody is a hamster monoclonal antibody (clone N22). Imiquimod (R837) and gardiquimod were purchased from Invivogen, CpG DNA (1826-CpG) was obtained from TIB Molbiol, and LPS ( 026:B6) was purchased from Sigma-Aldrich. Murine UNC93B (BC018388) was C-terminally fused with the Flag tag, followed by the TEV protease cleavage site (ENLYFQG) and the HA tag (UNC93B-HA, WT; see scheme in ). The point mutant UNC93B (H412R) was generated by sequential PCR with primers carrying the point mutation CC (His) to CC (Arg) and C-terminally fused with the Flag tag, followed by the TEV protease cleavage site and the HA tag (UNC93B-HA, H412R). UNC93B-HA WT and H412R were cloned into the retroviral vector pMSCVneo (Clontech Laboratories, Inc.), and stable cell lines in RAW or A20 B cells were established by retroviral transduction and selection with geneticin (see the following section). The TLR9 cDNA in pcDNA3.1 was provided by S. Bauer (Institut für Medizinische Mikrobiologie, München, Germany; ). C-terminally myc-tagged murine TLR3 (BC099937), TLR4 (BC029856), and TLR9 (AF348140) were generated by PCR and cloned into the pMSCVpuro (Clontech Laboratories, Inc.) or pcDNA3.1 vector (Invitrogen). The TLR chimeras were constructed by PCR “sewing” with cDNA corresponding to the following amino acids of the TLRs: TLR4-3-4, 1–625 of TLR4, 698–726 of TLR3, and 660–835 of TLR4; TLR3-4-3, 1–697 of TLR3, 626–659 of TLR4, and 727–905 of TLR3; TLR4-9-4, 1–625 of TLR4, 811–839 of TLR9, and 660–835 TLR4; and TLR9-4-9, 1–810 of TLR9, 626–659 of TLR4, 840–1032 of TLR9. All TLR chimeric constructs were C-terminally myc-tagged and cloned into the pMSCVpuro vector. The sequence of all constructs generated by PCR was verified. HEK 293-T cells were transfected with plasmids encoding VSV-G or Env, Gag-Pol, and pMSCV-UNC93B-HA (WT) or pMSCV-UNC93B-HA (H412R). 24 h after transfection, medium containing viral particles was collected, filtered through a 0.45-μm membrane, and added to RAW macrophages (VSV-G) or A20 B cells (Env) cells. Cells were spun for 2 h at 2,000 rpm, medium was changed, and cells were selected with 750 μg/ml geneticin (Invitrogen) 2 d after transduction. BM-DCs were prepared from C57BL/6, UNC93B mutant (3d), TLR7-deficient (TLR7), or TLR9-deficient (TLR9) mice, as previously described (). BM-DCs derived from wild-type (C57BL/6), UNC93B mutant (3d), TLR7, or TLR9 knockout mice were stimulated for 4 h with increasing concentrations of the TLR agonists LPS (TLR4), imiquimod (TLR7), or CpG DNA (TLR9). The conditioned medium was collected and analyzed by ELISA using the hamster anti–mouse/rat TNF antibody (BD Biosciences) as a capture antibody and a rabbit anti–mouse biotin-labeled secondary antibody (BD Biosciences). A20 B cells either not transduced or stably transduced with UNC93B-HA WT or H412R were metabolically labeled with [S]methionine/cysteine for 4 h (pulse). TLR agonists were added to the cells for the final hour of the pulse at the following concentrations: 10 μM imiquimod, 1 μM gardiquimod, and 1 μM CpG DNA. Cells were lysed in 1% digitonin lysis buffer and immunoprecipitation was performed as indicated in the figure legend. In brief, cells were starved for methionine and cysteine in Met/Cys-free DME (starvation medium) for 30 min, pulsed for different time periods, as indicated in the figure legends, with [S]methionine/cysteine (Perkin Elmer) in starvation medium supplemented with dialyzed HIFS, and chased with an excess of nonradioactive amino acids in regular DME for various time periods, as indicated in the figure legends. Cells were lysed in either of the following lysis buffers supplemented with the complete protease inhibitors (Roche), as indicated in the figure legends: RIPA (20 mM Tris-HCl, pH 7.4, 1 mM EDTA, 100 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, and 0.1% SDS) or buffer containing 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, and 5 mM EDTA with either 1% NP-40 or 1% digitonin as detergent. Lysates were equalized for incorporation of radioactive material with S counts in the trichloroacetic acid precipitate and immunoprecipitated with the indicated antibodies. Washes were performed with the same buffers used for lysis, except for digitonin lysates/immunoprecipitations, which were washed with 0.1–0.2% digitonin-containing buffer. Reimmunoprecipitations were performed as follows: protein–antibody complexes were dissolved by mild denaturation with 1% SDS and 1% β-mercaptoethanol for 1 h at 37°C. Subsequently, SDS and β-mercaptoethanol were diluted to 0.1% by addition of 1% NP-40 lysis buffer and reimmunoprecipitations were performed with the indicated antibodies. Immunoprecipitates were subjected to 10% SDS–PAGE without heating the samples, and polypeptides were visualized by fluorography. Digestions with Endo H and PNGase F were performed where indicated, in accordance with the manufacturer's instructions (New England BioLabs). In brief, cells were lysed in 1% NP-40, 1% digitonin, or RIPA buffer, and immunoprecipitation was performed with antibodies, as indicated in figure legends. The samples were subjected to 10% SDS–PAGE, transferred to a nitrocellulose membrane, and immunoblotted with the antibodies indicated. The procedure was adapted from . In brief, 4 billion RAW cells stably expressing UNC93B-HA (WT), UNC93B-HA (H412R), or no exogenous UNC93B protein (control cells) were lysed in 30 ml of ice-cold lysis buffer (1% digitonin, 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 5 mM EDTA, and complete protease inhibitors [Roche]), with rocking at 4°C for 1 h. The lysate was cleared of cell debris and nuclei by centrifugation at 20,000 for 15 min. UNC93B-HA and associated proteins were retrieved from 150 mg of cleared lysate by immunoprecipitation with 330 μl of compact anti-HA antibody beads. After incubation for 3 h, beads were extensively washed in wash buffer (the same composition as lysis buffer, except with 0.1% digitonin and without protease inhibitors). Bound material was eluted by incubation with 100 U TEV protease (Invitrogen) at 4°C overnight in 200 μl wash buffer. The eluate was exchanged into 20 mM NHCO, pH 8.0, with 0.1% SDS with the use of Sephadex G-25 resin (GE Healthcare) and concentrated in a Speed Vac (Savant). Reducing SDS loading buffer was added to the sample, and polypeptides were separated by 10% SDS–PAGE and revealed by silver staining. The bands of interest were excised, subjected to trypsinolysis, separated by liquid chromatography, analyzed by MS/MS, and database searched, as previously described (). Fig. S1 shows TNF secretion in response to TLR agonists in BM-DCs from wild-type versus UNC93B mutant mice. Fig. S2 shows that TLR7 and 9 coimmunoprecipitate with wild-type UNC93B in splenocytes from wild-type, but not UNC93B mutant, mice. Table S1 shows peptide sequences of UNC93B as identified by LC/MS/MS. Table S2 shows peptide sequences of TLR3, 7, 9, and 13 as identified by LC/MS/MS. The online version of this article is available at .

The crucial regulators of commitment to apoptotic cell death in mammals include three subclasses of the Bcl-2 family of proteins, each characterized by the presence of one or more Bcl-2 homology (BH) domains (; ; ). The prosurvival (or antiapoptotic) proteins, which contain three or four of the BH domains, are represented by Bcl-2, Bcl-x, Bcl-w, Mcl-1, and A1. The proapoptotic BH3-only members, which act as sensors of specific types of cellular stress, include Bid, Bim, Puma, Bad, Noxa, Bmf, Hrk, and Bik (). Acting downstream of both of these groups are the proapoptotic multidomain proteins, which contain BH1–3 and are represented by Bax, Bak, and perhaps Bok (; ). Once activated, Bax and Bak mediate permeability of the mitochondrial outer membrane, releasing proapoptotic factors, particularly cytochrome , that provoke caspase activation and the resulting rapid packaging of cell fragments for removal (). Although either Bax or Bak is required for apoptosis, their localization in healthy cells differs: Bak is an integral protein of the mitochondrial outer membrane, whereas Bax is predominantly cytosolic or loosely attached to the mitochondrial outer membrane. Commitment to apoptosis is governed by interactions between members of the Bcl-2 subclasses mediated by the amphipathic α-helical BH3 domain. Structural studies have revealed that BH3 peptides from Bak and Bad as well as the Bim polypeptide behave like ligands in binding to a hydrophobic groove on the surface of Bcl-x (; ). Indeed, BH3 peptides derived from each of the BH3-only proteins can mimic their full-length parent polypeptides in binding to prosurvival proteins, permeabilizing isolated mitochondria, and inducing apoptosis (; ; ). The interactions between Bcl-2 family members that allow the critical step of Bax and Bak activation remain poorly understood (). One widely discussed model posits that a subclass of BH3-only proteins termed activators, which are proposed to include Bim and Bid (after its truncation by caspases; ; ; ) and perhaps Puma (), can bind not only to the prosurvival proteins but also to Bax and Bak and provoke their activation. On this direct activation model (, left), the remaining BH3-only proteins, which are termed sensitizers, simply bind to the prosurvival proteins, lowering their capacity to sequester the activators. However, important aspects of this model remain problematic (see Discussion; ). An alternative view promoted by recent findings (; , ) is that the BH3-only proteins exclusively engage the prosurvival proteins, overcoming their sequestration of Bak or Bax (, right). Pertinent to this indirect activation model, certain BH3-only proteins interact selectively with subsets of the prosurvival proteins (): whereas Bim and Puma bound all five tightly, Noxa instead bound only Mcl-1 and A1, whereas Bad engaged only Bcl-2, Bcl-x, and Bcl-w (). Importantly, neutralization of multiple prosurvival proteins was required to induce apoptosis (). Furthermore, only certain prosurvival proteins engage Bak (): in healthy fibroblasts, both Mcl-1 and Bcl-x but not Bcl-2 could bind Bak and protect against Bak-mediated permeability (). In accord with the indirect activation model, Bak was freed by BH3-only proteins that bind tightly to both Mcl-1 and Bcl-x, such as Noxa plus Bad (). Thus, whether the BH3-only proteins activate Bak and Bax directly or indirectly (or both) remains to be established (). Because the Bcl-2 family primarily regulates the integrity of the mitochondrial outer membrane, it is essential to establish which interactions between its subclasses directly control mitochondrial permeability. In this study, we have explored whether the previously identified interactions of prosurvival proteins with Bak or those with BH3 peptides from BH3-only proteins (; ) can account for their ability to regulate mitochondrial permeability. Cytochrome release was assessed on two sources of mitochondria: those from eggs ( egg mitochondria [XEM]) were chosen because they are particularly robust and their permeabilization has been widely studied (; ; ; ), whereas mouse liver mitochondria (MLM) were selected because they lack Bax and thus exhibit Bak-dependent permeabilization (). We analyze our results both from the perspective of how prosurvival proteins prevent permeabilization and how BH3-only proteins act to permeabilize mitochondria. In accord with the indirect activation model, the findings indicate that prosurvival proteins act by interacting with mitochondrial components such as Bak (and perhaps Bax) and that BH3-only proteins must engage multiple prosurvival proteins to induce efficient permeabilization. To explore whether selective binding of BH3-only domains to prosurvival proteins could regulate mitochondrial integrity, we first established whether each of eight BH3 peptides could permeabilize mitochondria. We used long BH3 peptides (24–26 mers) with known affinities for the prosurvival proteins (Fig. S1, available at ; ). When MLM were incubated with each peptide for 2 h at 37°C, specificity was observed in that most of the peptides (Puma, Bmf, Bid, Bik, Bad, Hrk, and Bim), if added at 10 μm, promoted Bak oligomerization and cytochrome release, whereas the human NoxaBH3 was inactive ( and Fig. S2), as were peptides from both the BH3 domains of mouse Noxa (not depicted). The BH3 peptides exhibited similar specificity on XEM after a 2-h incubation at room temperature, although certain ones were up to 10-fold more active (). Notably, under our conditions, several BH3 peptides termed sensitizers (Puma, Bmf, Bik, Bad, and Hrk) promoted cytochrome release essentially as effectively as the putative activators Bid and Bim (see Discussion). Each of the BH3 peptides active in this assay exhibits high affinity for Bcl-x and Bcl-w, whereas the only inactive peptide, Noxa, binds only to Mcl-1 and A1 ( and S1 B; ; ). This suggests that the endogenous prosurvival proteins protecting both XEM and MLM are principally Bcl-x–like rather than Mcl-1–like. Indeed, by Western blot analysis, MLM contained readily detectable levels of Bcl-x, albeit undetectable Bcl-2, Bcl-w, or A1 and very little Mcl-1 ( and Fig. S3, A–C; available at ). Nevertheless, even these low levels of Mcl-1 seem to be functionally relevant because the Bad peptide, which does not bind Mcl-1 or A1, gave a less complete permeabilization of MLM than the Bim or Bid BH3 peptides. For the XEM, appropriate antibodies are not yet available for determining which of the five known prosurvival homologues () are present, but Bcl-x–like proteins probably predominate because BadBH3 but not NoxaBH3 permeabilized the XEM. Importantly, the permeabilizing activities of Bim and NoxaBH3 peptides were altered by mutations known to affect both their binding to prosurvival molecules and their proapoptotic activities. Thus, BimBH3 mutated at either a single invariant residue (L94A) or, at four conserved residues (4E: I89E, L94E, I97E, and F101E), no longer permeabilized XEM () or MLM (not depicted). These mutations decrease binding to prosurvival proteins at least 50-fold and abrogate proapoptotic activity in mouse embryonic fibroblasts (; ; ). Conversely, in the case of NoxaBH3, we previously showed that single (K35E or F32I) or double (K35E/F32I) mutations markedly increase binding to Bcl-x–like prosurvival proteins as well as proapoptotic activity (Fig. S1; ). Accordingly, the mutant peptides, unlike the wild-type NoxaBH3, could permeabilize both XEM and MLM (). Thus, both ablating the ability of Bim BH3 to bind prosurvival proteins and extending the range bound by NoxaBH3 have effects on mitochondrial permeability that parallel their known proapoptotic activities in cells. These findings argue that the ability to engage specific subsets of prosurvival proteins determines both the mitochondrial permeabilizing activity and proapoptotic behavior of BH3-only proteins. A recent study () reported results with MLM that differ in important respects from those shown in . Specifically, sensitizer BH3 peptides such as Bad and Puma were reported not to permeabilize MLM (Bim and Bid peptides were not tested). Because apparently incubated the mitochondria at room temperature (), whereas we used 37°C, we tested the effect of temperature on MLM permeabilization elicited by several peptides and by tBid (). Importantly, at room temperature (22°C), Puma and Bad released very little cytochrome , which is consistent with the study by , but neither did the Bim or Bid peptides. In contrast, all four peptides provoked release at 37°C. At 30 and 37°C, the Puma peptide was essentially as effective as the Bim or Bid peptides, whereas the Bad peptide was weaker, which is consistent with its more restricted binding pattern for prosurvival proteins ( and S1). Moreover, tBid was several times more active at the higher temperatures than at room temperature. Thus, the release of cytochrome from MLM involves a temperature-dependent step, but their permeabilization is not confined to Bim and Bid (see Discussion). We next explored how mitochondrial permeability was affected by the addition of recombinant prosurvival proteins. Bcl-x, Bcl-w, Mcl-1, and BHRF1 (the Bcl-2 homologue from Epstein-Barr virus) were selected because they have distinct binding profiles for BH3 peptides and/or Bak. Bcl-x and Bcl-w bind Bad but not Noxa, whereas Mcl-1 binds Noxa but not Bad (), and BHRF1 binds neither Bad nor Noxa ( and S1 B; unpublished data). Furthermore, whereas both Bcl-x and Mcl-1 bind Bak, Bcl-w does so only very poorly (). For Bcl-w, Mcl-1, and BHRF1, we had to use forms lacking a portion of the C-terminal hydrophobic segment because the full-length recombinant proteins were poorly soluble, whereas the truncated molecules are well-behaved proteins that retain full BH3 binding (Hinds, M., and P. Czabotar, personal communication; ; ). All four prosurvival proteins could block the tBid-induced permeability of XEM, and all except Bcl-w protected MLM (). The failure of Bcl-w to protect MLM is consistent with its poor ability to bind to Bak or to protect mouse cells from Bak-mediated apoptosis (). mice are insensitive to tBid or BH3 peptides (Fig. S3, D–F; ). In contrast, XEM permeability apparently is not mediated solely via Bak because XEM are protected by the addition of either Bcl-w () or Bcl-2 (), neither of which can bind Bak appreciably or protect cells from Bak-mediated apoptosis (). Because possesses homologues of Bax as well as Bak () and mitochondria of undifferentiated cells can contain high levels of Bax (), we surmise that XEM permeabilization is not solely Bak dependent because of resident Bax (or Bok). Although Bcl-x at low concentrations (∼20 nM) protected both types of mitochondria, 10–30-fold higher concentrations of the other prosurvival proteins were required. Their lower potency can be attributed largely to their C-terminal truncation, as Bcl-xΔC24 was around 10–30 times less active than full-length Bcl-x (). The cytosolic conformer of the Bcl-w prosurvival protein has its hydrophobic C-terminal domain tucked into the hydrophobic binding groove, but this domain can evert to insert into intracellular membranes (). Removal of its C-terminal residues facilitates binding of BH3-only proteins to the hydrophobic groove but decreases attachment to mitochondria (; ). As each prosurvival protein contains a hydrophobic C-terminal domain, the much stronger protection observed with full-length Bcl-x () probably reflects its enhanced ability to locate to the mitochondrial outer membrane. The failure of Bcl-w to protect MLM but not XEM from tBid () may reflect its poor affinity for Bak. Interestingly, high concentrations of Bcl-w did augment the protection from tBid conveyed by Bcl-x if Bcl-x was at a concentration that was only partially protective (, 10 nM Bcl-x in the 9th, 12th, and 15th lanes). This limited degree of protection may be caused by Bcl-w sequestration of tBid or by a slight ability of Bcl-w to bind Bak (). Similarly, a recent report that high concentrations of Bcl-w could block the tBid-induced permeabilization of MLM () might be a result of the sequestration of the low concentration of tBid added (which gave only 50% cytochrome release Together, these findings suggest that prosurvival proteins do not protect mitochondria simply by sequestering BH3 ligands (see Discussion). Presumably, Bcl-w can protect the XEM () by engaging a protein present specifically in those mitochondria (perhaps Bax). We assessed how the known affinities of these prosurvival proteins for each BH3 peptide ( and S1 B; ) correlated with their ability to block the permeabilization elicited by that peptide. If a prosurvival protein acted solely by sequestering the BH3 peptide, protection would be minimal where interaction with the peptide is weak. To test this, XEM were preincubated with increasing concentrations of a prosurvival protein and were challenged with the BH3 peptides shown () to be active ( and Fig. S4, available at ; and not depicted). Notably, although Mcl-1 does not bind the Bad BH3 peptide (), it efficiently blocked BadBH3-induced permeability. Likewise, despite the poor binding affinity of BHRF1 for the Bad, Bmf, and Hrk BH3 peptides (Fig. S1 B), BHRF1 efficiently blocked cytochrome release from XEM by each of them ( and S4). Mcl-1 and BHRF1 also blocked the permeabilizing effect of these BH3 peptides on MLM (). A model of how MLM are permeabilized by Bad and protected by Mcl-1 is shown in . In these cases, Mcl-1 and BHRF1 cannot protect mitochondria merely by sequestering the added BH3 peptide. Indeed, close examination of further results with XEM (Fig. S4) suggests that Mcl-1 and BHRF1 are actually more effective at blocking the activity of BH3 peptides that they cannot bind. For example, Mcl-1 protected XEM better from the Bad than from the Bim or Puma peptides, and BHRF1 protected XEM better from the Bad, Bmf, or Hrk peptides than from the Bim or Puma peptides. Together, these findings demonstrate that prosurvival function does not rely on the sequestration of BH3 ligands. Instead, the peptides that bind the prosurvival protein simply reduce its ability to bind its true target. As argued above for Bcl-x and Bcl-w (), Mcl-1 and BHRF1 must be acting on a component of the isolated mitochondria. Because both Mcl-1 and Bcl-x can bind Bak (; ) and both can prevent Bak-mediated apoptosis (; ), Bak is almost certainly their MLM target. How BHRF1 protects from Bak-mediated apoptosis is currently uncertain. A proponent of the direct activation model might argue that the sensitizer peptides induce permeabilization in these experiments by releasing endogenous Bim or tBid putatively bound to prosurvival proteins on the mitochondria. mice (). Moreover, shows that the absence of both Bim and Bid did not affect the ability of prosurvival proteins to protect the mitochondria from sensitizer peptides. In accord with this result, sensitive Western blots have failed to reveal any form of Bim, Bid, or Puma in wild-type MLM (Fig. S3, G–I). Thus, none of the putative activator BH3-only proteins appears to be required to induce permeabilization. To explore whether the binding profiles of BH3 peptides to prosurvival proteins regulate mitochondrial permeability, we established mitochondrial incubations in which the dominant acting prosurvival protein was either Bcl-x or Mcl-1 and exposed the mitochondria to either the BadBH3 peptide (to neutralize Bcl-x) or the Noxa one (to neutralize Mcl-1). We reasoned that only BH3 peptides that could bind to the governing prosurvival protein and, thus, prevent it from engaging its mitochondrial target would induce permeability. Bcl-x–protected mitochondria were provided by XEM supplemented with tBid and Bcl-x, whereas the Mcl-1–protected mitochondria were XEM supplemented with tBid and Mcl-1. Titrations showed that in these mixtures, either 30 nM Bcl-x or 300 nM Mcl-1 protected against at least 20 nM tBid (). Both types of supplemented XEM remained intact unless specific BH3 peptides were added. Notably, although NoxaBH3 did not permeabilize unsupplemented XEM (), it readily sensitized the Mcl-1–protected XEM to tBid (). Conversely, BadBH3 could permeabilize normal XEM but not the Mcl-1–protected XEM (). Thus, in these experiments, NoxaBH3 initiated permeability by directly binding and inactivating Mcl-1, whereas BadBH3 was inactive because it cannot bind Mcl-1. Conversely, the Bcl-x–protected XEM were susceptible to BadBH3 but not NoxaBH3. Furthermore, MLM supplemented in the same way as XEM gave equivalent results: Noxa but not Bad sensitized MLM protected by Mcl-1 to tBid, whereas Bad but not Noxa could disrupt MLM guarded by Bcl-x () or by Bcl-xΔC24 (not depicted). In contrast to Bad and Noxa, the Bim and Puma peptides, which can bind to both Mcl-1 and Bcl-x, induced permeability in MLM protected by either Mcl-1 or Bcl-x (, bottom). Thus, the contrasting behavior of the Bad and NoxaBH3 peptides with either XEM or MLM confirms that binding of BH3 ligands to specific prosurvival proteins is the major route to the disruption of mitochondria. In these experiments, it appears unlikely that tBid is acting as a direct activator of Bak because Fig. S5 (available at ) shows that the BadBH3 peptide, which does not associate with Bak, could substitute for tBid in an experiment equivalent to that in . Instead, tBid probably is required to permeabilize mitochondria in because it, together with the added BH3 peptide, can target all of the relevant prosurvival proteins, whether endogenous or added. Because only particular BH3 peptides could neutralize Mcl-1 and Bcl-x, we asked whether Bcl-w and BHRF1 behaved similarly by establishing mitochondrial mixtures in which each was the dominant prosurvival protein. Again, only specific BH3 peptides triggered permeabilization (). Because Bad or Puma but not Noxa can bind to Bcl-w ( B and S1 B), Bcl-w–mediated protection was overcome only by the Bad or Puma peptide. Similarly, BHRF1-mediated protection was overcome by the Puma but not the Bad or Noxa peptides, which is consistent with the ability of Puma but not Bad or Noxa to bind to BHRF1 ( B and S1 B). Thus, the binding of specific BH3 peptides to each prosurvival protein was required to overcome the mitochondrial protection bestowed by that prosurvival protein. Complementation between Noxa and BadBH3 domains has been observed in killing assays with cells (mouse embryonic fibroblasts) that express both Bcl-x–like and Mcl-1–like proteins (; ). Our aforementioned findings suggested that mitochondria protected by both types of prosurvival proteins would require both the Noxa and Bad peptides to induce permeability. To test their complementarity, we established mitochondria protected by both Bcl-x–like and Mcl-1 proteins. We reasoned that XEM predominantly contained only Bcl-x–like prosurvival proteins because BadBH3 readily induced their permeabilization ( and ). Thus, by simply supplementing XEM with Mcl-1, we generated dual-protected mitochondria. As predicted, neither the Bad nor Noxa peptide alone could permeabilize the dual-protected mitochondria even at high concentrations (30 μM), whereas the combination was active even at 1 μM (). Presumably, BadBH3 bound the endogenous Bcl-x–like proteins (as in and ), whereas NoxaBH3 neutralized the exogenous Mcl-1 (as in ). As the permeabilization in these experiments did not involve the addition of tBid (or indeed any other component), it appears that Bax/Bak activity can be triggered simply by inactivating both the Mcl-1 and Bcl-x–like proteins. It is also noteworthy that the endogenous Bcl-x–like proteins present in XEM behaved similarly to added recombinant Bcl-x (), supporting the physiological relevance of those experiments. Thus, the permeabilization of mitochondria protected by both Bcl-x–like and Mcl-1–like prosurvival proteins requires the neutralization of both types of guardians, as found for apoptosis in cells (, ). We next examined whether BH3 peptide–initiated mitochondrial permeabilization was caused by the disruption of Bak binding to prosurvival proteins. As in , MLM were incubated with tBid and either Mcl-1 or Bcl-x together with Bad, Noxa, or BimBH3 peptides (). The mitochondrial pellets were then examined for the binding of Bak to Mcl-1 or Bcl-x by coimmunoprecipitation. In the absence of BH3 peptides, Bak immunoprecipitated with Mcl-1 and Bcl-x, whereas in the presence of Bim peptide, Bak failed to associate with either Mcl-1 or Bcl-x. In contrast, Noxa specifically interfered with Bak binding to Mcl-1, whereas Bad interfered with its binding to Bcl-x. Importantly, in each case, the failure of Bak to bind its guardian correlated with the mitochondrial release of cytochrome . Thus, in these mitochondrial incubations, specific binding of BH3 peptides to prosurvival proteins blocks the sequestration of Bak, allowing it to permeabilize mitochondria. Interestingly, the binding of Noxa or Bim to Mcl-1 (which lacks 11 hydrophobic C-terminal residues) reduced its association with mitochondria, as indicated by the lower Mcl-1 levels immunoprecipitated from the mitochondrial pellet (, compare the third and fourth lanes with the first and second lanes). The Bad or Bim peptide did not reduce the association of full-length Bcl-x with mitochondria, presumably because its hydrophobic C terminus can insert into the mitochondrial outer membrane independent of binding to Bak. The specific effects of Bad and Noxa on Bak association with prosurvival proteins were also observed in whole cell lysates (). HA-tagged Bad or Noxa were stably expressed in HeLa cells, which contain functionally relevant levels of the two Bak guards Mcl-1 and Bcl-x. As previously observed for fibroblasts (; ), the cells remained viable because either Mcl-1 or Bcl-x is sufficient to sequester Bak. Notably, in the Bad-expressing cells, Bcl-x could bind Bad but could no longer bind Bak (, second lane), whereas in Noxa-expressing cells, Mcl-1 could bind Noxa but could not bind Bak (, sixth lane). Interestingly, the inability of one prosurvival guardian to bind Bak resulted in higher levels of Bak associating with the other guardian: for example, Noxa targeting of Mcl-1 lead to a higher Bak association with Bcl-x (, third lane) and vice versa for Bad (, fifth lane). This suggests that the Bak released from either of its two guards is bound by the other guard. #text The BH3 peptides, most of which were described previously (), were 24–34 residues long (Fig. S1 A). Caspase-8–cleaved human Bid (tBid) and human Bcl-x were produced as previously described (). Human Bcl-wA128EΔC10 and mouse Mcl-1ΔN151ΔC11, which were also produced as described previously (; ), contained five additional N-terminal residues (GPLGS) as a result of cloning. Recombinant BHRF1ΔC16 with N-terminal Flag tag was overexpressed in BL21-Codon Plus cells from a pET11a vector (Novagen) and the protein purified from cell lysates using Q-Sepharose ion-exchange chromatography followed by ammonium sulfate precipitation and Superdex G-75 gel filtration chromatography. Human Bcl-xΔC24 was a gift from P. Czabotar (The Walter and Eliza Hall Institute, Melbourne, Victoria, Australia; produced as described previously []). Bcl-xΔC23 (R11; a gift from J.R. Tata, National Institute for Medical Research, London, UK) was subcloned into pGEX-6P-3, and the protein was produced as described previously (; ). mouse liver as previously described (). For simplicity, these heavy membrane fractions are referred to here as XEM and MLM. For incubations with Bcl-2 family proteins or peptides, 1 mg/ml XEM and MLM were each suspended in buffer (100 mM KCl, 2.5 mM MgCl, 100 mM sucrose, 20 mM Hepes/KOH, pH 7.5, and 1 mM DTT). Recombinant prosurvival proteins were added, and samples were incubated for 20 min at 22°C for XEM and at 37°C for MLM before the addition of BH3-only reagents and further incubation for 2 h (MLM were also incubated at 22 and 30°C in as indicated). After centrifugation at 10,000 for 5 min at 4°C, the supernatant and pellet fractions were carefully separated. To assess mitochondrial permeability, each fraction was combined with loading buffer, and equivalent amounts (except as noted in ) were analyzed by SDS-PAGE and Western blotting for cytochrome (clone 7H8.2C12; BD Biosciences). For ease of viewing, the supernatant and pellet fractions were generally run on separate gels, which were then processed as matching pairs during each stage of electrophoresis, blotting, and detection and with no digital adjustment of the scanned images. Aliquots of MLM (125 or 25 μg of protein) and whole cell lysates of mouse embryonic fibroblasts (30 or 50 μg of protein) were analyzed by Western blotting for the presence of Bcl-x and Mcl-1 using antibodies raised against Bcl-x (rabbit polyclonal; BD Biosciences) and mMcl-1 (rat monoclonal clone 19C4-15) followed by incubation with anti–rabbit and anti–rat HRP-labeled secondary antibodies (Southern Biotechnology Associates, Inc.). The proteins were detected using ECL (GE Healthcare). HeLa cells stably expressing Bad or Noxa were provided by J. Fletcher (The Walter and Eliza Hall Institute, Melbourne, Victoria, Australia). MLM and HeLa cell lysates were prepared in lysis buffer (20 mM Tris, pH 7.4, 135 mM NaCl, 1.5 mM MgCl, 1 mM EGTA, and 10% glycerol) containing 1% Triton X-100 and supplemented with Complete protease inhibitors (Roche). Immunoprecipitation was performed using anti–Mcl-1 (rat monoclonal clone 14CH-20) and anti–Bcl-x (rat monoclonal clone 1C2; ), and immune complexes were captured with protein G–Sepharose. Proteins were resolved by SDS-PAGE, transferred onto polyvinylidene difluoride membranes, and detected by antibodies directed against Bak (rabbit polyclonal clone B5929; Sigma-Aldrich), Mcl-1 (rat monoclonal clone 19C4-15), anti–Bcl-x (mouse monoclonal clone 2H12; BD Biosciences), or HA (mouse monoclonal clone 16B12; BAbCO). Secondary antibodies used were anti–rabbit and anti–rat HRP-labeled antibodies (Southern Biotechnology Associates, Inc.) for Bak and Mcl-1, respectively, whereas Bcl-x and HA were detected using a goat anti–mouse IgG Fcγ fragment–specific HRP conjugate (Jackson ImmunoResearch Laboratories). Fig. S1 shows the sequence of each BH3 peptide used in this study and their binding affinities for four prosurvival proteins. Fig. S2 shows that MLM permeabilization by sensitizer BH3 peptides correlates with Bak cross-linking. In Fig. S3, Western blots show that MLM contain readily detectable levels of Bak, whereas Bax, Bcl-2, Bcl-w, A1, Bid, Bim, and Puma are undetectable. In Fig. S4, Mcl-1 and BHRF1 are most efficient at blocking the BH3 peptides for which they have least affinity. Fig. S5 shows that specific binding of BH3 peptides to Mcl-1 and Bcl-x results in mitochondrial permeabilization (as in A), even if BadBH3 replaces tBid as the initial permeabilizing agent. Online supplemental material is available at .

To be accurately partitioned during cell division and inherited by the daughter cells, organelles must double in size and divide during the cell cycle (; ; ). The division of mitochondria and chloroplasts, the two endosymbiotic organelles that are surrounded by inner and outer membranes, is uncoupled from cell division and requires the FtsZ-like and/or dynamin-related GTPases (). In contrast, the division of the Golgi apparatus, an organelle of the secretory pathway that is bound by a single membrane, is coupled to the cell division cycle and is served by a division machinery that is unique for this nonendosymbiotic organelle (; ). Similar to Golgi and in contrast to endosymbiotic organelles, peroxisomes derive from the endoplasmic reticulum () and are surrounded by a single membrane. On the other hand, akin to mitochondria and chloroplasts and contrary to Golgi, peroxisomes require dynamin-related GTPases for their division that is uncoupled from cell division (; ). Moreover, the peroxisomal and mitochondrial division machineries in mammalian cells share at least two essential protein components (). Although the mechanisms by which mitochondria, chloroplasts, and Golgi divide are well defined (; ; ), the molecular mechanism for the integration of multiple components of the peroxisomal division machinery remains to be established (; ). We study peroxisome division in the yeast . Similar to peroxisomes in humans and in other yeast species (), peroxisomes in do not grow and divide at the same time (). The growth of immature peroxisomal vesicles, termed P1–P5, which is accomplished by the stepwise import of distinct subsets of matrix proteins, and their development into mature peroxisomes P6 occur before completely assembled mature peroxisomes undergo division. The division of mature peroxisomes in is regulated by an unusual mechanism that controls membrane fission in response to a signal emanating from within the peroxisome (). The import of matrix proteins into different immature intermediates along the peroxisome assembly pathway provides them with an increasing fraction of the matrix proteins present in mature peroxisomes. The increase in the total mass of matrix proteins above a critical level causes the redistribution of a peroxisomal protein, acyl-CoA oxidase (Aox), from the matrix to the membrane. A substantial redistribution of Aox occurs only in mature peroxisomes, which contain the greatest percentage of matrix proteins. Inside mature peroxisomes, the membrane-bound pool of Aox interacts with Pex16p. Pex16p is a membrane-associated peroxin that negatively regulates the membrane fission event required for the division of immature peroxisomal vesicles, thereby preventing their excessive proliferation (). The interaction between membrane-attached Aox and Pex16p terminates the negative action of Pex16p on fission of the peroxisomal membrane, thereby allowing mature peroxisomes to divide. Akin to other membrane fission events (; ; ), fission of the peroxisomal membrane must be preceded by the destabilization of the membrane bilayer and strong membrane bending. These energetically unfavorable processes require several teams of proteins and a distinct set of membrane lipids, including phosphoinositides, phosphatidic acid (PA), and diacylglycerol (DAG; ; ; ; ; ). Here, we investigate how the interaction between Pex16p and Aox promotes the division of mature peroxisomes. We demonstrate that the Pex16p- and Aox-dependent intraperoxisomal signaling cascade activates the biosynthesis and transbilayer movement of a distinct set of membrane lipids. The resulting remodeling of the lipid repertoire of the membrane bilayer initiates the stepwise assembly of a multicomponent protein complex on the surface of the mature peroxisome. This newly assembled protein complex carries out membrane fission, thereby executing the terminal step of peroxisome division. In wild-type cells, the levels of PA and DAG in the peroxisomal membrane dramatically increased only during conversion of immature peroxisomal vesicles P5 to mature peroxisomes P6 (). These two cone-shaped lipids are potent inducers of membrane bending and fission (; ). In contrast, the level of lyso-PA (LPA), an inverted cone–shaped lipid (), in the membrane greatly reduced during conversion of P5 to P6 (). Importantly, the lack of Pex16p in mutant cells resulted in the accumulation of PA and DAG and led to the disappearance of LPA even in the membrane of immature peroxisomal vesicles P3 (). The mutation, which causes the excessive proliferation of immature peroxisomal vesicles (), impaired the conversion of P3 to P4 (unpublished data). On the contrary, the mutation, which averts peroxisome division by dramatically elevating the intraperoxisomal pool of Pex16p (), abolished the formation of PA and DAG and prevented the disappearance of LPA even in the membrane of P6 (). These findings suggest that the interaction between Aox and Pex16p at the matrix face of the membrane of mature peroxisomes activates the biosynthesis of PA and DAG and promotes the catabolism of LPA. To elucidate the mechanism that regulates the levels of PA, DAG, and LPA in the peroxisomal membrane, we reconstituted peroxisomal liposomes from detergent-solubilized peroxisomal membrane proteins (PMPs) and membrane lipids of immature peroxisomal vesicles P1. No Aox subunits are attached to the membrane inside these liposomes (), in which Pex16p is present only in its free form. In the membrane of P1 liposomes supplemented with [C]LPA as the only radiolabeled lipid, the decline in the level of [C]LPA coincided with the increase in the amount of newly synthesized [C]PA, which preceded the appearance of [C]DAG, only if these liposomes were reconstituted from PMPs immunodepleted of Pex16p (). In contrast, if [C]PA was used as the only radiolabeled membrane lipid for reconstituting P1 liposomes, the decline in its level coincided with the increase in the amount of newly synthesized [C]DAG even if the PMPs taken for liposome reconstitution were not immunodepleted of Pex16p (). Altogether, our findings provide evidence that the conversion of P5 to P6 is marked by the biosynthesis of PA and DAG in the peroxisomal membrane; PA and DAG are formed in a two-step biosynthetic pathway, which includes two consecutive enzymatic reactions catalyzed by an LPA acyltransferase (LPAAT) and a PA phosphatase (PAP; ); and Pex16p, a negative regulator of the division of immature peroxisomal vesicles (), inhibits LPAAT. The LPAAT and PAP reactions are the only reactions leading to the formation of PA and DAG, respectively, in the peroxisomal membrane. In fact, this membrane lacked the activities of all other enzymes that can promote the biosynthesis of PA or DAG (; ; ), including phospholipase D, inositol phosphosphingolipid phospholipase C (PLC), phosphoinositide-specific PLC, DAG kinase, inositol phosphorylceramide synthase, and inositolphosphotransferase 1 (Fig. S1, available at ). We purified LPAAT and PAP from the membrane of P6 (). Purified LPAAT and PAP were identified by mass spectrometry as Slc1p, an acylglycerol-3-phosphate acyltransferase (), and Dpp1p, a DAG pyrophosphate phosphatase (), respectively. Using highly purified peroxisomes of wild-type cells, we found that all six peroxisomal subforms have similar amounts of both Slc1p (LPAAT) and Dpp1p (PAP; ). Akin to the peroxisomal integral membrane protein Pex2p () and in contrast to the peroxisomal peripheral membrane protein Pex16p (), neither Slc1p (LPAAT) nor Dpp1p (PAP) was solubilized by either 1 M NaCl or 0.1 M NaCO (pH 11.0; ). Thus, both Slc1p (LPAAT) and Dpp1p (PAP) are integral membrane proteins. Furthermore, like Pex16p attached to the lumenal face () and unlike the peripheral membrane protein Pex19p on the cytosolic face of peroxisomes (; ), both Slc1p (LPAAT) and Dpp1p (PAP) were resistant to digestion by external protease added to intact peroxisomes (). Altogether, these data imply that, in all six peroxisomal subforms, both Slc1p (LPAAT) and Dpp1p (PAP) are integral membrane proteins that do not face the cytosol, being integrated into the lumenal leaflet of the membrane. Importantly, the lack of LPAAT in mutant cells abolished the formation of PA and DAG and prevented the disappearance of LPA in the membrane of P6 () and resulted in a reduced number of greatly enlarged mature peroxisomes (Fig. S2, A, B, G, and H, available at ). Moreover, the lack of PAP in mutant cells did not impair the Slc1p (LPAAT)-dependent biosynthesis of PA from LPA in the membrane of P6 (); prevented the conversion of PA to DAG in the membrane of P6 (); and resulted in fewer, but greatly enlarged, mature peroxisomes (Fig. S2, A, C, G, and H). Altogether, our findings provide evidence that both the Slc1p (LPAAT)-dependent formation of PA from LPA and the subsequent Dpp1p (PAP)-dependent biosynthesis of DAG from PA, which occur in the lumenal leaflet of the peroxisomal membrane only during conversion of P5 to P6, are essential for the division of P6, and although the biosynthesis of PA is necessary for the division of P6, the presence of PA alone is not sufficient for promoting this process, which also requires the biosynthesis of DAG. It remains to be established whether DAG alone stimulates peroxisome division or, alternatively, the simultaneous presence of PA and DAG in the membrane of P6 is mandatory for its fission. Because Pex16p inhibits LPAAT in the membranes of P1–P5, thereby preventing the formation of both PA and DAG, we sought to define the mechanism for the negative regulation of LPAAT by Pex16p in immature peroxisomal vesicles. Pex16p solubilized with the detergent n-octyl-β--glucopyranoside (n-OG) from the membranes of P1–P5 purified from wild-type cells was able to bind only to LPA, a substrate of LPAAT, but not to any other lipid tested (). In contrast, n-OG–soluble Pex16p of mature peroxisomes P6 did not bind to LPA if these peroxisomes were recovered from wild-type or , , and mutant strains (). All these strains lack LPA and carry both PA and DAG in the membranes of their division-competent mature peroxisomes ( and ). Of note, Pex16p is attached to the membranes of immature peroxisomal vesicles only in its free form, whereas all the Pex16p on the inner face of mature peroxisomes of wild-type or , , and mutant cells is titrated by its interaction with Aox (). Importantly, the interaction between Pex16p and Aox is not affected by n-OG. Altogether, these data suggest that the binding of Aox to Pex16p in mature peroxisomes of wild-type cells greatly decreases the affinity between Pex16p and LPA, thereby allowing LPA to enter the two-step biosynthetic pathway leading to the formation of PA and DAG. This hypothesis is supported by the observation that n-OG–soluble Pex16p of mature peroxisomes was capable of binding to LPA if these mature peroxisomes were purified from , , or strains (). All these mutant strains carry Pex16p in a free form that is not titrated by its interaction with Aox, are deficient in the division of mature peroxisomes, and accumulate a reduced number of greatly enlarged mature peroxisomes () that contain LPA but lack both PA and DAG ( and ). Our data suggest that LPA enters the two-step pathway for the biosynthesis of PA and DAG () only when the efficiency of its binding to Pex16p declines. Pex16p is a peripheral membrane protein that is attached only to the lumenal leaflet of the peroxisomal membrane (). Furthermore, it seems unlikely that LPA can translocate from the lumenal to the cytosolic leaflet of the peroxisomal membrane, as its spontaneous transbilayer movement is very slow (). Moreover, neither LPAAT nor PAP faces the cytosol, being integrated into the lumenal leaflet of the peroxisomal membrane (). Altogether, these findings imply that the biosynthesis of PA and DAG is spatially restricted to the lumenal leaflet of the peroxisomal membrane. To evaluate the arrangement of DAG between the two leaflets of the membrane bilayers in different peroxisomal subforms, we reconstituted two types of resealed peroxisomes, termed RPA and RPB, from osmotically lysed intact peroxisomes. RPA were reconstituted in a MES-based buffer at pH 5.5, whereas RPB was made in a Hepes-based buffer at pH 7.5. Similar to intact peroxisomes (), both RPA and RPB could float out of the most dense sucrose during centrifugation to equilibrium in sucrose density gradients (Fig. S3 A, available at ) and were bound by a single membrane (Fig. S3 C). In intact peroxisomes, Pex19p is a peripheral membrane protein that resides on the cytosolic face of the peroxisome, whereas the peripheral membrane protein Pex16p is attached to its lumenal face (Fig. S3, B and D). In RPA, most of Pex19p, but only a minor portion of Pex16p, was accessible to trypsin and to the corresponding antigen-specific IgG molecules exogenously added to this type of resealed peroxisomes (Fig. S3, B and D). Thus, the membrane delimiting most of the RPA species formed during peroxisome resealing was present in the outside-out orientation, whereas only a minor fraction of RPA species had their membrane resealed in the inside-out orientation. In contrast, in RPB, only a minor portion of Pex19p, but most of Pex16p, was accessible to trypsin and to the corresponding antigen-specific IgG molecules exogenously added to this type of resealed peroxisomes (Fig. S3, B and D). Hence, only a minor fraction of RPBs had their membrane resealed in the outside-out orientation, whereas the membrane delimiting most of the RPB species formed during peroxisome resealing was present in the inside-out orientation. Using a Pex19p-specific fluorescent probe, we calculated the percentages of outside-out– and inside-out–oriented species of RPA and RPB that were formed by resealing of osmotically lysed peroxisomal subforms P1–P6 (Fig. S3, E and F). The ability to calculate the percentage of outside-out– and inside-out–oriented species of RPA and RPB allowed us to calculate the percentage of DAG residing in the cytosolic and lumenal leaflets of the membrane bilayers in intact peroxisomes. The DAG-binding C1b domain of protein kinase C () labeled with the fluorophore Alexa Fluor 488 was used as a DAG-specific fluorescent probe. In intact P5, only 13 ± 4% of the total pool of DAG was detected in the cytosolic leaflet of the membrane bilayer (Fig. S3 H). Thus, DAG resides predominantly in the lumenal membrane leaflet of P5. In contrast, DAG is distributed symmetrically between the two leaflets of the membrane bilayer in mature peroxisomes P6. In fact, 57 ± 3% of this lipid resided in the cytosolic membrane leaflet of P6 (Fig. S3 H). We then used monoclonal antibodies to PS, a lipid that has a cylindrical shape (), to monitor its transbilayer distribution in the membranes of different peroxisomal subforms. PS in the membranes of immature peroxisomal vesicles P1–P3 resides predominantly in their cytosolic leaflets (Fig. S3 H). As peroxisomes mature, PS gradually moves from the cytosolic to the lumenal leaflets of their membranes. Indeed, only 15 ± 1% of this lipid resided in the cytosolic membrane leaflet of P6 (Fig. S3 H). In summary, the assembly of mature peroxisomes promotes the specific redistribution of DAG and PS between the two leaflets of the peroxisomal membrane. The movement of DAG from the lumenal to the cytosolic leaflet of the membrane bilayer coincides with the translocation of PS in the opposite direction. The levels of PC, a major glycerophospholipid of the peroxisomal membrane (), in P4, P5, and P6 peroxisomes of wild-type cells were substantially higher than in P1, P2, and P3 peroxisomes (). The observed increase in the levels of PC was not due to its de novo synthesis. In fact, the membranes of P3 and P4 did not contain PA and DAG (), two substrates for PC biosynthesis via the phosphatidylethanolamine (PE) methylation and cytidine diphosphate–choline pathways, respectively (). Thus, PC is transported to the membranes of P3 and P4 during their conversion to P4 and P5, respectively. Three established mechanisms of intracellular lipid transport to organellar membranes include transport catalyzed by cytosolic lipid transfer proteins (; ), vesicle-mediated transport (; ), and transport at regions of close apposition between specialized microdomains of the ER membrane and the membranes of the trans-Golgi or mitochondria (; ). Our data imply that PC is transferred from the donor membrane of a distinct subcompartment of the ER to the acceptor membranes of P3 and P4 associated with this subcompartment and that this transfer of PC requires the peroxisome-associated peroxin Pex2p, provides membranes of P3 and P4 with the bulk quantities of PC, and is essential for the conversion of P4 to P5. This hypothesis is based on the following findings. First, a distinct form of the ER copurifies with P3 and P4 peroxisomes and can be separated from them by treatment with EDTA (, ). Second, the P3- and P4-associated ER subcompartment can be distinguished from the free form of the ER by buoyant density and the total level of membrane glycerophospholipids (), as well as by protein spectrum (Fig. S4 B, available at ). Third, the mutation increases the levels of membrane glycerophospholipids in the P3- and P4-associated subcompartment of the ER (), substantially decreases the level of PC in P4 (Fig. S4 A), and impairs its conversion to P5 (unpublished data). It seems that PC in the peroxisomal membrane is a positive regulator of both LPAAT and PAP. In fact, the specific activities of these two membrane-bound enzymes in liposomes reconstituted from the Pex16p-immunodepleted PMPs and membrane lipids of P1, P2, and P3 were substantially lower than in liposomes reconstituted from membrane components of P4, P5, and P6 (). Of note, LPAAT and PAP activities detected in the membranes of these peroxisomal liposomes were proportional to the steady-state levels of PC recovered in these membranes (). Importantly, the positive effect of PC on both LPAAT and PAP could be reconstructed in four different types of the Pex16p-immunodepleted liposomes that were reconstituted from membrane components of P1, P2, or P3 and varied only in the quantities of PC present in their membranes (). Noteworthy, by rising the quantities of PC in the membranes of P1-, P2-, and P3-based liposomes to the levels comparable to those present in the membranes of P4-, P5-, and P6-based liposomes, both LPAAT and PAP could be substantially stimulated, matching their enzymatic activities in liposomes reconstituted from membrane components of P4, P5, and P6 (). Considering that all six peroxisomal subforms have similar amounts of both LPAAT and PAP (), the findings reported in support the notion that PC in the peroxisomal membrane activates these two enzymes. The protein Vps1p is essential for peroxisome division (). Vps1p is a member of the dynamin protein superfamily of large GTPases that carry out a broad range of functions, including organelle division and fusion, budding of transport vesicles, and cytokinesis (). Akin to its counterpart, Vps1p is required for peroxisome division. In fact, lack of this protein resulted in a reduced number of greatly enlarged peroxisomes (Fig. S2, A and D). Like most peroxisomes of wild-type cells, the majority of peroxisomes of cells could be pelleted by centrifugation at 20,000 . These 20,000- pelletable peroxisomes of cells were very similar to mature peroxisomes purified from wild-type cells in regards to buoyant density, spectra of matrix and membrane proteins, and lipid composition of their membranes (unpublished data). Morphometric analysis of random electron sections further confirmed that lack of Vps1p impairs the ability of completely assembled peroxisomes to divide, resulting in fewer, but greatly enlarged, mature peroxisomes (Fig. S2, G and H). Vps1p is mainly a cytosolic protein (). It can also be found in a variety of cellular locations, including the Golgi, peroxisomes, and vacuoles (; ; ). Likewise, most of Vps1p localized to the cytosol, whereas the minor portion of it was associated with both low-speed (20,000 ) and high-speed (200,000 ) pelletable organelles (Fig. S5 A, available at ). Using highly purified peroxisomal subforms of wild-type cells, we found that Vps1p was only present in division-competent mature peroxisomes (Fig. S5 B). In contrast, the division-incompetent immature peroxisomal vesicles P1–P5 lacked Vps1p (Fig. S5 B). The P6-associated form of Vps1p was solubilized completely by either 1 M NaCl or 0.1 M NaCO, pH 11.0, whereas the peroxisomal integral membrane protein Pex2p () was not (Fig. S5 C). Thus, Vps1p is a peripheral membrane protein. Furthermore, Vps1p of mature peroxisomes was digested by trypsin even in the absence of the detergent Triton X-100, whereas the membrane-enclosed protein thiolase was resistant to digestion by external protease added to intact peroxisomes (Fig. S5 D). Altogether, these data imply that the conversion of P5 to P6 in wild-type cells is marked by the recruitment of Vps1p from the cytosol to the surface of mature peroxisomes. Importantly, Vps1p was bound to division-competent mature peroxisomes of wild-type or , , and mutant strains (Fig. S5 E). In the membranes of mature peroxisomes of all these strains, LPA was converted to PA and DAG (). In contrast, Vps1p was not attached to mature peroxisomes of , , or mutant strains (Fig. S5 E). All these strains are deficient in the division of mature peroxisomes (), being unable to convert LPA to PA and DAG in the peroxisomal membrane ( and ). These findings suggest that the recruitment of Vps1p from the cytosol to the surface of mature peroxisomes relies on the Pex16p/Aox-dependent biosynthesis of PA and DAG in their membranes. To test whether Vps1p interacts with other components of the peroxisomal membrane, membrane proteins recovered after centrifugation of osmotically lysed mature peroxisomes of wild-type cells were treated with the thiol-cleavable cross-linker dithiobis(succinimidylpropionate) (DSP). These membrane proteins were immunoprecipitated with anti-Vps1p antibodies under denaturing, nonreducing conditions. The cross-linker was then cleaved with DTT, and the immunoprecipitated proteins were resolved by SDS-PAGE under reducing conditions, followed by silver staining. A cohort of proteins was specifically coimmunoprecipitated with Vps1p under these conditions (, lane 1), suggesting the existence of a Vps1p-containing complex on the outer face of the peroxisomal membrane. The following six components of this complex were identified by mass spectrometry: Vps1p, a dynamin-like GTPase that is required for the division of mature peroxisomes (see the previous section); Sla1p, a protein that regulates actin cytoskeleton organization and dynamics (); Abp1p, a protein that promotes F-actin assembly (); Act1p, a structural constituent of actin cytoskeleton in yeast (); the peroxin Pex19p, a protein required for the import and/or membrane assembly of numerous PMPs (; ); and the peroxin Pex10p, an integral PMP required for peroxisomal matrix protein import (). Importantly, antibodies specific to Pex10p and Pex19p, the two components of the Vps1p-containing complex, immunoprecipitated the same set of DSP-treated PMPs as anti-Vps1p antibodies did (, compare lanes 1, 3, and 7). Thus, all Vps1p, Sla1p, Abp1p, Act1p, Pex19p, and Pex10p form a single multicomponent complex and do not compose several subcomplexes formed by the association of a bait protein (i.e., Vps1p, Pex10p, or Pex19p) with different subsets of interacting protein partners. The Pex10p and Pex19p components of the Vps1p-containing complex were associated with all six peroxisomal subforms, whereas Vps1p itself was attached only to mature peroxisomes P6 (Fig. S5 B). Furthermore, lack of either Pex10p or Pex19p abolished the recruitment of Vps1p from the cytosol to the membrane of P6 (). Moreover, in the absence of Vps1p, none of the actin cytoskeleton–related components of the Vps1p-containing complex, including Sla1p, Abp1p, and Act1p, was bound to P6 (, lanes 4 and 8). Altogether, these findings imply that the Pex10p- and Pex19p-dependent recruitment of Vps1p from the cytosol to the surface of the mature peroxisome is mandatory for the attachment of Sla1p, Abp1p, and Act1p to this division-competent peroxisomal subform. The Sla1p and Abp1p components of the Vps1p-containing complex, but not its Act1p component, coimmunoprecipitated with Vps1p from the DSP-treated cytosolic fractions of wild-type, , and cells (, lanes 1, 3, and 4). Act1p coimmunoprecipitated with Vps1p, Sla1p, and Abp1p only if all these proteins were attached to the surface of mature peroxisomes (, lanes 1, 3, and 7). Hence, Vps1p, Sla1p, and Abp1p initially form a complex in the cytosol. This complex is then targeted from the cytosol to the surface of mature peroxisomes. Only after its binding to mature peroxisomes, the Vps1p–Sla1p–Abp1p complex is able to promote the attachment of Act1p to the peroxisomal membrane. Akin to Vps1p (see the previous section), each of the two other components of the Vps1p–Sla1p–Abp1p complex is required for peroxisome division. In fact, lack of either Sla1p or Abp1p resulted in a reduced number of greatly enlarged peroxisomes (Fig. S2, E–H). The Pex10p component of the Vps1p-containing complex is an integral PMP that could not be stripped off the peroxisomal membrane by either 1 M NaCl or 0.1 M NaCO, pH 11.0 (Fig. S5 C). On the contrary, its Pex19p component is a peripheral membrane protein on the outer face of peroxisomes that could be solubilized by either 1 M NaCl or 0.1 M NaCO, pH 11.0, and was sensitive to digestion by external protease added to intact peroxisomes (Fig. S5, C and D). Pex10p and Pex19p form a complex in the membrane of mature peroxisomes (, lanes 4 and 8) and of immature peroxisomal vesicles P1–P5 (not depicted). Thus, the Pex10p–Pex19p complex assembles in the peroxisomal membrane during the initial steps of the peroxisome assembly pathway. The attachment of Pex19p to the peroxisomal membrane requires Pex10p, as lack of Pex10p abolished the recruitment of Pex19p from the cytosol to the peroxisome (). Of note, Pex19p that was accumulated in the cytosol of cells did not coimmunoprecipitate with Vps1p, Sla1p, Abp1p, or Act1p (, lane 6). Furthermore, the Vps1p–Sla1p–Abp1p complex could be formed even in the cytosol of cells (, lane 4). These findings support the notion that, although cytosolic Pex19p is not required for the assembly of the Vps1p–Sla1p–Abp1p complex before its recruitment to the membrane, the membrane-bound form of Pex19p is mandatory for the attachment of the preformed Vps1p–Sla1p–Abp1p complex to the surface of mature peroxisomes. In summary, our data suggest that the assembly of the Vps1p–Sla1p–Abp1p complex in the cytosol precedes its attachment to the surface of division-competent mature peroxisomes P6 (). The Vps1p–Sla1p–Abp1p complex binds to P6 by interacting with Pex19p, a component of the Pex10p–Pex19p complex that is formed in the peroxisomal membrane during the earliest steps of peroxisome assembly and maturation. Only after it has been attached to the membrane of P6 is the Vps1p–Sla1p–Abp1p complex able to interact with Act1p, thereby promoting the recruitment of actin to the surface of these division-competent peroxisomes. This study and our published data () suggest the following model for peroxisome division in (). In immature peroxisomal vesicles P1–P5, Pex16p binds LPA in the lumenal leaflet of the peroxisomal membrane. The binding of Pex16p to LPA prevents the biosynthesis of PA and DAG in a two-step pathway, which includes two consecutive enzymatic reactions catalyzed by Slc1p (LPAAT) and Dpp1p (PAP; ). The stepwise import of distinct subsets of matrix proteins into immature peroxisomal vesicles P1–P5 provides them with an increasing fraction of the matrix proteins present in mature peroxisomes. The increase in the total mass of matrix proteins above a critical level, which occurs only inside mature peroxisomes, causes the redistribution of Aox from the matrix to the membrane and its subsequent binding to Pex16p. This, in turn, greatly decreases the affinity between Pex16p and LPA, thereby allowing LPA to enter the two-step biosynthetic pathway leading to the formation of PA and DAG. The glycerophospholipid PC, which is transferred to the peroxisomal membrane from the P3- and P4-associated subcompartment of the ER, activates both LPAAT and PAP. The resulting accumulation of PA and DAG in the lumenal leaflet of the membrane of mature peroxisomes triggers a cascade of events ultimately leading to peroxisome division. This cascade of events is initiated by the spontaneous flipping of DAG, which is known for its very fast transbilayer translocation, between the two membrane leaflets. The movement of DAG, a particularly potent cone-shaped inducer of membrane bending, from the lumenal to the cytosolic leaflet of the membrane bilayer coincides with the translocation of the glycerophospholipid PS in the opposite direction. This bidirectional movement of DAG and PS generates a lipid imbalance across the bilayer, which may promote the destabilization and bending of the membrane. The biosynthesis of PA and DAG in the membrane of mature peroxisomes and, perhaps, the bending of the membrane because of the bidirectional transbilayer movement of DAG and PS promote the docking of the Vps1p–Sla1p–Abp1p complex to the surface of mature peroxisomes. This preassembled in the cytosol protein complex binds to mature peroxisomes by interacting with the peroxin Pex19p. Pex19p is a component of the Pex10p–Pex19p complex that is formed in the peroxisomal membrane during the earliest steps of peroxisome assembly (). After its attachment to the peroxisomal membrane, the Vps1p–Sla1p–Abp1p complex interacts with Act1p, thereby recruiting this structural constituent of actin cytoskeleton to the surface of mature peroxisomes. The subsequent fission of the peroxisomal membrane leads to peroxisome division. It remains to be established how exactly Vps1p promotes peroxisome division. Initially, this dynamin-like GTPase interacts in the cytosol with Sla1p and Abp1p. Vps1p then functions in the attachment of the Vps1p–Sla1p–Abp1p protein complex to its docking factor Pex19p on the surface of mature peroxisomes, thereby promoting the subsequent recruitment of actin to the membrane of division-competent peroxisomes. Therefore, it seems unlikely that Vps1p acts only as a mechanochemical enzyme () whose GTPase activity provides the mechanical force required for membrane fission in the constricted neck. Our data suggest that this dynamin-like protein may rather function as a regulatory GTPase () whose GTP-bound form promotes the multistep assembly of the membrane fission machinery, initially in the cytosol and then on the surface of division-competent mature peroxisomes. This machinery includes the Sla1p, Abp1p, and Act1p components of actin cytoskeleton. The mechanism by which actin cytoskeleton regulates the terminal step of peroxisome division is currently being investigated. Similar to mitotic Golgi fragmentation (; ) and mitochondrial division during apoptosis (), peroxisome division is served by a protein team that is assembled on the peroxisomal surface in a stepwise fashion. The multicomponent protein machineries serving Golgi fragmentation and mitochondrial division are assembled in response to extraorganellar stimuli (; ). In contrast, the protein team that executes peroxisome division undergoes multistep assembly in response to an intraperoxisomal signaling cascade (). Although this Pex16p- and Aox-dependent signaling cascade is turned off inside immature peroxisomal vesicles, it is activated inside mature peroxisomes. Thus, it seems likely that the intraperoxisomal cascade for fine-tuning the fission of peroxisomal membrane is an intrinsic feature of the multistep peroxisome biogenesis program. Perhaps this program has evolved to separate the dramatic changes in the composition and architectural design of the membrane bilayer, all of which occur during peroxisome division, from the process of protein translocation across this bilayer, which takes place during peroxisome assembly. One of the benefits of using such strategy for the temporal separation of the processes of peroxisome assembly and division is that some of the membrane components can efficiently function in both processes. In fact, the peroxins Pex10p and Pex19p, known for their essential role in peroxisomal import of numerous matrix proteins and PMPs (), are also required for the assembly of the peroxisome division machinery on the surface of mature peroxisomes (). Our findings support the notion that a distinct set of lipid metabolic pathways operating in organellar membranes and specific changes in the distribution of some lipids across the membrane bilayers provide a driving force for organelle division (; ; ; ; ; ; ). It is tempting to speculate that, after its spontaneous flipping between the two leaflets of the peroxisomal membrane (), DAG undergoes the selective enrichment in distinct lipid domains that facilitate membrane fission through coordinated changes in local membrane curvature, initiate the assembly of the Vps1p-containing protein complexes on the surface of peroxisomes, and promote the clustering of these protein complexes at the membrane fission site. A challenge for the future will be to define the spatial distribution of DAG and Vps1p-containing protein complexes in the membrane of division-competent mature peroxisomes. The wild-type strain (); the mutant strains (), (), , and (); the single gene knock-out strains (); and the media, growth conditions, and genetic techniques for () have been previously described. Targeted integrative disruption of the , , , , , and genes was performed with the gene of , using a previously described modification of the sticky-end polymerase chain reaction procedure (). J.-M. Nicaud (Laboratoire de Microbiologie et de Génétique Moléculaire, Thiverval- Grignon, France) provided the and mutant strains. Antibodies to Pex2p (), Pex16p (), Pex19p (), and thiolase () have been previously described and were provided by R.A. Rachubinski (University of Alberta, Edmonton, Canada). Monospecific antibodies to Dpp1p, Pex10p, Slc1p, and Vps1p were raised in rabbit against their peptides GAPRPDMLARCRPMSWMRP, CRQGVREQNLLPIR, GRIFPQYCSVTAKKALKWYP, and MDKELISTVNKLQDALA, respectively. Purification of the DAG-binding C1b domain of protein kinase C () and its labeling with the fluorophore Alexa Fluor 488 () were performed as described previously. The GST-C1B vector was provided by A.C. Newton (University of California at San Diego, La Jolla, CA). SDS-PAGE and immunoblotting () were performed as described. Cholic acid (sodium salt), ergosterol, hydroxylapatite, n-OG, palmitoyl-CoA agarose, and Triton X-100 were purchased from Sigma-Aldrich. PIP Strips were obtained from Echelon Biosciences. Alexa Fluor 488 signal-amplification kit for fluorescein-conjugated probes was purchased from Invitrogen. Monoclonal anti-PS antibody was purchased from Upstate Biotechnology. Fluorescein-conjugated goat anti–rabbit IgG antibodies and fluorescein-conjugated goat anti–mouse IgM antibodies were obtained from Jackson ImmunoResearch Laboratories. -palmitoyl---sphingosine (ceramide), 1,2-dioleoyl--glycerol (DAG), 1,2-dioleoyl--glycero-3-phosphate (PA), 1,2-dioleoyl--glycero-3-phosphocholine (PC), 1-oleoyl-2-hydroxy--glycero-3-phosphate (LPA), 1,2-dioleoyl--glycero-3-phosphoethanolamine (PE), -α-phosphatidylinositol (PI), and 1,2-dioleoyl--glycero-3-(phospho--serine) (PS) were obtained from Avanti Polar Lipids, Inc. [C]-labeled lipids, HiTrap Blue HP, Resource Q, Resource S, and Superose 12 were obtained from GE Healthcare. The initial step in the subcellular fractionation of oleic acid-grown cells included the differential centrifugation of lysed and homogenized spheroplasts at 1,000 for 10 min at 4°C in a JS13.1 rotor (Beckman Coulter) to yield a postnuclear supernatant fraction. The postnuclear supernatant fraction was further subjected to differential centrifugation at 20,000 for 30 min at 4°C in a JS13.1 rotor to yield pellet (20KgP) and supernatant (20KgS) fractions. The 20KgS fraction was further subfractionated by differential centrifugation at 200,000 for 1 h at 4°C in a TLA110 rotor (Beckman Coulter) to yield pellet (200KgP) and supernatant (200KgS) fractions. To purify immature peroxisomal vesicles P1–P5, the 200KgP subcellular fraction was subjected to centrifugation on a discontinuous sucrose (18, 25, 30, 35, 40, and 53%; wt/wt) gradient at 120,000 for 18 h at 4°C in a SW28 rotor (Beckman Coulter). 36 fractions of 1 ml each were collected. Different subforms of immature peroxisomal vesicles peaked at densities of 1.18 g/cm (fraction 5; P5), 1.14 g/cm (fraction 15; P3 + P4), 1.11 g/cm (fraction 23; P1), and 1.09 g/cm (fraction 30; P2) were recovered (). The peak fractions containing immature peroxisomal vesicles P1, P2, P3 + P4, and P5 were recovered, and 4 vol of 0.5 M sucrose in buffer H (5 mM MES, pH 5.5, 1 mM KCl, 0.5 mM EDTA, 0.1% ethanol, and 1× protease inhibitor cocktail []) were added to each of them. Peroxisomes were pelleted onto a 150-μl cushion of 2 M sucrose in buffer H by centrifugation at 200,000 for 20 min at 4°C in a TLA110 rotor. Individual pellets of different subforms of immature peroxisomal vesicles were resuspended in 3 ml of 50% (wt/wt) sucrose in buffer H. For purification of immature peroxisomal vesicles P1 and P2, pellets of P1 and P2 resuspended in 50% (wt/wt) sucrose in buffer H were overlaid with 30, 28, 26, 24, 22, and 10% sucrose (all wt/wt in buffer H). After centrifugation at 120,000 for 18 h at 4°C in a SW28 rotor, 18 fractions of 2 ml each were collected. P1 and P2 were pelleted, resuspended, and subjected to a second flotation on the same multistep sucrose gradient. Gradients were fractionated into 2-ml fractions, and P1 and P2 were recovered () and used for biochemical analyses. For purification of immature peroxisomal vesicles P3 and P4, pellets of P3 and P4 resuspended in 50% (wt/wt) sucrose in buffer H were overlaid with 38, 35, 33, and 20% sucrose (all wt/wt in buffer H). After centrifugation at 120,000 for 18 h at 4°C in a SW28 rotor, 18 fractions of 2 ml each were collected. P3 and P4 were pelleted, resuspended in 3 ml of 50% (wt/wt) sucrose in buffer HE (20 mM MES, pH 5.5, 20 mM EDTA, and 0.1% ethanol), overlaid with 39, 37, 35, 33, and 20% sucrose (all wt/wt in buffer HE), and subjected to centrifugation as described. Gradients were fractionated into 2-ml fractions, and P3 and P4 were recovered and pelleted. After resuspension in 3 ml of 50% (wt/wt) sucrose in buffer H, P3 and P4 were again subjected to flotation on the second multistep sucrose gradient described. Gradients were fractionated into 2-ml fractions, and P3 and P4 were recovered () and used for biochemical analyses. Highly purified mature peroxisomes P6 were isolated from the 20KgP subcellular fraction by isopycnic centrifugation on a discontinuous sucrose gradient as described previously (). 4 vol of 0.5 M sucrose in buffer H were added to the peak peroxisomal fraction 4 recovered after isopycnic centrifugation on a discontinuous sucrose gradient. Peroxisomes were sedimented through a 150-μl cushion of 2 M sucrose in buffer H by centrifugation at 200,000 for 20 min at 4°C in a TLA110 rotor. The resultant pellet of mature peroxisomes P6 was resuspended in buffer H containing 1 M sorbitol and was subjected to further centrifugation on a linear 20–60% (wt/wt) sucrose gradient (in buffer H) at 197,000 for 18 h at 4°C in a SW41Ti rotor (Beckman Coulter). Peak peroxisomal fraction 5 equilibrating at a density of 1.21 g/cm was recovered, and peroxisomes were sedimented through a 150-μl cushion of 2 M sucrose in buffer H by centrifugation at 200,000 for 20 min at 4°C in a TLA110 rotor. Pellet of mature peroxisomes P6 was resuspended in 55% (wt/wt) sucrose in buffer H, overlaid with 50, 45, 40, 30, and 20% sucrose (all wt/wt in buffer H), and subjected to centrifugation at 120,000 for 18 h at 4°C in a SW28 rotor. 18 gradient fractions of 2 ml each were collected. Peak peroxisomal fraction 11 equilibrating at a density of 1.21 g/cm was recovered () and used for biochemical analyses. The free form of the ER () and the P3- and P4-associated subcompartment of the ER () were purified from cells as described previously. Subcellular fractionation of cells grown in glucose-containing YEPD medium and isolation of functional ER membranes were performed according to established procedures (). Highly purified peroxisomes were lysed by the addition of 10 vol of ice-cold LB buffer (20 mM Hepes-KOH, pH 8.0, 50 mM NaCl, and 1× protease inhibitor cocktail), followed by incubation on ice for 30 min with occasional agitation. The suspension was centrifuged at 200,000 for 20 min at 4°C in a TLA110 rotor. The pellet of membranes recovered after centrifugation of osmotically lysed peroxisomes was resuspended in ice-cold EB buffer (10 mM Hepes-KOH, pH 8.0, 5 mM EDTA, and 1× protease inhibitor cocktail) to a final concentration of 1.0 mg/ml. Equal aliquots of the suspension of membranes were then exposed to 1 M NaCl, 0.1 M NaCO, pH 11.0, or 0.5% (vol/vol) Triton X-100 (). After incubation on ice for 30 min with occasional agitation, the samples were subjected to centrifugation at 100,000 for 30 min at 4°C in a TLA110 rotor. Equal portions of the pellet and supernatant fractions were analyzed by SDS-PAGE, followed by immunoblotting. The pellet of highly purified mature peroxisomes was gently resuspended in ice-cold PPB buffer (5 mM MES, pH 5.5, 1 M sorbitol, 1 mM KCl, and 0.5 mM EDTA). Equal aliquots (10 or 20 μg of total protein) of these peroxisomes were incubated with 0, 5, 10, or 50 μg trypsin for 30 min on ice, either in the presence or absence of Triton X-100 at 0.5% (vol/vol) final concentration. The reaction was terminated by the addition of trichloroacetic acid to 10% final concentration. The protein precipitates were washed with ice-cold 80% (vol/vol) acetone, and equivalent fractions of each reaction were subjected to SDS-PAGE and immunoblotting. Highly purified peroxisomes were lysed by the addition of 10 vol of ice-cold LB buffer, followed by incubation on ice for 30 min with occasional agitation. The suspension was centrifuged at 200,000 for 20 min at 4°C in a TLA110 rotor. The recovered pellet of membranes that contained 1 mg of membrane protein was resuspended in 1.0 ml of chloroform/methanol (1:1; vol/vol). After incubation on ice for 15 min with occasional agitation, samples were subjected to centrifugation at 20,000 for 15 min at 4°C. The chloroform phase was separated and dried under nitrogen. The lipid film was dissolved in 100 μl of chloroform (for the analysis of DAG, ergosterol, and ceramide) or 100 μl of chloroform/methanol (1:1 [vol/vol]; for the analysis of PE, PA, PC, PI, PS, and LPA). 25 μl of each sample were spotted on 60-Å silica gel plates for TLC (Whatman). The lipids were developed in the following solvent systems: chloroform/acetone (4.6:0.4 [vol/vol]; for the analysis of DAG, ergosterol, and ceramide) and chloroform/methanol/water (65:25:4 [vol/vol]; for the analysis of PE, PA, PC, PI, PS, and LPA). All lipids were detected using 5% phosphomolybdic acid in ethanol and visualized by heating for 30 min at 110°C. Lipids were quantitated by densitometric analysis of TLC plates as described previously (), using lipid standards in the 0.1–0.5 μg range for calibration. For monitoring enzymatic activities of LPAAT and PAP, highly purified peroxisomes were lysed by the addition of 10 vol of ice-cold LB buffer, followed by incubation on ice for 30 min with occasional agitation. The suspension was centrifuged at 200,000 for 20 min at 4°C in a TLA110 rotor. The pellet of membranes recovered after centrifugation of osmotically lysed peroxisomes was resuspended in ice-cold buffer R (20 mM MES-KOH, pH 6.0, 150 mM NaCl, 5 mM DTT, and 10% glycerol) containing 1% (wt/vol) n-OG. After incubation on ice for 20 min with occasional agitation, the sample of detergent-solubilized PMPs was subjected to centrifugation at 100,000 for 20 min at 4°C in a TLA110 rotor. The resulting supernatant of solubilized PMPs was depleted of Pex16p by immunoaffinity chromatography under native conditions using anti-Pex16p antibodies covalently linked to protein A–Sepharose (). For the reconstitution of peroxisomal liposomes carrying Pex16p, detergent-solubilized PMPs immunodepleted of Pex16p were supplemented with Pex16p, which was purified from membranes of osmotically lysed immature peroxisomal vesicles P1 by immunoaffinity chromatography under native conditions using anti-Pex16p antibodies covalently linked to protein A–Sepharose (). After elution with buffer E (20 mM Hepes-KOH, pH 7.5, 250 mM MgCl, 5 mM DTT, and 10% glycerol) containing 1% (wt/vol) n-OG, purified Pex16p was dialyzed against buffer R supplemented with 1% (wt/vol) n-OG. For the reconstitution of peroxisomal liposomes lacking Pex16p, detergent-solubilized PMPs immunodepleted of Pex16p were supplemented only with buffer R containing 1% (wt/vol) n-OG. Detergent-solubilized PMPs immunodepleted of Pex16p and either supplemented or not supplemented with purified Pex16p in buffer R containing 1% (wt/vol) n-OG were then added to the films of unlabeled lipids, which were initially extracted from the membranes of highly purified peroxisomes using chloroform/methanol (1:1; vol/vol) and then dried down by a gentle stream of nitrogen. The lipid films were dissolved by gentle agitation for 20 min at room temperature. For monitoring LPAAT activity, the unlabeled lipids, which were extracted from the membranes of highly purified peroxisomes using chloroform/methanol (1:1; vol/vol), were supplemented with [C]-labeled LPA and unlabeled oleoyl-CoA (a cosubstrate of LPAAT) dissolved in chloroform/methanol (1:1; vol/vol). The mix of unlabeled membrane lipids and [C]-labeled LPA was then dried down by a gentle stream of nitrogen. For monitoring PAP activity, the unlabeled lipids, which were extracted from the membranes of highly purified peroxisomes using chloroform/methanol (1:1; vol/vol), were supplemented with [C]-labeled PA dissolved in chloroform/methanol (1:1; vol/vol). The mix of unlabeled membrane lipids and [C]-labeled PA was then dried down by a gentle stream of nitrogen. For evaluating the positive effect of PC on LPAAT and PAP, equal aliquots of unlabeled lipids extracted from the membranes of highly purified peroxisomes using chloroform/methanol (1:1; vol/vol) were first mixed with an appropriate [C]-labeled lipid substrate of LPAAT or PAP in chloroform/methanol (1:1; vol/vol) and were then supplemented with various quantities of PC dissolved in chloroform/methanol (1:1; vol/vol). The mix of unlabeled membrane lipids, a [C]-labeled lipid substrate, and unlabeled PC was then dried down by a gentle stream of nitrogen. The lipid films were finally dissolved by gentle agitation for 20 min at room temperature in buffer R containing detergent-solubilized PMPs, immunodepleted or not immunodepleted of Pex16p, in 1% (wt/vol) n-OG. To dilute the detergent n-OG below its critical micellar concentration, thereby promoting the formation of peroxisomal liposomes, 3 vol of buffer D (20 mM MES-KOH, pH 6.0, and 150 mM NaCl) was added to the mixture of detergent-solubilized PMPs, and membrane lipids were dissolved in buffer R containing 1% (wt/vol) n-OG. To remove the detergent, the samples were dialyzed in a Tube-O-Dialyzer (7.5-kD cutoff; Chemicon) against buffer D containing 0.1% Biobeads SM2 (Bio-Rad Laboratories). After overnight dialysis at 4°C, samples were transferred to the bottom of ultraclear centrifuge tubes (Beckman Coulter) and supplemented with 4 vol of 65% (wt/wt) sucrose in buffer D in order to adjust the sucrose concentration of the samples to 52% (wt/wt). Samples were overlaid with 40% and then with 20% sucrose (both wt/wt in buffer D) and, lastly, with buffer D alone. After centrifugation at 200,000 for 18 h at 4°C in a SW50.1 rotor (Beckman Coulter), 18 fractions of 275 μl each were collected. Peroxisomal liposomes were recovered at the 40%/20% sucrose interface. The recovered peroxisomal liposomes were transferred from ice to 26C. Samples were taken at the indicated times after the transfer. Lipids were extracted from the membrane and analyzed by TLC. To calculate the initial rates of the LPAAT and PAP reactions, the [C]-labeled LPA, PA, and DAG were separated by TLC and quantified by autoradiography. To evaluate the transbilayer distribution of DAG and PS in the membrane bilayers of different peroxisomal subforms, the suspension of highly purified peroxisomes in ice-cold H250S buffer (5 mM MES-KOH, pH 5.5, 250 mM sorbitol, 1 mM KCl, 0.5 mM EDTA, and 1× protease inhibitor cocktail) at 1 mg protein/ml was divided into two equal aliquots. One aliquot remained untreated, whereas peroxisomal vesicles in the other aliquot were lysed by the addition of 10 vol of ice-cold LB buffer, followed by incubation on ice for 30 min with occasional agitation. The suspension of lysed peroxisomes was divided into two equal aliquots. One aliquot was dialyzed in a Tube-O-Dialyzer (7.5-kD cutoff) against buffer MR (10 mM MES/KOH, pH 5.5, 1 mM KCl, and 0.5 mM EDTA) containing 250 mM sorbitol. The suspension of lysed peroxisomes in the other aliquot was dialyzed in a Tube-O-Dialyzer (7.5-kD cutoff) against buffer HR (10 mM Hepes/KOH, pH 7.5, 1 mM KCl, and 0.5 mM EDTA) containing 250 mM sorbitol. After overnight dialysis at 4°C, resealed peroxisomes RPA that were formed in the aliquot dialyzed against buffer MR containing 250 mM sorbitol and resealed peroxisomes RPB that were formed in the aliquot dialyzed against buffer HR containing 250 mM sorbitol were pelleted onto a 150-μl cushion of 2 M sucrose in buffer MR or HR, respectively, by centrifugation at 100,000 for 20 min at 4°C in a TLA110 rotor. Individual pellets of RPA and RPB were resuspended in 500 μl of 50% (wt/wt) sucrose in buffer MR or HR, respectively. The sample containing RPA was overlaid with 1.5 ml of 45% sucrose, 1 ml of 40% sucrose, 1 ml of 25% sucrose, and 1 ml of 10% sucrose (all wt/wt in buffer MR). The sample containing RPB was overlaid with 1.5 ml of 45% sucrose, 1 ml of 40% sucrose, 1 ml of 25% sucrose, and 1 ml of 10% sucrose (all wt/wt in buffer HR). Both samples were subjected to centrifugation at 200,000 for 18 h at 4°C in a SW50.1 rotor. Nine fractions of 555 μl each were collected. Resealed peroxisomes RPA and RPB floated to low density during centrifugation in the sucrose density gradient. Proteins from equal volumes of gradient fractions were analyzed by immunoblotting with antibodies to Pex16p and Pex19p. Equal volumes of gradient fractions were also subjected to lipid extraction, which was followed by TLC and visualization of lipids. Resealed peroxisomes RPA and RPB, which were recovered in the peak fractions of the flotation gradients, and a highly purified subform of the intact peroxisomes from which these two types of resealed peroxisomes were formed, were used to evaluate the orientation in which the membranes delimiting RPA and RPB were resealed. RPA and RPB were pelleted onto a 150-μl cushion of 2 M sucrose in buffer MR or HR, respectively, by centrifugation at 100,000 for 20 min at 4°C in a TLA110 rotor. Intact peroxisomes were pelleted onto a 150-μl cushion of 2 M sucrose in buffer H by centrifugation at 100,000 for 20 min at 4°C in a TLA110 rotor. Individual pellets of RPA, RPB, and intact peroxisomes were resuspended in ice-cold buffer H at 1 mg protein/ml. Serial dilutions of RPA, RPB, and intact peroxisomes in the range of 10–50 μg protein/ml were made in ice-cold buffer H. Anti-Pex16p rabbit IgG or anti-Pex19p rabbit IgG were added to concentrations 4 and 5 μg/ml, respectively. After incubation for 30 min on ice, samples were subjected to centrifugation at 100,000 for 10 min at 4°C in a TLA110 rotor. The pellets were resuspended in 200 μl of ice-cold buffer H and supplemented with fluorescein-conjugated goat anti–rabbit IgG. After incubation for 30 min on ice, samples were subjected to centrifugation at 100,000 for 10 min at 4°C in a TLA110 rotor. The pellets were resuspended in 200 μl of ice-cold buffer H and supplemented with Alexa Fluor 488 goat anti-fluorescein/Oregon green IgG at 15 μg/ml. After incubation for 30 min on ice, samples were subjected to centrifugation at 100,000 for 10 min at 4°C in a TLA110 rotor. The pellets were resuspended in 200 μl of ice-cold buffer H and supplemented with Alexa Fluor 488 chicken anti–goat IgG at 20 μg/ml. After incubation for 30 min on ice, samples were subjected to centrifugation at 100,000 for 10 min at 4°C in a TLA110 rotor. The pellets were resuspended in 200 μl of ice-cold buffer H and placed into the wells of a 96-well microplate. The fluorescence of samples was measured using the Victor 2 multilabel microplate fluorescence reader (Wallac) with filters set at 485 (±7.5) nm (excitation) and 510 (±5) nm (emission). Controls were made for each dilution of RPA, RPB, and intact peroxisomes. The controls included normal rabbit IgG at 4 or 5 μg/ml added instead of anti-Pex16p rabbit IgG or anti-Pex19p rabbit IgG, respectively. Background fluorescence, which was due to the nonspecific binding of rabbit IgG and/or fluorescein- or Alexa Fluor 488–labeled antibodies to the peroxisomal membrane, was subtracted. In intact peroxisomes, Pex19p is a peripheral membrane protein that resides on the outer (cytosolic) face of the peroxisome ( and Fig. S5, C and D). Because this protein is attached to the surface of intact peroxisomes, it is accessible to anti-Pex19p IgG exogenously added to these peroxisomes (Fig. S3 D, right). Importantly, the membranes of intact peroxisomes, RPA, and RPB are not permeable to the exogenously added IgG molecules. In fact, none of Pex16p, a peripheral membrane protein residing on the inner (lumenal) face of the peroxisome, in intact peroxisomes and only a minor portion of this protein in RPA was accessible to anti-Pex16p IgG (Fig. S3 D, left). The observed accessibility of the RPB-associated form of Pex16p to anti-Pex16p IgG was due to the inside-out orientation of the membrane delimiting most of the RPB species formed during peroxisome resealing. In addition, although the levels of Pex19p, a peripheral membrane protein residing on the peroxisomal surface, in intact peroxisomes, RPA, and RPB were very similar to each other (Fig. S3 B), only a minor portion of Pex19p in the mostly inside-out–oriented RPB species was accessible to anti-Pex19p IgG (Fig. S3 D, right). Altogether, these findings imply that, if the fluorescence for RPA (F)/fluorescence for intact peroxisomes (F) or fluorescence for RPB (F)/fluorescence for intact peroxisomes (F) ratio is calculated for Pex19p, it is equal to the fraction of the total pool of Pex19p that resides on the outer (cytosolic) face of those RPA or RPB species whose delimiting membranes acquired the outside-out orientation during their resealing. At the same time, the (F − F)/F or (F − F)/F ratio, if calculated for Pex19p, equals the fraction of Pex19p that resides on the inner (lumenal) face of those RPA or RPB species whose delimiting membranes acquired the inside-out orientation during their resealing. Hence, the F/F or F/F ratio for Pex19p is equal to the fraction of those RPA or RPB species that are present in the outside-out orientation ( and , respectively). Moreover, the (F − F)/F or (F − F)/F ratio for Pex19p equals the fraction of those RPA or RPB species that were resealed in the inside-out orientation ( and , respectively). Resealed peroxisomes RPA and RPB, which were recovered in the peak fractions of the flotation gradients, and a highly purified subform of the intact peroxisomes from which these two types of resealed peroxisomes were formed, were used to calculate the percentage of DAG and PS residing in the cytosolic and lumenal leaflets of the membrane bilayers in different peroxisomal subforms. RPA and RPB were pelleted onto a 150-μl cushion of 2 M sucrose in buffer MR or HR, respectively, by centrifugation at 100,000 for 20 min at 4°C in a TLA110 rotor. Intact peroxisomes were pelleted onto a 150-μl cushion of 2 M sucrose in buffer H by centrifugation at 100,000 for 20 min at 4°C in a TLA110 rotor. Individual pellets of RPA, RPB, and intact peroxisomes were resuspended in ice-cold buffer H at 1 mg protein/ml. Serial dilutions of RPA, RPB, and intact peroxisomes in the range of 10–50 μg protein/ml were made in ice-cold buffer H. The DAG-binding C1b domain of protein kinase C labeled with the fluorophore Alexa Fluor 488 or anti-PS mouse IgM were added to concentrations 5 and 1 μg/ml, respectively. After incubation for 30 min on ice, samples were subjected to centrifugation at 100,000 for 10 min at 4°C in a TLA110 rotor. For samples that were exposed to Alexa Fluor 488–tagged C1b domain, the pellets were resuspended in 200 μl of ice-cold buffer H and placed into the wells of a 96-well microplate. The fluorescence of these samples was measured using the Victor 2 multilabel microplate fluorescence reader with filters set at 485 (±7.5) nm (excitation) and 510 (±5) nm (emission). Controls for monitoring DAG were made for each dilution of intact peroxisomes P5 and P6 and of the P5- and P6-based RPA and RPB, all of which contained DAG ( and Fig. S3 A). The controls included the corresponding dilutions of intact peroxisomes P4 and of the P4-based RPA and RPB, all of which did not contain DAG (). Background fluorescence, which was due to the nonspecific binding of Alexa Fluor 488–tagged C1b domain to the peroxisomal membrane, was subtracted. For samples that were exposed to anti-PS mouse IgM, the pellets were resuspended in 200 μl of ice-cold buffer H and supplemented with fluorescein-conjugated goat anti–mouse IgM antibodies at 5 μg/ml. After incubation for 30 min on ice, samples were subjected to centrifugation at 100,000 for 10 min at 4°C in a TLA110 rotor. The pellets were resuspended in 200 μl of ice-cold buffer H and supplemented with Alexa Fluor 488 rabbit anti-fluorescein/Oregon green IgG at 15 μg/ml. After incubation for 30 min on ice, samples were subjected to centrifugation at 100,000 for 10 min at 4°C in a TLA110 rotor. The pellets were resuspended in 200 μl of ice-cold buffer H and supplemented with Alexa Fluor 488 goat anti–rabbit IgG at 20 μg/ml. After incubation for 30 min on ice, samples were subjected to centrifugation at 100,000 for 10 min at 4°C in a TLA110 rotor. The pellets were resuspended in 200 μl of ice-cold buffer H and placed into the wells of a 96-well microplate. The fluorescence of samples was measured using the Victor 2 multilabel microplate fluorescence reader with filters set at 485 (±7.5) nm (excitation) and 510 (±5) nm (emission). Controls were made for each dilution of RPA, RPB, and intact peroxisomes. The controls included normal mouse IgM at 1 μg/ml added instead of anti-PS mouse IgM. Background fluorescence, which was due to the nonspecific binding of mouse IgM and/or fluorescein- or Alexa Fluor 488–labeled antibodies to the peroxisomal membrane, was subtracted. The fraction of a monitored lipid, either DAG or PS, residing in the cytosolic leaflet of the membrane bilayer of the intact peroxisome can be calculated as follows: where F is the fluorescence of a lipid-specific fluorescent probe specifically bound to intact peroxisomes or to the species of RPA and RPB that are present in the outside-out orientation. In , F equals the fluorescence of this probe specifically bound to the outer (cytosolic) leaflet of the peroxisomal membrane bilayer delimiting intact peroxisomes or those species of RPA and RPB that were resealed in the outside-out orientation. Furthermore, F in is the fluorescence of a lipid-specific fluorescent reporter molecule that would, if it could, bind specifically to the inner (lumenal) leaflet of the peroxisomal membrane bilayer delimiting intact peroxisomes. F can be monitored by measuring the fluorescence of this reporter molecule bound to the surface of those species of RPA and RPB that were resealed in the inside-out orientation. The value of F, the fluorescence of a lipid-specific fluorescent reporter molecule specifically bound to the surface of RPA, can be calculated as follows: where is the fraction of the RPA species that are present in the outside-out orientation. The value of for each of the outside-out–oriented species of RPA formed during resealing of osmotically lysed peroxisomal subforms P1–P6 was calculated for a Pex19p-specific fluorescent reporter molecule as described. The values of for individual species of the P1- to P6-based RPA are presented in Fig. S3 (E and F). In , the value of for each of the inside-out–oriented species of RPA formed during resealing of osmotically lysed peroxisomal subforms P1–P6 was calculated for a Pex19p-specific fluorescent reporter molecule as described. The values of for individual species of P1- to P6-based RPA are presented in Fig. S3 (E and F). Based on , F can be calculated as follows: The value of F, the fluorescence of a lipid-specific fluorescent reporter molecule specifically bound to the surface of RPB, can be calculated as follows: where is the fraction of the RPB species that are present in the outside-out orientation. The value of for each of the outside-out–oriented species of RPB formed during resealing of osmotically lysed peroxisomal subforms P1–P6 was calculated for a Pex19p-specific fluorescent reporter molecule as described. The values of for individual species of the P1- to P6-based RPA are presented in Fig. S3 (E and F). In , the value of for each of the inside-out–oriented species of RPB formed during resealing of osmotically lysed peroxisomal subforms P1–P6 was calculated for a Pex19p-specific fluorescent reporter molecule as described. The values of for individual species of P1- to P6-based RPB are presented in Fig. S3 (E and F). Based on , F can be calculated as follows: Based on , for calculating the fraction of a monitored lipid, either DAG or PS, residing in the cytosolic leaflet of the membrane bilayer of the intact peroxisome can be rewritten as follows: Furthermore, based on , for calculating the fraction of a monitored lipid, either DAG or PS, residing in the cytosolic leaflet of the membrane bilayer of the intact peroxisome can be also rewritten as follows: For each of the intact peroxisomal subforms P1–P6, and were used for calculating the fraction of a monitored lipid, either DAG or PS, residing in the cytosolic leaflet of the membrane bilayer. To evaluate the lipid-binding specificity of Pex16p, the pellet of membranes recovered after centrifugation of osmotically lysed peroxisomes was resuspended in buffer TBSO (10 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 0.5% n-OG) and incubated for 30 min on ice. Samples were subjected to centrifugation at 100,000 for 30 min at 4°C in a TLA110 rotor. Under these conditions, n-OG completely solubilized the vast majority of all membrane proteins (). The supernatants of n-OG–solubilized proteins were then incubated at 5 μg/ml with the PIP Strips at 4°C overnight. After washing the PIP Strip five times for 5 min each with TBSO, Pex16p was detected by immunoblotting with anti-Pex16p antibodies. For purification of LPAAT, highly purified mature peroxisomes P6 were lysed by the addition of 10 vol of ice-cold LB buffer, followed by incubation on ice for 30 min with occasional agitation. The suspension was centrifuged at 200,000 for 20 min at 4°C in a TLA110 rotor. The pellet of membranes recovered after centrifugation of osmotically lysed peroxisomes was resuspended in ice-cold HAT + C buffer (20 mM Hepes-KOH, pH 7.4, 2 mM EDTA, 1 mM DTT, 10% [wt/vol] glycerol, and 0.5% [wt/vol] Na cholate) to a final concentration of 5.0 mg/ml. After incubation on ice for 30 min with occasional agitation, the sample was subjected to centrifugation at 100,000 for 30 min at 4°C in a TLA110 rotor. The recovered supernatant of detergent-solubilized PMPs was applied to a Resource S column, which was preequilibrated with 10 column volumes of HAT − C buffer (20 mM Hepes-KOH, pH 7.4, 2 mM EDTA, 1 mM DTT, and 10% [wt/vol] glycerol) followed by 1 column volume of HAT + C buffer. The column was washed with 4 column volumes of HAT + C buffer followed by elution of LPAAT activity with a linear 0–1 M NaCl gradient in HAT + C buffer. The peak of LPAAT activity eluted from the column in a NaCl concentration of 225 mM. LPAAT-containing fractions were combined, dialyzed against HPC + C buffer (20 mM Hepes-KOH, pH 7.4, 50 mM KCl, 25 mM KOAc, 3 mM MgCl, 2 mM MgOAc, 1 mM DTT, 10% [wt/vol] glycerol, and 0.5% [wt/vol] Na cholate), and applied to a palmitoyl-CoA agarose column, which was preequilibrated with 5 column volumes of HPC − C buffer (20 mM Hepes-KOH, pH 7.4, 50 mM KCl, 25 mM KOAc, 3 mM MgCl, 2 mM MgOAc, 1 mM DTT, and 10% [wt/vol] glycerol) followed by 1 column volume of HPC + C buffer. The column was washed with 9 column volumes of HPC + C buffer followed by 1 column volume of HPC + C buffer containing 0.5 mM free palmitoyl-CoA. The column was kept for 6 h at 4°C. LPAAT activity was eluted with 2 column volumes of HPC + C buffer containing 5 mM free palmitoyl-CoA followed by 2 column volumes of HPC + C buffer alone. The two LPAAT-containing eluates were pooled and stored at −80°C, before being analyzed by SDS-PAGE followed by silver staining and mass spectrometric identification of LPAAT. A summary of the purification of LPAAT is presented in . The overall purification of LPAAT over the Na-extracted membrane of mature peroxisomes P6 was 232-fold. For purification of PAP, highly purified mature peroxisomes P6 were lysed by the addition of 10 vol of ice-cold LB buffer, followed by incubation on ice for 30 min with occasional agitation. The suspension was centrifuged at 200,000 for 20 min at 4°C in a TLA110 rotor. The pellet of membranes recovered after centrifugation of osmotically lysed peroxisomes was resuspended in ice-cold TP + C buffer (20 mM Tris-HCl, pH 7.4, 3 mM MgCl, 2 mM MgOAc, 1 mM DTT, 10% [wt/vol] glycerol, and 0.5% [wt/vol] Na cholate) to a final concentration of 5.0 mg/ml. After incubation on ice for 30 min with occasional agitation, the sample was subjected to centrifugation at 100,000 for 30 min at 4°C in a TLA110 rotor. The recovered supernatant of detergent-solubilized PMPs was applied to a Resource Q column, which was preequilibrated with 5 column volumes of TP − C buffer (20 mM Tris-HCl, pH 7.4, 3 mM MgCl, 2 mM MgOAc, 1 mM DTT, and 10% [wt/vol] glycerol) followed by 1 column volume of TP + C buffer. The column was washed with 5 column volumes of TP + C buffer followed by elution of PAP activity with a linear 0–0.5 M NaCl gradient in TP + C buffer. The peak of PAP activity eluted from the column in a NaCl concentration of 110 mM. PAP-containing fractions were combined, dialyzed against PPS + C buffer (10 mM potassium phosphate, pH 7.0, 100 mM KCl, 3 mM MgCl, 2 mM MgOAc, 1 mM DTT, 10% [wt/vol] glycerol, and 0.5% [wt/vol] Na cholate), and applied to a HiTrap Blue HP column, which was preequilibrated with 5 column volumes of PPS − C buffer (10 mM potassium phosphate, pH 7.0, 100 mM KCl, 3 mM MgCl, 2 mM MgOAc, 1 mM DTT, and 10% [wt/vol] glycerol) followed by 1 column volume of PPS + C buffer. The column was washed with 5 column volumes of PPS + C buffer followed by elution of PAP activity with a linear 0.1–1.5 M KCl gradient in PPS + C buffer. The peak of PAP activity eluted from the column in a KCl concentration of 660 mM. PAP-containing fractions were pooled, dialyzed against PP + C buffer (10 mM potassium phosphate, pH 7.0, 5 mM MgCl, 5 mM MgOAc, 1 mM DTT, 10% [wt/vol] glycerol, and 0.5% [wt/vol] Na cholate), and applied to a hydroxylapatite column, which was preequilibrated with 5 column volumes of PP + C buffer. The column was washed with 3 column volumes of PP + C buffer followed by elution of PAP activity with 10 column volumes of a linear 10–200 mM potassium phosphate gradient in PP + C buffer. The peak of PAP activity eluted from the column in a potassium phosphate concentration of 95 mM. PAP-containing fractions were combined, dialyzed against TPM + C buffer (25 mM Tris-HCl, pH 7.4, 10 mM MgCl, 10 mM MgOAc, 1 mM DTT, 10% [wt/vol] glycerol, and 0.5% [wt/vol] Na cholate), and applied to a Superose 12 column, which was preequilibrated with 5 column volumes of TPM + C buffer. PAP was eluted from the column with TPM + C buffer. PAP-containing fractions were pooled and stored at −80°C before being analyzed by SDS-PAGE followed by silver staining and mass spectrometric identification of PAP. A summary of the purification of PAP is presented in . The overall purification of PAP over the Na-extracted membrane of mature peroxisomes P6 was 423-fold. For identifying components of the peroxisomal membrane that interact with Vps1p, Pex10p, or Pex19p, highly purified mature peroxisomes of wild-type and mutant strains were osmotically lysed by the addition of 10 vol of ice-cold LCC buffer (20 mM sodium phosphate buffer, pH 7.5, and 150 mM NaCl), followed by incubation on ice for 30 min with occasional agitation. The suspension was centrifuged at 200,000 for 20 min at 4°C in a TLA110 rotor. The pellet of membranes recovered after centrifugation of osmotically lysed peroxisomes was resuspended in ice-cold LCC buffer to a final concentration of 0.5 mg/ml. For identifying proteins that interact with Vps1p or Pex19p in the cytosol, wild- type and mutant cells were subjected to subcellular fractionation (see Subcellular fractionation…) to yield the 200S (cytosolic) fraction in buffer H containing 1 M sorbitol. 9 vol of ice-cold LCC buffer was added to the recovered cytosolic fraction. Cross-linking of membrane-associated or cytosolic proteins with the thiol-cleavable cross-linker DSP (Pierce Chemical Co.) was initiated by the addition of cross-linker (50 mM stock in DMSO) and continued for 1 h at 4°C. Cross-linking was quenched by the addition of 0.1 vol of 1 M Tris-HCl, pH 7.5, and incubation for 30 min at 4°C. SDS was added to 1.25%, and samples were warmed at 65°C for 20 min and then cooled to room temperature. 4 vol of 60 mM Tris-HCl, pH 7.4, 1.25% (vol/vol) Triton X-100, 190 mM NaCl, and 6 mM EDTA was added to the cooled samples, which were then cleared of any nonspecifically binding proteins by incubation for 20 min at 4° C with protein A–Sepharose washed five times with 10 mM Tris-HCl, pH 7.5. The cleared samples were then subjected to immunoprecipitation with anti-Vps1p, anti-Pex10p, or anti-Pex19p antibodies under denaturing, nonreducing conditions. These antibodies were covalently linked to protein A–Sepharose as described previously (). Bound proteins were washed five times with 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 1% (vol/vol) Triton X-100 and eluted with 2% SDS at 95°C for 5 min. Eluted proteins were analyzed by SDS- PAGE under reducing conditions, i.e., with DTT in the sample buffer, followed by silver staining. Proteins were resolved by SDS-PAGE and visualized by silver staining (). Protein bands were excised from the gel, reduced, alkylated, and in gel–digested with trypsin (). The proteins were identified by matrix-assisted laser desorption/ionization mass spectrometric peptide mapping (), using a Micromass time-of-flight mass spectrometer (Waters). Database searching using peptide masses was performed with the Mascot web-based search engine. Whole cells were fixed in 1.5% KMnO for 20 min at room temperature, dehydrated by successive incubations in increasing concentrations of ethanol, and embedded in Poly/Bed 812 epoxy resin (Polysciences). Ultrathin sections were cut using an Ultra-Cut E Microtome (Reichert-Jung). Silver/gold thin sections from the embedded blocks were examined in a transmission electron microscope (JEM-2000FX; JEOL). For morphometric analysis of random electron microscopic sections of cells, 12- × 14-cm prints and 8- × 10-cm negatives of 35–40 cell sections of each strain at 24,000–29,000 magnification were scanned and converted to digitized images with a ScanJet 4400c (Hewlett-Packard) and Photoshop 6.0 software (Adobe Systems, Inc.). Quantitation of digitized images was performed using the Discovery Series Quantity One 1-D Analysis Software (Bio-Rad Laboratories). Relative area of peroxisome section (as a percentage) was calculated as the area of peroxisome section/area of cell section × 100. Peroxisomes were counted in electron micrographs, and data are expressed as the number of peroxisomes per μm of cell section volume. Resealed peroxisomes RPA and RPB floated to low density during centrifugation in a multistep sucrose density gradient. A 200-μl aliquot of the peak fraction of purified RPA in MR buffer or a 200-μl aliquot of the peak fraction of purified RPB in HR buffer was mixed with 400 μl of ice-cold 150 mM sodium cacodylate buffer, pH 7.2, containing 3% glutaraldehyde. Immediately after mixing the sample and glutaraldehyde solution, 600 μl of 2% OsO in ice-cold CD buffer (100 mM sodium cacodylate, pH 7.2) was added. After a 2-h incubation on ice, the resealed peroxisomes RPA and RPB were sedimented at 100,000 for 20 min at 4°C in a TLS55 rotor (Beckman Coulter) onto a bed (25–50 μl) of hardened, low-melting 2.5% NuSieve GTG agarose (FMC). The pellet was postfixed in a solution of 1% OsO plus 2.5% KCrO in ice-cold CD buffer for 2 h on ice. The pellet was then rinsed twice with ice-cold CD buffer and exposed to 0.05% tannic acid in the same buffer. After a 30-min incubation on ice, the pellet was washed once with ice-cold CD buffer and three times with water. The pellet was incubated overnight with 2% uranyl acetate in water at 4°C and then washed three times with water. After dehydration in a graded ethanol series, the fixed and stained sample was embedded in Poly/Bed 812 epoxy resin (Polysciences). Silver/gold thin sections from the embedded blocks were examined in the JEM-2000FX transmission electron microscope. Fig. S1 shows that the peroxisomal membrane lacks the activities of enzymes that, in addition to LPAAT and PAP, can catalyze reactions resulting in the formation of PA or DAG. Fig. S2 demonstrates the effect of the , , , , and mutations on the size and number of peroxisomes. Fig. S3 documents the dynamics of changes in the transbilayer distribution of DAG and PS in the peroxisomal membrane during peroxisome maturation. Fig. S4 provides evidence that the Pex2p-dependent transfer of PC from a P3- and P4-associated subcompartment of the ER provides the peroxisomal membrane with the bulk quantities of this lipid. Fig. S5 shows that only division-competent mature peroxisomes recruit Vps1p from the cytosol to the outer face of their membrane. Online supplemental material is available at .

Retinoids (vitamin A and its derivatives) are critical for processes ranging from the immune response to neuronal plasticity, development, visual pigment generation, cell proliferation, and other essential physiological processes (for review see ; ). In animals, all retinoids must be acquired from the diet either as preformed vitamin A (all-trans-retinol) or must be formed from the provitamin A precursor, carotenoids. The dietary carotenoids are synthesized in plants, certain fungi, and bacteria, and, to become biologically active, must first be absorbed and then delivered to the site in the body where they are converted to vitamin A (for review see ). The β, β-carotene-15, 15′ monooxygenase (BCO) is the key enzyme in vitamin A formation, which catalyzes the centric cleavage of β-carotene to yield retinaldehyde (all-trans-retinal; ; ; ; ; ). However, until relatively recently, the identities of these enzymes in vertebrates and invertebrates were not known. In , BCO is encoded by (), and mutations in this gene disrupt retinoid production and phototransduction as a result of elimination of rhodopsin (). As carotenoids are highly lipophilic molecules, specific proteins must exist to transport them to specialized target tissues and to absorb the provitamin A into cells. It has been suggested that class B scavenger receptors may play important roles in the cellular uptake of carotenoids (for review see ). In , mutations in the gene (), which encodes a membrane protein homologous to the mammalian class B type I scavenger receptor (SR-BI; ), result in a defect in the uptake of carotenoids and synthesis of retinoids (). SR-BI plays critical roles in cholesterol and high-density lipoprotein metabolism and in maintaining plasma cholesterol levels (). SR-BI also mediates cellular uptake of free cholesterol (), triglycerides (), phospholipids (), and vitamin E (). Moreover, SR-BI is expressed in the human intestine (; ), where it is proposed to mediate absorption of dietary β-carotene (; ). The combination of these studies suggests that class B scavenger receptors may function as carotenoid receptors. Although the molecular mechanism through which class B scavenger receptors mediate absorption of carotenoids is not known, it might involve binding of carotenoid containing lipoproteins or micelles via the extracellular domain, separating the two transmembrane segments (; ), followed by uptake of carotenoids through a process independent of endocytosis (). The photopigment, rhodopsin, consists of a seven-transmembrane protein, opsin, and a chromophore (3-hydroxy-11-cis retinal and 11-cis retinal in and mammals, respectively), which is formed through metabolism of vitamin A (; ). In , light results in a cis- to trans-isomerization of the chromophore, and this transformation represents the only light-driven step during phototransduction. The all-trans-retinol is converted to 11-cis-retinal in pigment cells in a light-dependent, rather than an enzyme-dependent, manner (), whereas the pathway leading from dietary carotenoids to all-trans-retinal takes place outside of retina tissues (). Deprivation of vitamin A, either by depletion of dietary retinoids or as a result of mutations in the vitamin A pathway causes reductions in rhodopsin levels and defects in vision. In contrast to mammals, in , retinoids are not required for viability but appear to be required exclusively in the retina (). As such, represents a highly tractable animal model to study the metabolism of vitamin A in vivo. The gene encodes a BCO, which functions outside the retina for conversion of carotenoids to all-trans-retinal (; ; ). Thus, a key question concerns the identity of the scavenger receptor that is functionally coupled to NINAB for the uptake of carotenoids. It has been suggested that the class B scavenger receptor, NINAD, is the protein that functions in concert with NINAB (; ). However, expression is enriched in bodies, whereas expression is reported to be enriched in heads (; ), which questions how the two differentially expressed gene products are coupled. In the present study, we describe the isolation of the () locus, which encodes a new member of class B scavenger receptor family. Mutation of profoundly affected the visual response and production of rhodopsin, both of which were restored by providing all-trans-retinal to the diet. We found that functioned downstream of , in a step required for the conversion of carotenoids to vitamin A. The gene functioned outside of the retina and appeared to display a similar expression pattern as in fly heads. We provide evidence that and function in the same cells in vivo. Based on these results, we propose that the class B scavenger receptor, SANTA MARIA, is functionally coupled with the BCO enzyme, NINAB, in the conversion of carotenoids to retinaldehyde. In contrast to NINAB and SANTA MARIA, we show that the other class B scavenger receptor, NINAD functions in the uptake of carotenoids primarily in the midgut. Combined with our previous demonstration that the retinoid binding protein, PINTA (PDA [prolonged depolarization afterpotential] is not apparent), functions in the retinal pigment cells in the final step in the generation of the chromophore, we propose a pathway involving the NINAB, NINAD, PINTA, and SANTA MARIA proteins acting in multiple cell types in the conversion of carotenoids to the rhodopsin chromophore. To identify new genes in that functioned in the generation of rhodopsin and other aspects of phototransduction, we conducted a screen of chromosome 2 for homozygous viable mutations that caused a defect in electroretinogram (ERG) recordings (Fig. S1, available at ). ERGs are extracellular recordings that measure the summed retinal response to light. In , the chromophore stays bound to the light-activated metarhodopsin, and a second photon of light is required for the reconversion of the metarhodopsin to the inactive rhodopsin (; ). The major rhodopsin (Rh1) responds to either orange or blue light, whereas metarhodopsin responds effectively to orange light only. As a consequence, blue light causes stable activation of metarhodopsin, resulting in a PDA (). The PDA requires a molar excess of the active form of the metarhodopsin over the available arrestin, which is required to arrest the activity of the metarhodopsin (). Thus, when the Rh1 level is decreased, as occurs upon mutation of the structural gene for the Rh1 opsin (), a PDA is not produced (; ; ). One of mutant lines isolated in the ERG screen displayed a PDA-defective ERG phenotype, similar to that observed in () flies (). Mutations in three second-chromosomal genes, , , and , are known to reduce or eliminate the PDA (; ; ; ; ). The new mutation complemented , , or (unpublished data). Therefore, this mutation disrupted a new gene required for the generation of the PDA, which we refer to as . As the PDA phenotype is usually due to a reduction in the level of Rh1, we checked the Rh1 concentration and found that it was severely reduced in the mutant, as was the case in and flies (). We also checked () mRNA expression in using Northern blots and found that it was not reduced compared with wild type (). Thus, the reduction in Rh1 protein was not due to disruption in expression or stability of the mRNA. In addition to Rh1, there are four minor rhodopsins (Rh3–6) expressed in the retina (). These minor opsins are spatially localized in nonoverlapping subsets of the smaller R7 and R8 cells (; ; ; ; ). To address whether the mutation reduced the expression of an opsin other than Rh1, we checked the protein levels of Rh4 and found that the concentration of this protein was also diminished (). The levels of other photoreceptor proteins, such as the eye-enriched PLC (NORPA [no receptor potential A]) and the transient receptor potential channel, did not change (). Therefore, the mutation caused a reduction in the concentration of rhodopsins but did not result in a general defect in the expression of photoreceptor cell proteins. To identify the gene responsible for the phenotype, we mapped the site of the mutation to the 27F4 to 28A2 region (; see Materials and methods), which included 10 known or predicted genes () spanning the region between and . Among these 10 genes, the predicted amino acid sequences of three genes suggested that they were excellent candidates for encoding SANTA MARIA. Two of them ( and ) encode putative retinoid binding proteins, and the third () encodes a homologue of class B scavenger receptors. The predicted CG12789 protein shares 33% identity with the human SR-BI (hSR-BI; ; ); 26% identity with mouse CD36, the founding member of this family (); and 30% identity with the scavenger receptor NINAD, which also functions in rhodopsin biosynthesis (; ; ). Class B scavenger receptor family are suggested to consist of two transmembrane domains and cytoplasmic N- and C-termini (). To find out which of the three candidates was the gene, we introduced transgenes encoding , , and into flies. , expressed under the control of the promoter (), restored a wild-type PDA () and increased the level of the Rh1 protein in the mutant flies (;/+; ). The level of rhodopsin in these flies was lower than in wild-type, possibly because of the relative weakness of the promoter in some cell types. In contrast, neither the nor the transgenes rescued the PDA defect or increased Rh1 levels in the flies (unpublished data). Therefore, , which encodes a predicted class B scavenger receptor, is the gene. Some gene products that are essential for production or transport of the chromophore function in the retina, whereas others play roles outside the retina. The two proteins required in the retina are the retinoid binding protein, PINTA, which functions in pigment cells for chromophore synthesis (), and a oxidoreductase, NINAG, which is required in the compound eye for chromophore synthesis (). In contrast, two gene products that have been reported to operate outside of the retina for carotenoid metabolism are , which encodes a class B scavenger receptor (; ), and , which encodes a BCO (; ; ). These findings raise the question as to the tissue and cellular requirements for . To determine whether was required in the compound eye, we used two approaches. First, we generated mosaic flies using a mitotic recombination approach that leads to the generation of fully homozygous mutant eyes in otherwise heterozygous animals (). We found that the mosaic flies expressed normal levels of Rh1 (), indicating that was not required in the compound eye. Second, we tested for rescue of the phenotype, after expressing wild-type in the retina, using the / () system (). This approach results in expression of genes that are linked 3′ to the , to occur specifically under the control of the GAL4 transcription factor. Therefore, we generated transgenic flies, and introduced transgenes into these flies that direct expression of in different retinal cells. Normal Rh1 levels or a wild-type PDA were not restored in upon expression of throughout the eye () or exclusively in pigment cells () or photoreceptor cells (; ; and Fig. S2, A–C, available at ). The lack of rescue was not due to a problem with the transgene, as the phenotype was reversed in flies containing a in combination with the (). These results demonstrate that is required outside the retina for biosynthesis of rhodopsin. Neither flies nor mammals can synthesize β-carotene but must obtain this vitamin A/chromophore precursor from the diet. Both and , whose activities are required outside of the retina for rhodopsin biogenesis, function in the pathway from β-carotene to all-trans-retinal because both mutant phenotypes are rescued by supplementation of the food with all- trans-retinal (; ; and Fig. S2, D, E, G, and H). In contrast, PINTA and NINAG are required in the compound eyes and function subsequent to the generation of all-trans-retinal, as supplementation with all- trans-retinal does not restore Rh1 to wild-type levels in these mutants (; ). Because acts outside of the compound eyes, it may also function in a step necessary for the conversion of β-carotene to vitamin A. Therefore, we checked whether the phenotype could be rescued by addition of all-trans-retinal to the diet. As was the case with and , we found that supplementation of the food with all-trans-retinal (0.2 mM) restored the Rh1 levels and the PDA in flies (). Both and function in the generation of retinoids from carotenoids; however, only the phenotype, not the phenotype, is rescued by high doses of β-carotene ( and Fig. S2, F and I; ; ). NINAB is an essential enzyme necessary for all-trans-retinal production, whereas NINAD is a scavenger receptor, which promotes the uptake of carotenoids. This latter function can be bypassed by large concentrations of dietary carotenoids. Based on these data, it has been proposed that functions downstream of (). To test whether the phenotype was rescued by carotenoids, we fed the mutant flies 0.2 mM β-carotene. We found that addition of β-carotene did not rescue the phenotype (). NINAB is a BCO, which converts β-carotene to all-trans-retinal; therefore, NINAB would need to be coexpressed with a β-carotene receptor, to promote influx of β-carotene into the cells. The two class B scavenger receptors, SANTA MARIA and NINAD, are candidate proteins that could be functionally coexpressed with NINAB, and serve this role. To address whether NINAD or SANTA MARIA function in the same cells as NINAB, we used the / system (). To conduct these experiments, we generated , , and – transgenic flies (see Materials and methods) and introduced them into the , , and mutant backgrounds, along with , , and . Each of these and lines was effective because the , , or phenotype was rescued by the / transgenes corresponding to the same genes ( and ). We found that the and phenotypes were rescued by expression of one gene under the control of the other line. Specifically, expression of using the restored Rh1 levels in flies ( and ), whereas expression of under control of the rescued the phenotype (; and ). Indistinguishable results were obtained using two independent and three lines (unpublished data). These data indicate that is functionally coexpressed with in the same cells. In contrast to these results, the NINAB BCO did not function together with the other scavenger receptor, NINAD, as previously proposed (). Expression of or under control of either of two lines did not reduce the severity of the and phenotypes, respectively (; ; and not depicted). To determine the sensitivity of this analysis, we conducted a dilution experiment and found that we could detect 2% the wild-type levels of Rh1 (Fig. S3, available at ). Because the levels of Rh1 produced in ; flies, either in the presence or absence of , were both at the threshold for detection (≤2% wild-type levels; Fig. S3), we conclude that if there was any rescue with the , it was ≤2%. Furthermore, within the resolution of our analysis, expression of using the or did not increase Rh1 levels in flies ( and ). These results were not due to ineffectiveness of the or the , as the phenotype was rescued by cointroduction of these transgenes into the mutant flies (). Thus, was not functionally coexpressed with , consistent with the proposal that SANTA MARIA is the critical scavenger receptor operating in combination with NINAB. To find out which cell types express and , we first tested whether expression of these genes was enriched in bodies or heads, using a reporter. We prepared extracts from the heads and bodies of and flies and probed Western blots with anti-GFP antibodies. In flies, GFP was detected exclusively in the heads, whereas in flies, the GFP was found in both heads and bodies ( and not depicted). Because and appear to be functionally coexpressed, the two gene products may collaborate primarily in fly heads for the generation of all-trans-retinal. To address the cell types in adult heads expressing reporters under control of the and enhancer/promoters, we performed double-labeling experiments. We stained head sections obtained from and flies with anti-GFP antibodies, in combination with glial (anti-REPO) or neuronal markers (anti-ELAV). A GFP with a nuclear localization signal was used (GFP.nls), as REPO (; ) and ELAV (; ) are both nuclear proteins. We found that the GFP colocalized with ELAV and, to a lesser extent, with REPO (), which indicated that and were expressed in both neuronal cells and glia cells. The presumptive expression patterns for and , which were detected using the / system, appeared to reflect the patterns obtained by direct enhancer/promoter-reporter fusions, as we observed similar patterns of expression after staining the flies with anti-GAL4 and anti-GFP (unpublished data). To investigate the cell type in which and appear to function, we tested for rescue of the mutant phenotypes using the / approach. We found that expression of or in neurons or glia, using the or the , respectively, restored Rh1 levels in both and flies (; and ). However, the mutant phenotypes were not rescued by introduction of either or in combination with drivers that were expressed in other cell types and tissues, such as muscle cells, salivary glands, and fat bodies (unpublished data). These results suggest that and both function in neurons and glia cells. The finding that wild-type Rh1 levels are restored if and are expressed in either neurons or glia raises the possibility that the specific cell type expressing these two genes is not critical, as long as the two gene products are coordinately expressed. Therefore, we coexpressed and in the retinal pigment cells, which express the retinoid binding protein PINTA and normally function in the final step in the production of the chromophore—conversion of all-trans-retinol to 11-cis-retinal (). To conduct these experiments, we introduced into or flies a pigment cell (-) together with both and . However, in neither mutant was the reduced Rh1 level increased (Fig. S4, available at ). These results suggest that there are one or more additional components that are required for the - and –dependent conversion of carotenoids to vitamin A. Alternatively, we cannot exclude the possibility that there is a problem in the transport of carotenoids to the retina. Because NINAD and SANTA MARIA share considerable amino acid homology (30% identity) and both are required for the generation of retinoids from carotenoids, the two scavenger receptors might have the same molecular functions. To address whether NINAD and SANTA MARIA can functionally substitute for each other, we tested whether could rescue the phenotype if it was expressed in those cells that normally express . In addition, we performed the reciprocal experiment by expressing the transgene under control of the in flies. However, Rh1 levels were not restored either in flies containing the / transgenes or in flies containing the / transgenes (). A mammalian homologue of NINAD and SANTA MARIA, SR-BI, has been suggested to function in the uptake of a variety of lipids, including β-carotene (). Therefore, we considered the possibility that SR-BI may have the same molecular function as either NINAD or SANTA MARIA. To test this proposal, we introduced a transgene and expressed under control of the or the in or flies, respectively. However, expression of did not rescue either the or phenotypes (; ; and Fig. S5, available at ). Because the class B scavenger receptor, NINAD, is not functionally coexpressed with NINAB, it would appear that is required in cells distinct from those in which and function. The expression of has been detected in the midgut primordia in embryos (), raising the possibility that NINAD may function in the midgut for absorption of carotenoids into animals. To test whether is expressed in the midgut, we used a GFP reporter. We found that GFP fluorescence in / flies was detected almost exclusively in the midgut (). To address whether expression was enriched in gut, we prepared extracts from / fly heads, bodies, dissected guts, and bodies without guts and probed a Western blot with anti-GFP antibodies. To aid in the comparison, the extracts were prepared in a constant volume consisting of the same actual numbers of dissected guts and bodies, rather than the same total mass. GFP was detected in bodies and dissected guts, but not in heads (). Moreover, the GFP signal was dramatically reduced in bodies after removal of the guts. The results further supported the conclusion that was expressed primarily in the midgut. To address whether functioned in the midgut, we directed expression under control of , which drives expression in the midgut, as well as in the small intestines and Malpighian tubules (). We found that expression of under the control of fully restored the Rh1 levels in flies, whereas expression of using the did not increase Rh1 levels in flies ( and ). Based on these results, we propose that the NINAD functions in absorption of carotenoids in the midgut. In , a reduction in rhodopsin levels results from mutations affecting either the synthesis or transport of the opsin or chromophore subunits (; ). As such, genetic screens for mutations that affect rhodopsin levels provide an excellent opportunity to identify and characterize the roles of gene products required for production of the opsin, vitamin A, and the chromophore. Several genes required for rhodopsin biosynthesis have been previously reported, including those that are essential for steps involved in the synthesis or transport of the opsin (; ; ; ), all-trans-retinal from β-carotene (; ), and the chromophore from vitamin A (; ). However, there remained many questions concerning the cellular sites for the various steps in vitamin A/chromophore synthesis, the nature of the proteins that participate in the uptake, transport, and synthesis of the intermediates, and the identities of receptors and enzymes that functioned coordinately in the same cells. In animals, ranging from flies to humans, dietary β-carotene is the substrate for production of vitamin A, and the vitamin A is subsequently converted into the chromophore. The critical step in the conversion of β-carotene to vitamin A is the centric cleavage by BCO, which in is encoded by the gene (). The key question concerns the identity of the receptor protein that operates in concert with NINAB and is necessary for the uptake of carotenoids in the expressing cells. It has been suggested that the class B scavenger receptor encoded by serves this function (). However, we have found that the phenotype was not rescued by expression of wild-type in expressing cells. Moreover, we found that was expressed primarily in the heads, whereas was only detected in the bodies. As does not operate in concert with , there would appear to be another receptor that serves this function. In the current work, we identify a new class B scavenger receptor, SANTA MARIA, and provide evidence that it is functionally coupled to the NINAB BCO. In support of these conclusions, we found that SANTA MARIA is homologous to known class B scavenger receptors and mutations in disrupt the biogenesis of rhodopsin. Moreover, both and are expressed in fly heads and function in neurons and glia. Most important, expression of under control of the promoter rescued the phenotype and expression of using the promoter rescued the phenotype. Therefore, we suggest that carotenoids are taken up from circulation by SANTA MARIA, thereby providing the substrate for processing of carotenoids to all-trans-retinal by the NINAB BCO. We suggest that the mammalian class B scavenger receptors SR-BI and CD36 may also function in all-trans-retinal production, through coupling with a BCO. SR-BI and CD36 are expressed in the liver (; ) and in the intestines (; ) and appear to function in mediating absorption of β-carotene (; ). In mammals, mRNA is also detected in the small intestine and liver (; ; ), raising the possibility that one of these scavenger receptors is functional coupled with BCO in the conversion of β-carotene to all-trans-retinal. BCO is highly expressed in the retinal pigment epithelium of both human and monkey eyes (; ), suggesting that the biosynthesis of all-trans-retinal from carotenoids may occur in RPE cells, which are well known to function in the generation of the chromophore from all-trans-retinol (for review see ). Interestingly, CD36 also appears to be expressed in RPE cells, suggesting that it may participate in the metabolism of β-carotene to all-trans-retinal in RPE cells. However, the association of a specific mammalian scavenger receptor and a BCO has not been described. In principle, the conversion of carotenoids to all-trans-retinal could be a relatively simple process, which occurs in one cell type (e.g., cells in the gut) exclusively through the activity of a coupled scavenger receptor and a BCO. Although NINAB and SANTA MARIA function together for the centric cleavage of β-carotene, our data suggest that the production of all-trans-retinal from β-carotene is more complicated than previously envisioned. NINAD is another class B scavenger receptor required for the generation of all-trans-retinal; yet, NINAD and SANTA MARIA cannot substitute for each other. The gene is expressed primarily in the gut, and expression of using a driver that directs expression in the gut rescues the phenotype. Surprisingly, function can also be rescued by expression specifically in neurons (; unpublished data), demonstrating that expression of either in the midgut or specifically in neurons rescues function. Furthermore, and are expressed and function in glia and neurons distinct from those in which functions. The data from the current work enables the formulation of the pathway critical for the conversion of carotenoids to production of all-trans-retinal, which is subsequently metabolized into the chromophore (). The pathway begins with the uptake of dietary carotenoids into the gut, through a process that involves the NINAD scavenger receptor. The β-carotene does not appear to be metabolized in the gut or in the expressing neurons. Rather, the SANTA MARIA scavenger receptor and the NINAB BCO function coordinately in neurons and glia, for the uptake and centric cleavage of β-carotene. The all-trans-retinal is then metabolized into vitamin A and transferred to the retinal pigment cells, where it is converted into the chromophore, through a process involving the PINTA retinoid binding protein (). The NINAG oxidoreductase also participates in the production of the chromophore, in a step subsequent to the formation of vitamin A (; ), although it remains to be determined whether it functions in the retinal pigment cells or in photoreceptor cells. The proposed pathway is not yet complete. Because NINAD appears to operate upstream of NINAB/SANTA MARIA, there may be additional proteins that facilitate the uptake of carotenoids into the gut and in the transport to the neurons and glia that express NINAB/SANTA MARIA. In addition, there may be yet-to-be-identified dehydrogenases, as well as other proteins that participate in the uptake of the chromophore into photoreceptor cells. Given the evolutionary conservation of the known components that are required for vitamin A/chromophore production, these yet-to-be-identified proteins are also likely to participate in the carotenoid metabolic pathway in mammals. The mutant was isolated by performing EMS mutagenesis and screening for second-chromosome mutations affecting the PDA. To perform the screen (Fig. S1), we mutagenized an isogenized , stock with EMS as we have described recently (). The mutagenized flies were mated to DTS91, /CyO flies, and the homozygous viable F3 progeny were screened by performing ERGs, using blue and orange light, similar to the paradigm previously described (). The Bloomington Stock Center was the source for the second-chromosome deficiency kit and the following stocks: , , ; [][]CyO; [][], , , , P{-}3, P{-}TP1, P{[+]=}106, P{}, P{}, , and . W. Pak (Purdue University, West Lafayette, IN) provided , , and ; J. O'Tousa (University of Notre Dame, Notre Dame, IN) provided and flies; and C. Desplan (New York University, New York, NY) supplied the . The fly stocks generated were as follows: , ; /+; , [][]/[];[][], ,;/+, (pigment cell),, ,/,, ,;/+; , /+;,, /+;,, /+;,, ,;/+, ,;/+, ,;/+, ,;/+, ,;/+, ,;/+; , -/-, /, -/-, /-, ;,, ;,, ,;/+, ,/,, ,;/+; , ;/, ;/, ; /, /, /, /; , /, ,;/+, ,;/+. The mutation was crossed to the fly stocks that comprised the second-chromosome deficiency kit. The mutation was uncovered by ()-, which deleted 27E2 to 28D1, but not by other deficiency lines in the kit. To map the locus further, we used smaller deficiencies in the region and found that the mutation was uncovered by (27F4 to 28B1) and (27F3 to 28A1; ). Based on these data, we localized the mutation responsible for the phenotype to 27F4 to 28A1. This interval is ∼30 kb and included 10 genes spanning from to (). ERG recordings were performed as previously described (). In brief, two glass microelectrodes filled with Ringer's solution were inserted into small drops of electrode cream placed on the surfaces of the compound eye and the thorax. A Newport light projector (model 765) was used for stimulation. The ERGs were amplified with an electrometer (IE-210; Warner) and recorded with an A/D converter (MacLab/4s) and the Chart v3.4/s program (A/D Instruments). Five 5-s light pulses (orange, blue, blue, orange, and orange) were given to each fly, and the interval time between each pulse was 7 s. All recordings were performed at room temperature. To express the cDNA under the control of the () promoter and the promoter, the cDNA (EST clone RH67675; Drosophila Genomics Resource Center) was subcloned between the NotI and XbaI sites of the pCaspeR- vector () and same site of the pUAST vector (), respectively. To express SR-BI in flies, the human cDNA (EST clone 6384348; Invitrogen) were subcloned between the NotI and XbaI sites of pUAST. To express GAL4 under control of the , , and transcriptional control regions, we subcloned the following ∼2.0-kb genomic DNA fragments between the KpnI and BamHI sites of the GAL4 vector, pGATB (): , −1790 to +98 base pairs 5′ to the transcription starting site; , −2021 to +13 base pairs 5′ to the transcription starting site; and , −2030 to +30 base pairs 5′ to the transcription starting site. The promoter- fragments were excised from pGATB using KpnI and NotI and introduced between the same sites of pCaspeR4 (). The constructs were injected into of embryos, and transformants were identified on the basis of eye pigmentation. To perform Western blots, fly heads, bodies, and dissected guts were homogenized in SDS sample buffer with a Pellet Pestle (Kimble/Kontes). The proteins were fractionated by SDS-PAGE and transferred to Immobilon-P transfer membranes (Millipore) in Tris-glycine buffer. The blots were probed with mouse anti-tubulin primary antibodies (1:2,000 dilution; Developmental Studies Hybridoma Bank), mouse anti-Rh1 antibodies (1:2,000 dilution; Developmental Studies Hybridoma Bank), rabbit anti-Rh4 antibodies (1:1,000 dilution), rabbit anti-NORPA antibodies (1:2,000 dilution; ), rabbit anti-GFP antibodies (1:1,000; Santa Cruz Biotechnology, Inc.), or rabbit anti–SR-BI (1:1,000 dilution; Novus Biologicals) and, subsequently, with anti-rabbit or -mouse IgG peroxidase conjugate (Sigma-Aldrich). The signals were detected using ECL reagents (GE Healthcare). Total RNAs were prepared using Trizol reagent (Invitrogen), and the RNA samples (2 μg each) were fractionated on 3% formaldehyde and 1.2% agarose gels. The RNAs were transferred to nitrocellulose membranes and allowed to hybridize with P-labeled probes, which were prepared using random primers and a PCR product (nucleotides 300–900 of the cDNA) as the template. The hybridization was performed at 65°C in 7% SDS, 2 mM EDTA, and 0.5 M NaHPO, pH 7.2, and the membranes were washed at 65°C in 0.5× SSC and 0.1% SDS. Fig. S1 shows the genetic scheme to identify mutations on the second chromosome that disrupt the ERG. Fig. S2 shows that , , and function outside of retina for carotenoid metabolism. Fig. S3 shows that the Rh1 protein level in flies does not increase as a result of expression of under control of the . Fig. S4 shows that coexpression of and in the retinal pigment cells of or flies does not restore Rh1 levels. Fig. S5 shows that SR-BI is expressed in the / and / flies. Online supplemental material is available at .

In mature B lymphocytes, binding of antigen or antireceptor antibody to the B cell receptor (BCR) complex causes a sustained rise in intracellular free Ca, which, in turn, regulates proliferation and differentiation of the cells into either memory or antibody-secreting ones (for review see ). The increase in cytoplasmic Ca arises from two successive events: first, there occurs a transient Ca release from intracellular stores initiated by a rise in free inositol 1,4,5-trisphosphate (IP; for review see ). In turn, the emptying of Ca stores activates Icrac (Ca release activated) ion channels. The channels are Ca permeable, and Ca influx via these channels results in a sustained elevation of cytoplasmic free Ca (; for review see ). The molecular identity of the Icrac channels has yet to be fully determined, but recent studies demonstrate that the protein Orai1 (also called CRACM1) is an integral component of the channel and is associated with its Ca selectivity filter (; ). Orai1, a gene product identified in severe combined immunodeficiency patients (), was discovered to be important in Icrac function through RNAi screens (; ; ). Tyrosine kinase activity is absolutely required for activation of the BCR–Ca signaling pathway (for review see ). In the absence of kinase function, the sequence of molecular events linking BCR activation to IP production fails because the normal phosphorylation and activation of PLCγ, the enzyme that produces IP, does not occur. The BCR complex is a multimer consisting of membrane Ig associated with Ig α/β heterodimers. The BCR complex interacts with and is phosphorylated by one or more of the following members of the Src kinase family: Lyn, Fyn, and/or Blk. In addition to the Src kinase family, two other tyrosine kinases participate in the BCR–Ca signaling pathway: Syk (Syk family) and Btk (Tec family). All three tyrosine kinase families participate in PLCγ activation. Furthermore, tyrosine kinase–dependent activation of PLCγ is facilitated by adaptor or linker proteins such as Blnk (for review see ). Whether these enzymes play a role in the events that link the emptying of Ca stores to Icrac activation has not been investigated directly. However, indirect studies using pharmacological blockers of tyrosine kinase function have suggested that the enzymes may play a role in linking Ca store release and Ca influx (, ,). In this study, we provide direct evidence of a role for kinases in the link between Ca store emptying and Icrac activation, and we identify some of the specific enzymes involved. Two general classes of mechanisms have been proposed to link the store release of Ca to Icrac activation. The first class proposes that Icrac channels are structurally linked to the Ca-containing stores and that their activation depends on a conformational coupling between the store and the plasma membranes (; ; ). The second class of mechanisms proposes that second messenger molecules accomplish this link through Ca-dependent activation of target proteins. Over time, several plausible messenger proteins have been proposed: G proteins (; ; ; ), PKC (), tyrosine kinases (; ,; ), Ca influx factor (; ), inositol 1,3,4,5-tetrakisphosphate (), and cytochrome P-450 (). Recently, however, a compelling case has been developed that identifies stromal interaction molecule 1 (STIM1) as a messenger protein between Ca release and Icrac gating (; ; ). Simultaneous overexpression of STIM1 and Orai1 but not either alone facilitates the activation of Icrac (). STIM1 is a membrane-bound Ca-binding protein (its structure includes an EF hand) located in the ER. STIM1 acts as a Ca sensor: store Ca depletion causes STIM1 to cluster as discrete aggregates (puncta) that relocalize to ER membrane areas juxtaposed to the plasma membrane. STIM1 puncta colocalize with Orai1, and Ca release–activated channels open at or near the sites where this protein nexus forms (; ). To further our understanding of the mechanisms that couple Ca store release to Icrac gating, we investigated the features of Icrac and Ca store release in a chicken B lymphocytic cell line, DT40, in which the expression of specific kinases can be genetically manipulated (; ; ). Our results indicate that Lyn or Syk tyrosine kinases are absolutely required to link the Ca released from intracellular stores to Icrac gating: when the simultaneous expression of the enzymes is genetically suppressed, Icrac is not activated by the release of intracellular Ca stores. This effect is reversible and specific and can be mimicked by the application of drugs that suppress kinase enzymatic activity in wild-type cells. The kinase effects do not arise from changes in the expression levels of either Orai1 or STIM1. Moreover, the release of Ca does not change the level of the phosphorylation of Syk, a measure of its enzymatic activity, and only small changes are detected in Lyn. Therefore, tonic signals that generate small pools of phospho-Lyn and -Syk appear to be required for Icrac activation. i n v e s t i g a t e t h e r o l e o f t y r o s i n e k i n a s e s i n t h e m o l e c u l a r m e c h a n i s m s u n d e r l y i n g I c r a c a c t i v a t i o n , w e s t u d i e d t h e e l e c t r o p h y s i o l o g i c a l p r o p e r t i e s o f w i l d - t y p e a n d g e n e t i c a l l y m o d i f i e d D T 4 0 c e l l s e n g i n e e r e d t o s u p p r e s s t h e e x p r e s s i o n o f s p e c i f i c p r o t e i n s o f i n t e r e s t . In B lymphocytes, binding of antigen or antireceptor antibody to the BCR complex recruits and activates tyrosine kinases as part of the events that lead to the production of IP (; ; for reviews see ; ). Results presented in the current study indicate that these kinases also play a role in the link between the IP-dependent release of Ca from ER intracellular stores and activation of the Icrac ion channels in the plasma membrane. We compared the electrophysiological features of Icrac and changes in free cytoplasmic Ca between wild-type and Lyn- and Syk-deficient DT40 cells and found that cells simultaneously lacking Lyn and Syk tyrosine kinases fail to activate Icrac after the release of Ca stores. Interestingly, the genetic deletion of either enzyme alone did not prevent Icrac activation, revealing redundancy in the biological function of these kinases in this response. The physiological effects on Icrac and cytoplasmic Ca flux of pharmacological blockers of these enzymes leads us to suggest that the enzymes may operate in a sequential cascade with the Src family enzyme (Lyn) upstream from Syk. The failure to activate Icrac in the Lyn- and Syk-deficient cells reflects that protein phosphorylation is a required event in the mechanism that couples the emptying of Ca stores to channel activation or in the function of the channels themselves. These alternatives cannot now be unequivocally discerned. There are no previous studies on the electrical properties of lymphocytes with the genetically suppressed expression of tyrosine kinases. However, the hypothesis that tyrosine kinases may play a role in the link between Ca store release and Icrac activation has been proposed previously (; ,; ). Lymphocytes of mice in which the Btk-related Itk tyrosine kinase is knocked out respond to T cell receptor activation with a normal release of intracellular Ca stores, but Icrac activation by TG is less than in wild-type animals (). In addition to observations in animals subjected to genetic manipulation, a possible role for tyrosine kinases in lymphocyte Ca transduction paths has been suggested by experimental observations of some effects of the pharmacological inhibition of tyrosine kinase activity (; ; ; ). Currently, the mechanisms that couple Ca store release to Icrac gating fall into two main categories: physical coupling mechanisms propose that activation of the IP receptors are directly linked through conformational changes to the opening of Icrac channels. An alternative form of physical coupling, the vesicle fusion hypothesis proposes a fusionlike event between intracellular vesicles and the plasma membranes that delivers Icrac channels or their activators to the plasma membrane. Vesicle fusion is mediated by SNARE proteins. Messenger mechanisms propose that the released Ca is detected by particular cytoplasmic Ca sensor molecules that, in turn, gate channel activity. The direct coupling hypothesis was developed to explain findings that Icrac fails to occur after pharmacological treatments expected to disrupt cytoskeletal elements (for example, cytochalasin D [], calyculin A [], 2APB [], or clostridium C3 transferase injection []). Direct coupling supposes that molecular receptors in the ER membranes, IP receptors, or ryanodine receptors are linked to Icrac activation through conformational changes. In lymphocytes in particular, this hypothesis is unlikely because mutant DT40 cells lacking all IP receptor isoforms exhibit normal TG-activated Icrac () and ryanodine receptor knockout animals have normal T cell activation (). The vesicle fusion hypothesis (for review see ) was developed to explain the fact that interference with the function of all or even some of the various proteins that constitute the SNARE complex can block TG-dependent Ca elevation. Although it is evident that SNARE protein can alter the cell's Ca response, it remains questionable whether their specific activation by IP-linked events is required in the course of normal events linking Ca store release and Icrac gating (). Recent studies have identified two novel proteins, STIM1 and Orai1, that participate in the signaling between store Ca release and Icrac activation (; ; ). They are both integral membrane proteins located either in the ER membrane (STIM1) or in the plasma membrane (Orai1). Their simultaneous coexpression enhances Ca-dependent Icrac gating. These may be the first identified constituents of a larger family of proteins involved in the linkage between Ca store release and channel gating. The results we present here firmly indicate that the link between the release of Ca stores and Icrac activation in B cells also involves protein phosphorylation catalyzed by cytoplasmic tyrosine kinases. The target of phosphorylation remains to be identified in future work; it could be STIM1, Orai1, or other proteins yet to be discovered. However, we have demonstrated that kinase activity does not affect the expression levels of STIM1 or Orai1. We emphasize that our results suggest that the state of phosphorylation of a target protein is important, but we have no evidence, nor do we suggest that Ca sensing or channel gating is mediated by phosphorylation. We assayed kinase autophosphorylation as an index of enzymatic activity. Our biochemical results reveal that the activity of one of the critical kinases (Syk) does not change with Ca release, whereas the other (Lyn) changes only by a small extent and with a slow time course. We propose that phosphorylation is required as a tonic signal that sustains the normal function of its target protein rather than a phasic signal activated in the course of Ca release, Ca sensing, or channel gating. IP, TG, ionomycin, fibronectin, antiphosphotyrosine antibody 4G10, LavendustinA, and PP2 (4-amino-5-(4-chlorophenyl)-7-(-butyl)pyrazolo[3,4-] pyrimidine) were obtained from Calbiochem. The STIM1 antibody was purchased from BD Biosciences. M4, an anti-DT40 BCR chain mAb, was prepared as described previously (). Indo1–acetoxy-methyl and Fura2–acetoxy-methyl were obtained from Invitrogen. Phospho-Src family (P-Y416) and phospho–Zap-70 (P-Y319)/Syk (P-Y352) antibodies were purchased from Cell Signaling Technology. pLynGFP and pSykEGPF, which were gifts from A. Defranco (University of California, San Fransisco, San Fransisco, CA), are vectors that promote the expression of Lyn/GFP or Syk/EGFP proteins. DT40 cells were cultured at 37°C in 5% CO in RPMI 1640 with 10% FCS, 1% chicken serum, and 2 mM glutamine. (), and (), and (). Mutant cells were transfected with pLynGFP or pSykEGPF by electroporation and grown for 24 h after transformation before Ca imaging experiments. DT40 Ringer's solution consisted of 145 mM NaCl, 2.8 mM KCl, 10 mM CsCl, 10 mM CaCl, 2 mM MgCl, 10 mM glucose, and 10 mM Hepes-NaOH, pH 7.4. Tight-seal electrodes were filled with a quasi-intracellular solution of the composition 145 mM Cs glutamate, 8 mM NaCl, 1 mM MgCl, 2 mM MgATP, 10 mM EGTA, and added Ca to yield 90 nM of free Ca. Free Ca in the presence of the various Ca-binding molecules was calculated using WinMaxC. In some experiments, we tested the ion selectivity of Icrac by measuring the current in an extracellular solution consisting of 150 mM Na methanesulfonate, 10 mM Hepes, pH 7.4, and 2 mM EGTA with varying amounts of Ca, Ba, and Cd. Osmotic pressure was 300 mosM (adjusted with glucose). In only these experiments, we filled the tight-seal electrode with 128 mM Cs glutamate, 3.16 mM MgCl, and 10 mM EGTA and added Ca to yield 90 nM of free ion. Cells were washed free of incubation medium and resuspended at ∼10/ml in Ringer's solution. 200 μl of the cell suspension was applied onto a fibronectin-coated glass coverslip that formed the bottom of a recording chamber held on an upright microscope (Diaphot; Nikon) equipped with differential interference contrast optics. The cells were maintained at room temperature and continuously superfused with oxygenated recording solution. Drugs of interest were tested by adding them to the superfusing solution. To load, Indo1 was loaded into cells by incubating ∼3 × 10 cells/ml in RPMI 1640 with 10% FBS and 3 μM Indo1–acetoxy-methyl (Invitrogen) for 1 h at 37°C. The cells were then washed three times at room temperature with a DT40 Ringer's solution consisting of 125 mM NaCl, 5 mM KCl, 1 mM CaCl, 0.5 mM MgCl, 1 mM NaHPO 11 mM glucose, and 25 mM Hepes, pH 7.4, with 0.1% wt/vol BSA added. The last cell pellet was maintained on ice until just before measurements. To measure Ca, ∼2 × 10 Indo-loaded cells/ml were suspended in DT40 Ringer's solution with BSA and maintained at 37°C in a stirred cuvette in a fluorimeter (F-4500; Hitachi). Fluorescence emission intensity ratio at 400 and 500 nm (F400/F500) was measured continuously and assessed in response to stimulants injected into the stirred cuvette. In single-cell studies, we used the fluorescent dye Fura2 in a modified epifluorescent microscope. Cells were loaded by incubation in Fura2–acetoxy-methyl (Invitrogen) as described above for Indo1–acetoxy-methyl. Immediately after the last wash, cells were suspended in DT40 Ringer's solution at ∼10 cells/ml, and 200 μl was deposited on an imaging chamber held on the stage of the inverted microscope. The bottom of this chamber was a glass coverslip coated with 0.1 mg/ml poly--lysine, to which cells adhered. Cell fluorescence was excited at either 340 or 380 nm (narrow band interference filters selected with a rotating wheel; Sutter Instrument Co.). Cells were observed using a 20× 0.75 NA Fluor objective (Nikon), and images were captured through a 510 ± 40-nm bandpass optical window using a cooled high resolution CCD camera (MicroMAX 1300; Roper Scientific), an ST133 controller, and WinView32 software (Roper Scientific). Image pairs at 340 or 380 nm were acquired for 5 s each, separated by 0.5 s from one another. Image pairs were repeated at 30-s intervals over 20 min. To minimize fluorophore bleaching, an electronic shutter (LPS25; Vincent Associates) was used to restrict cell illumination to the time of image acquisition. In off-line analysis (Simple PCI software; Compix, Inc.), we selected images of individual cells that met three criteria: diameter within 10% of the population mean, low resting cytoplasmic Ca, and no detectable movement or loss of focus over the time course of the measurements. The mean gray level (at 10 bit resolution) of the emission intensity at 340 and 380 nm was measured for each selected cell at each time point and averaged over the cell population. Absolute cell emission intensity was calculated by subtracting the mean intensity of the background. The cell-bathing Ringer's solution in the microscope imaging chamber was exchanged throughout an experimental measurement at a rate of ∼0.5 ml/min. In addition, the medium specifically bathing cells in a selected field of view was controlled by local superfusion with solutions flowing from a 300-μm–inner diameter polyimide capillary tubing (PT Technologies) placed with its tip in the same plane as the cells and ∼300 μm away. This superfusing medium rapidly (≤200 ms) changed in the course of cytoplasmic Ca level measurements. The polyimide tubing was the exit port of micromanifold (MM-6; Warner Instruments) that allowed selection among six possible solutions. The identity and duration of flow of the superfusing solution were controlled with electronic switch valves (MPS-2; WPI Instruments). 1.0 × 10 DT40 cells were suspended in RPMI, stimulated by the addition of TG or M4, and sampled at different time points after stimulation. Cells were spun, lysis buffer was added, cells were spun again, and the supernatant was collected. 4G10 antibody was added to the supernatant and incubated for 1 h at 4°C on a rocker. Protein G beads were then added, and the suspension was incubated for an additional 2 h at 4°C. The beads with bound tyrosine-phosphorylated proteins were washed four times with lysis buffer and dissolved in standard SDS sample buffer with DTT. Proteins were separated in gradient 5–15% SDS-PAGE, blotted onto nitrocellulose, and reacted with antibodies. The reaction product was located on the blot with secondary antibodies linked to HRP and reacted with the chemiluminescent substrate Luminol. Within 1 min, luminescent images were acquired with a cooled high resolution CCD camera (MicroMAX 1300; Roper Scientific). In off-line image analysis (Simple PCI software; Compix), we measured the absolute gray level of each luminescent band.

Assembly of the human immunodeficiency virus type 1 (HIV) is a highly regulated process that requires the spatially and temporally coordinated recruitment of viral components, as well as key cellular proteins, to an appropriate membrane system. The site of HIV budding has been reported to vary for different host cells; in T cells and several nonhematopoietic cell lines, the majority of virus particles assemble at the cell surface, but in macrophages these events occur almost entirely on intracellular membranes (; ; ; ). Immunolabeling indicates that the intracellular sites of assembly in these cells are enriched in major histocompatibility complex class II (MHC-II) molecules, suggesting that they are MHC-II compartments (MIICs) where MHC-II molecules acquire their peptide cargo (). In support of this view, the intracellular virus-containing compartments (VCCs) also label for some markers of late endosomes, such as the tetraspanin CD63 (; ; ). Several studies with T cells, or cell lines such as HeLa, COS, or HEK293, have also demonstrated a limited and/or transient association of assembling HIV, or of Gag-based viruslike particles, with CD63-containing late endosomes (; ; ; ), suggesting an association between HIV assembly and late endosomes in most cells (). Nonetheless, many studies of T cells and nonhematopoietic cells producing HIV clearly show virus particles budding from the cell surface. This view is supported by several recent studies indicating that HIV Gag is targeted directly to the plasma membrane (; ), probably through a specific interaction of the matrix component of Gag with the plasma membrane lipid phosphatidylinositol 4,5-bis-phosphate (PIP; ; ), and that HIV particles only reach endosomes as a result of endocytosis of virions formed at the cell surface. It has been suggested that this may also be the case for the intracellular accumulations of HIV seen in macrophages (). Much of the evidence for HIV association with endosomes is based on immunolabeling for CD63. CD63 is a well-established marker for late endosomes, which are often multivesicular bodies (MVBs), where it is particularly enriched on the intralumenal vesicles (; ). Several other members of the tetraspanin superfamily have also been reported to be associated with MVB. In particular, CD53, CD81, and CD82 have been found to localize to MVB in B cells and are enriched on exosomes, which are the secreted MVB intralumenal vesicles derived from these cells (; ). Similarly, CD9 has been found on exosomes from dendritic cells (DCs; ). Like CD63, CD81 can be incorporated into the envelope of infectious HIV produced by human monocyte–derived macrophages (MDMs; ; ). Indeed, tetraspanin microdomains containing CD81 and CD63 have been proposed as sites of HIV assembly and release (; ). To further characterize the HIV assembly compartment in human primary MDMs, particularly in regard to its tetraspanin content, we have studied the distribution of CD81 and the related tetraspanins, CD9 and CD53. We show that in uninfected MDMs, these molecules are located at the plasma membrane and in intracellular structures with a complex array of internal CD81/CD9-labeled membrane vesicles and networks. These structures can be reached by antibodies internalized from the cell surface, but are distinct from early endosomes and from the late endosomes and lysosomes identified by CD63 and LAMP-1. On HIV-infected MDMs, virus particles bud into this intracellular CD81/CD9/CD53-containing compartment. Significantly, the VCCs are linked to the cell surface by narrow membrane channels. Notably, HIV infection recruits CD63 to the CD81/CD9/CD53-positive VCCs. Thus, it appears that the HIV assembly compartment in macrophages is not a conventional late endosome, but an internally sequestered plasma membrane domain to which the late endosome marker CD63 is recruited during virus assembly in HIV-infected cells. Previous analyses of HIV-infected MDMs indicated that most virus particles assemble on intracellular structures that contain markers for late endosomes, particularly CD63 (; ). Indeed, CD63 and another tetraspanin, CD81, are incorporated into the envelope of MDM-derived infectious HIV particles (; ). To examine in more detail the distribution of viral antigens compared with these cellular tetraspanins in infected macrophages, we immunolabeled whole cells or semithin (0.5 μm thick) cryosections for virus particles and for CD63, CD81, or the related tetraspanin, CD9 ( and Fig. S1, available at ). Immunofluorescence revealed the usual distribution for CD63 in MDMs (, top row; ), with the tubulovesicular pattern being reminiscent of the tubular lysosomes described in mouse macrophages (; ; ). In contrast, the viral antigens, identified with rabbit antibodies against the viral matrix protein p17 ( and S1 B) or p24 (not depicted) had a much more restricted distribution and appeared to be associated with a subset of the CD63-containing organelles. The distributions of CD81 and CD9 differed from that of CD63; mAbs to both CD81 and CD9 strongly labeled the plasma membrane of infected MDMs () and, in addition, stained several punctate intracellular structures (). In particular, the VCCs were strongly labeled for CD81 and CD9 (, and Fig. S1). This suggests that CD81 and CD9 more closely identify the HIV-containing subpopulation of CD63-labeled organelles. However, the distribution of CD81 and CD9 has not been looked at systematically in uninfected MDMs. Although many tetraspanins are prominently expressed at the plasma membrane, several have been found to be associated with intracellular compartments, including late endosomes. In particular, CD9, CD53, CD81, and CD82 have been localized to the intralumenal vesicles of MVBs or found on exosomes from B cell or DC lines (; ). We analyzed the distribution of these markers in uninfected MDMs by immunofluorescence microscopy. As in the HIV-infected MDMs, CD63 was found in a tubuloreticular network extending throughout the cells (). CD9, CD53, and CD81 showed a distinct distribution. On single confocal sections, CD9 and CD53, for example, were seen at the plasma membrane and also in discrete intracellular puncta (). These puncta measured up to 2–3 μm in diameter, and were larger than the CD63-containing structures, which did not exceed 0.5 μm diam. Typically, cells contained 20–30 of the larger CD9/CD53/CD81-labeled structures, often close to or extending to ∼2.5 μm from the base of the cell. shows by double labeling immunofluorescence that the majority of CD63 in uninfected MDMs was present on LAMP-1–positive tubules and vesicles. As suggested by the distributions described in the previous paragraph, there was no detectable overlap of the CD63-labeled network with the punctate staining for CD81 (). Instead, CD9, CD81, and CD53 colocalized extensively with each other, both at the cell surface and in the intracellular puncta, shown here by costaining of CD9 with CD53 or CD81 (, respectively). To rule out the possibility that these structures represent a subset of early endosomes, we costained the cells with an early endosome marker, EEA1. CD63 and CD9 showed only limited overlap with EEA1 (). The degree of colabeling was quantified using Volocity. This confirmed a high degree of colocalization between CD63 and LAMP-1 (94%), or between CD9 and CD81 or CD53 (84 and 85%, respectively). In contrast, there was minimal overlap between CD63 and CD81 or EEA1 (26 and 15%, respectively), and between CD9 and EEA1 (∼20%), indicating the presence of discrete structures containing either CD63 or the other tetraspanins, which are all largely distinct from early endosomes. To investigate whether the CD81/CD9/CD53-containing compartment in MDMs is related to the endosome system and whether it can be accessed from the plasma membrane, we performed antibody-feeding experiments. Uninfected MDMs were incubated for 3 h at 37°C with mAbs to CD81, CD9, or CD53, and then fixed, permeabilized, and processed for immunofluorescence labeling. The antibodies strongly stained the cell surface (Fig. S2, A and D, available at ). In addition, some of the antibodies were taken up into intracellular vesicles resembling the aforementioned puncta (Fig. S2), whereas an irrelevant control antibody did not enter the cells under these conditions (not depicted). To test whether the internalized antibodies could access the CD81/CD9/CD53-containing compartment, antibody-feeding experiments were combined with steady-state staining for different tetraspanins. shows that anti-CD81 or -CD53 mAbs could reach the CD9-positive compartment (), and internalized anti-CD9 mAbs colocalized with CD53 and CD81 (). Uptake of the antitetraspanin mAbs was temperature dependent, and labeling was restricted to the cell surface after incubation at 4°C (Fig. S2, B and E). This indicates that the CD81/CD9/CD53-containing compartment can be reached by antibodies fed from the cell surface at 37°C, although the mechanism of uptake remains unknown. To analyze the morphology of the MDM CD81/CD9/CD53-containing compartment in more detail, ultrathin cryosections were labeled with anti-CD9 or -CD81 mAbs and protein A-gold (PAG) and examined by EM. On sections labeled for CD9, abundant gold particles were found over cell surface microvilli and membrane ruffles. In addition, we occasionally observed labeling over three types of intracellular structures. The first type is large vacuole-like structures (1–1.5 μm diam) containing various internal membranes, ranging from 50 nm diam vesicles resembling the intralumenal vesicles of MVBs to larger membranes (). The second type is composed of extended membrane swirls, usually consisting of two closely apposed membranes separated by a narrow gap (). These are likely to be profiles of larger membrane sheets in the sections. Occasionally, such membranes were lined with an extended dense coat (), and some MVB-like structures had extensions of the apposed parallel membrane sheets (). The third type appeared like arrays of highly interconnected membranes (). These structures were not enclosed in a continuous limiting membrane, allowing extensions of the cytoplasm into the membrane meshwork and giving it a spongelike appearance. The compartments identified by the anti-CD9 or -CD81 antibodies differed from the structures labeled for CD63. In keeping with the many small vesicles and tubulovesicular structures seen by immunofluorescence, anti-CD63 antibodies labeled various small membrane vesicles and tubules, often close to the nucleus (). This, again, indicates that CD63 and the tetraspanins CD81, CD9, and CD53 are located in distinct structures in uninfected MDMs. Immunofluorescence staining of HIV-infected MDMs demonstrated that the virus, which was detected with anti-p17, was localized to VCCs that costained with CD81, CD9 ( and Fig. S1), and CD53 (not depicted), indicating that HIV assembles in the CD81/CD9/CD53 compartment. The VCCs appeared larger and more irregular than the CD81/CD9/CD53-containing puncta seen in uninfected MDMs, and were usually located higher in the cell (compare ), suggesting that HIV infection may expand the CD81/CD9/CD53-containing structures or cause them to coalesce. In addition to staining of the VCCs, some tetraspanin labeling was also observed in virus-negative intracellular puncta in the cells and at the plasma membrane, indicating that viral components are directed to a subset of the CD81/CD9/CD53-containing membranes. Little, if any, labeling for p17 was seen at the cell surface. EM immunolabeling of ultrathin cryosections of HIV-infected MDMs revealed VCCs that resembled the intracellular CD81- or CD9-labeled structures seen in uninfected cells. The sections in were double-labeled with 5 nm PAG marking p17 and 15 nm PAG identifying CD9 or CD81. Virus particles labeled for p17 could be seen in a variety of structures ranging from 0.5 μm to several micrometers in diameter (), some of which also contained internal vesicles and larger membranes that labeled strongly for CD9 () and resembled the multivesicular structures shown in . The VCCs in also has patches of flat coat on its limiting membrane (compare with Pelchen-Matthews et al., 2003) and a long, curved double-membrane sheet linking its two poles. Double labeling for virus and CD81 revealed similar structures. The VCC in contains some CD81-labeled small vesicles similar to MVB intralumenal vesicles. Two small VCCs in appear to be connected to extended CD81-labeled parallel membrane tracks (the extent of these membranes is shown at lower magnification in Fig. S3 A, available at ). Transmission EM of plastic-embedded, HIV-infected MDMs, where the ultrastructure of virus particles is much clearer, also indicated that HIV accumulated in intracellular structures that morphologically resembled the CD9- or CD81-labeled organelles observed on cryosections. Virions were found in VCCs in the juxtanuclear area (). Often, several such VCCs were located in close proximity, separated by thin membrane bridges, or with fingerlike membrane projections protruding into their interior (). Occasionally, coated pits could be seen on the limiting membranes. Other VCCs were associated with extended swirls of closely apposed membrane sheets (). shows part of a large VCC in the vicinity of an unusual structure consisting of highly interconnected membranes. Although this had an overall rounded appearance, it lacked a continuous limiting membrane, so that tracks of cytoplasm permeated deep into the membrane meshwork. This structure resembles the spongelike structures labeled for CD9 or CD81 (compare , respectively). In , a spongelike structure was located near a VCC, and many other VCCs were found close to, or appeared to arise from, such structures. For example, shows an extended area of interconnected membrane meshwork with peripheral pockets of virus particles; indeed, one virion can be seen in the center of the spongelike region (). Smaller interconnected membranes, which might be residues of spongelike elements, were also apparent in (asterisks). The presence of budding figures and immature virus particles within these structures, and association with the interconnected sponge membranes (, arrowheads), indicates that HIV particles are assembled on these membranes. Overall, spongelike structures of various sizes were found associated with ∼30% of the VCCs. Although these structures were relatively rare in the MDMs, their frequent association with HIV assembly sites indicates that they are involved in virus budding, perhaps providing a source of membranes to form the viral envelopes. Analysis of serial sections of HIV-infected macrophages by EM () further demonstrated the complexity of the virus assembly compartments. Although the end sections show relatively simple VCCs (), the central sections () reveal interconnected chambers and a large spongelike region surrounded by assembled viruses. Budding profiles and immature virus particles are apparent on several of the panels (), indicating that HIV assembly occurs directly into this compartment. Some buds can even be seen to arise from the spongelike structure in the core of the VCCs (). Significantly, although budding profiles and virus particles could be seen in intracellular VCCs, no such structures were found at the cell surface (Fig. S4 A, available at ). The serial section images in Fig. S4 show that even apparently dispersed VCCs are connected in some sections and associated with condensed spongelike structures. Some VCCs were connected to membrane tracks or parallel membrane sheets (; and Fig. S3), suggesting that they may be linked by elongated tubular extensions or via the closely apposed membrane sheets, and raised the possibility that some apparently intracellular VCCs may be connected to the cell surface. To test whether the VCCs are accessible from the cell surface, HIV-infected or uninfected MDMs fed for 3 h with anti-CD81 mAb, were cooled on ice and incubated for 1 h at 4°C in medium containing 10 mg/ml HRP. Subsequently, the cells were fixed and prepared for cryosectioning. Immunofluorescence staining of semithin cryosections for HIV p17, CD81, and HRP indicated that the HRP had reached the CD81-labeled VCC, even though this structure appeared to be located deep within the cell (). Similarly, HRP could access many of the puncta into which CD81 antibodies were internalized on uninfected cells (). Surface connections were also apparent on HIV-infected macrophages fixed in the presence of the electron-dense, cell-impermeant dye ruthenium red (RR). HIV particles located in apparently intracellular VCCs and clearly identified by their characteristic morphology were coated with a layer of RR (). The dye appeared to gain access to the virus particles through the narrow gaps between closely apposed membrane sheets, which could be traced from the cell surface to the deep VCCs (). Indeed, most of the VCCs in this image appear to be connected by narrow membrane tracks. The surface connections were particularly clear on serial sections. The images in show a narrow RR-stained channel linking three interconnected VCCs to the cell surface. Although the VCCs were accessible to a tracer from the extracellular medium, the connecting membrane tracks were always very narrow. The space between the membrane sheets appears to be ∼15–20 nm on Epon sections, or 25–30 nm on cryosections, which is far too narrow for virions (∼140 nm diam) to pass through. Although it appears that HIV targets the CD81/CD9/CD53 compartment, there are some differences between the CD81/CD9/CD53-labeled structures observed in uninfected MDMs and the VCCs in HIV-infected macrophages. The CD81/CD9/CD53-labeled structures were usually observed near the bottom of the cell (), whereas the VCCs were often found in higher, more medial confocal sections (). Furthermore, VCCs often appeared enlarged or expanded compared with the CD81/CD9/CD53-labeled structures in uninfected MDMs. The most striking difference, however, was in the distribution of CD63 relative to the CD81/CD9/CD53 compartment. On uninfected MDMs, there was no overlap between the late endosome marker CD63 and CD81, CD9, or CD53 (), whereas infected MDMs showed colabeling of viral p17 and CD63 ( and S1; ; ). This suggests that CD63 must be recruited to the VCCs in HIV-infected MDMs. To investigate whether CD63 trafficking over the cell surface has access to the VCCs, we incubated HIV-infected macrophages with an anti-CD63 mAb for 3 h at 37°C. Cells were then washed, fixed, permeabilized, and stained with antibodies against CD53 and p17. CD53 weakly stained the cell surface, and also labeled several discrete intracellular puncta (). In addition, it decorated the VCCs identified by anti-p17 labeling (). The internalized anti-CD63 mAb was not seen in the tubulovesicular structures that were so prominent in the steady-state staining studies (compare the staining pattern of fed anti-CD63 [] to steady state [ and ]). Instead, internalized anti-CD63 was seen in dispersed round vesicles and also localized strongly to the VCCs (). When we performed anti-CD63 feeding experiments on uninfected MDMs, the internalized anti-CD63 again did not reach the tubulovesicular network within the 3-h time course examined, but was seen in small vesicles dispersed throughout the cytoplasm (Fig. S5 A, available at ). In a small proportion of these vesicles, the internalized CD63 colocalized with EEA1 (Fig. S5 B) or with CD53 (Fig. S5 C), suggesting that at least a proportion of the cellular CD63 pool cycles over the plasma membrane and passes through early endosomes and the CD81/CD9/CD53 compartment. In HIV-infected cells, this cycling CD63 population may become trapped in the VCCs, perhaps by being incorporated into budding virions (). The distribution of HIV in infected MDMs has been largely determined using antibodies against the Gag protein. However, the production of infectious particles also requires the viral envelope protein (Env). We studied Env using the human anti-Env mAb 2G12, which binds an epitope covering several N-linked glycosylation sites () and appears to recognize post-ER forms of the viral glycoprotein (). To avoid binding of the mAb to macrophage Fc receptors, F(ab′) fragments of 2G12 were generated for immunofluorescence staining. Env was observed in the intracellular VCCs, where it colocalized with the capsid protein p24 (). Env staining in HIV-infected MDMs also overlapped closely with CD81 (). To determine whether Env itself can target the CD81 compartment, Env was expressed in uninfected MDMs by nucleofection. Env was observed almost exclusively in the intracellular CD81-containing compartment () and did not colocalize with the tubulovesicular late endosomes identified by CD63 (). As in uninfected MDMs (), CD63 and CD81 did not colocalize in the Env-transfected MDMs. Thus, the viral Env glycoprotein appears to be targeted to the CD81/CD9/CD53 compartment. In this study, we have further characterized the compartment where HIV assembles in MDMs. Although in macrophages HIV particles accumulate in structures with the appearance of vacuoles that contain the late endosome marker CD63, our results indicate that these compartments are not endosomes. Instead, we have shown that the virus assembly sites are marked by the tetraspanins CD81, CD9, and CD53. Immunofluorescence staining of uninfected MDMs indicated that CD81, CD9, and CD53 are present at the plasma membrane and in a population of intracellular puncta that are distinct from EEA1-containing early endosomes or the tubulovesicular late endosomes/lysosomes marked by CD63 and LAMP-1. EM analysis revealed a complex morphology for these structures, with multivesicular-like components, closely apposed extended membrane sheets, extensive interconnected spongelike membranes, and lipid deposits, which are structures that are clearly distinct from endocytic organelles. Notably, both CD9 and CD81 were located on the internal membranes of these structures, suggesting that, in MDMs at least, multivesicular organelles are heterogeneous; CD9 and CD81 do not mark endosomes and, in uninfected cells, CD63 does not localize to the multivesicular CD81/CD9/CD53 compartment. In addition, we now report that the CD81/CD9/CD53 compartment is linked to the cell surface, both in uninfected and HIV-infected cells, and so may be considered an internally sequestered plasma membrane domain. In MDMs infected with R5-tropic HIV, mature and immature virus particles and buds were seen in intracellular VCCs, as previously described (; ; ; ; ). In observations of a large number of cells on many sections, we saw HIV budding profiles at the limiting membrane of VCCs and on the spongelike membranes and immature virions in the interior ( and ), but not at the cell surface proper. This suggests that the vast majority of HIV particles are formed directly at this site, rather than reaching the intracellular location after endocytosis from the cell surface. Indeed, in infected MDMs, there were very few HIV particles at the cell surface available for endocytosis (Fig. S3 A), and virions were never seen in coated pits or in EEA1-positive early endosomes, as might be expected if virus accessed the VCCs by internalization from the cell surface. The HIV assembly compartment appeared to be derived from a subset of the structures marked by CD81, CD9, and CD53 because (a) immunofluorescence labeling revealed a close overlap of the intracellular virus with these markers, (b) immuno-EM of cryosections showed virus particles associated with structures resembling the CD9- and CD81-containing internal membranes in uninfected MDMs, and (c) an EM analysis of Epon sections of infected MDMs revealed VCCs associated with interconnected spongelike membranes or extended parallel membrane sheets. Although some of the spongelike membranes on Epon sections resemble the paracrystalline ER described in cells overexpressing certain ER proteins (), we did not observe costaining between CD53 or CD81 and the ER marker calnexin (unpublished data). HIV not only exploits the CD81/CD9/CD53 compartment for assembly, it also changes the compartment. Thus, virus infection expands and/or causes the coalescence of the CD81/CD9/CD53 structures, and it alters their distribution in infected cells. Most significantly, the VCCs also label for CD63, which is in agreement with previous studies (; ; ). However, our present studies indicate that CD63 only reaches the CD81/CD9/CD53 compartment after infection by HIV. Antibody-feeding experiments suggest that a pool of CD63 cycles over the cell surface and through CD81/CD9/CD53-labeled structures and that, in infected MDMs, this cycling CD63 can become trapped in the virus-containing CD81/CD9/CD53 compartment, perhaps through incorporation into HIV particles (). Macrophages are unusual in that virus assembly occurs almost exclusively in intracellular VCCs. Nonetheless, experiments in which MDMs were exposed to the fluid tracer HRP at 4°C or fixed in the presence of the membrane-impermeant dye RR indicated that the VCCs are part of an unusual network of surface-connected intracellular structures. As all of the membranes accessed by RR must be connected to the cell surface, they should strictly be classified as plasma membrane. Indeed, the intracellular, CD9-labeled membranes contain the plasma membrane lipid PIP, as demonstrated in preliminary experiments with the phospholipase Cδ1 PH domain linked to GFP (unpublished data). Thus, in MDMs, as in other cells, HIV components appear to be targeted to the PIP-containing plasma membrane, but this domain, though linked to the cell surface, is sequestered within the cell. These observations are consistent with recently published data indicating that HIV assembly does not occur on endosomes, but that the viral Gag protein is directly targeted to the plasma membrane (; ). To date, we have only seen the CD9- or CD81-labeled intracellular structures in MDMs or in some LPS-matured monocyte-derived DCs, but the compartment is rare and difficult to find, even in these cells. Thus, we cannot be sure whether similar structures exist in other cells, such as T cells or HeLa, which express the tetraspanins CD9 and CD81 at lower levels. The surface-connected structures in macrophages may be related to the surface-connected patocytosis compartment that is involved in the uptake and sequestration of aggregated cholesterol (). How the viral components are targeted to the intracellular CD81/CD9/CD53-containing membranes is unclear. Virus assembly may be stabilized at this compartment by the tetraspanins, as it has been suggested that HIV assembly may be gated through tetraspanin-enriched microdomains (). It seems unlikely, however, that virus targeting to tetraspanin domains alone accounts for targeting to the VCCs because high levels of CD9 or CD81 are also found at the cell surface in MDMs, whereas the bulk of CD63 is in distinct late endosomes/lysosomes. Indeed, even within the CD81/CD9/CD53 compartment, most of the tetraspanins appear to be sequestered on the internal membranes. When the HIV Env protein was expressed alone in MDMs after transfection, it localized to intracellular vesicles that costained with CD81, but did not relocate CD63 to these structures, suggesting that other viral proteins are required for CD63 accumulation in the virus assembly compartment. Assembly in the CD81/CD9/CD53 compartment appears to offer several advantages for HIV. The compartment contains large amounts of membrane that can be used by the virus to form its envelopes. The assembled virus particles may then be stored in an infectious form until they are released (). By EM, the channels that connect the virus-containing structures to the cell surface are very narrow (<20 nm on Epon sections), suggesting that although RR (Mr = 786) or HRP (40 kD) can pass through these gaps, the diffusion of larger antibody molecules, which are ∼10 nm long, may be impeded, and virus particles with diameters of 120–140 nm are retained in the cells. Generally, the spacing between the membrane sheets appeared very regular, even over extended distances ( and Fig. S3), suggesting that proteins lining the membrane—perhaps together with the tetraspanins—support these structures. In some regions, we observed flat coats and cytoskeletal meshworks, presumably actin, associated with the membrane tracks; these structures may further support membranes in the closely apposed configuration and perhaps regulate access. Indeed, treatment with cytochalasin D has been reported to disrupt the VCCs in MDMs (). Because the VCCs are connected to the cell surface, virions would be in contact with the extracellular, most likely neutral, pH and not the harsh acidic/degradative environment of endosomes, and any stimuli that would open the connecting membrane channels would lead to the rapid release of virus particles, together with CD81/CD9-containing exosomes and other membrane structures present in the VCCs. The possibility that HIV may be released from the CD81/CD9/CD53 structures in a regulated manner is also suggested by analogy to the interaction of HIV with DCs, where compartments marked by CD81 may play a role during antigen presentation. CD81 has been shown to redistribute to the immunological synapse between T cells and either B cells or DCs, in an antigen-dependent manner (). DCs, at least in tissue culture, are able to capture extracellular virus particles and later release them onto T cells for infection in trans (; ; ). The compartments into which HIV is internalized in DCs resemble the VCCs of MDMs in their size and complexity, internal vesicles and membranes that label for CD81 or CD9, clathrin-coated pits, and extended areas of thickened flat coats. Notably, the DC virus-containing structures also lack markers for early endosomes or late endosomes and lysosomes and stain only weakly for CD63 (). In DCs, HIV can access the compartment by uptake from the cell surface, whereas in MDMs, virus assembles directly into the CD81/CD9/CD53 compartment. The similarities between the viral compartments of DCs and MDMs suggest that the macrophage compartment may also be involved in virus transfer to target cells. Such processes are likely to involve complex membrane rearrangements. In this respect, the narrow membrane channels and surface connections from the MDM VCCs show some resemblance to the tubules induced during stimulation of murine DCs (; ). MDMs are primary cells, but, as they have undergone differentiation in culture, we cannot be sure that the structures we describe in this study are present in macrophages in vivo, though intracellular assembly of virus was observed with both HIV types 1 and 2 and in brain macrophages of an individual with AIDS encephalopathy (). Macrophages are widely distributed in all tissues and organs and are believed to be an important vehicle for spreading HIV infection to CD4 T cells (), perhaps because they sequester virus in intracellular compartments. Indeed, the HIV sequestered in MDMs can remain infectious for many days (). By analogy with studies on DCs, macrophages may likewise release HIV from the CD81/CD9/CD53-containing VCCs when they encounter T cells (). For HIV, virological synapses have been described for conjugates between DCs and T cells or between infected and uninfected T cells (), and macrophages have been shown to release HIV particles onto epithelial cells () or peripheral blood lymphocytes () in a directed manner. The CD81/CD9/CD53 compartment may function in the formation or activity of immunological synapses, and it is these properties that are exploited by HIV to facilitate cell–cell transfer through virological synapses. Moreover, the potential to hide virus intracellularly, beyond the reach of humoral immune responses, may have important implications for vaccine design and AIDS pathogenesis. The mouse mAb against CD63 (1B5 and IgG2b) has been previously described (); anti-CD81 (M38, IgG1) was provided by F. Berditchevski (University of Birmingham, Birmingham, UK); anti-CD53 (MEM-53 and IgG1) and anti-CD9 (MCA469GA and IgG2b) were purchased from Abcam and Serotec, respectively; anti–LAMP-1 (H4A3 and IgG1) and anti-EEA1 (clone 14 and IgG1) were obtained from BD Biosciences. mAbs to HIV p17 (4C9, mouse IgG2a, and ARP342 were obtained from R.B. Ferns and R.S. Tedder, Middlesex Hospital Medical School, London UK), p24 (mAbs 38:96K and EF7, both mouse IgG1, ARP365, and 366, respectively, from B. Wahren, National Bacteriological Laboratory, Stockholm, Sweden), and Env (human mAb 2G12, EVA3064; obtained from H. Katinger, Institute of Applied Microbiology, Vienna, Austria) were obtained through the National Institute for Biological Standards and Control Centralised Facility for AIDS Reagents, which is supported by the EU Program EVA and the UK Medical Research Council. F(ab′) fragments of 2G12 were generated using the ImmunoPure Preparation kit (Pierce). The rabbit antiserum UP595 against HIV p17 was provided by M. Malim (Guy's, King's and St Thomas' School of Medicine, London, UK), rabbit anti-HRP was purchased from Jackson ImmunoResearch Laboratories, rabbit anti–mouse bridging antibody was purchased from DakoCytomation, and biotin-conjugated anti-IgG2b was purchased from Southern Biotechnology Associates. Alexa Fluor–labeled fluorescent antibodies were obtained from Invitrogen, Cy5-streptavidin was purchased from Jackson ImmunoResearch Laboratories, and FITC goat anti–human was obtained from Pierce Chemical Co. PAG reagents were obtained from the EM Lab (Utrecht University, Utrecht, Netherlands). Peripheral blood mononuclear cells (PBMCs) were prepared from buffy coats from healthy donors (National Blood Service, Essex, UK), and adherent monocytes were isolated as previously described (). Monocytes were differentiated to MDMs in complete medium (RPMI 1640, 100 U/ml penicillin, 0.1 mg/ml streptomycin, and 10% human AB serum [PAA Laboratories] with 10 ng/ml of M-CSF [R&D Systems]) for 2 d, and then differentiated without M-CSF until required. HIV-1 was obtained from R. Shattock (St. George's Hospital, London, UK). A first virus stock was prepared by infection of PHA-stimulated PBMC (grown in RPMI 1640 with penicillin, streptomycin, 10% FCS, and 20 U/ml Il-2 [R&D Systems] and treated for 3 d with 0.5 μg/ml PHA-P). PBMC-grown HIV-1 was passaged once through MDMs to maintain tropism. Cell-free supernatants were harvested every 3 d, aliquoted, and stored in liquid nitrogen. Virus titers were determined on NP2-CD4-CCR5 cells as previously described (). For experiments, MDMs were infected 6 d after isolation with HIV-1 (2 FFU/cell) by spinoculation at 2.5 krpm for 2 h at RT, and cultured for various times as indicated. For infectivity assays, NP2-CD4-CCR5, were inoculated by spinoculation as above, and stained with X-Gal after 3 d (). MDMs were grown for 7 d, detached using trypsin/EDTA, and nucleofected with pSVIII HxB2 Env (), using a Nucleofector with the macrophage nucleofection kit and program Y-10 (Amaxa). MDMs were fixed by adding an equal volume of double-strength fixative, 6% PFA (TAAB Laboratories) to the culture medium for ∼30 min at RT. After washing, free-aldehyde groups were quenched with 50 mM NHCl, and the cells were permeabilized for 1 h in blocking buffer (0.1% saponin and 0.5% BSA in PBS) containing 10 μg/ml purified human IgG. Cells were stained with primary antibodies diluted in blocking buffer for 1 h, washed, and incubated for 1 h with appropriate combinations of fluorescent secondary antibodies. Coverslips were mounted in Mowiol (Merck), and examined directly or stored at –20°C. Staining was analyzed with a microscope (Optiphot-2; Nikon) equipped with an upright confocal laser scanner and a Plan Apochromat 60× oil-immersion lens (Nikon) and an upright confocal laser scanner (MRC 1024; Bio-Rad Laboratories). Images were acquired at RT using LaserSharp2000 (Bio-Rad Laboratories) and processed using Photoshop (Adobe) or ImageJ (National Institutes of Health). Volocity version 3.6.1 was used to measure colocalization. HIV-infected MDMs were fixed for 4 h in 2% PFA/3% glutaraldehyde in 0.1 M NaPi buffer, pH 7.4, postfixed for 1 h on ice in 1% OsO/1.5% K[Fe(CN)], and treated with 1.5% tannic acid (). To test accessibility to the cell surface, some samples were fixed for 4 h in the fixative containing 5 mg/ml RR, postfixed 2 h in 1% OsO/5 mg/ml RR, and stained for 1 h in 1% uranyl acetate. Samples were dehydrated in graded ethanols and embedded in Epon 812 (TAAB Laboratories). Ultrathin sections were cut on an Ultracut UCT ultramicrotome (Leica) and stained with lead citrate (sections of RR-treated cells remained unstained). For immunolabeling, HIV-infected MDMs or uninfected control cells were fixed by adding an equal volume of prewarmed double-strength fixative (8% PFA in 0.1 M NaPi buffer, pH7.4) directly into the culture medium. After 10 min, the medium was replaced with 4% PFA. Cells were rinsed in PBS containing 20 mM glycine, embedded in 12% gelatine, infiltrated with 2.3 M sucrose, and frozen in liquid nitrogen as previously described (). Semithin (0.5 μm) or ultrathin (50 nm) cryosections were cut on an Ultracut UCT microtome equipped with an EM FCS cryochamber. Fig. S1 shows immunofluorescence staining for HIV and CD63, CD81, or CD9 on semithin cryosections. Fig. S2 shows control experiments for the feeding of anti-CD81 or -CD9. Fig. S3 shows how closely apposed membrane sheets connect different VCCs, and Fig. S4 shows serial sections of VCCs in HIV-infected MDMs. Fig. S5 shows feeding of anti-CD63 on uninfected MDMs. The online version of this article is available at .

Bacterial protein toxins are useful probes to study endocytic mechanisms and intracellular trafficking pathways (; ; ; ). The VacA toxin ( of 90 kD) is an important virulence factor of this bacterium involved in the induction of gastric ulcers and cancer (). VacA is an A-B toxin that is transported from the cell surface to late endosomes (LEs), where it induces the formation of large vacuoles. The toxin also escapes from its endocytic pathway to target mitochondria (; ; ), but the precise mechanism of this event is still unknown. We have recently determined the sequence of events that leads to the endocytosis and intracellular trafficking of VacA in HeLa and human adenocarcinoma gastric cells (see for schematic representation of the sequence; ). The toxin first binds to the cell surface on lipid rafts (; ; ). Binding of VacA occurs preferentially on lipid rafts clustered on membrane protrusions containing filamentous actin (F-actin) structures that are regulated by the Rac1 GTPase (). Once bound to the cell surface, the toxin is rapidly internalized by a pinocytic mechanism that involves F-actin but is independent of clathrin, dynamin, and of the ARF6 GTPase. This pinocytic mechanism does not require tyrosine phosphorylation of cellular proteins and is not a macropinocytic process (). VacA pinocytosis is regulated by the Cdc42 GTPase activity (), as shown for the internalization of non–cross-linked glycosylphosphatidyl-anchored proteins (GPI-APs; ). Accordingly, VacA is internalized into GPI-AP–enriched early endosomal compartments (GEECs) that have been recently described (; ; for review see ). In contrast to GPI-APs, which, upon pinocytosis into GEECs, are recycled to the plasma membrane via recycling endosomes (), VacA is routed to early endosomes (EEs; EE antigen 1 [EEA1] positive; ) before reaching LEs (). To date, VacA is the only specific marker described for the GEEC–LE pathway. VacA-induced cell vacuolation occurring from LEs () is only slightly affected when microtubules are depolymerized (; ). This suggests that VacA intracellular trafficking toward LEs may be by a microtubule-independent process unlike that of receptor tyrosine kinases degraded into lysosomes (). Indeed, the transport of endocytosed molecules from sorting EEs to LEs is usually accomplished via endosomal carrier vesicles (also termed multivesicular bodies; ) moving on microtubules by the use of molecular motors (). Considering the specific pinocytic mechanism of VacA, we explored the possibility that this toxin may use a way other than microtubules to traffic from GEECs to LEs. In this study, we examined whether VacA intracellular trafficking toward LEs is an actin-dependent process, with VacA transported through endosomal compartments associated with polymerized actin. We then tested whether VacA intracellular trafficking requires the adaptor molecule CD2-associated protein (CD2AP; ; ), which is a good candidate to regulate the actin cytoskeleton and trafficking to LEs. The interaction of CD2AP with actin () and several actin-binding proteins (; ; ) favors a role for CD2AP in the dynamic actin remodeling process. This is also supported by its colocalization with the actin nucleation Arp2/3 complex and cortactin in actin dynamic-rich structures (leading edge of lamellipodia and comet tails; ; , ). Moreover, upon CD2AP overexpression in COS-7 cells, an increase of F-actin patches surrounded by CD2AP together with a decrease in stress fibers has been observed (; ). Several pieces of evidence support a role for CD2AP in the regulation of trafficking to LEs. The interaction of CD2AP with Rab4 and c-Cbl was implicated in the trafficking between EEs and LEs (). In accordance with this observation, have shown that T cells isolated from CD2AP mice failed to down-regulate the T cell receptor (TCR), although receptor internalization was not affected (). Finally, electron microscopic analysis of podocytes from CD2AP haploinsufficient mice revealed defects in the formation of multivesicular bodies (). The interaction of CD2AP with CAP-Z (capping protein of the Z line) and cortactin, two actin-binding proteins, is involved in down-regulation of the activated TCR and EGF receptor, suggesting a link between receptor degradation and actin polymerization (; ). In the present study, we explore the function of CD2AP for the sorting of VacA from GEECs to LEs. We have previously unraveled the four steps of the endocytic process that drives VacA from the plasma membrane to LEs in HeLa cells, as depicted in (). After binding to the cell by a Rac-dependent process (step 1) for 1 h at 4°C and upon warming to 37°C, VacA accumulates into GEECs within 10 min by a Cdc42-dependent nonmacropinocytic process (step 2), is enriched in the EEs within 30 min (step 3), and is transferred to LEs within 120 min (step 4). The aim of this study is to examine whether the dynamics of the actin cytoskeleton are required for the transfer of VacA from GEECs (step 2) to EEs (step 3) and LEs (step 4). We first tested for a polymerized actin requirement in VacA intracellular trafficking in HeLa cells using actin poisons. After VacA binding at 4°C, cells were warmed for 30 min to allow the internalization of VacA into GEECs and EEs (). As shown previously, VacA molecules were colocalized with EEs, but a sizable portion of VacA was still associated with GEECs ( and see Video 1 for 3D reconstructions; available at ). In a subsequent step, DMSO (for control), cytochalasin D (CD), or latrunculin B was added into the medium for the remaining 90 min required for the toxin to accumulate in LEs. In control cells, VacA was no longer associated with GEECs or EEs but colocalized with LEs ( and see Video 2 for 3D reconstructions). It is worth noting that VacA induced a strong clustering of LEs () as shown previously (). In contrast, in cells treated with actin poisons, a large portion of the pinocytosed VacA molecules did not reach LEs. Most of the internalized toxin molecules were found in GEECs based on cell peripheral localization of the toxin-labeled compartment and on the absence of EEA1 labeling associated with VacA (; and see Video 3 for 3D reconstructions). However, rare toxin molecules, which were probably already en route to LEs after 30 min of VacA pinocytosis, were found colocalized with lysosome-associated membrane protein 1 (LAMP1) compartments (). We next determined whether the transport of VacA from GEECs to LEs requires polymerized actin using a functional toxin test. VacA-induced cell vacuolation occurs from LEs, and we have previously shown that the vacuolation process itself does not depend on F-actin (). Therefore, we investigated whether VacA-induced cell vacuolation was impaired by the addition of CD during the course of VacA internalization (). When VacA internalization was started in the presence of CD, cells did not develop vacuoles (). This confirms that pinocytosis of VacA is a strictly F-actin–dependent process (; , ; ). When CD was added 15 or 30 min after the beginning of VacA internalization, vacuolation was again not observed (), although VacA was already in GEECs and EEs (). When CD was added 45 min after the beginning of VacA pinocytosis, a small but definite vacuolation was observed (). The addition of CD at later times was progressively less and less efficient at blocking VacA-induced cell vacuolation. Together, these data support a model in which the transport of VacA from GEECs to LEs via EEs is dependent on polymerized actin. Because the transport of VacA appears to depend on polymerized actin, we searched for the presence of F-actin structures associated with intracellular VacA-positive vesicles. VacA molecules internalized for 30 min were consistently present in vesicular compartments associated with F-actin structures (, insets 2–5). However, no polymerized actin structures were detected at the level of GEECs that still contained VacA (, inset 1). Polymerized actin structures associated with VacA-positive vesicles were localized most often in a polarized fashion onto the surface of the vesicles ( and see Video 4 for 3D reconstructions; available at ), resembling the dynamic F-actin structures found on rocketing endosomes (). To ascertain that F-actin structures at the tip of VacA-containing endosomes could be observed in vivo and induced an actin-based motility of vesicles, we intoxicated GFP-actin–transfected HeLa cells with a fluorescent-tagged VacA (Cy5-VacA [VacA toxin labeled with Cy5 dye]). GFP-actin–transfected HeLa cells were incubated for 1 h with Cy5-VacA at 4°C, washed, incubated for 30 min at 37°C, and analyzed by confocal live cell microscopy. In these conditions, vesicles containing Cy5-VacA and harboring F-actin structures were easily observed ( and S1). Furthermore, these structures were dynamically moving (Videos 5 and 6). To ascertain that vesicles containing VacA were moving by an actin-based motility mechanism, CD was added to Cy5-VacA–intoxicated HeLa cells after 20 min of endocytosis. As shown in Fig. S2, vesicles containing VacA had a threefold reduction of their motility in the presence of CD compared with the control preparation. This indicates that actin polymerization is required to efficiently propel VacA-containing endosomes. We next analyzed the type of vesicles containing VacA that harbored F-actin structures using specific markers of EEs (EEA1) or LEs (LAMP1). 30 min after the onset of VacA internalization, the vesicular compartments associated with F-actin stuctures and containing VacA were all labeled with EEA1 (; see Fig. S3 A for quantification and Video 7 for 3D reconstructions; available at ). LEs were clearly not associated with polymerized actin structures in these conditions ( and see Fig. S3 A for quantification). We then investigated whether VacA was required for the formation of F-actin structures. As shown in , after 30 min of internalization, VacA was associated with fluorescent dextran (a marker of fluid-phase uptake) in EEs, indicating that GEECs are implicated in fluid-phase uptake as previously shown (). Moreover, it has recently been shown that dextran is specifically taken up through a Cdc42-dependent mechanism like VacA (). In cells incubated with fluorescent dextran but without VacA, EEA1-positive vesicles containing dextran still exhibited polymerized actin structures at their tips (). CD2AP is a membrane-associated adaptor protein (; ) that has been implicated in the control of ligand intracellular trafficking toward the degradative pathway (; ; ; ). Importantly, CD2AP binds F-actin () and the Arp2/3 activator cortactin (). Therefore, we investigated whether CD2AP might be involved in VacA intracellular trafficking. We first investigated a possible colocalization between CD2AP and VacA. After 30 min of VacA endocytosis, both overexpressed and endogenous CD2AP were associated with VacA-containing vesicles (). Interestingly, CD2AP appeared to make a bridge between the surface of VacA-containing vesicles and the F-actin structures (). On the other hand, CD2AP was not detected in the GEECs that still contained VacA molecules (). Importantly, endogenous CD2AP and cortactin were colocalized (), suggesting that the two molecules could be involved in formation of the F-actin structures at the level of these intracellular vesicles. In light of our results, we next investigated whether CD2AP could be associated with dynamic F-actin structures in the cytosol. HeLa cells cotransfected with GFP-actin and DsRed-CD2AP but not treated with VacA were analyzed by live cell fluorescence microscopy. As shown in and in Video 8 (available at ), CD2AP was associated at the tip of F-actin structures that were moving randomly in the cytosol. To study the possible role of CD2AP in formation of the dynamic actin structures, we looked for dominant-negative effects by expressing different domains of CD2AP (). From its N to C termini, CD2AP contains three SH3 domains, a proline-rich region, and a coiled-coil domain (; ). Expression of full-length CD2AP in HeLa cells did not affect the formation of F-actin structures (). In contrast, expression of the first two SH3 domains of CD2AP (CD2AP-1–175) totally blocked the formation of F-actin structures compared with surrounding nontransfected cells, and the level of cortical F-actin was greater in transfected cells (). Expression of the C-terminal part of CD2AP (CD2AP-194–639) did not affect the formation of F-actin structures (; see Video 9 for 3D reconstructions of the effect of the three constructs and Fig. S3 B for quantifications; available at ). As a control for the specificity of the N-terminal domain of CD2AP, we expressed the SH3 domains of intersectin (Inter-SH3; ). As shown in , the expression of Inter-SH3 did not block the formation of F-actin structures. We next investigated the involvement of CD2AP in the formation of F-actin stuctures at the tips of EEs using an siRNA strategy that we have previously validated (). As shown in , the expression of CD2AP was decreased by nearly 90% in CD2AP-siRNA–treated HeLa cells. As observed in and Fig. S4 (available at ), CD2AP depletion led to a massive actin cytoskeleton remodeling with numerous stress fibers and a decrease in cortical actin labeling. These effects were not seen in cells transfected with control siRNA ( and S4). Importantly, in the CD2AP-depleted cells, no F-actin structures were observed associated with EEA1-positive compartments still present in the cytosol ( and S4). Altogether, these results indicate that the presence of F-actin structures on EEA1-positive EEs depends on CD2AP. We next studied the role of CD2AP-regulated F-actin structures in the trafficking of VacA from GEECs to LEs. To this aim, we investigated the effects caused by overexpression of the different CD2AP constructs on the transfer of the toxin from GEECs to EEs (after 30 min of pinocytosis) and to LEs (after 120 min of pinocytosis). In GFP-CD2AP–transfected cells, 30 min after VacA pinocytosis, the toxin was found in GEECs and in F-actin structure-associated EEs (). In contrast, overexpression of the first two SH3 domains of CD2AP (CD2AP-1–175) led to an increased labeling of VacA in GEECs ( [inset 2] and B). In these cells, VacA did not reach EEs compared with the nontransfected cells in the same field or with cells overexpressing Inter-SH3 (). Moreover, in the CD2AP-1–175-transfected cells, no vesicles associated with F-actin structures were observed at differences with the nontransfected cells or with the GFP-CD2AP–transfected cells (). Overexpression of the C-terminal part of CD2AP (CD2AP-194–639) did not lead to an increased labeling of VacA into GEECs, and all EEs containing VacA harbored F-actin structures at their tips (). Upon 120 min of pinocytosis, the expression of GFP-CD2AP did not affect the arrival of VacA into LEs (). In contrast, expression of the first two SH3 domains of CD2AP (CD2AP-1–175) blocked the arrival of VacA into LEs, and VacA remained associated with GEECs (). Expression of either the C-terminal part of CD2AP or of Inter-SH3 did not impair the arrival of VacA into LEs (). To further explore the role of CD2AP in VacA transport from GEECs to LEs, we studied whether VacA-induced cell vacuolation was affected in cells overexpressing CD2AP or truncated forms of that protein. The toxin-induced vacuolation of cells overexpressing the different CD2AP constructs was followed by videomicroscopy ( and Video 10; available at ), and the number of vacuolating cells was quantified (). Only the expression of CD2AP-1–175 was able to produce a substantial decrease in the formation of toxin-induced large vacuoles in cells ( and Video 10). To definitively establish the role of CD2AP in the intracellular trafficking of VacA toward LEs, we depleted HeLa cells of CD2AP using siRNA. In CD2AP-depleted cells, after 120 min of VacA endocytosis, the toxin did not reach LEs as it did in cells treated with the control siRNA. Instead, VacA remained within the GEEC (). Furthermore, the formation of toxin-induced large vacuoles was inhibited by 50% in CD2AP-depleted cells (). We verified that CD2AP depletion did not affect the pinocytosis of VacA using a biochemical assay (). Cells were treated with proteinase K to degrade extracellular VacA. The amount of remaining VacA corresponding to the toxin that has been internalized was determined by Western blotting. As a control, cells were treated with CD to block VacA pinocytosis as described previously (). Although CD totally blocked the pinocytosis of VacA (), no difference in the amounts of pinocytosed VacA between siRNA control and siRNA CD2AP–treated cells was observed (). Collectively, these results show that CD2AP is implicated in the sorting and transport of VacA from GEECs to LEs by regulating the formation of F-actin structures at the tip of EEs. #text HeLa cells (from human cervical carcinoma; a gift of T.L. Cover, Vanderbilt University, Nashville, TN) were cultured and transfected as previously described (; ). Full-length CD2AP cDNA () as well as sequences encoding for the truncated forms of CD2AP were amplified by PCR and subcloned into pEGFP-C1 or into pDsRed-C1 expression plasmids (BD Biosciences and CLONTECH Laboratories, Inc.) in frame with the sequence encoding GFP or DsRed. pEGFP–CD2AP-1–175 was obtained using site-directed mutagenesis as described previously (). The five tandem SH3 domains of intersectin tagged with GFP () were provided by P.S. McPherson (McGill University, Montreal, Canada). VacA was purified from the 60190 strain (49503; American Type Culture Collection) according to . Immediately before use on cells, purified VacA (2 μg/ml in all experiments) was acid activated as described previously (). Rabbit polyclonal IgG 958 anti-VacA and -VacA mAbs were gifts from T.L. Cover (). Polyclonal anti-CD2AP antibodies were purchased from Santa Cruz Biotechnology, Inc. Anticortactin mAb was obtained from Upstate Biotechnology. Anti–tubulin mAb was obtained from Sigma-Aldrich. Anti-EEA1 and -LAMP1 mAbs were purchased from BD Biosciences. Texas red–and Cy5-labeled secondary antibodies against rabbit IgG were purchased from Invitrogen, and anti–mouse secondary antibodies labeled with FITC or Cy5 were obtained from DakoCytomation. Phalloidin-FITC and -TRITC were purchased from Sigma-Aldrich, and Texas red–coupled phalloidin and Texas red–labeled or FITC-labeled 70-kD dextran were obtained from Invitrogen. Purified VacA toxin was labeled with Cy5 dye by incubation in borate buffer, pH 8.5, and labeled VacA was separated from unconjugated dye by gel filtration and eluted in PBS. Cy5-VacA was stored in melting ice and alkali activated () by drop-wise addition of 0.4 N NaOH immediately before use. To rule out the possibility that the labeling procedure may alter the biological behavior of the toxin, both the vacuolating activity and the intracellular trafficking of Cy5-VacA was compared with those of the native toxin (Fig. S5, available at ). Immunofluorescence studies and analysis by confocal microscopy were performed as previously described (; ) using a PL APO 63× NA 1.40 oil objective (TCS SP; Leica). The images were combined and merged using Photoshop software (Adobe). For endogenous CD2AP labeling, cells were fixed with methanol rather than PFA as described previously (). For LAMP1 detection, 0.1% Triton X-100 in the permeabilization buffer was replaced by 0.1% saponin. For 2D deconvolution or 3D reconstructions, fluorescence signals were recorded using a microscope (Axiovert 200; Carl Zeiss MicroImaging, Inc.) equipped with a shutter-controlled illumination system (HB0100 with Fluo Arc; Carl Zeiss MicroImaging, Inc.) and a cooled digital CCD camera (CoolSnap HQ; Roper Scientific). Images were reconstructed using MetaMorph 2.0 image analysis software (Universal Imaging Corp.) and QuickTime Pro 5 (Apple). For 3D reconstructions using confocal sections, ImageJ software was used (National Institutes of Health; .nih.gov/ij/). Studies of VacA endocytosis were performed as previously described (). In brief, VacA was first incubated with cells for 1 h at 4°C. After three washes in cold DME, cells were transferred to DMSO-DME, cytochalasin-DME, or latrunculin B–DME prewarmed to 37°C and incubated at 37°C for the times indicated in the figure legends before being processed for immunofluorescence analysis. CD (Sigma-Aldrich) and latrunculin B (Sigma-Aldrich) were used at 20 μM. The fluid-phase marker Texas red–labeled 70-kD dextran was used as previously described (). Quantification of colocalization was performed using the cell counter plugin in ImageJ. VacA-induced cell vacuolation and quantification of the degree of cell vacuolation by means of neutral red dye uptake assay were performed as previously described (). HeLa cells (40–50% confluent) cultivated in DME containing 10% FCS in 35-mm-diameter dishes were transfected with siRNA directed against CD2AP (Santa Cruz Biotechnology, Inc.) or a scramble siRNA (Eurogentec) using Oligofectamine (Invitrogen) as previously described (). Depletion of CD2AP expression was determined by Western blotting. Cells were filmed under constant conditions (5% CO at 37°C) with a motorized microscope (Axiovert 200; Carl Zeiss MicroImaging, Inc.) and a cooled digital CCD camera (CoolSnap HQ; Roper Scientific) and were observed by fluorescent or phase-contrast optics (LP Achroplan phase 2 20× NA 0.40 and plan-NEOFLUAR 40× NA 1.3 objectives lenses). Images were recorded and processed using MetaMorph 2.0 and QuickTime Pro5. HeLa cells pretreated with CD or transfected with siRNA were incubated with activated VacA at 4°C for 1 h, washed, and incubated for 2 h at 37°C to induce the internalization of VacA (t = 2 h). At the end of the incubations, cells were treated or not treated with 1 μg/ml proteinase K (Sigma-Aldrich) for 30 min at 4°C to degrade extracellular VacA. Cells were then washed extensively with medium containing 2 mM PMSF, 20% BSA, and 20% FCS to block the activity of proteinase K and then with medium containing 2 mM PMSF. Cells were scrapped in Laemmli buffer, and the amount of intracellular VacA (i.e., proteinase K resistant) was determined by Western blotting with a specific antibody. Fig. S1 shows the two cells from which two enlarged regions are presented in Video 5. Fig. S2 shows that treatment of cells by CD inhibits the motility of VacA-containing vesicles. Fig. S3 presents some quantifications of F-actin structures. Fig. S4 shows full fields of cells and some sequential confocal sections from the bottom to the top of the cells selected in B. Fig. S5 shows that the Cy5-VacA toxin has intracellular trafficking and vacuolating activity identical to that of the native cytotoxin. Videos 1, 2, and 3 are 360° rotations of 3D reconstructed HeLa cells presented in B, C, and E, respectively. Video 4 is a 360° rotation of a 3D reconstructed part of a HeLa cell processed as described in Video 5 shows that vesicles containing VacA exhibit F-actin structures and move rapidly in the cytosol. Video 6 shows a specific field of Video 5 focusing on the motility of a VacA-containing vesicle associated with an F-actin structure. Video 7 is a 360° rotation of a 3D reconstructed HeLa cell processed as described in A. Video 8 shows that CD2AP is associated with dynamic F-actin structures. Video 9 is a 360° rotation of 3D-reconstructed HeLa cells processed as described in (C–E). Video 10 shows that expression of the first two SH3 domains of CD2AP inhibits the vacuolation of LEs induced by VacA. Online supplemental material is available at .

Several lines of evidence have implicated the minor phospholipid phosphatidylinositol-4,5-bisphosphate (PtdInsP) as a key regulator of clathrin- and actin-dependent endocytosis. First, PtdInsP-binding domains are present in many endocytic proteins, including AP-2, Hip1R/Sla2p, epsin/Ent1p/Ent2p, AP180/YAP180, amphiphysin/Rvs161/167, and dynamin, and abrogating the PtdInsP binding ability of these proteins impairs endocytosis (; ; ; ; ; ; ; ). Second, several phosphoinositide kinases and phosphatases, which are involved in controlling the production and elimination of PtdInsP, physically interact with endocytic proteins (; ; ; ; ), and mutations in these enzymes often lead to altered PtdInsP levels and endocytic defects (; ; ). Third, PtdInsP has been implicated in actin assembly (), which is important for endocytosis in many cell types. Although these various observations suggest that modulation of PtdInsP levels is integral to coordination of endocytic events, little is known about how PtdInsP levels at endocytic sites change as a function of time, and how such changes might affect the functions of endocytic proteins at different steps of endocytic internalization. In the past several years, quantitative analysis of endocytic protein dynamics using multicolor real-time fluorescence microscopy and particle tracking algorithms has greatly contributed to our knowledge of the endocytic internalization process (). To extend this type of analysis to PtdInsP dynamics during endocytic internalization, probes that bind to PtdInsP specifically at endocytic sites are necessary. Evidence for PtdInsP enrichment at endocytic sites has not been obtained. GFP-tagged PLCδ-PH, which binds to PtdInsP with high specificity in vitro (), has been widely used in studying PtdInsP localization in live cells. However, when expressed in yeast at high levels, GFP-2xPLCδ-PH from rat labeled the yeast plasma membrane uniformly instead of showing the patchlike localization that characterizes endocytic proteins (). Recent studies identified a new PtdInsP binding domain, the AP180 N-terminal homology (ANTH) domain (; ). Interestingly, most ANTH domain–containing proteins are involved in endocytosis. Thus, it is possible that an ANTH domain could detect PtdInsP at endocytic sites, especially if it is expressed at endogenous levels. Although PtdInsP-binding domains have so far not been localized specifically to endocytic sites, synaptojanins, which are inositol-polyphosphate 5-phosphatases, have been detected at clathrin-coated endocytic intermediates at nerve terminals and at endocytic sites in yeast cells (; ; ; ). Therefore, high time resolution live-cell imaging of synaptojanin proteins could predict when PtdInsP levels change in vivo. Because overexpression of inositol-polyphosphate 5-phosphatases is expected to change PtdInsP levels, it is critical to perform this type of analysis on synaptojanin expressed at endogenous levels. However, because the fluorescence signal of endogenous synaptojanin proteins is often difficult to detect, quantitative analysis of endogenous synaptojanin protein dynamics has not yet been done. , provides the ability to test PtdInsP functions at endocytic sites. cells are two- to threefold higher than in wild-type cells (). cells also show a severe defect in both receptor-mediated and fluid-phase endocytosis (). Moreover, this mutant exhibits abnormal deep invaginations of the plasma membrane proposed to be generated by the formation of an endocytic bud, which cannot undergo fission (; ). However, the dynamics of endocytic proteins associated with these abnormal membrane structures has not been explored. In this study, we used two functional GFP-tagged protein probes, Sla2 ANTH-GFP and Sjl2-3GFP, expressed at endogenous levels to quantitatively analyze the dynamic turnover of PtdInsP during endocytic internalization in live cells. Furthermore, we also investigated how PtdInsP turnover affects different stages of endocytic vesicle formation by examining dynamics of various GFP-tagged endocytic proteins in synaptojanin mutants. To detect PtdInsP at endocytic sites in vivo, we first needed to develop and validate a probe. Sla2p is an essential protein for yeast endocytosis that contains an ANTH domain, a coiled-coil domain, and a talin-like domain (; ; ; ). Previously, we showed that Sla2p binds through its ANTH domain to PtdInsP specifically (). We fused GFP to the C terminus of the first 360 aa of Sla2p, which includes the ANTH domain but not the central coiled-coil domain or the talin-like domain. The wild-type gene was replaced with this construct so that ANTH-GFP would be expressed from the promoter. We will refer to the resulting strain as . As a control, we also generated a -GFP mutant, in which four lysine residues of Sla2 ANTH required for PtdInsP binding in vitro were changed to alanine (). Sla2 ANTH-GFP and Sla2 4K-A ANTH-GFP expression levels were shown to be similar by using antibodies against Sla2p or against GFP on immunoblots of cell extracts (unpublished data). Although strains are temperature sensitive (), exhibited normal growth at 37°C (). In addition, although endocytosis is completely inhibited in cells (), the endocytic membrane marker FM4-64 was internalized to the vacuole membrane in cells, albeit in 60 min (), which is slower than in wild-type cells (; ). We also examined the internalization rate of cells using the [S]methionine-labeled α-factor uptake assay. As shown in , the α-factor uptake rate for the strain was about half that of wild-type cells (). These results demonstrate that Sla2 ANTH-GFP is able to at least partially perform Sla2p's function in endocytic internalization. Previously, we showed that wild-type Sla2-GFP colocalizes with and exhibits similar dynamic behavior to the endocytic coat protein marker, Sla1p (). Both Sla2p and Sla1p form cortical patches at the plasma membrane, and then move off the cortex toward the cell center, which likely represents membrane invagination. Interestingly, Sla2 ANTH-GFP showed patchlike localization on the plasma membrane, whereas Sla2 4K-A ANTH-GFP was only detected in the cytoplasm ( and Video 1, available at ), indicating that the interaction of the ANTH domain with PtdInsP is crucial for the patchlike localization. The lifetime of Sla2 ANTH-GFP patches was 105 ± 28.7 s, which is ∼3 times longer than the lifetime of wild-type Sla2-GFP patches (36.6 ± 6.4 s). The longer lifetime of Sla2 ANTH-GFP patches is consistent with the reduced rate of endocytosis observed in this mutant. Furthermore, compared with Sla2-GFP, more Sla2 ANTH-GFP was observed in the cytoplasm, which is probably because the coiled-coil domain binds to other endocytic proteins, and the talin-like domain binds to F-actin (; ; ). These interactions may enhance the recruitment of full-length Sla2p to endocytic sites. cells, in which Sla2 4K-A ANTH-GFP does not localize to the cell cortex, showed a growth defect and complete endocytosis defect that was characteristic of cells (, A and C; ). Together, these data indicate that the Sla2 ANTH-GFP is sufficient for localization at the cell cortex, and that this localization is dependent on PtdInsP binding. To confirm that Sla2 ANTH-GFP patches correspond to endocytic sites, we performed two-color real-time microscopy. Most clathrin light chain 1 (Clc1)-GFP patches on the cell cortex colocalized with ANTH-mCherry (). Kymographs revealed that ANTH-mCherry joins the patches shortly after cortical Clc1-GFP appears in the patches (). As shown in , Sla2 ANTH-GFP and the endocytic coat protein marker Sla1-mCherry accumulate at the cell cortex with similar timing ( and Video 2, available at ), and then make an inward movement together when the cortical actin marker Abp1p joins the patch ( and Video 3). The yeast cortical actin patches are endocytic sites, and the inward movement of the endocytic coat represents membrane invagination (). These results suggest that Sla2 ANTH-GFP serves as a valid PtdInsP marker at endocytic sites in vivo. Furthermore, the synchronous dynamics of the Sla2 ANTH-GFP and Sla1-mCherry provide support for the idea that endocytic coat protein recruitment and subsequent disappearance at endocytic sites are, at least in part, the result of changes in PtdInsP levels. To further examine PtdInsP dynamics during endocytic internalization in wild-type cells, we tagged with GFP a phosphoinositide kinase and two phosphoinositide phosphatases, which control PtdInsP levels and have been implicated in endocytosis (; ; ). The GFP-tagged proteins were expressed from the endogenous promoters of the respective genes. Genetic studies confirmed that these GFP-tagged enzymes are functional (Fig. S1 A). The motivation for these studies was to monitor the dynamics of these enzymes at high temporal and spatial resolution to predict how local PtdInsP concentration changes may relate the specific steps in endocytic internalization. Mss4p is the only phosphatidylinositol-4-phosphate 5-kinase expressed in budding yeast (; ). Mss4-GFP was observed both in the cytoplasm and at the cell cortex (Fig. S1 B). Although Mss4-GFP showed a patch pattern at the plasma membrane, it was not polarized like endocytic proteins. Moreover, colocalization between Mss4-GFP and endocytic coat marker proteins, such as Ede1p and Sla1p, was not apparent (unpublished data). These results suggest that PtdInsP may not be selectively synthesized at sites of endocytosis. The yeast synaptojanins Sjl1p and Sjl2p, which are inositol-polyphosphate 5-phosphatases, have previously been implicated in endocytosis (). Although Sjl1-GFP mostly localized to the cytoplasm (Fig. S1 D), Sjl2-GFP appeared in polarized patch structures at the cell cortex (Fig. S1 C). Interestingly, more cortical Sjl1-GFP was observed in an strain (Fig. S1 E), which is consistent with the possibility that Sjl1p and Sjl2p may share functions. Based on these observations, Sjl2p appeared to be the most suitable probe for dynamic regulation of PtdInsP levels during endocytosis. Overexpressed Sjl2-GFP (lacking the final 37 Sjl2 amino acids) was previously shown to colocalize with Abp1-DsRed in live cells (). However, because DsRed is an obligate tetramer, actin dynamics may have been altered in those studies. Indeed, expression of Abp1-DsRed in our strain background caused abnormal actin patch dynamics (unpublished data). Abp1 tagged with GFP or monomeric RFP is a well-validated marker for endocytic actin structures (; ). Moreover, overexpression of Sjl2p may have altered PtdInsP levels or protein localization in the earlier study. To monitor Sjl2p dynamics quantitatively during different steps of endocytic internalization, expressing Sjl2p at endogenous levels is critical. We created a construct to fuse 3GFP to the Sjl2p C terminus, and expressed the fusion protein from the chromosomal locus ( and Video 4, available at ). Phenotypic analysis confirmed that Sjl2-3GFP is functional (Fig. S1 A). Sjl2-3GFP appeared at cortical patches. The lifetime of Sjl2-3GFP patches was ∼9.8 ± 1.4 s, which is shorter than the lifetime of the cortical actin marker Abp1 (14.7 ± 1.6 s). Sjl2-3GFP patches had an initial nonmotile phase, which was followed by a transition to a highly motile phase (). The fluorescence intensity of the Sjl2-3GFP patches developed in a very regular manner, reaching maximum intensity in only ∼3 s, indicating that Sjl2p is rapidly recruited to the cell cortex (). Sjl2-3GFP remained at its maximal level for ∼5 s, and then its levels started to decrease as it moved off of the cell cortex. We previously established analytical probes and methods for dissecting different steps of endocytic internalization in live cells (). Actin assembly (detected using Abp1-GFP) accompanies the slow inward movement of the endocytic coat (labeled by fluorescent Sla1p) off of the plasma membrane. The slow Sla1p movement, and the subsequent fast Abp1p movement, likely reflect membrane invagination and movement of the released endocytic vesicle, respectively (; ). To determine the exact endocytic steps during which Sjl2p is associated with endocytic sites, we imaged Sjl2-3GFP and Abp1-RFP simultaneously with a high temporal resolution of 1 s (Video 5, available at ). As shown in (A–C), Sjl2-3GFP appears 5–6 s after Abp1p, and they move rapidly away from plasma membrane together. Strikingly, alignment analysis showed that Sjl2p recruitment to endocytic sites starts only when both Sla1p and Abp1p intensities are close to their maximum levels (), suggesting that PtdInsP is only eliminated after endocytic coat and actin assembly have occurred. This result is in agreement with the possibility that PtdInsP plays a role in recruitment of the coat proteins and cytoskeleton proteins to endocytic sites. Sjl2p begins to accumulate ∼5 s after Abp1p, but rapidly reaches its maximum level at about the same time as Abp1p (). Sla1p intensity decreases as Sjl2p levels rapidly rise, supporting a role in uncoating (). Sjl2p remains at its maximum level, whereas Abp1p intensity decreases, implicating Sjl2p in actin disassembly. Furthermore, Sjl2p remains at its maximum intensity until the Abp1p/Sjl2p fast movement starts, which occurs after vesicle scission and uncoating (). Together, the highly regular dynamics of Sjl2p at endocytic sites suggests that PtdInsP levels are strictly controlled temporally during endocytotic internalization. We next tested the impact of altered PtdInsP levels on spatial and temporal aspects of endocytosis. To do this, we studied endocytic defects of a mutant in which PtdInsP cannot be properly broken down. Δ, is not viable, the double mutant is viable (). cells are two- to threefold higher than in wild-type cells (). Thus, this mutant provides an opportunity to determine how misregulation of PtdInsP levels affects endocytosis. In wild-type cells, the endocytic coat marker Sla1p arrives and disappears at patches with a lifetime 36.7 ± 8.5 s (; ). mutants, in addition to appearing on the cell cortex, Sla1-GFP patches were also observed inside the cells (). Both the internal and the cortical Sla1-GFP patches had similar lifetimes of 43.9 ± 5.8 s (). mutants (). Furthermore, turnover of Sla1-GFP patches at both locations was blocked by addition of the actin assembly inhibitor latrunculin A (lat A; ), indicating that the Sla1p patches on the cortex and inside of the cell exhibit actin assembly–dependent turnover. cells (; ; ). cells. Strikingly, all of the internal Sla1-GFP patches appeared to be associated with the abnormal internal membrane structures labeled by FM4-64 (). cells was probably not the result of endocytosis because no internal FM4-64 signal was observed in wild-type cells at this early time point (). cells with Alexa Fluor 594–α-factor (), which is a cargo for receptor-mediated endocytosis, and imaged both the Alexa Fluor signal and Sla1-GFP immediately after internalization was initiated by shifting cells from ice to room temperature. cells, but did not label internal structures in wild-type cells. This result strongly suggests that the internal membrane structures are continuous with the plasma membrane (). cells, multiple Sla1-GFP patches appear and disappear at different areas of the deep internal structures labeled by FM4-64 ( and Video 6, available at ). cells ( and Video 7). Multiple Abp1p patches appeared and disappeared on Pdr5-GFP–labeled internal structures (Video 7). cells were previously observed and proposed to result from failure of endocytic vesicle fission (; ). However, our two-color live-cell imaging analysis indicated that endocytic protein patches continuously assemble and disassemble on these structures, much as they do on the rest of plasma membrane. cells. Mss4p patches were also observed in the deeply invaginated membrane structures, suggesting PtdInsP can be produced on these structures (). background. Δ triple mutant is inviable (unpublished data), an strain is viable. In cells, Sla2 ANTH-GFP showed patch localization with a lifetime of 76.5 ± 13.2 s, both on the cell cortex and on the abnormal membrane invagination (). strain had a shorter lifetime than in wild-type cells (105 ± 28.7). strain, failure to carry out productive endocytic events results in premature disassembly of the endocytic machinery. Nevertheless, dynamic assembly of endocytic proteins on the abnormal membrane invaginations provides evidence that endocytic proteins do not recognize membrane compartments by virtue of where the membrane compartment localizes in the cell, but by virtue of the membrane's specific molecular components, including PtdInsP. Previously, based on the dynamic behavior of many different proteins during endocytic internalization, we proposed that the yeast endocytic machinery is composed of four protein modules (). mutants. The coat module consists of proteins that assemble on the plasma membrane and then internalize ∼200 nm before disassembling. cells (, A and C; and Videos 8–10, available at ). Strikingly, kymographs of the patches revealed that all of these proteins still undergo inward movement, indicating that invagination occurs in this mutant (). These results are consistent with the fact that Sjl2p is recruited to endocytic sites only after the slow Sla1p movement starts in wild-type cells (). cells (, A and B; and Videos 8 and 9). This may be because both Sla2p and Ent1p contain PtdInsP binding domains, whereas Sla1p does not (; ). cells, Sla2p and Ent1p disassembly are significantly delayed. Previous studies suggested that disassembly of the Sla1p–Pan1p–End3p complex is regulated by phosphorylation activity of the related protein kinases Ark1p and Prk1p (; ; Sekiya- Kawasaki et al., 2003). Together, these data suggest that there may be at least two kinds of mechanisms that regulate endocytic coat disassembly. One is based on protein phosphorylation by Ark1p and Prk1p, and the other is based on PtdInsP hydrolysis by Sjl2p. The myosin–WASP module nucleates actin polymerization and, rather than getting internalized, remains at the cell cortex during endocytosis (; ; ). Myo5p, which belongs to the type I myosin family, has a lifetime of 22.8 ± 5.9 s in cells, which is about double its lifetime in wild-type cells (, and Video 8; ). One mammalian type I myosin, Myo1c, has been shown to bind tightly and specifically to PtdInsP, as well as to inositol 1,4,5- trisphosphate (). More recently, Myo1c has been shown to bind phosphoinositides through a putative pleckstrin homology (PH) domain (). This putative PH domain is also conserved in Myo5p. Thus, the delay in Myo5p disassembly may be a direct effect of the PtdInsP turnover defect. Like the myosin–WASP module, actin and several other proteins of the actin-associated protein module are absolutely essential for yeast endocytosis. We used Abp1-GFP as a marker to monitor actin behavior. In mutants, actin assembly was slightly delayed, which is consistent with the observation that the endocytic coat module undergoes slow inward movement that is likely correlated with membrane invagination (). However, Abp1p showed a significantly longer patch lifetime (P < 0.01) in cells than in wild-type cells (, , and Video 8). Previously, we proposed that actin assembly promotes endocytic coat invagination and that actin disassembly may occur upon endocytic vesicle release (). Together, our results indicate that PtdInsP levels are important for actin regulation at endocytic sites. The scission module consists of the yeast amphiphysin homologues Rvs161p and Rvs167p, which can heterodimerize. In yeast, Rvs161p and Rvs167p arrive at endocytic sites after actin polymerization starts. After briefly remaining stationary in a patch, Rvs161p and Rvs167p move rapidly inward ∼100 nm. This rapid inward movement was proposed to correspond to vesicle scission (). In cells, Rvs167p still appears after Abp1p appears, but it remains at the sites 2–3 times longer than it does in wild-type cells (, and Video 8). cells exhibit a severe defect in both receptor-mediated and fluid-phase endocytosis, suggesting that vesicle scission may not occur in this mutant. cells, supporting this idea. Thus, these results suggest that PtdInsP turnover is important for Rvs161p/167p function in endocytic scission. #text Yeast strains used in this study are listed in Table S1 (available at ). C-terminal GFP and RFP tags were integrated by homologous recombination, as previously described (). We cloned using PBS-3xGFP–His vector, which was modified from the pBS-3xGFP–Trp vector (W.-L. Lee, University of Massachusetts, Amherst, MA; ). The resulting plasmids contain a fragment encoding the C terminus of Sjl2p fused in frame to a five Ala linker and triple GFP. Wild-type haploid cells (DDY 904) were transformed with this vector linearized by an appropriate restriction enzyme in the middle of the sequence. Stable His transformants were selected and screened for proper targeting by PCR. S-labeled α-factor was prepared as described in . The α-factor uptake assay was performed at 25°C, based on a continuous incubation protocol (). Cells were grown in YPD, harvested by centrifugation, and resuspended in internalization media (YPD media with 0.5% casamino acids and 1% BSA). At the indicated time points, aliquots were withdrawn and diluted in ice-cold buffer at pH 6.0 (total α-factor) or pH 1.1 (internalized α-factor). The samples were then filtered and radioactivity was measured in a scintillation counter. The results were expressed as the ratio of pH 1.1 cpm/pH 6.0 cpm for each time point to represent the percentage of internalization. FM4-64 staining was done in a flow chamber, as previously described (). Imaging was performed immediately after the addition of the dye, which was used at a concentration of 8 μM in SD-based media (). Alexa Fluor 594–α-factor imaging was done as previously described (). One-color live-cell imaging and two-color live-cell imaging were performed as previously described (). Cells were attached to concanavalin A–coated coverslips, which were sealed to slides with vacuum grease (Dow Corning). All imaging studies were performed at ∼25°C using a microscope (IX81; Olympus) equipped with 100×/NA 1.4 objectives and cameras (Orca II; Hamamatsu). Image analysis was performed with ImageJ (National Institutes of Health; ; ). Fig. S1 shows results of experiments demonstrating that Mss4-GFP, Sjl1-GFP, and Sjl2-3GFP are functional in vivo, and results of localization of these GFP-tagged proteins. Video 1 includes real-time videos of Sla2 ANTH-GFP and Sla2 4K-A ANTH-GFP dynamics in vivo. Video 2 is a two-color real-time video of Sla2 ANTH-GFP and Sla1-mCherry (provided by C. Toret, University of California, Berkeley, Berkeley, CA) dynamics in vivo. Video 3 is a two-color real-time video of Sla2 ANTH-GFP and Abp1-mRFP dynamics in vivo. Video 4 includes real-time videos of Sjl2-GFP or Sjl2-3GFP dynamics in vivo. Video 5 is a two-color real-time video of Sjl1-3GFP and Abp1-mRFP dynamics in vivo. cells. cells. cells. cells. cells. Table S1 is a list of yeast strains used in this study. The online version of this article is available at .

xref #text The origin of peroxisomes has long been matter of debate, and partially underscoring this controversy has been the mode by which peroxisome-destined proteins are synthesized and targeted within the cell. For instance, a major tenant of the previous “ER-vesiculation” model for peroxisome biogenesis was that all of the soluble and membrane bound protein constituents of the peroxisome were synthesized cotranslationally on the ER (). These nascent proteins were proposed to then be sequestered into an expanding vesicle that would eventually bud from the ER to produce a mature, functional peroxisome (). However, subsequent observations suggested that peroxisomal proteins were not synthesized on the ER but on free polyribosomes in the cytosol. These and other data led to the “growth and division” model for peroxisome biogenesis wherein peroxisomes, like mitochondria and chloroplasts, were considered to increase in size by the posttranslational import of their protein constituents and proliferate only through the division of preexisting peroxisomes (; ; ). Notably, the ER in the “growth and division” model was deemed only to be a source of membrane lipids for the enlargement of preexisting peroxisomes. Although for most of the past two decades the “growth and division” model has generally been considered the paradigm for peroxisome biogenesis, the recent monitoring of the sorting of various PMPs in evolutionarily diverse organisms has revealed that for at least a subset of these PMPs, referred to as group I PMPs (), the initial sorting site is the ER rather than the peroxisome membrane. Sorting of group I PMPs to and within the ER also appears to be mediated by several different mechanisms (). For instance, in mammalian cells, the group I PMP Pex16p is inserted cotranslationally into ER membranes and seems to be localized throughout the entire ER before its sorting to peroxisomes (). In the yeast , , and , group I PMPs Pex2p, -3p, and -16p are also initially targeted to the “general” ER (; ; ). However, unlike mammalian Pex16p, the ER targeting and insertion of these essential components of peroxisome assembly in does not require the Sec61p-dependent machinery for co- and posttranslational import of secretory proteins (). Furthermore, unlike mammalian Pex16p that remains in the general ER before its sorting to peroxisomes (), at least one of the group I PMPs in , namely, Pex3p, is directed from the general ER to a distinct subdomain of the ER (). This ER subdomain is referred to as the preperoxisomal template () and is considered to be the site where preperoxisomal carriers are formed. That is, after being segregated into the preperoxisomal template, Pex3p serves as a docking factor for Pex19p, a predominantly cytosolic protein (). The Pex3p-dependent recruitment of Pex19p from the cytosol to the outer face of the preperoxisomal template in is mandatory for the budding of small preperoxisomal vesicles (). These ER-derived carriers of Pex2p, -3p, -16p, and -19p lack secretory cargo proteins (). Although the mechanism responsible for segregating group I PMPs from secretory and ER resident membrane proteins in yeast remains to be established, it is noteworthy that the membrane of the ER-derived preperoxisomal vesicles in has unusual ergosterol- and ceramide-rich lipid domains (). These lipid domains are similar to detergent-resistant lipid domains in the membrane of ER, where glycosylphosphatidylinositol-anchored secretory proteins cluster and thereby segregate from other secretory proteins (). It is possible, therefore, that discrete lipid domains, perhaps ergosterol- and ceramide-rich lipid domains, in the membrane of yeast ER serve also as a sorting platform for segregating group I PMPs from secretory and ER resident membrane proteins. The resulting partitioning of group I PMPs into these membrane domains could also serve to generate an ER template for the formation of preperoxisomal vesicles. In contrast to yeast Pex2p, -3p, and -16p and mammalian Pex16p, other group I PMPs, such as ascorbate peroxidase (APX) in plant cells () and Pex13p in mouse dendritic cells (), can only be detected in a distinct portion of the ER, suggesting that they are targeted from the cytosol directly to a preexisting subdomain of the ER membrane. The terms peroxisomal ER (pER) and lamellar ER extension were coined for this ER found in plants and mice, respectively (; ). At least one notable difference between these two ER subdomains is that pER is considered to be a portion of rough ER membrane (), whereas the lamellar ER extension is a specialized domain in smooth ER membrane (). In plant cells, the cytosol-to-pER targeting of APX occurs posttranslationally and requires ATP as well as at least three components of the Hsp70 chaperone machinery (). Collectively, the aforementioned findings suggest that by segregating a distinct set of membrane proteins and lipids into specialized ER subdomains, plant and mouse dendritic cells have evolved a platform for the targeting of group I PMPs from the cytosol to the ER membrane. The existence of such a platform in the ER membrane could increase the efficiency of the ER-dependent, multistep process of peroxisome assembly in these cells. What structural features of group I PMPs are crucial for their sorting to the ER or to the peroxisomal membrane via either general ER or an ER subdomain remain to be determined. At present, it seems that the targeting of these PMPs from the cytosol to the ER membrane and their subsequent exit from the ER are mediated by two partially overlapping sets of sorting signals. One set of signals targets group I PMPs either co- or posttranslationally to the general ER or an ER subdomain, whereas the other set of signals act from within the ER lumen to sort these PMPs to the peroxisome (; ; ; ). Although all group I PMPs exit the ER via distinct preperoxisomal carriers that do not enter the classical secretory pathway of vesicular flow (; ), the morphology of these carriers in at least yeast and mammalian cells appears to differ (). In , , and , the ER-derived preperoxisomal carriers are small vesicles (; ; ). In contrast, the preperoxisomal carriers in mouse dendritic cells arise through direct en block protrusion of the specialized ER subdomain, the lamellar ER extension (). After reaching a considerable size, the lamellar extension detaches from the ER, giving rise to pleomorphic tubular-saccular carriers of group I PMPs. This detachment of preperoxisomal tubular-saccular carriers from the ER does not require coat protein complexes (COPs) I and II, which function in the formation of ER-derived carriers for secretory proteins (). It is noteworthy that, akin to ER-derived preperoxisomal carriers in yeast cells, all known types of transport carriers for secretory proteins are small vesicles in these cells (). On the contrary, just like the preperoxisomal carriers in mammalian cells, at least a subset of ER-to-Golgi carriers for many secretory proteins in these cells are pleomorphic tubular-saccular structures that are formed through direct en block protrusion of specialized domains in the ER membrane (). This fundamental difference in the morphology of ER-derived transport carriers is likely due to the difference in the spatial organization of transitional ER (tER), a specialized ER subdomain at which proteins are packaged into membrane-enclosed carriers. In the traditionally used model yeast organism, , the entire ER acts as tER, facilitating the budding of COPII-coated vesicles (). In contrast, the tER of mammalian cells is organized into discrete ER export sites (). It is therefore possible that by segregating a distinct set of membrane proteins and lipids into a specialized ER subdomain for the cytosol-to-ER targeting of group I PMPs (see the previous section), higher eukaryotic organisms have not only separated these domains from the sites for the ER targeting of secretory proteins but also developed a platform for the sculpturing of these pER subdomains into pleomorphic tubular-saccular carriers of PMPs. A critical evaluation of this hypothesis would require testing the spatial organization of the ER subdomains for the cytosol-to-ER targeting of group I PMPs and examining the morphology of ER-derived carriers for these PMPs in the yeast . Unlike and similar to mammals, has discrete tER export sites that give rise to a “conventional” mammalian-type secretory apparatus (). Presently, no solid data exist for the nature of the preperoxisomal carriers in plant cells, although, similar to mammals, the tER in these cells is restricted to discrete sites in the ER membrane (), suggesting that the organization of the pER subdomain as well as the formation of preperoxisomal carriers in plants is similar to that in mammals. Recent findings have provided strong evidence that, analogous to some organelles of the secretory endomembrane system, peroxisomes constitute a dynamic organelle population consisting of many structurally distinct compartments that differ in their import competency for various proteins. Moreover, it appears that the individual compartments of this peroxisomal endomembrane system undergo a multistep conversion to mature peroxisomes in a time-ordered manner. Two multistep pathways for peroxisome assembly and maturation have been described (). In , the posttranslational sorting of two partially overlapping sets of PMPs and a few matrix proteins converts two populations of ER-derived preperoxisomal vesicular carriers into the small (75–100 nm) peroxisomal vesicles P1 and P2 (). These vesicles then serve as the earliest intermediates in a multistep pathway that involves, at each step, the uptake of lipids and the selective import of matrix proteins, eventually resulting in the formation of a mature peroxisome referred to as P6 (). Overall, it seems that in and perhaps in other yeast, import machineries specific for different peroxisomal matrix proteins undergo a temporally ordered assembly in distinct vesicular intermediates along the peroxisome maturation pathway (). The plasticity of these import machineries is further underscored by the observation that the efficiency with which they recognize nonoverlapping targeting signals present on some of their protein substrates varies under different metabolic conditions. In fact, peroxisomal subforms present in yeast cells growing under conditions that induce peroxisome proliferation differ from basal, nonproliferated subforms with respect to the targeting sequence motifs that are used to direct the same protein to these different subforms of peroxisomes (). A quite different scenario orchestrates a multistep process of peroxisome assembly and maturation in mouse dendritic cells. Herein, the extrusion of the lamellar ER extensions is culminated by the detachment of pleomorphic tubular-saccular carriers of Pex13p from the ER (). Only after their separation from the ER are these preperoxisomal carriers able to recruit to their membranes the ATP binding cassette transporter protein PMP70 and, perhaps, the membrane components of the import machinery for peroxisomal matrix proteins (). This latter step of the peroxisome maturation pathway also results in the formation of the so-called peroxisomal reticulum. Only the peroxisomal reticulum is capable of importing at least two peroxisomal matrix proteins, namely, thiolase and catalase, directly from the cytosol (). Notably, these two peroxisomal matrix proteins do not fill the entire peroxisomal reticulum. Instead, they are sorted exclusively into mature globular peroxisomes that, during the final step in the peroxisome maturation pathway in mouse cells, bud from the peroxisomal reticulum (; ). It remains to be established whether other peroxisomal matrix proteins, similar to thiolase and catalase, are imported into the domain of the peroxisomal reticulum that gives rise to mature globular peroxisomes or whether, alternatively, these other matrix proteins in mouse cells are sorted to globular (mature) peroxisomes only after their budding from the peroxisomal reticulum. In both models for the multistep assembly and maturation of peroxisomes, the targeting of PMPs to the membrane of the early intermediates in a pathway precedes, and is mandatory for, the import of soluble peroxisomal proteins into the matrix of later intermediates. Because this strategy for peroxisome biogenesis has been conserved in the course of evolution, it seemingly provides an important advantage for the efficient, stepwise assembly of mature, metabolically active peroxisomes. It remains to be established whether, similar to a stepwise assembly of import machineries specific for different peroxisomal matrix proteins in yeast cells (), the import machineries for such proteins in mammalian cells can undergo a temporally ordered assembly in distinct intermediates along the peroxisome maturation pathway. It is also unclear at the moment whether the peroxisome maturation pathway acting in mammalian cells, akin to the pathway that functions in yeast cells (; ), includes fusion of any early pathway intermediates. It is tempting to speculate that such fusion of early pathway intermediates in yeast results in the formation of an ER–peroxisome intermediate compartment (ERPIC). Such a compartment could provide a template for the formation of downstream intermediates in the peroxisome assembly and maturation pathway and function in the sorting of PMPs from those escaped ER resident proteins that are retrieved by retrograde vesicular transport between the ERPIC and the ER. Both of these tentative functions of the ERPIC share similarity with the functions that have been proposed for the ER–Golgi intermediate compartment, also known as vesicular tubular clusters, which may regulate a bidirectional traffic of membrane-enclosed carriers through the classical secretory pathway (). Importantly, the resident proteins of the post-ER compartments in both the peroxisomal endomembrane system and the classical secretory system return to the ER in response to the treatment of yeast cells with brefeldin A, an inhibitor of COPI formation (). Thus, similar to its role in the secretory endomembrane system, yeast COPI can function in the retrieval of those ER resident proteins that had entered the peroxisomal endomembrane system by mistake. This is in contrast to COPI in cultured human fibroblasts, in which peroxisome-to-ER retrograde protein transport, if any, does not depend on COPI (; ). These findings further support the notion that yeast and higher eukaryotic organisms may use different strategies for the ER-dependent formation and maintenance of their peroxisomal endomembrane systems. #text In addition to their proposed role in the peroxisome-to-ER retrograde protein transport in virus-infected plant cells, both ARF1 and COPI can induce the proliferation of the peroxisomal endomembrane system in other evolutionarily diverse organisms by promoting the membrane scission event required for peroxisome division (). In fact, yeast mutants impaired in ARF1 and COPI, as well as mammalian cells deficient in COPI assembly, accumulate a reduced number of elongated tubular peroxisomes, consistent with impairment in peroxisome vesiculation (; ). Incubation of highly purified rat liver peroxisomes with cytosol results in specific binding of both ARF1 and COPI to the peroxisomal membrane, further supporting the notion that their recruitment from the cytosol in living cells is an initial event in the proliferation of the peroxisomal endomembrane system (). Moreover, similar to ARF1, the subtype 3 of yeast ARF also controls peroxisome division in vivo, although, in contrast to ARF1, in a negative fashion (). Collectively, these findings suggest that the peroxisomal endomembrane system and the classical secretory system of vesicular flow are served by a similar set of core protein components required for their communication with the ER and for their proliferation. The proliferation of the individual compartments of the peroxisomal endomembrane system is also driven by a peroxisome-specific protein machinery, which includes a distinct set of the PMPs and the dynamin-related proteins DLP1 (dynamin-like protein 1), DRP3A (dynamin-related protein 3A), and Vps1p (vacuolar protein sorting protein 1), recruited from the cytosol to the peroxisomal surface by their receptor Fis1p (; ). A challenge for the future will be to define how the interplay of all these protein components governs such proliferation under the different metabolic conditions in a given cell type or tissue. Importantly, peroxisome biogenesis appears to occur by way of a collaborative effort between two equally important pathways. The first pathway operates through the ER-dependent formation and maturation of the individual compartments of the peroxisomal endomembrane system, whereas the second pathway involves the precisely controlled division of these peroxisomal compartments. Growing evidence supports the view that cells have evolved at least two strategies for the coordination of compartment assembly and division in the peroxisomal endomembrane system. In the first strategy, the multistep growth and maturation of the ER-derived preperoxisomal carriers occurs before the completely assembled, mature peroxisomes undergo division (). In the second strategy, a significant increase in the number of preperoxisomal carriers, either by their en masse formation from the ER () or by the proliferation of a few preexisting carriers (; ), precedes the growth of these early peroxisomal precursors by membrane and matrix protein import and their conversion to mature, functional organelles containing a complete complement of peroxisomal proteins. Determining the relative contribution of these different mechanisms in the formation of peroxisomes in any given organism should now be more feasible through the use of live-cell, photo/pulse-chase labeling methods similar to that reported recently for a study of peroxisome biogenesis in mammalian cells (). Regardless of the strategies that evolutionarily distant organisms use for coordinating the assembly and division of individual compartments of the peroxisomal endomembrane system, the tubulation, constriction, and scission of these compartments is regulated, depending on the cellular and/or environmental conditions of a particular cell type, either by signals emanating from within these compartments () or by extraperoxisomal signals that are generated inside the cell in response to certain extracellular stimuli (). These intracellular signals include a distinct group of transcriptional factors that induce the transcription of genes encoding several proteins of the Pex11p family (). The peroxisome membrane bound Pex11p-type proteins then directly promote the proliferation of peroxisomal endomembrane compartments or activate peroxisome division indirectly by recruiting the dynamin-related proteins from the cytosol (). Furthermore, the division of the individual compartments of the peroxisomal endomembrane system must be preceded by the expansion of their membranes because of the acquisition of lipids. The ER, a principal site for the biosynthesis of phospholipids, is the most likely source of lipids for the growth of the peroxisomal membrane (), although oil bodies have been implicated also as a source of peroxisomal lipids in some organisms, e.g., germinated oilseeds () and (). It seems that in the bulk of phospholipids is transferred from the donor membrane of a specialized subcompartment of the ER to the closely apposed acceptor membranes of the two early intermediates, P3 and P4, in the peroxisome assembly pathway (). Although the mechanism responsible for such ER-to-peroxisomal membrane transfer of phospholipids via membrane contact sites remains to be established, several working models for the role of ER-associated lipid-transfer proteins in the establishment and functioning of such sites have recently been proposed (). These models should serve as a useful starting point for examining such events during peroxisome biogenesis. o w i n g e v i d e n c e s u p p o r t s t h e v i e w t h a t p e r o x i s o m e s c o n s t i t u t e a d y n a m i c e n d o m e m b r a n e s y s t e m t h a t o r i g i n a t e s f r o m t h e E R . A m a j o r c h a l l e n g e n o w i s t o i d e n t i f y t h e m o l e c u l a r p l a y e r s t h a t c o o r d i n a t e t h e f l o w o f m e m b r a n e - e n c l o s e d c a r r i e r s t h r o u g h t h e p e r o x i s o m a l e n d o m e m b r a n e s y s t e m . F u t u r e w o r k w i l l a i m a t u n d e r s t a n d i n g t h e s p a t i o t e m p o r a l d y n a m i c s a n d m o l e c u l a r m e c h a n i s m s u n d e r l y i n g t h i s m u l t i s t e p p r o c e s s i n e v o l u t i o n a r i l y d i v e r s e o r g a n i s m s . I t i s c o n c e i v a b l e t h a t t h e a n a l y s i s o f a v a r i e t y o f m o d e l o r g a n i s m s , i n c l u d i n g t i s s u e - c u l t u r e d h u m a n c e l l l i n e s a n d v a r i o u s y e a s t a n d p l a n t s p e c i e s , w i l l r e v e a l a s - y e t - u n k n o w n s t r a t e g i e s a n d m e c h a n i s m s g o v e r n i n g t h e b i o g e n e s i s o f t h e p e r o x i s o m a l e n d o m e m b r a n e s y s t e m a n d i t s r e l a t i o n s h i p w i t h t h e E R .

The ER is an important quality control site within the cell, where proteins destined for secretion or for sorting to post-Golgi organelles are monitored for proper folding and oligomeric assembly. Proteins that fail to fold or assemble are typically retained in the ER and, in some cases, retrotranslocated to the cytoplasm for proteosomal degradation (). Polytopic membrane proteins receive particular scrutiny in this regard. Indeed, many diseases are attributable to the failed ER export of mutant transmembrane proteins (). Yeast genetic analyses have identified several ER resident proteins that mediate the ER export of specific polytopic membrane proteins. For instance, Shr3 is required for the ER export of the yeast amino acid permeases. In mutants, these permeases are retained in the ER, whereas the transit of other polytopic proteins is unimpaired (). Similarly, Gsf2, Pho86, and Chs7, which are unrelated to Shr3 at the sequence level, are specifically required for the ER export of the hexose transporter Hxt1, the phosphate transporter Pho84, and the chitin synthase Chs3, respectively (). These export factors have been suggested either to direct the segregation of their target proteins into budding COPII vesicles for anterograde transport or, alternatively, to act as dedicated chaperones, regulating proper protein folding before transport. The yeast chitin synthase Chs3, which is a polytopic protein with six to eight predicted transmembrane domains, provides a genetic model for understanding mechanisms of transport through the secretory pathway. Chs3-mediated chitin deposition at the plasma membrane is highly regulated at the level of intracellular trafficking. Chs3 is maintained at steady state in an intracellular pool that may correspond to the TGN or endosomes (), and it is transported to the plasma membrane upon activation of the cell-integrity signaling pathway (). Mutants that impair cell wall chitin deposition have been found to block the plasma membrane delivery of Chs3 at different intracellular transport steps, whereas Chs7 mediates the ER export of Chs3, Chs5 and Chs6 direct its transport from the TGN to the plasma membrane, and Chs4 is required both for Chs3 activity at the cell surface and for its localization at the bud neck (for review see ). We describe a genomic analysis of factors that regulate the transport of Chs3 to the cell surface, and identify an unexpected role for protein palmitoylation in the ER export of Chs3. Palmitoylation, which is the thioester linkage of palmitate to selected cysteine residues, is one of several lipid modifications used for tethering proteins to membranes (). For transmembrane proteins, which are already embedded in the bilayer, the functional consequences of palmitoylation are not clear, though a role in directing segregation to membrane microdomains (lipid rafts) is often invoked. Enzymes for protein palmitoylation, which are called protein acyl transferases (PATs), were identified only recently by work in yeast (; ). The first two PATs to be characterized, Akr1 and Erf2, were found to contain a conserved zinc finger–like Asp-His-His-Cys (DHHC) domain, suggesting that this motif defines a larger PAT family. Yeast has seven DHHC proteins, whereas 23 are identifiable from the human genome. More recent reports have linked additional DHHC proteins to the palmitoylation of various substrates in both yeast and mammalian cells (for review see ). In this study, we find the uncharacterized yeast DHHC protein Pfa4 to be required for ER export by acting as the dedicated PAT for Chs3 palmitoylation. To identify additional genes required for Chs3 trafficking, we developed a fluorescence assay suitable for the large-scale screening of yeast deletion arrays, based on the binding of the fluorescent antifungal drug Calcofluor white (CW) to cell wall chitin (; ). Mutants that are defective for the transport of Chs3 to the cell surface produce little chitin, and thus exhibit low levels of fluorescence. We screened three independent gene-deletion collections in duplicate and calculated the median fluorescence intensity for each strain. As expected, the mutants with the lowest fluorescence values included cells, which lack chitin synthase III, and strains deleted for the known Chs3 transport factors (). In addition, the screen identified and , which are components of the cell-integrity pathway that stimulates the cell surface transport of Chs3, indicating that colony fluorescence values correlate well with independent measures of cell wall chitin (). Unexpectedly, cells deleted for the uncharacterized ORF YOL003c () consistently displayed fluorescence values comparable to mutants (). Pfa4 is predicted to be a 45-kD protein containing the signature DHHC cysteine-rich domain that has been linked to protein palmitoylation (). Like mutants, cells not only bind less CW but are also strikingly resistant to CW toxicity (). To determine if the low levels of cell wall chitin in mutants result from alterations in the intracellular transport of Chs3, we examined the subcellular localization of Chs3-GFP in strains and in mutants with known Chs3-trafficking defects. In wild-type cells, Chs3-GFP is present at the bud neck, bud tip, and intracellular compartments, whereas it is completely restricted to this latter compartment in mutants, which is consistent with a mislocalization to the TGN or endosomes (; ). In contrast, Chs3-GFP localized to intracellular rings in mutants, which are similar to those seen in mutants, where ER exit of Chs3 is blocked (). Colocalization with the ER marker Sec61 confirmed that Chs3-GFP resided primarily in the ER of cells (); however, unlike cells, a small proportion of cells also showed some Chs3-GFP at the bud neck or bud tip. The ER-localized pool of Chs3 is not unstable or targeted for degradation, as Chs3 is present at wild-type levels in mutants (). These results indicate that loss of cell surface Chs3 activity in mutants results from a defect in transport at the ER. Moreover, this transport defect is specific to Chs3, as localization of other yeast chitin synthases, Chs1 and Chs2, are unaltered in cells (). Because Pfa4 is a predicted PAT, we considered the possibility that ER export of Chs3 requires a Pfa4-mediated palmitoylation event. For the few DHHC PATs examined to date, mutation of the cysteine within the core DHHC tetrapeptide element has been found to abolish PAT activity both in vivo and in vitro (; ; ; ; ). Therefore, cysteine108 within the Pfa4 DHHC sequence was mutated to alanine. The Pfa4 mutant protein accumulated to wild-type levels (), indicating that the substitution does not destabilize Pfa4. Nevertheless, plasmid-borne failed to complement CW resistance and Chs3-GFP mislocalization in mutants (), suggesting these phenotypes are caused by a lack of Pfa4 enzymatic activity. Deletion of other DHHC proteins did not alter CW fluorescence (). To determine if Chs3 is the direct target of Pfa4, we tested Chs3 for palmitoylation using a modified acyl-biotin exchange assay (; ). This three-step protocol involves the following: blockade of free thiols with N-ethylmaleimide, hydroxylamine cleavage of palmitoylation thioester linkages, and thiol-specific biotinylation of the newly exposed cysteinyl thiols. To control for acyl-biotin exchange specificity, samples omitting the key hydroxylamine cleavage step were processed in parallel. We found that Chs3 is palmitoylated, and this modification is Pfa4-dependent, being abolished in and mutants (). In contrast, palmitoylation of the other chitin synthases, Chs1 and Chs2, or the ER export factor Chs7, could not be detected (unpublished data). Other PATs have been shown to copurify with their substrates (). Pfa4–Chs3 complexes could be detected by coimmunoprecipitation (), suggesting that Pfa4 interacts directly with Chs3 to mediate its palmitoylation. The observation that Chs3 is ER-localized and palmitoylated in cells () is consistent with Chs3 palmitoylation being an early, ER-localized event. To date, four of the seven yeast DHHC proteins have been shown to mediate protein palmitoylation in various cellular locations (; ; ; ; ). Our finding of Pfa4-dependent palmitoylation for Chs3 adds a fifth DHHC protein to this list. Two other yeast PATs, Erf2/4 and Swf1, are also known to function at the ER, but recognize distinct substrates. The emerging picture is, thus, of a family of PATs that are distinguished from one another both by intracellular localization and by substrate specificity. As both Chs7 and Pfa4 are required for Chs3 ER export, we considered the possibility that these two proteins act together. Heterooligomeric PATs have been identified that require binding partners for activity and stability (). Although Chs3 palmitoylation was reduced in cells, it clearly was not abolished (). Furthermore, Chs3 copurifies with its PAT even in the absence of Chs7 (), whereas Pfa4–Chs7 interactions could not be detected under similar conditions (not depicted). Therefore, Chs7 does not appear to be required for substrate recognition by Pfa4 and is unlikely to be a subunit of a dimeric PAT. We tested an alternative model, in which Chs7 preferentially interacts with lipid-modified Chs3 to promote its interaction with the COPII vesicle-budding machinery (). Although a Chs3–Chs7 physical interaction has not been yet reported, other polytopic yeast proteins, including Gap1 and the vacuolar H-ATPase, have been shown to interact with dedicated accessory factors at the ER during biosynthesis (; ). Using coimmunoprecipitation, we demonstrated that Chs3 and Chs7 do interact (, lane 3). In cells, the Chs3–Chs7 interaction was subtly reduced, but not eliminated (, lane 2), indicating that Chs3 palmitoylation is not required for recognition by Chs7. Recent work has suggested a chaperone function for Chs7 and the ER accessory proteins Shr3, Gsf2, and Pho86 (). When their cognate chaperones were absent, the substrate polytopic proteins were found to accumulate in the ER in high molecular mass aggregates, which were visualized by cross-linking with dithiobissuccinimidyl propionate (DSP). Using the DSP cross-linking protocol of , we examined Chs3 aggregation in cells (). mutants (), indicating that Chs7 chaperone function and Pfa4-mediated palmitoylation are both required to circumvent Chs3 aggregation. Despite the similarity of the and phenotypes, our data do not support models that Chs7 and Pfa4 act together as part of a complex or linear pathway. Instead, our results are best accommodated by models where Chs7 and Pfa4 act in parallel, mediating separate events that are both required for ER export. Nonetheless, defects in one pathway do appear to affect the other; Chs3 palmitoylation is reproducibly reduced in cells, and the Chs3–Chs7 interaction is decreased in cells. As both Chs7 and Pfa4 are required to circumvent Chs3 accumulation in high molecular mass aggregates, both appear to participate in the prerequisite folding of Chs3 that precedes ER export. Hydrophobic mismatch, resulting from an incompatibility between long transmembrane domains of polytopic proteins and the thinner ER bilayer, could explain why a membrane protein such as Chs3 requires both palmitoylation and chaperone association for export. ER chaperones have been hypothesized to shield hydrophobic regions of transmembrane domains to prevent protein aggregation (). Palmitoylation may also promote hydrophobic matching by targeting proteins to cholesterol-rich membrane microdomains, which provide a local region of higher bilayer thickness. It is interesting that Chs1 and Chs2, which are also polytopic proteins, require neither Chs7 nor Pfa4 for ER export. This suggests a requirement for chaperone association and palmitoylation for only a subset of membrane proteins. Several recent results suggest that the palmitoylation requirement for ER export may not be unique to Chs3. Our concurrent proteomic analysis of yeast protein palmitoylation indicates that several amino acid permeases are palmitoylated in a Pfa4-dependent manner (). Intriguingly, these polytopic proteins also require dedicated accessory proteins for their ER export (). It will be interesting to see if, as for Chs3, palmitoylation plays a role in permease trafficking. The link between palmitoylation and ER exit may hold true for at least some polytopic proteins in higher cells, as it was recently reported that functional cell surface expression of the nicotinic acetylcholine receptor requires an ER palmitoylation event (). Thus, palmitoylation may participate more generally in ER quality control mechanisms, particularly for newly synthesized polytopic integral membrane proteins. ext-link #text

The information encoded in the primary sequence of eukaryotic genomes is three-dimensionally organized within the cell nucleus. Although the underlying principles of this 3D genome organization are just beginning to be explored, emerging evidence indicates that it plays an important role in genomic functions. For example, gene expression is related to the 3D position of a locus within the overall nuclear volume, to associations of individual genes with specific nuclear compartments, and to spatial interactions between pairs of genes (; ; ; ). Given these functional interactions of individual genes, a key unanswered question is how a chromosome manages and coordinates the structural demands of multiple genes and their associated activities. Indeed, little is known about “higher order” DNA folding in the nucleus (), much less how this folding is related to the diverse information encoded in the underlying primary sequence. In mammals, the most widely known feature of genome 3D organization is the differential enrichment of euchromatin and heterochromatin in the nuclear interior and periphery, respectively. These patterns tend to be recapitulated by large-scale (∼5 Mb) gene-rich and -poor chromosome regions, which correspond to different cytogenetic chromosome bands (; ). The partitioning of chromosome regions according to gene density provides an overall framework for genome organization in the nucleus. However, these regions represent crude divisions of sequence. As such, they provide limited insights into the nuclear organization of specific sets of genes and how the chromatin polymer folds to accommodate different sequences. Completely sequenced mammalian genomes now allow for more precise and comprehensive studies of 3D genome organization within these large chromosomal regions. However, only a few pairs of well-annotated, closely linked genes have been localized relative to each other in nuclei (; ). Likewise, the handful of studies of chromosome regions typically have involved homogenous labeling across the region or probing a single locus in the region relative to the whole chromosome “territory” (; ; ; ; ). These studies have clearly demonstrated that chromosome architecture is related to gene density and gene activity and that this architecture is dynamic. However, the spatial relationships among the many different sequences within a large chromosome region remain poorly understood. We address how a series of multiple genes within a megabase-scale chromosomal region are organized relative to each other in the nucleus. We focused on a well-annotated gene-poor region on mouse () chromosome 14 (Mmu14). This 4.3-Mb region is enriched with genes that affect the development of multiple embryonic tissues, including the heart, skeleton, and various structures in the nervous system (). In the Mmu14 primary sequence, these genes are organized into small clusters separated by >400-kb stretches of gene-poor sequence called “gene deserts” (; ). We probed the nuclear structure of this entire region and found evidence for a dynamic, probabilistic framework for the 3D organization of multiple genes within a chromosome region. We selected a well-annotated, 4.3-Mb gene-poor region of distal Mmu14 to study 3D genome organization and chromatin folding across a contiguous chromosomal region (). Similar to other gene-poor regions, the selected Mmu14 region is later replicating and enriched with developmental genes (; ; ; ). The region's 19 genes are grouped into four distinct clusters separated by four gene deserts, all of which are 0.2–1.0 Mb long (; ). The content, order, and relative spacing of the genes in this region are conserved from primates to chickens, suggesting functional and structural constraints on gene organization in the primary sequence (). We first examined the 3D structure of this Mmu14 region in NIH-3T3 fibroblasts. Quantitative RT-PCR showed that loci in each of the Mmu14 gene clusters are expressed in this cell type, indicating that although this region is gene-poor, it is not silenced chromatin (). NIH-3T3 cells were probed by FISH according to the pattern of gene clusters and deserts in the primary sequence (). Because of their genomic size (0.2–1.0 Mb), the gene clusters and deserts are readily resolvable in interphase nuclei (). To directly compare cluster and desert positions, 23 bacterial artificial chromosomes (BACs) spanning the region were used as FISH probes (Fig. S1, available at ), differentiating all gene clusters from all deserts with a two-color labeling scheme (, red and green, respectively). Our initial survey of the Mmu14 clusters and deserts in NIH-3T3 nuclei revealed a striking partitioning of genic and nongenic sequences (). Moreover, we observed three basic patterns of 3D cluster–desert arrangement, plus combinations of those patterns (). One cluster–desert pattern was marked by alternating signals along a “striped” DNA fiber. This 3D arrangement reflects the linear organization of gene clusters and deserts in the primary sequence, albeit in a compacted state. End–end measurements of these structures in deconvolved epifluorescence images indicated a packing ratio of 300:1 relative to naked DNA (4.1 ± 0.5 μm/4.3 Mb), which is greater than the ∼40:1 packing ratio for 30-nm chromatin fibers (see Materials and methods). In the second conformation, all of the gene clusters were displaced to one side of all the deserts, compressing the striped fiber by a “zigzag” arrangement (∼800:1 packing ratio; 1.5 ± 0.3 μm end–end). The third structure was marked by close grouping of all gene clusters into a “hub” with peripherally arranged deserts (∼900:1 packing ratio; 1.4 ± 0.2 μm end–end). The latter two conformations indicate additional 3D levels of gene cluster organization, beyond their arrangement in the primary sequence. These conformations further reveal local domains of genic DNA within the nuclear chromosome territory, which are formed by the spatial aggregation of multiple gene clusters. To more rigorously classify these structures, we generated 3D deconvolved images and analyzed 132 chromosomes at multiple viewing angles. G phase cells were selected for this and all subsequent analyses to rule out the effects of cell cycle on architecture. To accomplish this, we included an additional probe for the earlier-replicating locus on Mmu4 (), and selected cells with a prereplication, singlet signal, which is indicative of G cells rather than those with a postreplication doublet signal. Scoring of cluster–desert patterns in these G cells indicated that the aforementioned three morphological classes were indeed predominant, representing 67% of chromosomes (). Unlike gene cluster hubs, desert hubs were rare (3%), indicating sequence-specific chromosome architecture. The remainder of chromosomes (20%) exhibited combinations of the three predominant conformations (e.g., half striped and half zigzag; , combo), suggesting transitional structures. The cluster–desert conformations were rarely the same for homologous chromosomes within a given cell (4% of cells). In addition, all combinations of different cluster–desert patterns within a cell were present at similar frequencies, which is consistent with independent folding of homologues in the same nucleus. Together, these findings indicate that Mmu14 regions form multiple, defined, and likely dynamic structures in nuclei. We confirmed the Mmu14 gene cluster–desert arrangements in additional cell populations and under different fixation conditions (Fig. S2, available at ). More than 500 NIH-3T3 cells were evaluated by higher throughput 2D analyses, which showed similar relative frequencies of the three predominant conformations. These were not affected by fixation protocol. Consistent with this, several studies have shown that FISH does not significantly affect chromatin organization at the size scale of these Mmu14 gene clusters and deserts (; ; ). In addition, the three predominant conformations were present at similar frequencies in both NIH-3T3 and primary mouse embryo fibroblasts (P ≥ 0.5, χ test; Fig. S2). Thus, they do not result from aneuploidy of immortalized NIH-3T3 fibroblasts. Collectively, the data strongly indicate specific 3D organizational states for the Mmu14 region gene clusters and deserts in fibroblast nuclei. We tested whether Mmu14 stripes, zigzags, and hubs truly reflect 3D organization based on the gene distribution pattern in the primary sequence or whether they are sequence-independent and simply a consequence of the alternating labeling scheme. The pattern of probe labels was shifted ∼250 kb down the chromosome so that each label no longer matched exclusively with gene clusters or deserts. This resulted in the increased overlap of the two labels in nuclei () and significant differences in the distributions of nuclear label patterns, as assessed by 2D scoring (P < 1 × 10, χ test; ). These differences largely resulted from fewer of the more highly folded conformations, zigzags and hubs of “red” label, which corresponded to gene clusters in the cluster–desert–matched labeling scheme. We found little change in the frequencies of striped fibers and hubs of “green” label (P = 0.8, χ test; ). However, the shifted labels would not be expected to affect these probe patterns if striped fibers represent a linear arrangement of chromosome sequence and if gene deserts form hubs at random frequencies. Thus, these findings provide strong evidence for nonrandom chromatin folding that is specifically matched to the gene cluster pattern in the primary sequence. If gene clusters and deserts establish Mmu14 region structure, then a chromosomal region with different primary sequence organization should also appear different when probed by a Mmu14 labeling scheme. To confirm this, we applied the Mmu14 labeling scheme to a homogeneously gene-dense region on Mmu15 that lacks gene deserts (). Hybridization to NIH-3T3 cells revealed an increase in striped fibers (P < 1 × 10, χ test; ). These were more decondensed (≤300-nm diam by FISH) than Mmu14 fibers (∼400-nm diam), resembling chromatin structures seen at other gene-rich sequences (; ; ). Importantly, the Mmu15 probes revealed fewer zigzag and red hub structures compared with Mmu14 cluster–desert probes (P < 1 × 10). Furthermore, the differentially red- or green-labeled Mmu15 regions, which mark sequences with similar gene densities, were found in hubs at similar frequencies (P = 0.2). This contrast with Mmu14 morphological patterns is consistent with higher order structures that are based on the pattern of gene distribution in the Mmu14 primary sequence. The dependence of Mmu14 region 3D conformation on genomic clusters of expressed genes suggests nonrandom organization in nuclei. We next investigated the arrangement of specific Mmu14 gene clusters and deserts relative to each other and compared their organization with theoretical models of chromosome organization that do not include functional information. We localized pairs of gene clusters or deserts that were separated by similar genomic distances in NIH-3T3 cells (). Two-color FISH of the proximal gene clusters, C1 and C2, produced signals that frequently overlap or abut each other (72%; ). In contrast, the flanking deserts (D1 and D2) contact each other in only 39% of chromosomes (). This differential organization was confirmed by 3D measurements of center–center distances (P < 1 × 10, Kolmogorov-Smirnov (KS) test; ). We also measured the distance between C1 and its flanking desert, D1, and found it similar to the C1–C2 distance, though C1 and C2 are further apart in the primary sequence (). Because a randomly folded 5-Mb region of chromatin has yet to be identified empirically, we compared our in situ data to a computational model of a randomly folded chromosomal region. This model includes all chromosomes in a mouse nucleus, with each chromosome represented as a polymer of connected 1-Mb spherical “domains,” which are similar in size to the gene clusters and deserts being studied (). The modeled domains are consistent with empirically observed ∼1-Mb chromatin foci or rosettes (; ; ), and they are depicted as elastic spheres to allow partial overlap (). The domains are connected by short DNA linkers defined by an entropic spring potential. Each chromosome territory is further bounded by a weak barrier potential to maintain a volume similar to an empirically determined average (; ). With no further constraints on domain positions, ∼400,000 Monte Carlo steps were calculated to independently move all chromosomal domains. Simulations of 50 different nuclei showed similar separations between C1 and C2 and between D1 and D2 in the virtual Mmu14 (P > 0.1, KS test; ). This contrasts markedly with our empirical measurements (). As expected, the model generated closer positions for C1 and D1 (P < 0.01, KS test), which are also closer in the primary sequence. These modeling and empirical data strongly support nonrandom, sequence-specific folding of the Mmu14 region in nuclei. To additionally verify nonrandom chromatin folding, we compared the Mmu14 region to a random-walk polymer model. For a random-walk polymer, the mean-squared distance between two points in 2D or 3D (e.g., nuclear distance) is proportional to their distance along the polymer (e.g., genomic distance; , dashed line; ). We measured the nuclear distance between C1 and successively more distal points in the Mmu14 region, ending 4.0 Mb away at D4. A simple linear relationship between the mean-squared nuclear distance and the genomic distance was not observed (, black lines). Rather, this relationship was multiphasic, with at least two transitions in slope. The initial 1.5 Mb (C1–C2) showed a consistent nuclear separation, producing a line with no slope and indicating highly nonrandom substructure. The central 2 Mb of the region produced a line with a positive slope, which is consistent with a short segment of random walk. The distal end of the Mmu14 region marked a transition to a negative slope, additionally suggesting that the distal end loops back toward the proximal end, which is similar to previously reported 2-Mb giant chromatin loops (). These multiple relationships indicate that the Mmu14 region is not folded by a simple random walk of the chromatin fiber, but contains specific subdomains with different folding properties. Chromosome folding that is based on the genomic distribution of genes suggests a relationship to gene activity. To explore this potential relationship, we examined the expression status of Mmu14 region genes in several different cell types and then compared expression states to the region's nuclear organization. First, quantitative RT-PCR of transcripts from chondrocytes, embryonic stem (ES), and T cells indicated at least twofold variations in expression levels of 7/19 genes across the region (). In no cell type was an entire gene cluster completely inactive, though a few individual genes were undetectable above background levels. The expression of at least one gene per cluster in multiple cell types is consistent with this region's diverse mixture of genes. Second, we examined Mmu14 region structure in the nuclei of these three diverse cell types. Cluster–desert arrangements similar to those detected in NIH-3T3 cells were found in all three cases, and these occurred at similar frequencies (P > 0.1, χ test; ). These data indicate that the predominant cluster–desert structures correlate with the activated expression status of the whole cluster rather than the expression levels of individual genes within the cluster. We next determined whether transcriptional activity at these active gene clusters affects cluster and desert nuclear organization. NIH-3T3 cells were treated with 5,6-dichlorobenzimidazole riboside (DRB) for 60 min to arrest transcription. This period of inhibition is sufficient for other regions of the genome to reorganize (). However, DRB treatment did not significantly change the frequencies of the predominant Mmu14 cluster–desert conformations in nuclei (P ≥ 0.5, χ test; ). Even after prolonged (5 h) DRB treatment, the predominant conformations remained, though overall nuclear morphology changed significantly (unpublished data). Thus, transcriptional elongation by itself does not maintain the predominant arrangements of Mmu14 gene clusters and deserts. The mammalian nucleus is organized into several functional compartments that are marked by accumulations of specific proteins. These compartments include the nucleolus, splicing factor–rich domains or “speckles,” and the nuclear periphery. Specific genes preferentially associate with these distinct compartments (e.g., ribosome DNA at nucleoli), and these associations are related to gene expression (; ). Given the tendency for the expressed Mmu14 gene clusters to aggregate in nuclei, we examined whether the clusters organize around nuclear compartments related to mRNA gene expression. Previous studies suggest that transcribing mRNA genes may converge at common nuclear sites, so-called transcription factories (; ). In cultured cell lines, including the NIH-3T3 cells used here, transcription sites are marked by thousands of small (∼70 nm) accumulations of nascent transcripts and RNA polymerase II (pol II; Fig. S3, available at ; ; ). Gene cluster associations with transcription factories might not be perturbed by DRB (), which halts pol II elongation rather than destabilizing pol II DNA binding (). Therefore, we determined whether Mmu14 gene clusters tend to associate with a common transcription site in uninhibited cells. Given the highly dispersed, complex pol II distribution in NIH-3T3 cells, we detected specific transcription sites for the Mmu14 gene clusters using RNA FISH. We examined the relative nuclear positions of transcribing clusters C1 and C2, and C2 and C4, each separated by ∼1.7 Mb (). For a given experiment in NIH-3T3 cells, transcripts from both of the probed clusters were detected in ∼60% of chromosomes (), which is consistent with variable expression levels reported for many homologous loci in the same nucleus (; ). In transcribing chromosomes, both separated and contacting transcript signals were detected (). Separated and contacting classes of signals were present at similar frequencies, indicating that transcribing gene clusters neither favor nor disfavor close nuclear aggregation. These findings argue against gene cluster associations in the Mmu14 region that are based solely on the clustering of genes at a common, small transcription factory. We note, however, that ∼50% of transcribing clusters are close enough that at times they could share a common transcription site. In addition to sites of transcription, genes can be functionally organized in the nucleus via association with larger nuclear domains. By fluorescence microscopy, such coassociating loci frequently appear to localize near each other rather than to directly overlap, similar to the gene clusters in the Mmu14 region. Though NIH-3T3 cells do not contain the large accumulations of pol II reported in primary cells, they do contain splicing factor–enriched domains or speckles (Fig. S3), which associate with multiple genes (; ). We examined whether the aggregated Mmu14 gene clusters organize around splicing factor speckles with triple-label experiments (Fig. S3). Multiple gene clusters did not align with or surround any splicing factor domains. Rather, they were typically localized to a different focal plane. Only 8% of Mmu14 signals contacted splicing factor domains (), which is similar to other gene-poor chromosome regions and loci that associate with splicing factor domains at random frequency (; ). In addition to nuclear domains associated with active genes, other nuclear regions are enriched with inactivated and gene-poor chromatin (; ). These include the heterochromatic centromeres and nuclear periphery. Quantitative 3D image analysis indicated that the Mmu14 region is most concentrated near the nuclear periphery (). Approximately half (51%) of Mmu14 regions localize within the nuclear zone defined by the outermost 10% of the nuclear radius, which represents only 27% of the nuclear volume (). Interestingly, this analysis also suggested that Mmu14 deserts localize more peripherally than gene clusters (). To further examine the organization of gene deserts with the nuclear periphery, we scored cluster and desert signals for association with the outermost edge of the nucleus, which was defined by lamin B receptor immunostain (). We found that deserts more frequently contact and align with the nuclear edge than gene clusters. Moreover, this enrichment was detected predominantly in chromosomes with zigzag and gene cluster hub conformations (P < 1 × 10, χ tests), in contrast with striped fibers (P = 0.2). These data indicate a preferential alignment of gene deserts with the edge of the nucleus, and suggest that this alignment plays a role in Mmu14 region folding. Though gene deserts preferentially align with the nuclear edge in zigzags and gene cluster hubs, we did not detect significant enrichment of any one folding pattern in the peripheral or more internal nuclear zones (P = 0.3, χ test; Fig. S4, available at ). 60% of zigzags, 54% of gene cluster hubs, and 46% of striped fibers contact the nuclear periphery. Of the chromosomes contacting the periphery, a similar proportion (75%) of each morphological class has multiple deserts contacting the periphery. In contrast, gene cluster contacts with the periphery do vary according to conformation. Fewer zigzags and gene cluster hubs than striped fibers have gene clusters aligned with the nuclear edge (). These findings suggest that clusters shift away from the nuclear periphery to form zigzags and gene cluster hubs. Half of the Mmu14 signals localize to the nuclear interior, where the most prominent heterochromatic domains are chromocenters. These centromere aggregates appear as bright spots in DAPI-stained mouse nuclei (Fig. S4; ). We found that 43% of the Mmu14 regions in the internal nuclear zone contact the edges of chromocenters (). However, the frequency of these contacts was distributed equally between gene clusters and deserts (Fig. S4). Thus, neither clusters nor deserts specifically align with chromocenters. These data suggest that Mmu14 conformation does not simply reflect a general association of gene deserts with heterochromatin and that other interior nuclear compartments may be linked to Mmu14 region folding instead. s h o w d e f i n e d n u c l e a r o r g a n i z a t i o n f o r t h e c o l l e c t i v e s e t o f g e n e s a c r o s s a 4 . 3 - M b c h r o m o s o m e r e g i o n . A r r a y e d g e n e c l u s t e r s a n d i n t e r v e n i n g g e n e d e s e r t s f o r m m u l t i p l e , b u t p r e d o m i n a n t , 3 D a r r a n g e m e n t s i n n u c l e i , t y p i c a l l y m a r k e d b y h u b s o f m u l t i p l e a s s o c i a t e d g e n e c l u s t e r s . T h o u g h g e n e c l u s t e r s a r e a c t i v a t e d a n d e x p r e s s e d , t h e i r n u c l e a r a g g r e g a t i o n i s n o t s i m p l y c o r r e l a t e d w i t h o n - g o i n g t r a n s c r i p t i o n , c o n s i s t e n t w i t h t h e d i v e r s e f u n c t i o n s o f m u l t i p l e g e n e s . G e n e - d e p l e t e d d e s e r t s p r e f e r e n t i a l l y a l i g n w i t h t h e n u c l e a r p e r i p h e r y , s u g g e s t i n g t h a t t h i s f u n c t i o n a l n u c l e a r c o m p a r t m e n t , a s w e l l a s g e n e d e s e r t s , p l a y a r o l e i n c h r o m o s o m e r e g i o n a r c h i t e c t u r e . C o l l e c t i v e l y , o u r f i n d i n g s s u g g e s t a s e q u e n c e - d e p e n d e n t , d y n a m i c 3 D f r a m e w o r k f o r t h e o r g a n i z a t i o n o f m u l t i p l e g e n e s w i t h i n a c h r o m o s o m e r e g i o n . NIH-3T3 fibroblasts (a gift from L. Lau, University of Illinois, Chicago, IL), primary mouse embryo fibroblasts, and SV-40–transformed chondrocytes (a gift from T. Barak, The Jackson Laboratory, Bar Harbor, ME) were grown at 37°C in DME (Invitrogen) supplemented with 10% heat-inactivated fetal bovine serum. Feeder-independent ES cells (XC749; a gift from G. Cox, The Jackson Laboratory) were cultured in Glasgow's minimal Essential Medium, 10% fetal bovine serum, and leukemia inhibitory factor. CD8 T cells (B/nx3; a gift from D. Roopenian, The Jackson Laboratory, Bar Harbor, ME) were grown in DME, 10% fetal bovine serum, and 50 U/ml interleukin 2. Where indicated, cells were treated with 30 μg/ml DRB (Calbiochem). RNA was isolated with TRIzol (Invitrogen), treated with RNase-free DNaseI (Ambion), and 1 μg RNA per cell type was reverse transcribed with random hexamers and SuperScript II (Invitrogen). Gene expression levels for biological replicates, as well as technical triplicates, were quantified using an ABI7500 and SYBR-green (Applied Biosystems). Primer pairs are listed in Table SI (available at ). Expression levels for each target gene were determined relative to the constitutive metabolic enzyme glucose phosphate isomerase ( using 2 , where ΔC = C target − C . Cells grown on coverslips were fixed with 4% formaldehyde according to two previously established protocols (; ). BAC pools were nick-translated with biotin-11-dUTP (Roche) or digoxigenin-16-dUTP (Roche; ). Fixed cells were base-hydrolyzed, heat denatured, and hybridized with 200 ng of each probe and 40 μg mouse CoT1 DNA (Invitrogen; ). Probes were detected with anti-digoxigenin antibody or avidin labeled with TRITC or fluorescein (Roche). Cells were counterstained with 1 μg/ml DAPI (Sigma-Aldrich) and mounted in Vectashield (Vector Laboratories). Cells fixed as in the previous section were immunostained with antibody CTD 4H8 (Upstate Biotechnology), rat anti-BrdU (Harlan SeraLab), guinea pig anti–lamin B receptor (a gift from L. Schultz, The Jackson Laboratory; ), or rabbit anti-SRm300 (a gift from B. Blencowe, University of Toronto, Toronto, Canada; ) as previously described (). They were then subsequently detected with Alexa Fluor 488–goat anti–mouse IgG (Invitrogen), FITC–goat anti–rat, Cy5–goat anti–guinea pig, or Cy5–goat anti–rabbit IgG (Jackson ImmunoResearch Laboratories). Cells were examined with an Axioplan2 (Carl Zeiss MicroImaging, Inc.) or a DMRE (Leica) microscope equipped with a filter wheel, triple-bandpass epifluorescence filter set (model 83000; Chroma Technology), and a 100×, 1.4 N.A., oil PlanApo objective, at 21 ± 1°C. Images were acquired with a Micromax (Princeton Instruments) camera and Metamorph imaging software (Universal Imaging), or a Zeiss Axiocam MRm and Axiovision 4.1. Image stacks were acquired at 0.1 μm intervals, deconvolved and rendered in 3D with AutoDeblurr (Media Cybernetics, Inc.). In some cases ( and Fig. S3), cells were also imaged with a confocal microscope (SP2; Leica) with a 100X, 1.4 N.A., oil PlanApo objective and with pinhole set at 1 Airy disc. The following definitions were applied for scoring morphological pattern: striped, at least six clearly alternating cluster–desert signals along a linear fiber; zigzag, >50% of cluster signals shifted in one direction from desert signals; gene cluster/red hub, central core of at least three clusters (or red foci with shifted labeling scheme), with peripheral deserts (or green foci) >180° around the cluster hub. Statistical differences in morphological pattern were determined by χ tests (). Packing ratios relative to naked DNA were determined from 3D images by measuring along the length of largest overall structure. Measurements of distances between specific gene clusters and deserts were compared by two-sided KS tests (). Image stacks were both manually (MetaMorph; Molecular Devices) and automatically segmented with custom written software (Khoros; AccuSoft; ), with similar results. For analysis of Mmu14 region radial distribution, gene cluster and desert signals in confocal image stacks were mapped relative to a series of 25 shells that expand from the center of the nucleus to the periphery (100; ). The spherical 1-Mb chromatin domain model (; ) was adapted to accommodate a diploid mouse genome in a spherical 8-μm-diam nucleus. In this model, each chromosome is described by a 1-Mb linear chain and 500-nm diam domains, similar to empirically observed replication foci (; ). 1-Mb domains are connected by 100-kb DNA linkers modeled with an entropic spring potential, which enforce a mean distance of 600 nm between consecutive domains. Different domains interact against each other by a slightly increasing exclusive potential that allows a certain amount of overlap. In addition, to maintain a mean chromosome territory volume similar to that measured in nuclei, each chain of domains representing a chromosome was surrounded by a weak barrier potential. Approximately 200,000 Metropolis Monte Carlo steps were calculated to independently move the positions of all chromosome domains and “relax” all chromosomes from a condensed, mitotic-like state. To ensure equilibrated interphase configurations, 200,000 additional steps were sampled (; ). To measure separation distances between each Mmu14 gene cluster and desert, the gravity centers of Mb domains corresponding to these regions in the virtual Mmu14 were identified for both homologues in 50 simulated nuclei. Fig. S1 shows BAC contigs used for probing Mmu14 and Mmu15 regions. Fig. S2 shows predominant cluster–desert conformations are independent of fixation conditions, sample size, and immortalization of fibroblasts. Fig. S3 shows gene cluster organization relative to pol II sites and splicing factor–rich domains. Fig. S4 shows cluster–desert organization relative to chromocenters. Table SI shows primer sets used for real time RT-PCR. Online supplemental material is available at .

The spindle assembly checkpoint (SAC) is activated during each mitosis to monitor the attachment of sister chromatids to the spindle (). Upon biorientation of all sister chromatid pairs, the SAC is switched off, and anaphase ensues. SAC components such as products of the (mitotic arrest deficient) and (budding uninhibited by benzymidazole) genes are recruited to kinetochores in prometaphase, where they monitor the attachment of microtubules and the tension that builds up between bipolarly attached sister chromatids (). Critical to the SAC is the interaction of Mad2 with Cdc20 (; ). The latter is a positive regulator of the anaphase-promoting complex or cyclosome, whose function is required for progression into anaphase (). In mitosis, Mad2 is continuously recruited to kinetochores and is released from these structures in a form that binds Cdc20 and sequesters it in an inactive form (, ; ). When all chromosomes are aligned on the metaphase plate, Cdc20 is reactivated, and the consequent activation of the anaphase-promoting complex or cyclosome triggers anaphase. Mad1 is required to recruit Mad2 at kinetochores and for efficient formation of the Mad2–Cdc20 complex. Two models have been proposed to explain the role of Mad1 in eliciting the formation of the Mad2–Cdc20 complex (for review see ; ; ). The Mad2 exchange model proposes that Mad1 recruits open Mad2 (O-Mad2) at the kinetochore and changes its conformation from O-Mad2 to closed Mad2 (C-Mad2). C-Mad2 then dissociates from Mad1 and binds Cdc20. This model depicts Mad1 as a catalyst of the conversion of O-Mad2 into C-Mad2, which, in turn, is required for Mad2 to bind Cdc20 (). However, the Mad2 exchange model is weakened by structural observations, indicating that Mad1 and Cdc20 bind the same pocket of Mad2. In the frame of the Mad2 exchange model, this implies that Mad1 and Cdc20 compete for Mad2 binding, which would rule out a role for Mad1 as a direct activator of Mad2 for Cdc20 binding (). The Mad2 template model resolves this difficulty by incorporating a remarkable property of Mad2: the ability of its two conformers, O- and C-Mad2, to bind each other in a conformational dimer (; ,). The model proposes that the kinetochore receptor of O-Mad2 is a tight complex between Mad1 and C-Mad2 (the Mad1–Mad2 core complex; ; ). Mad1 provides this very sturdy complex with an N-terminal kinetochore-targeting domain and a C-terminal Mad2-binding motif. The latter generates a stable form of kinetochore-bound C-Mad2 that acts as the O-Mad2 receptor. In the Mad2 template model, the C-Mad2 pool bound to Mad1 at the kinetochore and the O-Mad2 pool in the cytosol are distinct and nonexchanging. Thus, the model does not imply that Mad1 and Cdc20 compete for Mad2 binding, resolving the contradictions of the Mad2 exchange model (; ). Furthermore, the Mad2 template model provides a useful molecular framework to understand the existence of two distinct kinetochore pools of Mad2 revealed by FRAP (). Specifically, ∼50% of kinetochore Mad2 exchanges rapidly at unattached kinetochores, whereas a remaining 50% of Mad2 is stably bound (). The observation that Mad1 is also stable at unattached kinetochores (; ) prompted the suggestion that a stable Mad1–Mad2 complex might be involved in the recruitment of a cycling cytosolic fraction of Mad2 (). When combined with the molecular information of the Mad2 template model, these experiments suggest that the kinetochore cycle of Mad2 represents the rate of transformation of O-Mad2 into Cdc20-bound C-Mad2. An implication of the Mad2 template model is that the interaction of the O and C conformers of Mad2 facilitates the conversion of cytosolic O-Mad2 to Cdc20-bound C-Mad2. This might occur via mass action after concentrating O-Mad2 and Cdc20 at kinetochores or possibly catalytically by facilitating the structural conversion of Mad2 from O- to C-Mad2. Although initial evidence has been provided indicating that the interaction of O- and C-Mad2 is important for the SAC in HeLa cells (,), a more rigorous analysis is required. We studied the properties of the 's homologue of Mad2 (ScMad2), asking whether we could identify the biochemical and genetic properties supporting the Mad2 template model in mammalian cells. Our new genetic and biochemical evidence is completely consistent with the Mad2 template model. Two mutants of HsMad2 (the point mutant Arg133-Ala and the double point mutant Arg133-Glu/Gln134-Ala, abbreviated as Mad2 and Mad2, respectively) are impaired in the interaction between O- and C-Mad2 (; ,). The residues map to a solvent-exposed surface of Mad2, and their mutation does not significantly affect Mad2's structural stability (). The choice of using the Mad2 double mutant rather than the double alanine point mutant Mad2 arose because the recombinant form of the latter was largely insoluble, whereas good yields of Mad2 could be recovered from the soluble bacterial fraction (). Both Mad2 and Mad2 bind Mad1 and Cdc20 in vitro with identical affinity relative to wild-type Mad2 (Mad2; , ; ). Although the overexpression of Mad2 and Mad2 elicits a mitotic arrest (; ), near-physiological concentrations of these mutant proteins were unable to support the SAC in HeLa cells concomitantly depleted of Mad2 by RNAi (). To carry out more rigorous complementation experiments, we examined the effects of equivalent mutations on the SAC in . Arg133 and Gln134 of HsMad2 are conserved in evolution. The equivalent yeast residues are Arg126 and Gln127 (). We assayed the ability of ScMad2, ScMad2-Arg126-Ala (ScMad2), and ScMad2-Gln127-Ala (ScMad2) to restore the SAC deficiency caused by deleting in . Cells arrested in G1 with α factor were released in the cell cycle in the presence of nocodazole to activate the SAC. To assess SAC proficiency, we monitored the ability to arrest in mitosis and to prevent re-replication, the lack of rebudding, and the retention of sister chromatid cohesion. Wild-type cells completed DNA replication at ∼60 min after release from the G1 block in nocodazole and arrested as budded cells with 2C DNA content without rebudding or separating the sister chromatids (), which is indicative of an active SAC. Conversely, cells were unable to arrest, lost sister chromatid cohesion, rebudded, and re-replicated their DNA, which is indicative of a disrupted SAC. To test the complementation potential of different alleles, we integrated wild-type and mutant alleles at the locus of the strain. Mad2, Mad2, and Mad2 were expressed at similar levels, and their expression was essentially identical to that of endogenous Mad2 (see ). The expression of Mad2 in the strain fully restored the SAC (). However, the expression of Mad2 and Mad2 failed to complement the lack of . Cells expressing these proteins underwent sister chromatid separation, rebudding, and re-replication with timings that were very similar to those displayed by the bare strain. These results demonstrate that the ScMad2 surface containing Arg126 and Gln127 is essential for the SAC, confirming our previous observations in HeLa cells (). The residues equivalent to Arg126 and Gln127 (Arg133 and Gln134) in HsMad2 are part of an interface that mediates the interaction of the O- and C-Mad2 conformers (,). It is possible that the inability of Mad2 and Mad2 to support the SAC in results from the impairment of an equivalent O-Mad2–C-Mad2 interaction. However, it is unknown whether ScMad2 is endowed with the same unusual biochemical features that characterize HsMad2 and that include the ability to adopt two stable conformations and the ability of the opposite conformers to form a complex. Thus, we set out to address the important question of whether two interacting conformers of ScMad2 exist as shown previously for HsMad2 (). For this, we first tested the ability of purified recombinant ScMad2 to bind GST fusions of the Mad2-binding motifs of ScMad1 and ScCdc20 (GST-Mad1 and GST-Cdc20) in a solid phase binding assay (). Mad2 bound effectively to GST-Cdc20 and GST-Mad1 immobilized on glutathione–Sepharose (GSH) beads (, lanes 6 and 10). As a result of the ∼40% sequence identity between HsMad2 and ScMad2 and because the Mad2-binding motifs of Mad1 and Cdc20 conform to the same consensus sequence in different species (; ), we assume that ScMad2 adopts a C-Mad2 conformation when bound to ScMad1 and ScCdc20 similar to that adopted by HsMad2 when bound to its human partners. The deletion of 10 residues from the C terminus of HsMad2 (HsMad2) affects the structural stability of the C-Mad2 conformer while leaving the stability of O-Mad2 untouched, creating a constitutively open form of Mad2 (, ; ). Because the ability to reach the C-Mad2 conformation is critically required to bind Mad1 and Cdc20, Mad2 is inapt to bind Mad1 or Cdc20 even if the residues within the deletion are not in direct contact with Mad1 or Cdc20 (; , ; ). We created an equivalent mutant of ScMad2 (ScMad2) by deleting six residues from its C terminus (the C-terminal tail of ScMad2 is four residues shorter relative to HsMad2). Unlike ScMad2, pure recombinant ScMad2 was unable to bind GST-Mad1 or GST-Cdc20 (, lanes 7 and 11). Essentially identical results were obtained in solution using isothermal titration calorimetry (Table S1, available at ). Thus, ScMad2 is a constitutively open form of Mad2 that is unable to bind Mad1 or Cdc20 and is similar in all of these respects to HsMad2 (; ; ). We next tested whether the O- and C-Mad2 conformers of ScMad2 are capable of forming a complex like their human counterparts. First, we created C-Mad2 on solid phase by allowing ScMad2 to bind GST-Mad1 and GST-Cdc20 as in the experiments shown in (lanes 6 and 10). This time, however, after washing out the excess of unbound Mad2, we added Mad2 (i.e., O-Mad2) to see whether it could be retained on solid phase via an interaction with the previously bound C-Mad2. Indeed, ScMad2 was found on solid phase in a complex with GST-Mad1–C-Mad2 and GST-Cdc20– C-Mad2 (, lanes 8 and 12). This confirms the existence of an interaction of O- and C-Mad2 as described previously for HsMad2 (,). To test whether mutating Arg126 and Gln127 affects the interaction between O- and C-Mad2, we repeated this experiment with ScMad2 after its expression in bacteria and purification to homogeneity. Like Mad2, Mad2 bound effectively to GST-Mad1 and GST-Cdc20 (, lanes 6 and 10), indicating that this mutant can adopt the C-Mad2 conformation like Mad2. Indeed, Mad2 and Mad2 bound Mad1 and Cdc20 synthetic peptides with essentially identical affinities in isothermal titration calorimetry measurements (Table S1). Next, we incubated Mad2 with GST-Mad1 or GST-Cdc20 to create the closed conformer, washed away the excess of unbound Mad2, and added Mad2 (O-Mad2). This time, we failed to observe any retention of ScMad2 on solid phase (, lanes 8 and 12). Given our previous analyses of the effects of mutating Arg133 and Gln134 in HsMad2 (), the conservation of these residues in ScMad2, and the identity of results with ScMad2 and HsMad2, we conclude that Arg126 and Gln127 map to the O-Mad2–binding surface of C-Mad2 and that their concomitant mutation into glutamate and alanine, respectively, prevents this interaction. The interaction between O- and C-Mad2 was verified in solution using purified proteins (). Both ScMad2 (O-Mad2) and the complexes of Mad2 or Mad2 with the high affinity Cdc20 synthetic peptide (C-Mad2) eluted as apparent monomers from a Superdex-200 size-exclusion chromatography (SEC) column (), indicating that both O- and C-Mad2 are monomeric (although the C-Mad2 species forms dimers with the Cdc20 peptide, this is only a 17-residue segment that does not significantly change the Stokes' radius of Mad2). When combined stoichiometrically at 25°C for 60 min, ScMad2 and the Mad2–Cdc20 complex formed an O–C complex that eluted with a Stokes' radius larger than that of the individual species and compatible with the molecular mass expected for an ScMad2 dimer (∼48 kD, as the molecular mass of the monomeric protein is ∼24 kD; ). Fractions corresponding to this peak contained apparently stoichiometric amounts of each component. When ScMad2 was combined with Mad2–Cdc20 and the result of the incubation was analyzed by SEC, no equivalent shift in the elution profile was observed (). Although ScMad2 coeluted with Mad2–Cdc20, the comparison of the peak of elution with those of the individual proteins displayed in clarifies that this was simply caused by the overlap of two distinct peaks with essentially identical elution volumes. These results confirm that the RQEA mutation affects the interaction of C- with O-Mad2. We also investigated the state of the oligomerization of ScMad2. Pure ScMad2 eluted from a Superdex-75 SEC column as a monomer (). To assess whether this monomer is O-Mad2, as expected for Mad2 in the absence of Mad1 or Cdc20, we examined the effects of adding increasing concentrations of the Cdc20 synthetic peptide on the SEC profile of ScMad2. At a 4:1 Mad2/Cdc20 ratio, a dimeric species roughly engaging 50% of total Mad2 appeared (). This can be easily explained if we assume that Cdc20 transformed ∼1/4 of O-Mad2 in C-Mad2 and that this bound to an equimolar amount of O-Mad2, leaving half of the original O-Mad2 in the monomer peak. (As mentioned in the previous paragraph, the O–C dimer is actually a trimer if we consider Cdc20, but the latter does not contribute significantly to the elution profile of the O–C-Mad2 dimer.) Consistent with this hypothesis, at a 2:1 Mad2/Cdc20 ratio, most Mad2 eluted as a dimer (). When the Mad2/Cdc20 ratio was decreased to cause the conversion of more O-Mad2 to C-Mad2–Cdc20, the Mad2 dimer progressively disappeared, whereas a monomer peak corresponding to C-Mad2–Cdc20 accumulated (). (Again, this is technically a dimer whose elution is not significantly influenced by Cdc20). The elution volume of the C-Mad2 monomer was slightly but consistently retarded relative to that of the O-Mad2 monomer. Altogether, the experiments in and strongly suggest that the O and C conformers of ScMad2 form a dimeric complex like the one previously described for the equivalent conformers of HsMad2 (,). The similarity with HsMad2 extends to the fact that neither conformer forms dimers without the other conformer, contrary to the proposition that C-Mad2 forms dimers (). Furthermore, our data indicate that the interface containing Arg126 and Q127 of C-Mad2 is important for binding O-Mad2. In , we show that the converse is also true: namely, that a similar (but most likely not identical) interface in O-Mad2 is important for binding C-Mad2. Mad1 and Cdc20 bind the same Mad2 pocket and generate structurally similar C-Mad2 conformers. We have proposed that kinetochore recruitment of Mad2 requires a tight Mad1–Mad2 complex at the kinetochore whose C-Mad2 component offers the critical binding surface for cytosolic O-Mad2 (; ). To test whether this is also the case for ScMad2, we created a recombinant ScMad1–ScMad2 complex and purified it to homogeneity (). This complex lacks the kinetochore-binding domain of Mad1 (located in the N-terminal half of Mad1) but contains coiled-coil segments that mediate the dimerization of Mad1 and two Mad2-binding domains, one per Mad1 chain, that are required for high affinity binding of Mad2. The resulting complex contains a very stable tetrameric core, as shown previously for an equivalent human complex (, ). Although we have not analyzed the stability of the ScMad1–ScMad2 complex in detail, we failed to observe any changes in the relative stoichiometry of its components during purification, suggesting it is very stable. To test the ability of purified O-Mad2 to bind C-Mad2 in the Mad1–Mad2 complex, we covalently attached the AlexaFluor488 fluorophore to ScMad2, ScMad2, or ScMad2. We analyzed the resulting labeled proteins by SEC using a Superdex-200 column () after the elution of AlexaFluor488-labeled Mad2 at 280 and 495 nm (; black and green traces, respectively). Excitation of the AlexaFluor fluorophore at 300 nm using a UV trans-illuminator provided a useful means of detecting the labeled protein in elution fractions after SDS-PAGE separation. The same gels were also stained with Coomassie (; top and bottom gel sections). This revealed that all three proteins eluted from the SEC column apparently as monomers. We then mixed stoichiometric amounts of the fluorescent Mad2 species to the Mad1–Mad2 complex (at concentrations of ∼20 μM of monovalent Mad2 and 10 μM of divalent Mad1–Mad2 core complex) and, after a 1-h incubation, we analyzed the products by SEC. AlexaFluor-labeled ScMad2 and ScMad2 bound Mad1–Mad2 with high affinity, as evidenced by the essentially complete shift of the AlexaFluor fluorophore to fractions containing the Mad1–Mad2 core complex (). On the other hand, AlexaFluor-ScMad2 was unable to bind the Mad1–Mad2 core complex (). Addition of the Cdc20 synthetic peptide (at 200 μM) to “external” AlexaFluor–O-Mad2 prebound to the Mad1–Mad2 core complex resulted in dissociation of the AlexaFluor-labeled species in a low molecular weight complex, presumably in a complex with the Cdc20 peptide (). However, when the complex of O-Mad2 with Mad1–Mad2 was tested with Cdc20, neither Mad2 nor C-Mad2 that bound to Mad1 was released in a complex with Cdc20 (). This is consistent with the inability of Mad2 to bind Cdc20 and shows that the Mad1–Mad2 complex is stable and is not disrupted by Cdc20. Consistently, C-Mad2 in the Mad1–Mad2 complex did not dissociate from Mad1 if Cdc20 was added in the absence of external O-Mad2 (unpublished data). Overall, these results are indistinguishable from those previously described for HsMad2 and its interaction with Mad1 and Cdc20 () and indicate that the O-Mad2 conformer of ScMad2 binds the C-Mad2 conformer in the Mad1–Mad2 complex. Because Mad2 is unable to bind Mad1, whereas Mad2 is a normal Mad1 ligand, we conclude that Mad1 binding is not required for the binding reaction analyzed in . Rather, the interaction involves a surface predominantly or exclusively based on O- and C-Mad2. The experiments reported in show that O-Mad2 is unable to bind the wild-type C-Mad2 protein in the Mad1–Mad2 complex. Conversely, in , these mutations were shown to affect the binding to a functional O-Mad2 (ScMad2, whose deficiency consists uniquely in being unable to turn into C-Mad2). Thus, the surface containing Arg126 and Gln127 is involved in Mad2 dimerization both on the O and C conformers. Previous structural investigations demonstrated that human Mad2 has the O-Mad2 conformation, that Mad2 and Mad2 bound to Cdc20 or Mad1 are folded as C-Mad2, and that the two human conformers O- and C-Mad2 bind each other (, , ; ; ,). For instance, human O-Mad2 binds human Mad1–C-Mad2 (), which is a completely analogous reaction to that involving yeast proteins in . So far, we assumed that ScMad2 adopts O- and C-Mad2 conformations whose encounter results in their dimerization as for HsMad2. Although a direct structural investigation of the conformational states of ScMad2 goes beyond the purpose of this study, we wished to provide stronger evidence that ScMad2 adopts O- and C-Mad2 conformations like HsMad2. If ScMad2 has the O-Mad2 conformation previously characterized for HsMad2, it might be expected to bind human C-Mad2. To test this, we mixed stoichiometric amounts of human Mad1–C-Mad2 complex () with AlexaFluor-ScMad2 and analyzed the resulting species by SEC on a Superdex-200 column (). Confirming our expectation that ScMad2 has the same O-Mad2 conformation that was previously demonstrated for HsMad2 (), AlexaFluor-ScMad2 coeluted with the human Mad1–C-Mad2 complex. Binding was specific because AlexaFluor-ScMad2 that preincubated with the ScCdc20 synthetic peptide to create C-Mad2 failed to bind the human Mad1–Mad2 complex (). Conversely, if Mad1-bound ScMad2 has the same C-Mad2 conformation previously observed in the structure of the human Mad1–C-Mad2 complex (), it might be expected to bind human Mad2, which has been shown to fold as O-Mad2 (). Indeed, AlexaFluor-HsMad2 bound tightly to the Mad1–Mad2 complex (), strongly suggesting that the conformation of ScMad2 bound to ScMad1 is C-Mad2. Preincubation of HsMad2 with a synthetic peptide encompassing the Mad2-binding site of HsCdc20 (Cdc20) to create human C- Mad2 prevented its binding to the Mad1–Mad2 complex (). These experiments indicate that the interface mediating the interaction of O- with C-Mad2 is strongly conserved in evolution (as already exemplified by the conservation of R126 and Q127), underscoring the biological importance of Mad2 dimerization in checkpoint function. The results of our biochemical characterization of ScMad2 are consistent with the Mad2 template model (,). The latter depicts C-Mad2 stably bound to Mad1 (rather than Mad1 itself) as a trigger that is required for Mad2 to bind Cdc20 in living cells. In this view, the absolute requirements for Mad1 in activating Mad2 for Cdc20 binding are limited to its function in localizing a pool of C-Mad2 to the kinetochore. This form of C-Mad2 recruits O-Mad2 from the cytosol to assist in its transformation into C-Mad2 bound to Cdc20. To provide more evidence in favor of this model, we asked whether interfering with the O–C-Mad2 interaction weakens the SAC response. Because O-Mad2 binds C-Mad2 but is unable to be passed onto Cdc20 (), this mutant is expected to compete with the binding of O-Mad2 to C-Mad2, and we asked whether its expression perturbed the SAC response in . First, we determined that ScMad2 expressed from the endogenous promoter is unable to sustain the checkpoint in a strain (Fig. S1, A and B; available at ), which is consistent with a previous study () and with our own observation that ScMad2 is unable to bind Mad1 or Cdc20. To test our hypothesis, we expressed ScMad2 from the promoter and tested checkpoint function after the release of wild-type cells from a G1 arrest in nocodazole (). As expected, cells overexpressing ScMad2 but not those overexpressing full-length ScMad2 in a wild-type background lost sister chromatid cohesion and rebudded and re-replicated their chromosomes, which is indicative of a checkpoint defect. Thus, ScMad2 has a dominant-negative effect on the SAC analogous to that observed in vertebrate cells (; ; ), which is in agreement with our hypothesis that this mutant interferes with the interaction of Mad2 with Cdc20. Although we have been thus far unable to coimmunoprecipitate the complex between ScMad2 and the Mad1–Mad2 complex, we show that ScMad2 overexpressed from the promoter in a strain was unable to bind Mad1 (Fig. S1 C). As shown in and in Table S1, Mad2 and Mad2 bind Mad1 and Cdc20 effectively in vitro (the single mutants Mad2 and Mad2 bind equally well; unpublished data). However, the reintroduction of at the endogenous locus of a strain restored the SAC, whereas the expression of and failed to do so (). Because our model proposes that the interaction of the Mad2 conformers is essential to activate Mad2 for Cdc20 binding, Mad2 mutants that are unable to sustain this interaction should be unable to reach Cdc20 in living cells. We decided to assess the amounts of Mad1 and Cdc20 that bound to Mad2, Mad2, and Mad2 expressed in a background. The association of Mad1 with Mad2 is not regulated during the cell cycle (). To test the ability of Mad2 and its mutant variants to bind Mad1, we performed Mad1-myc18 immunoprecipitations (IPs) from cycling cells of and strains carrying the , , and alleles integrated at the locus. Consistent with their ability to bind Mad1 in vitro, Mad2 and Mad2 were found to associate with Mad1 in vivo as efficiently as Mad2 (). Unlike the levels of Mad1 and Mad2, which are essentially constant during the cell cycle, Cdc20 is a cell cycle–regulated protein whose destruction is required for mitotic exit (). Thus, we investigated the levels of Mad2 and its mutant variants associated with Cdc20 during mitosis, when expression is maximal (). For this, cells and cells carrying the , , and alleles integrated at the locus were arrested in G1 with α factor and were released into the cell cycle in the presence of nocodazole, which activates the SAC, promoting the formation of the Mad2–Cdc20 complex. All three strains reentered the cell cycle normally and completed DNA replication synchronously roughly 60 min after release from the arrest. To assess the amounts of Mad2 bound to Cdc20, IPs were performed from extracts of cells harvested 80 min after release from the G1 arrest. To confirm that these cells were in mitosis regardless of their genotype, we monitored the levels of the mitotic cyclin Clb2. These were found to be identical (), showing that both checkpoint-proficient () and checkpoint-deficient ( or ) cells had been harvested while in mitosis. Although Mad2 expressed in cells bound normally to myc-Cdc20, the binding of Mad2 and Mad2 to myc-Cdc20 was severely impaired (). In the converse experiment, we found greatly diminished levels of myc-Cdc20 in Mad2 IPs (). Overall, these results indicate that the differences in the amount of Mad2–Cdc20 complex in cells expressing Mad2, Mad2, or Mad2 must be the result of the inability of the mutant proteins to support the interaction of O- with C-Mad2. We conclude that this interaction represents a critical step in the activation of Mad2 in the SAC. In this study, we show that two critical features of Mad2—the ability to adopt open and closed conformations and dimerization of the open and closed conformers—are both likely to be conserved in all eukaryotes. Both features appear to be required for Mad2 to bind Cdc20 and to sustain the checkpoint. We provide convincing evidence that recombinant ScMad2 folds as an O-Mad2 monomer that changes its conformation to C-Mad2 upon binding Mad1 or Cdc20. Recombinant HsMad2 forms oligomers in the absence of Mad1 or Cdc20 (). We have reanalyzed the mechanism of oligomerization of HsMad2 and found that the Mad2 oligomers are O–C dimers created by the partial, spontaneous conversion of O-Mad2 into an “empty” C-Mad2 (i.e., devoid of Mad1 or Cdc20), which, in turn, binds the remaining O-Mad2 (). Thus, HsMad2 oligomerization in vitro () appears to be based on the same O–C Mad2 interaction supporting the checkpoint but in the absence of Mad2 ligands. We suspect that empty C-Mad2 is unlikely to be the active Mad2 species, as proposed recently (). Our skepticism in regarding empty C-Mad2 as the direct binder of Cdc20 is based on the fact that the specific closed conformation of the Mad2 C-terminal tail would prevent the loading of Cdc20 onto empty C-Mad2 (; ). It seems sensible to suggest that if empty C-Mad2 ever formed in living cells, it would then need to unfold its C-terminal tail to be able to bind Cdc20. The reason why recombinant HsMad2 rearranges spontaneously to create empty C-Mad2, whereas recombinant ScMad2 does not appear to do so, is currently unclear. Our work shows identical mechanisms of O–C oligomerization for ScMad2 and HsMad2. As for HsMad2 (), the O- and C-Mad2 conformers of ScMad2 bind each other, whereas neither of them forms oligomers on their own. These observations are inconsistent with the proposition that C-Mad2 forms C–C dimers (). However, our results are completely consistent with an earlier study from the same authors showing that C-Mad2 created using the Mad2-binding site of Cdc20 is a monomer (). It is interesting to observe that because O–O and C–C dimers are not observed, there is a logical requirement for the O- and C-Mad2 surfaces involved in the O–C interaction to be different. The identification of residues whose mutation into alanine prevents binding of the mutant C-Mad2 conformer to wild-type O-Mad2 while leaving unaltered the ability of the mutant O conformer to bind wild-type C-Mad2 () confirms the idea that the specificity of the O–C dimerization is caused by elements of structural asymmetry. Conversely, Arg126 and Gln127 belong to a class of “symmetric” residues whose mutation affects binding to the opposite conformer both in the O and C state (suggesting, but not implying, that some level of symmetry at the O–C interface might also be present). This gives us an opportunity to explain our choice for using single or double point mutants of Arg126 and Gln127 (Mad2 or Mad2 vs. Mad2) in different experiments. Although single point mutants Mad2 and Mad2 display significant residual binding to a wild-type version of the opposite conformer, the double mutant Mad2 is significantly more penetrant and devoid of any significant residual binding activity toward the opposite conformer of Mad2 (,). Thus, we will use Mad2 if we want to test the interaction of a mutant Mad2 conformer to the opposite conformation of Mad2. On the other hand, the single point mutants Mad2 or Mad2 will be sufficient to disrupt binding if both binding interfaces are mutated. Accordingly, and alleles that are reintroduced in a strain are unable to reconstitute the SAC. In vitro, the level of disruption of the O-Mad2–C-Mad2 interaction observed in this case is roughly similar to that observed when testing the double point mutant RQEA against a wild-type surface (). provides a schematic account of the Mad2 template model. A stable Mad1–C-Mad2 complex recruits O-Mad2 from the cytosol via the O–C-Mad2 interaction, favoring its transformation into C-Mad2 bound to Cdc20. Strong in vivo evidence for the Mad2 template model comes from FRAP experiments revealing the presence of two distinct and quantitatively equivalent kinetochore pools of Mad2 with fast and slow turnover (; ). In the molecular description of the Mad2 template model, the stable and mobile pools of Mad2 coincide with Mad1-bound C-Mad2 and C-Mad2–bound O-Mad2, respectively. Formal proof that these two pools account for the FRAP rates observed in vivo needs to be provided. Because our data show that the Mad2–Cdc20 interaction in yeast requires O-Mad2–C-Mad2 binding, it is puzzling that Mad2 and Cdc20 bind spontaneously in vitro. A possible explanation for this apparent discrepancy is that the noncatalyzed rate of formation of the Mad2–Cdc20 complex is too slow to allow the accumulation of Mad2–Cdc20 that is required to sustain the SAC (). We speculate that the mechanistic significance of the interaction of O-Mad2 with Mad1-bound C-Mad2 is that the latter acts as a catalyst for the otherwise slow transformation of O-Mad2 into Cdc20-bound C-Mad2. A large energy barrier (and correspondingly slow kinetics) is expected for the conformational change required to turn O-Mad2 into C-Mad2, which implies the reorganization of an entire β sheet (, ; ). C-Mad2 may trigger the reorganization of the C-terminal tail of O-Mad2, creating a structural intermediate for its conversion into C-Mad2. The enrichment of Cdc20 at kinetochores would also favor its capture in a C-Mad2–Cdc20 complex. This reaction may also be negatively regulated. A negative regulator of the SAC such as p31 (; ), which binds exclusively to C-Mad2 and, therefore, is likely to act as a competitor of O-Mad2, might be expected to decrease the levels of C-Mad2 available to bind O-Mad2. However, a functional homologue of this protein has not yet been identified in , so the generality of this hypothesis remains unclear. The fact that a mutational impairment of the O-Mad2–C-Mad2 interaction abrogates the SAC acts in support of the Mad2 template model. Our experiments are unable to distinguish whether the C-Mad2–O-Mad2 interaction specifically requires Mad1-bound C-Mad2. More specifically, it is possible that this interaction also involves Cdc20-bound C-Mad2 in a positive feedback loop () as we have suggested previously (,). Thus, it will now be essential to dissect the specific functions of the C-Mad2 pools bound to Mad1 and Cdc20. Standard genetic techniques were used to manipulate yeast strains (). All yeast strains were derivatives of W303 ( and ) and are listed in . Cells were grown in YEP medium (1% yeast extract, 2% bactopeptone, and 50 mg/l adenine) supplemented with 2% glucose (YEPD), 2% raffinose (YEPR), or 2% raffinose and 1% galactose (YEPRG). α Factor was used at 2 μg/ml, and nocodazole was used at 15 μg/ml. All strains were normally grown at 25°C. pGEX- and pGEX- contain the coding sequence of ScMad1 and Cdc20, respectively. pET43-–6His- contains the coding sequence of separated from 6His- by a ribosome-binding site. We generated pET43-6His-, pET43-6His-, and pET43-6His- by replacing the BamHI–EcoRI fragment of pET43-6His-Mad2 () with coding sequences of mutant alleles. The HindIII–BglII fragment containing the whole coding region plus ∼400 bp of upstream and ∼280 bp of downstream sequence was cloned in HindIII–BamHI of YIplac128. The resulting pSP42 plasmid was integrated at the locus by EcoRV digestion. and fusions as well as a allele under the control of ∼400 bp of promoter were cloned in YIplac128 upstream of ∼280 bp of terminator to generate pSP187, pSP385, and pSP384 plasmids, whose integration was directed to the locus by EcoRV digestion. Single integrations were checked by Southern analysis. Mutant alleles were generated using QuikChange (Stratagene). was tagged with the myc tag immediately before the stop codon by one-step gene tagging (). The strain has been described previously (). ScMad2–6His-ScMad1 was generated in BL21-c41(DE3). After metal affinity chromatography on a 1-ml HiTrap Chelating HP column (GE Healthcare), the protein was purified by ion exchange on a Resource Q column (GE Healthcare) and dialyzed in buffer L (20 mM Hepes, pH 7.5, 300 mM NaCl, 10% glycerol, and 1 mM EDTA). Mad2 proteins were expressed in BL21(DE3) and purified essentially as described previously for the human complex (). Proteins were labeled with AlexaFluor488 succimidyl ester reactive dye (Invitrogen) as described previously (), with a final dye/protein ratio of ∼0.5. Analytical SEC was performed on a SMART device (GE Healthcare) using Superdex-75 or -200 PC 3.2/30 columns equilibrated in buffer L. 10 μM of divalent ScMad2–6His-ScMad1 complex was incubated for 1 h at 25°C with 20 μM AlexaFluor-ScMad2 or AlexaFluor-ScMad2 mutants. The reactions were separated by SEC. Elution was performed at 40 ml/min. Custom-built synthetic peptides were purchased from Eurogentec. For IPs, cells were lysed with glass beads in 50 mM Hepes, pH 7.6, 75 mM KCl, 1 mM MgCl, 1 mM EGTA, 1 mM AEBSF, 0.5 mM DTT, 120 mM β-glycerophosphate, and 0.1% Triton X-100 supplemented with a cocktail of protease inhibitors (Complete; Boehringer). 1–2 mg of cleared extracts was incubated for 2 h with antibody directly cross-linked to protein A–Sepharose, except in the case of the anti-Mad2 IPs. The slurry was washed three times with lysis buffer. Protein extracts were run on 15% SDS-PAGE gels. For Western blot analysis, proteins were transferred to Protran membranes. myc18-Cdc20 and Mad1-myc18 were detected with monoclonal antibody 9E10. Anti-Clb2 polyclonal antibodies were a gift from W. Zachariae (Max Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany). Anti-ScMad2 polyclonal antibodies used for and SF1C were provided by K. Hardwick (Wellcome Trust Centre for Cell Biology, University of Edinburgh, Edinburgh, UK; ). Anti-ScMad2 polyclonal antibodies used for were produced locally. Secondary antibodies were purchased from GE Healthcare and Bio-Rad Laboratories. GST-ScMad1 and GST-ScCdc20 were expressed in BL21-c41(DE3). After lysis by sonication in buffer A (10 mM Hepes, pH 7.5, 100 mM NaCl, 1 mM DTT, and 0.5 mM EDTA), 1% Triton X-100 was added. The GST proteins were purified with GSH agarose (GE Healthcare). ∼8 μg GST, GST-ScMad1, or GST-ScCdc20 on beads were incubated for 1 h at RT with 40 μg ScMad2, ScMad2, or ScMad2 in 0.3 ml of buffer A (final concentrations of GST fusion protein and Mad2 were ∼1 and ∼5 μM, respectively). Beads were washed twice with 0.4 ml of buffer A supplemented with 1% Triton X-100, and bound proteins were separated by SDS-PAGE. Flow cytometric DNA quantitation was determined on a FACScan (Becton Dickinson) as described previously (). Sister chromatid separation was followed on ethanol-fixed cells by visualizing tetracycline-repressor GFP fusion proteins bound to tandem repeats of tet operators integrated at ∼35 kb away from the centromere of chromosome V (). Fig. S1 shows the characterization of Mad2 in . Table S1 provides data on isothermal titration calorimetry. Online supplemental material is available at .

Endomembrane compartments create specialized environments that are optimized for diverse reactions, including protein folding, quality control, processing, sorting, and turnover. How these compartments are established and maintained is a question of fundamental importance. Compartment size is somehow coupled to cell growth such that it increases as cells grow and remains roughly constant in postmitotic cells. On the other hand, during mitosis or differentiation, or in response to stress, secretory compartments can undergo extensive up- or down-regulation depending on cell type, developmental timing, or stress condition (; ; ; ; ). Thus, homeostatic mechanisms that maintain a proper balance of membrane input and output at each compartment must be sufficiently flexible to allow extensive, rapid, and reversible changes. Although signaling pathways, transcriptional events, and other higher order processes undoubtedly participate (; ), compartment homeostasis might fundamentally rely on intrinsic features of the vesicle coat machinery. Transport vesicles typically fuse shortly after their formation, indicating that vesicle formation is rate limiting. Thus, the net flux of membrane through the Golgi apparatus, for example, would be determined by the rates at which vesicle coats acting at the ER and endosomes contribute input of vesicles and the rates at which vesicle coats acting at the Golgi drive vesicle export. What, then, determines the rate of vesicle production? Assembly of the coat protein complex (COP) II coat on the ER membrane is initiated by guanine nucleotide exchange of the GTPase Sar1p (). The presence of Sar1p-GTP leads to the successive recruitment of the coat components Sec23p–Sec24p and Sec13p–Sec31p (). COPII coat assembly is opposed by GTP hydrolysis by Sar1p triggered by Sec23p GAP activity and amplified by the presence of Sec13p–Sec31p (; ). Although each step may play a regulatory role that influences the rate of vesicle production, the self-terminating property of COPII assembly suggests that additional factors stabilize the coat on the membrane and regulate overall rate. Evidence suggests that cargo molecules in the ER membrane contribute (; ; ). For example, synchronized export of vesicular stomatitus virus G protein (VSVG) stimulates COPII vesicle budding, and inhibition of protein synthesis depresses it (). If compartment homeostasis is determined by vesicle production rates and vesicle production rates are influenced by cargo concentration, then it is important to ask whether compartment size is a function of cargo abundance. Here, we test the hypothesis that Golgi residents, in the guise of cargo at the ER, influence the size of the Golgi apparatus by regulating COPII assembly and thereby determining the extent of membrane input to the Golgi. Osmotic stress causes Golgi collapse and dispersal of Golgi components in the ER; yet, after cell volume recovery, the Golgi apparatus completely reassembles (). Biogenesis of the Golgi apparatus from the ER is recapitulated using drug washout after sequential treatment with brefeldin A (BFA), to inhibit Arf1, and H89, to block COPII assembly. That is, the Golgi apparatus efficiently and synchronously reassembles from the ER (). Inhibitor reversal is rapid, as indicated by restoration of membrane recruitment of Sec13 and β-COP at the earliest time points tested (). During biogenesis, Golgi growth is rapid until steady state is reestablished, implying a temporary shift in the input/output balance at the Golgi that favors input. To test the involvement of increased COPII assembly, we used morphological assays to determine the levels of COPII and Golgi assembly in cells before and during BFA/H89 washout. In normal rat kidney (NRK) cells, the Golgi marker giantin yielded the expected juxtanuclear ribbon structure before treatment, a dispersed ER pattern upon BFA/H89 treatment, and, during washout, a punctate vesicular-tubular cluster (VTC) pattern followed by reestablishment of the juxtanuclear Golgi ribbon (). In the same cells, staining of the COPII component Sec13 was restricted to ER exit sites before treatment, absent during treatment, and reestablished at exit sites during washout (). Significantly, quantification (see Materials and methods) revealed a reproducible peak of COPII assembly (). The transient up-regulation of COPII assembly was threefold (P = 0.006) greater than steady-state levels and coincided with the rapid phase of giantin emergence from the ER. HeLa cells, which exhibit significantly slower Golgi assembly (), also yielded a peak in COPII assembly (twofold; P = 0.01), and this coincided in time with the delayed emergence of giantin from the ER, thereby confirming COPII up-regulation and its correlation with Golgi exit from the ER. Further, single HeLa cells expressing Sec13-GFP () and imaged live at consecutive 2-min intervals while undergoing Golgi assembly during BFA/H89 washout also exhibited transient up-regulation in COPII assembly (). In contrast, untreated control cells exhibited relatively stable levels of COPII assembly (). In sum, a transient up-regulation in COPII assembly occurs during de novo biogenesis of the Golgi from the ER and coincides with Golgi egress marked by giantin. The values shown represent, on a per-cell basis, the total above-threshold Sec13 fluorescence, which is a composite of the number of exit sites per cell, their size, and their intensity. Analysis of these parameters indicated that the change in total fluorescence was due to both increased size of exit sites and change in number (). Thus, up-regulation during Golgi biogenesis involves up-regulated assembly at preexisting sites as well as the formation of new ER exit sites. ER-localized Golgi proteins may activate COPII recruitment at new and preexisting sites. As a test, we used BFA treatment to induce redistribution of Golgi enzymes to the ER. Unlike the combined BFA/H89 treatment, BFA alone, which inhibits Arf1 guanine nucleotide exchange factor (; ), does not prevent COPII assembly (; ; ). Indeed, in BFA-treated NRK cells, Sec13 remained localized to exit sites, whereas the Golgi enzyme mannosidase II was dispersed in the ER (unpublished data). Importantly, Sec13 staining increased to a new, higher steady state (twofold; P = 0.002) during BFA treatment () and, during washout, returned to its original steady-state levels (). The increase was due to both increased exit site size (1.7-fold; P = 0.02) and the formation of new sites (1.3-fold; P = 0.0008). BFA-induced twofold up-regulation of COPII assembly involving preexisting and new sites was also observed in single cells using live imaging of Sec13-GFP (Fig. S1, available at ). Presumably, Golgi proteins induce COPII up-regulation while undergoing ER exit that, in the presence of BFA, is sustained because of the immediate recycling of Golgi enzymes back to the ER (; ). These results suggest that the Golgi protein expression level might influence COPII assembly. Therefore, the level of Sec13 at exit sites was determined in a HeLa cell line stably overexpressing the Golgi enzyme -acetylgalactosaminyl transferase-2 (GalNAcT2) coupled to GFP (T2-GFP). In overexpressing cells, T2-GFP was readily detected in both the Golgi and the ER (see ). The cell line also contained cells expressing little or no T2-GFP, and these cells provided an internal control sample. Strikingly, relative to these control cells, overexpressors exhibited increased levels (1.7-fold; P = 0.006) of Sec13 staining at exit sites (). Altogether, these results indicate that increased Golgi protein abundance in the ER triggers increased COPII assembly through new exit site formation and that, if sustained, even for a single Golgi protein, a new steady-state level is established. To address the mechanism of stimulated COPII recruitment by ER-localized Golgi proteins, we used a permeabilized cell assay in which COPII assembly is performed on salt-washed cells in the presence of nonhydrolysable GTP (). One possibility is that ER-localized Golgi proteins, either directly or indirectly, inhibit GTP hydrolysis by the Sar1 GTPase. For the COPI coat, such a mechanism might stabilize the coat on the membrane, thereby ensuring productive, i.e., cargo-loaded, vesicle formation (). If enhanced COPII recruitment by ER-localized Golgi components depends on inhibition of GTP hydrolysis, then nonhydrolysable GTP should negate the difference in COPII assembly between control and BFA-treated or T2-GFP–overexpressing cells. However, despite the presence of GTPγS, BFA-treated cells exhibited twofold-increased COPII recruitment, as detected by Sec13 staining at exit sites () and by Sec13 immunoblotting (). As expected, the assay itself was cytosol dependent, and enhanced Sec13 recovery was evident at all cytosol concentrations tested (). A cytosol-dependent and statistically significant increase in COPII recruitment was also observed in the presence of GTPγS in permeabilized cells overexpressing T2-GFP (see ; unpublished data). Assembly increase in the absence of GTP hydrolysis argues against a mechanism involving inhibition of Sar1-GTP hydrolysis. This conclusion is consistent with a direct measurement showing a slight elevation, rather than inhibition, of GTP hydrolysis by Sar1p in the presence of the yeast cargo receptor Emp47p (). To determine whether COPII components alone are sufficient for up-regulated COPII assembly, the permeabilized cell assay was performed using purified COPII in place of cytosol. Yeast COPII was chosen as a convenient source that has previously been shown to recapitulate aspects of assembly at exit sites in NRK cells (). Consistent with previous work, purified COPII yielded assembly both in the juxtanuclear region and at peripheral exit sites, as detected by anti-Sec24p antibody staining after addition of Sar1p, Sec23p–Sec24p, and Sec13p–Sec31p (). Significantly, the signal was enhanced in BFA-treated cells relative to controls, indicating that purified COPII is sufficient to yield up-regulated COPII assembly (). Furthermore, when assembly and detection using anti-Sec24p antibodies were performed under identical conditions, except that Sec13p–Sec31p was omitted, up-regulated assembly was again observed (). Because Sar1p, Sec23p, and Sec24p form a prebudding complex in the absence of Sec13p and Sec31p (; ), these results indicate that increased COPII assembly occurs at the level of the prebudding complex. Thus, the mechanism of increased COPII assembly stimulated by ER-localized Golgi proteins occurs in the absence of any additional cytoplasmic factors or GTP hydrolysis and requires only the inner COPII components. Based on these results, we considered the possibility that Sar1p alone might show increased binding. Indeed, although Sar1p yielded a diffuse ER localization rather than accumulation at exit sites (unpublished data), its membrane association was significantly increased in BFA-treated cells (). This suggests that Golgi components stimulate increased COPII assembly by promoting Sar1p binding possibly via direct interactions, as several Golgi proteins contain cytoplasmic domain dibasic motifs that interact with Sar1p and mediate their ER exit (). As a test, we first determined whether the GalNAcT2 cytoplasmic domain, which contains a cluster of basic residues, binds purified Sar1p. An immobilized peptide corresponding to the GalNAcT2 cytoplasmic domain yielded robust and specific binding to Sar1p, whereas alanine substitution of the basic residues blocked binding (). Binding occurred in the absence of added GTP but was enhanced by the presence of GTPγS. Thus, GalNAcT2 is capable of directly binding Sar1p, and it may stabilize Sar1p on the ER membrane, leading to enhanced COPII assembly. Next, we asked whether presence of the GalNAcT2 cytoplasmic domain peptide would inhibit up-regulation of COPII assembly in cells overexpressing T2-GFP. Increasing concentrations of the alanine-substituted control peptide and GalNAcT2 cytoplasmic domain peptide were added together with cytosol and GTPγS to salt-washed permeabilized cells, and T2-GFP and Sec13 levels were determined (images presented in Fig. S2, available at ). As expected, cells overexpressing T2-GFP yielded increased COPII assembly compared with adjacent low expressors and the alanine-substituted control peptide (R→A) exerted little or no effect (). In contrast, the GalNAcT2 peptide potently inhibited the COPII assembly up-regulation triggered by T2-GFP overexpression (). A moderate inhibition of basal COPII assembly was also observed at higher peptide concentrations, indicating the involvement of the Sar1p dibasic binding site in COPII recruitment generally. Analysis of the area and number of Sec13-labeled exit sites per cell indicated that inhibition occurred at both up-regulated, preexisting sites and at newly formed exit sites (Fig. S3). In conclusion, increased T2-GFP in the ER stimulates new COPII assembly via direct binding to Sar1p. This offers an important variation on recent work showing stabilization of Sec23p–Sec24p membrane contact by cargo in the presence of multiple rounds of GTP hydrolysis by Sar1p (; ). Further, these results reveal an elegant solution to the problem of transiently accelerated Golgi growth in the biogenesis assay. Increased availability of Golgi proteins in the ER stimulates COPII assembly via direct interaction with the coat, thereby favoring membrane input to the Golgi and Golgi growth. As ER levels of Golgi proteins subside, so does the rate of growth. Such a mechanism (hereafter referred to as variable coat assembly) may participate generally in Golgi homeostasis. That is, Golgi size may be determined by Golgi resident proteins regulating coat assembly at sites, such as ER, that directly impact the net Golgi input/output ratio. To test this, the size of the Golgi apparatus was quantified in cells exhibiting COPII up-regulation because of T2-GFP overexpression and compared with cells in the control sample. To estimate Golgi size, 3D confocal image sets were used to determine the volume of the staining corresponding to the Golgi marker giantin, as opposed to its staining intensity, and this value was normalized using total cell volume. Volume normalization reduced the variation due to normal Golgi growth and did not alter the experimental outcome. Strikingly, T2-GFP overexpressors exhibited up-regulated COPII assembly and yielded a mean normalized Golgi size that was 1.3-fold higher than control cells (). Note that giantin expression level, as determined by immunoblotting, was not altered by T2-GFP overexpression and that an identical apparent Golgi size increase was observed when the same analysis was performed using GM130 staining rather than giantin to determine Golgi size (1.3 ± 0.1 times greater in T2-GFP overexpressors). Further, the effect was not restricted to T2-GFP expression because both COPII assembly and Golgi size were also up-regulated in stable cells overexpressing GPP130-GFP (Fig. S4 A, available at ), a Golgi protein that, similar to GalNAcT2, contains cytoplasmic dibasic motifs (). These findings, which report normalized means derived from multiple experiments, were further supported when the data was analyzed on a cell-by-cell basis for single experiments using a correlation matrix. As shown for T2-GFP overexpression, when the Sec13 level at exit sites was plotted against the ratio of Golgi/cell size, it was clear that increased COPII was associated with increased Golgi size (). As controls for this experiment, Golgi size was determined upon overexpression of non-Golgi residents. First, we analyzed a stably transfected HeLa cell line expressing ts045-VSVG tagged with GFP (VSVG-GFP) at 37°C, a temperature that allows its trafficking to the cell surface. VSVG binds the COPII coat through a diacidic motif (; ), and synchronized movement of VSVG out of ER increases COPII vesicle production (). Consistent with this, compared with cells in a control sample, we observed a twofold increase in Sec13 levels at exit sites in cells expressing VSVG-GFP, yet there was no significant difference in Golgi size induced by VSVG-GFP expression (). Using GM130 to measure Golgi size yielded identical results. Analysis of the data from single experiments on a cell-by-cell basis also supported these findings. In contrast to the case for T2-GFP, the COPII up-regulation induced by VSVG expression correlated poorly, if at all, with Golgi size (). Even under conditions of high-level VSVG-GFP expression induced by transient transfection (1.5-fold higher than T2-GFP based on total fluorescence per cell), there was no significant change in apparent Golgi size (1.0 ± 0.1–fold). If the COPII assembly increase induced by VSVG-GFP corresponded to increased input to the Golgi, as would be expected, then VSVG-GFP must have also induced a compensatory increase in output to account for the lack of change in Golgi size. Indeed, a transient Golgi size increase was previously observed coincident with a wave of VSVG passing through the Golgi (; ), and we also observed increased Golgi size if we analyzed cells 20 min after releasing ER-accumulated VSVG-GFP using temperature shift (unpublished data). As further controls, we analyzed cells expressing the ER-localized protein Sec61-GFP () and the secreted protein albumin. Comparison of expressing to nonexpressing cells indicated that Sec61-GFP expression (at a level equal to T2-GFP based on total fluorescence per cell) altered neither COPII assembly nor Golgi size (Fig. S4 B) and that albumin expression modestly elevated COPII assembly but did not alter Golgi size (Fig. S4 C). In sum, these results suggest that COPII assembly can be induced by proteins rapidly exiting the ER and that, in the case of Golgi residents, this can lead to sustained changes in Golgi size. If T2-GFP influences Golgi size via its interaction with the COPII coat, then interfering with this interaction should alter both COPII assembly and Golgi size. Consistent with this, it was shown in in permeabilized cells that the GalNAcT2 cytoplasmic domain peptide, which binds Sar1p, blocks the up-regulation of COPII assembly induced by T2-GFP expression. Therefore, we microinjected this peptide or the alanine-substituted control peptide into nonexpressing and T2-GFP–expressing cells and, after 45 min, determined COPII assembly levels and Golgi size in cells marked by coinjected fluorescent dextran (). Microinjection itself led to an unexplained minimal increase in dissociated Golgi elements, but it did not alter steady-state changes in COPII assembly or Golgi size. That is, similar to noninjected cells, T2-GFP–overexpressing cells injected with the control peptide exhibited significant increases in both COPII assembly and Golgi size (, R→A). In contrast, the GalNAcT2 peptide blocked both the COPII assembly increase and the Golgi size increase (, wt) strongly suggesting that the Golgi size increase evident in T2-GFP–expressing cells is a direct consequence of the COPII–Golgi protein interaction. COPII assembly was regulated by the availability of Golgi proteins in the ER that was due to, in the case of T2-GFP, a direct interaction between the Golgi protein cytoplasmic domain and Sar1p. Transient availability of Golgi proteins in the ER triggered transient COPII up-regulation, explaining the burst of Golgi growth that accompanies Golgi biogenesis from the ER. Further, when sustained, increased availability of Golgi proteins in the ER established a new, higher steady-state level of COPII assembly, and this accompanied a stable increase in Golgi size. Thus, our work shows for the first time that variable coat assembly regulated by compartment residents significantly impacts organelle homeostasis. The binding of cargo to assembling coats is widely recognized to underlie the local enrichment of cargo at bud sites and, thus, sorting (; ). Less well recognized is that the capacity of cargo to bind coats influences coat assembly by increasing avidity of the coat for the membrane (; ; ; ). GDP bound Sar1p interacted with GalNAcT2, suggesting that, as shown in the model (), increased GalNAcT2 in the ER membrane recruited more Sar1p-GDP and, after Sec12p-mediated GTP exchange, this led to increased formation of Sec23–Sec24 prebudding complexes. Whereas Sar1 is diffusely localized on the ER membrane, assembled COPII is not. A simple explanation is that lateral interactions between coat components accounts for the concentration of COPII into new and larger sites. That is, up-regulated Sar1 recruitment increases COPII components on the ER membrane, which in turn form new and larger clusters. It was very recently shown that secretory cargo influences formation of ER-to-Golgi tubular carriers (). It is unclear whether this is mechanistically related to ER exit site formation, but it may be that cargo-induced new and/or larger ER exit sites give rise to tubular carriers. Cargo binding to the GDP form of Sar1p may be generally involved because, in addition to the dibasic-containing Golgi enzymes () and VSVG (), at least two SNAREs, Bet1p and Bos1p, also bind Sar1p-GDP (). Because all these proteins also bind Sar1p-GTP, an additional effect may be stabilization of Sar1p-GTP on the membrane. However, based on the failure of inhibition of GTP hydrolysis by Sar1p to induce COPII assembly up-regulation, Sar1p-GTP stabilization is not sufficient. Further, the best described cargo–COPII interactions involve binding sites on the Sec24p subunit (; ), and SNARE binding at these sites stabilizes membrane binding of Sec23–Sec24p complex during multiple rounds of Sar1p-GTP hydrolysis (). Thus, cargo–coat interactions exert multilevel control of coat assembly. Variable coat assembly regulated by compartment residents could profoundly impact organelle homeostasis (). Steady-state Golgi growth reflects the balance of all input and output reactions. Input reactions could be most sensitive to Golgi protein expression level, whereas output reactions might be most sensitive to the level of non-Golgi residents, such as proteins that rapidly recycle to the ER and newly synthesized proteins headed for distal compartments, including outside the cell, the cell surface, and lysosomes. If so, increased Golgi protein synthesis would cause increased Golgi growth because of increased input from the ER with a less dramatic change in exit from the Golgi. Because of the necessary presence of targeting and fusion factors accompanying the increased input from the ER, there is likely an increase in Golgi-to-ER recycling. An explanation for why this does not offset the size increase is that recycling involves less material and therefore less membrane. In contrast to Golgi protein synthesis, increased synthesis of plasma membrane proteins would cause a transient increase in Golgi size, as has been observed (; ), but little lasting change in Golgi growth. That is, any increase in input to the Golgi from the ER would shortly thereafter be offset by a corresponding increase in exit from the Golgi, possibly because of cargo–coat interactions at the TGN (). Thus, homeostasis of endomembrane compartments might be partly determined by the level of compartment residents because of their influence on coat assembly and vesicle trafficking kinetics. Paralleling the sensitivity of COPII assembly to VSVG levels, assembly of the clathrin–AP2 complex is sensitive to transferrin receptor and mannose-6 phosphate receptor levels (; ; ). Future work may reveal that the cargo sensitivity of all coats extends to the level of compartment residents and that the sum of the consequent interactions establishes compartment size. Interestingly, other complex biological structures, such as the yeast spindle pole body, show size regulation via expression level (; ). Variable coat assembly provides a straightforward way in which cells can respond to dramatic, reversible changes in the integrity of an organelle, such as the Golgi apparatus, and restore its original state. Conceivably, variable coat assembly also drives sustained changes in organelle size, such as those that occur during differentiation, by responding to coordinated changes in expression of organelle residents. NRK cells, HepG2 cells, HeLa cells, or HeLa cell lines stably expressing the GFP-tagged proteins GalNacT2 (T2-GFP), ts045-VSVG (VSVG-GFP), or GPP130 (GPP130-GFP) were maintained as described previously (). BFA (Sigma-Aldrich) and H89 (Toronto Research) were used in media at 2.5 μg/ml and 100 μM, respectively. For transient transfections, HeLa cells were transfected with plasmids encoding GFP-tagged Sec13 () using calcium phosphate or VSVG- or Sec61-GFP using Transfectol (GeneChoice). Immunofluorescence was performed as described previously (). The antibodies and their dilutions were mouse anti-giantin at 1:100 (); rabbit anti-Sec13 at 1:500 (); rabbit anti-Sec24p at 1:500 (provided by T.H. Lee, Carnegie Mellon University, Pittsburgh, PA); mouse anti–mannosidase II at 1:10,000 (Covance); rabbit anti-GM130 at 1:500 (); FITC-conjugated anti–albumin at 1:100 (Biotrend); Cy5-labeled goat anti–mouse or anti–rabbit at 1:500 (Zymed Laboratories); and rhodamine-labeled goat anti–rabbit at 1:500 (Zymed Laboratories). Microscopy was performed using a spinning disk confocal scan head equipped with three-line laser and independent excitation and emission filter wheels (PerkinElmer) and a 12-bit digital camera (Orca ER; Hamamatsu) mounted on a microscope (Axiovert 200; Carl Zeiss MicroImaging, Inc.) with a 100×, 1.4 NA apochromat oil-immersion objective (Carl Zeiss MicroImaging, Inc.). Single optical sections or sections at 0.3-μm spacing were acquired using ImagingSuite software (PerkinElmer). Individual experiments were performed with identical laser output levels, exposure times, and scaling. At least six representative fields, each containing 3–5 cells, were taken. Total fluorescence of Sec13 at peripheral sites was quantified using ImageJ on single optical sections as follows. For each experiment, a single fixed threshold was manually chosen (based on comparison to the original gray-scale images) and applied to all images. Individual cells were then selected with the free-hand tool, and the total above-threshold fluorescence was determined using either the analyze particles or measure functions. For quantification of Golgi size, the analyze particles function was used after thresholding to yield, on a per-cell basis, the sum of each section's Golgi area as determined by giantin or GM130 staining. Cell volume was estimated by summing the area in each section outlined manually based on the diffuse background staining of Sec13. To allow direct comparison of distinct experiments given small changes in staining intensity, COPII assembly and Golgi size values were normalized by dividing by the mean values of the entire dataset for a given experiment. To compare expression level of GFP constructs, single optical sections were acquired using identical settings and the 12-bit images were background subtracted using a fixed value. Means of the total fluorescence per cell were compared (>10 cells each). Live imaging was performed 2 d after transfection in Optimem (Invitrogen) containing 10% fetal bovine serum on a 37°C stage using 300-ms exposures every 2 min. For each time point, fluorescence intensity in objects was determined as just described. These values were then adjusted to correct for the slight rate of photobleaching based on the photobleaching rate of an identically captured movie for a control cell outside of the dataset. The values for a given cell were then normalized by dividing each by the mean value of the last five time points for that cell. This allowed direct comparison of distinct cells. Neither the correction nor the normalization altered the pattern of fluorescence changes. The morphological permeabilized cell COPII assembly assay was performed as described previously (). NRK cells at 70% confluence on 12-mm glass coverslips were treated with BFA for 30 min, washed 3 × 0.5 ml with cold DME, and washed 2 × 0.5 ml with cold KOAc buffer (115 mM KOAc, 2.5 mM MgOAc, 25 mM Hepes, pH 7.2, and 1 mM dithiothreitol). The washed cells were permeabilized 6 min at RT in 0.5 ml 0.03 mg/ml digitonin in KOAc buffer followed by 3 × 0.5 ml washes in cold KOAc buffer. After transfer to parafilm, the coverslips were incubated at 37°C for 10 min in 50 μl KOAc buffer containing either rabbit liver cytosol, at the indicated concentrations, or purified yeast Sar1p (1 μg/ml), Sec23p–Sec24p (3 μg/ml), and Sec13p–Sec31p (7 μg/ml). Where indicated, each reaction also contained 500 μM GTPγS and an ATP regeneration system (0.5 mM ATP, 0.5 mM UTP, 50 μM GTP, 5 mM creatine phosphate, 25 μg/ml creatine phosphokinase, 0.05 mM EGTA, and 0.5 mM MgCl). The coverslips were then washed in cold KOAc buffer and fixed and stained as described. The morphological assay was slightly modified for the peptide inhibition experiments. T2-GFP cells were used instead of NRK and, before use, the cytosol was preincubated at 4°C 1 h in 20-μl reactions containing 0.2 mg cytosol, 500 μM GTPγS, the ATP regeneration system, 1 mM PMSF, and synthetic peptides (see Peptide binding) at the indicated concentrations. The immunoblot-based assay was essentially the same except that cells were grown in 35-mm dishes and volumes were scaled accordingly. Instead of fixation, the washed cells were scraped into lysis buffer (1% TX-100, 2 mM EDTA, 50 mM Tris, pH 8.0, 1 mM PMSF, and 150 mM NaCl) and incubated on ice for 10 min with vortexing followed by centrifugation at 14,000 rpm for 5 min. The cleared lysate was precipitated using trichloroacetic acid and analyzed by immunoblotting () using enhanced chemiluminescence (Pierce Chemical Co.) with acquisition by the LAS-3000 imaging system (Fujifilm). Antibodies used were rabbit anti-Sec13 at 1:1,000 (), rabbit anti-GPP130 at 1:1,000 (), and peroxidase-conjugated anti–rabbit at 1:5,000 (Bio-Rad Laboratories). Synthetic peptides (MRRRSRC and MAAASAC) were purchased from GeneScript and coupled via the added C-terminal cysteine residue to Sepharose 6B beads according to previous work (; ) and manufacturer instructions (GE Healthcare) in reactions containing 100 μl 14 mg/ml peptide, 500 μl beads, and 600 μl coupling buffer (50 mM Tris, pH 7.3, and 0.5 M NaCl). Coupling efficiency was determined by the measuring 2-thiopyridone release at 343 nm. The beads (14-μl aliquots) were then blocked by a 40-min 4°C incubation in 0.5 ml 5 mM β-mercaptoethanol, 50 mM NaOAc, 0.5 M NaCl, pH 4.5, followed by washing and a 2-h 4°C incubation in binding buffer (20 mM Hepes, pH 7.2, 250 mM sorbitol, 70 mM KOAc, 1 mM Mg[OAc]2, and 1 mg/ml bovine serum albumin). For binding experiments, 0.5 μg Sar1p was preincubated for 30 min at 4°C in the presence or absence of 500 μM GTPγS in a total volume of 14 μl binding buffer. To this was added 16 μl buffer containing 7 μl beads containing 5 nmoles of peptide for 1 h at 4°C. The beads were washed 4 × 500 μl for 1 min each with binding buffer lacking albumin and analyzed for Sar1p recovery by immunoblot as described in the previous paragraph using rabbit anti-Sar1p antibody at 1:2,000. The MRRRSRC or MAAASAC peptides were microinjected into T2-GFP cells at 1 mM in water containing 0.25 mg/ml Alexa 568–conjugated dextran (Invitrogen) using a FemtoJet system with InjectMan-NI2 micromanipulator (Brinkman). The injected cells were maintained at 37°C in Optimem containing 10% serum for 45 min followed by analysis of COPII assembly using anti-Sec13 and Golgi size using anti-giantin as described in Image analysis. Fig. S1 shows up-regulation of COPII assembly at exit sites using representative frames and quantified data from live imaging experiments in which Sec13-GFP was visualized in control cells or upon BFA treatment. Fig. S2 and Fig. S3 show the specific inhibition of T2-GFP–induced COPII up-regulation by addition of the GalNAcT2 cytoplasmic domain peptide to the permeabilized cell assay. Representative images are presented in Fig. S2; total quantified Sec13 fluorescence is presented in ; and quantification of number, size, and intensity of fluorescent objects (ER exit sites) is presented in Fig. S3. Fig. S4 shows that Golgi size is increased by overexpression of the Golgi protein GPP130 but not by overexpression of the ER protein Sec61 or the secreted protein albumin using representative images and quantified data. Online supplemental material is available at .

Spinocerebellar ataxia 7 (SCA7) is a progressive autosomal dominant neurodegenerative disorder characterized by cerebellar ataxia and visual impairment () that is due to moderate to severe neuronal loss in the cerebellum and associated structures () and degeneration of cone and rod photoreceptors. The SCA7 gene product, ataxin-7 (ATXN7), is a component of the TBP-free TAF-containing complex (TFTC) and the SPT3/TAF9/GCN5 acetyltransferase complex (STAGA), which are implicated in several steps of transcriptional regulation, such as histone acetylation and recruitment of the preinitiation complex to promoters (; ). SCA7 belongs to a group of nine inherited neurodegenerative disorders caused by an unstable CAG repeat expansion in gene coding regions, leading to the elongation of a polyglutamine (polyQ) tract in the respective proteins (). PolyQ expansions confer toxic properties on mutant proteins, which accumulate aberrantly in neurons, leading to the formation of insoluble nuclear inclusions (NIs), a hallmark of polyQ diseases. Analysis of mouse models of polyQ disorders showed that accumulation in neuronal nuclei of proteins with expanded polyQ is an important step in pathogenesis (; ; ). In SCA7 transgenic and knockin mouse models, progressive retinal degeneration correlates with nuclear accumulation of mutant ATXN7 and altered transcription of photoreceptor genes (; ; ). Several mechanisms have been proposed to underlie polyQ toxicity in the nucleus (), including sequestration into NIs of nuclear proteins, such as transcription factors, nuclear body components, constituents of the ubiquitin-proteasome system (UPS) and chaperones, which might impair their functions. In the case of SCA7, it was recently shown that mutant ATXN7 alters the functions of the TFTC and STAGA complexes (; ; ). Accumulation of polyQ-expanded proteins in the nucleus may be due to a defect in protein folding, turnover, or degradation. As neurons are postmitotic and long-lived cells, failure to prevent the accumulation of toxic proteins may compromise their survival. Accordingly, molecules that prevent nuclear accumulation or increase the clearance of misfolded polyQ proteins were protective against polyQ toxicity in mouse models (; ; ). An understanding of the mechanisms whereby mutant proteins accumulate, aggregate, or are eliminated in the nucleus would help in the development of therapeutic strategies for these diseases. Previous studies showed that promyelocytic leukemia protein (PML) nuclear bodies colocalized with polyQ-containing proteins in nuclear matrix preparations of cells expressing ATXN7 or ataxin-1 (; ). Furthermore, the normal nuclear distribution of PML was altered by the expression of mutant ataxin-1 and -3 (; ). In the brains of patients with SCA7 or other polyQ disorders, PML bodies often colocalized with neuronal NIs (). Interestingly, colocalization of PML bodies occurred more frequently in small than in large NIs (), suggesting that PML bodies are associated with early steps of polyQ protein aggregation. PML bodies are multiprotein complexes distributed in a speckled pattern throughout the nucleus (; ; ), where they have been suggested to play a role in many cellular processes, such as transcriptional regulation, growth control, and apoptosis (; ). The heterogeneous composition and morphology of PML bodies has led to the suggestion that the different PML isoforms have distinct cellular functions (). Because of the variety of partner proteins found in PML bodies, it has also been proposed that PML bodies might be storage compartments for nuclear proteins (). However, clastosomes, a subset of PML bodies, contain components of the UPS and were suggested to be sites of protein degradation in the nucleus (). Clastosomes appear to be transient structures that assemble when proteolysis is highly active in the nucleus but disappear when the proteasome is inhibited (). Interestingly, showed that some PML bodies move about rapidly in the nucleus in an energy-dependent fashion and suggested that these dynamic PML bodies may be nuclear sensors of foreign or misfolded proteins, including those of viral origin, and would thus act like a subnuclear immune system. This is consistent with the observation that interferon (INF) induces PML expression and stimulates the formation of PML bodies (). In this study, we aimed at understanding the role of PML bodies in the fate of polyQ-expanded ATXN7 in the nucleus. Modifying the level and composition of PML bodies dramatically affects the aggregation of mutant ATXN7. We show that a specific isoform of PML, PML IV, orchestrates the recruitment of components of the UPS, chaperones, and mutant ATXN7 (and other expanded polyQ proteins) in specialized nuclear bodies that resemble clastosomes to promote their degradation. The capacity of the nucleus to degrade proteins can be modulated to prevent the toxic accumulation of expanded polyQ proteins. We previously showed that PML bodies colocalized with a subset of polyQ aggregates in the brains of patients with SCA7 (). To determine whether this was also the case in models of SCA7, we examined the brains of transgenic mice B7E2.B, which express mutant ATXN7 with 128 glutamines ubiquitously in the central nervous system and develop progressive ataxia (). Mutant ATXN7 accumulated in most neurons, including the cerebellar Purkinje cells () and retinal neurons (), leading to NI formation. A subset of large ATXN7 aggregates in Purkinje cells () and retinal ganglion neurons () contained PML bodies. Ganglion neurons also had small ATXN7 aggregates, which were either sparsely distributed in the nucleus or juxtaposed to large aggregates. PML bodies colocalized with small aggregates, in some cases forming a ring around the aggregate (, and d). Some PML bodies were juxtaposed to both small and large aggregates (). Colocalization of PML bodies with ATXN7 aggregates, notably small ones, is consistent with our previous study of SCA7 brains and raises the question of how PML bodies affect the aggregation of mutant ATXN7. To investigate the relationship between PML bodies and mutant ATXN7, we coexpressed full-length wild-type and mutant ATXN7 with isoforms of PML in neurons and in COS-7 cells. There are seven isoforms of PML, resulting from differential splicing, that form nuclear bodies with different morphologies (; ). We have examined the effect of exogenous expression of PML IV, which was previously shown to induce cellular senescence, apoptosis, and regulate p53 activity (; ) and may thus be related to neuronal dysfunction or death. The effects of PML III were examined to control for specificity. Endogenous ATXN7 was distributed homogeneously in a granular pattern throughout the nucleus and did not colocalize with the small endogenous nuclear bodies observed in neurons (). When PML IV was overexpressed in transfected neurons, both endogenous ATXN7 () and full-length wild-type (FL-10Q) exogenous ATXN7 () colocalized with PML IV in larger nuclear bodies. In COS-7 cells, mutant ATXN7 with 74Q (FL-74Q) was also distributed homogeneously in the nucleoplasm of the majority of transfected cells when expressed alone (unpublished data). In 10–20% of the cells, however, multiple dense nuclear aggregates were seen that contained endogenous PML bodies in their center (), as well as the Daxx protein, a resident PML body protein (not depicted). Exogenous PML III formed rod-shaped bodies and exogenous PML IV larger patchy structures, some of which had a ring-like shape (unpublished data). When PML III was coexpressed with FL-74Q, both rod-shaped PML III bodies and ATXN7 aggregates were observed, but they rarely colocalized (). In contrast, PML IV and mutant ATXN7 colocalized in large, rounder nuclear structures () that differed from the dense ATXN7 aggregates but resembled the PML IV bodies in that some had a ring-like shape (, h, arrowheads). In primary cultures of rat cortical neurons, PML IV also formed round nuclear bodies, but they were smaller and morphologically more homogenous than in COS-7 cells (). When mutant ATXN7 and PML IV were coexpressed in these neurons (), mutant ATXN7 was localized in the PML IV bodies. These data suggest that PML IV bodies recruit both endogenous ATXN7 and exogenous normal and mutant ATNX7. To determine whether PML IV recruits ATXN7 to the nuclear bodies, we performed video-recorded time-lapse experiments on living HeLa Kyoto cells expressing full-length ATXN7 with 100Q fused to EGFP (EGFP-FL-100Q) and PML IV fused to RFP. The cells were monitored from 6 to 22 h after transfection (i.e., for 16 h). illustrates the development of ATXN7-containing PML bodies over a period of 6 h and 40 min. In cells expressing EGFP-FL-100Q alone, the intensity of the green fluorescence increased over time up to 22 h after transfection but remained homogeneously distributed in the nucleoplasm (unpublished data), indicating that endogenous PML bodies do not form ATXN7-positive bodies. In the presence of RFP-PML IV (red), EGFP-FL-100Q fluorescence remained weak in the nucleoplasm but increased in PML IV bodies from the moment they formed (), indicating that the soluble form of mutant ATXN7 was recruited to the PML IV bodies. Small ATXN7-containing PML IV bodies then moved and coalesced into larger bodies. PML IV, therefore, recruits soluble mutant ATXN7 to nascent PML IV bodies, preventing its accumulation in the nucleus and formation of focal aggregates. Coalescence of small bodies into larger ones was also observed in cells expressing RFP-PML IV alone (unpublished data). To determine whether recruitment of ATXN7 to PML IV bodies was due to a physical interaction with PML IV in cell nuclei, we performed coimmunoprecipitation experiments. We first verified that two different anti-PML antibodies (monoclonal and polyclonal) immunoprecipitated PML IV (, lanes 2 and 3) and that the anti-ATXN7 monoclonal antibody 1C1 immunoprecipitated FL-10Q (, lane 2) and FL-74Q (, lane 2). The control antibodies (anti-myc and anti-Xpress) did not immunoprecipitate either protein (, A [lanes 4, 5, and 7] and B [lanes 3, 6, and 7]). When HA-tagged FL-74Q and PML IV were coexpressed and ATXN7 was immunoprecipitated with 1C1, PML IV coprecipitated in the complex (, lane 6). When PML IV was immunoprecipitated, HA-tagged FL-74Q coprecipitated in the complex (, lanes 4 and 5). FL-10Q and PML IV also coimmunoprecipitated (, lane 3). Therefore, not only mutant full-length ATXN7 but also its wild-type counterpart physically interact, directly or indirectly, with PML IV in nuclear bodies. We then determined the effect of PML IV on the biochemical properties of mutant ATXN7. When FL-74Q alone was expressed in COS-7 cells, both soluble and aggregated ATXN7 were detected. The soluble form of mutant ATXN7 appeared as a 150-kD protein on Western blots (, top). The aggregated form of the protein was retained on cellulose acetate filters in the filter retardation assay (, bottom), which evaluates the overall amount of SDS-insoluble amyloid fibers formed by the aggregation of proteins with expanded polyQ (). Aggregated ATXN7 was found in the pellet when cell homogenates were centrifuged but not in the supernatant. Coexpression with PML III very slightly increased the amounts of soluble and aggregated ATXN7 (, top and bottom). In contrast, coexpression with PML IV led to a strong decrease in the amounts of soluble and aggregated ATXN7 (, top and bottom). The supernatant obtained after centrifugation of cell homogenates was also devoid of SDS-insoluble ATXN7. The difference between the effects of PML III and IV was not due to different levels of expression of the isoforms (, top) or to differential effects on the transcription of the ATXN7 gene (). This suggests that PML IV bodies act directly on the ATXN7 protein by increasing its degradation. We then examined whether PML IV–dependent degradation of ATXN7 depends on the length of the polyQ tract. When expressed with PML IV, the level of wild-type FL-10Q also decreased (), whereas its level slightly increased when expressed in the presence of PML III. Degradation therefore does not depend on the presence of an expanded polyQ tract. To determine whether PML IV bodies also degrade other polyQ-containing proteins, we coexpressed PML IV with huntingtin (Htt)-exon1 harboring 125Q. PML IV strongly reduced the level of soluble mutant Htt-exon1 (). We also verified whether PML IV would degrade an unrelated nuclear protein, fragile X mental retardation protein (FMRP; ) without a polyQ sequence. The level of expression of exogenous FMRP was unaffected by coexpression with PML IV (). Intriguingly, a second faint band of lower molecular mass appeared to be decreased by PML IV. Its identity is unknown. To determine whether PML IV increased the rate of degradation of mutant ATXN7, we performed pulse-chase experiments in COS-7 cells in which mutant ATXN7 was expressed either alone or in combination with PML IV. The degradation of mutant ATXN7 expressed alone was relatively slow, in good agreement with a previously published pulse-chase experiment (). During the first 8 h of the chase period, only 28% of ATXN7 was degraded (). In the presence of PML IV, however, 83% of mutant ATXN7 was degraded after 8 h (). The difference between the levels of ATXN7 in the presence and in the absence of PML IV was significant at all time points, starting at 2 h, indicating that PML IV bodies accelerate the degradation of mutant ATXN7. To see whether the degradation of mutant ATXN7 in the presence of PML IV was mediated by proteasomes, COS-7 cells expressing FL-74Q in the presence and in the absence of PML IV were treated with 5 μM of the proteasome inhibitor MG132. Inhibition of the proteasome increased the amount of mutant ATXN7 in the cells coexpressing FL-74Q and PML IV (, compare third and fourth lanes). However, because PML IV was already active before MG132 treatment, it was not possible to accumulate ATXN7 to the level observed without PML. MG132 also increased the level of mutant ATXN7 expressed in the absence of PML IV by impeding normal turnover of the protein (, compare first and second lanes). We thus conclude that the effect of PML IV on the degradation of ATXN7 was mediated by the UPS. Under certain stress conditions, PML forms clastosomes that contain components of the UPS (). To determine whether PML IV bodies are clastosomes and to understand how they may clear ATXN7, they were analyzed for the presence of UPS components. S10a, a subunit of the 19S non-ATPase regulatory complex of the proteasome was not found in focal aggregates formed by mutant ATXN7 () but was highly colocalized with PML IV bodies when only PML IV was expressed (). Interestingly, when coexpressed with PML IV, mutant ATXN7 colocalized perfectly with the S10a subunit in PML IV bodies (). Similar observations were made with two other 19S proteasome subunits, S5a and S6b (Fig. S1, A and B, available at ); the latter was already detected in endogenous clastosomes (). The 20S catalytic core of the proteasome colocalized only with a subset of aggregates in cells expressing FL-74Q alone (), but colocalization was complete with PML IV bodies containing the mutant protein (), as well as with PML IV expressed alone (). This was also observed for subunit α2 of the 20S proteasome (Fig. S1 C). The same was true for the chaperone Hsp40, which recognizes misfolded polyQ proteins (). We next verified whether ATXN7 was ubiquitinylated in PML IV bodies. Nontransfected cells possessed only a few PML bodies labeled with an antibody against polyubiquitin that were possibly endogenous clastosomes (, arrowheads indicate colocalization). Aggregates of ATXN7, when expressed alone (), were only slightly labeled with the anti-polyubiquitin antibody, but ATXN7-containing PML IV bodies were strongly labeled (). Together, these data indicate that PML IV induces the assembly of a nuclear structure dedicated to protein degradation, reminiscent of clastosomes. They recruit both chaperones and proteasomes together with its substrates, such as mutant ATXN7, targeted for degradation by polyubiquitin. We investigated the nature of mutant ATXN7 in PML IV bodies by performing an ultrastructural analysis of transfected COS-7 cells. Because truncated mutant ATXN7 aggregates more rapidly than the full-length protein (unpublished data), we expressed Tr-100Q-EGFP (the first 232 amino acids of the protein containing the polyQ stretch fused to EGFP) in the presence or absence of PML IV. We examined ATXN7 distribution by electron microscopy after immunolabeling with a monoclonal anti-ATXN7 antibody and a secondary antibody coupled with 10-nm gold particles. Tr-100Q-EGFP, when expressed alone, was restricted to pale-stained fibrillary structures in the nucleus with fibers radiating from the edges of the aggregates (). When coexpressed with PML IV, Tr-100Q-EGFP colocalized with PML IV (colabeling with species-specific secondary antibodies coupled with 15-nm gold particles to detect PML) in round or oval structures, which are most likely PML IV bodies (, C and D; arrows [ATXN7] and arrowheads [PML] indicate colocalized gold particles). Importantly, no fibrillary structures were evident in this case (). This was confirmed by the absence of ATXN7 aggregates on the filter assay (). ATXN7 in PML IV bodies was therefore no longer fibrillary in structure and can no longer be considered to be aggregated. Soluble Tr-100Q-EGFP levels analyzed on Western blot were slightly decreased by PML IV expression (). To determine whether altering the amount of endogenous PML bodies affects the degradation or aggregation of mutant ATXN7, we treated cells expressing FL-74Q with cadmium chloride, reported to disrupt PML bodies (). Immunofluorescence analysis showed that 2 μM cadmium totally disrupted the endogenous PML bodies. Consequently, no colocalization between mutant ATXN7 aggregates and the PML protein was observed (), compared with the untreated cells (). A treatment with increasing concentrations of cadmium (0.25–2 μM) showed that the aggregation of mutant ATXN7 was dose dependent, as shown on filter assay (), suggesting that endogenous clastosomes are indeed involved in mutant ATXN7 degradation. Because β-INF was reported to up-regulate the expression of PML (), we investigated its effect on the PML-dependent degradation of mutant ATXN7. β-INF increased the size and number of PML bodies in nontransfected COS-7 cells (, a and b). Faint endogenous PML immunoreactivity was observed in aggregates of mutant ATXN7 in untreated cells (, arrowheads). β-INF treatment led to the formation of large, strongly immunoreactive PML bodies that colocalized with mutant ATXN7 (). These results also suggest that the presence of mutant ATXN7 promoted the coalescence and fusion of PML bodies into enlarged bodies resembling PML IV bodies, which colocalized perfectly with mutant ATXN7 (, compare b and g). Semiquantitative RT-PCR performed with primers that specifically amplify PML IV showed that the expression of PML IV was induced in β-INF–treated cells but not in control cells (, lane 2). Western blots show that β-INF increased the levels of several PML isoforms in both untransfected COS-7 cells and cells overexpressing exogenous ATXN7 in its wild-type FL-10Q or mutant form FL-74Q (). β-INF treatment decreased the amount of overexpressed soluble FL-10Q () and FL-74Q () but had no effect on the level of endogenous wild-type ATXN7 (). The PML-dependent degradation of ATXN7 induced by β-INF was mediated by the proteasome, as it was inhibited by MG132 (). Interestingly, insoluble mutant ATXN7 disappeared from cells treated with β-INF, as shown by the filter assay, but an insoluble PML-containing material was detected (, left, lane 2). This material was independent of ATXN7 expression but dependent on β-INF treatment and attributable to the increased expression of PML IV isoform (, right, lanes 2 and 3). Accumulation of misfolded polyQ-expansion proteins in neuronal nuclei, ultimately leading to the formation of NIs, is a key step in the pathogenesis of SCA7 and other polyQ-expansion diseases (). We have focused on PML bodies, known to be associated with NIs (), which are suggested to be privileged sites of nuclear protein degradation (), and have investigated their role in either the degradation or the accumulation of polyQ-containing protein. We report that modifying the level and composition of PML nuclear bodies has profound effects on the fate of disease-causing polyQ-expansion proteins in the nucleus. Seven PML isoforms have been described that share an NH-terminal region comprising the RBCC (RING-finger, B-box, coiled-coil domain) but differ in their COOH-terminal region because of alternative splicing (). The role of each isoform is not yet known, but they might be associated with different functions of the nuclear bodies (; ; ). We have explored the effect of two isoforms, PML III and IV, on the fate of mutant ATXN7 in the nucleus. Although these isoforms differ only by their short COOH-terminal extensions (71 amino acids in PML III and 63 amino acids in PML IV), they form nuclear bodies that have strikingly different morphologies and protein composition. PML IV bodies are round and larger than PML III bodies, which are rod shaped. PML IV bodies were highly enriched in proteasome components. They contained subunits of the 19S complex, implicated in unfolding of the substrates, as well as subunits of the 20S catalytic core complex. PML IV bodies also colocalized with the chaperone Hsp40 and polyubiquitinylated proteins and thus resemble clastosomes, a subset of endogenous PML bodies enriched in UPS components, suggested to be sites where nuclear proteins are degraded (). Thus, endogenous PML IV appears to be a component of endogenous clastosomes. Whether they also contain other isoforms of PML remains to be determined. Interestingly, only overexpressed PML IV relocalized endogenous as well as exogenous wild-type and mutant ATXN7 to PML IV bodies. Consequently, because of the presence of UPS components, PML IV actively increased the degradation of the soluble form of mutant ATXN7, which led to the disappearance of SDS-insoluble aggregates. We further demonstrated that the PML IV–dependent degradation of mutant ATXN7 is mediated by the proteasome, showing for the first time that exogenously expressed PML IV clastosomes indeed degrade a substrate, in our case, mutant ATXN7. Time-lapse experiments showed that PML IV acts on the soluble form of mutant ATXN7 to recruit it to PML IV bodies. Correspondingly, soluble PML IV physically interacts with the soluble forms of mutant and wild-type ATXN7. This suggests that PML IV convoys the soluble forms of ATXN7 to the bodies, where they are degraded by the UPS components, thus preventing the deleterious accumulation of mutant ATXN7. How PML IV selects the substrates for degradation remains to be elucidated. Interestingly, ATXN7 does not interact with PML III (unpublished data) and, consequently, it is not recruited to PML III bodies. However, the immediate early gene X1 (IEX-1), a stress response gene involved in apoptosis, was shown to colocalize with both PML III and IV and to coimmunoprecipitate with PML III but not PML IV (). These authors noted on Western blots positive coregulation of PML III with IEX-1 but a negative coregulation of PML IV. IEX-1 might therefore be another nuclear substrate for PML IV clastosomes. An investigation of how ATXN7 interacts with PML IV, directly or indirectly, might provide a clue as to how PML IV selects its substrates. The SDS-insoluble structures formed by PML IV, consistent with the fact that PML can self-interact and oligomerize (; ; ), might serve as a scaffold to bring together both the actors of the UPS and the substrates to be degraded. When the level of PML IV is increased by exogenous expression, the composition of the PML bodies changes, leading to the formation of enlarged clastosomes. This increases the capacity of the nucleus to degrade specific substrates, such as toxic polyQ protein. Others have shown that proteasome-dependent protein degradation occurs in nucleoplasmic foci that partially overlap or juxtapose with splicing speckles and PML bodies (). In a very recent study, a novel GTPase CRAG (CRAM-associated GTPase) induced, under cellular stress, an interaction with PML that formed the enlarged ring structures typical of PML bodies (). Interestingly, it was shown that polyQ accumulation and oxydative stress triggered association of CRAG and polyQ in the cytoplasm before translocation to the nucleus. Once inside the nucleus, CRAG promoted the degradation of polyQ in PML bodies. It would be interesting to know whether CRAG was also implicated in the formation of PML IV bodies in our cell culture model, or whether activation of CRAG by experimental stress can convert endogenous PML bodies that do not degrade ATXN7 to degradation-competent bodies, in the absence of overexpression or β-INF treatment. Our study with cadmium showed, for the first time, that some endogenous PML bodies, presumably endogenous clastosomes, actively participate in the degradation of mutant ATXN7 and, when disrupted, increase the amount of ATXN7 in cells. The slight increase in the amount of ATXN7 observed when PML III was overexpressed might also explain the disorganization of endogenous clastosomes or a change in their composition. These observations also suggest that endogenous clastosomes are not sufficiently active to prevent the aggregation of mutant ATXN7. The inhibitory effects on aggregation of increased PML IV expression raises the question of whether this would provide the basis of an interesting therapeutic strategy to treat SCA7 or other polyQ disorders. In the brains of patients, endogenous clastosomes might prevent the accumulation of mutant proteins for several decades before onset of aggregation. Later in the pathogenic process, the balance between clastosome activity and the level of accumulating polyQ proteins might be altered as a result of the disease process or a decrease in proteasome or chaperone activity during aging (; ), leading to the formation of polyQ aggregates. Because the mutant proteins are concentrated in the clastosomes, this might even be a site where the aggregation of proteolytic fragments of the proteins begins. The colocalization of PML bodies and small polyQ aggregates in the brain of patients and animal models supports this hypothesis. β-INF increased the number and size of PML bodies and decreased the levels of both soluble and aggregated mutant ATXN7. Although the debate concerning the relative contributions of soluble and aggregated polyQ proteins to cellular toxicity has not yet been resolved (), the reduction of either form of the protein could be beneficial to cells. The observation that only mutant but not endogenous ATXN7 is degraded in the presence of β-INF, which is an approved drug used in the treatment of multiple sclerosis (), reinforces the hypothesis that this substance may have therapeutic value, as it would not eliminate functional ATXN7. However, it affected the aggregation of mutant ATXN7 in cell cultures only when treatment began before or shortly after transfection with the ATXN7 constructs. This is consistent with the action of PML IV on the soluble form of mutant ATXN7 but not on preexisting aggregates, raising the question of when in the course of SCA7 or other polyQ diseases an effective therapy might be initiated. This should be explored in animal models. In conclusion, our study shows that the capacity of the nucleus to degrade specific protein substrates is intimately linked to the level and composition of PML bodies. In addition to providing a molecular basis for the accumulation of certain mutant proteins in the nucleus, these findings also suggest that protein degradation through clastosomes can be potentiated to prevent the nuclear accumulation of undesirable proteins, such as those with expanded polyQ, raising the hopes that this mechanism will be adapted to therapeutic ends. The SCA7 constructs were previously described (). FL-10Q and -74Q, cloned in pCDNA3, contain the full-length wild-type or expanded ATXN7 cDNA, respectively, with FLAG tags at the COOH terminus. FL-74Q contains, in addition, an HA tag at the NH terminus. Full-length and truncated ATXN7 containing 100Q were cloned in the pEGFP-N1 vector. A PML IV construct in the pTL2 vector was provided by P. Kastner (Institut de Génétique et de Biologie Moléculaire et Cellulaire, Strasbourg, France). PML IV was amplified by PCR (see the supplemental text, available at , for primer sequences) and cloned in pCDNA3. HcRFP-PML IV plasmid was generated by subcloning PML IV from pCDNA3 into pHcRed1-C1 vector (CLONTECH Laboratories, Inc.). All PML constructions were verified by sequencing. The PML III construct in the pSG5 vector was given by A. Dejean (Institut National de la Santé et de la Recherche Médicale, Paris, France). RNA was extracted from whole-cell lysates of transfected or untransfected COS-7 cells, using the RNAeasy kit (QIAGEN), and cDNAs were synthesized (ThermoScript RT-PCR System; Invitrogen) according to the manufacturer's instructions. PCR was performed using primers specific for the genes of interest. PCR primers for amplification of the SCA7 gene and for the PML IV–specific 3′ end are given in the supplemental text. The cyclophilin A gene was used as the reference. Cells were harvested in PBS, pelleted, and lysed 30 min on ice in lysis buffer containing 50 mM Tris, pH 8.8, 100 mM NaCl, 5 mM MgCl, 0.5% NP-40, 1 mM EDTA, and 250 IU/ml benzonase (Merck) supplemented with a cocktail of protease inhibitors (Complete and Pefabloc; Roche). Total extracts were centrifuged at 13,000 rpm for 10 min at 4°C to separate soluble proteins from aggregates and membranes. Supernatants (30 μg protein) were analyzed by Western blot. Pellets were washed with PBS and further incubated 30 min on ice in a pellet buffer containing 20 mM Tris, pH 8.0, 15 mM MgCl, and 250 IU/ml benzonase. Protein concentrations (supernatant and pellet) were determined by Bradford assay (Bio-Rad Laboratories). Samples were prepared for Western blot and for filter assay as previously described (; ). Pulse-chase experiments were performed 17 h after transfection of COS-7 cells with FL-74Q or with FL-74Q and PML IV. Cells were starved in methionine/cysteine-free DME containing 2% FBS for 1 h. Cells were labeled for 1 h with 200 μCi/ml of Pro-mix -S labeling mix (GE Healthcare) in starvation medium. After the pulse, cells were rinsed twice and chased for the indicated periods of time in DME with 5% FBS. For immunoprecipitation, cells were lysed in 50 mM Tris, pH 8.0, 300 mM NaCl, 1% NP-40, 5% glycerol, and protease inhibitors. Equal amounts of labeled lysates (12 × 10 cpm at time 0, determined after TCA precipitation) were immunoprecipitated with 3 μg anti-HA antibody (coupled to protein G–Sepharose beads) for 2 h at 4°C. Beads were washed four times with RIPA buffer, and bound proteins were directly denatured in 40 μl Laemmli buffer. Immunoprecipitates were resolved on 4–12% SDS gradient gels (Invitrogen), visualized by phosphoimaging, and quantified with Aida analysis software (Raytest GmbH). Cells plated on glass coverslips coated with poly-lysine (COS-7) or polyethyleneimine (cortical neurons) were fixed for 20 min with 4% paraformaldehyde. Primary neurons were fixed at 4 d after transfection. Permeabilization and further incubations with antibodies were done as previously described (). Cells were mounted with Fluoromount-G, and samples were observed with a confocal microscope (SP1; Leica) equipped with a 63×/1.32 NA objective. Leica confocal software was used to acquire images. For conventional analysis, we used a microscope (Axioplan-2; Carl Zeiss MicroImaging, Inc.) equipped with 25×/0.80 NA, 63×/1.40 NA, and 100×/1.40 NA objectives and a cooled mono 12-bits camera (Evolution QEi; Explora Nova). Fluo'Up software (Explora Nova) was used for image acquisition. 9-μm cryostat sections of fixed retina were prepared and used for immunofluorescence detection as described previously (). For Western blot and immunofluorescence analysis, we used the following antibodies and concentrations: 1C1 mouse anti-ATXN7 () at 1:5,000 (Western blot) and 1:1,000 (immunofluorescence), mouse anti-HA (Babco) at 1:10,000 (Western blot) and 1:4,000 (immunofluorescence), rabbit anti-ATXN7 (Affinity BioReagents, Inc.) at 1:500 (immunofluorescence), rabbit anti-ATXN7 affinity-purified polyclonal antibody 1261 at 1:100 (immunofluorescence on fixed retina only), mouse anti-PML (PG-M3; Santa Cruz Biotechnology, Inc.) at 1:500 (Western blot and immunofluorescence), rabbit anti-PML (H-238; Santa Cruz Biotechnology, Inc.) at 1:500 (Western blot and immunofluorescence), chicken anti-PML antibody at 1:500 (immunofluorescence on fixed retina only; a gift from H. de Thé, Centre National de la Recherche Scientifique, Paris, France), 1C2 anti-polyQ monoclonal antibody at 1:2,000 (Western blot), anti-FMRP 1C3 antibody () at 1:1,000 (Western blot), mouse anti-polyubiquitinylated proteins (FK2; BIOMOL Research Laboratories, Inc.) at 1:1,000 (immunofluorescence), rabbit anti-19S proteasome subunit S10a (PW-8225; BIOMOL Research Laboratories, Inc.) at 1:500 (immunofluorescence), rabbit anti-20S core (PW-8155; BIOMOL Research Laboratories, Inc.) at 1:1,000 (immunofluorescence), rabbit anti-Hsp40 (SPA-400; StressGen Biotechnologies) at 1:2,000 (immunofluorescence). For the filter assay, we used the same antibody concentrations as for Western blots. For immunofluorescence, secondary antibodies were as follows: Alexa 488–conjugated goat anti–mouse or goat anti–rabbit IgG (Invitrogen) used at 1:500 and CY3-conjugated goat anti–mouse or goat anti–rabbit IgG (Jackson ImmunoResearch Laboratories) used at 1:1,000. In , different secondary antibodies and dilutions were used: CY3-conjugated goat anti–rabbit IgG (1:200) and FITC-conjugated goat anti–chicken IgG (1:150). HeLa Kyoto cells (a gift from S. Narumiya, Kyoto University, Kyoto, Japan) grown on 35-mm microwell dishes (MatTek Corporation) were transfected with 1 μg FL-100Q-EGFP (expressing GFP-ATXN7 fusion protein) and 0.5 μg HcRFP-PML IV using FuGENE6 (Roche). At 6 h after transfection, cells were placed on the microscope in a chamber at 37°C and 5% CO. Several cells showing low levels of GFP-ATXN7 were selected and monitored every 10 min for 16 h. Images were acquired with an inverted microscope (DMIRE2; Leica) equipped with a 63×/1.35 NA objective and a camera (CoolSNAP fx; Roper Scientific). MetaMorph software (Molecular Devices) was used to control image acquisition and manipulation. COS-7 cells were transfected in 90-mm culture dishes with PML IV and FL-74Q or -10Q. Cells were harvested 42–45 h after transfection and lysed on ice in 150 mM salt buffer (50 mM Tris, pH 8.0, 150 mM NaCl, 1 mM EDTA, and 1% NP-40) supplemented with a cocktail of protease inhibitors (Complete and Pefabloc). Total extract was centrifuged at 13,000 rpm for 10 min at 4°C, and the supernatant was precleared with protein A– or protein G–Sepharose beads (GE Healthcare) for 1 h at 4°C. Precleared supernatant and protein A– or protein G–Sepharose beads coupled with the appropriate antibodies were incubated 2 h at 4°C on a rotating wheel for immunoprecipitation. Finally, the beads were washed five times with 500 mM salt buffer (50 mM Tris, pH 8.0, 500 mM NaCl, 1 mM EDTA, and 1% NP-40; supplemented with protease inhibitors), and bound proteins were directly denatured by heating for 5 min at 95°C in 30 μl Laemmli buffer. Samples were analyzed by Western blot. Cells were fixed with 4% paraformaldehyde in phosphate buffer, pH 7.4, for 1 h on ice and then postfixed with 1% osmium tetroxyde and embedded in Epon 812 (Taab). Ultrathin sections (50 nm) placed on nickel grids were immunogold labeled after embedding. We used primary antibodies mouse monoclonal anti–ATXN7 1C1 at 1:200 and rabbit polyclonal anti-PML at 1:250 and secondary antibodies conjugated to 10- or 15-nm gold particles (goat anti–mouse or goat anti–rabbit [Aurion]) at 1:100. After labeling, sections were postfixed in 1% glutaraldehyde and stained with uranyl acetate and lead citrate (see detailed protocol in the supplemental text). Cells were examined with an electron microscope (1200EX; Jeol) at 80 kV. Fig. S1 shows that PML IV bodies recruit 19S5a, 19S6b, and 20Sα2 proteasome subunits. As a consequence, the colocalization of these actors of the protein degradation pathway with mutant ATXN7 strongly increases in double-transfected cells compared with cells expressing mutant ATXN7 alone. Online supplemental material is available at .

DNA replication is a key process that is functionally perturbed during DNA damage–triggered apoptosis (). DNA damage triggers apoptosis in a replication-dependent way by activating the mitochondrial damage pathway in fibroblasts (). Chromosomal replication can be impaired by intrinsic replication errors or by external agents that cause DNA damage (; ). Checkpoint-sensing kinases detect abnormal replication structures and activate the Chk2 kinase, which stabilizes replication forks and promotes recovery from DNA damage by phosphorylating downstream endonucleases, helicases, and recombinases, demonstrating that DNA replication forks are activators and effectors of the checkpoint pathway in S phase (; ). In response to a variety of DNA lesions in eukaryotic cells, DNA damage–sensing kinases such as ataxia telangiectasia mutated (ATM), ATM and Rad-3 related (ATR), and DNA-dependent protein kinase are activated as checkpoint sensors that signal both the cell cycle and apoptosis machinery through the Chk1/2 checkpoint kinases (; ; ). ATM and Chk2 directly phosphorylate p53, a key regulator of cellular responses to genotoxic stress. Phospho-p53 can then dissociate with the inhibitor protein Mdm2 and, thus, is stabilized and transcriptionally activated for DNA damage responses (; ). The p53 protein also has a transcription-independent activity that potentiates cell death once transcription-dependent functions initiate this process (). Cytoplasmic p53 directly activates the proapoptotic protein Bax through direct interaction (; ). Several lines of evidence indicate that replication initiation is impaired in the early stages of apoptosis. First, apoptosis is induced by defects in the initiation of DNA replication as a result of the o mutation (). In addition, temperature-sensitive mutants cause a defect in a checkpoint and initiation of DNA replication (; ). Second, replication initiation proteins such as Cdc6 and Mcm3 are cleaved by caspase early in apoptosis (; ; ; ). Third, when the expression of proteins such as Cdc6, Mcm2, and Cdc45 is blocked by the siRNA technique, proliferation is inhibited, and apoptosis is induced in cancer cells (). Thus, replication fork collapse induced by interfering with the pre–replicative complex (RC) may be a general feature of the early stages of apoptosis. In a previous study, we showed that caspase-3–mediated cleavage of Cdc6 induces the nuclear retention of the truncated protein p49-tCdc6 (truncated Cdc6) and apoptosis (). We proposed that p49-tCdc6 acts as a dominant-negative inhibitor of replication that consequently induces and enhances apoptosis. In this study, we show that Cdc6 is also specifically cleaved during apoptosis by caspase-3 at another aspartic acid residue, D290, yielding a truncation protein (p32-tCdc6) that accumulates in the nucleus under conditions in which cyclin A/Cdk2 activity is up-regulated. Interestingly, the expression of p32-tCdc6 or p49-tCdc6 markedly increases apoptosis in etoposide-treated cells and induces apoptosis of untreated cells. Moreover, the expression of tCdc6 proteins induces activation of the ATM and ATR kinase. The expression of tCdc6 also suppresses the chromatin loading of the replication initiation factor Mcm2. Importantly, tCdc6-induced effects are prevented in cells that coexpress a version of Cdc6 that cannot be cleaved by caspase-3 (Cdc6-UM). Moreover, when ATM or ATR is down-expressed using the siRNA technique, caspase-3 activity is suppressed in TNF-α–induced cells. These results suggest that the caspase-mediated cleavage of Cdc6 destabilizes the pre-RC and, thus, that tCdc6 proteins impair replication initiation by acting as dominant-negative inhibitors. This, in turn, results in the disruption of chromatin structure and/or induction of DNA damage, leading to ATM/ATR kinase–mediated apoptosis. In etoposide-induced apoptosis of HeLa cells, the 62-kD Cdc6 protein is specifically cleaved into fragments of 49 and ∼36 kD, with kinetics similar to that of the proteolytic cleavage of poly (ADP-ribose) polymerase (PARP; ). Cleavage at SEVD/G produces p49-tCdc6. To identify the second Cdc6 cleavage site, we constructed mutants in which asparagine is substituted for aspartic acid at one candidate site (DQLD/S) and at three other sequences (AGKD/M, LVLD/E, and KGQD/V) that are all in a region where cleavage would produce a fragment of ∼36 kD. Recombinant caspase-3 efficiently cleaved wild-type (wt) Cdc6 (Cdc6-wt) and the D262N, D284N, and D295N mutant proteins into 49- and 32-kD fragments (). In contrast, the D290N mutant was not cleaved, and this cleavage site was further confirmed with the double mutant protein D442N/D290N, which was resistant to caspase-3 cleavage, indicating that aspartic acid residues 290 and 442 are caspase-3 cleavage sites during apoptosis (). Interestingly, a 15–amino acid residue region including the DQLD/S cleavage site is well conserved among man, mouse, and , whereas the region including the SEVD/G site is more loosely conserved (). Moreover, p32-tCdc6 and p49-tCdc6 contain all functional domains that are known to be essential for pre-RC loading (). We then assessed whether the p32-tCdc6 protein is functional for translocation into the cytoplasm because caspase cleavage removes the COOH-terminal nuclear export signal (NES) sequences. pCS2+GFP-tagged Cdc6-wt, tCdc6, or Cdc6-AAA, which is an unphosphorylatable mutant of Cdc6 on 54, 74, and 106 serines, was cotransfected with pCMV-mock, cyclin A, or dominant-negative Cdk2 into HeLa cells. Cdc6-wt was translocated to the cytoplasm when cyclin A was coexpressed, whereas Cdc6-AAA was not even if cyclin A was coexpressed (). Although the cytoplasmic translocation of Cdc6-wt was mainly regulated by cyclin A/Cdk2-mediated phosphorylation (), the cytoplasmic translocation of p32-tCdc6 was relatively independent of cyclin A/Cdk2 activity. The p32-tCdc6 protein was localized in the nucleus in 92% of the cells cotransfected with pCMV-mock, a higher percentage than for cells expressing p49-tCdc6 or Cdc6-wt (87 and 68%, respectively). The differential distribution of p32-tCdc6 and p49-tCdc6 was more significant when cyclin A was coexpressed. Under conditions of elevated cyclin A/Cdk2 activity, the cytoplasmic trans-localization of p32-tCdc6 was suppressed by 78% as compared with that of Cdc6-wt (94 − 16%, because 94% of Cdc6-wt and 16% of p32-tCdc6 was localized in the cytoplasm). Under the same conditions, the translocation of p49-tCdc6 was suppressed by 57% (94 − 37%), as 37% of p49-tCdc6 was localized in the cytoplasm. Thus, p32-tCdc6 was more strongly retained in the nucleus (by 21%; 78 − 57%) as compared with the p49-tCdc6 protein under conditions of up-regulated cyclin A/Cdk2 activity. These results indicated that the p49-tCdc6 protein might contain an additional NES-like sequence between amino acid residues 290 and 442. In contrast, the cytoplasmic export of p49-tCdc6 and p32-tCdc6 was almost completely prevented, as was the translocation of Cdc6-wt in cells expressing Cdk2-DN, resulting in nuclear localization in 99% of the transfected cells (). Together, our data indicate that the phosphorylation of Cdc6-wt and tCdc6 proteins is essential for their cytoplasmic translocation and that the loss of a NES-like sequence from p49-tCdc6 reduces phosphorylation-dependent translocation. Our results showed that the majority of Cdc6-wt was translocated to the cytoplasm, whereas the Cdc6-AAA mutant protein was mainly retained in the nucleus even if cyclin A was coexpressed in the cells (). Also, Cdc6-wt, p49-tCdc6, and p32-tCdc6 but not Cdc6-AAA were all similarly phosphorylated when they were prepared from cells that coexpressed cyclin A. Phosphorylation was prevented in cells expressing Cdk2-DN (). Thus, tCdc6 proteins were similarly phosphorylated, but their phosphorylation-dependent cytoplasmic translocation was impaired by the loss of NES-like sequences in the region of 300–315 amino acid sequences. We next tested the effect of p32-tCdc6 or p49-tCdc6 expression on apoptosis. Indeed, the expression of p32-tCdc6 notably enhanced etoposide-induced cell death to a greater extent than that of p49-tCdc6 (). The expression of either protein markedly increased the number of cells with apoptotic morphology (membrane blebbing and cell shrinkage) to 95 and 79%, respectively, whereas such changes occurred in <32% of cells expressing Cdc6-wt (). Similarly, the numbers of cells expressing tCdc6 that exhibited condensed DNA or that were Annexin V and propidium iodide (PI) also significantly increased after treatment with etoposide (). In all cases, the apoptosis-promoting effects of p32-tCdc6 expression were higher than those induced by p49-tCdc6 expression. In addition, when pCS2+GFP-tagged p49- and p32-tCdc6 (1.5 μg cDNA of each) were cotransfected, the apoptosis-promoting effect was slightly lower than that induced by p32-tCdc6 (3 μg cDNA) but significantly higher than that induced by the same amount of p49-tCdc6 cDNA (). Also, apoptosis was clearly induced in cells expressing p32-tCdc6 under normal HeLa cell culture conditions. The number of cells with apoptotic cell morphology increased by 78 and 63% from 24 to 48 h after transfection in p32-tCdc6– or p49-tCdc6–expressed cells, respectively, whereas the expression of Cdc6-wt had a minimal effect (). Moreover, the sub-G1 proportion of cells expressing GFP-tagged p32-tCdc6 increased to 21% over the course of 48 h, whereas <10% of cells expressing Cdc6-wt or the vector alone were in this sub-G1 proportion (). In addition, the Annexin V assay showed that the early apoptotic population of cells expressing p32-tCdc6 increased by three- to fourfold as compared with cells overexpressing Cdc6-wt or the vector 48 h after transfection (). When p49- and p32-tCdc6 were coexpressed in the cells, the apoptosis-inducing effect increased in a time-dependent manner over the 48 h (). The results indicated that in the coexpressed cells, the apoptosis-inducing effect is even higher than those induced by expressing p49- or p32-tCdc6 alone 48 h after transfection. Thus, the expression of tCdc6 proteins induces apoptosis in HeLa cells under normal cell culture conditions, and the differential effects of the two truncated versions of Cdc6 are consistent with their relative impairments in export from the nucleus. To assess the functional relevance of the caspase-mediated cleavage of Cdc6, we first tested its effect on pre-RC loading. Both p49- and p32-tCdc6 bound chromatin with kinetics similar to that of PARP cleavage in cells treated with etoposide or TNF-α (). The kinetics of binding indicated that p49-tCdc6 bound before p32-tCdc6. Moreover, chromatin-bound Mcm2 levels were strongly reduced, whereas the levels of chromatin-bound Orc2 remained unaltered in cells treated with etoposide or TNF-α. In addition, the reduced levels of chromatin-bound Mcm2 temporally correlated with elevations in the level of chromatin-bound tCdc6 proteins (). Thus, the caspase-dependent cleavage of chromatin-bound Cdc6 interfered with the binding of Mcm2 binding onto chromatin, but cleavage did not prevent tCdc6 proteins from binding chromatin. Also, ectopically expressed tCdc6 proteins were specifically incorporated onto chromatin and subsequently inhibited the binding of Mcm2 to chromatin, but they did not interfere with the loading of Orc2 onto chromatin in both asynchronized and early S phase cells (). Our results also indicate that the interaction of Mcm2 with tCdc6 proteins was markedly reduced as compared with the Mcm2–Cdc6-wt interaction in transfected cell extracts (). The cellular levels of Cdc6-wt and tCdc6 proteins were largely equivalent. In summary, these results indicate that the binding of tCdc6 proteins to chromatin interferes with the loading of DNA replication initiation factors, including Mcm2, onto replication complexes, thereby hindering the initiation of replication. We then determined whether the tCdc6-induced impairment of Mcm2 loading onto chromatin is functionally linked to ATM and ATR kinase activities (). Interestingly, in cells expressing tCdc6, ATR and ATM kinase activities significantly increased in a time-dependent fashion 24–36 h after transfection, whereas their cellular protein levels were minimally altered. Moreover, in cells expressing tCdc6, the phosphorylated levels of Chk1 on S345 residue and Chk2 on T68 also markedly increased, with kinetics similar to those of the increases in ATM and ATR kinase activities (). In addition, the extent of phosphorylation on the S15 residue of p53, which is phosphorylated by ATM kinase, increased in parallel with increases in the p53 protein level in cells expressing tCdc6. These results indicate that the expression of tCdc6 is likely to disturb chromatin structures and/or induce DNA damage, leading to ATM and ATR kinase activation and, thus, to p53 protein stabilization during apoptosis. The protein levels of Bax and p21, which are target molecules of p53, also increased under the same conditions, indicating that the increased p53 level was functionally linked to induce Bax and p21 expression in tCdc6-expressing HeLa cells (). We next assessed the mitochondrial levels of Bax in cells expressing Cdc6-wt or tCdct6, as Bax is known to translocate to mitochondria during apoptosis (). As shown in , the mitochondrial levels of Bax in cells expressing tCdc6 increased in a time-dependent manner, whereas the levels of mitochondrial VDAC, a mitochondrial marker protein, were unaltered. These results indicate that the mitochondrial translocation of Bax is promoted in cells expressing tCdc6 proteins. Interestingly, the levels of the p53–Bax complex were also notably elevated in cells expressing tCdc6, whereas the complex was almost undetectable in cells expressing the vector or Cdc6-wt (). Moreover, Bax in association with the p53 protein was detected as a double band in cells expressing tCdc6, whereas Bax that was unassociated with p53 was detected as a single band, suggesting that the upper band is a phosphorylated form (). These results also showed that the levels of mitochondrial Bax significantly increased in parallel with increases in the levels of p53 in transfected cells (). Interestingly, the levels of p53 in mitochondria also increased in concert with increases in mitochondrial Bax 36 h after transfection in tCdc6-expressed cells (). These results indicate that tCdc6 expression induces an elevation in the cellular levels of the p53–Bax complex and that this increase may be involved in the mitochondrial translocation of these proteins. We asked whether an uncleavable mutant of Cdc6 (Cdc6-UM) has the effects described for tCdc6 that disturb Mcm2 loading to the chromatin and activate DNA damage–sensing kinase activity and proapoptotic effect. As expected, Cdc6-UM was uncleaved, whereas Cdc6-wt was cleaved to p49-tCdc6 during TNF-α–induced apoptosis (). In addition, loading of the Mcm2 protein onto chromatin and ATR kinase activity were minimally influenced in cells expressing Cdc6-UM. In contrast, Mcm2 chromatin loading was suppressed, whereas ATR kinase activity was prominently up-regulated in cells expressing p49-tCdc6 or p32-tCdc6 (, B and C; top). However, the tCdc6-induced up-regulation of ATR kinase activity was markedly suppressed by coexpressing Cdc6-UM (, bottom). Moreover, the percentage of p32-tCdc6–expressing cells with apoptotic cell morphology decreased by ∼45% (79.7 − 34.5% = 45.2%) over the time course of 36–48 h after coexpression with Cdc6-UM (). These competition effects of Cdc6-UM on the tCdc6-induced events were similarly observed when the cells were coexpressed with Cdc6-wt and tCdc6 (). These results indicate that the expression of tCdc6 impairs Mcm2 loading onto chromatin, which is functionally linked to ATM/ATR kinase activation leading to apoptosis, and that these effects are blocked by the expression of Cdc6-UM. We next addressed the possibility that TNF-α–induced cell death is influenced by tCdc6 and Cdc6-UM. TNF-α–induced apoptosis, as measured by caspase-3 activity, was enhanced in cells expressing tCdc6 (). In contrast, the tCdc6-induced activation of caspase-3 was suppressed in cells expressing Cdc6-UM to a level that was lower than that exhibited by mock-transfected cells treated with TNF-α but was still much higher than that of unexposed control cells. Thus, we tested whether tCdc6 expression promotes ATM kinase activity during TNF-α–induced apoptosis. Interestingly, this activity was significantly elevated in HeLa cells 2 h after TNF-α treatment (). Moreover, TNF-α–induced kinase activation was clearly enhanced by about threefold in cells expressing p32-tCdc6 as compared with that in mock-transfected cells either untreated or treated with TNF-α. Again, TNF-α–induced caspase-3 activity was suppressed in cells expressing Cdc6-UM to levels even lower than those of cells expressing mock or Cdc6-wt (). Similarly, TNF-α–induced ATM kinase activity in cells expressing Cdc6-UM was at the basal level, which was similar to that in cells expressing mock (). Nevertheless, the kinase activity in Cdc6-UM–transfected cells was three- to fourfold lower than that in p32-tCdc6–expressed cells after TNF-α treatment. Thus, these results indicate that specific cleavage of Cdc6 is likely functionally linked to ATM kinase activation during TNF-α–induced cell death in HeLa cells. To assess whether ATM or ATR kinase activation might be functionally linked to apoptosis induced by expressing tCdc6 or treatment with TNF-α, the expression level of ATM or ATR kinase was specifically down-regulated using the siRNA technique. 24 h after transfection with siRNA for ATM or ATR, the cells were transfected with pCS2+GFP–Cdc6-wt or −p32-tCdc6. The results indicated that protein levels and the tCdc6-induced activation of ATM or ATR kinase activity were similarly down-regulated to basal levels in the cells by treatment with corresponding siRNA (). Moreover, the p32-tCdc6–induced activation of caspase-3 activity was markedly suppressed in the cells by treatment with siRNA of ATM or ATR (). Thus, apoptotic cell death induced by expressing tCdc6 was significantly suppressed by down-expressing ATM or ATR kinase in the cells. Interestingly, in the ATM or ATR down-expressed cells, the levels of caspase-3 activity were even lower than those of cells expressing mock or Cdc6-wt (). These results indicated that apoptotic cell death induced by transfection stress alone can be also suppressed by down-expressing ATM or ATR kinase in the cells. We next addressed whether the down-expression of ATM or ATR might also suppress the TNF-α–induced apoptosis. The TNF-α–induced caspase-3 activity was similarly suppressed by ∼50% as compared with those in the control siRNA-treated cells by down-expressing ATM or ATR in the cells (). These results indicate that activation of ATM and ATR kinase activity is commonly required for tCdc6- or TNF-α–induced apoptosis of the cells. d e s c r i b e a n o v e l f u n c t i o n a l r o l e f o r c a s p a s e - m e d i a t e d c l e a v a g e o f C d c 6 d u r i n g e t o p o s i d e - o r T N F - α – i n d u c e d a p o p t o s i s . I n p a r t i c u l a r , w e s h o w t h a t C d c 6 t r u n c a t i o n i m p a i r s M c m 2 l o a d i n g o n t o c h r o m a t i n a n d s u b s e q u e n t l y a c t i v a t e s D N A d a m a g e – s e n s i n g k i n a s e a c t i v i t i e s t h a t m a y b e l i n k e d t o p 5 3 – B a x - m e d i a t e d p r o g r a m m e d c e l l d e a t h . O u r f i n d i n g 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 f u n c t i o n a l r o l e o f C d c 6 i n a p o p t o s i s a n d a l s o f o r t C d c 6 - i n d u c e d D N A d a m a g e s i g n a l i n g i n e n h a n c i n g a p o p t o s i s t h a t i s i n d u c e d b y t r e a t m e n t w i t h e t o p o s i d e o r T N F - α . DME and FBS were obtained from Invitrogen. The transfection reagent Polyfect was purchased from QIAGEN, and Annexin V AlexaFluor568 was purchased from Roche Diagnostics. Human recombinant TNF-α was obtained from Biosource International. Other chemical reagents were purchased from Sigma-Aldrich. Human Cdc6 cDNA was provided by R.S. Williams (University of Texas Southwestern Medical Center, Dallas, TX; GenBank/EMBL/DDBJ accession no. ). Converting the Asp 262, 284, 290, 295, and 442 residues of Cdc6 to Asn was accomplished by altering the Asp codons GAT or GAC to the Asn codons AAT or AAC, respectively, by the megaprimer method (). The mutation was first introduced by PCR with the mutagenic primers (5′-TGGGAAGAACATGATGAGG-3′, 5′-GGTATTGAACGAGATGG-3′, 5′-TCAACTGAACAGCAAAG-3′, 5′-AGGCCAGAATGTATTG-3′, 5′-TCAGAAGTTAATGGTAACAGG-3′), the forward primer (5′-AAGGATCCCCTCAAACCCGATCCCAG-3′), and the reverse primer (5′-AAGGATCCT TAAGGCAATCCAG-3′), and the products were cloned into pCITE-4c for in vitro transcription and translation. Cdc6-AAA (S54A, S74A, and D106A), an unphosphorylatable mutant of Cdc6 by cyclin A/Cdk2 in which Ser 54, 74, and 106 residues of Cdc6 were mutated to Ala, was constructed by sequential single mutation using the PCR method (S54A, 5′-AGCCCTGCCTCTCGCCCCCAGGAAACGT-3′; S74A, 5′-CCCATTTACCTCCTTGTGCTCCACCAAAGCAAGG-3′; and S106A, 5′-AATCAGCTGACAATTAAGGCTCCTAGCAAAAGAGAACTAG-3′). All mutants were confirmed by DNA sequencing. To monitor mammalian expression, Cdc6 was inserted into the vector pCS2+GFP by amplifying full-length Cdc6 cDNA (encoding aa 1–560), p49-tCdc6 cDNA (encoding aa 1–442), or p32-tCdc6 cDNA (encoding aa 1–290). Full-length human cyclin A cDNA was generated by RT-PCR (using oligomers 5′-AAGGATCCAGCCTATTCTTTGGCC-3′ and 5′-AAGGATCCGGCAGCTGGCATCATTAATAC-3′ with BamHI sites at the 5′and 3′ ends) and was cloned into the pCMV expression vector. siRNA duplexes for ATM, ATR, and the control nontargeting siRNA were purchased from Dharmacon. The siRNA sequences (sense strand) for ATM and ATR were UAUAUCACCUGUUUGUUAGUU and GCAACUCGCCUAACAGAUAUU, respectively. 5 × 10 HeLa cells in 35-mm dishes were transfected with a final concentration of 100 nM siRNA duplexes using Oligofectamine (Invitrogen) according to the manufacturer's instructions. The in vitro transcription and translation of cDNAs encoding the human Cdc6 protein and the Cdc6 substitution mutants were performed with the TNT Coupled Reticulocyte Lysate System (Promega) according to the manufacturer's protocols. For in vitro cleavage by caspase-3, 2 μl [S]methionine-labeled Cdc6 was incubated with 100 ng recombinant caspase-3 as described previously (). Human cervical carcinoma (HeLa) cells were maintained at 37°C and 5% CO as a monolayer culture in DME supplemented with 10% (vol/vol) heat-inactivated FBS, 100 units/ml penicillin, 100 μg/ml streptomycin, and 250 ng/ml amphotericin B. HeLa cells were transfected in 35-mm dishes with 3 μg of the appropriate plasmids using 12 μl Polyfect according to the manufacturer's instructions. For apoptotic induction, cells were treated with 85 μM etoposide or 5 ng/ml TNF-α and 5 μg/ml cycloheximide. As TNF-α alone had no effect on HeLa cell viability, protein synthesis inhibitor cycloheximide was used together with TNF-α for the induction of apoptosis in HeLa cells (). For DAPI staining, cells were fixed in 4% PFA solution as described previously (). The Annexin V assay was performed as described previously (). The percentage of Annexin V–positive cells and PI-negative cells among GFP-expressing cells was determined by flow cytometry (Particle Analyzing System PAS-III; Partec). Transfected cells were selectively analyzed by gating a subpopulation by forward scatter versus side scatter followed by gating green fluorescence intensity. Because the percentage of Annexin V cells indicates the frequency of total apoptotic cells and the percentage of PI cells indicates the frequency of late apoptotic and necrotic cells, the apoptotic index of early apoptotic cell populations was calculated by subtracting the percentage of PI from the percentage of Annexin V. For determination of the sub-G1 fraction as an apoptotic index, cells expressing GFP-tagged proteins were collected over time and fixed in 75% ethanol, stained with 500 μl of 50 μg/ml PI solution, and subjected to FACS analysis. GFP-expressing cells were selectively analyzed for DNA content followed by gating green fluorescence intensity. HeLa cells expressing Cdc6 were collected and extracted in lysis buffer (0.5% Triton X-100, 20 mM Tris, pH 7.5, 2 mM MgCl, 1 mM DTT, 1 mM EGTA, 50 mM β-glycerophosphate, 25 mM NaF, 1 mM Na vanadate, 2 μg/ml leupeptin, 2 μg/ml pepstatin A, 1 μg/ml antipain, and 100 μg/ml PMSF). Cell lysates were centrifuged at 12,000 rpm for 15 min at 4°C, and the supernatants were collected. After adjusting the protein concentration, proteins were resolved by SDS-PAGE before Western blot analysis with appropriate antibodies. Monoclonal anti-Cdc6, anti-GFP, anti-ATM, anti-p53, and anti-Chk2 antibodies and polyclonal anti-PARP, anti–p-Cdc6 (S54), anti–cyclin A, anti-Cdk2, anti-Chk1, anti-ATM, anti-ATR, anti-Bax, and anti-VDAC antibodies were obtained from Santa Cruz Biotechnology, Inc. Polyclonal phospho-p53 (S15), phospho-Chk2 (T68), and phospho-Chk1 (S345) antibodies were purchased from Cell Signaling Technology. Polyclonal anti-Mcm2 antibody was obtained from BD Biosciences. Immune complexes were revealed using ECL Western blotting detection reagents (Intron Biotechnologies). For immunoprecipitation reaction, p53 or GFP was precipitated from cell lysates with an anti-p53 polyclonal antibody or anti-GFP monoclonal antibody overnight at 4°C with end-over-end mixing followed by incubation with protein A agarose for 2 h at 4°C. Immunoprecipitates were separated from supernatants by centrifugation and washed with lysis buffer. Proteins were extracted from agarose beads by resuspension in 1× laemmli gel-loading buffer and resolved by SDS-PAGE. For the detection of mitochondrial proteins, mitochondrial fractions were extracted in lysis buffer. Mitochondrial proteins were resolved by SDS-PAGE before Western blot analysis with anti-p53, anti-Bax, and anti-VDAC antibodies. Nuclear and mitochondrial extracts were prepared as described previously (). Chromatin was fractionated with Triton X-100 as described previously () with slight modifications. To synchronize 24 h after transfection, cells were treated with 2.5 mM hydroxyurea for 24 h. The nuclear fraction of HeLa cells (2 × 10) expressing Cdc6 was prepared and resuspended in 300 μl Cytoskeleton buffer containing 0.5% Triton X-100. For nuclease digestions, chromatin pellets were resuspended in lysis buffer including 25 units of benzonase nuclease and incubated at 37°C for 10 min followed by chilling on ice and centrifugation. 20 μg of the supernatants were boiled with loading buffer and resolved by 10% SDS-PAGE before Western blot analysis with anti-Cdc6, anti-Orc2, and anti-Mcm2 antibodies. ATM or ATR kinase was immunoprecipitated from tCdc6-overexpressing HeLa cells with anti-ATM or anti-ATR polyclonal antibody in buffer (0.5% Triton X-100, 20 mM Tris, pH 7.5, 2 mM MgCl, 1 mM DTT, 1 mM EGTA, 50 mM β-glycerophosphate, 25 mM NaF, 1 mM NaVO, 2 μg/ml leupeptin, 2 μg/ml pepstatin A, 100 μg/ml PMSF, and 1 μg/ml antipain) followed by incubation with protein A agarose at 4°C for 2 h. The reactions for ATM or ATR kinase activity were performed using 1 μg PHAS-1 (Calbiochem) as a substrate in reaction buffer (50 mM Tris, pH 7.5, 10 mM MgCl, 1 mM DTT, 1 mM EGTA, 50 mM β-glycerophosphate, 25 mM NaF, 0.1 mM NaVO, 1 μg/ml leupeptin, 1 μg/ml pepstatin A, 1 μg/ml antipain, and 100 μg/ml PMSF) containing immunoprecipitated ATM protein or ATR protein and 10 μCi γ-[P]ATP at 30°C for 30 min. The reacted samples were suspended in SDS loading buffer, resolved by SDS-PAGE, and detected by autoradiography. 50 μg of transfected cell lysates were incubated with 200 nM Ac-DEVD-AFC (BD Biosciences) in reaction buffer (20 mM Hepes, pH 7.4, 100 mM NaCl, 10 mM DTT, 0.1% CHAPS, and 10% sucrose) at 37°C for 1 h. The reaction was monitored by fluorescence emission at 535 nm (excitation at 405 nm).

In addition to being the most common cell state on earth (), quiescence, or G, is critically important in the development and survival of all organisms and plays a significant role in disease, such as cancer (). Quiescent cells are directly involved in the establishment and persistence of microbial infectious diseases such as tuberculosis () and cryptosporosis (). In complex eukaryotes, quiescence is essential for stem-cell maintenance () as well as for activating cells for wound healing () and sexual reproduction. Both wound healing and sexual reproduction require regulated exit from G (). Difficulties in isolating populations of quiescent cells have made a clear understanding of quiescence almost impossible. For example, in metazoans, quiescent cells usually closely associate with nonquiescent cells; thus, the isolation of pure populations of quiescent cells in quantities large enough for systems-level analysis has not usually been feasible (). In addition, stationary-phase (SP) cultures of the unicellular yeast include both quiescent and nonquiescent cells, but this heterogeneity was not recognized so that all cells in an SP culture were sometimes mistakenly referred to as quiescent (; ). This heterogeneity has led to serious confusion about the relationship between SP and quiescence in yeast. This confusion has been compounded by the lack of a concise definition of quiescence; quiescent yeast cells have been defined more by what they do not do than by what they do (). Nonetheless, a comparison of yeast cells in SP cultures with those in exponentially growing cultures has identified several characteristics associated with quiescence, including a thickened cell wall, decreased metabolic rate, and accumulation of a variety of storage molecules, such as trehalose and glycogen (). In addition, quiescent cells exhibit decreased transcription () and translation (), have characteristically condensed chromosomes (), and show increased resistance to a variety of stresses (). Because cells under other conditions can exhibit these characteristics, the hallmark of quiescent cells was proposed to be their ability to retain viability without added nutrients, including the ability to reproduce. We report here the separation of quiescent and nonquiescent cells from SP cultures by density-gradient centrifugation. Cells in the lower, denser fraction have characteristics of quiescent cells, suggesting that they are bona fide G cells. They are mostly unbudded daughter cells that retain both viability and the ability to reproduce. Furthermore, they synchronously reenter mitosis when inoculated into fresh medium. Additionally, they are very thermotolerant and highly refractive by phase-contrast microscopy. Cells in the less dense, upper fraction are heterogeneous. They include both budded and unbudded cells and do not synchronously reenter mitosis, but they all rapidly lose the ability to reproduce over time. They are less resistant to heat stress than quiescent cells and exhibit signs of oxidative stress and apoptosis. Microarray analysis also revealed distinct differences between the cell fractions. Transcripts encoding proteins required for response to water stress, fatty acid oxidation, and some metabolic pathways, including coenzyme metabolism, were more abundant in quiescent cells. Transcripts encoding genes involved in DNA recombination and transposition were more abundant in nonquiescent cells, consistent with a high percentage of these cells being apoptotic. Interestingly, the transcripts that differentiate quiescent and nonquiescent cells are not those that have previously been identified as SP transcripts (; ). The significance of this work is that this represents the first time quiescent, putative G cells have been isolated and studied from yeast SP cultures. It was unexpected that the quiescent cell fraction would consist mostly of daughter cells and that, although viable, nonquiescent cells would so rapidly lose their ability to reproduce. Based on these observations, we present models for the development of quiescent and nonquiescent cells during the growth of cultures to SP and their relationship to cells in the mitotic cell cycle. Previously, cells in yeast SP cultures were observed to be asynchronous and contain a mixture of unbudded and budded cells (; ). To further characterize different cell types within an SP culture, we sought a method that would separate the cell types. Separation of cells from SP yeast cultures by equilibrium density-gradient centrifugation revealed the presence of two fractions with different buoyant densities: ∼1.10 g/ml (the upper fraction) and 1.14 g/ml (the lower fraction; ). In contrast, cells from exponentially growing cultures formed a single band with a buoyant density of 1.12 g/ml. Two fractions were also observed for SP cultures of other strains, including W303, BY4742, and BY4743 (unpublished data). Additionally, several (, , , , , and ) of >50 mitochondrial mutants evaluated () were unable to form a lower band at 7 d (unpublished data), whereas other, functionally related mutants, such as , , , and , were able to do so. These differences in ability of mitochondrial mutants to form a lower band suggest that there are some as-yet-uncharacterized functional differences between these mitochondrial genes. We examined cultures to determine when the two fractions began to differentiate in the culture over time (). A single band of cells was detected 11 and 12 h after inoculation. Notably, glucose exhaustion also occurred at 12 h after inoculation (, arrow). A distinct broadening of the band of cells was detected by 19 h after inoculation, and two bands were observed by 20–24.5 h. The lower cell fraction continued to increase in density for at least 7 d, whereas the upper cell fraction exhibited a slight decrease in density between 3 and 7 d after inoculation. We conclude that the lower-fraction cells form after glucose exhaustion during the transition to growth on nonfermentable carbon sources. To determine whether there were morphological differences between the fractions, they were examined by light and electron microscopy. The more dense, lower-fraction cells were unbudded and highly refractive by phase-contrast microscopy () as well as uniform in size (4.6 ± 0.33 μm). These cells contained nuclei and electron-dense vacuoles but, strikingly, other internal organelles, including mitochondria and endoplasmic reticulum, were not detectable (, lower fraction, TEM). They also contained significant concentrations of glycogen (, lower fraction, TEM–phosphotungstic acid), which is known to increase after the diauxic shift (). The less dense, upper-fraction cells exhibited some distinct characteristics. They contained budded and unbudded cells and were dark by phase-contrast microscopy (, upper fraction, phase contrast) but had mean diameters similar to the lower-fraction cells (4.1 ± 0.34 μm). These cells had large vacuoles and numerous mitochondria and ER profiles but only trace amounts of glycogen (, upper fraction, TEM and TEM–phosphotungstic acid). In addition, their vacuoles contained numerous vesicles (, TEM–phosphotungstic acid, white arrow), suggesting enhanced autophagy (). We conclude that the two cell fractions are morphologically distinct, with the more uniform, lower-fraction cells exhibiting characteristics associated with quiescent cells. To determine whether these cell fractions were physiologically different, we tested for long-term viability, colony-forming capacity, and thermotolerance. Lower-fraction cells exhibited >87% viability over 28 d in culture, as detected by FUN-1 (2-chloro-4-[2,3-dihydro-3-methyl-{benzo-1,3-thiazol-2-yl}-methylidene]-1-phenylquinolinium iodide) uptake and flow cytometry (). Upper-fraction cells (>85%) were viable at 7 and 14 d but decreased in viability (to 47%) by 28 d. The two-cell fractions exhibited even larger differences in colony-forming capacity (). Essentially all viable lower-fraction cells could produce colonies at 7 d. This decreased to ∼65% at both 14 and 21 d and to 12% by 28 d. Because cells in SP cultures are known to become thermotolerant (), we wanted to determine whether these cell fractions differed in their response to heat shock. Cells from both fractions and exponential cultures were tested for their ability to survive a heat shock at 52°C (). Lower-fraction cells were the most heat resistant, retaining essentially 100% viability over 15 min. As expected, cells from exponentially growing cultures rapidly lost the ability to form colonies after heat shock. Upper-fraction cells were more resistant than cells from exponentially growing cultures but significantly less resistant than lower-fraction cells. Approximately 50% of the upper-fraction cells were unable to form colonies after 10–15 min at 52°C, decreasing to 25% after 20 min. We conclude that the lower fraction is uniformly thermotolerant, whereas the upper fraction is heterogeneous and contains a small percentage of thermotolerant cells. In eukaryotes, the hallmark of G cells obtained by serum starvation is their ability to synchronously reenter the mitotic cell cycle (). Microscopic analysis revealed that lower-fraction, quiescent cells budded synchronously after slightly more than 1 h (, black line). In contrast, nonquiescent, upper-fraction cells were not synchronous (, gray line). Cell counts (unpublished data), using a particle counter, demonstrated that the lag phase for both fractions was ∼4 h, consistent with previous studies (; ; ). For both the lower- and upper-fraction cells, we observed microscopically that the daughter cells did not separate from the mothers until the second cell division (4.5 h after refeeding). This could account for the time difference between budding and the increase in cell number. DNA-content analysis further demonstrated the synchrony of the quiescent cells (). The first cell cycle contains a single 1N peak at T that broadens and becomes an intermediate S-phase peak between 1 and 1.5 h. A 2N shoulder is evident by 2 h. The cells rapidly lost synchrony during the second cell cycle, with both 1N and 2N peaks present by 3.5 h (). Because the quiescent cells, like mammalian G cells, are obtained as a result of starvation and are synchronous upon reentry into the cell cycle, we conclude that they are also likely to be in a G state. Further work is needed to determine cell cycle–specific differences between cells in this state and those in the G phase of the mitotic cell cycle. Because lower-fraction, quiescent cells retained their reproductive capacity markedly better than the upper-fraction cells, we were interested in determining whether the quiescent cells were all of the same replicative age as assayed by their number of bud scars. Bud scars were measured by calcofluor white staining and flow cytometry. Flow cytometry gates were set to enrich for populations of cells with 0, 1, 2, or 3 or more bud scars (), and the number of bud scars on separated cells were verified microscopically (see the supplemental text, available at ). Microscopic evaluation of FACS-sorted cells revealed that 95% of cells sorted using the first gate (0 bud scar group) had 0 bud scars and 5% had 1 or 2 bud scars ( = 61). Of cells sorted using the second gate (1 bud scar group), 90% had 1 bud scar and 10% had 2 bud scars ( = 40). Finally, cells sorted using the third gate (2 bud scars), 81% had 2 bud scars, and 19% had 3 or more bud scars ( = 53). Therefore, the flow cytometry gates were most accurate for unbudded daughter cells. Flow cytometry of calcofluor-stained cells from unfractionated SP cultures revealed a broad peak of fluorescence that included cells with 0 bud scars, an extended region that included cells with 1 bud scar, and two additional peaks representing cells with 2 and 3 or more bud scars, respectively (). If all mother cells divided once after glucose exhaustion, 50% of cells in SP cultures would be expected to be unbudded daughter cells. A slightly higher than expected percentage of unbudded daughter cells (55%) was observed in cells from SP cultures (). Upper-fraction, nonquiescent cells exhibited a profile similar to that observed with SP cultures (, gray line; and ). In contrast, lower-fraction, quiescent cells, which were purified twice by density-gradient centrifugation, were predominantly daughter cells (91% had no detectable bud scars and only 9% had one bud scar; ). We conclude from these results that the quiescent fraction is predominantly composed of daughter cells produced in the first round of budding after glucose exhaustion, consistent with the appearance of the lower band () and that these cells comprised <50% of the cells in SP cultures in this separation. We also conclude that nonquiescent mother cells continue to bud during the postdiauxic phase, producing additional unbudded, nonquiescent daughters. This is consistent with the observation of budded cells in upper fraction 7 d after inoculation (). The colony-forming capacity of cells with different numbers of bud scars was also evaluated (). Quiescent, lower-fraction cells, which have either 0 or 1 bud scars were similar to cells from nonseparated, exponentially growing cultures in their ability to form colonies. There were no cells with >1 bud scar in this fraction. For nonquiescent, upper-fraction cells, all cells exhibited reduced colony-forming capacity; however, cells with 2 or more bud scars showed a slightly greater reduction. The overall loss of replicative capacity as compared with cells with 0 or 1 bud scars in nonquiescent cells suggests that the switch that limits the ability of these cells to reproduce occurs in most, if not all, upper-fraction cells and is relatively independent of replicative age. As expected, cells from SP cultures exhibited intermediate colony-forming capacities. Furthermore, the unbudded and singly budded cells in the lower, quiescent fraction are indistinguishable with respect to colony-forming capacity, suggesting that quiescence is not simply a property unique to virgin cells but, under some set of circumstances, singly budded, new mother cells can also acquire quiescent characteristics. Because glycogen accumulation occurred predominantly in lower-fraction, quiescent cells (, TEM– phosphotungstic acid), we tested a mutant that fails to accumulate glycogen because it lacks a glycogen-branching enzyme, to determine whether glycogen was responsible for the increase in density. There were no significant differences in cell number between upper fractions from parental or mutant strains (). However, the lower cell fraction from the mutant exhibited a small but significant decrease in cell number (). mutant strains did exhibit a slight decrease in density (), suggesting that glycogen accumulation does not contribute substantially to the increased density in these cells. These results are consistent with our finding that some mitochondrial mutants, such as and , which are also defective for glycogen accumulation ( Genome Database) are able to form lower bands (unpublished data). Chronologically aged SP cultures have been shown to display markers of apoptosis (; ). Aging cultures have also been demonstrated to accumulate ROS, which is typically a precursor to age-induced apoptosis (). We examined 7-, 14-, and 21-d-old cells to determine whether quiescent and nonquiescent cell fractions differed with respect to ROS (detected by dihydroethidium [DHE] staining) and three other markers for apoptosis (). ROS-positive cells were predominantly in the nonquiescent fraction. Lower-fraction, quiescent cells showed very low percentages of DHE-positive cells at all time points, reaching only slightly more than 15% by day 21 (). In contrast, 45% of the upper-fraction, nonquiescent cells were DHE-positive cells by day 7, increasing to >60% by 21 d (). As expected, cells from unfractionated SP cultures were intermediate between the quiescent and nonquiescent fractions (). ROSs have been shown to cause cell death during SP (; ), and it has been suggested that yeast caspase is required for generation of oxygen radicals during induction of apoptosis (; ). We were interested in determining whether ROS accumulation was an immediate response to glucose exhaustion at the diauxic shift and dependent on caspase encoded by . For this experiment, we tested both fractions from parental (BY4742) and mutant strains () over a 21-d period. Quiescent cells, as expected, exhibited very low ROS accumulation at all time points (). For the first 14 d, very few quiescent cells from either the parental or mutant strain were ROS positive and, by day 21, only 18% of parental cells and <10% of mutants exhibited ROS accumulation, indicating a slight dependency of ROS accumulation on Mca1p. As expected, upper-fraction, nonquiescent cells showed much more rapid and extensive ROS accumulation. However, this accumulation occurred 2 d after glucose exhaustion and, thus, is not an immediate response to the change in carbon source. ROS accumulation in the mutant was delayed significantly (P < 0.05) from that in the parental strain, but by day 21, ROS accumulation in both strains was indistinguishable. These results suggest that the function of Mca1p is transient, consistent with previous evidence for additional sources of caspase-like activity in aging yeast cultures (). Cells from nonseparated, SP cultures exhibited ROS accumulation that was typically intermediate between the two cell fractions (unpublished data). We conclude from these results that Mca1p has a positive but transient effect on ROS accumulation and that ROS accumulation is not fully induced by glucose exhaustion. Quiescent and nonquiescent cell fractions also differed with respect to other apoptosis-related markers: DNA fragmentation as detected by TUNEL staining () and externalization of phosphatidyl serine and loss of membrane integrity by Annexin V (AnnV) and propidium iodide (PI) staining (). TUNEL staining was more pronounced in upper- fraction, nonquiescent cells than in quiescent cells at both 7 and 21 d after inoculation (), indicating apoptotic DNA fragmentation in nonquiescent cells. Flow cytometric analysis of AnnV and PI staining allowed differentiation of nonquiescent cells into four subpopulations: nonstaining, viable cells; early apoptotic cells with intact membranes (AnnV positive); late apoptotic cells (AnnV + PI positive); and necrotic cells (PI positive). 11% of upper-fraction, nonquiescent cells were stained with either AnnV (early apoptosis) or PI (necrosis) by day 7, and >50% of these cells were stained by day 14 (), including ∼15% that were doubly stained, indicating late-apoptotic cells (). In contrast, <2% of quiescent cells were positively stained with AnnV by day 7, which increased to 6 and 14% by days 14 and 21, respectively (). We conclude that there is a significant difference between quiescent and nonquiescent cells with respect to the timing and extent of induction of apoptosis and necrosis. Furthermore, the stability of AnnV and PI staining at 14 and 21 d suggests that the nonquiescent cell fraction is heterogeneous with respect to the induction of apoptosis. We examined transcript abundance by microarray analysis in quiescent and nonquiescent cell fractions to gain further insight into these cell populations. Statistical ranking analysis of biological replicates revealed 68 and 266 transcripts that distinguished quiescent and nonquiescent fractions, respectively (P < 10; Table S1, available at ). Gene Ontology analysis revealed that 65 of the 266 transcripts in nonquiescent cells are known to be associated with DNA recombination (), including Ty-element transposition. These cells also accumulate transcripts encoding proteins involved in the resolution of stalled replication forks, including the DNA helicase Rrm3p (), the conserved Mus81p-resolvase binding partner Mms4p (; ; ), and the replication and DNA damage checkpoint proteins Mrc1p () and Rad17p (). Additionally, nonquiescent cells accumulate transcripts encoding Amn1p, a negative regulator of exit from mitosis that helps reset the cell cycle, and Clb2p, a G2 cyclin that binds to Amn1p (). Based on these observations, we hypothesize that nonquiescent cells are unable to arrest and that the consequences of entering S-phase under carbon starvation may contribute to the development of apoptosis in these cells. In the lower, quiescent fraction, the most significant Gene Ontology–process categories were response to water stress and energy metabolism (). Transcripts that differentiated quiescent from nonquiescent cells encode proteins involved in sensing and responding to ROS, such as Trr2p, a thioredoxin reductase involved in protection against oxidative stress and essential for survival during respiratory conditions (); Gtt1p, glutathione transferase; and Hyr1p (Gpx3), a glutathione peroxidase–like enzyme that functions as a hydroperoxide receptor involved in transducing a ROS-induced, redox signal to Yap1 (); Mcr1p, involved in ergosterol biosynthesis () and response to oxidative stress (); and Bcy1p, the regulatory subunit for cAMP-dependent protein kinase, which is essential for entry into SP () and activation of the oxidative stress response (). Quiescent cells also accumulate three transcripts involved in high-affinity iron transport (), siderophore activity (), and mitochondrial iron ion homeostasis (), underscoring the importance of iron, which is essential for the survival of mitochondrial function in cells in SP cultures (). We conclude from these results that studies of separated quiescent and nonquiescent cells from SP cultures provides cell type–specific information that cannot be determined from studies of unfractionated cultures alone. Two cell fractions are separable from yeast SP cultures by density-gradient centrifugation. Morphological and physiological characteristics, including DNA content and budding, indicate that cells in the denser, lower-fraction cells are quiescent and likely to be in a G state. The less dense, upper-fraction cells are heterogeneous, mostly nonquiescent, and, although viable as analyzed by FUN-1 staining, rapidly lose the ability to reproduce. About half of the nonquiescent cells in 14-d-old cultures appear to be apoptotic or necrotic, but a small percentage (∼10%) are thermotolerant, suggesting that there may be a few cells with quiescent properties. Further work is needed to determine the number of distinct populations present in the nonquiescent fraction and the regulation of reproductive capacity in these cells. These findings demonstrate, in the face of years of misconceptions about yeast SP cultures, that there is a quiescent and, likely, a G state in yeast that is very distinct from other cells in SP cultures. As cultures approach and enter SP, there is a process that leads to morphologically and physiologically unique populations of unbudded, quiescent daughter cells and nonquiescent cells. There is also a switch that significantly changes the reproductive capacity of mother cells in the nonquiescent population. Recent studies of stem cells in tissues and microniches () and on yeast colonies (; ) have demonstrated functional heterogeneity of cells as a function of position within a tissue or colony. Our observation of different life cycle trajectories within SP cultures is consistent with the hypothesis that both unicellular and metazoan species have evolved mechanisms for functional differentiation of cells in close proximity. This differentiation is likely to provide a selective advantage to microbes as well as to metazoans. More than 260 transcripts were identified that distinguish quiescent from nonquiescent cells. Interestingly, the most abundant transcripts that distinguish quiescent from nonquiescent cell fractions are not the same as those transcripts that distinguish SP cultures from exponentially growing cultures (; ; ). This important finding suggests that the transcriptional changes observed as cultures enter SP are likely to be physiological responses occurring in both quiescent and nonquiescent cells. One might consider that this is similar to a tissue-level response in metazoans. This further suggests that studying the development of cellular heterogeneity in living systems and the responses of functionally distinct populations at the genomic level will require examination of homogenous cell subpopulations in isolation. S288C (α ) was obtained from G. Fink (Massachusetts Institute of Technology, Cambridge, MA). S289 (α), an S288c diploid derivative, was obtained by mating. BY4742 (α Δ1 Δ0 Δ0 Δ0), BY4743 (αΔ1Δ1 Δ0Δ0 Δ0Δ0Δ0Δ0), and mca1 (α Δ1 Δ0 Δ0 Δ0 ) were obtained from Open Biosystems. W303 (α ) was obtained from L. Breeden (Fred Hutchinson Cancer Research Center, Seattle, WA). JC889-14B (α ) and the parental strain JC746-9D (α ) were obtained from J. Cannon (University of Missouri, Columbia, MO). Strains were cultured at 30°C with aeration in YPD + A (2% yeast extract, 1% peptone, 2% glucose, 0.04 mg/ml adenine, and 50 μg/ml ampicillin; ). RediGrad or Percoll density gradients (GE Healthcare) were prepared using the manufacturer's preformed gradient protocol with modifications. Percoll was diluted 9:1 (vol/vol) with 1.5 M NaCl, for a final NaCl concentration of 167 mM. To form the gradients, 10 ml of the Percoll solution was put into 15-ml Corex tubes and centrifuged at 13,800 RPM (19,240 ) for 15 min at 20°C. Approximately 2 × 10 cells (200 OD) were pelleted, resuspended in 1 ml Tris buffer, overlaid onto the preformed gradient, and centrifuged at 400 for 60 min in a tabletop centrifuge equipped with a swinging bucket rotor (Beckman Instruments) at 20°C. Fractions were collected, washed once in 40 ml Tris buffer, pelleted, and resuspended in ddH0 or conditioned medium for subsequent assays. For microarray and bud scar analysis, upper- and lower-fraction cells from the first fractionation were overlaid onto preformed gradients and centrifuged at 400 for 60 min, resulting in a two-step purification of the upper- and lower-fraction cells. For the time course separations (), glucose concentration was determined by using Precision glucose test strips (Precision Labs). For fraction quantification, ∼1.0 × 10 cells were pelleted and resuspended in 500 μl of Tris buffer and counted. For both the parental and glycogen mutant strains, 1.73 × 10 cells were overlaid onto three identical preformed gradients and centrifuged at 400 for 60 min in a tabletop centrifuge equipped with a swinging bucket rotor (Beckman Instruments) at 20°C. The upper cell fractions were collected into 50-ml conical tubes using a 1,000-ml pipetman. The remaining portion containing Percoll/NaCl/cells was vortexed and collected in separate 50-ml conical tubes. These fractions were washed once in 40 ml of Tris buffer, pelleted, and resuspended in 800 ml ddHO for quantification. Cell counts for each fraction were determined using the particle count and size analyzer. Values represent the mean (±SD) for three replicates per fraction per strain. A microscope (Optiphot; Nikon) equipped with a Phase 3 Plan 40 NA 0.70 DL objective was used for phase contrast light microscopy. Images were acquired with a digital camera (Coolpix 995; Nikon) affixed to the microscope with a 0.55× CCTV adaptor (Diagnostic Instruments). Differential interference contrast and fluorescent microscopy were done using an Axioskop 2 mot plus microscope (Carl Zeiss MicroImaging, Inc.) equipped with Plan-Apochromat 63× NA 1.40 oil (Carl Zeiss MicroImaging, Inc.) or Plan-Neofluar 100× NA 1.30 oil (Carl Zeiss MicroImaging, Inc.) objectives using DAPI (excitation, G 365, and emission, 445/50 [Carl Zeiss MicroImaging, Inc.], for Calcofluor white staining), FITC (excitation, BP 485/20, and emission, BP 515–565 [Carl Zeiss MicroImaging, Inc.], for AnnV staining), or Rhodamine (excitation, BP 546/12, and emission, FT 580 LP590 [Carl Zeiss MicroImaging, Inc.], for PI staining) filter sets. Images were acquired using AxioVision 4.4 software (Carl Zeiss MicroImaging, Inc.). Bud scar micrographs are composite images from Z stack acquisitions using AxioVision 4.4 software. Image postprocessing was done using Photoshop CS 8.0 and Illustrator CS 11.0.0 (Adobe). No image manipulations other than contrast, brightness, and color balance adjustments were used. For fixation of SP cultures (7 d after inoculation), ∼1 × 10 (100 OD) cells were pelleted and washed twice in distilled HO. Washed cells were resuspended in 1 ml of freshly prepared fixation solution (cold 2.5% EM grade glutaraldehyde in 0.1 M cold sodium cacodylate buffer, pH 7.2; Electron Microscopy Sciences) and fixed for 90 min on ice. The cells were pelleted, resuspended in 1 ml of fresh cacodylate buffer, and stored at 4°C until the cells were processed. Cells were embedded in Epon 812 (Shell) and examined in an electron microscope (CM12; Philips) as previously described (). Phosphotungstic acid staining to detect glycogen was performed as described previously (). For assays at 7, 14, and 21 d, ∼2 × 10 S288c cells from each fraction of separated cultures, cultures grown overnight, and “killed cells” ( and mutants grown for 14 or more days in YPD; ) were harvested and washed twice in glucose-Hepes buffer (2% [wt/vol] d-[+]-glucose [Sigma-Aldrich] and 10 mM Na-Hepes [Sigma-Aldrich]). Each sample was resuspended in 500 μl of the glucose-Hepes buffer containing 2 μl FUN-1 (Invitrogen) and incubated for 45–60 min at 30°C. Cells were diluted to 1 × 10 cells/ml in Isoton II (Beckman Instruments), and 30,000 cells per sample were analyzed with a flow cytometer (FACSCalibur; Becton Dickinson) using 488 nm excitation and collecting fluorescent emission with filters at 530/30 nm for FL-1 parameter and 585/42 nm for FL-2 parameter. CellQuest software (Becton Dickinson) was used for data collection and analysis. To establish flow cytometry parameters, 30,000 cells from cultures grown overnight (∼100% live) were assayed. Bright red (CIVS) and green fluorescent emissions were detected and plotted, and quadrants were established (Fig. S1 C, available at ) such that >99.9% of the fluorescent emissions from these positive control cells were contained in the top right quadrant. 30,000 killed cells stained with FUN-1 (Fig. S1 E), unstained overnight culture cells (Fig. S1 D), and unstained killed cells (Fig S1 F) were assayed and plotted as negative controls. >99% of all the negative control emissions were found in the bottom left quadrants. Error bars in represent the SD from the mean for = 4–5 measurements. Heteroscedastic, two-tailed tests were performed using Excel 2002 (Microsoft). Two-way ANOVAs were performed using Prism 4.01 (GraphPad Software). Unstained cells from samples used to assay live/dead status (see the previous section) were diluted, and 400 upper- or 200 lower-fraction cells were spread onto each of two YPD + A plates per sample and incubated at 30°C for 3 d. Colonies were counted, and the percentage of survival was calculated relative to a control sample of exponentially growing cells (overnight culture). Values represent the mean (±SEM) for 4–7 biological replicates. Heteroscedastic, two-tailed tests were performed using Excel 2002. Two-way ANOVAs were performed using Prism 4.01. Machine plating for colony-forming units () was done using a cell sorter (MoFlo; DakoCytomation) equipped with an automated stage. A YPD + A 10-cm plate was placed on the stage, and the sorter was adjusted so that the machine would deposit single cells in a 12 × 12 square grid on the plate. The plates were incubated for 3 d at 30°C. The percentage of colony-forming units was determined using the following equation: (the number of colonies that grew/144) × 100. Values represent the mean (±SD) for three replicate plates per sample. To test thermotolerance, 4 × 10 S288c haploid upper- and lower-fraction cells in glass test tubes were incubated for 0, 10, 15, or 20 min in a 52°C water bath. After the heat shock, the sample tubes were placed on ice until all samples were treated. For cell viability analysis, the cells were diluted and 400 cells from upper and 200 cells from lower fractions were spread onto each of two YPD + A plates per sample and incubated for 3 d at 30°C. The percentage of cells able to form colonies was calculated relative to plating identical numbers of untreated cells. 200 cells from S288c haploid cultures grown overnight and thermotolerance induced by incubation at 37°C for 1 h were used as a positive control for heat shock sensitivity. Values represent the mean (±SEM) for 5–6 biological replicates. tests and ANOVAs were performed as described in the previous sections. S288c haploid lower- and upper-fraction cells were examined microscopically for the presence of new buds. Three fields of ∼50–60 cells were examined per time point, and the budding percentage was calculated by comparing the number of cells with new buds to the total number of cells counted. DNA content was analyzed using a SYBR Green I staining protocol () with modifications. For analysis, 600 μl (∼4.5 × 10 cells) of S288c lower-fraction cells from a 7-d-old SP culture were harvested into tubes containing 900 μl of 100% ethanol (for a final ethanol concentration of 70%), fixed overnight at 4°C, washed twice in 1 ml Tris buffer, resuspended in 900 μl Tris buffer and 100 μl of a 10× RNase A solution (10 mg/ml RNase A and 100 mM NaOAc; Sigma-Aldrich), and incubated overnight at 37°C. Samples were pelleted, resuspended in 1 ml of freshly prepared pepsin solution (5 mg/ml pepsin in HO, pH adjusted, with 55 μl 1N HCl per milliliter of solution), and incubated at room temperature for 5 min. The samples were pelleted, washed twice in TE buffer (10 mM Tris base and 1 mM EDTA, pH 8.0), resuspended in 1 ml of a SYBR Green I staining solution (SYBR Green I stock solution [Invitrogen] diluted 1:10,000 in TE buffer and 0.25% Nonidet P40 [Sigma-Aldrich]), and stained overnight at 4°C in the dark. On the following day, the cells were washed twice in Tris buffer before dilution in Isoton II, and 30,000 cells per sample were analyzed on a FACSCalibur flow cytometer using 488 nm excitation and collecting fluorescent emission with filters at 530/30 nm for FL-1 parameter. CellQuest software was used for data collection and analysis. The voltage was adjusted to center the FL1-A G intensity peak at 200. Calcofluor white M2R 5 mM stock solution (Invitrogen) was diluted 1:200 in ddHO for a 25-μM working solution. Approximately 2 × 10 S288c cells from SP cultures and the separated upper- and lower-cell fractions were pelleted, resuspended in the working solution, and incubated for 90 min at room temperature in the dark. The samples were washed twice in ddHO. The cells were diluted to 1 × 10 cells/ml in Isoton II, and ∼30,000 cells per sample were analyzed with a flow cytometer (MoFlo; DakoCytomation) using 351 nm excitation and collecting fluorescent emission with filters at 450/65 nm for FL-6 parameter. Summit software (DakoCytomation) was used for data collection and analysis. DHE, TUNEL, AnnV, and PI staining were performed as described previously (, ). Cells were observed microscopically and quantitatively using flow cytometry. DHE stock solution (Invitrogen) was diluted 1:10 in PBS (Fluka) for a working solution. Approximately 1 × 10 S288c upper- and lower-fraction cells per sample were pelleted and resuspended in 100 μl of the YPD + A supernatant that had been filter sterilized. 1 μl DHE working solution was added to each sample and incubated for 3 min at room temperature in the dark. The samples were washed three times in PBS. The samples were diluted to 1 × 10 cells/ml in Isoton II, and 30,000 cells per sample were analyzed with a FACScan flow cytometer (CLONTECH Laboratories, Inc.) using 488 nm excitation and collecting fluorescent emission with filters at 585/42 nm for FL-1 parameters. CellQuest software was used for data collection and analysis. AnnV and PI costaining were done as described previously (). After staining, the samples were diluted to 1 × 10 cells/ml in Isoton II and 30,000 cells per sample were analyzed with a FACScan flow cytometer using 488 nm excitation and collecting fluorescent emission with filters at 530/30 nm for FL-1 parameter and 585/42 nm for FL-2 parameter. CellQuest software was used for data collection and analysis. Quadrants were established such that >99.9% of the autofluorescent emissions from unstained cells were contained in the bottom left quadrant. BY4742 cells were grown for 7 d and separated into upper and lower fractions using the two-step density-gradient protocol. RNA was isolated, and labeled cDNA was prepared and hybridized to 70-mer DNA oligonucleotide microarrays as described previously (). Transcript abundance was analyzed using GenePix 6.0 as described previously (). Six replicates from each fraction were used for a total of 12 microarrays. To identify genes strongly associated with either fraction, we calculated for each gene the difference between lower and upper gene expression for all microarrays in the dataset and then calculated the mean of these differences for each. If the mean difference for each gene is positive, this gene will be more highly expressed in the lower fraction. Conversely, if the mean paired difference is negative, this gene will be more highly expressed in the upper fraction. As a result, we transformed the mean difference in expression for each gene to a distribution () with − 1 degrees of freedom, where is the number microarrays, and generated a two-tailed p-value using the cumulative distribution function. We established a significance cutoff level of P = 0.001 to generate gene lists for the upper and lower fractions. The upper- and lower-fraction gene lists (Table S1) were queried, and annotations for gene processes were obtained from the Genome Database Gene Ontology (). A step-by-step description of this method can be found in the supplemental text. Fig. S1 shows flow cytometric scatter plots of FUN-1–stained upper- and lower-fraction cells from SP cultures, controls, and a representative micrographic image of a FUN-1–stained cell. Fig. S2 shows micrographs of bud scars from FACS sorting. The supplemental text gives a step-by-step description of the microarray ranking analysis. Table S1 provides gene lists for genes strongly associated with quiescent and nonquiescent cells from the microarray statistical ranking analysis. Online supplemental material is available at .

Development of the vertebrate skeleton ensues with the aggregation of mesenchymal cells into condensations. In endochondral ossification, these precartilaginous condensations prefigure the elements of the adult skeleton, and alterations in this program lead to a range of congenital skeletal abnormalities. Various members of the transforming growth factor-β family figure prominently at multiple stages within the skeletogenic program. Consequently, mutation or disruption of these genes or their receptors negatively impacts growth and development of the skeleton (; ). In particular, many of the bone morphogenetic proteins (BMPs) and growth and differentiation factors (GDFs) exhibit potent prochondrogenic activity. Indeed, several of the BMPs have been shown to stimulate ectopic cartilage formation both in vivo and in vitro, whereas disruption of , which is a BMP antagonist (of BMP2, -4, and, to a lesser extent, -7), leads to skeletal overgrowth (). Thus, it appears that BMPs may promote the expansion of cartilage, either by stimulating the proliferation of chondroprogenitors or chondroblasts, or by recruitment of cells with limited or no chondrogenic potential (). Downstream effectors of the BMP signaling cascade have been well characterized; however, the targets (direct and indirect) that underlie BMP action in chondrogenesis remain elusive. Other signaling molecules that influence the chondrogenic program include vitamin A and its metabolites, the retinoids (,). The retinoids act through the modulation of the transcriptional activity of the nuclear receptors for retinoic acid (RA), the RA receptors (RARs), and the retinoid X receptors. The transcriptional activity of these receptors is governed by RA availability, which is regulated largely by the combined actions of the enzymes involved in its synthesis and degradation, the ALDH1As and CYP26s, respectively. Indeed, the s, and other genes encoding enzymes involved in RA metabolism, are dynamically expressed during chondrogenesis. Furthermore, compound null mutants of the Rars and null mutants of , , or present with a spectrum of skeletal abnormalities (; ; ; ; ). The formation of precartilaginous condensations and the subsequent appearance of chondroblasts require the activity of , which is a transcription factor belonging to the Sry-related HMG box gene family (; ). In accordance with a proposed role for RA signaling in chondrogenesis, RA influences the expression and/or activity of (, ). In earlier studies, we demonstrated that mesenchymal cells isolated from a transgenic animal overexpressing a weak, constitutively active transgene exhibit skeletal defects, and this, in part, results from delayed or inhibited chondroblast differentiation (). Subsequent studies have indicated that the chondrogenic defect within the transgene mesenchyme is not rescued by the addition of BMP2 or -4, leading to the postulation that retinoid signaling may operate downstream of the BMP signaling pathway within this program (; ). Further series of experiments demonstrated that skeletal progenitor differentiation requires RAR-mediated repression, and that antagonism of RAR signaling in primary cultures of limb mesenchyme is accompanied by increased expression and activity (). In this study, we demonstrate that BMPs decrease RA availability, and that this is required for their prochondrogenic function. Furthermore, during autopod development, expression is dynamically expressed and influenced by BMP signaling. During this process, expression becomes progressively restricted to nonchondrogenic regions, where it likely serves to inhibit or suppress expression of a chondroblast phenotype. italic #text xref italic #text atRA, ketoconazole, diethyl aminobenzaldehyde (DEAB), cycloheximide, and actinomycin-D were obtained from Sigma-Aldrich; all compounds were dissolved in 95% ethanol, with the exception of ketoconazole, which was dissolved in deionized water. BMPs and GDFs were purchased from R&D Systems and resuspended according to the manufacturer's instructions. BMPs were added to media at a concentration of 10–20 ng/ml, whereas GDF5 was used at a concentration of 50 ng/ml to reflect its much lower activity (R&D Systems). AGN194310 was provided by R. Chandraratna (Vitae Pharmaceuticals, Irvine, CA). Reporter plasmids containing SOX9 binding sites (pGL3[4X48]) or trimerized RARE-luc were previously described (). pG5, which is a reporter that contains five GAL4 binding domains, was obtained from Promega. To generate GAL4–RAR(DEF) fusions, the C-terminal region of encompassing the ligand-binding domain was subcloned in-frame into pBIND (Promega). For expression in primary cells, full-length versions of murine (provided by M. Snead, University of California, Los Angeles, Los Angeles, CA) and were subcloned into a modified pSG5 (), and the construct containing an N-terminal EGFP fusion to full-length (provided by M. Privalsky, University of California, Davis, Davis, CA) was as previously described (). Mammalian expression plasmids containing and were provided by J. Wrana (University of Toronto, Toronto, Canada). Primary mesenchymal cultures from CD-1 E11.5 mouse limb buds were established as previously described () with the following modifications. For microarray analyses and other specified experiments, distal limb bud tips (subridge region) were dissected from E11.5 mouse embryos (the morning of the plug was considered E0.5) as described by , with the excised region extending 0.3–0.4 mm from the distal apex of the limb to the proximal cut edge. After proteolytic digestion, cells were filtered through a cell strainer (40 μM; BD Biosciences) to obtain a single cell suspension. Cells were pelleted and resuspended at a density of 2 × 10/ml and 12–15 10-μl aliquots of this suspension were plated into a 35-mm tissue culture dish (Nunc) and allowed to adhere for 1 h. After this period, 2 ml of culture medium consisting of 60% F12/40% DME and supplemented with 10% FBS (Qualified; Invitrogen) was added to each well with or without 20 ng/ml BMP4 (R&D Systems); this time was considered T = 0. Cultures were maintained for a period of up to 3 d; to minimize handling, culture media was replaced on alternate days. Cultures from either RARE-hsp68-lacZ (CD-1 background; provided by J. Greer, University of Alberta, Edmonton, CA) or Col2-EGFP mice (CD-1 background; derived from breeding of heterozygous transgenic males with CD-1 females) were established in a similar manner. Alcian blue staining, β-galactosidase staining, and the establishment and transient transfection of cultures derived from WL buds was performed as described in , ). In brief, cultures were transfected at day 0 and extracts for luciferase analysis were collected at day 2 unless indicated otherwise. For experiments involving primary limb mesenchymal cells alone or cells transfected with reporter genes only, factors or compounds were added at the time of media addition (T = 0). For experiments involving cotransfection with expression plasmids, factors and compounds were added 24 h after transfection. WL buds were collected in PSA from ∼E11.5 Col2-EGFP embryos. Affi-Gel blue beads (Bio-Rad Laboratories) soaked in either vehicle or BMP4 (20 ng/μl) for 2 h were transferred into the IDR of the limb buds. Limb buds were cultured on Nucleopore Track-Etch membranes (Whatman) at the air–liquid interface on top of stainless steel mesh in 12-well tissue culture plates. PSA was aspirated from each well and replaced with BGJb medium (Invitrogen) containing 10% FBS and antibiotics. The level of culture medium should not exceed the height of the membranes. Limbs were incubated under standard tissue culture conditions. After a 24-h incubation, EGFP expression was visualized using a dissection microscope (model MZ12; Leica) with epifluorescence. Limbs were subsequently fixed in 4% PFA before processing for whole-mount in situ hybridization. RNA was harvested from primary cultures using RNAeasy (QIAGEN) according to the manufacturer's instructions. For the zero time point, cells were allowed to attach for 1 h and were subsequently processed for RNA isolation. For other cultures, the media was gently aspirated, and any remaining media was blotted from the well before the addition of the lysis reagent. After isolation, the RNA was precipitated, and resuspended at 2 μg/ml; RNA quality was examined on a Bioanalyzer 2100 (Agilent), and the expression of and were measured using real-time PCR. To follow the expression of transcripts for , , and quantitative real-time PCR was performed using the 7900HT Sequence Detection System (Applied Biosystems). Primers and MGB probes (TaqMan) were designed using PrimerDesigner 2.0 (Applied Biosystems). The primer/probe sets used for detection of and were as described in and the following primer/probe sets were used for quantifying transcript abundance: forward primer, 5′-CTCCAACCTGCACGATTCCT-3′, reverse primer, 5′-CGGCTGAAGGCCTGCAT-3′, probe 6FAM-5′-CAGCGAAAGAAGGTG-3′-MGBNFQ. transcripts were detected using the forward primer, 5′-GGTATCCTCCGCAATGCAA-3′, reverse primer, 5′-GCGCATTTAAGGCATTGTAACA-3′, and probe, 6FAM-5′-CTGGGACAGTTTGGATC-3′-MGBNFQ. Primer and probe concentrations were optimized according to the manufacturer's instructions. Total RNA was isolated from primary cultures as described above. Quantification was performed using ∼25 ng of total RNA and the expression of all genes relative to endogenous was determined using TaqMan Ribosomal Control Reagents (Applied Biosystems) and the comparative C method as described in User Bulletin #2 (Applied Biosystems). Whole-mount in situ hybridization was performed on primary mesenchymal cultures as previously described () with the following modifications. After permeabilization using 10 μg/ml proteinase K in PBS supplemented with 0.05% Triton X-100, cells were postfixed in 4% PFA and 2% glutaraldehyde in PBS. Blocking and antibody incubations were performed in 1% blocking reagent (Roche) in 1× maleic acid buffer (100 mM maleic acid and 150 mM NaCl, pH 7.5). Riboprobes were synthesized in the presence of UTP-digoxigenin with the appropriate RNA polymerase and linearized template DNA, according to manufacturer's instructions (Roche). Riboprobe complementary to was generated from BamHI-linearized pBluescript containing 1.1 kb of the C-propeptide–encoding region of , and transcribed in vitro with T7 RNA polymerase. riboprobe was transcribed from EcoRI-linearized modified pBluescript containing a 400-bp fragment of the coding sequence using T3 RNA polymerase. Double in situ hybridizations were performed as above, but with the following modifications. Hybridization was performed using both DIG- and fluorescein-labeled RNA probes. Cultures were subsequently incubated overnight in fluorescein-labeled antibody (for detection of ). After incubation in NBT/BCIP/10% PVA–staining buffer, cultures were rinsed twice with PBS, and then fixed for 10 h in a 4% PFA/PBS solution. Cultures were again briefly rinsed twice in PBS, and the blocking antibody and staining steps were repeated, using an anti–DIG-AP antibody and INT/BCIP for detection of . Single and double whole-mount in situ hybridization on limb buds and cultured limbs was performed as described by . Double in situ hybridizations were performed as described in the previous paragraph, except that BCIP alone was used for detection of . To enhance visualization of the light blue (BCIP) and purple staining (NBT/BCIP), all images were adjusted to maximum hue in Photoshop (Adobe). Images of fixed cultures (in 70% ethanol or Tris-buffered salt solution with 0.1% Triton X-100 [TBTX]) were collected at room temperature using either a dissection microscope (Stemi SV11 Apo, S1.6× objective; Carl Zeiss MicroImaging, Inc.) or an inverted microscope (Axiovert S100; Carl Zeiss MicroImaging, Inc.) fitted with a CP-Achromat 5×, 0.12 NA, CP-Achromat 10×, 0.25 NA, LD Achrostigmat, 0.55 NA, 10× Fluar, 0.5 NA, and an LD Achroplan 40×, 0.6 NA. Monochromatic and color images were acquired from a QImaging Retiga Exi (12-bit) and a QImaging Retiga 1300i (12-bit) camera, respectively, using Openlab 4 software (Improvision). Live cell and organ culture images were collected at room temperature in culture media with either an Axiovert S100 microscope or a dissection microscope (MZ FLIII; Plan 1.0× objective; Leica). Photoshop was used to adjust some image levels. All luciferase assays were performed in quadruplicate and repeated using three distinct preparations of primary cells. Real-time PCR analysis was performed in duplicate and repeated a minimum of two times with independent RNA samples. Real-time PCR and luciferase reporter data were analyzed by one-way analysis of variance, followed by a Bonferroni posttest for multiple comparisons using GraphPad Prism, Version 4.0 (GraphPad Software, Inc.). Significance is represented as follows: *, P < 0.05; **, P < 0.01; #, P < 0.001. One representative experiment is shown for all luciferase and real-time PCR results.

Keratins are intermediate filament (IF) proteins that are preferentially expressed in epithelial cells and epidermal appendages (; ). In epithelial cells, the prominent IFs include keratins 1–20 (K1–K20), which are further classified into type I (K9–K20) and II (K1–K8) keratins, which form obligate, noncovalent type I/II keratin heteropolymers (; ). Unique keratin complements serve as cell-specific markers that distinguish different epithelial cell types. For example, basal epidermal keratinocytes preferentially express K5/K14, and suprabasal keratinocytes in the upper layer of the skin express K1/K10, whereas K8/K18 are the IF proteins of adult hepatocytes (; ). K8/K18 is also the prototype keratin pair of simple-type epithelia and, as such, is broadly expressed in epithelial components of glandular tissues, including the pancreas and intestine, with variable levels of K19, K20, and K7, depending on the cell type (). All IF proteins, including keratins, contain a central and conserved coiled coil–forming α-helical “rod” domain that is flanked by relatively nonconserved non–α-helical NH-terminal “head” and COOH-terminal “tail” domains (; ). The flanking head and tail domains are the more exposed portions of IF proteins, which explains why all IF phosphorylation sites reside in these domains (). Several in vivo K8/K18 phosphorylation sites have been identified that include K8 Ser23/Ser73/Ser431 and K18 Ser33/Ser52 (). K8/K18 hyperphosphorylation correlates with disease progression in patients with chronic liver disease (; ) and plays an essential role in regulating keratin filament organization, association with binding partners such as 14-3-3 proteins, and turnover (). Keratin mutations are associated with several skin, oral, esophageal, ocular, hair, and liver diseases that reflect the tissue-specific expression of the particular keratin (; ). The resulting disease-related tissue defects are manifestations of the clearly defined function of keratins that allows cells to cope with mechanical stresses. This keratin-related cytoprotective effect is most evident in the keratinocyte fragility phenotype of human epidermolysis bullosa simplex (EBS), which is caused by K5/K14 mutations, and is evident in the phenotypes of several animal models that lack or express mutant keratins (; ; ). Emerging evidence also indicates that keratins protect cells from nonmechanical injury via mechanisms that include keratin regulation of cell signaling cascades, regulation of susceptibility to apoptosis, and modulation of protein targeting to subcellular compartments (; ). For example, livers of K8- or K18-null mice or mice that express K18 Arg89-to-Cys (an EBS-like mutation) manifest a remarkable predisposition to injury and apoptosis (, ; ; ; ). K18 R89C and K14 R125 residues and their surrounding amino acids are highly conserved, and K14 R125 mutations cause the severest form of EBS and are the most common in keratin-related skin diseases (; ). Most human keratin-associated diseases are caused by autosomal-dominant keratin missense mutations with near complete penetrance, and most of these mutations are located at highly conserved regions at the ends of the rod domain (; ). Exceptions include mutations in K8/K18, which pose a risk for the subsequent development of cirrhosis and liver disease progression (, , 2005; ,), and may also be associated with inflammatory bowel disease (). All known human K8/K18 mutations do not involve the highly conserved ends of the rod domain. For example, the EBS-like K18 Arg89-to-Cys mutation, which causes hepatocyte fragility and predisposes to hepatocyte injury and apoptosis in mice (, , 2003), has not been found in humans, and it is hypothesized that such mutations are embryolethal (; ). The prevalence odds ratio for the association of K8/K18 mutations with human cirrhosis is 3.8 (95% confidence interval of 2.1–7.1), and the association with liver disease is highly significant when comparing a large American cohort of liver disease patients who underwent liver transplantation with a control group (P < 0.0001; ). In addition, a study using a large German patient cohort with chronic hepatitis C showed a significant association of exonic K8 variants with increased fibrosis (). However, direct evidence for the predisposition to liver injury via any natural human K8/K18 mutation has not been described. To address the in vivo significance of human liver disease–associated keratin mutations, we generated transgenic mice that overexpress wild-type (WT) or Gly61-to-Cys (G61C) human K8 (hK8) and compared their susceptibility to stress-induced liver injury. We targeted K8 G61 for the following reasons: (a) K8 G61 is highly conserved among type II keratins (), (b) K8 G61C is the second most prevalent among K8/K18 variants that are associated with cirrhosis and fibrosis progression (; ), and (c) in transfected cells, G61C interferes with keratin filament reorganization and cross-links hK8 under oxidative conditions (, ). The G61C mice unmasked an important relationship between K8 G61C mutation and K8 S73 phosphorylation by stress-activated protein kinases (SAPKs). This relationship was further explored by generating transgenic mice that overexpress K8 S73A. Transgene expression of the WT and G61C K8 mice was confirmed by blotting total liver homogenates with antibodies (Abs) specific to hK8, and by Coomassie staining of a cytoskeletal high salt extract (HSE) that shows the total liver keratin composition and distinguishes endogenous from exogenous keratins (). Importantly, protein levels of the endogenous mouse WT K8 and human G61C transgene product are similar, which mimics what is seen in human liver disease patients with heterozygous K8 G61C expression. Cross-linked hK8 is found only in G61C, but not WT, livers () because of the absence of Cys in WT hK8. Both WT and G61C mice were viable, fertile, and had no obvious phenotype under basal conditions. Previously described transgenic mice that overexpress high levels of WT K8, using the same WT hK8 we used, also do not have liver abnormalities under basal conditions, but develop pancreatic insufficiency that is possibly related to the WT hK8 expression level (). The K8 WT and G61C mice described herein do not have abnormal histology in their pancreata under basal conditions (unpublished data). We tested the consequence of G61C on susceptibility to liver injury using the Fas (which causes hepatocyte apoptosis; ) or microcystin-LR (MLR; which causes hemorrhagic hepatitis; ) injury models. Fas or MLR administration causes significant lethality (), and the G61C mice were markedly more susceptible to lethal liver injury as compared with nontransgenic and hK8 WT mice (∼80% G61C vs. ∼40% control; ). The increased lethality of G61C, as compared with WT mice, is caused by severe liver hemorrhage and apoptosis (). Although K8 G61C forms normal-appearing keratin filaments under basal conditions, Fas administration causes hepatocyte drop-off and a more prominent keratin filament collapse in G61C, as compared with WT mice (). Fas administration also modulates K8/K18 phosphorylation (), including an increase in K8 S73/S431 and a decrease in K18 S33 phosphorylation in all transgenic lines (). However, K8 S73 hyperphosphorylation () was significantly less in G61C, as compared with WT livers (65% less, as determined using quantitative blotting with anti-K8 pS73 Ab; not depicted). The increased apoptosis in K8 G61C-expressing hepatocytes was also confirmed by enhanced formation of the caspase-generated K18 fragment (). Therefore, the natural K8 G61C mutation causes a dramatic increase in transgenic mouse susceptibility to stress-induced liver injury. We compared the effect of K8 G61C (i.e., the patient-related variant) and K18 R89C mutation, or K8/K18 absence in transfected cells and mouse livers. K18 R89C is not a natural mutation, but K18 R89 is highly conserved among all IF proteins and the homologous residue is a mutation “hotspot” in epidermal keratin-related diseases such as EBS (). Overexpression of K18 R89C in transgenic mice causes hepatocyte keratin filament disruption, hepatocyte fragility, and enhanced susceptibility to liver injury and apoptosis (, ) and was the major clue that led us to search for, and then associate, K8/K18 variants with human liver disease (). When tested in transfected cells, K8 G61C behaves similarly to K18 R89C (a control for Cys mutation); both are highly insoluble and generate keratin cross-links upon oxidative challenge when compared with WT K8 and another natural K8 (R340H) variant (). The cross-linked K8 species are clearly detected in total lysates, but not cytosol (, lane 11 vs. 15), which indicates that most of these species are in the filament fraction. K8 G61C insolubility () and cross-linking after Fas or paraquat administration () were similarly noted in transgenic mouse livers. We then assessed hepatocyte fragility and the expression of several apoptosis-related proteins in K8 G61C mice as compared with K18 R89C, K8-, and K18-null mice. The K8- or K18-null mice, and K18 R89C mice, have a dramatically increased susceptibility to liver injury and apoptosis (; ; ). We examined the expression of several apoptosis-associated proteins because c-Flip protein, but not mRNA, was reported to be absent in K8-null mouse liver, and c-Flip absence may account for the increased susceptibility of K8-null livers to apoptosis (). We did not observe any difference in c-Flip (using two independent anti–c-Flip Abs) or several other apoptosis-associated protein levels between these four mouse groups and nontransgenic mouse groups (), which indicates that the previously reported results with c-Flip () were likely caused by differences in Ab specificity. We then examined hepatocyte fragility in K18-null, K8 WT, and K8 G61C mice upon liver perfusion because K8-null and the EBS-like K18 R89C livers have remarkably fragile hepatocytes (; ). Nontransgenic and K18 R89C mice were used as controls. Hepatocytes isolated from nontransgenic, K8 WT, and K8 G61C livers had 82–89% ( = 4–6 livers/genotype) viability as compared with hepatocytes from K18-null livers, which had only 16–22% viability ( = 3). Hence, the phenotypes of keratin-null or keratin mutant genotypes (summarized in ) suggest that K8 G61C alters hepatocyte function differently than when K18 is mutated at R89C or K8/K18 proteins are absent. This is despite the finding that both K18 R89C and K8 G61C decrease keratin solubility and cause cross-linking during oxidative conditions (). Human K8 includes three major in vivo phosphorylation sites (S23/S73/S431) that are conserved in mouse K8 (). S23 is phosphorylated under basal conditions, and S73/S431 are phosphorylated by SAPKs, such as p38, JNK, and p42 MAPK. p38 phosphorylates only S73 and generates a unique, slightly slower-migrating K8 species (termed HK8) upon SDS-PAGE, whereas JNK and p42 phosphorylate S73/S431 and generate both HK8 and K8 phosphospecies (; ; ). Given that the natural hK8 G433S alters K8 S431 in vitro phosphorylation by p42 MAPK (; ; likely caused by proximity), we tested the effect of several K8 mutations on K8 phosphorylation by p38/JNK/p42. K8 G61C blocks S73 phosphorylation by purified SAPKs, but does not completely eliminate S73 in vivo phosphorylation (likely via other kinases), as determined by blotting of transfected-cell total lysates () or transgenic mouse livers () with anti-K8 pS73–specific Ab. Some of the other K8 mutants also inhibited K8 phosphorylation (R453C [all three kinases] and G433S [p42]), but the most prominent effect was noted in G61C (). Moreover, K8 I62V, which is a variant found at higher frequency in controls as compared with liver disease patients (), has similar S73 in vitro phosphorylation by the SAPKs as WT K8 (unpublished data). We further substantiated the effect of K8 G61C by comparing K8 S73 phosphorylation in BHK cells cotransfected with WT or kinase-inactive p38. K8 G61C causes the near-complete absence of K8 S73 phosphorylation by transfected p38, which is similar to the near-absent phosphorylation of WT K8 upon kinase-inactive p38 transfection (). Hence, K8 G61C significantly inhibits K8 S73 phosphorylation in vivo by SAPKs. The effect of K8 G61C on S73 phosphorylation led us to hypothesize that the inability to phosphorylate S73 in K8 G61C mice is a major trigger for their increased susceptibility to apoptosis. We tested this hypothesis by generating transgenic mice that overexpress hK8 S73A and tested their predisposition to apoptosis. Expression of hK8 S73A was verified by blotting using anti-hK8–specific Ab and by detection of K8 S73 phosphorylation, after Fas administration, in WT but not S73A hK8 livers (). Fas administration increases K8 S431 phosphorylation in WT and S73A, but generates the K18 apoptotic fragment more prominently in S73A livers (). This suggests that inhibition of K8 S73 phosphorylation increases susceptibility to Fas-mediated liver injury, which was confirmed by the marked lethality of S73A, as compared with WT K8 and nontransgenic mice (). The increased lethality of K8 S73A mice is likely attributable to increased hemorrhage and apoptosis (), which mimics findings in K8 G61C mice (). K8 S73A and G61C mice share other features, including normal-appearing keratin filaments under basal conditions (), and similar cell viability of the isolated hepatocytes after liver perfusion (84–92% viability; = 5 livers). We hypothesized that G61C or S73A expression shunts phosphorylation from K8 S73 to other SAPK substrates after Fas stimulation. This hypothesis is based on the following: (a) cellular abundance of K8 (e.g., prominent Coomassie staining; ), (b) its role as an in vivo substrate for SAPKs (; ), (c) compensatory down-regulation of endogenous mK8 (i.e., the conserved SAPK K8 S73 substrate [S79 in mK8]) in response to mutant/WT hK8 overexpression ( and ), and (d) inhibition of K8 S73 phosphorylation in K8 G61C. We tested this hypothesis by first showing that p42/44, JNK1/2, and p38 phosphorylation (and, hence, activation) were markedly but similarly increased in response to Fas stimulation in WT and keratin mutant livers (), which supports previous reports that SAPKs play proapoptotic roles (; ; ). We then compared primary hepatocyte cultures for their susceptibility to apoptosis by testing for the presence of caspase-generated products after Fas stimulation (). The use of primary hepatocyte cultures eliminates the potential variability among mice and makes it feasible to test multiple time points. Cleaved caspases 3/7 were detected between 3 and 5 h, but, as expected from the intact animal studies ( and ), the cleaved caspase products were more evident in K8 G61C and S73A compared with WT hepatocytes (). Finally, we compared the phosphorylation of endogenous mouse proteins that are known to serve as SAPK substrates: c-Jun (S63/S73) by JNK, cAMP response element binding protein (CREB) and NF-κB p65 by p38, and p90RSK by p42/44 (; ; ). Although phosphorylation of NF-κB p65 was similar in K8 WT, G61C, or S73A hepatocytes after Fas treatment, phosphorylation of c-Jun, CREB, and p90RSK in K8 G61C and S73A hepatocytes was more pronounced and sustained when compared with WT hepatocytes (). The pattern of phosphorylated/activated SAPKs in K8 WT, G61C, or S73A hepatocytes after Fas treatment is relatively similar (), and activation by Fas is not as dramatic in primary cultures as it is in vivo (e.g., lanes 1 and 2 for phospho-JNK in , as contrasted with ), which is likely caused by the stress incurred upon hepatocyte isolation. Our data support the conclusion that increased phosphorylation of nonkeratin SAPK substrates in G61C and S73A hepatocytes reflects a shunting of phosphorylation toward these other substrates in association with caspase activation. The findings herein represent the first in vivo evidence that naturally occurring human K8 mutations can predispose to liver injury and apoptosis. Our results show that introducing hK8 G61C into mice predisposes to liver injury and apoptosis, and suggest the following sequence of events in response to stress: a G61C-mediated conformational change leads to the inability of K8 S73 to serve as a SAPK substrate, which creates an imbalance of kinase substrate availability, thereby predisposing to apoptosis and liver injury (). K8 G61C is the second most frequent liver disease–associated variant after R340H (; ). Mutations of IF proteins, including keratins, desmin, neurofilaments, and lamins, among others, are associated with a wide range of tissue-specific human diseases (; ). Compared with most other IF mutations, one unique characteristic of K8/K18 mutations is that they pose a risk of disease, rather than directly causing it (; ; ). The results herein fill an important missing link by experimentally demonstrating that a human disease-associated keratin mutation can, indeed, cause disease when an animal carrying the mutation is challenged by oxidative and other stresses. Therefore, the findings in K8 G61C transgenic mice provide strong supporting evidence for the human association studies that have been performed to date. The mechanisms of liver injury predisposition by different K8/K18 variants may be variant/disease-specific or may have mechanistic overlap. This is supported by the inability of other (non-G61C) K8 natural mutants (e.g., G52V, Y53H, I465-fs [], and I62V [not depicted]) to interfere with SAPK phosphorylation of K8. Furthermore, K18 R89C (in mice) alters a K8/K18 mechanical function (i.e., fragility predisposition) that is not shared by K8 G61C (). K18 R89C also primes hepatocytes to undergo apoptosis () and oxidative injury (), which are likely mechanisms that are shared by K8 G61C and K8 S73A. A common denominator for imparting protection from liver injury is keratin phosphorylation, which correlates with progression from chronic to end-stage liver disease (; ). This notion is supported by the finding that K18 S52A (S52 is a major hK18 phosphorylation site) expression predisposes to hepatotoxic injury in transgenic mice (), and by the results herein. The conformational link between K8 S73 phosphorylation and G61 is supported by the similar findings in K8 G61C and S73A mice, and by the inhibition of K8 S73 phosphorylation in vitro upon G61C mutation. This link is also supported by inhibition of binding of an Ab directed to a K8 G61C-containing epitope when S73 is phosphorylated (). Immunoblotting of liver homogenates from transgenic mice with this Ab showed its binding with WT and S73A, but not with G61C K8 homogenates (Fig. S1, available at ). K8 S73 phosphorylation also leads to a distinct retardation in migration during SDS-PAGE, which is specific to that phosphorylation site and is also seen when S73 is mutated to aspartate (; ). Hence, the phenotype we observe appears to be caused by mutation-related interference with a kinase–substrate interaction though other direct/indirect interference with potential stable keratin and kinase–kinase regulator interactions is possible. For example, p38 and JNK associate with keratins, but this association appears to be stoichiometrically limited and is more consistent with a kinase–substrate association (; ). In addition, both WT and S73A K8 coimmunoprecipitate with p38 MAPK, although we could not adequately test the K8 G61C mutant because of its limited solubility in mild detergents (unpublished data). Further support for a conformational link between K8 G61C and S73 phosphorylation is that the G61C mutation inhibits in vitro phosphorylation of K8 S73 by purified SAPKs, but does not significantly affect K8 S431 phosphorylation (). Lack of accessibility of SAPKs to K8 S73 upon G61C mutation may also be impacted by the formation of K8 cross-links in response to oxidative stress (; ). In this context, exposure to Fas or paraquat increases K8 cross-linked species that are otherwise barely detectable (). This reflects the change in the normally reducing cytoplasmic environment under stress, with production of reactive oxygen species during Fas-mediated apoptosis and other oxidative stresses as seen in rat hepatocytes () and T cell lines (). The effect of G61C on keratin solubility () may also independently contribute to kinase inaccessibility. K8 S73 phosphorylation is likely to be associated with several functions because it occurs during stress, apoptosis, and mitosis (; ; ). Previous studies in transfected cells () showed that K8 S73 phosphorylation promotes keratin filament reorganization (e.g., Ala substitution blocked stimulus-induced filament reorganization, which was rescued by Asp substitution). The multiplicity of keratin phosphorylation sites raises the untested possibility that site-specific phosphorylation can have a domino effect that “opens” the filaments to additional phosphorylation/dephosphorylation events, with consequent functional implications. The similarity of the K8 G61C and S73A transgenic mice phenotypes supports an important role for S73 as a phosphate sponge for SAPKs in normal tissues undergoing stress. Amongst the earliest descriptions of IFs potentially serving as “phosphate sinks” was the observation of significant vimentin and keratin hyperphosphorylation after short exposure of cells to okadaic acid (). Subsequent studies suggested that neurofilaments may serve as phosphate sinks (), which was supported by others, although in this case the sink model is not protective, as initially hypothesized (). We elected to use the term “sponge” instead of sink because it is more general, in that sponges are more easily transportable (i.e., dynamic) and not only collect spills but also allow their recovery (as free phosphates) with ready availability of fresh sponge capacity. We hypothesize that a phosphorylation sponge effect () may be detrimental or beneficial, in a context and phosphorylation site-specific fashion, and that in the case of K8 S73 the phosphorylation role is beneficial. This model does not exclude the possibility that some antiapoptotic kinase substrates may in fact become hypophosphorylated, but predicts that overall phosphorylation is shunted, with the net effect being enhanced apoptosis. The abundance of cytoplasmic keratins allows for plentiful sponge capacity. For example, K8/K18 can be easily seen by Coomassie staining upon HSE (), and they make up 5% of cultured colonocyte total proteins () and 0.2% of total mouse liver protein (; the 0.2% is an underestimate because it includes proteins from other resident nonkeratin-containing endothelial and Kupffer cells and some blood proteins). Also, K8 S73 is a unique and readily available SAPK substrate because it behaves as a switch that is either “on” or “off,” being completely unphosphorylated (off) under basal conditions but turning on via phosphorylation during apoptosis and cell injury (). SAPKs can play proapoptosis roles in several disorders, including liver (), neuronal (), and cardiac () disease. For example, disruption of JNK3 in mice results in resistance to excitotoxicity-induced neuronal apoptosis () and pharmacologic inhibition of p38 interferes with TNF-induced hepatocyte apoptosis (). Most of the known SAPK substrates are signaling molecules, such as protein kinases (e.g., MSKs, RSKs, and MNKs), transcription factors (e.g., c-Jun, Elk-1, c-Fos, and NF-κB), or apoptosis-associated proteins (e.g., Bim and Bad; ; ). Some protein kinases are phosphorylated and activated by SAPKs and then phosphorylate and activate transcription factors among other substrates. For example, p38 MAPK phosphorylates MSK1 that can phosphorylate the CREB transcription factor (). The increased phosphorylation of cellular proteins after Fas stimulation was not universal in G61C and S73A hepatocytes (e.g., phospho–NF-κB p65 was not altered; ). Several other substrates were tested, including MSKs, Elk-1, c-Fos, c-Myc, p53, and Bim, but the results were unrevealing because of the lack of cross reactivity of the respective Abs with mouse proteins (unpublished data). The K8 G61C/S73A-associated increase in c-Jun phosphorylation supports the enhanced susceptibility of G61C and S73A hepatocytes toward apoptosis, given that AP-1 members such as c-Jun are involved in proapoptotic or survival signaling depending on the cellular context and external stimulus. For example, stress-activated JNK phosphorylates c-Jun, which results in enhanced transcription of target genes (e.g., FasL) associated with apoptosis (), and c-Jun S63/73 mutations in mice protects neurons from apoptosis (). The role of CREB and p90RSK is well studied in cell survival, but less so in proapoptotic pathways. However, serial analysis of chromatin occupancy supports the involvement of CREB in several proapoptotic gene products such as DEDD, TRADD, GADD45γ, and Bim (). Several type II keratins contain a unique LL PL motif, which is LLPL in hK8 (i.e., S73-containing) or LLPL in K4/K5/K6 and hair keratins (; ). LL PL is phosphorylated during apoptosis and other stresses by SAPKs in several epithelial tissues, including the liver and intestine (K8) and the esophagus and skin (K4/K5/K6; ; ; ). Our findings suggest an important nonmechanical role for K8 in protecting hepatocytes from injury by serving as a phosphate sponge for SAPKs that can absorb some of their untoward effects (). This role may extend to keratins and their related diseases in other tissues where the K8 S73-containing LL PL motif is conserved, and is negatively impacted by G61C mutation and possibly other K8 or K18 liver disease–associated mutations. The reagents used include the following: MLR (Alexis Corp.), paraquat (Sigma-Aldrich), TdT-FragEL DNA fragmentation detection kit (Calbiochem), p38α MAPK (Upstate Biotechnology), JNK and p42 MAPK (Cell Signaling Technology), collagenase type II (Worthington Biochemical Corp.), and Lipofectamine (Invitrogen). All Abs to keratins and phosphokeratins were previously described (). Other Abs used were directed to Fas for mouse injection (BD Biosciences); to phospho- or nonphospho-p38, CREB, CREB pS133, p90RSK, phospho-p90RSK (human T359/S363 and mouse T348/S352), phospho-NF-κB p65 (human S536 and mouse S534), phospho-p42, JNK, and c-Jun (Cell Signaling Technology); to Fas for immunoblotting, FADD, and Bax (Upstate Biotechnology); and to Fas-ligand, nonphospho–c-Jun (Santa Cruz Biotechnology, inc.). Two independent Abs were used to detect mouse c-Flip; one directed to an NH-terminal region (SAEVIHQVEEALDTDE) that is 100% identical in human and mouse c-Flip (Upstate Biotechnology), and another directed to a mouse-specific COOH-terminal region (DKVYAWNSGVSSKEKYS) of c-Flip (Sigma-Aldrich). K8- and K18-null mice were provided by R. Oshima (The Burnham Institute, La Jolla, CA) and T. Magin (University of Bonn, Bonn, Germany), respectively. A transformer kit (CLONTECH Laboratories) was used to introduce single point mutations into a human (h) WT 12-kb genomic K8 clone (). The K8 genomic clone (provided by W. Franke, German Cancer Research Center, Heidelberg, Germany) includes endogenous regulatory elements that maintain tissue expression. Two mutant genomic constructs were generated (hK8 G61C or S73A), and both strands of the mutated region were sequenced to confirm the mutation. Fidelity of the mutant and WT constructs was verified by testing its expression by transient transfection into BHK cells. The 12-kb Sal I fragments of mutant or WT genomic hK8 DNA were then injected into pronuclei of fertilized FVB/n mouse eggs. Progeny mice carrying the hK8 transgene were chosen after PCR screening of tail genomic DNA, which was followed by breeding to select for germline transmission (primers of a 250-bp PCR fragment; 5′-GGCGGCGGCTATGGTGGGGCC-3′ and 5′-AGATGTGCATAGGGACCGGGA-3′). Two independent heterozygous mouse lines per construct were established and expanded (K8: WT and WT; G61C and G61C; and S73A and S73A), all in an FVB/n background, and then used for subsequent studies. The two transgenic lines for each construct had similar K8 expression and afforded near-identical results. For the lethality experiments, mice (age and sex matched) were fasted overnight, followed by intraperitoneal injection of Fas Ab (0.15 μg/g mouse body weight) or MLR (30 ng/g mouse body weight; ). Mice were killed by CO inhalation 4 h after Fas Ab injection, and their livers were isolated and processed for immunofluorescence, histology (HistoTec Laboratories), and TUNEL analyses (). For induction of oxidative injury, mice were fasted overnight then injected intraperitoneally with paraquat (70 μg/g mouse body weight; ), followed by harvesting of livers after 60 h. Hepatocyte isolation was performed by liver perfusion of three to six age- and sex-matched mice/genotype, using collagenase type II (). Cultured hepatocytes were treated with Fas Ab (0.5 μg/ml for 0.5–5 h), followed by preparation of cell lysates for immunoblotting. Keratins were isolated by HSE using liver pieces as previously described (). Alternatively, total liver homogenates were prepared by solubilizing in SDS-containing buffer. Proteins were separated by SDS-PAGE, followed by staining with Coomassie blue or transferal to membranes; they were then immunoblotted and visualized by enhanced chemiluminescence. Quantitative immunoblotting was performed using serial dilutions of the two samples to be compared and analyzed on the same gel. Immunofluorescence staining was done as previously described (), and fluorescence images were analyzed using a confocal microscope (MRC 1024ES; Bio-Rad Laboratories). BHK cells were transiently cotransfected (using Lipofectamine) with K18 WT and K8 (WT or mutant constructs; ) or K18 R89C and K8 WT (). 3 d after transfection, the cells were further cultured (37°C) in the presence or absence of 20 mM HO for 1 h, followed by isolation of a detergent-free, cytosolic soluble fraction or a total cell lysate (). The triple transient transfections with kinase active/inactive p38α and keratins were also performed in BHK cells (). In vitro phosphorylation was done as previously described (). K8/K18 immunoprecipitates, which were isolated from BHK-transfected cells, were washed with kinase-specific buffers (Cell Signaling Technology; Upstate Biotechnology), heated (90°C) to inactivate any bound kinase activity, and incubated with the kinases (p38, JNK, or p42) and γ-[P]ATP (). Kinase reactions were quenched by boiling in the presence of 2% SDS-containing buffer, which was followed by analysis by SDS-PAGE and autoradiography. Fig. S1 shows blotting with anti-hK8 G61C Ab that binds with WT and S73A, but not G61C K8, in transgenic liver homogenates. Online supplemental material is available at .

During the formation of synaptic connections, axons remodel and begin to assemble the machinery required for neurotransmitter release upon arrival to their synaptic targets. An intrinsic genetic program is likely to regulate the synthesis of synaptic components and their targeting to pre- and postsynaptic sites. It is becoming clear that the cross communication between the pre- and postsynaptic terminals is essential for the coordinated assembly at both sides of the synapse. Although great emphasis has been given to the role of membrane-bound proteins such as neuroligin–neurexin () and cadherins in this process, there is increasing evidence that secreted molecules such as Wnt, fibroblast growth factor (FGF), and TGFβ also play a crucial role in synaptic assembly and growth (; ; ; ; ). However, little is known about the mechanisms by which these signals regulate synapse formation and to what extent changes in synaptic assembly translate into function. Recent studies on neuroligin and neurexin have strengthened their role in synapse formation, as these molecules provide a local signal that stimulates synaptic assembly. Consistent with this notion, neuroligin and neurexin interact with components of the synaptic machinery (for review see ). In contrast, a distinct role for Wnts, FGFs, and thrombospondin has been proposed (). In this model, these secreted signals could indirectly regulate synaptic formation by accelerating neuronal maturation through changes in the transcription and/or translation of synaptic components (). Thus, synapses are formed through a sequence of events in which secreted factors stimulate neuronal maturation, thus, priming neurons for synapse formation followed by the focal action of membrane proteins that stimulate synaptic assembly. However, this model has not been fully tested and, hence, the precise role for secreted molecules in synapse formation remains to be established. Wnt signaling plays a key role in diverse aspects of neuronal connectivity by regulating axon guidance, dendritic development, axon remodeling and synapse formation (). In the cerebellum, is expressed in granule cells (GCs) at the time when mossy fiber (MF) axons, their presynaptic partners, reach the cerebellar cortex and make synaptic contact with GCs (). Upon contact, MF terminals are extensively remodeled, resulting in the formation of complex and elaborate structures called glomerular rosettes (). The extensive interdigitation of several GC dendrites into a single MF axon leads to a significant increase in the area of contact and is thought to contribute to some of the unusual functional properties observed at the MF-GC synapse (; ). These morphological changes are concurrent with the accumulation of presynaptic proteins and the formation of active zones. In -deficient mice, this extensive remodeling is affected, and the accumulation of the presynaptic protein synapsin I at glomerular rosettes is delayed (). Conversely, gain of function of Wnt7a increases the remodeling and clustering of synapsin I in MF axons in vitro (). Thus, Wnt7a functions as a retrograde GC signal that acts on MF axons to regulate presynaptic differentiation. The signaling pathway activated by Wnt7a during synaptogenesis remains unknown. During early development, Wnt proteins signal through at least three pathways (). Signaling through Frizzled and, in some instances through Ryk, has been shown to require the cytoplasmic scaffold protein Dishevelled (Dvl; ). However, the discovery of new Wnt receptors () raises the question of whether Dvl is an absolute requirement for all Wnt-mediated functions. In the β-catenin pathway, Wnts activate Dvl to inhibit the serine/threonine kinase Gsk-3β, resulting in the elevation of β-catenin levels and its subsequent translocation to the nucleus to regulate transcription (). In neurons, lithium, which is an inhibitor of GSK-3β, mimics the remodeling and presynaptic differentiation effects of Wnt7a suggesting that the canonical pathway might be involved (). However, the poor specificity of lithium as a Gsk-3β inhibitor poses the question as to whether Wnt7a signals through this pathway to regulate synapse formation. Thus, the downstream events leading to synapse formation induced by Wnts remain poorly understood. To address the mechanism by which Wnt7a regulates presynaptic differentiation and to determine the functional impact of deficit in Wnt signaling, we have examined the contribution of Dvl (). New studies have revealed that Dvl functions locally within a cellular compartment (; ) and can bind to microtubules to increase their stability in both dividing cells and differentiated neurons (). In addition, Dvl has been shown to regulate dendritic morphology in hippocampal neurons (). Loss of function of , which is one of the three mouse genes, results in abnormal behavior that is manifested by defects in social interactions (). However, the mechanism leading to this defect remains unexplored. Interestingly, although Dvl has been shown to be involved in synapse formation at the neuromuscular junction (), the role for Dvl1 at central synapses is unknown. mutant mice for possible defects in synapse formation and for the contribution of Dvl1 to Wnt7a function in this process. We found a functional interaction between Wnt7a and Dvl1 during the formation of the MF-GC synapse. mutant mice exhibit an enhanced defect in the localization of presynaptic proteins at MF terminals when compared with single mutant mice. Ultrastructural studies reveal that glomerular rosettes in the mutant are simpler, yet synapses still form with normal active zones. Importantly, electrophysiological recordings of double mutant mice reveal a decreased frequency of mEPSCs without changes in amplitude, indicating a defect in neurotransmitter release. Furthermore, we show by gain and loss-of-function studies that Dvl is necessary to regulate the formation of presynaptic clusters and synaptic vesicle recycling sites. Moreover, the presence of Dvl protein at presynaptic sites and its ability to increase the clustering of Bassoon, which is a cytomatrix protein involved in synaptic assembly, are consistent with the notion that Wnt regulates synaptic assembly through Dvl. Our studies also raise the interesting possibility of a role for Wnt signaling in synaptic function. To begin to dissect the mechanism by which Wnts regulate synapse formation, we first examined the distribution of Dvl in neurons and found that endogenous Dvl is present in synaptosomal fractions isolated from adult brains (). Interestingly, other components of the Wnt pathway, such as β-catenin and Gsk-3β, are also present in synaptosomes. The specificity of this fraction was confirmed by the presence of synaptic proteins such as PSD95 and synaptophysin (). To further test whether Dvl is present at synaptic sites, we examined the colocalization of Dvl with presynaptic markers in cultured hippocampal neurons. When expressed in hippocampal neurons, Dvl1-HA has a punctate distribution along the axon and colocalizes with endogenous synapsin I (). Importantly, endogenous Dvl is found in bright and small puncta, as well as in larger and diffuse areas along the axon (). Some of the bright Dvl puncta colocalizes with endogenous clusters of synapsin I and Bassoon (). To examine whether Dvl localizes to stable synaptic sites, Dvl1-HA and PSD95-GFP were expressed in 14 d in vitro (DIV) cultures. Some Dvl1 clusters colocalize with synapsin I in close apposition with PSD95-GFP (Fig. S1, available at ). Collectively, these findings suggest that Dvl is present presynaptically at central synapses. To test the function of Dvl in synapse formation, we examined the consequences of Wnt signaling on the formation of presynaptic clusters during the initial stages of synaptic assembly. Several studies have suggested that GC release factors, such as Wnt7a and FGF22, modulate synapse formation between MFs and GCs (; ). We tested the effect of GC conditioned media (CM) on the clustering of synapsin I, which provides a readout of presynaptic differentiation. We found that CM from cerebellar GCs induces the formation of numerous and very large clusters of synapsin I in MF cultures (Fig. S2 A, available at ). Wnt7a, which is released by GCs, also induces clustering, but to a lesser extent than GC-secreted factors (Fig. S2 A). Wnt7b, which is highly related to Wnt7a and is also expressed by cerebellar GCs (not shown), increases clustering of synapsin I puncta as observed with Wnt7a (Fig. S2 A). Thus, both endogenous Wnt7a and Wnt7b could regulate synapse formation between MFs and GCs in the cerebellum. The effect of Wnt7b could explain the rescue of the phenotype of the mutant at postnatal day (P) 15 (), as expression of Wnt7b increases in GCs at this stage (unpublished data). We tested whether inhibition of Wnt activity by addition of the secreted Wnt inhibitor Sfrp-1 (; ) was able to block the clustering activity of GC factors. Indeed, Sfrp-1 blocks most of the clustering activity of GCs (Fig. S2 B). These findings indicate that Wnts, possibly Wnt7a and Wnt7b, contribute significantly to the clustering activity of factors released by cerebellar GCs. To further examine the role of Wnts in presynaptic differentiation, we tested the effect of Wnts on several synaptic proteins. As Wnt7b CM exhibits a more reliable level of activity than Wnt7a CM, we have hereon used Wnt7b for gain-of-function studies. In MF axons, Wnt7b induces an 85% increase in the number of clusters labeled with VAMP2 (24 ± 2.3 clusters per 100 μm), compared with controls (13.3 ± 2.0 clusters per 100 μm; P < 0.05; ). Furthermore, the Wnt antagonist Sfrp-1 blocks Wnt7b activity, as the number of VAMP2 puncta (14.1 ± 1.0 clusters per 100 μm) is similar to control (). In addition, VAMP2 clusters are larger in cultures exposed to Wnt7b (80.4 ± 3.3 μm), compared with control (51.3 ± 1.8 μm) and Sfrp-1 (48.2 ± 2.8 μm; P < 0.05). We then examined the effect of Wnt on the clustering of Bassoon. In 10 DIV hippocampal cultures, Wnt7b induces a 28% increase in the number of Bassoon puncta with 22.8 ± 1.9 clusters per 100 μm, compared with 17.7 ± 1.5 clusters per 100 μm in control (). Conversely, Sfrp1 reduces the Wnt effect to almost control levels (18.2 ± 1.9 clusters per 100 μm). Western blot analysis shows that Wnt7b does not affect the level of presynaptic proteins such as synapsin I, VAMP2, Munc18, or syntaxin in cultured MFs (). In addition, β-catenin levels are unaffected (), suggesting that Wnt might signal through a β-catenin–independent pathway. Alternatively, Wnt could induce small changes in β-catenin levels, which are not detectable after overnight treatment. Collectively, these findings suggest that Wnt increases the clustering of presynaptic markers without affecting the overall levels of presynaptic proteins. To begin to address the mechanism by which Wnts regulate synaptogenesis, we examined the time course of Wnt action. We found that Wnt7b could induce clustering of VAMP2 within 15 min in MFs and hippocampal neurons without a visible effect on axon remodeling (Fig. S3, A–D, available at ). We then examined whether Dvl mimics the effect of Wnts on synaptic protein clustering. As explants of MFs cannot be easily transfected, we used hippocampal cultures. In EGFP-expressing control neurons, few synapsin I clusters are found along the axon (), which is similar to untransfected cells in these early cultures. In contrast, Dvl1-HA–expressing neurons exhibit a twofold increase in the number of synapsin I clusters (). Importantly, expression of Dvl also increases the number of Bassoon clusters by 50% (). Thus, Dvl1 mimics the effect of Wnts on the clustering of presynaptic markers, which is a readout of presynaptic differentiation. We then examined the consequence of loss of function. MFs isolated from wild-type mice contain 28.3 ± 1.6 clusters of VAMP2 per 100 μm (). In contrast, MFs from mutant mice exhibit a 57% decrease in the number of VAMP2 puncta (16.0 ± 1.2 clusters per 100 μm; P < 0.05; ). Furthermore, the average size of VAMP2 clusters in wild-type MFs is 53.9 ± 3.5 μm, whereas MFs from mutant have an average size of 37.3 ± 2.6 μm, demonstrating that the loss of function leads to a 44% decrease in the size of VAMP2 clusters (P < 0.01). Loss of function of both and in presynaptic differentiation also results in a significant decrease in the number of VAMP2 clusters to similar levels to mutant MFs (Fig. S3, E and F). This finding is consistent with the fact that isolated MFs from pontine nuclei have not yet been exposed to endogenous Wnt7a from GCs, as they were isolated before contacting GCs in the cerebellum. Interestingly, some presynaptic differentiation still occurs in mutant MFs. The expression of and in MFs (unpublished data) could contribute to the residual level of presynaptic differentiation in the mutant. Consistent with this view, we found that exposure of mutant MFs to Wnt7b does result in a small increase in the number of VAMP2 clusters, but does not fully rescue the phenotype to wild-type levels (). Collectively, these results demonstrate that is required for optimal Wnt signaling to mediate presynaptic differentiation. The aforementioned results suggest that Wnt signaling through Dvl1 regulates presynaptic differentiation by increasing the assembly of presynaptic sites. To test this directly, we have examined the effect of Wnt7b on synaptic vesicle recycling in MFs. Synaptic vesicle recycling was determined by the uptake of an antibody against the intralumenal domain of synaptotagmin (; ). Upon depolarization with a high potassium buffer, the synaptotagmin antibody is detected in vesicles corresponding to the readily releasable and reserve pools of synaptic vesicles, which have undergone exocytosis and endocytosis (; ). Consistent with a role in presynaptic assembly, Wnt7b increases the number of recycling puncta labeled by anti-synaptotagmin antibody by ∼61% (31.4 ± 2.2 puncta per 100 μm) when compared with controls (18.9 ± 2.8 puncta per 100 μm; P < 0.01; ). Importantly, the presence of Sfrp-1 significantly reduces the number of recycling vesicles when compared with Wnt7b CM (16.7 ± 3.1 puncta per 100 μm; ). Wnt7b also induces a significant increase (57%) in the size of recycling clusters (85.4 ± 4.7 μm) compared with control (49.1 ± 3.5 μm) or Sfrp-1–treated cultures (53.2 ± 2.8 μm; P < 0.05). Western blot analyses showed that Wnt7b does not affect the level of synaptotagmin protein (), indicating that the increased number of recycling puncta reflects a true increase in the recycling rate. Moreover, synaptic vesicle recycling measured by FM1-43 dye uptake also showed that Wnt7b increases recycling (). These results demonstrate that Wnt7b regulates the formation of functional presynaptic terminals. To test the requirement of Dvl1 in synaptic vesicle recycling, MFs from mutant mice were examined. We used pontine explants from P0 animals when MFs have not yet been in contact with -expressing GCs. MFs from mutant mice exhibit a 37% decrease in the number of recycling puncta (25 ± 1.4 puncta per 100 μm) compared with wild type (34 ± 2.0 puncta per 100 μm; P < 0.05; ). The recycling clusters are also significantly smaller in mutant MFs (22.9 ± 1.6 μm) compared with wild type (30.8 ± 2.2 μm; P < 0.05). These results indicate that is required in MF terminals for normal synaptic vesicle recycling. Our previous study demonstrated that -null mutant mice exhibit a defect in synaptic differentiation that is manifested by a delayed accumulation of synapsin I at MF terminals between P10 and P12 (). By P15, this defect is no longer evident (). This could be caused by compensation by other factors, possibly by other Wnts like Wnt7b, which are expressed in the cerebellum. Therefore, we decided to examine the consequence of the loss of function of during synapse formation. is highly expressed in the cerebellum (; ) and, thus, could mediate Wnt7a function. Importantly, MFs from mutants exhibit a defect in presynaptic differentiation. To test a possible genetic interaction between and , we also examined the cerebellum of double mutant mice. Interestingly, genetic studies in have shown that defects in hypomorphic wingless/Wnt mutants can be enhanced by mutations in (). Therefore, we predict that a concomitant loss of and would enhance the phenotype. We first analyzed , single mutant, and / double mutant mice for possible defects in the accumulation of synapsin I at P10. Indeed, mutants exhibit a defect in the accumulation of synapsin I at glomerular rosettes compared with wild type, as observed in the single mutant (). More importantly, double mutant mice exhibit a stronger defect than the single mutants at P10 (). Quantification of glomerular areas stained by synapsin I revealed a shift in the size distribution of synapsin I–stained rosettes in all the mutants (), which was made evident by the increase in the percentage of smaller (<40 μm) with a concomitant decrease of larger ones (>20 μm; P < 0.05 for single mutants, P < 0.01 for double mutants). We then examined glomerular rosettes at P15, which is the peak of cerebellar synaptogenesis. We found that the defect in presynaptic protein accumulation recovers by P15 in the single and mutants (). In contrast, double mutant mice still exhibit a significant defect in the accumulation of synapsin I at glomerular rosettes (). Quantification shows that double mutant mice exhibit an increase of ∼50% in the number of smaller synapsin I–stained glomerular rosettes (<40 μm) with a concomitant decrease in the number of larger ones (). Furthermore, other presynaptic proteins, such as synaptobrevin/VAMP2 (Fig. S4 A, available at ), SV2 (Fig. S4 C), and synaptophysin (Fig. S4 E), were also examined at P15. Although some defects are detected in large glomerular rosettes in single mutants (Fig. S4, B, D, and F), a clear defect is observed in the double mutant, as the number of small rosettes increases with a concomitant decrease in larger rosettes (Fig. S4, B, D, and F). These findings indicate that there is a genetic interaction between and at cerebellar synapses, as deficiency enhances the phenotype of in cerebellar glomerular rosettes. During postnatal development, a single MF makes contact with several GCs to form the glomerular rosette, a multisynaptic structure. Upon contact, GCs start to interdigitate into the MF terminal, resulting in increased perimeter and complexity of this terminal (). These morphological changes are correlated with synaptic maturation. Therefore, a delay in the accumulation of presynaptic proteins observed in and mutants could be caused by a delay in the maturation of MFs or by a decrease in the size of the presynaptic terminal. To investigate this more closely, we analyzed the morphology of glomerular rosettes by EM at P15. Our EM studies revealed that MF terminals have a simpler structure in the double mutant cerebella. We measured complexity, which was calculated based on perimeter squared divided by area (), as a parameter of the irregularity of the terminal. We found that the area and complexity of MF terminals () are not significantly different between wild-type (, A and A'), (, B and B') and (, C and C') single mutant mice (ranging from 32.30 ± 1.25 to 29.95 ± 1.09). In contrast, glomerular rosettes from double mutant mice exhibit a significant decrease in complexity (26.24 ± 0.63; P < 0.01; , D and D'). As the area of the terminals is not affected in double mutants, this finding indicates that the defect in complexity is caused by less interdigitation of GCs into the MF terminal. These results show that Wnt-Dvl signaling is required for the proper structural maturation of the MF-GC synapse in vivo. As the size of glomerular rosettes is unaffected in the mutants, the decrease in the area labeled with presynaptic proteins at glomerular rosettes () is caused by a defect in the localization of presynaptic markers. The simpler morphology of the glomerular rosettes and the delay in the accumulation of presynaptic proteins in double mutant mice led us to examine in detail for possible structural defects at the presynaptic terminal. Several parameters were measured, including the number of synaptic contacts per glomerular rosette, the number of docked and reserve pool vesicles, the width of synaptic cleft, the length of synaptic contact, and the thickness of the postsynaptic density. Docked vesicles were defined as those vesicles present within a distance of 50 nm of the active zone, whereas the reserve pool was defined as vesicles present up to 200 nm away from the active zone (). Morphometric analyses showed no significant differences in the number of synaptic contacts per glomerular rosette, in the number of docked or reserve pool vesicles between single mutant, double mutant and wild-type animals (). However, double mutant mice exhibit a mild, but significant, decrease in the width of the synaptic cleft (P < 0.01) and the postsynaptic density (P < 0.01; ). These results suggest that apart from a defect in complexity, active zones are structurally normal in single and double mutant mice. The changes in synaptic vesicle recycling in MFs observed in gain- and loss-of-function studies led us to test for possible functional defects at the MF-GC synapse. Acute cerebellar slices were used to analyze functional properties of the MF-GC synapse in the double mutant. We chose to perform electrophysiological recordings at P15, as this stage is the peak of cerebellar synaptogenesis. Importantly, P15 GCs are less heterogeneous than at earlier ages, thus, more consistent recordings can be obtained. Capacitance measurement during whole-cell voltage clamp recordings enables the measurement of estimated total membrane area (). We found that cerebellar GCs recorded from wild-type and double mutant mice had a similar membrane capacitance of 3.3 ± 0.3 pF ( = 10) and 2.9 ± 0.2 pF ( = 10), respectively. Thus, no significant differences in the size of cerebellar GCs between wild-type and double mutant mice were observed. The relatively small size of cerebellar GCs means that there is little dendritic filtering of the conductance change that results from activation of postsynaptic glutamate receptors. Therefore, analysis of miniature excitatory postsynaptic currents (mEPSCs) recorded in the presence of tetrodotoxin can be used to estimate the frequency of spontaneous vesicle fusion at the MF terminal. In whole-cell recordings from wild-type GCs, mEPSCs were clearly detected in 6 out of 10 recordings, whereas in the double mutant mEPSCs were detected in 8 out of 10 recordings. The mean mEPSC frequency in wild-type GCs was 0.23 ± 0.07 Hz (). In double mutant GCs, the mEPSC frequency was consistently lower than wild type, with an average mEPSC frequency of only 0.05 ± 0.02 Hz (, and G; P < 0.05; unpaired test). Interestingly, single or mutants showed mEPSC frequencies similar to wild type (not depicted). There were no significant differences in the mEPSC peak amplitudes (), 10–90% rise-time, or 50% decay recorded in the either of the single knockout strains or in double mutant mice (Table S1, available at ). Therefore, a similar number of postsynaptic glutamate receptors appear to be activated after vesicle release at the MF terminal. The kinetics of activation and deactivation of these receptors were not obviously affected in any of the mutants. These results show that deficiency in or alone does not affect synaptic function at this stage, but a deficiency in both and results in a significant decrease of mEPSC frequency, indicating a defect in the release of neurotransmitters. italic #text All mutant mice were maintained on a C57BL/6 background. double mutant mice were obtained from crosses of heterozygous (A. McMahon, Harvard University, Cambridge, MA) and mutant mice (T. Wynshaw-Boris, University of California, San Diego, La Jolla, CA). Genotypes were determined by three-primer PCR using tail DNA. For Wnt7a, the primers used were forward, 5′-TTCTCTTCGTGGTAGCTCTGGGTG-3′, reverse, 5`-CTCCTTCCCGAAGACAGTACGCTCT-3′, and the forward Neo primer 5`-AGGCCTACCCGCTTCCATTGCTCA-3′. For Dvl1, the primers used were forward 5`-CGCCGCCGATCCCCTCTC-3′, reverse, 5`-TCT GCCCAATTCCACCTGCTTCTT-3′, and the Neo primer 5`-AGGCCT- ACCCGCTTCCATTGCTCA-3′. mutant mice were dissected from newborn or P1 pups, cut into 100-μm explants, and cultured in 16-well LabTeck chamber slides, as previously described (). Explants were cultured for 2 d and then treated either overnight or for 15 min with control, Wnt7b CM, or Wnt7b CM plus Sfrp-1. Untreated mutant MFs were fixed after 3 d. Hippocampal cultures were prepared from E18 rats as previously described (). For transfections, a Nucleofector Amaxa system was used according to the manufacturer's protocol (Amaxa Biosystems). In brief, 3 × 10 neurons were electroporated, plated at 2 × 10 neurons per coverslip, and allowed to grow for 9 d before fixation. Constructs for transfections were EGFP, Dvl1-HA, and PSD95-GFP. CM from cerebellar GCs were prepared as described in the previous section and used to treat pontine explants. Control, Wnt7a-HA, and Wnt7b-HA CM were obtained from stably transfected Rat1b cells. Sfrp-1 CM in Fig. S2 was prepared from QT6 cells transiently transfected with myc-tagged Sfrp-1 construct (J. Nathans, Johns Hopkins University, Baltimore, MD) and used to block GC CM (). Neuronal culture medium (N2/B27) was added to cell lines and conditioned overnight. The levels of Wnt7b and Sfrp-1 proteins were assessed by Western blot. For other blocking experiments, 2.5 μg/ml of purified Sfrp-1 (R&D Systems) was preincubated with Wnt7b CM for 20 min at room temperature before addition to neurons. The intralumenal antisynaptotagmin antibody uptake assay was performed as previously described (a gift from P. de Camilli, Yale University, New Haven, CT; ) on MFs at day 3. In brief, pontine explants were incubated for 5 min at 37°C with warm high potassium depolarization buffer; Krebs-Ringer solution (125 mM NaCl, 25 mM Hepes, pH 7.4, 5 mM KCl, 2 mM CaCl, 1.2 mM MgSO, 1.2 mM KHPO, and 6 mM glucose) supplemented with 110 mM KCl and 10 μM APV and CNQX and contained either 10 μg/ml of rabbit antiintralumenal synaptotagmin antibody or 10 μM of fixable FM1-43 (Invitrogen). After incubation, cultures were washed three times in warm Krebs-Ringer buffer, and then fixed and processed for immunofluorescence or directly mounted for FM1-43. Images were captured with a microscope (model BX60; Olympus) or with a motorized microscope (Carl Zeiss MicroImaging, Inc.) using a charge-coupled device camera (Orca ER; Hamamatsu). The number, size, and intensity of synaptotagmin puncta larger than 0.16 μm were quantified using MetaMorph (Molecular Devices). As FM1-43 also labels nonsynaptic clusters, we used a more stringent threshold to select larger and brighter puncta, typically >0.20 μm. Axons were traced manually and the number of puncta per 100-μm neurite length and area (μm) were quantified. Each experiment was performed in triplicate. Hippocampal neurons and pontine explants were fixed with 4% paraformaldehyde, 4% sucrose in PBS (pontine explants have 0.5% EM grade glutaraldehyde added), and then permeabilized with 0.02% Triton X-100, followed by blocking with 5% BSA and incubation with primary antibodies for at least 1 h. When endogenous proteins were analyzed, neurons were fixed with methanol for 10 min at −20°C. Primary antibodies were against Dvl (), synaptobrevin/VAMP2 (Synaptic Systems), rabbit polyclonal intralumenal synaptotagmin (a gift from P. de Camilli), synapsin I (BD Biosciences), SV2 (Developmental Studies Hybridoma Bank), GAP-43 (Abcam), Bassoon (Bioquote Limited), and HA (Boehringer). The secondary antibodies Alexa Fluor 488 and 594 were obtained from Invitrogen. Each experiment was performed in triplicate and at least 15 images were taken per condition. The images were analyzed using MetaMorph. A threshold was set to capture clusters that were clearly distinguishable and did not merge with one another. The number, size, and intensity of the puncta were measured and Analysis of variance (ANOVA) statistical tests were performed. Cerebella from wild-type and mutant animals were fixed in 4% PFA, embedded in paraffin, and processed as previously described (). Antibodies against synapsin I, synaptophysin (CHEMICON International, Inc.), VAMP2, and SV2 were used. Localization of the proteins was visualized by the VECTASTAIN ABC peroxidase system (Vector Laboratories). Three independently bred animals were used for each of the four genotypes. An area of 2,700 glomerular rosettes per animal was measured. ANOVA statistical tests were performed. SDS-PAGE and immunoblotting were performed from pontine explants or from cerebella using standard methods. Pontine samples were analyzed for cytoplasmic synaptotagmin (Synaptic Systems), VAMP2, synapsin I, Munc18, and β-catenin (BD Biosciences). Gsk-3β (BD Biosciences) was used as a loading control, whereas P15 cerebellar homogenates were analyzed for CASK (CHEMICON International, Inc.), synaptophysin (CHEMICON International, Inc.), VAMP2, synaptosomal-associated protein of 25 kD (Synaptic Systems), synapsin I, β-catenin, and tubulin (Abcam). Synaptosomes were prepared as previously described (). In brief, adult mouse brains were homogenized in buffer A (0.32 M sucrose, 4 mM Hepes, pH 7.4, plus protease and phosphatase inhibitors) and centrifuged at 800 for 10 min. All procedures were performed at 4°C. The supernatant was clarified by centrifugation at 9,000 for 15 min, and the pellet containing myelin and synaptosomal and mitochondrial structures was resuspended in buffer A and layered on top of a discontinuous gradient containing 0.8/1.0/1.2 M sucrose in 4 mM Hepes, pH 7.4, plus protease inhibitors, and centrifuged at 82,500 for 90 min. Synaptic membranes were taken from a 1–1.2-M interface and resuspended in buffer B (0.32 M sucrose, 4 mM Hepes, pH 7.4, and 150 mM NaCl plus inhibitors). Proteins were quantified by Lowry assay. Small pieces from the internal granular layer of cerebellar lobes V, VI, and VIII of P15 animals were dissected and fixed in 3% EM grade glutaraldehyde in 0.1 M phosphate buffer, pH 7.3, at 4°C for at least 5 d. Samples from three wild-type, 3 , 3 , and 5 mice were analyzed. Tissues were processed at the EM Unit of King's College London. Photographs of whole glomerular rosettes were taken at 20,000×, whereas photographs of synapses were taken at 72,000×. Photographs were scanned and analyzed. Glomerular complexity was estimated using the formula perimeter/area (). Docked vesicles were considered as vesicles present within 50 nm (one vesicle diameter) of the active zone, whereas vesicles located up to 200 nm away from the docked vesicles were considered as the reserve pool (). ANOVA statistical tests were performed. Parasagittal slices (250 μm thick) were cut from P14–P15 cerebellar vermis of mice. The slicing solution contained the following: 125 mM NaCl, 2.5 mM KCl, 1 mM CaCl, 2 mM MgCl, 26 mM NaHCO, 1.25 mM NaHPO, 25 mM glucose, 0.02–0.08 mM -2-amino-5-phosphonopentanoic acid (-AP5), pH 7.4, when bubbled with 95% O and 5% CO. Patch-clamp recordings were made at room temperature (22–23°C), with an amplifier (Axopatch 700B; Axon Instruments). 50 mM picrotoxin and 1 mM strychnine was included in the recording solution to block GABA and glycine receptors, respectively, and in the presence 0.5 mM TTX to isolate mEPSCs. Patch pipette “internal” solution contained the following: 140 mM CsCl, 4 mM NaCl, 0.5 mM CaCl, 10 mM Hepes, 5 mM EGTA, and 2 mM Mg-ATP (adjusted to pH 7.3 with CsOH). Data analysis was performed using custom written software (Istvan Mody and Thorsten Hodapp, University of California, Las Angeles, Los Angeles, CA). Spontaneous mEPSCs were filtered at 2 kHz and digitized at 10 kHz. As all data were normally distributed (Shapiro-Wilk test), statistical differences between groups were tested using the test and considered significant at P < 0.05 (STATISTICA 5.1; StatSoft). Fig. S1 shows that Dvl1 colocalizes with synapsin I at synaptic sites. Fig. S2 shows that Wnt7a and Wnt7b have similar effects on synaptic vesicle clustering. Fig. ; double mutant (DKO) mossy fibers exhibit a defect in presynaptic clustering. Fig. S4 shows that ; double mutant exhibits a defect in the accumulation of several presynaptic proteins at cerebellar glomerular rosettes at P15. P15 animals. Online supplemental material is available at .

Neurons extend axons and dendrites, and these neurites select specific partners for establishing interneuronal connections. In the case of interactions between hippocampal pyramidal neurons, their axons are initially captured by filopodial protrusions from the dendrites of other neurons (; ; ), and the contact sites between the dendritic filopodia and axons gradually mature into synapses. At the same time, the filopodia are morphologically converted into the mushroom-shaped spines. Through these processes, stable axodendritic associations become established. In this type of neuron, dendrites do not appear to form functional contacts with other dendrites. Dendrites of some neurons even actively repel each other (; ) and avoid overlapping, which is called tiling (; ); multiple mechanisms seem to be involved in the tiling processes (; ; ). On the other hand, in limited classes of neurons, dendrodendritic synapses (; ) as well as dendrodendritic nonsynaptic junctions (; ) can form. These observations suggest that there are neuron type–specific mechanisms to promote or suppress the interactions between a selected pair of neurites. However, the question of how axons or dendrites can preferentially bind their specific partners remains to be answered. Formation of the contacts between axons and dendritic filopodia involves cadherin activities. Cadherins are homophilic adhesion molecules that function with their cytoplasmic (CP) partners, the catenins (). The cadherin–catenin complexes are accumulated at early axodendritic filopodial contacts and are retained in many of the mature synapses. Blockade of the cadherin–catenin system causes perturbation of synaptic differentiation (; ). This adhesion system was also shown to be important for the assembly of synaptic subcellular structures (), stabilization of synaptic contacts (), and activity-dependent synapse remodeling (; ). Based on their homophilic binding nature, cadherins theoretically can hold any combination of cells together, whether heterotypic or homotypic, if the cells express the same cadherin types (). In neurons, cadherins are localized in both axons and dendrites. Curiously, however, in many neurons such as hippocampal pyramidal neurons, although firm contacts between axons and dendritic spines are formed depending on cadherin activities, other types of contacts, such as dendrodendritic contacts, are not stabilized. We must ask why cadherins participate predominantly in the heterotypic (axodendritic) synaptic junctions but not in the homotypic dendrodendritic contacts even though this molecular family is in general used for linking the “like” cells. There should be some mechanisms for allowing cadherins to promote specifically axodendritic associations in these neurons. Some classes of molecules that have cell-binding activities are localized only in axons or dendrites, and their partners are present on the counter-neurites. For example, neuroligin is expressed by dendrites, whereas its ligand, neurexin, is localized in axons (). Such receptor–ligand systems should be able to facilitate the selective contacts between axon and dendrite but not those between dendrite and dendrite or axon and axon. In the case of neuroligin and neurexin, their molecular interactions have been implicated in synaptic differentiation (; ; ). Although the neuroligin–neurexin interaction has been shown to promote cell adhesion (; ), it remains to be defined whether these molecules are important for maintaining the physical associations between axons and dendrites. Nectins, forming a small subfamily of Ig domain proteins, show an asymmetrical distribution in synapses (). In the mossy fiber terminals of the hippocampus, nectin-1 (N1) is predominantly localized in the presynaptic membrane, and N3 is localized in the postsynaptic membrane, whereas their CP partner l-afadin is detected in both membranes (). Both N1 and N3 show homophilic binding abilities and can promote cell aggregation (; ). However, importantly, they also can bind one another heterophilically, and this heterotypic binding is much stronger than the homophilic one (; ; ). Furthermore, nectin interactions at cell–cell boundaries promote the recruitment of cadherin molecules to these sites (; ). These unique distributions and properties of nectins suggest that they may play an active role in the preferential contacts between axons and dendrites. In this study, we tested this idea and found that the trans-interaction of N1 and N3 indeed controlled the adhesion between these heterotypic neurites. Our results also explain why cadherins are active only for axodendritic connections. We first examined the distribution of N1 and N3 in rat hippocampal pyramidal neurons cultured for 4–6 d in vitro (DIV). Their axons and dendrites were identified by immunostaining for dephosphorylated tau () and MAP2 (), respectively. In isolated single neurons, both N1 and N3 were detected diffusely along their neurites, but they displayed distinct patterns of distribution: N1 was always more abundant in axons than in dendrites, whereas N3 was equally present in axons and dendrites (). When axodendritic contact formation began, both N1 and N3 became concentrated at early synaptic contacts formed on the tip of dendritic filopodia, overlapping with β-catenin, a representative of the cadherin–catenin complex. Concomitantly, the diffuse nectin signals disappeared from the axons (). Dendrites from different neurons did not form firm contacts to each other, but they occasionally happened to cross. At these dendrodendritic crossing points, N1 was not detectable (), and N3 was present around the dendrodendritic interfaces but was not particularly concentrated there (compare the faint N3 signals on these sites with those highly up-regulated at synaptic contacts on the same dendrite in ). β-Catenin was detected on some of the dendrodendritic crossing points () but not on all of them (<40% at 10 DIV). These dendrodendritic β-catenin accumulations became hardly detectable at later stages (e.g., at 14 DIV). In summary, N1 and N3 were preferentially concentrated at axodendritic interfaces. Once the nectin signals had been concentrated in the synapses, it became difficult to define whether these signals were derived from axons or dendrites. To determine their localization in mature neurons accurately, we transfected neurons with N1 or N3 cDNA, cultured them for 3 wk, and observed the distributions of the exogenously introduced nectins. Because of the overexpression, excess nectin signals were not restricted to synapses but were diffusely detected along neurites, allowing us to determine which neurites expressed these nectins. In N1-transfected cultures, N1 immunofluorescence signals were detected emanating from thin neurites that migrated on the culture plate as well as from those associated with dendritic processes (Fig. S1 a, available at ). The former population of neurites was identified as axonal because of their MAP2 negativity. Closer observations of the latter revealed that the N1 signals were localized along spine-free neurites running on the dendritic shaft (), suggesting that these also were axons. The majority of the dendritic spines in these cultures were N1 negative. On the other hand, N3 immunofluorescence signals evenly delineated the entire dendritic process, showing typical arrays of spines ( and Fig. S1 b), and their signals were barely detectable on MAP2-negative neurites. These findings indicate that N1 prefers to localize in axons at any developmental stage and that N3 localization becomes biased toward dendrites during development, which is consistent with the in vivo observation (). To study the role of N1 and N3 showing the aforementioned differential distribution, we examined the effects of N1 or N3 overexpression in more detail by observing neurons earlier after their transfection. Neurons transfected with N1 or N3 cDNA were cultured for 5–6 d and analyzed for the expression of exogenous nectins. Western blot analysis showed that the total level of N1 or N3 in these cultures significantly increased (Fig. S1 c). Immunostaining analysis revealed that exogenous N1 (exN1) was abundant in axons, but, as a result of overexpression, its relative level in dendrites appeared to have increased ( and ). Intriguingly, this N1 overexpression caused abnormal neurite patterning. Nontransfected pyramidal neurons, in principle, extended axons and dendrites radially from their soma (). In N1-overexpressing neurons, however, their axons often entwined around their own dendrites (, tau). Furthermore, many of their dendrites aberrantly touched each other, giving a looplike appearance (, MAP2; see for quantification). On the other hand, axons of neurons with exN3, which was distributed evenly among neurites (as seen for the endogenous N3 [enN3]), did not show such abnormal migration: when their axons happened to migrate onto their own bodies, they crossed them with a simple track (, arrow). A similar crossing was observed in nontransfected neurons, suggesting that this behavior was not caused by N3 overexpression. At 14 DIV, neurons expressing exN1 again exhibited extensive intraneuronal dendritic attachments, whereas those expressing N3 did so only at a minimum level (). We could not accurately trace axons in these older cultures, as tau distribution lost its continuity along the axons. These observations indicate that the overexpression of N1 but not N3 induced atypical sticking between neurites. To determine which domain (the extracellular [EC] or CP) was critical for the aforementioned activity of nectins, we expressed the EC domain of N1 (N1-EC) or N3 (N3-EC; ) in neurons and found that both constructs were not particularly effective in inducing aberrant neurite patterning (Fig. S2, a and b; available at ). Both of these molecules were detected on axons and dendrites, although N3-EC was homogenously distributed, whereas N1-EC tended to be clustered. We also expressed the CP domain of N1 (N1-CP) and N3 (N3-CP) but found no effects on neurite patterning. Both of these constructs tended to accumulate in the cell body regions, but a fraction of them also spread into neurites. Importantly, although N3-CP was widely distributed into MAP2-positive neurites, N1-CP was uniquely condensed along a single neurite extending from the soma of each neuron; these neurites had been identified as axons, as they did not react with anti-MAP2 antibodies except at the proximal region (Fig. S2, c and d). These results suggest that the CP domain of N1 was responsible for its axon-biased localization and also that this domain of N1 or its EC domain alone was not sufficient to exhibit the biological activities. To further investigate the roles of the EC or CP domains of nectins, we constructed chimeric molecules of N1 and N3, N13 and N31, by swapping their EC and CP domains (). When their transfectants were examined at 5–6 DIV, N13 (N1-EC + N3-CP) was detected on both axons and dendrites as clustered signals (), whereas the N31, having N3-EC + N1-CP, was localized more abundantly in axons (). This was reminiscent of the enN1 localization. These results support the idea that the CP domain is responsible for the axon-biased distribution of N1. The expression of N13 caused the severe entangling of axons along their own dendrites (), as in the case of N1 overexpression. In neurons expressing N31, their axons also showed a tendency to attach to their own dendrites (), but to a lesser extent as compared with the N13 expression (i.e., those axons simply crossed dendrites in most cases in contrast with the firm tangling of axons with dendrites in N13-expressing neurons). Aberrant attachment between dendritic processes was also induced by N13 expression but not N31 () and was confirmed by quantitative analysis (). These findings suggest that misexpression of the N1-EC domain, whether it is linked with its own CP domain or with the N3-CP domain, induces atypical neurite associations. We sought to understand how N1 misexpression induced the abnormal neurite interactions. Double immunostaining for N1 and N3 revealed that in N1-transfected neurons exhibiting atypical dendrodendritic contacts, exN1 molecules were concentrated together with enN3 at their contact sites (), suggesting that their heterophilic interactions were involved in inducing these phenotypes. In neurons transfected with N3, exN3 was unable to condense at the sites where their dendrites had happened to touch each other (). In these dendrites, exN3 was deposited along their noncontacting portions together with enN1. These observations suggest that only excess N1 molecules were able to accumulate themselves and their partner molecules into ectopic neurite contact sites and sustain their atypical associations. Whether exN1 also recruited the same nectin type remains to be determined because our antibodies to detect endogenous nectins could not distinguish between the exogenous and endogenous molecules. To test further whether the interaction between N1 and N3 was important for the aberrant neurite sticking, we mixed N1- and N3-transfected neurons in the same cultures. When transfectants with the different nectins happened to reside next to each other, their dendritic branches became deeply intermingled (). At their contact points, the two molecules were closely colocalized. Thus, the trans-interactions between overexpressed N1 and N3 molecules induced interneuronal dendrodendritic associations, which are not generally observed in hippocampal cultures except for their simple crossing (). Similar cocondensation of N1 and N3 was also found at axodendritic contact sites formed between these transfectants. For example, when an axon expressing exN1 migrates on other neurons with exN3, the exN3 molecules have sharply been concentrated along the axon (). In many such cases, noncontacting portions of the recipient neuron lost the exN3 signals, suggesting that the majority of N3 molecules expressed by the cell had been accumulated at the axodendritic contact sites. All of these results support the hypothesis that the N1–N3 interaction facilitates interneurite adhesions. In addition, we examined whether the nectin overexpression also affected synapse formation by immunostaining the aforementioned mixed cultures for synaptic markers and found that at the contact sites between N1-overexpressing axons and N3-overexpressing dendrites, the distribution pattern of synaptotagmin, a presynaptic marker, was not particularly altered (Fig. S3, available at ). This result suggests that the N1–N3 interactions enhance only the affinities between neurites but not their synaptogenesis, which probably requires additional machineries such as the neuroligin–neurexin interactions. We hypothesized that the role of enN1 localized in axons might be to promote the attachment of the axons to dendrites through its trans-interactions with dendritic N3 molecules. As a step to test this idea, we asked with which of the nectin types dendrites or axons preferred to interact. As neurite–neurite interfaces do not provide sufficient resolution for this analysis, we constructed a model system: we prepared HEK293 cells transfected with N1 (N1-293) or N3 (N3-293). These cells endogenously express N-cadherin, and the respective nectins were concentrated at their cell–cell boundaries (Fig. S4, available at ). We seeded neurons onto monolayers of these 293 transfectants and observed the distribution of each nectin at the interfaces between neurites and 293 cells. For accurate assessment of the specific effects of nectin expression, we used mosaic cultures of transfected and nontransfected 293 cells. Neurons extended their axons and dendrites onto the surfaces of these transfectants or nontransfectants. Immunostaining of these samples showed that when dendrites had attached onto N1-293 cells, N1 molecules derived from the N1-293 cells became intensely concentrated along the dendritic processes (). On N3-293 cells, the dendrites also recruited N3 molecules, but only faintly (; our antibodies against N1 and N3 could detect the antigens from these transfectants with similar fluorescence intensity; see Fig. S4). Thus, dendrites preferentially recruited N1 that was present on the counter–cell membranes. On the other hand, when axons had attached to N1- or N3-293 cells, these nectins were similarly concentrated along the axons, although not uniformly (), which is consistent with the finding that early axons expressed both N1 and N3. This result suggests that axonal N1 and N3 are ready to interact with the counter-nectins if these molecules are expressed on the surfaces of the adjacent cells. However, dendrites, which are the actual partners for axons, did not equally express these nectins and responded to them differentially, as shown above. Thus, as a result of the nonuniform distribution of N1 in neurites, a biased interacting system between dendritic N3 and axonal N1 seems to have been established (see ). To verify the hypothesis that the N1–N3 interaction regulates axodendritic association, we examined the effects of the genetic deficiency of N1 on neurite patterning by culturing hippocampal neurons isolated from N1 knockout mice (). Radial extension of axons and dendrites normally occurred in the mutant pyramidal neurons, and they did not display any aberrant patterning. However, when axodendritic contacts had begun, the mutant neurons came to exhibit atypical morphologies. Actin staining at 14–17 DIV revealed that their dendritic spines were unusually elongated or deformed, resulting in a smaller spine head (). In the cultures of wild-type neurons, their spine heads swelled, firmly attaching to axon fibers (, top). In contrast, in N1-deficient neurons, many of their dendritic spines, which exhibited filopodia-like morphology, did not associate with axons that could have been traced with their diffuse synaptotagmin signals (, bottom). All of these results suggest that the adhesive affinity between axons and dendritic spines, the major structures on pyramidal neurons to receive axonal input, was significantly reduced as a result of N1 deficiency. In more mature stages, synaptotagmin or β-catenin became concentrated onto their spine heads even in mutant neurons. Nevertheless, their signals were generally reduced, corresponding to the reduction in head size, and many of the spine or filopodial heads lost their association with synaptotagmin signals (). These results indicate that although N1 is dispensable for synapse formation, its absence impairs the normal process of axodendritic spine contacts and keeps them looser than usual even after their maturation. Because a role of nectins is to recruit cadherins, we tested whether the aforementioned activities of nectins involved cadherin actions. We found that whenever N1 and N3 were concentrated together at neurite contact sites, β-catenin was also recruited to these sites (see example in ), and N-cadherin showed a similar response (not depicted), confirming that nectin interactions promote cadherin-mediated adhesion. To examine how much the heterophilic N1–N3 and homophilic N1–N1 or N3–N3 interactions differ in their abilities to recruit β-catenin, we prepared mixed cultures of N1- and N3-293. In the original transfectants, either N1 or N3 was concentrated at cell–cell boundaries, although β-catenin was less tightly colocalized with N3 than with N1 (Fig. S4). In the mixed cultures, three types of interfaces—N1–N1, N3–N3, and N1–N3—were formed. Triple immunostaining for N1, N3, and β-catenin in these cultures showed the clear tendency that N1 and N3 were more intensely condensed together at the heterotypic boundaries between N1- and N3-293 cells than at the homotypic boundaries, and the β-catenin level proportionally increased in those heterotypic contact sites ( and Fig. S5, available at ). As a consequence, the junctional accumulation of these molecules was relatively decreased at N1–N1 or N3–N3 cell interfaces and, surprisingly, even disappeared from certain homotypic boundaries, causing the local separation of cells at these boundaries (Fig. S5). These results indicate that the homotypic and heterotypic cell boundaries compete for β-catenin recruitment and that the latter prevails. The aforementioned observations in a model system confirmed that the trans-interaction between N1 and N3 most effectively recruited cadherin–catenin complexes to cell contact sites. We also noticed that in N1 or N3-transfected neurons, the total level of N-cadherin slightly increased approximately two times per neuron in each transfectant (Fig. S1 c), suggesting that both nectins can stabilize cadherins irrespective of their abilities to induce excessive interneurite contacts. Therefore, we tested whether an increase in cadherin-dependent adhesiveness was sufficient to induce excessive neurite interactions by overexpressing N-cadherin in neurons. The total N-cadherin level per neuron increased two to three times in these cultures compared with untransfected cultures (). However, this N-cadherin overexpression had no effect on neurite patterning () even though cadherins had previously been shown to generate much stronger adhesiveness than nectins (), indicating that a simple increase in cadherin level or surface adhesiveness was not sufficient for inducing the atypical neurite interactions. Finally, we asked whether the cadherin–catenin adhesion system was required for the aforementioned actions of nectins. To test this possibility, we isolated hippocampal neurons from αN-catenin–deleted mutant mice in which cadherin activities were impaired () and transfected them with N1 cDNA. We first confirmed that mouse pyramidal neurons responded to N1 overexpression in a way similar to the rat ones (). Notably, when αN-catenin–deleted neurons had been used for transfection, their dendritic morphology was little affected by N1 overexpression (). We also found that the N-cadherin level was kept lower in αN-catenin–deficient neurons than in wild-type ones after the N1 transfection (Fig. S1 d). Thus, these results demonstrate that the cadherin–αN-catenin system was required for the actions of the nectins. We showed that N1 was preferentially localized in axons and that perturbation of its distribution by overexpression induced atypical associations between neurites. On the other hand, N3 was equally detected on both axons and dendrites, although this molecule appeared to prefer localizing on dendrites in mature neurons. Upon synaptogenesis, both N1 and N3 became concentrated together at axodendritic contact sites, whereas such condensation did not occur at the sites where dendrites crossed each other. In the 293 cell model system, the N1–N3 heterophilic interaction prevailed over the homophilic one, with more recruitment of β-catenin by the former. We also showed that dendrites more efficiently recruited N1 than N3 onto the counter–cell membranes, implying that the dendrites dominantly use their N3 molecules to interact with axons. Furthermore, N1-deleted neurons exhibited loosened associations between axons and dendritic spines. All of these results support the idea that the axon-biased localization of N1 and its trans-interaction with dendritic N3 plays a critical role in sustaining the normal association between axons and dendrites (). These actions of nectins required cadherin–catenin activities. Intriguingly, however, the overexpression of N-cadherin itself had no effect on neurite patterning. Thus, a cooperation of these heterophilic and homophilic adhesion systems is required for exerting their full activities, possibly generating unique mechanisms for linking the heterotypic pair of axonal and dendritic plasma membranes. The overexpression of N1 resulted in the excessive association of axons and dendrites derived from the same neuron. The formation of synapses by neurons onto themselves occurs normally, whose structures are known as autapses (; ). However, the overexpressed N1 appeared to have overly attracted axons and dendrites and, furthermore, induced atypical dendrodendritic contacts. Under these conditions, N1 molecules were leaked out to dendrites and ectopically condensed at dendrodendritic interfaces, recruiting enN3 to theses sites. This suggests that the mislocalization of excess N1 and its interaction with N3 was a primary cause for the induction of the dendrodendritic adhesions (). Once the level of N1 has increased in dendrites, this molecule should also be able to undergo substantial interactions with axonal N3, accounting for the excessive axodendritic associations (). This idea is supported by the observation that N13 exhibited similar effects. As N13 has the N3-CP domain, this chimeric molecule should have followed the N3 distribution, ensuring ectopic localization of the N1-EC domain to dendrites. Together, our results indicate that the proper localization of N1 is important for the correct neurite interactions. On the other hand, the overexpression of N3 had little effect. exN3 molecules did not accumulate at dendrodendritic interfaces, suggesting that these molecules cannot actively hold their attachments. It should be noted that the N3–N3 homophilic interactions are less effective in inducing cell aggregation than those of N1 (). We also noticed that N3 less efficiently recruited β-catenin to cell–cell contact sites compared with N1 in 293 cells; nevertheless, the total level of N-cadherin increased not only in N1- but also in N3-transfected neurons. Thus, we can speculate that this nectin can interact with the cadherin–catenin complex by itself but is unable to efficiently bring the complex into cell contact sites for some reason. Based on these observations, we suspect that N3 itself may not be a strong adhesion molecule and that it functions only significantly as a heterophilic partner for N1. Once the level of N3 has reached saturation with respect to N1, excess N3 molecules may not be able to exert additional biological effects. Although these two nectins were differentially distributed in axons and dendrites, they were not strictly confined to either of these neurites, particularly in early neurons. Thus, we can suppose that nectins or cadherins can also be used to promote dendrodendritic adhesion (). However, this form of adhesion was not observable unless N1 had been overexpressed. Is there any mechanism to exclude stable dendrodendritic attachment in the normal situation? We found that in mixed cultures of N1- and N3-expressing 293 cells, the N1–N3 boundaries collected greater amounts of N1 and N3 than their homophilic interfaces, sometimes causing cell separation at the latter interfaces. Such competition between the heterophilic and homophilic interactions of nectins also likely occurs in neurons, and the N1–N3 interactions at axodendritic interfaces could sweep away N1 or N3 from dendrodendritic interfaces, prohibiting their associations. Nectin interactions have been proposed to facilitate the accumulation of cadherins at cell–cell contact sites (; ). Consistent with this idea, β-catenin was always highly concentrated at the nectin-condensed sites in cultured neurons. At dendrodendritic crossing points where nectins were not concentrated, β-catenin was only transiently localized. Our results also showed that nectin overexpression could not induce aberrant neurite associations if αN-catenin–deficient neurons were used for transfection and, in addition, that for exhibiting the overexpression phenotype, N1 required the COOH-terminal domain that was the binding site for l-afadin, a mediator for the interaction between N1 and α-catenin (). These results suggest that nectins alone cannot function but that they need to interact with cadherin via the l-afadin–α-catenin complex. We further demonstrated that nectin overexpression up-regulated the N-cadherin level, but this effect was suppressed in αN-catenin–deficient neurons, suggesting the possibility that the role of N1 is to stabilize or up-regulate cadherin via binding with α-catenin. Importantly, however, the overexpression of N-cadherin itself had no effect on neurite patterning. Moreover, not only N1 but also N3 could up-regulate the N-cadherin level in their transfectants even though only N1 was active in altering neurite adhesiveness. These suggest that the real role of nectins was not simply to up-regulate the level of cadherins, although cadherin up-regulation might have been a prerequisite for the nectin actions. It is known that trans-nectin interactions activate small GTPases in their CP domain–dependent manners and also that these small GTPases can facilitate cadherin activities (; ; ). Thus, nectins may cooperate with cadherin through such physiological cross talks in addition to its up-regulation. A similar cooperation of cadherin and an Ig domain protein, echinoid, was found to occur in cell–cell adhesion in imaginal discs (). Intriguingly, the CP domain of echinoid resembles that of nectins, as both can bind l-afadin. Transfection experiments thus far published suggest that the cadherin–catenin complex can solely function for cell–cell adhesion (), but its activity seems to be modulated by these Ig domain molecules. Our in vitro analysis of N1-deleted neurons provides loss of function evidence that N1 is required for proper axodendritic interactions. In the absence of N1, the attachment of axons to dendritic spines appeared significantly loosened. In vivo analysis of the hippocampus in N1 knockout mice demonstrated that a population of axons from the dentate gyrus failed to terminate at the correct portions of CA3 neurons (), supporting the idea that N1 is required for axons to properly recognize and attach to their target dendrites. On the other hand, synaptic protein assembly more or less occurred in the N1-deleted neurons both in vitro and in vivo, indicating that nectins are dispensable for synapse formation itself. As a certain level of β-catenin is still detectable on the N1-deficient synapses, residual cadherin–catenin complexes or other cell adhesion molecules may serve or compensate for maintaining their remnant synaptic contacts. It would be intriguing to test the effects of the double knockout of the cadherin and nectin systems on synapse formation in future studies. Nectins are widely but not ubiquitously expressed in the brain (). Other ligand receptors may also play a role in regulating axodendritic associations. It is important to note that some classes of neurons can form dendrodendritic adherens junctions or synapses (; ). Therefore, our final goal should be to identify neuron type–specific mechanisms, which control the adhesive affinities between neurites, for a deeper understanding of interneuronal recognition mechanisms. Mice in which exon 2 of the N1 gene had been replaced with the neomycin resistance gene () were maintained on a C57/BL6 background. The genotyping methods for αN-catenin knockout mice were described previously (). Rat hippocampal neuronal cultures were prepared from embryonic day (E) 18 rat embryos by using previously described methods () with some modifications. In brief, hippocampi were dissociated by trypsinization and trituration and were plated at 5,000–10,000 cells/cm onto poly--lysine–coated glass coverslips. Cultures were maintained in DME F-12 with 2% B27 supplements (Invitrogen) and 5% horse serum or in neurobasal medium (Invitrogen) with 2% B27 supplements. Cytosine arabinoside was added after 3 d to inhibit glial proliferation. or wild-type mice were prepared from E16–17 embryos according to the methods used for the rat hippocampus. Cultures of neurons from αN-catenin knockout mice were prepared separately from individual mouse embryos at E16–17, and those of mutant neurons were selected after genotyping of the original embryos. Neurons were transfected with various DNA constructs by using an electroporation device (Nucleofector 1; Amaxa). For transfection, neurons were suspended at 250,000–3,000,000 cells/transfection in 40–100 μl of the Amaxa nucleofector solution and electroporated with 1–2.5 μg DNA. cDNAs for mouse N1α and N3α were used throughout the experiments. For the construction of N13, cDNA fragments encoding amino acids 1–379 of mouse N1α and amino acids 429–549 of mouse N3α were amplified by PCR by using primer sets of 5′-ATGGCTCGGATGGGGCTTGCCG-3′ and 5′-GCAGGGCCACTATGATCCCTCCGAC-3′ as well as 5′-GACGGACGTTTCGTGGAGA-3′ and 5′-TTAGACATACCACTCCCTCC-3′, respectively. For the construction of N31, cDNA fragments encoding amino acids 1–427 of mouse N3α and amino acids 381–516 of mouse N1α were amplified by PCR using primer sets of 5′-ATGGCGCGGACCCCGGG-3′ and 5′-GATAGCAGAATACCCCAGCTAAAA-3′ as well as 5′-GCCGGCACACCTTCAAG-3′ and 5′-CTACACATACCACTCTTTCTTG-3′, respectively. Obtained fragments were ligated through the underlined SalI sites. To construct the expression vector for N13 and N31, we subcloned each cDNA fragment into pCA-pA using a HindIII and NheI linker. The Flag-tagged nectin CP region of N1α (pCA-N1CP; 356–512 residues) and N3α (pCA-N3CP; 404–545 residues) were constructed by using pCA-Sig-pA. For the construction of pCA-N1EC-EGFP and pCA-N3EC-EGFP, cDNA fragments encoding amino acids 1–377 of mouse N1α and 1–425 of N3α were ligated into pCA-EGFP-pA, respectively. The generation of other constructs was described previously (). Cells on coverslips were fixed in 2–4% PFA in HBSS with 4% sucrose for 10–15 min at room temperature or 37°C. After treatment with 0.25% Triton X-100 in TBST (TBS with 0.005% Tween-20) for 5 min at room temperature, the cells were blocked with 5% skim milk in TBST at 37°C and exposed for 2 h to primary antibodies in 5% skim milk in TBST at room temperature or 37°C. Primary antibodies were visualized with goat fluorochrome-conjugated secondary antibodies. The fluorochromes used were AlexaFluor350, -488, -555, -647 (Invitrogen), and Cy3 (Chemicon). F-actin was visualized by use of AlexaFluor488-conjugated phalloidin (Invitrogen). Rabbit anti-N1 and anti-N3 antibodies were raised against the CP portion of mouse N1α and N3α proteins, respectively, and were affinity purified by using standard protocols. These antibodies cross reacted with rat nectins and were used to detect endogenous rat nectins. Rat monoclonal anti–mouse N1 (clone 48–12; MBL International Corporation) and anti–mouse N3 (clone 103-A1; MBL International Corporation) antibodies, which recognized the EC regions of N1 and N3, respectively, were used to detect exogenously introduced mouse nectins. These monoclonal antibodies did not immunocytochemically detect rat endogenous nectins in cultured neurons. Other antibodies used were mouse monoclonal anti-MAP2 antibody (clone HM-2; Sigma-Aldrich), rabbit anti-MAP2 antibody (Chemicon), mouse monoclonal anti–tau-1 antibody (clone PC1C6; Chemicon), mouse monoclonal anti–β-catenin antibody (clone 5H10; a gift from M.J. Wheelock, University of Nebraska, Omaha, NE), rabbit anti–β-catenin antibody (Sigma-Aldrich), rat anti-GFP antibody (Nacalai Tesque), rabbit anti-Flag antibody (Sigma-Aldrich), mouse monoclonal anti–N-cadherin antibody (Transduction Laboratories), and mouse antisynaptotagmin antibody (Chemicon). Neuronal cultures were prepared in 35-mm petri dishes, and their lysates were analyzed by SDS-PAGE in which the total protein concentration had been adjusted to be equal for each lane. Proteins were transferred to a nitrocellulose membrane, the membrane was blocked with 5% skim milk for 1 h, and membranes were incubated overnight at 4°C with anti-nectin or anti–N-cadherin antibodies in Can Get Signal solution (Toyobo). Blots were washed with TBS, incubated for 1 h in HRP-conjugated goat anti–mouse antiserum (1:5,000; Jackson ImmunoResearch Laboratories), and visualized by exposing to X-ray films after treatment with ECL Plus Substrate (GE Healthcare). The signals on the films were digitally scanned and analyzed by using Scion Image densitometric analysis. Images of neurons were obtained with a confocal microscope (LSM510; Carl Zeiss MicroImaging, Inc.) equipped with a 63× NA 1.4 or a 40× NA 1.3 lens using LSM510 software (Carl Zeiss MicroImaging, Inc.), and their morphology was analyzed with the same software and with Adobe Photoshop. For quantification of the dendritic arbor pattern, the number of dendrites that elongated directly from the cell body was first counted, in which measurement dendrites shorter than the diameter of neuronal soma were omitted. The dendrite processes were then manually traced to measure their length by LSM software; the two longest dendrites were chosen for this measurement. The number of branches protruding from these dendrites was also manually counted. To obtain the circle-crossing index of dendritic arbors, we superimposed a circle of 40 μm in diameter on the center of the cell body of each neuron. Then, the number of dendrites crossing the circle was counted and plotted; subsequently, Welch's t test was performed. In general, several neurons were randomly chosen from multiple culture plates for each assay. Neurons at 7–8 DIV were used for these analyses. Fig. S1 shows the localization of exogenous nectins in nectin-transfected mature neurons as well as the effects of nectin overexpression on the N-cadherin level. Fig. S2 shows the effects of expression of nectin mutants on neurite patterning. Fig. S3 shows the effects of nectin overexpression on synaptotagmin distribution. Fig. S4 shows nectin and β-catenin distribution in nectin-transfected 293 cells. Fig. S5 shows nectin and β-catenin distribution in a mixed culture of N1- and N3-transfected 293 cells. Online supplemental material is available at .

Natural killer (NK) cell activation is regulated by a balance between activating and inhibitory cell surface receptors (; ). Consistent with the missing self hypothesis (; ; ), NK cell cytotoxicity can be inhibited by engagement of inhibitory receptors specific for major histocompatibility complex (MHC) class I proteins, including killer Ig–like receptors (KIRs; ; ; ; ). Initiation of the inhibitory signal upon ligand binding requires the phosphorylation of two tyrosine residues within immunoreceptor tyrosine-based inhibition motifs (ITIMs) in the cytoplasmic domain. These phosphorylated tyrosines act as a recruitment site for SH2 domain–containing tyrosine phosphatases, including Src homology protein tyrosine phosphatase (SHP) 1 or 2 (; ; ; ). Several signaling molecules involved in NK cell activation can be targets for SHP-1– and SHP-2–mediated dephosphorylation, including Zap70, Syk, PLCγ, LAT, and SLP76 (for review see ). However, using a transfectant of YTS expressing KIR2DL1 fused to a substrate-trapping mutant of SHP-1, a guanine nucleotide exchange factor that regulates the actin cytoskeleton, Vav-1, was the only protein detected as a direct substrate for SHP-1 (). Downstream, inhibitory KIR2DL1 signaling prevents the assembly of a large complex of cytoskeletal-linked proteins required for cytotoxicity (). KIR phosphorylation after engagement of MHC class I protein on target cells has proved difficult to detect biochemically and has in some cases required addition of a phosphatase inhibitor, pervanadate, to facilitate its detection (). The most likely explanation for this is that only a small fraction of KIR is phosphorylated at any given moment. Thus, determining where and when inhibitory KIR signaling occurs is an essential next stage toward understanding how the balance of activating and inhibitory signals is assessed during NK cell surveillance. KIR and their corresponding MHC class I ligands, as well as many other receptor/ligand pairs, have been shown to cluster at the immunological synapse (IS) between NK cells and other cells (; ; , ; ; ). However, whether there is an importance in the segregation and patterning of proteins at an inhibitory NK cell IS, e.g., in influencing downstream signaling, remains unclear (). We set out to determine the supramolecular organization of the first step in inhibitory receptor signaling, phosphorylation of the cytoplasmic ITIMs of KIR2DL1. Förster resonance energy transfer (FRET) involves the nonradiative transfer of energy from an excited donor fluorophore to a nearby acceptor and can be used to detect macromolecular associations within cells on the nanometer scale (). Here, we image KIR phosphorylation at the NK cell IS using fluorescence lifetime imaging (FLIM) to report FRET. Rather than a small fraction of KIR being phosphorylated homogeneously across the IS, we unexpectedly observed that KIR phosphorylation is spatially restricted to discrete domains or microclusters within the IS. To visualize KIR signaling, we used a generic anti-phosphotyrosine mAb labeled with Cy3 as the acceptor for FRET from the donor GFP tagged to the cytoplasmic portion of the NK inhibitory receptor KIR2DL1. FRET will only be detected if the spatial separation of GFP and Cy3 fluorophores is no more than 9 nm (); thus, FRET will occur only when the anti-phosphotyrosine mAb is extremely close, i.e., bound, to KIR2DL1-GFP. The most robust way to detect FRET is through detecting a decrease in the fluorescence lifetime, τ, of the donor fluorophore, in this case, GFP (). Hence, KIR phosphorylation at the NK cell IS can be detected by comparing the fluorescence lifetime of GFP-tagged KIR2DL1 in unstained cell conjugates (donor only [D]) with the fluorescence lifetime of GFP-tagged KIR2DL1 in conjugates stained with Cy3-labeled anti-phosphotyrosine (donor in the presence of acceptor [DA]; ). An accumulation of KIR2DL1-GFP and phosphotyrosine is clearly visible at the IS between YTS/KIR2DL1-GFP and 221/Cw6. The fluorescence lifetime of GFP is reduced in samples stained with anti-phosphotyrosine compared with unstained samples (). The mean fluorescence lifetime where KIR clustered at the IS in conjugates stained for phosphotyrosine (DA) was 5–10% lower than unstained control cells (D; ). The fluorescence lifetime of GFP was not reduced in anti-phosphotyrosine–stained samples of YTS/KIR2DL1-GFP and 221/Cw3 that express a class I MHC protein not recognized by KIR2DL1 (unpublished data). However, KIR2DL1 does not cluster at these synapses with target cells lacking a cognate MHC ligand. Thus, to further test whether the observed decrease in GFP fluorescence lifetime at the IS between YTS/KIR2DL1-GFP and 221/Cw6 specifically reported KIR phosphorylation, a transfectant expressing a truncated ITIM-less KIR2DL1-GFP (YTS-TR) was used, where GFP was placed just upstream of the membrane-proximal ITIM and the rest of the KIR cytoplasmic tail deleted. Truncated KIR2DL1-GFP still clustered at the IS with target cells expressing cognate MHC protein (). Thus, signaling through ITIMs, or any other possible signaling motifs, in the cytoplasmic tail of KIR is not absolutely required to cluster KIR at the IS, consistent with previous observations (; ). There was clearly anti-phosphotyrosine staining at the synapse involving YTS expressing truncated KIR2DL1, as expected because KIR is not the only protein to be phosphorylated at a synapse, and indeed the lack of KIR signaling would allow a cytolytic synapse to persist. However, FRET could not be detected at the IS despite truncated KIR2DL1-GFP being clustered at the IS, where phosphotyrosine is also present (). To control for the possibility that our methodology would be sensitive to phosphorylated proteins that might associate with the cytoplasmic tail independently of ITIM phosphorylation, we also used a transfectant expressing KIR2DL1-GFP (Y281F, Y311F), in which the two ITIM tyrosines had been specifically altered to phenylalanine. Again, FRET could not be detected in synapses involving transfectants expressing this point-mutated ITIM-less KIR2DL1-GFP (), despite this receptor being clustered at the IS, where phosphotyrosine is also present. We next imaged whether or not KIR phosphorylation would persist after treatment with an Src family kinase inhibitor, PP2, or an Lck-specific inhibitor (). KIR2DL1-GFP clustered at the IS after treatment of NK cells with either inhibitor (), indicating that Src family kinase–mediated signaling is not necessary for KIR clustering, consistent with previous studies (). Anti-phosphotyrosine staining was also still apparent at the IS, consistent with phosphorylation of some proteins at the IS being mediated by other kinases. Despite KIR2DL1 receptor clustering and anti-phosphotyrosine staining, FRET was not detected in the presence of either PP2 or the Lck-specific inhibitor (). Therefore, the activity of Src family protein tyrosine kinases, and specifically Lck, are necessary for KIR phosphorylation at the IS. To examine the dynamics of KIR phosphorylation at the inhibitory NK cell IS, we imaged conjugates fixed after different times of coincubation. The fluorescence lifetime of KIR2DL1-GFP decreased in the presence of Cy3-tagged anti-phosphotyrosine mAb across all the times examined (). KIR signaling is thus sustained for some minutes, perhaps ensuring ongoing interruption of competing activating signals. Changes in the FRET efficiency relate to the relative extent of KIR phosphorylation within the synapse (; right). The extent of FRET detected at individual synapses varied, such that a relatively large degree of KIR phosphorylation occurred at some synapses, whereas little occurred at others. The mean FRET efficiency across several synapses was 4.3, 5.2, and 2.8% after 5, 10, and 20 min of coincubation, respectively (). Although the FRET efficiency does not directly report the number of activated KIRs, these numbers do suggest that only a surprisingly small fraction of KIR is phosphorylated at a given moment, explaining why KIR phosphorylation has been difficult to detect biochemically. Interestingly, there was no decrease in the fluorescence lifetime outside where KIR2DL1-GFP clustered at the intercellular contact (). Thus, KIR phosphorylation occurred only locally at the IS, in contrast to EGF receptor signaling, for example, which rapidly spreads throughout the cell membrane (). Even when a single NK cell engages two resistant target cells, KIR phosphorylation is restricted to the intercellular contact with both targets (, top) and does not spread to the remaining, unconjugated membrane. By not spreading KIR phosphorylation outside the IS, NK cell inhibition can be restricted to one target cell while maintaining an effective surveillance of another conjugated cell, allowing NK cells to survey both susceptible and resistant target cells simultaneously and respond appropriately (). We also found that a single target cell can effectively trigger KIR phosphorylation when bound to two NK cells simultaneously (, bottom), ensuring that accumulation of human leukocyte antigen (HLA)–C at one synapse would not make the target cell susceptible to attack by another bound NK cell. Analysis of the intensity and fluorescence lifetime values across the synapse revealed discrete domains with decreased lifetime within the aggregate of KIR2DL1-GFP at the IS (). Such “microclusters” of phosphorylated KIR did not specifically colocalize with an increased concentration of KIR2DL1-GFP, as determined by comparing the location of FRET (decreased lifetime) with the intensity of GFP (, top three panels). Domains of decreased fluorescence lifetime were not observed at synapses with cells expressing KIR-GFP lacking ITIMs (KIR-TR; , bottom). To statistically assess whether the observed microclusters with decreased lifetime could arise from natural statistical variation in the measured fluorescence lifetime, we quantitatively analyzed the standard deviation in the lifetime of KIR2DL1-GFP as a function of intensity compared with a simulated dataset, as previously described (). Effectively, the simulated data establishes a baseline in the variability of measured lifetime values that arise as a consequence of noise in the photon-counting process and not due to any real variation in the fluorescence lifetime of the sample. The standard deviation of the fluorescence lifetime of KIR2DL1-GFP at the inhibitory synapse in the presence of an acceptor is significantly greater than the simulated GFP or that observed for the ITIM-less KIR2DL1-GFP (), confirming that there are true differences in the fluorescence lifetime of KIR2DL1-GFP. Our analysis does not allow us to precisely determine the size of microclusters of phosphorylated KIR. However, taking these data together, structure in the organization of phosphorylated KIR must occur on a scale close to the limit of the spatial resolution of our method, i.e., 1 μm. This scale could represent the size of microclusters themselves or a distance between smaller microclusters. Very small microclusters, e.g., below the limit of resolution for optical microscopy (∼0.4 mm), would be less visible. Large clusters of phosphorylated KIR, i.e., 3–4 μm, were sometimes seen, which may result from an aggregation of smaller microclusters. For comparison, the size of the cluster of KIR2DL1-GFP was 5–8 μm. 3D imaging of FRET was obtained by fluorescence lifetime images being acquired every 0.5 μm throughout the conjugate, and the en face fluorescence intensity and lifetime at the IS was reconstructed (). Discrete regions of decreased fluorescence lifetime within the cluster of KIR were clearly visible. Thus, KIR signaling does not occur uniformly across the IS and, although there is potentially significant KIR signaling occurring in other areas of the synapse, the majority of KIR signaling appears spatially confined to discrete microclusters within the larger aggregation of receptor at the IS. We next set out to test for the presence of microclusters of signaling in live cell–cell conjugates. Visualization of KIR phosphorylation at the NK cell IS required accurate detection of low levels of FRET. Single photon counting is the method of choice for the most accurate measurements of fluorescence lifetime. However, single photon–counting FLIM is inherently slow () and thus it is not possible to extend our methodology to detect KIR phosphorylation in fast-moving live cell interactions. Instead, because we found that, consistent with previous studies (), Lck was necessary for KIR phosphorylation (), we set out to determine whether the kinase itself would accumulate in microclusters at the inhibitory NK cell IS. To assess the cellular distribution of Lck in live cells, YTS/KIR2DL1 cells were transfected to express Lck conjugated to monomeric YFP (mYFP). Western blotting confirmed the presence of Lck-mYFP at the expected size in these transfectants (unpublished data). Confirming that the chimera Lck-mYFP could be functional, expression of Lck-mYFP in JCam1.6 restored the ability of this Lck-deficient cell line to flux calcium after anti-CD3 mAb stimulation (unpublished data), consistent with previous observations (). Expression of Lck-mYFP did not alter the cytolytic response of YTS/KIR2DL1, as target cells expressing a noncognate MHC class I protein (221/Cw3) were still lysed by this transfectant, whereas target cells expressing a cognate MHC class I protein (221/Cw6) were resistant (unpublished data). In images, Lck-mYFP was localized at the YTS/KIR2DL1 plasma membrane as well as in an intracellular pool, as previously observed in peripheral blood NK cells () and T cells (). Soon after contact with target cells, small clusters of membrane-associated Lck are clearly visible at the IS (), consistent with previous studies observing clusters of Lck at NK cell synapses in fixed cells (). Small clusters of Lck-YFP were sustained for several minutes at the inhibitory NK cell IS (). The presence of small clusters of Lck being sustained at the inhibitory NK cell IS coupled with observations of microclusters of phosphorylated KIR is consistent with Lck being necessary for KIR phosphorylation and demonstrates further that NK cell signaling occurs within microclusters. We visualized signaling at an IS using FLIM to report FRET between GFP-tagged KIR and a fluorophore-tagged general anti-phosphotyrosine mAb. This technology can be applied to image the phosphorylation of any specific receptor at an intercellular contact. This methodology is of particular use where phosphospecific mAbs are not available and could be extended toward developing rapid screens for protein phosphorylation. Using this methodology, we found that KIR signaling occurs within discrete microclusters within the larger aggregation of protein at the NK cell IS. Imaging of T cell signaling events recently revealed microclusters of TCR colocalized with activated forms of Lck, ZAP-70, and LAT at the contact between a T cell and a lipid bilayer (; ; ; ), which builds on earlier work showing TCR signaling clusters assembled at contacts with antibody-coated coverslips (). Thus, our observations extend the generality of microcluster-mediated signaling by demonstrating their relevance to human NK cell inhibitory signaling. What then, is restricting KIR phosphorylation to discrete microclusters? Although the Singer-Nicolson fluid mosaic model has formed the basis of our understanding of the cell membrane for the past three decades, relatively recent evidence suggests the organization of both proteins and lipids to be much more complex than this (). Here, it is unlikely that lipid rafts facilitate the formation of microclusters of inhibitory signaling, as KIR2DL1 is excluded from such domains and in fact signals to block an accumulation of lipid rafts to the IS (; ). Alternatively, specific protein–protein interactions could create microclusters of KIR2DL1 phosphorylation or they might be a result of “membrane-skeleton corralling,” i.e., by the cytoplasmic domains of transmembrane proteins colliding with the cytoskeleton (). Such temporary confinement would be further enhanced by oligomerization, and indeed metal ions, known to dimerize KIR2DL1 in solution (), have been found to be necessary for KIR function and phosphorylation (; ). It is perhaps surprising that spatially confined inhibitory KIR signals are effective in inhibiting NK cell responses. However, the observation that KIR need not be continuously actively inhibiting over the entire synapse is consistent with the observation that a chimera of KIR tagged with an extracellular GFP was able to efficiently inhibit NK cell responses without being significantly clustered at the IS (). In general, spatially confined recruitment of kinases to the NK cell synapse may facilitate the transphosphorylation of numerous NK cell receptors within a microcluster. For example, trapped Lck, as recently visualized at the interface between a T cell and a mAb-coated coverslip (), could facilitate transphosphorylation between KIR and other NK cell receptors within discrete microclusters. Thus, one specific hypothesis for an efficient mechanism of KIR-mediated inhibition would be that activating NK cell receptors recruit Lck, which is then able to transphosphorylate local KIR if they are engaged by HLA-C on the opposing cell. This model therefore requires that signaling by NK cell–activating receptors occurs within microclusters, analogous to TCR signals. This could be caused by NK cell–activating ligands being organized into specific microdomains on the surface of the opposing target cell (). Thus, in this model, the spatial confinement of inhibitory receptor phosphorylation serves to focus inhibitory action, e.g., the dephosphorylation of Vav-1 (), to specific sites where the competing activating signals are triggered. Broadly, it would previously have been expected that KIR signaling occurs homogeneously across the cluster of KIR at the IS. Thus, the data presented here seeds new research into how the spatial confinement of receptor phosphorylation could influence the integration of activating and inhibitory signals by NK cells. A transfectant of the MHC-deficient human B-lymphoblastoid cell line, 721.221 expressing HLA-Cw6 (221/Cw6) has been described (). YTS, a subclone of the human tumor line YT, expressing KIR2DL1 (YTS/KIR2DL1; ), COOH-terminal GFP-tagged KIR2DL1 (YTS/KIR2DL1-GFP; ), or a truncated ITIM-less KIR2DL1-GFP (YTS-TR; ) where GFP was placed just upstream of the membrane-proximal ITIM and the rest of the KIR was cytoplasmic tail deleted have also been described. A mutant of KIR2DL1-GFP (Y281F and Y311F) in which the two ITIM tyrosines were mutated to phenylalanine was generated in pBABE (QuikChange Mutagenesis; Stratagene). The membrane-distal tyrosine was mutated to phenylalanine first using the forward primer 5′-GATATCATCGTGTTCACGGAACTTCC-3′ and its reverse complement. The membrane-proximal tyrosine was then mutated to phenylalanine using the forward primer 5′-CCTCAGGAGGTGACATTCACACAGTTGAATC-3′ and its reverse complement. The fidelity of the construct was confirmed by sequencing and expressed in YTS cells by retroviral transduction as described previously (). For Lck-mYFP, human was amplified by PCR from cDNA synthesized from PBL using primers 5′-CCCAAGCTTGCC ACCATGGGCTGTGGCTGCAGCTC-3′, including a HindIII restriction site, and 5′-GCGGTACCCCAGGCTGAGGCTGGTACTGGCCCTC-3′ to remove the stop codon and include a KpnI restriction site. The PCR product was first cloned into pCR2.1-TOPO (Invitrogen), and the correct sequence was confirmed and subcloned into the mammalian expression vector pcDNA 3.1/Hygro containing mYFP, i.e., with substitutions S65G, S72A, T203Y, and A206K (a gift from R. Tsien, University of California, San Diego, La Jolla, CA). This resulted in a construct encoding Lck and mYFP connected by an 11-amino-acid linker (Gly-Val-Pro-Ser-Ser-Asp-Pro-Pro-Val-Ala-Thr). YTS-KIR2DL1 and a variant of Jurkat lacking Lck, JCam1.6 (American Type Culture Collection), were each transfected to express Lck-mYFP by electroporation. Transfectants were grown in 0.8 mg/ml hygromycin, and Lck-mYFP–expressing cells were sorted by flow cytometry (FACSDiva; Becton Dickinson). For cell conjugation, 5 × 10 YTS/KIR2DL1, YTS/KIR2DL1-GFP, YTS-TR, or YTS-KIR2DL1 (Y281F and Y311F) was mixed with 5 × 10 221/Cw6 cells in 50 μl warm media and incubated at 37°C/5% CO for the indicated time. Cell conjugates were then fixed in 100 μl buffer containing paraformaldehyde and saponin (Cytofix/Cytoperm; Becton Dickinson) for 15 min at 4°C followed by 5 min at room temperature. Cells were washed in 0.1% Tween-20/PBS and blocked in 100 μl buffer containing saponin (Perm/Wash; Becton Dickinson) with 5% horse serum/3% BSA for 30 min at 4°C. Cell conjugates were stained in 100 μl of 10 μg/ml anti-phosphotyrosine mAb (clone 4G10; Upstate Biotechnology) tagged with Cy3 (Cy3/mAb ratio 8:1) for 2 h at 4°C. After washing, cell conjugates were gently resuspended and ∼8 μl was placed between a microscope slide (thickness 1 mm; Becton Dickinson) and a glass coverslip (thickness No. 1.5; Becton Dickinson). The Src family tyrosine kinase inhibitor PP2 (Calbiochem) and the Lck-specific inhibitor, 7-Cyclopentyl-5-(4-phenoxyphenyl)-7H-pyrrolo[2,3-d]pyrimidin-4-ylamine (; Sigma-Aldrich) were diluted to 5 μM in cell media. YTS transfectants were preincubated with inhibitors for 1 h at 37°C/5% CO before mixing with target cells for 10 min in the continuing presence of the inhibitor. The fluorescence lifetime of GFP-tagged KIR2DL1 was measured using time-correlated single photon counting (SPC-730; Becker & Hickl GmbH). Laser power was adjusted to give a mean photon count rate of ∼1 × 10 counts/s, and fluorescence lifetime images were acquired over 300 s. Fluorescence lifetimes were calculated for all pixels in the field of view (128 × 128 pixels; SPCImage). As the fluorescence intensity in the unconjugated (nonsynapse) membrane was very low, it was necessary to bin all photons from this region to accurately calculate the fluorescence lifetime. To achieve this, an in house–written fluorescence-decay program (written in Labview [National Instruments]) was used. The fluorescence lifetime for the synapse was also calculated using this program and agreed well with the fluorescence lifetime calculated using SPCImage. FRET efficiency images were calculated such that the FRET efficiency, E = 1 − τ/τ, where τ is the pixel-by-pixel fluorescence lifetime of the donor in the presence of the acceptor and τ is the mean fluorescence lifetime of the donor at the IS in the absence of the acceptor for all cells imaged (unstained controls). Mean FRET efficiencies at the IS were calculated where τ is the mean fluorescence lifetime of the donor in the presence of the acceptor and τ is the mean fluorescence lifetime of the donor in the absence of the acceptor. As fluorescence lifetime images were limited to 128 × 128 pixels, the limit in resolution of the photon counting detector, it was necessary to apply an interpolation method to obtain enlarged images of synapses at suitable resolution and size for publication. For this, the “nearest neighbor” interpolation method (Photoshop 7; Adobe) was applied, which sets the value (or color, in this case) of an interpolated point to the value of the nearest existing data point, effectively making the pixels bigger. For example, to enlarge 200%, one pixel will be enlarged to a 2 × 2 area of four pixels with the same color as the original pixel. This is the most appropriate method for interpolation of indexed images, i.e., images that map pixel values to colors, as it does not change the color information of the image and does not introduce any anti-aliasing, which would make edges appear smoother by averaging out pixels. An exception is that for reconstruction of the en face synapse (shown in ) the bicubic method of interpolation was applied. We carefully confirmed that the images of synapses did not appear different by applying this procedure, and analysis of fluorescence lifetime data was always performed on the raw, noninterpolated data. It is clear that random errors caused by instrumental drift or subtle variation in the biological sample will effect different lifetime images far more so than within a single image. It has been specifically calculated that interimage differences, using a frequency-based FLIM methodology, are typically one order of magnitude greater than intraimage variations (). We have addressed this in three ways: In all experiments, we compared image data taken from one specific experiment performed over one single day. We confirmed that the presence or absence of FRET was consistent in each sample over multiple independent experiments. The “breakpoint” in all images using a discrete scale for the fluorescence lifetime is specifically set at the point where FRET efficiency is 5% for that experiment. YTS/KIR2DL1 cells transfected to express Lck-mYFP and 221/Cw6 target were mixed in a glass-bottomed microscope chamber (Nunc) containing 200 μl of warm RPMI media. Live cell conjugates were imaged by resonance scanning confocal microscopy (DMIRE2/TCS SP2 RS) using a 63× water-immersion objective (NA 1.2). The microscope stage was housed within an environmental chamber (Solent Scientific) maintained at 37°C/5% CO. mYFP was excited using the 514-nm line of an argon laser. 3D views of cell conjugates and en face views of the synapse were reconstructed using Volocity.

Keratins (previously also called cytokeratins) are filament-forming proteins of epithelial cells and are essential for normal tissue structure and function. Keratin genes account for most of the intermediate filament genes in the human genome, making up the two largest sequence homology groups, type I and II, of this large multigene family. They are highly differentiation-specific in their expression patterns, implying functional differences. Mutations in most of them are now associated with specific tissue-fragility disorders, and antibodies to keratins are important markers of tissue differentiation and, therefore, tools in diagnostic pathology. Since the first keratins were sequenced and identified as type I and II intermediate filament proteins, the increasing numbers of keratins has provided an ongoing challenge for their clear identification and logical classification across species. The first attempt at providing a comprehensive keratin nomenclature dates back to 1982. used 2D isoelectric focusing and SDS-PAGE to map the keratin profiles of a large number of normal human epithelia, tumors, and cultured cells. They grouped the basic-to-neutral type II keratins as K1–K8 and the acidic type I keratins as K9–K19 (). Although not open-ended for type II keratins, this system has so far proven manageable, as the incorporation of a few novel type II keratins could be accomplished by the addition of discriminatory suffix letters to keratins exhibiting similar gel-electrophoretic properties (,; ). Moreover, the Moll nomenclature has not been further challenged by the “hard” α-keratins of hair and nail (hair keratins), as these keratins were named Ha (acidic, type I) or Hb (basic to neutral, type II) followed by a number, with H standing for hair (; , ). Overall, however, the present naming of keratins has not been systematic, and a reorganized and durable scheme is long overdue. Genome analyses have recently demonstrated that humans possess a total of 54 functional keratin genes, i.e., 28 type I and 26 type II keratins, forming two clusters of 27 genes each on chromosomes 17q21.2 and 12q13.13 (the gene for the type I keratin K18 being located in the type II keratin gene domain; , ; , ; ). Recognition of the extent of this large mammalian gene family led to a suggested revised nomenclature () based on an extended Moll system, K1–K8, and K9–K24 (, ; ; ), and conceptually close to an earlier proposal (). In this nomenclature, all human type I keratins were named Ka9 to KaX and all type II keratins were named Kb1 to KbY, thus, enabling type I and II keratins of other mammalian species to be added consecutively into this open-ended system. At the 2004 Gordon Conference on Intermediate Filaments in Oxford, an initiative to achieve international consensus led to the formation of a broad-based Keratin Nomenclature Committee that included active investigators in the keratin field and members of the Human Genome Nomenclature Committee (HGNC) and the Mouse Genome Nomenclature Committee. This committee evaluated several potential nomenclature schemes and, after extensive deliberation and consultation with other colleagues in the intermediate filament field, arrived at the consensus nomenclature system that is detailed in the following sections. To structure the new nomenclature system, the 54 human keratins and their genes are divided into three categories: epithelial keratins/genes, hair keratins/genes, and keratin pseudogenes. The nomenclature is also structured to allow for the inclusion of a fourth category of nonhuman epithelial and hair keratins of other mammalian species, whose genes are either absent or occur as pseudogenes in the human genome. For both type I and II keratins, these four categories are numerically arranged in the following order (): human epithelial keratins, human hair keratins, nonhuman epithelial/hair keratins, and human keratin pseudogenes. Because of historical reasons and the extensive number of existing publications, the Moll designation for the epithelial keratins K1–K8 and K9–K24 (, ; ; ; ) is retained, as is the existing HGNC gene designation scheme (i.e., ). #text Compared with the classical type I epithelial keratins, significantly more adjustments had to be made within the family of type II epithelial keratins K1–K8 (). The numbering of additional type II epithelial keratins begins with K71, after the nonhuman mammalian type I keratins (). Thus, K71–K74 were assigned to the four type II inner root sheath keratins K6irs1–4 (; ). The number of distinct variants of the gene has been a matter for some discussion, but sequencing of the human genome has revealed that there are, in fact, only three K6 variants, each encoded by their own gene (). These are now designated K6a (), K6b (), and K6c (). Because of its lack of conformity to the rules of the HGNC, the hair follicle–specific keratin K6hf (; ) has been repositioned after the last inner root sheath keratin K74 and so renamed K75 (). For similar reasons of identity and nomenclature conformity, the epidermal keratin K2e has been redesignated K2, and the palatal keratin K2p (,) has been renamed K76 (, column 4). Although the completion of sequencing and analysis of the human type I keratin gene domain did not reveal any new keratin genes, that of the type II keratin gene domain led to the detection of four hitherto unknown genes whose encoded proteins had previously been designated K1b, K5b, K6l, and Kb20 (, ; ; , column 1). Although there is only limited expression data available for K5b, K6l, and Kb20 (), keratin K1b has recently been demonstrated to be specifically expressed in eccrine sweat glands (). In the new nomenclature, these keratins are designated K77–K80 (, column 4). Collectively, K1–K8 and K71–K80 cover the twenty human type II epithelial keratins ( and ). Human keratin pseudogenes of both types were also included in the new nomenclature system, to give all-inclusive schemes of the type I and II keratin gene chromosomal domains, as shown in . Whereas the slots reserved for type II pseudogenes extend from 121–220, type I pseudogenes have an open system starting with 221 (), as this is the last category to be named. Presently, there are eight keratin pseudogenes designated – located within the type II keratin gene domain on chromosome 12q13.13 (, column 5). Of these, – are hair keratin pseudogenes and – are epithelial keratin pseudogenes (). Should some of these pseudogenes have active counterparts in other species, the encoded keratins will retain the numbering of the respective pseudogene without the suffix “,” and, depending on their type, will be included in the respective category of nonhuman epithelial/hair keratins. The positions above 128 can be used to include keratin pseudogenes identified in other mammals. The type I keratin gene domain contains two hair keratin and three epithelial keratin pseudogenes ( and ). Two of them, the hair keratin pseudogene and the epithelial keratin pseudogene , possess active gene counterparts in other species and have already been named accordingly. The remaining three genes were designated – (, column 5). It should be emphasized that in addition to the aforementioned type I and II keratin pseudogenes in the keratin gene domains on chromosomes 17 and 121, there are at least 61 processed pseudogenes for the type II keratin K8 and 77 for the type I keratin K18, which are dispersed throughout the human genome. Moreover, there are five processed pseudogenes for the type I keratin K19, which are single pseudogenes located on chromosomes 4, 6, and 10, with two pseudogenes located on chromosome 12. None of these contains an intact reading frame. Furthermore, the terminal segment on the human type I keratin gene domain spanning genes and () is inserted four times into different regions of chromosome 17. This gives rise to three unprocessed pseudogenes for K14 and K16, and four for K17, as well as four pseudogenes, which are all assumed to be nonfunctional (, ). The decision as to how these pseudogenes will be included into the respective lists of type I and II keratin pseudogenes will be left to the HGNC. i s m o d i f i e d a n d u n i f y i n g n o m e n c l a t u r e f o r m a m m a l i a n k e r a t i n s p r e s e r v e s t h e w i d e l y u s e d a n d b r o a d l y r e f e r e n c e d M o l l d e s i g n a t i o n s y s t e m f o r t h e c l a s s i c a l h u m a n e p i t h e l i a l k e r a t i n s K 1 – K 8 a n d K 9 – K 2 4 . T h e f e w c h a n g e s t h a t h a v e b e e n i n t r o d u c e d r e f l e c t c o n s t r a i n t s t o r e s t r i c t t h e p r o t e i n d e s i g n a t i o n t y p e K # $ ( # = a n u m b e r ; $ = a l e t t e r ) t o t r u e k e r a t i n i s o f o r m s . A c c o r d i n g l y , t h e n e w n o m e n c l a t u r e w i l l h a v e n o s i g n i f i c a n t i m p a c t o n c u r r e n t t e x t b o o k s a n d c o m m e r c i a l c a t a l o g s . M a j o r c h a n g e s a r e n e a r l y a l l r e s t r i c t e d t o r e c e n t l y i d e n t i f i e d e p i t h e l i a l k e r a t i n s . H o w e v e r , l a b o r a t o r i e s w o r k i n g o n h a i r k e r a t i n s w i l l h a v e t o , u n d e r t h e n e w s y s t e m , p a r t w i t h t h e 2 0 - y r - o l d d e s i g n a t i o n s H a X a n d H b Y a n d a d o p t n e w n o m e n c l a t u r e . A n e f f o r t h a s t h e r e f o r e b e e n m a d e t o p r e s e r v e t h e l a s t d i g i t b e t w e e n t h e o l d a n d t h e n e w d e s i g n a t i o n s , w h i c h i s h o p e d w i l l h e l p r e s e a r c h e r s t o a d a p t t h e n e w n a m e s . A l t h o u g h t h e n e w n o m e n c l a t u r e a s s e m b l e s t h e h u m a n t y p e I k e r a t i n s i n t o a n a l m o s t u n i n t e r r u p t e d s e r i e s b y p r e s e r v i n g t h e o r i g i n a l M o l l n o m e n c l a t u r e w h e r e p o s s i b l e , t h e r e i s a n u n a v o i d a b l e g a p i n t h e n u m b e r i n g o f h u m a n t y p e I I k e r a t i n s . S u f f i c i e n t s p a c e h a s b e e n l e f t i n t h e s y s t e m f o r k e r a t i n s o c c u r r i n g i n o t h e r m a m m a l i a n s p e c i e s , a s w e l l a s f o r k e r a t i n p s e u d o g e n e s . M o r e o v e r , w e s u g g e s t t h a t t h e t e r m “ k e r a t i n ” r a t h e r t h a n “ c y t o k e r a t i n ” b e u s e d a n d t h a t m a m m a l i a n o r t h o l o g u e s o f h u m a n k e r a t i n s b e g i v e n t h e s a m e n a m i n g s y s t e m . A s t h e r e v i s e d n o m e n c l a t u r e s h o u l d f a c i l i t a t e c o m m u n i c a t i o n a n d u n d e r s t a n d i n g w i t h i n t h e c o m m u n i t y i n t e r e s t e d i n k e r a t i n s a n d t h e i r d i s e a s e s , w e a d v o c a t e t h a t t h i s n e w s y s t e m b e u s e d i n a l l f u t u r e s t u d i e s .

Epithelial–mesenchymal transition (EMT) converts polarized epithelial cells to motile mesenchymal cells (). EMT operates during embryonic cell layer movements and tumor cell invasiveness (). During EMT, the epithelial proteins E-cadherin and zonula occludens 1 (ZO-1) are down-regulated, and the mesenchymal proteins vimentin, α-smooth muscle actin, and fibronectin are up-regulated. Receptor tyrosine kinase, Wnt, Notch, and TGF-β pathways trigger EMT (). TGF-β binds to receptor serine/threonine kinases, which activate intracellular Smad and other signaling pathways that regulate gene expression (). TGF-β inhibits epithelial cell growth, acting as a tumor suppressor, but it also promotes carcinoma progression and metastasis (). The tumor-promoting effects of TGF-β are based on its ability to induce (a) EMT, matrix invasiveness, and blood vessel intravasation by carcinoma cells (); (b) cytostatic effects on surveilling immune cells (); and (c) proangiogenic effects (). TGF-β elicits EMT and in vivo metastasis via Smads and complementary non-Smad effectors, such as Rho GTPases and p38 MAPK (). TGF-β represses inhibitor of differentiation (Id) 2 and 3 expression and induces expression of the Notch ligand Jagged-1, which are critical events during EMT (; ; ). Here, we describe the role of high mobility group (HMG) A2, also known as HMGI-C, as an effector of TGF-β that causes EMT. HMGA2 and -1 constitute a family of nuclear factors that bind AT-rich DNA sequences (; ). HMGA factors contribute to transcriptional regulation by organizing nucleoprotein complexes such as enhanceosomes (). HMGA2 is expressed during embryogenesis and becomes silent in adult tissues (). However, HMGA2 is abundantly expressed by transformed cells or tumors of mesenchymal and epithelial origin (for reviews see ; ). In contrast, depletion of HMGA2 by antisense cDNA in thyroid cells eliminates their transformation by myeloproliferative and Kirsten murine sarcoma viruses (). Here, we show that HMGA2 regulates the transcription factors Snail, Slug, Twist, and Id2, thus linking TGF-β signaling to regulators of tumor invasiveness and metastasis. Transcriptomic analysis of TGF-β–induced EMT in mammary epithelial NMuMG cells identified as a prominent TGF-β target (). mRNA increased after 2 and 8 h and returned to basal levels after 36 h of TGF-β stimulation (). In contrast, mRNA was not regulated and dropped significantly upon cell confluence at 36 h (). HMGA2 protein increased after 8 h and peaked at 12 h of TGF-β stimulation (). A TGF-β type I receptor kinase inhibitor (LY580276; ) did not affect basal HMGA2 levels, demonstrating the absence of autocrine TGF-β (). mRNA induction by TGF-β was not impaired by the protein synthesis inhibitor cycloheximide while it was blocked by the RNA polymerase inhibitor II actinomycin D (Fig. S1 A, available at ). A constitutively active form of the TGF-β type I receptor increased expression more efficiently than TGF-β itself, whereas a kinase-dead mutant of this receptor inhibited it (Fig. S1 B). When the constitutively active type I receptor was expressed at higher levels, it often failed to induce at higher levels than TGF-β (). This reflects mechanisms of pathway desensitization, as TGF-β signaling is controlled in a timed fashion by activation and inactivation of receptor and Smads. The results suggest that is a direct TGF-β target. Mouse promoter analysis showed that basal promoter activity varied according to the deletion construct used, and TGF-β stimulation led to a 2.5–3-fold induction (). Basal promoter activity and induction by TGF-β were lost when the proximal region containing TCC repeats was deleted. Sequence inspection of 4 kbp upstream from the transcription initiation site showed few noncanonical Smad binding elements between −700 and −100 bp (unpublished data). We now examine the role of these elements on transcriptional induction by TGF-β. mRNA induction and promoter activation by TGF-β was blocked in cells expressing dominant-negative Smad2 (Smad2 SA; ); Smad2 SA cannot be phosphorylated by the TGF-β type I receptor and blocks TGF-β– induced EMT (). Knockdown of Smad2 by 80% or Smad3 by 65% after RNAi had no effect on induction by TGF-β or on the EMT response ( and Fig. S1, C and D). However, knockdown of the common partner of Smad2 and -3, Smad4, by 95% effectively blocked induction by TGF-β and the EMT response ( and Fig. S1, C and D; ). The lack of effect by knockdown of Smad2 or -3 may indicate that the protein depletion achieved was insufficient. Alternatively, both Smad2 and -3 may be involved in EMT, as we previously proposed (), and for effective block of EMT, both Smad2 and -3 need to be depleted. Experiments are under way to test this possibility. In another cell line, metastatic breast cancer MDA-MB-231 cells, TGF-β weakly induced expression, and knockdown of Smad3 or -4 blocked this response, whereas knockdown of Smad2 did not ( and Fig. S1, E and F). Based on these data, it appears that single Smad3 or -4 knockdown is sufficient in blocking TGF-β–induced expression ( and Fig. S1 F). A more robust knockdown of Smad2 is needed to reach final conclusions about the role of this Smad isoform in regulation and EMT. Finally, immunoprecipitation of chromatin bound Smad4 from NMuMG cells confirmed a TGF-β–inducible association of Smad4 in the proximal (−195/+5) and upstream (−495/+245) but not in the distal (−3420/−3320) promoter region of the gene (). The −500 to +5 promoter region where Smad4 binds overlapped with putative Smad binding elements. The results establish that Smad signaling is involved in induction by TGF-β, with Smad4 clearly being implicated. However, we cannot conclude whether Smad4 cooperates with Smad2, Smad3, or both during regulation. To address the functional role of HMGA2, we established stable NMuMG clones inducibly expressing human HMGA2. Ectopic HMGA2 was expressed in the absence of an inducer, and induction increased its expression further (). Ectopic HMGA2 localized in the nucleus as expected, and its localization was not affected by TGF-β (). In the absence of TGF-β, HMGA2 clones grew slower than mock cells (). TGF-β inhibited growth of mock and HMGA2-expressing cells (). Thus, TGF-β induces growth arrest despite ectopic HMGA2 expression. Mock NMuMG clones treated with an inducer displayed characteristic polarized epithelial morphology (). TGF-β caused EMT, as mock cells acquired elongated, fibroblast-like morphology. HMGA2 clones were constitutively elongated and lost cell–cell contacts, suggesting induction of EMT, which was enhanced further by TGF-β (). EMT in HMGA2 clones was confirmed by visualizing actin cytoskeleton rearrangements and the loss of ZO-1 and E-cadherin from cell junctions () and by measuring the loss of expression of and mRNA (). Moreover, mRNAs of the mesenchymal markers , , and were constitutively expressed, and was increased to a lesser extent. Immunoblot analysis confirmed E-cadherin protein down-regulation and enhanced expression of mesenchymal N-cadherin in HMGA2 clones (). These experiments demonstrate that ectopic HMGA2 causes EMT. The fact that ectopic HMGA2 mimicked the TGF-β response () raises the question of whether HMGA2 activates autocrine TGF-β, leading to EMT. In mock NMuMG cells, the LY580276 inhibitor blocked TGF-β–mediated EMT, as cells kept a cortical actin distribution and did not down-regulate E-cadherin (Fig. S2, available at ). However, in HMGA2 clones, LY580276 had no effect on the elongated morphology, actin stress fiber network, or lack of E-cadherin. Similar results were obtained with a TGF-β–neutralizing antibody added to the medium of HMGA2 clones for several days (unpublished data). These experiments demonstrate that the profound effects HMGA2 shows on EMT cannot be accounted for by the induction of autocrine TGF-β that signals in a constitutive manner. Transfection of NMuMG cells with siRNA resulted in an ∼70% decrease in basal and TGF-β–induced mRNA expression (). An even stronger reduction was seen of the endogenous HMGA2 protein level (Fig. S3, available at ). Weak nuclear HMGA2 was seen in control cells; TGF-β dramatically increased the nuclear HMGA2 levels, as the cells became elongated and fibroblast like (Fig. S3). Upon HMGA2 depletion, its nuclear staining was barely detectable. Immunoblot analysis of endogenous HMGA2 protein upon chromatin extraction () was not efficient enough in these transfected cells to quantitatively monitor the degree of protein depletion upon RNAi (unpublished data). We reproducibly observed that the mammary epithelial cells enlarged their diameter by roughly 1.8–2.2-fold when HMGA2 was depleted (Fig. S3). This effect was specific for the siRNA, as we did not observe size changes by control (siLuc) or a panel of siRNAs that target unrelated genes in this cell line (Fig. S3 and unpublished data). TGF-β induced EMT in NMuMG cells transfected with control siRNA. In contrast, cells transfected with siRNA do not undergo EMT. Indeed, these cells maintained polarized morphology, ZO-1, and E-cadherin at their junctions and decreased the TGF-β–inducible levels of N-cadherin and (; and Fig. S3). Although knockdown restored to a large extent epithelial tight and adherence junctions, we observed only a weak block of total ZO-1 and E-cadherin protein down-regulation by TGF-β (unpublished data). We conclude that endogenous HMGA2 is required for TGF-β–induced EMT. The zinc-finger transcription factors Snail, Slug, δEF-1/ZEB-1, and SIP-1/ZEB-2 or the basic helix-loop-helix factor Twist repress expression during embryonic EMT () and promote tumor cell metastasis or cancer recurrence after therapy (; ). The extreme down-regulation of expression in HMGA2 clones () prompted us to analyze some of these transcriptional repressors. In parental NMuMG cells, and and, to a lesser extent, mRNA were induced by TGF-β (). Repressor expression was dramatically up-regulated in HMGA2 clones even in the absence of TGF-β. Similar to regulation of endogenous and mRNA, TGF-β induced and promoter activity (). Notably, cotransfection of HMGA2 enhanced and promoter activity to significantly higher levels than TGF-β stimulation alone. As specificity control, the Smad3/Smad4-dependent promoter reporter CAGA-Luc was not regulated by HMGA2 (unpublished data). Upon RNAi-mediated knockdown of , endogenous mRNA induction by TGF-β was reduced by ∼50% (). This explains why total E-cadherin and ZO-1 protein levels were still repressible by TGF-β after knockdown (unpublished data). We conclude that partial depletion of endogenous by RNAi is sufficient to restore epithelial differentiation in NMuMG cells and establishment of cell–cell junctions; however, it is not sufficient to block strongly enough induction by TGF-β. TGF-β down-regulates Id2 to induce EMT (; ). Id2 mRNA and protein expression were down-regulated in HMGA2 clones compared with mock cells, and TGF-β further repressed Id2 levels (). The results demonstrate that HMGA2 regulates , , , and , all key players of EMT. TGF-β induces Snail expression either via Smad3 or via MAPK signaling (; ). Our results add HMGA2 as a novel regulator of expression downstream of Smads. Whether HMGA2 cooperates with Smad3 or MAPK signals to induce is being explored. is another gene target of HMGA2 that is weakly induced by TGF-β. We conclude that TGF-β, via HMGA2, primarily affects the Snail pathway and, to a lesser extent, the Twist pathway. We describe a new target of TGF-β signaling, the nuclear factor HMGA2, and a new transcriptional circuitry that mediates EMT by TGF-β (). HMGA2 links TGF-β signaling to major factors of tumor invasiveness and metastasis. This work suggests that HMGA2 acts not only as an architectural chromatin factor as previously thought but also as a gene-specific regulator that responds to signals from extracellular factors. HMGA2 is overexpressed in a variety of tumors primarily of mesenchymal origin (for review see ). However, the mechanism of HMGA2 action is not yet known. This study demonstrates for the first time that HMGA2 is involved in EMT. Our results are consistent with the specific presence of HMGA2 at the invasive front of squamous carcinomas (), a place where EMT occurs during cancer progression. On the other hand, overexpression in MCF7 mammary carcinoma cells of HMGA1b, another member of the HMGA family, but not that of the related HMGA1a, led to invasive tumor growth in nude mice (). Histochemical and transcriptomic analysis of tumor samples from such mice indicated that HMGA1b induced expression of genes with links to EMT. Mouse mammary epithelial NMuMG, human hepatocarcinoma HepG2, human MDA-MB-231-Eco cells, and stable NMuMG clones expressing Smad2 SA (mutant with alanines in place of two C-terminal serines that become phosphorylated by receptors) have been described (; ; ). Adenoviruses expressing control GFP (Ad-GFP) were a gift from B. Vogelstein (Johns Hopkins Medical Institutions, Baltimore, MD); adenoviruses expressing C-terminally HA-tagged constitutively active TGF-β type I receptor (activin receptor-like kinase 5) ALK-5(TD) and HA-tagged kinase-inactive ALK5(KR) receptor were a gift from K. Miyazono (University of Tokyo, Tokyo, Japan) and have been described (). The mouse siRNA (available from GenBank/EMBL/DDBJ under accession no. ), which was a single RNA oligonucleotide (D-043585-03), and control siRNA against the luciferase reporter vector pGL2 (accession no. ) were obtained from Dharmacon Research, Inc. Human cDNA, cloned by PCR from total normal human mRNA in pcDNA3-HA C-terminally of the HA tag using EcoRI–XhoI as restriction sites, resulted in pcDNA3-HA-hHMGA2. The N-terminally HA-tagged hHMGA2 HindIII–XhoI fragment was subcloned in the inducible vector pMEP4 to produce pMEP4-HA-hHMGA2. The LY580276 inhibitor for the TGF-β type I receptor kinase was a gift from J.M. Yingling (Eli Lilly and Company, Indianapolis, IN; ). Recombinant mature TGF-β1 was obtained from PeproTech and TGF-β3 from K.K. Iwata (OSI Pharmaceuticals, Farmingdale, NY). TGF-β1 was used in most experiments and TGF-β3 in experiments with stable NMuMG and MDA-MB-231 clones for Smad shRNAs ( and Fig. S1). TGF-β1 and -β3 have indistinguishable effects on EMT or gene regulation. Cell monolayers were washed with PBS, trypsinized, and resuspended in PBS, and cell numbers were calculated using a Z1 cell counter (Beckman Coulter). Numbers are plotted as means from triplicate determinations with standard errors. pMEP4 and pMEP4-HA-hHMGA2 were transfected into NMuMG using Lipofectamine 2000 (Invitrogen). 2 d after transfection, cells were cultured in 400 μg/ml hygromycin-B (Calbiochem), and individual antibiotic-resistant clones were derived. For HMGA2 induction, cells growing in hygromycin-B were treated with 10 μM CdCl (Sigma-Aldrich) for 24 h, and cells were incubated with vehicle or TGF-β for another 24–48 h. NMuMG cells were transiently transfected with siRNAs using Dharmafect 4 (Dharmacon Research, Inc.). After 2 h of stimulation with TGF-β, cells were retransfected with siRNAs to a final concentration of 50 nM for 10 h (RNA assay) or 36 h (protein/immunofluorescence assay). NMuMG and MDA-MB-231-Eco clones with Smad knockdown were established after infection with retroviral supernatants derived from cells transfected with pRetroSuper-expressing shRNA against Smad2, -3, or -4 and were provided by M. van Dinther and P. ten Dijke (Leiden University Medical Center, Leiden, Netherlands; ). The shRNA oligonucleotide sequences were 5′-GGAGTGCGCTTGTATTACA-3′ (mouse/ human Smad2), 5′-CGTCAACACCAAGTGCATC-3′ (mouse/human Smad3), and 5′- CCAGCTACTTACCATCATA-3′ (mouse/human Smad4). Total DNA-free cellular RNA was extracted with the RNeasy kit (QIAGEN). RT-PCRs were performed as described previously () and analyzed using specific primers ( and ). Primers for mouse ′ () were used for reference. Lack of DNA contamination was verified by omitting reverse transcriptase (−RT). Quantitative real-time PCR reactions were done as described previously (). Gene expression levels were determined with the comparative Ct method using as reference. The ground condition was set to 1, and expression data are presented as bar graphs of mean values plus standard deviations. NMuMG cells were lysed in 140 mM NaCl, 1.5 mM MgCl, 10 mM Tris- HCl, pH 8.6, and 0.5% NP-40, supplemented with protease inhibitors. After centrifugation (3 min, 6,000 , 4°C), nuclear pellets were resuspended, vortexed for 30 s, and rotated at room temperature for 1 h in 5% perchloric acid. Perchloric acid supernatants (5 min, 6,000 , 25°C) were precipitated by 8 vol cold ethanol and centrifuged (15 min, 10,000 , 4°C), and pellets were resuspended in lysis buffer and analyzed by immunoblotting. Total protein extracts subjected to SDS-PAGE were analyzed by immunoblotting as described previously (). Mouse monoclonal anti–β-tubulin (T8535) and anti–β-actin (AC-15) were obtained from Sigma-Aldrich; mouse monoclonal anti-HA (12CA5) was obtained from Roche Applied Science; mouse monoclonal anti–E-cadherin (C20820) was obtained from BD Biosciences; rat monoclonal anti–ZO-1 (MAB1520) was obtained from Chemicon; and mouse monoclonal anti–histone H1 (AE-4), anti-Smad2/Smad3 (H-2), and anti-Smad4 (B-8) were obtained from Santa Cruz Biotechnology, Inc. Rabbit polyclonal anti-Hmga2 was raised against a synthetic peptide (CSPSKAAQKKAETIGE, where S maps at residue 59 of mouse Hmga2) and recognizes mouse but not human HMGA2. Secondary anti–mouse IgG and anti–rabbit IgG coupled to horseradish peroxidase were purchased from GE Healthcare. The enhanced chemiluminescence detection system was purchased from Santa Cruz Biotechnology, Inc. For immunofluorescence, cells were treated as indicated in the figure legends, fixed, and stained with tetramethyl-rhodamine isothiocyanate (TRITC)–labeled phalloidin (Sigma-Aldrich) or with rat anti–ZO-1, mouse anti–E-cadherin, and rabbit anti-HMGA2 antibodies as primary antibodies and TRITC-conjugated goat anti–rat IgG and FITC-conjugated anti–mouse or anti–rabbit IgG antibodies as secondary antibodies (Jackson ImmunoResearch Laboratories) as described previously (). Photomicrographs were obtained by a microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) with a digital camera (C4742-95; Hamamatsu), using Plan-neofluar 40×/0.75 objective lens (Carl Zeiss MicroImaging, Inc.) and photographing at ambient temperature in the absence of oil immersion. Primary images were acquired with the camera's QED software. Image memory content was reduced, and brightness contrast was adjusted using Photoshop 6.0 (Adobe). HepG2 cells were transiently transfected with calcium phosphate and a panel of mouse promoter luciferase constructs as described previously (). The and promoter luciferase constructs were provided by A. Cano (Universidad Autónoma de Madrid, Madrid, Spain) and L.R. Howe (Cornell University, New York, NY), respectively. All promoter constructs were cotransfected with the normalization reporter plasmid pCMV–β-Gal and the expression vector pcDNA3 (mock vector) or pcDNA3-HA-hHMGA2 for and promoter analysis or pcDNA3-HA-Smad2 SA for promoter analysis. The enhanced luciferase assay kit from BD Biosciences was used. Normalized promoter activity data are plotted in bar graphs representing mean values from triplicate determinations with standard deviations. Each independent experiment was repeated at least twice. The equivalent of 10 cells was used per ChIP reaction. Cross-linking was performed using 1% formaldehyde followed by neutralization with 0.125 M glycine. Cells were lysed in 1% SDS, 10 mM EDTA, 50 mM Tris, pH 8.1, and protease inhibitors on ice. DNA was sheared by sonication to 200–1,000 bp in length. Sonicated cell pellets were diluted 10 times in a buffer containing 0.01% SDS and were precleared with protein A–Sepharose in the presence of BSA and salmon sperm DNA before incubation with 5 μg rabbit anti-Smad4 antibody (H-552; Santa Cruz Biotechnology, Inc.) or preimmune rabbit antiserum as a negative control. Protein–DNA complexes were precipitated with protein A–Sepharose in the presence of BSA and salmon sperm DNA. Immunoprecipitated complexes were washed once with 150 mM NaCl, 0.2% SDS, 1% Triton X-100, 2 mM EDTA, and 20 mM Tris-HCl, pH 8.1; once with an identical buffer containing 500 mM NaCl; once with 0.25 M LiCl, 1% NP-40, 1% deoxycholate, 1 mM EDTA, and 10 mM Tris-HCl, pH 8.1; and twice with 10 mM Tris-EDTA. Immunoprecipitated complexes were eluted with 1% SDS and 0.1 M NaHCO. After reversal of cross-links by heating in 0.2 M NaCl, proteinase K treatment, purification by classical phenol-chloroform extraction, and ethanol precipitation, DNA pellets were resuspended in 20 μl of water. Input material before immunoprecipitation corresponds to one third of the immunoprecipitated material. Input DNA pellets were resuspended in 50 μl of water. PCR was performed using 2 μl of immunoprecipitated or 0.5 μl of input material using Taq DNA polymerase (Invitrogen). Amplification was performed for 30 cycles for immunoprecipitated DNA and 25 cycles for input DNA. Primer sets used to amplify different regions of the promoter are described in . Statistical analysis of real-time PCR quantification and promoter assays was performed by two-tailed paired test. Significance was considered at P ≤ 0.05. Fig. S1 demonstrates that is a direct gene target of TGF-β signaling and analyzes the effects of siRNA-mediated knockdown of Smads on EMT and expression. Fig. S2 demonstrates that ectopic HMGA2 does not induce autocrine TGF-β signaling. Fig. S3 analyzes the effects of siRNA-mediated knockdown of Hmga2 on its endogenous protein level and localization. Online supplemental material is available at .

In mammalian cells, DNA replication takes place in a defined spatio-temporal order. In general, euchromatic domains reside in the interior of the nucleus and replicate in early S phase, whereas heterochromatic domains localize to the nuclear periphery or near nucleoli and replicate late (; ). Although this spatio-temporal organization has been appreciated for some time, its functional significance is not understood. Because chromatin is assembled at the replication fork, temporal segregation could provide an important regulatory opportunity (). Indeed, reporter genes microinjected into mammalian nuclei at different times are assembled into different types of chromatin (), and the replication timing and subnuclear position of some genes is developmentally regulated (; ; ). By introducing G1-phase nuclei into a cell-free replication system, we previously demonstrated that the replication timing program is established at a discrete point during early G1 phase termed the timing decision point (TDP; ; , ). Intriguingly, the subnuclear spatial repositioning of chromosomal domains as well as the clustering of synchronously firing replication origins occurs during this same brief window of time. A similar early G1-phase event may regulate subnuclear position and replication timing in budding yeast (; ). What has not been clear is whether the replication program for constitutive heterochromatin is also reestablished in each cell cycle. In this study, we have examined the establishment of late replication for mouse pericentric heterochromatin. Chromocenters contain a large central core of pericentric heterochromatin consisting of γ-satellite DNA repeats packaged into chromatin that contains histone H3 trimethylated at lysine 9 (MeK9H3; ; ). Trimethylation, which is performed by the Suv39h1,2 histone methyltransferases, creates a high affinity (albeit context dependent) binding site for HP1α and -β, which become concentrated within chromocenters (; ; ). HP1 localization to chromocenters is a logical candidate for a replication timing determinant given the parallels between HP1 proteins and budding yeast silent chromatin (Sir) proteins (; ). Sir proteins concentrate at clusters of telomeres anchored to the nuclear periphery. Telomere clustering creates a sink for Sir proteins, sequestering them at the periphery and preventing them from silencing the active genome (). Telomeres replicate late, and Sir proteins are required for the late replication of some yeast replication origins (; ). We find that the replication timing program of chromocenters is reestablished coincident with the reorganization of pericentric heterochromatin into chromocenters. Thus, the TDP affects many types of chromatin simultaneously. However, Suv39h1,2-mediated trimethylation of K9H3 and the interaction of HP1 with chromatin were neither necessary nor sufficient for the establishment of late replication at the TDP. Instead, Suv39 was required for a partial delay of chromocenter replication relative to other late-replicating domains, demonstrating that a global timing program is established independently from additional factors that fine-tune the replication program. Mouse chromocenters are replicated in the second half of S phase (; ). To determine whether the replication timing of chromocenters is established during early G1 phase, mouse C127 cells were synchronized in mitosis and released into G1 phase for different time periods (G1 1, 2, and 3 h). Intact nuclei were introduced into egg extracts, and DNA synthesized in vitro was pulse labeled with biotin-dUTP at various times after the initiation of replication. FISH with a mouse γ-satellite (major satellite) DNA probe was used to visualize pericentric heterochromatin, and the colocalization of FISH signals with replicated DNA, which was identified by staining with labeled avidin, was monitored as an indication of chromocenter replication (). With G1 1- and 2-h nuclei, chromocenters were replicated at the earliest detectable signs of DNA synthesis, indicating a lack of temporal specificity. In contrast, replication of chromocenters within G1 3-h nuclei was significantly delayed, indicating that the late replication program for chromocenters was established between 2 and 3 h after mitosis. Because the overall rate of replication was identical between all three populations of nuclei (), the early replication of chromocenters in G1 1- and 2-h nuclei was not simply caused by an increased rate of DNA synthesis. This window of time (2–3 h after mitosis) is later than the previously characterized TDP in CHO cells (∼1–2 h after mitosis; ; , ). To determine whether this difference was the result of cell type–specific differences in a global TDP or to a distinctly later establishment for replication timing of constitutive heterochromatin, we evaluated the global spatio-temporal replication program in C127 (). Asynchronously growing cells were pulse labeled with BrdU and chased for a period of time (4–5 h) that was optimized to obtain the highest percentage of mitotic cells in late S phase during the BrdU pulse (∼50% of mitotic figures display BrdU label, with >95% of the label from late S phase). Nuclei from cells released into G1 phase for 1, 2, or 3 h were introduced into egg extract, and the sites of earliest DNA synthesis in vitro (first 30 min) were monitored by biotin-dUTP incorporation (). Colocalization of late-replicating BrdU label with biotin-labeled early in vitro DNA synthesis indicates a lack of temporal specificity. In these experiments, the presence or absence of colocalization could be clearly distinguished manually using a dual (red/green) filter (scoring for yellow coloration), obviating the need for the cumbersome colocalization analysis performed in . This difference may be caused by the enhanced preservation of 3D structure in nuclei that have not been denatured for FISH analysis. As shown in , most G1 1- and 2-h but very few 3-h nuclei displayed yellow foci, demonstrating that the overall replication timing program in mouse C127 cells is established between 2 and 3 h after mitosis. This later G1-phase timing for the TDP in C127 versus CHO cells may be the result of a 2-h longer G1 phase in these cells relative to CHO cells (unpublished data). To determine the time at which chromosome domains become repositioned after mitosis, late S BrdU label was tracked in G1 nuclei. Because a large percentage of late S-phase DNA synthesis takes place at the nuclear periphery, we quantified the percentage of nuclei that had repositioned the BrdU label to the nuclear periphery. As show in , with G1 1- and 2-h nuclei, <10% of BrdU-positive cells displayed the label in a peripheral pattern, whereas this percentage reached nearly 50% (a plateau level equivalent to that obtained at much later times in the cell cycle) by 3 h after mitosis. We conclude that although different cell types may reestablish replication timing at slightly different times after mitosis, the TDP is nonetheless coincident with the repositioning of chromosome domains and simultaneously affects many different types of chromatin. Although the large blocks of pericentric AT-rich satellite DNA are readily visible by FISH or DAPI staining throughout the cell cycle, we could nonetheless observe their reorganization into more regularly shaped structures during the TDP transition (). Because K9H3 trimethylation and binding of HP1 proteins are implicated in the assembly of heterochromatin (), we examined the presence of MeK9H3 and HP1 proteins within chromocenters as cells pass through the TDP. MeK9H3 was concentrated within DAPI-dense regions before the TDP (), and the total amount of MeK9H3 in cells was unchanged during this time (). Because we and others have shown that MeK20H4 is also enriched at pericentric heterochromatin (; ) and some studies suggest that this modification may be cell cycle regulated (; ), we also examined its abundance during the TDP, but no change was detected. Therefore, these two histone modifications within pericentric heterochromatin are not sufficient to establish late replication. Previous experiments in () and mammalian cells (; ) have demonstrated that the majority of HP1 dissociates from chromatin during mitosis and reassociates thereafter, making HP1 an interesting candidate for a protein involved in resetting replication timing at the TDP. We confirmed by both immunofluorescence and live cell imaging of GFP-tagged HP1 proteins that HP1α and -β were largely dispersed during mitosis (unpublished data). However, all detectable HP1α and -β rebound to chromatin by anaphase (not depicted) and could clearly be seen concentrated within the DAPI-dense regions in G1 1-h nuclei (). To determine whether any change in the affinity of HP1 proteins for chromatin coincided with the TDP, we extracted soluble cellular proteins from pre- and post-TDP cells with nonionic detergent at various salt concentrations (). These results revealed that approximately half of HP1 proteins were soluble or readily dissociated from metaphase chromatin. However, by 1 h after mitosis, all detectable HP1α and -β were very tightly associated with chromatin, with no detectable change in affinity at the TDP (). We conclude that the HP1–MeK9H3 interaction in pericentric heterochromatin takes place before the TDP and, therefore, is not sufficient to establish the late replication timing program of chromocenters. To investigate whether HP1 association is necessary for late replication, we took advantage of the cell-free nature of our system to remove HP1 from post-TDP chromatin before in vitro replication using a peptide mimicking the methylated H3 tail (). Nuclei from cells synchronized 3 h after mitosis were incubated with a trimethylated peptide consisting of the first 20 amino acids of histone H3 (). As controls, aliquots of the same nuclei were incubated with either the unmethylated form of the same peptide or no peptide. Incubation with the trimethylated but not the control peptide resulted in the solubilization of 30–40% of total HP1 protein () and the removal of almost all detectable HP1 at chromocenters (). In fact, HP1 remained bound to chromatin surrounding chromocenters but was selectively removed from the DAPI-dense chromocenters themselves. This indicates that MeK9H3 is a primary binding site for HP1 in chromocenters, whereas HP1 at other sites is bound to other components of chromatin known to tether HP1 (; ). These nuclei were then introduced into a egg extract, and the colocalization of the earliest in vitro DNA synthesis with γ-satellite DNA was evaluated. Depletion of HP1 at chromocenters had no significant effect on the timing of these domains or the total rate of in vitro DNA synthesis (). The aforementioned experiments demonstrate that the HP1– MeK9H3 interaction is not sufficient for establishing the late replication of chromocenters. This was surprising in light of recent links between chromatin structure and replication timing in both mammalian (; ) and yeast systems (for review see ) and the role of HP1 proteins in the formation of pericentric heterochromatin (). To address whether this interaction has any role in chromocenter replication timing, we examined chromocenter replication in cells lacking Suv39h1 and Suv39h2 (). Mouse embryonic fibroblasts (MEFs) derived from mice lacking both of these enzymes (double null; Suv39dn) have <30% of the total amount of cellular MeK9H3 and no detectable MeK9H3 within chromocenters (; ). In these cells, HP1 proteins remain tightly bound to chromatin but are depleted from chromocenters (Fig. S1, available at ). This is consistent with the results in indicating that HP1 is tethered to chromocenters through MeK9H3 but binds to other chromatin sites via other mechanisms. Reintroduction of Suv39 activity by stable transfection with a Suv39h1 expression vector () partially restores MeK9H3 and HP1 at chromocenters (Fig. S2), providing an “add-back” control to verify that any differences are the result of the loss of Suv39 activity. To determine the timing of chromocenter replication in these three cell lines, we used a retroactive synchrony method that is commonly used to analyze replication timing of specific gene sequences (). This method avoids the need for cumbersome cell line–specific synchronization methods that can perturb the cell cycle. After pulse-labeling nascent DNA with BrdU, cells were retroactively sorted by flow cytometry into populations in different stages of S phase (). Genomic DNA was isolated from each fraction, and nascent (BrdU substituted) DNA was immunoprecipitated with anti-BrdU antibodies. Aliquots of these nascent strand preparations were immobilized on nylon filters and hybridized with probes containing either the major or minor satellite DNA repeats () that characterize pericentric and centromeric DNA, respectively (). As controls, we monitored the replication of α- and β-globin genes (not depicted), which are early and late replicating, respectively, and mitochondrial DNA (), which replicates throughout the cell cycle and is equally represented in nascent DNA preparations from all cell cycle times (; ; ). As shown in , minor satellite DNA replicated at a distinctly earlier time during S phase than major satellite, but we could detect no significant difference in the replication program of these DNA sequences in either Suv39dn MEFs or the rescued add-back cell line. The molecular analyses in confirm the in vitro studies in and demonstrate that Suv39h1,2 and the HP1–MeK9H3 interaction are neither necessary nor sufficient for late replication of pericentric heterochromatin. However, we did notice a slight but not statistically significant advance in the replication time of major satellite DNA in D15 (). Small changes in replication timing are better revealed by cell-based assays. For example, the small differences in replication timing of imprinted and immunoglobulin genes can be detected with cell-based assays (; ) but not molecular analyses (). Thus, we evaluated the replication time of chromocenters in individual cells using a pulse-chase-pulse method that also does not require cell synchronization (). Cells were pulse labeled with 5′-chloro-2′-deoxyuridine (CldU), chased for different lengths of time, and subsequently pulse labeled with 5′-iodo-2′-deoxyuridine (IdU). Sites of CldU and IdU incorporation were detected by immunofluorescence with CldU- and IdU-specific antibodies (). The change in distribution of these two labels within individual nuclei after different chase times reveals the temporal order of these replication patterns. Exemplary merged CldU and IdU images in Suv39dn MEFs are shown in . The overall spatial arrangement and temporal order of the typical six mouse replication patterns (), which are described in detail in , were largely unchanged between wild-type and Suv39dn MEFs, demonstrating that MeK9H3 is not necessary for maintaining the overall spatio-temporal replication program. However, careful inspection of the patterns in Suv39dn MEFs revealed that a true pattern III was not observed (). Rather, it appeared as if chromocenter replication (normally specific to pattern IV) was already taking place in cells that were otherwise characteristic of pattern III. To specifically evaluate the time of chromocenter replication relative to the global replication program, we calculated the length of time it took for cells to progress from spatial replication patterns characteristic of early S phase to the time of chromocenter replication (). This was quantified as the percentage of nuclei that had CldU-labeled early S-phase patterns and IdU-labeled chromocenters at different chase times. These results revealed that chromocenter replication was advanced in the Suv39dn MEFs relative to wild-type MEFs by 10–15% of S phase. Reintroduction of Suv39 activity in the add-back cell line mostly restored this slight delay in chromocenter replication (). In the course of the experiments described in , we recognized the presence of a prominent body of chromatin that was intensely labeled with BrdU during pattern III DNA synthesis and replicated synchronously with chromocenters only in Suv39dn mutant MEFs (). This body also stained intensely with an antibody specific to trimethylated lysine 27 of histone H3 (), a modification that is highly enriched in the inactivated late-replicating X chromosome (; ). Given that these MEFs were derived from a female mouse embryo, it is very likely that this body is the late-replicating inactive X chromosome (Xi), and, for the purposes of discussion, we will refer to it as the Xi. Importantly, in wild-type MEFs () and in Suv39dn MEFs rescued by Suv39h1 add-back (), chromocenter replication took place distinctly after replication of the Xi. In contrast, Suv39dn MEFs replicated chromocenters simultaneously with replication of the Xi () in 75% of Xi-labeled cells. Because the Xi is not enriched for MeK9H3 () and chromocenters can be seen to replicate in an otherwise pattern III along with the last remnants of the small internal euchromatic foci (), we interpret these results as an advance of chromocenter replication rather than a delay in Xi replication. These results provide an internally controlled reference and, together with the data in , demonstrate that the Suv39-mediated trimethylation of H3K9 is responsible for a 10–15% delay in replication time. d e m o n s t r a t e t h a t t h e p r o p e r t y o f l a t e S - p h a s e r e p l i c a t i o n f o r c h r o m o c e n t e r s i s e s t a b l i s h e d d u r i n g e a r l y G 1 p h a s e c o i n c i d e n t w i t h t h e r e a s s e m b l y o f c e n t r o m e r e c l u s t e r s k n o w n a s c h r o m o c e n t e r s . T h i s e v e n t w a s c o i n c i d e n t w i t h t h e r e e s t a b l i s h m e n t o f a g l o b a l r e p l i c a t i o n t i m i n g p r o g r a m , d e m o n s t r a t i n g t h a t c o n s t i t u t i v e h e t e r o c h r o m a t i n i s s u b j e c t t o t h e s a m e d i s m a n t l i n g a n d r e a s s e m b l y o f r e p l i c a t i o n t i m i n g c o m p o n e n t s d u r i n g t h e c e l l c y c l e a s p r e v i o u s l y s h o w n f o r o t h e r c h r o m a t i n d o m a i n s . S u r p r i s i n g l y , t r i m e t h y l a t i o n o f K 9 H 3 a n d t h e h i g h a f f i n i t y b i n d i n g o f H P 1 t o p e r i c e n t r i c h e t e r o c h r o m a t i n w e r e n e i t h e r n e c e s s a r y n o r s u f f i c i e n t f o r l a t e r e p l i c a t i o n . H o w e v e r , i n c e l l s l a c k i n g S u v 3 9 , r e p l i c a t i o n t i m i n g o f c h r o m o c e n t e r s w a s s l i g h t l y a d v a n c e d r e l a t i v e t o w i l d - t y p e a n d S u v 3 9 - r e s c u e d c e l l s . T h e s e r e s u l t s r e v e a l s e p a r a t e g l o b a l v e r s u s f i n e - t u n i n g m e c h a n i s m s t h a t r e g u l a t e r e p l i c a t i o n t i m i n g a t p e r i c e n t r i c h e t e r o c h r o m a t i n . Mouse C127 cells were cultured and synchronized in mitosis as described previously (; for review see ). MEFs were cultured as described previously (), except with 10% FBS. Under these conditions, the lengths of S phase (measured as shown in ) were 7.2–9.4 h for wild type, 8.6–10.2 h for D15, and 9–9.7 h for D15 + Suv39h1. The Suv39h1-rescued D15 cell line was provided by J. Rice (University of Southern California, Los Angeles, CA). Intact nuclei were prepared from G1-phase C127 cells and introduced into egg extract as described previously (; ). Where indicated, nuclei were incubated with peptides in the described transport buffer containing 1.5% BSA. Peptides () were dissolved in distilled water at 10 mg/ml, and aliquoted stocks were stored at −70°C. At each time point, aliquots of reactions were removed and pulse labeled with 50 μM biotin-16-dUTP (Sigma-Aldrich) for 5–10 min. Reactions were stopped by diluting 1:10 in cold nuclear isolation buffer, and nuclei were adhered to glass slides and fixed as previously described (). Pericentric heterochromatin was detected by FISH with a γ-satellite DNA probe, and biotin-dUTP incorporation was detected with Texas red–conjugated streptavidin (GE Healthcare). The γ-satellite plasmid containing eight copies of the 234-bp satellite repeat () was labeled by nick translation with digoxigenin-11–dUTP (Life Technologies), and FISH was performed as previously described (). The pulse-chase-pulse protocol has been described in detail previously (; ). Note that the 10–15% replication timing advance was not reported in a study of embryonic stem cells that were deficient in Suv39 (), possibly because chromocenter domains are not as prominent in these embryonic stem cells. Immunofluorescence was performed as described previously (). Rat anti-HP1β antibody () was diluted 1:200 in blocking buffer (3% BSA in PBS/0.5% Tween). Rabbit anti-2xMeK9H3 () was diluted 1:1,000. AlexaFluor488- or -594–conjugated secondary antibodies (Invitrogen) were diluted ∼1:300 to 1:400. Preparation of whole cell extracts, chromatin isolation, and Western blotting were performed as described previously (). Stained specimens were observed with a microscope (Labophot-2; Nikon) equipped with a 100× 1.4 NA planApo oil immersion objective (Nikon), and epifluorescence images were collected with a CCD camera (SPOT RT Slider; Diagnostic Instruments). Deconvolution of stacked images collected at 0.5-μM intervals with QED Image software (Media Cybernetics) was processed with AutoDeblur software (AutoQuant Imaging, Inc.) using the Adaptive Blind setting. Confocal sections were obtained with a confocal microscope (MRC-1024; Bio-Rad Laboratories) mounted on a microscope (Eclipse 600; Nikon). Colocalization analysis was performed with LaserPix software (Bio-Rad Laboratories) as described previously (). Selected images were assembled using Adobe Photoshop. BrdU labeling, cell sorting, and immunoprecipitation of BrdU-labeled DNA was performed as described previously (). BrdU-labeled DNA from equal numbers of cells was immobilized on nylon membranes and hybridized with major (provided by N. Dillon, Medical Research Council, Clinical Sciences Centre, London, United Kingdom; ) and minor (pCR4 Min5-1; provided by T. Jenuwein, Research Institute of Molecular Pathology, Vienna, Austria; ) satellite DNA probes as well as a mouse mitochondrial probe (p501-1; provided by T. Brown, National Institutes of Health, Bethesda, MD; ). Probes were labeled by the random-priming method (Invitrogen). Membranes were hybridized and washed as described previously (for review see ), and relative counts per minute were obtained by phosphorimaging analysis (Molecular Dynamics). Values for major and minor satellite DNA were normalized to values for mitochondrial DNA hybridization, and the relative hybridization signal was presented as a percentage of the sum of these normalized values across all cell cycle fractions. Figs. S1 and S2 characterize the solubility of HP1 in Suv39dn MEFs and the restoration of HP1 binding to chromocenters when Suv39h1 activity is reintroduced into the Suv38h1,2-deficient MEFs. Online supplemental material is available at .

Cell cycle progression after DNA damage is rapidly halted by checkpoint controls, which are relaxed after the damage has been assessed and processed. Cells with misrepaired or unrepaired DNA lesions are eliminated by different cell death mechanisms (; ; ). One such mechanism is mitotic cell death (MCD), which is also known as “mitotic catastrophe,” a prominent but poorly defined form of cell death that is described mainly in morphological terms. MCD is an outcome of aberrant mitosis that results in the formation of giant multimicronucleated cells (; ). It is a major form of tumor cell death after treatments with ionizing radiation (IR) or certain chemotherapeutic agents (; ; ). MCD has been shown to prevail in cells with impaired G1, G2, prophase, and mitotic spindle checkpoint functions (; ; ). A prominent cell cycle checkpoint is activated by DNA double-strand breaks (DSBs) at the G2/M boundary. Activation of this checkpoint is driven by the nuclear protein kinase ataxia telangiectasia mutated (ATM), its downstream substrates p53 and the Chk1 and Chk2 kinases, Polo-like kinase 1 (Plk-1), and the p53-inducible proteins p21 and 14-3-3-σ. The p53-mediated arm of the G2/M checkpoint was shown to be pivotal in preventing MCD in DNA-damaged cells (; ), although some studies challenge this observation (; ). MCD has been assumed to result from the entry into mitosis of cells with unrepaired DNA damage, although a mechanism linking DNA lesions with mitotic abnormalities has yet to be uncovered. One of the early steps in the chain of events that culminates in MCD is cell entry into premature mitosis (; ; ). To date, evidence of premature mitosis in damaged cells relies primarily on the appearance of uneven chromatin condensation (UCC), which is the formation of hypercondensed chromatin aggregates at nucleolar sites (; ; ). The mechanisms underlying this phenomenon are unknown. During normal progression through mitosis, the structural reorganization of chromatin into condensed chromosomes entails the multiprotein complexes condensin I and II (; ; ). In vitro studies showed that condensin I possesses a DNA-stimulated ATPase activity and is capable of introducing constrained, positive supercoils into DNA (). This activity is believed to be essential for initiating the assembly of mitotic chromosomes and for proper assembly and orientation of centromeres (; ). The two condensin complexes are each composed of five subunits conserved from yeast to mammals (; ; ). The core complex common to both condensins consists of the structural maintenance of chromosomes (SMC) proteins CAP-E/SMC2 and CAP-C/SMC4. Two other members of this family, SMC1 and SMC3, form the core of the cohesin complex that plays a central role in sister chromatid cohesion (). Each condensin complex then contains a regulatory subcomplex consisting of three non-SMC proteins. In condensin I, these are CAP-D2, -G, and -H. CAP-D2 and -G possess a highly degenerate repeating motif known as the HEAT repeat (), whereas CAP-H belongs to a recently identified superfamily of proteins termed kleisins (). In condensin II, the regulatory subcomplex contains the proteins CAP-D3, -G2, and -H2 (; ; ). During interphase, both types of condensins appear to be localized in the cytosol and the nucleus (; ), with condensin I being predominantly cytosolic and condensin II being primarily nuclear (; ). The two condensin complexes localize in different places along mitotic chromosomes assembled in vitro and in vivo, suggesting distinct functions in chromosome architecture (; , ). Importantly, in vitro studies indicated that cyclin-dependent kinase 1 (Cdk1)–mediated phosphorylation of the non-SMC subunit set is required for chromosomal localization of condensin I and stimulation of its supercoiling activity (, ). Another prominent type of chromatin condensation noted in mammalian cells is apoptosis-related condensation. It is believed to be mediated primarily by the nuclear protein acinus (apoptotic chromatin condensation inducer in the nucleus) after its cleavage by activated caspase 3 (). Acinus lacks the DNase activities exhibited by other cellular factors that can induce apoptotic chromatin condensation via DNA fragmentation. Therefore, acinus functions as a “pure” regulator of apoptosis-related chromatin condensation (). In this study, we show that induction of the highly cytotoxic DSBs in DNA of cells with compromised p53-mediated G2/M checkpoint functions triggers UCC and premature mitosis. We demonstrate that the unscheduled activation of Cdk1 and the recruitment of activated condensin I to damaged chromatin are specifically involved in UCC formation. Condensin II and cohesin proteins are not involved in this process. Importantly, the acinus-mediated pathway, which is responsible for apoptotic chromatin condensation, was also not found to operate in UCC. Using a panel of tumor cell lines, we show that the defective damage-induced, p53-mediated G2/M checkpoint is an important but not a sole requirement for the activation of this pathway. DNA damage–induced MCD, with UCC as an early step, has been shown to occur at high rates in tumor cells in which cell cycle checkpoint functions are impaired (). We studied this phenomenon in human cervix carcinoma HeLa cells in which the G1 and G2 DNA damage checkpoint functions are compromised as a result of p53 inactivation by human papilloma virus E6 (). UCC was induced in exponentially growing HeLa cells by treatment with 10 Gy IR or 200 ng/ml neocarzinostatin (NCS), a radiomimetic agent that intercalates into cellular DNA and induces DSBs (). 24 h later, confocal microscopy revealed that cellular DNA was condensed into globular clumped structures typical of UCC ( and Fig. S1 A, available at ). This phenomenon was observed with two different DNA dyes: DAPI and Yo-Pro-1, a monomeric green fluorescent cyanine with a high affinity for double-stranded DNA (). Importantly, these globular structures contained the condensin core subunits SMC2 () and SMC4 (Fig. S2), whose nuclear localization changed after treatment from a diffuse, pannuclear pattern to a compact one that overlapped the UCC bodies. Western blotting analysis of chromatin fractions confirmed the recruitment of SMC2 and SMC4 to chromatin in response to DNA damage (). Accordingly, DNase-I treatment in situ led to dissociation of the UCC bodies and release of the associated SMC2 (). DSB-induced UCC was dose dependent and maximized at 24 h after treatment (Fig. S2). The percentage of cells exhibiting UCC increased from 5 ± 2% in untreated cells to 65.8 ± 25.1% 24 h after 200 ng/ml NCS (P < 0.01; = 6). Importantly, the spatial distribution of the cohesin subunit SMC1 was not altered (unpublished data), suggesting that the cohesin complex was not involved in this phenomenon. UV irradiation did not induce this process (Fig. S1 B), suggesting that it was associated specifically with DSBs. Importantly, a similar rate of UCC induction was observed in both ATM-proficient and ATM-deficient HeLa cells in which ATM had been knocked down using stable expression of the appropriate short hairpin RNA (shRNA; Fig. S3, available at ). According to our Nomarski images (), UCC structures colocalized with nucleoli, a phenomenon that has been previously reported (; ; ). However, although the mean number of nucleoli in untreated cells was about three per nucleus, most of the nuclei with damage-induced UCC contained a single nucleolar body (). These nucleolar bodies were significantly larger than normal nucleoli: their major axis was measured as 5.82 ± 0.93 μm in damaged cells versus 3.06 ± 1.35 μm in untreated cells (P < 0.001). To examine this phenomenon further, nucleoli were visualized by immunostaining with antibodies against the nucleolar proteins GNL3 (guanine nucleotide-binding proteinlike 3; nucleostemin; ; ) and nucleolin (). Although UV irradiation led to nucleolar disruption and release of nucleolar proteins into the nucleoplasm as previously described (; ), in NCS-treated cells, the nucleolar proteins remained within a single nucleolar body that colocalized with the UCC bodies (). HeLa cells are capable of activating the apoptotic cell death pathway (), making them suitable for determining whether the recruitment of condensin to chromatin is specific to UCC or also accompanies apoptotic chromatin condensation. HeLa cells were treated with NCS or staurosporine, a documented apoptosis inducer (), and the induction of apoptosis was monitored by following poly (ADP-ribose)polymerase-1 (PARP-1) cleavage (; ). The apoptotic chromatin condensation was morphologically distinct from damage-induced UCC and, unlike UCC and mitotic chromatin condensation, did not involve condensin recruitment (). In fact, in apoptotic cells, the chromatin-bound SMC2 fraction was reduced compared with untreated cells (), conceivably because of apoptosis-associated DNA fragmentation, which subsequently released SMC2 from chromatin. To further demonstrate the differential recruitment of SMC2 to chromatin upon different types of chromatin condensation, we compared its distribution with that of HP-1α (heterochromatin protein 1α). HP-1α is known to constitutively associate with condensed chromatin in interphase and mitotic cells (). We found that HP-1α was associated with condensed DNA in all three types of chromatin condensation that we studied (mitotic, apoptotic, and UCC), whereas SMC2 was loaded onto DNA in mitosis and upon UCC but not upon apoptotic chromatin condensation (). Interestingly, HP-1α and SMC2 were colocalized in normal mitotic chromosomes and in UCC bodies, but HP-1α was observed mainly in subcompartments of these structures, particularly in the inner core of the UCC bodies (Fig. S4, available at ). SMC2 and SMC4 constitute the common cores of the condensin I and II complexes (; ). The condensin I–related protein hCAP-H (human chromatin-associated protein H; kleisin-γ) and the condensin II–related protein hCAP-H2 (kleisin-β) are thought to be the first of the non-SMC subcomplex proteins to bind to SMC2/SMC4 heterodimers, subsequently recruiting other non-SMC subunits to form the complete type I or II condensin complexes, respectively (). We found that the condensin I–associated protein hCAP-H but not condensin II–related hCAP-H2 was associated with damage-induced UCC bodies (). Accordingly, biochemical analysis of cytosolic, nucleoplasmic, and chromatin fractions confirmed that hCAP-H, similar to condensin core subunits, was recruited to the damaged chromatin, but hCAP-H2 was not (). These results suggest that of the two condensin complexes, condensin I but not condensin II is involved in damage-induced UCC. In the aforementioned experiments, we noticed slower gel migration of chromatin-bound hCAP-H compared with soluble hCAP-H (), suggesting that chromatin-bound hCAP-H underwent posttranslational modifications. During mitotic chromatin condensation, the non-SMC subunits of condensin I and presumably condensin II undergo Cdk1/cyclin B–mediated phosphorylation, which is required for condensin targeting to chromatin and its subsequent functionality (, ). We asked whether hCAP-H that had been recruited to chromatin upon UCC was phosphorylated. We treated the chromatin fractions with λ-phosphatase before Western blotting analysis and found that this treatment indeed abolished the shift in hCAP-H migration (). This result suggests that similar to its recruitment to mitotic chromatin, condensin I is recruited to chromatin during UCC as an activated complex. UCC is a morphological hallmark of premature mitosis that is typical of cells with a defective G2/M checkpoint (). Our data suggest that recruitment of condensin I to chromatin is a biochemical marker of this process. To substantiate the link between a defective G2/M checkpoint and premature mitosis, we examined this process in HeLa cells, in which the G2/M checkpoint is compromised, and the human osteosarcoma cell line U2OS, in which p53 and the G2/M checkpoint are functional. Significantly, U2OS cells exhibited neither UCC nor condensin recruitment to chromatin after the same radiomimetic treatment that induced them in HeLa cells (), even after prolonged time periods (up to 48 h) following high NCS doses (≥200 ng/ml). One of the biochemical markers of the G2→M transition is elevated phosphorylation of histone 3 on Ser10, which has been associated with the loading of condensins onto chromatin and chromatin condensation at the early stages of mitosis (; ). We noticed that UCC-inducing treatments led to histone 3 hyperphosphorylation in HeLa but not in U2OS cells (), suggesting that DNA damage in HeLa cells indeed leads to premature mitosis. To substantiate this conclusion, we used FACS analyses to demonstrate the G2/M and mitotic fractions in both cell types after DNA damage. The results () showed that a vast majority of the HeLa cells were at G2/M 24 h after the treatment, with >70% of the cells being in mitosis. In contrast, the G2/M population in U2OS cells consisted mainly of cells at G2; here, the mitotic fraction decreased from 1.3 to 0.4% after NCS treatment. Accordingly, the G2/M arrest in U2OS cells was correlated with elevated levels of p53 and p21 (), which are both required for sustaining the G2/M checkpoint (). Because damage-induced premature mitosis in HeLa cells was characterized by histone 3 phosphorylation, a hallmark of normal mitosis, we examined whether premature mitosis was also associated with Cdk1 activation, which is an important requirement for the initiation of normal mitosis (). The activation status of Cdk1 can be monitored by comparing the extent of its phosphorylation on Tyr15 (inhibitory phosphorylation) and Thr161 (activating phosphorylation; ; ). Therefore, Cdk1 activation was compared in HeLa and U2OS cells after DNA damage. We found that in U2OS cells 24–48 h after treatment, Cdk1 exhibited the expected phosphorylation of Tyr15 that is typical for G2/M-arrested cells (). The level of cyclin B, Cdk1's regulatory subunit that is highly expressed at G2/M and decreases as the cells progress through mitosis, was also markedly elevated 24–48 h after NCS treatment and began to decrease thereafter (). These results indicate that the G2/M arrest in DNA-damaged U2OS cells lasted for at least 48 h after treatment. FACS analysis supported these conclusions (unpublished data). Interestingly, in HeLa cells, Cdk1 was hyperphosphorylated on both Tyr15 (inhibitory phosphorylation) and Thr161 (activating phosphorylation; ). Cyclin B levels were markedly elevated 24 h after NCS treatment but declined significantly thereafter (). We compared this unique Cdk1 phosphorylation pattern with Cdk1 phosphorylation in HeLa cells artificially arrested in mitosis using colcemid treatment. In metaphase-arrested cells, the expected dephosphorylation of Tyr15 occurred concomitantly with the hyperphosphorylation of Thr161 (). This experiment allowed us to demonstrate the sharp difference between this pattern of Cdk1 phosphorylation and the one observed in DNA-damaged cells, in which Cdk1 was hyperphosphorylated on both Thr161 and Tyr15 (). Cell fractionation further revealed that after NCS treatment, cyclin B/pT161-Cdk1 (activated complex) was predominantly nuclear, whereas inactivated Cdk1, which was tagged by pTyr15, was largely sequestered in the cytosol (). These findings indicate that the induction of DSBs leads to unscheduled Cdk1 activation with the appearance of two Cdk1 pools exhibiting distinct subcellular localizations. Collectively, our data provide insights into the mechanism of damage-induced UCC and suggest that the unscheduled activation of Cdk1, recruitment of condensin I to chromatin, and UCC together lead to (and represent) a hallmark of premature mitosis and could serve as useful markers for distinguishing premature from regular mitotic events. We asked whether cells that react to extensive damage with UCC and premature mitosis (HeLa) will activate a different death pathway compared with cells that activate the G2/M checkpoint in the face of such damage and do not exhibit premature mitosis (U2OS). The results are summarized in . 3 d after damage infliction, U2OS cells exhibited typical morphological and biochemical features of apoptosis: pyknosis with the condensation of chromatin, nuclear fragmentation, phosphorylation of Ser46 of p53 (), activation of caspase 3, and subsequent cleavage of its downstream substrates such as acinus (a marker of apoptosis-related chromatin condensation) and PARP (). On the other hand, the same DNA damage led in HeLa cells to the appearance of giant multimicronucleated cells, which is a hallmark of MCD (). Caspase 3, acinus, and PARP remained intact in these cells (). FACS analysis supported the occurrence of apoptosis in U2OS cells, with the sub-G1 fraction increasing approximately ninefold, whereas the same treatment in HeLa cells resulted in only a 1.7-fold increase in the sub-G1 fraction (unpublished data). The role of the p53-mediated G2/M checkpoint in MCD is currently unresolved (; ; ; ). Data reported here imply that a compromised p53-mediated G2/M checkpoint might be associated with MCD, but the role of p53 in preventing MCD requires further substantiation. We examined MCD in a panel of tumor cell lines with different p53 constitutions. We also stably knocked down p53 in U2OS cells by expressing in them an appropriate shRNA. The p53 status and functionality (measured by the ability to induce the gene encoding p21) in these cell lines are demonstrated in Fig. S5 A (available at ). After NCS treatment, UCC was observed in three cell lines in which p53 was impaired—HeLa, 293T, and HCT166 (p53) cells—but not in other p53-compromised cell lines ( and Fig. S5 B). Neither increasing the NCS dose (≥500 ng/ml) nor prolonging cell exposure to this drug (up to 48 h) changed these results (unpublished data). Accordingly, 72 h after NCS treatment, MCD was observed only in HeLa, 293T, and HCT166 (p53) cells ( and Fig. S5 C) but not in the cell lines that did not exhibit UCC. Notably, the highest levels of MCD was observed in HeLa cells (60–70%), whereas MCD levels in 293T and HCT166 (p53) cells reached only 15–20%. Apoptosis was the other prominent type of cell death noted to occur at high rates in all of these cell lines (with the exception of HeLa cells) after DNA damage (Fig. S5 C). MCD results from aberrant mitosis, which fails to produce proper chromosome alignment and subsequent chromosome segregation, and culminates in the formation of large polynucleated cells (; ). Abnormal mitosis in drug-treated or irradiated cells may proceed through several different pathways, but the final step is almost always the formation of nuclear envelopes around individual clusters of missegregated chromosomes (). At least two mechanisms have been proposed for MCD. The first one is based on the aberrant duplication of centrosomes that leads to multipolar mitosis and subsequent formation of micronuclei (). The second one is based on cell entry into premature mitosis, implying that cells proceed into mitosis before the completion of S or G2 phase. Such cells cannot properly compact their chromatin and segregate their chromosomes and must be eliminated. This mechanism offers an attractive explanation for DNA damage–induced MCD, but the evidence for premature mitosis in damaged cells has rested primarily on the morphological observations of UCC, whose mechanism is elusive (). Thus, elucidation of the molecular pathways underlying these processes is a prerequisite to understanding MCD. In this study, we provide some of the missing components of this process. Although damage-induced UCC appears, at least morphologically, to differ from other types of chromatin condensation in eukaryotic cells such as mitotic or apoptotic chromatin condensation, the question arises whether the mechanisms involved in mitotic or apoptotic chromatin condensations are involved in MCD-related UCC. We demonstrated that UCC engages some of the proteins involved in mitotic chromatin packaging but not the acinus-mediated mechanism that operates in apoptotic chromatin condensation. Condensin recruitment was found to be a common denominator of mitotic chromatin condensation and UCC, but UCC involved activated condensin I and not condensin II, thereby differing from mitosis, which entails the recruitment of both condensins to chromatin (; ). Because condensin recruitment is important for proper chromosome alignment and subsequent segregation, the exclusive targeting of condensin I to damaged chromatin might lead to the formation of unaligned, hypercondensed chromatin aggregates typical of UCC. Another difference between cell progression through normal mitosis and premature mitosis is unscheduled activation of Cdk1. In contrast to cells entering mitosis normally, Cdk1 in premature mitosis undergoes unscheduled activation, resulting in the appearance of two Cdk1 pools with different subcellular localizations. We showed that cyclin B/pThr161-activated Cdk1 is localized predominantly in the nucleus, whereas Cdk1, which is inactivated by Tyr15 phosphorylation, is sequestered primarily in the cytosol. This division of Cdk1 into two pools may reflect two populations of Cdk1 molecules, each characterized by different posttranslational modifications and each probably capable of responding to different stimuli or interacting with different regulators. DNA damage–induced UCC was noted previously to develop in close proximity to the nucleolar site (). Our findings, although confirming these observations, also indicate that the number of nucleoli in such cells was markedly reduced, mainly to a single enlarged nucleolus. We propose that the UCC might entail coalescence of the nucleoli. A similar change in nucleolar morphology resulting in a single, enlarged nucleolar body was shown in HeLa cells after DNA damage induced by the alkylating agent MNNG (), but the underlying mechanisms were unknown. Unscheduled activation of Cdk1 in cells with UCC might explain this phenomenon. Because Cdk1 was shown to play a role in the maintenance of functional nucleoli and perturbation of its activation was associated with dramatic changes in nucleolar organization (), it is reasonable to assume that the unscheduled activation of Cdk1 could lead to the changes in nucleolar morphology. Furthermore, alteration of the nucleolar morphology after DNA damage could result from the marked changes in the spatial distribution of condensin. It has been recently shown that condensin could take part in nucleolar organization by the arrangement of rDNA gene repeats into heterochromatic-like structures thorough its interactions with Sir2p (), a histone deacetylase that deacetylates histone tails to generate a hypoacetylated histone environment characteristic of heterochromatin. The UCC–MCD pathway was primarily observed in cells with compromised p53-mediated G2/M checkpoint functions (; ), although its pivotal role in preventing MCD has recently been questioned (; ). In our panel of several human cell lines with compromised p53, an impaired p53-mediated G2/M checkpoint was not the sole requirement for activation of the UCC–MCD pathway after extensive DNA damage. Presumably, the combination of compromised p53 and impaired function of other mediators of G2/M progression are responsible for this phenomenon. Among the possible candidates for such mediators are Chk1, Plk-1, and the Aurora A kinase, which were recently shown to play a key role in the reactivation of Cdk1 and onset of mitosis after DNA damage (). In addition, it has recently been shown that in cells with compromised p53, p73, another member of the p53-like family of proteins, can at least partly substitute for p53 in cell cycle regulation (), making p73 still another possible candidate for involvement in IR-induced MCD. It is noteworthy that ATM, the chief regulator of cellular responses to DSBs, was uninvolved in this pathway. In the face of DSBs, ATM mediates the pathways leading to cellular rescue and survival on the one hand and apoptosis on the other (; ). The ATM-dependent apoptotic response relies on functional p53 (; ). Our findings suggest that the UCC–MCD pathway is turned on in cells that are destined to die and cannot activate the ATM–p53-mediated apoptotic pathway. This is probably one of the very few DSB-triggered pathways that are ATM independent. This observation draws a line between DSB-induced processes that are under ATM jurisdiction and those that are not. In summary, our findings disclose a cell death pathway that is triggered by the extensive induction of DSBs, is ATM independent, and is distinct from the apoptotic pathway. It leads to the unscheduled activation of Cdk1 followed by premature mitosis characterized by UCC that is associated with recruitment of condensin I to chromatin and ultimately to MCD (). Thus, the unscheduled activation of Cdk1 and phosphorylation of the non-SMC subunits of condensin I could be sequentially linked. Considering the importance of this pathway in the response of tumor cells to radiotherapy and chemotherapy, further elucidation of its molecular steps is crucial for understanding the action of anticancer drugs and the rational design of therapeutic regimens. The following antibodies were used in this study: α-SMC2 (BL548) and α-SMC4 (BL551) were obtained from Bethyl Laboratories, Inc.; anti–hCAP-H and hCAP-H2 were provided by J.-M. Peters (Research Institute of Molecular Pathology, Vienna, Austria); α-SMC1 was a gift from R. Jessberger (Mount Sinai School of Medicine, New York, NY); α-nucleolin (7G2) was a gift from Y. Shav-Tal (Bar-Ilan University, Ramat-Gan, Israel); α–HP-1α was purchased from Chemicon; α-nucleostemin was obtained from R&D Systems; α-phospho-Cdk1 (pTyr15), α-phospho-Cdk1 (pThr161), α-phospho- p53 (Ser46), α-cleaved caspase 3 (5A1), and α–PARP-1 were obtained from Cell Signaling Technology, Inc.; anti-H2B, α-p53 (DO-1), α-Hsc70, and α-Cdc2 p34 were obtained from Santa Cruz Biotechnology, Inc.; α-p21 (C19) was obtained from Delta Biolabs; α-acinus (C terminus) and α-phospho-histone H3 were purchased from Upstate Biotechnology; α-tubulin was obtained from Sigma-Aldrich; and α–cyclin B was obtained from BD Transduction Laboratories. Secondary HRP-conjugated as well as rhodamine red– and FITC-conjugated antibodies were purchased from Jackson ImmunoResearch Laboratories. HeLa and U2OS cells were obtained from American Type Culture Collection, MDA-MD-231 and MDA-MB-435 cells were a gift from I. Tsarfaty (Tel Aviv University, Tel Aviv, Israel), and HCT116 cells and their p53-null variant (HCT116 p53 and HCT116 p53, respectively) were gifts from B. Vogelstein (Johns Hopkins University, Baltimore, MD). Cells were cultured in the recommended growth media under standard conditions. DNA damage was induced either by adding NCS to the cultures, by X irradiation using an irradiator (MGC40; Philips), or by UV-C irradiation using an irradiator (FLX-35M; Vilber Lourmat) at dose ranges of 30–50 J/m. Apoptosis and MCD were visualized by light microscopy. The cells were fixed in methanol, stained with Hemacolor reagents (Merck) as previously described (), and photographed under a light microscope (Eclipse TE 2000-5; Nikon) with a 20× NA 0.45 plain Fluor objective. All comparative images (treated vs. untreated samples) were obtained under identical microscope and camera settings. Cells grown on coverslips were treated with DNase-I diluted in Tris-buffered saline (25 mM Tris-HCl, pH 7.5) containing 1% BSA and supplemented with 5 mM MgSO for 1 h at 23°C) as described previously (). The cells were either directly processed for immunostaining or were processed after extraction of their cleaved DNA (with buffer containing 20 mM Tris-HCl, pH 7.5, 0.25 M ammonium sulfate, and 0.2 mM MgCl). Total cell lysates were prepared using radioimmunoprecipitation buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% IGEPAL, 0.1% SDS, and 0.5% sodium deoxycholate) supplemented with a mixture of protease and phosphatase inhibitors. Nuclear and cytosolic extracts were obtained as previously described (). In brief, cytosolic extracts were prepared using buffer A (10 mM Hepes, pH 7.9, 10 mM KCl, 1 mM EDTA, 1 mM EGTA, and 1 mM DTT supplemented with a mixture of protease and phosphatase inhibitors and 0.6% NP-40). Nuclear extracts were prepared by the dissolution of nuclei in buffer C (20 mM Hepes, pH 7.9, 400 mM KCl, 1 mM EDTA, 1 mM EGTA, 1 mM DTT, and a mixture of protease and phosphatase inhibitors) using vortex at 4°C for 45 min. Soluble chromatin fractions were subsequently prepared from the previous step by sonication (40% amplitude; three pulses at 20 s each on ice) in radioimmunoprecipitation buffer. Protein content was calibrated using the bicinchoninic acid protein assay reagent kit (Pierce Chemical Co.). The samples were subjected to standard Western blotting analysis. Immunoblots (polyvinylidene difluoride) were visualized using enhanced chemiluminescence (SuperSignal, Pierce Chemical Co.). Reactions were performed on solubilized chromatin fractions in 1× PTase buffer (50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 0.1 mM EGTA, 2 mM DTT, and 0.01% Brij 35) supplemented with MnCl using 400 U λ-protein phosphatase (λ-PPase; New England Biolabs, Inc.) per 100 μg of extracted proteins at 30°C for 1 h and were stopped by supplementation to 50 mM EDTA. The extracts were analyzed directly by immunoblotting. Cells were analyzed by two-dimensional flow cytometry after costaining with an antibody against phosphorylated histone 3 (H3-p, a mitotic marker) and propidium iodide (PI) for DNA quantitation. In brief, cells were fixed, permeabilized with 0.25% Triton X-100, incubated with the H3-p antibody, labeled with FITC-conjugated anti–rabbit IgG secondary antibody, treated with 5 μg/ml DNase-free RNase (37°C for 30 min), and stained with PI. Data were collected using FACSort flow cytometry (Becton Dickinson) with H3-p in the first and PI in the second dimension (10,000 events/sample). Cell cycle gating and analysis were performed using WinMDI software. Logarithmically growing cultures of HeLa cells were mitotically arrested by treatment with 0.1 μg/ml colcemid for 17 h as described previously (), collected by shake-off, and subjected to Western blotting analysis. FACS analysis confirmed that >90% of colcemid-treated cells were in M phase. p53 was stably knocked down in U20S cells by expressing the appropriate shRNA in them using the pRETRO-SUPER retroviral vector (). The sequence 5′-GATCCCCGACTCCAGTGGTAATCTACTTCAAGAGAGTAGATTACCACTG GAGTCTTTTTGGAAA-′3 was cloned in the vector using BglII and HindIII restriction sites. Data were analyzed using the two-tailed test. P values of <0.05 were considered statistically significant. Fig. S1 shows that IR and NCS but not UV trigger the recruitment of condensin core subunits to damaged chromatin and the formation of UCC bodies. Fig. S2 shows that condensin recruitment and UCC develop in HeLa cells as a delayed cellular response to severe DNA damage. Fig. S3 shows that ATM knockdown does not alter the rate of DSB-induced UCC. Fig. S4 shows that condensin and HP-1α strictly colocalize in normal mitosis but occupy different compartments in UCC bodies associated with premature mitosis. Fig. S5 shows that the defective p53-mediated G2/M checkpoint is important for condensin recruitment, UCC, and MCD but is not the sole requirement. Online supplemental material is available at .

Cytokinesis and septum deposition in budding yeast occur only after exit from mitosis has taken place. Exit from mitosis, which is defined biochemically as the destruction of mitotic cyclins such as Clb2p, depends on ubiquitin-mediated proteolysis. The anaphase-promoting complex, an E3 ligase, together with its activators Cdc20p and Hct1p, act sequentially to cause the proteolysis of Clb2p upon which cells enter G1 of a new cell division cycle (; ; ). An early observation indicating that cytokinesis depends on mitotic exit was the failure of cells to proceed to cytokinesis in the presence of a nondestructible Clb2p (). Furthermore, the mitotic exit network (MEN), which plays a role in the activation of Hct1p (; ), is implicated in cytokinesis and septum formation (for review see ). The MEN comprises Tem1p (a GTPase), Lte1p (a GTP/GDP exchange factor), Cdc15p, Cdc5p, Dbf2p, and Dbf20p (Ser/Thr kinases), Mob1p (a kinase), and Cdc14p (a phosphatase; ; ; ). Tem1p, Cdc15p, Cdc5p, and Dbf2p have been observed at the mother–daughter neck, linking them to the process of cytokinesis and/or septation (for review see ). For instance, the overexpression of mutant constructs of either Cdc15p or Cdc5p can give rise to cytokinesis defects, whereas a deletion in leads to a failure in cytokinesis. Both and mutants are defective in actomyosin ring constriction, whereas the mutant is also unable to form a proper septum. However, the exact roles of the MEN components at the neck with respect to cytokinesis and septum deposition remain undefined. Septum deposition starts with the laying down of a primary septum of chitin between the mother and daughter cells by chitin synthase 2 (Chs2p) during cytokinesis (for review see ; ; ; ). This occurs as the actomyosin ring contracts, which provides an inward force that leads to the invagination of the plasma membrane at the neck (; ). Chs2p, which is localized to the plasma membrane at the neck during this time, deposits a layer of chitin between the mother and daughter cells after the line of constriction of the plasma membrane (; ). The secondary septum is subsequently deposited around the primary septum from mother and daughter cells (for review see ; ; ; ). It was previously documented that Chs2p was found at the neck only in telophase (). Recently, it was shown that the stability of the Myo1p ring during cytokinesis is dependent on the presence of Chs2p (). Interestingly, the transcription of Chs2p peaks earlier on at G2/M phase (; ; ). These observations suggest that Chs2p neck localization and chitin deposition could be tightly coordinated with late mitotic events, as Chs2p normally arrives at the neck before Myo1p ring constriction. To date, it is not clear what determines the timely neck localization of Chs2p to the neck at the end of mitosis. Given the localization of Chs2p to the neck late in telophase (; ), we asked whether there exists a link between the mitotic exit and neck localization of Chs2p that could explain the occurrence of septum deposition subsequent to mitotic exit. In this study, we show that the translocation of Chs2p to the neck is dependent on mitotic kinase inactivation. Moreover, the inactivation of the mitotic kinase activity by Sic1p overexpression led to the localization of Chs2p to the neck in telophase in the absence of Cdc15p function, indicating that the MEN component of Cdc15p's role in Chs2p localization is through its function in promoting exit from mitosis. More importantly, premature inactivation of the mitotic kinase at metaphase was sufficient to send Chs2p to the neck. Overexpression of Sic1p resulted in the deposition of chitin in metaphase-arrested cells, implying that Chs2p prematurely localized at the neck was active. We further demonstrate that mitotic exit is required for the triggering of the COPII-dependent transport of Chs2p out of the ER. Our data provide evidence that the timely localization of Chs2p to the neck is regulated at the level of ER export by a decrease in mitotic kinase at the end of mitosis. It was previously published that Chs2p localizes late in telophase to the mother–daughter neck () ∼2.4 ± 0.8 min before spindle disassembly (). To further understand the regulation of Chs2p at the end of mitosis, we examined the neck localization of Chs2p over one cell cycle in wild-type cells carrying (FM119). FM119 cells arrested in G1 using α-factor were released into fresh medium at 24°C, and samples were collected for observation of the GFP fusion proteins and Clb2p and Cdc28p levels. Chs2p-YFP can be seen at the mother–daughter necks () at a peak of 17.33 ± 2.52% of the cells at 105 min after α-factor release (). This occurred upon Clb2p decrease from the 90- to 105-min time point (). The low percentage of cells exhibiting Chs2p-YFP signals was likely caused by the dynamics of Chs2p localization to the neck and its transport away for destruction shortly after. Indeed, from at least 10 time-lapsed sequences with 2-min intervals, it can be seen that Chs2p normally disappeared around 8 min (usually four frames after arriving at the neck) after going to the neck (unpublished data), which was in agreement with a previously published study (). We also examined in cells carrying (FM224) the relative timings of Clb2p destruction and Chs2p localization to the mother–daughter necks. The patterns of Clb2-YFP localization in our strain to the nuclei, spindle pole bodies (SPBs), spindles, and necks () are consistent with previously published data (; ) indicating that Clb2p driven from its endogenous promoter with only a single YFP fused to its C-terminal can be used for our subsequent studies. To enrich for cells going through mitosis, we examined FM224 cells that were synchronized in S phase by hydroxyurea (HU) arrest and were then released. Chs2p-YFP signals appeared only in cells in which Clb2p-YFP signals were low such that SPB, spindle, and neck signals were not visible but had low signals in the nuclei (). The correlation between Clb2p-YFP levels (Clb2p-YFP signals were counted as cells with clearly visible signals in nuclei, SPBs, spindle, and neck) and Chs2p-CFP signals supports this observation (). Furthermore, in at least 10 time-lapsed sequences, Chs2p-CFP was detected at the neck only upon a decrease in Clb2p-YFP signal intensity (; compare the intensity of Clb2p-YFP signals at 0 and 4 min, when faint Chs2p-CFP signals first appeared at the neck). This observation was consistent with the fixed samples, indicating that the decrease in Clb2p-YFP signals was not caused by bleaching during time-lapsed microscopy. Therefore, Chs2p neck localization correlates strongly with mitotic exit. This finding concurs with previous studies showing Chs2p neck localization in telophase (; ). showed recently that Sec3p-GFP, a component of the exocyst complex, can be found at the mother–daughter neck 3.90 ± 1.90 min before spindle breakdown and that its localization to the neck depends on mitotic exit. Proper Chs2p neck localization depends on Sec10p, another component of the exocyst complex (). However, in a mutant, Chs2p was still able to translocate to a region near the neck, albeit forming a fuzzy ring around the neck. These observations indicate that although Chs2p localization to a discrete band at the plasma membrane of the neck requires the exocyst complex, the regulation of Chs2p transport itself to the neck depends on determinants other than the exocyst complex. Given our aforementioned data showing a strong correlation between mitotic exit and Chs2p neck localization (), we tested whether Chs2p localization to the neck was dependent on mitotic exit in the temperature-sensitive mutant carrying (FM113). At 24°C, Chs2p-YFP in the cells exhibited similar dynamics as in wild-type cells (unpublished data). When cells (FM113) were arrested at 37°C in telophase, the cells could maintain Myo1p-CFP rings at the necks, although they were devoid of Chs2p-YFP neck signals (, top). When released from 37 to 24°C, a peak of 33.67 ± 4.93% (, top) of the cells exhibited Chs2p-YFP signals at the neck (, right; 60 min), upon Clb2p decrease (, bottom). This indicated that the translocation of Chs2p-YFP to the neck is an event downstream of Cdc15p function. To directly test the role of mitotic exit in translocating Chs2p-YFP to the neck, cells (FM145) were cultured at 37°C in yeast extract peptone YP)/raffinose (Raff) for 3 h to arrest the cells in telophase with high mitotic kinase. Galactose (Gal) was added to one half of the culture to induce Sic1p, whereas the other half remained in Raff. In the absence of Cdc15p function, Chs2p-YFP signals were detected at the necks (, right) only after Sic1p induction (, bottom). As there was a slight difficulty in triggering mitotic exit via Sic1p induction at 37°C (unpublished data), we saw a peak of 23.49 ± 7.71% of the Gal culture with Chs2p-YFP signals at the neck around 90 min (, top). Nonetheless, this was significantly higher than 0.61 ± 0.54% of the cells in the Raff culture with Chs2p-YFP neck signals (, top; 90 min). Our findings show that the localization of Chs2p to the neck does not depend on an active MEN component such as Cdc15p except for its role in triggering mitotic exit. To further assess the role of mitotic kinase in the localization of Chs2p to the neck, we next tested whether the inactivation of mitotic kinase at a point when Chs2p normally does not localize to the neck can prematurely send Chs2p to the neck. We arrested (FM143) cells at metaphase in YP/Raff containing nocodazole (Noc) at 24°C. Once the cells were arrested, half of the culture was maintained in YP/Raff, whereas Gal was added to a final concentration of 2% in the other to induce Sic1p expression (). Interestingly, the cells arrested in Noc (, top) and telophase ( and ) have a characteristic ringlike appearance, probably as a result of Chs2p-YFP's localization at the perinuclear ER (; see next section). After 90 min of Sic1p induction (), Chs2p-YFP localized to the mother–daughter necks at a peak of 51.89 ± 8.89% of the cells (). This is significantly higher than that in the control culture, where only 2.00 ± 1.73% of cells exhibited Chs2p-YFP signals at the necks. That Chs2p-YFP can be sent to the neck prematurely at metaphase upon Sic1p induction clearly indicates that mitotic exit was indeed sufficient to cause the translocation of Chs2p to the neck. Although it was possible to cause Chs2p to translocate to the neck by triggering mitotic exit in either telophase or metaphase ( and , respectively), it was not clear whether the prematurely localized Chs2p was in fact active at the neck. Using transmission EM, we examined the ultrastructural morphology of cells in which Sic1p was induced to assess whether Chs2p that prematurely localized can deposit a septum between the mother and daughter cells. Chs3p was previously shown to be involved in the formation of a remedial septum in the absence of proper Myo1p ring constriction (; ). strain (FM295). In experiments analogous to those described in the previous paragraph, FM295 cells were arrested in metaphase with Noc followed by the induction of Sic1p in one half of the culture, whereas the other half remained uninduced. In the Noc-arrested cells, the neck region was generally broad, with no septum laid across the neck (, top). In cells in which Sic1p was induced, electron-lucent septa formed across mother–daughter necks compared with the Raff control (, bottom; compare 120 min of Raff with Gal). 82.02 ± 14.31% of Sic1p-induced cells in three independent experiments had the thin septa compared with 5.11 ± 2.06% of uninduced cells (). Our results indicate that Sic1p induction during metaphase led to the deposition of chitin between the mother and daughter cells, most likely as a result of the premature neck localization of Chs2p (). From our aforementioned data, Chs2p-YFP in telophase- ( and ) and metaphase-arrested cells () localized to perinuclear structures before its neck localization. To determine the localization of Chs2p before its neck localization, we examined wild-type cells (FM182) arrested in metaphase. Chs2p-YFP signals colocalized with the perinuclear signals of Sec63p-CFP (, Noc), an ER marker (). This is consistent with immunofluorescence experiments showing perinuclear staining of Chs2p with anti-Chs2p antibodies (). Upon release from Noc, Chs2p-YFP signals dispersed from the ER and could be seen at the neck (, 100 min; and B). To exclude the possibility that the Chs2p-YFP we observed in Noc-arrested cells was an artifact of the metaphase arrest, we next tested whether Chs2p-YFP normally can be observed in the ER during progression through the cell division cycle. It was not easy to detect Chs2p-YFP in the ER in cells progressing through the cell division at 24°C, although the neck signals could be found in a few cells (unpublished data). This was most likely caused by the dynamic nature of Chs2p synthesis and export from the ER. To slow down the cell cycle, we first synchronized FM182 cells in S phase using HU, which is much earlier than metaphase, and then released the cells into YPD at the lower temperature of 16°C. As cells were released from the arrest, they progressed into the cell cycle at 16°C with an accumulation of cells with Chs2p-YFP signals that colocalized with Sec63p-CFP 270 min after the release (, bottom), confirming that in the normal cell cycle progression, Chs2p did in fact translocate into the ER and, thereafter, localized to the neck. To further characterize the transport of Chs2p from the ER to the neck, various mutant strains of the secretory pathway were used (). Temperature-sensitive mutants , , and together with the wild-type strain each carrying were generated. Cells were released into YPD at 37°C from a Noc arrest. Chs2p-YFP localized to the perinuclear ER at the Noc-arrested stage in all of the strains (, top). Chs2p-YFP signals at the neck in wild-type cells (FM119) peaked at 44.26 ± 11.67% (, bottom) around 50 min after release from Noc arrest. This coincided with mitotic exit, as judged by the decrease in Clb2 (, wild type [WT] at 50 min). However, the mutants that were treated similarly failed to transport Chs2p-YFP to the neck, although Clb2 levels had decreased. In cells (FM262), the loss of function of the GTP exchange factor required for budding of COPII vesicles prevented Chs2-YFP from exiting the ER, as evident from the perinuclear YFP signals (, at 50 min). In (FM260), some Chs2p-YFP signals were found as spots in the cytoplasm, whereas some Chs2p-YFP remained in the ER (, at 50 min), perhaps because of the fact that Sec18p is needed for the recycling of membranes to the ER for the continued export of Chs2p from the ER. In cells (FM258), only a peak of 3.56 ± 1.40% of cells exhibited Chs2p-YFP neck signals (, bottom). However, Chs2p-YFP signals had dispersed from the ER, giving rise to a punctate appearance (, compare at 50 min with at 50 min). This suggested that Chs2p-YFP had exited from the ER but that the transport vesicles carrying Chs2p-YFP through the secretory pathway failed to fuse to the plasma membrane at the neck. Although it had previously been hinted at by biochemical analysis () that Chs2p enters the ER during synthesis, we show that Chs2p colocalized with Sec63p, an ER marker during mitosis. Furthermore, Chs2p depends on the secretory pathway for its transport from the ER to the neck during mitotic exit, which is consistent with previous data (; ). To test whether the mitotic exit directly impinges upon the secretory pathway with respect to Chs2p transport to the neck, wild-type, , and strains each carrying were arrested in Noc in YP/Raff at 24°C. Each culture was then shifted to 37°C to inactivate the mutant components, and Sic1p was induced in one half of the culture for 60 min. At 37°C in the presence of Sic1p, the Chs2p-YFP signals could be detected at the neck (, top left) in 36.67 ± 6.03% of wild-type cells (FM143) compared with 9.00 ± 6.08% in control cells (, bottom). Although the inactivation of mitotic kinase activity was relatively inefficient at 37°C, the data are consistent with our aforementioned findings showing Chs2p localizing to the neck upon exit from mitosis ( ). In mutants (FM272), the induction of Sic1p failed to cause the transport of Chs2p to the neck, as only 2.33 ± 1.53% of cells showed neck signals after Gal addition compared with 0.67 ± 0.58% in the Raff culture (, bottom). Consistent with the role of Sec12p in activating the assembly of COPII components (), the FM272 cells exhibited mostly ER signals even in the presence of Sic1p induction (, top middle). In (FM268) cells, there were 12.00 ± 2.00% of cells with neck signals upon Gal induction as compared with 2.67 ± 0.58% without (, bottom), although some of the Chs2p-YFP signals at the neck appeared diffused (not depicted). This could be caused by the incomplete inactivation of sec2-59p during the early part of Gal induction, leading to the translocation of some of the vesicles to a region near the neck (). However, unlike the cells, the cells in the Gal culture exhibited punctate staining and a decrease in clear Chs2p-YFP ER signals (, top right), indicating that Chs2p-YFP had exited from the ER upon a decrease in mitotic kinase but that Chs2p-YFP was trapped in vesicles as a result of the mutation. Our data showing that Sic1p-induced Chs2p-YFP neck localization was abolished in a mutant supported the notion that the mitotic exit normally served to trigger the export of Chs2p from the ER in a Sec12p-dependent manner. The experiments with Sic1p-induced mitotic exit suggested that Chs2p was retained in the ER during mitosis and continues along the secretory pathway only upon mitotic exit. However, an alternative interpretation is possible. expression strongly peaks during mitosis (). Thus, cells arrested in metaphase or in telophase could be overexpressing , and so the accumulation of Chs2p in the ER may be an artifact of overproduction. For example, it was previously shown that Chs3p, when overexpressed, remained in the ER unless the chaperone Chs7p was coinduced (; ). To exclude this possibility, we first examined the Chs2p-YFP ER signals after a limited induction of Chs2p-YFP in a wild-type strain carrying an integrated construct (FM317). A short pulse of expression from the promoter would show a time-dependent decrease in the intensity of Chs2p-YFP signals in the ER if indeed there were continued export of Chs2p out of the ER in metaphase. FM317 cells were arrested in Noc and Gal induction performed for 1 h. shows that after 60 min of Gal induction, Chs2p-YFP signals could be seen. Subsequently, the induction was shut off by washing half of the culture into YPD with Noc and the other half into YPD without Noc. At 60 min after Gal shut-off and Noc release, 82.00 ± 8.18% of cells had ER signals, but, after 150 min, only 13.33 ± 6.11% of the cells exhibited ER signals (, bottom right). Furthermore, these cells that were released from Noc showed punctate spots (, bottom right), which were most likely Chs2p being transported away from the neck in vesicles at the end of cytokinesis (; ). However, in cells cultured continuously in the presence of Noc, 93.33 ± 3.06% of them exhibited characteristic ER signals even after 150 min of Gal shut-off and were devoid of the punctate spots (, top right). This observation supported the idea that Chs2p-YFP was retained in the ER in Noc-arrested cells. Our data showing the accumulation of Chs2p in the ER in the presence of high mitotic kinase and the export of Chs2p from the ER upon mitotic exit would predict that the induction of Chs2p-YFP expression in α-factor–arrested cells in which the mitotic kinase is at the lowest would lead to the unrestrained export of Chs2p. To test this, FM317 cells were arrested in α-factor, and Chs2p-YFP was induced with a short pulse of Gal addition followed by shut-off by washing the cells in YPD. It was observed that several of the α-factor–arrested cells exhibited punctate signals in the cytoplasm that did not colocalize with Sec63p-CFP (, left). These were presumably Chs2p-YFP that had been induced but that was not retained in the ER. These spots were unlike the characteristic ER localization in the control cells that were arrested in Noc (, right). The spots in the cytoplasm in G1 cells were confirmed to have originated from the ER using a strain (FM355; unpublished data). However, not all cells showed the cytoplasmic spots or Chs2p-YFP signals, indicating that perhaps the induction in G1 was not as uniform as in metaphase (, left). Nevertheless, collectively, our data showed that unlike in metaphase, the expression of Chs2p-YFP in G1 cells with low mitotic kinase led to unrestrained export from the ER. We next wanted to confirm that Chs2p-YFP was not in fact slowly exiting the ER in Noc-arrested cells. This is because a decrease in mitotic kinase activity is necessary for targeted exocytosis to the neck (), and, as such, if there were some Chs2p-YFP exiting from the ER in metaphase-arrested cells, the Chs2p-YFP could be delocalized and difficult to detect because of the absence of polarized transport. To eliminate this possibility, a (FM258) strain was arrested in Noc at 24°C, after which half of the culture was shifted to 37°C after washing out Noc. The other half was shifted to 37°C in the continued presence of Noc. If indeed Chs2p was continuously transported out of the ER during the metaphase arrest, the Chs2p-YFP signals from the ER would diminish, and the signals would be observed as punctate spots in the culture at 37°C in the continued presence of Noc as a result of the block at the stage. As can be seen in , in the control culture in which Noc was washed off at 37°C to allow for mitotic exit, a punctate appearance in the cells was evident at 60 min with a concomitant loss of ER signals (, bottom). By 90 min, most of the ER signals were no longer obvious, but punctate spots could be seen instead (, bottom). In the 37°C culture in which Noc was present throughout, ER signals were obvious (, top). This implied that there was no continuous export of Chs2p from the ER but that it was restrained in the ER during metaphase arrest. Intriguingly, there appeared to be areas of more intense YFP signals in the cells maintained at 37°C with Noc (, top; arrow). These spots of intensity were different from the punctate spots seen in the culture released from Noc (, bottom). They were unlikely to be localized in the Golgi compartments, as the patterns of localization did not coincide with Golgi markers such as Och1p-CFP or Sec7p-CFP (Fig. S1, top; available at ). However, these spots appeared to partially colocalize with Sec13p-CFP when the spinning disk confocal was used during image acquisition (Fig. S1, bottom right). This would suggest that Chs2p-YFP was accumulating, perhaps in potential ER exit sites where Sec13p had docked. The reason for the appearance of such spots in the background is unknown to us presently. Control experiments performed similarly with the wild-type strain (FM119) showed that cells maintained in the continued presence of Noc exhibited strong ER signals at 37°C throughout 90 min, whereas cells with Noc washed out had diminished ER signals with the appearance of neck signals at 30 min after Noc release. By 90 min, most of the ER signals had disappeared (Fig. S2). Collectively, these observations further confirmed the notion that Chs2p-YFP indeed remained in the ER during metaphase arrest. It had previously been documented that other cargoes of the secretory pathway such as α-factor and invertase are continuously transported via the secretory pathway during mitosis (; ). To examine whether the regulation of Chs2p exit from the ER at metaphase was indeed specific to Chs2p in our set-up, we examined Chs2p-YFP localization and invertase activity in Noc-arrested cells. Both wild-type (FM119) and (FM262) cells were arrested in Noc in YPD, after which the cells were shifted to 0.05% glucose medium in the presence of 2% sucrose and Noc to induce invertase at metaphase. (top left) shows Chs2p-YFP ER signals in the metaphase cells while invertase continued to be transported out of the ER, as observed from the periplasmic invertase activity (, bottom left; external). Transport of invertase to the periplasm was disrupted during metaphase only if was inactivated (, bottom right; external at 37°C). Our data indicated that the high mitotic kinase in metaphase specifically retained Chs2p in the ER but not other cargoes of the secretory pathway such as invertase. The timing of cytokinesis and septum formation in budding yeast, such that they occur late in mitosis, ensures that the mother and daughter cells are partitioned only after mitotic events are completed. For the completion of cytokinesis, cells from different organisms, including plant and higher eukaryotes, depend on membrane trafficking (; ; ; ). The evidence in budding yeast suggests that secretory vesicles are targeted to the mother–daughter neck late in mitosis (; ), perhaps for the delivery of membranes and proteins such as Chs2p (). The secretory vesicles are associated with the class V myosin, Myo2p, tropomyosin, Tpm1p, and Tpm2p, all of which are needed for transporting the vesicles along the actin cables to their destinations (; ). It is the orientation of actin cables that directs the secretory vesicles to their destinations (; ). Indeed, the actin cytoskeleton, which is reoriented toward the neck at the end of mitosis (), helps guide the post-Golgi secretory vesicles to the target membrane at the neck (; ). Furthermore, the presence of the exocyst complex on the target membrane facilitates the docking of the vesicles onto the destination (; ) during targeted transport. It is not clear what binds the timing of orientation of the actin cables toward the neck to the arrival of the exocyst complex at the neck at the end of mitosis such that secretory vesicles dock at the neck in time for cytokinesis (). However, the orientation of the actin cables toward the neck () and the localization of Sec3p, one of the components of the exocyst complex (), to the mother–daughter neck upon mitotic exit () point to the possibility that the status of the mitotic kinase activity contributes to the direction and timing of the arrival of the post-Golgi vesicles through its effects on both actin reorganization and Sec3p localization. In this study, we investigated the regulation of Chs2p that could account for its timely arrival at the mother–daughter neck during cytokinesis. The precision needed for its timely neck localization reflects the role of Chs2p not just for primary septum deposition (; ) but also for successful cytokinesis. It has been shown that Chs2p translocates to the neck before Myo1p ring constriction (), and a leads to abnormalities in Myo1p ring constriction (). The basis of this requirement was recently revealed when it was found that a failure to transport Chs2p to the neck leads to a loss of Myo1p ring integrity during cytokinesis (). We found that Chs2p-YFP, a cargo of the secretory pathway (; ; ), was regulated at the level of export from the ER ( ). This would imply that as cells progressed through metaphase to telophase and the mitotic kinase declined, Chs2p is triggered to exit from the ER via the COPII secretory pathway (). This, coupled with actin reorganization () and the arrival of Sec3p at the neck ∼5 min before cytokinesis (), leads to the localization of Chs2p to the neck 1 min after the exocyst complex (). Our observations, which place the export of Chs2p from the ER as being contingent upon mitotic exit ( and ), could explain the fact that Chs2p, which is transcribed in G2/M phase (; ; ) and is synthesized in the ER in metaphase (), is localized to the neck only late in mitosis (; ; ; ). The significance of the exact timing of Chs2p neck localization late in telophase is underscored by our finding that Sic1p overexpression in metaphase resulted in the premature Chs2p neck localization () and deposition of chitin between the mother and daughter cells (). With respect to the activity of Chs2p, had previously shown that Chs2p is isolated in a zymogen form that requires partial proteolysis for activation in vitro. However, found that Chs2p is in fact active in vitro before protease treatment, indicating that Chs2p may well be functional as an active form and that it is hyperactive upon proteolytic treatment. In vivo, it is not known whether Chs2p is activated before it reaches the mother–daughter neck, as no activators have been isolated to date. Our data showing chitin deposition between mother and daughter cells upon Sic1p induction () would mean that Chs2p was active at the neck in the presence of low mitotic kinase, albeit in metaphase. Whether its activation occurred along its passage to the neck or only upon reaching the neck is unknown. Nonetheless, our study further extends the previous proposal that Chs2p function is regulated posttranslationally by synthesis and degradation (). Indeed, an additional layer of control at Chs2p export from the ER, which is dependent on mitotic exit, is perhaps critical for ensuring that septum formation only occurs upon the completion of sister chromatid segregation and late mitotic events. An interesting point to note from this experiment is that complete septa were formed across the mother–daughter neck in metaphase cells expressing Sic1p (, 120 min of YP/Raff/Gal). As the closure of a primary septum depends on concomitant actomyosin ring contraction (; ), this would imply that actomyosin ring constriction had in fact occurred at metaphase. Indeed, we had observed Myo1p-CFP ring constriction in the Noc-arrested cells in which Sic1p was overexpressed (unpublished data). This could either be caused by the direct effect of the decrease in mitotic kinase or by the reduction in mitotic kinase that leads to the localization of MEN components to the neck, where they promote cytokinesis (for review see ). The tight coordination between mitotic exit, chitin deposition, and cytokinesis together with the interdependence of chitin deposition and actomyosin ring constriction underpin the proper partitioning of mother and daughter cells in a timely manner. In mammalian cells, intracellular export from the ER is halted during mitosis, as evident from studies on the number of ER exit sites (; ). Cdc2 kinase has been implicated in the disassembly of ER exit sites in mammalian cells, thereby inhibiting the secretory pathway during mitosis (). In the budding yeast, it remains to be seen how the export of Chs2p from the ER is affected by a decrease in mitotic kinase given that other cargoes such as invertase and α-factor are insensitive to mitotic kinase levels (; ; ). Interestingly, Chs2p has been found to be a potential substrate of the mitotic kinase (); however, the implication of that finding with respect to ER export has yet to be determined. Wild-type haploid W303a strain was used in this study. Cells were backcrossed to the W303a background at least three times. Cells were routinely grown in yeast extract peptone or selective medium supplemented with 2% dextrose at 24°C. For experiments requiring the Gal induction of Sic1p, cells were grown in yeast extract peptone supplemented with 2% Raff followed by the addition of Gal to a final concentration of 2%, unless otherwise stated. For experiments requiring synchronized cultures, exponential phase cells were diluted to 10 cells/ml in growth medium at 24°C and arrested using either α-factor at 0.4 μg/ml or HU at 0.2 M. After the cells were arrested, they were washed by filtration and resuspended in media at the required conditions as described in the various sections. For a typical Noc arrest, cells were first treated with 7.5 μg/ml Noc for 2 h followed by the further addition of 7.5 μg/ml for another 3 h. The drug was washed off by filtration or centrifugation of the cells. Cells were then sampled at intervals at the indicated times. Each experiment has been performed three times, and 100 cells were counted for each of the experiments unless otherwise stated. Graphs shown were plots of the mean values of three experiments with SD. A combination of standard molecular biology and molecular genetic techniques such as PCR-based tagging of endogenous genes and tetrad dissection were used to construct plasmids and strains with various genotypes (). The plasmids for the CFP and YFP cassettes were obtained from EUROSCARF. A list of primers used for the tagging of endogenous genes with CFP or YFP cassettes is shown in . lists the primers used for checking the correct insertion of the CFP or YFP-tagged PCR products. Strains US1363 and US107 were provided by U. Surana (Institute of Molecular and Cell Biology, Singapore); strains RSY263, RSY279, and RSY319 were provided by R. Schekman (University of California, Berkeley, Berkeley, CA); and strain NY26 was provided by P. Novick (Yale University School of Medicine, New Haven, CT). In brief, to make the construct, was synthesized by PCR from the ATG of to the terminator sequence from the pKT series of plasmids () using genomic DNA from FM182 as the template. The PCR fragment was cut using HindIII and BamHI, which were incorporated into the primers and ligated to the EcoRI–BamHI-cut promoter and YIplac204 cut open with EcoRI and HindIII. Western blot analyses were performed as previously described (). Anti-Cdc28p antibodies (Santa Cruz Biotechnology, Inc.) and anti-myc antibodies (Santa Cruz Biotechnology, Inc.) were used at a 1:1,000 dilution, and anti-Clb2 antibodies (Santa Cruz Biotechnology, Inc.) were used at a 1:5,000 dilution. An enhanced chemiluminescence kit (Pierce Chemical Co.) was used according to the manufacturer's recommendations. The invertase assay was performed as previously described () with several modifications. In brief, an overnight culture in YPD was arrested in Noc at 24°C in YPD, and invertase was induced by washing the cells into YP/2% sucrose/0.05% glucose for 2 h. At the appropriate time points, cells were collected and washed once in stop mix (10 mM NaN in 25 mM Tris-Cl, pH 7.4) before resuspending in 5.5 ml of fresh stop mix. 0.5 ml of the cell suspension was rested on ice for use in assaying the external invertase activity. To prepare the lysate for the total invertase activity, the rest of the cell suspension was spun down and resuspended in 2 ml spheroplast medium (1.4 M sorbitol, 0.1 M KPi, pH 7.5, and 5 mM NaN), and 80 U lyticase (Sigma-Aldrich) was added (3.34 μl from a 10-mg/ml stock). Spheroplasting was performed at 37°C for 45 min. 3 ml of lysis buffer (1.67% Triton X-100) was then added to lyse the cells. For sucrose conversion, 40 μl of intact cells or total lysate was combined with 35 μl sodium acetate (0.1 M, pH 5.0) and 25 μl of 0.5 M sucrose (Sigma-Aldrich), and the mixture was incubated at 37°C for 20 min. 150 μl of 0.2 M KHPO was then added, and the reaction was put on ice before boiling for 3 min. 750 μl of sterile water was added to make up to 1 ml, and, of this, 250 μl was used for the glucose assay. Essentially, 0.5 ml glucose assay reagent (Glucose Assay Kit; Sigma-Aldrich) together with o-Dionisidine (Sigma-Aldrich) was added, and the mixture was incubated at 37°C for 30 min. The reaction was stopped by adding 0.5 ml of 12 N HSO. The absorbance at 540 nm was then measured. To calculate the amount of glucose produced from sucrose, the following equation was used: milligrams of glucose = (ΔA of test) × (milligrams of glucose in standard)/(ΔA of standard). To calculate invertase activity, the following equation was used: activity = (milligrams of glucose/180) × 10 × 4 × (500/40)/20. Cells carrying CFP and YFP fusions were fixed briefly in KPF () and washed in PBS or observed directly without fixation after washing in PBS. Samples were observed using a microscope (IX81; Olympus), 60× NA 1.4 oil lens, and 1.5× Opitvar. Filter sets for the fluorescence proteins were purchased from Omega and Semrock, and images were captured using a CCD camera (CoolSnap HQ; Photometrics). Image acquisition was controlled by MetaMorph software (Molecular Devices). Typically, the exposure time for YFP was ∼0.8–1 s and for CFP was ∼0.6–0.8 s. In most experiments, unless otherwise stated, 100 cells were counted for each time point for three independent experiments, and the mean values were plotted with SD. For time-lapse microscopy, cells were mounted in complete minimal media, and images were captured at room temperature at 2-min intervals. Image J (National Institutes of Health) and Adobe Photoshop were used for production of the figures. Cells were pelleted and fixed overnight at 4°C in primary fixative (2% PFA, 2% glutaraldehyde, 40 mM phosphate buffer, and 0.5 mM magnesium chloride, pH 6.5). Postfixation of the cells was performed in 2% osmium tetroxide (2% osmium, 40 mM phosphate buffer, and 0.1% ferro-cyanide) for 1 h at room temperature. Subsequently, the cells were washed twice with 40 mM phosphate buffer and dehydrated using 50% ethanol for 20 min at room temperature followed by 75% ethanol overnight at 4°C and 95% ethanol for 20 min at room temperature. Another three changes of absolute ethanol for 30 min each at room temperature were performed. The cells were infiltrated using a series of ethanol–LVER (Low Viscosity Epoxy Resin) mixtures (7:3 followed by 5:5 and 3:7 ratios of ethanol/ LVER). The cells were left in the first and second embedding mixtures for 1 h at room temperature and then overnight in a 3:7 ethanol/ LVER mixture at room temperature. The cells were passed through three changes of pure embedding media for a period of 90 min at each change. After the third change, cells were resuspended in ∼200 μl LVER and transferred into BEEM capsules. The BEEM capsules were topped up with embedding media and spun at 4,000 rpm for 10 min at 40°C. Polymerization was allowed to occur overnight at 60°C. Ultrathin sections were obtained using an ultramicrotome (Ultracut; Leica), and the sections were stained with uranyl acetate for 20 min followed by lead citrate for 30 min. The sections were viewed using an electron microscope (JEM-1220; JEOL) at 100 kV. Fig. S1 shows that cells in prolonged Noc arrest at 37°C exhibit intense Chs2p-YFP spots that do not colocalize with Sec7p-CFP or Och1p-CFP. Fig. S2 shows that wild-type cells arrested in Noc at 37°C exhibit Chs2p-YFP ER signals. Online supplemental material is available at .

Self-incompatibility (SI) is one of the systems that prevent self-fertilization in flowering plants. SI is controlled by a multiallelic locus; specific pollen rejection results from the interaction of pollen and pistil determinants that have matching alleles (). In , the pistil S proteins () act as ligands, triggering increases in the cytosolic-free calcium concentration ([Ca]) in incompatible pollen (, ). The Ca-mediated signaling network results in the rapid inhibition of incompatible pollen tube growth. Within a few minutes of SI signals, reorganization and massive, sustained depolymerization of the pollen filamentous actin (F-actin) is induced (; ). Although the extent of F-actin depolymerization during SI is clearly sufficient to inhibit pollen tube growth, it appears to be a gross excess over the amount required to achieve this. This suggested possible additional functions for the alteration to actin dynamics. Unwanted cells are usually removed by programmed cell death (PCD). Many examples of PCD in plant development () and in responses to external stimuli (; ; ; ) have been documented. Features of PCD in animal cells include cytochrome leakage from the mitochondria, DNA fragmentation, and caspase activation. In animal cells, apoptosis is mediated by a caspase cascade. Activated caspases cleave numerous substrates, including endogenous nuclease inhibitors, resulting in the fragmentation of nuclear DNA. Although PCD should theoretically involve a caspase-3–like activity, no caspase homologues have been found in plants (). Despite this, there is good evidence for caspase-like activities in plant cells (; ). We recently reported that SI triggers a PCD cascade in incompatible pollen, which involves a caspase-3–like activity (). This provides a precise mechanism for the specific destruction of incompatible pollen. As the actin cytoskeleton is a major target and effector of signaling cascades in both animal and plant cells (; ), we explored a possible role for actin depolymerization in signaling to PCD. Recent evidence suggests that either stabilization or depolymerization of the actin cytoskeleton is adequate to induce PCD in yeast and some animal cells, depending on the cell type (; ; ; ; ; ; for review see ). It is postulated that the alteration of actin filament dynamics initiates or modulates the apoptotic signaling cascade (; ; ; ; ), thereby committing cells to die (for review see ). Because PCD is triggered by SI in incompatible pollen (), we hypothesized that early SI-induced actin depolymerization might play a role in acting as an upstream component in PCD activation. We have investigated this possibility using specific drugs to alter actin polymerization status in pollen tubes. Pollen tubes have a characteristic F-actin organization (), and SI induces the rapid depolymerization of actin filaments and bundles (). To test the hypothesis that SI-stimulated PCD is triggered by actin depolymerization, we used jasplakinolide (Jasp), which stabilizes actin filaments and stimulates polymerization (; ) and has been shown to inhibit pollen tube growth (). We found that 0.5 μM Jasp disrupted pollen tube actin organization, which induced actin filament bundling and aggregation (), and inhibited tip growth (unpublished data). The effect of Jasp on actin organization was essentially the opposite of that induced by SI. We examined whether transient treatments with Jasp might “rescue” SI-induced pollen from PCD by counteracting the actin depolymerization. One of the hallmark features of PCD, DNA fragmentation, is triggered by SI and involves a caspase-3–like activity (). This was used as a marker for PCD and was assessed using TUNEL (). The induction of SI, which is caused by exposing incompatible pollen to S proteins for 30 min, produced high levels of DNA fragmentation of 55.2%, which was only slightly lower than levels of 65.4% induced by SI for 8 h and significantly different from control levels of 16.8% (P < 0.001; ). These data indicated that the interaction of incompatible pollen with S proteins for 30 min was sufficient to trigger PCD. As we had previously demonstrated that 10 min of SI stimulated a 65% reduction in F-actin levels (), we allowed SI to progress for 10 min to ensure that a significant level of actin depolymerization had occurred and then added 0.5 μM Jasp for 20 min (see Pollen treatments). This consecutive treatment significantly reduced DNA fragmentation () by 32% (P = 0.04). Simultaneous treatment with SI and Jasp for 30 min also resulted in significantly reduced levels of DNA fragmentation (33.8%; P < 0.001; ). This demonstrates that Jasp can alleviate SI-induced PCD. These data indicate that the actin depolymerization triggered by SI plays a functional role in the initiation phase of PCD in pollen tubes. To further investigate whether actin polymerization status is involved in the initiation of PCD in pollen, we attempted to mimic the actin depolymerization induced by SI without the involvement of other SI initiation signals, such as changes to [Ca]. To clarify whether actin depolymerization itself could induce PCD, latrunculin B (LatB; ; ) was used to depolymerize the actin cytoskeleton in growing pollen tubes. LatB caused the rapid disruption of F-actin organization () and the inhibition of pollen tube growth (unpublished data). To confirm actin depolymerization, we measured F-actin levels (; ) after 10 min of LatB treatments. LatB caused the depolymerization of pollen tube actin filaments in a concentration-dependent manner (). Treatment with 0.1, 1, and 10 μM LatB caused significant reductions (P < 0.001) in F-actin levels when compared with controls. Treatment with 1 μM LatB, which induced 61.5% F-actin depolymerization, gave a similar reduction to that stimulated by SI (69%; ). To test whether LatB-induced actin depolymerization could trigger PCD, pollen tubes were treated with LatB, and levels of DNA fragmentation were assessed after 8 h. LatB stimulated DNA fragmentation in a concentration-dependent manner (). The incidence of DNA fragmentation at all concentrations of LatB was significantly different from control pollen tubes (P < 0.001) and increased from 6.9% in controls to 30.7% in the presence of 0.01 μM LatB and to 75.8% with 10 μM LatB (). Assuming that LatB does not have side effects, this indicated that actin depolymerization triggered DNA fragmentation independently of other SI signals. We also examined whether actin filament stabilization or net assembly of filaments from the profilin–actin pool might also induce PCD in pollen. Jasp stimulated DNA fragmentation in a concentration-dependent manner (). The incidence of DNA fragmentation in Jasp-treated pollen tubes increased from 10.4% in control samples to 58% in the presence of 1 μM Jasp (). As Jasp binds the phalloidin-binding site of actin, we could not quantify actin polymer levels for these treatments. Given the assumptions regarding the lack of Jasp and LatB side effects, the results indicate that changes in actin filament dynamics can induce DNA fragmentation. A hallmark feature of apoptosis and PCD is the involvement of caspases, which are responsible for initiating and executing cell death. We showed previously that SI-induced PCD is mediated by a caspase-3–like/DEVDase activity (). Therefore, we tested whether the DNA fragmentation stimulated by LatB and Jasp involved such an activity. We used the tetrapeptide Ac-DEVD-CHO (DEVD), which is a caspase-3 inhibitor (; ; ; ). Pollen was pretreated with either DEVD or the caspase 1 inhibitor Ac-YVAD-CHO (YVAD), which acts as a negative control, for 1 h before the addition of LatB or Jasp for 8 h. Pollen tubes pretreated with DEVD had significant (65.1–71.6%) reductions in the levels of DNA fragmentation compared with controls without pretreatment (P < 0.001; ). Pretreatment with YVAD had no significant effect on the levels of DNA fragmentation induced by LatB or Jasp (P = 0.2; ). Because DEVD can prevent LatB- or Jasp-induced DNA fragmentation, this implies that both actin depolymerization and stabilization or polymerization can stimulate the activation of a caspase-like enzyme upstream of DNA fragmentation. These data indicate that changes in actin filament dynamics are sufficient to induce a caspase-like (DEVDase) activity that results in PCD in pollen. To examine whether transient changes in actin filament levels or dynamics could serve as a signal involved in the initiation of PCD or whether sustained alterations are required, “washout” experiments were conducted using LatB treatments, as this most closely mimicked the effect of SI. Pollen tubes were treated with 0.1 and 1 μM LatB for 10 or 60 min, after which the drug was washed out and the incidence of DNA fragmentation was assessed. Actin filament levels were determined at the end of the LatB incubation period to demonstrate that depolymerization had taken place and also after washouts to establish whether repolymerization had occurred. As expected, F-actin levels were reduced in a concentration- and time-dependent manner (). The level of actin polymer after treatment with 0.1 μM LatB was reduced by 23.0% (P = 0.03) at 10 min and was further reduced by 50.2% (P < 0.001) by 60 min. The higher concentration of 1 μM LatB reduced actin levels by 53.1% (P < 0.001) at 10 min and by 69.2% (P < 0.001) at 60 min. After washouts, with the exception of the 1-μM 60-min treatment, F-actin levels returned to similar levels as found in untreated pollen, demonstrating that F-actin was only depolymerized transiently (). Pollen tubes treated with 0.1 μM LatB also resumed normal growth (unpublished data). DNA fragmentation correlated with the duration and extent of actin depolymerization. After 5- and 10-min treatments with 0.1 μM LatB, the effects were similar, and only small increases in DNA fragmentation were detected (P = 0.3; NS; ). With longer treatments of 0.1 μM LatB, DNA fragmentation increased to 57.1% at 30 min (P = 0.02) and to 64.0% at 60 min (P = 0.001). The 1-μM LatB treatments induced higher levels of DNA fragmentation than those found in the controls for all four time points (P < 0.001). However, there was no significant difference between the four treatments (P = 0.6; ), indicating that a threshold level of DNA fragmentation had been triggered. Taking these F-actin quantification and DNA fragmentation data into account, reducing F-actin levels to ∼50% for either 10 or 60 min gave a high incidence of DNA fragmentation even though F-actin levels returned to normal after washing. This indicates that during a brief 10-min period of actin depolymerization to this level, an irreversible “decision-making” step is made, pushing pollen into PCD. Because a DEVDase/caspase-like activity is involved, this implicates actin depolymerization in PCD initiation. Treatments that did not reduce F-actin to this level (e.g., 0.1 μM LatB for 10 min) had a low incidence of DNA fragmentation (24.3%), which was only slightly higher than the control level (P = 0.02). Therefore, we can say with some confidence that here, the reduced incidence of DNA fragmentation was the result of a transient actin depolymerization that was insufficient to cause PCD. Thus, the threshold for PCD induction is somewhere between a 23 and 46.9% reduction in F-actin levels for as little as 10 min. Our data indicate that <50% actin depolymerization for 10 min is sufficient to induce PCD, achieving levels of DNA fragmentation very similar to that induced by 8 h of LatB (P = 0.4; NS). Notably, this threshold level of actin depolymerization required to initiate PCD using LatB is somewhat lower than the 58 and 69% reduction in F-actin levels resulting from 5- and 10-min SI inductions, respectively (). As LatB treatments appeared to mimic the SI–PCD response, we examined whether LatB-induced DNA fragmentation relied on a caspase-like/DEVDase activity similar to that induced by SI (). DNA fragmentation was normally induced by 1-μM LatB treatments, whereas DEVD pretreatment prevented this (), and levels of DNA fragmentation were not significantly different from untreated samples (P = 0.13). These data provide strong evidence that even quite transient LatB-induced actin depolymerization is sufficient to induce a caspase-3–like/DEVDase activity upstream of DNA fragmentation. Because DEVD had no significant effect on the amount of actin depolymerization caused by LatB (), this established that DEVD does not interfere with the action of LatB. We have also used the caspase-3 substrate Ac-DEVD-AMC to establish this caspase-like activity more directly. Pollen tubes treated with 1 μM LatB for 6 h exhibited an increase in DEVDase activity of 54 ± 5% ( = 4) compared with control samples containing germination medium (GM) only. This level of caspase activity was similar to that induced by either SI or 0.5-μM Jasp treatments (unpublished data). Thus, LatB-stimulated pollen extracts exhibited markedly increased levels of caspase-3–like activity, indicating cleavage of the substrate by a DEVDase activity. The activation of caspases commits the cell to die; therefore, our data implicate alterations in actin depolymerization as playing a functional role in stimulating or regulating the PCD cascade in pollen. Because both LatB and Jasp inhibit tip growth (; ; ; ), we wished to establish that PCD induction was not indirectly caused by the cessation of tip growth. We used caffeine, which perturbs tip growth () but does not affect the actin cytoskeleton of pollen tubes (; ). Pollen tubes treated with 3 mM caffeine for 10 or 60 min did not significantly alter F-actin levels compared with controls (112 ± 12 and 112 ± 2%, respectively; = 3; P = 0.06). Caffeine-treated pollen tubes had low DNA fragmentation levels that were not significantly different from controls (15 ± 3% at 10 min, P = 0.2; and 23 ± 4% at 60 min, P = 0.06; = 3). This demonstrates that the LatB/Jasp-induced DNA fragmentation is not merely a consequence of the inhibition of growth but is caused by changes in actin polymer levels or assembly dynamics. We also investigated whether counteracting the effect of LatB-induced depolymerization with Jasp might prevent progression into PCD. Pollen tubes were treated in a similar manner to the SI experiment. They were first treated with 0.1 μM LatB for 10 min to ensure that actin depolymerization had occurred; 0.5 μM Jasp was then added for 20 min, after which the drugs were washed out. Thus, the total length of time that the pollen was exposed to LatB for this consecutive treatment was 30 min. This treatment resulted in a 45% reduction (P = 0.002) in the level of DNA fragmentation compared with LatB alone for 30 min (). We also performed treatments whereby LatB and Jasp were added simultaneously; this had a similar significant effect with a 51.7% (P = 0.007) reduction in the incidence of PCD to 19.8%, which was not significantly different (P = 0.2) from control levels of 11.3% (). These data show that Jasp can counteract the actin depolymerization induced by LatB and, thereby, rescues pollen from entry into PCD. The actin cytoskeleton has been identified as a major target and effector of signaling cascades in both animal and plant cells (; ). SI may be viewed as the triggering of a signaling cascade that takes the cell into PCD. In this study, we provide the first account of a causal link between actin polymer levels or dynamics and the initiation of PCD in a plant cell. Our study provides a significant advance in our understanding of the mechanisms involved in early SI and of the initiation of PCD in plant cells because very little is known about the early events involved in PCD in plants. Our data provide evidence that relatively transient but substantial F-actin depolymerization can trigger PCD, which is mediated by a caspase-3–like activity. We previously demonstrated that the SI response in pollen results in a 69% reduction in F-actin levels within 10 min () and, independently, that SI also induces PCD (). In this study, we have demonstrated that a transient reduction in F-actin levels of <50%, which is independent of SI induction, is sufficient to induce PCD. This provides evidence for a link between the signaling cascades involving actin dynamics and PCD, showing that actin depolymerization is sufficient to induce a caspase-like/DEVDase activity resulting in PCD. Furthermore, together with previous data (), they demonstrate that during an incompatible SI response, the degree of actin depolymerization is more than adequate to act as an intermediary signal to PCD. Because Jasp can alleviate SI-induced DNA fragmentation, presumably by interfering with or counteracting the actin depolymerization induced by SI, this provides further, more direct evidence for the involvement of actin depolymerization in SI-induced PCD. Thus, rapid depolymerization of the actin cytoskeleton induced by SI not only inhibits pollen tip growth but also acts upstream of a signaling cascade that is involved in initiating PCD. Our data support the hypothesis that the F-actin depolymerization and PCD observed during SI function together in a signaling network that prevents incompatible pollen from affecting fertilization. Although LatB or cytochalasin D (CD) has previously been reported to implicate actin depolymerization in PCD during embryogenesis in plants (), the drug treatments extended over a long period of time (6 d), and disruption of the actin cytoskeleton for this length of time will almost certainly result in defects in embryogenesis; this in itself could result in PCD. Thus, although the data suggested a possible involvement of the actin cytoskeleton in PCD, they provided no clear evidence for a functional role for actin depolymerization in PCD. Our findings establish a fundamentally important role for the actin cytoskeleton as a sensor of cellular stress that is common to many eukaryotic cells, as various links between changes in actin dynamics and PCD induction have previously been reported for animals and yeast (; ; ; ; ; ). Several studies have shown that alterations to the actin cytoskeleton (either stabilization or depolymerization, depending on the cell type involved) can play a functional role as an effector in the initiation of PCD. For example, CD resulted in DNA fragmentation and caspase-3 activity in Jurkat T cells and hippocampal neurons (), and Jasp enhanced and accelerated apoptosis induced by cytokine withdrawal in the IL-2–dependent T cell line CTLL-20 and induced a caspase-3 activity (). Because these data demonstrate that presumed changes to actin polymer levels or dynamics can stimulate caspase-3 activity, which is an effector caspase triggered early in PCD, they provide evidence that actin depolymerization or stabilization is sufficient to act as an intermediary signal early in the PCD signaling cascade. Other studies in yeast and some mammalian cells have also shown that decreasing actin filament turnover (by using Jasp to stabilize actin) induces apoptosis (; ; ; ). Thus, it has been proposed that altering actin dynamics or the rates of assembly and disassembly rather than the level of polymeric actin in cells is sufficient to signal the initiation of apoptosis (; ; ; ; ; ; ; ; ; for review see ). Interestingly, although actin stabilization stimulates PCD in yeast, actin depolymerization has the reverse effect, leading to increased viability (), which is not the case in pollen. This suggests that although there are clear parallels, the specific mechanisms are likely to differ between plants, animals, and yeast. This is borne out by several studies with mammalian cells, in which actin depolymerization can induce or promote apoptosis (; ; ; ; ), whereas actin stabilization can inhibit or delay apoptosis (; ). Indeed, it has been shown in Jurkat T cells that either increased actin depolymerization or polymerization using CD or Jasp can enhance apoptosis (; ). This appears to be the situation in pollen, in which either LatB or Jasp can induce PCD. Furthermore, it has recently been shown that Jasp can alleviate apoptosis in ischemic kidney cells; ischemia normally correlates with a substantial depolymerization of the actin cytoskeleton (). Similarly, phalloidin treatment, which stabilizes F-actin, prevented cisplatin-mediated actin depolymerization and apoptosis in porcine kidney proximal tubule cells (). This is exactly what we have demonstrated here; Jasp effectively reverses the effects of S-protein treatment, presumably by stabilizing actin filaments against depolymerization or counteracting the calcium-induced disassembly. Therefore, it would appear that a precise level of actin polymer or exact flux of subunits through the polymer pool is necessary for normal growth and cell viability of pollen (; ); when this is substantially perturbed, PCD is initiated. It has been proposed that it is the alteration of actin dynamics (i.e., the rate of actin polymerization and depolymerization) that modulates the transduction of the apoptotic signal (). The alterations to actin providing the sensory mechanism could involve either changes in polymer levels, changes in the flux of actin through the filament pool, or both. However, although in some mammalian cells, alterations to F-actin status alone (either polymerization or depolymerization, depending on the cell type) are sufficient to induce apoptosis, in other cell types, other apoptotic stimuli are required in addition to drug treatments affecting actin dynamics (; ; ). Thus, there are likely to be quite important and subtle differences in the mechanisms operating to initiate apoptosis/PCD in different cell types. One notable and important difference between plant pollen and yeast or mammalian cells is the high ratio of globular actin (G-actin) to F-actin in actively growing cells. Measurements from maize and pollen indicate that just 5–10% of the total actin protein is present in the filamentous form (; ). In comparison, budding yeast cells are considered to have the majority of the total actin pool in filamentous form (). These observations further suggest that certain differences in apoptosis initiation and/or responses to cytoskeletal drugs may relate to the endogenous balance between monomer and polymer in different eukaryotic cells. Importantly, our study, which used washouts to achieve short-term actin depolymerization or stabilization, is, to our knowledge, the first to demonstrate that transient (10–60 min) alterations to actin dynamics can trigger PCD. Most of the aforementioned studies have used rather long-term treatments with Jasp or LatB for between 8 and 48 h to induce apoptosis (; ; ). To our knowledge, this is the only study to quantify the changes in F-actin levels required to trigger PCD. Only two other studies attempted to quantify the actin polymerization status (; ). One of these studies reports that measurements after CD treatment of CTLL-20 and Jurkat T cells indicated an increase in the amount of monomeric or G-actin and a decrease in the amount of F-actin, suggesting the promotion of actin filament disruption and depolymerization (). However, no data are shown. The other study () measured G- and F-actin levels after camptothecin-induced apoptosis in HL-60 cells and found that G-actin was decreased in TUNEL-negative cells at 2 h, indicating that polymerization was increased early in apoptosis. In TUNEL-positive cells (in which PCD had already occurred), G-actin levels were increased, indicating depolymerization. These data are consistent with camptothecin- and Jasp-induced actin polymerization occurring before PCD. Our data go further than this because we have estimated the levels of actin depolymerization required to induce PCD in pollen. In this study, we show that 50% depolymerization of the actin cytoskeleton for just 10 min is sufficient to trigger substantial numbers of cells to undergo PCD. The alterations to actin could involve either changes in polymer levels, changes in the flux of actin through the filament pool, or both. Because virtually no biochemical analysis of actin dynamics leading to PCD has been performed, it would be of considerable interest to establish which mechanisms are involved and whether they vary depending on cell type. Similarly, several different actin-binding proteins have been suggested to play a role during the initiation or execution phases of PCD, but cause and effect relationships are difficult to decipher. We recently identified pollen gelsolin, PrABP80, which is a calcium-stimulated actin filament severing and depolymerizing protein that could potentially be involved in SI-mediated actin depolymerization (). This is of interest, as gelsolin is implicated in modulating apoptosis in animal cells (for review see ). How exactly actin polymerization status can affect apoptosis induction is not yet established in any organism. However, there is evidence that changes to actin dynamics interact with apoptotic signaling cascades. It has been suggested that disruption of the cytoskeleton might promote the release of caspases, enabling their activation or, alternatively, disrupting mitochondria and causing the release of cytochrome , caspases, or caspase activators (). In mammalian cells, there is evidence that either actin depolymerization () or stabilization (; ; ) can induce caspase-3 activation; actin depolymerization using CD has been shown to regulate nitric oxide–stimulated apoptosis by modulating PI3-kinase, PKC, and MAPK signaling (). In yeast, it has been proposed that the actin cytoskeleton could act as a regulator for reactive oxygen species (ROS) release from mitochondria, as Jasp can stimulate changes in the levels of ROS (). We have shown here that actin depolymerization or polymerization is sufficient to induce a caspase-3–like activity in pollen. Whether ROS is involved in SI is not yet known, but it is known that the rapid disruption of mitochondria is triggered by SI (). Furthermore, we have previously shown that a MAPK, p56, is activated by SI (). These provide possible targets for signaling links between the actin cytoskeleton and PCD in pollen. Thus, our data clearly demonstrate that actin acts as a sensor for signals and that its dynamics play a key role in signaling to initiate PCD in plant cells. Pollen of , the field poppy, was germinated and grown in vitro in liquid GM (0.01% HBO, 0.01% KNO, 0.01% Mg(NO)-6HO, 0.036% CaCl-2HO, and 13.5% sucrose) as described previously () at 25°C. Pollen was grown for 1 h before any treatments were applied. For SI treatments, recombinant proteins were produced by cloning the nucleotide sequences specifying the mature peptide of the S, S, and S alleles of the gene (pPRS100, pPRS300, and pPRS800) into the expression vector pMS119 as described previously (). Expression and purification of the proteins were performed as described previously (). SI was induced by adding recombinant S proteins (final concentration of 5 μg/ml) to pollen growing in vitro as described previously (). Fixed pollen tubes were labeled with a Deadend Fluorometric TUNEL kit (Promega) according to the manufacturer's instructions. Pollen tubes were scored for DNA fragmentation (50 tubes per treatment; = 3) using a fluorescence microscope (T300; Nikon) and a 60× plan-Apo 1.4 NA oil objective (Nikon). Capture and analysis of images was performed at 20°C using a camera (SenSys KAF1400-G2; Photometrics) and an image analysis system (Quips PathVysion; Applied Imaging). Composite images were prepared using Adobe Photoshop 8.0. We also used the fluorogenic caspase-3 substrate Ac-DEVD-AMC (Calbiochem) to establish this caspase-like activity more directly. Pollen tubes were treated with 1 μM LatB and 0.5 μM Jasp or GM for 6 h, and protein extracts were made from them by grinding in extraction buffer (50 mM sodium acetate and 10 mM -cysteine, pH 6.0). 10 μg of total protein was incubated with 50 μM Ac-DEVD-AMC at 27°C, and fluorescence was monitored at 460 nm using a time-resolved fluorescence plate reader (FLUOstar OPTIMA; BMG LABTECH). Fluorescence at 480 nm indicated cleavage of the substrate by a DEVDase activity. Relative fluorescence units were expressed as the percent increase relative to the control after 4 h. Pollen was grown on GM medium solidified with 1.2% wt/vol agarose, treated with actin inhibitors at various concentrations and different times as described in the figure legends and text, and fixed with 400 μM 3-maleimidobenzoic acid -hydroxysuccinimide ester (MBS; Pierce Chemical Co.) for 6 min followed by 2% PFA for 40 min. Pollen tubes were washed three times in actin-stabilizing buffer (100 mM Pipes, pH 6.8, 1 mM MgCl, 1 mM CaCl, and 75 mM KCl) and permeabilized with TBS-Tween-DTT (50 mM Tris, pH 7.4, 200 mM NaCl, 0.05% Tween 20, and 5 mM DTT) for 15 min. For Jasp-treated pollen, pollen tubes were grown, treated with 0.5 μm Jasp for 10 min, fixed with PFA and MBS, and prepared for immunolocalization. Samples were washed three times in actin-stabilizing buffer and once in MES buffer (5 mM MES, pH 5.0, 5 mM EGTA, and 0.4 M mannitol). Cell walls were digested with 0.1% cellulase and 0.1% macerozyme in MES buffer containing 0.1 mM PMSF for 10 min. Pollen tubes were washed once in MES, washed twice in TBS, permeabilized in 0.1% Triton X-100 in TBS for 10 min, and washed in TBS + 1% BSA. Samples were then incubated with a 1:250 dilution of anti-actin antibody () overnight at 4°C. Unbound primary antibody was washed out, and the pollen tubes were incubated for 2 h at RT in anti–rabbit FITC antibody (1:200 dilution). All images were collected at 20°C with a 60× plan-Apo 1.4 NA oil objective (Nikon) using a Radiance 2000 laser-scanning system with the 488-nm excitation line of a 50-mW Ar laser. Images are optical projections of z series (0.5-μm sections). Images were exported, and composite images were prepared using Adobe Photoshop 8.0. Pollen F-actin levels were determined using the modified phalloidin-binding assay described previously (). In brief, LatB-treated pollen tubes were fixed (1.55 M sucrose, 0.1% NP-40, and 600 μM MBS) for 1 h and washed in GM + 0.05% NP-40. Fixed pollen tubes were gradually exchanged into TBS + sucrose (TBSS; 50 mM Tris, pH 7.4, 200 mM NaCl, and 400 mM sucrose) plus 0.05% NP-40 and 2 mM DTT. The TBSS was aspirated from pollen samples, and 50 μl of 2 μM AlexaFluor 488 phalloidin (Invitrogen) in TBSS + 0.05% NP-40 was added to label F-actin at 4°C overnight. After washing, bound phalloidin was extracted in 1 ml methanol overnight at 4°C. The amount of F-actin in the samples was determined by fluorimetry with excitation at 492 nm and emission at 514 nm using a spectrofluorometer (QM-2000-SE; Photon Technology International). Fluorescence values were calculated per 100,000 pollen grains/tube and expressed as the percentage of the relevant control ± SEM. F-actin levels from control samples without LatB treatment were normalized to 100%.

Cells can reach specific target tissues through the general circulation by different mechanisms. Homing has been studied extensively both in vitro and in vivo with different cell types, such as leukocytes or hematopoietic stem cells (; ); it is believed to rely on adhesion molecules and cytokines receptors by a multistep cascade, consisting of a rolling process followed by firm adhesion and transmigration into the surrounding tissue (; ). The repertoire and the level of cytokine expression by the target tissue, as well as expression of the relative receptors on endothelium, influence the efficiency of homing. For example, stromal-derived factor (SDF) 1 favors the arrest of progenitors on vascular endothelium, whereas interleukin (IL) 8 promotes stem cell mobilization from the marrow (; ). Although these mechanisms have been elucidated to a large extent for leukocytes and hematopoietic stem cells, far less is known for other types of stem cells. Mesoangioblasts were recently characterized as a population of vessel-associated stem cells that differentiate into several mesoderm cell types, including skeletal muscle (). They have been shown to restore to a significant extent muscle structure and function in a mouse model of muscular dystrophy (). One main reason for the partial effect of mesoangioblasts in this model is likely to be ascribed to the limited ability of these cells to reach and colonize the muscle, depending in turn on incomplete adhesion and extravasation. Mesoangioblast extravasation must be directed by selective mesoangioblast–endothelial cell recognition. Microarray analysis revealed that mesoangioblasts express E-selectin, β7 integrin, AlCAM, several cytokines receptors, and CD44 but lack many of the leukocyte molecules implicated in transmigration (). To increase the efficiency of muscle repair by mesoangioblasts, it would be essential to increase their migration to skeletal muscle, with the additional benefit of reducing unspecific trapping in the capillary filters of the body, such as liver and lung. Here, we report that expression of α4 integrin and exposure of cells to SDF-1 or TNF-α improve up to fivefold migration of wild-type (WT) mesoangioblasts to the dystrophic muscles and consequent production of new fibers that express the normal copy of the mutated gene. These results elucidate the migration mechanism of mesoangioblasts and open a therapeutic opportunity for improving efficacy of cell therapy in muscular dystrophy. Mesoangioblasts (clone D16; ) were starved for 12 h in the absence of serum and then subjected to transmigration assays using transwell chambers, as described previously (). In a first series of experiments, we created an artificial environment where mesoangioblasts would face an activated endothelial layer separating them from differentiating muscle cells (much as it happens in regenerating muscle) or specific cytokines with known chemoattractive potency. As shown in (left), mesoangioblasts were unable to cross endothelium-coated filters in the absence of stimuli. However, multinucleated myotubes but not undifferentiated myoblasts induced active migration of mesoangioblasts but not primary fibroblasts. This effect was slightly stronger than that induced by FGF, used as a positive control. (right) shows a representative field of the lower side of the filter, fixed and stained. We also performed the transmigration assay through endothelium-coated filters in the presence of a panel of cytokines for which mesoangioblasts have receptors (). The presence of SDF-1 and TNF-α in the lower chamber caused a tenfold increase of mesoangioblast migration, a significantly more robust effect than that elicited by FGF or high mobility group box (HMGB) 1, previously described as an enhancer of mesoangioblast migration (); IL-1 had little effect, whereas IL-6 and -10 had no effect (, left). Mesoangioblasts stained at the lower side of the filter, confirming the positive transmigration effect of SDF-1 and TNF-α (, right). This is in accordance with the fact that myotubes secrete these cytokines at a higher rate than myoblasts. As shown in , SDF-1 or TNF-α as well as IL-6 and monocyte chemotactic protein 1 were detected at a higher level in the supernatant from myotubes than in the corresponding supernatant from myoblasts. These data suggest that differentiated myotubes may secrete growth factors or cytokines such as SDF-1 or TNF-α that favor mesoangioblast transmigration by either a chemotactic effect and/or by modifying the endothelium barrier. To assess the ability of mesoangioblasts to migrate in vivo from the vessel lumen to muscle interstitial tissue (see ), we injected 5 × 10 GFP-labeled mesoangioblasts into the right femoral artery of WT C57 (previously injected intramuscularly with cardiotoxin [ctx]), X chromosome–linked muscular dystrophy (mdx), and α-SG–null mice. Mice were killed 6, 12, or 24 h after injection, the muscles (quadriceps, gastrocnemius, and tibialis) from treated (right) and contralateral legs as well as filter organs (liver, spleen, and lung) were collected, and RNA was extracted. The percentage of migrated mesoangioblasts in each recipient organ was calculated by real-time PCR for GFP expression as a percentage of the value corresponding to the total number of injected cells. As shown in , mesoangioblasts were able to migrate to the muscles of the treated legs; ∼10% of injected cells could be recovered, whereas most of injected cells were retained in the different filter organs without reaching the contralateral muscles. We set our time of analysis at 6 h after injection for the in vivo experiments, as we observed that with the exception of ctx-treated mice, the number of injected cells remains constant for the first 12 h after the injection and thereafter decreases to varying extents (up to 50% of the value observed at 6 h; ) to increase again in the following days (see ). Interestingly, mesoangioblasts migrate more efficiently to muscles of ctx-treated normal mice or of α-SG–null mice (, top and bottom) than to muscles of the mdx mouse (, middle). As shown in , the presence of a higher concentration of TNF-α (and to a minor extent SDF-1) in muscles from ctx-treated WT or α-SG–null mice than in mdx mice could explain the preferential migration of mesoangioblasts. Although other cytokines were found to be more abundant in the muscles of mdx mice, these did not affect in vitro transmigration of mesoangioblasts (IL-1, -6, and -10) or, alternatively, mesoangioblasts do not express the corresponding receptors (monocyte chemotactic protein 1 and IL-8). Mesoangioblasts were injected in the femoral artery of WT (ctx injected), mdx, or α-SG–null mice at 2 or 8 mo of age, and different organs were collected after 6 h. As shown in , the number of mesoangioblasts that reached the damaged muscles was reduced to half when the injection was performed on old mice. This effect could be due to different adhesion molecules expressed by vessels in the muscles of young and old mice. Alexa 488–labeled mAbs anti–MAdCAM-1 and anti–E- or anti–P-selectin were injected in the tail vein of α-SG–null mice, and after 20 min, their accumulation was studied with intravital video microscopy. As shown in , E- and P-selectin were barely detected in the muscles of old α-SG–null mice. In contrast, in young α-SG–null mice, E- and P-selectin were clearly detected on capillaries and small vessels in the muscles. MAdCAM-1 and E- or P-selectin were not detected in muscles from old mdx mice, whereas only MAdCAM-1 was slightly expressed in young mdx mice (unpublished data). mRNA expression analysis confirmed that E- and P-selectins are highly expressed in muscles of 2-mo-old α-SG–null mice but barely detectable in 8-mo-old mice (). These results show that in addition to the reduction of the microvasculature that accompanies fat and connective tissue accumulation in aging, dystrophic muscles, reduced expression of adhesion molecules may further impair stem cell migration. Different types of stem cells have been shown to circulate and to be able to differentiate into skeletal muscle, thus representing potential candidates for a cell therapy protocol. Therefore, we compared migration to muscles of mesoangioblasts with mesenchymal and neural stem cells. 2.5 × 10 EGFP mesenchymal, neural, or mesoangioblast stem cells were injected into the femoral artery of mdx or α-SG–null mice, and after 6 h, quadriceps, gastrocnemius, and tibialis from the treated leg and liver, spleen, and lung were collected, and the percentage of donor cells was calculated by real-time PCR for GFP. As shown in , mesoangioblasts reached the muscles of the injected leg in higher numbers than the other stem cell populations in both animal models (left). These data show that mesoangioblasts naturally migrate to dystrophic muscle in comparison with the other cells tested. Interestingly, mesenchymal stem cells were recovered preferentially in the lungs (, right). Experiments with acutely isolated, noncultivated EGFP-hematopoietic stem cells were also performed. Even though results are not directly comparable with those obtained with in vitro–expanded cells, we observed that most hematopoietic cells were retained in the liver and the spleen (unpublished data). In an attempt to increase the number of mesoangioblasts that migrate to skeletal muscles, we followed two approaches. First, we pretreated mesoangioblasts with the cytokines SDF-1 and TNF-α for 12 h before challenging them in the transwell assay. The results, shown in , clearly indicated that pretreatment with either SDF-1 and TNF-α stimulated mesoangioblast migration to the same extent observed when the same cytokines were present in the lower chamber. Therefore, SDF-1– or TNF-α–pretreated GFP mesoangioblasts were injected through the femoral artery of ctx-pretreated WT, mdx, or α-SG–null mice, and after 6 h, muscles and filter organs were collected and analyzed by real-time PCR for presence of GFP. As shown in , both SDF-1 and TNF-α had a modest effect on WT mice (probably because of the acute inflammation induced by ctx) but increased about twofold the migration of mesoangioblasts to the injured muscles of dystrophic mice. Consistently, the number of mesoangioblasts detected in the filter organs was reduced to almost half of control. We then attempted to identify the surface molecules that may be responsible for the changes observed in mesoangioblast migration after pretreatment with SDF-1 or TNF-α. The possible change in the expression of several candidate surface molecules was first investigated by FACS analysis (). Many of the tested molecules were not changed by treatment with either SDF-1 or TNF-α. However, CD44 expression was increased more than twofold by TNF-α and not by SDF-1, whereas FGFR1 expression was increased only by SDF-1. Although these two molecules are involved in migration processes and could explain part of our effects, we performed a microarray analysis for 125 genes, including surface receptors, to look for more candidates genes induced in mesoangioblasts by SDF-1 or TNF-α. 16 genes were modified after pretreatment of mesoangioblasts with either SDF-1 or TNF-α (). Caspase-8 (Casp8), caveolin 1 (cav-1), CD44, cadherin 1 (Cdh1), metalloproteinase (MMP) 13, or vascular cell adhesion molecule (VCAM) 1 expression was increased two- to threefold only by TNF-α pretreatment, whereas Itgαv (αv integrin), MMP-14, or neural cell adhesion molecule (NCAM) 1 were induced approximately twofold only by SDF-1 pretreatment. Pretreatment with either TNF-α or SDF-1 increased expression of intercellular adhesion molecule (ICAM) 1, NCAM-2, different MMPs, platelet endothelial cell adhesion molecule (PECAM), and trombospondins. To verify the role in mesoangioblast transmigration of those molecules that appeared to be more potently induced by TNF-α or SDF-1, we repeated the transwell assay in presence of antibodies against VCAM-1, αv integrins, ICAM-1, or CD44 or MMP inhibitor GM1489. As shown in , antibodies against CD44 partially inhibited transmigration induced by TNF-α, whereas antibodies against αv integrin inhibited SDF-1–induced transmigration. Blocking MMP activity with GM1489 dramatically inhibited transmigration induced by both molecules. Together, these data suggest that TNF-α and SDF-1 increase mesoangioblast transmigration in vitro and migration to muscles in vivo by increasing expression of several molecules, among which CD44 and MMP appeared the most relevant in the case of TNF-α and αv integrin and MMP appeared most relevant in the case of SDF-1. As revealed by microarray analysis (), mesoangioblasts do not express some of the key molecules that control rolling and extravasation of leukocytes and different stem cells such as L-selectin, α4 integrin, or β2 integrin (). Therefore, we transfected mesoangioblasts with vectors expressing these molecules and EGFP, which allowed FACS selection of transfected cells. To verify the expression of these proteins, a Western blot for total lysates of transfected mesoangioblasts were performed (, left). Correct surface expression of these proteins was also verified by flow cytometry analysis (, right). Finally, mesoangioblasts transfected with the different constructs or pretreated with TNF-α or SDF-1 were subjected to differentiation assays. Expression of L-selectin or α4 integrin did not interfere with skeletal muscle differentiation of mesoangioblasts in vitro, whereas expression of β2 integrin inhibited differentiation (, bottom). To assess whether the expression of these molecules could improve mesoangioblast migration, cells transfected with one, two, or all three constructs were subjected to transmigration in vitro assay through activated endothelium. As shown in , α4 (but nor β2) integrin or L-selectin expression increased mesoangioblast transmigration six- to eightfold, whether or not C2C12-derived myotubes were present in the lower chamber. Thus, expression of the counterpart ligands for these molecules at the surface of the activated endothelium was sufficient to allow mesoangioblast transmigration independent of the presence of myotubes, with a chemoattractant-independent mechanism. Cotransfection of two or all constructs had only a slightly cumulative effect, which was not statistically significant. When α4 integrin– or L-selectin–expressing mesoangioblasts were injected into the femoral artery of ctx-treated WT, mdx, or α-SG–null mice, the number of cells that reached downstream muscles was increased approximately twofold with some variability in different mice (). In general, α4 integrin appeared to be more efficient than L-selectin. Therefore, expression of α4 integrins and L-selectin by mesoangioblasts improves their migration, possibly by favoring their interaction with the ligands expressed on the endothelium surface. We then tested the possibility of additive or synergistic effects of cytokines and adhesion molecules. Mesoangioblasts, previously transfected with α4 integrin or/and L-selectin constructs, were pretreated for 12 h with TNF-α or SDF-1 and then subjected to transmigration assay through endothelium. As shown in , some combinations were able to improve mesoangioblast transmigration in vitro, the most efficient of which was TNF-α treatment of α4 integrin–expressing cells, which migrated in vitro 15-fold more efficiently than control. Therefore, these cells were injected into the femoral artery of the different animal models: results showed that almost 50% of injected cells reached the injured muscles after 6 h, with a concomitant reduction in the number mesoangioblasts detected in filter organs (). To follow these modified mesoangioblasts, we performed immunohistology analysis of the tibialis anterior from an α-SG–null mouse whose femoral artery had been previously (6 h) injected with control or modified mesoangioblasts. As shown in (right), TNF-α–pretreated, α4 integrin–transfected mesoangioblasts (GFP, right) were able to cross the endothelial barrier of vessels (stained with PECAM, right) and localized in the surrounding muscle tissue (green, left) outside of and sometimes underneath laminin (red, left) in a position typical of satellite cells. Untreated mesoangioblasts were less efficient in crossing the vessel wall, and many fewer were found around the muscle fibers. To monitor the distribution of injected cells at 48 h after injection, muscle sections of transplanted mice were stained for GFP, laminin, and PECAM. As shown in , all GFP-modified mesoangioblasts were localized in the muscle tissue, mainly outside of the basal lamina and far from vessels, but in some case underneath it, in a position typical of satellite cells. Indeed, as shown in , wk after injection, we detected GFP mesoangioblasts expressing the satellite cell Myf-5 (; , left) that occupied the typical position of satellite cells (, right). 5 × 10 TNF-α–pretreated, α4 integrin–expressing mesoangioblasts were injected into the femoral artery of α-SG–null mice, and after 6 h or 1, 15, 30, 60, or 120 d, the tibialis anterior from treated legs was collected and analyzed by real-time-PCR for GFP or α-SG. As shown in , after an initial modest decline, the number of both control and modified mesoangioblasts began to increase by 15 d after the injection and reached a plateau by 1 mo. At all time points, modified mesoangioblasts were four- to sixfold more numerous than control mesoangioblasts. Likewise, α-SG became detectable ∼15 d after injection, when mesoangioblasts had been incorporated into regenerating fibers and continued to increase thereafter, again at a much higher level in muscles injected with modified mesoangioblasts. As a matter of fact, 4 mo after a single injection, the level of α-SG expression in the tibialis of injected α-SG–null mice was almost 60% of that detected in the corresponding tibialis from a WT mouse. At the same time, the level of expression of α-SG in the tibialis injected with control mesoangioblasts was only 10% of a WT mouse. In agreement with RT-PCR data, α-SG was detected in almost all (>90%) the muscle fibers of the tibialis injected with treated mesoangioblasts, whereas a lower percentage (∼20% with a weaker staining) of positive fibers was present in the tibialis injected with control mesoangioblasts (). The newly synthesized α-SG protein was also analyzed by Western blot. As shown in , after 4 mo, the quantity of protein detected was approximately sixfold higher in muscles injected with treated mesoangioblasts than with control cells. To test whether increased expression of α-SG would correspond to increased motility, mice were subjected to a running test on a treadmill. The number of times that WT mice, α-SG–null mice, either untreated or treated with control or modified (TNF-α–pretreated, α4 integrin–expressing) mesoangioblasts, fell into the grid when running was recorded. shows that α-SG–null mice injected with modified mesoangioblasts fell into the grid less frequently than noninjected or control mesoangioblast–injected α-SG–null mice. After this exercise test, many of the null mice appeared to be extremely fatigued and took a relatively long time to recover from the exercise, whereas treated mice seemed to recover from the effort much faster. Together, these data show that mesoangioblasts pretreated with TNF-α and transfected with α4 integrins are not only able to more efficiently reach the damaged muscle but also to reconstitute SG-positive fibers with much higher efficiency and improved muscle function. To test whether a similar protocol would also be suitable for human cells, we isolated human mesoangioblasts from small vessels of biopsies from normal human adult muscle. These cells have morphology and gene expression similar to their mouse counterparts, a finite lifespan, and a similar ability to differentiate into different types of muscle cells, osteoblasts, and adipocytes. Human mesoangioblasts were serum starved and subjected to transmigration assays using transwell chambers. As shown in , human mesoangioblasts were also unable to cross endothelium-coated filters in absence of stimuli. However, multinucleated myotubes or pretreatment of cells with TNF-α or SDF-1 increased human mesoangioblast migration four- to eightfold without significant enhancement of human fibroblast migration. We then transfected human mesoangioblasts (that also do not express these molecules; unpublished data) with vectors expressing L-selectin, α4 integrin, or β2 integrin and, after transfection, selected the GFP-positive cells with FACS. To verify the correct surface expression of these proteins, transfected human mesoangioblasts were also analyzed by flow cytometry analysis (). shows that pretreatment of α4 integrin– or L-selectin–expressing human mesoangioblasts with TNF-α or SDF-1 increased their ability to cross the endothelial barrier in transwell assays by ∼18-fold. Therefore, human mesoangioblasts, pretreated with TNF-α and transfected with vectors encoding for α4 integrin or L-selectin were injected into the femoral artery of mdx-SCID mice that do not reject human cells. shows that almost 40% of human modified mesoangioblasts reached the muscles 6 h after intrafemoral artery injection, with a concomitant reduction in the number of cells detected in filter organs. Together, these data show that it is possible to increase human mesoangioblast migration to dystrophic muscle with an experimental protocol similar to that developed for mouse mesoangioblasts. #text Anti-GFP antibody and anti-isotype control antibodies were purchased from Chemicon. Anti–α4, anti–αv, and anti–β2 integrins; anti–L-selectin; anti-CD44; anti-FGFR1; anti–VCAM-1; and anti–ICAM-1 antibodies were obtained from Serotec. Anti–L-selectin mAb Mel-14 was obtained from American Type Culture Collection. Anti–E- and anti–P-selectin mAbs (RME-1 and RMP-1) were obtained as previously described (; ). Anti–MAdCAM-1 and anti-PECAM were provided by E. Butcher (Stanford University, Stanford, CA) and E. Dejana (FIRC Institute of Molecular Oncology, Milan, Italy). Anti–α-SG and anti-myf5 antibodies were purchased from NovoCastra. mdx and C57BL10 WT mice were purchased from Charles River Laboratories, and homo-EGFP mice were provided by A. Nagy (Mount Sinai Hospital, Toronto, Canada; ). α-SG–null mice were a gift from K. Campbell (University of Iowa, Iowa City, IA; ). All mice were handled following institutional guidelines. C2C12 mouse myoblasts or L6 rat myoblasts were cultured in Dulbecco's minimal essential medium plus 20% FBS and induced to differentiate in Dulbecco's minimal essential medium plus 2% FBS. D16 mice mesoangioblasts were isolated from the dorsal aorta of mouse embryos (embryonic day 9.5), cloned, and expanded as described previously (). Mesoangioblasts transduced with a lentiviral vector encoding the GFP protein were used for some experiments. Murine microvascular endothelial H5V cells (a gift form E. Dejana, FIRC Institute of Molecular Oncology) and mouse embryonic fibroblasts 3T3 were grown in Dulbecco's minimal essential medium plus 10% FBS and starved for 12 h before functional experiments. Mesenchymal and hematopoietic stem cells were isolated from homo-EGFP mice using the SpinSep TM kit (StemCell Technologies, Inc.) and checked for purity according to the manufacturer's instructions. Neural stem cells were isolated from the forebrains of embryos (embryonic day 11.5) as previously described (). Murine or human β2 integrin, L-selectin, and α4 integrin mRNA cloned in a pT7T3D-Pac1 vector were purchased from Research Genetics. Inserts were cloned into pIRES2-EGFP expression vector from CLONTECH Laboratories, Inc., and transfected into mice or human mesoangioblasts with Lipofectamine (Invitrogen). GFP-positive mesoangioblasts were sorted and used for the different experiments. Treated mesoangioblasts were cocultured with L6 cells in low serum (2%). After 3 d, dishes were fixed with 4% PFA, and the number of mouse nuclei stained with DAPI inside myosin-positive cells was counted. The percentage of differentiation was calculated against the total number of cells. Total RNA from different organs was isolated with TRIzol protocol (Invitrogen) and reverse transcribed by Taqman kit (Platinum Taq DNA polymerase; Invitrogen). RT-PCR was performed for E- and P-selectin (E-selectin primers: Fw, CCATGTTCCACGTCAAGGACC; Rw, GCCATGTGATAGGCCACAG; and P-selectin primers: Fw, CCAGTATCTAGACCCGAAGG; Rw, GCCATTCTCTCTCCTCGAAACAC), and mRNA expression was analyzed by agarose gel. Alternatively, real-time quantitative PCR was done on a real-time PCR system (Mx3000P; Stratagene). Each cDNA sample was amplified in duplicate by using the SYBR Green Supermix (Bio-Rad Laboratories) for GFP or ɛ chromosome detection (GFP primers: Fw, AAGTTCATCTGCACCACCG; Rw, TCCTTGAAGAAGATGGTGCG; and ɛ chromosome primers: Fw, AAGCGACCCATGAACGCATT; Rw, TTCGGGTATTTCTCTCTGTC) or the Taqman Universal PCR master mixture containing AmpliTaq Gold DNA with commercial primers (Applied Biosystems) for α-SG detection. Data are expressed as the percentage of migrated cells, which is calculated by comparing the concentration of target genes in our sample with the total input of injected cells. For the Taqman assay, the level of α-SG detected represents the specific signal detected in the sample (mesoangioblast- injected α-SG–null mouse tibialis anterior) compared with a positive control (WT mouse tibialis anterior; i.e., 100% of positive fibers). 8-μm transwell filters (Corning) were coated with 1% gelatin, and endothelial cells H5V (previously preactivated by 12 h of exposure to TNF-α) were plated to confluence on them for 24 h. Confluence of the endothelial monolayer was assessed by measuring the diffusion of BSA from the upper to the lower chamber. At the same time, C2C12 were grown on a p24w plate with or without differentiated medium for 0–4 d. 10 mice mesoangioblasts, WT or pretreated or transfected, were plated in Dulbecco's minimal essential medium containing 2% serum on the upper side of transwell chamber 24 h after plating H5V cells, and the chamber was then moved to the wells containing differentiated C2C12 or different cytokines. After 6 h of transmigration, migrated mesoangioblasts on the lower side of the filter were fixed in 4% paraformaldehyde, stained with crystal violet, and counted using an inverted microscope (five random fields of the lower face of the transwell membrane at 20× magnification). The results show migrated cells as a percentage of the total number of input cells. 2- or 8-mo-old WT (injected slowly through all the anterior surface of the tibialis with a 27-gauge needle containing 100 μl of 5 μM ctx [Sigma-Aldrich] 12 h before treatment), mdx, or α-SG–null mice were injected through the right femoral artery with 5 × 10 mouse mesoangioblasts. After 6, 12, or 24 h, animals were killed, and different muscles (quadriceps, gastrocnemius, and tibialis) or filtered organs (liver, lung, and spleen) were collected. RNA was extracted, and a real-time PCR for GFP was performed in all the samples as described (see Real-time PCR). Data are represented as a percentage of migrated cells (percentage of GFP detected) to the different organs relative to the input value. 50 μg of Alexa 488–labeled mAbs anti–E- or anti–P-selectin or anti–MAdCAM-1 were injected through the tail vein of α-SG–null mice. 20 min later, the animal was anesthetized and perfused through the left ventricle with cold PBS (). Blood vessels of striated muscles were visualized with a silicon-intensified target video camera (VE-1000 SIT; Dage-MTI) and monitor (SSM-125CE; Sony). For mRNA analysis, total RNA was extracted from treated mesoangioblasts using TRIzol. cDNA was prepared by reverse-transcription reaction and hybridized to the GEArray gene expression array MM-010 (SuperArray) for extracellular matrix and adhesion molecules according to the manufacturer's instruction. Data were subjected to densitometric analysis using GEArray Expression Analysis Suite (SuperArray). RNA levels were expressed as the relative density after normalizing the hybridization signal to β-actin. Mesoangioblasts were transfected with the different constructs (see Constructs) and lysed directly in Laemmli buffer on ice. Lysates were resolved on 10% SDS-PAGE under reducing conditions, and proteins were transferred to nitrocellulose membrane (Hybond-ECL; GE Healthcare). Membranes were revealed with anti-GFP and anti–α4 or anti–β2 integrins or anti–L-selectin antibodies. Mesoangioblasts pretreated with different cytokines or transfected with distinct constructs were detached with PBS plus 5 mM EDTA on ice and incubated with antibodies against the different surface molecules for 30 min at 4°C. Cells were then incubated with a Alexa 488–conjugated anti–mouse Ig. Finally, fluorescent samples were analyzed in a FACSCalibur flow cytometer (Becton Dickinson). Supernatants from fibroblasts, myoblasts, or myotube cultures were collected after 4 d of culture and analyzed with the Beadlyte mouse multicytokine-detection system (Upstate) for measuring cytokines according to the manufacturer's instructions. Muscles from different animal models were homogenized and cytokines were measured using the same kit protocol. Tibialis anterior muscles of α-SG–null mice were removed from mice previously injected with mesoangioblasts and frozen in liquid N cooled isopentane. Serial muscle sections were fixed with 4% PFA, permeabilized, satured, and immunostained as previously described () with anti-GFP, anti–α-SG, anti-laminin, anti-myosin, anti-PECAM, or anti-myf5 antibodies. Alexa 488 or 594 or Cyane (Invitrogen) were used as secondary staining Ig and DAPI for nuclear staining. Isotype control antibodies were used as negative staining. Images were taken with a microscope (S100 TV; Carl Zeiss MicroImaging, Inc.). α-SG–null mice ( = 4 for each of the four groups) injected or not with control or treated mesoangioblasts were run on a treadmill (Columbus Instruments) set at 7 m/min for 2 min before and after 1 mo from injection. WT C57 mice are also presented as healthy control. The back of the treadmill was equipped with a grid that discharged a mild current, a stimulus designed to motivate the animal to keep running on the treadmill. Performance was measured by the number of times a mouse failed to stay on the running belt and fell into the stimulus grid. Running time was limited to 2 min because the majority of the α-SG–null mice could not run any longer. Statistical significance of the differences between the percentage values was assessed by using the Kruskal-Wallis one-way analysis of variance by ranks test. α represents the significance in each independent case.

The syncytial myofiber is the functional unit of skeletal muscle. As with many other adult tissues, skeletal muscle contains resident stem cells, termed satellite cells because of their location on the periphery of myofibers under the surrounding basal lamina (; for review see ). Satellite cells proliferate to provide myonuclei to growing myofibers before becoming quiescent in mature muscle (). Fulfillment of satellite cell functions of maintenance, hypertrophy, and repair requires that they first be activated to enter the cell cycle and produce large numbers of myoblast progeny. Most of these subsequently differentiate and either fuse with existing myofibers or form new myotubes (), but others adopt an alternative fate and self-renew to maintain the satellite cell pool (; ). What controls the crucial transition of satellite cells from quiescence to proliferation remains largely unknown. Various stimuli released from crushed myofibers, invading macrophages, and connective tissue have been proposed to trigger satellite cell activation (for review see ). One of these signals is hepatocyte growth factor (HGF; ), and members of the FGF family have also been proposed to recruit satellite cells to division (; ). Because the receptors for both HGF () and FGF are members of the tyrosine kinase family of receptors, attention has been directed toward classical kinase-mediated signaling in control of satellite cell activation (; ; ) and proliferation in myogenic cells (; ). Over the last few years, the importance of sphingolipid signaling has begun to be understood (for review see ). Sphingosine-1-phosphate (S1P) and homologous phosphorylated long-chain sphingoids act in diverse organisms such as mammals, worms, flies, slime mold, yeast, and plants. In mammals, S1P is mitogenic for several cell types, including fibroblasts and endothelial cells (; ; ). In addition to this role in cell division and survival, S1P is involved in processes such as calcium homeostasis and cell migration during angiogenesis and cardiac development, as well as in the adult immune system (for review see ). In contrast, other sphingolipids such as ceramide and sphingosine are associated with cell growth arrest, stress responses, and apoptosis (for review see ). Recent studies have revealed that sphingolipids are also active in muscle. Defects in muscle development and integrity in are observed when the level of S1P is perturbed by targeted deletion of S1P lyase, an enzyme responsible for the irreversible degradation of S1P (). Sphingosine can also affect muscle contraction by modulating the function of certain calcium channels (for review see ), whereas ceramide has an inhibitory effect on insulin-like growth factor I–induced protein synthesis in mouse myogenic C2C12 cells (). S1P affects Ca homeostasis and cytoskeletal rearrangement () in C2 cells and stimulates differentiation in this cell line (). Sphingomyelin is a precursor of ceramide, sphingosine, and S1P, which are all components of the same metabolic pathway (). Sphingomyelin is one of the major lipid components of cell membranes in animals, with significant amounts in the plasma membrane (for review see ). Although the location of sphingomyelin used to generate signaling molecules such as S1P is unclear, the inner leaflet of the plasma membrane has been suggested as a source (; ; ; ). We have recently demonstrated the presence of high concentrations of sphingomyelin in the plasma membrane of quiescent, but not proliferating, satellite cells (). Here, we focus on the role of S1P as a novel regulator of satellite cells and muscle regeneration. S1P can induce satellite cells to enter the cell cycle, whereas inhibiting the sphingolipid signaling cascade that generates S1P significantly reduces the number of satellite cells able to divide in response to mitogen stimulation. Sphingomyelin is hydrolyzed by neutral sphingomyelinase (N-SMase) and can then be further metabolized to S1P, to mediate the mitogenic signal. By use of a novel combination of the sphingomyelin binding protein lysenin (; ; ) as a cytochemical probe, together with selective sphingomyelin digestion, we demonstrate that the main source of S1P is sphingomyelin located in the inner leaflet of the plasma membrane. Crucially, inhibiting S1P production after muscle damage greatly perturbs subsequent muscle regeneration. Together, our observations show the central role that sphingolipid signaling plays in mediating the entry of satellite cells into the cell cycle and muscle regeneration. #text Precise control of skeletal muscle size, along with rapid and appropriate repair and maintenance, is predominantly a function of the satellite cell population and demands a well-controlled transition between quiescence and proliferation. Although several factors involved in this process during muscle regeneration have been described, the process is far from fully understood (for review see ). Here, we show that this mitogenic signal is in part mediated in satellite cells by the cleavage of sphingomyelin at the inner leaflet of the plasma membrane that is ultimately metabolized to generate S1P, which in turn stimulates entry into the cell cycle. As would be expected from these observations, inhibition of S1P synthesis perturbs muscle regeneration in vivo. S1P can function in two ways: either as an agonist for specific cell surface receptors or as a second messenger. S1P was shown to be a ligand for the EDG-1 (S1P) receptor () and is now recognized as an agonist for five (S1P through S1P) G-protein–coupled cell surface receptors. S1P through S1P are coupled to different G-proteins and so the relative abundance of specific receptors and G-proteins allows S1P to have heterogeneous effects. The C2 myogenic cell line was originally derived from regenerating adult muscle () and has been shown to express S1P through S1P (). S1P, acting through S1P, induces a small decrease in the number of cells proliferating in response to serum and stimulates the appearance of proteins associated with differentiation (). Our results are consistent with those of where they overlap because, in our hands, S1P did not prevent differentiation in cycling C2C12 cells, but we did not determine whether it actively promoted it. Although the S1P receptors present on satellite cells are presently unknown, analysis has shown that S1P mRNA is present in quiescent satellite cells, but the levels are dramatically lower in dividing cells (Montarras, D., and Buckingham, M., personal communication). Intriguingly, overexpression of S1P in cycling C2C12 cells suppresses markers of myogenic differentiation elicited by S1P (). S1P has also been implicated as a second messenger for cell growth and survival independently of its receptors (; ; ). For example, it has recently been shown that VEGF binding to VEGF receptor 2 stimulates endothelial cell growth through protein kinase C, which leads to the activation of sphingosine kinase 1 and the generation of S1P. The extracellular signal–regulated kinase (ERK)–MAPK pathway is then activated by S1P, resulting in cell division (). Because HGF also binds a tyrosine kinase type receptor, it is possible that the activation of sphingosine kinase 1 and subsequent production of S1P may also stimulate ERK–MAPK, already shown to be involved in satellite cell activation (; ; ). In addition to HGF (), TNF-α has recently been proposed to be able to activate satellite cells, as systemically delivered TNF-α enhances BrdU incorporation into skeletal muscles, presumably by the intermediary of satellite cells (). Although TNF-α can work through serum response factor, it also activates sphingolipid signaling in various cell types and induces sphingomyelin cleavage (; ). Therefore, lipid signaling may act as a second messenger or via receptor-mediated pathways, or a combination of the two, in induction of satellite cell proliferation. In contrast to the mitogenic effects of S1P, ceramide is associated with cell growth arrest, stress responses, and apoptosis in several cell types (for review see ). In muscle, ceramide has been shown to suppress the hypertrophic effects of insulin-like growth factor I () as well as that of insulin in C2C12 cells (). As it has been proposed that the relative levels of these interconvertible metabolites can determine cell fate (), it is possible that S1P and ceramide also exert contrary effects in satellite cells. Higher ceramide levels may be involved in the maintenance of satellite cell quiescence, whereas increasing S1P levels would act to overcome this block, inducing activation and subsequent proliferation. Although both DMS and GW4869 effectively reduce the mitogenic effect of serum, neither of them was able to totally prevent the activation of all myogenic cells. There is increasing evidence that satellite cells may represent a heterogeneous population (), so it is possible that they have different requirements for sphingolipid signaling. Another possible explanation is that the pharmacological inhibitors do not completely block the target enzyme, which would have a significant influence when the ratio of S1P to ceramide, rather than their absolute amounts, is proposed to be the crucial factor in the cellular response (). It is also probable that there are other pathways that control satellite cells in addition to S1P-mediated signaling (; ) and that signal redundancy and cross talk between pathways may reduce the effects of specifically inhibiting a particular signaling route. Sphingolipid metabolites can be generated from two main sources: de novo synthesis of ceramide or cleavage of sphingomyelin. The site of the subcellular pool of sphingomyelin that is used for signaling, however, is unclear. Although sphingomyelin is mostly located at the outer leaflet of the plasma membrane (), several lines of evidence suggest that there is a distinct pool of signaling sphingomyelin in the inner leaflet, which acts as a reservoir for generating sphingolipids (; ; ). In particular, labeling sphingomyelin in various compartments with H before stimulation and determining the location and amount of labeled metabolites () or using subcellular fractionation to locate membranes depleted of sphingomyelin () both indicate the inner leaflet as the source. The enzyme mainly responsible for N-SMase activity outside the CNS, sphingomyelin phosphodiesterase 2 (), is primarily located in the plasma membrane (). Furthermore, N-SMase activity predominantly occurs in the plasma membrane fraction, and receptor-coupled sphingomyelin degradation has been shown to mostly occur at this site (). The novel technique that we have used, combining probing with the sphingomyelin binding protein lysenin (for review see ) with selective sphingomyelin digestion, permits separate microscopic visualization of the dynamics of sphingomyelin in either the outer or inner leaflets on an individual cell basis. Using this method, we show that the inner leaflet of the plasma membrane contains a pool of sphingomyelin that can be recruited for signaling during induction of myogenic cell proliferation. This reduction in sphingomyelin levels in the inner leaflet in response to CFCS stimulation was caused by N-SMase catalyzed cleavage, as the N-SMase inhibitor GW4869 () both prevented the transient drop in sphingomyelin and drastically reduced the number of cells entering the cell cycle. These observations strongly indicate that the inner leaflet is the source of the mitogenic S1P, consistent with and extending previous studies (; ; ). In conclusion, we demonstrate that sphingolipid signaling plays a central role in adult stem cell biology. Upon stimulation, sphingomyelin in the inner leaflet of the plasma membrane is cleaved by N-SMase, and subsequent metabolism results in an increased level of S1P, which acts as a mitogen for muscle satellite cells. As would be predicted from these observations, inhibition of S1P generation perturbs muscle regeneration. Indeed, a recently described mutation in choline kinase β gene, central to phospholipid biosynthesis, results in a rapidly progressive muscular dystrophy (). Coordinated control of satellite cell function is crucial to the regenerative responsiveness of muscle both during aging () and in hereditary myopathies. Lipid signaling potentially offers new targets for therapeutic intervention that could augment/restore satellite cell function and may well be involved in the control of related stem cell systems. The C2 () subclone C2C12 was maintained in DME containing 20% FCS, 4 mM -glutamine, 100 U/ml penicillin, and 100 μg/ml streptomycin at 37°C in 5% CO. To prepare reserve cells, proliferating C2C12 cells were switched to DME containing 1% horse serum for 5 d, after which few cells were still dividing as shown by BrdU incorporation. CFCS was prepared by incubating 10 ml FCS with 1 g of activated charcoal (Sigma-Aldrich) overnight at 4°C before the charcoal was removed by centrifugation and filtration. Cells were stimulated with either 0.5–2% CFCS or S1P (BIOMOL Research Laboratories, Inc.) in DME containing 4 mg/ml fatty acid–free BSA (Sigma-Aldrich) for 18 h, and 10 μM BrdU was added for 6 h before fixation with 4% PFA/PBS. Cells were preincubated with inhibitors for 45 min before serum stimulation. DMS (BIOMOL Research Laboratories, Inc.) stock solution was prepared in DMSO and used at a final concentration of 2–10 μM. GW4869 (Calbiochem) was stored as a 1.5 mM suspension in DMSO and solubilized by the addition of methane sulphonic acid to a final concentration of 0.25%, immediately before use at 10–20 μM. FB1 (BIOMOL Research Laboratories, Inc.) was dissolved in distilled water and used at 10–20 μM. Adult (>8 wk of age) C57Bl/6 or C57B16/DBA2 mice were killed by cervical dislocation before the extensor digitorum longus muscles were carefully removed. Myofibers were isolated in DME as described previously () and cultured in either DME/BSA or DME with 10% [vol/vol] horse serum and 0.5% [vol/vol] CEE [MP Biomedicals]) at 37°C in 5% CO. Drug treatments were performed in the same way as for C2C12 cells, except that BrdU was added to the medium from the beginning of culture. Cells were treated with 100 mU/ml bSMase from (Sigma-Aldrich) for 2 h at 37°C in 5% CO either before or after fixation with 4% PFA. When GW4869 was used, the cells were washed in DME after bSMase treatment. Fixed cells were blocked with 0.5% BSA in PBS and incubated with 0.2–10 μg/ml lysenin (provided by S. Kawashima and N. Ohta, Zenyaku Kogyo, Tokyo, Japan; Peptide Institute, Inc.) in 0.5% BSA in PBS for 1 h. Lysenin was then visualized with an anti-lysenin antibody (Peptide Institute, Inc.). To elicit muscle damage, both TA muscles were injected with 20 μl of the snake venom cardiotoxin (Sigma-Aldrich) together with India ink to mark the injection site. The right TA also received DMS (1 μmol/kg of body weight) dissolved in DMSO, whereas the left served as a control, receiving only DMSO. 7 d later, both muscles were removed and cryosectioned, and separate sections were analyzed by hematoxylin and eosin staining or immunostaining for eMyHC. The area occupied by regenerating myofibers was then determined by using Scion Image (Scion Corp.) from several such experiments, and the data were pooled and expressed as a percentage ± SEM of the total area assayed. For BrdU detection, fixed C2C12 cells or myofibers were permeabilized with 0.5% (vol/vol) Triton X-100 and treated with 3 N hydrochloric acid for 10 min at room temperature. For immunostaining for PCNA, myofibers were treated with 100% methanol. Frozen sections were fixed with 10% formalin/PBS for 30 min at room temperature and then autoclaved for 10 min in 20 mM Tris-HCl buffer, pH 9.0. After these treatments, samples were incubated for 2 h at room temperature with primary antibodies in 0.5% BSA/PBS, followed by fluorochrome (Alexafluor 488 with Alexafluor 594 or TRITC)–conjugated secondary antibodies (Invitrogen) before mounting in Fluoromount fluorescent mounting medium (DakoCytomation) containing 100 ng/ml DAPI. Primary antibodies used were monoclonal rat anti-BrdU (clone BU1/75; Abcam), monoclonal mouse anti-PCNA (clone PC10; DakoCytomation), monoclonal Pax7 (Developmental Studies Hybridoma Bank), monoclonal eMyHC (clone F1.652; American Type Culture Collection), monoclonal MyoD (clone 5.8A; DakoCytomation), polyclonal rabbit anti-lysenin (Peptide Institute, Inc.), and polyclonal MyoD (Santa Cruz Biotechnology, Inc.). Immunostained cells and myofibers were viewed on an epifluorescence microscope (model Axiophot; Carl Zeiss MicroImaging, Inc.) using 10×/0.30 Ph1-, 20×/0.50 Ph2-, and 40×/0.75 Ph2-PlanNeofluar lenses. Digital images were acquired with a charge-coupled device camera (model RTE/CCD-1300-Y; Princeton Instruments, Inc.) at −10°C using MetaMorph software version 4.5r5 (Universal Imaging Corp.). Images were optimized globally for contrast and brightness and assembled into figures using Photoshop 7.0.1 (Adobe).

Elimination of pathogens by phagocytosis is an essential component of the innate immune response. Phagocytosis is a complex sequence of signaling and cytoskeletal remodeling events that culminates in the engulfment of microorganisms into a vacuole. The process is initiated by recognition of ligands on the surface of the pathogen by specialized phagocytic receptors. Progressive zippering of receptors to multiple ligands on the target particle drives the apposition of the host cell membrane to the surface of the pathogen. In this specialized area of contact, known as the phagosomal cup, receptor clustering unleashes a signaling cascade that ultimately promotes actin polymerization, pseudopod extension, and particle internalization. Phosphoinositides play a critical role in the initiation of phagocytosis. Phosphatidylinositol-4,5-bisphosphate (PtdInsP) undergoes a biphasic change at the phagocytic cup: an initial, transient increase that is followed by its virtual disappearance by the time phagosomal sealing is complete (). These changes appear to be essential for successful completion of phagocytosis, as interference with PtdInsP biosynthesis or catabolism impairs particle engulfment (; ; ; ). Conversion to phosphatidylinositol-3,4,5-trisphosphate (PtdInsP) is partly responsible for the disappearance of PtdInsP from the phagosomal cup. Accordingly, formation of PtdInsP can be readily detected at the base of the nascent phagosome (), and inhibition of phosphatidylinositol-3-kinases, which are responsible for its synthesis, effectively blocks the uptake of phagocytic particles that are >3 μm (). Remarkably, the reported changes in PtdInsP and PtdInsP are confined to the phagocytic cup, without detectable alteration of the inositides in the unengaged (bulk) plasma membrane (PM). Biological membranes are generally regarded as fluid mosaics wherein lipids or clustered lipid microdomains can diffuse freely (). At physiological temperatures, such unrestricted lateral diffusion would result in rapid redistribution and homogenization of phospholipids. It is therefore unclear how gradients of inositides can be sustained for the time required for phagosome formation, which can exceed 3–4 min for large particles. Two possibilities can be envisaged: first, the lipids may be continuously generated at sites of phagocytosis and, although able to diffuse, they may be rapidly hydrolyzed as they leave the cup. Thus, a dynamic steady-state gradient could be achieved. Alternatively, the diffusion of lipids at sites of phagocytosis may be restricted, differing from their mobility in the bulk of the plasmalemma. Mobility may be restricted within or across the boundary of the phagocytic cup. Lipids are important determinants of the distribution of membrane proteins. Extrinsic proteins can associate with lipid headgroups, and transmembrane proteins segregate into microdomains according to the nature of the surrounding lipids. Several important signal transduction proteins associate with membranes by inserting their acyl and prenyl moieties into the hydrophobic domain of the bilayer. It can be anticipated that changes in the lipidic composition of the membrane during phagocytosis would have important consequences on the distribution and hence the activity of signaling proteins. In fact, phosphoinositides are thought to contribute to signal transduction by recruiting adaptor and effector proteins to sites of phagocytosis (; ; ; ; ). Clearly, the distribution and mobility of lipids and lipid-associated proteins is critical for vectorial transduction of signals during phagocytosis. However, the mobility of specific lipids in native membranes is difficult to analyze. Introduction of fluorescent moieties can alter the size, charge, and/or conformation of their headgroup or tail, and defined labeled lipids are rapidly converted to other chemical species. An alternative method frequently used to study lipids in cells, namely, the expression of fluorescent chimeric proteins containing specific lipid binding domains, is of limited use to study mobility. The limitation stems from the fact that the complex formed between the lipid and the chimera is in rapid dynamic equilibrium, with dissociation occurring much faster than the movement of the lipid in the plane of the membrane (). Because of these limitations and because of their importance in signal transduction, we decided to analyze instead the mobility of lipid-linked proteins at the phagosomal cup. FRAP was used for this purpose. Various constructs were used that targeted either the inner or outer monolayer of the plasmalemma and that resided preferentially or, alternatively, were excluded from areas rich in saturated lipids. Using large phagocytic targets and a combination of bright-field and confocal fluorescence microscopy, we were able to establish that the mobility of saturated lipids is drastically reduced at the phagocytic cup by a process that requires receptor-induced tyrosine phosphorylation. The mobility of membrane-associated molecules in activated macrophages was studied earlier by sedimentation of suspended cells onto IgG-coated surfaces (). This system has distinct optical advantages, as the membrane becomes activated at a fixed, predictable focal plane. However, this model of abortive phagocytosis does not recapitulate all aspects of the engulfment process and may involve components of cell spreading onto the substratum. On the other hand, phagocytosis of small particles is not amenable to the study of lipid mobility because of the rapidity of the internalization event and the small cup size (Fig. S1, available at ). As an alternative, we used large (8.3-μm diameter) particles as phagocytic targets. The size of these particles is similar to that of apoptotic cells that are commonly ingested by macrophages (). The distribution of PM-GFP in macrophages is shown in . PM-GFP is a chimeric construct of the N-terminal 10 amino acids from Lyn with GFP. The N-terminal sequence of Lyn directs myristoylation and palmitoylation of the chimera, which targets the fluorescent protein to the inner monolayer of the plasmalemma (). Transverse (x vs. y; ) and sagittal (x vs. z; ) sections of the cells confirm that PM-GFP is largely plasmalemmal, although varying amounts of endomembrane staining can be seen, depending on the expression level. The mobility of PM-GFP was initially assessed by FRAP in unstimulated cells. As shown in and particularly in the three-dimensional reconstruction of , the optical setup used bleached a nearly circular area of ∼2 μm in diameter within 1–2 s. Under the conditions of our experiments, ∼10–20% of the original intensity remained after bleaching (). In otherwise untreated cells, the fluorescence recovered almost completely (, squares; and ); in eight determinations, the mobile fraction (MF) averaged 1.10 ± 0.04 (these and all subsequent data are presented as means ± 1 SEM of the indicated number of determinations). Recovery was half maximal ( ) after 15 ± 2 s, indicative of a diffusion coefficient of 1.5 × 10 cm/s. This value is very similar to that we find for phospholipids in these cells. As shown in and , fluorescently labeled phosphatidylserine and phosphatidylethanolamine recover from photobleaching with comparable kinetics, yielding diffusion coefficients 2.4 × 10 and 2.2 × 10 cm/s, respectively. Together, these findings imply that the mobility of PM-GFP in the membrane is limited by association of its acyl chains with other constituents of the bilayer and not by drag imposed by the GFP itself. Accordingly, the diffusion coefficient of free GFP in the cytosol has been estimated at 2.5 − 3.0 × 10 cm/s (), much faster than that of the diacylated construct. RAW cells were exposed to IgG-opsonized latex beads to assess the mobility of PM-GFP at the phagocytic cup (). The considerable time required for complete engulfment of the large beads (≥6 min; Fig. S1) enabled us to perform photobleaching and measure the recovery of fluorescence before phagocytosis was completed. shows that the area of the membrane engaged in particle recognition and engulfment (the cup) was readily identifiable and sufficiently large to accommodate the ∼2-μm bleaching zone that was used. To normalize the recovery for changes in focal plane, photobleaching, or de novo delivery of probe, recovery was also measured at an area of unengaged membrane of the same cell. Although theoretically the MF cannot exceed 100%, membrane convolution at sites of ingestion can occasionally cause the MF to exceed this value. Typical results are shown in . Although the unengaged area of the membrane behaved as described for the membrane of resting cells (MF = 0.8–1), the fraction of mobile PM-GFP at the cup was markedly decreased (MF = 0.36 ± 0.01; = 8). The diffusion rate of the remaining MF was indistinguishable from the PM-GFP in the bulk membrane. These findings suggest that the mobility of acylated proteins is diminished in the vicinity of the Fcγ receptors associated with and activated by the phagocytic particle. Phagosomes undergo membrane remodeling after sealing, a consequence of active fusion and fission events (). We found that for large beads such as those used in the present study, remodeling starts even before phagocytosis is completed (Fig. S2, available at ). It was therefore important to ascertain that the failure of PM-GFP fluorescence to recover was a reflection of immobility and not its removal from the membrane. To this end, we performed careful quantitation of the rate of disappearance of PM-GFP and of another membrane marker, glycosylphosphatidylinositol (GPI)-anchored YFP during phagocytosis of large beads. Up to 40% of the fluorescence is lost from the base of the cup in 3 min (Fig. S2). We therefore limited our FRAP experiments to the initial 100 s, when the loss by remodeling is modest. Stimulatory Fcγ receptors bearing an immunoreceptor tyrosine-based activation motif associate with and become phosphorylated by Src-family kinases, including Lyn. Because the N-terminal sequence used to target PM-GFP to the membrane was derived from Lyn, we considered the possibility that a specific, direct interaction with the receptor complex might account for the reduced mobility of the fluorescent probe. As an alternative strategy to target GFP to the inner aspect of the PM, we attached the C-terminal 9 amino acids of H-Ras to GFP. In addition to the prenylation that is characteristic of all Ras isoforms, the C terminus of H-Ras is doubly acylated, and the chimeric GFP comprising the 9 amino acids of H-Ras (GFP-tH) undergoes the same posttranslational modifications (). As a result, GFP-tH targets almost exclusively to the plasmalemma in BHK cells (), as well as in RAW macrophages (). As shown in (solid squares) and , when photobleached in resting macrophages or in the unengaged region of cells engulfing beads, GFP-tH recovered rapidly ( = 10 ± 0.5 s; = 8) and extensively (MF = 0.78 ± 0.01). In contrast, at the phagosomal cup, a sizable fraction of GFP-tH was immobile during the period analyzed (MF = 0.37 ± 0.01; = 8). The diffusion of the MF was similar to that of the unengaged control membrane (). Therefore, two different acylated GFP probes displayed reduced mobility in nascent phagosomes. Because the C terminus of H-Ras is not anticipated to interact with Fc receptors, it is unlikely that direct association with the receptor complex is responsible for the immobilization of either probe. The preceding data indicate that two different probes anchored to the cytosolic aspect of the membrane exhibit reduced mobility at sites of phagosome formation. One possible obstacle to the movement of inner membrane–associated GFP is the actin cytoskeleton. Accumulation of actin and other cytoskeletal proteins is a well-established feature of phagosome generation (). We therefore considered whether the partial immobilization of PM-GFP and GFP-tH at the cup resulted from steric hindrance by actin-associated proteins. It has been shown that preventing actin polymerization with cytochalasin D does not abrogate bead engagement or the subsequent tyrosine phosphorylation of Fc receptors (). In accordance with these findings, we found that when large opsonized beads were added to cytochalasin-treated cells, cup formation was evident (), enabling us to perform FRAP with minimal actin polymerization. In cells treated with cytochalasin D, the mobility of PM-GFP in the bulk membrane was unaffected (MF = 0.84; ). However, the area of the membrane engaged in phagocytosis still showed reduced mobility (). To further study this phenomenon, we took advantage of the observation that under normal conditions actin dissociates from the membrane as the phagosome seals. The accumulation of F-actin at the base of the cup and its dissociation from the membrane of recently formed phagosomes is documented dynamically in Video 1 (available at ). Note that the amount of residual F-actin associated with formed phagosomes is minute, much lower than that of the unengaged PM. Because of the premature membrane remodeling observed during engulfment of large beads, smaller particles (3.1 μm) were used to ensure sufficient retention of the probe in sealed phagosomes (). We were thus able to compare the lateral mobility of PM-GFP in actin-depleted sealed phagosomal and PMs. As shown in , the MF of the probe in the sealed vacuole remained considerably lower than that of the bulk plasmalemma (MF = 0.26 ± 0.02 and 0.9 ± 0.02, respectively; = 8). Qualitatively similar results were obtained using 8.3-μm beads, but the results were less reliable because of the small amount of fluorescence remaining after sealing (unpublished data). Given the small amount of actin that remains associated with the formed phagosome, it is unlikely that the reduced mobility of lipid-associated probes is attributable to physical hindrance by the cytoskeleton. Lipids containing saturated acyl chains are thought to localize preferentially in sphingolipid and cholesterol-enriched microdomains often called “rafts” (). Cross-linked Fcγ receptors are thought to cluster in similar microdomains (). It therefore seemed likely that the reduced mobility of PM-GFP and GFP-tH could result from trapping in poorly mobile raft aggregates at the phagocytic cup. To test this hypothesis, we measured the lateral mobility of GPI-GFP and -YFP, as GPI-linked proteins partition selectively in rafts (). When expressed in macrophages, GPI-GFP is present largely in the plasmalemma (). That the protein is anchored to the outer monolayer was verified by its accessibility to anti-GFP antibodies added extracellularly to intact cells (unpublished data). Most of the GPI-anchored probe was mobile in resting cells (MF = 0.81 ± 0.04; = 8) and in the unengaged regions of the membrane of cells performing phagocytosis (). The diffusion rate was similar to that of PM-GFP (). More important, the MF and rate of diffusion of GPI-GFP or -YFP were virtually identical at the phagocytic cup and elsewhere in the unengaged membrane (). Therefore, the behavior of GPI-anchored probes is distinctly different from that of PM-GFP and GFP-tH. It has recently become apparent that different types of lipid microdomains (rafts) can coexist in cells (). Therefore, the differential behavior of GPI- and PM-GFP does not necessarily rule out raft involvement in reducing the mobility of PM-GFP. To further explore the role of lipid microdomains, we transfected cells with a lipid-anchored construct that is largely excluded from the rafts. GFP-tK was constructed by adding the C-terminal 17 residues of K-Ras to GFP. This portion of the hypervariable domain of K-Ras includes the prenylation CAAX box plus a polycationic sequence that directs the resulting chimera to anionic lipids of the cytosolic face of the plasmalemma (). As anticipated, GFP-tK was predominantly found at the cell membrane (unpublished data). In unstimulated cells, as well as in unengaged regions of the membrane of cells performing phagocytosis, GFP-tK was highly mobile (; MF = 1.13 ± 0.03; = 9 ± 1 s; = 6). Its mobility was only marginally lower at the phagocytic cup (MF = 0.90 ± 0.03; = 7.8 ± 1.4 s; = 6). The differential behavior of the various lipid-associated proteins tested suggests that individual microdomains have distinct mobility within nascent phagosomes. To further test the role of lipid microdomains, we used methyl-β-cyclodextrin (MβCD) to remove cholesterol from the membrane (). Cholesterol is essential for the formation of most lipid rafts, and its removal consistently leads to their destabilization (). Treatment of RAW cells with MβCD as described in Materials and methods resulted in sizable removal of plasmalemmal cholesterol, which could be readily visualized by staining the cells with filipin (). The total cellular content of cholesterol, determined using a cholesterol oxidase-based spectroscopic assay, was reduced by 50% after treatment with MβCD. Of note, extraction of cholesterol did not affect the mobility of PM-GFP in otherwise untreated cells. The MF (1.27 ± 0.1; = 7) and the diffusion rate ( = 23 ± 4 s) were altered only marginally by pretreatment with MβCD ( and ). Cholesterol depletion had no discernible effect on the ability of RAW cells to ingest opsonized beads, consistent with earlier findings (); note, however, that inhibitory effects of MβCD on some types of phagocytosis have also been reported (; ). More important, the immobilization of a large fraction of PM-GFP at the phagocytic cup persisted in cholesterol-depleted cells (; MF = 0.44 ± 0.02; = 7). These findings imply that normal cholesterol content is not essential for the preservation of the microdomains that experience reduced mobility at sites of phagocytosis. Further evidence that sphingolipid and cholesterol-rich microdomains do not mediate the immobilization of diacylated proteins was obtained by studying the behavior of LAT (linker for activation of T cells). This adaptor is a transmembrane protein known to associate preferentially with such microdomains. As shown in , GFP-tagged LAT localized to the PM, as reported for the native protein and for the fluorescent chimera in lymphoid cells (). Of note, the mobility of LAT at the phagocytic cup was not different from that measured elsewhere in the cell ( and ). These observations suggest that cholesterol-enriched rafts are not noticeably immobilized in the vicinity of engaged Fcγ receptors. We next investigated whether the alteration in the mobility of lipid-associated proteins during phagocytosis is a passive consequence of receptor clustering at the cup or requires active signaling. Tyrosine phosphorylation of the Fcγ receptors by Src-family kinases is one of the earliest events in the signaling cascade and is essential for progression of phagocytosis. Inhibition of Src-family kinases with inhibitors such as PP1 and PP2 precludes particle internalization (; ), yet does not prevent receptor–ligand association and formation of a well-defined phagocytic cup (Fig. S3, B and E, available at ). We were therefore able to assess the mobility of both PM-GFP and GPI-YFP at the cup of PP1-inhibited cells. The effectiveness of the kinase inhibitor was verified by its ability to prevent the PLC-mediated hydrolysis of PtdInsP (measured using a specific pleckstrin homology [PH] domain; Fig. S3, B and E), a tyrosine phosphorylation–dependent event, and by the virtually complete inhibition of particle engulfment despite the formation of stable incipient cups. As illustrated in , the mobility of the lipid-associated probe was only marginally reduced at the cup, compared with the bulk, unengaged membrane. In eight experiments using PM-GFP, the MF was 0.76 ± 0.03 in the former and 0.96 ± 0.03 in the latter. Similarly, the fraction and half-time of GPI-YFP recovery in the presence of PP1 were not significantly altered ( and ). The possible role of phosphatidylinositol 3-kinase in controlling lipid mobility was also investigated. We initially confirmed that under the conditions used, the inhibitor LY294002 impaired phosphatidylinositol 3-kinase activity, as it prevented the accumulation of 3′-phosphorylated polyphosphoinositides normally observed at the cup during the early stages of phagocytosis (Fig. S3, C and F). As reported earlier (; ), treatment with inhibitors of this kinase arrested the development of phagosomes at an intermediate stage, where cup formation is evident but sealing is impaired, particularly in the case of large beads such as those used in this study (Fig. S3, C and F). As expected, LY294002 had no discernible effect on the mobility of the exofacial marker GPI-YFP. Interestingly, the immobilization of PM-GFP normally seen in untreated cells persisted in the presence of the inhibitor (MF = 0.46), as illustrated in and quantified in . These findings imply that generation of 3′-phosphorylated inositides is not essential to retain acylated molecules in the inner aspect of the phagocytic cup. The objective of our experiments was to assess the mobility of lipid-associated proteins in the phagosome and to define its determinants. The fluorescent proteins used in this study, such as PM-GFP and GFP-tH, are suitable probes to measure the contribution of the hydrophobic moiety of lipid-anchored molecules such as diacylated Src-family kinases and small GTPases, which are critical for the onset and development of phagocytosis. The molecular weight, acyl chain composition, and membrane disposition of the probes is very similar to that of the endogenous signaling molecules, yet they greatly simplify the analysis by obviating protein–protein interactions. In addition, acylated fluorescent proteins are arguably good models to analyze the mobility of lipids in the plane of the bilayer. At first glance, it may appear that attachment of the comparatively large protein moiety to the acyl chains would greatly reduce the lateral mobility of the complex, compared with that of endogenous phospholipids. However, hydrophobic interactions within the bilayer appear to be, by far, the main impediment to the lateral displacement of lipids and lipid-anchored proteins. We calculated diffusion coefficients of ∼2.3 × 10 cm/s for labeled phospholipids and 1.5 × 10 cm/s for diacylated GFP in RAW cells. Such coefficients are nearly three orders of magnitude lower than that reported for GFP in water (9 × 10 cm/s), which is reduced only three- to fourfold when the protein is expressed in intracellular compartments, including the cytosol (; ). Therefore, attachment of a GFP moiety would be expected to contribute minimally to the mobility of the acyl chains in the plane of the membrane. The differential behavior of GPI-YFP, a probe located on the outer monolayer of the plasmalemma, and PM-GFP, an inner monolayer probe, could be attributed to a unique steric hindrance on the cytosolic face of the membrane. The actin cytoskeleton would be an obvious candidate for such a physical obstacle. Cytoskeletal proteins may restrict the motion of the GFP moiety and in extreme cases, corral it within domains that are fenced in. Several lines of evidence argue against this. First, although actin and its associated proteins do indeed accumulate at the base of the cup during the initial stages of phagocytosis, they subsequently detach (Video 1). In fact, in the case of large beads such as those used in our experiments, the density of actin below the forming phagosome drops below the levels of the unengaged membrane, where the mobility of the lipid-associated probe remains high. Moreover, lipid immobilization persisted in the region subtending the particle in cells treated with cytochalasin D to preclude actin polymerization. Moreover, in the absence of the inhibitor, an extreme situation is reached after the phagosome seals, when actin is no longer detectable on its membrane, yet the reduction in lipid mobility persisted (). Second, the mobility of GFP-tK was altered much less than that of GFP-tH or PM-GFP (). Because the size and disposition of the protein moiety of the probes with respect to the membrane are similar in all cases, fencing in by cytoskeletal elements is an unlikely explanation for the altered mobility. Still, a contribution of the actin network to the mobility of the probes cannot be entirely ruled out. If it exists, such a steric hindrance would contribute little to the mobility of free lipids but would nevertheless affect lipid-associated proteins, whether the association is covalent, as in the case of Src-family kinases, or electrostatic, as in the case of proteins bearing PH domains. We believe that immobilization of lipid microdomains in the vicinity of clustered Fcγ receptors is the most likely explanation for our observations. Lipid rafts, cholesterol-rich microdomains that also contain glycosphingolipids, are often invoked in the context of signal transduction by immunoreceptors (). Both PM-GFP and GFP-tH would be expected to partition into such rafts, and coalescence of the latter around activated receptors may have contributed to immobilization of the probes. However, our observations do not fit the conventional model of the raft on two accounts. First, the mobility of GPI-YFP, which is predicted to reside in rafts, was similar in the cup and elsewhere in the membrane. Second, extraction of 50% of the total cellular cholesterol and likely an even greater fraction of the plasmalemmal cholesterol had little effect on the immobilization of PM-GFP at the cup. Two explanations can be considered: that coalescence of lipid microdomains is not the mechanism underlying the change in lipid mobility and that unique microdomains that do not conform to the conventional sphingolipid and cholesterol-rich raft are responsible. In this regard, it is noteworthy that unlike the situation reported for immunoreceptors in lymphoid and basophilic cells (; ), extraction of cholesterol does not impair Fcγ receptor signaling, leading to phagocytosis (; ). Passive coalescence of lipid microdomains cannot explain the sensitivity of the immobilization to inhibitors of Src-family kinases, which provide the earliest signal in the phagocytic cascade. We propose that events that follow tyrosine phosphorylation contribute to the assembly of lipid microdomains. It is conceivable that the recruitment of adaptor molecules with lipid-interacting moieties facilitates the coalescence of specific lipids, thereby reducing their mobility. Two types of adaptors could fulfill this function by different mechanisms. Transmembrane molecules that preferentially associate with saturated lipids such as LAT can be recruited to activated receptor complexes (). The lipid annulus associated with LAT or similar adaptors could interact with and reduce the mobility of lipids with saturated chains in the immediate vicinity of the activated receptor complex. Of note, the mobility of LAT itself has been documented to be reduced when T cell receptors are stimulated (). However, LAT is not an essential adaptor of Fc receptors in phagocytes, which use other adaptors such as Gab2, Grb2, and CrkII. Accordingly, we found that LAT is not recruited or immobilized at sites of phagocytosis. Other, unidentified transmembrane adaptors may nevertheless cause the immobilization of selected lipids near the activated receptors, but soluble adaptors could similarly be involved. Adaptors bearing lipid binding domains would be recruited to the receptor complex by the former and would stabilize lipids in its vicinity. Several adaptors possessing PH, ENTH (epsin N-terminal homology), or VHS (Vps27, Hrs, and Stam) domains are known to exist, and some of these, such as Gab2, have been reported to associate with Fcγ receptors (). Some signaling molecules, such as Vav or PLCγ, become part of the activated receptor complex and contain lipid binding PH domains. Together, the proteins that cluster around activated receptors can cause immobilization of defined lipids. Importantly, the acyl moieties of phosphoinositides would facilitate accumulation and immobilization of other lipids with saturated chains and of proteins like diacylated Src-family kinases or GTPases. Polystyrene beads (3.1 and 8.3 μm in diameter) were obtained from Bangs Laboratories. FuGene 6 was purchased from Roche Molecular Biochemicals. PP1 was obtained from BIOMOL Research Laboratories, Inc., LY294002 from Calbiochem, and α-MEM from Wisent, Inc. Cy3-labeled secondary antibodies were obtained from Jackson ImmunoResearch Laboratories. Rhodamine-phalloidin and the Amplex red cholesterol assay kit were obtained from Invitrogen. Human IgG, MβCD, filipin, and all other reagents were obtained from Sigma-Aldrich. The headgroup-labeled lipids nitrobenzoxadiazole (NBD)-phosphatidylserine and boron dipyrromethene difluoride (BODIPY)–phosphatidylethanolamine were purchased from Avanti Polar Lipids, Inc., and Invitrogen, respectively. The synthetic medium used for fluorescence determinations consisted of 140 mM NaCl, 3 mM KCl, 10 mM glucose, 20 mM Hepes, 1 mM MgCl, and 1 mM CaCl, pH 7.4 (290 ± 5 mosM). PM-GFP encodes the 10 amino acid myristoylation/palmitoylation sequence from Lyn fused to enhanced GFP (). GFP-tH consists of the C-terminal 9 amino acids of H-Ras fused to the C terminus of GFP. GFP-tK consists of the C-terminal 17 amino acids of K-Ras fused to the C terminus of GFP. These C-terminal regions comprise the complete targeting domains of H- and K-Ras, respectively (). GPI-GFP and -YFP encode the 26 amino acid signal sequence of insulin fused to enhanced GFP or YFP, followed by the 43 amino acid GPI sequence motif of decay-accelerating factor (). Construction of the LAT-GFP plasmid was detailed in . For localization of PtdInsP and PtdInsP, we used enhanced GFP fusions of the PH domains of PLCδ or Akt, respectively, both gifts from T. Meyer (Stanford University, Stanford, CA). The macrophage RAW264.7 cell line was obtained from American Type Culture Collection. These macrophages, referred to hereafter and throughout as RAW cells, were cultured in α-MEM supplemented with 10% fetal calf serum at 37°C under a humidified 5% CO atmosphere. Cells were trypsinized and seeded onto 2.5-cm glass coverslips at ∼30% confluence. Cells were transiently transfected by lipofection using FuGene 6 or by electroporation with the Nucleofector system (Amaxa) according to the manufacturers' directions and used within 16–24 h of transfection. Polystyrene beads were opsonized with 1 mg/ml human IgG by incubation for at least 1 h at 37°C, followed by three washes with PBS. To initiate phagocytosis, opsonized beads were allowed to sediment on RAW cells grown on coverslips and bathed in synthetic medium at 37°C. RAW macrophages were grown on glass coverslips to ∼30% confluence. 50 μl of either lipid (dissolved in chloroform) were added to 9.5 ml of synthetic medium supplemented with 1 ml bovine serum albumin and mixed vigorously. Cells were overlaid with the lipid suspension and incubated at 4°C for 60 min. Cells were then washed and warmed with medium at 37°C before measuring fluorescence. To analyze the rate of recovery, we compared the fluorescence of the bleached area to that of an adjacent unbleached area of the same cell with similar fluorescence intensity. For each time point, the fluorescence of the bleached area was normalized to that of the corresponding control (unbleached) area to correct for possible drift of the focal plane or photobleaching incurred during the low-light sampling. For reference, images of the entire field were acquired immediately before and at the end of the experiment. All FRAP measurements were performed at 37°C. Data were fit to a simple diffusion, zero flow model () using the formulawhere the fluorescence intensity () at a given time () is related to the maximal fluorescence ( ) and the half-time of maximal recovery ( ). Using this equation, recovery curves were fit by least squares using Prism 4 (GraphPad Software, Inc.). In all cases, this method provided highly significant R values that were comparable to and generally larger than those obtained using other models for single or multiple components of diffusion or flow (; ). of the recovery curves as previously described (eq. 19 in ). To label F-actin, cells were washed twice with PBS and fixed with 4% paraformaldehyde for 1 h. The cells were next permeabilized in 0.1% Triton X-100 and 100 mM glycine in PBS for 10 min and stained with a 1:400 dilution of rhodamine-phalloidin for 1 h. Where specified, actin polymerization was impaired by pretreating the cells with 2 μM cytochalasin D for 10 min at 37°C as previously described (). The inhibitor was maintained in the medium throughout the fluorescence determinations. Where indicated, the cells were incubated with 10 mM MβCD for 30 min at 37°C to remove cholesterol. To verify the effectiveness of the treatment, cells were then fixed with 4% paraformaldehyde and incubated with 25 μg/ml of filipin, a cholesterol binding fluorescent probe, in PBS containing 1 mM MgCl and CaCl for 1 h at 4°C. Cells were then washed twice with PBS, and images were acquired with excitation at 488 nm and emission at 500–550 nm. More quantitative estimates of cholesterol content were made using the Amplex red cholesterol assay kit, according to the manufacturer's instructions. To inhibit phosphatidylinositol 3-kinases or Src-family kinases, RAW cells were exposed to either 100 μM LY294002 for 30 min or 10 μM PP1 for 1 h, respectively, at 37°C before the fluorescence determinations. Inhibitors were maintained in the medium throughout the microscopy experiments. Fig. S1 provides an estimation of the time required for the phagocytic internalization of beads of different sizes by RAW macrophages. Fig. S2 shows the persistence of inner and outer leaflet membrane probes during phagocytosis. Fig. S3 shows the distribution of phospholipid probes for PtdInsP and PtdInsP in the presence of Src-family kinase and PI3K inhibitors during phagocytosis. Video 1 shows the transient remodeling of actin during the phagocytosis of an 8-μm bead. Online supplemental material is available at .

α-Herpesviruses are a subfamily of the herpesviruses containing closely related human and animal pathogens, including human herpes simplex virus 1 (HSV-1; cold sores, corneal blindness, and encephalitis) and important animal viruses such as the porcine pseudorabies virus (PRV) and bovine herpesvirus 1 (BoHV-1; respiratory symptoms, abortions, and/or neurological symptoms). Many of the disease symptoms observed after infection with α-herpesviruses are associated with their neurotropic behavior, including their ability to establish lifelong cycles of latency and reactivation in the peripheral nervous system of their host (; ). Primary replication of most α-herpesviruses occurs in epithelial cells of the upper respiratory tract. Sensory neurons of the trigeminal ganglion (TG) that innervate these epithelial cells are predominant target cells for HSV-1, PRV, and BoHV-1 (; ; ). Entrance of HSV and PRV in the axons of these sensory neurons is thought to be initiated by an interaction of the viral envelope glycoprotein D (gD) with its receptor nectin-1, followed by fusion of the viral envelope with the axolemma, which is mediated by viral proteins gB, gD, gH, and gL (; ; ; ; ; ). Fusion of the viral envelope with the axolemma is followed by retrograde transport of the capsid and a part of the associated tegument to the cell nucleus by means of microtubule-associated fast axonal transport (; ; ). After entry of the DNA into the nucleus, either a full replication cycle is initiated, leading to the formation of new virions, or a latent infection is established (). Newly produced virions, during primary infection or after reactivation, are transported in the anterograde direction along the axon, followed by virus release at the axon terminus (; , ). Recent data indicate that virus egress in axons may not be limited to the axon terminus but also seems to occur at scattered sites along the axon shaft in a manner that remains not fully understood (; ; ). Despite the obvious importance of TG neurons as predominant target cells and sites of latency/reactivation events for many α-herpesviruses, a detailed study of the interactions between α-herpesviruses and this pathogenetically important cell type has been hampered by the lack of easy-to-handle, homologous in vitro systems. We recently established such a homologous in vitro two-chamber system, based on the “Campenot” system, to study the interaction between porcine TG neurons and the porcine α-herpesvirus PRV (; ). Using this in vitro model, we report that PRV induces, via its gD envelope protein, the formation of presynaptic boutons (varicosities) along the axon shaft of infected TG neurons. Varicosities are swellings along neuronal axons where synaptic vesicles, mitochondria, and ER accumulate (). They are able to form synaptic contacts with contacting nonneuronal cells and other axons (), but they also seem to play an important role in nonsynaptic communication in the nervous system by the release of neurotransmitters directly in the extrasynaptic space (; ). We observed that nonneuronal cells aligning the axon shaft of infected TG neurons were frequently infected, and the first infected nonneuronal cells were almost invariably located in close proximity to the varicosities. This suggests that virus-induced varicosities may serve as axon exit sites for the virus to infect neighboring cells. The two-chamber system to study interactions of PRV with porcine TG neurons is mounted on a coverslip and consists of an inner chamber, in which the neuronal culture (composed of neuronal and nonneuronal cells) is seeded, and an outer chamber, and the two are separated from each other by a virus- and medium-impermeable silicon barrier (). After 2–3 wk of cultivation of trigeminal neurons in the inner chamber, axonal outgrowth through the silicon barrier into the outer chamber was detected by light microscopy. Addition of 2 × 10 plaque-forming units (PFUs) of PRV to the outer chamber resulted in exclusive infection of trigeminal neuronal cell bodies in the inner chamber, as described previously (). Surprisingly, axons of PRV-infected neurons (24 h after inoculation) showed a massive amount of bouton-like axonal swellings, which were rarely detected on axons of noninfected cell bodies (). Double immunofluorescent stainings using a neuronal cell marker (neurofilament) and a marker for synaptic vesicles (synaptophysin; ) confirmed that the swellings are presynaptic boutons, also called varicosities. Labeling of the PRV-induced varicosities with FM1-43, a fluorescent marker for firing neurons, indicated that the synaptic transmission at the varicosities is intact (). Varicosities were formed between 3 and 6 h after inoculation and were observed in >70% of the axons at both early (6 h after inoculation) and late (24 h after inoculation) stages of infection. In mock-treated cultures, only 12% of the axons showed varicosities (). Varicosities could be induced by UV-inactivated PRV (), indicating that the trigger for varicosity formation occurs early in infection, either during virus attachment or entry, before the onset of viral protein production. To assess whether induction of varicosities involves the virus entry–essential viral envelope proteins gB or gD, isogenic stocks of phenotypically and genotypically gBnull- and gDnull-PRV were prepared. Virus stocks were obtained by inoculating phenotypically complemented gBnull- and gDnull-PRV stocks on noncomplementing swine testicle cells and harvesting the progeny virus from the culture supernatant. SDS-PAGE and Western blotting confirmed that gB and gD were absent from the gBnull- and gDnull-PRV stocks, respectively (). Relative virus particle numbers in the gBnull- and gDnull-PRV stocks were determined as described previously (; ), by comparing the amount of gB and gD in the deleted stocks and in serial dilutions of wild-type (WT)–PRV stock by optical densitometry after Western blotting, as described in Materials and methods. Virus quantities of gBnull- and gDnull-PRV corresponding to the amount of particles present in 2 × 10 PFUs of WT-PRV were added to the outer chamber of the two-chamber system, and the percentages of axons displaying varicosities were determined at 24 h after inoculation. Varicosity induction by the gBnull virus was comparable to that of WT-PRV (68 vs. 72%, respectively; ). However, when axons were inoculated with the gDnull virus, the percentage of axons with varicosities was comparable to that of the mock-treated control (14 vs. 12%, respectively). As expected, neither null virus induced infection of the TG neurons, as neither was able to enter the cells (unpublished data). To exclude the possibility that the lack of synapse induction by the gDnull virus was due to an insufficient amount of virus added, the experiment was repeated with 10-fold more gDnull virus, and similar results were obtained (unpublished data). In conclusion, a gDnull-PRV strain is unable to induce the formation of varicosities, suggesting that virion gD triggers this process. To address whether gD alone could induce varicosities, various concentrations of a truncated, soluble form of PRV-gD (PRV-gDt; ) were added to the outer chamber of the in vitro model and incubated with the axons for 24 h, followed by neurofilament staining and quantification of the number of axons that showed varicosities. Incubation resulted in a dose-dependent increase in the number of axons carrying varicosities (). The addition of 5 μg PRV-gDt/ml or more resulted in 60–70% axons with varicosities (), which is comparable to the percentages observed by the addition of WT-PRV. Entry of α-herpesviruses, like HSV-1 or PRV, into sensory neurons is believed to be mediated by interaction of viral envelope protein gD with nectin-1 (; ; ; ). Therefore, we hypothesized that gD-mediated induction of varicosities results from an interaction with nectin-1. To test this hypothesis, we determined whether addition of various concentrations of an antibody that binds the ectodomain of nectin-1 (CK6; Krummenacher et al., 2000) was also able to trigger varicosity formation. As with soluble gD, a 24-h incubation period with monoclonal anti–nectin-1 antibody in the outer chamber again resulted in a clear dose-dependent response. For antibody CK6, 100 μg/ml caused varicosities to form on 70% of the axons (), again similar to the percentage observed by addition of WT-PRV. The addition of 100 μg/ml of CK41, a monoclonal antibody directed against another epitope on the ectodomain of nectin-1, also resulted in induction of varicosity formation (unpublished data). The addition of 100 μg/ml of an isotype-matched control IgG1 antibody (13D12) did not result in an increase in the percentage of axons with varicosities compared with mock-treated two-chamber systems. In conclusion, addition of recombinant gD or anti–nectin-1 antibodies to the axons of TG neurons is sufficient to induce the formation of varicosities. Nectin-1 has been shown to signal via small Rho GTPase signaling pathways in MDCK and L cells (, ). In addition, Cdc42 Rho GTPase and MAPK signaling pathways have been suggested to be involved in varicosity formation (; ; ). Therefore, to determine which signaling pathways may be involved in PRV-induced varicosity formation, the effect of inhibitors of different signaling pathways (Rho GTPase and MAPK signaling pathways) were tested for their effect on PRV-induced varicosity formation. A broad-range inhibitor of small Rho GTPase signaling (50 ng/ml toxin B) as well as specific inhibitors of Rho (30 μM Y27632), Rac1 (100 μM Rac1 inhibitor), or Cdc42 (2 μM secramine A) signaling were used. The role of MAPK signaling was examined by using inhibitors for extracellular signal–regulated kinase (ERK) signaling (10 μM U0126), JNK signaling (20 μM JNK inhibitor II), and p38 signaling (20 μM SB203580). General inhibition of small Rho GTPase signaling as well as specific inhibition of Cdc42 signaling suppressed varicosity formation to a level that was not significantly different from mock-infected cultures, whereas specific inhibition of Rho or Rac1 signaling had no obvious effect (). Inhibition of p38 MAPK signaling also resulted in a strong reduction of varicosity formation, in contrast to inhibition of ERK or JNK MAPK signaling (). These data suggest that PRV activates varicosity formation via signaling pathways dependent on Rho GTPase (in particular Cdc42) and MAPK (in particular p38). The inner chamber of the TG cultures does not consist solely of TG neurons but also contains many nonneuronal cells, which form a monolayer in the inner chamber, in which the TG neurons are dispersed. When the inner chamber of TG neuronal cultures was stained for PRV antigens at 24 h (or later) after inoculation with WT-PRV in the outer chamber, viral antigen–positive nonneuronal cells were observed aligning the axons of PRV-infected TG neurons. Interestingly, single-infected nonneuronal cells were almost invariably (88%) juxtaposed to varicosities (). Spread of PRV from varicosities to neighboring nonneuronal cells could not be blocked by neutralizing antibodies (unpublished data), indicating that it occurs via direct cell–cell spread. These data indicate that egress of infectious virus along the axon shaft occurs specifically at varicosities. The neurotropic behavior of α-herpesviruses is of crucial importance for the pathogenicity of these viruses, allowing them to establish lifelong latency and cause central nervous disorders, encephalitis, and recurrent disease. Neurons of the TG represent crucial target neurons for many α-herpesviruses, including HSV-1, PRV, and BoHV-1. The exact cell biology underlying the interactions of α-herpesviruses with neurons, especially TG neurons, remains far from fully understood. Here, we used an in vitro model to study the interaction of an α-herpesvirus (PRV) with TG neurons of its corresponding host (the pig). We report a novel aspect of the cell biology of α-herpesvirus interaction with TG neurons. We found that the interaction of PRV with axons of porcine TG neurons triggers the formation of synaptic boutons (varicosities) along the axons of these neurons. To our knowledge, this is the first paper reporting that a virus infection stimulates the formation of varicosities. In addition, we show that the viral envelope protein gD is responsible for the induction of varicosities, probably via an interaction with the entry receptor nectin-1 or nectin-like molecules; that Cdc42 Rho GTPase and p38 MAPK signaling pathways are involved; and that virus egress along the axon shaft of infected TG neurons occurs frequently and specifically at these varicosities. These observations open the intriguing possibility that the virus has evolved a strategy to facilitate spread of progeny virus from TG neurons by inducing varicosities at the time of virus attachment and entry in axons. An important question is how exactly PRV induces the formation of varicosities in porcine TG neurons. We found that addition of UV-inactivated PRV (which is able to adhere to and penetrate cells but does not start up viral gene expression) or a strain of PRV that lacks the envelope protein gB (which is able to adhere to but not penetrate cells) to the axons of TG neurons still resulted in varicosity formation. This shows that varicosity formation is triggered during virus attachment to the axons and does not require infection. Interestingly, a PRV strain that lacks gD in its envelope (which is also able to adhere to but not penetrate cells) did not induce varicosities, demonstrating an involvement of gD in this process. Moreover, addition of recombinant soluble gD to the axons of TG neurons was sufficient to trigger varicosity formation. Together, these data indicate that interaction of PRV envelope protein gD with axons of TG neurons during virus attachment provides the trigger necessary for subsequent formation of varicosities. How does the interaction of gD with the surface of axons lead to varicosity formation? Three classes of receptors for α-herpesvirus gD proteins have been described to date: one belongs to the tumor necrosis factor receptor family (i.e., herpes virus entry mediator), another class belongs to the immunoglobulin superfamily (including nectin-1 and -2), and another type of receptor consists of modified heparan sulfate (; ; ; ; ; ). Nectin-1 has been suggested to serve as the gD receptor on sensory neurons (like TG neurons) for both HSV and PRV (Haarr et al., 2001; ; ; ). Nectins are cell adhesion molecules that play important roles in the formation of many types of cell–cell junctions, including synapses (). We found that, for both of two different nectin-1–specific antibodies, addition to the axons of TG neurons (as a surrogate ligand for nectin-1 instead of gD) resulted in the formation of varicosities. This suggests that the interaction between gD and nectin-1 on the axons of TG neurons provides the trigger that ultimately culminates in the formation of varicosities. This would be consistent with observations made by , who showed that stimulation of nectin-1 in mouse hippocampus neurons resulted in a substantial increase in the number of synaptophysin-positive varicosities. However, at present, we cannot rule out the possibility that other nectins or nectin-like molecules, in addition to or instead of nectin-1 might be relevant to the gD-mediated induction of varicosities. Porcine nectin-1 is thus far the only porcine entry receptor that has been reported for PRV (), but different human forms of the nectin family, like nectin-1, nectin-2, and necl-5 (poliovirus receptor), have been reported to be gD-binding entry receptors for PRV (, ). In addition to a crucial role of the interaction between gD and nectin-1 and/or other members of the nectin family for the PRV-induced formation of varicosities, we also found that Cdc42 Rho GTPase as well as p38 MAPK signaling pathways are involved. Inhibition of these signaling pathways strongly suppressed PRV-induced varicosity formation. These data are consistent with recent indications that the Cdc42 (but not Rho or Rac) signaling pathway is involved in serotonin-induced varicosity formation in sensory neurons () and that MAPK signaling pathways, such as p38, may play important roles during development of varicosities (; ). Further in line with our current observations, it has been reported that nectin-1 may signal through Rho GTPase pathways, such as Cdc42 and Rac, in MDCK and L cells (, ). Our current findings that an α-herpesvirus, via its envelope gD, stimulates formation of synaptophysin-positive varicosities and that this process involves Cdc42 and p38 MAPK signaling may, therefore, constitute a valuable tool to further dissect the underlying molecular mechanisms of varicosity formation and the role of nectin-1 and other nectin-like molecules herein. α-Herpesviruses have evolved different strategies to modulate the host cell to facilitate virus spread, such as gE-mediated targeting of virus particles to cell junctions for spread in polarized cells (; ) and intercellular virus spread via US3-induced actin rearrangements and cell projections (). Although future in vivo experiments will be required to fully delineate the consequences of our current findings for α-herpesvirus pathogenesis, we propose that induction of axonal varicosities by PRV early in infection of TG neurons may be part of the viral strategy to later promote efficient spread. Varicosity-mediated spread may lead to egress of α-herpesviruses along the axon shaft but may also enhance spread of the virus to mucosal surfaces, a crucial step in α-herpesvirus pathogenesis. Trigeminal sensory nerve fibers may extend beyond the basement membrane to nearly reach the epithelial surface (). These intraepithelial terminal regions of the axon have penetrated the basement membrane and lost their myelin sheath and are prone to varicosity formation. Strings of varicosities have been reported to occur in these intraepithelial terminal areas of the TG axons in vivo (, ). The exact function of α-herpesvirus egress along the entire length of axon shafts is not understood, but PRV egress along axons shafts has clearly been demonstrated in vivo in rats and was found to occur via direct cell–cell spread, which is in line with our current findings (). Interestingly, this axonal egress in vivo as well as in vitro was found to occur at scattered sites (; ). Given our current finding that axonal egress almost exclusively occurs at varicosities, we hypothesize that these scattered sites may correspond to varicosities. In further support of this, a recent paper suggests that axonal egress of HSV-1 occurs via varicosities in human fetal dorsal root ganglia neurons in vitro (). Transport of progeny herpesvirus particles in the axon is believed to either occur via secretory vesicles containing fully matured virions or as subvirion particles, where capsids and envelope proteins are transported separately and only assemble into mature virions at synapses (). In our opinion, both scenarios can lead to specific axonal egress at varicosities, as these structures have been shown to accumulate secretory vesicles and may also constitute functional synapses (). Virus-induced varicosities endured for at least several days (>72 h; unpublished data). It remains to be determined whether and when they disappear after the initial infection. Although speculative at this time, it is possible that the virus is able to reinduce varicosities over and over again when needed during latency/reactivation cycles. Reactivations can lead to production and release of infectious virus, which in turn may attach to and enter new TG neurons, thereby inducing varicosity formation and further promoting virus spread. It is possible that virus-induced formation of varicosities has consequences beyond virus spread. At least a subpopulation of synaptophysin-positive varicosities function as sites for neurotransmitter release along the axon of different types of neurons (; ; ). In line with this, we found that the virus-induced varicosities stain positive for FM1-43, characteristic of intact synaptic transmission. Synaptic transmission was also intact in varicosities induced by UV-inactivated PRV or recombinant gD of PRV (unpublished data). In this context, it is interesting to note that α-herpesvirus infections have been associated with hyperexcitability of neurons, possibly involved in acquired epilepsy after HSV encephalitis (). Future investigations will be designed to further unravel whether the PRV-induced formation of synaptically effective varicosities lead to changes in excitability of neurons. In conclusion, we have found that entry of an α-herpesvirus in neurons of the TG of its natural host is associated with the induction of synaptic varicosities along the axons. Virus-induced varicosity induction depends on viral envelope protein gD and on Cdc42 Rho GTPase and p38 MAPK signaling pathways, and the virus uses the varicosities as axonal egress sites to spread to neighboring cells. WT-PRV strain Becker, WT-PRV strain Kaplan, and isogenic deletion mutants gBnull and gDnull were used (; ; ; ). Stocks of phenotypically complemented gBnull and gDnull viruses were grown on gB- and gD-complementing cell lines. Stocks of phenotypically and genotypically gBnull and gDnull viruses were produced by a single round of infection of phenotypically complemented gBnull and gDnull viruses on noncomplementing swine testicle cells and harvesting the progeny virus from the supernatant. Monoclonal mouse anti-gB (1C11) and anti-gD (13D12) antibodies and polyclonal porcine FITC-labeled anti-PRV antibodies were produced as previously described (). The monoclonal neuronal markers mouse anti–neurofilament 68 and rabbit anti–neurofilament 200 and the monoclonal synapse marker mouse anti-synaptophysin were purchased from Sigma-Aldrich. FITC- and Texas red–labeled goat anti–mouse antibodies and Texas red–labeled goat anti–rabbit antibodies were obtained from Invitrogen. Biotinylated sheep anti–mouse IgG and a streptavidin-biotinylated horseradish peroxidase complex were purchased from GE Healthcare. Inhibitors toxin B, Y27632, Rac1 inhibitor (NSC23766), U0126, SB203580, and JNK inhibitor II were obtained from Calbiochem. Secramine A was used as a specific inhibitor for Cdc42, as described previously (). PRV-gDt () and two monoclonal mouse anti–human antibodies directed against different epitopes on the ectodomain of nectin-1 (CK6 and CK41; ) were used. The number of virus particles in the gBnull and gDnull stock was estimated by optical densitometry, as described previously (; ). Equal volumes of a serial dilution of a WT-PRV stock with a known titer and of stocks of the genotypically and phenotypically gBnull- and gDnull-PRV strains were subjected to SDS-PAGE under nonreducing conditions and Western blotting, followed by detection of gB or gD using monoclonal gB- and gD-specific antibodies, biotinylated secondary sheep anti–mouse antibodies, streptavidin-biotinylated horseradish peroxidase complex, and 3,3′-diaminobenzidine (Sigma-Aldrich) for gB or enhanced chemiluminescence (ECL Western blotting analysis system; GE Healthcare) for gD. All antibodies were diluted in PBS with 0.1% Tween-20 (Sigma-Aldrich), and blots were washed three times in PBS with 0.1% Tween-20 in between different antibody incubations and after the final antibody incubation. Relative amounts of gB and gD in the gDnull and gBnull stock, respectively, were compared with the amount of gB and gD present in the WT stocks using Quantity One 1-D analysis software (Bio-Rad Laboratories). After being washed in PBS, neuronal cultures in the inner and outer chamber of the two-chamber model were fixed in 100% methanol for 20 min at −20°C, except for cultures that were labeled for synaptophysin, which were fixed using 4% paraformaldehyde in PBS for 10 min and subsequently permeabilized in 0.2% Triton X-100 (Sigma-Aldrich) in PBS for 2 min. All antibodies were diluted in PBS to 1:100. Cells were incubated with each antibody for 1 h at 37°C and were washed two times in PBS in between all incubation steps and after the last incubation step. When necessary, nuclei were stained using 10 μg/ml Hoechst 33342 (Invitrogen) for 10 min before the final washing steps. Porcine TG neurons were obtained as described previously () and seeded in an in vitro model based on the “Campenot” system (), which allows simulation of the in vivo route of neuronal infection (). In brief, porcine TG were excised from 4–6-wk-old piglets and dissociated by enzymatic digestion with 0.2% collagenase A (Roche). The harvested cells were resuspended in culture medium (basic culture medium without glutamine) and supplemented with 30 ng/ml nerve growth factor (Sigma-Aldrich) and seeded in the inner chamber of an in vitro two-chamber model. The two-chamber system consists of a polyallomer tube that is fixed with silicon grease on a collagen-coated cover glass inserted in a 6-well plate (Nunc). The inside of the tube forms the inner chamber, and the outside forms the outer chamber. The silicon barrier prevents diffusion of medium or virus from one chamber to the other (). 1 d after seeding, cultures were washed with RPMI (Invitrogen) to remove nonadherent cells and, from then on, culture medium was changed three times a week. After 2–3 wk of cultivation, when clear axon growth could be observed in the outer chamber, the outer chamber was inoculated with 2 × 10 PFUs of WT-PRV or with an equivalent number of UV-inactivated, gBnull- or gDnull-PRV particles. In some experiments, the outer chamber was incubated with soluble PRV-gD (ranging from 0.001 to 10 μg/ml), with antibodies directed against nectin-1 (CK6, ranging from 0.1 to 100 μg/ml, or CK41, 100 μg/ml) or with an isotype-matched (IgG1) control antibody directed to the viral envelope gD (100 μg/ml 13D12; ; ). The firing capacity of the induced varicosities was determined by loading the neurons with FM1-43 (Invitrogen), basically as described before (). After 2–3 wk in culture, the outer chamber was treated for 24 h with 2 × 10 PFUs of WT-PRV, an equivalent number of UV-inactivated PRV particles, or 10 μg/ml soluble PRV-gD. Then, the inner chamber was washed with Hanks' balanced salt solution supplemented with 100 mM KCl and 1.5 mM CaCl for 1 min. The neurons were incubated with culture medium containing 100 mM KCl and 20 μM FM1-43 for 10 min. After being washed with Hanks' balanced salt solution for 15 min, cultures were mounted on coverslips without fixation and examined by confocal microscopy. To examine the effect of different inhibitors on varicosity formation, both the inner and outer chamber of the two-chamber system were pretreated with culture medium supplemented with the respective inhibitor for 2 h. Afterward, the outer chamber was incubated with UV-inactivated PRV particles equivalent to 2 × 10 PFUs of WT-PRV in the presence of the inhibitor. After an incubation period of 16 h, the two-chamber system was methanol fixed and stained, and the percentage of axons showing varicosities was determined as described (see Cultivation, inoculation, and analysis…). Single-infected cells were scored as juxtaposed to a varicosity when the signal of the viral antigens (FITC signal) contacted the varicosity (Texas red signal), as seen in two-chamber models fixed at 24 h after inoculation with 2 × 10 PFUs of WT-PRV and stained with polyclonal FITC-labeled anti-PRV antibodies and the neuronal cell marker anti–neurofilament 68 (Texas red). 50 single-infected cells were analyzed, and data shown represent the mean ± SEM of four assays. Stainings were analyzed on a laser-scanning spectrum confocal system (TCS SP2; Leica Microsystems GmbH) linked to a microscope (DM IRBE; Leica Microsystems GmbH). Images were taken using a 63× oil objective (NA 1.40–0.60) at room temperature and using confocal acquisition software (Leica Microsystems GmbH). Adjustments of brightness and contrast were applied to the entire images and were done using Photoshop (Adobe). The mean percentages of axons displaying varicosities after the different treatments were compared with an analysis of variance and a least significant difference post hoc test for a multiple comparison of means (α = 0.05).

Cell proliferation in the multicellular organism is tightly controlled through the cooperative efforts of numerous microenvironmental cues, including soluble growth factors and adhesion to the ECM. One potential point of integration between growth factor and adhesive signaling is in the focal adhesion (). Focal adhesions are structures that arise during the binding and clustering of integrins and serve to physically link the actin cytoskeleton to the underlying ECM. Because they also contain numerous growth factor receptors and signaling proteins, focal adhesions have been proposed as localized sites where growth factor and adhesion signaling converge (for reviews see ; ). FAK is a key effector in focal adhesion signaling and a potential integrator of integrin- and growth factor–mediated proliferative signaling. It is rapidly phosphorylated after integrin ligation (; ; ), which stimulates its kinase activity (; ) and triggers the activation of signaling pathways involved in modulating focal adhesions and their surrounding cytoskeletal structures (; ). Given its central role in adhesion signaling, it is not surprising that numerous studies have demonstrated a regulatory role for FAK in cell cycle progression (; ; ). Such studies have shown that FAK overexpression drives G/S phase cell cycle progression, whereas dominant–negative FAK mutants, such as FRNK, or anti-FAK antibodies block the cell cycle at the G/S phase boundary (; ; ; ). Mechanistically, FAK overexpression appears to enhance the transcriptional activation of cyclin D1 (). FAK appears to regulate the G cell cycle machinery through numerous signaling pathways. In endothelial cells (EC), FAK is required for sustained ERK activity downstream of VEGF stimulation (). Additionally, FAK regulates the activity of the Rho GTPase RhoA, which is also required for sustained ERK signaling (; ; ). Importantly, although FAK signaling clearly modulates cell cycle progression, it does not appear to be required, as FAK−/− cells and cells treated with FAK RNAi still proliferate (; ). Thus, the role of FAK in adhesion-regulated proliferation is likely to be multifaceted, and may depend on the adhesive context in which FAK signaling occurs. To conceptually dissect how FAK might regulate adhesion-dependent proliferation, it is necessary to define adhesion more precisely. Although cell adhesion is initiated by integrin binding to ECM ligands, it involves numerous other processes, such as integrin clustering, focal adhesion maturation, and cell spreading and flattening against the substrate, each of which appears to be involved in regulating proliferation. Integrin ligation and clustering, although necessary for the proliferation of adherent cells, is not sufficient to support cell cycle progression. Proliferation also requires that the ECM allows cells to physically spread against the substrate; cells that are prevented from spreading or flattening against the ECM are growth arrested (). Interestingly, these changes in cell spreading appear to be required for RhoA-mediated cytoskeletal tension and focal adhesions to develop (; ), and inhibiting cytoskeletal tension and focal adhesion formation appear to abolish proliferation in spread cells (; ). Thus, changes in integrin ligation, cell spreading, cytoskeletal tension, and focal adhesion formation are clearly interdependent, and have all been implicated in growth regulation. Because of the prominent role of FAK in multiple aspects of the adhesive processes, including focal adhesion development (), spreading (; ), and mechanical tension (), FAK may serve as a critical point of integration for transducing each of these adhesive processes into a coordinated biological response, such as proliferation. However, despite the involvement of FAK in the various aspects of adhesion, how FAK functions to regulate proliferation under different adhesive contexts is ill defined. By examining the proliferative effects of modulating FAK in different adhesive contexts, we have found that FAK plays a dual role in regulating growth. In contexts of high adhesion, FAK activity and proliferation are high. In low ECM ligand or low cell-spreading contexts, normally growth-arrested cells can be induced to proliferate by activating FAK. Surprisingly, the growth inhibition in these low adhesive states is mediated by inactive FAK, as loss of FAK in either FAK−/− cells or FRNK-expressing cells dysregulated adhesion-dependent growth control. Full-length, kinase-dead FAK-Y397F, in contrast to FRNK, rescued adhesion-dependent growth regulation, suggesting the possibility that the N terminus of FAK may mediate the growth inhibitory function. The uncontrolled growth after loss of FAK was mediated through an increase in RhoA signaling and cytoskeletal tension. Thus, FAK appears to transduce both high adhesive signals, to stimulate proliferation, and low adhesive signals, to arrest growth. This dual nature highlights FAK as a central control point for growth regulation, and underscores its critical role in integrating the multiple adhesive, mechanical, and biochemical functions of focal adhesions. To begin to explore the role of FAK in regulating proliferation, we first established the dependence of bovine pulmonary artery EC proliferation on growth factors and adhesion. Cells were G synchronized at confluence, replated under various growth factor or adhesive conditions, and assayed for proliferation by tracking BrdU incorporation as a marker of S phase entry. As expected, when ECs were exposed to low serum (0.01%) or grown on surfaces coated with a low density of fibronectin (0.1 μg/ml), cell proliferation was inhibited compared with cells grown in high serum (5%) or on surfaces coated with a high density of fibronectin (25 μg/ml; ). To examine whether the serum or fibronectin concentrations affected focal adhesion formation, we analyzed vinculin distribution by immunofluorescence. Whereas cells grown in high serum formed large, well-defined focal adhesions (), cells cultured in low serum showed reduced focal adhesion number and area (). Fibronectin concentration affected focal adhesion formation to an even greater extent (). This correlation between proliferation and focal adhesion area in both serum- and adhesion-regulated growth suggested the possibility that FAK might be involved in both growth factor– and adhesion-mediated proliferation. To begin to explore this possibility, we examined whether overexpression of FAK could overcome the proliferation block caused by either low serum or low-density fibronectin. G-synchronized ECs were transduced with a recombinant adenovirus containing wild-type FAK, resulting in FAK overexpression and constitutive autophosphorylation. FAK overexpression did not rescue the growth arrest caused by low serum and did not affect proliferation induced by high serum (). In contrast, cells plated on low-density fibronectin dramatically increased proliferation upon FAK overexpression (). These findings suggest the possibility that FAK mediates the proliferative signals initiated by adhesion, but not by growth factors. Cell adhesion involves many different steps, including integrin ligation and clustering and cell spreading and flattening against the substrate (). Decreasing fibronectin density not only decreased integrin clustering and focal adhesion formation, but also impaired cell spreading (). Because changes in cell spreading can directly regulate cell proliferation, despite the presence of excess extracellular matrix, we examined whether FAK is also involved in the regulation of cell proliferation by changes in cell shape. To specifically modulate cell shape without altering fibronectin density and integrin clustering, we used microcontact printing to generate micrometer-scale islands coated with a high density of fibronectin, separated by nonadhesive regions such that the size of the islands dictated the degree of cell spreading. ECs seeded onto small, square islands (625 μm) remained relatively unspread, whereas ECs seeded onto uniformly coated surfaces spread to an average of 2,000 μm (). Measurement of S phase entry under these conditions demonstrated that the unspread cells could not proliferate (). Substantially fewer and smaller focal adhesions formed in the growth-arrested unspread cells compared with spread controls (), suggesting the possibility that alterations in focal adhesion architecture and/or signaling may also underlie proliferative regulation by cell spreading. To examine whether cell spreading specifically affected FAK activity, we measured FAK phosphorylation at tyrosine 397 in these cells. At early time points after replating, attachment, spreading, and FAK phosphorylation at Y397 was similar between spread and unspread cells (). At later time points, unspread cells showed progressively lower FAK phosphorylation while spread cells transiently increased FAK activation (). These data suggested the possibility that FAK signaling may be fundamentally different in spread versus unspread cells and that FAK may be directly involved in the proliferation response of cells to changes in cell spreading. To explore this possibility, cells were transduced with wild-type FAK adenovirus and cultured on the micropatterned substrates. FAK overexpression increased proliferation as compared with a GFP control (). Because FAK overexpression appears to rescue proliferation that was inhibited both by low-density fibronectin and by reduced cell spreading, but not by low serum, FAK appears to be specifically involved in proliferative signals mediated by adhesive cues. In physiologic settings, however, the primary mode of adhesion-mediated arrest in ECs is mediated by confluence of the monolayer, not through changes in ligand density or cell area. To test whether FAK signaling is involved in confluence-induced arrest, we expressed FAK in monolayer cultures. FAK overexpression increased proliferation in cells arrested by traditional contact inhibition (). Together these studies suggest that FAK may be involved in several of the means by which adhesion regulates proliferation. The stimulation of proliferation by FAK overexpression suggests at least two possible models for adhesion-regulated proliferation. The first, and predominantly accepted, model is that FAK activity triggered by adhesion stimulates proliferation (; ). A second, equally plausible model is that inactive FAK in cells with limited adhesion or spreading inhibits proliferation. To begin to address these possibilities, we examined the proliferative response of cells completely lacking FAK. G-synchronized FAK−/− mouse embryo fibroblasts were seeded onto micropatterned islands of various sizes or onto unpatterned surfaces, where the cells ranged in size from 625 μm to fully spread (∼2,500 μm; ). Well-spread FAK−/− cells proliferated maximally, as expected (). Surprisingly, unspread FAK−/− cells also proliferated (), indicating that loss of FAK may have eliminated adhesion-dependent proliferative control mechanisms. To address this, we examined the effect of reexpressing FAK on proliferation. FAK reexpression to endogenous levels, which resulted in the rescue of the spreading-dependent FAK autophosphorylation seen in ECs (), inhibited proliferation only in unspread cells, rescued normal adhesion-dependent growth control, and confirmed that the loss of growth control was specific to loss of FAK (). The constitutive proliferation in FAK−/− cells suggests that one important and previously undescribed function of FAK is to limit proliferation in low adhesive conditions. However, although the micropatterned substrates provide a precise quantitative method to control adhesion, fibroblasts are typically adhesion-regulated in a 3D microenvironment. In this context, we cultured the FAK−/− and FAK-reexpressing fibroblasts in 3D collagen gels, where cell proliferation is often suppressed. Consistent with the micropatterning studies, FAK−/− cells continued to proliferate at higher levels in the collagen gel, whereas FAK reexpression rescued growth suppression (). As with ECs, highly overexpressing FAK to severalfold above endogenous levels in the FAK-reexpressing fibroblasts increased proliferation in unspread conditions (unpublished data). Thus, it appears that a delicate balance of FAK expression is needed for proliferative control. Because the FAK−/− and FAK-reexpressing cell lines are immortalized, and known compensatory changes in signaling pathways might have affected our interpretation of the proliferative effect, we next examined whether the same inhibitory role of FAK in proliferation might operate in normal nonimmortalized cells. To address this question, we generated recombinant adenoviruses to express the well-characterized dominant–negative FAK construct FRNK; consisting of amino acids 668–1,053 of wild-type FAK (), as well as a shorter C-terminal construct of FAK containing only the focal adhesion–targeting (FAT) domain (amino acids 919–1,053; ). We also generated an autophosphorylation-defective FAK mutant (FAK-Y397F) in adenovirus (). Infecting cells with the FAK adenovirus causes overexpression of FAK that is highly phosphorylated (), whereas expression of FRNK, FAT, and FAK-Y397F down-regulates endogenous FAK phosphorylation (). Previous studies have shown that FRNK and FAT displace endogenous FAK from adhesions (). We have confirmed these findings in our system. Cells expressing GFP, FRNK, FAK, and FAK-Y397F were fractionated into Triton X-100–soluble and –insoluble pools and blotted for FAK. FRNK decreased total FAK in the insoluble pool and phosphorylated FAK to nearly undetectable levels (). Similarly, FRNK expression also decreased the amount of total FAK (and phosphorylated FAK) that coimmunoprecipitated with paxillin (unpublished data). Because FRNK, FAT, and FAK-Y397F all contain the C-terminal FAT region, lack kinase activity, compete to displace endogenous full-length FAK from the focal adhesion, and thereby decrease endogenous FAK phosphorylation, we postulated that expression of these dominant–negative mutants might have the same proliferative effects as seen in the FAK−/− cells. To examine this possibility, ECs were transduced with recombinant adenoviruses to express FRNK, FAT, or FAK-Y397F, cultured on small islands of fibronectin, and assayed for proliferation by BrdU incorporation. As compared with GFP and FAK, as negative and positive controls, respectively, FRNK increased proliferation (). Expressing the FAT construct also relieved the proliferation arrest induced by restricted adhesion. Interestingly, FAK-Y397F did not induce cell proliferation. FAK or FRNK expression also released cells from growth arrest in monolayer cultures, but did not rescue proliferation in cells placed in suspension (unpublished data). Although the various FAK constructs increased proliferation relative to a GFP control in conditions of low adhesion, cell proliferation in a highly adhesive environment was not dramatically affected by expression of the FAK constructs (). Although the stimulatory effects of wild-type FAK expression on proliferation is consistent with previous studies (; ), the loss of adhesion-dependent proliferative control in FAK−/−, FRNK-, or FAT-expressing cells suggests that, in addition, inactive FAK might function to actively inhibit proliferation. FRNK and FAT may relieve this inhibition by displacing inactive FAK from the adhesion, whereas FAK phosphorylation might do so via a different mechanism. In support of this hypothesis, overexpressing the inactive FAK-Y397F in FAK−/− cells, like wild-type FAK, rescued adhesion-mediated growth control (). Together, these results uncover a previously undescribed function of FAK as a negative growth regulator, and, in particular, support a model whereby inactive FAK within adhesions inhibits proliferation. As an initial characterization of the proliferative mechanisms induced by FAK or FRNK, we examined the role of downstream MAPK and Src signaling pathways. Although most extracellular signals regulate proliferation through the regulation of MAPK-dependent signals in the G phase of the cell cycle, others have been reported to occur at different levels (). Because FAK is known to have a very close association with the nonreceptor tyrosine kinase Src, which is another important proliferative signaling protein, we also examined whether FAK- or FRNK-induced proliferation were Src dependent. G-synchronized cells were transduced with adenoviruses to express FAK, FRNK, or GFP, seeded onto 625-μm islands of fibronectin, and treated with 10 μM of the MEK inhibitor UO126, 25 μM JNK inhibitor I, 1 μM of the p38 inhibitor SB203580, or 1 μM of the Src inhibitor PP2. Although inhibiting MEK or JNK activity completely blocked FAK- and FRNK-induced proliferation, the p38 inhibitor had no effect ( and not depicted). Interestingly, FAK- and FRNK-expressing cells responded differently to PP2 treatment. The FAK-mediated increase in cell proliferation was blocked by inhibiting Src, but FRNK-mediated proliferation was not (). These findings suggest a divergence of signaling pathways between the proliferative effects mediated by FAK activation and those mediated by loss of FAK. Because the dysregulation of adhesion-dependent growth control by FAK down-regulation has not been previously described, we chose to further investigate the molecular mechanisms underlying this process. Our initial studies indicated that focal adhesions are significantly larger in conditions that promoted proliferation than in those that arrested growth. Therefore, we explored whether the size of focal adhesions in spread and unspread cells was also affected by the expression of FAK, FRNK, FAT, and FAK-Y397F. FRNK and FAT expression both dramatically increased focal adhesion area in unspread cells, but not in well-spread cells (), mirroring their effects on proliferation. FAK and the Y397F mutant increased focal adhesion size, but to a lesser extent. Focal adhesion size has been shown to depend on RhoA signaling (; ), suggesting that changes in FAK signaling may modulate RhoA activity. To test this possibility, we examined RhoA activity in FRNK-, FAK-, or FAK-Y397F–expressing ECs. Cells were transduced with recombinant adenoviruses, replated onto 625-μm square patterns or onto surfaces uniformly coated with fibronectin, and lysed 6 h after replating. Using the RhoA pull-down assay to measure GTP-bound RhoA, we found that FRNK expression increased RhoA activity compared with GFP-expressing control cells both in spread and unspread conditions, whereas FAK or FAK-Y397F expression had little to no effect (). Likewise, the FAK−/− cells showed higher RhoA activity than FAK-reexpressing cells (). To address whether RhoA was directly involved in the dysregulation of proliferative control induced by loss of FAK signaling, we examined the effects of inhibiting the RhoA effector ROCK in FRNK-expressing cells. ROCK inhibition with 50 μM Y-27632 blocked the FRNK-induced increase in proliferation in unspread cells (). This effect was specific to the release of growth inhibition by FRNK, as Y-27632 treatment did not inhibit proliferation rates in well-spread cells (). Similarly, FAK−/− cells treated with Y-27632 also regained adhesion-dependent growth control. That is, cell proliferation was low in unspread cells and high in spread cells in the presence of the ROCK inhibitor (). Collectively, these data suggest a signaling pathway whereby lack of FAK or displacing endogenous FAK from focal adhesions causes an increase in RhoA activity, and this increase, in turn, is required for loss of the growth control normally observed in low adhesive conditions. To determine whether changes in RhoA signaling are sufficient to directly affect proliferation, we overexpressed a constitutively active form of RhoA (RhoA-V14) in unspread ECs. RhoA-V14 dramatically increased stress fiber formation () and was sufficient to overcome the spreading-regulated block in proliferation (). High RhoA also released cells from proliferation arrest induced by confluence (unpublished data). This effect was mediated through the RhoA effector ROCK, as treatment with Y-27632 abrogated the RhoA-V14–induced proliferation (). This ROCK activity was not only necessary but also sufficient to induce proliferation, as expression of a constitutively active ROCK (ROCK-Δ3) also bypassed the shape-dependent control mechanism (). As with RhoA-V14 overexpression, ROCK-Δ3 overexpression had no effect in well-spread cells (). One important consequence of RhoA and ROCK signaling is in mediating changes in myosin-regulated cytoskeletal tension (; ; ). To address whether FRNK-induced signaling altered focal adhesion structure and proliferation via RhoA-mediated changes in cytoskeletal tension, we assessed myosin phosphorylation in cells expressing FRNK. FRNK expression dramatically increased the amount of phosphomyosin compared with GFP controls (). Although this suggests that FRNK might be functioning to increase cytoskeletal tension in unspread cells, myosin phosphorylation is not always associated with the development of tension. To directly measure the tension transmitted across the focal adhesion onto the underlying substrate, we used a previously described microfabricated force sensor (), consisting of an array of vertically placed elastomeric microneedles. These microneedles report the traction force exerted by cells on the underlying substrate. Thus, we directly measured the tension generated in unspread cells expressing FAK, FRNK, FAK-Y397F, or a GFP control. Notably, only FRNK expression increased traction force (). FAK expression showed no differences in tension, whereas expression of FAK-Y397F decreased tension. Collectively, these data support a novel role for FAK in growth control, in which loss of FAK signaling can induce RhoA-mediated cytoskeletal tension, leading to the loss of adhesion-dependent control of cell proliferation. In this study, we demonstrate that FAK plays a key role in the regulation of proliferation by cell adhesion, whether modulated by ECM density, cell spreading, confluence, or 3D culture. FAK overexpression has been shown to increase proliferation in previous studies (; ). We find that FAK exerts not only stimulatory but also inhibitory effects on proliferation. The inhibitory function of FAK is lost in FAK−/− cells and, importantly, rescued when FAK is reexpressed. Interestingly, expressing the C-terminal fragments of FAK (FRNK or FAT) also dysregulated the inhibitory function of FAK, whereas the full-length, kinase-dead mutant (FAK-Y397F) could rescue growth inhibition. These data suggest that the inhibitory function of FAK lies in its N-terminal domain. Given that we and others find that FRNK and FAT displace endogenous full-length FAK from focal adhesions (), these C-terminal constructs might interfere with FAK function by competitively inhibiting the targeting of cellular FAK to the focal adhesion, suggesting the interesting possibility that a pool of inactive FAK may normally function to inhibit proliferation through these interactions, and suggests a model whereby FAK acts within adhesions as a graded sensor that transduces adhesive signals to regulate the cell cycle (). High FAK activation caused by high adhesion or by high FAK expression stimulates proliferation, whereas minimal adhesion prevents FAK activation and yields inactive complexes that inhibit proliferation. Interestingly, observed an unexpected enhancement of soft agar colony formation in v-Src–transformed cells lacking FAK that was subsequently prevented by FAK reexpression, suggesting the possibility that FAK may play a negative regulatory role in transformation. Notably, this occurred in a low adhesive environment. An alternative model for the proliferative response to both up- and down-regulation of FAK is the possibility that dynamic cycling of FAK activation and deactivation is required for growth inhibition. Repeated cycles of FAK phosphorylation and dephosphorylation appear to be important for cell migration, as both decreasing and increasing FAK activity reduce migration (; ). Thus, both stimulatory and inhibitory roles for FAK may be an inherent feature of its function in numerous cellular processes. RhoA is a critical regulator of focal adhesion formation (; ). Our results also demonstrate that RhoA plays a role in the dysregulation of growth control in cells lacking FAK. Both FRNK-expressing cells and FAK−/− cells exhibit high RhoA activity that appears to be both necessary and sufficient for the observed proliferative effect (), supporting studies suggesting that RhoA promotes cell cycle progression (). Although we show that the RhoA effector ROCK is important in our system, RhoA-mediated mDia signaling also appears to be sufficient to induce proliferation (), suggesting that numerous RhoA signals may regulate growth. The mechanism by which FRNK and loss of FAK might up-regulate RhoA remains to be defined, although a simple mechanism may be that FRNK opposes the suppression of RhoA activity by endogenous FAK. The ability of FAK to down-regulate RhoA activity is well documented (), and it has been shown that FAK may interact with the Rho GTPase-activating protein (GAP) GRAF () and phosphorylate p190RhoGAP (). It is possible that under different adhesive contexts, such as high or low ECM ligand density or high or low cell spreading, FAK may alter its interaction with Rho GAPs or Rho GEFs and, thus, modulate RhoA activity and proliferation. It has long been known that changes in cell shape and the associated changes in cytoskeletal tension are required for proliferation (; ; ; ). We show that FAK transduces cell shape into proliferative signals. Interestingly, although FAK has been implicated as a mechanosensor where increasing tension leads to FAK activation (), we show that FAK also alters the cytoskeletal tension and forces experienced at the adhesion. Expression of FRNK, through its effects on RhoA, increases myosin-based cytoskeletal tension, confirming earlier suggestions from the Parsons group that FRNK might increase cellular contractility (). It has been previously observed that FRNK also increases focal adhesion size (). Our findings would suggest that these changes in focal adhesions are actually mediated by increased cytoskeletal tension, as focal adhesion maturation is induced by mechanical stress (; ; ). Thus, it appears that FAK both responds to and causes changes in mechanical force, and the latter links changes in cell adhesion to changes in cell mechanics and proliferation. These two reciprocal functions likely provide the mechanochemical feedback that is required for tightly integrating the mechanical and biochemical dynamics of cell adhesion. The role of FAK in cell proliferation has implications for human physiology and pathology, where FAK protein overexpression has been found in invasive human tumors (; ). This has led to the suggestion that targeting FAK might reduce cancer proliferation, migration, and invasion. However, it is now clear that the model whereby FAK is strictly a stimulatory molecule for proliferation is oversimplified. In fact, FAK down-regulation can increase tumor cell motility, invasion, and metastasis (; ), and we speculate that it may also extend to include increased proliferation. Thus, simply eliminating FAK function in cancer settings may be detrimental, and recognizing these additional layers in FAK function may reveal how cells can interpret complex adhesive contexts into a well-adapted response. For many adherent cell types, both integrin ligation and cell spreading are required to support proliferation. Because focal adhesion architecture and, likely, the focal adhesion character are different in spread and unspread cells, it is probable that focal adhesions formed under these various adhesive or mechanical contexts transmit different signals, leading to potentially divergent cellular behaviors. Importantly, FAK appears to be a central regulator of adhesion-mediated proliferation, whether signaled by spreading, confluence, ligand density, or 3D matrix architecture, where it can transduce both stimulatory and inhibitory proliferative signals. Understanding how this single molecule can play such a central role in many complex interactions will uncover important insights into how cells navigate and respond to their adhesive and mechanical environments in physiologically meaningful ways. Bovine pulmonary artery ECs (VEC Technologies, Inc.) were cultured in low glucose DME containing 2 mM glutamine, 100 units/ml penicillin, 100 μg/ml streptomycin, and 5% bovine serum (all from Invitrogen). ECs were maintained in a humidified 10% CO incubator. FAK−/− and FAK-reexpressing mouse embryo fibroblasts were a gift from S. Hanks (Vanderbilt University, Nashville, TN) and were cultured in DME containing 4,500 mg of D-glucose/ml, 2 mM glutamine, 100 units/ml penicillin, 100 μg/ml streptomycin, 0.25 μg of amphotericin B/ml (all from Invitrogen), and 10% fetal bovine serum (Atlanta Biologicals) and were maintained at 37°C in a humidified 5% CO incubator. The following reagents were purchased from the given suppliers: human fibronectin (Invitrogen); Y-27632 (Calbiochem), PP2 (Calbiochem), JNK inhibitor I, UO126 (Calbiochem), anti-vinculin clone hVin-1 (Sigma-Aldrich), TRITC-conjugated phalloidin (Sigma-Aldrich), anti-RhoA (Santa Cruz Biotechnology, Inc.), phospho-Y397-FAK antibody (BioSource International), total FAK antibody (Cell Signaling Technology), phospho-S18/S19 MLC antibody (Cell Signaling Technology), and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) antibody (Abcam). For F-actin stains, cells were fixed with 4% paraformaldehyde in PBS. F-actin was visualized by incubating samples with fluorophore-conjugated phalloidin (Invitrogen). Quantitative analysis of focal adhesions was performed as previously described (). In brief, cells were incubated for 1 min in ice-cold cytoskeleton buffer (50 mM NaCl, 150 mM sucrose, 3 mM MgCl, 1 μg/ml aprotinin, 1 μg/ml leupeptin, 1 μg/ml pepstatin, and 2 mM PMSF), followed by 1 min in cytoskeleton buffer supplemented with 0.5% Triton X-100. Detergent-extracted cells were fixed in 4% paraformaldehyde in PBS, washed, and incubated with a primary antibody to vinculin (Sigma-Aldrich). After incubation with Alexa Fluor 594–conjugated secondary antibodies (Invitrogen), quantitative microscopy of focal adhesion proteins was performed using a charge-coupled device camera (Orca; Hamamatsu) attached to an inverted microscope (model TE2000; Nikon) using a 100×, 1.4 NA, oil immersion objective with a 400-ms exposure time at RT. Images were obtained and processed using IPLab software (Scanalytics); original images were filtered and binarized to subtract background fluorescence, and then segmented with a threshold of 0.25 μm to quantify the area of individual adhesions. Approximately 100–150 cells were analyzed per experimental condition. Triton X-100 soluble and insoluble pools were generated by washing cells with ice-cold TBS, followed by a 5-min wash with Triton extraction buffer (50 mM NaCl, 150 mM sucrose, 3 mM MgCl, 0.5% Triton X-100, 1 μg/ml aprotinin, 1 μg/ml leupeptin, 1 μg/ml pepstatin, and 2 mM PMSF). The soluble fraction was collected, mixed with Laemmli sample buffer, and boiled. The remaining Triton-insoluble fraction was collected by scraping directly into 1× Laemmli sample buffer and then boiled. Soluble and insoluble fractions were run on SDS-PAGE gels and blotted. 3D collagen I gels were prepared by mixing M199 (Invitrogen), NaHCO (0.035% wt/vol; Sigma-Aldrich), 10 mM Hepes buffer (Invitrogen), rat tail collagen I (BD Biosciences), and distilled water with the pH adjusted to 7.4. Synchronized FAK−/− and FAK-reexpressing cells were seeded into a 2.4-mg/ml collagen gel at a concentration of 16,000 cells/ml followed by gelation at 37°C for 30 min. Cells were incubated for 22 h in the presence of radiolabeled thymidine (MP Biomedicals), after which the cells were lysed and DNA was precipitated with 16 M NaOH containing 0.25% Triton X-100. Radioactivity counts were measured using a scintillation counter (Beckman Coulter). Blank collagen gels were used to measure background residual thymidine. To generate stamps for microcontact printing of proteins, a prepolymer of poly(dimethylsiloxane) (PDMS; Sylgard 184; Dow Corning) was poured over a photolithographically generated master, as previously described (). Stamps were immersed for 1 h in 50 μg/ml fibronectin, washed three times in water, and blown dry under nitrogen. Coated stamps were placed in conformal contact with a surface-oxidized PDMS-coated glass coverslip. Stamped coverslips were immersed in 0.2% Pluronic F127 (BASF) in PBS for 1 h and washed. FAK, FRNK, FAT, FAK-Y397F, RhoA-V14, ROCK-Δ3, and GFP recombinant adenoviruses were constructed using the AdEasy XL system (Stratagene) according to manufacturer's instructions. RhoA cDNAs were obtained from M. Philips (New York University Medical Center, New York, NY) and P. Burbelo (Georgetown University, Washington, DC). ROCK cDNAs were obtained from S. Narumiya (Kyoto University, Kyoto, Japan). In brief, cDNAs were subcloned into the pShuttle-IRES-GFP1 vector, and then cotransformed with the pADEASY1 plasmid. After homologous recombination, plasmids were used to transfect human embryonic kidney 293 cells. High titer preparations of recombinant adenovirus were generated by CsCl density gradient centrifugation. In viral infection experiments, viral MOI resulting in a transduction efficiency of at least 80% was added to cells. ECs were G synchronized by holding the cells at confluence for 2 d. FAK−/− and FAK-reexpressing cells were synchronized by 60-h serum starvation. Cells were then trypsinized and replated in the presence of BrdU (GE Healthcare). Cells were fixed at 22 h and stained for BrdU incorporation using a monoclonal antibody directed against BrdU (GE Healthcare). Cells were counterstained with Hoechst 33342 (Invitrogen). RhoA-GTP levels were measured by pull-down assay (). In brief, cells were washed with cold TBS, scraped into lysis buffer (25 mM Hepes, pH 7.5, 15 mM NaCl, 1% Igepal CA-630, 5 mM MgCl, 1 mM EDTA, 10% glycerol, 10 μg/ml aprotinin, 10 μg/ml leupeptin, 10 μg/ml pepstatin, and 2 mM PMSF). Cleared lysates were incubated with 30 μg GST–rhotekin-binding domain–agarose beads (Upstate Biotechnology) for 45 min at 4°C, centrifuged, washed, and eluted by boiling in SDS-PAGE buffer containing 5% β-mercaptoethanol for 5 min. RhoA was detected by Western blotting using a monoclonal antibody to RhoA (Santa Cruz Biotechnology, Inc.). The level of RhoA activity in different samples was determined by normalizing the amount of rhotekin-binding domain–bound RhoA to the total amount of RhoA in cell lysates. Cells were washed in TBS and lysed in cold modified RIPA buffer (50 mM Tris-HCl, pH 7.4, 1% Igepal CA-630, 0.25% deoxycholate, 150 mM NaCl, 1 mM EDTA, 1 mM PMSF, 1 mM orthovanadate, 1 mM NaF, and 1 μg/ml each aprotinin, leupeptin, and pepstatin). Proteins were separated by denaturing SDS-PAGE electroblotted onto PVDF, blocked with 5% milk in TBS, immunoblotted with specific primary antibodies, and detected using horseradish peroxidase–conjugated secondary antibodies (Jackson ImmunoResearch Laboratories) and SuperSignal West Dura (Pierce Chemical Co.) as a chemiluminescent substrate. Densitometric analysis was performed using a VersaDoc imaging system with QuantityOne software (Bio-Rad Laboratories). Microfabricated post array detectors (mPADs) were fabricated as previously described (; ). mPADs used in these studies were 11 μm tall and 3 μm in diameter, with 9 μm center–center spacing. To control cell spreading on microneedle tips, the tips were stamped with fibronectin using microcontact printing (), and nonstamped regions were blocked with 0.2% Pluronic F127 (BASF). ECs expressing either GFP, FRNK, FAK, or FAK-Y397F were cultured on the mPADs for 22 h, after which the samples were fixed with 4% paraformaldehyde in PBS. Fibronectin was stained with goat anti-fibronectin antibody (ICN Biomedicals) and the nuclei were stained with Hoechst 33342. The samples were imaged using an Axiovert 200M (Carl Zeiss MicroImaging, Inc.) with the Apotome module, equipped with 63× Plan-Apochromat, 1.4 NA, oil immersion objective, an Axiocam camera, and Axiovision software (Carl Zeiss MicroImaging, Inc.). A Matlab program (The MathWorks) was used to obtain tractional force from the acquired images. At least six cells were used in force measurements in each condition.

Synaptic plasticity, both in the short- and long-term, enables the brain to adapt in response to environmental inputs. Substantial evidence indicates that the strength, number, and morphology of synapses can be changed by neuronal activity. Indeed, stimulation of hippocampal slices under conditions that induce long-term potentiation (LTP) not only increases the efficacy of individual synapses but also increases overall synapse number, the length of active zones (for review see ), and the number and length of active zones (; ); it also modifies spine morphology (). It is now widely accepted that neuronal activity enhances local synthesis and secretion of neurotrophins, most notably brain-derived neurotrophic factor (BDNF), which in turn play a crucial role in synaptic transmission and plasticity (; ). Acute application of BDNF has been found to rapidly enhance synaptic transmission and transmitter release and to mediate increased synapse sprouting, which is similar to that seen after strong activity (Jovanovic et al., 2000). Mice expressing reduced levels of BDNF exhibit a dramatic decrease in the number of docked vesicles per synapse, pronounced synaptic fatigue, and deficits in synaptic sprouting (; ). Similarly, mice lacking the BDNF receptor TrkB exhibit a decreased number of both docked and total synaptic vesicles (SVs) per synapse, as well as a decrease in overall synapse number (; ; ). TrkB receptors are localized at synapses; therefore, they are well positioned to rapidly regulate synapse form and function after activation (; ; ). Both BDNF and TrkB mutant mice exhibit impaired induction of LTP, resulting in large part from defects in presynaptic function (). Despite a rapid growth in this field in recent years, the molecular mechanisms that mediate changes in the structure and function of synapses remain largely unknown. Evidence suggests that classic cell adhesion molecules such as cadherins, integrins, and immunoglobulin domain–containing proteins, as well as neurexins and neuroligins, play a large role in regulating synapse formation (for review see ). Cadherins in particular have been well studied and shown to play a role in regulating synapse formation, function, and plasticity. Cadherins are localized at synapses just adjacent to active zones () and are linked to the actin cytoskeleton via β- and α-catenin (for review see ). The cadherin–catenin complex is therefore well situated to coordinate pre- and postsynaptic structural changes, as well as to facilitate the formation and maintenance of synaptic junctions. Consistent with this, cadherins and catenins have been shown to be important for localizing SVs to presynaptic compartments () and for modulating the shape and formation of postsynaptic spines (; ). Synaptic activity has been reported to modify the conformational state of N-cadherin (), and N-cadherin levels are significantly elevated during late LTP when new synapses are formed (). Furthermore, impairment of cadherin function inhibits the induction, but not the maintenance of LTP (). Both BDNF–TrkB and cadherin–β-catenin complexes have been shown to play an important role in regulating the number of SVs at individual presynaptic compartments, as well as regulating the overall number of synapses. Therefore, we hypothesized that the synaptic effects observed after activation of BDNF–TrkB signals might be mediated by the modulation of cadherin–β-catenin interactions. In this work, we demonstrate that acute treatment of cultured hippocampal neurons with BDNF results in a transient dispersal of SVs into perisynaptic regions, followed by a relatively sustained enhancement of SV cluster splitting. Therefore, we conclude that BDNF enhances SV mobility. Long-term treatment of neurons with BDNF resulted in an increase in both the density of SV clusters along the axon and the density of synapses, which are identified by the colocalization of SV clusters with the postsynaptic marker PSD-95. We show that BDNF treatment dissolves cadherin–β-catenin complexes by promoting the phosphorylation of β-catenin on tyrosine residue 654. Interestingly, artificially maintaining cadherin–β-catenin interactions by introducing a β-catenin point mutant that cannot be phosphorylated on this tyrosine residue (β-catenin Y654F) abolished both the enhanced mobilization of SVs and the BDNF-mediated increase in the overall density of SV clusters and synapses. xref #text xref #text Rat hippocampi from E18 fetal rats were prepared as previously described () and plated at a density of 130 cells/mm. For time-lapse studies, neurons were transfected using Effectene (QIAGEN) transfection at 10 days in vitro (DIV), or as indicated, and examined 2 d later. For long-term BDNF studies, cultures were transfected at 10 DIV, treated with 100 ng/ml BDNF at 11 DIV, and examined at 14 DIV. Protein extracts were prepared from 12 DIV primary hippocampal cultures treated with either media alone or 100 ng/ml BDNF for 10 or 30 min. Extracts were immunoprecipitated with monoclonal anti–β-catenin (Zymed Laboratories) and immunoblots were probed with anti-phosphotyrosine (4G10; Cell Signaling Solutions) and anti–N-cadherin (a gift from D. Colman, McGill University, Montreal, Quebec, Canada). Proteins were visualized using enhanced chemiluminescence. Exposed film was scanned and the brightness and contrast of entire images was moderately adjusted using Photoshop (Adobe) after recommended, scientifically acceptable procedures, and no information was obscured or eliminated from the original (). For immunohistochemistry, neuronal cultures were fixed in 4% paraformaldehyde/4% sucrose for 10 min, permeabilized in 0.1% Triton X-100 for 10 min, and then blocked in 10% goat serum for 1 h at room temperature. Primary antibodies were applied in 1% goat serum overnight at 4°C, and secondary antibodies were applied in 1% goat serum for 1 h at room temperature. The primary antibodies used were mouse anti-synaptophysin (Sigma-Aldrich) and mouse anti–PSD-95 (Affinity BioReagents, Inc.); the secondary antibodies used were Alexa Fluor 488 and Texas red–conjugated goat anti–mouse or goat anti–rabbit IgGs (Invitrogen). = at least 10 neurons per condition from at least three separate cultures. #text italic #text Video 1 shows that there is minimal variability in stably localized synaptophysin-GFP puncta over time in untreated cells. Video 2 shows that BDNF induces SV dispersal and increased splitting of SV clusters from stably localized synaptophysin-GFP–labeled puncta. Video 3 shows that BDNF induces SV diffusion and the splitting of SV clusters. Online supplemental material is available at .

Protocadherins are transmembrane glycoproteins that contain six or more conserved cadherin-repeats (EC domains) in their extracellular domains. They constitute a large subfamily of the cadherin superfamily (; ; ). Classical cadherins are Ca-dependent, homophilic cell–cell adhesion molecules with five EC domains. Their adhesion activities rely on two common features: the conserved Trp2 in the first cadherin-repeat domain (EC1), which is necessary for homophilic binding, and the conserved catenin-binding motifs in the cytoplasmic domain, which are required for signaling and linkage to the actin cytoskeleton (). Much less is known about the adhesion properties of protocadherins. Protocadherins do not have the Trp2 residue in the extracellular domain or the catenin-binding motifs in the cytoplasmic domain (). It is not even entirely clear whether they function as adhesion molecules or have evolved to perform different cellular functions. Some protocadherins exhibit weak cell aggregation activity when overexpressed in L cells, whereas others do not (; ; ; ; ). It is not clear whether the weak cell aggregation mediated by a few of these protocadherins reflects true cell adhesion function at physiological levels of expression. More direct and thorough studies, like those that have been performed on classical cadherins, are needed to establish the adhesion properties of a protocadherin. paraxial protocadherin (PAPC) is a protocadherin that has been shown to play an essential role in the convergence and extension movements of paraxial mesoderm and in the establishment of somite boundaries during the early development of embryos (, ). It is first expressed in Spemann's organizer at the onset of gastrulation, and is later expressed in the paraxial trunk mesoderm. By stage 14, PAPC is expressed in stripes and prefigures the forming somites. PAPC also induces the sorting out of blastomeres, which was taken as evidence that it functions as a homophilic cell-adhesion molecule. Recently, two groups reported that PAPC interacts with Frizzled-7 (Xfz7) and can activate RhoA and JNK signaling via the noncanonical Wnt pathway to regulate tissue separation or convergent extension (; ). C-cadherin is a classical cadherin that mediates cell–cell adhesion between blastomeres. It is expressed both maternally and zygotically in all cell types throughout the early stages of embryonic development (; ; ), and plays essential roles in the maintenance of embryo integrity () and in morphogenetic cell movements (; ). Inhibition of C-cadherin adhesion activity by dominant-negative (DN) C-cadherins causes failure of blastopore closure (). Furthermore, the adhesion activity of C-cadherin at the blastomere surface is down-regulated during activin-induced elongation of animal cap explants, a process believed to mimic the convergence and extension cell movements during gastrulation (). Disrupting the down-regulation of C-cadherin adhesion activity by an activating antibody blocks animal cap elongation (). These findings demonstrate that dynamic regulation of C-cadherin adhesion activity plays a pivotal role in embryonic tissue morphogenesis. However, the mechanism by which C-cadherin activity is regulated during morphogenesis is unknown. Because PAPC, like C-cadherin, is a cadherin with a role in cell sorting and convergence and extension morphogenetic cell movements during gastrulation, we chose it as an interesting model protocadherin to investigate. We first undertook a thorough examination of the adhesion properties of PAPC, including an analysis of domains required for its function. We also investigated the mechanism by which PAPC mediates cell sorting in the embryo and its relationship to C-cadherin–mediated adhesion. Finally, we asked how PAPC and C-cadherin cooperate to regulate tissue morphogenesis in the embryo. CHO cells do not express endogenous cadherins and have been successfully used for studying the adhesion activities of classical cadherins (; ; ). We generated stable CHO cell lines that express full-length PAPC (FL-PAPC), a cytoplasmic tail–deleted form of PAPC (M-PAPC; ), or GFP as control, and examined their cell aggregation properties and their capacity to adhere to a substrate of purified PAPC protein. We tested M-PAPC as well as FL-PAPC because M-PAPC has been reported to have stronger cell sorting activity than FL-PAPC (). Both FL-PAPC and M-PAPC were expressed on the surface of CHO cells, as demonstrated by accessibility to trypsinization, surface biotinylation, and immunofluorescence staining of intact cells (Fig. S1, available at ). Surprisingly, in cell aggregation assays, neither FL-PAPC-CHO cells nor M-PAPC-CHO cells aggregated to any extent compared with mock control cells (GFP-CHO; ). For comparison, a stable cell line expressing C-cadherin, C-CHO, aggregated over time in the same experiment (). We also performed cell attachment flow assays using a purified PAPC protein as an adhesion substrate (PAPC-EC.Fc, which is the extracellular domain of PAPC fused with human IgG Fc, see Fig. S2). Although C-cadherin–expressing cells (C-CHO) adhered strongly to a C-cadherin substrate (C-cad-EC.Fc), neither FL-PAPC-CHO cells nor M-PAPC-CHO cells adhered to the PAPC substrate (). In addition to PAPC-EC.Fc, we also used another adhesion substrate, purified soluble PAPC with a C-terminal 6×His-tag (PAPC-EC.His; Fig. S2), which forms higher-order oligomers rather than dimers (unpublished data) and may have a conformation different from PAPC-EC.Fc. However, PAPC-expressing cells did not adhere to PAPC-EC.His either (unpublished data). Therefore, PAPC does not mediate homophilic cell adhesion in CHO cells. It is possible that CHO cells lack the necessary cytoplasmic factors for PAPC-mediated adhesion. Therefore, we also prepared multiple stable PAPC-expressing cell lines using different kinds of cell types, including human epithelial A431 cells (), XTC cells (), MBA-MD231, SW480, and MCF7 cells, and several others (unpublished data). None of these PAPC-expressing cell lines exhibited cell adhesion activity to purified PAPC substrates, but they were all able to adhere strongly to either purified E- or C-cadherin substrates via their endogenous cadherins. We also examined the adhesion activity of PAPC in blastomeres, in which it has been shown to mediate cell sorting (). FL-PAPC, M-PAPC, or GFP was expressed in embryos by mRNA injection, and blastomeres dissociated from isolated animal caps were tested for adhesion to either PAPC-substrates or the C-cadherin substrate. Blastomeres expressing PAPC did not adhere to either PAPC substrate, whereas the same blastomeres were able to adhere strongly to a 10 times less concentrated C-cadherin substrate (). Adhesion assays were also performed on blastomeres from the dorsal trunk mesoderm of stage 12 embryos, in which endogenous PAPC is expressed, but these blastomeres did not adhere to PAPC substrates either (). In summary, both the blastomere adhesion assays and the cell culture adhesion assays indicate that PAPC does not function effectively as a homophilic cell–cell adhesion molecule. The lack of intrinsic cell adhesion activity for PAPC appeared inconsistent with the reported cell sorting activity of PAPC in embryos (). Therefore we tried to reproduce the cell sorting assays as described by , using both cell dispersal assays and reaggregation assays. In cell dispersal assays, with GFP mRNA alone injected into a single blastomere at the 32-cell stage, labeled cells extensively interspersed with surrounding unlabeled cells at a later stage of development (). In contrast, cells derived from FL-PAPC or M-PAPC mRNA-injected blastomeres (with a GFP lineage tracer) formed tight patches and maintained sharp boundaries with their unlabeled neighbors (), confirming the cell sorting activity of both FL-PAPC and M-PAPC. Moreover, in dissociation and reaggregation assays, blastomeres dissociated from FL-PAPC or M-PAPC mRNA-injected embryos (with a GFP lineage tracer) nicely sorted out from blastomeres obtained from uninjected embryos (), whereas blastomeres from embryos in which GFP mRNA alone was injected uniformly mixed with uninjected blastomeres (). Notably, FL-PAPC has the same activity in inducing cell sorting as M-PAPC because the same amount of mRNA was injected in every experiment. This appears different from an earlier study that found M-PAPC mRNA to be seven times more efficient in inducing cell sorting (). However, in that study, protein expression levels were not measured because anti-PAPC antibodies were not yet available. Indeed, the FL-PAPC construct used in the previous study produces seven times less protein in embryos than the M-PAPC construct (Fig. S3, available at ). This original FL-PAPC construct differs from the M-PAPC construct because it retains both the 5′ and 3′ untranslated regions (UTRs). For our experiments, we removed the 3′ (and 5′) UTR, which results in similar protein expression levels for FL-PAPC and M-PAPC (Fig. S3 D–F) and, thus, higher cell-sorting activity for FL-PAPC (Fig. S3, G–I). Therefore, the cytoplasmic tail is not required for the cell-sorting activity of PAPC. Furthermore, expression of the membrane-bound cytoplasmic domain of PAPC had no detectable affect on cell sorting or on M-PAPC–induced cell sorting (Fig. S4). It is important to note that the PAPC-expressing cells sorted to the outside of the aggregates in clusters. This was apparent in a surface view (), but was also confirmed by bisection of the aggregates (). According to Steinberg's differential adhesion theory (; ), cells with weaker adhesion strength tend to sort to the periphery of coaggregates. This suggests that PAPC-expressing cells have weaker adhesion strength than uninjected cells. Indeed, in blastomere aggregation assays, FL-PAPC– or M-PAPC–expressing cells only formed small aggregates () compared with control GFP-expressing cells (), showing that PAPC-expressing cells exhibit less overall cell adhesion activity. Therefore, both FL-PAPC and M-PAPC may induce cell sorting by down-regulating the overall cell adhesion strength. Knowing that PAPC does not mediate cell adhesion itself and yet can induce cell sorting, we hypothesized that PAPC changes the adhesion activity of other adhesion molecules. The best candidate is C-cadherin, because C-cadherin is expressed throughout the embryo in early stages of development and has been shown to be necessary for blastomere adhesion (). In addition, it has been shown that C-cadherin activity can be down-regulated by growth factors such as activin (). To examine whether PAPC regulates C-cadherin adhesion, we performed blastomere adhesion assays under conditions used to detect activin-regulation of C-cadherin adhesion activity. Blastomeres obtained from FL-PAPC or M-PAPC mRNA-injected embryos exhibited significantly decreased levels of C-cadherin–mediated adhesion, which was equivalent to only ∼40% of the control level exhibited by blastomeres from GFP mRNA-injected embryos (). A decrease in C-cadherin adhesion could be caused either by decreased C-cadherin protein level at the cell surface or by decreased intrinsic adhesion activity of C-cadherin. PAPC expression did not change the overall levels of C-cadherin protein in whole embryos (, lane 1–2) or in animal cap blastomeres used for adhesion assays (, lane 3–4). To determine whether the surface level of C-cadherin was changed by PAPC expression, we treated dissociated blastomeres with trypsin-EDTA to remove cell surface C-cadherin. PAPC expression did not alter the amount of trypsin-accessible C-cadherin (, lane 5–8). Previous work has shown that a specific C-cadherin–activating antibody, AA5, can reverse activin-regulation of C-cadherin–mediated adhesion at the cell surface, demonstrating intrinsic regulation of C-cadherin adhesion activity by activin (). AA5 similarly reversed PAPC-regulation of C-cadherin–mediated adhesion (). Therefore, PAPC functions by decreasing intrinsic C-cadherin adhesion activity at the cell surface. The effect of PAPC expression on C-cadherin–mediated adhesion is specific because blastomere adhesion to fibronectin or to antibodies against a nonspecific, exogenously expressed cell surface protein, human interleukin 2 receptor α (IL2Rα), is not affected by PAPC expression (). To determine whether the down-regulation of C-cadherin activity is the cause of PAPC-induced cell sorting, we asked whether increasing C-cadherin expression levels could reverse PAPC-induced cell sorting. In cell dispersal assays with increasing amounts of C-cadherin mRNA coinjected with M-PAPC (and GFP) mRNA, the GFP-labeled cell population gradually changed from a tight patch to a loose patch and, eventually, to total mixing with their uninjected neighbors (). Therefore, overexpression of C-cadherin reverts PAPC-induced cell sorting. This result provides additional evidence that the down-regulation of C-cadherin adhesion by PAPC, rather than added PAPC-mediated adhesion, causes cell sorting. If PAPC increased cell adhesion, coexpression of C-cadherin should bolster cell sorting instead of blocking it. Overexpression of M-PAPC in the animal hemisphere of the embryo consistently caused failure of blastopore closure during gastrulation (defect rate 30/30; , middle column), which is a phenotype similar to that caused by DN C-cadherin expression (). The same was observed for FL-PAPC mRNA injection (unpublished data). This phenotype is specific because injection of GFP mRNA did not cause any defect at the same stage (defect rate 0/50; , left column). Co-injection of C-cadherin mRNA, along with PAPC mRNA, rescued the blastopore closure defect significantly (defect rate 6/51; , right column). These results suggest that ectopic PAPC expression exerts its overall gastrulation phenotype by down-regulating C-cadherin activity. To determine whether endogenous PAPC functions to inhibit C-cadherin adhesion activity, we also did loss-of-function studies. It is known that at the early stage of gastrulation, PAPC expression is limited to the dorsal marginal zone (DMZ; ). We found that the blastomeres obtained from the DMZ of stage 10.5 embryos exhibit significantly lower C-cadherin adhesion level than those from the ventral marginal zone (VMZ; , first two columns). We effectively knocked down endogenous PAPC expression using PAPC-specific morpholinos (PAPCMO; , lanes 5–6). Knocking down of endogenous PAPC expression results in a significant increase in the level of C-cadherin–mediated adhesion of the DMZ blastomeres to that of the ventral blastomeres (, third column), suggesting that PAPC is responsible for the lower level of C-cadherin–mediated adhesion in the DMZ of control morpholino (COMO)–injected embryos. Moreover, the C-cadherin–activating mAb AA5 can also increase the adhesion of the DMZ blastomeres (, last column), indicating that the lower level of C-cadherin–mediated adhesion in the DMZ blastomeres is caused by specific down-regulation of the adhesion activity of the C-cadherin protein. In the whole embryo, loss of PAPC expression by PAPCMO injection leads to a blastopore closure defect in stage 12.5 embryos (, PAPCMO), indicating defects in morphogenetic movements of gastrulation. This defect is specifically caused by the loss of PAPC, because it can be rescued by a morpholino-resistant form of FL-PAPC (, PAPCMO+FL). We then asked whether this gastrulation defect occurred because of the lack of down-regulation of C-cadherin adhesion activity by PAPC. Indeed, a DN C-cadherin mutant, the cytoplasmic tail of C-cadherin (), was able to rescue the PAPCMO-defect to the same extent as the morpholino-resistant FL-PAPC (, PAPCMO + Ctail). Both decreased the blastopore size from ∼67% of total embryo diameter in PAPCMO-embryos to ∼37% of embryo diameter in rescued embryos. These results strongly suggest that PAPC functions in vivo to down-regulate C-cadherin adhesion activity and that this function of PAPC is required for proper morphogenetic cell movements during gastrulation. Activin, which is a TGFβ family growth factor, is a mesoderm inducer that induces elongation of animal cap explants, a process mimicking the convergence and extension movements that normally occur during gastrulation (). Activin down-regulates C-cadherin activity without changing the level of C-cadherin, and this down-regulation is necessary for induction of animal cap elongation (; ). We therefore examined whether PAPC plays a role in activin regulation of adhesion and induction of morphogenesis. Activin treatment of animal caps induced PAPC expression in 1–2 h compared with untreated animal caps (), which is similar to the time required for the down-regulation of C-cadherin adhesion activity by activin (). Injection of PAPC morpholinos (PAPCMO) significantly reduced both activin-induced PAPC expression and endogenous PAPC expression compared with a control morpholino (COMO; ). In PAPCMO-injected embryos, activin-treatment failed to down-regulate C-cadherin adhesion activity (, columns 3 and 4), in contrast to significant decrease of C-cadherin adhesion activity in COMO-injected embryos (, columns 1 and 2). Furthermore, coinjection of a morpholino-resistant form of FL-PAPC with PAPCMO (, column 5) resulted in significant down-regulation of C-cadherin adhesion, even without activin treatment, which is similar to injection of FL-PAPC mRNA alone (, column 6). The level of down-regulation by PAPC is comparable to the down-regulation caused by activin. These results demonstrate that PAPC is necessary as well as sufficient to mediate activin-induced down-regulation of C-cadherin adhesion activity in blastomeres. We also asked whether PAPC expression is required for activin-induced elongation of animal cap explants. Animal caps excised from COMO-injected embryos fully elongated (20/20) in response to activin treatment (, ), whereas explants from PAPCMO-injected embryos fell into two groups: no elongation (18/30; , ) and partial elongation without significant narrowing (12/30; , ). As shown in (bottom), the explants from COMO-injected embryos had high levels of PAPC, whereas the “partial-elongation” group of explants from PAPCMO-injected embryos had lower, but detectable, levels of PAPC expression, and the “no-elongation” group of explants from PAPCMO-injected embryos had no detectable PAPC expression. Hence, strong inhibition of PAPC expression blocked elongation, whereas partial reduction in its expression partially blocked elongation. On the other hand, animal caps from PAPC mRNA-injected embryos did not elongate in the absence of activin treatment (unpublished data). Therefore, PAPC expression is necessary, but not sufficient, for activin-induced animal cap elongation. Two recent studies reported that PAPC functionally interacts with Frizzled-7 (Xfz7) –mediated Wnt/planar cell polarity pathway to control tissue separation behavior () and convergent extension movements (). Because M-PAPC, unlike FL-PAPC, does not induce tissue separation when coexpressed with Xfz7 (), but is still capable of decreasing C-cadherin–mediated adhesion suggests that the mechanism of PAPC-dependent tissue separation is different from that of PAPC-inhibition of C-cadherin adhesion activity. Nonetheless, we decided to directly test whether Xfz7 mediates the C-cadherin down-regulation activity of PAPC. Two methods were used to disrupt Xfz7 function: Xfz7 morpholinos (Xfz7MO) and cytoplasmic domain–deleted DN form of Xfz7 (DN-Xfz7). Both Xfz7MO and DN-Xfz7 had been successfully used previously to interfere with Xfz7 function in embryos (; ). Xfz7MO- or DN-Xfz7–injected embryos developed severe gastrulation defects, and failed to form a normal axis at the neurula stage (), indicating that both the morpholinos and the DN construct effectively interfered with Xfz7 function. However, coinjection of Xfz7MO or DN-Xfz7 mRNA with PAPC mRNA into embryos had no affect on the ability of PAPC to decrease C-cadherin adhesion activity compared with COMO coinjection (). Moreover, Xfz7MO or DN-Xfz7 coinjection did not block PAPC-mediated cell sorting (). Thus, interference with Xfz7 function does not affect PAPC-mediated down-regulation of C-cadherin adhesion. Furthermore, we did not observe any change of C-cadherin adhesion activity because of Xfz7 expression or the sorting out of Xfz7-expressing cells (Fig. S5, available at ). Therefore, PAPC regulates C-cadherin adhesion activity and cell sorting independent of Xfz7 signaling. Previous studies on protocadherins, including PAPC, have either assumed or suggested that they function by mediating cell–cell adhesion (; ). This notion has been based primarily on the presence of cadherin EC domains, but in some cases also based on limited evidence for adhesive function. In the case of PAPC, the evidence for adhesion was that it caused cell sorting out, which is a common consequence of adhesion molecule function (). To our surprise, we found no evidence that PAPC functions as a bona fide homophilic cell–cell adhesion molecule. First, cells that express PAPC at their surfaces, either in tissue culture or from embryos, exhibit no detectable adhesion to purified PAPC proteins. Second, there is no detectable aggregation of PAPC-expressing cells, indicating no adhesive interactions between PAPC molecules, even when both are presented on the surface of living cells. Third, PAPC- expressing blastomeres exhibit less aggregation activity than non–PAPC-expressing blastomeres and sort to the outside of coaggregates with non–PAPC-expressing blastomeres, suggesting a decrease, rather than an increase, in the cell-adhesive strength of PAPC-expressing cells. Furthermore, overexpression of C-cadherin counteracts, rather than reinforces, PAPC-mediated cell sorting, consistent with the notion that PAPC does not mediate cell adhesion. Although we cannot exclude the possibility that PAPC has weak homophilic binding activity undetectable in our adhesion assays or that PAPC mediates adhesion in some cell systems other than the ones we tested, it is clear that its cell sorting activity in the embryo is not mediated by PAPC-mediated increase in cell–cell adhesion. A key finding of this study is that PAPC down-regulates C-cadherin adhesion activity to cause cell sorting and contribute to morphogenetic movements. PAPC expression causes a significant decrease in blastomere adhesion to purified C-cadherin protein (). In addition, PAPC-induced cell sorting in embryos is reversed by coexpression of C-cadherin, consistent with the view that decreased C-cadherin adhesion in PAPC-expressing cells is the cause of cell sorting (). Furthermore, PAPC-induced gastrulation defects in embryos phenocopies the adhesion defect caused by DN C-cadherin (), and overexpression of C-cadherin rescues PAPC-induced gastrulation defects (). More importantly, knocking-down of endogenous PAPC results in increased C-cadherin adhesion activity in corresponding tissue, and loss-of-PAPC-function defects can be rescued by decreasing C-cadherin–mediated adhesion using a DN C-cadherin construct (). We find that both M-PAPC, the cytoplasmic domain deletion mutant of PAPC, and FL-PAPC have the same activity in cell sorting and regulation of adhesion. This differs from a previous study in which M-PAPC was found to have higher cell-sorting activity than wild type FL-PAPC (). Our results using anti-PAPC antibodies indicate that this was caused by differences in the levels of PAPC protein expression. Removal of the 3′UTR from the FL-PAPC mRNA resulted in higher protein expression and similar sorting and adhesion regulation activity as M-PAPC. Therefore, the cytoplasmic domain of PAPC is not required for its function in regulation of adhesion and induction of cell sorting. The cytoplasmic domain of PAPC is probably involved in other signaling functions of PAPC. Recent studies found that FL-PAPC interacts and cooperates with Frizzled-7 (Xfz7) and activates RhoA and JNK in regulation of tissue separation, as well as convergent extension movements (; ). The cytoplasmic domain appears to be required because M-PAPC cannot induce tissue separation together with Xfz7 (). Furthermore, an earlier study has found that FL-PAPC, but not M-PAPC, promotes elongation of animal cap explants that are treated with low activin, suggesting a requirement for the cytoplasmic domain in induction of morphogenetic movements (). The cytoplasmic domain of PAPC contains a region of 25 amino acid residues (aa 816–840) that is highly conserved across species and present in other protocadherins. This region might be important for mediating interactions with unknown cytoplasmic factors involved in Xfz7-mediated signal transduction events. Although Xfz7 can mediate signaling events induced by PAPC (; ), it does not appear to be involved in PAPC regulation of C-cadherin adhesion activity. The regulation of C-cadherin by PAPC does not require the cytoplasmic domain, in contrast to Xfz7-mediated PAPC-control of tissue separation and convergent extension movements (). Moreover, interference with Xfz7 expression or function has no affect on the ability of PAPC to down-regulate C-cadherin adhesion activity or to induce cell sorting (). Furthermore, overexpression of Xfz7 does not affect C-cadherin–mediated adhesion nor induce cell-sorting behavior in embryos (Fig. S5). Therefore, PAPC regulates C-cadherin adhesion activity and cell sorting independent of Xfz7. The molecular mechanism by which PAPC down-regulates C-cadherin activity is not yet understood. One possibility is that PAPC could interact with C-cadherin directly and influence its adhesive conformation and activity, but in preliminary experiments, we have not yet observed significant amounts of C-cadherin coimmunoprecipitated with PAPC from detergent lysates of embryos or PAPC/C-cadherin–expressing CHO cells (unpublished data). Moreover, stable expression of PAPC in C-cadherin–expressing CHO cells does not appear to significantly change C-cadherin adhesion activity, suggesting that a more complicated mechanism is involved in regulation in embryo blastomeres. For example, it is possible that PAPC interacts with another membrane protein that either links PAPC to C-cadherin or transduces a signal from PAPC to regulate C-cadherin. If we are able to identify such a membrane protein in future studies, it will be interesting to determine whether it can reconstitute PAPC regulation of C-cadherin in CHO cells, as it does in blastomeres. Activin is a TGFβ family member that induces mesodermal gene expression in embryos. Activin treatment triggers animal cap explants to elongate, a process mimicking convergence and extension movements during gastrulation. Little is known about the mechanism of activin-induced animal cap elongation. Activin has been reported to decrease C-cadherin adhesion activity in animal cap explants (), and reversing this down-regulation with a C-cadherin–activating antibody blocks activin-induced animal cap elongation (). In the present study, we demonstrate that activin induces PAPC expression and that PAPC expression is necessary for activin-regulation of C-cadherin adhesion activity, as well as activin induction of animal cap elongation. PAPC has also been reported to be required for animal cap elongation induced by another TGFβ family growth factor, BVg1 (). However, PAPC expression alone is not sufficient to induce animal cap elongation (unpublished data). These results suggest that additional signals resulting from activin induction are required to elicit morphogenetic cell movements. We propose a model for how activin or other TGFβ family members present in the embryo regulate cell adhesion and induce morphogenesis (). Activin induces PAPC expression and PAPC down-regulates C-cadherin adhesion activity. Dynamic regulation of C-cadherin–mediated cell– cell adhesion is required for convergence and extension cell movements (). PAPC probably also contributes to morphogenesis via Frizzled-7–mediated planar cell polarity pathway, perhaps via cell polarization. Because PAPC expression alone is not sufficient to induce animal cap elongation, activin probably induces expression of additional factors that participate in the generation of convergent extension movements. Protocadherins represent a huge subfamily of molecules in the cadherin superfamily in vertebrates, and have been implicated in several biological processes, especially in the nervous system (; ; ). Protocadherins have neither the known conserved cadherin interfaces for homophilic adhesion, including the Trp2 residue, that mediate adhesion of classical cadherins nor do they have the catenin-binding motifs required for cytoskeletal interactions in their cytoplasmic domains (). To date, the protocadherins that have been studied either exhibit no adhesion activity or have been suggested to mediate weak adhesion based on limited evidence (; ; ; ; ). Whether the weak interaction between some of these protocadherins represents bona fide cell–cell adhesion or is actually involved in other functions such as signal transduction remains unclear. The protocadherin α proteins (Pcdhα) bind the secreted protein reelin and mediate reelin signaling via the nonreceptor tyrosine kinase Fyn that binds to their cytoplasmic domains (). Therefore, protocadherins can function as receptors for extracellular ligands that mediate signal transduction into the cell. An important finding of our study is that a protocadherin can modify cell adhesion by regulating the adhesion activity of a classical cadherin. This could be a potential general mechanism for how protocadherins affect cell adhesion. In fact, Angst et al. speculate that Pcdhα may regulate N-cadherin function in neurons (for review see ), but it would be interesting to test whether Pcdhα regulates N-cadherin–mediated adhesion. Two other protocadherins, neural fold protocadherin and axial protocadherin, also induce cell sorting like PAPC (; ; ), but their intrinsic adhesion activities have not been directly tested. One possibility is that they, like PAPC, regulate adhesion activities of other adhesion molecules. To determine whether any of these protocadherins function as bona fide adhesion molecules, direct careful examination of their adhesive functions will be required. The plasmids pCS2+/FL-PAPC, pCS2+/M-PAPC, and pCS2+/DN-PAPC were provided by E. DeRobertis (University of California, Los Angeles, Los Angeles, CA; ). The 3′- and 5′- UTR of the FL-PAPC cDNA were removed to generate pCS2+/FL(-5′ and -3′) for FL-PAPC mRNA production. For eukaryotic expression, the FL-PAPC coding sequence was amplified by PCR and inserted into the NheI–XhoI site of pCDNA6-V5-His/A vector (Invitrogen), and the M-PAPC coding sequence excised from pCS2+/M-PAPC was inserted into the EcoRI–NotI site of pCDNA-V5-His/A. The coding sequence for the membrane-bound cytoplasmic domain of PAPC (TMC; aa 680–979) was amplified by PCR and inserted to the XhoI–XbaI site of the pCS2+/DN-PAPC to replace the DN-PAPC coding sequence for expression. For production of recombinant soluble PAPC proteins, PAPC extracellular domain (PAPC-EC; aa 1–685) was amplified by PCR and inserted into the HindIII–XbaI site of the pEE14-Fc vector described previously (), resulting in a soluble PAPC protein with a C-terminal human IgG Fc fusion. To prepare 6×His-tagged PAPC-EC construct, PAPC-EC was first cloned into the NheI–XbaI site of pCDNA6-V5-His/A vector. The whole PAPC-EC coding sequence plus the V5-His tag sequence was then excised with NheI and PmeI and inserted into the XbaI–SmaI site of pEE14 vector. pCS2+/NLS-GFP encodes a nucleus-localized GFP and was a gift from L. Davidson (University of Pittsburgh, Pittsburgh, PA). pT3TS/Xfz7 encodes the full-length Xfz7 and pT3TS/DN-Xfz7 encodes a DN-Xfz7 that lacks the cytoplasmic domain. Both were gifts from M. Marsden (University of Waterloo, Ontario, Canada) and were originally constructed by S. Sumanas (University of Minnesota, Minneapolis, MN; ). pSP64T/C-cad and pSP36T/Ctail () were used to make C-cadherin mRNA and C-cadherin cytoplasmic tail RNA (DN form), respectively. All transfections were performed with Lipofectamine or Lipofectamine 2000 (Invitrogen) following the manufacturer's instruction. For stable transfections, cells cotransfected with pCS2+ constructs and pCDNA3 (containing G418-resistant gene) were selected against 0.8 mg/ml G418; cells transfected with pEE14 constructs were selected against 25 μM methionine sulfoximine; cells transfected with pCDNA6 constructs were selected against 5 μg/ml blasticidine. Recombinant C-cad-EC.Fc, C-cad-EC.His, and human E-cad-EC.Fc were purified from conditioned media, as previously described (; ). PAPC-EC.Fc and PAPC-EC.His were similarly purified. After initial purification by protein A or Ni-NTA affinity chromatography, all proteins were further purified on a HiTrap Q ion-exchange column (Invitrogen). Anti-PAPC mAbs, 11A6 and 28F12, were generated against purified PAPC-EC at the Monoclonal Antibody and Hybridoma Facility at Memorial Sloan-Kettering Cancer Center. Anti–C-cadherin mAbs, 6B6, and the activating antibody AA5 have been previously described (; ). Cells were mock treated or treated with 100 μg/ml trypsin and 2 mM EDTA at 4°C for 20 min. Digestion was terminated by washing 3× with PBS containing 2 mg/ml soybean trypsin inhibitor, and lysed directly in SDS-PAGE sample buffer. The trypsinization of blastomeres was carried out as previously described (). Cell surface biotinylation was performed with Sulfo-NHS-ss-Biotin (Pierce Chemical Co.) following the manufacturer's instruction. Biotinylated cells were lysed in PBS/1% NP-40/protease inhibitors (Roche), and biotinylated proteins were pulled down with streptavidin beads. The cell aggregation assay was previously described (). The cell attachment flow assay was performed as previously described (; ), with the following minor modifications: glass capillary tubes were first coated with 5 mg/ml purified goat anti–human IgG (Fc specific; Jackson ImmunoResearch Laboratories, Inc.) before loading of the Fc-fused adhesion substrates (0.1 mg/ml). Blastomere aggregation assays were performed as previously described (). Blastomere adhesion assays were performed as previously described (), with the following modifications: (a) 15 μl of 0.1 mg/ml purified PAPC-EC.Fc, 0.1 mg/ml PAPC-EC.His, 3–10 μg/ml C-cad-EC.Fc, 50 μg/ml fibronectin, or 0.5 mg/ml anti-IL2R mAb BB10 was used for substrate coating, and 1% BSA was used for substrate blocking; and (b) animal caps (at least five) were excised at stage 9 to obtain blastomeres. As needed, the dissociated blastomeres were treated with 5 ng/ml activin for 1 h and/or with 1 μg/ml AA5 F for 30 min. The adhesion strength of blastomeres was measured by the ratio of the number of blastomeres remaining attached after shaking (Nt) versus the number before shaking (No). At least four independent experiments were performed for each sample. The SEM was plotted as error bars. Capped mRNAs were synthesized using the Riboprobe in vitro transcription systems (Promega). Two PAPC morpholinos () and two Xfz7 morpholinos () have been described, which were ordered from Gene Tools, LLC. In each case, a 1:1 mix of the two morpholinos was used for injection. The same amount of standard control morpholino (Gene Tools, LLC) was injected to control embryos. All experimental protocols that involved the use of were approved by the Animal Care and Use Committee at the University of Virginia. eggs and embryos were obtained and handled by standard techniques (). Standard Nieuwkoop staging of embryos was used (). Microinjection of mRNAs or morpholinos was performed at the 2–4–cell stage, as previously described (). Typically, 1–2 ng mRNA or 40–80 ng of morpholinos were injected into the animal pole of the embryos. For animal cap elongation assays, animal caps were excised at stage 8, treated with 5 ng/ml activin for 75 min in 1× Modified Barth's Saline (), rinsed, and further incubated in 1× Modified Barth's Saline at 16°C overnight. Both assays were performed as previously described (). NLS-GFP mRNA (200–400 pg/embryo) was coinjected with PAPC mRNA as a lineage tracer. In brief, for reaggregation assays, RNAs (with tracer) were injected into 4-cell stage embryos. At stage 9, animal caps were excised and cap-blastomeres dissociated from injected and uninjected embryos were mixed at a 1:2 ratio and rocked overnight in Ca-containing media. The aggregates, with or without bisection, were examined under fluorescence microscope. For dispersal assays, sample mRNA, together with tracer, was injected into one blastomere at the animal hemisphere of 32-cell stage embryos. After stage 13, the injected embryos were observed under fluorescence microscope for distribution of GFP-labeled blastomeres. All images were acquired at room temperature. Light images of embryos were acquired with a digital camera (model G6; Canon; at ∼4× optical zoom) mounted on a dissecting microscope (Stemi SVII; Carl Zeiss MicroImaging, Inc.) with lens magnification set between 0.8 and 1.6×. Fluorescence images of embryos were acquired with a color LCD camera (SPOT Insight; Diagnostic Instruments) mounted on an inverted microscope (Diaphot; Nikon) with an objective lens (Plan 4; Nikon) and a Fitz filter. Immunofluorescence microscopy was performed on an Axioplan2 microscope with a Neoplan 20× objective lens and Cys3 filter (all from Carl Zeiss MicroImaging, Inc.). Images were acquired with a digital cameral (model C4742-95; Hamamatsu) and with Openlab 4.0 (Improvision) software. Fig. S1 shows surface expression of PAPC in CHO cells. Fig. S2 shows the purity of PAPC adhesion substrates. Fig. S3 shows that the 3′UTR of FL-PAPC inhibits its protein expression. Fig. S4 shows the membrane-bound cytoplasmic domain of PAPC does not induce cell sorting or affect M-PAPC–induced cell sorting. Fig. S5 shows that Xfz7 expression does not decrease C-cadherin–mediated adhesion or induce cell sorting. Online supplemental material is available at .

Progress in functional proteomics of the nucleolus and nucleolar protein complexes has begun to reveal the complex pathway of eukaryotic ribosome biogenesis. Advances in affinity purification methods and mass spectrometry have allowed the identification of several multiprotein complexes containing hundreds of proteins involved in the production of nucleolar preribosomes. Integration of these results by bioinformatical approaches have led to a tentative map of ribosome assembly pathways (; ; ). However, ribosomal proteins (r-proteins) were largely excluded from such analyses because of their tendency to bind nonspecifically to isolated protein complexes (). Based on previous seminal biochemical analyses of early and late stage preribosomes isolated from mammalian and yeast nucleoli, it is generally assumed that a distinct set of r-proteins (the early assembly group) binds to the pre-ribosomal RNA (rRNA) during or immediately after transcription (; ; ; ). This notion received further support from an immunoelectron microscopic study of spread rRNA genes (“Christmas trees”) of showing the direct association of the r-proteins S14 and L4 with the growing pre-rRNA fibrils (). More recently, five r-proteins (S4, S6, S7, S9, and S14) were identified by proteomic analyses as integral components of the small subunit (SSU) processome of yeast (; a similar RNP complex was identified and termed 90S particle by ). The SSU processome is an RNP particle that is thought to correspond to the large electron-dense terminal knobs seen in electron microscopic spreads of yeast chromatin at the free (5′) ends of nascent pre-rRNA transcripts (). Hence, the SSU processome–associated r-proteins can be classified as early binding proteins. In fact, three of them (S4, S6, and S14) belong to the early assembly group identified in mammalian cells, as does L4, although it is absent from the SSU processome (). The process of ribosome biogenesis can be spatially subdivided and assigned to different nucleolar compartments. Nucleoli contain three morphologically distinct components, which reflect the vectorial process of ribosome production. Synthesis, modification, and initial cleavage steps of the pre-rRNA take place within the fibrillar components (fibrillar center [FC] and the surrounding dense fibrillar component [DFC]), whereas later processing steps are performed in the granular component (GC; for review see ). It is generally assumed that preribosome assembly begins in the DFC of mammalian cells with the formation of 80S RNP complexes containing full-length pre-rRNAs, early binding r-proteins, and numerous nonribosomal proteins and processing and assembly factors (; ; ). A similar scheme applies for yeast (; ), albeit the occurrence of cotranscriptional processing events in these cells has complicated the identification of primary preribosomes containing the full-length RNA transcript (). The notion that 80S preribosomes are assembled in the DFC was recently challenged by a study that analyzed the spatial distribution of several preribosome-associated nonribosomal proteins (). The authors proposed that the most likely site of integration of proteins such as Imp3, Imp4, and Mpp10 into the emerging SSU processome is not the DFC but rather the DFC–GC interface of mammalian nucleoli (for a detailed characterization of the human SSU processome, see ). We have used similar cell biological approaches to clarify the intranucleolar localization of all r-proteins considered early binding and to examine the question of whether they actually accompany the pre-rRNAs on their intranucleolar pathway from the very beginning, as suggested by biochemical analyses. We have cloned cDNAs coding for the early binding human r-proteins S4, S6, S7, S9, S14, and L4. In addition, we have raised antibodies against recombinant S14 to examine the distribution of endogenous S14. The r-proteins were transiently expressed as GFP fusion proteins in Hep2 cells with GFP fused to either the N or the C terminus. To address potential localization problems caused by GFP, because of sterical hindrance or the duration of GFP maturation, the proteins were also fused to a fast maturing YFP (Venus; ) and to a short myc tag that was detected by immunofluorescence. With the appropriate tag, all chimeric r-proteins displayed an intracellular distribution consistent with the well-known fact that ribosomal subunits are assembled in nucleoli and then exported to the cytoplasm (). The GFP fusion proteins S4, S6, S9, S14, and L4 (GFP C-terminal), as well as the corresponding Venus fusions and the myc-tagged proteins S7, S9, S14, and L4 (myc N-terminal), localized in the nucleolus and the cytoplasm, with very little detectable signal in the nucleoplasm. None of the N-terminal GFP fusion proteins localized correctly, either not being directed to the nucleolus or not being exported to the cytoplasm. To verify that the distribution of the expressed fusion proteins mirrored that of their endogenous counterparts, we exemplarily raised an antibody against recombinant S14. of S14 = 16,273). When probed by immunofluorescence, the antibodies labeled nucleoli and the cytoplasm () in a pattern comparable to that produced by the expression of Venus-tagged S14 (). The characteristic nucleolar/cytoplasmic distribution pattern was also seen in a stably transfected Hep2 cell line expressing S4-GFP (unpublished data). As judged from the growth and division of the transfected cells, the tagged r-proteins did not interfere with basic cellular processes. To prove that S4-GFP was actually incorporated into cytoplasmic ribosomes, we isolated ribosomal subunits from the cell line followed by SDS-PAGE and immunoblotting experiments with antibodies against GFP. As shown in , the antibodies recognized a polypeptide band of the small (40S) but not of the large (60S) ribosomal subunits with an apparent molecular mass of ∼55 kD, in agreement with the combined molecular masses of S4 (30 kD) and GFP (27 kD). To correlate the distribution of r-proteins with specific nucleolar components in living cells, we simultaneously expressed fibrillarin fused to monomeric red fluorescent protein (Fib-mRFP) and an r-protein fused to GFP. Fibrillarin is an established marker protein of the DFC of the nucleolus. High- resolution confocal microscopy revealed a clear topological separation between fibrillarin and all examined r-proteins, as shown for L4 and S4 (, rows i and j). The r-proteins appeared to fill the space between the fibrillarin-positive structures, which corresponds to the GC. The assignment of the r-proteins to the GC of the nucleolus was further confirmed by their colocalization with B23-mRFP, an established marker of the GC (, row i, insets). Treatment of cells with low doses of actinomycin D (AMD) inhibits RNA polymerase I–mediated transcription and causes the visible segregation of nucleolar components into closely apposed yet distinct structural entities (). None of the r-proteins colocalized with the marker protein fibrillarin in the segregated DFC, which formed cap-like structures apposed to the GC. Instead, as shown exemplarily for endogenous S14 and S6-GFP, the r-proteins were located in the larger and almost spherical component adjacent to the DFC (, rows a and b). The identity of this component as the GC was ascertained by coexpression of tagged marker proteins for the DFC (fibrillarin-RFP) and the GC (B23-GFP; ″, inset). Furthermore, the distribution of the r-proteins coincided with B23 after treatment of cells with low doses of AMD (unpublished data). It is a distinct possibility that upon AMD treatment the previously synthesized RNA molecules including prematurely released transcripts migrate from the DFC into the GC, where they accumulate (). Therefore, the absence of r-proteins from the DFC in AMD-treated cells could simply reflect the depletion of preribosomes from the DFC and not the inability of r-proteins to bind to early pre-rRNAs. To circumvent the interpretive problems inherent in the AMD experiments, we have exposed cells to the casein kinase 2 inhibitor 5,6-dichloro-1-β--ribofuranosylbenzimidazole (DRB), which does not inhibit transcription of the rRNA genes but causes the unraveling of nucleoli, along with the physical separation of the fibrillar components (FC/DFC) from the GC (). When cells expressing GFP- or Venus-tagged r-proteins were treated with DRB and counterstained with antibodies against fibrillarin to mark the DFC, the r-proteins were absent from the DFC, which often formed necklace-like structures, but were present in spherical GC remnants distributed throughout the nucleoplasm (, rows c and d). In independent experiments, we have confirmed the GC origin of the spherical masses based on the presence of B23 (unpublished data). Nucleoli disintegrate during mitosis of mammalian cells. Reassembly is a stepwise process beginning with the formation of the DFC at the chromosomal nucleolus organizing regions (NORs), followed by the emergence of the GC. Kinetic analyses have shown that the DFC marker fibrillarin accumulates at the NORs several minutes before GC markers (). When we performed live-cell microscopy with cells expressing S4-GFP and fibrillarin-mRFP, we observed a corresponding delay between the early recruitment of fibrillarin and of S4 at the NORs (, row e). On the other hand, the reassembly kinetics of S4 were comparable to that of B23 (unpublished data). We conclude that, during postmitotic nucleologenesis, S4 behaves like a bona fide component of the GC. Next, we examined the intranucleolar localization of myc-tagged r-proteins and endogenous S14 by immunogold electron microscopy. We used pre- and postembedding labeling protocols, as both approaches have different assets and drawbacks. Irrespective of the method used or of the specific r-protein examined, the results were essentially identical. All the early binding r-proteins localized throughout the GC but were not detectable in the FC/DFC regions (for S9, S14, and L4 examples, see ). The apparent absence of r-proteins from FC and DFC was not a consequence of limited antibody accessibility. This was shown by the effective decoration of the DFC with antibodies against fibrillarin () and of the FC with antibodies against RNA polymerase I, an established marker of the FC (). Notably, the localization of early binding r-proteins described in this study resembles the intranucleolar distribution of several nonribosomal proteins, which are integral components of the 80S preribosome (). To compare the distribution of both protein classes directly, we have coexpressed one of the nonribosomal proteins, Imp3, together with either the r-protein S4 or the DFC marker fibrillarin using different fluorescent tags (Fig. S1, available at ). GFP-Imp3 was largely excluded from the DFC, in agreement with previously published data (). In contrast, GFP-Imp3 and S4-Venus colocalized extensively in the GC of nucleoli. Interestingly, the outermost regions of the GC contained relatively low concentrations of Imp3 as compared with S4, suggesting that the removal of nonribosomal protein and final maturation steps of ribosomal particles are confined to more peripheral regions of nucleoli. Collectively, our results are difficult to reconcile with the generally held view that early binding r-proteins associate with pre-rRNAs during or immediately after transcription. If this were the case, the r-proteins should be readily detectable in the DFC, where both elongating and full-length primary transcripts are present at relatively high concentrations, as shown by in situ hybridization studies at the light and electron microscopic level (, ; ). Rather, our data provide evidence that r-proteins do not assemble into preribosomes until pre-rRNAs migrate from the DFC to the GC. In this aspect, our results substantiate the model of preribosome formation at the DFC–GC border proposed by . It will be interesting to find out whether this step corresponds to a specific pre-rRNA cleavage event. In any event, our finding that r-proteins are first detectable in the GC, but that early processing events at both ends of the pre-rRNA already occur within the DFC (; note that 3′ processing requires template-released pre-rRNAs), makes it unlikely, at least in human cells, that the r-proteins associate cotranscriptionally with elongating pre-rRNA transcripts, as has been concluded from electron microscopic immunolocalizations on Miller-type chromatin spreads (). However, at the moment we cannot exclude the formal possibility that the free ends of nascent pre-rRNA transcripts are the sites where assembly of preribosomes takes place, as suggested by , similar to the situation described in yeast (). If this were the case, the nascent transcript fibrils of the rDNA transcription units must all be orientated toward the GC with their free ends aligned at the DFC–GC boundary. At the moment, virtually nothing is known about the spatial arrangement, compaction, and orientation of the rDNA transcription units in nucleoli of live mammalian cells. Furthermore, a structural correlate to the large (∼35–45 nm) terminal yeast knob presumed to represent a fully assembled SSU processome has not yet been identified in chromatin spreads of higher eukaryotic cells (). The abrupt appearance of granules at the DFC–GC interface is indicative of profound structural rearrangements of the nascent ribosomes. We propose that it is the binding of r-proteins that assists and stabilizes the correct folding of the rRNA (; ) and is thus causally involved in the transition from a more extended to a granular character of the preribosomes. What mechanisms might prevent the recruitment of r-proteins to early pre-rRNAs in the DFC, which spend ∼20 min in this compartment before they appear in the GC (; )? We envisage that the association of the large number of small nucleolar RNPs (snoRNPs) with the nascent transcripts competes with the binding of r-proteins. Two classes of snoRNPs mediate the site-specific 2--methylation and the isomerization of uridines to pseudouridines of the rRNAs (for review see ). During or immediately after synthesis, the primary pre-rRNA transcripts are modified by transient interactions with ∼200 different snoRNA–protein complexes with methyltransferase (box C/D snoRNPs) or pseudouridylase (box H/ACA snoRNPs) activities. Each modification-guide RNA is associated with a specific set of proteins, and it is reasonable to assume that binding of numerous snoRNPs of considerable size not only prevents folding of the pre-rRNA () but also prevents interactions with r-proteins. The majority of snoRNPs are located in the DFC, as shown by the intranucleolar distribution of their core proteins and by in situ hybridization experiments with probes complementary to snoRNAs (; ; ; ). It has been proposed that, upon snoRNP dissociation, the pre-rRNAs are extensively refolded by energy-requiring processes mediated by a plethora of transacting protein factors (). Our data suggest that r-proteins are involved in these structural rearrangements and that the actual assembly of ribosome subunits in human cells, i.e., the association of r-proteins with pre-rRNAs, occurs in the GC and is spatially separate from the sites of synthesis, chemical modification, and early processing of the pre-rRNA in the DFC. It has proven difficult to deduce from the results of biochemical analyses alone the order of early assembly steps that lead to preribosomes in yeast and even more so in higher eukaryotes. In situ and in vivo analyses of the spatial distribution of specific preribosome-associated factors will be an important complement in defining pathways of ribosome assembly. The following cDNAs were cloned by RT-PCR using total RNA isolated from Hep2 cells: S4 (gi:39812410), S6 (gi:20381195), S7 (gi:15431308), S9 (gi:550022), S14 (gi:14141191), L4 (gi:16579884), fibrillarin (gi:14763877), B23 (gi: 12803184), and Imp3 (gi:70908369). The cDNAs were cloned into the pCR2.1-Topo vector (Invitrogen), sequenced, and subcloned into the following vectors: pEGFP-N1, pEGFP-C1 or -C3, pDsRed-Monomer-N1, pCMV-Myc (all obtained from CLONTECH Laboratories, Inc.), pQ-C-His, and pVenus. pVenus was produced by site-directed mutagenesis of pEYFP (CLONTECH Laboratories, Inc.) as described by . Transfections were performed using Effectene (QIAGEN) or Fugene (Roche) according to the manufacturers' instructions. Transiently transfected cells were analyzed 24–72 h after transfection. The clone of Hep2 cells stably expressing S4-GFP was isolated after a 3-wk selection period with 0.75 mg/ml G418. Hybridoma supernatant S14-39 was raised against His-tagged S14. S14-His fusion protein was expressed from the vector pQ-C-His in XLI-blue and purified using Talon metal affinity resin (BD Biosciences). Further antibodies used were mAb 72B9 against fibrillarin (), autoimmune serum S18 against pol I (), anti-GFP (Sigma-Aldrich), c-Myc mAb (BD Biosciences), and QM (C-17) against L10 (Santa Cruz Biotechnology, Inc.). Hep2 cells grown on coverslips were fixed with 2% formaldehyde in PBS and permeabilized with 0.2% Triton X-100 in PBS and incubated for 30 min with the primary antibodies and 15 min with the appropriate Texas red–conjugated secondary antibodies (Dianova). In some experiments, cells were treated with 0.04 μg/ml AMD for 2–3 h or with 50 μg/ml DRB for 5–6 h before fixation or live-cell imaging. For live-cell microscopy, cells were grown on glass-bottomed dishes (WPI). Images were taken with a confocal laser-scanning microscope (TCS-SP or TCS-SP2; Leica) equipped with 63×/1.4 NA oil-immersion objectives and a 37°C/5% CO incubation chamber. High-resolution live-cell images (; and Fig. S1) were subject to noise reduction and background subtraction using ImageJ (NIH; ). Images were merged and assembled in Photoshop (Adobe). Ribosomes from ∼10 Hep2 cells were isolated as previously described (). Ribosomal subunits were separated by sucrose gradient (10–40%) centrifugation for 16 h at 23,000 rpm in an SW41 rotor (Beckman Coulter). Fractions containing the small and large ribosomal subunits were analyzed by SDS-PAGE, and immunoblots were performed as previously described (). Fig. S1 shows the localization of IMP3 in the granular compartment of the nucleolus. Online supplemental material is available at .

Chemotaxis is a pivotal response of many cells to spatial cues (; ; ). It plays important roles in diverse functions, such as finding nutrients in prokaryotes, forming multicellular structures in protozoa, and tracking bacterial infections in neutrophils (; ; ). Research on directional movement by external cues in eukaryotes is dominated by chemoattraction, which is the movement toward the chemical compound. Repellents play an important role in morphogenesis, especially during embryonic development (; Schmitt et al., 2005). Cell movement during chick primitive streak formation is controlled by FGF8-mediated chemorepulsion of the cells away from the streak, followed by chemoattraction toward the FGF4 signal produced by the forming notochord (). Axon guidance during spinal chord development away from the roof plate is regulated by multiple repellents, such as BMP (), and by the attractant netrin toward the floor plate (). The mechanism by which repellents work is not well known (). We envision that a critical step of the signal transduction pathway for cell movement is stimulated by a chemoattractant and inhibited by a repellent. It is essential that this hypothetical step is somehow connected with cell polarity to obtain directional movement. cells have been instrumental in resolving the mechanism by which cells sense and respond to chemoattractants. It has been shown that phosphatidylinositol-3,4,5-trisphosphate (PIP), which is formed at the side of the cell closest to the source of chemoattractant, is a very strong inducer of pseudopod extensions (; ; ; ; ). cells are known to be repelled by unidentified compounds that are secreted by starving cells (; ), indicating that cells have a mechanism to process repellents. Previously, we have shown that several analogues of the attractant cAMP behave as a repellent (). The analogues mediate their effect through binding to the surface cAMP receptor cAR1 (), and they can be polar (3′deoxy, 3′amino-cAMP; 3′NH-cAMP) or lipophilic (8-para-chlorphenylthio-cAMP; 8CPT-cAMP). The analogues induce many signaling responses that are essentially identical to the responses induced by cAMP, including activation and adaptation of adenylyl and guanylyl cyclase (; , ). We show that these analogues inhibit PLC, contrary to activation of PLC by cAMP. As a consequence, they induce dominant PIP signaling in the rear of the cell, by which cells move away from the repellent. cells were stimulated with a micropipette containing either the agonist cAMP or the commercially available antagonist 8CPT-cAMP. The cells moved toward the pipette with cAMP, but did not move effectively toward the pipette with 8CPT-cAMP, and actually moved away from the pipette (Fig. S1 and Videos 1 and 2, available at ). Experiments have been repeated with 3′NH-cAMP, yielding the same results as with 8CPT-cAMP (unpublished data). shows four frames from a movie in which cells were stimulated with two pipettes containing cAMP and 8CPT-cAMP, respectively (Video 3). In buffer, cells move in random directions (, 1 min), and cells move away from the pipette with 8CPT-cAMP (, 16 min). Upon application of the pipette with cAMP (cAMP and 8CPT-cAMP; , 26 min) cells moved in nearly random directions. However, upon withdrawal of the pipette containing 8CPT-cAMP, cells immediately moved toward the pipette with cAMP (, 38 min). The trajectories of the cells were analyzed. Data are presented as the chemotaxis index, which is the distance moved in the direction of the gradient (“upgradient”) divided by the total distance moved in 30-s intervals. Data from Video 3 are presented in , and the means and the SEMs for six independent experiments are presented in . Wild-type cells show an excellent chemotactic response toward cAMP, with a chemotaxis index of 0.81 ± 0.05. Cells are not attracted to the pipette containing 8CPT-cAMP, but instead exhibit a significant negative chemotaxis index of −0.52 ± 0.04 (P< 0.005). The chemotaxis index of cells stimulated simultaneously with cAMP and 8CPT-cAMP is −0.18 ± 0.11, indicating that 8CPT-cAMP antagonizes the positive chemotaxis toward cAMP and cAMP antagonizes the negative chemotaxis induced by 8CPT-cAMP. Finally, starting with stimulation by the two pipettes, upon withdrawal of the pipette with 8CPT-cAMP the chemotaxis index toward cAMP rapidly increases to 0.72 ± 0.06. The results demonstrate that 8CPT-cAMP is a repellent that can reversibly inhibit the chemotactic response to cAMP. cells move using actin filaments in the front of the cell, which induce the formation of local pseudopodia, and actomyosin filaments in the rear of the cell, which inhibit pseudopod formation and retract the uropod. We coexpressed Myosin-RFP and the filamentous actin-binding protein LimE-GFP from a single plasmid. A pipette with cAMP induces the expected movement of the cells upgradient with LimE-GFP localized in the front and Myosin-RFP in the rear of the cell ( and Video 4, available at ). Interestingly, the localization of LimE-GFP in the protruding front and Myosin-RFP in the retracting back is identical in cells stimulated with 8CPT-cAMP, except that the front is downgradient and cells move away from the pipette ( and Video 5). To investigate the mechanism by which 8CPT-cAMP induces negative chemotaxis, wild-type cells expressing the PIP detector PHcracGFP were stimulated with cAMP and 8CPT-cAMP. Similar to previous investigations (; ), a pipette with cAMP induces strong localization of PHcracGFP to the plasma membrane at the upgradient side of the cell. Pseudopodia are extended from PHcracGFP-containing areas and cells move upgradient toward the pipette ( and Video 6, available at ). 8CPT-cAMP also induces strong localization of PHcracGFP at the plasma membrane, but with opposite polarity compared with cAMP, which is downgradient ( and Video 7). Cells extend pseudopodia from these PHcracGFP-containing areas, and therefore move away from the pipette with 8CPT-cAMP. The size of the PHcracGFP patches induced by 8CPT-cAMP (9.0 ± 0.43 μm) is only slightly larger than the patches induced by cAMP (6.6 ± 0.17 μm), indicating that 8CPT-cAMP effectively reverses the PIP polarity. PIP is formed by PI3-kinase (PI3K) and degraded by PTEN that, in cAMP gradients, are localized at the leading edge and the rear of the cell, respectively. In 8CPT-cAMP gradients, the localization of PI3K and PTEN is reversed compared with cAMP gradients (). -null cells toward cAMP and 8CPT-cAMP. In -null cells, two PI3Ks are deleted that, together, mediate the vast majority of cAMP-stimulated PIP production (; ; ). These experiments are possible because PI3K is not essential for chemotaxis, and directional sensing can be mediated by other pathways (; ; ; ; ; ). shows that -null cells exhibit a good chemotactic response toward a pipette with cAMP (chemotaxis index is 0.80 ± 0.13). null cells do not exhibit a significant negative or positive response to 8CPT-cAMP (chemotaxis index is 0.11 ± 0.12). -null cells effectively move toward cAMP and are not inhibited by 8CPT-cAMP ( and Video 8, available at ), indicating that PI3K is essential for the repellent activity of 8CPT-cAMP and for the inhibitory effect of 8CPT-cAMP on cAMP chemoattraction. The molecular mechanism by which cAMP mediates PIP accumulation upgradient in cells has been well described. PI3K is activated and enriched upgradient in the cell, whereas the PIP-degrading enzyme PTEN strongly localizes downgradient in the cell (; ). PTEN has been demonstrated to bind to phosphatidylinositol-3,4,5-trisphosphate (PIP), suggesting that PIP is depleted upgradient in the cell (). This depletion of PIP could be induced by several nonexclusive methods, such as the observed conversion of PIP to PIP upgradient by PI3K (; ), but also by the conversion of PIP to InsP and DAG by PLC, which is known to be activated by cAMP (; ). We propose a mechanism by which 8CPT-cAMP could revert the polarity of chemotactic sensing that is based on the observation that cAMP stimulates PLC, whereas 8CPT-cAMP inhibits this enzyme (; ; ; supporting biochemical data are presented in Fig. S2, available at ). Upgradient stimulation of PLC by cAMP will lead to local depletion of PIP, and thereby prevent PTEN binding, by which the upgradient PIP accumulation is stabilized. In contrast, the upgradient inhibition of PLC by 8CPT-cAMP will lead to the local accumulation of PIP, thereby inducing PTEN binding and upgradient PIP degradation (). This relatively simple model for polarity reversal predicts that 8CPT-cAMP does not induce polarity switching in -null cells. cells contain a single gene encoding a PLCδ isoform (), which, like PI3K, is instrumental but not essential for chemotaxis (). Expression of GFP-tagged reporter proteins in -null cells reveal, as predicted, cytosolic localization of PH-cracGFP and enhanced PTEN-GFP expression at the membrane in cAMP and 8CPT-cAMP gradients (Fig. S3). As presented in and Video 9, -null cells show a similar chemotactic response toward 8CPT-cAMP as -null cells: they move in random directions in the presence of 8CPT-cAMP alone and, subsequently, move effectively toward an additional pipette with cAMP. This indicates that PLC is also essential for mediating the inhibitory effect of 8CPT-cAMP, as is PI3K. Finally, -null cells were investigated, showing that these cells are attracted toward cAMP, but are not repelled by 8CPT-cAMP (unpublished data). A scheme for PIP-mediated chemotaxis reversal by 8CPT-cAMP consists of three parts (). The basis is a PLC/PIP polarity switch. In , PLC is regulated by the activating Gα2 and inhibitory Gα1, which, in a gradient of attractant or repellent, will determine the polarity of the PIP gradient. The attractant cAMP shows predominant activation of PLC, leading to lower PIP levels upgradient, while the repellent 8CPT-cAMP inhibits PLC, leading to higher PIP levels upgradient. The resulting gradients of PIP and colocalized PTEN mediate opposite gradients of PIP, leading to the localized polymerization of actin. The gradients of localized PTEN and PI3K are stabilized because PTEN accumulates at the site of its product PIP, whereas PI3K accumulates at sites of its effector, PIP-induced F-actin. This mutually spatial exclusion of PI3K and PTEN will result in symmetry breaking, by which small spatial differences in the underlying polarity gradient can be amplified to the observed strong PIP gradient. Although PI3K and PLC are not essential for chemotaxis, the results clearly demonstrate that local formation of PIP is a very strong inducer of pseudopod formation, such that the cells can even move downgradient, overruling any upgradient signaling that 8CPT-cAMP may induce. In our model, a compound is a repellent because it binds to a receptor that is preferentially coupled to PLC via an inhibitory G protein, whereas it is an attractant when the receptor is coupled to a stimulatory G protein. The regulation of PLC by the stimulatory G2 and inhibitory G1 forms the basis for the polarity switch, and it allows the cell to respond to chemical gradient with repulsion or attraction. This polarity switch may be used by the cell during development. cells grow on bacteria. Cells starved for <1 h secrete unidentified compounds that induce repulsion of the cells, by which cells may find bacteria in a larger area (; ). Cells starved for ∼5 h secrete cAMP, to which they are attracted and which allows the cells to form a multicellular structure. Interestingly, G1 is expressed throughout development, whereas G2 is nearly absent during early starvation and expressed only after ∼4 h (). Thus, in early starved cells with the predominant inhibitory G1, the PLC–PI3K system is pruned for repulsion, whereas it becomes a system for attraction by expression of the stimulatory G2 during late starvation. The mechanism of polarity reversal of PLC–PI3K signaling could be instrumental in mammalian cells to navigate in complex chemotactic gradients. During development, many cells, such as neurons and gonads, are projected in the body by mixtures of attractants and repellents (; Schmitt et al., 2005). Observations on the action of Slit2 may be instrumental. Slit2 is a repellent for neuronal cells (; ). In contrast, Slit2 does not affect the direction of movement of vascular smooth muscle cells, but strongly inhibits PDGF-stimulated chemotaxis by inhibition of PDGF stimulation of Rac1 (; ). It is possible that, in neuronal cells, Slit2 induces a polar inhibition of Rac1, thereby inducing repulsion, whereas in vascular smooth muscle cells Slit2 induces uniform inhibition of Rac1 and is therefore not a repellent, but only an inhibitor of chemoattractants. Rac1 is known to be regulated by PIP3 in mammalian (; ) and () cells. The observed simplicity by which PLC-mediated polarity inversion of PI3K signaling in converts attraction to repulsion may provide a single mechanism to integrate complex positive and negative chemotactic signals during development. The plasmid pWF38 (PHcracGFP) expressing the 700-bp N-terminal PH domain of CRAC fused to GFP (), and plasmids expressing PI3K2-GFP (; ) and PTEN-GFP () were provided by P. Devreotes (Johns Hopkins University School of Medicine, Baltimore, MD). Plasmid 339-3 expressing mRFPmars () was provided by A. Muller-Taubenberger (Ludwig Maximilians University Munich, Munich, Germany). Plasmid pBIG-GFP-myo expressing a GFP fusion with myosin heavy chain II () was a gift from T. Egelhoff (Case Western Reserve University, Cleveland, OH). -null cells were provided by R. Firtel (University of California, San Diego, La Jolla, CA). Plasmid LB15B expressing LimE-GFP and Myo-RFP was constructed as follows. The neomycin resistance gene of MB74 was exchanged for the HPH hygromycin resistance gene that was preceded by an actin 15 promotor and terminated with a cabA terminator. The DNA coding for the actin-binding domain of LimE (aa 1–145) was cloned behind an actin 15 promoter and 5 adenosines, which serve as the Kozak sequence. It was followed by a SpeI site (coding for Thr and Ser) and the complete open reading frame of GFP (S65T variant), followed by a stop codon and an actin 8 terminator. This yielded the plasmid MB74hyg-LimE-GFP. The gene encoding the monomeric red fluorescent protein mRFPmars () was amplified by PCR on plasmid DNA. The gene was preceded by a NgoMIV site, an actin 15 promotor, and 5 adenosines, and was followed by a BamHI site (encoding Gly and Ser), the sequence encoding aa 2–2116 of myosin heavy chain, the myosin terminator from the vector pBIG-GFP-myo (), and a NgoMIV site. Finally, the gene encoding the mRFPmars-myosin fusion was released using the NgoMIV site and cloned into the single NgoMIV site of MB74hyg-LimE-GFP. The strain AX3 was used as wild-type control in all experiments. The mutants strains used are the -null strain 1.19 (), the -null strain GMP1 (), and -null cells (). Cells were grown in shaking culture in HG5 medium (containing per liter: 14.3 g oxoid peptone, 7.15 g bacto yeast extract, 1.36 g NaHPO× 12HO, 0.49 g KHPO, 10.0 g glucose) at a density between 5 × 10 and 6 × 10 cells/ml. Cells were harvested by centrifugation for 3 min at 300 g, washed in PB (10 mM KHPO/NaHPO, pH 6.5), and starved in PB in 6-well plates (Nunc) for 5h. Cells were then resuspended in PB, centrifuged, and washed once in PB, and resuspended in PB at a density of 6 × 10 cells/ml. Unless otherwise mentioned, digital images of cells in PB at room temperature were captured at 10-s time intervals over 45 min. Videos 1, 2, 4, and 5–7 were captured using a confocal laser scanning microscope (LSM 510 META-NLO; Carl Zeiss Microimaging, Inc.) equipped with a 63×/NA 1.4 objective (Plan-Apochromatic; Carl Zeiss Microimaging, Inc.). For excitation of the fluorochromes, GFP (S65T variant), and mRFPmars, a 488-nm argon/krypton laser and a 543-nm helium laser were used, respectively. The fluorescence was filtered through a BP500-530 IR and a LP560 filter, and was detected by a photomultiplier tube. The field of observation is 206 × 206 μm; Videos 1 and 2 present the phase-contrast channel, whereas the fluorescent channel is shown in Videos 4–7. For Videos 3 and 9, an inverted light microscope (Type CK40 with a LWD A240 20×/NA 0.4 objective; Olympus) fitted with a charge-coupled device camera (TK-C1381; JVC) was used. Digital images were captured on a PC using VirtualDub software and Indeo video 5.10 compression. The field of observation is 358 × 269 μm. Video 8 was captured using a 10× numerical aperture 0.25 objective, and presents a selected field of the same size, namely 358 × 269 μm. For all individual videos, specific time periods were selected that start at the moment the pipette was lowered to the plane just above the cells. In the phase-contrast videos, the pipette tips are visible as dark triangular shadows. In the fluorescence videos (Videos 4–7), the place of the pipette tip is indicated with an asterisk. The chemotaxis index, which is defined as the ratio of the cell displacement in the direction of the gradient and its total traveled distance, was determined for ∼25 cells in a video, as follows. First, the position of the centroid of a cell was determined with ImageJ (National Institutes of Health; ) for frames at 30-s intervals, yielding a series of coordinates for that cell. Using these coordinates, the chemotaxis index of each 30-s step was calculated and averaged, yielding the chemotaxis index for that cell in the movie. The data shown are the average and SEM of the chemotaxis indices from at least three independent experiments, with ∼25 cells per experiment. Fig. S1 shows cell trajectories of wild-type cells in a gradient of cAMP and 8CPT-cAMP, revealing that cells are attracted toward cAMP, but repelled from 8CPT-cAMP. Fig. S2 shows inhibition of PLC signaling by the antagonist 3′NH-cAMP. 8CPT-cAMP has similar properties to 3′NH-cAMP. Fig. S3 shows the localization of PHcrac-GFP, PTEN-GFP, and PI3K-GFP in -null cells in a gradient of cAMP or 8CPT-cAMP. Video 1 shows chemotaxis toward a pipette with cAMP. Video 2 shows chemotaxis away from a pipette with 8CPT-cAMP. Video 3 shows cell movement in gradients of 8CPT-cAMP and cAMP+8CPT-cAMP, followed by movement in only cAMP. Video 4 shows the localization of F-actin at the leading edge and myosin in the back of cells chemotaxing toward cAMP. Video 5 shows the localization of F-actin at the leading edge and myosin in the back of cells chemotaxing away from 8CPT-cAMP. Video 6 shows the localization of PHcracGFP (detecting PIP) at the leading edge of cells chemotaxing toward cAMP. Video 7 shows the localization of PHcracGFP (detecting PIP) at the leading edge of cells chemotaxing away from 8CPT-cAMP. Video 8 shows chemotaxis of -null cells toward cAMP in the presence of 8CPT-cAMP. Video 9 shows chemotaxis of -null cells toward cAMP in the presence of 8CPT-cAMP. The online version of this article is available at .

In eukaryotes, sister chromatid cohesion depends on the ringlike cohesin complex, consisting of four subunits (SMC1, SMC3, SCC3, and SCC1/MCD1/RAD21 or the α-kleisin). Cohesins are recruited to chromosomes before DNA replication, a process that requires the SCC2–SCC4 complex and the assembly of preRC (; ; ; ; ; ). However, the recruitment alone is not sufficient for sister chromatid cohesion because yeast mutants lacking the acetyltransferase Eco1/Ctf7 (or Eso1 in fission yeast) exhibit defective cohesion despite cohesins continuing to localize to the chromosomes (; ; ). The mechanism involving Eco1/Ctf7 seems to be conserved, as its orthologues have been identified in () and humans (; ). Eco1/Ctf7 family proteins exhibit acetyltransferase activity and modify several cohesion proteins in vitro (; ; ). However, the acetyltransferase activity may not be required for sister chromatid cohesion (; ). It has been suggested that the interactions with other proteins, rather than the acetyltransferase activity, is important for sister chromatid cohesion. In budding yeast, Eco1/Ctf7 interacts genetically and physically with many proteins involved in DNA replication (; ; ; ; ), and its physical interaction with PCNA is required for sister chromatid cohesion (). After it is established, sister chromatid cohesion is maintained until anaphase. In yeast, cohesins locate along the entire chromosome in S, G2, and M phase and hold sister chromatids together along their entire length. However, in higher eukaryotes, most of the cohesins are removed from the chromosome arms in prophase by the so-called “prophase pathway” (). This step of cohesin removal depends on Wapl (; ) and the phosphorylation of the SA1/2 subunit of the cohesin complex (), which is catalyzed by the pololike kinase 1 (Plk1) and aurora B kinase. The cohesins at the kinetochores and at some heterochromatin regions are protected from this prophase pathway. Proteins such as Sgo1, PP2A, and Bub1 have been implicated in this protection (; ; , ; ; ). The second step of cohesin removal is catalyzed by separase, which cleaves the α-kleisin and triggers the final separation of sister chromatids in anaphase. Interestingly, a genetic screen in revealed a putative acetyltransferase called San (). San is essential for life and the mutant exhibits apparent sister chromatid cohesion defects. In , San associates with Nat1 and Ard1, both of which are subunits of the N-terminal acetyltransferase A (NatA), which is conserved from yeast to human (). NatA has been extensively characterized in yeast, and no cohesion phenotype has been reported. Furthermore, the closest yeast homologue to San is Nat5. Although Nat5 is also found in the yeast NatA complex, its deletion causes no detectable phenotype (). Therefore, the requirement of San-like protein for sister chromatid cohesion is not conserved in budding yeast. In this study, we found that depletion of San in HeLa cells also causes precocious sister chromatid separation. The depletion causes cohesin to dissociate from the centromeres in mitosis without affecting the localization of Sgo1. Different from the findings in , most San proteins do not associate with the NatA complex and, unlike San, the NatA complex is not required for sister chromatid cohesion. In addition, recombinant San exhibits acetyltransferase activity on its own and is capable of modifying several chromosome-associated proteins. Rescue experiments indicate that the enzymatic activity of San is required for sister chromatid cohesion. Cells depleted of San exhibit normal Sgo1 localization, but no detectable cohesin complexes at the mitotic centromeres. Finally, depletion of Plk1 rescued sister chromatid cohesion in the San-depleted cells along the chromosome arms, but not at or near the centromeres. This result indicates that San is not required for the establishment of the cohesion along the chromosome arms. It is, however, necessary for centromeric cohesion in human cells. A single human cDNA (gi 13376735) was identified by BLAST search using the protein sequence of San. The full-length cDNA, encoding a protein of 169 aa, was cloned from a human fetal thymus cDNA library (CLONTECH Laboratories, Inc.) by PCR and the sequence was confirmed by analyzing at least three different clones (unpublished data). This sequence is identical to the recently described human homologue of San (). Full-length recombinant San was produced in , polyclonal antibody was generated using the recombinant San as the antigen, and the antiserum was affinity purified before use. As shown in Fig. S1 (available at ), the antibody detected a single protein band of ∼20 kD in HTC116, 293T, and HeLa cell lysates, which comigrated with the in vitro–synthesized protein in wheat germ extract. To investigate whether the loss of San causes any sister chromatid cohesion defect, we depleted endogenous San from HeLa cells by siRNA (). Mitotic chromosome spreads were prepared and analyzed. Remarkably, ∼78% of the spreads prepared from the San-depleted cells exhibited precocious chromatid separation, whereas only ∼14% of the mock-treated controls showed a similar phenotype (). In addition to the dramatic increase in the unpaired chromatids, San-siRNA cells also exhibited a noticeable delay in cell division. FACS analysis showed that San-siRNA cells of 4N DNA content accumulated from 15 to 60% (). Indirect immunofluorescent microscopy revealed that the incidences of multipole spindles (type II) and scattering chromosomes (type III) increased from 10 to 48% and from 4 to 30% of the mitotic cells, respectively (). The enrichment of 4N cells is likely caused by metaphase arrest because the level of cyclin B in these cells remained high, whereas the level of cyclin A was low (Fig. S2, available at ). All these phenotypes indicate that depletion of San causes premature sister chromatid separation in HeLa cells. In , San was found exclusively in the NatA complex, implying a role of the N-terminal acetyltransferase activity in sister chromatid cohesion (). To investigate any role of the human NatA complex in sister chromatid cohesion, we performed loss-of-function studies of the two subunits of the complex, NatH and Ard1 (; ). Depletion of Ard1 and NatH by siRNA caused a rapid loss of viability in 3 and 5 d, respectively (unpublished data). This was supported by the FACS analysis, which revealed many cells with DNA content smaller than 2N (). The lethality suggested that the depletion was effective. For the purpose of direct comparison with the result obtained in San-siRNA cells, we performed siRNA-depletion for both Ard1 and NatH and prepared chromosome spreads 5 d after the first transfection of the siRNA oligonucleotides. The depletion effect was further confirmed by Western blot (). In all cases, we did not observe any cohesion defects (). In fact, there were less unpaired sister chromatids than the mock-transfected controls. Therefore, it seems that the NatA complex is not involved in sister chromatid cohesion. However, we could not exclude the possibility that the essential function of the NatA complex, which causes lethality in its absence, overshadows its role in sister chromatid cohesion in these analyses. Interestingly, siRNA-depletion of NatH reproducibly reduced the expression levels of both San and Ard1 (). It seems that the optimal protein levels of San and Ard1 depend on NatH. In addition, in approximately half of the depletion experiments, a reduction of San expression was also observed in the cells depleted of Ard1, such as the one shown in . This was likely caused by the observation that cells depleted of Ard1 were dying rapidly and San might be degraded in some experiments, depending on when the samples were harvested. In all cases, depletion of San did not affect the expression of Ard1 or NatH. Because depletions of San and Ard1/NatH caused different phenotypes, we asked whether San associates with the NatA complex in HeLa cells. To this end, we fused HA-tag to the N terminus of San and overexpressed it in 293T cells. The interactions between HA–San and endogenous NatH and Ard1 were analyzed by coimmunoprecipitation with anti-HA beads. Both NatH and Ard1 precipitated with HA-San and the amounts pulled down were >10% of their respective inputs (, lane 9). Similarly, both San and Ard1 coimmunoprecipitated with overexpressed HA-NatH (, lane 3); and both San and NatH coimmunoprecipitated with overexpressed HA-Ard1 (, lane 6). The interaction seemed to be stable, as it survived stringent washing conditions with up to 500 mM NaCl. A similar observation has recently been reported using 293 cells (). However, the amounts of endogenous San pulled down by HA-NatH and HA-Ard1 were noticeably <10% of their respective inputs, despite the fact that the overexpressed HA-NatH and HA-Ard1 were virtually depleted by HA-beads (, lanes 2 and 5, respectively). This suggested that a significant pool of San might not associate with the NatA complex. To further demonstrate whether San also exists outside of the NatA complex, we performed sucrose gradient analysis () and size exclusion chromatography () to examine whether endogenous San cofractionates with the NatA complex. In both analyses, NatH and Ard1 were detected in the same fractions. On the other hand, over 80% of San was found in the fractions excluding the NatA complex. This indicates that the majority of San either is not in the NatA complex or only weakly associates with the complex. Together with the loss-of-function studies, it appears that San mediates sister chromatid cohesion independent of the NatA complex. To determine whether San is an acetyltransferase independent of the NatA complex, we incubated purified chromosome pellet with recombinant San in the acetyltransferase activity assay. Immunoblot analysis failed to detect NatH and Ard1 in these chromosome pellets, which was expected because both proteins localize primarily in the cytoplasm. As expected, many proteins were acetylated in a San-dependent manner (, lane 2), although their identities have not yet been determined. Furthermore, San is also autoacetylated (, lane 3), which is a common feature of many acetyltransferases. Collectively, these results indicate that San, by itself, is an acetyltransferase. To generate a mutant San defective in acetyltransferase activity, we substituted tyrosine-124 with phenylalanine because this tyrosine is conserved among several well-characterized acetyltransferases. Based on the atomic structure studies, mutational analyses, and/or enzymatic kinetics studies of serotonin -acetyltransferase (SNAT), aminoglycoside 6′--acetyltransferase (ACC6), and HPA1/2 (; ; ), this tyrosine is required for catalysis. Furthermore, SNAT is predicted to be the closest structural homologue of San by 3D-PSSM (). As expected, this mutation reduced the level of autoacetylation by approximately threefold, as determined by quantifying the C signals of lanes 3 and 5 in . Similarly, the Y124F mutant only slightly increased the level of labeling above the background (, compare lane 4 with 1), and the C signals of lane 4 in the area above San were reduced by approximately ninefold. The difference in reduction between the autoacetylation and the acetylation of the chromosome substrates can be explained if the autoacetylation occurs via an intramolecular mechanism. Unlike an intermolecular mechanism, the reaction rate of an intramolecular mechanism will not be affected by the diminishing concentration of the substrate. Therefore, during the 1-h incubation time, the autoacetylation could be faster and more complete than the acetylation of the chromosome substrates. To determine whether autoacetylation occurs via an intramolecular mechanism, we constructed a mutant San lacking the C-terminal 10 aa (ΔC10). This mutant migrates faster than the full-length San on SDS-PAGE and remains active in vitro (, lanes 6 and 7). The mutant also rescued sister chromatid cohesion when expressed in the San-depleted cells (unpublished data). When incubated with ΔC10, the catalytic-defective Y124F mutant remained unlabeled (, lane 8), indicating that the autoacetylation, indeed, occurs via an intramolecular mechanism. Therefore, we have constructed a catalytic-defective San mutant by a single substitution at the conserved tyrosine-124. To determine whether the acetyltransferase activity of San is required for sister chromatid cohesion, we performed a rescue experiment. To this end, we first constructed a stable cell line where the expression of a shRNA targeting San was induced by doxycycline. The design of the shRNA is identical to the siRNA oligonucleotides that successfully deplete San in HeLa cells (). After screening ∼200 stable lines, one San-shRNA cell line that was integrated with three copies of shRNA units was isolated. After induction for ∼4–5 d, the level of endogenous San was reduced ∼90% (). The depletion increased the percentage of the mitotic spreads with unpaired sister chromatids from 9 to 57% (; P = 3 × 10; = 5 in a paired test). This, again, confirmed that San is required for sister chromatid cohesion. Using this conditional shRNA cell line, we tested whether sister chromatid cohesion could be rescued by the wild-type or Y124F mutant San. The Y124F mutant was chosen because it is based on an established substitution that has been shown to specifically affect catalysis in several acetyltransferases. The cells were transfected with the scrambled rescue constructs (his-San* and his6-San-Y124F*) at the time of induction. After two rounds of transfections with 0.8 μg of the rescue construct, cells were harvested on day five. As shown in , transfecting GFP did not rescue the cohesion defect (48 vs. 57%; P = 0.09; = 5). On the other hand, expression of the wild-type San significantly reduced the cohesion defect from 48 to 19% (P = 0.002; = 5). The <100% transfection efficiency was likely responsible for not reducing the defects to the background level of 9% detected in the uninduced cells. On the other hand, the Y124F mutant only partially reduced the defects to 33%, which is not significantly different from 48% obtained with GFP (P = 0.06; = 5), but significantly greater than 19% achieved by the wild type (P = 0.004; = 5). The partial rescue is most likely caused by the residual activity of the Y124F mutant described in . The effect of this residual activity might be further augmented by the high expression level of this mutant (, compare lane 6 with 10). We titrated the amount of transfected DNA from 0.2–1.2 μg and measured the rescue effects (). Within this range of DNA, we did not detect any variations in transfection efficiency by transfecting GFP under the same vector (unpublished data). With the same amount of DNA, the expression level of the Y124F mutant was reproducibly higher than that of the wild-type San, perhaps caused by a difference in mRNA or protein stability. To compare the effects of the rescue constructs expressing at the same level, we measured the signals of San and actin with a densitometry reader, calculated the ratio of the signals of San to actin, normalized the ratios derived from the rescue constructs to the ratio of endogenous San, and plotted the rescue effects of the wild-type and mutant San against their normalized expression levels (). When the wild-type San was expressed at a slightly higher rate than the endogenous San, sister chromatid cohesion was rescued with only ∼20% the spreads exhibiting unpaired chromatids. At the same expression level, the Y124F mutant had little, if any, effect. When the expression of Y124F increased to approximately threefold of the endogenous San, we observed a rescue effect similar to that achieved by the wild-type San expressing near the endogenous level. Therefore, the Y124F mutant is approximately threefold less effective at rescuing sister chromatid cohesion in San-siRNA cells. Remarkably, the Y124F mutant is also threefold less active than the wild-type San, as determined by autoacetylation (). Therefore, we conclude that the acetyltransferase activity of San is required for its function in mediating stable sister chromatid cohesion. Although the results from the aforementioned experiments indicate that San and its acetyltransferase activity are required for stable sister chromatid cohesion in HeLa cells, the mechanism remains elusive. Delineating the regulation of San may shed some light on this. To this end, we analyzed the expression levels of San in different phases of the cell cycle. HeLa cells were synchronized at the G1/S transition and c-metaphase by double-thymidine and thymidine–nocodazole treatment, respectively. Next, the cultures were synchronously released into the cell cycle. Cells were withdrawn from the cultures at various time points, and the levels of San and an array of cell cycle markers were determined by immunoblot assay. As shown in (A and B), the expression levels of San are constant throughout the cell cycle. Next, we tested whether the cellular localization of San is regulated. We performed cellular fractionation and detect San only in the cytoplasmic fraction (). To confirm this localization, we also performed indirect immunofluorescence microscopy using our affinity-purified polyclonal antibody. As shown in , the antibody is specific to San because it detects strong signals in the uninduced San-shRNA cells (−Dox), but only weak signals upon induction (+Dox). Enlarged images are shown in the insets, and, consistent with the cellular fractionation, San signals were detected in the cytoplasm in the uninduced cells. On the contrary, the weak background signals distributed uniformly in the induced cells. Using this antibody, we examined the localization of San in various phases of the cell cycle. San localizes to the cytoplasm in interphase and is excluded from chromosomes in metaphase and anaphase (). Similar observations were confirmed with HA-tagged San, which was transiently overexpressed in HeLa cells (unpublished data). Although we could not exclude the possibility that a small undetectable fraction of San is inside the nucleus, the cytoplasmic localization implies that San may facilitate sister chromatid cohesion either directly and only during mitosis after the breakdown of the nuclear envelope or indirectly in interphase by acetylating a cytoplasmic factor, which is shuttled into the nucleus. Studies in suggested that San might be involved in the centromeric cohesion. To investigate whether the same is true in HeLa cells, we analyzed whether the localization of Sgo1, which is one of the factors required for cohesion at the centromeres, was affected by depletion of San. Using an antibody described previously (), we examined the localization of Sgo1 in the San-shRNA cell line under both the induced and uninduced conditions. We first confirmed the previous finding that Sgo1 localizes to the centromeres. As shown in , the nuclear localization of shugoshin was detected in some, but not all, interphase cells. This is consistent with the fact that Sgo1 is a substrate of the anaphase-promoting complex (APC/C) and is degraded only in early G1 cells (). In prophase cells, punctuated staining of Sgo1 was detected, which roughly colocalized with the kinetochores illuminated by crest serum. Centromeric Sgo1 was also detected in metaphase cells. However, in early anaphase, although the overall levels of Sgo1 remained high, the signals on chromosomes had greatly diminished. By late anaphase or telophase, Sgo1 was barely detectable. Similar analysis revealed the same dynamics of Sgo1 in cells depleted of San. As shown in , we detected strong centromeric Sgo1 signals in the cells with either multipolar spindles or scattering chromosomes (), indicating that, as in , Sgo1 localizes to the centromeres in a San- independent manner. Next, we determined whether the centromeric localization of the cohesin complexes was affected by depletion of San. To this end, we used a stable cell line expressing a C-terminal GFP-tagged SMC1. The tagged SMC1 is a good reporter for the cohesin dynamics for the following reasons. First, the expression of SMC1-GFP is well below the endogenous level of SMC1 (Fig. S3 A, lane 1, available at ). Second, SMC1-GFP interacts with SCC1 in a coimmunoprecipitation assay (Fig. S3 A, lane 2) and cosediments with the cohesin complex on a sucrose gradient (Fig. S3 B). Third, SMC1-GFP localizes to chromosomes in interphase, and most of them disassociate from chromosomes in mitosis ( and Fig. S3 C). Fourth, the chromosome localization of SMC-GFP depends on the presence of SCC1 (). Finally, SMC1-GFP can be detected at or near the centromeres in mitosis (). All of these observations are expected for the cohesin complex. As shown in Fig. S3 E, we were able to effectively deplete San in this cell line without affecting the protein levels of SCC1 and SMC1-GFP. The depletion increased the percentage of the mitotic spreads with unpaired chromatids from 2 to 58%. Next, the depleted cells were extracted with detergent and fixed to analyze the localization of SMC1-GFP. As shown in , SMC1-GFP localized to chromosomes during interphase in both mock and San siRNA-depleted cells. Therefore, San is not required for the overall association of the cohesin complex with chromosomes. On the other hand, SMC1-GFP was no longer detected on the mitotic chromosomes in the absence of San (). By analyzing the mitotic chromosome spreads (), cohesin was no longer detected at the centromeres in San-siRNA cells. Therefore, it seems that San is required for the mitotic localization of cohesin at the centromeres. To determine whether San is specifically involved in centromeric cohesion or is also required for cohesion at chromosome arms, we inactivated the prophase pathway so that cohesion at chromosome arms could be examined in mitosis. To this end, both Plk1 and San were depleted separately and simultaneously (), and the resulting chromosome spreads were analyzed. We observed three major types of mitotic chromosome spreads. The “paired” type contained mostly the paired chromatids. This type could be further classified into two subtypes. One consisted of the butterfly-shaped chromosomes observed mostly in the mock-treated samples (, Mock). The other subtype had tightly paired chromosomes with cohesion along their entire length. This subtype is observed in the Plk1-depleted samples (, Plk1). The “separated” type contained mostly completely separated chromatids observed in the San-depleted samples (, San). In the double-depleted samples, many of the spreads contained chromatids that were separated only in the middle of the chromosomes (, San&Plk1). This type of spread was called the “puffed” type. Remarkably, the separated region always contained the centromeres, as revealed by staining the spreads with the crest serum and Sgo1, which illuminate the kinetochores (Fig. S4, available at ). The distances between the kinetochores of these puffed chromatids in the San&Plk1-depleted samples were significantly extended compared with the paired chromatids in the mock and Plk1-depleted samples (). Therefore, the “puffed” type represents the chromosomes with defective cohesion specifically at their centromeres. As expected, Plk1 depletion resulted in 83% of the mitotic spreads displaying mostly tightly paired chromosomes (). This depletion also reproducibly caused 10% of the “puffed” type, which may reflect a background level of defective cohesion at the centromeres in HeLa cells. In the absence of cohesion at the arms, this background defect would have contributed to the “separated” type observed in 21% of the mock-treated cells. In addition, we also detected ∼7% of the “separated” type in the Plk1 depletion samples. These cells grossly failed to establish or maintain sister chromatid cohesion. Remarkably, double depletion of San and Plk1 significantly reduced the “separated” type from 60 to 14%in the San-depleted samples (P = 0.001; = 3). At the same time, the “puffed” type significantly increased from 10% (the background level in the Plk1-depleted cells) to 36% (P = 0.03; = 3) and the “paired” type slightly increased from 38% (mostly the butterfly-shaped chromosomes) to 50% (mostly the tightly paired chromosomes; P = 0.08; = 3). The rescue of cohesion at the chromosome arms indicates that San is not required for establishment or maintenance of cohesion at these regions. On the other hand, the failure to rescue the centromeric cohesion suggests that San is necessary for the establishment and/or maintenance of the cohesion at the centromeres. Depletion of San causes sister chromatids to separate prematurely. This result can be interpreted in two ways. The simplest explanation is that San is required for sister chromatid cohesion. Alternatively, depletion of San may cause prolonged mitotic arrest, which allows enough time for the prophase pathway to remove even centromeric cohesion. We consider the second explanation unlikely for the following two reasons: first, the actual length of mitotic arrest was <12 h in our experiments because we removed most of the loosely attached mitotic cells by a shake-off during a medium change 12 h before harvesting the cells; second, the prophase pathway mainly depends on Plk1 to remove cohesins from chromosomes. However, in the double-depletion experiment, cohesion between the chromosome arms is rescued, whereas the centromeres remain separated (). This directly argues against any significant role of the prophase pathway in mediating the premature chromatid separation in cells depleted of San. Collectively, these data strongly support that San is required for stable sister chromatid cohesion. Because the homologue of San is required for cohesion in , but not in budding yeast, this function of San may be conserved among only metazoans. The implication of San in sister chromatid cohesion again raises the question of whether the activity itself is required. To test this, we generated a catalytic-defective mutant San by a single mutation at the conserved tyrosine-124. Based on the extensive studies of several established acetyltransferases, this conserved tyrosine is involved in catalysis (; ; ). As expected, the Y124F substitution reduces the autoacetylation of San about threefold. Because the autoacetylation occurs via an intramolecular mechanism (), a reduction of autoacetylation indicates a reduction in the enzymatic activity rather than substrate interaction. Furthermore, although the residual activity is less desirable for the rescue experiment, it does indicate that the substitution does not grossly disrupt the conformation of San. Consistent with the notion that the acetyltransferase activity of San is required for sister chromatid cohesion, the Y124F mutant, even expressed at a higher level, failed to rescue the cohesion as efficiently as the wild-type San (). Collectively, our results strongly suggest that the acetyltransferase activity of San is required for its function in sister chromatid cohesion. In HeLa cells, the localization of Sgo1 appears unchanged after the depletion of San (). This is similar to what was reported in (). Nonetheless, the centromeric cohesion is compromised based on the “puffed” phenotype in cells depleted of both San and Plk1 and the lack of cohesin at the mitotic centromeres in cells depleted of San. How does San mediate the centromeric cohesion? One possibility is that San is required for the establishment of sister chromatid cohesion specifically at the centromeres. In the absence of San, the centromeric cohesion may never be properly established, thus cannot be rescued by inactivating the prophase pathway. This scenario is attractive because genetic studies suggest that acetyltransferase Eco1/Ctf7 may be involved in the establishment of sister chromatid cohesion during S phase. It is possible that in metazoans, the establishment of cohesion may use two different acetyltransferases. However, San seems to localize to the cytoplasm in interphase cells (). This localization is also conserved in (). In contrast to San, the members of the Eco1/Ctf7 family localize directly on interphase chromosomes in yeast and human cells (; ; ). Therefore, it is unlikely that San plays a direct role in the establishment of sister chromatid cohesion in interphase. It may, however, modify a cohesion establishment factor in the cytoplasm, which is then shuttled into the nucleus. Alternatively, San may be involved in the maintenance of the centromeric cohesion in mitosis. This is consistent with the apparently normal cohesin localization in interphase cells (). If this is the case, the epistasis of the San and Plk1 depletions suggests that San may function downstream of Plk1. However, because Plk1 seems to phosphorylate SA2 directly to remove cohesin from chromosomes (), a linear pathway, which places San between Plk1 and cohesin, becomes less plausible. On the other hand, San and Plk1 may work independently to stabilize cohesins on the mitotic centromeres, in which both the dephosphorylated status of SA2 and the acetylated-status of a San substrate are required for stable centromeric cohesion. This scenario will satisfy the epistasis described in . Interestingly, a recent study demonstrated that a phosphorylation of histone H3 may be required for sister chromatid cohesion in a Sgo1-independent manner (). It is possible that San may also mediate this histone phosphorylation event. The identification of the “cohesion substrate” of San will ultimately shed light on the mechanism. The polyclonal rabbit antibody to human San was raised by Genemed Synthesis, Inc. using His-tagged recombinant San as the antigen. The resulting crude serum was affinity-purified before use in immunoblot and indirect immunofluorescent staining. Antibodies to NatH and Ard1 were provided by J.R. Lillehaug (University of Bergen, Bergen, Norway; ). Antibody to Sgo1 was a gift from H. Yu (University of Texas Southwestern Medical Center, Dallas, TX; ). Antibodies to securin () and ESCO2 () have been previously described. Antibodies to cyclin B1, cyclin A2, α-tubulin, phospho-H3, and topoisomerase II-α were purchased from Santa Cruz Biotechnology. The CREST serum was purchased from Immunovision. FITC, Cy3, and Alexa Fluor 488–labeled secondary antibodies were purchased from Invitrogen. HeLa and 293T cells were grown in DME, whereas HCT116 cells were cultured in McCoy's 5A, both supplemented with 10% FBS. The GFP-SMC1–stable HeLa cell line was maintained in DME containing 0.5 mg/ml G418 (Sigma-Aldrich). HeLa cells were synchronized at G1/S or c-metaphase by double thymidine block or thymidine–nocodazole arrest, respectively (). The cellular fractionation was performed as previously described (). To knockdown San and NatH transiently in HeLa cells, two consecutive transfections were performed on days one and two by the calcium phosphate method, and cells were harvested on day five for analysis. To deplete Plk1, a previously described RNA duplex () was introduced into HeLa cells on day four. The RNA oligonucleotides (sense/antisense) synthesized by Thermo Fisher Scientific to deplete NatH, Ard1, and San are ACCUUGGCUAUGAAAGGActt/AUCCUUUCAUAGCCAAGGUtt, UGGGAAGAUUGUGGGGUActt/AUACCCCACAAUCUUCCCAtt, and GACAAGUUCUACAAUUAGtt/AUCCUUGUAGAACUUGUCAtt, respectively. To deplete San from the SMC1-GFP cells, we used GGCUAGGAAUAGGAACUAAtt/UUAGUUCCUAUUCCUAGCCtt. The DNA oligonucleotides GATCCCGTGACAAGTTCTACAATTAGTTCAAGAGAATCCTTGTAGAACTTGTCATTTTTA and AGCTTAAAAATGACAAGTTCTACAAGGATTCTCTTGAACTAATTGTAGAACTTGTCACGG synthesized by IDT were annealed and cloned into BglII and HindIII sites of the pSuperior.puro vector from OligoEngine to generate a single-unit shRNA construct. The shRNA unit, containing an H1 promoter and San-shRNA flanked by XhoI and SalI sites, was amplified by PCR and cloned into the SalI site of the single-unit shRNA construct to generate the two-unit shRNA construct. This procedure was repeated again to generate the three-unit shRNA construct that was used in this study. The inducible San-shRNA construct and pcDNA6/TR (Invitrogen) were cotransfected into HeLa cells using Lipofectamine (Invitrogen) according to the manufacturer's protocol. The next day, cells were selected with 1 μg/ml puromycin and 3 μg/ml blasticidin. 3 wk later, individual clones were picked up, and the expression of shRNA was induced with 3 μg/ml doxycycline for 4 d. A clone showing the best knockdown effect was used for the rescue experiment. In the rescue experiment, the rescuing plasmids were transfected into the inducible San-shRNA cell line right before the induction. The second round of DNA transfection was performed the next day to boost the transfection efficiency, and cells were harvested 4 d later for analysis. All the rescue constructs were based on pCS2 and tagged with His tag. To make their transcripts resistant to San-shRNA, the DNA sequence “AATGACAAGTTCTACAAGGAT” in San cDNA was scrambled to “AAGAAATTTAAAGA,” which does not change the resulting amino acid sequence (the substitutions are underlined). The SMC1 gene, with GFP fused in-frame at its C-terminal end, was cloned into the pIRESneo3 vector (CLONTECH Laboratories, Inc.) for eukaryotic gene expression. The construct was transfected into HeLa cells, and a G418-resistant stable cell line expressing SMC1-GFP was generated and used for further experiments. The assay was performed as previously described (). The final concentration for C-labeled acetyl-CoA is ∼1 mM. The chromosome pellet was prepared as previously described (). The procedure to prepare mitotic chromosome spreads has been previously described (). To prepare the spreads for immunofluorescent staining, the cells were incubated in hypertonic buffer (60–70 mM KCl) for 10 min and cytospun onto the slides. For statistical comparison, the percentage of the mitotic spreads exhibiting various status of sister chromatid cohesion was calculated. After repeating at least three trials, the statistic significance was determined by a paired two-sided test and indicated as the P value. Fig. S1 shows the cloning of the human homologue of San. Fig. S2 shows that San-siRNA cells were arrested in metaphase. Fig. S3 shows that the endogenous San was depleted by siRNA in HeLa cells stably expressing SMC1-GFP, and that SMC1-GFP faithfully reported the dynamics of the cohesin complex. Fig. S4 shows that Sgo1 localizes next to the crest signals on the chromosome spreads with both paired and unpaired chromatids. The online version of this article is available at .

Balanced chromosome partitioning during anaphase relies on the prior establishment of sister chromatid cohesion, which takes place concomitantly to DNA replication. Sister chromatid cohesion is essential for bipolar attachment of chromosomes to the mitotic spindle and depends on a cohesin complex formed by the Smc1, Smc3, Mcd1/Scc1, and Scc3 proteins. The Pds5 protein binds less tightly to this core complex but also contributes to sister chromatid cohesion (). To undergo chromosome segregation in anaphase, cohesin must be removed from chromosomes. This occurs through two distinct routes in higher eukaryotes, where a “prophase pathway” involving Polo and Aurora B kinases promotes the dissociation of most cohesin from chromosome arms as they condense. The remaining cohesin is removed at the metaphase-to-anaphase transition by separase that operates the proteolytic cleavage of Mcd1, in turn stimulated by Polo kinase–dependent Mcd1 phosphorylation (; ). Such a prophase pathway does not seem to exist in budding and fission yeasts, where separase appears fully responsible for cohesin dissociation along the entire chromosome (). Because of its irreversible nature, sister chromatid separation is tightly regulated and inhibited by several checkpoint mechanisms. Separase activation, for example, is finely tuned by its association with securin, which acts both as a molecular chaperone contributing to separase activation and as an inhibitor of its protease activity (; ). Anaphase-promoting complex (APC)–dependent ubiquitylation of securin triggers its destruction, which is essential for anaphase onset (), and both DNA and spindle damage inhibit anaphase by stabilizing securin (). In budding yeast, the morphogenesis checkpoint prevents the onset of anaphase in case of budding defects or alterations of the actin cytoskeleton. This depends on the Swe1 kinase that triggers the inhibitory phosphorylation of Cdk1 (). By investigating how the morphogenesis checkpoint controls sister chromatid separation, we found that neither securin inactivation nor forced Mcd1 cleavage are sufficient to allow anaphase when the morphogenesis checkpoint is activated. Rather, the protein phosphatase PP2A associated with its regulatory subunit Cdc55 is necessary to inhibit sister chromatid separation under these circumstances. Altogether, our data highlight a novel mechanism for controlling sister chromatid severing and segregation that involves the PP2A-regulated release of cohesion. High levels of a truncated version of the budding yeast p21-activated kinase Cla4 (Cla4t) activate the morphogenesis checkpoint by inhibiting endogenous Cla4 and its paralogue Ste20 (), which share essential functions in bud neck formation, septin ring assembly, and cytokinesis (). Upon overexpression from the promoter, haploid yeast cells arrest with wide bud necks, replicated chromosomes, undivided nuclei, short metaphase spindles, and high levels of the securin Pds1 (). In addition, they markedly delay activation of the Polo kinase Cdc5 (Fig. S1, available at ), suggesting that they arrest in G2. As deletion is sufficient to allow anaphase in most mutants arresting in mitosis, we asked whether it could bypass the G2 arrest caused by high Cla4t levels. Elutriated G1 cells of a strain with four copies of the construct integrated in the genome (4X ) were released into the cell cycle in the presence of galactose. As expected, DNA replication () and bipolar spindle formation () took place normally in these conditions, whereas bud neck formation was abnormal because of overexpression (not depicted). Surprisingly, pericentromeric chromosomal sequences marked by a tet operator array that binds TetR-GFP () could not separate in these cells (), indicating that sister chromatid separation did not occur. Nuclear division and spindle elongation did not take place throughout the course of the experiment (), similar to 4X cells under the same conditions (). Thus, deletion of is not sufficient to bypass the G2 arrest caused by high levels of Cla4t. As shown in (D–F), latrunculin-A (Lat-A), which activates the morphogenesis checkpoint by depolymerizing the actin cytoskeleton, induced, like Cla4t, a securin-independent G2 arrest. In fact, cells released from a G1 arrest in the presence of Lat-A did not bud () but replicated DNA () and formed bipolar spindles (). However, neither wild-type nor cells underwent sister chromatid separation, nuclear division, or spindle elongation (). In contrast, the same events took place promptly in the morphogenesis checkpoint–defective cells, which also exited mitosis and entered a new round of DNA replication, as indicated by the appearance of 4C DNA contents (). Altogether, these data indicate that the morphogenesis checkpoint appears to prevent the onset of anaphase independently of securin. Besides its inhibitory function, securin also has a positive role in separase activation in several eukaryotic systems, prompting us to test whether Cla4t overproduction might impair Pds1 interaction with the Esp1 separase and/or Esp1 nuclear import. Wild-type, 4X , and 4X cells expressing HA-tagged Pds1 (Pds1-HA) and myc-tagged Esp1 (Esp1-myc18) were grown in raffinose, arrested in G1 by α-factor, and released in the presence of galactose, followed by the analysis of Pds1 and Esp1 nuclear localization and physical interaction. As shown in , budding was delayed in 4X cells compared with wild type, but kinetics of Pds1-HA and Esp1-myc18 nuclear accumulation were similar in the two strains. Thus, Pds1 can still act as an Esp1 molecular chaperone in the presence of high levels of Cla4t. Accordingly, similar levels of Esp1-myc18 were immunoprecipitated with Pds1-HA from both wild-type and 4X cell extracts (). Although lack of securin did not allow chromatid separation upon morphogenesis checkpoint activation, ectopic cohesin cleavage could be expected to trigger nuclear division in the same conditions. We engineered 4X cells to express a Mcd1–tobacco etch virus (TEV) variant, where the Esp1 cleavage site at position 268 is replaced by the recognition sequence for the TEV protease (). We then introduced in the same cells the TEV protease coding sequence under the control of the promoter. These cells grow normally under uninduced conditions because the Mcd1-TEV variant can be cleaved by separase at position 180, whereas it is cleaved and fully removed from chromosomes upon TEV induction even if separase is inactive. Small G1 cells of this strain were elutriated and released in the presence of galactose to trigger expression of both Cla4t and TEV. Remarkably, nuclear division did not take place (), suggesting that cohesin cleavage might be insufficient to allow chromosome segregation in 4X cells. Conversely, as previously reported (), cohesin cleavage by the TEV protease was sufficient to trigger anaphase in cells depleted for Cdc20 (), the APC regulatory subunit essential for Pds1 proteolysis and anaphase onset (). Thus, cohesin cleavage seems to be sufficient to trigger anaphase in metaphase-arrested cells but not in cells arrested in G2 by the morphogenesis checkpoint. Because it was formally possible that the lack of nuclear division in 4X cells was due to inefficient cohesin cleavage, we analyzed the kinetics of cohesin cleavage by the TEV protease in 4X versus wild-type cells after release from G1 in the presence of galactose. Full length of Mcd1-TEV tagged with 3 HA epitopes at the C terminus (Mcd1-HA3) and its cleavage product by separase (at position 180) were detectable in both strains in cycling cells and at time 0 (). Upon galactose addition, kinetics of TEV production, as well as appearance of the TEV-induced Mcd1-HA3 cleavage product (at position 268), were similar in the two strains. However, disappearance of full-length Mcd1 and its separase-induced cleavage product, which can both be cleaved by TEV, was slower in 4X than wild-type cells (). This might be due to delayed activation of the Polo/Cdc5 kinase, which stimulates Mcd1 cleavage (), in 4X versus wild-type cells. In spite of that, most, if not all, Mcd1-HA3 was cleaved by 3 h in 4X cells, but nuclear division occurred only in a small fraction of them (). In contrast, >75% of wild-type cells had accomplished nuclear division under the same conditions. Therefore, other mechanisms besides cohesin-mediated sister chromatid cohesion likely contribute to prevent chromosome segregation when the morphogenesis checkpoint is active. Because mitotic Cdks regulate spindle assembly and microtubule dynamics, the morphogenesis checkpoint might delay nuclear division through spindle misfunction. Upon bipolar attachment of sister kinetochores to microtubules, spindle forces overwhelm centromeric cohesion, leading to precocious separation of sister centromeres before anaphase (), thus providing a readout for spindle function. We found that sister centromeres of chromosome 15 could separate concomitantly with spindle formation in the presence of Lat-B (), suggesting that spindle forces are normal. Because kinetochore inactivation by the mutation prevents kinetochore–microtubule attachment without affecting spindle formation and elongation (), we also asked whether spindle elongation could take place in cells under morphogenesis checkpoint activation. We induced morphogenetic defects by using a temperature- sensitive mutation, which alters a guanine-nucleotide exchange factor for the GTPase Cdc42 that is required for budding (). Upon release of synchronized G1 cells at 37°C, cells arrested in G2 as unbudded with undivided nuclei and short metaphase spindles. Lack of kinetochore attachment in cells was sufficient to allow spindle elongation (), suggesting that spindle dynamics is not affected by morphogenetic defects. Therefore, residual sister chromatid cohesion, rather than a misfunctional spindle, is likely responsible for preventing chromosome segregation in the absence of Mcd1 upon morphogenesis checkpoint activation. Cdc55 is one of the two regulatory subunits of yeast protein phosphatase PP2A and was previously implicated in maintaining sister chromatid cohesion in response to spindle defects (). This prompted us to test whether deletion could allow sister chromatid separation in Cla4t-overexpressing cells. Elutriated G1 cells of a 4X strain carrying the tetO/tetR-GFP constructs for monitoring sister chromatid separation were released into the cell cycle in the presence of galactose. As shown in , deletion of partially rescued the cytokinetic defects caused by high Cla4t levels, indicated by reaccumulation of a small fraction of cells with 1C DNA contents at the end of the first cell cycle. Most cells, however, displayed abnormal bud necks characteristic of 4X cells. In spite of that, they underwent efficient sister chromatid separation and nuclear division (), suggesting that Cdc55 prevents anaphase onset when p21-activated kinases are inactive. Nuclear division could also be induced in 4X cells by expressing a mutant form of the Pph21 catalytic subunit (Pph21-L369Δ; Fig. S2, available at ) that was shown to preferentially fail to interact with Cdc55 (). Therefore, chromatid cohesion upon morphogenesis checkpoint activation requires the protein phosphatase PP2A bound to Cdc55. The catalytic and structural PP2A subunits can form mutually exclusive complexes with either one of the regulatory subunits Cdc55 and Rts1 (). PP2A and its human counterpart have recently been shown to prevent precocious dissociation of centromeres both in mitosis and in meiosis I (; ). In an experiment similar to the one described for , we found that pericentromeric sequences could not separate in the majority of 4X cells (). When pericentromeric regions did split (∼25% of the cells), GFP dots were always found very close to each other () and nuclear division was negligible (), suggesting that PP2A plays a minor role, compared with PP2A, in controlling chromatid cohesion under these circumstances. Because Cdc55 and Rts1 compete for binding to the other PP2A subunits, sister chromatid separation in the absence of Cdc55 could be ascribed to increased levels of the PP2A complex. To investigate this possibility, we asked whether 4X cells lacking both Cdc55 and Rts1 could undergo anaphase. Elutriated G1 cells of the 4X strain released in the presence of galactose progressed into the cell cycle very slowly, as a result of budding and replication defects (). In spite of that, those that could finish chromosome replication underwent efficient dissociation of sister chromatids and nuclear division (), suggesting that anaphase onset in 4X cells lacking Cdc55 is not due to increased levels of PP2A activity. We then asked whether PP2A also controls sister chromatid cohesion in other conditions that activate the morphogenesis checkpoint. Wild-type and cells were arrested in G1 by α-factor and then released in the presence of Lat-A. In these conditions, neither wild-type nor cells budded throughout the course of the experiment (). As expected, wild-type cells accumulated with 2C DNA contents, unsevered sister chromatids, undivided nuclei, and short metaphase spindles (). Strikingly, sister chromatids separated efficiently in cells under the same conditions, thus allowing spindles to elongate and nuclei to divide (). Finally, because deletion causes by itself morphogenetic defects and Swe1 stabilization at low temperatures (; ), we asked whether the mutant could separate sister chromatids at 16°C. At this temperature, cells showed prominent morphogenetic defects (not depicted), but nevertheless could split chromatids and divide nuclei, albeit with a delay compared with wild-type cells (). To directly compare the effects of cohesin inactivation and lack of PP2A on sister chromatid separation of cells with morphogenetic defects, we used the temperature-sensitive allele, which inactivates Mcd1 and advances sister chromatid separation relative to wild type at the restrictive temperature (). G1-arrested cells either lacking or carrying the allele were released at 37°C. cells could efficiently separate chromosome V arm sequences, although with a delay compared with cells, but did not elongate spindles or divide nuclei (). In contrast, cells underwent complete chromosome segregation under the same conditions (). Accordingly, the distance between separating chromatids at 150 min after release was significantly higher in cells than in cells (). Therefore, some residual chromatid cohesion likely persists even when cohesin is inactivated and PP2A plays a crucial role in controlling sister chromatid separation when the morphogenesis checkpoint is activated. Although ectopic cohesin cleavage did not allow nuclear division during morphogenesis checkpoint activation, deletion might still allow anaphase onset in these conditions through cohesin cleavage. To test this possibility, , , and cells were arrested in G1 by α-factor and then released at 37°C, followed by analysis of cell cycle parameters () and Mcd1 cleavage by separase (). As expected, cells arrested with 2C DNA contents, unseparated sister chromatids, and metaphase spindles, whereas most cells underwent anaphase and spindle elongation and eventually exited mitosis and rereplicated their chromosomes, accumulating DNA contents higher than 2C (), suggesting that lack of Swe1 overrides cells' ability to sense morphogenetic defects. Interestingly, cells could also undergo anaphase in the same conditions, albeit with a delay compared with cells, but remained mostly arrested with 2C DNA contents. The Mcd1 cleavage product, which was readily apparent in cells and preceded sister chromatid separation, was mostly negligible in cells (). Nevertheless, chromatin staining of Mcd1 after chromosome spreading revealed that cohesin remained bound to chromatin in wild-type cells (not depicted) but had dissociated from the chromosomes in nuclei of cells that underwent anaphase (). Thus, sister chromatid separation and Mcd1 dissociation from chromosomes in cells under morphogenesis checkpoint activation do not seem to correlate with separase-dependent cleavage of cohesin. Accordingly, the Mcd1 cleavage product was not detectable in 4X cells undergoing anaphase in the presence of galactose, similar to 4X cells (Fig. S3, available at ), and Mcd1 disappeared from the nuclei of 4X cells in anaphase (Fig. S3 E). Mcd1 displacement from chromatin did not correlate with increased Mcd1 phosphorylation, which could instead be detected as electrophoretic mobility shift in nocodazole-arrested cells (Fig. S3 D). It is interesting to note that deletion in Cla4t-overexpressing cells caused rapid Pds1 and Clb2 proteolysis, as well as appearance of the Mcd1 cleavage product, whereas Pds1 and Clb2 remained mostly stable upon deletion of (unpublished data). Unlike in cells under morphogenesis checkpoint activation, sister chromatid separation in nocodazole-treated cells was accompanied by Pds1 degradation Mcd1 cleavage, although with a delay compared with the spindle checkpoint–defective cells (Fig. S4, available at ). Therefore, PP2A contributes to maintaining sister chromatid cohesion in nocodazole by impinging on the same targets of the spindle assembly checkpoint, as recently suggested by others (). In contrast, PP2A likely prevents sister chromatid separation in G2 through a different mechanism. PP2A has recently been shown to prevent Cdc14 early anaphase release from the nucleolus through Net1 dephosphorylation (). Cdc14 can in turn trigger Pds1 proteolysis in nocodazole-arrested cells (), and this mechanism has been proposed to be responsible for the precocious dissociation of sister chromatids in nocodazole-treated cells (). We therefore asked whether Cdc14 was released from the nucleolus in cells with morphogenetic defects and necessary for their onset of anaphase. Wild-type and cells were arrested in G1 by α-factor and released in the presence of Lat-B. In situ immunostaining of Cdc14 showed that anaphase took place in cells before Cdc14 release from the nucleolus (). In addition, analysis of anaphase cells 150 min after release revealed that a high fraction of them (68.3%; = 120) had undergone anaphase with Cdc14 in the nucleolus (), suggesting that premature Cdc14 release is not responsible for sister chromatid separation in these cells. To test whether Cdc14 was required for the onset of anaphase in mutants with morphogenetic defects, we inactivated Cdc14 in cells with the temperature-sensitive allele. As a control for Cdc14 inactivation, we analyzed the subcellular localization of the Swi5 transcription factor, whose nuclear import in telophase is strictly dependent on its dephosphorylation by Cdc14 (). strains expressing a myc-tagged Swi5 protein were synchronized in G1 by α-factor and released at 37°C to analyze, over time, budding kinetics, Swi5 localization, and nuclear division. Swi5 was cytoplasmic in both strains throughout most of the cell cycle. cells, indicating that Cdc14 had been inactivated (). Lack of Cdc55 allowed a fraction of cells to divide nuclei irrespective of Cdc14 function (), indicating that Cdc14 is dispensable for the onset of anaphase in these conditions. Accordingly, Cdc14 was also insufficient to promote sister chromatid separation in - overexpressing cells carrying the dominant allele, which encodes a hyperactive Cdc14 variant with reduced affinity to its inhibitor Net1 (; ). Although we did not detect any increase in Mcd1 phosphorylation in versus wild-type cells overproducing Cla4t (Fig. S3), it was still possible that PP2A could prevent sister chromatid separation by counteracting the Cdc5-mediated phosphorylation of a small fraction of Mcd1 or other cohesin subunits. However, inactivation of Cdc5 with the temperature-sensitive allele did not prevent anaphase in cells (), suggesting that Cdc5 is not required for this process. Timely sister chromatid segregation, especially of ribosomal DNA and chromosome sequences far from centromeres, depends on condensin and DNA topoisomerase II (; ; ; ; ). We therefore tested the effects of the temperature-sensitive and mutations, affecting condensin and DNA topoisomerase II, respectively, on the unscheduled anaphase of cells. cells mostly prevented anaphase (), suggesting that the presence of topological linkages prevents sister chromatid separation under these conditions. Consistently, the presence of the allele could partially rescue the cold sensitivity of cells (), which is presumably due to unscheduled sister chromatid separation in the presence of morphogenetic defects. We then asked whether morphogenetic defects could arrest the cell cycle in a stage where topological linkages are not resolved, using an assay that allows detection of accumulation of catenated forms of a circular minichromosome (). Unlike mutants, however, neither () nor cells (not depicted) accumulated minichromosome topoisomers. Although we cannot exclude the possibility that the behavior of natural chromosomes is different from that of minichromosomes, the delay of nuclear division caused by the morphogenetic checkpoint does not seem to be accompanied by lack of decatenation. If PP2A acts as an inhibitor of sister chromatid separation, increasing its dosage might delay the onset of anaphase. We therefore introduced into the genome of otherwise wild-type cells multiple copies of a galactose-inducible construct. Parental and transformed strains growing in raffinose were arrested in G1 with α-factor and released in the presence of galactose. We then monitored separation of the tetO array located 13 kb away from , as well as spindle formation and elongation (). overexpression did not affect bipolar spindle formation but delayed sister chromatid separation, nuclear division, and spindle elongation, causing cells to accumulate in G2. This delay did not depend on functional securin, as deletion did not accelerate the onset of anaphase in -overexpressing cells. High levels of Cdc55 delayed sister chromatid separation at both pericentromeric and telomeric regions (unpublished data), suggesting that PP2A prevents dissociation of sister chromatids along their length. If PP2A acted as anaphase inhibitor independently of securin, we could expect that simultaneous loss of Pds1 and Cdc55 might have additive effects, allowing precocious separation of sister chromatids during the unperturbed cell cycle. Indeed, concomitant deletion of and turned out to be lethal (unpublished data). It has been well established that morphogenetic defects, such as lack of actin polarization or budding, cause a G2 arrest in budding yeast because of the inhibitory phosphorylation of Cdk1 on tyrosine 19 by the Swe1 kinase (). This inhibitory phosphorylation likely involves only a small pool of mitotic Cdks. In fact, the morphogenesis checkpoint arrests the cell cycle after spindle formation, whereas complete inactivation of all mitotic Cdks by mutations or overexpression prevents spindle pole body separation and bipolar spindle assembly (). We show here that the morphogenesis checkpoint prevents sister chromatid separation independently of securin because cells treated with Lat-A or overexpressing the dominant-negative allele do not attempt anaphase. Our data also indicate that morphogenesis checkpoint activation does not delay separase association to securin and its nuclear import, which depends on Pds1 phosphorylation by Cdks (; ), consistent with only a minor pool of mitotic Cdks being inactivated under these conditions. Inactivation of Mcd1 through the temperature-sensitive allele or its ectopic cleavage also turned out to be insufficient for anaphase and chromosome segregation under morphogenesis checkpoint activation, raising the possibility that either spindle function is compromised or residual cohesion persists on chromosomes after Mcd1 inactivation. Because in our assays spindle forces seem normal, we favor the second interpretation. Whether residual cohesion depends on other cohesin subunits or on other proteins remains to be established. Cohesin-independent chromatid linkages have been reported for repetitive sequences (; ; ), and a role for condensin in chromatid cohesion has been recently described (). Swe1-mediated phosphorylation of mitotic Cdks could prevent the release of these linkages in addition to inhibiting securin degradation. Although a direct role for mitotic Cdks in dismantling sister chromatid cohesion has not been reported so far, Cdks are required at different levels for Polo kinase activation, which in turn contributes to dissociation of sister chromatids by phosphorylating the cohesin Mcd1 and enhancing its susceptibility to cleavage by separase (). In addition, in higher eukaryotic cells Polo and Aurora B kinases promote the prophase pathway of cohesin dissociation from chromosome arms that is independent of securin degradation and relies on phosphorylation of the SA2 cohesin subunit (; ). In budding yeast, mitotic Cdks activate the Polo kinase through several mechanisms, including transcription (), phosphorylation (), and inhibition of proteolysis (). It is therefore not surprising that Cdc5 activation is dramatically delayed in response to the morphogenesis checkpoint. The failure to timely activate Cdc5 could contribute to the lack of sister separation in these conditions but cannot be the only culprit. In fact, Cdc5 inactivation leads to inefficient separation of telomeric regions but has no or little effect on that of centromeric and arm sequences (). In addition, Cdc5 is not required for the onset of anaphase of cells. If the failure to separate sister chromatids when the morphogenetic checkpoint is active were merely due to delayed Cdc5 activation, anaphase should be resumed by ectopic Mcd1 cleavage, which we show not to be the case. Therefore, sister chromatid cohesion seems to be maintained by the morphogenesis checkpoint through a previously unanticipated mechanism that does not depend only on securin stabilization and Polo kinase inactivation. We find that inactivation of the protein phosphatase PP2A is sufficient to allow sister chromatid separation when the morphogenesis checkpoint is activated. Unlike upon deletion of , which completely abolishes the cell's ability to respond to morphogenetic defects, this is not achieved through switch off of checkpoint signaling, because lack of PP2A activates by itself the checkpoint and induces Swe1 stabilization by causing morphogenetic defects (). In agreement with a critical function for PP2A in controlling sister separation when the morphogenesis checkpoint is active, deletion of turned out to be lethal for and mutants (unpublished data), whose morphogenesis defects are known to activate the checkpoint (). Recently, PP2A bound to Rts1/B56, the other regulatory subunit, has been found to protect centromeric cohesion during mitosis and meiosis I, in both yeast and human cells (; ). In our experimental conditions, PP2A seems to have only a minor role, perhaps restricted to centromeric regions, in preventing chromatid dissociation. Cdc55 was previously implicated in maintaining sister chromatid cohesion in response to activation of the spindle assembly checkpoint (), suggesting that PP2A acts as anaphase inhibitor in several conditions. However, in nocodazole-treated cells, sister chromatid separation is accompanied by Mcd1 proteolytic cleavage (; this study), whereas we could not find evidence for such event in cells undergoing anaphase in the presence of morphogenetic defects. In agreement with our data, Cdc55 has recently been shown to prevent chromatid separation independently of securin degradation and Mcd1 cleavage in cells with telomeric DNA lesions (). How could PP2A prevent sister chromatid separation in G2? For instance, it could regulate a pathway of cohesin removal similar to the prophase pathway of higher eukaryotic cells, although so far Mcd1 cleavage by separase seems to be the only necessary and sufficient event for cohesin removal from yeast chromosomes (). If PP2A were to inhibit cohesin dissociation independently of Mcd1 cleavage, its inactivation could allow anaphase in the absence of separase. In contrast to recently published data (), we find that both the mutation () and overexpression of nondegradable Pds1 () prevent cells from undergoing anaphase (unpublished data), suggesting that separase is still required for sister chromatid separation in the absence of PP2A. It should be noted, however, that separase has additional functions that are unrelated to its role in Mcd1 cleavage (; ; ; ). Interestingly, Cdc55 has recently been shown to interact physically with Esp1 and to prevent the early anaphase release of Cdc14 by causing dephosphorylation of its inhibitor Net1 (). This raises the possibility that lack of PP2A causes the unscheduled activation of Cdh1/APC, and thereby Pds1 degradation, by promoting Cdc14 release. Although this could partly explain the separation of sister chromatids in nocodazole-treated cells, we show here that nuclear division of cells when the morphogenesis checkpoint is active is independent of Cdc14 function, suggesting that PP2A must have other roles, besides inhibiting Cdc14 dissociation from Net1, before the onset of anaphase. Therefore, a more direct role of PP2A in controlling sister chromatid separation in G2 must be invoked. double mutants, where sister chromatid separation could be so premature as to cause lethal chromosome missegregation. In addition, overexpression delays chromatid dissociation independently of securin. In agreement with a crucial function as anaphase inhibitor, PP2A phosphatase activity decreases at the onset of anaphase (). An obvious candidate for being dephosphorylated by PP2A to prevent sister chromatid dissociation was Mcd1, especially in light of recent data indicating that the other PP2A complex, PP2A, prevents precocious loss of centromeric cohesion by counteracting Mcd1 phosphorylation by Polo kinase (; ). However, as discussed above, PP2A might target other proteins beside Mcd1. For instance, it could dephosphorylate other cohesin subunits and prevent cohesin unloading through a pathway analogous to the vertebrate prophase pathway. Despite the efforts, we could not detect any difference in the electrophoretic mobility of other cohesin subunits, such as Scc3 and Pds5, in versus wild-type cells (unpublished data). Alternatively, PP2A could regulate other chromatin-bound proteins, such as the condensin complex. It is worth mentioning that the human condensin HCP-6 interacts with and is dephosphorylated by PP2A bound to the B subunit (). Finally, another putative target of PP2A might be Esp1, which interacts physically with Cdc55 (). Although separase has been proposed to down-regulate PP2A activity, separase regulation of by PP2A can also be envisaged. In summary, a crucial role for PP2A in maintaining sister chromatid cohesion in response to several stress conditions is emerging, making it a key factor for preserving genome stability. Mutations in PP2A B subunit, the Cdc55 counterpart, cause chromosome segregation defects (), and mammalian PP2A is considered to be a principal guardian against malignant transformation (). Understanding the molecular mechanisms by which PP2A controls the onset of anaphase under different conditions might shed light on processes that prevent chromosome missegregation, which is intimately linked to tumorigenesis. All yeast strains (Table S1, available at ) were derivatives of or were backcrossed at least three times to W303 (). Cells were grown in YEP medium (1% yeast extract, 2% bactopeptone, and 50 mg/l adenine) supplemented with 2% glucose (YEPD), 2% raffinose (YEPR), or 2% raffinose and 1% galactose (YEPRG). Unless otherwise stated, α-factor, nocodazole, Lat-A, and Lat-B were used at 3 μg/ml, 15 μg/ml, 0.1 mM, and 0.2 mM, respectively. For galactose induction of α-factor–synchronized cells, galactose was added half an hour before release. cells were grown in synthetic medium lacking methionine, whereas the promoter was shut off by resuspending cells in YEPD medium supplemented with 2 mM methionine. To clone under the promoter (plasmid pSP376), a BglII–PstI PCR product containing the coding region and 140 bp of downstream sequence was cloned in the BamHI–PstI site of a –bearing YIplac211 vector. pSP376 integration was directed to the locus by BglII digestion. Copy number of the integrated plasmid was verified by Southern analysis. , , , and chromosomal deletion were generated by one-step gene replacement (). Immunoprecipitations were performed as described by ; lysis buffer was supplemented with 0.1% Triton X-100 (Fluka). Cdc5 kinase assays were performed according to . For Western blot analysis, TCA protein extracts were prepared according to . Nondenaturing protein extracts were prepared according to . Proteins transferred to Protran membranes (Schleicher and Schuell) were probed with 9E10 mAb for myc-tagged proteins, with 12CA5 or 16B12 mAb (Babco) for HA-tagged proteins, and with polyclonal antibodies against Clb2 and Swi6. Secondary antibodies were obtained from GE Healthcare, and proteins were detected by an enhanced chemiluminescence system according to the manufacturer. Flow cytometric DNA quantitation, in situ immunofluorescence, and chromosome spreads were performed according to . Nuclear division was scored with a fluorescent microscope on cells stained with propidium iodide. Visualization of Tet operators using GFP was performed as described in . Catenation assays were performed according to . To detect spindle formation and elongation, α-tubulin immunostaining was performed with the YOL34 monoclonal antibody (Serotec) followed by indirect immunofluorescence using rhodamine-conjugated anti-rat antibody (1:100; Pierce Chemical Co.). Cdc14 immunostaining was performed with sc-12045 polyclonal antibodies (Santa Cruz Biotechnology, Inc.) followed by indirect immunofluorescence using CY3-conjugated anti-goat antibody (GE Healthcare). Immunostaining of myc- and HA-tagged proteins was done by incubation with the 9E10 mAb and 16B12 mAb (Babco), respectively, followed by indirect immunofluorescence using CY3-conjugated goat anti–mouse antibody (GE Healthcare). Digital images were acquired on a fluorescent microscope (Eclipse E600; Nikon) equipped with a charge-coupled device camera (DC350F; Leica) at 20°C with an oil 100× 1.3 NA Plan Fluor objective (Nikon), using FW4000 software (Leica). Fig. S1 shows that Cdc5 protein levels and associated kinase are delayed in response to morphogenetic defects. Fig. S2 shows that mutations in the catalytic subunit of PP2A that impair its interaction with Cdc55 promote nuclear division when the morphogenesis checkpoint is activated. Fig. S3 shows that Mcd1 falls off chromatin but its proteolytic cleavage is undetectable in cells overexpressing Cla4t. Fig. S4 shows that sister separation in the absence of Cdc55 upon nocodazole treatment correlates with Mcd1 cleavage. Table S1 describes the genotypes of strains used in this work. Online supplemental material is available at .

ER-associated degradation (ERAD) is an important component of ER quality control whereby unwanted proteins that are misfolded, misassembled, or metabolically excessive are recognized and returned to the cytosol by a process called retrotranslocation or dislocation (; ; ; ). Once exposed to the cytosol, ERAD-targeted proteins are ubiquitinated and subsequently degraded by the cytosolic proteasome. The fact that dysfunction in ERAD causes human diseases () and many viral proteins hijack this pathway to evade detection by the immune system (; ) highlights its importance. Since ERAD was first appreciated over a decade ago, several key players have been identified, particularly from a study of yeast (). However, our knowledge of how ERAD substrates are specifically recognized and extracted from the ER lumen remains incomplete. Highly relevant to this question, recent studies have demonstrated that distinct protein complexes are formed at the ER membrane that are involved in the recognition, ubiquitination, and extraction of specific substrate classes (; ). Although only a few have been implicated in ERAD, ubiquitin (Ub) E3 ligases clearly play a central role in the organization of different ER membrane complexes involved in ERAD of distinct substrate classes. For example, yeast E3 ligase Hrd1p/Der3p is a key component of a core membrane complex that processes substrates with lumenal lesions, the so-called ERAD-L pathway. This core complex includes membrane protein Hrd3p (; ; ) that recruits lumenal folding sensor Yos9p (; ) as well as the membrane protein Ubx2p that recruits the cytosolic cdc48 ATPase complex (). On the other hand, Doa10p, another well-characterized yeast E3 ligase implicated in ERAD, is a key and central component of a core membrane complex that processes ERAD substrates with lesions in their cytoplasmic domains, a so-called ERAD-C pathway (; ). This Doa10p complex includes Ubc7 and its membrane anchor Cue1 as well as cdc48 and its cofactors. However, the specific factors that are capable of recognizing the defect in the cytoplasmic tail of a substrate have not been defined. Nevertheless, substrate ubiquitination as specifically rendered by the E3 ligase is required for both pathways to completely remove the ERAD target from the ER by the cdc48 ATPase complex (; ; ). Within this basic framework of how different substrates are targeted for ERAD, several critical questions remain. For example, in the context of each pathway, how do E3 ligases impose substrate specificity, and at which step of ERAD does substrate ubiquitination occur? Whether different pathways defined in yeast such as the ERAD-L and ERAD-C pathways are conserved in mammals is not well established. However, the fact that most components of ERAD defined in yeast have functional homologues in mammals suggests evolutionary conservation. In agreement with this hypothesis, ER membrane core complexes, including E3 ligases that link ERAD substrates to ubiquitination and extraction machinery, have been defined in human cell studies (; ). However, the mammalian ERAD mechanism is clearly more complex. For example, three Der1p homologues have been defined in mammals, which are designated as Derlin1, 2, and 3. Derlin1 but not Derlin2 plays a central role in ERAD of major histocompatibility complex (MHC) class I heavy chain (HC) by human cytomegalovirus protein US11 (; ). In contrast, both Derlin2 and 3 are associated with EDEM (ER degradation–enhancing α-mannosidase–like protein) and p97 (cdc48 in yeast) and are functionally required for ERAD of NHK (null Hong Kong), a misfolded glycosylated luminal protein in the ER (). Higher eukaryotic cells presumably have many additional E3 ligases participating in ERAD compared with yeast. For example, mammals have a homologue of the yeast RING-H2–type E3 ligase Hrd1p called HRD-1 (; ; ). However, mammals have an additional RING-H2–type E3 ligase not found in yeast called gp78 (; ). Interestingly, both HRD1 and gp78 are found in the same multiprotein ER membrane complex containing Derlin1 and p97 (). Whether they are responsible for distinct subsets of ERAD substrates or share the same substrates is not yet clear. Furthermore, multiple lines of evidence indicate that US2 and US11, two human cytomegalovirus-encoded immune evasion proteins, use distinct ERAD pathways to target HC for ERAD (; ). However, neither HRD1 nor gp78 is required for US2- or US11-induced ERAD, suggesting that they may recruit novel E3s (; ). Collectively, these findings clearly suggest that mammals have evolved very specific and highly regulated ERAD pathways servicing distinct substrates. Immune evasion protein mK3 is a viral E3 ligase encoded by mouse γ-herpesvirus 68 (). It was previously shown that mK3 specifically targets nascent HC for rapid degradation in an ubiquitination-proteasome–dependent manner (; ). The RING-CH domain presumably conferring mK3 ligase activity is highly conserved among members of the K3 family, which includes a group of viral proteins encoded by herpesviruses and poxviruses (; ). Cellular K3 homologues have also been detected, although their physiological function remains to be defined (). Interestingly, one of the cellular K3 homologues in human named MARCH VI (TEB4) has been recently identified as a homologue of Doa10p based on its ER localization and conserved cytosolic RING-CH domain with E3 ligase activity (; ; ). However, whether MARCH VI is a functional homologue of Doa10p remains to be determined. We previously demonstrated that the specific recognition of HC by mK3 required its association with transporter associated with antigen processing (TAP; ; ), a transporter of peptides into the ER lumen (). mK3 is also physically associated with p97 and Derlin1, suggesting that mK3 exploits a physiological ERAD pathway (). Curiously, polyubiquitinated HCs detected in the presence of mK3 were membrane bound, but lysine (K) residues in the HC tail were not required for mK3-induced degradation (). Based on these observations and similar findings with other ERAD substrates, a partial dislocation model was proposed whereby ectodomain residues of the substrate are ubiquitinated after exposure to the cytosol (; ). Recent findings also mandated that N-terminal or non-K forms of ubiquitination be considered (for review see ; ). In this study, we present unequivocal evidence that the cytoplasmic tail of HC is directly ubiquitinated in the presence of mK3 and that S or T residues are sufficient to induce the ubiquitination and rapid degradation of HC. These findings implicate a novel chemical mechanism of substrate ubiquitination via ester linkages. Mechanistically, our findings demonstrate that mK3-induced ubiquitination does not require partial dislocation of the N terminus of HC. Alternatively, our data support a model whereby mK3 substrates are dislocated via their C terminus after tail ubiquitination. These combined observations explain how the mK3 ligase can function as a central component of a core ER membrane complex constructed for the specific targeting of HC for ERAD. How ERAD substrates, especially ER membrane proteins, are extracted from the ER membrane remains elusive, although it has been known for a long time that ubiquitination of substrates is pivotal. We previously demonstrated that mK3 induces polyubiquitination and rapid degradation of HC with K-less tails (). Given the fact that all of the functional ubiquitination components or domains, including mK3, are located outside of the ER, this finding and similar findings with other ERAD substrates (; ; ) raised the possibility that K residues in the HC ectodomain were being ubiquitinated. The presumption is that ERAD substrates can be partially dislocated to the cytosol before ubiquitination. If this is the case for mK3-induced ERAD of HC, identifying which K residues in the ectodomain of HC are ubiquitinated would lend strong support to a partial dislocation model. To determine whether and which K residues on the HC are required for ubiquitination, we took a mutagenesis approach using L as a prototypic MHC class I HC. As shown in , a total of nine Ks were replaced by arginines (Rs) singly or in groups based on HC domain structure. Because our previous study had shown that tail Ks were dispensable for mK3-mediated Ub conjugation and rapid degradation of HC (), all ectodomain K mutations were introduced on a K-less tail template. Surface expression and TAP association were confirmed for all mutants to rule out gross misfolding. We next coexpressed each L K mutant with mK3 to test its capacity to be mK3 regulated. The results were unequivocal and unexpected. None of the individual mutations, groups of mutations, or even totally K-less L were found to have a substantial impact on mK3-mediated HC down-regulation (unpublished data). For example, in the presence of mK3, surface K-less L was reduced >25-fold () and rapidly degraded with kinetics similar to wild-type (wt) L (). Thus, mK3 down-regulation of HC is clearly not compromised by the absence of K residues in the substrate. To establish physiological relevance, we next tested whether the degradation of K-less HC was also dependent on Ub conjugation and proteasome activity. As expected, the surface down-regulation and the polyubiquitination of K-less L were both detected only in the presence of wt mK3 but not the RING mutant (C48G, C51G mutations in its RING-CH domain; and , respectively). More importantly, the appearance of polyubiquitinated forms of K-less L in the presence of mK3 correlated with a lower steady-state level of unconjugated molecules. In contrast, in cells expressing the mK3 RING mutant, K-less L molecules were not Ub conjugated and were dramatically stabilized (). Thus, similar to wt L, down-regulation of K-less L was dependent on the polyubiquitination of HC facilitated by mK3. Moreover, compared with wt L, K-less L showed a strikingly similar pattern of ubiquitination (, lanes 9–12 vs. lanes 3–6). When treated with proteasome inhibitor, the polyubiquitinated conjugates of K-less L accumulated (, top; lanes 10 and 12 vs. lanes 9 and 11, respectively), and unconjugated K-less L was stabilized in cells expressing mK3 (, bottom; lanes 10 and 12 vs. lanes 9 and 11, respectively). These combined findings demonstrate that degradation of K-less HC, like wt HC, is ubiquitination and proteasome dependent. The implication of the aforementioned findings is that there must be sites other than the ɛ-NH group of K on HC for mK3-induced ubiquitination. As recently reviewed, a few recent studies have suggested that the first Ub moiety can be fused linearly to the α-NH group on the N-terminal residue of select substrates (for review see ). To test the likelihood and consequences of the N-terminal ubiquitination of HC by mK3, four different Ub/HC fusion proteins were made. More specifically, fusion proteins with either a wt or K-less Ub moiety were attached to the N terminus of either wt or K-less L. For these fusion proteins, the G76 residue of Ub was replaced with valine to avoid detection by deubiquitinating enzymes (; ; ). When expressed in WT3 cells, these fusion proteins were glycosylated with N-linked endoglycosidase H (endo H)–sensitive glycans, assembled normally with β2m, and attained native class I folding based on their detection with conformation-dependent mAb. Thus, these fusion proteins clearly have the appropriate topology in the ER. Our rationale for this approach was the prediction that adding a K-less Ub to the N terminus might be unfavorable for polyubiquitination and ERAD based on the findings that the addition of large N-terminal tags (e.g., 6× myc or GFP) to the substrate prevented the degradation of proteins known to be capable of N-terminal ubiquitination (). Alternatively, fusion of a wt Ub moiety on HC might facilitate ERAD because it should be readily polyubiquitinated if or when it gains access to the cytosol. The rationale of making fusion proteins with K-less L as well as wt L was to determine whether the removal of Ks might promote N-terminal ubiquitination. To test these possible scenarios, a pulse-chase experiment was conducted with cells expressing each of the four different fusion proteins to compare their relative stability in the presence or absence of mK3. As shown in , in the absence of mK3, both wt and K-less Ub/L fusion proteins were quite stable throughout the chase time (lanes 1–4; and not depicted). This demonstrates that N-terminal ubiquitination of HC alone is not sufficient to trigger a rapid degradation. A likely explanation is that the N terminus of HC does not gain access to the cytosol in the absence of mK3. However, in the presence of mK3, both wt Ub and K-less Ub/L fusion proteins were rapidly degraded (, lanes 5–8; and not depicted). More strikingly, all of these fusion proteins displayed similar ubiquitination patterns in cells with mK3 (unpublished data). These results argue against an N-terminal Ub conjugation. However, how these fusion proteins were polyubiquitinated, especially the one with K-less Ub fused to K-less L, remained unknown. To solve this puzzle, a proteinase K assay was adopted to determine the site of the polyubiquitinated portion of completely K-less Ub/L fusion proteins. It was formally possible that the N terminus of K-less Ub/K-less L was polyubiquitinated after partial dislocation. If this was the case, the cytosolic-exposed N termini of fusion proteins as well as their poly-Ub forms would be sensitive to proteinase K digestion. To test this possibility, the ER fraction from cells expressing K-less Ub/K-less L was isolated by ultracentrifugation. The isolated ER fraction was then subjected to proteinase K digestion followed by immunoprecipitation of HC and blotting for Ub. As expected, Ub/L fusion HCs were observed only in the pellet fraction but not in the supernatant (), indicating that the Ub/L fusion protein is ER membrane bound, thus validating the ER isolation. Again, the poly-Ub bands were observed associated with HC precipitates in mK3-expressing cells but not in the cells without mK3 (, top; lanes 9 and 10 vs. lanes 3 and 4; poly-Ub forms are demarcated with asterisks). Also as expected, the polyubiquitinated and unconjugated HC bands were comparably sensitive to endo H, demonstrating that mK3 was modifying ER membrane–bound newly synthesized fusion proteins (, lane 10 vs. lane 9). Upon digestion of the membrane fraction by proteinase K, domains exposed to the cytosol should be cleaved. Indeed, when treated with proteinase K, the K-less Ub/K-less L fusion protein was slightly reduced in size, resulting in a faster gel migration (, Ub blot; lanes 5 and 6 vs. lanes 3 and 4 and lanes 11 and 12 vs. lanes 9 and 10, respectively). Furthermore, the proteinase K–treated fusion protein was only detected by an antibody against Ub but not by an antibody to the L cytoplasmic tail (, lanes 11 and 12). Thus, proteinase K cleaved the HC tail of the fusion protein, demonstrating that it was in the cytosol, whereas the N-terminal Ub moiety was undisturbed, suggesting that it resided in the lumen. More strikingly, the poly-Ub conjugation bands also disappeared when the tail was cleaved by proteinase K (, Ub blot; lanes 11 and 12 vs. lanes 9 and 10), suggesting that the cytoplasmic tail of Ub/L fusion protein is the site of poly-Ub chain formation induced by mK3. To rule out the possibility that the fusion proteins used an alternative pathway, a similar experiment was conducted with cells expressing either wt or K-less L. Again, we observed that poly-Ub forms of both wt and K-less L are ER membrane associated and disappear when the HC tail is removed ( and not depicted), suggesting that for both Ub/L fusion protein and wt L, the cytoplasmic tail is the site of Ub conjugation. Although we considered it unlikely, the possibility remained that proteinase K was removing poly-Ub from the N terminus of L or Ub/L that occurred after partial dislocation to the cytosol. If this was the case, the cytoplasmic tail would not have to be the site of ubiquitination. To further demonstrate that the tail is the site of ubiquitination, a thrombin (TMB) cleavage site, LVPRGG, was engineered into the tail of L right after the basic cluster KRRRNT proximal to the transmembrane domain. The K residue in this basic cluster was replaced by R. This molecule (L TMB) was normally glycosylated, β2m assembled, and folded when expressed in WT3 cells (unpublished data). In the presence of mK3, L TMB molecules were polyubiquitinated in a pattern similar to L (). Lysates from cells expressing mK3 and wt L or L TMB were immunoprecipitated with antibodies specific for the lumenal domain of L followed by TMB treatment. As expected, in the presence of TMB, only the tails of L TMB molecules and not wt L were cleaved, as demonstrated by their faster migration and by blotting with the tail-specific antibody (, lanes 7 and 8 vs. lanes 5 and 6). In comparison, wt L molecules were resistant to the TMB treatment (, lanes 3 and 4 vs. lanes 1 and 2), indicating a specific and successful TMB cleavage. More importantly, along with the cleavage of the tail of L, the polyubiquitinated forms of L TMB also disappeared (, Ub blot; lanes 7 and 8). Because TMB treatment should not affect N-terminal polyubiquitination, these experiments along with the proteinase K experiments clearly defined the HC tail and not the N terminus as the site of ubiquitination. Importantly, the same observations were obtained with proteinase K–treated K-less L or tail K-less L TMB (unpublished data). Thus, the clear implication was that ubiquitination of HC by mK3 likely occurs on non-K residues. Notably, it was recently reported that the ubiquitination of class I HC induced by MIR1 (also called kK3), an E3 ligase of Kaposi's sarcoma-associated herpesvirus (KSHV), requires a C residue in the cytoplasmic tail of MHC-I molecules (). To test whether the mK3-induced ubiquitination of HC, like the MIR1-induced ubiquitination of HC, is dependent on a C residue in the tail of HC, a new L mutant was made. The L tail has a single C at position 336 that is conserved among most mouse HCs. Thus, to test the importance of the tail C residue, a C336S point mutation was introduced onto a tail K-less L template (). WT3 cells were cotransduced with this construct, termed L tail KC-less, and mK3. In the presence of mK3, this KC-less mutant displayed prominent polyubiquitination () and rapid degradation () similar to wt L. Thus, unlike MIR1, a C residue in the HC tail is not required for HC to be a substrate for mK3-induced ubiquitination and degradation. To extend this finding and rule out the possibility that C residues in the HC ectodomain are required for mK3 ubiquitination, L HC molecules were immunoprecipitated and eluted in the presence of the thioester-reducing agent 2-mercaptoethanol (2ME; ). It should be noted that this is the same treatment used to disrupt the aforementioned ubiquitination of HC by KSHV protein MIR1 (). Dissociation of the Ig to heavy and light chains (, lanes 1–4 vs. lanes 5–8; indicated by arrows) was an internal control demonstrating the effectiveness of the reduction. However, comparison of the sample with and without 2ME treatment showed no reduction in the levels of polyubiquitination of wt or K-less L. These combined mutagenesis and chemical findings demonstrate that mK3-induced polyubiquitination does not require C residues in its HC substrate. Although it was predicted that they would have insufficient stability, T and S residues have hydroxyl groups that could potentially form ester bonds with Ub (; for review see ). To test this hypothesis, we made a construct without any of the four potential residues in the tail for Ub conjugation. In total, three Ks, 4 Ss, 1 T, and 1 C in the tail of L were mutated (sequence shown in ), but the ectodomain and transmembrane portion of wt L were unchanged. The resulting construct, termed KCST-less L tail, was stably expressed in WT3 cells with and without mK3. The ubiquitination status as well as the steady-state level of this L mutant was examined. As shown in , polyubiquitination of this KCST-less mutant was undetectable compared with wt L (left; third and fourth lanes vs. first and second lanes), although both cell lines expressed similar amounts of mK3 (, right). Concomitant with substantially diminished polyubiquitination, the KCST-less mutant was considerably more stable than wt L (). To rule out the possibility that resistance of this mutant to mK3 is the result of impaired mK3 interaction, we demonstrated that the KCST-less mutant coimmunoprecipitated with mK3 similar to wt L (unpublished data). Thus, the removal of Ks and all residues with the potential to form ester bonds from the HC tail did not affect its ability to associate with mK3, only its ability to be ubiquitinated by mK3. The fact that the polyubiquitined forms disappeared when KCST residues were removed from only the tail of L and not the rest of the protein corroborated the conclusion that the tail is the site of ubiquitination. Furthermore, the fact that the KC-less tail mutant was ubiquitinated by mK3 but the tail KCST-less was not clearly implicated S and/or T residues in mK3 function. We next compared the ability of individual residues to restore mK3-induced ubiquitination in the KCST-less tail. For these comparisons, one K (K337), one C (C336), one S (S329), or one T (T313) was added back to the KCST-less tail mutant of L (). These locations were selected because they are highly conserved among HC alleles. Because the T313 residue is membrane proximal, whereas the other potential Ub sites were clustered toward the C terminus of the tail, we also constructed a T337 mutant. Strikingly, a single S or T residue near the C terminus of the HC tail resulted in substantial polyubiquitination similar to the polyubiquitination of a single K in the presence of mK3 (; left). In contrast, the single C residue near the C terminus or the membrane-proximal T residue only showed very weak or undetectable ubiquitination. These findings suggested that hydoxylated amino acids (either S or T) can be sites of Ub conjugation but that location toward the C terminus of the tail is critical (; left). Importantly, similar levels of mK3 were expressed in these cells (; right). Consistent with Ub conjugation, the S329, T337, and K337 mutants were more rapidly degraded in mK3-expressing cells compared with the KCST-less tail mutant (). The fact that the 1S or 1T tail was polyubiquitinated and degraded in the absence of K residues largely ruled out the likelihood that S or T residues are required for other modifications, such as phosphorylation (). However, the possibility remained that the observed ubiquitination of a K-less ERAD substrate might be the result of the ubiquitination of adaptor proteins (; ). To definitively demonstrate that the HC is directly ubiquitinated, precipitates of the 1K or 1S L proteins were boiled in 0.5% SDS and 10 mM DTT and reprecipitated with antibody to denatured L. As shown in , polyubiquitination of 1K or 1S tails was unaffected by denaturation, whereas β2m assembly was eliminated. This result demonstrated that either hydroxylated or amide amino acid side chains can be directly Ub conjugated by mK3. If this conclusion is true, the linkage between the hydroxyl side chain of S and the C-terminal glycine residue of Ub would be an ester bond. To test the existence of this linkage, we compared the stability of the bonds between Ub and 1K or 1S HCs after exposure to mild alkaline hydrolysis. It is known that ester bonds are more labile under these conditions than amide bonds, such as those formed between K residues and Ub (; ). Indeed, the ubiquitinated forms of the 1S mutant were undetectable upon incubation with either 1 M hydroxylamine, pH 9.0, or 0.1 M NaOH (, top and bottom, respectively). Similarly, ubiquitinated forms of the 1T mutant were also labile to mild alkaline treatment (unpublished data). As expected, the 1K mutant was stable to treatment with either reagent (). These data strongly support the hypothesis that mK3 is capable of facilitating the novel conjugation of ubiquitination to an S or T residue in the cytoplasmic tail of L via an ester linkage. The mechanism of how substrates are retrotranslocated or dislocated from ER membrane is still poorly understood. It is known that ubiquitination plays a crucial role in this action, but when it happens and whether a direct Ub conjugation to the substrate is required are not clear. For soluble lumenal substrates, mutations in the ubiquitination system result in substrates accumulating in the ER (), suggesting that the ubiquitination of substrates is required for their dislocation. However, because lumenal substrates are separated from the ubiquitination apparatus by a membrane, the substrate ubiquitination obviously does not initiate their dislocation. On the other hand, for integral membrane proteins, one would think that its cytosolic portion might be the most convenient site for Ub conjugation, which could subsequently recruit Ub-binding protein and the ATPase complex required for complete extraction of the substrate. However, for several typical type I membrane protein substrates, such as orphan subunit TCRα and MHC-I HC in the presence of US2 or US11, K residues in the substrate's tail are dispensable for their dislocation (; ; ). In the specific case of US2-induced ERAD of HC, internal K residues were found to be required for dislocation (). However, for the other aforementioned ERAD substrates, the replacement of all internal Ks on these ERAD substrates did not affect their dislocation, suggesting a non-K–mediated form of ubiquitination. One possible explanation is N-terminal ubiquitination, in which the first Ub moiety is fused linearly to the α-NH group of the substrate N terminus. This type of ubiquitination has been demonstrated for >12 proteins and, thus, provides an alternative mode of ubiquitination that may be particularly important for the regulation of proteins that have no accessible K residues (). In ERAD, however, such a process has not been well established, especially for soluble lumenal substrates or type I membrane substrates with their N termini in the ER lumen. For these types of substrates, a partial dislocation of their N termini to the cytosolic side is presumably required before the addition of Ub because all of the functional ubiquitination components are located in the cytosol. Recently, attached Ub to the N terminus of wt and K-less HCs and found that it did not accelerate their dislocation in the presence of US11. This finding suggests that N-terminal ubiquitination does not play a role in US11-initiated extraction of HCs from the ER membrane. An alternative explanation for the dislocation of K-less membrane proteins is that an adaptor protein is ubiquitinated instead of direct ubiquitination of the substrate. Ubiquitination on an adaptor could link the associated substrate with the extraction machinery, resulting in complete extraction of the substrate. Interestingly, this mechanism is supported by recent studies of the dislocation of K-less HC by US11 (). However, the ultimate confirmation for such a model will have to await identification of the ubiquitinated adaptor protein. An important consideration regarding conclusions using K-less substrates is whether ubiquitination can occur via non-K residues. For example, it has been reported that MIR1 ligase (also called kK3) of KSHV can ubiquitinate C residues in the HC tail via thioester bonds, thus targeting it for degradation through an endocytic–lysosome pathway (). In the case of mK3-induced HC ubiquitination and degradation, we previously showed that the substitution of all Ks in the tail of MHC-I HC did not affect the ubiquitination status or the degradation of these molecules (). Nevertheless, deletion of the tail abolished the ubiquitination of HC, leading to its stabilization in the ER (). Indeed, a similar phenotype has also been found in the US11-induced turnover of MHC-I HC (; ). Our erroneous explanation for these observations was that the tail of HC is required for the initiation of dislocation but not for direct tail ubiquitination. In the present study, a complete K-less L and a K-less Ub/K-less L fusion protein were found to be polyubiquitinated and rapidly degraded in the presence of mK3. This finding suggested that neither lumenal Ks nor an α-NH group of the N terminus of HC are essential Ub conjugation sites required by mK3. More importantly, we demonstrated that polyubiquitinated HCs become undetectable when the tail was removed by proteinase K digestion or via an engineered TMB cleavage site. These findings demonstrated that the HC tail is the site of ubiquitination. Of note, these observations were also made with wt L molecules, demonstrating that it is a normal process used by native protein and not merely a forced pathway used by substrates having no available Ks. Furthermore, we found that mutations of all K and C residues of the L tail did not affect mK3-induced ERAD. However, the additional removal of S and T residues in the tail abolished mK3-induced polyubiquitination and stabilized the HC. These findings suggested that one or both of the hydroxylated amino acids were sufficient for ubiquitination. To extend these observations using a better defined approach, individual K, C, S, and T residues were reintroduced into L with a KCST-less tail. Remarkably, HCs with tails having only one S, T, or K were polyubiquitinated and were susceptible to ERAD in the presence of mK3. However, location was found to be important in that all efficacious ubiquitination sites were clustered toward the C terminus of the HC tail. In contrast, a tail with a membrane-proximal T residue displayed no detectable ubiquitination. S and T residues are potential phosphorylation sites that could be required to recruit a putative HC-associated protein that is ubiquitinated. To demonstrate that HC is itself ubiquitinated, HCs with single S or K in their tails were denatured and reprecipitated with a class I–specific mAb. In both cases, ubiquitinated HCs were still detected after treatment, demonstrating that Ub moieties can be covalently attached to either S, T, or K residues in the HC tail in the presence of mK3. In strong support of this conclusion, 2-ME that specifically disrupts thioester bonds formed with C residues had no effect on the Ub conjugation to S or K residues of the HC tail in the presence of mK3. Alternatively, mild alkaline hydrolysis with sodium hydroxylamine or NaOH, reagents that are effective in cleaving ester bonds and not amide bonds, removed detectable mK3-mediated ubiquitination via S or T but not K residues. Thus, in the presence of mK3, ester bonds are formed that covalently link Ub to S or T residues in the tail of the HC substrate, thereby defining a novel form of ubiquitination. It will be interesting to determine whether the ability to use S and T as well as K residues reflects a unique, direct interaction of mK3 with the HC substrate or, alternatively, may involve distinct cellular components such as E2 conjugation enzymes or other necessary cofactors. The ubiquitination pattern induced by mK3 is somewhat surprising. Even when the HC substrate has a wt tail, the Ub and Ub forms are predominant. It is noteworthy that also detected HCs with predominantly Ub and Ub forms and few, if any, higher than Ub in the presence of mK3. Because the proteasome requires Ub to provide sufficient avidity for binding substrates for degradation (), we have considered the notion that multimers higher than Ub may be unstable and, thus, difficult to detect in steady-state assays. Indeed, when proteasome inhibitor was used, Ub and Ub multimers can be detected ( and ). However, proteasome inhibitors result in only modest increases in forms higher than Ub, which is likely the result of the fact that the 2–3-h treatment does not appreciably affect steady-state levels of ubiquitination. Interestingly, tails containing a single S, T, or K residue displayed an even greater predominance of Ub forms, although Ub and higher multimers show up on longer exposure. In any case, HC tails with only one S, T, or K residue are viable ERAD substrates in the presence of mK3, implying they are polyubiquitinated to Ub. It is also interesting that in the presence of mK3, the minimum ubiquitination forms of HCs are Ub, regardless of whether it is a wt or tail 1S, 1T, or 1K substrate. This may imply that mK3 facilitates the transfer of a premade Ub or Ub chain from cognate E2 to the substrates. The findings reported in this study clarify the molecular basis of mK3 substrate specificity. Our previous study failed to detect an HC-specific tail sequence required for mK3 interaction (). The finding of alternative sites for mK3-mediated ubiquitination reported here is consistent with this conclusion. Interestingly, the primary binding partner for mK3 is TAP, the transporter responsible for dislocating proteasome-processed peptides into the ER lumen, and a mutagenesis study showed that it is the C terminus of mK3 that binds TAP (). Most class I HC alleles require a physical association with TAP after assembly with β2m and before the binding of high affinity peptide ligands (). These combined findings suggest a proximity model whereby the specific recognition of HC by mK3 is achieved by TAP binding to the mK3 C terminus, which then orients the N-terminal mK3 RING domain such that it can only interact with the HC tail. Thus, TAP is the key mK3 adaptor protein juxtaposing the mK3 RING domain with the HC tail to confer substrate specificity. The demonstration here of the importance of the location of ubiquitination sites within the tail lends strong support to this proximity model. This model provides a viable general strategy for allowing E3s to maintain their requisite substrate specificity while facilitating the ubiquitination of a highly polymorphic substrate such as HC. With the required juxtaposition of mK3 RING and HC tail, the necessary cellular ubiquitination components, including an appropriate E2, can then be recruited, resulting in ubiquitination of the HC tail. This core ER membrane complex centered on mK3 can also include Derlin1, although it appears to have a redundant function (; and unpublished data). ATPase p97 is also recruited to this mK3 core membrane complex, and this association clearly has an impact on mK3-induced ERAD (). The aforementioned findings suggest that the viral mK3 ligase assimilates a core membrane complex in mouse cells similar to ERAD complexes described in yeast. In this context, it is interesting to compare properties and outstanding questions regarding the mK3 membrane complex with the aforementioned yeast Doa10p (ERAD-C) and Hrd1p (ERAD-L) complexes (; ; ). HC tail ubiquitination by mK3 suggests that it mimics an ERAD-C pathway induced by substrates with a cytosolic lesion (; ), and ERAD-C does not require Der1p, as mK3 does not require the Der1p homologue Derlin1 (; and unpublished data). Models such as ERAD-C and the mK3 model would not require the partial dislocation of substrate before ubiquitination, and polyubiquitination of the tail could be the initiating signal for dislocation. It is intriguing to speculate that mK3 may also be a component of the dislocation channel, if indeed one is required for this pathway. In the context of an ERAD-L–like model, it is unclear whether molecular interactions are required in the ER lumen to promote substrate dislocation. Although mK3 has only 12 amino acids in the ER lumen, it could disrupt transmembrane interactions between proteins within the core membrane complex, and this disruption could induce ER protein folding sensors (chaperones) to bind the ectodomain of HC and facilitate dislocation. Defining the molecular composition of the dislocation pore and the signals that initiate substrate dislocation are key questions for any of these Ub ligase–induced membrane complexes involved in distinct ERAD pathways. In any case, findings with viral ligase mK3 suggest that it can assemble a unique core ER membrane complex that specifically detects MHC class I HCs and targets them for ERAD. In summary, we report that the mK3 ligase directly ubiquitinates its HC substrate at tail S, T, or K residues. This observation obviates the need for the partial dislocation of HC before ubiquitination, suggesting a vectorial exit from the ER. It will be interesting to determine whether other ERAD pathways involving transmembrane protein substrates might also involve tail ubiquitination using non-K residues. Furthermore, the fact that mK3 has numerous viral (including MIR1) and cellular homologues makes it attractive to speculate that other ubiquitination-regulated processes use similar nonconventional methods of Ub conjugation. Mouse B6/WT3 (WT3, ) cells were described previously (). 293T cells () were used for the production of ecotropic retrovirus. All cells were maintained in complete RPMI 1640 with 10% FCS (HyClone) as described previously (). Retrovirus-containing supernatants were produced using the Vpack vector system (Stratagene) with transient transfection of 293T cells. Cells transduced by the pMIN-containing virus were enriched by geneticine selection, whereas GFP cells from pMIG-transduced lines were enriched by cell sorting. Where indicated, cells were cultured for 24 h with 125 U/ml of mouse IFN-γ (Biosource International) and for 2–3 h with 30–60 μM of the proteasome inhibitor MG132 (Boston Biochem) or 5 μM clasto-lactacystin β-lactone (Calbiochem) before harvesting with trypsin-EDTA. All flow cytometric analyses were performed as previously described (). Two retroviral expression vectors, pMSCV.IRES.GFP (pMIG) and pMSCV.IRES.neo (pMIN; ), were used to express mK3 and L constructs, respectively. mK3 sequence was obtained by PCR amplification of the K3 gene from a γHV68 subclone (). Both mK3 and L mutants were generated by site-directed mutagenesis (Stratagene). Ub/L fusion constructs were made by overlapping PCR, wherein the uncleavable wt and K-less Ub sequences (wt Ub and K-less Ub, with replacement of glycine 76 to valine) were PCR amplified from the constructs pLZRS-Ub G76V-HLA.A2 and pLZRS-Ub G76V-HLA.A2 (). The correct sequences for all of the constructs were confirmed by DNA sequencing. Rabbit anti-mK3, Ub antibodies, β-actin (AC-74) antibodies, and mAbs 30-5-7 and 64-3-7 to folded and open forms of MHC class I L were previously described (). Antibodies (Ra20873) to the cytoplasmic tail of L were produced in rabbits immunized with the cytoplasmic tail peptide (). Immunoprecipitation and immunoblotting were conducted as previously described (). In brief, cells were lysed in PBS buffer containing 1% NP-40, 20 mM iodoacetamide (Sigma-Aldrich), and Complete mini protease inhibitors (Roche) or 0.4 mM PMSF (Sigma-Aldrich). Postnuclear lysates were incubated with protein A–Sepharose beads (Sigma-Aldrich) and antibodies. After washing beads four times with PBS/iodoacetamide buffer containing 0.15% NP-40, immunoprecipitates were eluted from protein A by boiling for 3 min in lithium dodecyl sulfate sample buffer (Invitrogen). For endo H treatment, immunoprecipitates were eluted in 10 mM Tris-HCl, pH 6.8, with 0.5% SDS, and the elutes then were mixed with an equal volume of 100 mM sodium acetate, pH 5.4, and incubated with 1 μU endo H (MP Biomedicals) at 37°C for 2 h. Immunoblotting was performed after SDS-PAGE separation of precipitated proteins or cell lysates as previously described (). Specific proteins were visualized by chemiluminescence using the ECL system (GE Healthcare). After 30 min of preincubation in Cys- and Met-free medium (MEM-Earle's with 5% dialyzed FCS), cells were pulse labeled with Express [S]Cys/Met labeling mix (Perkin Elmer) at 150 μCi/ml for 10–15 min. Chase was initiated by the addition of an excess of unlabeled Cys/Met (5 mM each). Immunoprecipitation was performed as described in the previous section. Samples were subjected to SDS-PAGE, and gels were treated with Amplify (GE Healthcare), dried, and exposed to BioMax-MR film (Kodak). Cells were incubated in cold hypotonic extraction buffer containing 10 mM Hepes, pH 7.8, 25 mM KCl, and 1 mM EGTA for 20 min and were resuspended in isotonic extraction buffer containing 10 mM Hepes, pH 7.8, 250 mM sucrose, 25 mM KCl, and 1 mM EGTA before they were broken with 20 strokes of Dounce homogenizer on ice. The homogenate was then subjected to serial centrifugations at 1,000 for 10 min, 10,000 for 10 min, and 100,000 for 60 min, which were all performed at 4°C. The 100,000 pellet was resuspended in PBS with 20 mM iodoacetamide. Two aliquots of the suspension were incubated on ice with or without 10 μg/ml proteinase K (Invitrogen) for 20 min. After the digestion was stopped by 2 mM PMSF, NP-40 was added to a final concentration of 1% to disrupt the membranes. PMSF and NP-40 were also added into 100,000 supernatant fractions. Subsequent immunoprecipitation and immunoblotting were performed as described above in the Immunoprecipitation and immunoblots section. Immunoprecipitates were boiled in 0.5% SDS for 3 min followed by incubation in either 1 M sodium hydroxylamine, pH 9, for 4 h at 37°C or sodium hydroxide (0.1 M NaOH) for 2 h at 37°C. Mock-treated samples were incubated with PBS. Before SDS-PAGE analysis, the sodium hydroxylamine–treated samples were dialyzed against PBS overnight at 4°C using 3500 MWCO MINI dialysis units (Pierce Chemical Co.).

Sensitivity and resistance to apoptosis are to a large degree regulated by pro- and anti-apoptotic members of the Bcl-2 protein family. How this regulation is achieved is under intense investigation (; ). Structural and functional similarities divide the Bcl-2 family into three groups. The pro-apoptotic multidomain proteins (containing the Bcl-2 homology [BH] domains 1–3) Bax and/or Bak are required for mitochondrial permeabilization during apoptosis (; ; ; ). The activation of Bax/Bak is caused consecutively to the activation of the BH3-only group of Bcl-2 family proteins (which have in common only the short BH3 domain; ; ). The anti-apoptotic group of Bcl-2 proteins contains Bcl-2, Bcl-x, Bcl-w, Mcl-1, and A1 (). The various interactions between these proteins are crucial for the life-death decision. However, many details of these interactions are still unclear. Direct binding of Bax to Bcl-2 was demonstrated early on (), but the significance of this interaction is questionable because it depends on the presence of certain detergents () and because Bcl-2 is localized on intracellular membranes, whereas Bax is largely soluble in the cytosol (; ). Bak, on the other hand, is an integral protein of the outer mitochondrial membrane, and it has recently been demonstrated to be sequestered there and kept inactive by the two anti-apoptotic Bcl-2 proteins, Mcl-1 and Bcl-x (). Eight BH3-only proteins are known. It is firmly established that their BH3 domains can bind to anti-apoptotic Bcl-2 proteins, which prevents their activating Bax/Bak (; ; ). This has led to the proposition that BH3-only proteins induce apoptosis, at least in part, through the neutralization of Bcl-2-like proteins. This view has gained support by the recent demonstration of a selectivity in binding between BH3-only proteins and Bcl-2-like proteins (), reproduced in another study using a different technical approach (). Although both studies are potentially limited by the use of BH3 domain peptides rather than whole proteins, the results are intriguing. The selectivity of binding found in these studies could explain the varying apoptosis-inducing potency of BH3-only proteins, and the model has been elegantly confirmed by the demonstration that combining BH3-only proteins that can bind to Bcl-x and Mcl-1 leads to the release and, presumably, the auto-activation of Bak (). The above studies have engendered two models of BH3-only protein action. One model (the direct binding model) proposes that the BH3-only proteins Bim and tBid (and perhaps Puma) can directly bind and activate Bax/Bak (; ), whereas the remaining BH3-only proteins (Bik, Puma, Noxa, Bad, Bmf, and Hrk) can only sensitize, i.e., release Bim and tBid from their site of sequestration to Bcl-2-like proteins (and Bim/tBid would then go on to activate Bax/Bak). Although it is difficult to demonstrate the interaction of Bim/tBid with Bax/Bak in intact cells, this model has received support from studies with purified proteins and artificial membranes (, ). The second model (the displacement model) proposes that Bax/Bak can auto-activate once the inhibition imposed by Bcl-2-like proteins has been removed by BH3-only proteins, but that no direct interaction between BH3-only proteins and Bax/Bak is required (; ). Although this model has been supported experimentally for Bak (see above), it is still difficult to see how this mechanism could work for Bax. One step in the activation of Bax has to be the mitochondrial translocation. Although this Bax translocation and insertion is seen during apoptosis induction (; ), there is no signal known that could initiate this, and no interaction partner is known that could determine Bax translocation. BH3-only proteins are activated upon apoptotic stimuli from the outside. However, the study of BH3-only protein action in intact cells is not easy. One problem is that most BH3-only proteins are subject to post-translational regulation, and their activation and activity are very difficult to measure. For instance, Bim is expressed in normal cells and is activated upon a suitable stimulus, but is it unclear how Bim is activated on a molecular level. Furthermore, apoptotic stimuli may activate more than one BH3-only protein, and this may depend on the cell type used. Lastly, most apoptotic stimuli have massive side effects. UV irradiation, for instance, will activate BH3-only proteins in many cells but also activate other pathways, such as stress kinases and the DNA-damage response, confounding the observed results. To escape these problems and to analyze the consequences of BH3-only protein activation in human cells, we used a strategy of inducible expression of Bim in HeLa epithelial cells. The inducible expression of Bim was sufficient to induce apoptosis. This enabled us to investigate the interaction of Bim with other Bcl-2 proteins and the events triggered by Bim at the cell's mitochondria. We first tested the major isoform of Bim, Bim. In transient transfection experiments of T-REx-HeLa cells (a cell line stably carrying the tetracycline repressor, in which expression of the construct is induced by tetracycline [tet]) Bim induced apoptosis of transfected cells (∼50% dead cells after 24 h in the presence of tet). Upon transfection and selection, nine clones were obtained that stably carried the tet-inducible Bim construct. All of these clones expressed Bim, and the expression was induced to varying degrees by tet addition. The expression exceeded in all clones by far the endogenous levels in T-REx-HeLa cells, which is easily detectable by Western blotting (unpublished data). However, tet-induced expression of Bim completely failed to induce apoptosis in these cells (unpublished data). Although this was unexpected, it might be explained by the fact that Bim is in many cells expressed at relatively high levels and is probably activated by molecular events that have not been fully worked out. We thus turned to the expression of Bim (mouse Bim was used, as at the time the identity of human Bim was not unequivocal in the databases). Bim is the shortest of the regularly expressed Bim isoforms, is normally expressed at very low levels, and is not subject to any known post-translational activation (; ). When expressed transiently as above, Bim was more active than Bim (∼80% dead cells at 24 h), confirming earlier results (). Two stable cell lines carrying the inducible Bim construct were analyzed in detail. There was no major difference in the parameters analyzed between these two lines. In these clones, tet addition induced apoptosis in a dose-dependent manner (Fig. S1 a, available at ). Cell death was rapid, with ∼50% of cells containing active caspase-3 after 2 h (Fig. S1 b) and ∼90% of cells containing condensed nuclei after 4–6 h (not depicted). We then proceeded to analyze the molecular events during apoptosis induction by Bim. Both models of BH3-only protein action, especially the displacement model, predict that Bim will induce apoptosis through binding to anti-apoptotic Bcl-2 proteins. We therefore tested the association of Bim with Bcl-2 by immunoprecipitation (IP). IP with antibodies specific for Bim precipitated Bim from both clones ( and Fig. S3 a). A fraction of Bcl-2, Mcl-1, and Bcl-x was coprecipitated with Bim from uninduced cells (), suggesting that the “leaky” Bim that was expressed in these cells was associated with Bcl-2, Mcl-1, and Bcl-x (see for a specificity control). Induction with tet led to the increase in the amount of Bim isolated (). However, the amount of Bcl-2, Mcl-1, and Bcl-x coprecipitated with Bim appeared to be essentially the same as in uninduced cells, suggesting that the induced, apoptosis-inducing fraction of Bim is not bound to Bcl-2, Mcl-1, or Bcl-x. Although Bcl-w and A1 were detectable in the cells by Western blotting, they were not seen in IP products, suggesting that none or very little of these proteins was bound to Bim (). These results indicate that, in order to induce apoptosis, Bim does not need to bind to detectable amounts of anti-apoptotic Bcl-2 proteins in HeLa cells. To test this in a different cellular system, 293 human fibroblasts carrying the tet repressor (T-REx-293 cells) were transiently transfected with the Bim construct, and binding of Bim to Bcl-2 family proteins was tested. As in the case of HeLa cells, small amounts of Bcl-2 and Mcl-1 could be coprecipitated with Bim, and a very faint band was seen when probing for Bcl-w (Bcl-x was undetectable in these cells). However, as in the HeLa clones, the amounts precipitated appeared essentially the same when Bim expression was induced by tet (). Notably, we could detect some Bax by co-IP in these experiments, when higher amounts of mitochondria were used or when mitochondria or whole-cell lysates were used from the T-REx-293 cells transiently transfected with the Bim construct (see Fig. S3, b–d). Weak binding of Bim (unlike Bim or Bim) to Bax in detergent lysates has been observed before but was found not to be required for apoptosis induction, as a Bim mutant that had lost this binding capacity still induced apoptosis equally well (; ). As an additional control, Bim cells were transiently transfected with an expression construct for Bcl-2. As expected, the transfected cells were protected against Bim-induced apoptosis (Fig. S2, available at ). However, in cells overexpressing Bcl-2, tet-induced Bim clearly associated with Bcl-2 as judged by the increased amount of Bcl-2 coprecipitated with Bim (, compare lanes where the IP pellet was loaded). This indicates that in the situation where Bcl-2 acts to block Bim-induced apoptosis, it does so via direct binding of Bim. Conversely, when the apoptosis-inducing activity of Bim is not blocked, no binding between Bim and anti-apoptotic Bcl-2 proteins can be detected. In the attempt to understand the molecular function of Bim we next analyzed the subcellular localization of induced Bim. In uninduced cells, a small amount of Bim was detectable in the fraction containing mitochondria. Tet-induced Bim was initially also found in mitochondria, and only when higher levels were reached (especially in the 3C4 clone) part of it was seen in the cytosol ( and Fig. S4, respectively, for the two clones tested; there was some difference between the clones as to localization, which may be linked to different amounts of Bim induced or possibly to different mitochondrial import capacity [see below]). Intriguingly, mitochondrial Bim was integrally inserted into mitochondrial membranes, as measured by alkaline treatment of mitochondria (which allows separation of integral membrane proteins from soluble/peripherally attached proteins; ). Some (endogenous) Bim was also found expressed in mitochondria, possibly a fraction that had been activated but blocked by anti-apoptotic proteins. Bim was also inserted into mitochondrial membranes, as was Bax (). In vitro import experiments using radiolabeled Bim proteins confirmed these results. Both in vitro–translated Bim and Bim proteins became associated with isolated HeLa mitochondria upon incubation in a typical mitochondrial import reaction (). When mitochondria were then treated with protease the Bim proteins were readily digested, suggesting an exposed localization at the outer membrane of mitochondria. Alkaline extraction experiments following the import experiments indicate that Bim was integrally inserted into mitochondrial membranes (). Similar results were obtained when yeast mitochondria (from ) were used (not depicted). The exposed localization and the membrane integration of Bim imply that Bim is integrally inserted into the mitochondrial outer membrane, suggesting that it translocates to the surface of the mitochondria (most likely directed by an intrinsic mitochondrial targeting signal), where it is inserted into the lipid bilayer of the outer membrane through the activity of the TOM complex (the general translocase of the mitochondrial outer membrane). Bim thus seems to follow the main import pathway for mitochondrial outer membrane proteins (). We noticed in these experiments that the expression of Bim caused an early shift of Bax to mitochondria. This was detectable as both an increase in mitochondrial Bax and, perhaps more distinctly, a cytosolic depletion of Bax that was clearly visible a few hours after Bim induction ( and Fig. S4, available at ). Similar results were obtained when T-REx-293 cells were transiently transfected with wild-type Bim construct and induced with tet. No translocation of Bax to the mitochondria was seen when the empty expression vector was used (unpublished data). The activation of Bax has been measured by a conformational change at the N terminus () or the C-terminal part (), or as oligomerization at mitochondria (). Bands corresponding to the described Bax dimers and trimers were found when mitochondrial proteins were cross-linked upon Bim induction (). Bax activation as measured by staining for N- or C-terminal conformational change was also detected (not depicted). In order better to visualize mitochondrial translocation of Bax, we used an EGFP-Bax fusion protein that has been described before to translocate to mitochondria during apoptosis (). Laser-scanning microscopy confirmed that tet-induced Bim was predominantly located on the cell's mitochondria (). Upon transient transfection EGFP-Bax was localized in the cytosol, but upon induction of Bim translocated to the mitochondria where it colocalized with Bim (). It is worth noting that mitochondrial Bax at all times appeared to be inserted into mitochondrial membranes rather than be loosely attached (). Thus, the sole manipulation of tet-dependent induction of Bim causes the translocation of Bax to mitochondria, accompanied by Bax activation. These results suggested that the mitochondrial translocation, but not the interaction with anti-apoptotic Bcl-2 proteins, was the key event in apoptosis induction by Bim. We therefore pursued the two-pronged strategy of analyzing Bim mutants that were engineered to differ either in translocation or in their capacity to interact with Bcl-2. gives a schematic representation of the mutants used. It has been noted before that the C terminus of Bim (which is part of the protein in all major splice forms) contains a potential membrane insertion sequence, and it has been shown early on that overexpressed Bim is targeted to intracellular membranes (). Further, our above data suggest that active Bim is a tail-anchored protein, which is targeted to mitochondria through this hydrophobic C-terminal domain (see ). To analyze the role of mitochondrial translocation of Bim, three mutants were generated. The first one, Bim(1–88), was generated by introducing a stop codon after amino acid 88, deleting the hydrophobic C terminus. The second mutant (BimTom5) was engineered to consist of the amino acids 1–84 of Bim followed by the mitochondrial tail-anchor sequence of yeast Tom5 (which is imported into the outer mitochondrial membrane), a well-investigated member of the mitochondrial outer membrane protein translocation complex (). The third mutant (BimΔΔTom5) differed from the second by the deletion of two amino acid residues in the BH3 domain predicted to abolish its pro-apoptotic activity (). Localization of the mutants in transient transfection of T-REx-293 cells was as predicted (): Bim(1–88) was detected almost exclusively in the cytosol, whereas BimTom5 was mostly found on mitochondria. The majority of BimΔΔTom5 was mitochondrial with a cytosolic fraction, possibly an overflow from mitochondria as this mutant was expressed at higher levels. BimΔΔ (full-length Bim containing the BH3 domain double deletion, ) was also targeted to mitochondria, as was Bim(D69A). Co-expression of Bcl-2 failed to direct Bim(1–88) to mitochondria (). Laser-scanning microscopy was used to confirm these results with respect to Bim, Bim(1–88), and BimTom5 (). The apoptosis-inducing activity of the mutants was tested by transient transfection of T-REx-293 cells, where the expression was regulated by tet. As shown in , Bim(1–88) had completely lost its apoptosis inducing activity, while retargeting of this mutant by a mitochondrial targeting sequence (BimTom5) restored this activity not completely but to a substantial degree. The BH3 domain deletion (BimΔΔTom5) again almost abolished it ( shows the results for caspase-3 activation; similar results were seen when apoptotic morphology was assessed [Fig. S5 a, available at ]). Therefore, mitochondrial targeting and an intact BH3 domain are required for apoptosis induction by Bim in this system. In the second line of experiments, mutations were made to the BH3 domain of Bim and their effect on binding to Bcl-2 and on apoptosis induction was tested. The double deletion mentioned above and three point mutations were introduced in Bim (). When binding to Bcl-2 was tested by cotransfection of Bcl-2 and the respective Bim mutant into T-REx-293 cells, interaction with Bcl-2 (measured by co-IP) was seen for Bim wild type, BimTom5 and, albeit to a lesser extent, Bim(1–88) and Bim(R66E). It has to be pointed out that this assay only tests for interaction capacity and not actual interaction. Bim(1–88) is almost exclusively located in the cytosol, while Bcl-2 is membrane associated (compare with ), so the co-IP very likely relies on an interaction enabled only upon detergent solubilization. No binding to Bcl-2 was detected for the mutants BimΔΔ, BimΔΔTom5, Bim(D69A), and Bim(L64A) (). BimΔΔ and the point mutants showed almost no apoptosis inducing activity in this assay except for Bim(D69A), which conserved over half of the apoptosis inducing activity ( and Fig. S5 b). This group of mutants therefore contains two examples that did not show correlation between Bcl-2 binding and apoptosis induction predicted by the displacement model. Bim(R66E) shows considerable binding to overexpressed Bcl-2 () but fails to induce apoptosis, while Bim(D69A) fails to bind to Bcl-2 but has considerable apoptosis-inducing activity. When the mutant was expressed in T-Rex-293 cells, Bim(D69A)-IP brought down a small amount of endogenous Mcl-1. However, this amount was essentially the same when Mcl-1 was overexpressed (). This suggests that the small amount of coprecipitated Mcl-1 was bound to endogenous Bim, which is also found in mitochondria and cannot be separated in this assay (see ). The binding of Bim(D69A) to Mcl-1 is thus either abolished or at least very strongly reduced. As predicted by all models from the lack of binding, Bcl-2 was unable to protect TRex HeLa cells against the overexpression of Bim(D69A) in transient transfection assays, although it protected against Bim-induced apoptosis (not depicted). Notably, although Bim was found to bind to Bax, Bim(D69A) failed to do so when whole-cell lysates of T-REx-293 cells transiently transfected with the Bim mutant construct were used for Bim-IP (see Fig. S3 d). A summary of the results is given in . These results suggested that mitochondrial targeting rather than binding to anti-apoptotic Bcl-2 proteins was the determinant of apoptosis induction through Bim. To test this hypothesis in an independent system, we turned to the yeast . Although it is a matter of dispute whether yeast possesses a system for apoptotic response (; ), it is clear that the expression of Bax can kill yeast cells (), which can be detected by growth arrest and which is accompanied by mitochondrial hyperpolarization (). Notably, yeast has no recognized Bcl-2 family proteins. We therefore examined whether Bim would enhance Bax-dependent killing in yeast (which would have to occur in the absence of anti-apoptotic Bcl-2 proteins). Bim was expressed from a constitutive promoter and Bax was placed under the control of the tet-off system (i.e., Bax was expressed in the absence of tet). Yeast cells tolerated the expression of Bim without any growth delay or other detectable alteration, and no effect was seen in cells that carried the (switched-off) inducible Bax plasmid in the presence of tet (). At 48 h after removal of tet from the liquid culture, there was a clear reduction in growth that was the same in cells that expressed Bax and in cells that expressed Bim and Bax (). No further growth was observed in these cells at later time points (not depicted). When cell density was measured at earlier times, we noted that although Bax expression on its own had no significant growth-delaying effect at 24–30 h (P = 0.511; Bax vs. wt), there was a clear and significant reduction of cell growth at this point when Bim and Bax were coexpressed (; (P = 0.002, Bim + Bax vs. wt). This suggested that the expression of Bim, although not affecting growth on its own, accelerated the growth inhibition conferred by Bax. Notably, the expression of Bim(1–88) or BimΔΔ together with Bax failed to enhance the Bax effect (unpublished data). A similar acceleration/enhancement was seen when yeast were assayed for the second criterion, i.e., mitochondrial membrane potential. No change in uptake of the potential-sensitive dye rhodamine123 was seen due to the expression of Bim or in yeast carrying the uninduced Bax construct (, and unpublished data). Induction of Bax caused a partial shift of the population to higher fluorescence, as described previously (), indicative of mitochondrial hyperpolarization. When Bax was induced in Bim-expressing cells this shift was more pronounced, suggesting that Bim either increased the population of yeast cells with hyperpolarized mitochondria or increased mitochondrial membrane potential in the individual cells. Dot blot analysis of the data showed that the observed shift in rhodamine123 uptake was largely due to the appearance of a distinct population of yeast cells, which was increased in size in cells expressing both Bim and Bax as compared with Bax alone (). When Bim(1–88) or BimΔΔ were coexpressed with Bax, no enhancement of the Bax effect on mitochondrial membrane potential was seen (unpublished data). These data thus indicate a death-inducing effect of Bim in yeast cells, which depends on Bax but not on anti-apoptotic Bcl-2 proteins. In this study we analyzed molecular events during apoptosis induction by Bim. Although no evidence was found for binding of Bim to anti-apoptotic Bcl-2 proteins, we describe the mitochondrial translocation and insertion of Bim, which was sufficient to recruit and activate cytosolic Bax. Analysis of Bim mutants showed that apoptosis induction correlated with mitochondrial localization but not the ability to bind to Bcl-2. In yeast, Bim enhanced the death-inducing activity of Bax. Surprisingly, high-level expression of Bim failed to induce apoptosis when expressed in cells stably carrying the inducible construct. This may be a result of some experimental selection process (although the clones were established and selected alongside the Bim clones), or expression of some post-translational regulation mechanism that controls Bim but not Bim. Only low levels of Bim are normally found in any cell that has been investigated. This is presumably linked to the fact that Bim is not subject to any post-translational regulation and is immediately active when expressed. Although it is conceivable that the apoptosis-inducing activity of Bim is substantially different from that one of Bim, there is little evidence to suggest that. Bim contains the same BH3 domain and the same hydrophobic C terminus as the other major splice forms, whereas it lacks the DLC1-binding site and other parts of the longer forms. Bim was found to be, unlike the other forms, able to bind to Bax in certain detergents; however, a BH3 domain point mutant that had lost the ability to bind Bax was unaltered in its capacity to induce apoptosis, indicating that this binding is not physiologically relevant (; ). We also observed some binding of Bim to Bax in co-IP experiments. However, Bim(D69A) showed no Bax binding, confirming the view that this interaction is not required for apoptosis induction by Bim. When we investigated the association of Bim with mitochondria, some Bim was also detected on mitochondria, and this fraction of Bim was also inserted in the mitochondrial outer membrane (and presumably kept inactive by Bcl-2-like proteins). Furthermore, both Bim and Bim were imported into isolated mitochondria. There seems no reason to assume that, once both are active, Bim acts in any way different from Bim. Bim can bind Bcl-2 and Bcl-2-like proteins very well, and it was therefore surprising to see that tet-induced Bim failed to do so in any detectable way. Perhaps there is the possibility that the amount of anti-apoptotic Bcl-2-like proteins that needs to be neutralized is so small as not to be detectable. However, there is no evidence to support this, and all models of interactions between Bcl-2 family members are based on solidly detectable results. Importantly, when Bcl-2 was overexpressed to inhibit Bim-induced apoptosis, a direct interaction was seen. The data thus confirm the view that Bcl-2 blocks BH3-only protein-induced apoptosis by direct binding. However, Bim does not appear to induce apoptosis by binding to Bcl-2-like anti-apoptotic proteins. This conclusion is also supported by the analysis of mutants in either mitochondrial localization or BH3 domain, where apoptosis induction was strictly correlated with mitochondrial localization but not with the ability to bind Bcl-2 (; a different Bim mutant that had lost the ability to bind Bcl-2 but still induced apoptosis has been published previously []). Finally, in yeast, Bim enhanced the death-inducing activity of Bax, in the bona fide absence of Bcl-2-like proteins. Although each of these findings on its own may have their limitations, in their sum they strongly suggest that Bax activation and apoptosis by Bim occurs in the absence of binding to anti-apoptotic Bcl-2 proteins. Although Bax can be demonstrated to be bound to Bcl-2-like proteins in some detergents (), it is difficult to know how much of this interaction is physiological. The displacement model of BH3-only protein function predicts that Bax is bound to Bcl-2 until this interaction is broken up by BH3-only proteins (leaving aside for the moment the question of different localization; []). We tested the interaction between Bax and Bcl-2-like proteins by Bax-IP in lysates of mitochondria containing 1% Triton X-100 (which allows co-IP of Bax and Bcl-2). Some Bcl-2 was coprecipitated; the amount of Bcl-x was greater (as reported []), and some Mcl-1 was co-IPed as well (Fig. S3 b). There was no clear evidence of a Bim-dependent reduction of these interactions. However, the evidence is probably not conclusive because much more Bax is found on mitochondria upon Bim induction. Importantly, Bak has been found to be sequestered by Mcl-1 and Bcl-x, and an elegant model has been proposed how Bcl-x and Mcl-1 could restrain Bak and stop it from auto-activating (). Very recently, predictions of both models of BH3-only function have been put to the test, and neither model has come away unscathed. The direct activation model has it that no Bax/Bak dependent apoptosis can occur in the absence of activator BH3-only proteins, but taking away these proteins led either only to a reduction of apoptosis () or had no effect (). The Bak auto-activation proposed by the replacement model has also been queried because mutants of Bax and Bak that have lost detectable binding to anti-apoptotic Bcl-2-like proteins were functionally unaltered (). In any case, although there is the possibility that Bax is also sequestered and kept inactive by anti-apoptotic Bcl-2 proteins, it seems clear that the activation of Bax has to involve the additional step of mitochondrial translocation. Conflicting results have been reported regarding the question which domain of Bax is important for this process (; ; ), and Bax can permeabilize artificial membranes in the absence of other proteins (), suggesting that Bax is not regularly imported into mitochondria but uses an alternative way of translocation. However, some signal must trigger this translocation. Although a peptide representing the BH3 domain of Bim (or Bid) can activate Bax (), high concentrations of peptide are required, and recent work shows that tethering Bid peptide to membranes strongly enhances its Bax-activating activity (). Similarly, we detected no killing activity in Bim(1–88) (lacking the mitochondrial targeting domain), but substantial activity in BimTom5 (targeted to the outer mitochondrial membrane); the differences seen between studies using peptides and BH3-only protein may therefore simply be a matter of abundance. The essential activity of Bim thus appears to be the activation of Bax. Although some interaction between the proteins can be demonstrated, mutant analysis shows that this detectable interaction is not required for apoptosis induction. Furthermore, experiments by many investigators with chemical cross-linkers typically showed only dimeric and trimeric Bax on mitochondrial membranes, arguing against a complex containing Bax and BH3-only proteins (although large complexes perhaps would go unnoticed). It is completely unclear how Bim conveys an activation signal to cytosolic Bax. Perhaps Bim causes changes to composition and topology of the outer mitochondrial membrane that will initiate an alteration in Bax. At least in most cells, some Bax is found on mitochondria, possibly in exchange and balance with cytosolic Bax. It seems therefore conceivable that the insertion of Bim into mitochondria somehow affects mitochondrial Bax and removes it from the reaction, causing recruitment of Bax from the cytosol until a critical concentration is reached in mitochondria and Bax auto-activates. A final point is the question whether the behavior we saw for Bim also holds true for other BH3-only proteins. Published studies, including analyses of gene-deficient mice, indicate that Bim, Bid, and Puma are the strongest apoptosis inducers within this group. This may be because, as suggested, the capacity of directly activating Bax () or their affinity for all anti-apoptotic Bcl-2 proteins rather than only a subset (; ). However, another point can now be made for mitochondrial insertion. It has been noted before that active Bid localizes to mitochondrial membranes (; ), and mitochondrial import of Bim was directly shown in this study. Of the other BH3-only proteins known, only Puma has a consensus mitochondrial tail-anchor sequence. A third possibility for the differing activity thus appears to be the ability to localize to mitochondrial membranes. T-REx-293 cells transiently transfected with different mBim mutants and T-REx-HeLa cell lines (Invitrogen) that stably express the tetracycline repressor from the pCDNA6/TR vector were cultured in DME supplemented with 10% fetal calf serum (Tet negative; PAA Laboratories), 5 μg/ml blasticidin, 50 μg/ml gentamycin and 20 μg/ml vancomycin at 37°C/5% CO. T-REx-HeLa cell lines stably expressing human hBim or murine mBim were additionally supplemented with zeocin (125 μg/ml). Induction of mBim mutant proteins was performed 7–24 h after electroporation of T-REx-293 cells. For immunoprecipitation of Bim mutants, T-REx-293 cells were cotransfected with a pEF-hBcl-2 construct (a gift from Dr. David Huang, Walter and Eliza Hall Institute for Medical Research, Victoria, Australia) or with pCDNA-hMcl-1 (a gift from Dr. Joseph Opferman, St. Jude Children's Research Hospital, Memphis, TN). Stable T-REx-HeLa cell lines expressing mBim protein were established using pcDNA4 vectors coding for mBim (see below) according to the manufacturer's instructions (Invitrogen). The genes coding for mBim were amplified by PCR and subcloned into the pcDNA4/TO/-HisA vector (Invitrogen) with a stop codon at the 3-prime end to avoid transcription of the His tag. For generation of the Bim-BH3 mutant proteins, the QuickChange II Site-Directed Mutagenesis kit (Stratagene) was used. C-terminal truncated Bim protein (Bim(1–88)) was obtained by PCR. Replacement of the C-terminal domain (aa 85–110) including the hydrophobic part of mBim with the hydrophobic domain of the yeast Tom5 protein (aa 23–50) was performed by two-step PCR. mBimΔΔTom5 was generated using the same mutation strategy as above. Cells were harvested in isotonic mitochondrial buffer (MB) (210 mM mannitol, 70 mM sucrose, 1 mM EDTA, and 10 mM Hepes [pH 7.5]) supplemented with 1× complete protease inhibitor cocktail (Roche) and the mitochondria were isolated as described earlier () All fractions (membrane and soluble fractions) were stored in MB-EGTA buffer at −80°C until further analyses. Protein concentrations were determined by the Bradford assay in MB-EGTA-1% Triton buffer. Intracellular localization of Bim and Bax during induction of Bim in T-REx-HeLa cells or of Bim mutants in T-REx-293 cells was analyzed by loading equal protein amounts of the membrane fraction and the soluble fraction on 12.5% SDS-PAA gels. Bim and Bax were detected with anti-Bim (Sigma-Aldrich) or anti-Bax antibodies (clone 6A7; Upstate Biotechnology). A monoclonal 20E8 antibody against cytochrome oxidase subunit IV (CoxIV; MoBiTec) and a monoclonal antibody directed against caspase-8 (1C12 clone; Cell Signaling Technology) were used to detect marker proteins. The insertion of Bax and Bim into the mitochondrial membrane during Bim induction was tested by alkali extraction of the mitochondrial pellets as described earlier (). The membranes were pelleted by centrifugation (260,000 for 1 h at 4°C); proteins (the alkali-sensitive fraction) were precipitated by 12% TCA and the corresponding volume of supernatants and mitochondrial pellets were separated by SDS-PAGE. Cells or isolated mitochondria were extracted in lysis buffer (50 mM Hepes [pH 7.5], 150 mM NaCl, 1 mM EDTA, 10% [vol/vol] glycerol, and 1% [vol/vol] Triton X-100) supplemented with a protease inhibitor cocktail for 30 min on ice. The lysate was spun at 13.000 for 10 min. For immunoprecipitation of Bim, protein G–Sepharose beads were washed twice with lysis buffer followed by preincubation with monoclonal rat anti-Bim antibody (3C5 clone, a gift from Dr. Andreas Strasser, Walter and Eliza Hall Institute for Medical Research, Victoria, Australia) for 1 h at 4°C. As a control an antibody directed against mBmf (provided by Dr. Strasser) was used. The beads were incubated in the presence of 45–200 μg of proteins overnight with constant agitation at 4°C. After washing four times with lysis/1% (vol/vol) Triton X-100 buffer (see above), bound protein was eluted off the beads by boiling in 2× Laemmli sample buffer and resolved by SDS-PAGE. Bax-IP was performed with 130 μg of mitochondria as described by the manufacturer with a human-specific polyclonal Bax antibody (1:100; Cell Signaling Technology) using the same buffers as decribed above. Detection of Bim, Bcl-2, Mcl-1, Bcl-w, A1, Bcl-x, and Bax was done by Western blotting using antibodies directed against Bim and Bax (see above). Bcl-2, Mcl-1 (clone 22), and Bcl-x(clone 2H12) antibodies were from BD Biosciences; Bcl-w (clone 31H4) and A1 antibodies were from Cell Signaling Technology. After induction of Bim or Bim mutants, cells were washed in PBS before fixation in 4% formaldehyde for 20 min at room temperature. Cells were washed in PBS and incubated in the presence of monoclonal anti-active-caspase-3 antibody (BD Biosciences) in permeabilization buffer (0.5% [wt/vol] BSA and 0.5% [wt/vol] saponin in PBS). Cells were washed in permeabilization buffer and incubated with species-specific Cy3-conjugated secondary antibody (Dianova). Flow cytometry was performed using a FACSCalibur (Becton Dickinson). Mitochondria were isolated from clone A2 6 h after tetracycline induction (control: untreated) and were incubated either with disuccinimidyl subernate (DSS, Pierce Chemical Co.) or bis(sulfosuccinimidyl) subernate (BS; Pierce Chemical Co.) as decribed earlier (). Samples were directly dissolved in Laemmli sample buffer and then analyzed by Western blotting with an anti-Bax antibody (Clone 3; BD Biosciences). For localization of EGFP-Bax during Bim induction, Bim clone A2 was transiently transfected with an expression plasmid for EGFP-Bax (a gift from Dr. Christoph Borner, University of Freiburg, Freiburg, Germany). Cells were then seeded onto coverslips for overnight culture, and mitochondria were stained with MitoTracker Orange CMTMRos (Invitrogen) 30 min before washing with fresh media and tetracycline induction of Bim. To prevent detachment of cells, 25 μM of zVAD-fmk was added to block caspase-dependent apoptosis. Cells were fixed in 3.7% formaldehyde for 15 min, permeabilized with 0.1% Triton X-100 for 4 min, and blocked with 5% heat-inactivated donkey serum for 10 min. Primary antibody incubations with rabbit anti-Bim (StressGen Biotechnologies) were performed for 40 min, followed by incubation with secondary donkey anti–rabbit-Cy5 antibody (1:150; Dianova) for 30 min. 0.5% BSA-PBS was used for blocking, antibody incubations, and wash steps, and coverslips were mounted in Mowiol. For localization of Bim and Bim mutants, T-REx-293 cells were transiently transfected with the respective expression plasmid, seeded onto coverslips and induced 7 h later for further 15 h with tet, stained for mitochondria, and processed as described. Cells were stained with rat anti-Bim (1:100, clone 3C5), followed by donkey anti–rat Alexa-488 (1:200; Invitrogen). Radiolabeled (S-Met) Bim and Bim proteins were generated using the TNT reticulocyte transcription/translation system (Promega). Mitochondrial import experiments were performed essentially as described previously (). The coding sequence of mouse Bax was amplified by PCR and cloned into the tet-off plasmid pCM189 (). Mouse Bim and Bim mutants were inserted into the constitutive-expression vector p415-ADH, respectively (). The plasmids were transformed into the strain EGY-48. Yeast cells were grown in SD medium containing 2% lactate and 1 μg/ml tetracycline. For induction of Bax, cells were shifted to medium without tetracycline (start at OD = 0.1). After ∼8 h, cultures were diluted to OD = 0.1 (no differences of growth were seen at this point in time between cultures), and culture was continued. OD was measured at indicated time points. Significance of the observed differences was tested by ANOVA followed by the least square difference post-hoc test on data from independent experiments. For FACS analysis of the mitochondrial membrane potential, yeast cells were incubated with 5 μM rhodamine123 for 30 min at 30°C in the dark. Subsequently, cells were isolated by centrifugation, washed in PBS, and resuspended in PBS. Stained cells were immediately analyzed in a FACSCalibur cytometer (Becton Dickinson). Fig. S1 shows induction of apoptosis by tet-dependent Bim induction. Fig. S2 shows how Bcl-2 overexpression blocks Bim-induced apoptosis in the A2 Bim clone. Fig. S3 describes interaction of Bax, Bim/Bim(D69A), and prosurvival Bcl-2-like proteins by IP. Fig. S4 shows subcellular localization of Bax and Bim in the 3C4 Bim clone. Fig. S5 shows pro-apoptotic activity of Bim and Bim mutants by Hoechst staining. Online supplemental material is available at .

Zinc is a structural constituent of a great number of proteins, including enzymes belonging to cellular signaling pathways and transcription factors, and it is essential for their biological activity (; ). Zinc has a variety of effects on the immune and nervous systems in vivo and vitro, and these effects mainly depend on the zinc concentration (; ). Many researchers have reported that immune function decreases after zinc depletion. Zinc-deficient mice exhibit reduced natural killer cell–mediated cytotoxic activity, antibody-mediated responses, and host defense against pathogens and tumors (; ; ). The requirement for zinc is most likely because of its essential constitutive role in maintaining the conformation or enzymatic activity of many important components of these processes, including enzymes, transcription factors, and signaling molecules. On the other hand, zinc itself is cytotoxic: zinc induces apoptosis in T and B cells (; ) and neuronal death (; ). Therefore, the intracellular zinc concentration is tightly controlled by zinc importers (ZIPs/SLA39s; ), exporters (zinc transporters/SLC30s; ), and binding proteins such as metallothioneins (). In addition, zinc-sensing molecules such as metal response element–binding transcription factor-1 respond to free zinc levels by regulating gene expression to maintain zinc homeostasis (). Zinc has been shown to act as a neurotransmitter (; ). In neurons, exocytotic stimuli induce zinc release into the surrounding milieu and its uptake into the cytoplasm through gated zinc channels on neighboring cells. Synaptically released zinc probably travels to adjacent cells such as postsynaptic neurons and glial cells and functions as a modulator and mediator of cell-to-cell signaling (; ; ). In this role, zinc acts as an autocrine or paracrine, transcellular, transmembrane signaling factor, like a neurotransmitter. Zinc mimics the actions of hormones, growth factors, and cytokines, which suggests that zinc may act on intracellular signaling molecules (). In fact, zinc is a well-known inhibitor of protein tyrosine phosphatases (). The inhibition constant is reported to be in the nanomolar range (). In addition, zinc affects the regulation of transcription factors. Zinc can induce the expression of some genes, including those coding for molecules involved in zinc homeostasis, like zinc transporters and metallothioneins (). The gene expression of metallothioneins by zinc is regulated by metal response element–binding transcription factor-1 (). We previously reported that the nuclear localization of the transcription factor Snail is dependent on the zinc transporter Zip6, suggesting that zinc plays a role in the nuclear localization of Snail and may act as an intracellular signaling molecule (). This notion was further supported by the finding that toll-like receptor 4–mediated dendritic cell maturation is, at least in part, dependent on a toll-like receptor 4–induced decrease in intracellular free zinc (). Collectively, this evidence suggests that zinc may act as an intracellular signaling molecule. However, the toll-like receptor 4–mediated decrease in intracellular free zinc is dependent on the change in the expression profile of zinc transporters. Therefore, it remains unknown whether zinc acts as an intracellular second messenger like calcium and cAMP. A second messenger is defined as a molecule whose intracellular status is directly altered by extracellular stimuli and that can transduce the extracellular stimuli into intracellular signaling events. In this study, we report that an extracellular stimulus such as high affinity IgE receptor (Fcɛ receptor I [FcɛRI]) cross- linking directly induces a release of free zinc from the area of the ER in mast cells, a phenomenon we call the zinc wave. The zinc wave occurred in a manner dependent on calcium influx and MAPK/extracellular signal-regulated kinase (ERK) kinase (MEK) activation. Based on our results, we suggest that one of the roles of the zinc wave is to inhibit phosphatase activity, resulting in the modulation of MAPK activation and the expression of the genes for interleukin-6 (IL-6) and TNFα in mast cells. Our results show that zinc is a novel intracellular second messenger. To investigate whether the level of intracellular free zinc changes after FcɛRI stimulation, we observed its level over time using the zinc indicator Newport green DCF. The fluorescent signal was observed mainly in the cytoplasm, and a gradual enhancement of fluorescence intensity was observed several minutes after stimulation in the center of the cell rather than in its peripheral region, suggesting that the zinc was released from intracellular stores (; and Video 1, available at ). No obvious change was seen in the absence of stimulation ( and Video 2). The cell-impermeable zinc chelator diethylenetriaminepentaacetic acid (DTPA) did not inhibit the enhancement of the Newport green fluorescence, further supporting the idea that the zinc was released from intracellular stores (). Treatment of the cells by 10 μM of cell-permeable metal chelator -tetrakis (2-pyridylmethyl) ethylenediamine (TPEN) decreased the enhancement of Newport green fluorescence, but 10 μM of chelators for copper (ammonium tetrathiomolybdate), iron (2,2′-dipyridyl), or manganese (-aminosalicylic acid) did not, indicating that the increased fluorescence was specific for changes in zinc levels (). These results were supported by the relative increase of fluorescence intensity of Newport green for each metal ion (for Zn = 1.0, Cu = −0.35, Fe = 0.30, and Mn = 0.22, respectively). This increase in intracellular free zinc was observed several minutes after the stimulation, in contrast to the rapid increase in intracellular calcium, which occurred seconds afterward (). These results indicated that the FcɛRI stimulation induced an increase in intracellular free zinc. We called this phenomenon the zinc wave. FcɛRI-mediated signal transduction occurs by two pathways: the Lyn–Syk–SLP-76–PLCγ2 pathway and the Fyn–Gab2 (Grb2-associated binder 2) pathway (; ). The Lyn–Syk–SLP-76–PLCγ2 pathway is required for inositol 1,4,5-triphosphate receptor (IPR)–dependent calcium release (), whereas the Fyn–Gab2 pathway has little or no effect on intracellular calcium (). To investigate which pathway is required for the FcɛRI-induced zinc wave, we used mast cells defective in various signaling molecules. As shown in , the zinc wave was diminished in Syk- and PLCγ2-deficient mast cells but not in Gab2-deficient mast cells. Thus, the zinc wave was induced by FcɛRI via the Syk–PLCγ2-dependent pathway. Because Syk–PLCγ2 is required for calcium signaling and the zinc wave was observed after calcium influx, we next investigated whether the zinc wave requires calcium signaling. As shown in , under calcium-free conditions, the zinc wave was suppressed. Furthermore, Xestospongin C (an inhibitor of IPR) inhibited the zinc wave (), and this suppression was reversed by simultaneous treatment with a calcium ionophore, ionomycin (). These results suggested that the entry of external calcium induced by FcɛRI stimulation is required to trigger the zinc wave. However, ionomycin, which induces calcium influx, could not induce the zinc wave by itself (). Collectively, these results indicated that calcium is essential but not sufficient for the FcɛRI-induced zinc wave. To reveal what signals, in addition to calcium, are required for the zinc wave, we tested whether the Ras–MAPK pathway is involved by examining the effect of the MEK inhibitors PD98059 and U0126. As shown in , both PD98059 and U0126 inhibited the zinc wave without any effect on the calcium influx. These results indicated that both calcium and MEK signaling are involved in the zinc wave. Consistent with this notion, as shown in , simultaneous treatment with ionomycin and EGF induced the zinc wave, although neither treatment alone did so. Furthermore, the ionomycin- and EGF-induced zinc wave was blocked by a MEK inhibitor. We confirmed that EGF alone could not induce calcium influx (unpublished data). These results clearly indicated that the zinc wave is regulated by calcium- and MEK-dependent pathways. Fluorescence laser scanning is known to induce reactive oxygen species (), which may induce the release of free zinc from zinc-binding proteins such as metallothionein, independent of receptor-mediated signaling events. To identify the subcellular regions where the zinc wave originates while avoiding secondary effects caused by cell stress, we used a thin-layer illumination microscope system that was based on the total internal reflection fluorescence (TIRF) microscope (). By recording images with highly sensitive CCD video cameras, we could detect fluorescence signals with a lower laser power than is usual for observation under a confocal laser microscope. Using this modified TIRF microscope, we observed the FcɛRI stimulation–dependent zinc wave (). To learn where the zinc wave was generated, we performed 3D imaging of mast cells costained with Newport green and the ER-specific marker ER-tracker red. As shown in , the fluorescence signal of the Newport green was enhanced mainly in the perinuclear and nuclear areas after FcɛRI stimulation. Importantly, the fluorescence intensity of the Newport green initially increased specifically in the area stained by the ER marker (, 0.5 min). This observation suggested that the zinc wave most likely originated in regions that include the ER, although we do not neglect the possible involvement of other sources, such as the nucleus. To determine whether the zinc wave can affect signaling events, we investigated the effect of zinc depletion by TPEN and of enforced zinc influx using the zinc ionophore pyrithione on FcɛRI-induced signaling events. TPEN treatment inhibited the zinc wave (), whereas pyrithione/zinc (Py/Zn) treatment rapidly increased the free zinc in cells (). TPEN treatment decreased the FcɛRI-induced IL-6 and TNFα mRNA expression () and the activation of ERK and JNK (). In particular, 5–60 min after stimulation, ERK activation was still impaired (). On the other hand, Py/Zn prolonged the expression of IL-6 and TNFα mRNA as well as the activation of the MAPKs induced by FcɛRI stimulation (). We confirmed that the Py/Zn-induced sustained MAPK activation was canceled by TPEN (). These results suggested that one of roles of the zinc wave is to regulate the duration of MAPK activation and modulation of the late phase of these signaling events. Zinc is reported to inhibit phosphatase activity (); therefore, one likely target of the zinc wave is phosphatase. In fact, Py/Zn treatment enhanced the total FcɛRI-induced tyrosine phosphorylation in mast cells (). Furthermore, the phosphatase activity of the FcɛRI-stimulated mast cells was inhibited by the addition of Py/Zn, and this inhibitory effect was rescued by TPEN (). To verify this result, we tested whether zinc directly inhibits tyrosine phosphatase activity using mast cell lysate. The results showed that zinc addition inhibited tyrosine phosphatase activity in a dose-dependent manner (). Collectively, the results support the hypothesis that the zinc wave plays a role, at least in part, in the modulation of signaling efficiency, with phosphatase as one of its targets. Several studies have shown that many secretory cells, such as nerve, pancreatic, and mast cells, contain granules with high concentrations of zinc (; ; ). In nerve cells, zinc released by exocytotic stimuli can be taken up into the cytoplasm, resulting in an increase of free zinc (; ; ; ). In these studies, the source of the increased zinc ion was extracellular. In contrast to these observations, we showed that the source of the zinc wave induced by FcɛR1 stimulation was intracellular, not extracellular. The possibility of extracellular zinc influx was unlikely because the cell-impermeable zinc chelator DTPA did not inhibit the zinc wave (), and as shown in , the zinc wave was still observed in mast cells derived from Gab2-deficient mice, which are defective in FcɛRI-mediated degranulation (), negating the possibility that zinc released from mast cells upon degranulation was the source of the zinc wave. Moreover, we showed that upon FcɛRI stimulation, the zinc wave originated in the perinuclear region that includes the ER (). This observation was confirmed using a thin-layer illumination microscope and ER marker. The zinc wave was first seen in the region that was stained with an ER marker, such as ER-tracker red, and then the increased free zinc was observed in the nucleus (). At present, however, it remains possible that other intracellular compartments, including the nucleus and mitochondria, contribute to the zinc wave. Although the precise intracellular source of the zinc wave is still uncertain, it would be interesting if the zinc wave originates in the ER, considering that calcium signaling is elicited through IPR stimulation on the ER membrane. Although the mechanism is still an open question, an attractive hypothesis is that the integration of calcium- and MEK-dependent events causes the activation of unknown molecules capable of stimulating IPR-like molecules or ZIP family members present on the ER membrane to elicit the zinc wave. The interpretation of many of the effects of zinc on phosphorylation-dependent signal transduction events, including those in insulin signaling, requires an evaluation of whether or not they occur at physiological zinc concentrations (). However, it is still unclear how zinc regulates signaling events downstream of receptor-mediated activation that lead to changes in biological activity, such as cytokine production. Regarding cytokine gene expression, we previously showed that the FcɛRI-induced gene expression of IL-6 and TNFα requires zinc-dependent mechanisms because the pretreatment of mast cells with TPEN inhibited this gene induction, at least in part, by inhibiting PKC activation (). Our previous results clearly showed the presence of zinc-dependent mechanisms in the PKC–nuclear factor κB signaling pathway (). Because Py/Zn treatment, which mimicked the zinc wave, induced neither PKC nor nuclear factor κB activation (unpublished data), that requirement for zinc is most likely in its previously recognized role as a constituent of signaling molecules essential for maintaining their proper conformation or enzymatic activity. In this study, we show that free zinc at the levels elicited by the zinc wave after FcɛRI stimulation enhances the transcription of the genes for the cytokines IL-6 and TNFα. This enhancing effect of free zinc was cancelled by simultaneous treatment with TPEN, negating the possibility that this was an artificial effect of adding zinc to the culture medium. Furthermore, we showed that free zinc could inhibit tyrosine phosphatase activity, suggesting that tyrosine phosphatases are possible targets of the zinc wave. Calcium signaling is rapidly induced seconds after stimulation, whereas the zinc wave was elicited in minutes. We showed that the zinc wave might modulate signaling events by affecting several molecules, including tyrosine phosphatase activity, which is consistent with reports that zinc inhibits phosphatase activity (). It is generally thought that the final output of signaling is dependent not only on the quality of the signal but also on its quantity (for instance, its duration and the signal strength). For example, NGF-induced dendrite outgrowth is totally dependent on the duration of MAPK activation (; ). We hypothesize that one of the roles of the zinc wave is to modulate signaling quantity, thereby playing a critical role in determining the final output of the signaling pathway. Zinc is a structural constituent of a great number of proteins, including enzymes of cellular signaling pathways and transcription factors, and it is essential for their biological activity. In these cases, zinc binds tightly to proteins containing the zinc finger motif and maintains their structure. However, zinc has not been thought to play a role as an intracellular second messenger capable of transducing extracellular stimuli into intracellular signaling pathways, like calcium and cAMP. In neurons, exocytotic stimuli induce zinc release into the surrounding milieu, and it is then taken up into the cytoplasm of neighboring cells through gated zinc channels. In this case, the action of zinc is very similar to that of neurotransmitters, which are stored in membrane-enclosed synaptic vesicles and released by exocytosis, activating postsynaptic cells through transmitter-gated ion channels (; ; ; ; ). However, the action of zinc as a neurotransmitter is different from the conventional concept of a second messenger. In this study, we showed that an extracellular stimulus such as FcɛRI stimulation induced an increase in intracellular free zinc, which we called the zinc wave, originating in the region of the ER. Furthermore, the zinc wave was observed under conditions in which either the extracellular zinc influx or the exocytosis of granules, which are rich in zinc, was inhibited. Collectively, our observations indicate that the zinc wave is a completely different phenomenon from that already reported in neurons; rather, zinc is a novel intracellular second messenger. This conclusion is drawn from the following results. An extracellular stimulus, such as FcɛRI cross-linking, directly induced an increase in intracellular free zinc, the zinc wave. The source of zinc was an intracellular compartment, possibly the ER. Free zinc at a level similar to that elicited by the zinc wave affects intracellular signaling molecules, such as tyrosine phosphatase, and, therefore, it could modulate the final output triggered by extracellular stimuli. We previously showed that the Stat3-Liv1 (Zip6) cascade is critically involved in the epithelial-mesenchymal transition and is required for the nuclear localization of Snail1, a zinc finger–containing repressor (). In addition, toll-like receptor–mediated signaling decreases the intracellular free zinc in dendritic cells, and this decrease is required for dendritic cell activation (), suggesting that zinc acts as a signaling molecule. An important difference between our current observations and our previous findings is that the zinc wave was observed several minutes after the stimulation, whereas the change in free zinc induced by toll-like receptor ligand was observed several hours after stimulation. In addition, the latter was totally dependent on a change in zinc transporter expression. We propose that intracellular zinc signaling can be classified into at least two categories: one is late zinc signaling that is dependent on a transcriptional change in zinc transporter expression, and the other is the zinc wave, an early zinc signaling pathway that is directly induced by an extracellular stimulus, such as FcɛRI. Under the latter condition, zinc acts as an intracellular second messenger capable of directly transducing the extracellular stimulus into intracellular signaling events. cAMP was the first intracellular second messenger to be discovered, by ; calcium was the second. At present, a limited number of intracellular signaling effectors or modes are known, including cAMP, calcium, NO, lipid mediators, G proteins, and related molecular mediators, protein phosphorylation, and dephosphorylation (). We do not know whether the zinc wave occurs in cell types other than mast cells, and this important issue remains to be resolved in the future. Nevertheless, our results support the idea that zinc is a novel second messenger/signaling ion that has the potential to influence many aspects of cellular signaling through its effect on zinc-binding proteins because there are many transcription factors and enzymes containing zinc-binding sites. This novel finding yields new insight into the areas of cell signaling and biological response. Bone marrow–derived mast cells (BMMCs) were prepared as described previously (). Syk-, Gab2-, and PLCγ2-deficient mice, which are crosses of C57BL6 and 129Sv, were generated as described previously (; ; ). Syk-deficient mice were provided by V.L.J. Tybulewicz (Division of Immune Cell Biology, National Institute for Medical Research, London, UK). TPEN (a = 10 M, 10 M, 10 M, and 10 M for Zn, Fe, Cu, and Mn, respectively; ; ), 2,2′-dipyridyl (a = 10 M for Fe; ), -aminosalicylic acid (Mn chelator; ), DTPA (membrane-impermeable zinc chelator; ; ), Xestospongin C, and ionomycin were purchased from Sigma-Aldrich. Ammonium tetrathiomolybdate (Cu chelator; ) was purchased from Sigma-Aldrich. Newport green DCF diacetate, Fluo-4 AM, ER-tracker red, and pyrithione were purchased from Invitrogen. The relative increase of fluorescence intensity of Newport green DCF is calculated from the fluorescence intensity of 1 μM ion-containing solution and that of reference solution containing 10 μM EGTA and 10 μM TPEN. PD98059 was obtained from Calbiochem. U0126 was obtained from Cell Signaling, and EGF was purchased from PeproTech. The rabbit antiactive MAPK, anti-ERK1/2, and antiactive JNK polyclonal antibody were purchased from Promega. The rabbit anti-JNK1 polyclonal antibody was purchased from Santa Cruz Biotechnology, Inc.. 1 × 10/ml BMMCs were sensitized with 1 μg/ml IgE (anti-DNP IgE clone SPE-7; Sigma-Aldrich) for 6 h at 37°C. IgE-sensitized mast cells were washed three times, resuspended in Tyrode's buffer (10 mM Hepes, pH 7.4, 130 mM NaCl, 5 mM KCl, 1.4 mM CaCl, 1 mM MgCl, and 5.6 mM glucose), allowed to adhere to a poly--lysine–coated glass-bottom dish, and incubated with 10 μM Newport green or 5 μM Fluo-4 for 30 min at 37°C. These dyes are cell permeant. Surplus fluorescence indicator and floating cells were removed by at least three washes with Tyrode's buffer. Cells were stimulated with 100 ng/ml dinitrophenylated human serum albumin (Sigma-Aldrich) at 37°C. The images of fluorescent signals were captured every 10 or 30 s by an inverted microscope (Axiovert 200 MO; Carl Zeiss MicroImaging, Inc.) with an oil plan Neofluar 100× NA 1.3 objective (Carl Zeiss MicroImaging, Inc.), CCD camera (CoolSnap HQ; Roper Scientific), and the system control application SlideBook (Intelligent Imaging Innovation) at 25°C. Obtained images were processed with Photoshop software (Adobe) to adjust for size and contrast. Newport green and ER tracker were detected using a thin-layer illumination microscope system based on the TIRF microscope () at 25°C. For illumination, two laser lines of 488 nm (Sapphire 488–20-OPS; Coherent) and 558 nm (YA11-558; Megaopto) were directed through a 100× NA 1.45 oil immersion objective (PLAPAO100XOTIRFM; Olympus) to an inverted microscope (IX-81; Olympus). The fluorescence images were collected at every 0.5-μm slice from the bottom to the top of the cells during scanning along the z direction using two electron-bombarded CCD cameras (C-7190-23; Hamamatsu Photonics), each equipped with an image intensifier (C8600-05; Hamamatsu Photonics). The images were captured using AQUACOSMOS software (Hamamatsu Photonics) and processed to obtain 3D images using an averaging method of AQUACOSMOS software, a deconvolution method of Volicity software (Improvision), and γ adjustments (Adobe Photoshop). Anti-DNP IgE-sensitized mast cells were washed, resuspended in Tyrode's buffer, and incubated with 10 μM Newport green for 30 min in suspension at 37°C. The cells were washed twice with Tyrode's buffer and incubated with the indicated metal chelators at 10 μM for 10 min at 37°C. The cells were stimulated with DNP–human serum albumin for 15 min at 37°C and fixed with 4% PFA in PBS on ice. The intensity of the Newport green fluorescence was analyzed by flow cytometry. Anti-DNP IgE-sensitized BMMCs were stimulated with DNP–human serum albumin at 37°C. After the indicated times, the cells were harvested and lysed with lysis buffer (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1% NP-40, proteinase inhibitors, 5 μg/ml pepstatin, and 10 μg/ml leupeptin) for 30 min at 4°C and spun at 12,000 at 4°C for 30 min. The eluted and reduced samples were resolved by SDS-PAGE using a 4–20% gradient polyacrylamide gel (Dai-ichi Kagaku) and transferred to a polyvinylidene difluoride membrane (Immobilon-P; Millipore). For immunoblotting, the membranes were incubated with antiphosphotyrosine, antiphospho-ERK, antiphospho-JNK, anti-ERK, or anti-JNK. After the reaction with the first antibody, the membranes were incubated with HRP-conjugated anti–mouse or rabbit IgG (Zymed Laboratories) for 1 h at room temperature. After extensive washing of the membranes, immunoreactive proteins were visualized using the Renaissance chemiluminescence system (Dupont NEN) according to the manufacturer's recommendations. The chemiluminescence images of the polyvinylidene difluoride membranes were captured with a chemiluminescence and fluorescence imaging system (LAS-1000; Fuji) and analyzed with Image Gauge software (Fuji). Cells were homogenized, and total RNA was isolated with the RNeasy Protect kit (QIAGEN) according to the manufacturer's instructions. For standard RT-PCR, cDNA was synthesized from 500 ng of the total RNA by incubation with reverse transcriptase (ReverTra Ace; Toyobo) and 500 ng of oligonucleotide (dT) primer for 30 min at 42°C (Invitrogen). A portion of the cDNA (typically a 1/20 volume) was used for the standard PCR to detect IL-6, TNFα, and glyceraldehyde-3-phosphate dehydrogenase. 25 cycles of PCR were performed with 0.5 U rTaq DNA polymerase and 10 pmol of gene-specific sense and antisense primers. Amplified segments of RT-PCR for IL-6, TNFα, and glyceraldehyde-3-phosphate dehydrogenase were 141 bp, 175 bp, and 227 bp, respectively. Primers used in these experiments were purchased from Invitrogen, and sequences were as follows: IL-6, forward primer (5′-GAGGATACCACTCCCAACAGACC-3′) and reverse primer (5′-AAGTGCATCATCGTTGTTCATACA-3′); TNFα, forward primer (5′-CATCTTCTCAAAATTCGAGTGACAA-3′) and reverse primer (5′-TGGGAGTAGACAAGGTACAACCC-3′); and glyceraldehyde-3-phosphate dehydrogenase, forward primer (5′-TTCACCACCATGGAGAAGGCCG-3′) and reverse primer (5′-GGCATGGACTGTGGTCATGA-3′). Anti-DNP IgE-sensitized BMMCs were stimulated with DNP–human serum albumin at 37°C for 30 min and lysed with lysis buffer as described in Cell lysates and immunoblotting. The cell lysate was analyzed using the Tyrosine Phosphatase Assay System (Promega) according to the manufacturer's recommendations. Videos 1 and 2 show time-lapse images of Newport green in mast cells. Online supplemental material is available at .

Skeletal muscle exhibits a tremendous capacity for regeneration in response to muscle injury induced by disease, trauma, or intensive exercise. The process of muscle regeneration involves two distinct phases: a degenerative phase, in which inflammatory cells infiltrate the site of injury and play a critical role in the phagocytosis of necrotic myofibers, and a regenerative phase, in which muscle stem cells are activated, proliferate, differentiate, and fuse to form new myofibers (; ; ; for review see ). Crushed muscle fibers produce mitogenic factors, and, in addition, activated macrophages also secrete various cytokines and growth factors that are important for the activation and proliferation of satellite cells. Normally, the regenerated muscle is functionally identical to uninjured muscle, but if the regeneration process is compromised, muscle tissue is replaced by scar tissue. Defects in the immune response (; ), altered production or signaling by growth factors (; ; ), or disruption of molecules that regulate cellular proliferation (; ; ) result in impaired muscle regeneration. Satellite cells (the skeletal muscle stem cells) are primarily responsible for regeneration (; ; ). Although normally quiescent, satellite cells are activated upon muscle damage and reenter the cell cycle, providing a pool of proliferating myoblasts that differentiate and fuse to form new myofibers, leading to the complete regeneration of damaged muscles. Quiescent satellite cells do not express the myogenic basic helix-loop-helix (bHLH) factor MyoD () but are positive for Pax7 () and CD34 (). Upon activation by damage, satellite cells proliferate and rapidly up-regulate the expression of myogenic factors MyoD and Myf5, whereas myogenin and MRF4 are up-regulated later during differentiation (). Several pathways, including hepatocyte growth factor (; ), myostatin (), Notch (), and p38α/β MAPK (), regulate the transition of satellite cells from quiescence to an activated state. In addition, various cytokines and growth factors such as interleukin-4, transforming growth factor-β, and insulin-like growth factor (; ; ) regulate proliferation and differentiation of myogenic precursor cells. Among the various pathways that regulate distinct steps of satellite cell activation, Notch signaling regulates not only the transition of satellite cells from quiescence to actively proliferating myoblasts but also plays a role in subsequent myogenic differentiation (; ; ; ; ). The Notch signaling pathway plays an important role in cellular proliferation, differentiation, and apoptosis (; ; ). Notch signaling is triggered by the interaction between Notch ligands and receptors. This interaction results in the proteolytic cleavage of Notch, which allows the release and translocation of Notch intracellular domain (NICD) into the nucleus, where it interacts with the DNA-bound transcription factor CBF-1/RBPJk (recombination signal sequence–binding protein for Jκ). This association results in the recruitment of coactivators and the activation of downstream target genes Hes1, Hes5, Hey1, Hey2, and HeyL (; ; ), which are members of the bHLH family of transcription factors. Because Notch signaling regulates different biological processes, it is not surprising that the pathway is tightly regulated at different levels, including expression of the ligands, translocation of Notch into the nucleus, and regulation of NICD activity in the nucleus (; ; ). Stra13/DEC1/SHARP-2 is a bHLH transcription factor that is expressed in a number of cell types during mouse embryogenesis (). Similar to Hes and Hey proteins, Stra13 contains an orange (O) domain, which is characteristically seen in all members of the bHLH-O transcriptional repressor subfamily (). However, Stra13 differs from the Hes and Hey families in having distinct DNA-binding properties as well as transcriptional repression mechanisms (; ). Stra13 expression is inducible by several stimuli in different cell types, and gain of function in cultured cells has demonstrated its role in the regulation of cellular differentiation, cell cycle arrest, stress response, and apoptosis. For instance, the overexpression of Stra13 in many cell types causes cell cycle arrest (; ; ; ) and promotes neurogenic and chondrogenic differentiation (; ) but inhibits adipogenesis (). We have previously shown that Stra13-null mice exhibit defects in T cell activation, which leads to the development of a lupus-like autoimmune disorder (H. ) as well as defects in radiation-induced apoptosis (). However, the role of Stra13 in the development or function of other cell types has not been determined in vivo. Because Stra13 is expressed in skeletal muscle (), in this study, we investigated its function in myogenesis and muscle regeneration in vivo. We found that although Stra13 is expressed in embryonic and adult muscles, it is dispensable for skeletal muscle development. However, analysis of Stra13 muscle upon injury revealed a striking defect in muscle regeneration characterized by degenerated myotubes, increased mononuclear cells, and fibrosis. Interestingly, Notch signaling is elevated in Stra13 primary myoblasts as well as in regenerating muscle. Strikingly, blocking Notch signaling ex vivo in Stra13 myoblasts reduces cellular proliferation and enhances myogenic differentiation. Moreover, the inhibition of Notch signaling at late stages subsequent to satellite cell activation in vivo results in improved regeneration with a concomitant reduction in cellular proliferation in Stra13 mice. We demonstrate that Stra13 antagonizes the Notch-imposed inhibition of myogenic differentiation and inhibits Notch activity in vitro by interaction with the intracellular domain of Notch1, resulting in its reduced association with CBF-1. We conclude that Stra13 plays an important role in the regulation of postnatal myogenesis by antagonizing Notch signaling. sup #text Notch signaling plays a prominent role in the regeneration of skeletal muscle (), and insufficient Notch activity in ageing satellite cells leads to reduced regenerative potential (). However, sustained Notch activity is incompatible with myogenic differentiation (; ; ). Subsequent to satellite cell activation, Notch signaling is attenuated during differentiation by up-regulation and asymmetric localization of its antagonist Numb (). Because alterations in Notch activity can have profound effects on the regenerative response, it is likely that Notch signaling is controlled at multiple steps. The results of this study reveal an unanticipated function for Stra13 as an inhibitor of Notch transcriptional activity. We demonstrate that Stra13 regulates myoblast proliferation and differentiation by modulating Notch signaling and is required for efficient muscle regeneration. Several lines of evidence suggest that Stra13, like Numb, is critical for the attenuation of Notch signaling in satellite cells. First, Stra13 primary myoblasts exhibit enhanced proliferation and defective differentiation, a phenotype that is remarkably similar to the expression of constitutively active Notch in myoblasts (; ). Second, Notch reporter activity and the expression of its downstream target Hey1 are increased in Stra13 myoblasts. The increased Notch activity in proliferating Stra13 myoblasts suggest that persistent Notch activity is not likely to be secondary to defective differentiation but rather causal to it. Note that Stra13 is expressed at low levels in proliferating satellite cells when Notch activity is high, and its expression is up-regulated during terminal differentiation concomitant with reduced Notch signaling. Third, treatment with the Notch antagonist Jagged-Fc resulted in reduced proliferation and enhanced myotube formation in Stra13 myoblasts. Fourth, Notch-induced inhibition of myogenic differentiation in C2C12 cells can be partially overcome by the forced expression of Stra13. Finally, Stra13 interacts with NICD and inhibits its transcriptional activity. Consistent with these observations in primary myoblasts, increased levels of activated Notch and Hey1 are also evident in regenerating Stra13 muscle. As predicted with elevated Notch signaling, the number of myogenic precursor cells is increased, indicating elevated satellite cell activation and myogenic precursor cell expansion during regeneration in Stra13 muscle. However, in contrast to the differentiation defect of Stra13 myoblasts in vitro, myogenic differentiation is not perturbed in Stra13 regenerating muscle. These observations suggest that in vivo, compensatory mechanisms allow myogenic differentiation to proceed despite elevated Notch activity. Several possibilities may account for this paradox. During regeneration, both MyoD and activated Notch levels are elevated in the mutant tissue. It is possible that the increase in Notch activity is insufficient to completely block MyoD levels (; ), and, thus, despite elevated Notch activity, myogenic differentiation in vivo proceeds normally. Moreover, elevated Notch1 and Hey1 are apparent in newly formed myofibers in the mutant tissue. Although the role of Notch signaling in the proliferation and differentiation of progenitor cells is well established, its potential function in differentiated cells is unclear. Because Notch activity is high in the mutants in newly formed myofibers that have already differentiated, it is possible that under these conditions, Notch is unable to inhibit MyoD levels and block the differentiation of progenitor cells. However, sustained Notch activity in newly formed myofibers may result in their degeneration, leading to defective regeneration. In support of this possibility, the injection of Jagged-Fc into injured muscles of Stra13 mice at day 7 results in improved histopathology, as seen by reduced necrosis and Ki67 cells, suggesting that deregulated Notch signaling in vivo is indeed one underlying basis of the defective regeneration of Stra13 mice. An alternative possibility for the relatively normal differentiation during regeneration could be through the altered production of growth factors and cytokines by injured muscle and immune cells that regulate satellite cell activation, proliferation, and differentiation. For instance, we have previously demonstrated the aberrant CD4 T cell activation (H. ) and reduced production of several cytokines, including interleukin-4 in Stra13 mice, which is required for effective regeneration (). Activation of peritoneal macrophages is also impaired in Stra13 mice, resulting in the reduced production of cytokines interleukin-1β and TNF-α (unpublished data). Macrophages play a critical role in muscle regeneration not only in the clearance of necrotic debris but also at late stages in the fusion of myofibers, and TNF-α receptor mutants are impaired in their regenerative response (). It is conceivable that the inflammatory reaction and expression of various cytokines after muscle injury may be altered in Stra13 mice. The reduced production of inhibitory cytokines that normally block myogenesis may allow myotube formation in vivo but would obviously have no impact on the differentiation of cultured myoblasts in vitro. Therefore, it is likely that Stra13 regulates postnatal myogenesis through cross talk with distinct pathways, and the cumulative effects of muscle–cell intrinsic and extrinsic changes lead to diminished regenerative capacity in vivo. Few inhibitors of Notch signaling have been identified, which include the cytoplasmic regulator Numb as well as the nuclear antagonists Msx2-interacting nuclear target protein (MINT), Hairless, and Nubbin (; ). Unlike Numb, which attenuates Notch signaling by interacting with NICD and regulating the endocytosis and ubiquitination of Notch, Stra13 represses Notch signaling in the nucleus by reducing its interaction with CBF-1. Because Notch activity needs to be precisely regulated spatially and temporally during muscle regeneration, it is likely controlled at multiple steps. Whether Numb and Stra13 function at distinct temporal phases or, alternatively, function at the same time but at distinct locations to ensure the complete inhibition of Notch activity remains to be determined. The intracellular domain of Notch is constitutively active in the nucleus and bypasses ligand requirement. Because Stra13 antagonizes the activity of NICD both in vitro as well as functionally in myogenic C2C12 cells, our results suggest that Stra13 may directly repress Notch signaling independently of the effects on ligand expression. Nevertheless, we cannot exclude the possibility that in vivo, Stra13 may regulate Notch signaling at additional levels, including the transcriptional regulation of Delta, regulation of Numb expression or localization, or by dimerization with Hes or Hey proteins. It will be of interest to precisely define these changes in order to fully understand the molecular mechanisms by which Stra13 regulates skeletal muscle regeneration. Stra13 mice have been previously described (H. ). 3-mo-old littermate WT and Stra13 mice were injured by applying a metal probe precooled in dry ice for 10 s. At least two mice were analyzed per time point (2, 5, 10, and 16 d after injury) in each experiment, and regeneration studies were performed four times. All animal protocols followed institutional guidelines. To inhibit Notch signaling, the quadriceps muscles were subjected to freeze injury. 7 d after injury, 15 μl (1.5 μg) Jagged-Fc fusion protein (R&D Systems) was injected into six sites of injured muscles. PBS (in 0.1% BSA) was injected in the contralateral injured muscle. 3 d after injection, muscles were collected and analyzed histologically. Anti-Stra13 (Dec1) antibody was provided by B. Yan (University of Rhode Island, Kingston, RI). Anti-MyoD, anti-myogenin, and anti-Notch1 (mN1A and C-20) were obtained from Santa Cruz Biotechnology, Inc. Antibody against activated Notch1 (Val1744) was purchased from Cell Signaling, anti-Ki67 was purchased from Novocastra, and anti-Hey1 was obtained from Chemicon. Anti-BrdU, anti-MHC, and anti–β-actin were purchased from Sigma-Aldrich, anti-Hes1 was a gift from Y.N. Jan (University of California, San Francisco, San Francisco, CA), anti-EF1α was purchased from Upstate Biotechnology, and anti-Pax7 and anti-eMHC were obtained from Developmental Studies Hybridoma Bank. Serial cross sections were collected along the length of muscle, and one in every six slides was stained with hematoxylin and eosin (HE). Sections from both genotypes with comparable damaged areas were used for histological and immunohistochemical analysis. Immunohistochemistry was performed as described previously (H. ). In brief, paraffin-embedded sections were incubated with primary antibody, and staining without primary antibody served as a negative control. After washes, slides were incubated with biotinylated secondary antibodies and developed using VECTASTAIN Elite ABC kit (Vector Laboratories). Masson's trichrome staining was performed using a kit from Diagnostic Biosystems. Phase-contrast microscopy was performed using a microscope (Eclipse TS100; Nikon) with plan Fluor 10× NA 0.3 and 20× NA 0.45 objectives (Nikon) at room temperature, and images were captured using a camera (2.2.1 Spot RT Color; Diagnostic Instruments) and Spot software (version 3.5.9; Diagnostic Instruments). All figures were prepared using Photoshop version 7.0 (Adobe). Quadriceps muscles from four mice of each genotype were embedded in paraffin. Sections from the center of the muscle were stained with HE, and the cross section area of individual myofibers was measured using ImageJ software (version 1.36b; National Institutes of Health). For all analysis, at least three to four mice per time point or treatment were used. An unpaired test was used for all statistical calculations, and P-values of <0.05 were considered to be statistically significant. Frozen sections of tibialis anterior muscle from 1-mo-old mice were stained with anti-Pax7 antibody and counterstained with hematoxylin. Both Pax7 nuclei and myonuclei were counted from several random fields for each animal, and the percentage of satellite cells was calculated as the number of Pax7 cells per 100 myonuclei. Myonuclei were identified based on their location being closely associated with myofibers, and interstitial nuclei were excluded from the counting. C3H10T1/2 and 293T were maintained in DME containing 10% FBS. Transient transfections were performed using LipofectAMINE Plus (Invitrogen). Empty expression vectors were added to normalize the amount of DNA in each well. Luciferase assays were performed using the dual luciferase system (Promega). C2C12 cells were maintained in DME containing 20% FBS. To establish stable cell lines, C2C12 cells were transfected with a vector expressing N1IC. Cells were selected with G418, and several independent colonies were tested for N1IC expression. For rescue experiments, two independent N1IC-expressing cell lines were transduced with a vector alone or retroviral vector expressing Stra13 (pBabe-Stra13) and selected with puromycin. After selection, the bulk population of infected cells was switched to differentiation medium (DME with 2% horse serum) for 4 d and immunostained with anti-MHC antibody. Primary myoblasts were isolated as previously described (). In brief, hindlimb muscles were dissected from 3-wk-old mice and digested with collagenase. Single-cell suspensions were resuspended in growth medium (20% FBS in F-10 medium supplemented with 5 ng/ml basic FGF) and plated on collagen-coated cell culture dishes. After two to three passages, >95% of cells were myogenic, as seen by staining with Pax7, c-met, and desmin antibodies. All assays were performed with primary myoblasts within passages four to seven, and three to four independent isolates were used for each assay. For differentiation assays, myoblasts were cultured in differentiation medium (DME with 5% horse serum) and harvested at the times indicated. The percentage of differentiated cells was calculated by counting the number of MHC-stained myocytes/total number of nuclei. The fusion index was calculated as the percentage of cells containing more than two nuclei within MHC cells. For proliferation assays, myoblasts in growth medium were incubated with 20 μM BrdU for 3.5 h and immunostained with an antibody against BrdU. For rescue experiments, Stra13 myoblasts were infected with pBabe or pBabe-Stra13. Infected cells were selected with 1 μg/ml puromycin for 2 d, expanded, and subjected to proliferation and differentiation assays. To inhibit Notch signaling, WT and Stra13 myoblasts were cultured in the presence of 1 μg/ml Jagged-Fc (R&D Systems) and subjected to proliferation and differentiation assays. Coimmunoprecipitation and GST pull-down assays were performed as described previously (). For coimmunoprecipitation assays with transfected plasmids, 400 μg of cell lysate was incubated with 2 μg of antibodies followed by the addition of protein A/G plus agarose beads. Proteins were eluted and subjected to SDS-PAGE. Western blotting was performed with anti-myc, anti-Flag, or anti-Stra13 antibodies. For GST pull- down assays, N1IC was translated in vitro and labeled with [S]methionine using the rabbit reticulocyte in vitro transcription-translation system (TNT; Promega). S-labeled N1IC was incubated with purified GST-Stra13 () or GST in binding buffer. Samples were run on SDS gels and detected by autoradiography. Total RNA was extracted using TRIzol (Invitrogen) or with an RNeasy kit (QIAGEN) and treated with Rnase-free DNaseI (TURBO DNA-free kit; Ambion). Quantitative real-time PCR was performed using the Quantitect SYBR green PCR kit (QIAGEN) with 10 ng cDNA as a template. Each sample was amplified in triplicate in a thermocycler (ABI Prism 7900HT; Applied Biosystems). PCR cycling conditions were as follows: 95°C for 15 min and 40 cycles of 95°C for 15 s, 55°C for 30 s, and 72°C for 30 s. Amplicon size was confirmed by agarose gel electrophoresis. Quantitative PCR standard curves were constructed by using serial dilutions of total RNA isolated from day 5 regenerating muscle. All data were normalized to glyceraldehyde-3-phosphate dehydrogenase. The primer pairs used were as follows: Hey1 (CTTGAGTTCGGCTCTGTGTTCC and GATGCCTCTCCGTCTTTTCCT), Stra13 (TACAAGCTGGTGATTTGTCGG and CTGGGAAGATTTCAGGTCCCG), and glyceraldehyde-3-phosphate dehydrogenase (AGGAGCGAGACCCCACTAACAT and GTGAAGACACCAGTAGACTCCACG). 20 μg of total RNA was run on formaldehyde-agarose gels and hybridized with P-labeled cDNA probes for Stra13 and 36B4.

Across eukaryotes, a protein trio comprising a Rab protein, a member of the family of small GTPases that regulate exchange between membrane compartments, a myosin motor, notably myosin V (MyoV), and a linker/adaptor protein powers organelle motility and polarized secretion (; ; ). For example, HeLa and MDCK cells recycle endocytosed cell surface receptors through a recycling endosome, the return leg mediated by Rab11 together with MyoV and the Rab11 adaptor/linker protein, family interacting protein 2 (FIP2) (). photoreceptors are typical polarized epithelial cells and morphogenesis of their photosensory membrane organelles, rhabdomeres, is driven by a late-pupal surge of secretory traffic that greatly expands the apical plasma membrane in a column of closely packed, rhodopsin-rich photosensitive microvilli. We recently found that Rab11 mediates membrane transport to developing rhabdomeres (), prompting us to ask if MyoV (; ) and dRip11, and FIP2 () also participate in morphogenic secretory transport. Numerous observations link MyoV to polarized membrane transport (). Budding yeast lacking essential MyoV, Myo2p, accumulate cytoplasmic post-Golgi secretory vesicles; secretion continues in mutants, but is not correctly targeted to the growing bud (). Melanocytes of mouse mutants lacking MyoVa fail to properly localize melanosome pigment organelles to the actin-rich cell periphery; expression of a MyoVa C-terminal fragment (MyoVa-CT) that displaces endogenous MyoVa from melanosomes mimics MyoVa loss (). Expression of MyoVa-CT similarly inhibits melanosome motility () and HeLa cell transferrin receptor recycling (; ; ). Notably, in polarized MDCK cells, MyoVb-CT selectively disrupts Rab11-dependent apical, but not basolateral, membrane recycling (). Parallel loss-of-function phenotypes suggest MyoV and Rab11 cooperate in membrane transport. Loss of either activity inhibits recycling of CXCR2 chemokine and M muscarinic acetylcholine receptors (; , ). Similarly, MyoV or Rab11 reduction prevents biogenesis of apical cannicular membranes in polarized hepatocytes () and decreases glutamate receptor 1 (GluR1) subunit delivery to developing synapses of hippocampal cells in culture (). Direct interaction between rabbit Rab11a and MyoVb is detected in yeast two-hybrid screens (), and deletion of MyoVb-CT's Rab11 binding sequence neutralizes its dominant-negative impact on GluR1 delivery in hippocampal neurons, suggesting MyoVb binds Rab11 in GluR1 trafficking (). Genetic interaction between Myo2p and Sec4p mutants is consistent with direct or close cooperation (). In addition to MyoV, Rab11 interacts with Rab11-FIPs at a signature Rab11 binding domain (RBD) (). Class I FIPs contain a C2 domain that targets recycling vesicles to the plasma membrane (), and truncated FIPs lacking the C2 domain inhibit receptor recycling (; ; ; ). encodes a single class I FIP, dRip11 (), but its function has not been reported. The genome includes a single MyoV gene, () (; ). embryos receive substantial maternal MyoV and the protein is ubiquitously expressed throughout development, including the adult retina, where it localizes to the base of the rhabdomere (). Mutants lacking MyoV show strong developmental delays and substantial late larval lethality. Surprisingly, rare homozygous mutant escapers showed normal embryogenesis and cellular architecture, suggesting MyoV is dispensable for the wide range of membrane trafficking that supports normal development (). Actin staining of mutant eyes showed apparently normal rhabdomeres and adult mutants were normally phototaxic, suggesting that MyoV does not play an obvious role in rhabdomere development or photoreception (). In this paper, we investigate the role of MyoV and dRip11 in the polarized membrane transport that builds rhabdomeres. We find both are essential. In MyoV mutants, rhodopsin 1 (Rh1) is not delivered to the growing rhabdomere, but instead accumulates in photoreceptor cytoplasm; rhodopsin-bearing vesicles, and the Rab11 and dRip11 they carry, do not approach the rhabdomere base. dRip11 loss similarly impairs secretory transport, delocalizing MyoV and Rab11 and promoting cytoplasmic Rh1. MyoV mutant photoreceptors also develop supernumerary rhabdomeres ectopically positioned within basolateral plasma membrane, suggesting MyoV-mediated transport suppresses formation of inappropriate rhabdomere primordia. photoreceptors harness an evolutionarily conserved protein trio to deliver polarized apical membrane traffic in cellular morphogenesis. Morphogenic secretory traffic targets the rhabdomere base, the plasma membrane at the cytoplasmic ends of the sensory microvilli (). Cytoplasm adjacent to the rhabdomere base is permeated by a dense microfilament brush, the rhabdomere terminal web (RTW), which extends from the rhabdomere base deep into photoreceptor cytoplasm (; ). Microfilaments are poorly preserved in chemically fixed tissue, but distinct “RTW cytoplasm” is manifest as organelle-poor cytoplasm behind the rhabdomere (, red highlight). RTW cytoplasm excludes even ribosomes, whose absence contributes to the light, clear appearance of RTW cytoplasm. Biosynthetic ER (blue highlight) and Golgi (green highlight) are distributed the length of the cell, in close proximity to the RTW's cytoplasmic terminus. The rhabdomere base differentiates in mid-pupal photoreceptors as the photoreceptor apical membrane resolves to a central Moesin-rich rhabdomere primordium surrounded by a Crumbs-rich supporting domain (; ; ). Once founded, the rhabdomere base organizes the RTW and receives morphogenic traffic. The stalk accepts little traffic, focusing exocytosis to the rhabdomere. The stalk links the rhabdomere to the retina's junctional network and projects it into an apical lumen, the IRS, aligned to the eye's optical axis (). Membrane transport in light-adapted late pupal photoreceptors is dynamic, with biosynthetic and endocytic traffic reflected in numerous, complex membrane compartments (). Post-Golgi secretory traffic is carried in tubular vesicles, approximately 100 nm across; endocytosed membrane gathers in multivesicular bodies. Complex membrane forms are common at the rhabdomere base, likely a consequence of extensive membrane fusion. Confocal immunofluorescence localizes Rab11 to puncta throughout photoreceptor cytoplasm, with a prominent concentration at the rhabdomere base () (). dRip11 immunolocalization resembled Rab11, with cytoplasmic puncta and localization at the rhabdomere base (). Note that Rab11 and dRip11 lie within RTW cytoplasm, overlapping the actin brush extending from the rhabdomere's curving base. MyoV concentrates across the rhabdomere base of late pupal photoreceptors, often appearing strongest at the sides (). Like Rab11 and dRip11, MyoV staining was strongly within RTW cytoplasm. Cytoplasmic MyoV was lightly diffuse with scattered brighter puncta. Westerns and genetic removal verified antibodies (Fig. S1, available at ). We used binary yeast two-hybrid assays to evaluate potential interactions among MyoV, Rab11, and dRip11. In AH109 yeast, we found strong interaction between full-length Rab11 and three truncated MyoV tail proteins; in PJ69, the shortest tail fragment tested, aa 1383–1800 did not support colony growth, while longer fragments aa 922–1800 and aa 1063–1800 did (Fig. S2 A, available at ). β-galactosidase expression in PJ69 detects Rab11 interaction for all three MyoV constructs (Fig. S2 B). To investigate potential MyoV function, we examined Rab11 and dRip11 distribution in normal and MyoV mutant photoreceptors (). MyoV bright crescents marked rhabdomere bases of normal photoreceptors (). Rab11 () and dRip11 () colocalize with MyoV at the rhabdomere base, well within RTW cytoplasm, often appearing as a lumpy and interrupted crescent overlapping MyoV. In mutant photoreceptors MyoV is strongly reduced, showing virtually no cytoplasmic and weak rhabdomere base staining. Mutant photoreceptor actin cytoskeletons appear largely normal, with well-formed RTWs, suggesting cytoskeletal organization withstands strong MyoV reduction and that RTW cytoplasm is largely intact. MyoV reduction depletes Rab11 () and dRip11 () from the rhabdomere base. In mutant photoreceptors, Rab11 and dRip11 are excluded from RTW cytoplasm, standing off from the rhabdomere base. They otherwise retain a generally punctate, membrane-associated appearance, suggesting Rab11 and dRip11 do not require MyoV for vesicle attachment. Although not quantitated, Rab11 and dRip11 staining often seem enhanced in mutant photoreceptors. Delocalization of Rab11 and dRip11 from the rhabdomere base could reflect failure of secretory vesicles to transit the RTW, so we asked if Rh1 transport was also defective in MyoV mutant photoreceptors. In normal photoreceptors, efficient transport to the growing rhabdomere keeps pace with biosynthesis and cytoplasmic Rh1 is limited to scattered puncta. Rh1 concentrates in rhabdomeres and in Rh1-positive large vesicles, multivesicular bodies containing endocytosed Rh1 (). By contrast, abundant cytoplasmic Rh1 accumulates in photoreceptors of rare homozygous escapers; late pupal escaper rhabdomeres show virtually no Rh1 (), though occasional adult escaper rhabdomeres show some Rh1. In mosaic eyes containing large homozygous patches (e.g., covering roughly a quarter of the eye), mutant photoreceptors likewise show cytoplasmic Rh1 and small rhabdomeres (). These rhabdomeres contain some Rh1, indicating limited transport still operates, presumably mediated by MyoV persisting after recombination renders cells genetically null. In smaller patches containing a handful of mutant cells, cytoplasmic Rh1 is also observed, but rhabdomeres are larger and strongly Rh1 positive (). We speculate clone size correlates roughly with MyoV reduction, with larger clones founded early in development diluting perdurant MyoV among more progeny cells and allowing for increased MyoV turnover; conversely, cells in small clones are milder hypomorphs. has a P-element insertion on first exon, 4 bp after the start codon. Consistent with a hypomorphic allele, faint but detectable MyoV is observed in Western blots of homozygous (not depicted), although it appears absent in immunostained eyes. In accord with , we observe substantial larval lethality and developmental delay in MyoV null homozygotes. Escapers are rare; 84 of 86 individually tracked homozygotes died as midsized, apparently second instar larvae over the three weeks after hatching. Overall, we find escaper nulls with a frequency of ∼2%. We propose that abundant, maternally supplied MyoV is long-lived and severely reduced protein numbers can still affect membrane delivery, albeit at a reduced rate. Chance fluctuation in maternal MyoV in amount and/or distribution into cellularizing embryos may confer just enough activity to sustain development. Little cytoplasmic Rh1 was observed in photoreceptors homozygous for hypomorphic (), or in photoreceptors expressing MyoV-RNAi (). We also reduced MyoV activity by expressing MyoV-CT during rhabdomere morphogenesis, and this resulted in strong accumulation of cytoplasmic Rh1 and rhabdomere reduction (). A striking phenotype of MyoV reduction is the appearance of ectopic rhabdomeres. In addition to principal rhabdomeres of relatively normal size and shape, extending up to 80 microns in length, MyoV mutant photoreceptors often sport rhabdomere “patches”, typically a few microns deep with profiles ranging from small, malformed microvillar groups to well-formed rhabdomeres as large or larger than principal rhabdomeres and inappropriately positioned in basolateral membrane (). Ectopic rhabdomeres are prominent in hypomorphs ( and ), mosaic patches ( and ), and in eyes in which MyoV RNAi is driven using GMR-Gal4, which begins expression at the onset of photoreceptor differentiation in the retinal epithelium ( and ). Ectopic rhabdomeres are present, but less common, in null homozygotes (). We propose limited MyoV activity is required to fully manifest ectopic rhabdomeres. We used electron microscopy in order investigate the impact of MyoV loss in cytoplasmic organization. Abnormal vesicles crowd MyoV mutant photoreceptor cytoplasm (). These appear empty with irregular profiles typically 200–400 nm across, substantially larger than normal secretory vesicles. Whether these vesicles result from homotypic fusion of stalled and undelivered post-Golgi traffic or if MyoV shapes the morphology of post-Golgi vesicles, for example pulling buds from TGN tubules, remains to be determined. Ultrastructurally, ectopic rhabdomere microvilli fully resemble those of normal rhabdomeres in dimension and packing; like normal rhabdomeres, a well-ordered line of membrane loops mark ectopic rhabdomere bases. Often, but not always, plasma membrane immediately adjacent to ectopic rhabdomeres has the “flattened” appearance characteristic of the stalk domain that surrounds normal rhabdomeres. Cross sections of large, well-organized ectopic rhabdomeres can rival normal rhabdomeres (). Ectopic rhabdomeres first become apparent in mid-pupal MyoV hypomorphs with the appearance of isolated actin-rich plasma membrane patches that resemble the rhabdomere primordium of the normal rhabdomere. MyoV-CT expression at 45% pupal development using relatively brief 20-min heat shocks produces ectopic rhabdomeres (unpublished data). As true for normal rhabdomeres, activated Moesin defines the ectopic rhabdomere base. Ectopic rhabdomeres are often, but not always, flanked by Crb-positive stalk membrane and ectopic cell–cell junctions marked by β-catenin and E-cadherin (). A set of apical membrane proteins is thus mislocalized when MyoV is reduced. As noted by , the genome contains a single class I FIP, dRip11, the product of homozygous lethal (CG6606). Entire dRip11 aligns with human class I FIPs, Rip11, Rab11-FIP2, and RCP (Fig. S3, available at ). As true for FIPs generally, dRip11 is not significantly similar to other FIPs outside of these shared homology domains. Notably, dRip11 does not contain the NPF domains that couple Rab11-FIP2 to Reps1, an EH domain protein that promotes endocytosis (; ). Similar to RCP, dRip11 contains two PEST domains that potentially promote rapid degradation (). dRip11 also conserves Rab11-FIP2 Ser258, phosphorylated by MARK2 in MDCK cells, and corresponding to dRip11 S262 (). To test for dRip11 function, we made mosaic eyes containing patches. In normal photoreceptors, Rab11 and dRip11 colocalize extensively and concentrate at the rhabdomere base and cytoplasmic vesicles (). Occasional puncta stain for only one of the proteins, suggesting they may not be obligatory partners. MyoV and dRip11 likewise colocalize at the rhabdomere base of normal cells (). Anti-dRip11 staining is strongly reduced in mutant photoreceptors. Strikingly, in dRip11-deficient cells, Rab11 and MyoV are lost from the rhabdomere base. Rhabdomeres are reduced and the actin cytoskeleton is disturbed, with diffuse cytoplasmic F-actin not observed in wild-type cells. We assessed dRip11's role in Rh1 transport during development using confocal immunohistochemistry to localize Rh1 in wild-type and mutant photoreceptors. Like MyoV and Rab11 loss, dRip11 mutant photoreceptors show loss of rhabdomere Rh1 staining and cytoplasmic accumulation throughout cytoplasm (). Rhabdomeres are strongly reduced in dRip11 mutant cells and photoreceptor cytoplasm fills with a profusion of abnormal vesicles. As Rab11-FIP C-terminal is dominant-negative for normal membrane traffic in mammalian cells in culture (; ), we asked if expression of dRab11-CT also impairs Rh1 transport in developing photoreceptors. Late pupal expression of a dRab11-CT (700–821) GFP fusion protein, dRip11-CT-GFP, delocalized Rab11 (Fig. S4, available at ). dRip11-CT-GFP contains the RBD, and our two-hybrid results showed dRab11-CT (726–821) interacts with Rab11. Diffuse but occasionally still punctate Rab11 colocalizes with dRip11-CT-GFP (Fig. S4 A, arrow). Rh1 accumulates in the cytoplasm and there is no Rh1 staining in rhabdomeres. Cytoplasmic F-actin accumulation is also seen. Electron micrographs showed massive accumulation of abnormal vesicles (unpublished data). dRip11-CT-GFP overexpression strongly phenocopies patches of mosaic eyes. photoreceptors, like many polarized epithelial cells, greatly amplify their apical membranes during terminal differentiation via targeted membrane delivery. Here we show that a protein trio (Rab11, dRip11, and MyoV) mediates this morphogenic secretory traffic. MyoV normally concentrates at the base and its loss causes three notable phenotypes of compromised apical transport: Rab11 and dRip11 delocalize from the base, Rh1 accumulates in photoreceptor cytoplasm, and ectopic rhabdomeres are formed. dRip11, the sole class I Rab-FIP (), is also required for normal Rh1 transport; its loss delocalizes Rab11 and MyoV. Together with our previous demonstration that Rab11 is essential for photoreceptor secretory traffic (), we propose MyoV pulls post-Golgi secretory vesicles, marked for rhabdomere delivery by Rab11 and dRip11, through an exclusionary subcortical cytoskeletal web along polarized microfilaments leading directly to the exocytic targeting patch at the rhabdomere base (). The RTW's role as both a barrier and a carrier for morphogenic traffic is an instance of a general theme of a dynamic regulatory role of the subcortical cytoskeleton in secretion (; ; ). Myosin S1 decoration shows RTW filaments are oriented with plus-ends at the membrane, a correct orientation for MyoV-based secretory transport (), and disruption of the actin cytoskeleton prevents the morphogenic traffic that rebuilds crab rhabdomeres at dusk (; ). The RTW's strong polarization and anchorage to a secretory targeting patch resembles the polarized actin cables that mediate budding yeast secretory traffic (). Absorptive and secretory epithelial cell specialists often regulate apical membrane activity by dynamic, Rab11-dependent exchange of plasma membrane with recycling endosomes. For example, gastric parietal cells meet demand for additional acid secretion by Rab11-, Rab11-FIP2-, and MyoV-dependent delivery of additional H/K ATPase pumps to the cell surface from a recycling endosome (; ; ). Like GPCRs generally, Rh1 is endocytosed upon stimulation but appears to be degraded rather than recycled back to the rhabdomere (; ; ). photoreceptor Rab11-dependent transport appears to be principally devoted to delivery of newly synthesized cargo from the TGN to the plasma membrane, a conserved Rab11 activity () now seen to further parallel recycling transport. Ectopic rhabdomeres in hypomorphs suggest MyoV normally suppresses the establishment of inappropriate rhabdomere primordia; oncefounded ectopic rhabdomeres develop in concert with principal rhabdomeres, presumably drawing from the same secretory traffic. We speculate MyoV normally drives traffic to the differentiating rhabdomere primordium and that positive feedback driven by the incorporation of morphogenic determinants, perhaps proteins that anchor and promote RTW development, gives the original, “true” apical membrane an overwhelming growth advantage, starving weak, inappropriate sites. MyoV reduction might diminish this advantage, allowing ectopic foci to capture sufficient morphogenic traffic to assemble a rhabdomere patch. Our observation here that MyoV is required for normal rhabdomere development differs from Mermall and colleague's report of normal rhabdomeres in mutant eyes (). However, long ribbons of the principal rhabdomeres dominate phalloidin-stained longitudinal sections, and ectopic rhabdomeres, often patches a few microns across, are not prominent. Mermall et al.'s supplementary L, a tangential section, shows actin-bright profiles apart from the principal rhabdomeres—potentially ectopic rhabdomeres. Massive biosynthetic traffic in late pupal photoreceptors sensitizes cells to compromise of efficient, accurate transport and accumulation of cytoplasmic Rh1 reflects an inability of transport to keep pace with biosynthesis. dRip11 loss inhibits secretory transport and misolcalizes Rab11 and MyoV. We suggest dRip11 couples two broad streams of membrane transport, Rab11- and MyoV-dependent activities, to drive morphogenic secretory traffic. Results here are consistent with previously demonstrated roles for FIPs as contributors to membrane targeting (; ), and as scaffolds for the growing Rab11 effector ensemble (). Similar to chromaffin cells, where MyoV only partially overlaps with secretory vesicles (), MyoV and Rab11 only partially overlap in developing photoreceptors, and it is likely MyoV transports multiple and changing cargoes. Rab11 participates in both constitutive and Ca-regulated secretion (), and both cargo binding and Ca regulate MyoV activity (; ; ). Rhabdomere morphogenesis utilizes constitutive exocytosis, with substantial rhabdomere growth before Rh1 expression and photoresponse Ca influx. Rhabdomeres likewise develop normally in the dark, indicating light-dependent Ca elevation is not required for MyoV morphogenic transport. We propose dRip11, in proximity to MyoV via their mutual binding to Rab11 on post-Golgi secretory vesicles, interprets or conveys non-Ca-stimulated MyoV activation, promoting developmental MyoV secretory transport. Flies were reared at 20°C, 12 h light/12 h dark, on standard cornmeal agar food. White-eye flies () were used as wild type. Two MyoV null lines, , predicted to introduce a non-sense mutation at residue Gln1052, and (; Δ28)), a deletion mutation removing 5′ and adjacent , with transgenically restored, were provided by the Cooley and Mooseker labs (). To generate mosaic eyes was recombined onto FRT42B. virgins were crossed to to generate retinal clonal patches. A transposon allele, balanced over CyO, was obtained from the Bloomington Stock Center (#14094; Bloomington, IN). No homozygous animals were recovered in the original stock. However, upon outcrossing to or recombination onto FRT42D, homozygous mutant animals survived, suggesting the original chromosome contains a second site lethal(s). A transposon allele, balanced over , was obtained from the Bloomington Stock Center (#13742), and was combined. has a P-element insertion 83 bp upstream from the start codon. Males of the genotype (B-# 5579) were crossed to females of the genotype virgins to generate mitotic clones. To generate whole dRip11 reduced eyes, we homozygosed in the eye using the EGUF method (). Males of the genotype were crossed to virgin . Each of two constructs encoding MyoV (nt 1501–1860 and 4801–5160) inverted repeats (IR-MyoV) was inserted into pWIZ vector () to produce hairpin RNAs under UAS control. As MyoV CT is dominant-negative for membrane transport, we inserted MyoV CT (aa 768–1800) into pUAST-GFP vector, permitting UAS expression. As overexpression of a Rab11-FIP C-terminal fragment is dominant-negative in mammalian cells, dRip11 C-terminal (aa 700–832) was inserted into pUAST-GFP vector. These constructs were transformed into and insertion strains containing a single copy of each transgene were generated by standard methods. Rabbit affinity-purified anti-MyoV polyclonal rabbit antibody was generated against a unique MyoV tail domain peptide (aa 1783–1800: EDIELPSHLNLDEFLTKI) (Bethyl Laboratories). 6xHis-dRip11 C-terminal 133-aa (700–832) fusion protein was expressed in cells, purified on polyhistidine affinity resin (QIAGEN). Rabbit affinity-purified anti-dRip11 antibody was raised against this fusion protein (Bethyl Laboratories). strains AH109 (CLONTECH Laboratories, Inc.) and PJ69 () were co-transformed with the indicated constructs. Transformation and cell growth were done according to Matchmaker two-hybrid protocols (BD Biosciences). MyoV–Rab11 protein interaction was assayed via coimmunoprecipitation using a Matchmaker Co-IP kit (CLONTECH Laboratories, Inc.). Biotinylated HA-tagged MyoV (aa 1063–1800) and c-Myc–tagged Rab11 were produced from pGADT7-MyoV (aa 1063–180) and pGBKT7-Rab11 (full length), respectively, using a TT T7 Quick Coupled Transcription/Translation System (Promega) and biotinylated lysine (Promega Transcend tRNA) as directed by the manufacturer. pGBKT7-53 and pGADT7-T (CLONTECH Laboratories, Inc.) encoding murine p53 and simian virus large T antigen, respectively, were similarly produced as controls. Products were pairwise mixed with final concentrations of 200 μM GTP-γ-S, 4 mM Mg and 1X protease inhibitor (Pierce Chemical Co.), which are present in the subsequent steps. After 1 or 4 h incubation at room temperature (RT), 10 μl anti-Myc was added for an additional 1 h at RT. Protein A–Sepahrose (3 μl) was then added to each sample. After incubation for an additional 1 h at RT, samples were centrifuged and washed four times with Wash Buffer 1 and two times with Wash Buffer 2 (both buffers from CLONTECH Laboratories, Inc.) at 4°C. Synthesized protein, unbound protein, and immunoprecipitated protein were analyzed on 12.5% SDS-PAGE gel followed by Western Blot using HRP-conjugated streptavidin (Kirkegaard & Perry Laboratories) detected using ECL detection reagents (GE Healthcare). Fly heads were dissected and immersed in T-Per tissue protein extraction reagent (Pierce Chemical Co.) with additional 8 M urea. After homogenizing fly heads, SDS-PAGE loading buffer was immediately added into the lysates, boiled for 5 min, and then centrifuged at 164,000 . The supernatant was separated on 7.5% gel and transferred to PVDF membrane (Millipore), which was then probed with rabbit anti-Myo V (1:10,000), rat anti-Myo V (1:10,000), or anti-dRip11 (1:10,000). Equivalent protein loading for each sample was verified by probing the membrane with mouse anti-actin (1:10,000), a gift from Dr. J. Lessard (Cincinnati Children's Hospital Medical Center, Cincinnati, OH). Fixation and staining methods are described in . Primary antisera were: mouse anti-Rab11(1:250; ), rabbit anti-Rh1 (1:1,000; ), anti-Rh1 (4C5) mouse monoclonal (DSHB), chicken anti-GFP (1:2,000; Chemicon), rabbit anti-MyoV (1:1,000; this paper), rat anti-MyoV (1:1,000; ) and dRip11 (1:1,000; this paper). Secondary antibodies were anti–mouse, –rabbit and/or –chicken labeled with Alexa488, -568, -647 (1:300; Molecular Probes), Cy2 (1:300; GE Healthcare), or biotin-conjugated anti–rabbit (1:150), followed by streptavidin labeled with Alexa488, -647, or Texas red (1:2,000). Samples were mounted in glycerol/PBS with 1% -propyl gallate to reduce fading and imaged at room temperature using a confocal microscope (MRC1024; Bio-Rad Laboratories) (Nikon 60×, 1.4 NA lens) using LaserSharp software (Bio-Rad Laboratories). To minimize bleed-through, each signal in double- or triple-stained samples was imaged separately using a single line and then merged. Acquired images were processed by ImageJ (National Institutes of Health, Bethesda, MD) and/or Photoshop7. Image processing was fully compliant with guidelines for proper digital image handling (). Conventional electron microscopic methods were as described in , ). Samples were observed on an electron microscope (model 300; Philips). Fig. S1 shows verification of MyoV and dRip11 antibodies. Fig. S2 shows that Rab11 interacts with MyoV and dRip11. Fig. S3 shows Clustal alignment of dRip11, human Rab11-FIP2, human Rip11, and human RCP. Fig. S4 shows overexpression of dRip11 C-terminal phenocopies . Online supplemental material is available at .

Microvilli are actin-rich membrane protrusions common to all transporting and sensory epithelial cell types. The brush border (BB) domain found at the apex of the enterocyte consists of thousands of tightly packed microvilli that extend off of the apical surface to a uniform length. In the BB, each microvillus is supported by a polarized bundle of actin filaments that are linked to the overlying apical membrane by an ensemble of the membrane-binding motor protein, myosin-1a (Myo1a, originally brush border myosin I; ; ). While the cellular role of this array remains unclear, BBs in mice lacking Myo1a demonstrate a variety of defects; among the most striking are herniations of apical membrane, irregularities in microvillar packing, and abnormal variability in microvillar length (). The mechanistic details underlying these phenotypes and specifically, the involvement of Myo1a motor activity, has yet to be elucidated. Microvilli share structural features with other actin-rich membrane protrusions such as filopodia and stereocilia. In all of these cases, parallel bundles of actin filaments provide the structural core and mechanical support for the membrane extension (). The uniform polarity of filaments in these structures suggests that supporting bundles could serve as tracks for the polarized movement of motor proteins. Indeed, recent studies have demonstrated that myosin-X undergoes motor-driven movements toward the tips of filopodia (). In a similar manner, the motor activity of myosin-XVa is thought to drive its accumulation at the tips of stereocilia (; ). Moreover, recent data from demonstrate that, in the context of the kidney proximal tubule, acute hypertension induces a redistribution in myosin-VI (Myo6) immunoreactivity from microvillar tip to base, implying that this motor may use the core bundle as a track for minus end–directed movement. While Myo1a demonstrates mechanical activity in the sliding filament assay (), there is currently no data on the ability of Myo1a to move apical membrane or translate membrane components along microvillar actin bundles. Based on the geometry of the microvillus () and the known mechanical properties of Myo1a (), we predict that the ensemble of Myo1a in the microvillus represents a contractile array, exerting plus end–directed force on the apical membrane. In vivo, these forces could be engaged for the intramicrovillar trafficking of membrane lipids and proteins, or perhaps other unexplored roles. We sought to test our prediction by investigating isolated BBs under conditions expected to stimulate the mechanical activity of Myo1a. Data from time-lapse imaging studies, ultrastructural analysis, and biochemical assays all indicate that the microvillar array of Myo1a is mechanically active and exerts substantial plus end–directed force on the apical membrane. In isolated BBs, this force is manifest as a rapid plus end–directed sliding of apical membrane along microvillar actin bundles. The translation of apical membrane ultimately results in the release of vesicles from microvillar tips. These studies demonstrate that microvillar actin bundles are a suitable substrate for myosin-based movement and that Myo1a produces mechanical force sufficient to power the movement of apical membrane over the actin cytoskeleton. Intriguingly, these data may also provide a mechanism for the appearance of BB membrane vesicles in the lumen of the small intestine (). Thus, in addition to providing a means for amplifying apical surface area, we propose that microvilli function as actomyosin contractile arrays, powering the release of BB membrane vesicles into the intestinal lumen. This activity may have implications with regard to the general efficiency of nutrient processing and other critical aspects of gastrointestinal physiology. To determine if microvillar Myo1a is mechanically active, we first examined the impact of ATP on the structure of BBs isolated from rat small intestine. For these studies, phalloidin-stabilized BBs were exposed to saturating (2 mM) levels of ATP in the presence of 1 mM EGTA; this chelator was included because Ca is known to depress the mechanical activity of Myo1a in vitro, via effects on bound calmodulin light chains (). BBs were fixed 5 min after ATP addition, labeled with ConA, and examined using laser scanning confocal microscopy. These observations revealed that ATP treatment induced a striking accumulation of apical membrane at microvillar tips (i.e., actin bundle plus-ends; ). This was confirmed quantitatively as a significant loss in the correlation coefficient calculated from the BB membrane and actin probe fluorescence signals (). In many BBs, we also observed terminal web contraction, an established form of myosin-II (Myo2) mechanical activity (; ; ). To examine the dynamics of apical membrane redistribution induced by ATP, we used time-lapse differential interference contrast (DIC) microscopy. BBs were immobilized within a flowcell and imaged with the long axis of microvilli parallel to the focal plane (). When 2 mM ATP was perfused through the flowcell we observed a robust, centrifugal expansion of the refractile bands that represent apical membrane (; Video 1, available at ). Kymographs formed from a line that bisects the BB () clearly show that the bands of apical membrane translate away from the center of the structure. Movement began immediately after ATP addition and continued rapidly in the first minute of observation. After ∼1 min, movement slowed as membrane reached a distance equivalent to the original position of microvillar tips (, white dashed lines). Although these observations suggest that ATP stimulates the translation of apical membrane along microvillar actin bundles, we sought to confirm this by performing time-lapse fluorescence microscopy of BBs labeled with probes specific to the apical membrane and actin cytoskeleton. Indeed, spinning disk confocal microscopy (SDCM) of double-labeled BBs confirmed that ATP stimulates a robust translation of apical membrane toward microvillar tips (; Video 2). Kymographs of the resulting time series demonstrated that the length of microvillar actin bundles is unaffected during this process (). Moreover, when the velocity of membrane movement was tallied from multiple kymographs, we obtained a mean membrane translation velocity of 19.2 ± 6.1 nm/s (). These data indicate that ATP induces movement of apical membrane with a polarity and velocity that are consistent with the activity of a class I myosin (). Kymograph analysis of SDCM time-lapse data shows that the band representing apical membrane narrows as it translates toward microvillar tips, indicating an overall loss of BB-associated membrane after ATP addition (). One explanation for this loss is that membrane is pushed off of microvillar tips, resulting in its vesiculation or “shedding” from the BB. Indeed, the DIC images presented above () reveal that ATP-stimulated membrane movement is accompanied by an extensive release of material resembling small vesicles (, arrowheads at 68 s). Consistent with these data, SDCM images that are contrast adjusted to enable visualization of dim material reveal a cloud of vesicles in solution around the BB after ATP addition (). When total solution fluorescence is monitored throughout the time lapse, a burst in signal intensity is observed coinciding with the addition of ATP (). These data indicate that membrane shedding occurs in parallel with membrane translation. We expect membrane shedding to take place at microvillar tips, as membrane accumulates in this region after ATP addition (– ). To further characterize the BB response to ATP and unambiguously determine the location of membrane release, we performed ultrastructural analysis on ATP-treated BBs. Transmission electron microscopy (TEM) revealed extensive vesiculation of the apical membrane at microvillar tips in the presence of ATP (). Vesiculation was widespread and observed at the tip of nearly every microvillus (). High magnification panels revealed that vesiculating membrane was devoid of electron-dense material and exhibited a diameter comparable to that of the microvillus (∼100 nm; ). ATP treatment also resulted in the exposure of a greater length of actin bundle at the base of microvilli. In the absence of ATP, microvillar actin bundles typically extend into the terminal web by only a small fraction of their total length (10–20% of total length; , red region). After ATP activation, microvillar bundles in most BBs were observed with ≥50% of their total length exposed and extending into the terminal web (, red region). In other cases, we observed actin bundles that were completely stripped of membrane, yet positioned adjacent to an accumulation of vesicles (unpublished data). These ultrastructural analyses show that ATP-induced membrane translation gives rise to membrane shedding from microvillar tips. Our ultrastructural observations suggest that the vesicles released from microvillar tips after ATP treatment represent parcels of apical membrane. To further characterize the composition and properties of shed vesicles, we used differential centrifugation to fractionate ATP-treated BBs. When BBs were incubated in the presence or absence of ATP and then sedimented at 5,000 , membrane markers (sucrase-isomaltase [SI] and alkaline phosphatase [AP]) and Myo1a were released into the supernatant only in the presence of ATP (). Moreover, all of the SI and AP in the 5,000- supernatant sedimented at 100,000 , suggesting that these components are indeed membrane associated (). A significant fraction (∼20%) of the Myo1a released with ATP addition also sedimented at 100,000 , indicating that a subpopulation of this motor remains bound to the apical membrane during the time course of ATP-induced membrane translation and shedding (). The appearance of vesicles in ATP-treated preparations was also confirmed using TEM analysis; an abundance of vesiculated membrane material appeared in the 5,000- supernatant only after BBs were treated with ATP (). In addition, examination of the 100,000- pellet showed that this fraction is enriched in small vesicles ranging from 50–200 nm in diameter, comparable to those observed in ultrastructural analyses of ATP-treated BBs (). To further investigate the mechanism of BB membrane shedding, we devised a quantitative membrane shedding assay (MSA). This assay allowed us to quantify the release of apical membrane vesicles produced by a large population of BBs, under a variety of biochemical conditions and perturbations. BBs were labeled with the lipophilic fluorescent dye AM1-43 and then activated to shed with the addition of ATP. After 2 min, vesiculated membrane was separated from BB remnants with a centrifugation step and the resulting supernatant transferred to a 96-well microplate for the measurement of AM1-43 fluorescence (). Importantly, AM1-43 labeling appeared uniform within and between individual BBs (), and the ATP-dependent signal scaled linearly with the quantity of BB in the reaction (). To determine if ATP-induced membrane translation and shedding are powered by an ATPase such as Myo1a, we used the MSA to examine the nucleotide dependence of this activity. In the absence of ATP, ADP and PPi were unable to activate membrane shedding (). Next, we determined if ATP hydrolysis was required to activate membrane shedding by carrying out the MSA with the nonhydrolyzable ATP analogues, ATPγS and AMP-PNP. These analogues produced only a fraction of the response observed for an equivalent concentration of ATP, indicating that hydrolysis is required to fully activate membrane shedding (). The remnant response observed in the case of both analogues (∼10–20% of ATP control) could be due to the low concentration of contaminating ATP in these analogue preparations. Alternatively, it may indicate that ATP binding alone accounts for a small fraction of total membrane shedding. Myosin mechanical activity is known to demonstrate Michaelis-Menten dependence with respect to ATP concentration (). If BB membrane shedding is powered by a motor such as Myo1a, it should also demonstrate such dependence. To test this, we examined the rate of membrane shedding over a wide range of ATP concentrations. Indeed, these experiments revealed that shedding () exhibits hyperbolic dependence with respect to ATP concentration. Fitting these data to a Michaelis-Menten model (normalized shedding = V[ATP]/K + [ATP]) yields a V of 1.2 ± 0.1 mM and a K (for ATP) of 396 ± 103 μM. We also examined the impact of the hydrolysis product, ADP, on shedding activity in the presence of saturating levels of ATP. In vitro biochemical studies on other myosin motors have established that ADP serves as a pure competitive inhibitor of ATPase activity (). Consistent with this prediction, the presence of 2 mM ADP increased the K with respect to ATP approximately fivefold, while V was unaffected (, inset). This shift in K enabled us to calculate an ADP K of 554 μM (from K = K [ADP]/K′ − K). This value is slightly higher but comparable to that determined for smooth muscle Myo2 (in the sliding filament assay), a myosin that exhibits high affinity for ADP in a manner similar to Myo1a (; ; ). The nucleotide dependence of membrane shedding suggests that this process is powered by a myosin ATPase. Thus, we wanted to determine if Myo2, the most abundant myosin in the BB, plays a role in driving this activity. For these experiments, we performed the MSA in the presence of 50 μM blebbistatin, an established Myo2 inhibitor (; ). Consistent with the known localization of Myo2 in the terminal web, blebbistatin had no observable impact on the extent of membrane shedding from microvillar tips (). Moreover, in the presence of blebbistatin, BBs lacked the tight curvature normally produced by ATP addition (compare with ), indicating that terminal web contraction (; ; ; ), and thus Myo2 activity, were indeed inhibited. These data indicate that ATP-induced membrane translation and shedding are Myo2-independent activities. Given its established membrane binding potential (; ), high density in the microvillus (), and polarity of movement (), Myo1a is the most obvious candidate for powering the plus end–directed movement and shedding of membrane induced by ATP. Indeed, our previous studies have shown that BBs from Myo1a knock-out (KO) mice are not “damaged” to the same extent as wild-type (WT) BBs when exposed to ATP (). Thus, we investigated the involvement of Myo1a in the membrane shedding response by performing the MSA with BBs isolated from the small intestine of Myo1a knock-out (KO) mice () and age-matched wild-type (WT) controls (). Strikingly, the shedding response of Myo1a KO BBs was only a small fraction (∼5%) of that exhibited by WT BBs (). The difference in activity was not due to disparity in AM1-43 labeling efficiency as labeled WT and KO BBs exhibited comparable levels of fluorescence (). Moreover, visual observation of the response to ATP confirmed that KO BBs failed to carry out the plus end–directed membrane translation and shedding normally observed in WT BBs (). Together these data strongly indicate that Myo1a is required for the ATP-stimulated membrane movement and vesiculation observed in isolated BBs. In this paper, we describe a novel form of microvillar contractility that can be reactivated by exposing native, isolated BBs to ATP. Reactivation is manifest as the plus end–directed movement of apical membrane along microvillar core actin bundles, and eventually, the accumulation of membrane at microvillar tips. Upon reaching the tips, the membrane no longer maintains contact with the underlying actin cytoskeleton and vesiculation is favored; small vesicles that contain Myo1a and are enriched in other apical membrane markers are released into solution (). The translation and shedding of membrane requires ATP hydrolysis, demonstrates the nucleotide dependence expected for a myosin ATPase, and is substantially depressed in the absence of Myo1a. Intriguingly, the membrane translation velocities (; 19.2 ± 6.1 nm/s) measured here are nearly identical to velocities measured in sliding filament assays with Myo1a immobilized on lipid-coated coverslips (). Myo1a is a good candidate for driving this novel form of motility, as previous studies have established that this motor can bind directly to apical membrane lipids (; ) and proteins (), is present at a very high concentration (>70 μM) in the microvillus (), and moves toward the plus-ends of actin filaments in vitro (). Thus, in combination with previous studies, our data strongly indicate that Myo1a is the motor that powers ATP-stimulated membrane translation and shedding in isolated BBs. Our findings demonstrate that the polarized actin bundles that support microvilli serve as tracks for myosin-based motor activity. Although these studies focus on Myo1a, a reasonable extension of this is that other myosins in the BB () may also use core actin bundles as tracks for directed transport. Consistent with this idea, recent studies in the context of kidney proximal tubule reveal that hypertension induces the redistribution of Myo6 immunoreactivity along the microvillar axis (). Our data also show that microvillar Myo1a is mechanochemically active and generates force in its native environment. This suggests that Myo1a may function as more than a passive “linker” serving to stabilize membrane/cytoskeleton interactions (). Finally, while general models for myosin-I function have always included some form of mechanical activity involving membrane rearrangement or movement (; ), direct evidence for these functions has been lacking. The data presented here demonstrate directly, in a native system, that Myo1a is capable of producing mechanical forces sufficient to power the movement of cellular membranes over the actin cytoskeleton. A number of previous biochemical studies have documented the existence of small vesicles in the lumen of the small intestine (; ; ; ; ; ). Although these vesicles are enriched in nutrient-processing enzymes (e.g., SI and AP) in a manner similar to BB membrane, their mechanism of release remains unclear. Vesiculation from the tips of microvilli has also been captured in ultrastructural studies of intact enterocytes (), suggesting that lumenal vesicles may originate from microvillar membrane. We propose that the Myo1a-dependent mechanical activity described in this paper provides a mechanism for the formation and release of vesicles from enterocyte BBs in vivo (). This model becomes even more appealing if we consider that membrane shedding from microvilli in vivo is accelerated when enterocytes are treated with antibodies against SI (), a transmembrane disaccharidase that interacts directly with Myo1a in a raft-like complex in the microvillar membrane (). The shedding of vesicles from the plasma membrane is a well-documented activity performed by a variety of epithelial cell types under normal and pathological conditions (). In the gastrointestinal tract, the release of membrane vesicles laden with nutrient-processing enzymes could serve to increase the effective apical membrane surface area; such vesicles would allow processing to begin before nutrients reached the actual surface of the enterocyte (). Others have proposed that BB membrane shedding may allow the enterocyte to continually modify its apical membrane composition (). This form of plasticity may be a critical aspect of the enterocyte response to the shifting demands in nutrient processing and absorption that are commonplace in the small intestine. Does the membrane shedding process described here represent a general function for microvilli found on other polarized cell types? Although the expression of Myo1a is restricted to the gastrointestinal tract and inner ear (; ; ; ), other closely related class I myosins (Myo1b, Myo1c, and Myo1d) are more widely expressed in polarized cells from a variety of tissues (including kidney, liver, and pancreas) and in some cases are known to localize to microvilli (). Interestingly, previous studies have established that all of these tissues release vesicles into their lumens (). Thus, further studies are required to determine whether the activity described in this paper represents a general function for class I myosins expressed in other cell types, or a phenomenon specific to the gastrointestinal tract. In recent studies with the Myo1a KO mouse, we observed herniations of the apical domain, where large regions of BB membrane were detached from underlying microvillar actin bundles (). We originally proposed that these herniations arise due to a lack of membrane/cytoskeleton adhesion normally provided by Myo1a. However, our new findings suggest an alternative explanation: herniations may represent excess apical membrane in the BB. In KO enterocytes, the apical sorting machinery continues to deliver apical domain components to the BB. However, in the absence of Myo1a and its associated plus end–directed mechanical activity, membrane release from microvillar tips may be slowed, resulting in the accumulation of membrane in the BB, and ultimately the formation of membrane herniations. Early ultrastructural studies of the enterocyte BB revealed the presence of a prominent electron-dense plaque at the distal tips of microvilli (). To date, the role of the tip complex remains unclear, but by analogy to similar structures in stereocilia and filopodia, we suspect that at least one function may be in controlling the dimensions and dynamics of the core actin bundle (). Does the tip complex also play a role in the process of membrane vesiculation from microvillar tips as described in this paper? It remains possible that an additional role for this complex may be in the formation of vesicles, perhaps through promoting and/or stabilizing the high curvature of membrane that envelops the microvillus tip. This would be analogous to the activity of viral proteins such as VSV-M, which are known to be involved in curving membrane through direct interactions with lipids (). Alternatively, this complex may contain proteins that actively promote fission and vesicle release. In either case, the tip complex must be dynamically reassembled after vesicle release or somehow left behind during the process. The former idea finds support in recent proteomic studies (), which show that one of the few proteins known to localize to microvillar tips, Eps8 (), is found in vesicles released from the apical surface of cultured intestinal epithelial cells. Early experiments with isolated BBs established that ATP induced a contraction of the junctional band of actin filaments surrounding the terminal web region (). A number of groups actively investigated this contractility and the involvement of Myo2, which at the time was the only known force generator in the BB (; ; ; ). The findings presented here demonstrate that terminal web contraction and microvillar membrane translation/shedding are independent and separable activities (). Thus, we propose that the BB contains two distinct actomyosin contractile arrays: a Myo2-based array that powers the contraction of the junctional band surrounding the terminal web region, and a Myo1a-based array in microvilli that exerts plus end–directed forces on the apical membrane (). Myo1a was first visualized in microvilli in the form of lateral bridges () at about the same time the early studies on BB contractility began (around 1976; ). However, the motor properties of Myo1a were not discovered until the late 1980s (). We believe that this lag in the experimental timeline may explain why microvillar membrane translation and shedding were not described before this work. Interestingly, evidence of ATP-induced membrane vesiculation consistent with that reported here can indeed be seen in ultramicrographs from studies on isolated BBs and microvilli performed before Myo1a was recognized as a bona fide motor protein (; ; ). The classic view of microvillar function suggests these structures serve to enhance the efficiency of nutrient uptake by amplifying the apical membrane surface area available for nutrient processing and absorption. The work presented here suggests that microvilli also function as actomyosin contractile arrays, allowing for the plus end–directed movement and shedding of BB membrane from microvillar tips. Because this process may have considerable implications with regard to gastrointestinal physiology, future studies will focus on investigating how Myo1a contributes to BB membrane shedding in vivo. BBs were collected with a modification of the method of . All procedures involving animals were performed under the auspices of Vanderbilt University Medical Center (VUMC) Institutional Animal Care and Use Committee. All chemicals were obtained from Sigma-Aldrich unless otherwise noted. Intestines were dissected from adult animals (Sprague-Dawley rats or wild-type/Myo1a KO mice), flushed with ice-cold saline (150 mM NaCl, 2 mM imidazole-Cl, and 0.02% Na-Azide), and stirred in dissociation solution (DS; 200 mM sucrose, 0.02% Na-Azide, 12 mM EDTA-K, 18.9 mM KHPO, and 78 mM NaHPO, pH 7.2) for 20 min. Released cells were washed with multiple cycles of sedimentation (200 for 10 min; X-15R centrifuge, Beckman Coulter) and resuspension in fresh DS. Cleaned cell pellets were resuspended in homogenization buffer (HB; 10 mM imidazole, 4 mM EDTA-K, 1 mM EGTA-K, 0.02% Na-Azide, 1 mM DTT, and 1 mM Pefabloc-SC, pH 7.2) and homogenized in a Waring blender with 4 × 15-s pulses. BBs were collected from the homogenate by centrifugation at 1,000 for 10 min; BB pellets were then washed in solution A (75 mM KCl, 10 mM imidazole, 1 mM EGTA, 5 mM MgCl, and 0.02% Na-Azide, pH 7.2) and sucrose was added to 50% final concentration. Samples were overlayed with 40% sucrose and centrifuged at 130,000 for 1 h at 4°C in an ultracentrifuge (L8-70M; Beckman Coulter). BBs were collected from the 40/50% interface, resuspended in solution B (150 mM KCl, 20 mM imidazole, 2 mM EGTA, 5 mM MgCl, 0.02% Na-Azide, 1 mM DTT, and 1 mM Pefabloc-SC, pH 7.2), and stored on ice. Protein concentrations were determined using the Coomassie Blue Assay (Pierce Chemical Co.). Isolated BBs were incubated in solution B in the presence or absence of 2 mM ATP for 5 min at room temperature, then fixed for 15 min at room temperature with 4% paraformaldehyde in PBS (137 mM NaCl, 7 mM NaHPO, and 3 mM NaHPO, pH 7.2). BBs were then washed with fresh PBS, stained with Alexa488-phalloidin (1:200) and either TRITC-ConA (1:200; Molecular Probes) or AM1-43 (1:100; Biotium Inc.) overnight on ice. BBs were washed three times in PBS and mounted on slides. Confocal micrographs were acquired on a laser scanning confocal microscope (FV-1000; Olympus) (100×/1.3 N.A. Plan Apo objective). All images were contrast enhanced, pseudo-colored, and cropped using ImageJ software (v. 1.36b; National Institutes of Health, Bethesda, MD); figures were assembled using PowerPoint X (Microsoft). Pearson correlation values were calculated from BB confocal images as follows. In brief, red and green color channels were separated and converted to 8-bit grayscale images. Each image was normalized with the Enhance Contrast function, such that 0.5% of all pixels achieved saturation. Correlation coefficients were then extracted using the Colocalization Finder plug-in (C. Laummonerie and J. Mutterer, Institute of Plant Molecular Biology, Strasbourg, France). For time-lapse microscopy, BBs in solution B were added to flow cells constructed of #1 cover glass (Corning) separated by two parallel strips of double-sided Scotch tape and allowed to adsorb the glass surface for 2 min at room temperature. Loosely bound and unbound BBs were washed out of the flow cell with several volumes of fresh solution B. Flow cells were imaged using an inverted microscope (TE2000; Nikon) equipped with either DIC optics (100×/1.3 N.A. Plan Fluor objective, CoolSnap HQ CCD camera [Roper Scientific], ImagePro Express software [Media Cybernetics]), or a spinning disk confocal head (QLC100; Visitech) (100×/1.4 N.A. Plan Apo objective, Cascade 512B CCD camera [Roper Scientific], IPLab software [BD Biosciences]). Images used to calculate velocity histogram data were acquired on a Leica TCS SP5 laser-scanning confocal (63×/1.4 N.A. Plan Apo objective). To activate BBs, solution B supplemented with 2 mM ATP was pipetted into the flow cell during image acquisition. Montages of time-lapse data were created using ImageJ, and kymographs were generated using the ImageJ Multiple Kymograph plug-in (J. Rietdorf and A. Seitz, European Molecular Biology Laboratory, Heidelberg, Germany). BBs (200 μg total protein) were suspended in solution B and incubated with or without 2 mM ATP for 10 min at room temperature. Samples were centrifuged at 5,000 for 10 min at 4°C to separate shed membrane (supernatant) from intact BBs and BB fragments (pellet). To isolate the shed membrane vesicles, the 5,000- supernatant was spun at 100,000 for 2 h at 4°C in an ultracentrifuge (TL-100; Beckman Coulter) using a TLA 120.2 rotor. 5,000- and 100,000- pellets were resuspended in solution B to volumes equivalent to the original reaction. The resulting fractions were prepared for immunoblot analysis and electron microscopy as described below. All EM reagents were purchased from Electron Microscopy Sciences. For the ultrastructural examination of ATP-treated BBs, BBs were incubated in solution B in the presence or absence of 2 mM ATP for 5 min at room temperature. BBs were fixed in 0.1 M Na-phosphate buffer (pH 7.0) containing 2% glutaraldehyde and 2 mg/ml Tannic acid on ice for 1 h. BB were washed with 0.1 M Na-phosphate buffer (pH 7.0) and post-fixed with 1% OsO in 0.1 M Na-phosphate buffer (pH 6.0) for 30 min on ice. BBs were then washed in cold water and stained overnight in 1% uranyl acetate. Samples were then dehydrated with a graded ethanol series followed by 100% propylene dioxide. BBs were infiltrated with a 1:1 epon/propylene dioxide mix for 6 h, and then placed in fresh EPON for 2 h and baked at 60°C overnight. Ultrathin sections were cut on an ultra-microtome (Leica). For negative stain, BB fractions were deposited on Formvar coated grids and stained with 1% uranyl acetate. All grids were imaged on a transmission electron microscope (CM12; Phillips) equipped with 8-bit grayscale digital image capture capabilities (1024 × 1184 pixels). Protein fractions were separated using NuPAGE Bis-Tris 4–12% gradient gels (Invitrogen) run in Morpholineethanesulfonic acid buffer. Proteins were transferred to nitrocellulose membranes at 30V overnight at 4°C. Stock solutions of primary antibodies were used as follows: anti-Myo1a (CX-1 ascites fluid; a gift from Mark Mooseker, Yale University, New Haven, CT; ), anti–rat SI (a gift from Andreas Quaroni, Cornell University, Ithaca, NY), and anti-CaM (Upstate Biotechnologies) were all diluted 1:1,000; anti-AP (Sigma-Aldrich) was diluted 1:2,500. Secondary antibodies (Promega) were diluted 1:5,000. Immunogens were visualized using ECL reagents according to manufacturer's protocol (GE Healthcare). BBs were stained with AM1-43 (1:100; Biotium Inc.) and unlabeled phalloidin (1:100; Invitrogen) overnight on ice, and then washed with fresh solution B to remove excess dye. Shedding assay reactions were performed in triplicate; BBs were resuspended in solution B at 0.02 mg/ml and then stimulated with the addition of ATP (or other ligands as indicated) to 2 mM. Reactions were incubated for 2 min at room temperature and then subject to centrifugation at 5000 for 5 min at 4°C. The resulting 5,000- supernatant was transferred to a flat-bottomed, 96-well plate (Corning) and fluorescence was measured using a microplate reader (Synergy HT; Bio-Tek) using 485/15-nm excitation and 590/20-nm emission detection. Detector gain was set so that the brightest well was near the upper limit of the detector range. Control BBs (not exposed to ATP) were processed in parallel and provided an estimate of the reaction background. For each condition, the triplicate fluorescence values were averaged and background subtracted to derive the ATP-dependent response. Nucleotide analogues ATPγS and AMP-PNP were purchased from Roche Applied Science or Sigma-Aldrich. A test was used to analyze data in . A Mann-Whitney test for unpaired data was used to analyze data in ; P < 0.05 was considered significant. Video 1 shows time-lapse DIC microscopy of an isolated BB shedding membrane in response to ATP. Video 2 shows time-lapse spinning disk confocal microscopy of two isolated BBs shedding membrane in response to ATP. Online supplemental material is available at .

Cell migration is a key step in many physiological and pathological processes, such as wound repair, embryonic development, tissue regeneration, angiogenesis, and metastasis (; ). In an attempt to understand this complex process, migration has been viewed as a multiple-step cycle, where migrating cells become highly polarized and display sequential morphological changes (; ). These include extension of protrusions at the cell front, formation of stable adhesions at the leading edge, reorientation of the Golgi and the microtubule organizing center (MTOC) toward the leading edge, translocation of the cell body in the direction of the movement, and focal adhesion (FA) release and retraction at the cell rear (). These steps are easily observed in slow-moving cells, such as fibroblasts and endothelial cells (). The acquisition of a highly polarized phenotype is not generally regarded as a step in the migration process but, rather, as a concomitant and essential event to cell migration (; ). Migration is a very complex process that requires the spatial and temporal integration of different signaling components (for reviews see ; ), many of which are not yet understood. Among them, the Rho family of small GTPases is one of the master regulators of cell motility, as they control both actin cytoskeleton remodeling as well as FA formation and turnover. FAs are one of the key elements in migration (; ; ; ; ). They are formed by the recruitment of cytoskeletal and signaling proteins to the sites where integrins attach to the ECM. FAs serve as anchorage points for stress fibers (; ). Highly regulated, polarized actin polymerization is involved in the protrusion formation at the leading edge of the cell, whereas detachment and retraction of the cell rear involves myosin-dependent contraction of stress fibers in the tail (; ). Both FAs and stress fibers are regulated by Rho GTPases. By coordinating the activation of several effectors and, ultimately, actin polymerization, Cdc42 triggers filopodia and Rac regulates lamellipodia protrusion and membrane ruffles, whereas Rho regulates formation of stress fibers and cell contractility in the cell body, and hence Rho activity is high at the cell contractile tail (; ). Importantly, it has recently been reported that Rho activity is also high at a sharp band immediately adjacent to the leading edge of cells migrating out of a wounded monolayer (; ). On the other hand, Rac and Cdc42 regulate the initial recruitment of cytoskeletal and signaling proteins into small focal complexes, whereas Rho controls the maturation of these complexes into bigger FAs, as well as FA lifetime (; ; ; ). Many other players are involved in the complex regulation of cell migration (; ), among them, caveolae and caveolin (), although in a controversial manner. Caveolae are specialized plasma membrane microdomains with a flask-shaped, invaginated morphology, highly enriched in cholesterol and sphingolipids (). Caveolin is their principal structural component. There are three proteins encoded by the caveolin gene family (caveolin-1, -2, and -3). Caveolin-1 and -2 are coexpressed in numerous cell types, whereas caveolin-3 is muscle specific. It has been proposed that caveolin-1 could play an important role in cell motility by controlling polarization of signaling molecules (; ). Supporting this notion, caveolin-1 is linked to the actin cytoskeleton through filamin () and has been proposed to associate with a certain subset of integrins (; ). On the other hand, caveolin-1 and caveolae present a polarized distribution in migrating endothelial cells (; ). Moreover, several reports have shown that migration is affected by changes in caveolin-1 expression levels, although in a controversial manner. Some data suggest that caveolin-1 promotes cell migration. Thus, knock down of caveolin-1 expression correlated with a decrease in the chemotaxis of endothelial cells, astrocytes, and multiple myeloma cells (; ; ). However, other studies indicate that caveolin-1 could be a negative regulator. For instance, restoration of caveolin-1 expression in MTLn3 cells reduces lamellipodia formation and chemotactic migration (). Likewise, caveolin-1 knockdown increased directed migration toward sphingosine-1 phosphate in bovine aortic endothelial cells (). Although some of these discrepancies could be ascribed to technical or cell type specificity issues, it seems important to ascertain what kind of role, if any, caveolin-1 plays in the coordinated processes of polarization and migration. We decided to explore this issue by using fibroblasts from caveolin-1–deficient mice. These cells do not express any caveolins, as caveolin-2 is degraded in the absence of caveolin-1 through the proteasomal pathway (), and caveolin-3 is muscle specific. Our results demonstrate that caveolin-1 plays an essential role in the acquisition of a polarized phenotype and, accordingly, in directional cell migration, including both intrinsic persistence of migration and chemotaxis. It does so by regulating the activation of Src, which in turn regulates signaling by Rho GTPases. We compared the morphological phenotype of wild-type (WT) and caveolin-1–deficient mouse embryonic fibroblasts (MEFs; ) spread on fibronectin (Fn) by immunofluorescence confocal analysis. As shown in , Cav1 MEFs displayed a remarkable morphological change. Most of the WT MEFs exhibited a polarized morphology, with an elongated, polygonal shape (see , left). However, Cav1 MEFs adopted a nonpolarized, rounded shape and displayed an aberrant architecture of the actin cytoskeleton. Stress fibers presented a peripheral, concentric localization, creating cortical actin ring structures, whereas in WT MEFs, stress fibers were organized in bundles aligned along the long axis of the cell. We also detected an alteration in the formation of vinculin-stained FAs. These structures were smaller, much like focal complexes (), and more abundant compared with normal fibroblasts. In addition, adhesions were distributed over the entire ventral surface, whereas in WT MEFs, FAs were mostly located at cell edges. Interestingly, Cav1 MEFs presented abundant homogeneously distributed vinculin, i.e., not associated with adhesive complexes (, top middle), consistent with the altered, immature focal complexes phenotype observed in these cells. Paxillin staining was very similar to that of vinculin (see and Fig. S4, available at ). To quantify morphological changes detected in Cav1 MEFs, we calculated the elliptical factor (EF), which is defined as the ratio between the longest and the shortest axis in the cell. In WT MEFs, EF was significantly higher than that of Cav1 MEFs (). Importantly, a much higher WT population displayed an EF >2 (indicative of a polarized morphology) compared with the Cav1 population (). Similar results were obtained with fibroblasts prepared from a Cav1 conditional knockout mouse () versus their WT littermates (see , bottom; quantified in Fig. S3, b and c) . Fibroblasts prepared from these two strains of Cav1 mice (; ) have been used throughout this study. Cell area measurements showed that Cav1 cells do not exhibit a delay in cell spreading, and the morphological changes described were still present at late time points after spreading (Fig. S1). Reexpression of caveolin-1 in Cav1 MEFs () restored the normal EF (; and see ), thus showing that caveolin-1 is required for elongation. Altogether, these results suggest that caveolin-1 is required for the acquisition of an elongated morphology, stress fiber architecture, and FA formation in fibroblasts. Polarization usually correlates with, but is not equivalent to, elongation. To directly measure cell polarity stimulated by a directional stimulus, we assayed for MTOC polarization in fibroblasts migrating out of a wounded monolayer. MTOC reorientation toward the leading edge was substantially reduced in Cav1 MEFs, and Cav1 reexpression restored polarity to levels achieved in WT cells (; quantified in ). Collectively, these results demonstrate a defect in the establishment of cell polarity in the caveolin-1–deficient fibroblasts and establish a correlation between elongation defects and polarization changes in these cells. Impaired polarization and defects in the stress fiber and FA pattern in Cav1 fibroblasts suggest alterations in the activity of the GTPases of the Rho family, as they are the principal regulators of these processes (; ; ; ). To test this hypothesis, we performed pull-down assays to determine the level of activation of Rho, Rac, and Cdc42 in WT and Cav1 MEFs. We found that caveolin-deficient MEFs showed a notable decrease in basal Rho activity and a significant increase in Rac and Cdc42 activity (). To confirm these biochemical observations, we performed several assays designed to test biological consequences downstream of GTPases, namely, protrusion formation and lifetime of nascent adhesions at protruding areas. Time-lapse video recording of paxillin-GFP–expressing MEFs revealed that the lifetime of newly forming adhesions at protruding regions was twofold shorter in Cav1 MEFs compared with WT cells ( and Videos 1 and 2, available at ) and was rescued by caveolin-1 expression (). Increased adhesion turnover speed at cell protrusions in Cav1 MEFs is consistent with reduced Rho activity, which could be essential for the maturation of small focal complexes into FAs. In this regard, increased Rac and Cdc42 activity fits with the higher number of immature focal complexes observed in the Cav1 MEFs (, magnified areas). To test whether hyperactivation of Rac and Cdc42 has consequences in membrane protrusion, we performed time-lapse analysis of the protrusive–retractile activity at cell edges. Both protrusive and retractile activities were significantly higher in Cav1 MEFs ( and Videos 3 and 4). Careful observation of these videos revealed that protrusive activity in WT MEFs is highly directional, i.e., it only occurs in the direction of movement, whereas retraction occurs preferentially at the rear end. However, in Cav1 MEFs, protrusion and retraction is higher and occurs concomitantly throughout the cell perimeter ( and Videos 3 and 4). Quantification of the protrusive and retractile area showed a significant increase in Cav1 MEFs compared with WT MEFs (). Altogether, these results indicated that changes in the activity of Rho GTPases could be responsible for the morphological defects observed in Cav1 fibroblasts. To test this hypothesis, we reconstituted the normal GTPase pattern of activity by expressing a constitutively active mutant of Rho (GFP-Rho G14V) or dominant-negative mutants of Rac (GFP-Rac T17N) or Cdc42 (GFP-Cdc42 T17N) in Cav1 fibroblasts. Expression of each individual construct was able to restore the WT elongated phenotype (Fig. S2, available at ). This result most likely reflects the fact that alteration of the proper reciprocal balance between Rho GTPases could account for the polarization defects of Cav1 cells. A role for caveolin-1 in cell motility has been previously suggested, although contradictory effects have been reported. In an attempt to elucidate this issue, we performed several migration assays with Cav1 MEFs. As a first approach, we measured the velocity of random migration of these cells on a Fn matrix, using time-lapse microscopy. We found that Cav1 MEFs moved subtly faster compared with WT MEFs (). However, important differences were observed in the pattern of migration of Cav1 versus WT MEFs (Videos 5 and 6, available at ). WT cells showed an intrinsic directionality or persistency of migration, i.e., they tend to migrate in the same direction for a sustained period of time without turning even in the absence of a chemotactic gradient, whereas Cav1 MEFs completely lost directionality. This observation was confirmed by measuring the trajectory of each individual cell during a 10-h migration period by tracking its centroid from the time-lapse video. To clearly visualize the differences, cell movement paths were reproduced on composite panels (). Interestingly, we found that Cav1 MEFs displayed much shorter net translocation (the shortest linear distance from the starting point to the end point of the time-lapse recording) than the WT cells, which showed longer paths and migrated on a straighter way. To quantify these differences, we measured the directional persistence of the cells as estimated by the index of directionality (ID; i.e., the ratio of the net distance divided by the total distance traveled by the cell). The Cav1 MEFs showed a significant reduction in the ID () that was restored by caveolin-1 reexpression (; and Video 7). These results show that caveolin-1 contributes to persistent migration, i.e., the cell's internal sense of directionality. To explore directional migration stimulated by external stimuli, we performed wound-healing and chemotaxis assays. Both wound closure (, c and d; and Videos 8 and 9) and the chemotactic response in a transwell assay () were hampered in caveolin-1–deficient MEFs, and reconstituting caveolin-1 expression rescued both responses (Video 10). Collectively, these data demonstrate that caveolin is required in fibroblasts for persistency of migration in the absence of an external chemotactic cue and for directional migration in the presence of an external stimulus, whereas it slightly slows down the velocity of random migration. Caveolin-1 is a substrate for nonreceptor tyrosine kinases, including Src. In fact, caveolin-1 was first described as the major substrate for Src in v-src transformed cell lines (). Furthermore, it has been described that caveolin phosphorylated on Tyr 14 (pY14-Cav1) can inhibit Src through the recruitment of C-terminal Src kinase (Csk; ; ). Thus, we hypothesized that in the absence of caveolin-1, Src activity could be affected. We evaluated the Tyr 418 phosphorylation state of Src in Cav1 MEFs and found an increase in the basal activation of Src (; quantified in ), accompanied by a reduction in the abundance of this protein (), consistent with the reported increased degradation of active versus inactive Src (). Increased Src activity in Cav1 MEFs suggested that Src could be involved in the morphological changes observed in the absence of caveolin-1. To test this hypothesis, we blocked Src activation by treating the Cav1 MEFs with two inhibitors of Src family kinases, PP2 and SU6656. Interestingly enough, Src inhibition restored the morphological phenotype of the WT MEFs (; and Fig. S4). Because Src regulates caveolin-1 through phosphorylation of Tyr 14, one requirement of our hypothesis is that expression of a nonphosphorylable mutant of caveolin-1 in Cav1 MEFs should not restore the normal phenotype. Accordingly, cells stably transfected with the Y14F caveolin-1 mutant () did not restore the cell morphological changes (), MTOC polarization (), wound closure (), directionally persistent migration (), or the chemotactic response (), which were restored by WT Cav1 in all cases. These results highlight the importance of caveolin-1 Tyr 14 phosphorylation by Src in the polarization and directed motility of mouse fibroblasts. Several links have been reported between the activity of Src kinase and Rho GTPases. Thus, Src is involved in activation of Rac (; ) and Cdc42 (; ; ) and can inhibit Rho through activation of p190RhoGAP (; ; ). Therefore, increased Src activity could explain the reduction in Rho GTP levels observed in Cav1 fibroblasts through activation of p190RhoGAP. Both total protein and tyrosine phosphorylated p190RhoGAP levels are mildly elevated in Cav1 cells (Fig. S5, available at ), supporting this notion. To test this hypothesis, we expressed a dominant-negative mutant of p190RhoGAP in Cav1 fibroblasts and measured the EF. Importantly, GFP-p190 R1283A–expressing Cav1 fibroblasts showed an elongated shape (Fig. S3). Moreover, knocking down p190RhoGAP expression using RNA interference () led to a recovery of the morphological changes (; and Fig. S4) and persistent migration defects (; quantified in ). Both Src inhibition and RhoGAP knockdown individually were able to rescue the various phenotypes to the levels achieved in WT cells, and they did not display an additive effect when combining both treatments (; and Fig. S4). These results suggest that a linear pathway of Src and p190RhoGAP activities drives Rho down-regulation in cells lacking caveolin-1. In this regard, expression of WT caveolin-1, but not the Y14F caveolin-1 mutant, restored Rho GTP loading in Cav1 MEFs (). These findings suggest an important role of caveolin-1 and Src in the establishment of cell polarity by regulating the activity of Rho. Our results support a model in which, in the absence of caveolin-1 expression, Src activity may become insensitive to Csk modulation; increased Src could inactivate Rho in a p190RhoGAP-dependent manner, resulting in profound defects in cell polarization and directed migration. To explore whether the differences in the migratory behavior observed ex vivo have any correspondence in vivo, we performed wound-healing experiments in the skin of 11-wk-old WT and Cav1 mice. Wound healing is a complex process involving growth factors, ECM, and epidermal and mesenchymal cells undergoing directional migration, proliferation, and differentiation events (). Two 3.5-mm-diameter circular punch biopsies were performed on the back skin of five WT and five Cav1 mice. The rate of wound healing was monitored as the percentage of the initial wound area left open with time after the punch was made. Cav1 mice showed a significantly slower wound healing rate compared with WT littermates (). As examples, shows representative images of WT and Cav1 wounds 1 and 7 d after the wounds were made. These results indicate that the skin of Cav1 mice has an impairment in wound healing. Provided the hyperproliferative phenotype of Cav1 mice (; ), these in vivo findings support our ex vivo observations that caveolin-1 is required for directional migration. Caveolin-1 is one of the numerous intracellular signaling molecules that have been implicated in cell migration. However, there is no consensus on its role in controlling cell motility, as both positive (; ; ) and negative (; ) regulatory effects have been reported. Technical or cell type specificity issues most likely account for this discrepancy. However, provided the relevance of both caveolin-1 and cell migration in both physiology and many pathological conditions, it seems important to further investigate this controversy. In this study, we analyze fibroblasts prepared from two types of Cav1 mouse (; ) in a variety of polarization and migration assays, including random and directional migration approaches. We have obtained evidence for the requirement of caveolin-1 for the establishment of a polarized and elongated morphology, internal persistency of migration, and externally stimulated directional migration. We show that caveolin-1 regulates polarity and directional migration by affecting the activation patterns of Src and Rho GTPases. Our results suggest a model in which the modulation of Src activation by the caveolin-1–Csk module (; ) is crucial for the establishment of cell polarity and directional migration. In caveolin-1–deficient cells, Src activity is constitutively high, leading to a p190RhoGAP- dependent decrease in the levels of GTP-loaded Rho and subsequent defects in the arrangement of the actin cytoskeleton. The first and evident feature we observed in the Cav1 fibroblasts is a profound change in their morphology and in the actin cytoskeleton architecture. Most of them were rounded and showed stress fibers localized in the cell periphery, creating concentric cortical actin rings. This phenotype suggested alterations in the activity of the GTPases of the Rho family, as they are the principal regulators of polarity and cytoskeletal rearrangements (; ; ; ). We found that Cav1 fibroblasts showed a dramatic decrease in basal Rho activity and a significant increase in Cdc42 and Rac activity, the latter consistent with the reported data for Rac activity upon caveolin-1 silencing (). On the other hand, Cav1 fibroblasts presented shorter lifetime of nascent adhesions at protrusive areas, compared with the control cells. This is consistent with the low levels of Rho GTP loading, as adhesion turnover inversely correlates with Rho activity (). In addition, these adhesive complexes were smaller and more abundant in Cav1 than in WT fibroblasts. This fact indicates a possible blockage in the maturation of these structures, which is also dependent on Rho (; ). The increase in Rac and Cdc42 GTP levels also supports this phenotype, as both GTPases are involved in the creation of new substrate contacts and recruitment of cytoskeletal and signaling proteins into nascent small focal complexes (; ). Likewise, we found an enhanced, nondirectional protrusive activity in the whole periphery of caveolin-1–deficient cells, in agreement with the altered pattern of activity of Rho GTPases. In fact, both Rho inactivation () and Rac activation promote membrane protrusion (; ), and there is evidence of a mutual antagonism between both GTPases (; ; ; ; ). Interestingly, the use of a fluorescence resonance energy transfer biosensor to measure local activation of Rho has recently established that Rho is active in a sharp band immediately adjacent to the leading edge of migrating cells, both randomly and in a wounded monolayer (; ). This suggests that Rho is also required for directional migration, and not only for cell contraction, as it was previously established. The dramatic decrease of Rho activity in caveolin-1–deficient cells together with the lack of directionality in these cells is consistent with these reports. The fact that each individual Rho GTPase mutant (active Rho or inactive Rac/Cdc42) restore the normal polarized phenotype is intriguing. The pattern of activity of all three Rho GTPases is altered in caveolin-deficient cells, which complicates the interpretation of this result. It could be explained by the mutual inhibition between the frontness (Rac/Cdc42 dependent) and backness (Rho dependent) pathways (; ; ). Therefore, this result does not necessarily imply that alterations in the GTP loading of each GTPase play a direct role in the phenotype observed in Cav1 fibroblasts, as this could be an indirect effect, through a negative feedback on the opposing pole of the cell. For Rac, apart from increased GTP loading, enhanced plasma membrane targeting in caveolin-1–deficient cells () could be partially responsible for the phenotype observed. In any case, alteration of the normal reciprocal balance between Rho GTPases most likely contributes to the aberrant cellular morphology and migratory behavior of Cav1 cells. Consistent with the morphological phenotype, Cav1 fibroblasts show a remarkable defect in directional migration, both externally stimulated and internally persistent. Persistence is the intrinsic propensity of the cells to continue migrating in the same direction in the absence of exogenous stimuli (). Recently, showed that high Rac activity renders a random migration pattern, whereas decreasing Rac activity switches cell migration from random to directionally persistent. Consistently, we found a mild but significant increase in the velocity of random migration and a dramatic switch toward random migration in Cav1 fibroblasts, concomitant with increased Rac GTP levels. Thus, increased Rac activity and/or enhanced Rac membrane targeting in caveolin-1–deficient cells () would render nonpolarized increased protrusive activity throughout the whole cell perimeter and, consequently, loss of directionally persistent cell migration. Previous reports where caveolin-1 expression inhibited cell migration (; ; ) fit with our result that Cav1 cells migrate slightly faster in the absence of an external chemotactic cue. Therefore, previous contradictory results could be reconciled by a detailed analysis of random versus directional migration in caveolin-1–deficient cells. In other cases, divergent results can be ascribed to technical or cell-specific issues. In this regard, a recent report showed that caveolin-1–deficient aortic smooth muscle cells are slightly more migratory than the control cells (). However, these cells express caveolin-3, which could compensate the absence of caveolin-1. Moreover, these cells were isolated from Ink4a mice, which show increased proliferation (), which could influence the observed results. The phenotypic deficiencies reported in Cav1 fibroblasts are dependent on caveolin-1, as restoring its expression rescues the WT phenotype. Importantly, this recovery is dependent on caveolin-1 Tyr 14, which is the substrate for Src phosphorylation (). This phosphorylation is important for fine-tuning of Src activation, as pY14-Cav1 binds to and activates Csk, which phosphorylates inhibitory Tyr 527, resulting in Src inactivation, closing a negative feedback loop (; ). These results predict that in the absence of pY14-Cav1, Src should be insensitive to modulation by Csk. Accordingly, we find increased Src activation in Cav1 fibroblasts, and when these cells were reconstituted with a Src-insensitive Y14F caveolin-1 mutant, the normal phenotype was not restored. Increased Src activation was accompanied by a reduction in the protein levels, consistent with previous reports that Src activation leads to increased polyubiquitination and proteasome-dependent degradation (). In spite of protein reduction levels, increased Src activation is functionally relevant because pharmacological inhibition of Src in Cav1 fibroblasts restored the normal phenotype. Mechanistically, we have found that alterations in the signaling of both Src kinase and Rho GTPases are responsible for the aberrant morphology and migratory pattern of Cav1 cells. Our results and previous reports allow us to envisage a model in which caveolin-1 influences signaling of Rho GTPases through activated Src. Src regulates the activity of the three major Rho GTPases studied here. It can activate Cdc42 through many pathways, including the exchange factors FRG (), Vav2 (), and C3G (in collaboration with Crk, Rap1, and FRG; ). It activates Rac through the exchange factors Dock180 (in collaboration with p130Cas and Crk; ), Tiam1, Vav2 (), and FRG (through Cdc42 and Vav2; ). However, Src inhibits Rho activity through activation of p190RhoGAP (; ; ). We present evidence that p190RhoGAP is involved in caveolin-mediated alterations of the cell polarity, allowing us to propose this model. The ex vivo migratory defects reported here are most likely related to the in vivo impaired angiogenic response displayed by Cav1 mice (; ) and could account for it. Angiogenesis is a complex phenomenon involving both migration and proliferation of endothelial cells (). In spite of increased cellular proliferation, Matrigel plugs implanted in Cav1 mice showed reduced blood vessel formation (), suggesting that decreased migration is most likely responsible for the angiogenic defects. In good agreement, knocking down caveolin-1 expression diminishes the chemotactic response in endothelial cells in vitro (). We also report here an impaired in vivo wound healing in Cav1 mice. Wound healing is also a complex process that depends on directional migration, proliferation, and differentiation of epidermal and mesenchymal cells (). Because enhanced cell proliferation of Cav1 mice (; ) would favor the wound-healing process, our results support the notion that decreased directional migration accounts for decreased wound healing observed in vivo. However, caveolin-1 appears to be dispensable for normal embryonic development, as Cav1 mice are viable (; ). Although embryonic development of these animals has not been studied in detail, most likely other signaling pathways involved in the complex migratory process (; ) are compensating for caveolin-1 absence. Caveolin-deficient animals display a wide range of phenotypes (), and it seems feasible that some of them might be related to the migratory deficiencies reported here. Elucidating such phenotypes is an interesting goal for future work. Caveolin-1–deficient mice strain Cav-1/J and their WT littermates were obtained from The Jackson Laboratory. Mice were housed and maintained in a barrier facility at our institute (Centro Nacional de Investigaciones Cardiovasculares, Madrid, Spain), which approved the animal protocols. Pathogen-free procedures are used in all mouse rooms. Quarterly health-monitoring reports have been negative for all pathogens in accordance with Federation of European Laboratory Animal Science Associations recommendations. Mice were kept on a 12:12-h light–dark cycle, with ad libitum access to food and water. MEFs () and thymus fibroblasts () from Cav1 and Cav1 littermate mice have been used throughout this study. M.P. Lisanti (Thomas Jefferson University, Philadelphia, PA) and R.G.W. Anderson (University of Texas Southwestern Medical Center, Dallas, TX) provided MEFs and thymus fibroblasts from Cav and Cav littermate mice, respectively. All experiments were performed with both cell types, except from the MTOC polarization assay, Rho GTPase pull downs, and FA lifetime estimations, which were performed only with the MEFs. Cells were maintained in culture in DME supplemented with 10% FBS, 100 U/ml penicillin, and 100 μg/ml streptomycin. Plasmids encoding pEGFP-p190RhoGAP R1283A (); pEGFP-paxillin (); and GFP-tagged Rac T17N, Cdc42 T17N, and Rho G14V () were previously described. K. Burridge (University of North Carolina at Chapel Hill, Chapel Hill, NC) provided GFP-p190 R1283A. C-terminal Flag-tagged mouse caveolin-1 was cut with BamH1–EcoRI, blunt-ended, and ligated to blunt-ended MIGR1 EcoRI, a bicistronic, GFP-expressing retroviral vector (). C-terminal Flag tag caveolin-1 Y14F was cut with BglII–BamH1 and ligated to BglII site in MIGR1. NIH 3T3 or 293T/17 cells were transfected (using Lipofectamine 2000 or calcium phosphate method) with MIGR1, MIGR1–caveolin-1 or MIGR1–caveolin-1Y14F and packaging plasmid pSVψ2. 48 h later, supernatants were filtered and added to Cav1 MEFs in DME plus 10% FBS containing 4 μg/ml polybrene. 48 h later, GFP-positive cells were sorted using a cell sorter (DakoCytomation). Levels of caveolin-1 and caveolin-1 Y14F were similar to endogenous caveolin-1 in MEFs (). Mouse p190RhoGAP (available from GenBank/EMBL/DDBJ under accession no. ) targeting sequence (nucleotides 2935–2953, 5′-gttatggacgcaacattaa-3′) and control, nontargeting sequence (5′-gcgcgctttgtaggattcg-3′) were cloned into short hairpin RNA (shRNA) vector pSuper.Retro.Neo+GFP (Oligoengine) to generate pSuperRetroGFP-p190-2935 and pSuperRetroGFP-Control vectors. Retroviral supernatants were generated by transfecting 293T/17 cells with each shRNA and pSVψ2 vector using Fugene 6 (Roche) transfection reagent. Cav-1 MEFs were infected with retroviral supernatants as previously indicated, and high GFP-expressing cells were sorted (∼15% of the cell population). mAb against p190RhoGAP was purchased from BD Biosciences. pAb against Cdc42 and mAb against Rho were purchased from Santa Cruz Biotechnology, Inc. Anti-Rac, anti-Src and anti-phosphotyrosine (4G10) mAbs were obtained from Upstate Biotechnology, mAb for vinculin was obtained from Sigma-Aldrich, and anti-Src(pY418) phosphospecific pAb was purchased from Biosource International. Anti-paxillin mAb was purchased from Invitrogen. Rhodamine phalloidin and wheat germ agglutinin labeled with tetramethyl rhodamine were purchased from Invitrogen. Alexa 594 and FITC-conjugated antibody and peroxidase-conjugated goat anti–rabbit and anti–mouse IgG were obtained from Jackson ImmunoResearch Laboratories. Src kinase inhibitors SU6656 and PP2, as well as PP3 control, were obtained from Calbiochem. Fn was purified from human plasma as described previously (). Cells were attached to glass coverslips precoated with 5 μg/ml Fn for different times. Cells were fixed with 2% formaldehyde-PBS for 20 min, permeabilized in 0.2% Triton X-100 in PBS for 10 min, and blocked with 10% normal goat serum before staining. Anti-vinculin/paxillin antibodies followed by FITC/Alexa594-conjugated anti-IgG were used to stain FAs. Actin cytoskeleton was stained with rhodamine phalloidin. For polarity determination, membrane of fixed cells was stained with 10 μg/ml wheat germ agglutinin for 30 min at RT. Images were acquired using a confocal microscope (Radiance 2100; Bio-Rad Laboratories, Inc.). Then, cells were outlined using the Kirsch edge detection algorithm, which is included in MetaMorph software (Universal Imaging Corp.). Outlines were checked and corrected by hand if necessary. Using MetaMorph's integrated morphometry analysis function, we determined the EF (length/breadth) of cells as a measure of elongation. Transwell cell culture chambers containing polycarbonate membrane with 8-μm pore size (Corning Costar Corp.) were coated with 5 μg/ml Fn. Cells were starved for 24 h before experiments. After trypsinization and dilution, 15,000 cells in DME plus 0.2% BSA were added to the top Transwell chamber. The bottom compartment was filled with 500 μl DME plus 10% FBS, and the assembly was incubated at 37°C for 4 h, to allow cell migration. In the negative controls, bottom chambers were filled with DME plus 0.2% BSA. After incubation, the membranes were washed with PBS, and cells were fixed in 2% formaldehyde and stained with Hoechst 33342 (Sigma-Aldrich) according to the manufacturer's protocols. Cells that did not migrate were gently removed from the top surface, and cells that had migrated to the bottom of the membrane were counted in five random fields using a fluorescence microscope (Axiovert 200 M SP LSM5; Carl Zeiss MicroImaging, Inc.). Cells were transiently transfected with pEGFP-paxillin 2 d before the experiment and were plated on a chambered coverglass (eight chambers; Lab-Tek) coated with 5 μg/ml Fn. After 45 min, chambers were filled with Optimem (Invitrogen) supplemented with 10% FBS, sealed using vacuum grease and a glass plate, and transferred to a microscope heated to 37°C. Image series of FA dynamics were acquired using a 60× 1.4 NA Plan Apo objective lens on an inverted microscope (Eclipse TE300; Nikon) heated with an airstream incubator (Nevtek) to 37°C. The microscope was equipped with a robotic stage with linear position feedback encoders on the x, y, and z axes (MS-2000; Applied Scientific Instruments) to allow image series to be collected at different stage positions over time. Images were captured using an cooled charge-coupled device camera (Orca II; Hamamatsu). Fluorescent and phase-contrast images were taken in rapid succession at multiple positions, evenly distributed over the chambers to exclude differences due to experimental variation, at 2.5-min intervals for 2.5 h. Phase-contrast image series of cell migratory behavior were collected on a similar microscope system, except using a 20× 0.5 NA Plan objective lens and a 0.5 NA extra-long working distance condenser, and were captured with a 10-bit chilled charge-coupled device camera (Orca 285; Hamamatsu). For image processing and analyses, we used MetaMorph software. Individual FAs were clearly distinguishable in the GFP-paxillin time-lapse image series. The time of formation and disappearance of the nascent adhesions from protruding areas of different cell types were monitored by eye and double-blind method. From the noted times of appearance and disappearance, the lifetimes of nascent adhesions in WT versus Cav1 cells were determined, averaged, and plotted, with the error bars indicating the SEM (Videos 1 and 2). To determine cell trajectories in phase-contrast time-lapse image series (cells filmed at 8-min intervals for 10 h), the centroids of the cell nuclei were followed (Videos 5–7). To automate this and allow the unbiased analysis of many cells in multiple time lapses, a program was written in Matlab (Mathworks; ), which segments images based on pixel intensity and determines the presence of nuclei based on phase density, size, and shape. Nuclei are then linked in consecutive frames using a neural network algorithm, and cells tracked for less than five consecutive frames are automatically discarded. Detection fidelity in our experiments was usually >90%, which was confirmed by eye for each individual time lapse. To prevent erratic conclusions because of false results generated by this automated analysis, a smaller number of random cells from several videos was tracked using the track objects function in MetaMorph, leading to the same results. Fluorescence time-lapse images of GFP-paxillin were used to outline cells by thresholding for pixels with high intensity. Areas covered by cells in consecutive images in the time series were subtracted to determine the percentage change of the total cell area contributing to protrusion or retraction occurring in the 2.5 min elapsed between the two images (Videos 3 and 4). Plotted are means of protrusion over the first 1.5 h of imaging from six different WT and seven different Cav1 cells. Cells were grown to confluence on coverglass chambers. Confluent monolayers were scraped with a 0.1–2 μl pipette tip. Wound closure was monitored by time-lapse video microscopy. Images were taken at 45-min intervals for 48 h (Videos 8–10). For MTOC polarization assay, cells were seeded on coverslips, and wounded monolayers were fixed 2, 4, 6, and 8 h after scraping. MTOC were localized by immunolabeling using anti-pericentrin antibodies. Only the first row of the wound edge was measured. Cells in which MTOC was contained by the quadrant facing the wound were scored positive (, diagram). Two full-thickness punch biopsies extending through the epidermis and dermis (punch diameter 3.5 mm) were performed on the back of five WT and five Cav1 mice (11 wk of age) after depilation. Mice were anesthetized before wound creation. The wound-healing rate was calculated as the percentage of initial wound area with time. After capturing the images with a digital camera (DFC490; Leica), we determined the area of the wound with image analysis software (Leica IM50 Image Manager). Rho, Rac, and Cdc42 activity were determined by pull-down assays as described before (; ). Src activity was assayed by Western blotting. Low-density WT and Cav1 MEFs were lysed in RIPA buffer (10 mM Tris-HCl, pH 7.2, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate, 150 mM NaCl, 5 mM EDTA, and 3 mM EGTA) with phosphatase inhibitors (10 mM NaPO, 1 mM NaVO, 30 mM NaF, and 3 mM β-glycerophosphate) and protease inhibitors (1 μg/ml aprotinin, 1 μg/ml leupeptin, and 10 mM PMSF). Lysates were spun down at 13,000 for 10 min, and 500 μg of total lysate was used to immunoprecipitate p190RhoGAP with a mAb against it for 2 h. Protein G–Sepharose was added for another 2 h, and immunocomplexes were washed five times and run on an SDS-PAGE. Tubulin, phosphorylated, and total p190RhoGAP bands were quantified with ImageGauge 4.0 (FujiFilm). Statistical significance was determined using a test with OriginPro7 software (OriginLab Co.). P values < 0.05 were considered significant. Fig. S1 shows that absence of caveolin-1 does not delay cell spreading. Fig. S2 shows that constitutively active Rho and dominant-negative mutants of Rac and Cdc42 restore normal polarity in Cav1 fibroblasts. Fig. S3 shows that a dominant-negative p190RhoGAP construct rescues polarity in Cav1 fibroblasts. Fig. S4 demonstrates that Src inhibition and p190RhoGAP knockdown in Cav1 MEFs restore the WT morphology pattern. Fig. S5 shows that total and tyrosine phosphorylated p190RhoGAP are elevated in Cav1 MEFs. Video 1 shows adhesion turnover in WT MEFs. Video 2 shows adhesion turnover in Cav1 MEFs. Video 3 shows protrusion– retraction analysis in WT MEF. Video 4 shows protrusion–retraction analysis in Cav1 MEFs. Video 5 displays random migration of WT MEFs. Video 6 demonstrates random migration of Cav1 MEFs. Video 7 shows random migration of Cav1 MEFs reconstituted with Cav1. Video 8 shows in vitro wound-healing assays in WT s. Video 9 shows in vitro wound-healing assays in Cav1 MEFs. Video 10 demonstrates in vitro wound-healing assays in Cav1 MEFs reconstituted with Cav1. Online supplemental material is available at .

Glucose-stimulated insulin release displays a biphasic pattern in both in vitro and in vivo systems (; ). This pattern consists of a rapidly initiated and transient first phase preceding a sustained second phase. The ability of glucose to evoke first-phase release is shared by other stimuli (such as high KCl stimulation), resulting in membrane depolarization followed by increased cytosolic Ca, whereas only fuel secretagogues are able to initiate second-phase insulin release (). Electrophysiological experiments in single β cells have shown that first-phase release reflects Ca-dependent exocytosis of primed granules in a readily releasable pool of granules, whereas second-phase release involves an ATP-dependent release of granules that may be located further from the release site in a reserve pool (; ). These results suggest that the two phases of release subject insulin granules to nonsynonymous regulatory mechanisms. Fundamental components of secretory machinery, such as SNARE, required for the docking and fusion of vesicles in neuronal cells (), are expressed in pancreatic β cells and play an important role in insulin exocytosis (; ; ). Although the function of SNAREs in docking and fusion during exocytosis is already established (; ), the distinct role of SNAREs in the individual phases of insulin release remains unclear. Interestingly, the expression of t-SNARE, syntaxin 1A/HPC-1 (Synt1A; ; ), and its cognate SNARE partners, synaptosome-associated protein of 25 kD (SNAP-25) and vesicle-associated membrane protein 2 (VAMP2), reportedly decreased in islets of the Goto-Kakizaki rat, an animal model for human type 2 diabetes (; ; ), and in type 2 diabetic patients (). Because type 2 diabetes is associated with disturbances in the release pattern manifested as the selective loss of first-phase release (; ; ), SNAREs may have a specialized role in phasic insulin exocytosis. In the present study, we used Synt1A mice and total internal reflection fluorescence (TIRF) imaging to investigate a potential role for Synt1A in first-phase insulin release. Synt1A pancreatic β cells displayed no fusion from previously docked granules in first-phase release, whereas fusion from newcomers, which are responsible for second-phase release, was still preserved. Thus, we propose a new model for biphasic insulin release wherein docking and fusion of insulin granules is Synt1A dependent during the first phase but Synt1A independent during the second phase. We initially analyzed the dynamic interaction between insulin granules and Synt1A in control mouse β cells using dual-color TIRF microscopy (TIRFM). Expression of GFP-tagged insulin allowed insulin granule observation, and Synt1A was detected by a TAT-conjugated Cy3-tagged mAb. Here, we chose not to use a conventional overexpression approach, such as Synt1A tagged with GFP or RFP, because overexpression of syntaxin disturbs the function of endogenous syntaxin molecules (). Therefore, to analyze the interaction between insulin granules and Synt1A clusters during biphasic insulin release, we labeled the endogenous Synt1A clusters with TAT antibody. As previously reported (), TAT-conjugated Cy3-labeled anti-Synt1A antibody was rapidly transduced into living β cells (unpublished data). We ensured that TAT-conjugated Cy3-labeled anti-Synt1A antibody specifically labeled endogenous Synt1A clusters in the plasma membrane. Cells treated with TAT-conjugated Cy3 anti-Synt1A mAb for 50 min were fixed and immunostained with anti-Synt1A pAb. As shown in Fig. S1 (available at ), there was overlapping of Synt1A clusters labeled with TAT-conjugated Cy3 anti-Synt1A mAb (red) and those stained with anti-Synt1A pAb (green). In addition, it should be noted that most endogenous Synt1A was labeled with TAT antibody. Pancreatic β cells that expressed insulin-GFP (Fig. S1, green) and were treated with TAT-conjugated Cy3 anti-Synt1A antibody (red) were stimulated by 22 mM glucose. Dual-color TIRF images were obtained every 300 ms (). Approximately 75% of insulin granule fusion during the first phase (<4 min after stimulation) involved previously docked rather than newcomer granules (). We observed that most fusion events involving previously docked granules occurred at the site of Synt1A clusters (), whereas fusion from newcomers occurred at sites distinct from the Synt1A clusters. There was no significant difference in the number of fusion events between control (see ) and TAT-conjugated Cy3 anti-Synt1A mAb–treated β cells (): the total number of fusion events from previously docked granules in wild-type (WT) versus TAT-treated cells was 18.2 ± 1.8 versus 14.9 ± 3.1 in 0–4 min (P = NS; > 5 cells), suggesting that the introduction of TAT-conjugated Cy3 anti-Synt1A mAb into β cells does not affect insulin exocytosis. These results suggest that first-phase release heavily involves a Synt1A-based SNARE complex, whereas second-phase release is chiefly independent of a Synt1A-based SNARE complex. If Synt1A is essential for docking and fusing insulin granules specifically during the first phase, the deletion of Synt1A may cause reduction in first-phase but not second-phase insulin release. To examine this hypothesis, we used β cells from Synt1A mice () as a context for analyzing docking and fusion of insulin granules by TIRFM. We first investigated Synt1A protein levels in Synt1A versus WT mouse pancreatic islets. Fig. S2 (available at ) shows the lack of Synt1A protein expression in Synt1A islets. Expression of Synt1B was not observed in either Synt1A or WT islets, in accord with the report that Synt1B is expressed at very low levels in control β cells (), although the brain abundantly expresses Synt1B (). We found no difference between WT and Synt1A islets in expression levels of other plasma membrane proteins, such as Synt3, Synt4, the other SNAREs, and related proteins SNAP-25, VAMP2, and Munc18. We then examined the pancreatic islets morphologically (Fig. S3). We found that paraffin-embedded pancreatic tissue sections showed insulin immunofluorescence patterns typical for β cells with no notable difference between the Synt1A and WT islets (Fig. S3, A and B). EM of pancreatic β cells also revealed that cell size, total number of granules per section, and mean granule diameter were similar between WT and Synt1A β cells (Fig. S3, C–F). Thus, Synt1A β cells displayed specific Synt1A protein depletion but were similar to WT cells in these other traits assayed. We examined the docking status of insulin granules in Synt1A β cells using TIRFM with immunostaining for insulin (). Because evanescent field illumination reaches a <100-nm-thick layer immediately adjacent to the cover glass under our TIRF conditions, TIRFM illuminates only the plasma membrane with its associated organelles, such as synaptic vesicles (), secretory granules (), and glucose transporter 4 (GLUT4) vesicles (), where a cell adheres tightly to the cover glass. We interpret the individual fluorescent spots shown in the TIRF image in to be equivalent to morphologically docked granules (see Materials and methods). We rarely observed morphologically docked granules in Synt1A β cells (number of docked granules: 253.3 ± 10.2 vs. 12.3 ± 2.2 granules per 200 μm in WT and Synt1A β cells, respectively; = 12 cells; P < 0.0001). Plasma membrane staining with a lipophilic dye ensured that the Synt1A β cells adhered tightly to the cover glass (unpublished data). To confirm the TIRFM data, we used EM to examine insulin granules that were morphologically docked to the plasma membrane. Using EM, granules at their shortest distance of <10 nm from the plasma membrane qualified as morphologically docked granules (; ). The number of morphologically docked granules observed by EM was significantly reduced in Synt1A β cells (9.6 ± 1.5 vs. 0.8 ± 0.2 granules per 10 μm of plasma membrane in WT and Synt1A cells, respectively; = 12 cells; P < 0.0001). Along with the results of the morphometric analysis, these data suggest that Synt1A deficiency specifically impairs the docking of insulin granules to the plasma membrane. We explored the effects of Synt1A deficiency on the dynamic motion of single insulin granules. In agreement with what has been reported for rat β cells (), we found that in WT mouse β cells, fusion of insulin granules with the plasma membrane during first-phase release mainly involved previously docked granules (; and Video 3, available at ). In contrast, because Synt1A β cells have fewer docked granules, TIRF analysis in these cells showed that the fusion from previously docked granules was severely abolished (; and Video 4). Despite an appreciable number of fusion events from previously docked granules in WT β cells, there was no fusion from previously docked granules in Synt1A β cells (18.2 ± 1.8 vs.0 in 0–4 min, WT vs. Synt1A; ). However, some fusion from newcomer granules was observed during the first phase even in Synt1A β cells. During second-phase release (>4 min), there was no significant difference in the total number of newcomer fusion events between WT and Synt1A β cells (WT, 43.1 ± 5.0, and Synt1A, 49.8 ± 3.7, during 4–17 min; P = NS; = 10 cells; ). ELISA data evaluating endogenous insulin release from perfused WT and Synt1A β cells () were compatible with the TIRFM data. The small peak of first-phase release from Synt1A β cells shown in perfusion analysis is inferred to be composed of fusion from newcomers. Both the amplitude and time course of the glucose-induced rise in intracellular Ca concentration ([Ca]) measured using Fura-2 were similar between WT and Synt1A β cells (), suggesting that glucose metabolism and ATP production were normally processed in Synt1A β cells and that Synt1A does not affect the activity of the -type Ca channels. This disagrees with the results of other groups (; ; ), but the reason for the discrepancy is unknown. We performed rescue experiments to confirm Synt1A function in the docking and fusing of granules during first-phase release. We restored Synt1A protein expression to Synt1A β cells by infecting them with an adenovirus encoding Synt1A, Adex1CA Synt1A (Ax-Synt1A; ). The number of Synt1A clusters was considerably restored, although to still subnormal levels (270.8 ± 13.0 vs. 212.4 ± 15.7, WT vs. Ax-Synt1A–infected Synt1A cells; P < 0.05). In accordance with restored Synt1A cluster levels, the number of docked insulin granules in Ax-Synt1A–infected Synt1A cells was restored (261.1 ± 13.6 vs. 230.0 ± 12.0, WT vs. Ax-Synt1A–infected Synt1A cells; P = NS). Infection of Adex1CA Synt1A did not alter the number of SNAP25 clusters () that interact with Synt1 clusters (). We then performed TIRFM analysis of the docking and fusion of insulin granules stimulated by 22 mM glucose in Ax-Synt1A–infected Synt1A β cells. This analysis showed a substantial increase in fusion events from previously docked granules ( and Video 5, available at ). The total number of fusion events from previously docked granules during the first phase in Ax-Synt1A–infected Synt1A β cells was restored (18.2 ± 1.8 vs. 12.7 ± 3.3 in 0–4 min, WT vs. Ax-Synt1A– infected Synt1A cells; P = NS). Synt1A restoration did not affect fusion events from newcomers during the second phase. In addition, we examined the interaction between insulin granules and Synt1A clusters labeled with TAT-conjugated Cy3 anti-Synt1A mAb in Ax-Synt1A–infected Synt1A cells. Dual-color TIRFM showed that previously docked granules fused at the site of the Synt1A clusters during the first phase; during the second phase, newcomer granules fused external to the Synt1A clusters (). This was also observed in WT β cells. These data support a model where Synt1A clusters are required for previously docked granules to dock and fuse during the first phase but dispensable for newcomers to dock and fuse during the second phase. As shown in , the fusion of newcomers during the second phase was well preserved in the absence of Synt1A. Yet the question remained of whether other syntaxin isoforms might be functioning in second-phase release, as pancreatic β cells do express detectable levels of plasma membrane–localized syntaxin isoforms, such as Synt3 and -4 (; ). To investigate whether these membrane syntaxins Synt3 and -4 are involved in the second phase, we used TAT fusion proteins that encode the Synt3-H3 (TAT-Synt3-H3) and Synt4-H3 (TAT-Synt4-H3) domains. We previously reported that the recombinant Synt1A SNARE motif (H3 domain) fused to TAT (TAT-Synt1A-H3) rapidly transduced into MIN6 β cells, inhibiting insulin release (). Because the syntaxin H3 domain contributes to one of the four α-helical bundles in the SNARE core complex (), a large molar excess of the Synt1A-H3 domain fused to TAT interrupted the formation of functional SNARE complexes (), as previously reported in other systems (; ). We therefore used TAT-H3 of each syntaxin isoform to perform dominant-negative type experiments. We first produced TAT fusion proteins encoding the Synt3-H3 (TAT-Synt3-H3) and Synt4-H3 (TAT-Synt4-H3) domains. In addition, we produced TAT fusion proteins that encoded the Synt1A-H3 (TAT-Synt1A-H3) and Synt1B-H3 (TAT-Synt1B-H3) domains. A non–coiled-coil domain of ELKS, which has no effect on insulin exocytosis, composed the peptide fusion in our TAT-Control (). As shown in (C and D), the transduction of TAT-Synt3-H3 and TAT-Synt4-H3 into WT β cells reduced the number of fusion events from previously docked granules during the first phase to ∼58 and ∼59% that of control levels, respectively. Second-phase release, which consisted mostly of newcomers, was unaffected by the Synt3-H3 and Synt4-H3 constructs expressed. TAT-Control treatment had no effect on either phase (). However, TAT-Synt1A-H3 treatment strongly reduced the total number of fusion events from previously docked granules during the first phase to ∼23% that of control levels, while showing no effect on second-phase release (). These data are consistent with our results from Synt1A β cells. Synt1B does not express in β cells, but TAT-Synt1B-H3 treatment showed results similar to those in the TAT-Synt1A-H3 treatment, reducing the total number of fusion events from previously docked granules during the first phase to ∼28% that of control levels (). This may be a reflection of a higher homology of Synt1B-H3 to Synt1A-H3. Overall, these findings suggest that these other syntaxin family members are not involved in second-phase release. Because Synt1A β cells exhibit reduced first-phase insulin release, these mice would be expected to develop diabetes. The Goto-Kakizaki rat model for human type 2 diabetes is known to be defective in first-phase insulin release and displays hyperglycemia (). In contrast, we found that Synt1A mice did not show any significant hyperglycemia; fasting blood glucose levels of Synt1A mice were not different from those of WT mice (Synt1A, 63.9 ± 4.3 mg/dl [ = 7], vs. WT, 65.1 ± 3.3 mg/dl [ = 11]; P = NS). However, the oral glucose tolerance test did show impaired glucose tolerance in Synt1A mice (). 30 min after challenge, blood glucose levels in Synt1A mice were significantly higher than in WT mice (Synt1A, 385.0 ± 14.1 mg/dl [ = 7], vs. WT, 286.3 ± 10.4 mg/dl [ = 11]; P < 0.0001). In agreement with these data, we found serum insulin levels to be lower in Synt1A than in WT mice at 30 min after challenge (). Thus, Synt1A mice displayed an impaired glucose tolerance but not marked hyperglycemia. Our dual-color TIRFM approach has shown that during first- phase release insulin granules fuse at the site of Synt1A clusters, but during second-phase release the granules fuse external to Synt1A clusters. We previously found that granules fusing during the first phase originated mostly from morphologically previously docked granules, whereas granules fusing during the second phase arose from newcomers that were originally stored intracellularly (). We also reported that previously docked insulin granules were colocalized with Synt1A clusters in the plasma membrane of MIN6 β cells (). Collectively, these findings suggested that Synt1A is probably essential for docking and fusing insulin granules during the first phase; however, no direct evidence existed to verify this. Recently, it was reported that other isoforms of the syntaxin family might be associated with biphasic insulin release (; ). We therefore used Synt1A mice to directly address how Synt1A functions in granule docking and fusing in biphasic insulin exocytosis. First, we examined the docking status of insulin granules in Synt1A β cells. TIRFM and EM analysis in Synt1A β cells documented a marked reduction of the number of granules docked onto the plasma membrane. Because granules fused during the first phase originated from docked granules, as expected, TIRFM revealed that there was no fusion from docked granules during the first phase in knockout cells. However, fusion from newcomers was still preserved in Synt1A β cells under glucose stimulation. Consistent with these data, perfusion analysis of Synt1A β cells showed a marked reduction in first-phase insulin release but no change in second-phase release. Furthermore, restoration of Synt1A to subnormal levels via the adenoviral vector in Synt1A β cells restored the insulin granules docked onto the plasma membrane, accompanied by an appreciable number of fusion events from these granules. Thus, our data provide direct evidence that Synt1A is essential for docking and fusion of insulin granules during first-phase release. The docking status of synaptic vesicles in the brain hippocampus showed no difference between WT and Synt1A mice (). The reason for this discrepancy between brain and pancreatic β cells is unknown, but it may be due to the expression of Synt1B, which is highly homologous to Synt1A and is abundant in brain cells () but not in pancreatic β cells (). Although the function of Synt1B may not be equal to that of Synt1A in pancreatic β cells (), the brain may have either a tremendous safety network or a different system from pancreatic β cells that permits Synt1B or other homologues to compensate for the lack of Synt1A in brain tissue. Although our data specify a requirement for Synt1A during first-phase release, we still do not know whether other isoforms of the syntaxin family participate in the first phase. WT β cells transduced with TAT-Synt3-H3 and TAT-Synt4-H3, which function in a dominant-negative manner to the corresponding syntaxin isoforms, showed reduction to some extent in the fusion events from previously docked granules during the first phase (). Yet, as no docked insulin granules were seen on the plasma membrane in Synt1A mice, it is difficult to conclude that both Synt3 and -4 are associated with first-phase exocytosis. Rather, we assume that the reduction of fusion events during the first phase by TAT-Synt3-H3 and TAT-Synt4-H3 treatment may reflect the homology of their amino acid sequence to Synt1A-H3. Nevertheless, it remains to be empirically determined if, and how, the other plasma membrane syntaxins contribute to the first phase. Fusion from newcomer granules was not altered at all regardless of Synt1A deletion, indicating that this type of fusion may occur via some mechanism other than the Synt1A-based SNARE complex. Indeed, granule behavior between the first and second phases of release is quite different. As previously reported, upon reaching the plasma membrane, newcomers fused immediately (<50 ms), whereas granules previously docked on the plasma membrane stayed at the same place for a relatively long time (). To examine the possibility that other syntaxin isoforms are involved in second-phase exocytosis, we performed dominant-negative type experiments with TAT-Synt3-H3 and TAT-Synt4-H3 and demonstrated that there was no correlation between syntaxins and second-phase exocytosis. In agreement with our data, noted only a slight decrease in second-phase insulin release in Synt4 mice. Thus, it is plausible that the first phase is Synt1A dependent but the second phase is Synt1A independent, as depicted in the . A similar phenomenon exists in neurotransmitter release, where neural t-SNARE SNAP-25 is essential for evoked neurotransmitter release but nonessential for nonevoked release (). Furthermore, fusion via neural v-SNARE VAMP2 in evoked neurotransmitter release differs from spontaneous neurotransmitter release (). Of course, we do not know whether an evoked or spontaneous release in neurons is compatible with the first and second phases of insulin release in pancreatic β cells, respectively, but there must be some unknown mechanism (other than the SNARE-mediated docking and fusion reactions) in different cell types. Further studies will be required to identify the specific molecules involved in newcomer fusion. In the present study, we also examined the in vivo effect of Synt1A ablation followed by reduced first-phase insulin release on glucose homeostasis. The lack of first-phase insulin release is a main manifestation of type 2 diabetes (; ; ). This phenomenon is quite similar to the insulin release pattern observed in the perfusion of Synt1A β cells. Therefore, we expected Synt1A mice to become diabetic. The in vivo studies showed that Synt1A mice had impaired oral glucose tolerance and decreased serum insulin levels; however, there was no marked hyperglycemia. Thus, factors other than Synt1A depletion may be required to drive hyperglycemia. Williams-Beuren syndrome () is an interesting clinical case when considering the role of Synt1A in diabetes pathogenesis. This syndrome is a multisystem developmental disorder caused by the hemizygous deletion of a 1.5-million-bp region of chromosome 7q11.23 (), which includes the Synt1A gene (). Only some Williams-Beuren syndrome patients exhibit impaired glucose tolerance (). This may be due to the hemizygous deletion; however, it is also postulated that deletion of only Synt1A is not enough to cause abnormal glucose homeostasis. In summary, the present study has provided the first documentation that first-phase insulin release is Synt1A dependent, but second-phase release is Synt1A independent, highlighting that the two phases differ not only spatially but also mechanistically. In a physiological context, our data supporting glucose intolerance in Synt1A mice in vivo encourage therapeutic consideration of the significance of Synt1A in first-phase insulin release. We generated Synt1A mice as previously described (). The genotyping of mice was performed by PCR. Mice were backcrossed with strain C57BL/6 over at least five generations and were used at the age of 10–14 wk. Animal experiments were approved by the Kyorin University Animal Care Committee. Pancreatic islets of Langerhans were isolated from male WT and Synt1A mice by collagenase digestion as described previously (). Isolated islets were dispersed in calcium-free Krebs-Ringer buffer (KRB) containing 1 mM EGTA and cultured on fibronectin-coated (KOKEN Co.), high refractive index cover glass (Olympus) in RPMI 1640 medium (Invitrogen) supplemented with 10% FBS (Invitrogen), 200 U/ml penicillin, and 200 μg/ml streptomycin at 37°C in an atmosphere of 5% CO. To label the insulin secretory granules, pancreatic β cells were infected with recombinant adenovirus Adex1CA insulin-GFP as described previously (). For Synt1A rescue experiments, cells were infected with Adex1CA Synt1A () before being infected with Adex1CA insulin-GFP. Experiments were performed 2 d after the final infection. Proteins were extracted from mouse whole brain or mouse pancreatic islets and immunoblotted as previously described (). Anti-Synt1A mAb and anti-Synt1B pAb were obtained also as previously described (). Antibodies against Synt3 (Synaptic Systems GmbH), Synt4 (BD Biosciences), SNAP-25 (Wako), VAMP2 (Wako), and Munc18 (BD Biosciences) were purchased from commercial sources. WT and Synt1A β cells cultured on high refractive index glass were fixed and made permeable with 2% paraformaldehyde/0.1% Triton X-100 and were processed for immunohistochemistry as described previously (). Cells were labeled with anti-insulin mAb (Sigma-Aldrich), Synt1A, and SNAP-25 and processed with goat anti–mouse IgG conjugated to Alexa Fluor 488 (Invitrogen; ). Immunofluorescence was detected by TIRFM. This procedure allowed us to evaluate the number of docked insulin granules and clusters of Synt1A and SNAP-25. EM was performed by conventional methods as previously described (). Tissues were fixed in phosphate-buffered 2.5% glutaraldehyde, pH 7.4, postosmicated, dehydrated with graded alcohols, and embedded in Epon 812. After staining with uranyl acetate and lead citrate, ultrathin sections were examined with a transmission electron microscope (TEM-1010C; JEOL). In EM, granules at their shortest distance of <10 nm from the plasma membrane were qualified as morphologically docked granules (). For the analysis of islet size and β cell mass, paraffin-embedded pancreas sections (10 μm) were labeled with anti-insulin antibody and detected by an avidin-biotin-peroxidase technique (Vector Laboratories). Sections were collected at 500-μm intervals from tissue blocks, and all islets in the sections were analyzed as islet area over total pancreatic area. Images were acquired with a microscope (IX70; Olympus) that was equipped with a charge-coupled device (CCD) camera and analyzed with MetaMorph software (Universal Imaging Corp.). TAT-conjugated Cy3-labeled anti-Synt1A antibody was prepared as described elsewhere (). In brief, anti-Synt1A mAb was labeled with Cy3 by use of a Fluoro Link antibody Cy3 labeling kit (GE Healthcare), according to the manufacturer's instructions. The Cy3-labeled antibody was dialyzed against 0.1 M borate buffer and was incubated with a fivefold molar excess of a cross-linker, sulfosuccinimidyl 6-(3′-[2-pyridyldithio]-propionamido) hexanoate (Pierce Chemical Co.) for 3 h at room temperature. The conjugated antibody was separated from the free cross-linker by gel filtration eluted with 5 mM Hanks' Hepes buffer, pH 7.2. A 10-fold molar excess of TAT protein transduction domain (PTD) peptide (GYGRKKRRQRRRGGGC) was added to the conjugated antibody, and the mixture was incubated overnight at 4°C. The TAT-conjugated antibody was separated from the free TAT PTD peptide by gel filtration eluted with 5 mM Hanks' Hepes buffer. On the day of TIRFM experiments, Adex1CA insulin-GFP–infected cells were treated with ∼120 μg/ml TAT-conjugated Cy3-labeled anti-Synt1A mAb for 50 min as described previously (). To produce constructs in which the TAT PTD peptide is located at the N terminus of Synt1A-H3 (aa 202–265), Synt1B-H3 (aa 201–264), Synt3-H3 (aa 201–264), Synt4-H3 (aa 210–273), or control peptides (non–coiled-coil domain of ELKS; aa 324–403; ), the coding region that corresponds to rat Synt1A-H3, Synt1B-H3, Synt3-H3, Synt4-H3, or control peptides was amplified by PCR by using oligonucleotide primers, including the nucleotide sequence against the TAT PTD peptide (YGRKKRRQRRR) in each sense primer, as described previously (). PCR products were subcloned into a pPROEX HTa bacterial expression vector (Invitrogen) with an additional His tag at the N terminus. The resulting products were confirmed by an automated DNA sequencer (GE Healthcare). TAT fusion proteins in the pPROEX HTa vector were expressed in a DH5α strain by induction with isopropyl-b--thiogalactopyranoside for 5 h at 37°C. The recombinant proteins were extracted with 8 M urea in 50 mM Tris and 100 mM KCl, pH 8.0. Urea extracts were incubated with Ni-NTA-agarose (QIAGEN) before washing, and stepwise removal of urea was performed to allow renaturation of bound protein. Proteins were eluted from Ni-agarose by 200 mM imidazole and were desalted on a PD-10 column (GE Healthcare) with Hanks' balanced salt solution (Invitrogen). The Olympus total internal reflection system was used with a high-aperture objective lens (Apo 100× OHR; NA 1.65; Olympus) as previously published (). To observe GFP or Alexa Fluor 488 alone, we used a 488-nm laser line for excitation and a 515-nm long-pass filter for the barrier. Diiodomethane sulfur immersion oil ( = 1.81; Cargille Laboratories) was used to make contact between the objective lens and the high refractive index cover glass. Light propagates through the cover glass at an angle measured as 65° and undergoes total internal reflection at the glass–cell interface. The refractive indices for glass ( = 1.8 at 488 nm) and cells ( = 1.37) predict an evanescent field declining e-fold within 44 nm from the interface and to ∼10% within 100 nm. A granule 100 nm from the interface would be illuminated too dimly to be visible under our conditions. Thus, we look barely 100 nm into the cell, a distance comparable to the thickness of ultrathin sections cut for EM (). In an evanescent field declining e-fold within 44 nm, a granule at 80% brightness would have a vertical distance of 9.6 nm from the plasma membrane and qualify as a morphologically docked granule (granule distance from plasma membrane <10 nm in EM studies; ). Images were projected onto a CCD camera (DV887DCSBV; Andor) operated with MetaMorph version 6.3. Images were acquired at 300-ms intervals. For real-time images of GFP-tagged insulin granule motion by TIRFM, treated β cells were placed on the high refractive index glass, mounted in an open chamber, and incubated for 30 min at 37°C in KRB containing 110 mM NaCl, 4.4 mM KCl, 1.45 mM KHPO, 1.2 mM MgSO, 2.3 mM calcium gluconate, 4.8 mM NaHCO, 2.2 mM glucose, 10 mM Hepes, pH 7.4, and 0.3% bovine serum albumin. Cells were then transferred to the thermostat-controlled stage (37°C) of TIRFM, and stimulation with glucose was achieved by the addition of 52 mM glucose-KRB into the chamber for a final concentration of 22 mM glucose. Most analyses, including tracking (single projection of different images) and area calculations were performed using MetaMorph software. To analyze the data, fusion events were manually selected, and the mean fluorescence intensity of individual granules in a 1 μm × 1 μm square placed over the granule center was calculated. The number of fusion events was manually counted while looping ∼5,000 frame time lapses. To observe the fluorescence of GFP and Cy3 simultaneously, we used the 488-nm laser line for excitation and an image splitter (Optical Insight) that divided the green and red components of the images with a 565-nm dichroic mirror (Q565; Chroma Technology Corp.), passing the green component through a 530 ± 15 nm bandpass filter (HQ530/30 m; Chroma Technology Corp.) and the red component through a 630 nm ± 25 nm bandpass filter (HQ630/50 m [Chroma Technology Corp.]; ). Images were then projected side by side onto a CCD camera. The two images were brought into focus in the same plane by adding weak lenses to one channel, and they were brought into register by careful adjustment of the mirrors in the image splitter. Before each experimental session, we took an alignment image that showed density by means of scattered 90 nm TetraSpeck fluorescent beads (Invitrogen). They were visible in both the green and red channels, and thus provided markers in the x-y plane. Beads in the two images were brought into superposition by shifting one image using MetaMorph software. β cells were housed in a small chamber (∼5 × 10 cells/chamber) and perfused with KRB (2.2 mM glucose) for 60 min at a flow rate of 0.5 ml/min at 37°C before collecting fractions. Insulin release was stimulated by 22 mM glucose. Fractions were collected at 1-min intervals. Insulin release in aliquots of media was measured by an insulin ELISA kit (Morinaga). β cells were loaded with 2 μM fura-2 acetoxymethyl ester (Fura-2 AM; Invitrogen) for 30 min at 37°C in KRB (2.2 mM glucose) and washed and incubated for an additional 15 min with KRB. Coverslips were mounted on an ARGUS/HiSCA system (Hamamatsu Photonics). Fura-2 fluorescence was detected by the cooled CCD camera after excitation at 340 nm (F340) and 380 nm (F380), and the ratio image (F340/F380) was calculated with the ARGUS/HiSCA system. Male mice age 10–14 wk were fasted for 14–15 h before the test. Glucose was administered orally at 2 g glucose/kg body weight. Blood samples were collected from a tail vein at 0, 30, 60, 90, and 120 min after loading. Blood glucose levels were measured by Glutest R (Sanwa Kagaku Kenkyusyo). Plasma insulin levels were measured by an insulin ELISA kit. Fig. S1 shows a TIRF image of Synt1A clusters in the plasma membrane labeled with TAT-conjugated Cy3-labeled anti-Synt1A mAb and stained with anti-Synt1A pAb. Fig. S2 shows an immunoblot analysis of Synt1A and other SNARE proteins in the brain and pancreatic islets from WT and Synt1A mice. Fig. S3 shows microscopic examination of pancreatic islets in WT and Synt1A mice. Video 1 displays dual-color TIRF images of GFP-tagged insulin granules and Cy3-labeled Synt1A clusters during first-phase insulin release (0–4 min after glucose stimulation). Video 2 shows dual-color TIRFM of GFP-tagged insulin granules and Cy3-labeled Synt1A clusters during second-phase insulin release (>4 min after glucose stimulation). Video 3 shows TIRFM of GFP-tagged insulin granule motion in the WT mouse β cell under 22 mM glucose stimulation. Video 4 shows TIRFM of GFP-tagged insulin granule motion in the Synt1A mouse β cell under 22 mM glucose stimulation. Video 5 shows TIRFM of GFP-tagged insulin granule motion in the Synt1A mouse β cell infected with Ax-Synt1A under 22 mM glucose stimulation. Online supplemental material is available at .