Involvement of the Polycomb-group gene Ring1B in the specification of the anterior-posterior axis in mice

The Polycomb group (PcG) of genes was first identified by their requirement for the maintenance of the stable repression of Hox genes during the development of Drosophila melanogaster (Paro, 1995; Pirotta, 1997). Biochemical and immunohistochemical analyses indicate that Drosophila PcG gene products function as large multimeric protein complexes that are thought to act by changing the local chromatin structure (Paro, 1995; Pirotta, 1997). Synergistic genetic interactions between mutant alleles of different Drosophila PcG genes indicate their ability to affect the expression of Hox genes in a gene dose-dependent manner and are in line with the action of PcG gene products in multimeric protein complexes (Jurgens, 1985; Franke et al., 1992; Shao et al., 1999).

Mammalian genes that are structurally and functionally related to Drosophila PcG genes have been identified. The PcG gene products form parts of two different complexes. One complex contains the product of Eed (the ortholog of the Drosophila extra sex combs gene), and the products of Enx1 (Ezh2 – Mouse Genome Informatics) and Enx2 (the orthologs of the Drosophila enhancer of zeste gene). Eed interacts with histone deacetylases (Schumacher et al., 1996; Laible et al., 1997; van Lohuizen et al., 1998; Sewalt et al., 1998; van der Vlag and Otte, 1999). The second complex include the products of the paralogs of a subset of PcG genes. These genes exist as gene groups like Mel18 (Zfp144 – Mouse Genome Informatics) and Bmi1; M33 (Cbx2 – Mouse Genome Informatics), Mpc2 (Cbx4 – Mouse Genome Informatics) and Pc3 (Cbx8 – Mouse Genome Informatics); Rae28 (Edr1 – Mouse Genome Informatics)/Mph1 and Mph2 (Edr2 – Mouse Genome Informatics); and Ring1A (Ring1 – Mouse Genome Informatics) and Ring1B (Rnf2 – Mouse Genome Informatics) (Tagawa et al., 1990; van Lohuizen et al., 1991; Pearce et al., 1992; Satijn et al., 1997a; Hemenway et al., 2000; Nomura et al., 1994; Gunster et al., 1997; Schoorlemmer et al., 1997; Satijn et al., 1997b; Hemenway et al., 1998). This complex is similar to the Polycomb repressive complex 1 (PRC1) recently identified in Drosophila (Shao et al., 1999; Gebuhr et al., 2000). Indeed, mice bearing mutations in Mel18, Bmi1, M33, Mph1 or Ring1A show homeotic transformations of the axial skeleton and alterations in the expression patterns of Hox genes (Akasaka et al., 1996; van der Lugt et al., 1994; Core et al., 1997; Takihara et al., 1997; del Mar Lorente et al., 2000). Interestingly, Ring1A mutants exhibit anterior transformations, while the others show posterior transformations (del Mar Lorente et al., 2000). In addition, mice with compound mutations in these genes show synergistic enhancement of the phenotypes associated with the individual genes evidencing genetic interactions between these mutations (Bel et al., 1998; Akasaka et al., 2001). This complex is required for the maintenance of the stable repression of Hox genes, as indicated by progressive derepressions of Hoxb3 expression in migrating facial neural crest cells in Mel18/Bmi1 compound and Rae28/Mph1 mutant embryos (Tomotsune et al., 2000; Akasaka et al., 2001).

To gain further insight in mammalian PcG function, we searched for Mel18 interactors and found Ring1B. The Ring1B protein is closely related to the product of PcG gene Ring1A, and was first identified by its interaction with the product of M33 (Schoorlemmer et al., 1997). Ring1B was also found to interact with Bmi1 and Mph2 in yeasts (Hemenway et al., 1998). Here we provide evidence for the presence of Ring1B in a complex containing M33, Mel18 and Rae28/Mph1 in extracts prepared from 11.5 days post coitus (dpc) mouse embryo, which were found associated to chromosomal DNA. These observations strongly suggest that Ring1B is a component of the mammalian counterpart of Drosophila PRC1 complex, as Drosophila Ring protein is included in PRC1 (Saurin et al., 2001). To test this possibility functionally, we attempted to generate a loss-of-function mouse model for Ring1B. Here, we describe a mouse line bearing a hypomorphic mutation for Ring1B and show that this mutation is accompanied by limited homeotic transformations of the axial skeleton and alterations in Hox gene expression. In addition, we show a genetic interaction between the Ring1B and Mel18 mutations. We also performed a gain-of-function experiment by in ovo electroporation. Overexpression of Ring1B protein in the developing neural tube results in the repression of chick Hoxb9 expression. These results demonstrate that Ring1B plays a role in the anteroposterior (AP) specification of the paraxial mesoderm and neural tube in vertebrate embryos.

A Mel18 full-length cDNA was fused in-frame to the lexA-coding sequence in vector pBTM116 (Fields and Bartel, 2001). The L40 yeast strain containing lexA-HIS3 and lexA-lacZ reporter genes was first transformed with pBTM116-Mel18, and then with the mouse 11.0 dpc embryo Matchmaker cDNA library in pGAD10 (Gal4 activation domain fusion vector, Clontech), using the lithium acetate method. Six hundred independent colonies were isolated from 7×10⁶ transformants on selective media plate lacking leucine, tryptophan and histidine and supplemented with 5 mM 3-aminotriazole. All these clones were restreaked onto the selection media and assayed for β-galactosidase activity by a filter assay (Vojtek et al., 1993). cDNA fragments were amplified by direct colony PCR from the His+lacZ+ positive clones, identified by sequencing and searched for using NCBI similarity search programs (BLASTn). Twenty clones, which were possible interactors of Mel18, were cured of the LexA fusion plasmid by growth on synthetic medium lacking leucine. Plasmid loss was confirmed by plating on both Leu-, Trp+ and Leu-, trp- plates. The selected clones grew only on trp+ media. The plasmid DNA was then prepared from those positive clones of interest and transformed into E. coli DH5α.

To test for interaction specificity, each pGAD10 variant containing Mel18, Bmi1, Ring1A, Ring1B, Mph2 or lamin was transformed into L40 (MATa), and each pBTM116 variant containing Mel18, Bmi1, Ring1A, Ring1B or Mph2 was transformed into AMR70 (MATα). Then the transformed clones were mated to each other in all possible combinations on YPAD plates. The Leu+Trp+ diploids were selected on Leu-Trp- plate and assayed for β-galactosidase activity using a filter assay.

The full-length and deleted Mel18 and Ring1B cDNAs were subcloned in the pcDNA3 vector (Invitrogen Japan, Tokyo, Japan). RNA was synthesized with 500 ng of supercoiled plasmids and translated in the presence of 10 μCi of [³⁵S] methionine (Amersham Biosciences, Tokyo, Japan) using the TNT T7 Quick Coupled Reticulocyte Lysate System (Promega, Tokyo, Japan). Ten micrograms of plasmids containing GST fused to either intact or truncated Mel18 or Ring1B cDNAs or GST alone (as a negative control) were pre-bound to 10 μl of a 1:1 suspension of Glutathion Sepharose 4 Fast Flow (Amersham Biosciences, Tokyo, Japan) and then incubated on a rotating wheel with the in vitro translation mixture in 500 μl of NETN buffer (0.5% NP-40, 20 mM Tris-Cl pH 8.0, 11 mM NaCl, 1 mM EDTA and 100 mM PMSF) for 1 hour at 4°C Bound proteins were eluted by boiling in SDS sample buffer and were analyzed on 11% polyacrylamide gels. Gel images were obtained by BAS-2000II Bio-Imaging Analyzer (FUJIFILM, Tokyo, Japan).

The following antibodies were used in this study:

mouse anti-mouse Ring1B monoclonal antibody (clone 3) (Atsuta et al., 2001);

rabbit anti-mouse Ring1B antiserum (Garcia et al., 1999);

rabbit anti-mouse Ring1A antiserum (Schoorlemmer et al., 1997);

mouse anti-mouse Rae28/Mph1 monoclonal antibody (K. I., H. M. and H. K., unpublished);

rabbit anti-mouse Rae28/Mph1 antiserum (Y. T., unpublished);

goat anti-mouse Mel18 purchased from Santa Cruz Biotechnology (Santa Cruz Biotechnology, Santa Cruz, CA);

mouse anti-mouse Bmi1 monoclonal antibody (Alkema et al., 1997);

rabbit anti-mouse M33 antiserum (Schoorlemmer et al., 1997);

rabbit anti-demethyl-HistoneH3 antiserum (Upstate Biotechnology, Lake Placid, NY) (Nakayama et al., 2001);

mouse anti-Bip monoclonal antibody (StressGen Biotechnologies, Victoria, B.C., Canada);

goat anti-Lamin B antibody (Santa Cruz Biotechnology, Santa Cruz, CA);

anti-HA epitope monoclonal antibody (Santa Cruz Biotechnology, Santa Cruz, CA);

rabbit anti-CENP-B antibody (BN1) (Kitagawa et al., 1995); and

goat anti-centromeric antigen (ACA) antiserum (Yoda et al., 1996).

Total cellular extracts were prepared from 11.5 dpc mouse embryos. A single embryo was solublized by sonication in 1 ml of TE (10 mM Tris-HCl, pH 8.0, 1 mM EDTA) using a Handy Sonic model UR-20P (TOMY seiko, Tokyo, Japan) followed by centrifugation at 9100 g for 5 minutes at 4°C to remove insoluble material. For immunoprecipitation, lysates were incubated with 1/2 volume of hybridoma culture supernatants containing either anti-Ring1B or anti-Rae28/Mph1 monoclonal antibodies, or with 1/30 volume of anti-Mel18 antiserum for 1 hour at 4°C with rocking. Immunocomplexes were captured with Protein A Sepharose beads (Amersham). Western blotting was performed as described previously (Atsuta et al., 2001).

U2-OS human osteosarcoma cells were used for immunofluorescence staining analyses as described previously (Atsuta et al., 2001). For the analysis of mitotic cells, cells were fixed with 70% acetone at –20°C for 30 minutes and air dried. Fixed cells were incubated with blocking solution for 1 hour at room temperature, and incubated with the mixture of rabbit anti-CENP-B antibody or ACA serum (1:200 dilution) and anti-Ring1B antibody for 2 hours at 37°C. Cells were washed with phosphate-buffered saline containing 0.1% Tween 20 (PBST) for 5 minutes at room temperature four times, and then incubated with fluorescein isothiocyanate-conjugated anti-rabbit IgG (1:160 dilution) (MBL, Nagoya, Japan) or fluorescein isothiocyanate-conjugated anti-human IgG (1:160 dilution) (MBL, Nagoya, Japan) and rhodamine-conjugated anti-mouse IgG (1:200 dilution) (MBL, Nagoya, Japan) diluted with PBST supplemented with 10% fetal bovine serum at 37°C for 1 hour. Cells on coverslips were then washed four times with PBST for 5 minutes at room temperature. Chromosomal DNA was stained with DAPI (1 μg/ml). Images were obtained with a fluorescence microscope (Carl Zeiss Japan, Tokyo, Japan) equipped with a cooled-CCD camera (PXL, Photometrics, CA) and IPLab software (Signal Analytics).

Mouse embryos at 12.5 dpc were minced with scissors and subjected to chemical crosslinking with 1% formaldehyde in phosphate-buffered saline (PBS) for 30 minutes. Isolation of chromatin fraction was performed as described elsewhere (Orlando et al., 1997).

A modified Ring1B allele was generated by gene targeting in R1 embryonic stem (ES) cells. Genomic Ring1B sequences were isolated from a mouse 129/SVJ genomic library. The targeting vector contains as homology arms, an EcoRI-SacI fragment containing two exons encoding the RING finger of Ring1B and a flanking EcoRI-HindIII fragment. MC1Neo and HSVtk genes were used as selectable cassettes for positive and negative selections, respectively (Fig. 6A). Thirty micrograms of the linearized plasmid were electroporated into 2×10⁷ R1 ES cells. Two hundred and forty double resistant R1 clones were analyzed by Southern blot. One of the clones showed an alteration in the Ring1B locus and was used to generate germline chimeras as described (Nagy et al., 1993).

We generated homozygous Ring1B mutant mice by interbreeding heterozygous animals that had been backcrossed eight times into C57BL/6J. Mel18-deficient mice (Akasaka et al., 1996) from both C57BL/6J genetic background were also used in this study. Ring1B and Mel18 heterozygous mice in C57BL/6J genetic background were interbred to generate compound homozygous mutant mice. Skeletal phenotypes of Ring1B mutants were also investigated on the compound genetic background between 129/SvJ derived from R1 ES cells and C57BL/6J. Ring1B and Mel18 mutations were backcrossed onto C57BL/6J twice (N2). N2 heterozygotes were intercrossed to generate homozygous mutants. This hybrid line was maintained by brother-sister mating for at least ten generations.

Ring1B mutant mice were genotyped by PCR using the following oligonucleotides (Fig. 5A): RB22, 5′-GTTGAAGACTTCATTGTACC-3′ and RB18, 5′-GGAAGTCACATCTTAGCAGG-3′ for the Ring1B wild-type allele; and Neo2, 5′-CTTCCATCCGAGTACGTGCT-3′ and RB11, 5′-TTATCTTCTGCTCCACTACC-3′ for the mutant allele. Oligonucleotides used for genotyping Mel18 mutant mice have been described previously (Akasaka et al., 2001).

Skeletal analysis of 18.5 dpc or new born mice and in situ hybridization to 11.5 dpc embryo sections were performed as described previously (Kessel and Gruss, 1991). Hoxb4 from Dr R. Krumlauf, Hoxb6 from Dr K. Schughart and Hoxb8 from Dr J. Deschampus were used (Akasaka et al., 1996). Whole-mount in situ hybridization was performed as described previously with slight modifications (Wilkinson and Nieto, 1993). Briefly, maleic acid buffer (0.5 M maleic acid, 0.15 M NaCl, pH 7.5) was used for washes after antibody reaction. Chicken Hoxb9 provided by Dr A. Kuroiwa, and mouse Mel18 and Ring1B probes were used.

cDNA fragments encoding the entire ORF of mouse Mel18 and Ring1B were amplified by PCR using primers that provided XhoI sites on both ends of the fragments for subcloning into the XhoI site of pCXN2 (kindly provided by Dr Jun-ichi Miyazaki, Osaka University, Japan) (Niwa et al., 1991). Empty pCXN2 vector was used as a negative control. In ovo electroporation was performed as described previously (Sakamoto et al., 1998; Nakamura and Funahashi, 2001). Expression vectors were dissolved at 2 μg/μl in PBS and mixed with 1/10 volume of 2 μg/μl β-galactosidase expression vector, pCHO110 (Amersham Biosciences, Tokyo, Japan), or green fluorescence protein (GFP) expression vector, pcDNA-GFP, and 1/20 volume of saturated Nile Blue solution. Fertilized chick eggs were purchased from Shiroyama chicken farm (Kanagawa, Japan) and cultured at 37.2°C before and after electroporation. The stages of chicken embryos were determined by counting somite number. After electroporation, embryos were re-incubated for varying length of time to allow expression of the introduced foreign DNA into one side of the neural tube and subsequently dissected and fixed in 4% paraformaldehyde in PBS.

Among 63 Mel18-binding proteins identified in the screening of cDNAs from an 11.0 dpc mouse embryo library, 22, 8 and two cDNA clones encoded parts of Ring1B, Mph2 and Rae28/Mph1 proteins, respectively (T. A., unpublished). In order to asses the specificity of the interactions, we performed yeast two hybrid mating assays (Fig. 1A). Bmi1, which is known to interact with Ring1B and Mph2, was used as positive control and Lamin as a negative control (Hemenway et al., 1998). Because Ring1A, a structurally related protein to Ring1B, was never identified in our screening of Mel18 interactors, we also examined the interaction of Mel18 and Ring1A. As visualized by β-galactosidase activity, Mel18 interacted strongly with Ring1B and Mph2 but hardly with itself, Bmi1 or Ring1A in our experimental conditions.

To determine whether the Mel18 protein can directly interact with Ring1B, we used an in vitro GST pull-down assay (Fig. 1B). Radiolabeled full-length or truncated Mel18 and Ring1B proteins were synthesized in vitro and tested for binding activity to respective GST fusion proteins with full-length Ring1B (GST-Ring1B), a C-terminal-truncated Ring1B (GST-Ring1BΔC) (amino acids 1-188), N-terminal-truncated Ring1B (GST-Ring1BΔN) (amino acids 189-365) and Mel18 lacking PEST region (GST-Mel18ΔPEST) (amino acids 1-233). Radiolabeled full-length Mel18 (Mel18fl) was able to bind to GST-Ring1B and GST-Ring1BΔC, and weakly to GST-Ring1BΔN but exhibited no appreciable binding to GST-Sepharose alone (Fig. 1B, part a). Because we could not obtain a full-length Mel18 protein fused to GST, we instead prepared a GST-Mel18ΔPEST protein. Full-length Ring1B (Ring1Bfl) bound to GST-Mel18ΔPEST but not to GST-Sepharose (Fig. 1B, part b). These observations indicated potential direct interaction of Mel18 and Ring1B. Ring1BΔC fragment was capable to bind to GST-Mel18ΔPEST more efficiently than Ring1BΔN fragment (Fig. 1B, part).

To obtain evidence for the in vivo association of Ring1B with mammalian PcG proteins in mammalian cells, we performed immunoprecipitation analysis using extracts from 11.5 dpc mouse embryos. We used anti-Mel18, anti-Ring1B or anti-Mph1 antibodies and preimmune rabbit serum as a negative control. A band with a mobility corresponding to 40 kDa was recognized with anti-Ring1B antibodies in the material immunoprecipitated by anti-Mel18, anti-Ring1B and anti-Rae28/Mph1 antibodies (Fig. 2A). This shows that protein complexes isolated from murine embryos containing Mel18 or Rae28/Mph1 also contain Ring1B.

Additional evidence for the in vivo association of Ring1B with Mel18 and other PcG proteins was obtained from their subcellular localization in U2-OS cells by indirect immunofluorescence (Fig. 2B-F). Endogenous Ring1B colocalized extensively with Mel18, Bmi1, Ring1A and Rae28/Mph1 in subnuclear speckles, previously termed PcG bodies, in interphase nuclei of U2-OS cells (Fig. 2B-E) (Saurin et al., 1998). Because a specific antibody against Mph2 was not available, we used U2-OS cells stably expressing Myc-tagged Mph2. In these cells endogenous Ring1B colocalized with Myc-tagged Mph2 (Fig. 2F). Taken together, these observations indicate that the PcG complexes in PcG bodies of interphase cells contain Ring1B.

Evidence for the association of PcG complexes to chromosomal DNA is available only for D. melanogaster (Zink and Paro, 1989; Orlando and Paro, 1993). To see whether mouse PcG complexes are also associated with chromosomal DNA, we analyzed formaldehyde crosslinked embryonic tissues after fractionation by cesium chloride isopyknic centrifugation for the presence of PcG proteins. Coomassie Brilliant Blue G-250 (CBB) staining showed most of the protein in fractions 1 to 3 but also in fractions 4 to 8 (Fig. 3B), whereas most chromosomal DNA was in fractions 5 to 7 (Fig. 3A). As a control, we studied the distribution of Bip, a luminal protein in the endoplasmic reticulum (ER) and nuclear envelope, which is thought not to interact with chromosomal DNA (Gething, 1999). The results showed that Bip was found in fractions 1 to 4, but it was absent from fractions 5 to 7 (Fig. 3C). A similar observation was obtained for Grp78, another ER luminal protein (Y. F., unpublished) (Gething, 1999). From the reciprocal distribution of chromosomal DNA and ER luminal proteins, we conclude that fractions 5 to 7 contain most of the chromatin-associated proteins. Distribution of dimethyl-Histone H3 to fractions 5 to 7 but not to fractions 1 and 2 revealed its similar distribution to chromosomal DNA (Fig. 3D) (Nakayama et al., 2001). This suggests that proteins that are not associated with chromatin are mainly distributed into fractions 1 and 2. Western blot analysis of these fractions showed that Ring1B and M33 were present in fractions 1 to 6 and Rae28/Mph1 in fractions 1 to 7 (Fig. 3D-F). Therefore, specific amounts of Ring1B, M33 and Rae28/Mph1 are associated with chromosomal DNA. Indeed, we have found that a significant amount of chromosomal DNA from fraction 6 could be co-immunoprecipitated by antibodies against Ring1B and Rae28/Mph1 (Y. F., unpublished). Our results also indicate that significant amount of Ring1B, M33 and Rae28/Mph1 are not closely associated with chromatin.

It is known that in tissue culture cells RING1 and BMI1-containing PcG bodies associate with pericentromeric heterochromatin in a cell cycle-dependent manner (Saurin et al., 1998). Therefore, we asked what the localization of Ring1B would be during the various phases of the cell cycle. In the interphase nuclei of U2-OS cells, Ring1B showed a speckled distribution associated to centromeric regions, just as it was found for Ring1, Bmi1 and M33 (Fig. 4A,B) (Saurin et al., 1998; Voncken et al., 1999; Wang et al., 1997). At mitosis, the chromosomal signal of Ring1B was very much reduced. However, upon chromosome condensation at prophase, Ring1B proteins were still associated with condensed chromosomes (Fig. 4C). At prometaphase, following disassembly of the nuclear envelop and nuclear matrix, Ring1B staining was significantly weaker to that in interphase nuclei and dispersed into the cytoplasm, although a weak signal was still seen associated with part of pericentromeric chromatin regions (broken line in Fig. 4D). At anaphase, a clearer Ring1B signal was found associated to chromosomal structures, but most of the signal was still found evenly distributed in a speckle-free cytoplasm (Fig. 4E). Thus, the localization of Ring1B at pericentromeric heterochromatin is cell cycle regulated in a similar manner to that of Ring1 or Bmi1 (Saurin et al., 1998; Voncken et al., 1999).

To start analyzing functionally Ring1B, we tried to inactivate the Ring1B locus in mice. In a gene targeting experiment, we obtained one clone with a modified Ring1B locus. Southern blot and sequence analysis of this locus showed that the selectable Neo cassette, together with some 0.8 kb sequences in one of the homology arms present in the targeting construct, were unexpectedly inserted between the exons labeled 3 and 4 (Fig. 5A-C). Because no other alterations were observed, there was the possibility that this modified Ring1B locus would synthesize an intact protein. We analyzed Ring1B gene expression in mice homozygous for the mutation and found that a cDNA encoding the full Ring1B ORF could be identified by RT-PCR (data not shown). However, the levels of Ring1B protein in these mice were significantly reduced compared with those in wild-type animals (Fig. 5D). Quantification of Ring1B signals in a Western blot of both homozygous and heterozygous mutant embryos, using the signals of LaminB to normalize for loading differences, showed that the levels of Ring1B in homozygous Ring1B mutant embryos were about 30% of those in heterozygous embryos. The reduced levels of Ring1B in the mutant mice, therefore, make it likely that this Ring1B allele is a hypomorphic one and is therefore designated as Ring1B^reduced (Ring1B^red hereafter). Ring1B^red homozygotes were healthy and fertile.

Perturbation of PcG function in mice characteristically results in homeotic transformations and other alterations of the axial skeleton (Akasaka et al., 1996; van der Lugt et al., 1994; Core et al., 1997; Takihara et al., 1997; del Mar Lorente et al., 2000). Therefore, if Ring1B participates in such a function, it is possible that similar defects could be seen in Ring1B^red mice. We found reproducible alterations in the cervicothoracic boundary of the axial skeleton of homozygous Ring1B^red mice, but not wild type. Thus, in 61% of the mice, we found ectopic ribs associated with the 7th cervical (C7) vertebra (Fig. 6A, parts b and f, Fig. 6B). At a lower penetrance, 40% of mice, we found that the prominent spinous process characteristic of the 2nd thoracic vertebrae (T2) was on the T1. These alterations were seen even in heterozygous Ring1B^red mice, although at low frequency (Fig. 6B). Formation of an additional ossification center in the cranial region of the sternum indicated anterior shift of the sternum in Ring1B^red homozygotes (Fig. 6A, part f). Holes in the ossification centers in the upper region of the xyphoid process were observed (Fig. 6A, part f). This is also seen in Hoxc4 mutants (Saegusa et al., 1996). Thus, these alterations in the axial skeleton and sternum seen in Ring1B^red mice can be interpreted as posteriorizations of the vertebral identities: C7→T1 and T1→T2.

As the anterior boundaries of Hoxb4, Hoxb6 and Hoxb8 expression in the paraxial mesoderm demarcate the prospective vertebrae C2, C7 and T1, respectively, we asked whether their expression patterns would be altered in Ring1B^red mice. The analysis of 11.5 dpc embryos homozygous for the Ring1B^red allele showed weak but reproducible signals for Hoxb4 and Hoxb8 in the C1 and C7 prevertebrae, respectively, whereas no signals were detected in the wild-type littermates (Fig. 7). Ectopic expression of Hoxb6 was weakly seen in the prospective neural arch of C6 prevertebra (Fig. 7). The expression of Hoxb4, Hoxb6 and Hoxb8 in the neural tube was not significantly altered.

Thus, the homeotic transformations of the axial skeleton and the anteriorization of the rostral boundaries of Hox gene expression in the paraxial mesoderm seen in this hypomorphic Ring1B mouse line suggest that Ring1B plays a role in PcG function in mice.

To evaluate the involvement of Ring1B in PcG function further, we investigated a possible genetic interaction with Mel18 by analyzing the axial skeleton of the offspring of crosses of Ring1B^red/+Mel18^+/– mice. Mel18^–/– mice with C57BL/6J genetic background now show a more homogeneous phenotype than when in a hybrid genetic background that we described previously and do not survive birth (Akasaka et al., 1996). The formation of ectopic arches in the occipital bone caused by the ectopic segmentation of the occipital bone, which was not observed in the hybrid genetic background, was now seen in every mouse (Fig. 6A, part c). The C1 vertebra was thinner than that of mice of a mixed genetic background and anterior shifts around the thoracolumbar, lumbosacral and sacrocaudal transition regions were uniformly seen in all Mel18^–/– mice. Interestingly, ectopic ribs on the seventh cervical vertebrae were less prominent and penetrant (Fig. 6B).

Animals double homozygous for mutations in the Ring1B^red and Mel18 loci do not survive birth and showed homogeneous phenotypes, very similar to that of Mel18-null mice, except for the cervicothoracic region, in which almost perfect ribs were bilaterally articulated with an anteriorly shifted sternum in 50% of the mice. A unilateral association was seen in 8.6% of Ring1B^red/red and was never seen in Mel18 null mice (Fig. 6A, part d, part h, Fig. 6B). In summary, the phenotypes at the cervicothoracic level are more severe than the phenotypes of both single mutants in the C57BL/6J background, which provides genetic evidence for an interaction between mutations at the genes encoding Ring1B and Mel18.

By contrast, the posterior transformations seen in the axial skeleton of Mel18^–/– mice were variably restored in Ring1B^red/redMel18^–/– mice on the compound background between 129/SvJ derived from R1 ES cells and C57BL/6J, while enhanced phenotypes were also seen with less penetrance (Fig. 8). Double homozygotes were healthy and fertile on the compound background. We found three groups of mice homozygous for both Ring1B^red and Mel18, depending on the alterations in their axial skeletons. One group, comprising about half of the mice (7/15) of the mice showed phenotypes almost identical to those of Mel18^–/– mice, except for the unilateral association of the seventh rib to the sternum (Fig. 8A, parts f, l and r). A second group, made with the other half of the mice (8/15), showed partial restoration of the posterior homeotic transformations typically seen in Mel18^–/– to the wild-type phenotype. A representative phenotype of three mice in the latter group is the one shown in Fig. 8A (parts d, j and p). They showed normal occipital bones and C1 vertebra, with seven sternovertebral ribs and normally positioned lumbosacral and sacrocaudal boundaries, and anterior shift the thoracolumabar boundary (as represented by lack of 13th rib) (Fig. 8A, parts d, j and p). Thus, this group exhibited a wild-type axial skeleton, except for the alterations in the thoracolumbar transitional zone. Five mice in the latter group, represented by the specimen shown in Fig. 8A (parts e, k and q) exhibited posterior transformations of the occipital, cervical and upper thoracic regions of the axial skeleton characteristics of Mel18^–/– mice. However, the lower thoracic, lumbar, sacral and caudal regions had a wild-type appearance. Therefore, the hybrid background does not allow the Ring1B mutation to lead to a clearly penetrant patterning defect. It also weakens the penetrance of abnormal phenotypes because of the Mel18 mutation at specific AP levels, and it leads to a partial rescue of the Mel18 mutant defects at all AP levels in the compound mutants. Strong interference of strain specific modifiers(s) in the hybrid background probably does not allow for the verification of genetic interaction observed in the C57BL/6.

To examine the effect of a gain of function of Ring1B on Hox gene expression, we unilaterally overexpressed it in the developing neural tube of chicken embryos by in ovo electroporation (Fig. 9A, part a). For 0-5 somites embryos, only 10% of transfected embryos survived in our experimental conditions and most of surviving embryos exhibited extensive morphological defects in the neural tube. Thus, we used embryos older than the 6-somite stage. By means of this method, mouse Ring1B and Mel18 were successfully expressed in the developing spinal cord 24 hours after electroporation, but less intensely in the hindbrain (Fig. 9B, part a, Fig. 9C, part a). We investigated the expression of chicken Hoxb9 in the neural tube because its expression extends up to the level of 9th or 10th somite where stable ectopic expression of mouse Ring1B and Mel18 was seen (Fig. 9A, parts b and c). We first studied the expression of chicken Hoxb9 gene 48 hours after electroporation of chick embryos of various stages of development. In 6-12 somite embryos, we observed a significant posteriorization of the anterior boundary of chicken Hoxb9 expression in 67% of efficiently transfected embryos (Fig. 9B,D). The downregulation of chicken Hoxb9 expression by Ring1B usually extended along several somites (parentheses in Fig. 9B, parts b-d). The introduction of empty vector occasionally resulted in a subtle posteriorization of chicken Hoxb9 of about half somite width (4%) (Fig. 9D) (M. S., unpublished). The Ring1B-mediated repression of chicken Hoxb9 expression was much weaker when 12-18 somites embryos were electroporated (only 25% showed a posterior shift) and no effect was seen on 18-24 somite embryos (Fig. 9D). By analyzing embryos at 24, 36 and 48 hours after electroporation, we found it took 36 hours for the repression of chicken Hoxb9 expression to occur (Fig. 9E).

Similar experiments with Mel18 showed a very mild posteriorization of the anterior boundary of chicken Hoxb9 expression that affected only a somite in only 25% of the efficiently transfected embryos (Fig. 9C,D).

In this study, we provide ample evidence to indicate that Ring1B is a functional component of mammalian PcG complexes. Thus, co-immunoprecipitation and subnuclear localization studies show that Ring1B is found in protein complexes together with other PcG proteins such as Mel18, M33 and Rae28/Mph1. In addition, a hypomorphic loss-of-function mouse model revealed alterations in the axial skeleton and in the expression pattern of some Hox genes, consistent with a PcG function. Conversely, transient gain-of-function experiments in chick embryos showed a role of Ring1B in the repression of Hox gene expression. Finally we showed a genetic interaction between Ring1B and Mel18 resulting in the enhancement of Mel18-deficient phenotypes by the reduction of Ring1B dose on the C57BL/6J background. Taken together, these observations strongly qualify Ring1B as a mammalian PcG gene product.

The phenotype of the Ring1B mutant mice described here show some differences with those of mice with mutations in other PcG genes. Thus, the axial skeletal alterations in Ring1B^red/red mice are restricted to the cervicothoracic transitional zone, while they are seen along the entire axis in other PcG mutant mice (Akasaka et al., 1996; van der Lugt et al., 1994; Core et al., 1997; Takihara et al., 1997; del Mar Lorente et al., 2000). We believe this apparent difference does not imply that Ring1B function is restricted to the prospective cervicothoracic transitional region. First, the mutant mice show a reduction of Ring1B levels, whereas all other PcG mutant mice bear null mutations, and it is possible that specification events in that region are the most sensitive to Ring1B dose. Second, Ring1B^red/red embryos show a de-repression of Hoxb4 mesodermal expression in C1 prevertebra. The fact that neither a transformation of cervical vertebrae C1 to C2 nor any other alterations that affect these vertebrae are observed could be explained if the deregulation of other Hox genes required for C1 specification are not significantly affected in Ring1B^red/red mice. Third, the effect that the low levels of Ring1B has on the phenotype of the axial skeleton of mice lacking Mel18 affects regions throughout the entire axis in hybrid genetic background.

Present and previous biochemical studies indicate that Ring1B is found in protein complexes containing other PcG proteins, such as M33, RYBP, Mel18 and Rae28/Mph1 (Schoorlemmeret al., 1997; Garcia et al., 1999). Furthermore, we have shown for the first time that in mammals these PcG proteins are closely associated to chromosomal DNA, as it had been observed before in D. melanogaster, while significant quantities of them are not closely associated with chromatin (Zink and Paro, 1989; Orlando and Paro, 1993). Ring1B associates with a PcG multimeric complex through multiple protein-protein interactions, as suggested for other components of PcG complexes (Garcia et al., 1999). We have shown here that Ring1B binds Mel18 and Bmi1 via its N-terminal half, which includes a RING finger motif. However, it has been shown that Ring1B binds M33 and RYBP via C-terminal half (Schoorlemmer et al., 1997; Garcia et al., 1999). The structural similarities of Ring1B-binding domains of Mel18 and Bmi1 (94% identity of amino acids), suggest that they may associate with the same interacting surface on Ring1B. As Mel18 and Bmi1 are known to function redundantly during AP specification and protection against apoptosis, biological impacts might be similar between Ring1B/Mel18 and Ring1B/Bmi1 interactions (Akasaka et al., 2001). However, Hox gene expression is uniquely affected in Mel18- and Bmi1-deficient mice (Akasaka et al., 1996; van der Lugt et al., 1996). This suggests that Ring1B might preferentially interact with either Mel18 or Bmi1 in Hox gene-dependent manner. This preference could also be influenced by the other components of the PcG complexes as different sets of Hox genes are also affected in M33- and Rae28/Mph1-deficient mice (Core et al., 1997; Takihara et al., 1997).

Novel genetic interactions between Ring1B^red and Mel18-null mutations upon the compound genetic background suggest the polygenic regulation of mammalian PcG complexes. Restorations of posterior transformations caused by Mel18 null-mutation by Ring1B^red mutation suggests presence of compensatory machinery. In particular, Bmi1 could function to compensate Mel18 mutation. As Bmi1 is known to interact with not only Ring1B but also Rae28/Mph1 and Mph2, multimerization of Bmi1 and its interactors could be modified by unknown polymorphic elements (Hemenway et al., 1998). Molecular basis for genetic modifiers for PcG is not known; however, genes encoding constituents of PcG complex or proteins required for post-translational modification of PcG proteins could be involved (Voncken et al., 1999).

The repression of chicken Hoxb9 expression by transient high levels of mouse Ring1B and Mel18 in chicken embryos is in good agreement with the posteriorization of the rostral boundaries of Hox gene expression seen in Bmi1 and Ring1A transgenic mice (Alkema et al., 1995; del Mar Lorente et al., 2000). These data are also consistent with previous observations that most PcG proteins, including Ring1 family proteins act as transcriptional repressors when recruited to reporter constructs (Satijn et al., 1997b; Schoorlemmer et al., 1997). The anterior boundary of chicken Hoxb9 expression in the neural tube at the level of prospective 9th somite is already established in seven-somite embryos (Fig. 9A, part b). Thus, the overexpression of mouse Ring1B and Mel18 results in the active silencing of a chicken Hoxb9 gene, which is already transcriptionally active. Such an effect may result from the recruiting of ectopic Ring1B and Mel18 to regulatory regions of chicken Hoxb9, which, in turn, may alter locally the structure of the chromatin. However, overexpression of Ring1B does not affect chicken Hoxb9 expression in later stage embryos (Fig. 8C). Biochemical studies using purified PcG complexes and chromatin remodeling factors of the SWI/SNF provide clues on how PcG and trxG proteins counteract (Shao et al., 1999). PcG complexes act as a molecular lock when tethered on nucleosomes prior to exposure to the remodeling factors. Conversely, once exposed to SWI/SNF factors, PcG-mediated repression does not occur. If this is the case, the inability of Ring1B to repress chicken Hoxb9 expression at later stages in development may be due to the establishment of a trxG-dependent activation of chicken Hoxb9. Thus, the accessibility to, or the activity of Ring1B-containing PcG complexes on Hox clusters might be regulated in a developmental stage-dependent manner, as suggested in D. melanogaster (Orlando et al., 1998).

We are grateful to Drs R. Krumlauf, K. Schughart, J. Deschampus, A. Kuroiwa, J. Miyazaki and T. Aoe for the mouse Hoxb4 cDNA clone, mouse Hoxb6 cDNA clone, mouse Hoxb8 cDNA clone, chicken Hoxb9, pCXN2 vector and anti-Bip monoclonal antibody, respectively; to Dr K. Katsube for initial instructions for in ovo electroporation; to Drs T. Oda and M. Muramatsu for initial instructions for yeast two hybrid screening; and to Dr M. van Lohuizen for communication of unpublished data and anti-Bmi1 monoclonal antibody. We also thank Ms Sanae Takeda, Misao Uchida and Mr Shozo Sugimori for their help. This project was supported by Special Coordination Funds for Promoting Science and Technology from the Ministry of Education, Culture, Sports, Science and Technology, the Japanese Government; and by grants from the Naito Foundation, Mochida Foundation, Kanae Foundation and Uehara Memorial Foundation. Work in M. V.’s laboratory is supported by the Ministry of Science and Technology of Spain (grant PB97-1238) and the CAM (grant 08.1/0050/2000).