Histone H3K27 methylation modulates the dynamics of FANCD2 on chromatin to facilitate NHEJ and genome stability

Over the past few decades, accumulating evidence has supported a critical role for aberrant epigenetic regulation and alterations in histone modification in the development of diverse cancers (Allis and Jenuwein, 2016; Lewis et al., 2013; Dawson and Kouzarides, 2012; Jones and Baylin, 2007). One histone modification of particular interest is di- and tri-methylation of lysine 27 on histone H3 (H3K27me2/3), a classic epigenetic marker for transcriptional gene silencing (Pirrotta, 1997; Cao et al., 2002; Margueron and Reinberg, 2011; Xiong et al., 2016). Excessive levels of H3K27me2/3 due to over-expression and gain-of-function mutation of the methyltransferase EZH2 have been found in a wide variety of tumours (Albert and Helin, 2010; Ezponda and Licht, 2014). However, a global decrease in H3K27me2/3 due to mutations of EZH2 that disable its methylation activity have also been identified in leukaemia and myeloid malignancies (Ernst et al., 2010; Kim et al., 2014). Direct mutation of histone H3 lysine 27 is involved in non-brainstem paediatric glioblastoma (non-BS-PG) and diffuse intrinsic pontine glioma (DIPG) (Wu et al., 2012; Schwartzentruber et al., 2012). Nearly 80% of DIPGs harbour mutations in canonical (H3.1) or variant (H3.3) histone H3 genes, with lysine 27 substituted by methionine (H3K27M) (Schwartzentruber et al., 2012). Unlike the replication-independent histone H3.3 variant, H3.1 is the major H3 variant and can be incorporated in a replication- or repair-dependent manner (Tagami et al., 2004; Polo et al., 2006). H3K27M mutants bind and inhibit the catalytic SET domain of EZH2 (Justin et al., 2016), resulting in a global loss of H3K27me2/3 (Lewis et al., 2013; Chan et al., 2013). In addition to the known role of the modification in transcription (Cao et al., 2002; Xiong et al., 2016; Pirrotta, 2009; Margueron and Reinberg, 2011), it remains to be understood what role H3.1K27M plays in tumorigenesis.

Genome instability is a hallmark of tumour progression (Hanahan and Weinberg, 2011; Zhao et al., 2013; Liu et al., 2012; Wang et al., 2015) and the DNA damage repair machinery is the major guardian of genome integrity in mammalian cells (Li et al., 2016; Schultz et al., 2000). Of all types of DNA damage, DNA double-strand breaks (DSBs) are the most difficult to repair and therefore pose the greatest challenge to cells if left unrepaired (Halazonetis et al., 2008). DNA non-homologous end-joining (NHEJ) is the major DSB rejoining pathway that occurs in all cell cycle stages, while homologous recombination (HR) functions in S/G2 phase. PRC2 proteins have been shown to accumulate at sites of DSBs and knockdown of any component could result in an increase in radiation sensitivity (Chou et al., 2010; O'Hagan et al., 2008; Seiler et al., 2011; Johnson et al., 2015; Campbell et al., 2013). PRC2 was also required for ionizing-radiation-induced H2AK119ub foci and DSB-induced transcriptional silencing, thus promoting DSB repair (Kakarougkas et al., 2014). Moreover, a recent study has found the PRC2 complex to be involved in the regulation of Fanconi anaemia (FA) pathway activation by modulating chromatin states (Vierra et al., 2017). However, it remains unclear whether H3K27me2/3 also participates in DSB repair or which repair pathway is responsible. A more detailed elucidation of the function of H3K27me2/3 in the DNA damage response could help further explicate the mechanism of hypo-H3K27me2/3 in tumorigenesis.

In this study, we constructed the H3.1K27M mutant and established a hypo-H3K27me2/3 cell line to explore the role of H3K27me2/3 in DSB repair. We provide direct evidence that H3.1K27M or hypo-H3K27me2/3 significantly reduced 53BP1 foci number and NHEJ repair efficiency. Further investigation showed that H3.1K27M interacted with FANCD2 and loss of H3K27me2/3 promotes FANCD2 association on chromatin, which in turn upregulates recruitment of Tip60 (KAT5), elevates local H4K16ac levels and blocks H4K20me2, the docking site for 53BP1 (van Nuland and Gozani, 2016; Panier and Boulton, 2014). Our results demonstrate that deficiencies in DSB repair, stemming from H3.1K27M and hypo-H3K27me2/3, lead to genome instability, thus providing a new perspective on the pathogenesis of cancers related to H3.1K27M or low H3K27me2/3. Moreover, the reduction of H3K27me2/3 levels promoted the sensitivity of cells towards bleomycin treatment, which provides a novel therapeutic strategy for relevant cancers.

Chromatin remodelling is believed to be a critical step for DSB repair. In many cases, chromatin structure and nucleosomal organization often represent a barrier to the DSB repair machinery (Price and D'Andrea, 2013). This model has been supported by various lines of evidence, with our recent work and others showing that a reduction in condensed heterochromatin promotes DSB repair (Goodarzi et al., 2008; Chiolo et al., 2011; Chen et al., 2015; An et al., 2017). Histone H3 with hyper-methylation at lysine 27 causes another type of repressive chromatin and is often associated with gene silencing via polycomb group protein (PcG) repression (Pirrotta, 1998, 1997; Simon and Kingston, 2009, 2013). The PRC2 complex has been reported to be related to DSB repair (Campbell et al., 2013; Chou et al., 2010; Kakarougkas et al., 2014; Vierra et al., 2017), however, it is not clear whether the aberrant changes in H3K27me2/3 levels caused by pathological events, such as mutations in EZH2 or presence of H3.1K27M, may directly affect DSB repair efficiency, thus altering genome stability and potentially leading to carcinogenesis. To test this hypothesis, we established stable cell lines ectopically expressing haemagglutinin (HA)-tagged histone H3.1 wild-type (WT) or H3.1K27M mutant by transfecting human primary dermal fibroblasts (HDFs) with lentiviruses (Fig. S1A). In accordance with previous studies on histone H3.3 (Lewis et al., 2013; Funato et al., 2014; Chan et al., 2013; Bender et al., 2013), exogenous expression of H3.1K27M caused a marked reduction in H3K27 di- and tri-methylation (H3K27me2/3) (Fig. 1A). To examine whether the presence or formation of foci containing 53BP1, which serves as a marker for DSBs and promotes NHEJ-mediated DSB repair (Panier and Boulton, 2014), was affected, cell lines expressing H3.1WT and H3.1K27M at different time points after ionizing radiation (IR) treatment (X-ray, 8 Gy) were tested by immunofluorescence staining. There was a 40% reduction in the number of 53BP1 foci per nucleus at 5 min, and an ∼60% reduction at 30 min post IR in H3.1K27M mutant cell lines compared with H3.1WT cells (Fig. 1B,C). This reduction did not seem to be due to altered expression of 53BP1, as the level of 53BP1 was similar in cells expressing H3.1K27M versus H3.1WT before or after induction of DNA damage with bleomycin, a chemotherapeutic drug that produces DSB lesions (Aouida and Ramotar, 2010) (Fig. 1A, Fig. S1B).

To determine whether the reduction of 53BP1 foci number was due to the hypo-H3K27me2/3 state in H3.1K27M cells, we established a stable cell line with EZH2 knockdown. As expected, the number of 53BP1 foci was reduced by 40% and 41% at 5 min and 30 min after DSB induction, respectively, in cells stably expressing EZH2 shRNA (Fig. 1A-C). Collectively, these results suggest that the presence of H3.1K27M or a decreased H3K27me2/3 level prevents the formation of 53BP1 foci after IR treatment. We next examined the status of DNA-PKcs phosphorylation at T2609 (pDNA-PKcs), a modification of the key NHEJ factor that facilitates DNA end-processing and serves as a proxy for NHEJ-dependent DSB repair (Lieber, 2008). As shown in Fig. 1D-F, the number of pDNA-PKcs foci per nucleus was decreased in H3.1K27M cells and also in cells stably expressing EZH2 shRNA, which was consistent with the behaviour of 53BP1 and further suggested that NHEJ repair was inhibited in these cells.

To further determine the impact of hypo-H3K27me3 on NHEJ efficiency, we used a reporter cell line with a previously described GFP-based NHEJ reporter cassette (Mao et al., 2008, 2011). The reporter contains the GFP gene with a rat Pem1 intron (GFP-Pem1) that is interrupted by an adenoviral exon flanked by two non-palindromic I-Sce1 recognition sites in an opposite orientation; the gene is only activated when I-Sce1-induced DSBs are successfully repaired by NHEJ (Fig. 2A). The DNA-PKcs inhibitor NU7441 was used as a control to confirm the efficiency of the NHEJ reporter system (Fig. 2B). NHEJ efficiency was reduced by 30-40% when H3K27me3 levels were downregulated by either EZH2 siRNA (Fig. 2C) or an EZH2 inhibitor (GSK343) (Fig. 2D). We also tested the impact of other core subunits of the PRC2 complex in the modulation of NHEJ. Depletion of SUZ12 resulted in a similar inhibition of NHEJ to EZH2 siRNA or inhibitor (Fig. S1C). By contrast, siRNA against EED increased NHEJ efficiency over 2-fold (Fig. S1D). Since EED could also act independently of the PRC2 complex, for example, by interacting with histone deacetylases (HDACs) resulting in H3K27me3-independent repression (Ai et al., 2017), the observed change in NHEJ efficiency by EED knockdown may act via a different mechanism to knockdown or inhibit EZH2 or SUZ12.

To test whether the reduced NHEJ efficiency by EZH2 siRNA or inhibitor was due to loss of 53BP1 binding at DSB sites under hypo-H3K27me3 conditions, we performed a chromatin immunoprecipitation (ChIP) assay using NHEJ reporter cells. Primers were designed adjacent to the I-Sce1 recognition sites to enable the detection of chromatin-associated proteins in the vicinity of DSBs (ChIP-on-break) (Fig. S1E). Consistent with our previous immunostaining assay, the ChIP result showed that 53BP1 binding was decreased over 90% upon DSB induction by I-Sce1 (2 h) in hypo-H3K27me3 cells treated with EZH2 inhibitor (GSK343) (Fig. 2E,F), confirming that the reduction of NHEJ efficiency was due to dissociation of 53BP1.

We wondered whether the H3.1K27M- or hypo-H3K27me3-induced reduction in NHEJ was also accompanied by increased efficiency of HR, the other major DSB repair pathway (Chapman et al., 2012). The RAD51 foci formation was therefore investigated. However, we failed to detect an significant increase in the number of RAD51 foci per nucleus in H3.1K27M cells at 5 min, 30 min or 60 min post DNA damage induction (bleomycin 5 μg/ml, 4 h) (Fig. S1F). The similar results were also observed in cells with depletion of EZH2 (Fig. S1G). To further confirm our observation, we chose to use the previously described homologous recombination (HR) reporter system (Mao et al., 2008, Mao et al., 2011) (Fig. S1H) to examine whether HR efficiency was altered by the downregulation of H3K27me3 levels. Briefly, the HR reporter cassette contains two copies of mutated GFP-Pem1. The GFP exon in the first copy is inactivated by a 22-nt deletion and the insertion of two inverted I-Sce1 recognition sites, while the second copy is inactivated by the lack of a promoter, an ATG and the second exon of GFP (Fig. S1H). The deletion guarantees that GFP cannot be reconstituted by NHEJ. When I-Sce1-induced DSBs are repaired by HR, functional GFP will be reconstituted by gene conversion between the first GFP-Pem1 exons in the two mutated copies. The efficiency of the HR reporter system was confirmed by siRNA-mediated knockdown of BRCA2, a key component in HR repair (Powell et al., 2002) (Fig. S1I). The results showed that HR efficiency was moderately upregulated under hypo-H3K27me3 conditions after either EZH2 knockdown (Fig. S1J) or inhibition (Fig. S1K). Meanwhile, we also detected changes in H3K36me2/3 in these cells and found no obvious alterations (Fig. S1J,K, Fig. 2C,D). H3K36me3 has been reported to be involved in HR repair (Aymard et al., 2014) whereas H3K36me2 was shown to improve the association of early NHEJ repair components upon DSB induction (Fnu et al., 2011). Our results suggest that H3K36me2/3 is unlikely to be involved in the observed alteration of NHEJ or HR repair efficiency. Taken together, our results suggest that hypo-H3K27me3 induced either by H3.1K27M or loss of function of EZH2 leads to reduced DSB repair capacity through NHEJ rather than HR.

Since decreased H3K27me3 significantly inhibited NHEJ efficiency without an obvious increase in HR repair, we speculated that the induced DSBs might not be repaired efficiently, leading to cell death. We then evaluated the impact of H3.1K27M on apoptosis. In agreement with previous studies on H3.3K27M (Funato et al., 2014), the H3.1K27M mutation in HDFs increased the percentage of early apoptotic cells (Fig. 3A and Fig. S2A, P<0.05). Moreover, after exposure to the DSB-inducing chemotherapeutic drug bleomycin, there was a nearly 3-fold increase in early apoptotic cells in H3.1K27M compared with H3.1WT (Fig. 3A and Fig. S2A, P<0.01). The increased apoptosis seems to be due to the presence of tumour suppressor p53 (TP53), as upregulation of p53 protein was detected in H3.1K27M cells and was further increased after bleomycin treatment (Fig. 3B). When H3.1K27M cells were treated with p53 inhibitor (pifithrin-α), the percentage of early apoptotic cells was significantly decreased (Fig. 3C) and the number of proliferating cells was elevated nearly 2-fold at day 6 (Fig. 3D). The average number of γH2AX, RAD51 and pDNA-PKcs foci per nucleus showed a minor shift in these cells compared with levels in H3.1WT without bleomycin treatment, indicating that spontaneous replication stress may occur in H3.1K27M or hypo-H3K27me3 cells (Fig. S2B), leading to slower proliferation of H3.1K27M cells compared with H3.1WT cells (Fig. S2C).

The p53-dependent apoptosis of H3.1K27M cells indicated that genome instability may be involved, since p53 plays a central role in maintaining a stable genome in the face of toxic insults (Eischen, 2016). To test this idea, we first examined whether there were any changes in the foci of the phosphorylated histone variantH2AX (γH2AX-S139), a marker for DSB signalling cascade, after bleomycin treatment (5 μg/ml, 4 h). We found that both H3.1WT and H3.1K27M cells showed active γH2AX foci formation after bleomycin treatment (5 μg/ml, 4 h), which were eliminated more slowly in H3.1K27M cells compared with H3.1WT cells (Fig. S2D), supporting slower repair kinetics of drug-induced DSBs in H3.1K27M cells.

In order to evaluate the outcome of the unusual DSB response in H3.1K27M cells, we further tested its effects on genome stability. We performed exon sequencing of H3.1WT and H3.1K27M cells and found 24% more insertions and deletions (InDels) in H3.1K27M cells than in H3.1WT cells (36 versus 29, respectively). Moreover, there were also increased copy number variations (CNVs) as both gains and losses in H3.1K27M cells compared with H3.1WT cells (Fig. 3E,F). We reasoned that the increased rate of InDel mutations could be an indication that the genome of H3.1K27M cells was more fragile with loss of H3K27me2/3. Indeed, spontaneous replication stress was observed in these cells (Fig. S2B), which provides preferential sensitivity for the common fragile sites (CFSs) (Glover et al., 2017; Irony-Tur Sinai and Kerem, 2018). Collectively, these results indicate that the impaired DSB response in hypo-H3K27me2/3 cells eventually led to p53-dependent apoptosis and the accumulation of more genetic injuries.

Because H3K27me2/3 is known to be associated with transcriptional gene repression, we first performed RNA sequencing to determine whether the transcription of DNA-repair-related genes was altered in the presence of H3.1K27M. Consistent with the role of H3K27me2/3 in gene repression, there were 26.5% more upregulated genes than downregulated genes in H3.1K27M cells (Fig. S3A,B). There were similar changes of transcription profiles in cells stably expressing H3.1K27A, a mutation nontoxic to PRC2 and global H3K27me3 (Lewis et al., 2013), as well as in EZH2-knockdown cells (Fig. S3A,B). The gene ontology (GO) enrichment analysis for biological processes did not show a significant enrichment of DNA damage repair-related pathways in any of the three cell lines (Fig. S3C). A list of overlapping genes up- or downregulated more than 2-fold in H3.1K27M/A and EZH2 knockdown cells (Fig. S3D) is provided in Table S1, and none of these genes is known to be involved in DNA damage repair, in particular in the NHEJ pathway. We next performed western blot analysis with whole-cell extracts to compare the protein levels of key NHEJ repair genes in H3.1WT and H3.1K27M cells, and found no obvious alterations (Fig. 3B). RT-qPCR for the genes tested by western blotting was also analysed, and none of them showed changes over 2-fold (Fig. S3E). Collectively, these results suggest that hypo-H3K27me2/3-induced inhibition of NHEJ does not function by regulating DNA-repair-related gene expression.

Since our results indicated that the hypo-H3K27me2/3-induced inhibition of NHEJ did not function by regulating gene expression at either transcriptional or translational levels, we speculated that H3.1K27M might affect DSB repair efficiency by modulating the association of key repair proteins on adjacent chromatins. To further explore the underlying mechanisms by which hypo-H3K27me3 inhibits NHEJ, we conducted mass spectrometry (MS) to identify proteins specifically bound to exogenous H3.1K27 mutants (Fig. S4A-F, Table S2). Interestingly, we found one protein of particular interest, FANCD2, which has been reported to participate in DSB repair and facilitate DNA replication (Federico et al., 2016; Kais et al., 2016; Lossaint et al., 2013; Moldovan and D'Andrea, 2009; Kim and D'Andrea, 2012; Kottemann and Smogorzewska, 2013; Madireddy et al., 2016; Sato et al., 2012). We verified its interaction with H3.1K27M through co-immunoprecipitation. As shown in Fig. 4A, exogenous H3.1K27M showed much stronger association than H3.1WT with FANCD2 in nuclear extracts. Since nuclear extracts contain only free proteins (including histones) dissociated from chromatin, we next tested the association of FANCD2 with chromatin-incorporated H3.1K27M using nuclear pellets from HDF stable cell lines. As we expected, the results were similar to those using nuclear extracts (Fig. 4B).

Since we observed signs of replication stress in H3.1K27M cells (Fig. S2B), the EdU fluorescence signal (APC. mean) of S-phase cells was then examined. There was ∼20% higher EdU signal in H3.1K27 mutant cells (Fig. S5A), which demonstrated replication stress in these cells. We next examined the recruitment of MCM3 and Chk1 (Ge and Blow, 2010; Ge et al., 2007) by H3.1K27M. MCM3 is a replication licensing factor (RLF) and was previously reported to interact with FANCD2 upon activation of ATR signalling independently of FANCD2 monoubiquitylation (Lossaint et al., 2013), which restrains incorrect DNA synthesis under replication stress (Lossaint et al., 2013). Chk1 is another downstream effector of ATR signalling which blocks progression through S phase and triggers the G2/M checkpoint, thus preventing cells with DNA damage from entering mitosis (Xiao et al., 2003). In concert with the replication stress, an increased binding of MCM3 and Chk1 to mutant H3.1 was indeed observed in our co-immunoprecipitation assay (Fig. 4A). Taken together, these results demonstrate that the presence of H3.1K27M promotes the association of FANCD2 as well as the recruitment of MCM3 and CHK1 on chromatin.

It has been reported that the histone acetyltransferase Tip60 constitutively interacts with FANCD2 (Hejna et al., 2008). We wondered whether increased binding of FANCD2 on H3.1K27M promoted the association of Tip60 and change in H4K16ac, which has been suggested to prevent the binding of 53BP1 (Renaud et al., 2016). We therefore performed co-immunoprecipitation assays using nuclear pellets from HDF cells stably expressing H3.1WT or H3.1K27M. As shown in Fig. 4B, more Tip60 was recruited to chromatin-bound H3.1K27M than to chromatin-bound H3.1WT and the corresponding H4K16ac histone modification was upregulated accordingly.

The above results suggest a stronger affinity of H3.1K27M to FANCD2. However, it is unclear whether the enhanced association of these proteins is caused by the exogenous expression of H3.1K27M or the subsequent downregulation of H3K27me3. To resolve this, HDF or HL7702 H3.1WT cells were treated with an EZH2 inhibitor (GSK343) and then subjected to co-immunoprecipitation assays using nuclear extracts or pellets. As shown in Fig. 4C,D, GSK343 reduced H3K27me3 levels on exogenous H3.1WT and promoted the association of FANCD2, MCM3, Chk1 (also known as CHEK1) and Tip60 (KAT5). Moreover, the level of H4K16ac bound to hypo-H3K27me3 was significantly increased, whereas the local level of H4K20me2, a docking site of 53BP1, decreased as expected (Panier and Boulton, 2014) (Fig. 4D). Collectively, these results argue that loss of H3K27me3 is responsible for the enhanced association of FANCD2 with the chromatin and for the recruitment of Tip60. The resulting hypo-H4K20me2 levels and steric inhibition by upregulated H4K16ac levels both impede 53BP1 binding, thus inhibiting NHEJ in hypo-H3K27me3 cells.

We then wondered whether hypo-H3K27me2/3 chromatin altered the general distribution of FANCD2 in the nuclei. To do this, we fractionated the nuclei into chromatin and nucleoplasm (Mendez and Stillman, 2000; Yu et al., 2012) and analysed the distribution of FANCD2. The results showed that more FANCD2 was bound to the chromatin (P3) in H3.1K27M cells compared with H3.1WT cells (Fig. 4E), suggesting the chromatin association of FANCD2 was significantly altered. Furthermore, FANCD2 immunostaining showed that the fluorescence intensity of FANCD2 on fixed chromatin was significantly higher in H3.1K27M cells with or without bleomycin treatment (Fig. 4F,G). Taken together, these results suggested that H3.1K27M promoted the association strength of FANCD2 with the chromatin.

Since FANCD2 has been shown to act as a histone chaperone and maintain the mobility of histone H3 following treatment with an interstrand DNA crosslinker, mitomycin C (MMC) (Sato et al., 2012), we wondered whether the binding of FANCD2 to H3.1K27M could alter the mobility and exchange activity of histone H3. Surprisingly, the results of fluorescence recovery after photobleaching (FRAP) showed that the recovery kinetics of H3.1K27M-incorporated histones were obviously slower than control, with or without MMC treatment (Fig. S6A). The mobility of H3.1K27M-incorporated histones was inhibited without MMC treatment, indicating that it is unlikely to be a consequence of the increased binding of FANCD2 on these chromatins, and that the presence of FANCD2 on H3.1K27M chromatins is not sufficient to promote ICL repair since the mobility of H3.1K27M histones was almost the same before and after MMC treatment (Fig. S6A). To further examine the potential effects of hypo-H3K27me2/3 on ICL repair efficiency, we analysed the sensitivity of control and knockdown cells to MMC. The results showed that both H3.1K27M and shEZH2 cell lines were more sensitive to MMC compared with the control cell lines (HDF, H3.1WT and shNC) (Fig. S6B). Moreover, the monoubiquitylation of FANCD2 and FANCI, a central step in the activation of FA pathway (Smogorzewska et al., 2007; Sims et al., 2007; Garcia-Higuera et al., 2001), was also inhibited in H3.1K27M cells after MMC treatment (Fig. S6C). All these results suggest that ICL repair efficiency is inhibited in hypo-H3K27me2/3 HDF cell lines.

To provide further evidence of the role of FANCD2 in 53BP1 distribution in hypo-H3K27me3 cells, we attempted to rescue the formation of 53BP1 foci in H3.1K27M cells by knocking down FANCD2. As shown in Fig. 5A, H3.1K27M cells with siRNA-induced FANCD2 knockdown were more competent at 53BP1 foci formation compared with siNC-treated control cells at 5 min, 30 min and 60 min post bleomycin treatment, which was rescued to a level comparable to that in H3.1WT cells (Fig. 5A). To confirm the results, we analysed NHEJ repair efficiency using the reporter cell line (Fig. 2A), and the results showed that siRNA-mediated FANCD2 knockdown rescued the NHEJ efficiency in hypo-H3K27me2/3 cells (GSK343 treatment, 1 μM for 7 days) (Fig. 5B). We also examined the effect of FANCD2 knockdown on H3.1K27M cell apoptosis, and the apoptosis rate of H3.1K27M cells transfected with siFANCD2 was reduced after bleomycin treatment compared with levels in siNC cells (Fig. 5C; Fig. S5B) (P<0.05), suggesting a recovery of DSB repair efficiency. Interestingly, we performed co-immunoprecipitations and found that FANCD2 knockdown in H3.1K27M cells resulted in reduced association of Tip60 and MCM3; consequently, the K16ac modification decreased while K20me2 levels increased on the associated histone H4 (Fig. 5D). Altogether, these findings confirm that the association between H3.1K27M and FANCD2 is a key player in the inhibition of 53BP1 focus formation, as well as NHEJ efficiency.

Previous studies have reported that PRC2 is recruited to laser-induced DNA damage and contributes to DSB repair (Campbell et al., 2013; Chou et al., 2010; Kakarougkas et al., 2014). However, it remained controversial whether the corresponding histone modification, H3K27me2/3, participated in the process (Chou et al., 2010; Campbell et al., 2013). PRC2 has been found to be involved in DSB-induced transcription silencing as well as chromatin-state modulation in FA pathway activation (Kakarougkas et al., 2014; Vierra et al., 2017). However, it remains to be discovered whether PRC2 affects DSB repair by other means than acting as a transcriptional gene repressor. In this study, we constructed specific K-to-M point mutation of lysine 27 on DNA replication-dependent histone variant H3.1, which specifically reduced global H3K27me2/3 levels. Our results provide direct evidence that H3K27me2/3 is required for DSB repair and that loss of this histone modification results in ineffective NHEJ (Figs 1 and 2) as well as higher sensitivity to drug-induced DSBs (Fig. 3A).

We provide multiple lines of evidence that FANCD2 plays a central role in the hypo-H3K27me2/3-mediated inhibition of DSB repair (Figs 4-6). The FA pathway has been suggested to be critical in maintaining the timely, efficient and correct restoration of chromosomal integrity (Moldovan and D'Andrea, 2009), and FA components participate in the replication-dependent repair of multiple types of DNA lesions, via HR, nucleotide excision repair (NER) and translesion synthesis (TLS) (Moldovan and D'Andrea, 2009; Niedzwiedz et al., 2004; Sato et al., 2012).

FANCD2 was also found to interact transiently with MCM proteins and to stabilize stalled replication forks (Lossaint et al., 2013; Karanja et al., 2014). Our results show that the hypo-H3K27me2/3-mediated binding of FANCD2 is accompanied by the recruitment of the DNA replication factors MCM3 and Chk1 (Fig. 4A-D), suggesting that H3K27me2/3 may not only participate in DSB repair. A recent study showed that FANCD2 facilitates replication through common fragile sites (CFSs) (Madireddy et al., 2016), thus contributing to the maintenance of CFS stability. Our studies suggested that hypo-H3K27me2/3 alters the dynamics of FANCD2 binding and increases genome instability (Fig. 4A-D and Fig. 3E-G). The preferential binding of FANCD2 to hypo-H3K27me2/3 chromatin altered its distribution in the nuclei (Fig. 4E-G), which may deplete the pool of free FANCD2 in these cells, creating a phenotype similar to that seen upon loss of FANCD2. The overload of FANCD2 on chromatin may be correlated to the replication stress in hypo-H3K27me2/3 cells, which was accompanied by an involuntary response to p53 checkpoint and increased HR potential (Fig. S2D). HR has been found to be error-prone when rebuilding replisomes under replication stress (Carr and Lambert, 2013), which may also contribute to the genome instability observed in H3.1K27M cells (Fig. 3E-F). In addition, although the increased binding of Tip60 and the resulting heightened H4K16ac have been reported to alter DSB repair pathway choice by restricting the recruitment of 53BP1 and promoting the HR repair pathway (Tang et al., 2013), we found a non-significant increase of HR efficiency in hypo-H3K27me2/3 cells (Fig. S1E-K). We argue that the depletion of ‘free’ FANCD2 in hypo-H3K27me2/3 cells (Fig. 4E-G) might have restrained the HR repair pathway around collapsed replication forks. Rationally, the overall DSB repair capability was not enough to effectively repair DNA damage under hypo-H3K27me2/3 conditions, and the situation became visibly deteriorated once facing external DSB stimulations.

Our study also provides evidence for a novel role of hypo-H3K27me2/3 in tumorigenesis. It was demonstrated that EZH2 loss-of-function mutations or histone H3K27 mutations contribute to tumorigenesis by regulating transcription (Conway et al., 2015; Dawson and Kouzarides, 2012; Ezponda and Licht, 2014; Bender et al., 2013; Piunti et al., 2017; Mohammad et al., 2017), since H3K27me2/3 is a well-known repressor (Cao et al., 2002). Our findings provide a second perspective on the mechanism by which hypo-H3K27me2/3 or the presence of H3.1K27M mutant may contribute to cancer development: they impede effective DNA repair and alter genome homeostasis, as evidenced by genome instability in H3.1K27M cells and its association with a decline in overall DSB repair efficiency (Fig. 3E,F and Fig. S2A). The increased rate of InDel mutations suggest that the genome of H3.1K27M cells is possibly more fragile, with consistent loss of H3K27me2/3, supported by the presence of markers indicating spontaneous replication stress in these cells (Figs S2B and S5A). Replication stress was shown to result in preferential sensitivity for the CFSs (Glover et al., 2017; Irony-Tur Sinai and Kerem, 2018). Meanwhile, our results suggest that the histone H3.1K27M or H3.3K27M mutations are not sufficient to induce cancer (Funato et al., 2014) but rather place cells in a precancerous state characterized by replication stress and spontaneous DNA damage (Fig. S2D and Fig. S5A), where ineffective DSB repair promotes genome instability (Figs 1-3). We argue that cells harbouring more genome aberrations are more likely to die unless important tumour repressors are inactivated or oncogenes are activated (Funato et al., 2014; Castel et al., 2015; Bender et al., 2013); however, extensive adaption to hypo-H3K27me2/3- or H3.1K27M-induced replication stress could largely enhance accessibility to malignancy of the pre-cancerous cells and help them to avoid apoptosis. Indeed, when treated with the p53 inhibitor pifithrin-α, H3.1K27M cells showed decreased apoptosis and increased cell proliferation (Fig. 3C,D). Another line of evidence that supports this conclusion is that although H3.1K27M and H3.3K27M mutations were discovered in over 70% of DIPG patients, exogenous expression of H3.3K27M alone was insufficient to induce neoplastic transformation of neural progenitor cells (Funato et al., 2014). The combined transfection of PDGFRA and knockdown of p53 synergize with H3.3K27M to promote cell proliferation and survival (Funato et al., 2014). Although our HDF model expressing exogenous H3.1K27M may not fully represent the case of the endogenous H3.1K27M mutant regulation in DIPG, both studies at least support a consistent role of H3K27me2/3 in the control of cell survival and proliferation. Our further observation that a decrease in H3K27me2/3 showed enhanced sensitivity to bleomycin treatment (Fig. 3A) might suggest a therapeutic strategy to eliminate aberrant cells with precancerous lesions in such patients. Similar strategies could be applied to the treatment of cancers with aberrant hyper-H3K27me2/3. H3K27me2/3 levels could be reduced using EZH2 inhibitors (such as GSK343) before administration of chemotherapeutic drugs (such as bleomycin) or radiotherapy, which might strengthen the curative effect by promoting the sensitivity of tumour cells.

In summary, our study attempted to provide new insights into the underlying mechanism by which hypo-H3K27me2/3 affects tumorigenesis, thus enabling the development of novel anti-cancer treatments. The results confirm that the maintenance of normal cellular levels of H3K27me2/3 is critical as both overabundance and insufficiency of this histone modification are associated with malignancy.

Human cell lines were cultured in DMEM (Gibco) medium supplemented with 10% fetal bovine serum (FBS, Gibco), 100 U/ml penicillin and 100 μg/ml streptomycin (Gibco) at 37°C in 5% CO₂. Human primary dermal fibroblasts (HDFs) and the reporter cell lines for NHEJ and HR were kindly provided by Dr Zhiyong Mao (Tongji University, China) (Wang et al., 2015; Mao et al., 2011). HL7702 cell line (Wang et al., 2017) was purchased from the Chinese Academy of Sciences. The two reporter cell lines used for analysing NHEJ and HR, HCA2-I9a and HCA2-H15c, were grown in DMEM (Gibco) medium supplemented with 10% fetal bovine serum (FBS, Gibco), 1% non-essential amino acids (Gibco), and 100 μg/ml streptomycin (Gibco) in a Hera240i incubator with 5% CO₂ and 3% O₂ at 37°C. For EZH2 inhibition, HDF cells were cultured in growth medium with 1 mM GSK343 (Selleckchem) for 7 days. To induce DSBs, cells were treated with 8Gy X-ray or cultured in growth medium with 5 μg/ml bleomycin (Selleckchem) for 4 h.

The coding sequence of histone H3.1 was cloned from a human cDNA library and used in point mutagenesis to produce H3.1K27M/A mutants, which were then tagged with HA-coding sequence at the 3′-terminus and cloned into pEF-CSII-IRES-EGFP. These plasmids were used for lentivirus particle production. The H3.1 WT or H3.1K27M/A mutant sequences were cloned into pCAG-EGFP-N1 vector and fused with EGFP at C-termini. The resulting plasmids were used to transfect HL-7702 cells. Short hairpin interference RNA (shRNA) against EZH2 was synthesized and ligated into GV-248 (Genechem), which was driven by a U6 promoter and could be sorted by EGFP as a positive reporter.

shRNA sequence targeting EZH2 is as follows: 5′- GGACTAGGGAGTGTTCGGTG -3′. For siRNA-mediated knockdown, the following target sequences were used: siFANCD2, 5′-UACCUCAAGUGUAUCCAUG-3′; siEZH2-1, 5′- CCCAACAUAGAUGGACCAAAU -3′; siEZH2-2, 5′- GCUAGGUUAAUUGGGACCAAA -3′; siEED, 5′- AGCCAUGGAAAUGCUAUCAAUG -3′; siSUZ12, 5′- ACCTGTATGCTGTTTGTAGAAA -3′. All siRNAs and negative control (NC) siRNA were purchased from GenePharma (Shanghai, China).

The lentivirus vectors and helper vectors (pCMV-dR8.91 and VSV-G) were transfected into HEK293t cells. After 48 h, the cell culture supernatants were harvested, filtered and concentrated. Virus titres were measured and stored at −86°C as aliquots. To generate stable cell lines expressing exogenous HA-tagged H3.1WT or H3.1K27M/A mutants, the concentrated lentiviruses (∼10⁷ IFU) were added to 100 µl HDF or HL7702 single-cell suspensions (10⁵ cells) for 1.5 h at 37°C in 5% CO₂. The suspensions were then diluted with 2 ml growth medium and cultured in 3.5 cm dishes for 24 h before replacing with fresh medium. After 5 to 7 days, the transduction-positive cells were sorted by flow cytometry using a GFP reporter (excitation/emission at 488/509 nm) as a selection marker. The percentage of positive cells could reach over 90% after two to three rounds of FACS sorting.

The GFP-tagged vectors with CAG promoters were transduced into HL7702 cells using Lipofectamine 2000 reagent (Invitrogen). 48 h after transfection, positive cells were selected using G418 (3.5 mg/ml, Invitrogen) for 10 to 14 days before sorting. The G418 concentration was reduced to 0.5 mg/ml as selection pressure.

To transfect siRNA into HDF stable cell lines, 5×10⁵ cells were inoculated into 10 cm dishes 48 h prior to transfection with 5 μg siRNA using the electroporation kit (V4XP-2032) on a Lonza 4D-Nucleofector System (Cologne, Germany) (program DT-130). The transfected cells were cultured for 48 h before the second round of transfection and cultivated 48 h later for FACS, immunofluorescence staining or co-immunoprecipitation.

Cells were collected and disrupted in 1.0% NP-40 cell lysis buffer [50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1.0% NP-40, 0.2 mM EDTA, 20% glycerol, 0.1% SDS supplemented with protease and phosphatase inhibitors (Roche)]. After incubation for 30 min on ice, the lysates were combined with 5× loading buffer containing β-2-mercaptoethanol and denatured by boiling. The proteins were separated by SDS-PAGE (He et al., 2016). Primary antibodies against the following proteins were used at the indicated dilutions: HA, 1:1000 (Abcam, ab9110); γ-H2AX, 1:1000 (Abcam, ab20669); Ligase IV 1:1000 (Abclonal, A1743); Ku 70, 1:1000 (Abclonal, A7330); Ku 80, 1:1000 (Abclonal, A5862); XRCC4, 1:1000 (Abclonal, A7539); 53BP1, 1:500 (Novus, NB-100-305); DNA-PKcs, 1:1000 (Abcam, ab70250); RAD51, 1:1000 (Abcam, ab176458); RPA2, 1:1000 (Abclonal, A2389); BRCA1, 1:1000 (Cell Signaling Technology, 9010S); MRE11, 1:1000 (Abclonal, A2559); FANCD2, 1:500 (Bioworld, BS6604) and 1:100 (Santa Cruz Biotechnology, sc-20022); FANCI, 1:100 (Santa Cruz Biotechnology, sc-271316); EZH2, 1:1000 (Cell Signaling Technology, 5246S); GFP, 1:1000 (Abcam, ab290); MCM3, 1:1000 (Abcam, ab4460); CHK1, 1:1000 (Proteintech, 10362-1-AP); TIP60, 1:1000 (Abcam, ab23886); H3, 1:1000 (Abcam ab1791); H3K27me3, 1:1000 (Abcam ab6002); H3K27me2, 1:1000 (Abcam, ab24684); H3K27ac, 1:1000 (Abcam, ab4729); H4K20me2, 1:1000 (Abcam, ab9052); H4K16ac, 1:1000 (Millipore, 17-10101); H3K36me3, 1:1000 (Abcam, ab9050); H3K36me2, 1:1000 (Abcam ab9049). GAPDH at 1:2000 (Novus, NB300-221) and β-actin at 1:5000 (Santa Cruz Biotechnology, SC47778) were used for internal loading control of total proteins.

Cells were plated and cultivated on glass coverslips, fixed in 4% formaldehyde for 15 min at room temperature and permeabilized in 0.2% Triton X-100 for 10 min. After blocking with 3% BSA in PBS, the cells were stained at 4°C overnight with antibodies against: 53BP1 (Novus, 1:400), pDNA-PKcs (phospho T2609) (Abcam, 1:400), γH2AX (Abcam, 1:400) or RAD51 (Abcam, 1:400). Primary antibodies were detected by incubation with anti-rabbit or anti-mouse Alexa Fluor 594 secondary antibodies (Invitrogen) for 30 min at room temperature. The cells were incubated with DAPI (1000×, Invitrogen) for 5 min at room temperature to visualize the nucleus. Slides were examined at 63× magnification using a fluorescence microscope (Leica) with a set parameter for each series of experiments (Pan et al., 2015). Image processing and analysis were performed using Image-Pro Plus software (Media Cybernetics). At least 50 cells were counted for each time point.

FRAP was performed as described (Kimura et al., 2006; Sato et al., 2012) using a confocal microscope (Leica TCS SP5 II). Cells were treated with or without 50 ng/ml MMC for 16 h (Sato et al., 2012). Three to five confocal images of a field containing 1-3 nuclei were collected and one half of each nucleus was bleached using 100% transmission at 488 nm. Images were obtained using the setting at 5 min intervals. The relative fluorescence intensity of 6 nuclei were analysed for each group.

A total of 5×10⁴ cells were plated in each well of a 6-well plate in triplicate for each cell line. After 24 h, MMC (Selleckchem) was added at final concentrations of 0 nM, 25 nM, 50 nM, 100 nM, 200 nM and 500 nM. Drug-containing medium was replaced 2 h later. Cells were counted after 5 days and the cell numbers at each dose of drug were divided by the cell number in the untreated sample to calculate the percentage survival (Peng et al., 2014; Kim et al., 2013).

In vivo analysis of NHEJ and HR efficiency was performed as described previously (Mao et al., 2011). Briefly, the reporter cell lines were co-transfected with I-Sce1, DsRed and the target siRNA in triplicate for each group. After incubation for 72 h, cells were collected and resuspended in ice-cold PBS for analysis on a FACSVerse (BD BioSciences). At least 20,000 cells were counted, and the data were further analysed using FlowJo software (Ashland).

Cells were treated in triplicate for each group with or without 10 μg/ml bleomycin (Selleckchem) for 24 h before collection. After washing twice with ice-cold PBS, cells were resuspended in 500 μl 1×Binding Buffer (eBioscience) at a concentration of 1×10⁶ cells/ml, and incubated with 5 μl APC-Annexin V (eBioscience) for 15 min followed by a 5 min incubation with 5 μl propidium iodide (eBioscience) at room temperature in the dark. Flow cytometry was performed within 1 h on a FACSverse (BD Bioscences). At least 10,000 cells were counted and the data was further analysed using FlowJo software (Ashland).

For genome DNA extraction, no less than 1×10⁶ cells were collected and DNA was purified using TIANamp genome extract kit (Tiangen Inc, DP304). The resulted genome DNA was tested for integrity by agarose gel electrophoresis and quantified using Qubit (Thermo Fisher Scientific), sheared with a sonicator (Covaris LE220) to generate 150-200 bp fragments and targeted for DNA library construction by Sureselect^XT reagent kit (Agilent). The enriched DNA libraries were quantified by the Qubit 2.0 fluorometer dsDNA HS Assay (Thermo Fisher Scientific) and analysed for size distribution using a BioAnalyzer 2100 (Agilent). Finally, the libraries were sequenced using an Illumina HiSeq X-10 system following protocols for 2×150 paired-end sequencing using the Genome Center at WuXi App Tec. (Shanghai, China) Sequencing depth was set at 100×.

Sequence reads were mapped to the build hg19 using Burrows-Wheeler Aligner version 0.7.15 (Li and Durbin, 2009). After the duplicated reads were marked using Picard, sequence reads were realigned using GATK (McKenna et al., 2010). IndelRealigner and quality scores were recalibrated by GATK Table Recalibration. InDels were called by VarScan (Koboldt et al., 2012) version 2.4.0 with parameter, min-coverage 10 and only outputting the high-confidence InDels. The false positives of high-confidence InDels were filtered by bam-readcount and fpfilter.pl of the VarScan package. CNVs were called by EXCAVATOR2 (D'Aurizio et al., 2016) using default parameters and the ideogram were drawn by Circos (Krzywinski et al., 2009). Exon sequencing data have been deposited with the European Nucleotide Archive (ENA) under accession number PRJEB21013. Genome Analysis Toolkit (GATK) used for exon-seq data process is available at https://software.broadinstitute.org/gatk/.

Total RNA was extracted using RNAiso plus reagent (TAKARA, Cat#9109), dissolved in DEPC-treated water and evaluated by the ratio of 28S/18S (RIN>8, Agilent 2100) for RNA integrity. RNA concentrations were detected by Qubit (Thermo Fisher Scientific) and 2 μg RNA per sample was used as input materials. Sequencing libraries were generated using NEBNext^®Ultra™ RNA Library Prep Kit for Illumina^® (NEB, USA) following the manual instructions and assessed for quality on the Agilent Bioanalyzer 2100 system. Index codes were added to attribute sequences to each sample, and clustering of the coded samples was performed on a cBot Cluster Generation System using TruSeq SR Cluster Kit v3-cBot-HS (Illumina). After cluster generation, sequencing was performed on an Illumina Hi-seq X-10 platform or HiSeq 2000 instrument (Novogene, Beijing, China) to generate 150 bp pair-ended reads. Sequence reads were trimmed to exclude adaptors and low quality bases using Cutadapt software (version 1.10, parameter: –q 20 –m 50), aligned to mm10 by TopHat software (v.2.0.13) using the default parameter to reserve only the highest MAPQ and paired by samtools view –q 50 –f 0×2 for further analysis. Prediction of differentially expressed genes was performed by Cuffdiff2 (v.2.2.1) with the filtration of fold change ≥2.0 and q-value <0.05. GO term enrichment analysis is an open source collaborative initiative available from David: https://david.ncifcrf.gov. RNA sequencing data have been deposited with the NCBI Gene Expression Omnibus (GEO) under accession number GSE98768.

No less than 1×10⁷ cells were collected and incubated with 37% formaldehyde diluted to a 1% final concentration for 15 min at room temperature in order to crosslink the protein-DNA complex. Formaldehyde was quenched by adding 1 M glycine diluted to a final concentration of 125 mM and samples were rocked for 5 min at room temperature. Cell pellets were resuspended in cell lysis buffer [5 mM HEPES, pH 8.0, 85 mM KCl, 1% NP-40, protease inhibitors (EDTA-free, Roche)] and incubated on ice for 15 min. To enhance nuclei release, the cell suspension was homogenized using a glass Dounce homogenizer (type B) and centrifuged at 1500 rpm for 5 min at 4°C. The pellets were resuspended in nuclei lysis buffer [50 mM Tris-Cl pH 8.1, 10 mM EDTA, 1% SDS, protease inhibitors (EDTA-free, Roche)], sonicated for 30 min (30 s on/30 s off) using a Bioruptor Twin (UCD-400, Diagenode, Denville, NJ) to shear the DNAs to an average length of 200-500 bp, and centrifuged at 20,000 g for 10 min. The supernatants were collected and quantified by Qubit assay (Invitrogen). Suitable amount of chromatins was diluted with 4-fold dilution buffer (50 mM Tris-HCl, pH7.4, 150 mM NaCl, 1%NP-40, 0.25% deoxycholic acid, 1 mM EDTA pH8, protease inhibitors) and incubated on a rocker at 4°C overnight with 2 μg of one of the following antibodies: H3K27me3 (Abcam, ab6002), 53BP1 (Novus, NB-100-305). Protein A/G-agarose (20 μl, Santa Cruz Biotechnology sc-2003) was added and incubation continued for 2 h before extensively washing with wash buffer 1 (50 mM Tris-HCl, pH7.4, 150 mM NaCl, 1%NP-40, 0.25% Deoxycholic acid, 1 mM EDTA pH8), wash buffer 2 (100 mM Tris-Cl pH 9.0, 500 mM LiCl, 1% NP-40, 1% deoxycholic acid), wash buffer 3 (100 mM Tris-HCl, pH 9.0, 500 mM LiCl, 1% NP-40, 1% deoxycholic acid, 150 mM NaCl) and TE buffer. The immune-captured chromatins were eluted with elution buffer (50 mM NaHCO₃, 1% SDS) and the resulting DNA reverse-crosslinked by incubating at 65°C overnight. Finally, the immunoprecipated DNA was treated with RNaseA and proteinase K and purified by ethanol precipitation with the assistance of nucleic acid carriers [in house with 5 μg/μl linear polyacrylamide (LPA)].

Quantitative PCR (q-PCR) was performed using a Bio-Rad IQ5 device. The primers used were: forward, 5′- CCTGAAGATTTGGGGGATTGTGCTTC -3′; reverse, 5′- CTTGGAAACACCCATGTTGAAATATC -3′.

Stable cell lines expressing exogenous histone mutants were processed in a required manner before they were cultivated. At least 1×10⁷ cells were collected and resuspended in 5-fold packed-cell volumes of hypotonic buffer [10 mM HEPES, pH 7.9, 1.5 mM MgCl2, 10 mM KCl, 0.5 mM DTT, 0.2 mM PMSF and proteinase inhibitors (EDTA-free, Roche)]. After incubation on ice for 10 min, the cell suspensions were transferred into a Dounce homogenizer (type A) and lysed with ∼10 strokes before centrifugation at 1000 g for 15 min. The pelleted nuclei were resuspended in low-salt buffer (20 mM HEPES, pH 7.9, 25% glycerol, 20 mM KCl, 1.5 mM MgCl₂, 0.2 mM EDTA, 0.5 mM DTT, 0.2 mM PMSF and proteinase inhibitors) and mixed with 1/4 volume of high-salt buffer (20 mM HEPES, pH 7.9, 25% glycerol, 1.2 M KCl, 1.5 mM MgCl₂, 0.2 mM EDTA, 0.5 mM DTT, 0.2 mM PMSF and proteinase inhibitors) on a rocker for 30 min at 4°C. Finally, centrifugation was performed at 25,000 g for 15 min at 4°C to separate nuclear extracts (NE, the supernatants) and nuclear pellets (NP, the pellets). The supernatants were further diluted with HEPES-glycerol buffer to level KCl to 150 mM, and the pellets were digested with benzonase buffer [50 mM Tris-HCl, pH 7.4, 0.5% NP-40, 10% glycerol, 150 mM NaCl and 2 mM MgCl₂, protease and phosphatase inhibitors, 25 U/ml Benzonase (Sigma)] at 4°C overnight before the co-immunoprecipitation experiments.

Co-IPs were performed by adding GFP-Trap beads (Chromo Tek) or Anti-HA affinity gel (Biotool) to either of the two kinds of nuclear components (NEs or NPs) according to the tags in different cell lines and incubating for 2 h at 4°C. Unbound proteins were extensively cleaned by adding washing buffer (20 mM HEPES, pH 7.9, 20% glycerol, 100 mM KCl, 0.2 mM EDTA, 0.5 mM DTT, 0.2 mM PMSF and proteinase inhibitors). The resulted proteins were subjected to western blot analysis.

Chromatin isolation assay was performed as previously described (Yu et al., 2012; Mendez and Stillman, 2000). Briefly, cells were resuspended in buffer A (10 mM HEPES, pH 7.9, 10 mM KCl, 1.5 mM MgCl₂, 0.34 M sucrose, 10% glycerol, 0.1% Triton X-100, 1 mM DTT and proteinase inhibitors) for 5 min on ice and then centrifuged (5 min, 1300 g, 4°C) to collect the nuclei. The nuclei were washed once in buffer A and lysed in buffer B (3 mM EDTA, 0.2 mM EGTA, 1 mM DTT and proteinase inhibitors) for 30 min. After centrifugation (5 min, 1700 g, 4°C), the supernatants (S2) were collected and the pellets were washed once in buffer B and centrifuged again to collect the final chromatin fraction (P3). Both S2 and P3 were subjected to western blot analysis.

We thank Prof. Or Gozani (Stanford University, USA) and Prof. Jian-Quan Ni (Tsinghua University, China) for their kind and generous help with manuscript preparation.