Inhibition of KDM6 activity during murine ESC differentiation induces DNA damage

Embryonic development with coordinated lineage specification of stem and progenitor cells requires dynamic packing and unfolding of chromatin to initiate and manifest tissue-specific gene expression patterns (Teif et al., 2012). This process includes the reversible modification of histone tails. In ESCs, distinct histone tail methylation patterns are associated with pluripotent and differentiated states. Among these modifications, tri-methylation (me3) of histone 3 residue K4 (H3K4), di-methylation (me2) of H3K79 and mono-methylation (me1) of H4K20 show developmental stage-specific dynamics that correlate with gene activation, whereas the H3K27me2 and H3K27me3 modifications are associated with gene silencing (Zhou et al., 2011). Although H3K4me3 and H3K27me3 are associated with opposing transcriptional states, they are at times found on the same target sites (Azuara et al., 2006; Bernstein et al., 2006; Roh et al., 2006). This bivalent modification state primes developmental genes for rapid transcriptional activation or repression (Bernstein et al., 2006).

The formation of the H3K27me3 mark is catalysed by the Polycomb Repressive Complex (PRC)2. H3K27me3 is not randomly distributed along the genome but decorates large domains exceeding 100 kb that comprise high gene and short interspersed nuclear element (SINE) densities (Pauler et al., 2009). Chromatin immunoprecipitation sequencing (ChIP-seq) analyses have revealed that a continous H3K27me3 marking across the body of genes corresponds to transcriptional inhibition, whereas a H3K27me3 peak around the transcriptional start site marks bivalent domains and, in contrast to the classic concept of H3K27me3 function, a peak in the promoter region is associated with active transcription (Young et al., 2011). H3K27me3 is also enriched in subtelomeric regions and in long-terminal-repeat retrotransposons (Leeb et al., 2010; Rosenfeld et al., 2009). Demethylation of H3K27 by specific demethylases represents a means to regulate the coordinated transcriptional activation of lineage-affiliated genes in response to developmental signals, whereas unused programs become packed into heterochromatin (Pedersen and Helin, 2010). Recently, ‘ubiquitously transcribed tetratricopeptide repeat’ on chromosome X (UTX; also known as KDM6A) and jumonji D3 (JMJD3; also known as KDM6B) have been identified as histone H3K27-specific demethylases (Agger et al., 2007). KDM6A and KDM6B catalyse the transition of H3K27me3 and, in vitro, also H3K27me2 to H3K27me1 (Agger et al., 2007; De Santa et al., 2007; Lan et al., 2007; Lee et al., 2007). Both KDM6 proteins are components of the mixed lineage leukemia (MLL)–COMPASS-like complex (Cho et al., 2007; De Santa et al., 2007; Issaeva et al., 2007). A paralog of the X-chromosome-located KDM6A, termed UTY (also known as KDM6C), is located on the Y-chromosome. It is so far unknown whether UTY possesses in vivo demethylase activity (Hong et al., 2007; Walport et al., 2014). KDM6A and KDM6B take part in the regulation of gene expression programs during development (Burgold et al., 2008,, 2012; Lee et al., 2012; Seenundun et al., 2010). Embryos in which KDM6B has been knocked out die perinatally owing to respiratory failure. A catalytically inactive KDM6B mutant fails to rescue this phenotype, supporting the notion that KDM6B enzymatic activity is crucial for the establishment of respiratory function (Burgold et al., 2012). Knockout (KO) embryos of KDM6A fail to activate expression of the early mesodermal marker brachyury (gene symbol T), and embryos show severe defects in the development of the notochord, hematopoietic and cardiac tissues (Lee et al., 2012). The KDM6A-KO phenotype is partially rescued by UTY in male embryos or by expression of a catalytically inactive KDM6A protein, suggesting a demethylase-independent function of KDM6A (Lee et al., 2012; Morales Torres et al., 2013; Thieme et al., 2013; Wang et al., 2012). So far it is unknown to what extent the demethylase activities of KDM6A and KDM6B can compensate for each other. A recent report suggests that KDM6B and UTY can rescue the loss of KDM6A in a demethylase-dependent and -independent manner, respectively (Morales Torres et al., 2013).

To analyse the functional consequences of combined KDM6A and KDM6B inhibition in ESCs and during ESC differentiation, we used the recently described specific small-molecule inhibitor GSK-J4 (Kruidenier et al., 2012) and employed KDM6A-KO ESCs and knockdown (KD) of KDM6B. We report that differentiating but not undifferentiated ESCs are crucially dependent on KDM6A and KDM6B enzymatic activities. In differentiating ESCs, combined KDM6A and KDM6B inhibition induces DNA damage.

To analyse the requirement for the combined enzymatic activities of KDM6A and KDM6B in undifferentiated and differentiating wild-type (WT) ESC embryoid bodies, we employed the specific cell-permeable small-molecule inhibitor GSK-J4 (Kruidenier et al., 2012). As assessed by western blot analyses using an H3K27me3-specific antibody, a two- to three-fold increase in H3K27me3 levels was observed in undifferentiated and differentiating ESC cultures following treatment with GSK-J4. In contrast, no changes in H3K27 methylation were observed when cells were treated with DMSO or with GSK-J5, an inactive regio-isomer of GSK-J4 (Fig. 1A; see Fig. S1A for densitometry analyses). The effect of GSK-J4 was restricted to the H3K27me3 mark as H3K27me2, H3K4me3 and HK4me2 levels remained unaltered (Fig. 1A; see Fig. S1A for densitometry analyses). The elevated H3K27me3 levels cannot be attributed to the loss of KDM6A or KDM6B expression as KDM6A and KDM6B were expressed at constant or higher levels in GSK-J4-treated ESCs and embryoid bodies (Fig. S1B). UTY was expressed at constant levels in GSK-J4-treated ESCs and embryoid bodies (Fig. S1B). To analyse the effect of combined KDM6A and KDM6B inhibition on ESC proliferation, we monitored cumulative population doublings in the presence or absence of GSK-J4. We observed that in comparison to GSK-J5- or DMSO-treated ESCs, GSK-J4 treatment had no effect on the cumulative population doublings (Fig. 1B, left). Similarly, ESC pluripotency as assessed by alkaline-phosphatase-positive colony formation, RNA expression of Oct4, Sox2 and Nanog, and SSEA-1 surface staining remained unaffected by GSK-J4 treatment (Fig. S1C–E). In contrast, GSK-J4 affected ESCs that had been cultured under differentiation conditions as the cellularity of embryoid bodies rapidly declined under GSK-J4 treatment (Fig. 1B, right). Although continuous GSK-J4 treatment of undifferentiated ESCs did not yield increased apoptotic cell numbers (Fig. 1C, left), we noticed a two- to three-fold increase in apoptotic cell numbers at day 2 in differentiating cultures when treated with GSK-J4 (Fig. 1C, right). In agreement with previously published work (Qi et al., 2015), caspase-3 activation can be observed during embryoid body formation as early as 2 days after leukemia inhibitory factor (LIF) withdrawal, consistent with an increased apoptotic rate. We, however, noticed higher cleaved caspase-3 levels in GSKJ-4-treated day 2 embryoid bodies than in untreated cells (Fig. 1D). Furthermore, the cell cycle distributions of undifferentiated ESCs cultured in the presence of GSK-J4 were comparable to the distribution of untreated control cultures (Fig. 1E, left). In contrast and consistent with the reduced cell numbers, we observed an accumulation of cells in G1 along with a reduced fraction of S-phase cells in GSK-J4-treated embryoid body cultures (Fig. 1E, right). In conclusion, inhibition of KDM6 protein activity affected viability and the cell cycle progression of differentiating ESCs but not of undifferentiated ESCs.

Next we sought to elucidate the underlying cause of the GSK-J4-induced effects during ESC differentiation. To this end, we first performed whole transcriptome analyses of ESC differentiation cultures that had either been treated with GSK-J4, GSK-J5 or DMSO (Fig. 2A). We chose to analyse an early time point (day 0.75) of differentiation to avoid apoptotic cells. In total, 303 genes were differentially expressed [false discovery rate (FDR)<0.05] in the presence of GSK-J4 (Fig. 2A; Table S1). Fewer genes were down- (75) than upregulated (228). Gene ontology analysis (Fig. 2B) revealed that a fraction of the differentially regulated genes were categorised as being associated with developmental processes (77). Additional germ-layer-specific differentiation markers that had not been identified in the global expression analyses showed no differential expression under GSK-J4 treatment (Fig. S2A). A substantial fraction of responsive genes was categorised as being associated with cell death (44) or cell proliferation (33). Consistent with the observed accumulation of GSK-J4-treated differentiating ESCs in the G1 phase, Ccnd2 (cyclin D2) was downregulated, whereas the cyclin-dependent kinase inhibitor Cdkn1a (p21) and the retinoblastoma protein RB1 were upregulated (Table S1). We further noticed that several of the cell-death- and/or cell-proliferation-associated upregulated genes (Mdm4, Polk, Btg2, p21CIP, Trp53inp1, Ddit3, Ddit4, cdc42bpg, RB1, Rbl2) were also associated with DNA damage. Quantitative real-time (qRT)-PCR analyses for a limited gene set confirmed the results obtained from global transcriptome analyses (Table S1, Fig. S2B).

DNA damage signalling is a potent inducer of cell cycle inhibition and apoptosis (Horn and Vousden, 2007). Therefore, we asked whether DNA damage response (DDR) signalling was activated in ESCs differentiating in the presence of GSK-J4. Ataxia telangiectasia mutated kinase (ATM) and ataxia telangiectasia and Rad3-related kinase (ATR) are central components of DNA damage signalling, and they phosphorylate a multitude of substrates, including H2AX at serine residue 139 (S139) or p53 at serine residue 15 (S15) in response to DNA damage (Matsuoka et al., 2007). Phosphorylated H2AX (S139) (γH2AX) marks DNA double-strand breaks (DSB), whereas phosphorylated p53 (at S15) is involved in the regulation of cell cycle progression and survival upon DNA damage (Horn and Vousden, 2007; Paull et al., 2000). We therefore analysed γH2AX and phosphorylated p53 (S15) levels in response to GSK-J4 treatment, and separated protein extracts of ESCs or embryoid bodies that had been either cultured in the presence of DMSO, GSK-J5 or GSK-J4 by using SDS-PAGE, which was followed by western blotting (Fig. 2C). Probing of the membranes with antibodies specific for γH2AX or phosphorylated p53 (S15) revealed that in agreement with a previous report, basal γH2AX levels were higher in ESCs than in embryoid bodies (Banáth et al., 2009). Similarly, phosphorylated p53 (S15) levels were also elevated in ESCs. No increase in γH2AX or phosphorylated p53 (S15) levels was observed as a consequence of GSK-J4 treatment in undifferentiated ESCs. In contrast, elevated γH2AX and phosphorylated p53 (S15) levels were detected in GSK-J4-treated embryoid bodies. Phosphorylation of p53 at serine 15 leads to reduced binding of MDM2 and as a consequence to stabilisation of the p53 protein (Canman et al., 1998; Banin et al., 1998). In agreement with this, we noticed increased p53 protein levels in GSK-J4-treated embryoid bodies, whereas p53 levels remained unaltered in GSK-J4-treated ESCs (Fig. 2C). Additional experiments employing an antibody specific for substrates phosphorylated by ATM and ATR revealed at least seven differentially phosphorylated ATM and ATR substrates in embryoid bodies that had been treated with GSK-J4 (Fig. S2C).

Next we asked whether the increased DNA damage signalling in the presence of GSK-J4 is paralleled by an increased incidence of DNA breaks. We therefore performed alkaline comet assays with cells isolated from embryoid bodies that had been differentiated in the presence of GSK-J4, GSK-J5 or DMSO (Fig. 2D). To avoid comets originating from apoptotic cells that could perturb the results (Choucroun et al., 2001), we analysed cells from 1-day-old embryoid bodies because apoptosis was not detectable at this time point (Fig. 1C,D). As a positive control for DNA damage induction, we used cells that were isolated from day-1 embryoid bodies after exposure to 6 Gy of ionizing radiation (IR). GSK-J4-treated embryoid bodies showed a strong increase in tail moment values as compared to those formed from GSK-J5- or DMSO-treated cells. The tail moment value of GSK-J4-treated cells was comparable to the tail moment value of 6-Gy-irradiated cells. Analyses of comet formation in embryoid bodies that had been cultured in the presence of DMSO, GSK-J5 or GSK-J4 for 12 or 18 h showed a time-dependent increase in tail moment values in the GSK-J4-treated cells (Fig. S2D). The underlying cause for the increase in DNA damage might be an impaired capacity of the GSK-J4-treated cells to repair DNA damage. To start addressing this, we tested whether one of the main pathways of DSB repair, homologous recombination, is impaired in the presence of GSK-J4. To this end, we employed a previously described system with chromosomally integrated SceI-inducible DSBs (Pierce et al., 2001). The results show that homologous recombination is utilised in differentiating ESCs and that it is unaltered in the presence of GSK-J4 (Fig. 2E). The MRE11 inhibitor Mirin reduced homologous recombination frequencies, as expected. We conclude that the increase in DNA damage signalling in differentiating ESCs in the presence of GSK-J4 is a consequence of DNA damage and that impaired homologous recombination is not the underlying cause for increased DNA damage.

To analyse whether the effect of GSK-J4 on viability, cell cycle progression and DNA damage signalling is a direct consequence of combined inhibition of KDM6A and KDM6B, or a consequence of off-target effects, we knocked down KDM6B in KDM6A-KO ESCs. The knockout of KDM6A led to a complete loss of KDM6A expression on RNA and protein levels, and the knockdown of KDM6B was evident in RNA and protein levels (Fig. 3A,B; Fig. S3A). Consistent with the results obtained with GSK-J4-treated WT ESCs, proliferation and cell cycle distribution as well as SSEA1 surface marker expression of KDM6A-KO KDM6B-KD ESCs were identical to those of control ESCs (Fig. 3A; Fig. S3B,C). In contrast, embryoid body differentiation of KDM6A-KO KDM6B-KD ESCs showed reduced cell numbers, along with an increased frequency of cells in G1/G0 (Fig. 3A; Fig. S3B). KDM6B KD in WT embryoid bodies did not lead to reduced cell numbers or to increased G1 frequencies (Fig. S3D,E). Taken together, these data indicate that the combined abrogation of KDM6A and KDM6B causes defective cell cycle progression. In addition, we observed elevated γH2AX, phosphorylated p53 (S15) and H3K27me3 levels in KDM6A-KO KDM6B-KD embryoid bodies, whereas H3K27me2 and H3K27me1 levels were constant (Fig. 3B; Fig. S3F). In agreement with increased global γH2AX levels in KDM6A-KO KDM6B-KD embryoid bodies, the frequency of γH2AX-positive foci in KDM6A-KO KDM6B-KD embryoid bodies was increased (Fig. S3G). Next, we analysed comet formation and found that differentiating KDM6A-KO KDM6B-KD ESCs showed increased tail moment values (Fig. 3C). Next, we asked whether the global increase in H3K27me3 levels in the knockout and knockdown cells leads to increased H3K27me1, H3K27me2 or H3K27me3 levels in promoter regions of putative KDM6 target genes (Fig. 3D). We included Sox2 in the analyses as it becomes tri-methylated at H3K27 during later time points of differentiation (Obier et al., 2015). HoxB1 and Pou2f3 were analysed because both were upregulated following GSK-J4 treatment (Table S1). ChIP analyses showed that H3K27 methylation levels were similar in wild-type and KDM6A-KO KDM6B-KD cells. Thus, these limited ChIP analyses revealed no increase in H3K27me3 in promoter regions. Finally, as previous work on Drosophila UTX indicates that knockdown of UTX leads to increased irradiation sensitivity (Zhang et al., 2013), we assessed the radiation sensitivity in wild-type and KDM6A-KO KDM6B-KD cells. As shown in Fig. 3E, exposure to 2.5 Gy increased apoptotic cell fractions in wild-type and KDM6A-KO KDM6B-KD cells. The increase in late apoptotic cells (annexinV and propidium iodide double positive) is moderately elevated in KDM6A-KO and KDM6A-KO KDM6B-KD cells.

Collectively, these data indicate that the combined abrogation of KDM6A and KDM6B leads to DNA damage, and to DNA damage signalling in differentiating ESCs.

To ask whether KDM6B or H3K27me3 associate with sites of DNA damage, we analysed the sub-nuclear distribution of H3K27me3 and KDM6B in relation to γH2AX damage foci by immunofluorescence staining (Fig. 4). Consistent with western blot analyses, we observed elevated H3K27me3 levels in GSK-J4-treated ESCs and embryoid bodies (Fig. 4A, lower panels). Also we noticed a higher frequency of γH2AX-positive nuclei in GSK-J5-treated ESCs compared to GSK-J5-treated embryoid bodies where γH2AX+ foci were infrequent. Visual inspection of the immunofluorescence-stained specimens revealed no colocalisation of γH2AX-positive nuclear foci and H3K27me3 signals in GSK-J5-treated ESCs (Fig. 4A, left top panel). GSK-J4 treatment uniformly increased H3K27me3 signals in ESCs and embryoid bodies, resulting in an overlay of γH2AX and H3K27me3 signals, which are unlikely to represent true colocalisation. In agreement with this, the quantification of signal overlap in ESCs and embryoid bodies by using Manders' colocalisation coefficient (Manders et al., 1993) showed only a moderate overlap of the γH2AX and H3K27me3 signals independent of treatment conditions, whereas Pearson's analyses for linear dependence of the two signals (Manders et al., 1993) indicated no correlation (Fig. 4C,D). In ESCs and embryoid bodies, KDM6B localised to the nucleus and cytoplasm, independently of treatment (Fig. 4B). Visual inspection or coefficient analyses (Fig. 4C,D) did not reveal colocalisation of γH2AX-positive nuclear foci with KDM6B.

As GSK-J4 could, as part of its DNA damaging action, impinge on the correct localisation of KDM6A or KDM6B, we assessed whether H3K27me3 or KDM6A and KDM6B colocalised with γH2AX-positive sites of laser-microirradiation-induced DNA damage. We exposed 0.74-day embryoid body cells to microirradiation. As microirradiation severely damaged GSK-J4-treated embryoid bodies, only GSK-J5-treated cells could be analysed. Fig. 5A shows that, following microirradiation, H3K27me3 and γH2AX signals colocalised in differentiating ESCs. We next asked whether H3K27me3 accumulated at microirradiation-induced DNA damage sites in undifferentiated ESCs. Fig. 5B shows that after microirradiation of ESCs, we did not observe colocalisation of H3K27me3 with γH2AX-positive areas. Time course analysis of the subnuclear distribution of KDM6A or KDM6B did not reveal co-localisation at γH2AX-positive sites (Fig. 6). In conclusion, and in agreement with previous reports on UV-induced damage (Mosammaparast et al., 2013; Šustácˇková et al., 2012), we did not observe colocalisation between KDM6B or KDM6A and γH2AX at sites of laser-microirradiation-induced DNA damage. These data are consistent with a differential requirement for H3K27me3 at sites of DNA damage in undifferentiated and differentiating ESCs.

As we observed the colocalisation of H3K27me3 and γH2AX at laser-induced sites of DNA damage, we assessed whether a complete lack of the H3K27me3 modification influences the DDR in GSK-J4-treated embryoid bodies. To address this question, we employed Eed-KO ESCs which, as previously reported (Schoeftner et al., 2006) and shown in Fig. S4A, are devoid of H3K27me3 owing to the lack of functional PRC2. Eed-KO ESCs can form embryoid bodies and express KDM6A and KDM6B (Fig. S4B). Similar to WT ESCs, treatment of Eed-KO ESCs with DMSO or GSK-J5 had no effect on ESC differentiation. In contrast to WT ESCs, however, Eed-KO ESC differentiation in the presence of GSK-J4 was only moderately affected (Fig. 7A). Consistent with this observation, we noticed only a mild decline of population doubling values of GSK-J4-treated Eed-KO embryoid bodies, whereas population doubling values of GSK-J4-treated WT embryoid bodies were strongly reduced (Fig. 7B). We conclude that the absence of Eed attenuates the GSK-J4 effect on cell numbers. We next determined the frequency of γH2AX-positive foci in WT and Eed-KO embryoid bodies following GSK-J5 or GSK-J4 treatment as a read out of DSB formation. Consistent with previous reports showing decreased DSB repair in cells lacking EZH2 (Campbell et al., 2013), an integral component of PRC2, we observed that numbers of γH2AX-positive foci were markedly lower in GSK-J4-treated Eed-KO embryoid bodies as compared to those in treated WT embryoid bodies (Fig. 7C). In summary, the effects of treatment with GSK-J4 are attenuated in differentiating Eed-KO ESCs, suggesting that H3K27me3 is required to elicit DDR in the presence of GSK-J4.

In this study, by using a KDM6A- and KDM6B-specific enzymatic inhibitor (GSK-J4), and by employing KDM6A-KO KDM6B-KD ESCs, we analysed the role of the KDM6 demethylase activity in undifferentiated ESCs and during ESC differentiation. Unexpectedly, we observed that inhibition of KDM6 led to cell death in early differentiating, but not in undifferentiated, ESCs. Global gene expression analyses in differentiating ESCs revealed that only a limited set of genes responded to the GSK-J4 treatment, with more genes up- than downregulated. Several of the upregulated genes encode factors that are associated with DDR. Consistently, GSK-J4 incubation caused severe DNA damage, which preceded G1 arrest and cell death. H3K27me3 but not KDM6A or KDM6B colocalised with γH2AX-positive areas that marked DSBs in laser-microirradiated nuclei of differentiating ESCs. In contrast, a colocalisation of H3K27me3 signals with laser-irradiation-induced γH2AX-positive areas was not observed. Finally, we showed that the GSK-J4-induced DDR was attenuated in differentiating Eed-KO ESCs.

The H3K27me3 mark is generally associated with gene silencing (Zhou et al., 2011). In agreement, lack of H3K27me3 in undifferentiated ESCs – i.e. due to knockout of Eed results in de-repression of PRC2 target genes (Boyer et al., 2006; Elliott et al., 1998). Although we observed higher global H3K27me3 levels in response to GSK-J4 treatment in differentiating ESCs, gene expression analyses showed that only a small set of genes was differentially expressed. Published whole genome ChIP-seq analyses show that H3K27me3 forms broad local enrichments (BLOCS) in gene-rich regions (Pauler et al., 2009). BLOCS alternate with H3K27me3-deficient gene-poor regions. It is currently not known whether low H3K27me3 levels in gene-poor regions are actively maintained by KDM-family proteins. If this is the case, elevated H3K27me3 levels specifically in gene-poor regions in response to GSK-J4 treatment could explain the small number of differentially regulated genes. In support of this notion, we did not find increased H3K27 methylation in the promoter region of putative KDM6 target genes in KDM6A-KO KDM6-KD cells, despite a global increase in H3K27me3 levels. These results further support previous findings indicating a demethylase-independent function of KDM6 for the regulation of developmental gene expression (Kang et al., 2015; Lee et al., 2012). In contrast to what might be expected upon global upregulation of a repressive chromatin mark, the majority of differentially expressed genes were expressed at higher levels following GSK-J4 treatment. Mis-expression of cell-differentiation-associated genes could lead to failed differentiation and, as a result, to cell death. However, only a minority of the genes that were either up- or downregulated were categorised as being associated with cell differentiation or developmental processes. Incorrect elimination of surplus cells during embryonic development could also be a consequence of ill-executed developmental programs (Coucouvanis and Martin, 1995; Debnath et al., 2002). In analogy to embryonic development, apoptosis occurs in floating embryoid body cultures during a process termed cavitation, which forms a proamniotic-like cavity (He et al., 2010; Qi et al., 2012). Bnip3, a BH3-only proapoptotic protein, is of key importance for this process (Qi et al., 2015). As Bnip3 was among the downregulated genes in our global expression analysis (Table S1), we consider it unlikely that normal differentiation was switched to a global proapoptotic program upon inhibition of KDM6 proteins. Our global gene expression analyses were performed at a time point when DNA damage was already evident in GSK-J4-treated embryoid bodies. It is therefore conceivable that the underlying cause for the observed gene expression pattern was cellular stress. In agreement with this, expression of multiple DNA-damage-associated genes was upregulated following GSK-J4 treatment. We consider it more likely that DNA damage rather than failed differentiation is the underlying cause for the observed cell death of differentiating ESCs in response to GSK-J4.

Our experiments using Eed-KO ESCs suggest that the H3K27me3 mark is necessary to elicit a DDR during differentiation under KDM6 inhibition. In the absence of H3K27 methylation in Eed-KO ESCs, the frequency of γH2AX-positive foci under GSK-J4 treatment was comparable to the frequency in untreated control cells. Nevertheless, the population doublings of Eed-KO embryoid body cultures were not as those in controls. This indicates that the GSK-J4 inhibitor has off-target effects. In support of this interpretation, the effect of combined KDM6A KO and KDM6B KD is less pronounced. In agreement with additional GSK-J4 targets, it has been reported that GSK-J4 also targets H3K4-specific KDM5-family members (Heinemann et al., 2014; Kamikawa and Donohoe, 2015). Though the inhibitor concentration we used did not affect H3K4 methylation, additional, so far unidentified, targets of GSK-J4 are possible. However, part of the GSK-J4 effect has to involve H3K27 methylation as we see attenuation of treatment effects in differentiating Eed-KO ESCs.

Recently, Shpargel et al. have reported that KDM6A KDM6B double-KO embryos can develop to term and that repressed genes lose H3K27 methylation and are then expressed (Shpargel et al., 2014). The strategy used in the approach by Shpargel et al. involved the deletion of KDM6B exons 14–21. KDM6B-KO mice generated using this strategy show perinatal lethality owing to defects in lung development (Satoh et al., 2010). In contrast, the phenotype of mice with a complete loss of KDM6B protein expression (due to the deletion of the ATG site and the introduction of a frame shift mutation) was more severe as embryos died at embryonic day (E)6.5 (Ohtani et al., 2013). This indicates that depending on the knockout strategy, the phenotype of KDM6B KO shows variation. Therefore, it is possible that KDM6B exerts additional, so far undescribed, functions in early development. Our own results show no effect of KDM6B knockdown on wild-type embryoid body formation. As discussed above, we observed a milder effect of combined KDM6A KO and KDM6B KD compared to that upon GSK-J4 treatment. Therefore, a complete genomic abrogation of KDM6B – e.g. using the strategy described by Ohtani and colleagues – might yield a more severe phenotype.

A previous report reveals that members of the PRC2 H3K27 methyltransferase complex and of PRC1 are recruited to sites of DSBs (Gieni et al., 2011). Data on PRC function during DDR suggest that PRCs play a role downstream of ATM (Vissers et al., 2012). Consistently, sensitisation to ionising radiation or UV is reported as a consequence of deletions of PRC2 components such as EZH2 (Campbell et al., 2013; Chou et al., 2010). In agreement with an increased sensitivity to DNA damage in the absence of H3K27me3, we found higher frequencies of γH2AX-positive foci in control-treated differentiating Eed-KO ESCs than in differentiating WT ESCs. Although the colocalisation of PRCs with sites of DNA damage is well established, data linking H3K27me3 directly to DNA damage foci are inconsistent. Two recent studies show increased H3K27me3 at UV-light-induced DNA damage sites (Chou et al., 2010; O'Hagan et al., 2008), whereas two other studies have reported no colocalisation in response to UV or ionising radiation (Campbell et al., 2013; Šustácˇková et al., 2012). This discrepancy could derive from varying experimental conditions. Our own work shows colocalisation of γH2AX and H3K27me3 marks in differentiating ESCs upon laser microirradiation, indicating that H3K27me3 is part of the DDR at sites of DNA damage. We did not find colocalisation of H3K27me3 with laser-microirradiation-induced γH2AX-positive areas in undifferentiated ESCs, pointing towards a differential requirement for H3K27me3 at sites of DNA damage in undifferentiated and differentiating ESCs. To what extent the differential localisation of H3K27me3 at sites of DNA damage in undifferentiated and differentiating ESCs is linked to the differential response of these cells to KDM6 inhibition remains elusive.

In agreement with recent literature (Mosammaparast et al., 2013), we did not find the colocalisation of KDM6B with γH2AX-positive DNA damage, and we extend this observation to KDM6A. Although we cannot exclude the possibility that KDM6A and KDM6B are recruited to γH2AX-positive foci at a level below the detection limit of immunofluorescence microscopy analyses, our observations are so far consistent with an indirect role of KDM6 activity in DNA repair.

A possible cause for the increased DNA damage in differentiating ESCs upon KDM6 inhibition could be the deregulation of cellular processes such as cell cycle regulation. Our results show that GSK-J4-treated or KDM6A-KO and KDM6B-KD cells increase G1 cell frequencies. In parallel, cell cycle regulators (p21, RB1) are upregulated. Because RB1 and p21 are downstream effectors of DNA damage (Sperka et al., 2012), we consider it more likely that the elevated frequencies of G1 cells is a consequence of DNA damage rather than a cause. Another cellular process that might be deregulated upon KDM6 inhibition is replication. In this regard, recent data (Piunti et al., 2014) are of interest, which show that the PRC2 core component EZH2 localises at replication forks and that abrogation of PRC1 core component Ring1B expression induces DNA damage. To what extent H3K27me2 and H3K27me3 methylation, the product of the EZH2 enzymatic activity, are necessary for the progression of replication forks is an interesting subject of future analyses.

Inefficient repair of basal DNA damage could be another cause for the accumulation of DNA lesions. Seminal studies revealed that undifferentiated ESCs are distinct from differentiated cells with respect to DDR. In this regard, we showed that the determination of γH2AX-positive foci frequencies in undifferentiated mouse ESCs does not reflect true DNA damage frequencies, because the foci are associated with global chromatin decondensation rather than pre-existing DNA damage (Banáth et al., 2009). In our analyses, we confirmed high frequencies of γH2AX-positive foci by immunofluorescence staining and high levels of γH2AX in western blot analyses of undifferentiated ESCs. γH2AX levels and foci frequencies decline upon differentiation. Another feature that distinguishes undifferentiated ESCs from differentiating ESCs is the choice of DNA damage repair pathways. ESCs remove ionising-radiation-induced DNA damage through homologous recombination rather than by error-prone non-homologous end joining (NHEJ) (Fortini et al., 2013). To our knowledge, no detailed experimental data on the DNA repair mechanism during ESC differentiation are available. It is, however, tempting to speculate that KDM6 plays a role in the transition from the use of homologous recombination in undifferentiated cells towards the use of NHEJ in differentiated ESCs. In this context, the co-presence of PTIP, KDM6A and KDM6B in the MLL2 complex might be relevant (Cloos et al., 2008). PTIP is also known for its interaction with 53BP1, and it is required for 53BP1-mediated inhibition of homologous recombination (Callen et al., 2013). So far however, it is unknown whether KDM6A and/or KDM6B are involved in PTIP-mediated homologous recombination inhibition. Our own data show that homologous recombination is utilised by differentiating ESCs to repair endonuclease-mediated DNA damage; however, we did not observe reduced homologous recombination frequencies in the presence of GSK-J4. Further analyses will show whether NHEJ is inhibited by GSK-J4 treatment. In light of a potential role for KDM6 proteins in switching from homologous recombination to NHEJ during differentiation, a report on the transcriptional regulation of the NHEJ factor Ku80 by Drosophila UTX in response to DNA damage (Zhang et al., 2013) could also be relevant. In addition, under UTX KD and exposure to ionising radiation, decreased cell numbers were observed. Our own data show only moderate effects of ionising radiation on cell survival, indicating species- and cell-specific effects of KDM6 proteins.

In summary, we observed that inhibition of KDM6 activity results in increased DNA damage, activation of the DDR and increased cell death in ESCs that had begun to initiate embryoid body differentiation. Our findings suggest that KDM6 protein activity is necessary to prevent DDR and for survival at the exit from ESC pluripotency.

ESC lines [R1, J1 and J1 Eed KO (Schoeftner et al., 2006)] were cultured on mitomycin C (MitC)-inactivated MEFs in high glucose Dulbecco's modified Eagle's medium (DMEM) (Sigma-Aldrich) supplemented with 15% ESC-grade fetal bovine serum (FBS; Biochrom, Berlin, Germany), 0.1 mM β-mercaptoethanol (Sigma-Aldrich), 2 mM l-glutamine (GE Lifesciences), 1 mM sodium pyruvate (GE Lifesciences), 100 U/ml penicillin (GE Lifesciences) 0.1 mg/ml streptomycin (GE Lifesciences), 1% non-essential amino acids (GE Lifesciences) and LIF. For analysis of cumulative population doublings of ESC cultures, feeder-depleted ESCs (separated by passing over gelatin-coated plates) were seeded onto MitC-inactivated MEFs at a density of 0.5×10⁶ in 6-cm culture dishes. After 2 days of culture, cells were trypsinised, and live cell numbers were determined by Trypan Blue staining. Subsequently, 0.5×10⁶ ESCs were reseeded and cultured for a total of three passages. The population doubling level at each passage was determined according to the following equation: x=log10[log10(N1)−log10(NH)], where N1 is the inoculum number, NH is the cell harvest number and x is the population doubling value (Cristofalo et al., 1998). New population doubling values were added to previous values to calculate cumulative population doubling values. For ESC differentiation, feeder-depleted ESCs were cultured in embryoid body medium (high-glucose DMEM; Sigma-Aldrich), 10% pre-selected FCS (Gibco, Life Technologies, Darmstadt, Germany), 0.1 mM β-mercaptoethanol (Sigma-Aldrich), 2 mM l-glutamine (GE Lifesciences), 1 mM sodium pyruvate (GE Lifesciences), 100 U/ml penicillin (GE Lifesciences), 0.1 mg/ml streptomycin (GE Lifesciences), 1% nonessential amino acids (GE Lifesciences) in non-coated bacteriological Petri dishes. Differentiating ESCs formed embryoid bodies. For harvesting, embryoid bodies were transferred to 15-ml Falcon tubes and allowed to sediment. For preparation of single-cell suspensions, medium was replaced by Trypsin-EDTA (Invitrogen, Life Technologies) followed by 3 min of incubation at 37°C. Live cell numbers were determined by Trypan Blue staining. To generate embryoid bodies and determine population doubling levels in embryoid bodies, ESCs were seeded in embryoid body medium at a density of 1×10⁵–2×10⁵ cells/ml. Embryoid bodies were harvested at 1, 2 or 3 days of differentiation, single-cell suspensions were prepared and live cell numbers were determined. Population doubling levels were calculated as described above. The GSK-J4 (Tocris Bioscience, Bristol, UK) concentration used throughout this study was determined in a dose-finding assay monitoring R1 ESC viability and H3K27me3 levels. Starting from the maximum concentration (30 µM) for KDM6 protein inhibition (Kruidenier et al., 2012), the concentration was gradually reduced to 0.9 µM (twofold dilution steps). A concentration of 1.8 µM yielded a maximum H3K27me3 level without interfering with cell viability. Based on this observation, GSK-J4 or the control compound GSK-J5 (Cayman Chemical) were both added at a concentration of 1.8 µM to ESC and embryoid body cultures. Irradiation of ESC and embryoid body cultures was performed in a Faxitron CP-160 X-ray radiation cabinet (160 kV, 6.3 mA, 0.71 Gy/min, filter: 0.5 mm Cu). Unless otherwise specified, all experiments were performed with R1 ESCs.

Single-cell suspensions of ESC or embryoid body cultures were washed twice with PBS before lysis in 2× Laemmli buffer (0.5 M Tris HCl, pH 6.8, 0.4% SDS, 2% glycerol, 0.5% β-mercaptoethanol and Bromophenol Blue). Lysates were adjusted to 10⁶ cells/125 µl and heated (5 min, 95°C). Proteins were separated by using SDS PAGE and blotted onto nitrocellulose membranes (Schleicher & Schuell, Dassel, Germany). Membranes were blocked in PBS, 0.05% Tween, 5% milk powder (30 min, room temperature) followed by overnight incubation with primary antibodies. Membranes were washed three times with PBS followed by incubation with secondary antibodies (1 h, room temperature). After three washes in PBS, membranes were incubated with Amersham ECL Select Western blot Detection Reagent (GE Lifesciences). Chemiluminescence was detected using a Chemidoc XRS low light imager (Bio-Rad Laboratories, Munich, Germany). If phospho-specific antibodies were used, PBS was replaced by TBS. The intensity of protein bands was determined using ImageJ software (http://rsb.info.nih.gov/ij/).

For staining of ESCs, cells were grown on MitC-treated MEFs on cover slips for two days. Next, ESCs were fixed with 4% PFA (10 min, room temperature) followed by 3 washes with PBS. For preparation of tissue sections, embryoid bodies were transferred to Falcon tubes and allowed to sediment. After 2 washes with PBS embryoid bodies were fixed in 4% PFA for 3 h and dehydrated overnight in a 20% sucrose solution (4°C). embryoid bodies were embedded in Tissue-Tek (Sakura, Leiden, NL), 5-μm sections were cut and mounted onto superfrost microscope slides. Fixed ESC or embryoid body sections were permeabilised (0.1% Triton X-100 in PBS) for 30 min at room temperature followed by incubation in blocking solution (20% FCS, 0.1% Triton X-100 in PBS) for 30 min. Subsequently, samples were incubated overnight at 4°C with primary antibodies diluted in blocking solution. Samples were washed three times with PBS before incubation (1 h, room temperature) with secondary antibodies diluted in blocking solution. After incubation, samples were washed three times. DAPI (5 µg/ml) was added to the last washing step and was followed by mounting in Mowiol. Microscopy-based inspection and imaging of samples was performed on an inverted confocal laser scanning microscope (LSM780, Zeiss). For determination of γH2AX frequencies, microscopic images of γH2AX-stained embryoid body sections were prepared using an inverted epifluorescence Microscope (Zeiss). Average numbers of γH2AX-positive foci per nucleus were determined from tissue sections of 15–30 embryoid bodies per condition (total of 100–650 cells).

Analyses of cell cycle phase propidium iodide staining of single-cell suspensions were performed. Briefly, cells were washed twice with cold PBS, permeabilised (70% EtOH, 30 min, −20°C), washed twice with PBS and incubated (30 min, 37°C) with propidium iodide (10 µg/µl) and RNase (10 µg/µl) before fluorescence-activated cell sorting (FACS) analysis. Frequencies of apoptotic cells were determined by annexinV–phycoerythrin Apoptosis Detection Kit I (BD Biosciences). Analyses were performed on a flow cytometer (FACS Calibur, BD Biosciences). Cell cycle distribution was determined using ModFit software (Verity Software House, Topsham, ME).

Comet assays were performed with the OxiSelect™ Comet Assay Kit (Cell Biolabs, San Diego, CA). Briefly, after trypsinisation, single-cell suspensions of embryoid body cultures that were either control or had been treated with GSK-J4 were allowed to recover by incubation (15 min, 37°C) in the presence or absence of inhibitor. For ionising radiation exposure, embryoid bodies were transferred to embryoid body culture medium without FCS before irradiation and immediately kept at 4°C after irradiation to avoid DNA repair. Subsequently, cells were washed, resuspended in pre-chilled 1× PBS (Sigma-Aldrich) and mixed with liquefied OxiSelect™ comet agarose. 10⁴ cells/well were pipetted onto OxiSelect™ Comet well slides and incubated (15 min, 4°C, in the dark). Slides were transferred into pre-chilled lysis buffer and incubated (60 min), followed by brief incubation in H₂O and finally in alkaline solution (30 min, 4°C, in the dark). Alkaline electrophoresis was performed at 4°C with alkaline electrophoresis solution (30 min, 1 V/cm, 300 mA). After electrophoresis, slides were washed three times in pre-chilled H₂O followed by incubation in ice-cold 70% EtOH (5 min). Slides were air-dried and stained with Vista Green DNA Dye. Microscopy-based inspection and imaging of samples was performed with an inverted epifluorescence microscope (Zeiss). For comet quantification, >50 nuclei were imaged per condition. Comet analyses were performed using CASP software (Końca et al., 2003) (CASP Lab).

Total RNA was extracted using RNeasy Mini (Qiagen) or peqGOLD RNA PURE (Peqlab, Erlangen, Germany) kits. RNA quality was assessed with the RNA 6000 nano kit using the Bioanalyzer 2100 instrument (Agilent). Before microarray hybridisation, 100 ng of total RNA was amplified using the WT Expression kit (Ambion, Life Technologies), followed by single stranded (ss)DNA labelling (WT Terminal Labeling Kit; Affymetrix). Samples were hybridised to GeneChip Mouse Gene 2.0 ST arrays (Affymetrix). Fluorescence intensities were scanned with a GeneChip Scanner 3000 7G (Affymetrix). Image processing (microarray grid definition, feature intensity readout) and CEL file generation were performed with GeneChip Operating Software (Affymetrix). Raw microarray data were background corrected, normalised and summarised to probeset expression values using the Robust Microarray Average (RMA) algorithm (Bolstad et al., 2003; Irizarry et al., 2003). Differentially expressed probe sets were detected by estimating the false discovery rates (FDR) with an empirical Bayesian methodology employing lognormal–normal data modelling (Kendziorski et al., 2003). Significant probe sets (FDR<0.05) were subjected to gene-set enrichment analysis using WebGestalt (Wang et al., 2013). Analyses were performed in the R environment (http://www.r-project.org) using Bioconductor (http://www.bioconductor.org) packages ‘affy’ and ‘embryoid body arrays’. Microarray data were deposited in MIAME-compliant form at Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo) under entry GSE49886.

KDM6A-KO ESCs were generated by transfection of R1 KDM6A conditional mutant ESCs (UTX FD), which contain KDM6A exon 3 flanked by loxP sites with the pCAG-CreERT2 vector (Addgene entry 14797) followed by 2 days of selection with Tamoxifen. Single colonies were picked and screened by using PCR for successful deletion of exon 3. For lentiviral small hairpin (sh)RNA transduction, viral particles were produced in 293T cells. Subsequently, ESCs were inoculated two consecutive times with lentiviral particles (24 h inoculation time each), followed by puromycin (Invitrogen, Life Technologies, Darmstadt, Germany) (1 µg/ml) selection for 4 days. ESCs and embryoid bodies were kept under puromycin selection. Lentiviral pLKO.1-puro vectors (Sigma-Aldrich) were as follows: SHC002 (Scr-shRNA), NM_001017426.1-5159s1c1 (KDM6B-shRNA1), NM_001017426.1-3013s1c1 (KDM6B-shRNA2).

For generation of the DR-GFP reporter transgenic cells (used to measure DSB homologous repair), R1 ESCs were transfected with the pHPRT-DRGFP construct (Addgene entry 26476) (Pierce et al., 2001), followed by 8 days of selection with puromycin (Invitrogen, Life Technologies) (1 µg/ml). Bulk cultures were used for subsequent experiments. For transfection with the pCBASceI construct (Addgene entry 26477) (Richardson et al., 1998), 10⁶ DR-GFP reporter cells were electroporated with 2 µg of pCBASceI vector using an Amaxa Nucleofector. After SceI transfection, cells were cultured in ESC medium for 18 h and then transferred to differentiation medium containing GSK-J4 (1.8 µM), GSK-J5 (1.8 µM) or Mirin (100 µM) – a specific MRE11 inhibitor (Dupré et al., 2008). The frequency of GFP-positive cells was assessed with flow cytometry 24 h after the induction of differentiation.

HA-tagged human KDMA6A (hUTX) was PCR-amplified from the pMSCV-HA-hUTX-puro vector to introduce a 5′NheI and a 3′SacII restriction site, respectively. The PCR product was NheI–SacII-digested, followed by ligation into the corresponding restriction sites of the pB-EF1-MCS-IRES-NEO Piggybac vector (System Biosciences, Mountain View, CA) multiple cloning site to generate the pB-EF1-HA-UTX-IRES-NEO vector. The pB-EF1-HA-UTX-IRES-NEO vector was electroporated into R1 KDM6A-KO ESCs together with the transposase coding pBase vector. Neomycin-selection (Invitrogen, Life Technologies) (800 µg/ml) was started 1 day after transfection. Single colonies were picked after 2 weeks of selection and analysed for HA–UTX expression by western blotting and reverse transcriptase PCR analysis.

For laser microirradiation of adherent embryoid bodies, ESCs were cultured in embryoid body medium for 24 h to allow embryoid body formation. The embryoid bodies were subsequently transferred to x-well tissue culture chambers (Sarstedt). The culture medium was supplemented with 5 µM BrdU for 24 h to sensitise the cells to DSB generation by UV-A laser (λ=337 nm) irradiation. Experiments were performed with a Leica ASLMD laser microdissection microscope (laser power set to 35%). DMSO, GSK-J5 or GSK-J4 was added 18 h before irradiation. For laser microirradiation of ESCs, cells were cultured for 24 h in the presence of 5 µM BrdU. GSK-J4 or GSK-J5 was added 18 h prior to irradiation. Cells were seeded into x-well tissue culture chambers 4 h before irradiation. Immediately before laser irradiation, the culture medium was changed to Phenol-Red-free HEPES-buffered embryoid body or ESC medium. Following irradiation of ∼100 cells (a procedure taking 10 min), embryoid bodies and ESCs were fixed in 3.7% formaldehyde and subjected to immunofluorescence staining. For time course analyses of KDM6A or KDM6B localisation after laser microirradiation, embryoid bodies were irradiated as described above. After irradiation medium had been exchanged for normal embryoid body culture medium, embryoid bodies were incubated under standard conditions (37°C, 5% CO₂) for the indicated time periods; 20–30 nuclei were analysed per condition.

Total RNA was extracted using RNeasy Mini (Qiagen) or peqGOLD RNA PURE (Peqlab) kits. 1 µg of total RNA was used for cDNA synthesis using the First Strand cDNA Synthesis Kit (Fermentas, Life Technologies). Quantitative real-time (qRT)-PCR reactions were performed using the ABsolute SYBR Green Mix (Thermo Scientific, Life Technologies) in a Roche Light Cycler 480 II (Roche Diagnostics) under the following conditions: initial denaturation for 15 min at 95°C, followed by repetitive cycles of 10 s at 95°C, 20 s at 60°C and 30 s at 72°C. CT values were normalised to two housekeeper genes (GAPDH and RPL4), which were pre-selected based on their stable expression using geNorm software (Vandesompele et al., 2002). Primer sequences were: KDM6A forward 5′-GCTGGAACAGCTGGAAAGTC-3′, reverse 5′-GAGTCAACTGTTGGCCCATT-3′; KDM6A exon 3 forward 5′-CTGAAGGGAAAGTGGAGTCTG-3′, reverse 5′-TCGACATAAAGCACCTCCTG-3′; KDM6B forward 5′-GGAAGCCACAGCTACAGGAG-3′, reverse 5′-CCACCAGGAACCAGTCAAGT-3′; UTY forward 5′-ATAGTGTCCAGACAGCTTCA-3′, reverse 5′-GAGGTAGGAATACGTAAGAA-3′; Oct4 forward 5′-AGGCCCGGAAGAGAAAGCGAACTA-3′, reverse 5′-TGGGGGCAGAGGAAAGGATACAGC-3′; Sox2 forward 5′-GCGGAGTGGAAACTTTTGTCC-3′, reverse 5′-CGGGAAGCGTGTACTTATCCTT-3′; Nanog forward 5′-TCTTCCTGGTCCCCACAGTTT-3′, reverse 5′-GCAAGAATAGTTCTCGGGATGAA-3′; GAPDH forward 5′-TGGAGAAACCTGCCAAGTATG-3′, reverse 5′-TCATACCAG GAAATGAGCTTGA-3′; RPL4 forward 5′-TTGGGTTGTATTCACTCTGCG-3′, reverse 5′-CAGACCAGTGCTGAGTCTTGG-3′; Polk forward 5′-AGCTCAAATTACCAGCCAGCA-3′, reverse 5′-GGTTGTCCCTCATTTCCACAG-3′; Btg2 forward 5′-ATGAGCCACGGGAAGAGAAC-3′, reverse 5′-GCCCTACTGAAAACCTTGAGTC-3′; Mdm4 forward 5′-TTCGGAACAAATTAGTCAGGTGC-3′, reverse 5′-AGTGCATTACCTCTTTCATGGTG-3′; p21^CIP forward 5′-CACAGCTCAGTGGACTGGAA-3′, reverse 5′-ACCCTAGACCCACAATGCAG-3′; FoxA2 forward 5′-TAGCGGAGGCAAGAAGACC-3′, reverse 5′-CTTAGGCCACCTCGCTTGT-3′; Gata4 forward 5′-CCCTACCCAGCCTACATGG-3′, reverse 5′-ACATATCGAGATTGGGGTGTCT-3′; Sox17 forward 5′-GATGCGGGATACGCCAGTG-3′, reverse 5′-CCACCACCTCGCCTTTCAC-3′; Nestin forward 5′-CAGAGAGGCGCTGGAACAGAGATT-3′, reverse 5′-AGACATAGGTGGGATGGGAGTGCT-3′; Sox1 forward 5′-ACAGATGCAACCGATGCACC-3′, reverse 5′-TGGAGTTGTACTGCAGGGCG-3′; FLk-1 forward 5′-CCTGGTCAAACAGCTCATCA-3′, reverse 5′-AAGCGTCTGCCTCAATCACT-3′, brachyury (T) forward 5′-GCTCTAAGGAACCACCGGTCATC-3′, reverse 5′-ATGGGACTGCAGCATGGACAG-3′.

Cells were stained using an Alkaline Phosphatase Detection Kit (Millipore).

Single-cell suspensions of feeder-depleted ESCs were washed with ice-cold FACS buffer (PBS, 0.1% BSA) and stained with an anti-mouse SSEA-1 antibody (BioLegend) (45 min, 4°C). Cells were washed with PBS and incubated with a secondary PE–Cy7–antibody (BD Biosciences) (45 min). Subsequently cells were washed with PBS and analyses were performed on a flow cytometer (FACSCalibur, BD).

For a list of antibodies used in this study see Table S2.

ChIP experiments were performed as described previously (Dahl and Collas, 2009) with modifications. Briefly, 1.2×10⁶ single-cell suspensions of embryoid bodies at day 3 were cross-linked with 16% formaldehyde. After nuclei extraction, sonication was performed using a Covaris M220 focused-ultrasonicator (20 min; 10% duty factor). 100 µl of chromatin solution was used as input sample and for immunoprecipitation with antibodies against histone 3 (2.5 μg, Abcam), H3K27me3 (2.5 μg, Diagenode), H3K27me2 (2.5 µg, Abcam), H3K27me1 (2.5 µg, Diagenode) or IgG isotype control (2.5 μg, Abcam). Genomic DNAs were isolated by phenol chloroform extraction and ethanol precipitation. Promoter-specific primers were: Sox2 forward 5′-TTTAGGGTAAGGTACTGGGAAGG-3′, reverse 5′-GAGCCCGGGAAATTCTTTTA-3′; Hoxb1 forward 5′-CTCTGGTCCCTTCTTTCC-3′, reverse 5′-GGCCAGAGTTTGGCAGTC-3′; Pou2f3 forward 5′-GGTAAGCGGTTGGAAGGCAAG-3′, reverse 5′-GGCTTCTGCCTAGTTGGGCTC-3′.

Colocalisation in image pairs was assessed from scatter plots of midnuclear confocal images as described previously (von Mikecz et al., 2000) and quantified using the ZEN black software (Zeiss). For each treatment condition, at least four different ESC colonies (at least 15 nuclei per colony) or sections of four individual embryoid bodies (at least 20 nuclei per section) were analysed. Background thresholds for each channel were determined as mean intensity value of the image region of interest (ROI) plus two times the standard deviation. The colocalisation coefficients for two signals were determined according to Manders, which provides a value range between 0 and 1 (0, no colocalisation; 1, all pixels colocalise) and according to Pearson, which provides a value range between −1 and 1 (−1, negative correlation of signal intensities; 0, no correlation of signal intensities; 1, positive correlation of signal intensities) (Manders et al., 1993).

Eed-KO ESCs were kindly provided by Dr A. Wutz (Institute for Molecular Health Sciences, Zurich, Switzerland). Conditional mutant UTX ESCs were kindly provided by Dr Konstantinos Anastassiadis (BIOTEC, Technische Universität Dresden, Dresden, Germany). The pMSCV-HA-hUTX-puro vector was kindly provided by Robert Slany (Lehrstuhl für Genetik, Erlangen, Germany). We thank Veronika Hornich Ursula Sauer for excellent technical assistance and Dr Detlef Schindler, Dr Elizabeth D. Martinez, Dr Stanislaw Gorski, Dr Almut Horch, Dr Tcholpon Djuzenova, Dr Jens Vanselow and Dr Peter Hemmerich for helpful discussions.