A positive feedback mechanism ensures proper assembly of the functional inner centromere during mitosis in human cells

The inner centromere region of a mitotic chromosome critically regulates sister chromatid cohesion and kinetochore–microtubule attachments. However, the molecular mechanism underlying inner centromere assembly remains elusive. Here, using CRISPR/Cas9-based gene editing in HeLa cells, we disrupted the interaction of Shugoshin 1 (Sgo1) with histone H2A phosphorylated on Thr-120 (H2ApT120) to selectively release Sgo1 from mitotic centromeres. Interestingly, cells expressing the H2ApT120-binding defective mutant of Sgo1 have an elevated rate of chromosome missegregation accompanied by weakened centromeric cohesion and decreased centromere accumulation of the chromosomal passenger complex (CPC), an integral part of the inner centromere and a key player in the correction of erroneous kinetochore–microtubule attachments. When artificially tethered to centromeres, a Sgo1 mutant defective in binding protein phosphatase 2A (PP2A) is not able to support proper centromeric cohesion and CPC accumulation, indicating that the Sgo1–PP2A interaction is essential for the integrity of mitotic centromeres. We further provide evidence indicating that Sgo1 protects centromeric cohesin to create a binding site for the histone H3–associated protein kinase Haspin, which not only inhibits the cohesin release factor Wapl and thereby strengthens centromeric cohesion but also phosphorylates histone H3 at Thr-3 to position CPC at inner centromeres. Taken together, our findings reveal a positive feedback–based mechanism that ensures proper assembly of the functional inner centromere during mitosis. They further suggest a causal link between centromeric cohesion defects and chromosomal instability in cancer cells.

Error-free chromosome segregation in mitosis requires timely resolution of sister chromatid cohesion and correct attachment of kinetochores to spindle microtubules. The centromere is a highly specialized chromatin region where sister chromatids are held together and the kinetochore is assembled during mitosis. At the intersection of the inter-kinetochore (inter-KT) 3 axis and the inter-sister chromatid axis is the inner centromere region. By acting as a platform to recruit various proteins, the inner centromere plays a key role in the regulation of sister chromatid cohesion and kinetochore-microtubule (KT-MT) attachments (1). Impaired integrity of inner centromeres causes chromosome missegregation, leading to chromosomal instability (CIN), which is a hallmark of cancer cells and may contribute to tumorigenesis (2). However, at the molecular level, how the functional inner centromere is assembled remains largely elusive.
Sgo1 is an important cohesin protector that predominantly localizes to centromeres in mitosis (11,(22)(23)(24). Sgo1 localizes to inner centromeres in a stepwise manner (Fig. 7) (25). First, through binding histone H2A, which is phosphorylated at thre-onine 120 (H2ApT120) by the outer kinetochore-localized kinase Bub1 (26 -29), Sgo1 is recruited to two KT-proximal centromere regions under the inner layer of kinetochores. In a second step, Sgo1 moves to inner centromeres, where it binds cohesin in a manner that is strongly enhanced by Cyclin-dependent kinase 1 (Cdk1) phosphorylation of Sgo1 (19,30). Previous studies have reported that Sgo1 collaborates with protein phosphatase 2A (PP2A) to antagonize phosphorylation of SA2 and Sororin, thereby preventing Wapl-dependent cohesin removal from chromosomes (13, 30 -34). Exogenous expression of a Sgo1 mutant defective in binding PP2A can hardly prevent premature sister chromatid separation induced by RNAi-mediated depletion of endogenous Sgo1 in human cells (32). However, it is controversial whether Bub1-dependent centromere localization of Sgo1 plays a role in maintaining centromeric cohesion in mammals (23,24,26,28,(35)(36)(37). Sgo1 also promotes centromeric accumulation of the chromosomal passenger complex (CPC), an integral part of the inner centromere and a key regulator of KT-MT attachments that consists of Aurora B kinase and the regulatory subunits Survivin, Borealin, and INCENP (38). Although it has been proposed that Sgo1 can bring the CPC to inner centromeres through direct interaction with Borealin that has been phosphorylated by Cdk1 (39), the mechanism by which Sgo1 targets CPC to centromeres remains incompletely addressed in mammalian cells (1,41).
In this study, we show that selective delocalization of Sgo1 from mitotic centromeres by disrupting the H2ApT120 -Sgo1 interaction results in loosened centromeric cohesion, which is accompanied by decreased centromeric localization of CPC and an elevated rate of chromosome missegregation. We further reveal the molecular mechanism by which Sgo1 and Haspin cooperate to allow proper assembly of the functional inner centromere and ensure high-fidelity chromosome segregation.

Loss of centromeric Sgo1 weakens cohesion at mitotic centromeres
Sgo1 directly binds H2ApT120 through a conserved SGO motif in its C-terminal region (27,29). Mutation of a conserved basic residue, lysine 492 to alanine (K492A), in its SGO motif prevents exogenously expressed Sgo1 from binding H2ApT120 and localizing to mitotic centromeres (26,32). To study the role of Sgo1 at centromeres, we set out to make the K492A mutation in endogenous Sgo1 by CRISPR/Cas9-mediated genome editing in HeLa cells (50). We obtained two clones, 3-9 and 3-1, in which the K492A mutation was confirmed by genomic DNA sequencing (Fig. 1A). Unless otherwise stated, we used clone 3-9 as the Sgo1-K492A mutant cells for the following studies.
Immunoblotting of lysates from mitotic cells arrested with the microtubule destabilizer nocodazole showed that the Sgo1-K492A mutant protein was expressed at a level comparable with that of WT Sgo1 in control HeLa cells (Fig. 1B). Immunofluorescence microscopy demonstrated that the Sgo1-K492A mutant failed to localize at mitotic centromeres, whereas H2ApT120 remained unaffected (Fig. 1C). Inspection of chromosome spreads prepared from nocodazole-arrested mitotic cells showed that the Sgo1-K492A mutant displayed diffuse signals on chromosome arms (Fig. S1A), likely because of its capability of binding cohesin (19,30). We also noticed that sister chromatids in the Sgo1-K492A mutant cells remained paired after 3-h treatment with nocodazole, indicating that loss of centromeric Sgo1 does not prevent the establishment of sister chromatid cohesion. Moreover, the percentage of cells with partly closed chromosome arms was higher in Sgo1-K492A cells than in control HeLa cells (Fig. 1D), which is in line with the impaired chromosome arm resolution in Bub1-depleted or -inhibited cells (23,24,51).
We next examined whether Sgo1-K492A cells have defects in sister chromatid cohesion. We found that Sgo1-K492A cells were strongly impaired in maintaining chromosome alignment on the metaphase plate during the sustained metaphase arrest induced by MG132 (Fig. 1, E and F), a proteasome inhibitor that prevents degradation of Cyclin B and Securin and therefore inhibits Separase activation. Inspection of chromosome spreads prepared from MG132-arrested mitotic cells revealed a strong increase in premature sister chromatid separation in Sgo1-K492A cells (Fig. 1, G and H). After 8-h treatment with MG132, the percentage of cells with cohesion loss increased from 1.2% in control HeLa cells to 23.2%-29.0% in Sgo1-K492A cells. Moreover, the percentage of cells with mild premature sister chromatid separation was also obviously higher in Sgo1-K492A cells (33.9%-42.0%) than in control HeLa cells (14.3%). These cohesion defects resemble an accelerated "cohesion fatigue" phenotype (52)(53)(54)(55).
We then measured the inter-KT distance on chromosome spreads prepared from cells that were arrested in mitosis with 3-h treatment with nocodazole. Interestingly, the inter-KT distances of mitotic chromosome spreads were at least 17.3% further apart in Sgo1-K492A cells than in control HeLa cells ( Fig.  1I and Fig. S1B), indicative of weakened centromeric cohesion. Thus, H2ApT120-mediated centromeric localization of Sgo1 is required for the maintenance of proper sister chromatid cohesion at mitotic centromeres.

Loss of centromeric Sgo1 leads to defective mitosis progression and chromosome congression
We next investigated the effect of loss of centromeric Sgo1 on cell proliferation and mitosis progression. We found that, under unperturbed conditions, the Sgo1-K492A mutation did not obviously affect cell proliferation (Fig. S2A). Interestingly, compared with that of control HeLa cells, the proliferation of Sgo1-K492A cells was more sensitive to clinically relevant low doses of paclitaxel, a microtubule poison widely used as a classic chemotherapy drug (56). This result is in line with a previous study showing that inhibition of Bub1 kinase activity by smallmolecule inhibitors caused remarkable impairment of chromo-Mechanism for inner centromere assembly some segregation and cell proliferation upon treatment with low doses of paclitaxel (51).
Time-lapse live imaging of cells stably expressing histone H2B fused to GFP (H2B-GFP) showed that, during unperturbed mitosis, the duration of mitosis in Sgo1-K492A cells (38.7 min, n ϭ 126) was only mildly longer than that in control HeLa cells (34.8 min, n ϭ 115). Interestingly, there were strong mitosis progression defects in Sgo1-K492A cells during the recovery from mitotic arrest induced by nocodazole treatment for 10 h (Fig. 2, A and B, and Movies S1-S4). Following chromosome biorientation, most (91%) control HeLa cells underwent anaphase onset at 62.7 Ϯ 3.2 min, on average, after nocodazole washout. In contrast, Sgo1-K492A cells showed strong mitotic arrest with complex chromosome behaviors that could be classified into two categories. Although 16.5% of Sgo1-K492A cells (type I) behaved like control HeLa cells, the remaining 83.5% of cells (type II) were defective in chromosome congression and underwent strikingly prolonged mitosis. These type II cells either died in mitosis or partitioned chromosomes into two or more masses and aberrantly exited mitosis without anaphase onset at 482.1 Ϯ 35.8 min, on average, after nocodazole washout.
We further monitored chromosome behavior when cells entered mitosis in the presence of MG132. We found that 3% and 18.2% of control HeLa cells and Sgo1-K492A cells were not able to achieve metaphase chromosome alignment, respectively (Fig. S2B). Moreover, among cells that were able to achieve metaphase chromosome alignment and remained alive, Sgo1-K492A cells began to exhibit irreversible chromosome scattering from the metaphase plate much earlier than control For clone 3-9, the genomic DNA PCR fragments were subcloned and sequenced. All 20 bacterial colonies showed the desired Sgo1-K492A mutation. For control HeLa cells and clone 3-1, the genomic DNA PCR fragments were sequenced directly. The sgRNA target DNA sequence preceding a 5Ј-NGG protospacer adjacent motif is shown. Multiple silent mutations were introduced into the repair template to prevent sgRNA targeting. B, HeLa cells and the indicated Sgo1-K492A mutant clones were treated with 0.33 M nocodazole for 12 h. Then mitotic cell lysates were immunoblotted with the indicated antibodies. C, HeLa cells and the indicated Sgo1-K492A clones were immunostained with the anti-human centromere autoantibody (ACA) and antibodies for Sgo1 and H2ApT120. D, HeLa and Sgo1-K492A cells were treated with 0.1 M nocodazole for 3 h. Mitotic chromosome spreads were stained with anti-human centromere autoantibody and DAPI. The chromosome morphology was classified and quantified in around 200 cells. Means and ranges are shown (n ϭ 2). E and F, HeLa and Sgo1-K492A clones were exposed to MG132, fixed at the indicated time points for CENP-C and DNA staining, and quantified in around 100 cells (E). Example images are shown (F). G and H, HeLa and Sgo1-K492A clones were exposed to MG132 for 8 h.

Mechanism for inner centromere assembly
HeLa cells (Fig. 2, C and D), which is in line with their centromeric cohesion defects. Thus, selective delocalization of Sgo1 from mitotic centromeres causes mitosis progression defects, particularly in achieving and maintaining chromosome alignment on the metaphase plate.

Loss of centromeric Sgo1 causes defects in correcting erroneous KT-MT attachments and accumulating CPC at mitotic centromeres
Inspection of paraformaldehyde (PFA)-fixed asynchronous Sgo1-K492A cells demonstrated an increased rate (8.3%-10.7%) of lagging chromosomes relative to control HeLa cells (3.7%) (Fig. 3, A and B). Lagging chromosomes are a hallmark of CIN, arising from persistent errors in KT-MT attachments (57). To determine whether Sgo1-K492A cells are defective in correcting KT-MT attachment errors, we performed S-trityl-L-cysteine (STLC) release assays (58). STLC is a kinesin-5/Eg5 inhibitor that prevents centrosome separation during mitotic entry, resulting in the formation of monopolar spindles with erroneously attached chromosomes (59). We treated cells with STLC to accumulate monopolar mitoses and then released them into MG132 to allow bipolar spindle formation and chromosome alignment. Examination of fixed cells showed that Sgo1-K492A cells were impaired in aligning chromosomes on the metaphase plate ( Fig. 3C and Fig. S3).
We further used live imaging to monitor chromosome alignment and segregation when cells were released from transient mitotic arrest induced by STLC treatment for 5 h. We found that most control HeLa cells underwent metaphase chromosome biorientation, followed by subsequent anaphase onset at 96.3 Ϯ 3.2 min, on average, after STLC washout. In contrast, 34.7% of Sgo1-K492A cells were defective in chromosome congression and underwent prolonged mitotic duration (Fig. 3, D and E, and Movies S5-S7), reminiscent of the "type II" cells observed upon release from nocodazole (Fig. 2, A and B). Moreover, upon STLC release, the percentage of anaphase cells with lagging chromosomes increased from 5.2% in HeLa cells to 9.9%-14.7% in Sgo1-K492A cells (Fig. 3F). These results suggest that Sgo1-K492A cells are defective in correcting erroneous KT-MT attachments.
Aurora B kinase accumulates at mitotic inner centromeres and plays an important role in promoting chromosome biorientation, mainly by phosphorylating kinetochore substrates to

Mechanism for inner centromere assembly
release improperly attached microtubules (60,61). Immunofluorescence microscopy demonstrated that, compared with control HeLa cells, Aurora B was less concentrated at centromeres in Sgo1-K492A cells and rather, displayed diffuse signals along the length of chromosomes (Fig. 3G). The ratio of the intensity of Aurora B versus CENP-C, a component protein of the constitutive centromere-associated network at inner kinetochores, was reduced by 33.8%-32.7% in Sgo1-K492A cells (Fig. 3H). By measuring the relative intensity of Aurora B staining at centromeres and on arms, we found that Aurora B was 40 -50% less enriched at centromeres in Sgo1-K492A cells (Fig. 3I). Thus, Sgo1-K492A cells are defective in accumulating Aurora B at mitotic centromeres, which might account for the impaired error correction efficiency.

The Sgo1-PP2A interaction is required to protect cohesion and localize CPC at mitotic centromeres
We exogenously expressed Sgo1 C-terminally fused to GFP (Sgo1-GFP) in Sgo1-K492A cells. As expected, Sgo1-GFP mainly localized to mitotic centromeres and largely restored proper inter-KT distance and centromeric localization of Aurora B, whereas the Sgo1-K492A-GFP mutant failed to do so (Fig. 4, A-C, and Fig. S4, A and B). These results validate the specificity of the centromere defects in Sgo1-K492A cells.
We next examined whether the interactions with cohesin and PP2A are important for Sgo1 function at mitotic centromeres. Previous studies showed that mutation of threonine 346 to alanine (T346A) in the cohesin-binding region (residues 313-353) does not affect the H2ApT120 -Sgo1 interaction but perturbs Sgo1 binding to the Scc1-SA2 interface and prevents Sgo1 from localizing to the inner centromere (19,26,30). Moreover, mutation of asparagine 61 to isoleucine (N61I) in the N-terminal coiled-coil region perturbs Sgo1 binding to PP2A and prevents Sgo1 from localizing to mitotic centromeres (32,62,63). To obtain equal levels of various Sgo1 proteins at the same location in the centromere region, we expressed Sgo1 as a fusion protein with the centromeric targeting domain of CENP-B (CB in short where necessary) (28,62), which binds a 17-bp CENP-B box motif within the ␣-satellite repeats of human centromeres (64 -66). As expected, we found that expression of CB-Sgo1-GFP restored the proper inter-KT dis-

Mechanism for inner centromere assembly
tance and centromeric localization of Aurora B in Sgo1-K492A cells (Fig. 4, D-F, and Fig. S4, C and D). Similar results were observed for CB-Sgo1-T346A-GFP as well as the CB-Sgo1-⌬313-353-GFP mutant lacking the cohesin-binding region. In contrast, CB-Sgo1-N61I-GFP was largely impaired in doing so. Similarly, CB-Sgo1-K492A-GFP, but not CB-Sgo1-N61I/ K492A-GFP, was able to support proper localization of CPC at centromeres (Fig. S4, E-H). Thus, when tethered to centromeres, the interaction with PP2A, but not H2ApT120 and cohesin, is required for Sgo1 to support proper cohesion and CPC at centromeres of Sgo1-K492A cells.

The Sgo1-PP2A interaction is required to concentrate H3pT3 at mitotic centromeres
We then investigated how the Sgo1-K492A mutation causes delocalization of CPC from mitotic centromeres. Immunofluorescence microscopy demonstrated that H3pT3 was enriched at the inner centromere in control HeLa cells, where it colocalized with Aurora B (Fig. 5, A-C), consistent with H3pT3 as the nucleosomal docking site for CPC (47)(48)(49). Interestingly, in Sgo1-K492A cells, the ratio of the intensity of H3pT3 versus CENP-C was reduced by 50%-60.8%, whereas that of H3pT3 at centromeres versus on arms was reduced by 62.8%-64.5%.

Mechanism for inner centromere assembly
Moreover, exogenous expression of Sgo1-GFP, but not Sgo1-K492A-GFP, restored centromeric H3pT3 in Sgo1-K492A cells (Fig. S5, A-C), validating the specificity of the phenotype. These data suggest that the defect in accumulating H3pT3 at mitotic centromeres might account for the reduced centromeric localization of Aurora B in Sgo1-K492A cells. Indeed, expression of Haspin as an enhanced GFP (EGFP) and CENP-B fusion protein (EGFP-CB-Haspin) in Sgo1-K492A cells effectively restored centromeric accumulation of H3pT3 (Fig. S5, D and E) and Aurora B (Fig. S5, F and G), as well as proper inter-KT distance (Fig. S5H).

The centromeric level of cohesin correlates with centromere accumulation of H3pT3 and CPC
Our results suggest that the centromeric cohesion defects in Sgo1-K492A cells might account for the reduced accumulation of H3pT3 and CPC at mitotic centromeres. To test this speculation, we sought to compromise the strength of centromeric cohesion by other approaches. Vertebrate cells express two paralogs of SA protein, SA1 and SA2 (68,69), which are redundantly required for sister chromatid cohesion and cell proliferation (70). A previous study showed that although SA1 is required for telomere and arm cohesion, SA2 is required for centromeric cohesion (71). In line with this, we recently showed that SA2 depletion by RNAi weakens centromeric cohesion, whereas sister chromatids remain not obviously compromised (44). We further found that centromeric accumulation of H3pT3 and Aurora B in SA2-depleted cells was significantly reduced (Fig. 6, A-D). Partial depletion of Scc1 by RNAi also strongly reduced centromeric H3pT3 in HeLa cells (Fig. S6,  A-C). Consistently, a previous study showed that accumulation of H3pT3 and Aurora B at the inner centromere is clearly reduced in mouse embryonic fibroblast cells prepared from Pds5B knockout mice (72). These results suggest a causal link between weakened centromeric cohesion and decreased H3pT3 and CPC at mitotic centromeres.
Conversely, we examined whether centromeric H3pT3 and CPC can be enhanced when cohesin is artificially tethered to centromeres. We found that expression of Scc1 as a CENP-B fusion protein (CB-Scc1-GFP) efficiently accumulated H3pT3 at the CENP-B loci in centromeres (Fig. 6, E-G), indicating recruitment of Haspin by Scc1. Moreover, CB-Scc1-GFP effectively recruited Aurora B to centromeres, presumably through H3pT3 (Fig. 6, H-J). We also noticed that tethering SA2 to centromeres as a CB-fusion protein (CB-SA2-GFP) only marginally increased centromeric H3pT3 (Fig. S6, D-F), suggesting that SA2 is not directly involved in the recruitment of Haspin. Taken together, these results demonstrate the requirement for centromeric cohesin in the enrichment of H3pT3 and CPC at mitotic centromeres.

Discussion
Structural and functional integrity of the inner centromere region is critical for coordination of sister chromatid cohesion and KT-MT attachment during mitosis. The role for Sgo1 in the assembly of inner centromere has been largely unclear (73), partly because of the massive loss of sister chromatid cohesion upon Sgo1 depletion by the commonly used RNAi technology (11,(22)(23)(24). In this study, we used CRISPR/Cas9-based genome editing to generate a mutant of endogenous Sgo1 in HeLa cells that cannot bind phosphorylated histone H2A, thereby disrupting its centromere localization. This mutant Sgo1 was employed as a tool to assess the function of centromeric Sgo1 for inner centromere assembly.
In contrast to the suggestion that Bub1-dependent localization of Sgo1 to centromeres during mitosis is not required to maintain cohesion (35), we find that loss of H2ApT120-dependent centromeric Sgo1 results in weakened centromeric cohesion. Our data support a direct role of Bub1 in protecting centromeric cohesion by generating H2ApT120 rather than by activating the spindle checkpoint (36). Our finding of the failure in supporting proper centromeric cohesion by centromere targeting of the PP2A-binding-deficient Sgo1 mutant is in line with the role of Sgo1-associated PP2A in cohesion protection (13, 30 -34). When exogenous Sgo1 is artificially tethered to centromeres, its interaction with cohesin appears dispensable for the maintenance of proper centromeric cohesion in cells in which endogenous Sgo1 is delocalized from centromeres. However, in the physiological state, the cohesin-bound pool of Sgo1 renders PP2A in close proximity to the cohesin complex at inner centromeres and would promote centromeric cohesion (19,26,29,30).
We did not observe a strong effect of loss of H2ApT120-dependent Sgo1 localization at centromeres on mitosis progression in an unperturbed situation, except for an increased frequency of chromosome missegregation. This seems to be in line with what was observed in HeLa cell treatment with the smallmolecule inhibitors of Bub1 kinase (51) as well as in a mouse mutant that lacks Bub1 kinase activity (28). Then, during unperturbed mitosis, why does loss of centromeric Sgo1 not show strong cohesion loss? During unperturbed mitosis, chromosome arm cohesion, which is mediated by the residual cohesin resistant to the prophase pathway of cohesin removal, persists throughout metaphase and is sufficient to maintain sister chromatid cohesion (9). Thus, local disassociation of cohesin from centromeres may not be sufficient to cause global cohesion loss in unperturbed mitosis as long as chromosome arm cohesion is not compromised. This may explain the difference in the severity of cohesion loss between Sgo1 deletion and centromeric Sgo1 delocalization. Strikingly, Sgo1-K492A cells are strongly defective in tolerating prolonged mitosis arrest. We reason that prolonged mitotic arrest allows time for further removal of cohesin from chromosome arms, rendering the Mechanism for inner centromere assembly strength of centromeric cohesion more critical to resist the sustained spindle pulling force and maintain sister chromatid cohesion until anaphase onset.
Although Sgo1 may interact with Borealin to directly target CPC to centromeres (39), our data indicate that, by protecting centromeric cohesin to provide a binding site for the histone H3 kinase Haspin, Sgo1 also indirectly positions CPC at the inner centromere to facilitate correction of erroneous KT-MT attachments. We previously showed that Haspin associates with the cohesin complex by binding Pds5B, thereby protecting centromeric cohesin to retain Sgo1 at inner centromeres (44). Taken together, we propose a model in which Bub1-mediated H2ApT120 enables centromeric localization of Sgo1, which not only protects centromeric cohesin, largely through PP2A-me-diated dephosphorylation of SA2 and Sororin (30,31), but also triggers a positive feedback loop in which Sgo1 promotes cohesin-mediated centromeric localization of Haspin to further enhance the strength of centromere cohesion and ensure Sgo1 localization at inner centromeres (Fig. 7). Thus, assembly of the functional inner centromere requires a positive feedback network involving Sgo1 and Haspin that accumulates cohesin and CPC at the centromere to guarantee precise chromosome segregation in mitosis. This study not only reveals the molecular mechanism by which two histone marks, H2ApT120 and H3pT3, cooperate to establish the inner centromere (47) but also provides important insight into the complexity of the centromere signaling network that coordinates various dynamic processes of mitosis (1).

Mechanism for inner centromere assembly
Defects that moderately impair chromosome segregation may allow cancer cells with CIN to become established (74,75). Compromised sister chromatid cohesion is proposed to be involved in CIN in cancer cells, but the underlying molecular mechanism is not clear (76). Our results suggest a causal link between weakened centromeric cohesion and reduced accumulation of H3pT3-CPC at mitotic centromeres. This link may account for the close correlation between compromised sister chromatid cohesion and increased chromosome segregation errors in cancer cells (40). Our data may also explain why SA2-deficient cells display decreased centromeric Aurora B, increased KT-MT attachment stability, and an elevated rate of chromosome missegregation (77). Recent studies of the cancer genome identified recurrent mutations in cohesin subunits and regulators in a wide range of human cancers (70, 78 -80), including the most frequently mutated cohesin subunit, SA2 (81,82). It will be interesting to investigate, in the future, whether the cohesin-dependent Haspin-H3pT3-CPC pathway is widely impaired in chromosomally instable cancer cells with sister chromatid cohesion defects.

Cell culture, plasmids, siRNAs, transfection, and drug treatments
All cells were cultured in Dulbecco's modified Eagle's medium supplemented with 1% penicillin/streptomycin and 10% fetal bovine serum (Gibco) and maintained at 37°C with 5% CO 2 . Cells stably expressing H2B-GFP were isolated and maintained in 3.0 and 2.0 g/ml blasticidin (Sigma), respectively. To mea-sure the effect of paclitaxel treatment on cell proliferation, ϳ4 ϫ 10 4 HeLa cells or Sgo1-K492A cells were plated on 6-well plates (Falcon) and cultured in DMSO or 1-4 nM paclitaxel for 7 days. Surviving cells were digested with trypsin (Life Technologies) and resuspended into culture medium, and then the number of surviving cells were analyzed using an automated cell counter (Thermo Fisher).
To make pBos-CENP-B-GFP, the H2B fragment in pBos-H2B-GFP (Clontech) was replaced with the KpnI/BamHI-digested PCR fragments encoding the centromere-targeting domain (residues 1-163) of CENP-B. The plasmid for Sgo1-GFP was constructed similarly. To make pBos-CB-Sgo1/Scc1/ PP2A_C␣-GFP constructs, the PCR fragments encoding Sgo1, Scc1, and PP2A_C␣ were subcloned into the BamHI site of pBos-CENP-B-GFP. To make pEGFP-CB-Haspin, the CENP-B fragment (residues 1-163) was also inserted into the BglII/Hin-dIII sites of pEGFP-Haspin. To make pMyc-PP2A_C␣, the PCR-amplified DNA fragments of PP2A_C␣ were first subcloned into the pDONR201 vector using Gateway Technology (Invitrogen), and then the corresponding fragment in the entry vector was transferred into a Gateway-compatible destination vector that harbors an N-terminal Myc tag. All point mutations were introduced with the QuikChange II XL site-directed mutagenesis kit (Agilent Technologies). All plasmids were sequenced to verify the desired mutations and absence of unintended mutations. The SA2 siRNA (5Ј-CCGAAUGAAUGGU-CAUCACdTdT-3Ј), Scc1 siRNAs (#1, 5Ј-AUACCUUCUUGC-AGACUGUdTdT-3Ј; #2, 5Ј-GCACUACUACUUCUAACCU-dTdT-3Ј), and control siRNA were ordered from RiboBio. Plas- Sgo1 is recruited to KT-proximal centromeres by Bub1-generated H2ApT120 and is then driven to inner centromeres where it binds cohesin. Largely through antagonizing the phosphorylation of cohesin and Sororin, Sgo1-bound PP2A protects centromeric cohesin to provide a binding site for Haspin, which not only feeds back to further protect cohesin and retain Sgo1 at the inner centromere but also recruits the CPC through H3pT3.

Mechanism for inner centromere assembly
mid and siRNA transfections were done with FuGENE 6 (Promega) and Oligofectamine or Lipofectamine RNAiMAX (Invitrogen), respectively. Cells were arrested in S phase or at the G 1 /S boundary by single or double thymidine (2 mM, Calbiochem) treatment, respectively, or in a prometaphase-like state with 0.1-3.3 M nocodazole (Selleckchem). Other drugs used in this study were STLC (5 M, Tocris Bioscience), MG132 (10 M, Sigma), and paclitaxel (MedChem Express). Mitotic cells were collected by selective detachment with "shake-off."

CRISPR/Cas9-mediated editing of the Sgo1 gene in HeLa cells
Single-guide RNA (sgRNA) for the human Sgo1 gene was ordered as oligonucleotides, annealed, and cloned into the dual Cas9 and sgRNA expression vector pX330 (Addgene, 42230) with BbsI sites. To make the K492A mutation in endogenous Sgo1, the pX330 plasmid encoding Cas9 and an sgRNA targeting a sequence (5Ј-TTACAGGAAACTGAGAAGAG-3Ј) close to Lys-492 of Sgo1 was cotransfected into HeLa cells with a single-stranded oligodeoxynucleotide as the homology-directed repair template. After 24-h incubation, the cells were treated with the DNA ligase IV inhibitor Scr7 (5 M, Selleckchem) for another 24 h to increase the efficiency of HDR-mediated genome editing. Then the cells were split individually to make a clonal cell line with selection using 1.0 g/ml puromycin for 3 days. Individual clones with an undetectable centromeric Sgo1 immunofluorescence signal were isolated. The genomic DNA fragments were PCR-amplified and sequenced to confirm the gene disruption (for clone 3-1). Alternatively, the PCR products were subcloned into pBluescriptII (Ϫ) with an EcoRV site and transformed into competent Escherichia coli cells (DH5␣), and then 20 positive bacterial colonies were sequenced (for clone [3][4][5][6][7][8][9]. The PCR primers were as follows: forward, 5Ј-ACACCACCTG-AAACTCAGCAGT-3Ј; reverse, 5Ј-AGGTTTAGGCAGCATA-AGAAATCG-3Ј. The sgRNA-resistant single-stranded oligodeoxynucleotide with the K492A mutation was ordered from Integrated DNA Technologies (5Ј-AATTGGTGTGTTTTA-CCATAACTTGGTAGGGAAGAGTAAGTTAATATTGGG-ATGCTTACATTATGCCTGAGATCTCTTTTTACTCTTAC-AGGgcACTccGgAGgGGaGACCCTTTTACAGATTTGTG-TTTTTTGAATTCTCCTATTTTCAAGCAGAAAAAGGA-TTTGAGACGTTCTAAAAAAAGTATGAA-3Ј). Multiple silent mutations (shown in lowercase) in the sgRNA target sequence and the protospacer-adjacent motif were introduced into the repair template to prevent sgRNA targeting.

Fluorescence microscopy, time-lapse live-cell imaging, and statistical analysis
Cells cultured on coverslips were fixed with 2% PFA in PBS for 10 min, followed by extraction with 0.5% Triton X-100 in PBS for 5 min, or fixed with 2% PFA for 10 min and then extracted with 1% Triton X-100 for 10 min. To produce chromosome spreads, mitotic cells obtained by selective detachment were incubated in 75 mM KCl for 10 min. After attachment to glass coverslips by Cytospin (Cytospin 4, Thermo Scientific) at 1500 rpm for 5 min, chromosome spreads were fixed with 2% PFA in PBS for 10 min, followed by extraction with 0.5% Triton X-100 in PBS for 5 min, or pre-extracted with 0.3% Triton X-100 in PBS for 5 min, followed by fixation with 4% PFA in PBS for 20 min. Fixed cells and chromosome spreads were stained with primary antibodies for 1-2 h and secondary antibodies for 1 h, all with 3% BSA in PBS with 0.5% Triton X-100 and at room temperature. DNA was stained for 10 min with DAPI. Fluorescence microscopy was carried out at room temperature using a Nikon Eclipse Ni microscope with a Plan Apo Fluor ϫ60 oil (numerical aperture 1.4) objective lens and a Clara charge-coupled device (Andor Technology). The inter-KT distance was measured using the inner kinetochore marker CENP-C on over 20 kinetochores per cell in at least 20 cells. Distance was determined by drawing a line from the outer kinetochore extending to the outer edge of its sister kinetochore. The length of the line was calculated using the imaging software of NIS-Elements BR (Nikon). Quantification of fluorescent intensity was carried out with ImageJ (National Institutes of Health) using images obtained with identical illumination settings. Briefly, on chromosome spreads, the average pixel intensity of H3pT3, Aurora B, or CENP-C staining at centromeres, defined as circular regions including paired centromeres, or on chromosome arms (except for CENP-C) was determined using ImageJ. After background correction, the ratio of centromeric H3pT3/CENP-C, centromeric Aurora B/CENP-C, centromeric H3pT3/arm H3pT3, or centromere Aurora B/arm Aurora B intensity was calculated for each centromere. Time-lapse live-cell imaging was carried out with the GE DV Elite Applied Precision DeltaVision system (GE Healthcare) equipped with Olympus oil objectives of ϫ40 (NA 1.35) UApo/340 and an API Custom Scientific CMOS camera and Resolve3D softWoRx imaging software. Cells expressing H2B-GFP were plated in four-chamber glass-bottomed 35-mm dishes (Cellvis) coated with poly-D-lysine and filmed in a climate-controlled and humidified environment (37°C and 5% CO 2 ). Images were captured every 2 min (Movies S1-S6; Fig. 2,

Mechanism for inner centromere assembly
A and B, and Fig. 3, D and E) or every 5 min (Fig. 2, C and D). The acquired images were processed using Adobe Photoshop and Adobe Illustrator. Statistical analyses were performed with a two-tailed unpaired Student's t test in GraphPad Prism 6. A p value of less than 0.05 was considered significant.