A potential role for human cohesin in mitotic spindle aster assembly.

The cohesin multiprotein complex containing SMC1, SMC3, Scc3 (SA), and Scc1 (Rad21) is required for sister chromatid cohesion in eukaryotes. Although metazoan cohesin associates with chromosomes and was shown to function in the establishment of sister chromatid cohesion during interphase, the majority of cohesin was found to be off chromosomes and reside in the cytoplasm in metaphase. Despite its dissociation from chromosomes, however, microinjection of an antibody against human SMC1 led to disorganization of the metaphase plate and cell cycle arrest, indicating that human cohesin still plays an important role in metaphase. To address the mitotic function of human cohesin, the subcellular localization of cohesin components was reexamined in human cells. Interestingly, we found that cohesin localizes to the spindle poles during mitosis and interacts with NuMA, a spindle pole-associated factor required for mitotic spindle organization. The interaction with NuMA persists during interphase. Similar to NuMA, a significant amount of cohesin was found to associate with the nuclear matrix. Furthermore, in the absence of cohesin, mitotic spindle asters failed to form in vitro. Our results raise the intriguing possibility that in addition to its well demonstrated function in sister chromatid cohesion, cohesin may be involved in spindle assembly during mitosis.

Proper chromosome segregation in mitosis is essential for the maintenance of chromosome integrity. Sister chromatid cohesion, which ensures the pairing of the two sister chromatids, is a prerequisite for proper chromosome segregation at anaphase. Cohesin is a multiprotein complex required for sister chromatid cohesion in eukaryotes. Cohesin was first characterized in Saccharomyces cerevisiae, which is composed of two structural maintenance of chromosome (SMC) 1 family proteins, Smc1p and Smc3p, as well as the non-SMC components Scc1 and Scc3p (Refs.1-3 and see review in Ref. 4). A similar complex was identified in Schizosaccharomyces pombe, Xenopus, and humans (5)(6)(7). In S. cerevisiae, cohesin functions in both the establishment and maintenance of sister chromatid cohesion from S phase to metaphase (1,2). Consistently, cohesin in S. cerevisiae stays associated with chromosomes from the late G 1 phase until the onset of anaphase (3). Scc1p is a homolog of Rad21 identified earlier in S. pombe (8). Proteolytic cleavage of Scc1p by the cysteine protease separin, which is indirectly promoted by the anaphase-promoting complex at the end of metaphase, is responsible for the destruction of cohesin and the initiation of sister chromatid segregation in anaphase (9). A similar observation was made in S. pombe, although S. pombe cohesin stays associated with chromosomes throughout the cell cycle (7). Studies of S. cerevisiae identified the in vivo binding sites of cohesin on the chromosome arms at estimated intervals of 13 kb with the highest concentration at the centromeric regions (10 -12). Similar centromeric clustering of cohesin was also observed in S. pombe (7).
In metazoans, the protein composition of cohesin is conserved, although there are two Scc3p homologs, stromal antigen 1 (SA1) and 2 (SA2), forming separate cohesin complexes in the cell (5,6). Interestingly, chromosomal association of cohesin in metazoans differs from that of yeast cohesin. In a Xenopus oocyte cell-free system, depletion of cohesin during interphase prior to entering mitosis inhibited subsequent mitotic cohesion of sister chromatids, suggesting that the complex is required for establishing cohesion following DNA replication (13). In contrast to S. cerevisiae, however, metazoan cohesin was found to dissociate from chromosomes prior to entering mitosis (6,13,14). Therefore, it is unclear how sister chromatid cohesion is maintained during metaphase. A recent study using Myc-tagged human RAD21 (hRAD21/Scc1) suggested that at least a small amount of hRAD21 is in the centromeric regions during metaphase and is subject to a similar proteolytic cleavage as in yeast (15). Similar localization was reported with the endogenous hRAD21 protein (16). In Xenopus, SA1 was shown to localize between two sister chromatids in vitro, suggesting that a low amount of cohesin may continue to associate with metaphase chromosomes in higher eukaryotes, reminiscent of buttons on a jacket securing a few critical regions of sister chromatids together (5). It was reasoned that the chromosomes of higher eukaryotes must condense to a higher level and that the presence of too many cohesin molecules may interfere with efficient chromosome condensation. However, the observed localization represents a very minor population of cohesin (estimated to be ϳ5%) (5), and the role of metazoan cohesin in metaphase cohesion has not been clearly demonstrated. Furthermore, it is not known whether cytoplasmic cohesin plays any role in mitosis.
Despite the fact that the majority of cohesin is in the cytoplasm in mitotic cells, we previously showed that the injection of an antibody specific for hSMC1 into human mitotic cells blocked the progression of metaphase and led to disorganization of the metaphase plate, suggesting a role for cohesin in mitosis (14). However, its mechanism was unknown. The same antibody failed to detect any hSMC1 at the interface between the two sister chromatids, which would be the predicted site for cohesin localization. To further address the mitotic role of human cohesin, we characterized the human cohesin complex containing hSMC1, hSMC3, hRAD21, and SA1/SA2 and performed detailed immunostaining and biochemical analyses of cohesin subcellular localization in human cells. Here we report that cohesin localizes at the spindle poles during mitosis and interacts with the nuclear mitotic apparatus protein (NuMA) and more weakly but specifically with dynein and ␤-tubulin. Similar to NuMA, cohesin also interacts with the nuclear matrix during interphase. Further analysis showed that cohesin is required for mitotic spindle aster assembly in vitro. Our studies suggest a novel function of cohesin in mitotic spindle organization.

EXPERIMENTAL PROCEDURES
Cell Lines and Cell Culture-HeLa cells were grown in Dulbecco's modified Eagle's medium (Sigma) supplemented with 10% fetal bovine serum, L-glutamate, and penicillin/streptomycin.
Synchronization of HeLa Cells-HeLa cells were synchronized to the S and G 2 phases by a double thymidine block, and to G 1 phase by an additional nocodazole treatment as previously described (17). The DNA content profile from cells synchronized at each cell cycle stage was analyzed by a FACSCalibur (BD PharMingen). For FACS analysis, cells were fixed in 75% ethanol at Ϫ80°C. Cells were then stained with 5 g/ml propidium iodide in the presence of 300 g/ml RNase A.
Antibodies-Rabbit polyclonal antibodies were raised against recombinant polypeptides expressed in Escherichia coli corresponding to the middle and C-terminal domains of hSMC1 (amino acids 402-894 and 888 -1233, respectively), the N-terminal domain of hSMC3 (amino acids 1-303), the N-and C-terminal domains of human SA1 designated as SA1N and SA1C (amino acids 1-230 and 1028 -1258, respectively), the C-terminal domain of hRAD21 (amino acids 286 -543), and the N terminus of NuMA (amino acids 1-307). Antibodies were affinity-purified with the same recombinant polypeptides purified and cross-linked to Affi-Gel (Bio-Rad) or expressed as glutathione S-transferase fusion proteins crosslinked to glutathione beads (Amersham Pharmacia Biotech.) with dimethyl pimelimidate (Sigma). Anti-Xenopus RAD21 (XRAD21) peptide antibody against the conserved C-terminal domain was kindly provided by Dr. Tatsuya Hirano at Cold Spring Harbor Laboratory (13). Anti-DNA ligase III antibody was generously provided by Dr. Thomas Lindahl at the Imperial Cancer Research Fund, United Kingdom. Anti-BRCA1 antibody was kindly provided by Dr. Ramin Shiekhattar at the Wistar Institute, Philadelphia, PA. Mouse monoclonal antibodies specific for NuMA (Calbiochem), for the dynein 70-kDa subunit (Sigma), and for ␤-tubulin (Sigma) were also used for colocalization studies and Western blot analyses. Secondary antibodies included goat anti-rabbit IgG antibody conjugated with Cy3 (Jackson Laboratory), and donkey anti-mouse IgG antibody conjugated with fluorescein (Vector Laboratory).
Immunoprecipitation and Western Blot Analysis-Large scale immunoprecipitation and peptide sequencing protocols used for the identification of hRAD21 are similar to those described (14). Coimmunoprecipitation was performed as previously described (14). The immunoprecipitated proteins were analyzed by SDS-PAGE and silver staining or transferred to nitrocellulose membrane and analyzed by Western blotting as described previously (14,17). Western blots were developed by either colorimetric reaction (Promega) or enhanced chemiluminescence (ECL kit, Amersham Pharmacia Biotech).
Sucrose Gradient Ultracentrifugation-To estimate sedimentation coefficient values, crude HeLa cell nuclear extracts in 0.4 M salt were subjected to 5-30% sucrose gradient centrifugation at 155,000 ϫ g for 15 h with a SW50.1 rotor (Beckman Instruments). Each fraction was trichloroacetic acid-precipitated and analyzed by SDS-PAGE and Western blotting. Standard proteins bovine serum albumin (4.6 S), catalase (11.3 S), and ferritin (16.5 S) were run in the same gradient multiple times to confirm the linearity of the gradient.
Immunofluorescent Staining Analysis-Immunofluorescent staining of cells was performed as described previously (14,17). Immunofluorescent image analysis was performed using a Zeiss Axioplan 2 microscope with a Photometrics Sensys Camera. J/C Cittert iterative digital deconvolution was performed using the Zeiss KS 300 program. Alternatively, the analysis was carried out using an Olympus IX70 with the Mag-naFire digital charge-coupled device camera system.
Cell Extraction-In situ cell extraction was performed essentially as described (17). Cell extraction was performed either in situ on coverslips for immunostaining analysis or in test tubes for Western blot analysis. The same number of cells were used for the extraction in test tubes. Briefly, the cells were washed with phosphate-buffered saline and cytoskeleton (CSK) buffer (10 mM Pipes, pH 7.0, 100 mM NaCl, 300 mM sucrose, and 3 mM MgCl 2 ). Cells were extracted using CSK buffer with 0.5% Triton-X for 5 min at 4°C to remove soluble cytoplasmic proteins. Cells were then treated with an extraction buffer (42.5 mM Tris-HCl, pH 8.3, 8.5 mM NaCl, 2.6 mM MgCl 2 , 1% Tween 20, and 0.5% sodium deoxycholate) for 5 min at 4°C to remove cytoskeletal proteins. The cells were subsequently treated with CSK buffer (containing 2 mM CaCl 2 and 2 mM MgCl 2 ), 0.5% Triton X, and 100 g/ml DNase I (Worthington) for 30 min at 37°C. The cells were then washed with 0.25 M ammonium sulfate in CSK. In a control sample, cells were treated with the same buffer without DNase I. Between each step, cells were centrifuged at 3,000 rpm, and the supernatant was recovered as the CSK extraction and DNase fractions. The pellet was washed twice in CSK buffer containing 0.5% Triton X, resuspended in SDS sample buffer, and subjected to SDS-PAGE and Western blot analysis. The experiments were repeated at least three times. The intensities of the protein bands in the Western blots from at least three experiments were quantitated by NIH IMAGE.
In Vitro Mitotic Aster Assembly Assay-The in vitro aster assembly assay was performed using the protocol of Gaglio et al. (18 -20). Briefly, HeLa cells were synchronized at mitosis as described above. Mitotic cells were collected by shake-off and incubated with 20 g/ml cytochalasin B. Cells were then washed with phosphate-buffered saline and resuspended in KHM buffer (78 mM KCl, 50 mM Hepes, pH 7.0, 4 mM MgCl 2 , 2 mM EGTA, 1 mM dithiothreitol, and 0.2 mM 4-(2-aminoethyl)benzenesulfonyl fluoride) containing cytochalasin B at a concentration of ϳ3 ϫ 10 7 cells/ml. Cells were Dounce-homogenized, and the crude extract was subjected to ultracentrifugation at 100,000 ϫ g for 15 min. at 4°C. The supernatant was collected, and a fraction of it was subjected to immunodepletion. Approximately 10 -20 g of either preimmune IgG purified with protein A beads, antigen affinity-purified anti-hSMC1, anti-hSMC3, anti-SA-1N, anti-hRAD21, or anti-BRCA1 antibody were coupled to protein A beads and incubated with 20 l of mitotic extracts for 45 min at 4°C. The beads were spun down, the supernatants were collected, and the depletion process was repeated. The final supernatants were passed through empty spin columns to remove any remaining beads. The supernatants were incubated for 60 min at 30°C in the presence of 2.5 mM ATP and 10 M taxol for the in vitro aster assembly. After the reaction, a small portion of each sample was dropped onto a coverslip and subjected to immunofluorescent staining with an antibody specific for ␤-tubulin (Sigma) or co-stained with anti-NuMA antibody. The remaining portions of the samples were used for Western blot analysis to detect the presence of hSMC1, ␤-tubulin, and NuMA. Mitotic asters assembled in a control reaction without immunodepletion were sedimented at 10,000 ϫ g for 15 min at 4°C as described (18), and the presence of NuMA and hSMC1 in the supernatant and pellet was analyzed by Western blot.

RESULTS
The hSMC1-hSMC3 Heterodimer Is in the Cohesin Complex in Both Interphase and Mitosis-Previously, we identified the stable heterodimeric complex of hSMC1 and hSMC3 in human cells (14). Reciprocal coimmunoprecipitation of the endogenous hSMC1 and hSMC3 proteins from HeLa nuclear extracts indicates that the SMC molecules are engaged in a highly stable heterodimeric complex resistant to a mild denaturant (2 M guanidine) in an equimolar ratio in the absence of any other associated proteins (Fig. 1A). Because microinjection of an an-tibody against hSMC1 blocked the progression of metaphase in the previous study (14), we first wanted to determine whether hSMC1-hSMC3 is still in the cohesin complex in mitotic cells. Under the coimmunoprecipitation condition, including a 1 M salt wash, human cohesin containing hRAD21 and SA1/SA2 is coimmunopurified with antibody specific for hSMC1 ( Fig. 1, B and C). The association of hRAD21 and SA1/SA2 with hSMC1-hSMC3, although resistant to the 1 M salt wash, is sensitive to 2 M guanidine unlike the interaction between hSMC1 and hSMC3 (Fig. 1A). Therefore, an hSMC1-hSMC3 heterodimer is a core subunit of cohesin, and hRAD21 and SA1/SA2 associate with it. The presence of hRad21 was also confirmed by peptide sequence analysis of the 115-kDa protein corresponding to hRAD21 (KINHL (amino acid 210) and KGGEADNLDEFLK (amino acid 410)) (GenBank TM accession number D38551) (21). The amounts of hRAD21 and SA1/SA2 coimmunoprecipitated with anti-hSMC1 antibody are comparable between S-and M-phase extracts, indicating that hSMC1-hSMC3 is part of cohesin during both S phase and mitosis ( Fig. 1, B and C). Anti-SA1N antibody (directed against the N terminus of SA1) detected both SA1 and SA2, whereas anti-SA1C (against the C terminus) antibody detected a single protein species (SA1), consistent with the different C-terminal sequences in SA1 and SA2 (Fig. 1, C and D) (5). Because bovine SMC1 and SMC3 was found to be part of the recombination repair complex, RC-1 (22), the presence of DNA ligase III in the hSMC1 coimmunoprecipitation was also examined. DNA ligase III was not detected in our coimmunoprecipitation from HeLa nuclear extracts, indicating that the RC-1 repair complex represents a very minor complex in human cells (data not shown). Taken together, these results indicate that hSMC1-hSMC3 is in the cohesin complex in both S phase and mitotic cells at similar levels. The specificity of the antigen affinity-purified antibodies specific for hSMC1, hSMC3, hRAD21, and SA1N and SA1C was demonstrated by Western blot analysis using crude HeLa extracts from various cell cycle stages (14) as well as mitotic whole cell extracts (Fig. 1D).
Cohesin Localizes to Spindle Poles during Mitosis and Interacts with NuMA, Dynein, and ␤-Tubulin-Using the above antibodies, we next tested whether cohesin localizes to specific sites in mitotic cells. We performed a partial in situ extraction of soluble proteins for this purpose (CSK extraction, see "Exper-imental Procedures"). We found that a subpopulation of cohesin localizes to spindle poles in mitotic cells, both in metaphase and anaphase (Fig. 2). Spindle pole staining was detected using antibodies against hSMC1 (produced in two different rabbits), hSMC3, SA1N (which also recognizes SA2), and SA1C ( Fig. 2 and data not shown). The corresponding preimmune antisera and antigen-depleted antibodies did not show spindle pole staining, indicating that the staining is antigen-specific (data not shown). The spindle pole localization was verified by co-staining with both ␤and ␥-tubulin antibodies (data not shown). Thus, cohesin is present at spindle poles during mitosis.
Based on the above results, we next tested for potential interactions of cohesin with proteins that are known to localize to the spindle poles. We found that cohesin specifically interacts with NuMA. NuMA plays an important role in mitotic spindle organization in vivo and in vitro (20,23). NuMA interacts with dynein/dynactin and ␤-tubulin and is targeted to the FIG. 1. Human cohesin, containing a hSMC1/hSMC3 heterodimer, SA1/SA2, and hRAD21 is a major complex containing hSMC1/ hSMC3. A, reciprocal immunoprecipitation of the hSMC1/hSMC3 heterodimeric complex. Antibodies specific for hSMC1 and hSMC3 were used to immunoprecipitate the endogenous SMC proteins from HeLa nuclear extracts. Immunoprecipitates were washed with 2 M guanidine-HCl, and proteins remaining on antibody beads were visualized by silver staining. The polypeptides corresponding to hSMC1 and hSMC3 are indicated. The antibodies used for immunoprecipitation are indicated at the top. IgH, IgG heavy chain; M, molecular weight marker. B, coimmunoprecipitation of hRAD21 with anti-hSMC1 antibody. HeLa extracts from S phase (S) and M phase (M) were immunoprecipitated with an anti-hSMC1 antibody and washed with 1 M salt. The precipitated materials were subjected to Western blot analysis using an antibody specific for RAD21.

FIG. 2. Human cohesin localizes to spindle poles in mitosis.
Immunofluorescent staining of spindle poles with antibodies specific for hSMC1 and SA1N. DNA was visualized by DAPI staining. Cells were pre-extracted with CSK buffer. Mitotic spindle poles are indicated by white arrowheads. In merged images (panels 3 and 6), SA1 or hSMC1 (red) and DAPI (cyan) are shown. mitotic spindle poles by the dynein/dynactin-dependent transport mechanism (23,24). An anti-hSMC3 antibody co-precipitated NuMA from mitotic HeLa extracts, which is resistant to a 1 M salt wash (Fig. 3, lane 4). Neither the protein A beads alone nor the preimmune antibody precipitated NuMA (Fig. 3,  lanes 2 and 3). Because NuMA interacts with dynein and ␤-tubulin, cohesin interaction with these proteins was also tested. Anti-hSMC3 antibody coimmunoprecipitated dynein and ␤-tubulin from mitotic extracts (Fig. 3, lanes 5-15). No dynein or ␤-tubulin was precipitated with protein A beads alone or with preimmune IgG (lanes 6, 7, 9, 10, 13, and 14). In contrast to the NuMA interaction, however, these interactions were entirely sensitive to 1 M salt, indicating that the interaction is weaker and may be indirect (Fig. 3, lanes 8 and 15).
The interaction of NuMA and cohesin was also examined in S phase (Fig. 4, A and B). Anti-hSMC1 antibody co-precipitated NuMA from both S and M phase extracts (Fig. 4A, lanes 5 and  6). In contrast, CNAP1 (hCAP-D2), a component of the condensin complex that is present abundantly in the mitotic cytoplasm as well as on chromosomes (17), failed to co-precipitate with hSMC1, demonstrating the specificity of the coimmunoprecipitation (Fig. 4A, lanes 9 -12). Consistently, NuMA reciprocally coimmunoprecipitated hSMC1 from both S-and Mphase extracts (Fig. 4B). Furthermore, SA1/SA2, hRAD21, and NuMA co-precipitated each other in a 1 M salt-resistant manner (data not shown). Interestingly, among the multiple species of NuMA, only a subspecies of NuMA appears to tightly interact with cohesin (Fig. 3, compare lanes 1 and 4; Fig. 4A, compare lanes 1, 2, 5, and 6). The one that preferentially interacts with cohesin in a 1 M salt-resistant manner differs between S and M phases, a faster migrating species in S phase and a slower one in M phase (Fig. 4A, lanes 5 and 6). NuMA was shown to be hyperphosphorylated in a mitosis-specific manner by cdc2 kinase, which results in slower migration of the protein in SDS-PAGE (25). This phosphorylation was shown to be critical for the proper targeting of NuMA to spindle microtubules and is thus important for the function of NuMA in spindle organization (26). Therefore, our results strongly suggest that cohesin specifically interacts with NuMA localized at the spindle structure in mitosis (Fig. 3, lane 4; Fig. 4A, lane 6). However, the tight interaction was only seen with a subpopulation of each form of NuMA (Fig. 4A, compare 1 M salt and guanidine eluate), and the peak of NuMA in sucrose density gradient centrifugation did not coincide with that of cohesin, indicating that NuMA is not a component of cohesin (Fig. 4C). Taken together, the identified interaction of cohesin with the known spindle pole-associated protein NuMA, as well as weaker but specific interactions with dynein and ␤-tubulin, further supports the observed localization of cohesin at the spindle poles.
Human Cohesin Interacts with the Nuclear Matrix-NuMA was shown to associate with the nuclear matrix during inter-phase (27). Because cohesin interacts with NuMA in interphase, we next tested whether cohesin also associates with the nuclear matrix. We performed serial extractions of cells on coverslips with detergent and DNase I to visualize the proteins tightly associated with the nuclear matrix (28) (Fig. 5). Completion of DNA digestion and extraction was monitored by the disappearance of DAPI staining (Fig. 5, panel 4). The results revealed that a significant amount of hSMC1 associates tightly with the nuclear matrix. Similar results were obtained with antibodies against hSMC3, hRAD21, and SA1/SA2 (data not shown). No significant difference was observed between the cells extracted with or without high salt after DNase I treatment. Thus, it is unlikely that the observed nuclear matrix localization  5 and 6). B, reciprocal immunoprecipitation of hSMC1 with anti-NuMA antibody. Lanes 1 and 4, input; lanes 2 and 5, immunoprecipitated with protein A beads alone (no antibody); and lanes 3 and 6, immunoprecipitated with anti-NuMA antibody as indicated at the top. hSMC1 signal is indicated. Lanes 1-3, mitotic extract; lanes 4 -6, S phase extract. C, Western blot analysis of fractions from 5 to 30% sucrose gradient fractionation of crude HeLa S phase extract with antibodies specific for hSMC1, NuMA, and RAD21. FIG. 3. The hSMC3 protein interacts with NuMA, dynein, and ␤-tubulin in mitosis. Western blot analyses of anti-hSMC3 coimmunoprecipitation probed with anti-NuMA, anti-dynein, and anti-␤-tubulin antibodies. HeLa mitotic extracts were immunoprecipitated with an anti-hSMC3 antibody, and the precipitated material was subjected to multiple low salt washes and eluted with 1 M salt followed by 2 M guanidine-HCl. As indicated at the bottom, proteins eluted by each treatment (1 M salt and 2 M guanidine-HCl) were trichloroacetic acidprecipitated and subjected to Western blot analysis. As indicated at the top, protein A beads without an antibody (lanes 2, 6, 9, and 13) or preimmune IgG on beads (lanes 3, 7, 10, and 14) were incubated with the same extract and were treated in an identical manner. Lanes 1, 5, and 12, an input extract as a positive control for the Western blot.
is due to the precipitation of cohesin proteins. The nuclear matrix association of cohesin contrasts with the observation noted with condensin, the other SMC-containing complex, the nuclear localization of which was sensitive to DNase I treatment (17).
We next performed the same experiments in a test tube. These extractions with salt and detergent as well as DNase I removed soluble cytoplasmic and nuclear proteins as well as DNA-associated proteins in a stepwise manner, leaving proteins that tightly associate with the nuclear matrix. Eluted material from each step was subjected to Western blot analysis with anti-hSMC1 antibody (Fig. 6A). We found three populations of hSMC1. The first population of hSMC1 eluted in the soluble fractions (Fig. 6A, CSK and Extraction, lanes 2 and 3). A second population of hSMC1 was co-eluted with DNA after DNase I digestion (Fig. 6A, lane 4). This elution was dependent on DNase I (Fig. 6A, lane 9). A third population of hSMC1 remained associated with the nuclear matrix (Fig. 6A, lane 5). In the same experiments, cytoplasmic dynein was removed in the first extraction while NuMA remained in the last nuclear matrix fraction, consistent with their subcellular localization (Fig. 6B). When extractions from S and G 2 phase cells were compared, we found that cohesin dissociation from chromosomes is initiated in the G 2 phase. In G 2 phase, significantly less hSMC1 and hSMC3 were found in the DNA fraction compared with S phase, indicating that they are dissociating from chromosomes during G 2 phase (Fig. 6C, compare lanes 4 and 9). Despite their dissociation from chromosomes, however, both hSMC1 and hSMC3 remain associated with the nuclear matrix (Fig. 6C, lane  10). Similar changes were observed with hRAD21 and SA1/SA2 (data not shown). Proper synchronization of cells to S and G 2 phases was confirmed by FACS analysis (Fig. 6D). Importantly, at 7 h after release from the thymidine block (the prospective time point for G 2 phase), cells were clearly in interphase and not in prophase (Fig. 6E). Therefore, this is different from the results in Xenopus embryos in which cohesin dissociation was shown to occur in prometaphase (6). This may reflect the differences in cell cycling between somatic and embryonic cells. Similar experiments with G 1 -synchronized cells revealed that cohesin associates with chromosomes, although at a lower level than in S phase, and with the nuclear matrix (data not shown). These results indicate that the association of cohesin with chromosomes is greatest in S phase and that its dissociation from chromosome begins during G 2 phase, whereas a subpopulation remains associated with the nuclear matrix throughout interphase.
Cohesin Is Required for Mitotic Aster Assembly in Vitro-Localization of human cohesin at the spindle poles raises the possibility that cohesin may function in the spindle organization during mitosis. To test this hypothesis, in vitro mitotic aster assembly was performed using mitotic HeLa extracts, and the effect of cohesin depletion was examined. In the control assay, NuMA and the major population of hSMC1 co-sedimented with assembled asters consistent with the localization of cohesin at the spindle poles in mitotic cells (Fig. 7A). The extracts were immunodepleted with comparable amounts of affinity-purified antibodies against hSMC1, hSMC3, SA1N, anti-BRCA1, or purified preimmune IgG prior to the assembly reaction. Depletion of hSMC1 from the extracts was confirmed by Western blot analysis (Fig. 7B). Anti-hSMC1-hSMC3, and SA1N antibodies efficiently depleted hSMC1, whereas preimmune IgG and anti-BRCA1 failed to deplete hSMC1. The in vitro aster assembly was carried out using these immunodepleted extracts (Fig. 7C). Aster assembly is ATP-dependent, because omitting ATP abolished any aster or visible microtubule formation (Fig. 7C, panel 1). While preimmune IgG-and anti-BRCA1 depletion had no significant effect on aster formation, depletion with the anti-hSMC1, anti-hSMC3, or anti-SA1N antibody severely inhibited aster assembly. Occasionally, we observed a few irregular, smaller asters and short microtubules (Fig. 7C, panels 4, 5, and 6). Anti-BRCA1 antibody was used for comparison because BRCA1 was previously shown to localize to spindle poles and to play a role in centrosome replication (29,30). The control experiments with preimmune IgG and anti-BRCA1 antibody demonstrate that the effect of immunodepletion is cohesin-specific. The failure to assemble asters in the cohesin-depleted extracts was apparently not due to indirect depletion of NuMA or ␤-tubulin, because the amounts of NuMA and ␤-tubulin were comparable in the control and experimental extracts as assessed by Western blot analysis (Fig. 7D). This is consistent with the observed in vivo interaction, in which only a subpopulation of NuMA and ␤-tubulin is involved in the interaction with cohesin (Figs. 3 and 4).
To quantify the efficiency of aster assembly in immunodepleted extracts, equal amounts of assembly reactions were spotted on coverslips and co-stained for ␤-tubulin and NuMA. In an in vitro assay, NuMA localizes to the polar end of the spindles at the core of the spindle aster, which can be used as a marker for the mitotic aster (18 -20). Mitotic extracts were immunodepleted with the preimmune IgG, anti-hSMC1, or anti-hRAD21 antibody (Fig. 8). Similar to antibodies against other cohesin components, anti-hRAD21 immunodepleted cohesin from mitotic extracts (Fig. 8A, lane 3). Two hundred ␤-tubulin/NuMA-positive signals were examined on coverslips in areas randomly chosen under the microscope, and the number of asters with normal morphology was counted (see Fig. 8 legend for details). The average percentages of normal asters in depleted extracts are shown in Fig. 8B. In preimmune-depleted extracts, normal asters constituted more than 95% of ␤-tubulin/NuMA-positive spots, whereas cohesin-depleted extracts showed less than 10% normal asters (Fig. 8B). Fig. 9 shows the typical assembly products by preimmune-depleted and cohesindepleted extracts. While clusters of asters with bright ␤-tubulin and NuMA staining at the central core were seen in the preimmune-depleted assembly similar to the undepleted control, only a few irregular spindles with weak ␤-tubulin staining and associated NuMA signals were seen in cohesin-depleted extracts. Taken together, these results indicate that cohesin is required for mitotic aster formation in vitro. DISCUSSION Although cohesin in both yeast and metazoans appears to play a role in the establishment of sister chromatid cohesion after replication during S phase, there are differences regard-  Fig. 8A were used to assemble asters in vitro. The same volumes of reaction mixtures were placed on coverslips and co-stained with anti-␤-tubulin and anti-NuMA antibodies (see Fig. 9 legend). We examined 200 ␤-tubulin-positive NuMA spots and counted normal asters among them by randomly choosing areas on the coverslip under the microscope. This was repeated three times on multiple coverslips from two independent experiments. The criteria for normal asters in this experiment are: 1) distinct clustering of ␤-tubulin at the center; 2) co-localization of ␤-tubulin with NuMA; and 3) at least four spindles emanating from the aster in at least three different directions (see Fig. 9). Percentages of normal asters were averaged with standard deviation and are shown in the bar graph.
ing the timing of cohesin dissociation from chromosomes and cohesin function during mitosis. Namely, cohesin functions in sister chromatid cohesion until the onset of anaphase in yeast, whereas a majority of it dissociates from chromosomes prior to mitosis in metazoans. However, our previous antibody microinjection studies suggested that hSMC1 is still involved in the progression of mitosis, possibly in metaphase plate organization in human mitotic cells (14). Does cytoplasmic cohesin, which no longer interacts with chromosomes, play any role in mitotic chromosome organization? Our current study demonstrates the mitosis-specific localization of cohesin at the spindle poles and a specific interaction between cohesin and NuMA, a factor critical for spindle organization. Finally, we show that cohesin depletion inhibits mitotic spindle aster assembly in vitro. Based on these results, we propose that cohesin may play an additional role in mitotic chromosome organization by acting on mitotic spindle structures.
Localization of Cohesin at the Mitotic Spindle Poles and the Nuclear Matrix-Analysis of the subcellular localization of a protein by using immunofluorescent staining has often proven to be valuable in deducing its potential function in the cell. It is important, however, to utilize multiple antibodies because staining patterns can vary significantly depending upon the antigen epitopes, especially for large and/or highly structured proteins. Thus, our criteria for determining specific protein localization are the following. 1) Western blot analysis shows highly specific reactivity to the target protein. 2) At least two different batches of antibodies exhibit the same staining pattern, preferably with antibodies against different domains of the same protein. 3) If the protein is in a complex, antibodies against other subunits exhibit the same staining pattern; and finally, 4) the protein biochemically interacts with other proteins that localize to the same place.
We first established the specificity of antibodies for the cohesin components by Western analysis against the crude HeLa extracts. The recent studies indicated that the non-SMC components of cohesin, SA1 and RAD21, localize to the interface between the two sister chromatids, particularly the centromeric region, during metaphase (5,15,16). However, our antibodies against all four cohesin components, including the anti-hSMC1 antibody previously used for antibody microinjection (14), failed to detect any specific localization of the endogenous cohesin components in this location. Instead, we found that a significant population of cohesin localizes to spindle poles. The spindle pole localization of cohesin was confirmed by staining with multiple antibodies against three of the cohesin subunits and was antigen-specific. Interestingly, a previous report demonstrated that hSMC1 interacts with hsHec1, which was shown to localize to the spindle poles (37). Consistent with the spindle pole localization, we observed in mitosis a strong interaction of cohesin with a subpopulation of NuMA, the largest species that most likely represents the mitosis-specific, hyperphosphorylated form (25,26). Weaker interactions of cohesin with dynein and ␤-tubulin, which may be indirectly mediated by NuMA, were also observed. Although our antibody against hRAD21 failed to stain the mitotic spindle poles, NuMA-coimmunoprecipitated hRAD21 from mitotic extracts and cohesin depletion by an anti-hRAD21 antibody had a similar inhibitory effect on mitotic aster assembly. Thus, the lack of staining is most likely due to epitope inaccessibility of our antibody. Consistent with this notion, hRad21 was recently observed at mitotic spindles using a different antibody (16).
In interphase cells, sequential extraction revealed that not all cohesin associates with DNA, but some is soluble while a third population associates with the nuclear matrix. Dissociation of all four cohesin components from chromosomes is initiated during G 2 phase earlier than previously observed in a Xenopus system, which may be due to differences in cell cycling between somatic cells and embryos (6). Consistent with the nuclear matrix association, the interaction between cohesin and NuMA persists dur-  D, F, and H) with NuMA in red, ␤-tubulin in green, and overlap in yellow. Upper panels are the assembly with preimmune-depleted extract, and the lower panels are with an anti-hRAD21depleted extract as indicated on the left. Because the ␤-tubulin staining of the normal asters in the preimmune-depleted extract assembly is so much brighter than that of the irregular asters in the cohesin-depleted extract assembly, the former image was exposed three times less than the latter to visualize asters clearly. ing interphase (25,26). Although not well defined, the nuclear matrix has been implicated in molecule transport as well as in a higher order chromosome organization (31)(32)(33). Our results raise the interesting possibility that cohesin-mediated chromosome organization may occur in the context of the nuclear matrix. Consistent with this hypothesis, cohesin was reported to play a role in establishing the boundary elements at the HMR locus in S. cerevisiae (34). Boundary elements are thought to exert their function by interacting with the nuclear matrix (35,36). Alternatively, the nuclear matrix-associated cohesin may constitute a separate population of cohesin, which may be dedicated to participation in spindle organization in mitosis.
The Possible Role of Cohesin in Mitotic Spindle Organization-To determine the potential function of cohesin found at the spindle apparatus, an in vitro aster assembly assay was carried out using HeLa mitotic extracts. This assay was used previously to demonstrate the function of proteins localized at the spindle poles, such as NuMA, Eg5, and dynein in mitotic spindle organization (18 -20, 23). These results were validated by the in vivo data in Xenopus (23). Using this assay, we demonstrated that mitotic aster assembly was inhibited in a cohesin depletion-specific manner. Aster assembly was inhibited by immunodepletion with an antibody against hSMC1, hSMC3, SA1/SA2, or hRAD21, but not with preimmune IgG, indicating that the cohesin complex is involved in the assembly of mitotic asters. Although BRCA1 was shown to localize to spindle poles and play an important role in centrosome replication (29,30), immunodepletion using anti-BRCA1 antibody did not deplete cohesin and had no effect on mitotic spindle aster assembly, indicating that the inhibition of aster formation is cohesin-specific and not due to a nonspecific depletion effect of any protein at spindle poles. Furthermore, cohesin antibodies failed to co-deplete significant amounts of NuMA and ␤-tubulin in our system, consistent with the fact that only a subpopulation of NuMA and ␤-tubulin interacts with cohesin. This result indicates that the abolishment of aster assembly is not due to the lack of NuMA or ␤-tubulin. This idea is further supported by the observation that NuMA is still associated with the irregularly shaped asters in the cohesin-depleted assembly reactions, indicating that depletion of cohesin does not affect NuMA localization. Rather, it is possible that cohesin is recruited to the spindle poles by NuMA. Because ␤-tubulin staining of the irregular asters was much weaker, cohesin may be required for the assembly or maintenance of spindles. This effect is specific to cohesin and/or other factors that interact with and therefore are co-depleted with cohesin. We have preliminary evidence that cohesin may interact in a mitosis-specific manner with other factors that may be important partners for the mitotic function of cohesin. 2 Further study is necessary to investigate the NuMA-cohesin interaction as well as whether cohesin interacts with other proteins important for mitotic spindle organization, such as the component(s) of centrosomes and/or pericentrosomal structures. Interestingly, the kinesin-related proteins KIF3A and 3B and their associated protein SMAP were previously reported to interact with hSMC3 (HCAP), also suggesting the potential function of cohe-sin in the context of the microtubule network (38). Although currently it is not clear how cohesin participates in mitotic spindle aster assembly, our results indicate the multifunctionality of cohesin in the cell.