Juxtaposed Polycomb complexes co-regulate vertebral identity

Originally discovered in Drosophila, Polycomb (PcG) proteins are required for epigenetic repression of Hox gene transcription during metazoan development (Francis and Kingston,2001; Ringrose and Paro,2004). In mouse embryos, defects in PcG function cause derepression of Hox gene transcription in individual somites anterior to the wild-type Hox expression boundary (van der Lugt et al., 1994; Akasaka et al., 1996; Schumacher et al., 1996; van der Lugt et al., 1996; Coré et al.,1997; Takihara et al.,1997; Katoh-Fukui et al.,1998; del Mar Lorente et al.,2000; Suzuki et al.,2002; Voncken et al.,2003; Isono et al.,2005). This alters the identity of the affected anterior somites toward a more posterior fate and manifests as posterior homeotic transformation of the corresponding vertebrae.

At least two PcG complexes with diverse composition and function in chromatin remodeling have been identified in mammals(Otte and Kwaks, 2003). The Polycomb repressive complex 1 (PRC1) involves the paralogous PcG proteins BMI1/MEL18 (PCGF2 - Mouse Genome Informatics), M33 (CBX2 - Mouse Genome Informatics)/PC2 (PCSK2 - Mouse Genome Informatics)/PC3 (PCSK1 - Mouse Genome Informatics), RAE28 (PHC1 - Mouse Genome Informatics)/MPH2, and RING1A (RING1- Mouse Genome Informatics)/RING1B (RNF2 - Mouse Genome Informatics)(Alkema et al., 1997a; Alkema et al., 1997b; Gunster et al., 1997; Satijn et al., 1997; Schoorlemmer et al., 1997; Hashimoto et al., 1998; Hemenway et al., 1998; Satijn and Otte, 1999; Levine et al., 2002; Suzuki et al., 2002). Evidence for PRC1-mediated chromatin modification derived from ubiquitylation at lysine 119 of histone H2A (H2A-K119) (de Napoles et al., 2004; Wang, H. et al., 2004; Cao et al.,2005). A second PcG complex, PRC2, encompasses EED, the histone methyltransferase EZH2, the zinc finger protein SUZ12, the histone-binding proteins RBAP46/RBAP48, and the histone deacetylase HDAC1(Denisenko et al., 1998; Sewalt et al., 1998; van Lohuizen et al., 1998; van der Vlag and Otte, 1999; Cao et al., 2002; Kuzmichev et al., 2002; Kuzmichev et al., 2004). Several EED isoforms, generated by alternate translation start site usage of eed mRNA, differentially engage in PRC2-related complexes (PRC2/3/4),targeting the histone methyltransferase activity of EZH2 to H3-K27 or H1-K26(Kuzmichev et al., 2002; Kuzmichev et al., 2004; Kuzmichev et al., 2005). PcG complexes bind to cis-acting Polycomb response elements (PREs), which encompass several hundred base pairs and are necessary and sufficient for PcG-mediated repression of target genes(Pirrotta et al., 2003). Whereas the function of several PREs has been delineated in Drosophila, similar elements await characterization in mammals.

Studies in Drosophila revealed strong genetic interaction in many,but not all, pairwise combinations of PcG mutant alleles(Jürgens, 1985; Kennison and Tamkun, 1988; Adler et al., 1989; Adler et al., 1991; Paro and Hogness, 1991; Cheng et al., 1994; Campbell et al., 1995; Bajusz et al., 2001). For example, Psc and Pc engage in Drosophila PRC1(Shao et al., 1999), and intercrosses between mutant alleles significantly enhanced the homeotic phenotypes compared with the single mutants(Campbell et al., 1995). Likewise, synergistic interaction between mutant alleles of their murine homologs, Bmi1 and M33, was evident from ectopic Hox gene expression and posterior homeotic transformations across multiple adjacent somites and vertebrae, respectively (Bel et al., 1998).

Molecular support for cooperative interaction between PcG complexes derived from the transient association of the ESC/EZ/PHO complex and PRC1 in preblastoderm embryos (Poux et al.,2001). Furthermore, EZH2, in a complex with Esc, methylated histone H3-K27 in the vicinity of PREs, resulting in Pc binding to this epigenetic mark and repression of Hox genes(Cao et al., 2002; Czermin et al., 2002; Müller et al., 2002; Wang, L. et al., 2004). In conjunction with similar findings in mammalian cell lines and embryonic stem cells (Kuzmichev et al., 2002; Cao et al., 2005; Fujimura et al., 2006), these results supported a model of hierarchical recruitment of PRC1 to target genes upon binding to PRC2/3/4-methylated H3-K27. Although mammalian cell lines exhibited co-localization of BMI1 with components of PRC2/3/4 in a cell-cycle dependent manner (Hernandez-Munoz et al.,2005), transient interaction between the core PcG complexes has not been reported at the molecular level. Thus, the functional relationship between PRC1 and PRC2/3/4 in the regulation of vertebral identity remains elusive.

The present study interrogated the genetic and molecular interplay of BMI1 and EED, pivotal constituents of PRC1 and PRC2/3/4, respectively, in axial Hox gene repression in the mouse embryo. Bmi1 (B cell-specific Mo-MLV integration site 1) encodes a 324 amino acid ring finger protein, which interacts directly with RAE28, RING1A, RING1B and M33(Alkema et al., 1997a; Gunster et al., 1997; Satijn et al., 1997). In conjunction with RING1A, BMI1 is required for ubiquitylation of H2A-K119(de Napoles et al., 2004; Wang, H. et al., 2004; Cao et al., 2005). A loss-of-function allele of Bmi1 demonstrated ectopic Hox gene expression and highly penetrant, dosage-sensitive posterior homeotic transformations along the vertebral column(van der Lugt et al., 1994; van der Lugt et al., 1996). Bmi1-deficient mice died between birth and 20 weeks of age and displayed neurological and hematopoietic abnormalities(van der Lugt et al., 1994; Lessard et al., 1999).

eed (embryonic ectoderm development) encodes a 535 amino acid protein with five WD motifs and is likely to adopt a toroidal β-propeller structure (Schumacher et al.,1996; Denisenko and Bomsztyk,1997; Ng et al.,1997; Schumacher et al.,1998; Sewalt et al.,1998; Kuzmichev et al.,2004). Integrity of the WD motifs is essential for protein interaction, and two point mutations, represented by the l7Rn5^3345SB null allele and the l7Rn5^1989SB hypomorphic allele, disrupted EED interaction with HDACs and EZH2 (Denisenko et al.,1998; Sewalt et al.,1998; van Lohuizen et al.,1998; van der Vlag and Otte,1999). A pivotal role for EED in PRC2/3/4 function was evident from global H3-K27 methylation defects in eed mutant embryonic and trophoblast stem cells (Montgomery et al.,2005). Homozygosity for the l7Rn5^3345SB null allele caused embryonic lethality and anteroposterior (AP) patterning defects of the primitive streak at gastrulation(Faust et al., 1995; Faust et al., 1998). By contrast, animals homozygous for the l7Rn5^1989SBhypomorphic allele were viable and manifested highly penetrant,dosage-sensitive posterior homeotic transformations along the vertebral column, as well as ectopic Hox gene expression(Schumacher et al., 1996; Wang et al., 2002). Additional developmental functions of EED included the regulation of random and imprinted X chromosome inactivation(Wang et al., 2001; Mak et al., 2002; Plath et al., 2003; Silva et al., 2003; Kalantry and Magnuson, 2006; Kalantry et al., 2006) and genomic imprinting (Mager et al.,2003). Furthermore, eed mutant animals exhibited hematopoietic defects in the bone marrow and thymus(Lessard et al., 1999; Richie et al., 2002).

Surprisingly, despite a significant overlap in homeotic phenotypes and co-regulation of Hox genes, the present genetic analysis implicated eed and Bmi1 in parallel pathways, which converged at the level of Hox gene regulation. EED and BMI1 engaged in separate, but juxtaposed complexes at Hox target loci. While both complexes contain YY1 as a DNA-binding factor, EED, but not BMI1, associated with methylated H3-K27. The combined genetic, biochemical and molecular results form the basis for a model of PcG complex recruitment and retention in mammalian Hox gene clusters.

l7Rn5^3345SB and l7Rn5^1989SBrepresent eed null and hypomorphic alleles, respectively, generated by N-ethyl-N-nitrosourea mutagenesis(Schumacher et al., 1996; Rinchik and Carpenter, 1999). The T¹⁰⁴⁰→C transition in the l7Rn5^3345SBallele disrupted a diagnostic AluI site used for genotyping(Schumacher et al., 1996). PCR amplification with primers 5′-GTTGGCCATGGAAATGTTA (forward) and 5′-CTATGCATTCTCAGAACAC (reverse) created a diagnostic MseI site in the context of the T¹⁰³¹→A transversion in the l7Rn5^1989SB allele. The Bmi1 null mutation was generated by gene targeting and detected by PCR as reported(van der Lugt et al., 1994). All mice were on a partially congenic FVB/NJ background (N5). For the developmental studies, embryos were recovered at embryonic day (E) 8.5 or 12.5 from timed matings. Noon of the day of the appearance of the vaginal plug was considered E0.5.

Whole-mount preparations of P0 skeletons were stained with 0.015% Alcian Blue 8GX (Sigma) and 0.005% Alizarin Red S (Sigma) and cleared by alkaline digestion in potassium hydroxide for 7-10 days as described previously(van der Lugt et al.,1994).

Immunohistochemistry and streptavidin-biotin immunoperoxidase detection on parasagittal 5 μm sections from E12.5 embryos were performed as described recently (Mok et al., 2004). mRNA in situ hybridization to parasagittal 7 μm sections from E12.5 embryos was conducted according to a standard protocol with minor modifications(Neubüser et al., 1995). Depending on the expression level, the alkaline phosphatase reaction was extended up to 4 weeks with weekly changes of the substrate solution to enhance signal intensity.

Digoxigenin-11-UTP-labeled antisense cRNA probes were prepared as reported(Wilkinson and Nieto, 1993). Antisense cRNA probes were generated from plasmids described previously: eed (Faust et al.,1995), Hoxa5(Colberg-Poley et al., 1985), Hoxa7 (Yu et al.,1995), Hoxb3 (Sham et al., 1992), Hoxb4(Ramirez-Solis et al., 1993), Hoxb6 (Schughart et al.,1988) and Hoxd4(Gaunt et al., 1989). In addition, a 408 base pair (bp) Hoxc8 probe (corresponding to nucleotides 274-682 of GenBank Accession number NM010466) and a 741 bp Bmi1 probe (corresponding to nucleotides 1529-2270 of GenBank Accession number XM130022) were employed. Sense eed and Bmi1cRNA probes served as negative controls.

Whole-mount in situ hybridization was performed as described(Wilkinson and Nieto, 1993). The alkaline phosphatase detection reactions generally reached completion within 3 days. Images were captured from flat-mount preparations of embryos under coverslips.

E12.5 embryos were dissected and eviscerated and limbs and heads were removed. The remaining trunks were pooled (n=2-3) and sonicated in lysis buffer [20 mmol/l Tris pH 7.5, 150 mmol/l NaCl, 1 mmol/l EDTA, 1 mmol/l EGTA, 1% Triton X-100 with Complete Mini Protease Inhibitor (Roche)]. Two milligrams of protein lysate were incubated overnight with an α-Eed orα-Bmi1 antibody covalently coupled to beads [ProFound Mammalian Co-Immunoprecipitation Kit (Pierce)] or with an α-Bmi1 antibody preincubated with protein G/A agarose beads (Oncogene). As a control, lysates were incubated with free beads. Elutions were separated on SDS-PAGE gels and transferred onto PVDF membrane (BioRad) for western blot analysis. Membranes were blocked with 5% BSA in TBST (20 mmol/l Tris, pH 7.6, 137 mmol/l NaCl,0.1% Tween-20) or ReliaBlot reagent (Bethyl Laboratories) for 2 hours at room temperature and incubated overnight at 4°C with α-EED, α-BMI1,α-RING1B, α-EZH2, α-H3M2K27, or α-YY1 antibody. Following incubation with appropriate horseradish peroxidase-conjugated secondary antibodies, (co-)immunoprecipitated proteins were detected by chemiluminescence (ECL reagent, Santa Cruz Biotechnology).

Chromatin immunoprecipitation (ChIP) was performed with the ChIP Assay Kit(Upstate). Input consisted of eviscerated E12.5 embryos without limbs and heads. For spatial analysis of EED and BMI1 binding to Hox regulatory regions,trunks were dissected further and cut rostral to the anterior expression boundaries of Hoxa5 (pv3) or Hoxc8 (pv12). Pools of two to three entire trunks, or pools of five to ten anterior and posterior fragments were incubated in 1% formaldehyde for 10 minutes at 37°C. Following crosslinking, tissues were washed and sonicated to shear the DNA to lengths between 200 and 1000 bp. Sonicated supernatants were precleared with salmon sperm DNA/protein A (or protein G) agarose beads, and incubated overnight with approximately 5 μg antibody. The chromatin-antibody complex was collected with salmon sperm DNA/protein A (or protein G) agarose. Crosslinking was reversed with sodium chloride at 65°C overnight. Phenol:chloroform-extracted chromatin was employed for PCR amplification of Hox promoter sequences. Hoxa5: 5′-CGTAACCGTCAGTGGGAGA (forward)and 5′-GGAATCATTTCGTGGCTGTG (reverse), 5′-ACCCAACTCCCCCATTAGTG and 5′-TGTGCTTGATTTGTGGCTCG (reverse). Hoxc8:5′-TGAGCACAATAGCCCACACG (forward) and 5′-TGGGAGAACCTGAGAGCAAAG(reverse), 5′-GCCCTGACTTACTAACTCTCACCTC (forward) and 5′-GCTCTCTGCTCACTGTCGGTAG (reverse). The promoter region ofβ-actin served as a negative control using primers 5′-ACAAGGGCGGAGGCTATTC (forward) and 5′-AGGCAACTTTCGGAACGG(reverse).

The following antibodies were employed: α-EED (rabbit polyclonal,raised against EED residues 123-140, custom-generated by Bethyl Laboratories),α-BMI1 antibody (rabbit polyclonal, raised against BMI1 residues 229-314, custom-generated by Bethyl Laboratories; mouse monoclonal, Upstate;goat polyclonal, Abcam; goat polyclonal, Santa Cruz Biotechnology),α-RING1B (mouse monoclonal, kindly provided by Haruhiko Koseki; rabbit polyclonal, Abcam), α-EZH2 (rabbit polyclonal, Upstate; rabbit polyclonal, Abcam; rabbit polyclonal, Bethyl Laboratories), α-H3M2K27(rabbit polyclonal, Upstate), α-H3M3K27 (rabbit polyclonal, Upstate),α-YY1 (rabbit polyclonal, Abcam; rabbit polyclonal, Santa Cruz Biotechnology), and α-FPN1 (rabbit polyclonal, raised against residues 553-568 at the FPN1 carboxy terminus, custom-generated by Bethyl Laboratories;see also Mok et al. (Mok et al.,2006).

Previous reports, mostly based on northern analysis of adult tissues and cell lines, indicated widespread mRNA expression of eed and Bmi1 (Schumacher et al.,1996; Alkema et al.,1997a; Denisenko and Bomsztyk,1997; Gunster et al.,1997; Denisenko et al.,1998; Rietzler et al.,1998; Schumacher et al.,1998; Sewalt et al.,1998; Peytavi et al.,1999). mRNA in situ hybridization studies detected coexpression of eed and Bmi1 in somites and neuroectoderm in E8.5 wild-type embryos (Fig. 1A,E). Likewise,at E12.5, prevertebrae and neuroectoderm exhibited coexpression of eed and Bmi1 mRNA (Fig. 1B,F). Control experiments with eed and Bmi1sense probes revealed no signal on sectioned E12.5 embryos(Fig. 1C,G). Immunohistochemistry confirmed EED and BMI1 protein expression in the nuclei of E12.5 prevertebrae (Fig. 1D,H).

Crosses between eed heterozygotes, i.e. l7Rn5^3345SB/+ or l7Rn5^1989SB/+, with Bmi1heterozygotes, followed by inter se crosses of double heterozygotes, yielded the various double mutant genotypes for this study. Homozygosity for the l7Rn5^3345SB allele caused lethality at gastrulation(Faust et al., 1995), whereas l7Rn5^1989SB homozygotes were viable and fertile(Schumacher et al., 1996). Consequently, results from eed homozygotes at E12.5 and P0 encompassed the l7Rn5^1989SB allele. Twenty percent of the eed;Bmi1 double homozygotes survived to term, but the severely runted pups died invariably during the first 24 hours. All other genotypes segregated with the expected Mendelian ratio and resulted in viable offspring (data not shown).

l7Rn5^3345SB or l7Rn5^1989SBheterozygotes presented no significant difference in the penetrance of homeotic transformations and ectopic Hox gene expression in transwith Bmi1 heterozygosity or homozygosity (data not shown). In the absence of allele-specific effects, data for l7Rn5^3345SBand l7Rn5^1989SB were combined. These results also suggested haploinsufficiency of the l7Rn5^1989SB allele in the homeotic pathways, which contrasted with its dominant-negative function in carcinogen-induced T-cell lymphoma development(Richie et al., 2002).

Analysis of whole-mount skeletal preparations provided a powerful means for quantitative characterization of the functional relationship between eed and Bmi1 in axial patterning. By genetic criteria, the overlapping, dosage-sensitive transformations in eed and Bmi1 single mutants (van der Lugt et al., 1994; Schumacher et al., 1996) reflect a common, sensitized pathway(s) susceptible to changes in phenotypic penetrance and expressivity in the presence of a non-allelic modifier. Synergistic interaction between eed and Bmi1 would manifest as supra-additive increases in the penetrance of axial transformations, as well as expression of novel phenotypes, similar to Bmi1;M33 double mutants(Bel et al., 1998). Alternatively, in case of sole dependence of PRC1 recruitment on PRC2/3/4-mediated methylation of H3-K27(Cao et al., 2002; Czermin et al., 2002; Kuzmichev et al., 2002; Müller et al., 2002; Wang, L. et al., 2004; Cao et al., 2005), eed;Bmi1 double mutants would exhibit the eedsingle mutant phenotype.

Fig. 2A summarizes the results of 668 skeletal preparations comprising all nine genotypes obtained from eed;Bmi1 double heterozygous intercrosses. In agreement with previous studies (van der Lugt et al., 1994; Schumacher et al.,1996), mutant skeletons exhibited dosage-sensitive, posterior transformations along the entire AP axis, which ranged between 70 and 100% in eed and Bmi1 homozygotes(Fig. 2A). Strikingly, the penetrance of the skeletal transformations in eed;Bmi1double heterozygotes reflected a strictly additive effect of the single heterozygous phenotypes (Fig. 2A). For example, the L6→S1 transformation was detected in 25 and 27% of eed^+/-;+/+ and +/+;Bmi1^+/-skeletons, respectively, compared with 54% in double heterozygotes.

The penetrance of all homeotic transformations approached 100% in eed^-/-;Bmi1^-/- double homozygotes(Fig. 2A). Although this hindered the evaluation of potential supra-additive increases in penetrance,changes in phenotypic expressivity, another hallmark of synergistic genetic interaction, could readily be ascertained. Regardless of the severity of the double mutant genotype, the transformations remained restricted to individual axial segments and were phenotypically identical to those detected in eed and Bmi1 single mutants(Fig. 3C-G). This included the cervical region, which represented a focal point of M33×Bmi1 synergy and manifested transformations across multiple vertebrae (Bel et al.,1998). Furthermore, two ectopic ossification centers rostral to the first cervical vertebra and broadening of the neural arch of C1 in Bmi1 homozygotes represented known Bmi1-specific defects(van der Lugt et al., 1994)with phenotypic expression not enhanced by eed mutant alleles(Fig. 3C,D). Finally, unlike crosses between M33 and Bmi1, the axial and appendicular skeleton in eed;Bmi1 double mutants did not present any novel phenotypes compared with the single mutants and wild-type controls (data not shown).

Thus, by contrast to intercrosses of PRC1 mutant alleles, genetic interaction between constituents of PRC1 and PRC2/3/4 did not result in synergistic phenotypes. The double mutant phenotype also rendered sole dependence of PRC1 recruitment on PRC2/3/4-mediated methylation of H3-K27 unlikely. Rather, in the context of a sensitized PcG pathway(s), additive increases in the penetrance of homeotic transformations are most consistent with parallel function of EED and BMI1 in the regulation of vertebral identity.

To account for the haploinsufficiency of eed and Bmi1function, as well as to detect potential cross- and para-regulatory interferences of Hox gene transcription(Gould et al., 1997; Mann, 1997), 237 embryos encompassing all nine eed×Bmi1 genotypes yielded sections for 352 mRNA in situ hybridizations(Fig. 2B, Fig. 3A,B). In agreement with previous reports (Gaunt, 1988; Dressler and Gruss, 1989; Gaunt et al., 1989; Gaunt et al., 1990; Kessel and Gruss, 1991; Sham et al., 1992; Eid et al., 1993; Akasaka et al., 1996), the following anterior boundaries of highlevel Hox gene expression were detected in the prevertebrae (pv) of E12.5 embryos: pv3 (Hoxa5), pv10(Hoxa7), pv1 (Hoxb3), pv2 (Hoxb4), pv7(Hoxb6), pv12 (Hoxc8) and pv2 (Hoxd4)(Fig. 3A,B and data not shown). Likewise, discrete levels of Hoxa5, Hoxb4 and Hoxd4expression were frequently detected in the prevertebra directly rostral to the anterior boundary (Fig. 3A,Band data not shown).

Compared with wild-type littermates, E12.5 eed mutant embryos displayed ectopic expression of Hoxa5, Hoxb3, Hoxb6 and Hoxc8 in prevertebrae, whereas Hoxa7, Hoxb4 and Hoxd4 expression appeared unaffected(Fig. 2B, Fig. 3A,B and data not shown). Therefore, EED functions as a negative regulator of Hox gene transcription,corroborating a previous study in E11.5 embryos(Wang et al., 2002). To ascertain a potential role for EED in early Hox gene repression, the expression of Hoxa5 and Hoxb6 was examined by whole-mount in situ hybridization in E8.5 eed mutant embryos(Fig. 4). l7Rn5^{1989SB/1989SB} and wild-type embryos presented the same anterior Hoxa5 and Hoxb6 expression boundary at the level of somites 5 and 8, respectively. These results contrasted with ectopic Hoxa5 and Hoxb6 expression in l7Rn5^{1989SB/1989SB} embryos at E12.5(Fig. 2B). Thus, similarly to Bmi1 and other PcG genes, EED is required for maintenance but not initiation of Hox gene expression.

In agreement with previous studies(van der Lugt et al., 1996; Bel et al., 1998),+/+;Bmi1^-/- embryos manifested ectopic expression of Hoxa5, Hoxb6 and Hoxc8(Fig. 2B, Fig. 3A,B). This contrasted with downregulation of Hoxb6 as well as normal expression levels of Hoxa5 and Hoxc8 in Bmi1-deficient murine embryonic fibroblasts (MEFs) (Cao et al.,2005) and suggests changes in PcG-mediated repression in response to cellular differentiation. Four Hox probes were selected for the analysis in eed;Bmi1 double mutants based on the presence (Hoxa5,Hoxb6, Hoxc8) or absence (Hoxb4) of ectopic expression in the single mutants. As shown in Fig. 2B, the penetrance of ectopic Hoxa5, Hoxb6 and Hoxc8 expression increased with the severity of the genotype and reached 100% in eed;Bmi1 double homozygotes. By contrast,the phenotypic expressivity was unaffected and ectopic expression remained confined to a single prevertebra rostral to the wild-type anterior expression boundary (Fig. 2B, Fig. 3A and data not shown). Furthermore, ectopic Hoxb4 expression was not detected, regardless of the genotypic severity (Fig. 2B, Fig. 3B).

In conclusion, additive effects in genetically sensitized double heterozygotes, confinement of ectopic Hox gene expression and homeotic transformations to single segments, and absence of novel phenotypes strongly support the notion that eed and Bmi1 govern parallel pathways converging at the level of Hox gene regulation.

An antibody raised against residues 123-140 of the EED amino terminus precipitated three distinct isoforms of approximately 50 and 75 kDA from E12.5 trunk (Fig. 5), representing three of the four EED isoforms previously reported in 293 cells(Kuzmichev et al., 2004). In addition to EZH2 and YY1, dimethylated H3-K27 co-immunoprecipitated with EED(Fig. 5). Immunoprecipitation identified three BMI1 isoforms of approximately 39-41 kDA. BMI1 was found in a complex with RING1B, but not dimethylated H3-K27. Similar to the EED complex,the BMI1 complex also contained YY1 (Fig. 5). It should be emphasized that all (co-)immunoprecipitating bands were detected by at least two antibodies against different epitopes. Strikingly, while dimethylated H3-K27 engaged in the EED complex,trimethylated H3-K27 did not appear to associate with either the EED or the BMI1 complex. Importantly, reciprocal co-immunoprecipitation detected EED and BMI1 in separate protein complexes.

Ectopic expression in mutant embryos revealed Hoxc8 and Hoxa5 as downstream targets of EED and BMI1 function(Fig. 2B, Fig. 3A). ChIP detected EED and BMI1 binding immediately upstream of the Hoxc8 transcribed region near putative promoter elements (Fig. 6A,B). The binding sites could not be separated, indicating close proximity of the complexes (data not shown). EED and BMI1 binding also clustered within a small fragment 1.5 kb upstream of the Hoxc8transcription start site (Fig. 6A,B), suggesting long-range juxtaposition of heterologous PcG complexes. Similar to EED and BMI1, YY1 localized to both regions. In support of YY1 binding to Hox regulatory regions, inspection of the mouse genome sequence (NCBI Build 35) revealed clusters of putative YY1 binding sites in both regions a and b, including TGTCCATTAG and CCCCCATTCC (region a), as well as ACACCATGGC,TTTCCATTAG and TCCCCATAAA (region b). CCAT represents the core of the YY1 consensus binding site, while flanking sequences exhibited significant tolerance for multiple nucleotides(Shrivastava and Calame, 1994; Mihaly et al., 1998). EED,BMI1 and YY1 also co-localized approximately 1.5 kb upstream of the transcription start site of Hoxa5(Fig. 6B). In addition to PcG binding, ChIP detected trimethylated H3-K27 throughout the regulatory regions of Hoxc8 and Hoxa5 (Fig. 6B). Furthermore, dimethylated H3-K27 localized to region b of Hoxc8. The presence of the dimethylated histone domain in other regions could not be determined conclusively. ChIP using an antibody against ferroportin 1, a membrane-bound iron exporter(Mok et al., 2004), confirmed the specificity of the experiments and yielded no amplification of Hoxc8 and Hoxa5 genomic fragments(Fig. 6B). As an additional control, EED and BMI1 complexes did not localize to the β-actinpromoter. Conversely, ChIP detected YY1 at the β-actin promoter,which contained a putative YY1 binding site CACCCATCGC. This finding is consistent with a context-dependent role of YY1 as a transcriptional activator (Gordon et al.,2006).

Spatial regulation of EED and BMI1 binding to Hox regulatory regions was evident from ChIP analysis of dissected anterior and posterior regions of E12.5 trunk. In agreement with transcriptional silencing of Hoxc8 and Hoxa5, EED and BMI1 binding was detected upstream of these loci in anterior regions of the trunk (Fig. 6C). By contrast, EED and BMI1 binding was absent from posterior regions of the trunk, where Hoxc8 and Hoxa5 are transcribed. These findings implicate PcG complexes in Hox gene repression in anterior regions of the AP axis, consistent with a recent study(Fujimura et al., 2006).

The combined interpretation of the co-immunoprecipitation and ChiP results indicates that trimethylated H3-K27 did not form a complex with EED or BMI1,despite co-localization of the three proteins in Hox regulatory regions. By contrast, co-immunoprecipitation demonstrated physical association of the EED complex with dimethylated H3-K27. In aggregate, the results support a model in which EED- and BMI1-containing chromatin remodeling complexes exist as separate, but juxtaposed, biochemical entities at Hox target loci.

While recent years have witnessed significant progress in elucidating PcG complex interaction in Drosophila development and mammalian cell lines (Francis and Kingston,2001; Lund and van Lohuizen,2004; Ringrose and Paro,2004), little is known about their functional interdependency in Hox gene regulation in the mouse embryo. Consistent with a common temporal requirement for PRC1 and PRC2/3/4, the present study demonstrated EED function in maintenance but not initiation of Hox gene repression, similar to BMI1/MEL18 and RAE28 (Tomotsune et al.,2000; Akasaka et al.,2001). eed and Bmi1 single mutants revealed ectopic expression of a similar set of Hox genes and common posterior transformations. Notably, this represented the most significant overlap in molecular and vertebral phenotypes among the murine PcG genes analyzed thus far, including the BMI1 homolog MEL18(van der Lugt et al., 1994; Akasaka et al., 1996; Schumacher et al., 1996; van der Lugt et al., 1996; Coré et al., 1997; Takihara et al., 1997; Katoh-Fukui et al., 1998; del Mar Lorente et al., 2000; Suzuki et al., 2002; Voncken et al., 2003; Isono et al., 2005). The distinct similarities in EED and BMI1 phenotypes predicted strong genetic interaction in double mutant crosses. Indeed, mutant alleles of Drosophila Psc significantly enhanced the homeotic phenotypes in esc^+/- embryos devoid of maternal esc(Campbell et al., 1995).

Surprisingly, the present study revealed additive increases in the penetrance of homeotic transformations in eed;Bmi1 double heterozygotes compared with the single mutants. Furthermore, regardless of the severity of the eed;Bmi1 genotype, ectopic Hox gene expression and homeotic transformations remained confined to single segments,and novel phenotypes were not detected. Conceivably, compensatory activity by PcG paralogs could mask synergistic interactions in eed;Bmi1double mutants. Whereas the mouse genome appears devoid of an EED homolog,BMI1 and MEL18 display 70% protein sequence identity. However, BMI1 and MEL18 exerted only partially overlapping functions during mouse development and co-regulated a similar, but not identical, set of Hox genes(van der Lugt et al., 1994; Akasaka et al., 1996; Akasaka et al., 2001). In addition, BMI1, but not MEL18, promoted the E3 ligase activity of RING1B in vitro (Cao et al., 2005). Most importantly, the same Bmi1 allele employed in this study synergistically enhanced the deficiency for M33(Bel et al., 1998), rendering compensatory activity of MEL18 unlikely. Furthermore, the Bmi1;M33 double mutant crosses provide testimony to the inherent capacity of the mammalian PcG repressor system to express synergistic phenotypes upon mutational disruption.

The genetic interaction between eed and Bmi1 in the regulation of vertebral identity also differs from studies in Drosophila embryos and mammalian cell lines, which suggest hierarchical PcG complex recruitment (Cao et al., 2002; Czermin et al.,2002; Kuzmichev et al.,2002; Müller et al.,2002; Wang, L. et al.,2004; Cao et al.,2005). Genetically, this should render the phenotypes of eed;Bmi1 double mutants indistinguishable from the eed single mutants. Therefore, rather than a synergistic or strictly hierarchical interplay of the core PcG complexes in the regulation of vertebral identity, the genetic interaction between eed and Bmi1 mutant alleles is most consistent with parallel pathways converging at the level of Hox gene repression.

Toward elucidation of the molecular mechanisms governing PcG pathway convergence, co-immunoprecipitation detected EED and BMI1 in separate complexes with EZH2 and RING1B, respectively, consistent with formation of PRC2/3/4 and PRC1 at E12.5. Interestingly, the zinc finger transcription factor YY1 co-immunoprecipitated with both EED and BMI1, suggesting developmental co-existence of YY1 in heterologous PcG core complexes. Previous studies in Drosophila and Xenopus embryos, as well as mammalian cell lines, revealed a context-dependent proclivity of YY1 to associate with constituents of PRC1 or PRC2/3/4(García et al., 1999; Poux et al., 2001; Satijn et al., 2001; Levine et al., 2002; Mak et al., 2002; Atchison et al., 2003; Jin et al., 2003; Lorente et al., 2006). Direct support for mammalian YY1 as a PcG protein derived from axial homeotic transformations in Yy1 mutant mice and genetic interaction with RING1A (Lorente et al.,2006).

Clusters of putative YY1/Pho DNA-binding sites have been detected near mammalian Hox genes as well as in Drosophila PREs(Brown et al., 1998; Fritsch et al., 1999; Mihaly et al., 1998; Gilthorpe et al., 2002). Indeed, EED, BMI1 and YY1 associated with DNA fragments at the Hoxc8and Hoxa5 locus, which harbored several putative YY1-binding sites within fewer than 50 bp. This strongly supports the notion that YY1 bestows sequence-specific DNA binding on heterologous PcG complexes, and, by virtue of clustered YY1-binding sites, governs their juxtaposition in Hox regulatory regions in the mouse embryo. Interestingly, PLZF (promyelocytic leukaemia zinc finger) also binds to multiple cis-acting elements near Hox transcription units and interacts directly with BMI1 in vivo(Barna et al., 2002). Whether both PLZF and YY1 form a common complex with BMI1 or, alternatively, whether YY1 or PLZF interact with BMI1, depending on the Hox target locus, awaits investigation.

Juxtaposed EED and BMI1 complexes associated with DNA fragments immediately upstream of the transcribed regions of Hoxc8 and Hoxa5. These findings are consistent with binding of both PcG complexes to the Hoxc13 and Hoxb8 promoter regions in MEFs and E12.5 mouse embryos, respectively (Cao et al.,2005; Fujimura et al.,2006). Furthermore, components of the general transcription machinery interact directly with constituents of PRC1, supporting a role for PcG targeting of the core promoter(Breiling et al., 2001; Saurin et al., 2001). Detection of EED and BMI1 complexes 1-1.5 kb upstream of the Hoxc8and Hoxa5 transcribed regions could suggest looping of PcG complexes from the core promoter. Indeed, PLZF dimers or trimers form loops between several PLZF-binding sites in Hoxd11 regulatory regions(Barna et al., 2002). In support of a similar mechanism, all DNA fragments associated with EED and BMI1 at the Hoxc8 and Hoxa5 locus harbor putative YY1-binding sites. However, to the best of our knowledge, formation of YY1 multimers has not been reported. Alternatively, the present results could implicate long-range PcG complex formation and histone modifications in Hox regulatory regions. Accordingly, multiple YY1-binding sites would anchor arrays of independent PcG complexes. In support of this notion, large continuous stretches of histone modifications across the Hox clusters contrast with punctate chromatin domains at reference loci(Bernstein et al., 2005). As an example for the unusual chromatin organization of the mammalian Hox clusters, a transcriptionally active chromatin domain governed by dimethylated H3-K4 spanned nearly 60 kb in length from Hoxa1 to the Hoxa7locus. Conceivably, long-range assembly of PcG complexes might form equally large repressed chromatin domains across transcriptionally silent Hox genes in mouse embryos.

Consistent with the presence of stable heterochromatin domains, ChIP detected both di- and trimethylated H3-K27 in the vicinity of EED and BMI1 complex binding in Hox regulatory regions. Strikingly, while trimethylated H3-K27 did not appear to stably associate with either PcG complex,dimethylated H3-K27 co-immunoprecipitated with EED, but not BMI1, from E12.5 trunk. Thus, while methylated H3-K27 plays a pivotal role in PRC1 recruitment(Cao et al., 2002; Czermin et al., 2002; Kuzmichev et al., 2002; Müller et al., 2002; Wang, L. et al., 2004; Cao et al., 2005), it does not permanently associate with this complex at later stages of Hox gene silencing. Beyond co-localization at potential target loci in mammalian embryonic stem cells and embryonic fibroblasts, as well as in Drosophila Kc and S2 cells (Boyer et al., 2006; Bracken et al., 2006; Lee et al., 2006; Schwartz et al., 2006; Tolhuis et al., 2006), the present study demonstrates direct physical interaction of PRC2/3/4 with dimethylated H3-K27 in differentiating somites. Therefore, at least in the context of Hox regulatory regions, the close proximity of PcG complexes and trimethylated H3-K27 does not equate to a stable physical association between PcG complexes and this modified histone site. Hence, defining the physical relationship between PcG complexes and histone sites at the recently identified target loci necessitates the complementation of the ChIP results by co-immunoprecipitation experiments in mouse embryos(Boyer et al., 2006; Bracken et al., 2006; Lee et al., 2006; Schwartz et al., 2006; Tolhuis et al., 2006).

The aggregate findings are depicted in a dynamic model of PcG complex assembly in mouse Hox gene clusters (Fig. 7). During the early maintenance phase of axial Hox gene repression, presumably between E9 and 10(Yu et al., 1995; Yu et al., 1998; Deschamps et al., 1999; Tomotsune et al., 2000),PRC2/3/4 binds to Hox target loci via YY1 and methylates H3-K27. In turn, the H3-K27 mark serves as a binding site for PRC1 to Hox target loci. However,additive homeotic phenotypes in eed;Bmi1 double mutants render sole dependence of PRC1 recruitment on methylated H3-K27 unlikely. Rather, YY1 binding sites, and as recently shown noncoding RNA molecules(Bernstein et al., 2006; Sanchez-Elsner et al., 2006),represent strong candidates for cooperative interaction with the H3-K27 mark in recruitment of PRC1 to target loci. Upon dissociation from methylated H3-K27, possibly governed by conformational changes, PRC1 retention at Hox regulatory regions would predominantly depend on YY1-mediated DNA binding. Juxtaposed PRC2/3/4, bound to DNA via YY1, stably re-associates with dimethylated H3-K27 throughout the maintenance phase of Hox gene repression and somite differentiation. In support of this notion, two recent studies demonstrated that PRC2/3/4-mediated methylation of H3-K27 prevents reactivation of silenced target loci during cellular differentiation(Kalantry and Magnuson, 2006; Kalantry et al., 2006). Thus,the model implicates a cooperative interplay between epigenetic and genetic elements in the recruitment and retention of juxtaposed PcG complexes in mammalian Hox gene clusters, delineating a molecular definition of PcG pathway convergence in the regulation of vertebral identity.

The authors are grateful to Maarten van Lohuizen for generously providing the Bmi1 mutant mice, as well as critical reading of the manuscript. Jacqueline Deschamps, Denis Duboule, Peter Gruss, Stanley Korsmeyer and Robb Krumlauf kindly provided the plasmids for the Hox probes. We thank Haruhiko Koseki for the RING1B antibody, Thomas Guenther for help with mRNA in situ hybridization, and Gerard Karsenty for access to microscopes and imaging equipment. S.Y.K. received partial support from a NIGMS predoctoral training grant. This work was supported by grants from the NIH, the Merck Genome Research Institute and the Curtis Hankamer Basic Research Fund to A.S.