Polycomb group proteins Ring1A/B are functionally linked to the core transcriptional regulatory circuitry to maintain ES cell identity

Embryonic stem (ES) cells derived from the inner cell mass (ICM) of mammalian blastocyst-stage embryos have unlimited growth potential while maintaining pluripotency - the ability to differentiate into all tissue types except the placenta (Evans and Kaufman,1981; Martin,1981). These properties of ES cells are maintained by symmetrical self-renewal, producing two identical stem cell daughters upon cell division(Burdon et al., 2002). The maintenance of ES cell pluripotency is thought to involve a transcriptional regulatory hierarchy including the transcription factors Oct3/4 (Pou5f1 -Mouse Genome Informatics), Sox2 and Nanog, which may be central components judging by their unique expression patterns and essential roles during early murine development (Avilion et al.,2003; Chambers et al.,2003; Mitsui et al.,2003; Nichols et al.,1998).

Recent genome-wide chromatin immunoprecipitation (ChIP) analyses revealed that OCT4, SOX2 and NANOG co-occupy the promoters of a large group of genes in human ES cells (Boyer et al.,2005), suggesting that these factors form a core regulatory feedback circuit, in which all three factors regulate the expression of themselves as well as of each other(Catena et al., 2004; Kuroda et al., 2005; Okumura-Nakanishi et al.,2005; Rodda et al.,2005). This positive-feedback loop promotes self-renewal of pluripotent ES cells by repressing transcription factors involved in differentiation and development, whilst likely activating the expression of genes involved in ES cell maintenance(Boyer et al., 2005).

The execution of differentiation programs in ES cells is likely to be preceded by interruption of the positive-feedback loop by developmental regulators such as Cdx2 and Gata6. The expression of Cdx2 and Gata6 is repressed by Oct3/4 and Nanog in undifferentiated ES cells(Boyer et al., 2005; Loh et al., 2006; Mitsui et al., 2003; Niwa et al., 2000; Niwa et al., 2005), and the enforced expression of Cdx2 and Gata6 quickly shuts down the positive loop and promotes a rapid transition from the undifferentiated to the differentiated state (Fujikura et al., 2002; Niwa et al., 2005). Therefore,the positive-feedback loops and developmental regulators are reciprocally engaged to maintain ES cell identity; however, the molecular mechanisms underlying this reciprocal interaction are not fully understood.

The Polycomb group (PcG) of proteins mediate heritable silencing of developmental regulators in metazoans, participating in one of two distinct multimeric protein complexes, the Polycomb repressive complexes 1 (PRC1) and 2(PRC2) (Cao et al., 2002; Czermin et al., 2002; Kuzmichev et al., 2002; Muller et al., 2002; Shao et al., 1999). In mammals, the core PRC2 is composed of Eed, Ezh2 and Suz12 and catalyses trimethylation of histone H3 at lysine 27 (H3K27), which in turn is thought to provide a recruitment site for PRC1 (Cao et al., 2002; Czermin et al.,2002; Fischle et al.,2003; Kuzmichev et al.,2002; Min et al.,2003). The core PRC1 is composed of orthologs of Drosophila Polycomb (Cbx2, Cbx4 and Cbx8), Posterior sex combs [Mel18(Pcgf2) and Bmi1], Sex comb extra (Ring1A and Ring1B, also known as Ring1 and Rnf2, respectively - Mouse Genome Informatics) and Polyhomeotic (Phc1, Phc2 and Phc3). Recent studies demonstrate that mono-ubiquitylation of histone H2A at lysine 119 is important in PcG-mediated silencing, with Ring1A/B functioning as the E3 ligase in this reaction(de Napoles et al., 2004; Wang et al., 2004).

Mouse and human ES cells have recently been analyzed by genome-wide ChIP,and PRC1 and PRC2 have been shown to repress genes involved in processes including development, transcriptional regulation and morphogenesis, via direct interactions with target genes(Boyer et al., 2006; Lee et al., 2006). Notably,PRC2 has been shown to share target genes with OCT4, SOX2 and/or NANOG in human ES cells (Lee et al.,2006). However, it is still unclear whether PcG and the core transcription network are functionally linked to regulate expression of their target genes.

In this study, we addressed the role of PRC1 in mouse ES cell maintenance and its functional interaction with the core transcriptional regulatory circuitry. We find that PRC1 is essential for the maintenance of ES cell identity and for the repression of developmental regulators by inhibiting chromatin remodeling. We go on to show that Ring1A/B-mediated PcG silencing is Oct3/4-dependent, whereas it is abolished by developmental cues resulting in Gata6 activation. Collectively, our data suggest that Ring1A/B-mediated Polycomb silencing functions downstream of the core transcriptional regulatory circuitry to maintain ES cell self-renewal.

Eed-KO, Dnmt1-KO, Oct3/4 conditional KO (ZHBTc4)ES cell lines and the ES cell line expressing a Gata6-GR fusion (G6GR) were described previously (Azuara et al.,2006; Lei et al.,1996; Niwa et al.,2000; Shimosato et al.,2007). Generation of mutant Ring1A and Ring1Bflox alleles were described previously(Cales et al., 2008; del Mar Lorente et al., 2000). Rosa26::CreERT2 transgenic mice were purchased from Artemis Pharmaceuticals (Seibler et al.,2003). Ring1A^-/-;Ring1B^fl/fl; Rosa26::CreERT2 ES cells were derived from blastocysts. Male ES cells were used in this study.

IP (Isono et al., 2005a)and ChIP (Orlando et al.,1997) were performed as previously described. Immunoprecipitated and input DNA were quantified by real-time PCR. Primer and probe sequences are available upon request. Antibodies used in this study are listed in Table 1.

RNA extraction and cDNA synthesis were performed as described previously(Isono et al., 2005b). Quantitative real-time PCR was carried out using the SYBR Green (Stratagene,Agilent Technologies, Santa Clara, CA) or Taqman (Biosearch Technologies,Novato, CA) method and amplifications detected with an Mx3005P (Stratagene, La Jolla, CA). Primer and probe sequences are available upon request.

Total RNA was extracted using Trizol (Invitrogen, Carlsbad, CA) and purified with RNeasy separation columns (Qiagen, Hilden, Germany). First-strand cDNA was synthesized and hybridized to Affymetrix GeneChip Mouse Genome 430 2.0 arrays (Affymetrix, Santa Clara, CA) to assess and compare the overall gene expression profiles.

To obtain normalized intensities from at least two slides, the quantile normalization method was used for every feature on the array(Bolstad et al., 2003). We calculated the log of the ratio of intensity in the knockout (KO) samples to the intensity in the respective control samples. Probes were not applied for further analysis when signals were at insignificant levels in control and KO samples. The expression change of a gene was calculated using the geometric mean of all probes aligned on the gene.

The microarray and ChIP-chip data are available in the NCBI Gene Expression Omnibus (GEO) under the series GSE10573 [NCBI GEO] with sample accession numbers GSM265040 to GSM265045, GSM266065 to GSM266067, GSM266076, GSM266077,GSM266115, GSM266837 and GSM266838.

We obtained Pearson product-moment correlation coefficients of the logarithms of expression changes between respective KO ES cells. The 95%confidence intervals of correlation coefficients were calculated using Z transformation. Eigenvalues and eigenvectors of the distribution in scatter diagrams were calculated using principal component analysis with software R(http://www.r-project.org/).

We performed GO analysis using our in-house programs written in Python and C++ and GO data retrieved from the Gene Ontology database(http://www.geneontology.org),KEGG(http://www.genome.jp/kegg/)and others (Auernhammer and Melmed,2000; Heinrich et al.,2003). The version of the dataset used was Oct 27th, 2006,submitted by Mouse Genome Informatics (MGI). We aligned microarray probes on mouse genes and assigned GO terms on all probes using these alignments. The significance of each GO term was determined using Fisher's exact test and Bonferroni adjustment for multiple testing. The P-value reflects the likelihood that we would observe such enrichment or higher by chance. Subsequent statistical examinations were also conducted using Fisher's exact test.

ChIP-on-chip analysis of Ring1B binding was carried out using the Mouse Promoter ChIP-on-chip Microarray Set (G4490A; Agilent Technologies). ES cells were subjected to ChIP assay using anti-Ring1B antibody as described(Fujimura et al., 2006). Purified immunoprecipitated and input DNA were subjected to blunt ligation with linker oligo DNA, linker-mediated PCR (LM-PCR), labeling, hybridization and washing following the Agilent mammalian ChIP-on-chip protocol. Scanned images were quantified with Agilent Feature Extraction software under standard conditions.

Assignment of regions bound by Ring1B around transcription start sites(TSSs) was carried out using direct sequence alignment on the mouse genome database (NCBI version 36). The location of Ring1B-bound regions was compared with a set of transcripts derived from the MGI database. We assigned bound regions that were within -8 kb to +2 kb of the TSS. Alignments on mouse genome and TSSs of genes were retrieved from Ensembl(http://www.ensembl.org).

The measured intensity ratios (IP/input: fold enrichment) were calculated,and the maximum value of the ratios in each promoter region (-8 kb to +2 kb around TSS) of a gene was used to represent the fold enrichment of the gene. Fold enrichment was calculated only for probes whose signals both from IP and input DNAs were significant (P<10^-3).

To investigate the role of PRC1 for maintenance of mouse ES cell identity,it was necessary to generate Ring1A/B double-knockout (dKO) ES cells because Ring1B single-knockout (Ring1B-KO) ES cells can be cultured for >20 passages and exhibit ES cell morphology(de Napoles et al., 2004; Fujimura et al., 2006) (data not shown). Thus, we established Ring1A^-/-;Ring1B^fl/fl; Rosa26::CreERT2 ES cell lines, in which Ring1Bcould be conditionally deleted by 4-hydroxy tamoxifen (OHT) treatment. Ring1B protein levels were dramatically depleted within 48 hours of OHT administration (Fig. 1A), and loss of Ring1B resulted in reduced levels of other PRC1 components Mel18,Phc1/2 and Cbx2 (Fujimura et al.,2006; Leeb and Wutz,2007). At a global level, PRC1-regulated mono-ubiquitylated histone H2A (H2Aub1) was rapidly depleted within 48 hours following OHT treatment (Fig. 1B). We thus concluded that PRC1 could be conditionally depleted in this ES cell line by OHT.

In the Ring1A/B-dKO, in contrast to the Ring1B-KO ES cells, proliferation was halted and the cells gradually lost typical ES cell morphology after OHT administration (Fig. 1C). Moreover, genome-wide mRNA analysis revealed preferential derepression of genes involved in differentiation and/or developmental processes (Fig. 1D; see also Table S1 in the supplementary material). These observations, considered together with structural and biochemical similarities of Ring1A and Ring1B(Buchwald et al., 2006), led us to hypothesize a compensatory role of Ring1A for Ring1B in the repression of developmental genes in ES cells. This idea is partly supported by the increased expression of Ring1A protein observed in the Ring1B-KO ES cells (see Fig. S1 in the supplementary material). We first compared gene expression between the Ring1A/B-dKO and Ring1B-KO ES cells by microarray analyses. We found that 491 genes were derepressed more than 2-fold in Ring1B-KO (constitutive), whereas in Ring1A/B-dKO(day 4) ES cells, 999 genes were derepressed (see Table S2 in the supplementary material). Fold expression changes for respective probes in Ring1A/B-dKO and Ring1B-KO ES cells, determined against the parental or wild-type cells, were plotted on a scatter diagram and the correlation was calculated according to Pearson (see Fig. S2 in the supplementary material). We found a strong correlation (r=0.386) in total calculable genes (see Fig. S2 in the supplementary material). This result indicates significant overlap of genes derepressed in Ring1A/B-dKO and Ring1B-KO ES cells. The level of derepression was much higher in Ring1A/B-dKO than in Ring1B-KO ES cells, as represented by differences in variance (see Fig. S2 in the supplementary material). On average, developmental genes were 1.389-fold derepressed in Ring1A/B-dKO ES cells, but 1.046-fold in Ring1B-KO (see Table S3 in the supplementary material). We confirmed these quantitative differences by evaluating the expression levels of several developmental regulators including Gata6 and Cdx2 by quantitative RT-PCR. These genes were significantly derepressed by conditional depletion of Ring1B, but the degree of derepression was higher in the Ring1A/B-dKO than in the Ring1B-KO cells(Fig. 1E). Therefore, Ring1A and Ring1B appear to act in a compensatory manner to repress the expression of developmental regulators in ES cells and consequently contribute to the maintenance of ES cells in an undifferentiated state. The phenotypic differences between Ring1B-KO and Ring1A/B-dKO ES cells are likely to be due to exaggerated derepression of developmental regulators such as Gata6 and Cdx2 in the dKO cells.

We next examined whether the derepression of developmental regulators is accompanied by disruption of the core transcriptional regulatory circuitry in ES cells. We performed multicolor immunofluorescence analysis for Oct3/4 and for Gata4, which is also expressed in primitive endoderm. Oct3/4 was expressed relatively uniformly in all of the control cells and Gata4 was also expressed in most of the cells (Fig. 1F,upper panels). Four days after OHT administration, we found striking heterogeneity in Oct3/4 expression (Fig. 1F, lower panels). In the example illustrated, most of the cells are compacted; however, a subset of cells at the edge spread out from the colony and exhibit epithelial cell morphology. Most of these cells expressed Gata4 but not Oct3/4 (Fig. 1F,arrowheads in lower right panel), which is likely to be indicative of the onset of spontaneous differentiation. Taken together with the gene expression analysis, these results indicate that Ring1A/B contribute crucially to the repression of ES cell differentiation and therefore to the maintenance of ES cell identity.

We next used ChIP analysis to determine whether genes derepressed in Ring1A/B-dKO ES cells were direct targets of PRC1. As shown in Fig. 2A, we observed binding of Ring1B and Phc1, another component of PRC1, to Hoxb8, Gata6, Cdx2,Zic1 and T, all of which are derepressed in the Ring1A/B-dKO (Fig. 1E). Binding to all these genes was significantly reduced 2 days after administration of OHT in Ring1A^-/-;Ring1B^fl/fl;Rosa26::CreERT2ES cells, suggesting that Ring1B is essential for the establishment of PRC1 at their respective loci.

We next examined whether genes derepressed in Ring1A/B-dKO ES cells were bound by Ring1B at their promoters using a ChIP-chip approach(Fig. 2B; see Tables S4 and S5 in the supplementary material). We identified almost the same set of Ring1B targets that had been reported previously, if a certain threshold is adopted to distinguish genes bound by Ring1B (Boyer et al., 2006) (data not shown). We further clarified linear correlations between the degree of Ring1B binding and derepression in Ring1B-KO and Ring1A/B-dKO ES cells(Fig. 2B). These results indicate that Ring1A/B generally repress transcription by directly binding to the target loci in a dose-dependent manner.

Recent studies have demonstrated that PcG targets in ES cells are often characterized by a unique chromatin configuration, being simultaneously enriched for histone modifications associated with gene activity [histone H3 lysine 4 trimethylation (H3K4me3) and lysine 9/14 acetylation (H3Ac)] and modifications associated with PcG-mediated repression [specifically H3K27 trimethylation (H3K27me3)] (Azuara et al.,2006; Bernstein et al.,2006). With this in mind, we investigated changes in chromatin configuration upon Ring1A/B depletion. At a global level, PRC1-mediated H2Aub1 was rapidly depleted (Fig. 1B). By contrast, there was no detectable change in overall levels of either H3K4me3, H3Ac, H3K27me3 or PRC2 components (see Fig. S3 in the supplementary material). We then analyzed promoter regions of selected PcG target loci derepressed in Ring1A/B-dKO ES cells by ChIP. In addition to histone modifications, we analyzed binding of Eed and non-phosphorylated RNA polymerase II (RNAPII) (Fig. 2C). Levels of H3Ac, H3K4me3 and RNAPII binding were significantly increased, whereas those of Eed and H3K27me3 were decreased. Although the molecular mechanism for the decrease in Eed binding upon Ring1A/B depletion is unclear, it is possible that changes in chromatin structure caused by Ring1A/B depletion might secondarily affect Eed binding.

The above results demonstrate that Ring1A/B depletion converts the local chromatin from an inactive into an active configuration. This would suggest that the engagement of PRC1 is important to ensure robust silencing within chromatin domains that are predisposed to transcriptional activation(Azuara et al., 2006; Bernstein et al., 2006).

Given that Ring1A/B are required for the maintenance of ES cell identity,we next examined the relationship between Ring1A/B and the core transcriptional regulatory circuitry in ES cells, because a previous study demonstrated that OCT3/4, SOX2 and NANOG co-occupy a significant subset of PRC2 target genes in human ES cells (Lee et al., 2006). In fact, inactive genes bound by OCT3/4, SOX2 and/or NANOG in human ES cells are overrepresented among those genes more than 2-fold derepressed in Ring1A/B-dKO mouse ES cells (see Fig. S4 in the supplementary material).

To directly test whether Ring1A/B mediate transcriptional silencing by the core transcriptional circuitry we made use of Oct3/4 conditional knockout ES cells (ZHBTc4) (Niwa et al.,2000), comparing changes in gene expression in Oct3/4-KO and Ring1A/B-dKO cells. Because most Oct3/4-KO ES cells begin to exhibit trophectoderm-like morphology within 2 to 3 days after induction (Niwa et al., 2000),we analyzed RNA from ES cells 1 day after tetracycline (Tc) treatment, at which time Oct3/4 protein is extensively depleted, thus minimizing the contribution of secondary changes in gene expression resulting from differentiation. As controls, we also analyzed gene expression in Eed- and Dnmt1-KO ES cells. Dnmt1-KO ES cells self-renew but fail to undergo differentiation upon induction(Lei et al., 1996). Fold changes for respective probes determined against the parental cells were distributed on scatter diagrams and the correlation among respective KO ES cells was calculated (Fig. 3A). We found a strong correlation in total calculable genes between Oct3/4-KO and Ring1A/B-dKO ES cells (r=0.279). Overall gene expression in Ring1A/B-dKO ES cells also exhibited a strong correlation (r=0.359) with the Eed-KO, which might represent functional engagement of PRC1 and PRC2. By contrast, we found no correlation of the Dnmt1-KO with either the Ring1A/B-dKO(r=0.078) or Oct3/4-KO (r=-0.001). This analysis indicates that a large number of genes in ES cells are concurrently repressed by Oct3/4 and Ring1A/B.

Next we tested which genes regulated by Oct3/4 and Ring1A/B are important in maintaining ES cell identity. For this purpose, we extended the comparative gene expression analysis into sorted genes based on GO term categories. We found a comparable correlation in genes involved in regulation of transcription, transcription, development and apoptosis(Fig. 3B). Notably, the highest correlation was seen in genes involved in signaling pathways for Notch and Lif, both of which are implicated in stem cell maintenance(Androutsellis-Theotokis et al.,2006; Williams et al.,1988).

To test whether this observed correlation is statistically significant, we investigated the average expression changes caused by Ring1A/B depletion in genes more than 2-fold derepressed and repressed by Oct3/4 depletion. We further estimated the correlation of the expression changes with the degree of Ring1B binding to the respective genes. On average, derepressed genes in Oct3/4-KO cells were significantly derepressed in the Ring1A/B-dKO, as represented by a value at the zero point on the x-axis (P=1.23×10^-39)(Fig. 3C, red circle; see also Table S6 in the supplementary material). The degree of derepression in Ring1A/B-dKO showed a linear correlation with the degree of Ring1B binding (Fig. 3C, red circles). Concordantly, 120 out of 670 genes repressed by Oct3/4 were bound by Ring1B(Fig. 3D). By contrast,repressed genes in the Oct3/4-KO were only slightly repressed in the Ring1A/B-dKO, and these were genes that bound less Ring1B, whereas this was not the case at genes bound by Ring1B at intermediate or high levels(Fig. 3C, blue circles). Taken together with the spontaneous differentiation observed in Ring1A/B-dKO ES cells, Ring1A/B appear to be functionally linked with Oct3/4 in mediating ES cell identity.

To examine the molecular basis for the functional link between Oct3/4 and Ring1A/B, we used ChIP to investigate the effect of Oct3/4 deletion on the levels of Ring1B at selected targets bound by both Ring1B and Oct3/4 and/or Nanog (Boyer et al., 2005; Loh et al., 2006). Of the selected genes Cdx2, Hand1, Gata6 and Hoxb4 were derepressed 1 day after Oct3/4 depletion, whereas T, Otx2 and Hoxb8 were not (Fig. 4A). Ring1B binding was significantly reduced irrespective of transcriptional status, suggesting Oct3/4-mediated regulation Ring1B binding to the chromatin(Fig. 4B). We extended the analysis to examine whether this hierarchical link is applicable to other Ring1B target genes by the ChIP-chip approach. As shown in Fig. 4C, Ring1B binding to the promoter regions of the target genes was, on average, significantly reduced 2 days after Tc treatment of ZHBTc4 ES cells. Therefore, binding of Ring1B to the chromatin in ES cells is generally dependent on Oct3/4. It has been reported that chromatin binding of Ring1B is also regulated by PRC2 functions(Boyer et al., 2006). We thus extended the analysis to address whether Oct3/4-dependent chromatin-binding of Ring1B involves PRC2, and found that binding of Eed to these genes was significantly reduced as well (Fig. 4B). Taken together, these results indicate that Oct3/4 mediates local engagement of PRC1 and PRC2.

Since enforced Ring1A/B depletion led to a rapid increase in the H3K4me3 level at the target genes (Fig. 2C), we next investigated changes in the degree of H3K4me3 at these genes upon Oct3/4 depletion. Notably, the level of H3K4me3 was increased at Cdx2, Hand1, Gata6 and Hoxb4, whereas it was reduced or unchanged at T, Otx2 and Hoxb8, consistent with increased gene expression (Fig. 4B). This indicates that the global reduction of PcG binding is not the sole mechanism for the changes in gene expression profile observed in ES cells upon Oct3/4 depletion. Oct3/4 has been suggested to upregulate some target genes and downregulate others. Otx2 is one gene that has been experimentally verified to be upregulated directly by Oct3/4(Boyer et al., 2005; Loh et al., 2006; Matoba et al., 2006). Moreover, the co-occupancy of promoters by PRC2 and OCT4/SOX2/NANOG has been demonstrated only at the transcriptionally repressed genes in human ES cells(Boyer et al., 2006). Considered together with the phenotypic difference between Oct3/4-KO and Ring1A/B-dKO ES cells, the global reduction in PcG binding might be a part of the mechanism for the differentiation of Oct3/4-KO ES cells.

Finally, we examined whether the binding of Oct3/4 depends on Ring1A/B. The levels of Oct3/4 binding to the PcG target sites were either unchanged or slightly decreased 2 to 4 days after OHT treatment of the Ring1A/B-dKO ES cells (Fig. 4D). Considering that the overall level of Oct3/4 decreases slightly 4 days after OHT treatment (see Fig. S3 in the supplementary material), we conclude that Ring1A/B are not directly required for the binding of Oct3/4 to the target sites.

In summary, loss of Oct3/4 consistently results in the reduction of Ring1B and Eed binding at PcG target genes in ES cells. Oct3/4 may maintain the repression of essential developmental regulators such as Cdx2 and Gata6 (Fujikura et al.,2002; Niwa et al.,2005) by maintaining local engagement of PRC1 and PRC2.

To determine the molecular mechanism for the global reduction of Ring1B binding upon Oct3/4 depletion, we investigated the effect of Oct3/4 deletion on the level of PRC1 and PRC2 proteins. Although Ring1B expression was only minimally affected during the first 48 hours of Tc treatment, expression of Phc1, Eed and Suz12 was significantly reduced(Fig. 5A). The decrease in Phc1 and PRC2 proteins was accompanied by a significant reduction in their respective transcript levels, whereas this was not the case for other PRC1 components, including Ring1B and Bmi1(Fig. 5B). Therefore, Oct3/4 regulates the expression of PRC1 and PRC2 components, and this may partly involve transcriptional regulation.

We also investigated whether the physical interaction of Ring1B with the Rex1 (Zfp42 - Mouse Genome Informatics) complex(Wang et al., 2006) could be extended to Oct3/4. Significant amounts of Oct3/4 and Ring1B as well as Rybp,a Ring1B-binding protein (Garcia et al.,1999), were found to form complexes in ES cells, whereas the PRC2 protein Suz12 did not co-immunoprecipitate with either Oct3/4 or Ring1B(Fig. 5C). Since reciprocal co-immunoprecipitation of Oct3/4 and Ring1B was not affected by the addition of ethidium bromide, which is known to disrupt protein-DNA interactions without affecting protein-protein interactions(Lai and Herr, 1992)(Fig. 5C, right), this interaction is not mediated by genomic DNA. This result suggests that the local binding of PRC1 to chromatin might involve direct interactions between PRC1 and protein complexes that include Nanog and/or Oct3/4. Taken together,these results suggest that PRC1 is linked to the core transcriptional regulatory circuitry at multiple levels.

We went on to address whether PRC1/2 binding depends solely on Oct3/4 or on the regulatory system for ES cell self-renewal. Various differentiation cues have been demonstrated to disrupt ES cell self-renewal maintained by the core transcriptional regulatory circuitry. For example, Gata6 has been thought of as a downstream effector of Nanog, and enforced Gata6 expression induces the differentiation program towards extraembryonic endoderm(Chazaud et al., 2006; Fujikura et al., 2002). We therefore examined the effects of the enforced expression of Gata6 on PRC1/2 engagement at Ring1B targets using ES cells expressing a Gata6-GR fusion protein (G6GR) (Shimosato et al.,2007). As previously described, G6GR cells rapidly lose ES-cell-like morphology and show dispersed visceral endoderm-like morphology upon Gata6 activation by the administration of dexamethasone (Dex) (see Fig. S5 in the supplementary material).

We first analyzed the effect of Gata6 activation on the level of PRC1 and PRC2 proteins (Fig. 6A). The level of Ring1B protein was unaffected or only minimally affected during the first 2 days of Dex treatment, but was slightly decreased by day 3. The levels of Phc1, Ezh2, Eed and Suz12 proteins were significantly reduced after Dex treatment, whereas H3K27me3 was unaffected.

Next we used microarray analysis to evaluate the effect of enforced Gata6 expression on gene expression, and compared this profile with those of Oct3/4-KO and Ring1A/B-dKO ES cells. We found a comparable correlation between Gata6-differentiated and Ring1A/B-dKO ES cells(r=0.297), between Oct3/4-KO and Gata6-differentiated ES cells (r=0.318), and between Oct3/4-KO and Ring1A/B-dKO ES cells (r=0.279), not only in terms of total calculable genes but also in genes involved in development, transcription,apoptosis and cell cycle (Fig. 6B). Therefore, a subset of genes induced by enforced Gata6 activation is correlated with those regulated by Oct3/4 and Ring1A/B,suggesting an extension of the functional link to Gata6.

We then investigated whether Gata6 activation affects the levels of Ring1B and Eed at selected Polycomb targets. Of the selected genes, the expression level was increased for Gata6, Cdx2, Zic1 and T, whereas it was reduced or unchanged for Hoxb8 and Hand1 upon Gata6 expression (data not shown). Ring1B binding was significantly reduced irrespective of transcriptional status(Fig. 6C). Phc1 and Eed binding was also reduced (Fig. 6C),consistent with the observed reduction in protein levels(Fig. 6A). These results indicate that Gata6 activation impacts upon gene expression in similar manner to Oct3/4 depletion. Therefore, it is likely that the local engagement of PRC1 depends on the core transcriptional regulatory circuitry rather than solely on Oct3/4 in ES cells.

In this study, we first demonstrate that Ring1A/B are required to maintain ES cells in an undifferentiated state by repressing the expression of developmental regulators that direct differentiation of ES cells. This process involves local inhibition of chromatin remodeling via direct binding of PRC1. We further show that Ring1A/B-mediated PcG silencing is hierarchically linked to the core transcriptional regulatory circuitry and that this linkage can be abolished by developmental cues that negatively regulate the core circuitry(Fig. 7). Developmental regulators repressed by this epistatic link, such as Cdx2 and Gata6, are shown to be predisposed for active transcription. Active and reversible repression by this epistatic link is important to potentiate ES cells so that they can respond to developmental cues appropriately and,consequently, it may underpin the maintenance of pluripotency. Therefore, our data show that Ring1A/B are instrumental for the core transcriptional regulatory circuitry to maintain ES cell identity.

These and previous studies suggest that PcG proteins are linked to the core transcriptional regulatory circuitry at multiple levels(Fig. 3)(Lee et al., 2006). Oct3/4 is likely to recruit Ring1B to its targets via direct interactions and also to induce the expression of PRC1 components via a transcriptional regulatory mechanism (Fig. 5). Moreover,it is notable that Oct3/4 loss displaces Ring1B from most of its target genes,which are not necessarily functional targets of Oct3/4(Fig. 4C). This prompts us to postulate an activity that modulates Ring1B recruitment under the regulation of Oct3/4. Indeed, the RING1 and YY1 binding protein, Rybp, potentially fulfils such a linking role between Oct3/4 and Ring1B because Rybp is able to form complexes with both proteins (Fig. 5C) (Wang et al.,2006). Such multiple interactions might enable coordinated displacement of PRC1 and PRC2 from their target genes upon disruption of the core circuitry by differentiation cues. Since forced depletion of Ring1A/B leads to spontaneous differentiation of ES cells,dissociation of PRC1 from the targets may be functionally implicated in the differentiation process. We presume that the dissociation of PRC1 and PRC2 is a prerequisite for the subsequent association of other chromatin modifiers such as Trithorax group proteins, which catalyze local hypertrimethylation of H3K4 upon Oct3/4 depletion (Dou et al.,2005; Wysocka et al.,2003). This is supported by our results shown in Fig. 2C, and by previous experiments showing that the SWI-SNF complex is unable to remodel polynucleosomal templates bound by PRC1 in vitro(Shao et al., 1999). Therefore, the global enhancement of chromatin remodeling at developmental genes might be one of the essential events in promoting proper differentiation of ES cells.

The reversibility of Polycomb binding to the targets regulated by the core transcriptional regulatory circuitry and by differentiation cues might confer self-renewing and differentiation capacities to ES cells. Intriguingly, PcG silencing has been suggested to be involved in the function and maintenance of tissue stem and cancer cells, which are also characterized by both self-renewal and differentiation potency(Lessard and Sauvageau, 2003; Molofsky et al., 2003; Ohta et al., 2002; Park et al., 2003; Villa et al., 2007). For example, Bmi1 loss promotes differentiation of hematopoietic stem cells (HSCs) and premature senescence of neural stem cells, whereas forced expression of Bmi1 enhances symmetrical cell division of HSCs(Iwama et al., 2004; Molofsky et al., 2005). Recently, it has been reported that knockdown of SUZ12 in acute promyelocytic leukemic cells results in myeloid differentiation(Villa et al., 2007). It is thus likely that similar molecular mechanisms identified in ES cells that involve Ring1A/B might operate in the maintenance and differentiation of various tissue stem cells and cancer cells. Since Oct3/4 and Nanog are not expressed in most somatic cells, other downstream effectors expressed in common among the stem cells might be more directly involved in the regulation of Polycomb binding. Alternatively, other factors specifically expressed in tissue and/or cancer stem cells might substitute for the action of Oct3/4 or Nanog. Further studies will be needed to clarify these issues.

We thank Hitoshi Niwa for providing ZHBTc4, G6GR and ZHTc6 cells. This work was supported in part by a grant from the Genome Network Project (to H.K.) and a grant-in-aid for Scientific Research on Priority Areas (17045038, to M.E.)from the Ministry of Education, Culture, Sports, Science and Technology,Japan.