The histone methyltransferase NSD3 contributes to sister chromatid cohesion and to cohesin loading at mitotic exit

To ensure that replicated DNA is correctly transmitted to the daughter cells during mitosis, sister chromatids need to be held together until all of the chromosomes are correctly bi-oriented toward the opposite spindle poles. Cohesin is a multisubunit protein complex with a ring-like structure, and it can topologically link two chromatin fibers (Nasmyth and Haering, 2009). Throughout the cell cycle, the complex contributes to the dynamic regulation of genome organization, transcription, DNA damage repair and sister chromatid cohesion. The core of the complex is made up of the three core subunits SMC1A, SMC3 and RAD21, and these are further bound by regulatory factors such as the Scc3 homologues SA-1 and SA-2 (also known as STAG1 and STAG2) (Makrantoni and Marston, 2018; Nishiyama, 2019). Cohesin is loaded onto chromatin during exit from mitosis by the kollerin complex (composed of NIPBL and MAU2) (Ciosk et al., 2000; Gillespie and Hirano, 2004; Litwin and Wysocki, 2018; Takahashi et al., 2004; Tonkin et al., 2004; Watrin et al., 2006). Nipped-B-like protein (NIPBL, also known as SCC2) is responsible for cohesin loading, whereas MAU2 sister chromatid cohesion factor (MAU2, also known as SCC4) facilitates the binding of that protein onto chromatin (Chao et al., 2015; Hinshaw et al., 2015; Murayama and Uhlmann, 2014; Parenti et al., 2020). Before DNA replication, cohesin and chromatin association is dynamic, and it is actively removed by the WAPL cohesin release factor (WAPL) aided by scaffolding proteins PDS5A and PDS5B (Chan et al., 2012; Gerlich et al., 2006; Kueng et al., 2006; Lopez-Serra et al., 2013; Ouyang et al., 2013, 2016). Afterwards, during DNA replication, sister chromatid cohesion is established when a fraction of cohesin stably associates with chromatin owing to the acetylation of SMC3 by the ESCO1 and ESCO2 acetyltransferases (Alomer et al., 2017; Ben-Shahar et al., 2008; Hou and Zou, 2005; Rowland et al., 2009; Ünal et al., 2008; Zhang et al., 2008). This process is accompanied by a specific cohesion-loading mechanism wherein kollerin binds to a phosphorylated form of the MCM2–MCM7 pre-replication complex (Zheng et al., 2018). Acetylated SMC3 allows for the subsequent binding of sororin (also known as cell division cycle associated 5; CDCA5) to cohesin complexes, where it antagonizes WAPL anti-cohesive activity until mitosis begins (Nishiyama et al., 2010; Ouyang et al., 2016; Rankin et al., 2005). Upon mitotic entry in human cells, the mitotic kinases CDK1, PLK1 and Aurora B phosphorylate sororin and the cohesin components. This makes cohesin sensitive to the activity of WAPL, and in a process known as the prophase pathway (Nishiyama et al., 2013; Sumara et al., 2002), the cohesin molecules disassociate from chromosomes arms. During this process, sister chromatid cohesion is protected from WAPL-mediated dissociation at the centromere by the proteins shugoshin 1 (SGO1) and the histone H3 associated protein kinase (HASPIN). SGO1 competes with WAPL for binding to the cohesin ring, whereas the SGO1 partner PP2A, the holoenzyme protein phosphatase 2A, is thought to counterbalance any phosphorylation of cohesin or sororin (Kitajima et al., 2006; Liu et al., 2013b). At the same time, HASPIN binds to PDS5 and prevents cohesin from interacting with WAPL (Liang et al., 2018; Zhou et al., 2017). Once all kinetochores are properly attached to microtubules during metaphase, the spindle assembly checkpoint (SAC) is turned off, and this leads to the activation of the endoproteinase separase that cleaves the RAD21 cohesin subunit, thereby causing the cohesin ring to open and the two sets of chromosomes to segregate (London and Biggins, 2014). Of note, in the case of prolonged mitotic arrest with centromeres under spindle tension, chromosomes are subjected to an unscheduled dissociation of the sister chromatids, a process referred to as cohesion fatigue (Daum et al., 2011; Stevens et al., 2011).

In a previous work, we reported the presence of dimethylated H3K4 at the centromere was correlated with the prevention of premature sister chromatid separation (PSCS) during mitosis (Eot-Houllier et al., 2008). In determining whether histone-lysine N-methyltransferases are involved in sister chromatid cohesion, we identified the possible contribution of the H3K36-specific methyltransferase NSD3 to this regulation. NSD3 belongs to the nuclear receptor-binding SET domain protein subfamily, a group with three members: NSD1, NSD2, also known as Wolf-Hirschhorn syndrome candidate 1 (WHSC1), and NSD3, also known as WHSC1-like 1 (WHSC1L1). NSD methyltransferases act as oncoproteins in different types of cancers (Han et al., 2018; Lucio-Eterovic and Carpenter, 2011), and they are considered to be specific for mono- or di-methylation of H3K36 (Li et al., 2021, 2009; Rahman et al., 2011; Wagner and Carpenter, 2012; Yuan et al., 2021), an epigenetic mark involved in gene transcription, RNA alternative splicing, and DNA replication, repair and methylation. However, the contribution of NSD3 to this methylation seems to be lower than that of NSD2 (Linares-Saldana et al., 2021). In addition to their SET domains, NSD family members are characterized by the presence of seven domains that bind to modified histones. Five of these are plant homeodomain (PHD) domains, which recognize specific DNA sequences together with histone post-translational modifications (methylated lysine or arginine, and acetylated lysine), and the other two are proline- and tryptophan-rich (PWWP) domains, which bind methylated lysine residues (Angrand et al., 2001; Bennett et al., 2017; Kim et al., 2006; Vermeulen et al., 2010; Wu et al., 2011; Zhang et al., 2021).

In somatic cells, two alternative NSD3 mRNAs, composed of 24 and 10 exons, lead to the expression of two isoforms, NSD3-long (NSD3-L) and NSD3-short (NSD3-s), respectively (Angrand et al., 2001; Kim et al., 2006; Zhou et al., 2010), whereas a third isoform WHISTLE (found in testes) is specifically expressed from a downstream promoter (Kim et al., 2006). The long isoform NSD3-L contains 1437 amino acids (aa), and the SET methyltransferase domain is located in its C-terminus. The short isoform NSD3-s (647 aa) is translated from an alternatively spliced mRNA and, because it lacks the SET domain, it also lacks methyltransferase activity. NSD3-s also differs NSD3-L at its C-terminal (aa 620–647), and in having a single PWWP domain. That single NSD3-s PWWP domain is required for the function of BRD4–NSD3-s–CHD8 complex to sustain leukemia cell proliferation (Shen et al., 2015).

Here, we report that inactivation of the methyltransferase SET family member NSD3 results in PSCS during early mitosis. We show that NSD3 is required for cohesion in G2 phase before mitotic entry, and that it interacts with kollerin and improves cohesin loading at mitotic exit. We also describe that NSD3 loads onto chromatin in early anaphase, prior to kollerin recruitment. Finally, we also demonstrate that the role of NSD3 in cohesin loading is mediated by the NSD3-L long isoform, and that an active SET methyltransferase domain is required.

In order to identify new methylation effectors involved in regulating sister chromatid cohesion, we performed an inactivation screen with RNA interference (RNAi) to identify defective mitotic cohesion in 14 human SET domain-containing methyltransferases. For each methyltransferase, three different siRNAs were transfected in HeLa Kyoto cells. We prepared mitotic chromosome spreads, and analyzed them for mitotic cohesion defects (Fig. 1A). NSD3 was the only methyltransferase tested that produced a significantly increased proportion of cells displaying separated sister chromatids during prometaphase with all three tested siRNA (Fig. 1B; Fig. S1). The siRNAs used for the screen, 1, 2 and 3 targeted NSD3 exons 5, 6, and 2, respectively. As expected, given that the short isoform of NSD3 results from an alternative splicing event between exons 9 and 10, these siRNAs decreased expression of both isoforms (Fig. 1C). This screen therefore revealed that depleting NSD3 causes defective sister chromatid cohesion, and that this is detectable in mitosis.

As NSD3 is involved in gene expression regulation, the cohesion defects that we observed could be caused by an altered expression of cohesin components or regulators. To test for this, we undertook immunoblot experiments to analyze the total amounts of the core cohesin subunits RAD21, SA-2 and SMC1A, the cohesin loader MAU2, and the cohesin regulators sororin, ESCO2 and WAPL. We observed that NSD3 depletion did not alter the expression levels of any of these cohesin subunits and partners (Fig. S2). Moreover, these experiments also indicated that global di-methylation levels of H3K36 remained unaffected upon NSD3 inactivation (Fig. S2), indicating that the contribution of NSD3 to sister chromatid cohesion is independent of both the global disruption of H3K36 di-methylation and the altered expression of cohesin components and regulators.

We went on to examine the proportions of mitotic cells in the population, and saw that, depending on the siRNA used, NSD3 depletion caused a significant increase in the mitotic index (Fig. 1D). An increase in the proportion of mitotic cells which exhibit separated sister chromatids, such as that observed with NSD3 inactivation, can arise either because of an accumulation of prometaphase cells with defective mitotic cohesion, or because the cells have prematurely entered into anaphase. In order to discriminate between these two possibilities, we analyzed the expression of two mitotic markers whose levels decrease when cells enter anaphase. The first was cyclin B1, whose levels increase starting at late S-phase, culminate during the first steps of mitosis, then rapidly degrade prior to anaphase onset (Pines and Hunter, 1989). The second marker was phosphorylation of H3S10, a process which begins when the cells enter prophase and which is then rapidly erased after the transition from metaphase to anaphase (Gurley et al., 1978; Hendzel et al., 1997). Following depletion with the two siRNAs that induced the strongest mitotic index increase, we observed an accumulation of both of these markers, indicating that cells were blocked in a prometaphase-like state (Fig. 1C). To confirm that this arrest resulted from SAC activity, we treated NSD3-depleted cells with the SAC inhibitor reversine for 3 h (Fig. 1E). In this case, the levels of both mitotic markers were similar to those in cells treated with the control siRNA. Moreover, we also treated cells with the microtubule-depolymerizing agent nocodazole for 6 h, artificially ensuring that the SAC remained active, and there was again no difference after NSD3 depletion (Fig. 1F). It is thus clear that NSD3 depletion does not directly affect SAC activity. Taken together, our results strongly suggest that cohesion defects induced by NSD3 depletion cause a SAC-dependent arrest during prometaphase in HeLa cells.

We next looked at which stage of the cohesion cycle is affected by NSD3 depletion. Because NSD3 inactivation induces PSCS, we excluded the idea that it contributes to the cohesion-releasing prophase pathway, as its depletion would lead to the opposite phenotype (a two-chromatid chromosome with cohesive arms rather than an X-shape). We therefore looked for the presence and localization of the centromeric cohesion protector SGO1. After NSD3 depletion, we detected as much SGO1 at the centromeres of separated sister chromatids as we saw in control cells having attached sister chromatids (Fig. 2A). However, two pools of SGO1 that are localized at the inner and outer centromeres, respectively, have been described (Liu et al., 2013a). The inner pool is required to protect centromeric cohesion, but the function of the outer pool remains elusive (Liu et al., 2013b; Zhou et al., 2017). In the absence of NSD3, sister chromatid separation prevented us from determining whether the centromeric SGO1 we observed corresponded to one or both of these pools. Protection of centromeric sister chromatids also depends on the kinase HASPIN, via WAPL phosphorylation and the interactions of HASPIN with PDS5 and WAPL (Goto et al., 2017; Liang et al., 2018; Zhou et al., 2017). Whether the molecular mechanism involves SGO1 or HASPIN, the protection of centromeric cohesion occurs by counteracting WAPL. Mitotic cohesion fatigue is also sensitive to the presence of WAPL (Daum et al., 2011). We therefore explored the effects of WAPL depletion in the absence of NSD3. To verify the functional efficiency of such a depletion, we confirmed that co-depletion of HeLa cells with both WAPL and SGO1 siRNA prevented PSCS, whereas cells depleted for SGO1 alone caused it (Fig. 2B, left). As an additional control, we also showed that the co-depletion of WAPL and RAD21 fails to rescue PSCS that had been induced by the lack of the core cohesin components (Fig. 2B, middle). Using the same set-up, we were able to show that WAPL depletion did not rescue PSCS induced by NSD3 depletion (Fig. 2B, right). Thus, the contribution of NSD3 to sister chromatid cohesion does not involve counteracting cohesin-dissociation activity of WAPL. Together, these results indicate that NSD3 is not involved in maintaining sister chromatid cohesion during mitosis and suggest that its role occurs in interphase.

To test this hypothesis, we then explored whether the defective cohesion seen after NSD3 depletion is also detected in interphasic cells before mitotic entry. To this end, we analyzed sister chromatid cohesion in G2-synchronized cells by DNA fluorescent in situ hybridization (FISH) using a probe specific for the centromeric region of chromosome 11 (Fig. 2C). Cohesion was assessed by measuring the distances between paired DNA FISH signals in control cells and in cells depleted for NSD3, with RAD21 depletion used as a positive control for G2 cohesion defects (Fig. 2C, left and middle). The mean distance between paired dots was 0.52 µm in the control cells, and this increased to 0.67 µm upon NSD3 depletion and to 0.68 µm in RAD21-depleted cells (Fig. 2C, right). These results demonstrate that NSD3 depletion leads to defective sister chromatid cohesion in G2 cells, thereby indicating that the contribution of NSD3 to cohesion must take place before cells enter mitosis.

We next addressed whether NSD3 could be involved in the loading of cohesin and/or kollerin onto chromatin, a process that takes place at mitotic exit and early in G1 phase. If so, the cohesion defects we observed 72 h after transfection with siRNA NSD3 would be the consequence of a defect in cohesin loading occurring at the exit of the previous mitosis, ∼48 h after NSD3 depletion. Thus, for these particular experiments, control and NSD3-depleted cells were synchronized by a single thymidine arrest-and-release, followed by mitotic arrest induced by nocodazole, and released from nocodazole-mediated arrest at 48 h post transfection. Then, cells were harvested at different time points and fractionated to obtain chromatin fractions (Fig. 3A; see Materials and Methods section for details). We next analyzed these by SDS-PAGE and immunoblotting (Fig. 3B). Under these experimental conditions, NSD3 inactivation did not impact kinetics of cell mitotic exit, because both the unloading of the CAP-D2 condensin complex subunit I from chromatin and the dephosphorylation of serine 10 on histone H3 occurred with identical kinetics in the control and NSD3-inactivated cells (Fig. 3B,C). By contrast, the accumulation of cohesin subunits (RAD21, SA-2 and SMC1A) onto chromatin over time was lower in NSD3-depleted cells than in control cells. Remarkably, a similar decrease was also observed for the kollerin subunit MAU2. These decreased protein signals were not due to reduced amounts of chromatin, as both the chromatin-associated enzyme topoisomerase II and histone H3 were present at comparable levels in the control and NSD3-depleted cells. The effect cannot be attributed to a decrease in total proteins either, because whole-cell extracts had constant protein levels (Fig. 3B, left panels). This experiment thus show that NSD3 inactivation leads to reduced recruitment of cohesin and MAU2 onto chromatin during mitotic exit, suggesting that NSD3 contributes to this process.

Next, we hypothesized that NSD3 interacts physically with the kollerin complex to recruit it to chromatin complexes. Using HeLa cell chromatin extracts treated with Turbonuclease to limit indirect interaction due to protein–DNA bridging, we immunoprecipitated endogenous NSD3, and we observed co-precipitation of both MAU2 and (albeit to a lesser extent) NIPBL (Fig. 3D). In a reciprocal experiment using the anti-NIPBL antibody, MAU2 was strongly co-immunoprecipitated as expected, and so was NSD3. To confirm this interaction, we used a cell line that allows for the inducible expression of full-length NSD3 fused to an emerald green fluorescent protein (EmGFP) tag (Fig. S3). Using an anti-GFP antibody, we could immunoprecipitate EmGFP–NSD3 together with MAU2 and NIPBL only when EmGFP–NSD3 was induced (Fig. 3E). Our results therefore show that NSD3 is able to interact physically with the kollerin complex.

As shown in Fig. 3B and C, immunoblot analysis of chromatin fractions revealed that both the long and short NSD3 variants are absent from chromatin in mitosis-arrested control cells, and that they are then progressively recruited onto chromatin upon release from mitotic arrest and progression to the G1 phase. This behavior is similar to that of kollerin and most of cohesin. Our next step was to more precisely describe the dynamics of chromatin loading of NSD3 relative to those of kollerin and cohesin at the single-cell level by fluorescence microscopy approaches.

Having checked the suitability of the anti-NSD3 antibody for indirect immunofluorescence experiments (Fig. S4A), we monitored the localization of NSD3 at different stages of the cell cycle, using chromosome morphology and H3S10 phosphorylation to identify the different mitotic stages. To assess the presence of NSD3 on chromatin, we performed immunostaining experiments with and without detergent-based pre-extraction of the soluble pool of proteins (Fig. S4B,C). As it is the case for many proteins that bind chromatin, we found that NSD3 localizes in the nucleus during interphase, is evicted from chromatin in prophase, then re-loads onto chromatin from anaphase onwards. Finally, to ensure that NSD3 localization is not altered by fixation, we performed live-cell imaging in a cell line expressing EmGFP–NSD3 and H2B–mCherry, and were able to confirm that EmGFP–NSD3 presents the same localization profile and dynamics as those revealed by immunofluorescence experiment (Movies 1–3).

We then analyzed the presence of EmGFP–NSD3 on chromatin from metaphase to mitotic exit, comparing its signal to those obtained with anti-RAD21 and anti-MAU2 antibodies, used as cohesin and kollerin markers, respectively. We saw that EmGFP–NSD3 signals were detected on chromatin prior to those of MAU2 and RAD21, indicating that NSD3 loads onto chromatin before MAU2 and RAD21 (Fig. 4A,B). NSD3 is therefore removed from chromatin after mitosis entry and is reloaded back at early anaphase shortly before kollerin and cohesin are recruited.

Our work shows that siRNA-mediated inactivation of NSD3 results in the reduction of cohesin and kollerin loading onto chromatin as well as cohesion defects in both interphase and mitotic cells. As we had used siRNA targeting both the long and short NSD3 isoforms, we decided to explore whether one or both isoforms participate in these processes. To this end, we designed siRNAs directed against the distinct 3′-UTR domain of each isoform (two different siRNAs per isoform). These siRNAs enabled the depletion of each form in an efficient and selective manner, as shown by immunoblotting experiments (Fig. 5A). We then compared the ability of these isoform-specific siRNAs to induce PSCS in mitotic cells. As shown in Fig. 5B, both of the siRNAs targeting the long isoform resulted in defective cohesion, whereas those targeting the short form did not despite efficient NSD3-s depletion. We next analyzed the consequence of isoform-specific depletion on SAC-dependent arrest by using the most efficient siRNA for each isoform, NSD3-s-1 and NSD3-L-2. Depletion of NSD3-s did not have any effect on the levels of phosphorylated H3S10 (Fig. 5C). In contrast, when NSD3-L was absent, phosphorylated H3S10 signals accumulated strongly, and were sensitive to a treatment with reversine (Fig. 5C). These results show that only depletion of the long form of NSD3 results in the accumulation of cells in mitosis as well as defects in sister chromatid cohesion. This strongly suggests that NSD3-L is the one NSD3 isoform involved in cohesion regulation.

We therefore went on to compare the effect of depleting each NSD3 isoform on the levels of chromatin-bound MAU2 and cohesin by subcellular fractionation experiments (Fig. 5D). Fractionation efficiency was verified with α-tubulin, a cytosolic protein present in the soluble fraction but not in the chromatin-associated fraction, and with histone H3, a core component of chromatin, which is not detected in the soluble fraction. As expected, RAD21 and MAU2 are present in both soluble and chromatin fractions in the control cell extracts. However, depletion of NSD3-L induced a decrease of MAU2 and RAD21 amounts in the chromatin fraction while, at the same time, these were conversely increased in the soluble fractions (Fig. 5D). This supports the hypothesis that the NSD3 long isoform (but not the short NSD3-s) is the one that contributes to cohesin loading.

In order to further strengthen this view and exclude any off-target effect that the siRNAs might have, we tested whether expression of exogenous NSD3-L could rescue defective sister chromatid cohesion induced by depletion of its endogenous counterpart. For this purpose, we performed rescue experiments using the cell line expressing EmGFP–NSD3-L with the siRNA NSD3-2, which targets both isoforms, as well as with NSD3-L-2, which is specific for the long variant only. We used western blots to verify proper expression of exogenous EmGFP–NSD3-L and efficient depletion of endogenous NSD3 isoforms (Fig. S5). Then, we determined the percentage of prometaphase cells exhibiting PSCS for each condition. To take into account only EmGFP–NSD3-L-positive cells in rescue experiments and to detect sister chromatid separation concomitantly, we deposited swollen cells onto a glass slide with a cytocentrifuge from the same populations used for western blots, and then labeled NSD3 and the centromere by immunofluorescence with an anti-GFP antibody and a CREST serum, respectively. As expected, given that NSD3 is not associated with chromatin during prometaphase, EmGFP–NSD3-L was diffusely localized around the chromosome (Fig. 5E). In the absence of doxycycline, we confirmed that depletion of both NSD3 isoforms and of just the long form results in PSCS (Fig. 5F). By contrast, in cells where wild-type EmGFP–NSD3-L was expressed, the proportion of mitotic cells displaying PSCS was significantly reduced (Fig. 5F). This indicates that ectopic correction of NSD3-L protein level rescues defective mitotic cohesion to a large extent, thereby establishing NSD3-L as the isoform required for proper sister chromatid cohesion.

Finally, we addressed whether NSD3 methyltransferase activity is required for sister chromatid cohesion. To this end, we generated another cell line expressing a catalytically inactive NSD3-L mutant by replacing tyrosine 1261 with an alanine, because this residue is inside the catalytic pocket of the SET domain (Li et al., 2021). When the same rescue experiment as the one described above was performed with these cells, expression of the EmGFP–NSD3-Y1261A mutant did not attenuate the cohesion defects induced by endogenous NSD3 depletion (Fig. 5F). Therefore, the methyltransferase activity of the long NSD3 isoform is required to ensure regulation of sister chromatid cohesion.

Sister chromatid cohesion is essential for proper chromosome segregation. In this work, we show that depletion of the methyltransferase NSD3 leads to a defect in this process and that this is characterized by a mitotic delay dependent on the SAC. Because the key WAPL regulator is unable to rescue PSCS induced by depletion of NSD3, we ruled out the possibility that the role of NSD3 in cohesion is related to processes involving cohesin dissociation or cohesion maintenance. Because of its involvement in kollerin recruitment and in cohesin loading onto chromatin, we propose that defective kollerin recruitment during mitotic exit is the cause of the PSCS observed in mitotic cells upon NSD3 inactivation. In agreement with this interpretation, the defects observed in mitotic sister cohesion are incomplete, and depending on the siRNA used, involve 20–40% of mitotic cells. In that regard, we noticed that the siRNA that induced the strongest phenotype is not the one leading to the most efficient NSD3 extinction (Fig. 1C,E). However, given that NSD3 is involved in gene expression and genome organization (Wagner and Carpenter, 2012; Shen et al., 2015) and in the regulation of cohesin (this study), we cannot rule out the possibility that a more efficient extinction of NSD3 leads to pleiotropic and indirect effects that would partially prevent sister chromatid dissociation. Nevertheless, the level of mitotic defects induced by all the NSD3 siRNAs used in this study are consistent with those usually observed in vertebrate cells upon depletion of kollerin or of other proteins known to be involved in cohesin loading during mitotic exit (Hahn et al., 2013; Ritchie et al., 2008; Watrin et al., 2006). However, cohesin loading occurs not only at mitotic exit, but also in S-phase (Zheng et al., 2018). In that case, cohesin loading is dependent on MCM2–MCM7 and involves kollerin, whose alteration leads to cohesion defects before mitotic entry. Thus, the mitotic cohesion defects observed when NSD3 is absent might also be attributed to its possible contribution to kollerin loading during the pathway specific to the S-phase. Moreover, because NSD3 depletion is likely to have pleiotropic effects due to its role in chromatin modification, we cannot definitively rule out the hypothesis that NSD3 is also involved in other interphasic cohesion-related processes.

Looking for a contribution for NSD3 in cohesion during interphase, we show that NSD3 interacts with kollerin on chromatin. We also show that NSD3 is required for the recruitment of MAU2 to chromatin during mitotic exit. As a consequence, NSD3 also contributes to the loading of cohesin onto chromatin. In full agreement with this function, we showed that NSD3 recruitment to chromatin occurs during early anaphase, slightly before those of MAU2 and RAD21 (Watrin et al., 2006). The physical interaction we uncovered between NSD3 and kollerin is therefore unlikely to occur prior to their targeting to chromatin. Instead, NSD3 might promote the binding of kollerin to chromatin during mitotic exit. Consistent with this possibility, the MAU2 homolog in budding yeast, Scc4, is necessary for loading Scc2 onto chromatin in vivo, but it has no affinity for DNA per se, implying that a protein receptor is required (Chao et al., 2015; Hinshaw et al., 2015; Murayama and Uhlmann, 2014). It has also been reported that the transcription regulator BRD4 co-immunoprecipitates with NSD3 and with the kollerin complex through its extra-terminal domain, stabilizing NIPBL on chromatin in vertebrate cells (Linares-Saldana et al., 2021). In RN2 human acute myeloid leukaemia cells, BRD4 recruits NSD3 on chromatin at active promoters and enhancers across the genome (Shen et al., 2015), and these domains are known to be enriched with cohesin (DeMare et al., 2013; Heidari et al., 2014; Kagey et al., 2010). In Drosophila melanogaster, the BRD4 homologue Fs(1)h promotes the association of Nipped-B and RAD21 with enhancers and promoter regions close to the replication origin, and genetic interactions exist between genes which code for these three proteins (Pherson et al., 2019). It is thus tempting to propose that BRD4 and NSD3 act together to load kollerin onto chromatin. However, we showed that NSD3 depletion decreased RAD21 level on chromatin and induced PSCS, which was not rescued by WAPL depletion. By contrast, the absence of BRD4 does not decrease the level of RAD21 on chromatin in mouse embryonic stem cells and induces differentiation defects in neural crest progenitors that is rescued by WAPL depletion (Linares-Saldana et al., 2021). Moreover, the link between NSD3 and BRD4 depends on the short NSD3 isoform at the functional level (Shen et al., 2015), whereas we have demonstrated that the long NSD3 isoform is responsible for the regulation of sister chromatid cohesion. Thus, the putative interplay between BRD4 and NSD3 in chromatin organization remains to be clarified. Future studies will also be necessary to precise the molecular mechanism of NSD3-dependent kollerin recruitment, and notably to confirm its specificity, to elucidate whether it occurs through direct interaction with the kollerin complex or by providing a chromatin context that is suitable for its binding. Interestingly, it has already been reported that NSD3 can play such an adaptor role, bridging the BET domain of BRD4 and the CHD8 chromatin remodeling factor (Shen et al., 2015). In Schizosaccharomyces pombe, the chromatin remodeling complex RSC (‘remodels the structure of chromatin’) acts as a chromatin receptor by physically interacting with the Scc2–Scc4 complex. It also promotes nucleosome eviction in order to render the DNA naked, thus facilitating cohesin loading (Lopez-Serra et al., 2014; Muñoz et al., 2022, 2019, 2020). Moreover, other chromatin remodelers have been described as contributing to cohesin loading (Eid et al., 2015; Hakimi et al., 2002; Kagey et al., 2010; Kernohan et al., 2010; Muñoz et al., 2022; Pherson et al., 2019). Looking for a link between NSD3 and chromatin remodeling is therefore a promising way to explore the contribution of NSD3 to kollerin chromatin association.

By targeting specific isoforms, we were able to show that the contribution of NSD3 to sister chromatid cohesion and to chromatin-association of kollerin and cohesin actually only involves the full-length form, which contains the methyltransferase activity. Moreover, we also showed that, unlike its wild-type counterpart, a catalytically inactive version of NSD3-L was not able to rescue cohesion defects. These results imply that NSD3-L regulates sister cohesion by methylating one or more substrates that have not yet been identified. Interestingly, the methyltransferase Suv4-20h2 has been previously shown to specifically regulate cohesin loading on pericentromeric heterochromatin, which is known to have low transcriptional activity (Hahn et al., 2013). This suggests that complementary cohesin-loading regulating pathways do co-exist according to the targeted chromatin domains. Because kollerin seems to require partners for its chromatin interactions (Chao et al., 2015; Hinshaw et al., 2015; Murayama and Uhlmann, 2014), one can speculate that the catalytic activity of NSD3-L might act to recruit kollerin at specific chromatin sites, which is consistent with the partial defects observed in kollerin loading and in mitotic sister cohesion. As for right now, the histone H3K36 is the only described substrate of NSD3. Thus, H3K36me2 stands as an attractive candidate to further investigate the contribution of NSD3 to the recruitment of cohesin loaders and cohesin onto chromatin. In agreement with this possibility, H3K36me2 is found in active gene promoters, where NIPBL is also enriched in human cells (Kuo et al., 2011; Zhu et al., 2016; Zuin et al., 2014). Such a methylated mark which is stable throughout cell division could be essential for the epigenetic inheritance of some cohesion-loading sites.

Primary antibodies used in this work with dilutions for western blotting (WB) and immunofluorescence (IF) are indicated in Table S1. Horseradish peroxidase coupled secondary antibodies (Jackson ImmunoResearch) were used at 1:5000 and 1:25,000 dilution for WB detection of mouse and rabbit antibodies, respectively. Alexa Fluor-coupled secondary antibodies from Invitrogen were used at 1:1000 dilution for IF detection.

A fragment containing an inducible TRE-tight promoter and a LAP tag (a FLAG tag followed by an EmGFP tag) was inserted at the XhoI site of the plasmid pBSKDB-CAG-rtTA2sM2-IRES-tSkid-IRES-Neo (Addgene, #62346), thereby generating the vector pGEH_ind_LAP-C. In this vector, the coding sequences of NSD3-s, NSD3-L and NSD3-L-Y1261A were each cloned in-frame in the C-terminal of the LAP tag after PCR amplification from the plasmids pMSCV_MigR1_NSD3short and pMSCV_MigR1_NSD3long (both kindly provided by Prof. Christopher R. Vakoc, Cold Spring Harbor Laboratory, Cancer Center, USA) (Shen et al., 2015). For the construction of NSD3-L-Y1261A, the canonical codon sequence ‘TAT’ generating Y1261 was replaced by ‘GCT’ on the PCR fragments used for the cloning reaction, leading to the generation of A1261. All PCR reactions were undertaken using Q5 high-fidelity polymerases (M0493S, NEB), and all constructions and intermediates were generated using a NEBuilder HiFi DNA assembly cloning kit (E5520S, NEB) according to the manufacturer's instructions. Plasmids and their maps can be provided upon request.

For generation of HeLa EmGFP–NSD3-L and EmGFP–NSD3-L-Y1261A cell lines, 5 µg of the corresponding plasmids supplemented with 10 µl of P3000 in 125 µl Opti-MEM (31985062, Thermo Fisher Scientific), was mixed with 7.5 µl Lipofectamine 3000 reagent in 125 µl Opti-MEM. After 5 min incubation at room temperature, the plasmid-containing mix was transfected in HeLa Kyoto cells (see below) seeded the day before in 2.25 ml of complete medium (see below) in six-well plates. After 24 h, the cells were trypsinized and either 1/10 or 1/50 of the cells were seeded in culture dishes (150 mm) with 25 ml of complete medium containing 1 mg/ml of geneticin selective antibiotic (11811023, Thermo Fisher Scientific). Selection lasted for 2–3 weeks, with replacement of the antibiotic-supplemented medium every 2–3 days. The cells were then induced for EmGFP–NSD3 expression with 2 µg/ml of doxycycline (D9891-1G, Merck) for 48 h, and GFP-positive cells were FACS-sorted at the Biosit CytomeTRI platform. They were cultured for another 2 weeks under selection pressure but without induction, and the cell lines were then subcloned by limiting dilution in 96-well plates. After 2 weeks of selection, EmGFP–NSD3 expression was evaluated in ∼30 clones for each NSD3 variant. Despite this process, no homogeneous clonal cell lines could be obtained. For the EmGFP–NSD3-s construct, the clones obtained always expressed the fused protein constitutively. In contrast (and as expected), expression of EmGFP–NSD3-L and EmGFP–NSD3-L-Y1261A in the selected cell lines was inducible. A HeLa H2B–mCherry EmGFP–NSD3-L cell line was generated by transfecting EmGFP–NSD3-L cells with a pH2B_mCherry_IRES_puro2 plasmid (Addgene, #21045) as described above, with a 15-day selection and 0.5 µg/ml puromycin.

All experiments discussed in this paper were performed with HeLa Kyoto cells or cell lines constructed from those cells. All cell lines were cultured at 37°C in an incubator supplied with 5% CO₂ in Dulbecco's modified Eagle's medium (DMEM) with glutamine analogue GlutaMAX (31966047, Thermo Fisher Scientific), supplemented with 10% fetal bovine serum (FBS) and a cocktail of penicillin-streptomycin antibiotics (15140-122, Thermo Fisher Scientific) at 100 U/ml and 100 µg/ml final concentrations, respectively (denoted complete medium). Stable cell lines were maintained in culture with 1 mg/ml geneticin. Doxycycline was used at 1 µg/ml in medium to induce expression of the exogenous tagged proteins, and was replaced with fresh doxycycline-containing medium after 48 h for experiments requiring 72 h induction. For SAC experiments, cells were incubated with 1 µM reversine (S7588, Selleck Chemicals) for 3 h or with 100 ng/ml nocodazole (M1404, Merck) for 6 h. To synchronize cells, thymidine (T1895-1G, Merck) was added to the medium 8 h post-transfection at a final concentration of 2 mM, and the cells were cultured for 24 h. To arrest the cells in prometaphase, this was followed by two successive 3-min PBS washes, then incubation with complete medium supplemented with 100 ng/ml nocodazole for another 16 h. Cells were harvested by mitotic shake-off, washed as described for the previous release, and seeded in fresh complete medium on new plates. To enrich the cells in G2, they were subjected to a double thymidine block (16 h each time), followed by a release of 6 h. Before use in synchronization experiments, all media were pre-warmed to 37°C.

All siRNAs used in this study are described in Table S2. Unless otherwise stated, HeLa cells were transfected with 20 nM siRNA for 72 h with Qiagen HiPerFect reagent according to the manufacturer's recommendations. As a general guideline, for a six-well transfection, 10 µl of reagent and 2.5 µl of 20 µM stock siRNA were mixed in 87.5 µl of Opti-MEM. This was added to 300,000 HeLa cells suspended in 2.4 ml of complete medium. Medium was replaced after 48 h of transfection, and cells were trypsinized and diluted if necessary. For WAPL co-depletion, cells were transfected with Luc or the indicated siRNA, then 24 h later a second transfection was performed with Luc or WAPL siRNA, and cell culture was prolonged for 48 h. The siRNA used are described below, and were purchased from Qiagen or Dharmacon. A listing of those used for methyltransferase screening can be provided upon request.

Cells harvested from a six-well plate were subjected to hypotonic shock in 4.5 ml of 75 mM KCl for 15 min at room temperature, then 500 µl of Carnoy's fixative solution (at a ratio of 3:1 methanol to acetic acid) was added. Cells were centrifuged at 250 g and the pellet was resuspended in 5 ml of the fixative. This operation was repeated three more times and the pellet was kept overnight at −20°C in 300 µl of Carnoy's fixative. We then dropped 30 µl of cells onto a dry slide, let them dry for 2 h at room temperature, and incubated the slides for 5 min in fresh 5% GIEMSA solution (1.09203, Merck) diluted in 100 ml Gurr buffer (10582-013, Thermo Fisher Scientific). After three to five washes in distilled water, Giemsa-stained cells were mounted with Entellan (1.07960, Sigma-Aldrich). For immunofluorescence in chromosome spreads, 50,000 cells swollen after hypotonic shock in 200 µl of KCl 75 mM were cytospun on a slide for 5 min at 900 rpm (Cytospin 4, Thermo Fisher Scientific). These cells were then fixed in 3% paraformaldehyde in PBS for further immunolabeling.

When stated, soluble contents of cells were pre-extracted before fixation by incubation for 1 min in 0.1% Triton X-100 (T8787-50ML, Merck) diluted in 1× PBS (10010-015, Thermo Fisher Scientific). Cells were fixed for 10 min in 4% paraformaldehyde (15710, Electron Microscopy Science) diluted in 1× PBS (pH 7.2–7.4). Slides or coverslips were then washed three times for 5 min in 1× PBS, permeabilized with 0.1% Triton X-100 for 10 min, washed three times in 1× PBS, then blocked by incubation with 5% fetal calf serum (FCS; CVFSVF00-01, Eurobio Scientific) in 1× PBS for 1 h at room temperature. This last solution was used to dilute primary and secondary antibodies. Slides or coverslips were then incubated overnight at 4°C with primary antibodies, washed three times with 1× PBS, and further incubated for 1 h at room temperature with fluorochrome-conjugated secondary antibodies. The DNA was stained with DAPI (100 ng/ml in PBS), and slides were mounted in ProLong Gold medium (P36982, Thermo Fisher Scientific).

Images were acquired with a Zeiss Axio Imager M2 epifluorescence microscope equipped with Zeiss Plan-Apochromat 40×/1.3 and 63×/1.40 oil objectives, a CoolSNAP HQ² CCD camera (Photometrics), and Zeiss AxioVision software (version 4.2). Signals were quantified using the Fiji image processing package (Schindelin et al., 2012).

DNA FISH was performed as previously described (Schmitz et al., 2007), except that cells were fixed with Carnoy's fixative (as described above), then spread on glass sides before being processed. Analysis was only done on pairs for which the dots could be clearly resolved in the same focal plane. The probe 5′-AgGgTtTcAgAgCtGcTc-3′ used for FISH was coupled with an Alexa 488 fluorophore and targets a complementary alpha-satellite sequence specific to the centromeric region of chromosome 11. In that probe sequence, uppercase letters correspond to DNA, whereas lowercase letters indicate locked nucleic acids, which are modified RNA nucleotides in which the ribose moiety is modified, with an extra bridge connecting the 2′ oxygen and 4′ carbon.

HeLa H2B–mCherry EmGFP–NSD3-L cells were seeded on a Lab-Tek chamber slide (155409, Thermo Fisher Scientific) in 500 µl DMEM complete medium supplemented with 2 µg/ml doxycycline, then incubated overnight as described above. Afterwards they were supplemented with 10 µM of the CDK1 inhibitor RO-3306 (S7747, Selleck Chemicals) and incubated for 6 h. Cells were washed three times with DMEM preheated to 37°C, then the medium was replaced with 500 µl of Leibovitz's L-15 medium (11415064, Thermo Fisher Scientific), also preheated to the same temperature and supplemented with 20% FBS, penicillin-streptomycin antibiotics (as above), and 2 mM L-glutamine (25030081, Thermo Fisher Scientific). Cells were imaged at 37°C every 3 min for 10 h, with 15 stacks spaced by 1 µm at each acquisition time. They were irradiated with lasers to excite H2B–mCherry (561 nm) and EmGFP–NSD3-L (488 nm). Images were acquired using a spinning disk system consisting of a Leica DMi8 microscope equipped with a 63×/1.4 oil objective, a CSU-X1 spinning-disk unit (Yokogawa), and an Evolve EMCCD camera (Photometrics). The microscope was controlled using its dedicated software and the Inscoper imaging suite.

For whole-cell extracts, proteins were extracted by directly resuspending cell pellets in Laemmli buffer (60 mM Tris-HCl pH 6.8, 10% glycerol, 2% SDS, 0.05% Bromophenol Blue and 5% β-mercaptoethanol). For fractionation, cells were collected by trypsinization and washed once with ice-cold PBS. The final cell pellet was resuspended in extraction buffer (20 mM Tris-HCl pH 7.5, 100 mM sodium chloride, 5 mM magnesium chloride, 0.2% NP-40, 10% glycerol and 0.5 mM dithiothreitol) supplemented with EDTA-free protease inhibitor tablets (5892953001, Merck) and a homemade phosphatase inhibitor cocktail (final concentrations of 5 mM sodium fluoride, 10 mM β-glycerophosphate, 1 mM sodium pyrophosphate and 0.2 mM sodium orthovanadate). Cells were lysed on ice by ten passages through a 27-gauge needle. The resulting lysates were incubated for 10 min on ice, an aliquot was collected as a whole-cell extract, and the remaining lysates were centrifuged at 12,000 g for 5 min at 4°C in order to collect the soluble protein extracts. The chromatin-containing pellet was washed four times with the same extraction buffer at 12,000 g for 5 min at 4°C and then resuspended in Laemmli buffer.

For immunoblotting, lysed cells were heated for 5 min at 95°C. Samples were then subjected to SDS-PAGE electrophoresis in 4–15% or 4–20% polyacrylamide gradient gels (4561085 or 4561095, Bio-Rad) and a Bio-Rad Trans-Blot Turbo transfer system was used to transfer them onto PVDF membranes (1704156, Bio-Rad). Following saturation for 1 h in 1× PBS containing 5% milk and 0.1% Tween 20 (655204-100ML, Merck), the membranes were incubated overnight at 4°C with primary antibodies, then for 1 h at room temperature with horseradish peroxidase (HRP)-conjugated secondary antibodies, according to standard procedures. For NSD3 western blot analysis, saturation and incubation were performed similarly, except that the saturation and incubation buffers were supplemented with 10% (instead of 5%) milk and with 150 mM NaCl. Membranes were treated with the Immobilon ECL Ultra Western HRP substrate (WBKLS0500, Merck), and signals were detected using an Amersham Imager 680 (GE Healthcare) or using Amersham Hyperfilm ECL film (GE28-9068-35, Merck). Quantification of band signal intensity was done with the analyze gels tool in the Fiji software.

Cells were resuspended at 4.10⁷ cells/ml of buffer A (10 mM HEPES pH 7.9, 10 mM potassium chloride, 1.5 mM magnesium chloride, 0.34 M sucrose, 10% glycerol and 0.5 mM dithiothreitol), to which the same volume of buffer A supplemented with 0.2% NP-40 was then added. After 10 min incubation on ice, the extracts were centrifugated at 1300 g for 5 min at 4°C. The pellets were washed for 1 min in buffer A to a concentration equivalent to 20,000 cells/ml. After a similar centrifugation, they were resuspended to 40,000 cells/ml in IP buffer (10 mM HEPES pH 7.9, 10 mM potassium chloride, 140 mM sodium chloride, 1.5 mM magnesium chloride, 10% glycerol and 0.5 mM dithiothreitol). The lysate was supplemented with calcium chloride to a final concentration of 3 mM and chromatin was solubilized for 2 h at 4°C by digestion with Turbonuclease (T4330, Merck) to a final concentration of 250 U/ml. The reaction was then arrested by addition of a 25× stop buffer (125 mM EGTA and 10 mM EDTA) to a final concentration of 1×. Solubilized chromatin was collected by centrifugation at 13,000 g for 20 min at 4°C. For immunoprecipitation against endogenous proteins, 250 µl of chromatin extracts were incubated overnight with 5 µg of IgG, anti-NSD3 or anti-NIPBL rabbit antibodies followed by 1 h incubation at 4°C on a rotating wheel with 40 µl of a 1:1 mix of Dynabead protein A (10001D, Life Technologies) and G (10003D, LifeTechnologies) magnetic beads. Beads were then washed four times in 500 µl of IP buffer without protease inhibitors and eluted in 80 µl of Laemmli buffer. For immunoprecipitation of the induced EmGFP–NSD3-L proteins, the protocol was the same except that chromatin solubilization with Turbonuclease was performed for 1 h only and was not arrested with the stop buffer. Instead, 25 µl of ChromoTek GFP-Trap M-270 magnetic particles (gtd-20, Proteintech) were immediately added to 250 µl of the extracts. This was incubated for 1 h on a rotating wheel at 4°C, and before elution, was washed four times in 500 µl ice-cooled wash buffer (10 mM Tris/HCl pH 7.5, 150 mM NaCl, 0.05% NP-40, and 0.5 mM EDTA). All buffers except the wash and elution buffers were supplemented with protease inhibitors.

All statistical analysis was performed using GraphPad Prism software (v6.05). For the comparison of the means of the proportions showing PSCS or the mitotic index between NSD3-depleted cell lines, we performed one-way analysis of variance (ANOVA), or in the case of Fig. 2B, two-way ANOVA. These were followed by Dunnett post-hoc analysis, assuming the normal law for the repartition of the means. For Fig. 2B, left and middle panels, we performed a χ-square analysis of the experiment with Yate's correction. Comparisons of fluorescence intensity from representative experiments were analyzed using a nonparametric Kruskal–Wallis test followed by Dunn's multiple comparison correction. For distance comparison in the DNA FISH experiments, one-way ANOVA was followed by Bonferroni's multiple comparisons test. For all tests, an alpha risk of 0.05 was used.

We would like to give special thanks to Dr Rozenn Gallais for her contributions to this study. For providing reagents, we thank Prof. Christopher Vakoc (Cold Spring Harbor Laboratory, USA), Dr Christophe Escudé (National Museum of Natural History, France), Prof. Jan-Michael Peters (Research Institute of Molecular Pathology, Austria), Prof. Yoshinori Watanabe (University of Tokyo, Japan) and Dr Isabelle Bahon-Riedinger (Rennes University Hospital, France). We also thank Laurent Deleurme of the CytomeTRI flow-cytometry platform at BIOSIT (SFR UMS CNRS 3480 – INSERM 018) for the FACS cell sorting of the EmGFP–NSD3 HeLa cell lines. We thank Juliana Berland for suggestions on the manuscript.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.261014.reviewer-comments.pdf.