Genetic interactions and dosage effects of Polycomb group genes in mice

In eukaryotic cells, specific transcription factors are needed to activate or repress gene expression through binding to promoter regions and their presence is then required to sustain activation or repression of target genes. However, the regulation of the homeotic genes seems to be an exception, since factors that activate or repress their transcription are only present during a short period of time in early development (Harding and Levine, 1988; Irish et al., 1989). In Drosophila, the domains of homeotic gene expression along the anteroposterior axis is determined by the transient-acting segmentation genes early in embryogenesis (Nusslein-Volhard and Wieschaus, 1980; Harding and Levine, 1988; Casares and Sanchez-Herrero, 1995). Two classes of genes have been identified that act to maintain homeotic gene transcription in the appropriate segments throughout development. The trithorax group (trx-G) sustains the active state in cells where the homeotic gene was originally expressed (for a review, see Kennison, 1993). The Polycomb group (Pc-G) acts to maintain the repressed state in cells where the homeotic gene originally was inactive (for a review, see Paro, 1993). In Polycomb (Pc) mutant embryos from Drosophila, homeotic genes are originally expressed in their appropriate segments but at mid-embryogenesis they are ectopically expressed, which results in the transformation of larval segments towards the identity of more posterior ones, demonstrating that Pc is required to maintain but not to establish the expression domains of the homeotic genes (Wedeen et al., 1986). Genetic analysis has demonstrated that the Pc-G form a large family of

approximately 40 members and both genetic and biochemical evidences have shown that Pc-G gene products constitute large multiprotein complexes that can act through cis-elements called Polycomb responsive elements (PREs) to repress specific target genes (Simon et al., 1993; Chan et al., 1994). The Pc-G proteins are thought to promote stable heterochromatin-like structure that are inherited through many cell divisions (for a review see Pirrotta, 1997). The hypothesis that Pc-G proteins could be involved in chromatin configuration first came from the identification of a conserved chromodomain within Pc. This motive has also been found in HP1, an heterochromatin-associated protein (James and Elgin, 1986) encoded by the Su(va)205 gene (Paro and Hogness, 1991). The co-localization of several Pc-G proteins to homeotic loci on polytene chromosomes (Jürgens, 1985; Dura et al., 1985; Zink and Paro, 1989; DeCamillis et al., 1992; Martin and Adler, 1993; Lonie et al., 1994; Carrington and Jones, 1996; Franke et al., 1992), and the association of Pc to an inactive HOM/lacZ reporter transgene (Müller and Bienz, 1991; Castelli-Gair et al., 1992; Busturia and Bienz, 1993) have led to the proposition that, in Drosophila, the Pc-G multimeric complex interacts with inactive HOM genes and promotes a local remodelling of chromatin leading to the repression of transcription (for a review, see Pirrotta, 1997).

In mice, increasing number of genes sharing structural similarity to members of the Drosophila Pc-G have been identified: the bmi1 (van Lohuizen et al., 1991a,b), mel18 (Tagawa et al., 1990), rae28 (Nomura et al., 1994), M33 (Pearce et al., 1992) and MPc2 (Alkema et al., 1997b) genes. It has been shown that, in mammals, Pc-G genes exist as very related pairs such as bmi1/mel18, M33/MPc2 and ENX1/ENX2 and Hph1/rae28/Hph2 (van Lohuizen et al., 1991b; Akasaka et al., 1996; Hobert et al., 1996; Alkema et al., 1997b; Coré et al., 1997; Takihara et al., 1997). Studies of Pc-G null-mutant mice (van der Lugt et al., 1994; Akasaka et al., 1996; Coré et al., 1997; Takihara et al., 1997) or transgenic mice overexpressing a Pc-G gene (Alkema et al., 1995) have established that changes in the level of individual Pc-G genes lead to skeletal homeotic transformations and revealed both gene-specific as well as common phenotypes. Interestingly these transformations are paralleled by both uniquely affected Hox genes as well as Hox genes that are commonly changed in all Pc-G mutant mice analysed to date. For example in bmi1 mutant embryos, the anterior limit of expression of Hoxc5 is shifted anteriorly by one segment in the somitic mesoderm whereas the expression of this gene is not affected in mel-18 mutant mice. On the contrary, Hoxc8 is shifted in both mutant mice (Akasaka et al., 1996; van der Lugt et al., 1996). Recently the existence of a mammalian Polycomb complex has been biochemically demonstrated, raising the possibility that Pc-G products act, as in Drosophila, in a multimeric complex to perform their silencing function (Alkema et al., 1997a,b; Gunster et al., 1997; Satijn et al., 1997).

In order to obtain genetic evidence for such a common function of the Pc-G genes in mice, we have interbred mice deficient for two non-homologous Pc-G genes. Here we show that mice deficient for both bmi1 and M33 present stronger homeotic transformations of the axial skeleton as compared to each single Pc-G mutant, indicating strong dosage interactions between the two Pc-G genes. Moreover, double mutant mice present additional skeletal transformations that are not seen in single mutant mice, demonstrating that these genes act in synergy during normal development. In double mutant mice, the skeletal transformations are strongly correlated with an enhanced shift of the anterior limit of expression of specific Hox genes in the somitic mesoderm. These results demonstrate that, in mice, the Polycomb group genes act in synergy to control the nested expression pattern of specific Hox genes in somitic mesodermal tissues during development.

Production of M33^−/− (129Ola/balb/c hybrid background) and bmi1^−/− (129Ola/C57Bl/6 hybrid background) mice were already described (van der Lugt et al., 1996; Coré et al., 1997). In timed pregnancies, the day of appearance of vaginal plug was taken as 0.5 dpc. at noon. Pregnant females were killed at the desired gestation time and embryos were collected from the decidua. Amnion was used for genotype analysis.

Whole-mount skeletons of 17.5 dpc animals were stained as described by Conlon and Rossant (1992). Briefly, embryos were skinned, fixed in EtOH for 6-8 hours and then incubated in Alcian blue overnight. Tissues remaining were digested by incubation in 2% KOH 3-4 hours. Coloration of bones was achieved using Alizarine red in 1% KOH for one day. Skeletons were kept in EtOH:glycerol 50:50.

Whole-mount 9.5 dpc embryos were fixed in 4% paraformaldehyde. Hybridization was performed with digoxigenin-labelled probes as described in Henrique et al. (1995). Genotype were determined from yolk sac DNA using PCR analysis. In situ hybridization on serial sections were performed as already described (van der Lugt et al., 1996). The following probes were used: Hoxd4, the full cDNA from D. Duboule; Hoxc9, a 638 bp EcoRI-XhoI fragment containing 3′ untranslated sequence and part of the first exon from A. Awgulewitsch; Hoxc8, a 180 bp SalI-AvaI fragment, from P. Gruss.

M33 and bmi1 heterozygous mice (Coré et al., 1997; van der Lugt et al., 1994) were interbred to generate double homozygous mutant offspring. New-born mice were investigated for genotype by PCR analysis. Only about 2% (4/209) of littermates were of M33^−/−bmi1^−/− genotype whereas the other genotypes were found at the expected Mendelian frequency indicating that most of the double mutant mice did not survive embryogenesis. However, when offspring were genotyped at 10.5 and 11.5 dpc, all the genotypes were found with normal Mendelian frequency, indicating that embryonic lethality was occurring solely in double mutant mice between 11.5 dpc and birth.

It has been previously shown that mice lacking the bmi1 gene present posterior transformations at several positions along the axial skeleton (van der Lugt et al., 1994). Skeletal transformations were also seen in M33^−/− mice (Coré et al., 1997). Some of them are identical to those induced in bmi1^−/− mice and other are specific to the M33 deficiency. In order to investigate potential genetic interactions between these two genes, we analysed the skeletal defects in newborns and embryos lacking both genes. In double M33 bmi1 mutants, we observe an enhanced phenotype as compared to the single mutant animals since stronger skeletal alterations are observed at several positions along the anteroposterior axis (Table 1). Additionally, specific alterations of the skull and the clavicle are detected exclusively in the double mutant mice.

Fig. 1B-F shows lateral views of the cervical region of mutant mice. The malformations in the cervical region of bmi1- and M33-deficient mice have been described in detail previously (van der Lugt et al., 1994; Coré et al., 1997). In short, in bmi1^−/− mice, an extra piece of bone is seen rostrally to the atlas; the neural arches of the atlas (C1) and of the axis (C2) are broadened and split, and the seventh cervical vertebra (C7) presents a rib fused to the first thoracic vertebra (T1) in 50% of the mice (Fig. 1C; Table 1). The M33^−/− mice present a fusion between the atlas and the exoccipital bone and the anterior arch of the atlas is shifted to the axis (C2) indicating a transformation of C2 toward a C1 identity. Those transformations lead to a reduced number of cervical vertebras, six in M33^−/− instead of seven in wild-type or in bmi1^−/− mice (Fig. 1B; Table 1). In the double mutant mice, this region is more severely affected as compared to both single mutants: the atlas is strongly affected, presenting a reduced size (Fig. 1E) or an atypical shape (Fig. 1F), and can be fused to the C2 and C3 cervical vertebrae (Fig. 1F). The posterior cervical vertebrae adopt a progressive thoracic identity since the seventh vertebra (C7) now exhibits a rib fused on T1 in 100% of the mice and since the C6 to C3 cervical vertebrae have rudimentary rib buds (Fig. 1F). Analysis of compound bmi1^−/−M33^+/− and bmi1^+/−M33^−/− animals reveal intermediate alterations as compared to the single and double mutant mice: For example, we observe a partial fusion in between C2 and C1 (Fig. 1D see arrow) whereas this fusion is complete in double mutants (Fig. 1E,F).

Both single mutants exhibit six thoracic ribs attached to the sternum (Fig. 1H) instead of seven in a wild-type mouse (Fig. 1G). Additionally, the number of sternebrae is reduced from six in a wild-type mouse to five in the single mutants. The double mutant mice show an enhancement of these thoracic malformations since only five or four ribs remain attached to the sternum (Fig. 1I). In addition, the number of sternebrae is reduced to four or three in the double mutant (Fig. 1I). The thoracic ribs adopt progressively a more posterior identity since the total number of vertebrosternal ribs is progressively reduced (Table 1). In the lumbar and sacral regions, we observe similarly progressive posterior transformations (Table 1). Furthermore, the significant enhancement of malformations in double mutant mice indicates that Bmi1 and M33 products act together, in a dose-dependent manner, to specify the axial skeleton.

In addition to the enhanced malformations in the axial skeleton, the double mutant mice present defects in the skull and in the clavicle that were never observed in any of the two single mutants (Fig. 2). In wild-type mice, the exoccipital bone of the skull is broad and the two lateral parts are fused together (Fig. 2A). In the double mutant mice, the size of this bone is strongly reduced and the two lateral parts are now almost separated (Fig. 2C). This malformation is dose dependent since the M33^−/−bmi^+/− compound mutant shows an intermediate reduction of size and since the separation of the lateral parts is less complete than in the double mutant (Fig. 2B). An additional new defect is visible in the limbs since the length of the clavicle is reduced in the double mutant (Fig. 2E) compared to the wild type (Fig. 2D) or to the single mutant mice.

Hox genes are crucial to the establishment of the mammalian body plan (McGinnis and Krumlauf, 1992). It has been shown previously that the expression pattern of several Hox genes is affected by the loss of function of a single Pc-G gene in mice (Akasaka et al., 1996; van der Lugt et al., 1996; Coré et al., 1997; Takihara et al., 1997). However, for each Pc-G mutant, only a subset of Hox genes was affected and the anterior boundary of the affected Hox genes was shifted by only one segment in the paraxial mesoderm (see Introduction; Table 2). In order to see if the BMI-1 and M33 proteins cooperate in regulating the expression boundaries of Hox genes, we have analysed the expression pattern of specific Hox genes in double M33 bmi1 mutant embryos. We chose to test Hox genes that showed a misregulation in one or two of the single mutants (Hoxc8, Hoxc9) and Hox genes that were not affected in any of the single mutants (Hoxb1, Hoxd4, Hoxd11) (see Table 2).

Hoxc9 expression normally spans pv16 to the most caudal part of the embryo in the somitic mesoderm and pv10 to the caudal end in the neural tube at 35-somite stage (Erselius et al., 1990) (Fig. 3A). In bmi1^−/− embryos, the boundary is shifted to pv15 in the mesoderm (van der Lugt et al., 1996) whereas we did not detect any change of this boundary in M33^−/− embryos. In the double mutants, the expression pattern of Hoxc9 is shifted anteriorly by 4 segments, up to pv11 in the somitic mesoderm while it is not affected in the neuroectoderm (Fig. 3F). Examination of compound heterozygotes reveals that, in bmi1^+/−M33^−/− embryos, the anterior limit of Hoxc9 expression is unaffected while, in bmi1^−/−M33^+/− embryos, a 1-segment anterior shift is detected up to pv15 in the paraxial mesoderm (Fig. 3D,E). Furthermore, this boundary is also not affected in the trans-heterozygote bmi^+/−M33^+/− embryos (Fig. 3B). These observations indicate that the expression of Hoxc9 is more sensitive to regulation by the bmi1 gene product than by the M33 protein.

Interestingly, when the expression domain of the Hoxc9 gene is essayed earlier, at the 20-somite stage, an anterior shift is detected not only in the mesoderm but also in the neurectoderm (Fig. 3G,H). The difference between the two stages raises the possibility that the expression domain of this gene is also regulated during early development by the Pc-G genes in the neurectoderm. Another possibility is that this neuroectodermal shift can be interpreted as a temporal advance in the caudal to rostral spreading of this gene in the double mutant, as is suggested by our recent studies on M33^−/− mice (S. B. et al., unpublished data).

It has been shown that the anterior limit of expression of the Hoxc8 gene is affected both in bmi1 deficient mice and in mice overexpressing this Pc-G gene (van der Lugt et al., 1996) indicating that Hoxc8 is sensitive to doses of this Pc-G gene. We have tested the pattern of expression of this gene on 12.5 dpc sagittal section of bmi1^−/−M33^−/− embryos and controls (Fig. 4). In bmi1^−/−, one prevertebrae shift is detected in the mesoderm, up to pv11 (van der Lugt et al., 1996). Surprisingly, in contrast to our previous results (Coré et al., 1997), we now detected a 1-somite anterior shift in the M33^−/− embryos (not shown). This observation most likely is caused by differences in genetic background between the two experiments. However, the anterior boundary of Hoxc8 expression is shifted anteriorly by three segments in the mesoderm of double mutant embryos (Fig. 4E,F), indicating an enhancement of Hoxc8 misregulation in the double mutants. In the trans-heterozygotes, a weak signal is detected in pv11 (Fig. 4B). The compound mutant embryos present the same boundaries as the corresponding single mutants; i.e., the mesodermal boundary of bmi1^−/−M33^+/− is shifted up to pv11 as in bmi1^−/− embryos (Fig. 4C) and the mesodermal boundary of bmi1^+/−M33^−/− embryos is shifted up to pv10 as in the M33^−/− embryos (Fig. 4D). Thus, in contrast to the Hoxc9 gene, Hoxc8 appears more sensitive to regulation by the M33 protein than to the BMI-1 product. However, the significantly enhanced shift in the Hoxc8 expression boundary in double mutant embryos clearly indicates that these two Pc-G proteins are both required to properly regulate Hoxc8 in the mesoderm.

The boundaries of both Hoxd11, Hoxb1 and Hoxd4 genes are unaffected in M33^−/− and bmi1^−/− mice. We checked for their expression in 9.5 dpc double mutant embryos, to see if abrogation of two independent Pc-G gene products may now affect the regulation of these genes. As seen in Fig. 5, the anterior limit of Hoxd4 expression remains unaffected both in the mesoderm and in the neurectoderm of double mutant embryos (Fig. 5B). Similarly, the anterior limit of Hoxd11 and Hoxb1 expression patterns remains unaffected in double mutant embryos at the same developmental stage (data not shown). Thus, these results indicate that Hox genes that are not misregulated in one of the two single mutant are either affected in the double mutant. These observations suggest that, as in Drosophila, different Pc-G complexes could be required to regulate different sets of Hox genes.

Most of our knowledge of the molecular function of Polycomb group genes (Pc-G) is derived from experiments in Drosophila. Pc-G mutation analysis has revealed that these genes act together to repress homeotic gene expression throughout development (Jürgens, 1985; McKeon and Brock, 1991; Simon et al., 1992). Initially, genetic data indicated that the Pc-Pc-G genes act together in the same molecular process, indicative of direct interactions between their gene products. This is illustrated by the more severe phenotypes in heterozygous combinations of several different Pc-G genes (Jürgens, 1985). These genetic interactions were further supported by the in vivo co-immunolocalization of different Pc-G proteins on polytene chromosomes (DeCamillis et al., 1992; Rastelli et al., 1993) and by immunoprecipitation experiments demonstrating that Pc and Ph are constituents of a large multiprotein complex containing approximately 10-15 different proteins (Franke et al., 1992). In mice several Pc-G proteins, including M33 and BMI-1, show a complete overlapping subnuclear speckled pattern in interphase nuclei. In addition, the coimmunoprecipitation of BMI-1, MEL18, M33 and MPH1/RAE28 proteins from embryonic extracts supports the existence of a mammalian Pc-G complex (Alkema et al., 1997a). Protein interaction analysis using the yeast two-hybrid screen allowed the mapping of domains in the MPH1/RAE28 protein involved in homodimerization and in direct association with BMI-1 and MEL-18 (Alkema et al., 1997a), demonstrating direct interactions between several Pc-G proteins. In order to genetically test the function of the Pc-G complex in mouse, we have generated double Polycomb mutants and analysed the skeletal transformations as well as the expression pattern of specific Hox genes during development.

Our results demonstrate that at least two members of the Polycomb family, M33 and bmi1, co-operate in vivo to control axial patterning in mouse and illustrate that this co-operation is dose-dependent. Moreover, simultaneous inactivation of both genes leads to enhanced misexpression of several Hox genes during embryogenesis. However, other Hox genes remain unaffected by the compound mutation, reinforcing the hypothesis that different types of Pc-G complexes might exist, controlling different subsets of Hox genes during somitic mesoderm development.

Double M33^−/−bmi1^−/− mutant mice reveal extended posterior transformations of the axial skeleton as well as new skeletal transformations. We previously reported a gene-dosage effect of bmi1 on the specification of vertebral identity and the severity of homeotic transformations (van der Lugt et al., 1994; Alkema et al., 1995). Here we show that the severity of malformations in the craniocervical region increases as the number of wild-type copies of M33 and bmi1 decreases. Indeed, M33^−/−bmi1^+/− mice exhibit an intermediate altered phenotype in the cervical region as compared to each single mutant and to double homozygous mutant mice (see Fig. 1). However, in the thoracic region, these intermediate phenotypes are not seen and only double mutant mice present an aggravated phenotype. These data show that, as in Drosophila (Jürgens, 1985), the murine Pc-G genes act synergistically in a dose-dependent manner. However, skeletal transformations resulting from single or double Pc-G gene mutations in mice are less severe and penetrant than the extreme posterior transformations observed in most of the Pc-G mutants in Drosophila. This observation and the fact that most mammalian Pc-G genes identified so far have a close relative (bmi1/mel18; M33/MPc2; Enx1/Enx2) suggest that Pc-G genes act in a more redundant way in mice compared to the fly, to control Hox gene expression pattern.

It has been previously shown that inactivation of either M33 or bmi1 alters similar structures in the axial skeleton, but some of these structures, like the atlas or the axis in the cervical region, can be transformed differently in each mutant mouse (van der Lugt et al., 1994; Coré et al., 1997). One possible explanation for these different phenotypes could be that the two genes are involved in controlling the correct identity of the same structure by regulating overlapping but different subsets of Hox genes within the corresponding somites. Nevertheless, when the two Pc-G genes are simultaneously deleted, the altered phenotype is aggravated in the same direction, toward stronger posteriorisation of the structure, revealing the synergistic effect of the combined mutation. This is clearly visible in the sternum: whereas both M33^−/− and bmi1^−/− mice exhibit the same malformation corresponding to a reduced number of sternebrae, the M33^−/−bmi1^−/− mutant shows a more dramatic alteration of this region characterised by an increased reduction of the sternebrae. Similarly, while the seventh cervical vertebra (C7) is never affected in the M33^−/− mutant and only slightly affected in the bmi1^−/− mutant, the inactivation of both genes leads to a C7 transformation towards T1 in 100% of the double mutant mice. The cooperative interactions between bmi1 and M33 mutants are indicative of their involvement in the same molecular process and are in line with interactions of the respective gene products in a larger protein complex (Alkema et al., 1997a). However, biochemical studies and two-hybrid experiments failed to show a direct interaction between BMI-1 and M33 (Alkema et al., 1997a). This could indicate that BMI-1 and M33 may require additional Pc-G proteins to be able to interact, reinforcing the need of multiprotein complexes to repress Hox gene expression.

Additionally, analysis of the Pc-G double mutants revealed the appearance of new abnormalities in the skull and the clavicle that have not been observed in single Pc-G mutant mice. These phenotypes emphasises the synergistic effect of the two Pc-G genes and suggest that the loss of two members of the Pc-G multimeric complexes probably reflects either an increased number of derepressed Hox gene targets or a combinatorial effect of several Hox genes expressed in ectopic positions.

When Pc-G mutant mice are compared, loss of each Pc-G gene shows a unique subset of affected Hox genes (see Table 2). For example, in M33 mutant mice, we have only been able to detect anterior shifts for Hoxa3 and, in some cases (this work), for Hoxc8. mel18^−/−, bmi1^−/− and rae28^−/− mice present a more extensive overlap in affected Hox genes, encompassing 1 pv anterior shifts of Hoxa5, and Hoxc8. However, Hoxc6 and Hoxc5 are uniquely affected in bmi1^−/− mice, while Hoxa7 and Hoxd4 are only affected in mel18^−/− mice and Hoxb5 is unaffected in all those mutant mice (Table 2). In M33 bmi1 double mutant mice, we show that the anterior limit of expression of at least two Hox genes, Hoxc9 and Hoxc8, is significantly more severely shifted as compared to both single mutants, demonstrating the additive effect of Pc-G products in maintaining the boundaries of selected Hox genes. One striking observation is that deleting 3 gene doses of these Pc-G genes does not increase the derepressive effect of Hoxc8 or Hoxc9 expression; full deficiency of M33 and bmi1 is required to induce extensive ectopic expression of these two genes in mesodermal tissues. Nevertheless, according to the differential expression of Hoxc8 and Hoxc9 in both single mutants, it seems that Hoxc8 is more sensitive to M33 regulation since we observe a one-segment anterior shift in M33^−/−bmi1^+/− (pv10) whereas that shift is not visible in bmi1^−/−M33^+/− (pv11); reciprocally, Hoxc9 is more sensitive to bmi1 regulation.

In contrast, we show that some other Hox genes like Hoxb1, Hoxd4 and Hoxd11 are not affected either in single or in double mutant mice. This suggests that several murine multimeric Pc-G protein complexes of different composition might exist that differ in their affinities for specific Hox genes. Alternatively, since in mammals Pc-G genes exist as highly related gene pairs such as mel18/bmi1, Enx1/Enx2, M33/MPc2, hPc1/hPc2 and Hph1/rae28/Hph2 (reviewed in Gould, 1997), a potential redundancy likely exists. This suggests that the homologous gene complements part of the function and thus maintains the boundaries of expression for a subset of Hox genes. Analysis of double mutant mice for a related pair such as mel18 and bmi1 should clarify the degree of redundancy. Differential effects of Pc-G mutations on Hox gene expression in different tissues like the somitic mesoderm and the neural tube, also suggest that different Pc-G complexes may regulate a specific subset of Hox genes in a tissue-dependent manner.

Biochemical evidence for the existence of separate multiprotein Pc-G complexes in mice has recently been obtained (van Lohuizen et al., 1998). In addition, this is also supported by other studies in Drosophila. Immunocolocalization of Pc-G proteins on Drosophila polytene chromosomes reveals that the Pc-G products do not bind strictly to the same targets but can display overlapping binding sites (Rastelli et al., 1993). Moreover, recent studies have demonstrated, by immunoprecipitation experiments using in vivo cross-linked chromatin, the existence of Pc-G protein complexes of different compositions on different target genes (Strutt and Paro, 1997).

All together, our data demonstrate that, as in Drosophila, Pc-G products in mammals act in synergy to control the expression pattern boundaries of Hox genes and further support the importance of a dose-dependent Pc-G regulatory function to specify the axial skeleton identity in mice.