Distinct roles of Polycomb group gene products in transcriptionally repressed and active domains of Hoxb8

In Drosophila, the regionally restricted expression of homeotic genes (Hox genes) is maintained by two groups of proteins: the trithorax group(trxG) and Polycomb group (PcG). The trxG gene products are required to maintain the activity of Hox genes in the appropriate segments, whereas the PcG are involved in their repression(Paro, 1995; Pirrotta, 1997). Indeed, some of the PcG genes have been identified as suppressors of trxG mutations(Kennison and Tamkun, 1988). The trxG and PcG genes encode nuclear factors that, by forming multimeric protein complexes on the chromatin, freeze transcriptional states determined early in embryogenesis. A major function of trxG gene products appears to concern the remodeling of chromatin structure, as several of these genes encode subunits of the SWI/SNF chromatin remodeling complex and associate with catalytic activities that modify core histone tails(Milne et al., 2002; Nakamura et al., 2002; Tamkun et al., 1992). PcG gene products have been shown to form at least two types of multimeric protein complexes (Shao et al., 1999; Czermin et al., 2002; Muller et al., 2002; Cao et al., 2002). The first type of complex, known as class 1, includes Extra sex combs (Esc) and Enhancer of zeste [E(z)], and its association with histone deacetylase and methyltransferase activities suggests a function in the modification of histone tails (Czermin et al.,2002; Muller et al.,2002; Cao et al.,2002). The second type of complex (class 2) contains Polycomb(Pc), Posterior sex combs (Psc), Polyhomeotic (Ph), Dring (Sex comb extra;Sce) and Sex comb on midleg (Scm) in Drosophila(Shao et al., 1999). It has been shown that, by interacting with DNA regulatory sequences termed PcG response elements (PREs), this complex can maintain the transcriptional silence of target genes. Recently, it has been reported that class 1 PcG-mediated trimethylation of histone H3 lysine 27 (H3-K27) serves as a signal for the recruitment of class 2 PcG complexes, which in turn may silence target genes flanked by PREs (Cavalli and Paro, 1998; Wang et al.,2004b). Purified class 2 core complexes efficiently block SWI/SNF-dependent remodeling of nucleosomal arrays and subsequent transcription in vitro (King et al.,2002; Levine et al.,2002; Shao et al.,1999).

In mammals, it has been demonstrated that the expression of Hox cluster genes depends on the control of mammalian homologues of trxG and both classes of PcG gene products. Silencing of several Hox gene expressions has been observed in mice lacking Mll, a mammalian homologue of Drosophila trx, between 8.5 and 9.5 days post coitus (dpc)(Yu et al., 1998). Conversely,a hypomorphic mutation of Eed, a homologue of Drosophila Esc, revealed its involvement in repressing Hox gene expression(Schumacher et al., 1996). As Mll and class 1 PcG complex are known to be associated with activities that modify histone tails(van der Vlag and Otte, 1999),such modifications could be involved in maintaining spatially restricted expression of Hox cluster genes. Indeed, it has been shown that the Mll protein regulates the acetylation of lysine 4 of histone H3 (H3-K4) at several Hox genes, whereas Ezh2, a homologue of E(z), mediates H3-K27 trimethylation(Milne et al., 2002; Nakamura et al., 2002). The involvement of class 2 PcG in this maintenance has been revealed previously by the effect of mutations in Bmi1 (Pcgf4 - Mouse Genome Informatics), Rnf110 (Mel18; Pcgf2 - Mouse Genome Informatics), Phc1 (rae28) and Rnf2(Ring1B) on Hox gene expression(Akasaka et al., 1996; van der Lugt et al., 1996; Suzuki et al., 2002; Takihara et al., 1997). Notably, in mice doubly deficient for Rnf110 and Bmi1,homologues of Drosophila Psc, Hoxb6 expression was normally localized in caudal tissues of 8.5 dpc embryos, but was progressively de-repressed cranially thereafter (Akasaka et al.,2001). Thus, class 2 PcG proteins may participate in maintaining transcriptionally silent states of Hox genes outside their expression domains. However, to date, the molecular mechanisms used by mammalian trxG and PcG gene products to regulate Hox genes is not well understood, in part because the relationship between PcG protein binding and histone tail modifications has not yet been widely documented around the Hox loci in developing embryos. We have addressed this issue by documenting the association of PcG proteins,H3-K9 acetylation, H3-K4 methylation (marks of transcriptionally active chromatin) and H3-K27 trimethylation (a mark of transcriptionally inactive chromatin) to the genomic region flanking Hoxb8, a Hox gene known to require the class 2 PcG proteins for its posterior restriction.

We show that the association of PcG proteins, H3-K9 acetylation, H3-K4 methylation and H3-K27 trimethylation around Hoxb8 differs in embryonic tissues expressing and not expressing the gene. By using mutant alleles for Rnf2, which encode constituents of class 2 PcG complexes,and Suz12 causing a strong decrease in H3-K27 methylation, we show that the recruitment of the class 2 PcG complex, mediated by trimethylated H3-K27, plays a decisive role in maintaining the repression of Hox genes outside their expression domain, as it is the case in Drosophila. The positive role of class 2 PcG complex proteins in the transcriptionally active domain was shown to involve the regulation of H3-K9 acetylation.

The generation of Rnf2^fl allele and Rnf2 mutant ES cells has been described previously (de Napoles et al., 2004). The generation of Suz12-deficient mice and ES cells will be described in a separate publication. CRE-mediated deletion of Rnf2 in MEF cell lines using adenovirus expressing the CRE recombinase (AdCre)was carried out as described (Kanegae et al., 1995). AdCre virus was concentrated to be 2.0×10^-9 i.f.u./ml and mouse embryonic fibroblast (MEFs) were infected with AdCre virus at MOI5.

Antibodies used are listed in the Table 1.

Embryonic tissues of interest were dissected from embryos, at the different developmental stages, under a stereomicroscope. Chromatin fraction from the embryonic tissues was purified by CsCl isopycnic centrifugation as described in Fig. S1 in the supplementary material(Orlando et al., 1997). This process was bypassed in some cases. Correct amounts of NaCl and NP-40 were added to the chromatin fraction or whole cell lysates (WCE) in order to perform optimal immunoprecipitation for each antibody. Precleared protein extracts were incubated with the correct amount of antibodies, at 4°C with rocking, for 2 hours to overnight. Immune complexes were captured after 3 hours incubation with Protein A Sepharose beads.

To isolate genomic DNA from immune complexes, beads were treated with 50μg/ml of RNaseA at 37°C for 30 minutes followed by overnight incubation with 500 μg/ml proteinase K/0.5% SDS at 37°C. After 3 hours heating at 65°C for reverse crosslinking, supernatants were collected, extracted by phenol-chloroform and concentrated by ethanol precipitation. Genomic DNA was also isolated from the original chromatin fraction or WCE through the same procedure as described above and designated as `Input' DNA (see Figs 1, 2, 3). To measure the DNA yield after immunoprecipitation, the aliquots of immunoprecipitated DNA were electrophoresed for 5 minutes in an agarose gel, next to serially diluted input DNA and band-intensities were compared after ethidium bromide staining(see Fig. S2A in the supplementary material).

Equivalent amounts of immunoprecipitated DNA to that of `Input' DNA loaded in lane `1' were subjected to PCR reactions. Usually, 10 to 20 ng of genomic DNA was used. Mock-immunoprecipitated DNA (A- and P-) derived from the same volume of the chromatin fraction as used for anti-Rnf2 immunoprecipitation were subjected to the PCR. To carry out semi-quantitative PCR, serially diluted `Input' DNA and immunoprecipitated DNA were used as templates. The relative quantity of each genomic region in immunoprecipitated genomic DNA was estimated by referring to the serial dilutions of `Input' DNA isolated from the initial lysates and an enrichment value was determined. Every series of experiments were performed at least three times (see Fig. S3 in the supplementary material). Primers used in this study are listed in Table 2. ChIP analysis by using ES cells was performed as described (Isono et al., 2005a).

RNA extraction from ES cells, reverse transcription and PCR reaction was performed as described previously (Isono et al., 2005a). Quantity of total cellular RNA subjected to reverse transcriptase (RT)-PCR for Hox genes was adjusted by the expression ofβ-actin. Hoxb8 expression shown in Fig. 3A was quantified by referring to Gapdh expression by real-time PCR analyses using Mx3005P multiplex quantitative PCR systems (Stratagene). Preparation of whole-cell extracts from embryos, ES cells and MEFs and western blot analysis were performed as described previously (Isono et al., 2005a).

First, the issue of whether PcG-silencing of Hox genes in developing embryos involves the direct binding of PcG gene products to Hox genes was addressed. PcG associations at genomic regions flanking Hoxb8 were compared between transcriptionally repressed cranial and active caudal tissues by ChIP using embryonic tissues at 12.5 dpc. As the rostral boundary of Hoxb8 expression is at the level of the 7th prevertebrae in the paraxial mesoderm and at the caudal hindbrain in the neural tube, embryos were dissected transversely at the levels of the pinna primordium and posterior edges of the forelimb buds, after the removal of internal viscera and hindlimb buds. Only cranial and caudal tissues were subjected to the analysis(Fig. 1A). Chromatin fractions of the respective embryonic tissues were purified and subjected to immunoprecipitation using antibodies against Rnf2, Phc1, Ring1, Cbx2 and Rnf110 (see Fig. S1 in the supplementary material)(Suzuki et al., 2002). Quantity of immunoprecipitated genomic DNA was measured by referring to serially diluted genomic DNA isolated from the initial lysates (see Fig. S2A in the supplementary material). Usually, for a given volume of chromatin fraction, the amount of DNA immunoprecipitated by anti-Rnf2 antibodies was four- to eightfold higher than that brought down in the absence of primary antibody (see Fig. S2A in the supplementary material). Amounts of DNA from respective tissues were further adjusted by PCR for the tbx2 gene,which is also bound by Rnf2 in embryonic tissues (see Fig. S2B in the supplementary material). Equivalent amounts of DNA immunoprecipitated from the anterior and posterior tissues were subjected to semi-quantitative PCR using pairs of primers defining discrete regions in the Hoxb locus(Fig. 1F, top; see Table 2). Serially diluted genomic DNA isolated from the initial lysates was also subjected to the PCR to evaluate the enrichment value for each region of Hoxb8 in immunoprecipitated materials (Fig. 1B-E). The adam34 gene, which is expressed exclusively in adult testes, is not bound by any of Rnf2, Ring1, Phc1, Cbx2 or Rnf110 in embryonic tissues and thus turned out to serve as a negative control(Brachvogel et al., 2002)(Fig. 1B-E).

The association of Rnf2 with regions 1, 2, 3, 9, 10, 11, 13, 15, 16 and D1 was more than four times different in cranial and caudal tissues, whereas this was not the case for regions 4, 5, 6, 7, 8, 12, 14, 19, 20, 21, 22, 23, 24, 25 or D2 (Fig. 1B,F, Supplemental Fig. 3). In region 17 and 18, no significant association of Rnf2 was observed. Regions 1 and 2, and 10 and 11, correspond to the Hoxb8 regulatory regions BH1100 and KA that have previously been identified by transgenic approaches; regions 3 and 13 include the promoters of Hoxb8 and Hoxb7, respectively(Charité et al., 1995; Vogels et al., 1993). Therefore, Rnf2 differentially binds to the proximal cis-regulatory elements of 5′ Hoxb genes in cranial and caudal embryonic tissues. Likewise, a significant differential association of Rnf2 to the distal element(DE) located between Hoxb4 and Hoxb5(Oosterveen et al., 2003) was also seen in cranial and caudal tissues. Similar to Rnf2, Ring1 association was seen in cranial tissues, except for regions 3 and 7(Fig. 1D). In contrast to Rnf2 binding, the association of Phc1 extended through all regions examined except for region 8, without any obvious differences between cranial and caudal tissues (Fig. 1C). Nevertheless, in region 3, the association did appear to be significantly stronger in the caudal tissues. Furthermore, there was no significant difference between cranial and caudal tissues with respect to the chromatin association of Cbx2 (Fig. 1D). Therefore, these results indicate that Phc1 and Cbx2 are bound to Hoxb8 genomic region irrespective of the transcriptional state of the gene. Rnf110 association to regions 3 and 10 was seen in the cranial tissues only, but to regions 1, 2, 7 and 14 it was seen in both cranial and caudal tissues (Fig. 1E). These experiments were performed three times with similar results (see Fig. S3 in the supplementary material). In summary, the complete form of the class 2 PcG complexes predominantly associate with Hoxb8 in tissues where the gene is repressed, whereas form(s) lacking at least the Rnf2 component also bind in tissues actively expressing the Hox gene.

Following this, the issue of the degree of H3-K9 acetylation and H3-K4 methylation at Hoxb8 genomic regions in expressing and non-expressing embryonic tissues, was addressed. These epigenetic marks have been shown to be one of the prerequisites for the efficient recruitment of TBP through the interaction of TAF1 with the H3-K9 and K14 acetylated residues(Agalioti et al., 2002). A linear correlation between histone modifications and the transcriptional state of Hox genes has been seen in developing embryos at Hoxd4 locus and MEFs that are derived from Mll mutants and wild-type controls(Rastegar et al., 2004; Milne et al., 2002). Therefore, as H3-K9 acetylation and H3-K4 methylation may be prerequisites for active transcription of Hox genes, these extents were compared between anterior and posterior embryonic tissues. H3-K9 acetylation and H3-K4 methylation around the Hoxb8 was more abundant in the transcriptionally active caudal embryonic part than in the non-expressing anterior part as reported by Rastegar et al.(Rastegar et al., 2004)(Fig. 2A,B).

The methylation of H3-K9 and H3-K27 constitutes imprints for transcriptionally silent chromatin and is involved in many functions,including the formation of centromeric heterochromatin, X inactivation,PcG-mediated gene silencing and transcriptional repression at euchromatic positions (Fischle et al.,2003; Lachner et al.,2003). In particular, trimethylation of H3-K27 has been shown to be mediated by class 1 PcG complexes (Cao et al., 2002). The degree of H3-K9 and H3-K27 methylation at regions 2, 3 and 8 was compared between cranial and caudal embryonic regions. At all the regions examined, H3-K27 trimethylation was more abundant in the transcriptionally silent cranial embryonic part than in the Hox-expressing caudal part (Fig. 2B). No significant levels of di- or trimethylation of H3-K9 or dimethylation of H3-K27 were seen at any of the regions examined (Y.F. and H.K.,unpublished).

Loss of spatial restriction of Hoxb6 expression, between 8.5 and 9.5 dpc, has been reported as resulting from the Rnf110/Bmi1double mutation (Akasaka et al.,2001). Similarly, Mll deficiency progressively silences Hox gene expression within this time window(Yu et al., 1998). This suggests that this developmental stage might be the crucial period when PcG association and H3-K9 acetylation at Hoxb8 play their role in modulating gene expression.

Prior to the ChIP analyses, we re-examined the expression of Hoxb8gene in the cranial and caudal tissues in quantitative manner by using real-time PCR (Fig. 3A). As reported previously by extensive in situ hybridization analyses, caudally restricted expression of Hoxb8 was already established at 8.0 dpc and its relative quantity progressively increased until 12.5 dpc(Deschamps and Wijgerde,1993). We carried out the kinetic analyses of the Rnf2 association, H3-K9 acetylation and H3-K27 trimethylation in early to later developmental stages across the Hoxb8 promoter region (region 3 in Fig. 1F). In 8.0 dpc embryos,no Rnf2 association was seen at the six-somite stage. It was first seen,exclusively in cranial tissue, at the eight-somite stage(Fig. 3B). This was also the case at later stages up to 12.5 dpc (Fig. 3C-E). The relative quantity of Rnf2 association gradually increased and reached a maximal level at 10.5 dpc(Fig. 3E). H3-K9 acetylation was observed in the cranial region at the six-somite stage and in both cranial and caudal regions at the eight-somite stage. It was present at higher levels in the caudal region in 9.5 and 10.5 dpc embryos than in 8.0 dpc. H3-K27 trimethylation at region 3 continued to be seen in the cranial, but not in the caudal, region from 8.5 to 12.5 dpc. In summary, a differential association of Rnf2 with the Hoxb8 promoter region, region 3, was established from 8.0 dpc onwards, and reached completion around 10.5 dpc(Fig. 3G)(Deschamps and Wijgerde,1993). Relative amounts of acetylated H3-K9 in this region also increased up to 12.5 dpc. Likewise, differential trimethylation of H3-K27 was already established at 9.5 dpc and maintained to 12.5 dpc. These observations indicate that the Rnf2 association, H3-K9 acetylation and H3-K27 trimethylation may be involved to maintain the spatially restricted expression of the Hoxb8.

Differential association of Rnf2 to Hoxb8 genomic surrounding in tissues not expressing Hoxb8 suggests its repressive role in Hoxb8 transcription. This hypothesis is supported by our previous experiments that used an hypomorphic allele of Rnf2, in which the de-repression of several Hox genes was seen at 11.5 dpc(Suzuki et al., 2002). However, it was not determined whether Rnf2 association was functionally involved in the maintenance rather than early establishment of this repressed status. To address this question, we conditionally depleted functional Rnf2 after the expression domain of Hoxb8 was established. Because of early embryonic lethality of Rnf2-null mice, we generated a conditional allele designated as Rnf2^fl in which exon 2,containing the ATG initiation codon, was flanked with loxP sites(Voncken et al., 2003; de Napoles et al., 2004). Primary MEFs were derived from the cranial part of Rnf2^fl/fl 9.5 dpc embryos in which Hoxb8expression was repressed. Subsequent deletion of Rnf2 was achieved by infection of MEF cultures with adenovirus expressing the CRE recombinase(Ad-Cre) as described (de Napoles et al.,2004). Rnf2 depletion was evident in western analysis of infected cell culture (Fig. 4A) and took 2 days for completion. The analysis of Hoxb8 expression continued for 4 days after Ad-Cre infection. Two independent experiments demonstrated more than a 16-fold increase of Hoxb8 expression in infected MEFs compared with the uninfected control. Therefore Rnf2 association is essential for the maintenance of the transcriptional repression of Hoxb8.

After this, the repressive role of Rnf2 in undifferentiated embryonic stem(ES) cells was investigated because Hox repression was accompanied by both chromatin condensation of the Hox cluster and the location of the Hox genes within the chromosome territory, whereas decondensation and reorganization accompanied Hox gene expression both in ES cells and the developing neural tube (Chambeyron and Bickmore,2004; Chambeyron et al.,2005).

Seven Rnf2^fl/fl ES cells have been derived(de Napoles et al., 2004). Rnf2^-/- ES cell derivatives of one of these lines, the male ES cell line, 13-3, were generated by CRE-mediated excision of exon 2(Fig. 4B). Loss of functional Rnf2 resulted in de-repression of Hoxa1, Hoxa7, Hoxb3, Hoxb8, Hoxb13,Hoxc9, Hoxd8 and Hoxd10 but not of Hoxc8(Fig. 4C). To test if de-repression of Hox genes is a direct consequence of mutating Rnf2,complementation experiments were carried out, transfecting mutant ES cells with a construct expressing Myc-tagged Rnf2. As shown in Fig. 4B, transfected mutant ES cells expressed transgene-encoded Rnf2 at a level equivalent to half that of endogenous Rnf2. RT-PCR analysis revealed that the expression of Hoxa1,Hoxb3, Hoxb8 and Hoxd8 was repressed albeit not to the same degree as in the parental Rnf2^fl/fl ES cells(Fig. 4C). These results confirm that Rnf2 association is required to mediate the transcriptional repression of Hox genes in ES cells.

Next, the method by which Rnf2 deficiency impacts on the functions of class 2 PcG complexes was investigated. As it has been suggested that Ring1 is an important component in the stabilization of the Polycomb core complex in Sf9 cells (Francis et al., 2001),the expression levels of other components of class 2 PcG complexes in Rnf2^-/- ES cells was examined. The expression of both Phc1 and Cbx2 gene products was obviously reduced in Rnf2^-/- ES cells (Fig. 4D; M.E. and H.K.,unpublished). By contrast, the expression of RYBP, another Rnf2-binding protein, which is not found in hPRC-H complex, was not altered(Garcia et al., 1999)(Fig. 4D). As the transcription of Phc1 and Cbx2 was not altered in Rnf2^-/- ES cells, Rnf2 loss specifically affects the expression of Phc1 and Cbx2. It is thus likely that Rnf2 impacts the Hox expression by regulating the stability of the class 2 PcG complexes.

The coincidence of Rnf2 association and H3-K27 trimethylation in transcriptionally repressed region further prompted us to ask whether de-repression of Hox genes in Rnf2^-/- ES cells involves a change of H3-K27 trimethylation across the Hox genomic regions. The expression of Suz12 and Ezh2 were not significantly changed in Rnf2^-/- ES cells (Fig. 4D). Concordant with this result, ChIP analyses revealed that local level of H3-K27 trimethylation at Hox promoter regions were almost unchanged (Fig. 4E). Therefore,Rnf2 deficiency affects Hox gene expressions without changing local H3-K27 trimethylation.

In Drosophila, trimethylation on H3-K27 has been shown to facilitate the recruitment of class 2 PcG complexes via direct interaction between trimethylated H3-K27 and the chromodomain of Pc(Cao et al., 2002; Czermin et al., 2002; Muller et al., 2002). Therefore, Hox gene expression and Rnf2 association at the Hoxblocus, in the absence of trimethylated H3-K27, was examined using Suz12^-/- ES cells. We have independently generated a loss-of-function allele of Suz12 and the homozygous mutants exhibited a phenotype almost identical to that reported by Pasini et al.(Pasini et al., 2004) (K.I. and H.K., unpublished). Suz12^-/- ES cells were derived from crosses of heterozygous mutants. The absence of Suz12 reduced the levels of H3-K27 tri- and dimethylation to less than 10% of the wild type but did not significantly change H3-K9 methylation(Pasini et al., 2004). First,the association of class 1 PcG and H3-K27 trimethylation at the Hox promotor regions was investigated. Anti-Suz12, -Eed and -trimethylated H3-K27 antibodies specifically immunoprecipitated significant amounts of Hox promoter fragments from the wild-type ES cells (Fig. 5A). In Suz12^-/- ES cells, no Hox DNA could be detected at all upon anti-Suz12 immunoprecipitation. The loss of Suz12 significantly reduced Eed association and H3-K27 trimethylation at Hox genes. Therefore, class 1 PcG complexes associate locally to Hox genes and mediate local H3-K27 trimethylation in undifferentiated ES cells. Second, the expression of Hoxa1, Hoxa4, Hoxb1, Hoxb3, Hoxb4, Hoxb6, Hoxb8, Hoxb9and Hoxc6 was compared between wild-type and Suz12^-/- ES cells by RT-PCR. Suz12^-/-ES cells were shown to express more Hox gene transcripts than the wild type(Fig. 5B). Therefore, local H3-K27 trimethylation mediated by class 1 PcG complexes, or association of the complexes, are likely to be required in order to mediate repression of Hox genes. We went on to examine Rnf2 association in Suz12^-/-ES cells. Rnf2 association to the Hox promoters was significantly reduced in Suz12^-/- ES cells (Fig. 5A). Therefore H3-K27 trimethylation may be a prerequisite for the association of Rnf2 with Hox regulatory regions, as already demonstrated in Drosophila. The role of class 1 PcG complexes in the recruitment of the class 2 complexes via protein-protein interactions is not necessarily excluded.

Results from this study also show that PcG gene products associate to Hoxb8 genomic regions in the transcriptionally active caudal embryonic tissues. This is in line with the recently uncovered positive function of class 2 PcG complexes (de Graaff et al., 2003). Null mutations in the Rnf110 and Phc1 loci have been shown to decrease the transcription level of endogenous Hoxb1, and to severely impair the transcription from Hox/lacZ reporters and knock-in loci(de Graaff et al., 2003). In addition, a positive action of PcG complexes explains the drop in gene expression levels within the Hoxb8 expression domains known to occur around 9.5 dpc in Bmi1/Rnf110 and Phc1/Phc2 double mutants (Akasaka et al., 2001; de Graaff et al., 2003; Isono et al., 2005b). As the level of histone H3 acetylation correlates with Hox gene expression in Mll mutants and controls, and Mll and Bmi1 proteins display discrete subnuclear colocalization, we investigated whether the drop in Hoxb8expression level in Rnf110/Bmi1 and Phc1/Phc2 double mutants involved a change in H3-K9 acetylation (Hanson et al.,1999; Milne et al.,2002; de Graaff et al.,2003). The degree of H3-K9 acetylation in the caudal region of Rnf110^-/-Bmi1^-/- embryos was much lower than in Rnf110^+/+Bmi1^+/-,whereas the H3-K9 acetylation at the β-actin promoter was almost equivalent (Fig. 6A). Similarly, H3-K9 acetylation level in the caudal part of Phc1^-/-Phc2^-/- was equivalent to the cranial part whereas craniocaudally graded acetylation was seen in Phc1^+/-Phc2^+/- embryos(Fig. 6B). Therefore association of class 2 PcG gene products to Hoxb8 is required for the maintenance of H3-K9 acetylation in transcriptionally active regions.

The main outcome of this study was to show that binding of a specific,Rnf2-containing form of the class 2 PcG complex, as well as H3-K27 trimethylation marking inactive chromatin, correlates with the maintenance of transcriptional silencing of a Hox gene in developing embryos. Moreover, the results demonstrated that genetic impairment of both PcG binding, and H3-K27 trimethylation leads to Hox gene derepression, and that H3-K27 trimethylation is required for PcG binding. In addition, we showed that the establishment of differential PcG binding and histone marks in expressing and non-expressing embryonic tissues occur in the same developmental time window as when Hox genes are deregulated in PcG mutants.

Rnf2 association to known regulatory elements of the Hoxb8 gene is seen predominantly in transcriptionally silent anterior embryonic tissues,whereas the binding of other PcG class 2 members, Phc1 and Cbx2, is observed at all AP levels, irrespective of transcriptional status. This implies that different forms of class 2 PcG complexes bind to the Hoxb genomic region in embryonic domains where the gene is transcriptionally active and repressed. This is reminiscent of previous findings in the Engrailed/Inv/GeneVIcomplex in Drosophila SL-2 cells, where the Pc protein is exclusively associated with transcriptionally silent genes, while Ph and Psc are present irrespective of the transcriptional status(Strutt and Paro, 1997). Therefore the complete, `perfect' form of the class 2 PcG core complex may mediate transcriptional repression more efficiently than form(s) lacking the Rnf2 component. If this is the case, incorporation of the Rnf2 component into the complex might be a limiting process to mediate transcriptional repression and regulate its stability (Francis et al., 2001). It is also possible that the role of Rnf2 is mediated through its E3 ubiquitin ligase activity directed to histone H2A(Wang et al., 2004a; de Napoles et al., 2004).

Transcriptional repression of Hox genes in the developing embryo has been shown to correlate with the association of Rnf2-containing class 2 PcG complexes and H3-K27 trimethylation. De-repression of Hox genes in Rnf2 and Suz12 mutant cells reveal the requirement of both Rnf2 association and H3-K27 trimethylation in the mediation of this transcriptional repression. As Rnf2 association to Hox genes is reduced in Suz12 mutant ES cells and Rnf2 mutation alters Hox expression without changing local levels of H3-K27 trimethylation, H3-K27 trimethylation mediated by class 1 PcG complexes at Hox genes may facilitate subsequent binding of Rnf2-containing PcG complexes. Recruitment of Rnf2-containing PcG complexes may in turn prevent the access of nucleosome remodeling factors, such as SWI/SNF complex, leading to the formation of a repressed chromatin status (Shao et al.,1999; Levine et al.,2002). Therefore, molecular circuitry underlying PcG silencing of Hox genes seems to have been evolutionarily conserved between Drosophila and mammals. It is also notable that Cbx2, a homologue of Drosophila Pc, binds to Hoxb8 in transcriptionally active embryonic tissues, despite the lack of histone H3 trimethylated at K27. This is consistent with previous biochemical data that have shown the association of purified or reconstituted PcG complexes with the nucleosomal templates lacking histone tails (Shao et al.,1999). The implication of these findings is that there are at least two different means by which class 2 PcG complexes bind to the chromatin, and that the association, which involves trimethylated H3-K27,mediates the repression at the Hox genes in vivo(Cao et al., 2002; Czermin et al., 2002; Muller et al., 2002).

The maintenance of regionally restricted expression of Hox genes is likely to involve H3-K9 acetylation and H3-K4 methylation(Milne et al., 2002; Rastegar et al., 2004). We have shown that these modifications of the histone tail increases craniocaudally along the axis. Although the transcriptionally active posterior tissues of 9.5 dpc and older embryos are more heavily acetylated at H3-K9 than the anterior, non-Hox expressing tissues, some acetylation of H3-K9 at Hoxb8 is seen in anterior regions where Hoxb8 expression is repressed at early and later developmental stages. De-repression of Hoxb8 expression upon depletion of Rnf2 in MEFs derived from the cranial part of 9.5 dpc embryos suggests the involvement of Rnf2-containing class 2 PcG complexes to mediate this transcriptional repression. Therefore,our data suggest that the associations of Rnf2-containing PcG complexes and acetylated H3-K9 may counteract each other and cooperate to maintain the anterior boundaries of Hoxb8 expression at mid-gestational stages and later. This is consistent with the antagonistic properties of Mll and Bmi1 mutations (Hanson et al.,1999). Moreover, the establishment of the differential binding of the Rnf2 and H3-K9 acetylation at Hoxb8 during embryogenesis temporally coincides with de-repression of that Hox gene in Bmi1/Rnf110 and Phc1/Phc2 double homozygotes, and loss of its transcription in Mll homozygotes(Akasaka et al., 2001; Yu et al., 1998; Isono et al., 2005b). Intriguingly, class 2 PcG complexes, which lack the Rnf2 component, are also involved in the maintenance of H3-K9 acetylation in embryonic tissues where Hox genes are expressed. This is consistent with predominant subnuclear localization of several PcG proteins in the perichromatin compartment where most pre-mRNA synthesis takes place (Cmarco et al., 2003). The molecular mechanisms underlying this positive action remain unaddressed.

In conclusion, class 2 PcG gene products play distinct roles in embryonic territories, which are silent or active for Hoxb8 transcription, by forming complexes of different composition. Interaction between class 1 and class 2 PcG complexes mediated by trimethylated H3-K27 play decisive roles in Hox gene repression outside their expression domains, as seen in Drosophila. In addition, within the Hox expression domain, class 2 PcG complexes are involved in maintaining a transcriptionally active status,independent of H3-K27 trimethylation.