GSK3-mediated CLASP2 phosphorylation modulates kinetochore dynamics

Kinetochores (KTs) are large protein complexes that connect mitotic chromosomes to spindle microtubules (MTs) and are essential for chromosome alignment and segregation. Despite enormous progress in revealing KT composition and structure, how KT–MT interactions remain dynamic throughout mitosis to enable chromosome movement and correction of attachment errors, which can result in genetic instability, remains incompletely understood (Godek et al., 2015; Heald and Khodjakov, 2015).

KT–MT attachment is predominantly mediated by the NDC80 complex that forms a dynamic and disordered MT-binding interface (Zaytsev et al., 2014). NDC80 is inhibited by Aurora-B-mediated multisite phosphorylation (DeLuca et al., 2011; Zaytsev et al., 2015), and a concept has emerged that shifting of the NDC80 MT-binding interface in and out of spatially separated zones dominated by either kinase or phosphatase activities underlies tension-induced hardening of KT–MT interactions (Liu et al., 2010; Welburn et al., 2010). Although the NDC80 complex is required for stable KT–MT attachment and force generation (DeLuca et al., 2005), other MT-binding activities localize to KTs (Godek et al., 2015), but the contribution of these proteins to KT–MT interactions remains unclear.

CLASPs are amongst the outermost KT proteins (Wan et al., 2009) that bind to CENP-E through their C-terminal domain (Maffini et al., 2009) and associate along MTs directly and to growing MT plus ends through interactions with EB1 (Mimori-Kiyosue et al., 2005; Wittmann and Waterman-Storer, 2005). CLASPs are required for mitosis (Maiato et al., 2003; Mimori-Kiyosue et al., 2006; Pereira et al., 2006) and may promote MT polymerization at the KT (Maiato et al., 2005; Young et al., 2014). The activity of CLASPs is controlled by phosphorylation predominantly in two intrinsically disordered regions. Phosphorylation by Cdk1 and Plk1 near to the KT-binding domain (Fig. 1A) contributes to normal chromosome dynamics and may stabilize KT–MT interactions, although the precise molecular mechanism remains elusive (Maia et al., 2012). In addition, we have previously identified multisite phosphorylation by Cdk1 and GSK3 (of which there are two isoforms encoded by GSK3A and GSK3B) near to the central SxIP EB1-binding motifs (where x indicates any amino acid) (Kumar et al., 2009). Although a mitotic role for GSK3 in chromosome segregation has been proposed (Tighe et al., 2007; Wakefield et al., 2003), no mitosis-specific GSK3 substrates have been identified. Here, we demonstrate that Cdk1 and GSK3-dependent phosphorylation control CLASP2 MT-binding throughout the cell cycle, and we report the effects of GSK3 phosphorylation-site mutants on KT and chromosome dynamics that are consistent with local KT-associated activation of CLASP–MT binding.

Phosphorylation by Cdk1 and GSK3 strongly inhibits binding of an N-terminally truncated CLASP2 [comprising residues 497-1515; CLASP2(497-1515)] construct to MTs (Kumar et al., 2012). This raises a paradox of how CLASP2 can contribute to KT–MT interactions if the CLASP2 MT-binding activity is globally turned off during mitosis. CLASP2(497-1515) contains the EB1-binding SxIP motifs and a domain that we have identified as being required for direct CLASP2–MT binding (Fig. 1A) (Wittmann and Waterman-Storer, 2005). Recent X-ray crystallography confirms that this domain (TOG3) has a TOG-like fold that is characteristic for tubulin interactions (Maki et al., 2015). Indeed, mutation of positively charged residues in TOG3 surface loops specifically abolished binding along MTs without inhibiting plus-end tracking (Fig. 1C,D). This indicates that, in addition to SxIP-motif-mediated interactions with EB1 on growing MT ends, TOG3 contributes to direct CLASP2–MT interactions in interphase cells. However, the full-length isoforms CLASP2γ and CLASP2α, generated by alternative splicing, contain additional N-terminal TOG domains (Fig. 1A), but how TOG1 and TOG2 in mammalian CLASP2 contribute to MT binding is unclear (Patel et al., 2012). Therefore, to test whether MT association of the longest CLASP2α isoform is cell cycle regulated, we measured EGFP–CLASP2α on MT ends as a function of cell cycle phases (Fig. 2A,C). Similar to truncated CLASP2(497-1515), EGFP–CLASP2α binding to MT ends decreased after nuclear envelope breakdown, was lowest in metaphase, and increased again during anaphase, probably as a result of Cdk1 inactivation. A raise above interphase levels during telophase further suggests an excess of phosphatase activity during exit from mitosis.

CLASP2 is phosphorylated in two intrinsically disordered regions: surrounding the SxIP EB1-interaction motifs and adjacent to the KT-binding domain (Fig. 1A). Compared with shorter CLASP2 constructs, the GSK3-mediated upshift of the ∼170 kDa endogenous full-length CLASP2α in gels is difficult to discern (Fig. 1B) (Kumar et al., 2009). In contrast, in metaphase-arrested cells, a larger upshift of CLASP2α in gels indicated substantial additional phosphorylation at other non-GSK3 sites during mitosis, as previously reported (Maia et al., 2012). Thus, to test whether mitotic CLASP2α MT-binding inhibition was due to GSK3 multisite phosphorylation at the previously identified sites, we compared the MT-end-association of non-phosphorylatable or phosphomimetic versions of full-length or truncated CLASP2 in metaphase-arrested cells (Fig. 2B,D). Non-phosphorylatable EGFP–CLASP2α(9×S/A) (comprising mutation of nine serine residues to alanine) was enriched several-fold on MT ends in metaphase. In contrast, the difference between wild-type and phosphomimetic EGFP–CLASP2α(8×S/D) (comprising mutation of eight serine residues to aspartic acid) was small and not significant. EGFP–CLASP2(497-1515) constructs behaved in a similar manner to full-length CLASP2α, except that wild-type EGFP–CLASP2(497-1515) showed slightly decreased MT binding, indicating that the N-terminal TOG domains only contribute to a small extent to CLASP2 binding to MT ends in mitotic cells. These data demonstrate that both direct and indirect MT interactions of all CLASP2 isoforms are greatly reduced in metaphase and are predominantly inhibited by GSK3-mediated phosphorylation in the central SxIP motif domain.

Because the KT-binding domain at the CLASP2 C-terminus is far from the central GSK3-phosphorylated region (Fig. 1A), we next tested whether GSK3 phosphorylation influences CLASP2 binding to KTs. Interestingly, compared with wild-type, non-phosphorylatable EGFP–CLASP2α(9×S/A) was enriched several-fold at KTs (Fig. 3A,B). However, nocodazole-induced MT depolymerization completely eliminated this enrichment. We therefore concluded that this increase reflects increased binding of EGFP–CLASP2α(9×S/A) to growing MT ends at KTs rather than direct KT binding itself.

Consistent with an enrichment at growing KT–MTs, EGFP–CLASP2α(9×S/A) fluorescence was often increased at one sister KT and not the other (Fig. 3A). During sister KT oscillations, only the KT moving away from the pole (anti-poleward) is expected to associate with polymerizing MT ends, and thus bind to increased amounts of +TIPs. To test if EGFP–CLASP2α(9×S/A) indeed identified polymerizing KT MTs, we analyzed time-lapse sequences of mitotic cells expressing EGFP–CLASP2α constructs and mCherry–CENPA to mark KTs. Indeed, following the movements of individual KT pairs, EGFP–CLASP2α(9×S/A) was highly enriched at anti-poleward-moving KTs and switched to the opposite KT during direction reversal (Fig. 3C), indicating a high level of coordination of the MT polymerization state with KT movement. However, wild-type EGFP–CLASP2α did not appear to be enriched at the polymerizing KT MT bundle. We therefore quantified the relative enrichment of EGFP–CLASP2α by dividing the fluorescence intensity of the brighter KT by the dimmer one and normalizing to the mCherry–CENPA signal to correct for KT movements in and out of focus (Fig. 3D). As expected, this sister KT intensity ratio was high for EGFP–CLASP2α(9×S/A) (Fig. 3E). In contrast, the intensity ratio of KT-bound wild-type EGFP–CLASP2α or the phosphomimetic 8×S/D variant was close to one and not significantly different. This confirms that the amount of KT-bound CLASP2α is independent of the polymerization state of the KT–MT bundle (Pereira et al., 2006).

In order to contribute to KT–MT interactions, we thus hypothesized that CLASP2α MT binding ought to be locally activated at KTs, presumably through local KT-associated dephosphorylation. Unfortunately, neither published (Watanabe et al., 2009) nor our own antibodies against the CLASP2 GSK3-phosphorylated motif S(p)QGCS(p)REAS(p) were sufficiently specific to visualize the local CLASP2 phosphorylation state by immunofluorescence. Instead, we tested the ensuing prediction that phosphorylation-deficient CLASP2α mutants act as dominant negatives by occupying KT binding sites and disrupting local control of CLASP2 binding to MTs at the KT.

As a readout for tension between KTs and a proxy for the strength of KT–MT interactions (Waters et al., 1996), we measured the distance between mCherry–CENPA-labeled sister KTs (ΔKT; Fig. 3D) at sub-resolution accuracy in cells expressing EGFP–CLASP2α phosphorylation-site mutants. In control cells, the average ΔKT was 1.13±0.14 µm (mean±s.d.) (Fig. 4A). Pharmacological inhibition of MT dynamics led to an expected relaxation of tension across KT pairs and a resulting ∼25–30% ΔKT decrease (nocodazole, 0.86±0.09 µm; taxol, 0.79±0.1 µm) (Magidson et al., 2016; Maresca and Salmon, 2009). Cells expressing wild-type EGFP–CLASP2α were indistinguishable from controls (1.13±0.14 µm). In contrast, in cells expressing the phosphomimetic 8×S/D variant, ΔKT was significantly reduced by ∼10% to 1.05±0.17 µm. This is similar to the ΔKT decrease in CLASP1 CLASP2 double-knockdown cells (Mimori-Kiyosue et al., 2006) and consistent with the idea that non-phosphorylated CLASP2 partially contributes to KT–MT interactions and resulting tension across the KT pair. In contrast, a subtle increase in ΔKT in EGFP–CLASP2α(9×S/A)-expressing cells (1.17±0.21 µm) was not statistically significant. However, the distribution of ΔKT values in EGFP–CLASP2α(9×S/A)-expressing cells was skewed toward larger values (Fig. 4B), and the variance of the EGFP–CLASP2α(9×S/A) population was statistically significantly increased compared with that of wild type (P<0.0001, Brown–Forsythe test), unlike the variance of the EGFP–CLASP2α(8×S/D) population, which was not different (P=0.28). Taken together, these data indicate that although CLASP2α in the cytoplasm is hyperphosphorylated during metaphase, a population of KT-bound CLASP2α is likely to be locally dephosphorylated and contributes measurably to KT–KT tension. These findings are also consistent with a GSK3-inhibitor-induced ΔKT increase (Tighe et al., 2007) and suggest that CLASP2 is an important KT-associated GSK3 substrate. In addition, the large outlier ΔKT values in cells that expressed the non-phosphorylatable mutant suggest that there are instances at which KT-bound CLASP2 binding to MTs must be reduced as part of normal metaphase KT dynamics.

To test how the CLASP2 phosphorylation state influences KT dynamics, we analyzed time-lapse sequences of metaphase cells expressing mCherry–CENPA. In control cells, KT pairs oscillated back and forth while retaining a roughly perpendicular orientation to the metaphase plate, indicating a balanced attachment of the sister KTs to either spindle pole (Fig. 4C). However, in cells expressing CLASP2α phosphorylation-site mutants, we occasionally observed highly unusual KT dynamics. In EGFP–CLASP2α(8×S/D) cells, KT pairs sometimes drifted out of the metaphase plate, accompanied by a substantial decrease in ΔKT and a loss of perpendicular orientation relative to the metaphase plate, consistent with temporary loss of MT attachment to one of the sister KTs (Fig. 4C,D). In contrast, in EGFP–CLASP2α(9×S/A) cells, KT pairs sometimes appeared to be pulled out of the metaphase plate, stretching KT pairs to generate almost twice the average ΔKT in control cells (Fig. 4C,D). This is consistent with our prediction that CLASP2 phosphorylation by GSK3 reduces KT–MT interactions. In ∼25 cells examined per condition, we estimate a ∼threefold increase in these abnormal KT dynamics compared with controls. However, these events were too infrequent to unambiguously establish that these correlated with the expression of CLASP2α phosphorylation-site mutants, and we were unable to reliably detect such anomalous KT behavior by computer-assisted tracking of the entire KT population (Jaqaman et al., 2010). Even if these unusual KT dynamics are not correlated with CLASP2 MT-binding activity, it is important to note that such non-standard KT behavior occurs in metaphase cells.

Because even small errors in KT–MT attachment can result in chromosome segregation defects and aneuploidy (Thompson and Compton, 2011), we tested how expression of EGFP–CLASP2α phosphorylation-site mutants affect mitotic fidelity. To increase the percentage of cells in anaphase, we released cells from a G2–M block, and fixed and counted cells that exhibited lagging chromosomes in anaphase or chromosome bridges in telophase (Fig. 5A). Expression of either phosphorylation-site mutant notably increased these segregation defects, while expression of wild-type EGFP–CLASP2α had no effect. Similarly, only wild-type EGFP–CLASP2α partially rescued mitotic phenotypes resulting from expression of shRNA against CLASP2, but phosphorylation site mutants did not (Fig. 5B).

In this paper, we examine the consequences of GSK3-mediated phosphorylation in the MT-binding domain of CLASP2 during mitosis. We show that the MT-binding activity of all CLASP2 isoforms is inhibited globally during metaphase, and we demonstrate the consequences of CLASP2 phosphorylation-site mutants on KT–KT tension and on the fidelity of chromosome segregation. However, neither the GSK3-mediated CLASP2 phosphorylation state nor the polymerization state of the KT–MT bundle appeared to affect CLASP2 localization to KTs. We find that only the non-phosphorylatable CLASP2α mutant was strongly enriched at growing KT MT bundles. This is similar to what has been observed for EB1 in Ptk1 cells (Tirnauer et al., 2002), and it is thus likely that this enrichment is due to increased binding to EB1. This is further consistent with the increased MT-end tracking of non-phosphorylatable CLASP2α during metaphase and indicates dephosphorylation of wild-type CLASP2α near to the KT. Of note, even though the KT area increased in nocodazole-treated cells, in contrast to previous reports (Pereira et al., 2006), we were unable to detect an increase of the amount of KT-bound EGFP–CLASP2α per unit area when MTs were depolymerized, which could be due to differences in methodology. We measured EGFP–CLASP2α fluorescence enrichment on KTs compared to the signal in the cytoplasm in live cells, and not by immunofluorescence in which a substantial amount of cytoplasmic EGFP–CLASP2α may be extracted during fixation. Unfortunately, our KT dynamics data rely on the expression of dominant-negative CLASP2α GSK3 phosphorylation-site mutants as we did not succeed in stably replacing endogenous CLASP2 activity. Possibly because of the presence of endogenous CLASP2, the overall effect on KT dynamics was small and, because it remains technically challenging to detect and quantify abnormal KT dynamics in mammalian cells, we were unable to unambiguously dissect how GSK3-mediated CLASP2 phosphorylation contributes to correct chromosome segregation. However, because the predominant MT-binding activity of KTs resides within the NDC80 complex, it is unsurprising that KT-dynamics defects resulting from disruption of CLASP2 phosphoregulation were subtle, and the precise mechanism by which GSK3-mediated regulation of CLASP2 MT-binding contributes to the robustness of error-free chromosome segregation remains to be determined.

Nevertheless, our data support a mitotic function of these GSK3 phosphorylation sites at KTs and further support a model in which GSK3-mediated phosphoregulation of CLASP2 interactions with MT plus ends contributes to the fine-tuning of the KT–MT interaction. We propose that CLASP2 is globally hyperphosphorylated during mitosis by GSK3 (Kumar et al., 2012), which inhibits its MT-binding activity. CLASP2-mediated KT–MT interactions therefore imply a local activation of CLASP2–MT binding that we believe could result from local dephosphorylation of GSK3 sites of only KT-bound CLASP2 molecules (Fig. 6). This proposed local activation of CLASP2–MT binding by KT-associated dephosphorylation is conceptually similar to the tension-regulated phosphorylation of the MT-binding state of the NDC80 complex (Welburn et al., 2010; Zaytsev et al., 2015). However, CLASP2 localization to the outer KT places it in the zone of proposed high phosphatase activity (Funabiki and Wynne, 2013) that is possibly independent of tension-mediated intra-KT structural rearrangements (Welburn et al., 2010). Although our data are consistent with this idea and we see very little CLASP2α localization to spindle MTs in metaphase, it should be noted that we cannot rule out other indirect effects of altered CLASP2–MT interactions. For example, stabilization of anti-parallel MT bundles could increase outward pushing forces and increase tension between sister KTs without directly altering KT–MT interactions (Liu et al., 2009). In addition to simply contributing to the KT–MT linkage, KT-bound active CLASP2 is also likely to promote polymerization of KT MTs (Funk et al., 2014; Maiato et al., 2005), which could indirectly alter KT-pair dynamics. Non-phosphorylatable CLASP2 could result in chromosome alignment defects by aberrantly pushing KT pairs out of the metaphase plate. Finally, how GSK3-mediated control of CLASP2–MT binding is integrated with Plk1-mediated CLASP2 phosphorylation that stabilizes and fine-tunes KT–MT attachment remains to be determined (Maia et al., 2012).

H2B–mCherry or CENPA–mCherry pLenti6/V5-DEST (Invitrogen) vectors were generated by subcloning from Addgene plasmid #21044 or pEGFP–CENPA (from Alexey Khodjakov, Wadsworth Center, New York State Department of Health, Albany, NY). CLASP2α was subcloned from pEGFP-CLASP2α (from Irina Kaverina, Vanderbilt University Medical Center, Nashville, TN) into pAd/CMV/V5-DEST (Invitrogen). Wild-type and non-phosphorylatable 9×S/A CLASP2(497-1515) constructs have been described previously (Kumar et al., 2012). CLASP2(8×S/D) and mutated CLASP2α constructs were generated by swapping fragments from XbaI/BamH1 digests.

HaCaT cells were maintained, and lenti- and adenovirus particles produced and utilized as described previously (Stehbens et al., 2014). Cells were infected with lentiviral non-targeting shRNAs or shRNAs against CLASP2 and selected for 3 days with puromycin (Invitrogen) starting 48 h after infection. For immunoblotting, cells were lysed in a protease, kinase and phosphatase inhibitor buffer (Kumar et al., 2009). Rat anti-CLASP2 antibodies were from Absea Biotechnology Ltd. (KT68, 1:1000). Secondary horseradish peroxidase (HRP)-conjugated antibodies were from Jackson ImmunoResearch Laboratories. Cells were synchronized by incubation with 9 µM RO-3306 (EMD BioScience) for 16 h, washed three times with PBS, fixed in 4% paraformaldehyde after 90-120 min and stained with 500 nM propidium iodide (Invitrogen, P3566).

Spinning disk confocal microscopy in sealed imaging chambers was as described previously (Ettinger and Wittmann, 2014; Stehbens et al., 2012). For fast KT dynamics, an iXon EMCCD camera (Andor) and increased disk speed (3500 rpm) were used to achieve sufficient frame rates for multiple z-slices.

For all fluorescence measurements of different fluorescently tagged CLASP2 constructs, cells were selected that fell within the same range of fluorescence intensity and thus expression levels. To quantify MT-bound EGFP–CLASP2, fluorescence intensity was measured in a 3-pixel-wide intensity profile perpendicular to and near the MT plus end. After subtraction of the camera offset, this was fitted with a Gaussian function and total intensity integrated around the mean±3 s.d., which encompasses 99.7% of the integrated area, and then normalized to the expression level by dividing by the local cytoplasm intensity defined as the y-offset of the fitted Gaussian function. Similarly, the sister KT ratio was calculated by fitting a 3-pixel-wide intensity profile with a double Gaussian function, dividing the integrated intensity of the brighter KT by the dimmer one, and normalized for the CENPA–mCherry ratio to correct for focus differences. The relative amount of KT-bound EGFP–CLASP2 was measured in a small region of interest on the KT (I_KT) and a local matching region in the cytoplasm (I_Cyto), as well as dark background outside the cell (I_Dark), and ratios were calculated as: (I_KT−I_Dark)/(I_Cyto−I_Dark).

Statistical analysis used Excel (Microsoft) and the Analyse-it add-in (Analyse-it Software, Ltd). Box plots show median (lines), first and third quartile (boxes), observations within 1.5× the interquartile range (whiskers), outliers or all data points (dots), and 95% confidence intervals (notches). P-values were calculated using the Tukey–Kramer honest significant difference (HSD) test. Least square curve fitting was performed using the Solver function in Excel.

We thank Alexey Khodjakov and Irina Kaverina for gifts of reagents.