Multiple epigenetic maintenance factors implicated by the loss of Mll2 in mouse development

Until recently, epigenetic mechanisms have been equated with the inheritance of heterochromatin, exemplified by DNA methylation in mammals(Wolffe and Matzke, 1999; Egger et al., 2004). Evidence that transcriptionally active states of chromatin can also be epigenetically maintained is now accumulating (Roguev et al., 2001; Noma and Grewal,2002; Jaenisch and Bird,2003; Wysocka et al.,2005). This paradigm shift began with the realization that DNA methylation is secondary to a more universal mechanism for epigenetic silencing based on methylation of histone 3 lysine 9 (H3 K9) for constitutive heterochromatin (Rea et al.,2000; Lachner et al.,2001; Bannister et al.,2001) and on H3 K27 for facultative heterochromatin(Cao et al., 2002; Czermin et al., 2002; Kuzmichev et al., 2002; Muller et al., 2002).

Active chromatin states may also be epigenetically maintained because alternative histone lysine methylations, mainly at H3 K4, characterize active chromatin and preclude the lysine methylations that characterize inactive chromatin (Noma et al., 2001; Litt et al., 2001). The functional opposition of these two classes of histone lysine methylation was further supported by linkage to the antagonism between Polycomb- and trithorax-Group (PcG and trxG) action(Brock and Fisher, 2005; Ringrose and Paro, 2004). The association of trxG action with H3 K4 methylation(Roguev et al., 2001; Krogan et al., 2002; Milne et al., 2002; Nagy et al., 2002; Nakamura et al., 2002; Beisel et al., 2002) and PcG action with H3 K27 methylation (Cao et al., 2002; Czermin et al.,2002; Kuzmichev et al.,2002; Muller et al.,2002) provides a molecular explanation for this antagonism.

The silencing methylations at both H3 K9 and H3 K27 are epigenetically maintained via positive-feedback loops. The relevant methyltransferases [SUV39 and E(Z), respectively] associate with a protein (HP1 and Polycomb,respectively) that recognizes the methylated epitope. The enzyme is thereby associated with the chromatin that it has methylated, to methylate it further and propagate the silent state.

Recent evidence indicates that maintenance of active chromatin occurs in the same way, because a constituent protein of all H3 K4 methyltransferase complexes, WDR5/SWD3 (Roguev et al.,2001; Krogan et al.,2002; Nagy et al.,2002; Wysocka et al.,2003; Hughes et al.,2004; Yokoyama et al.,2004; Dou et al.,2005; Lee and Skalnik,2005), binds to methylated H3 K4(Wysocka et al., 2005).

These observations support the polarization model, which is based on opposing positive feedback loops (Jaenisch and Bird, 2003). Both active and heterochromatic states involve several positive feedback loops that reinforce the local status quo. Hence,each state has an implicit epigenetic status that adds stability and reduces the chances of inadvertent transitions to the other state.

The polarization model poses questions that can now be addressed. Do uncommitted neutral chromatin states exist? How are active and silenced states specified? Are transitions between active and heterochromatic states used to regulate gene expression and if so, how? These questions are particularly relevant to the study of development because the transition from the totipotency of the zygote to differentiated states in the adult is reflected by changes in the epigenetic status of chromatin(Jaenisch and Bird, 2003).

S. pombe has only one enzyme for each of the active and heterochromatic lysine methylation sites(Roguev et al., 2003; Sanders et al., 2004; Cam et al., 2005). To date, no evidence for specific gene regulation by histone methylation in either S. pombe or S. cerevisiae has been observed. Higher eukaryotes have several enzymes for each site of lysine methylation. SET domain sequence alignments suggest that the mouse genome encodes at least six H3 K4 methyltransferases and at least seven H3 K9 methyltransferases(Fig. 1). Because gene-specific regulation by epigenesis during development has been anticipated but not yet defined, we decided to look among these SET domain factors in mouse development.

Few lysine methyltransferases have been analyzed for their role during metazoan development. Of the H3 K4 methyltransferases(Fig. 1), only Mll(mixed lineage leukaemia; Mll1 – Mouse Genome Informatics),which is the mouse homolog of Drosophila trithorax, has been mutated in the mouse. The gene has been independently mutated in the mouse three times, each different mutation producing a different lethal phenotype when homozygous. Fusion of lacZ into exon 3 caused embryonic lethality after E10.5 when homozygous, with pleiotropic defects in many tissues and disturbed Hox gene expression (Yu et al.,1995; Hess et al.,1997; Yu et al.,1998; Hanson et al.,1999). This mutant allele expresses the first 458 amino acids of Mll, including the conserved AT hooks region, both protein phosphatase 2A interaction sites and SNL1 (speckled nuclear localization signal) fused to the tetramerizing protein, β-galactosidase (see Fig. S1 in the supplementary material). Heterozygotes showed a hypofertile and mildly homeotic phenotype. By contrast, truncation at exon 5, which permits expression of the above regions plus the MT (CxxC methyltransferase homology) domain, was lethal at the two-cell stage in homozygotes, with a mildly homeotic heterozygous phenotype without reported hypofertility(Ayton et al., 2001). Replacement of exons 12-14 truncated the transcript in the middle of the conserved PHD finger region and caused embryonic lethality around E13.5,without a reported heterozygous phenotype(Yagi et al., 1998). It is not clear whether any of these mutations resulted in a complete loss of function. However, it is clear that Mll is required for definitive hematopoiesis(Ernst et al., 2004a) and sustains expression of certain Hox genes, particularly Hoxa7, Hoxa9and Hoxc8 (Yu et al.,1998; Hanson et al.,1999; Ernst et al.,2004b).

Mammals have a second trithorax homologue, Mll2, as well as two other similar genes Mll3 and Mll4, and two more genes, Set1a and Set1b, which contain very similar SET domains (Fig. 1). Mlland Mll2 are closely related proteins (see Fig. S1 in the supplementary material), having arisen from a duplication that includes the upstream [Plzf (Zbtb16 – Mouse Genome Informatics) and Plzf2] and downstream (U2af1 and U2af1l4) genes(FitzGerald and Diaz, 1999). The functional relationship between Mll and Mll2 is not known. Both proteins are very large, share the same architecture(Fig. 2A; see Fig. S1 in the supplementary material) and are expressed from CpG islands in a nearly, but not completely, ubiquitous manner, including in ES cells and all major tissues, as determined by northern analysis(FitzGerald and Diaz, 1999)(data not shown). Partial characterizations of associated proteins indicate that they both reside in similar complexes(Hughes et al., 2004; Yokoyama et al., 2004; Dou et al., 2005), as do Set1a/b (Wysocka et al., 2003; Lee and Skalnik, 2005).

Because Mll and Mll2 are orthologous, a comparison of their functional roles in mouse development represents a good way to look for evidence of epigenetic regulation in specific gene expression. Therefore we created a multipurpose allele for Mll2. By conversion of the allele from one state to another, we established the null phenotype in mouse development and in ES cells.

A 16.4 kb fragment of Mll2 genomic DNA including exon 1 to exon 10 was isolated from a 129Sv ES cell-phage library. The targeting vector was constructed by inserting an FRT flanked cassette including the splice acceptor sequence from the second exon of the engrailed 2 gene, EMCV IRES, a LacZ-neo fusion and SV40 early polyadenylation signal(Testa et al., 2004) into intron 1 by recombineering (Angrand et al.,1999). The cassette is flanked on the 3′ end by a loxP site and a second loxP site replaced an ApaL1 site in intron 2. Two correctly targeted E14 ES cell clones were used to generate chimeric mice that were subsequently bred to C57BL/6 mice. Offspring from Mll2^+/– crosses and individual embryos were genotyped by PCR with the following primers that produced a 1.7 kb product from the targeted allele: primer 34,5′-GGGCTGACCGCTTCCTCGTGCTTTAC-3′; primer 36,5′-GGAGAACAGTTGTGGGGAGATGGGTC-3′. Homozygote Mll2 embryos were distinguished from heterozygotes with the following primers that produced a 914 bp product from the wild-type allele and no product from the targeted allele: primer 31, 5′-CTCTCTGGTTCTAAGGTAGAGTG-3′ and primer 36. To generate Mll2 F/+ mice, we crossed Mll2^+/–mice to the hATPC-Flpe line (Rodriguez et al., 2000). Subsequent crossing to the PGK-Crem line(Lallemand et al., 1998)produced Mll2 FC/+ mice. Mll2 F/+ and Mll2 FC/+mice were genotyped by PCR with the following primers: primer 145,5′-CGGAGGAAGAGAGCAGTGACG-3′; primer 147,5′-GGACAGGAGTCACATCTGCTAGG-3′. The products were 1554 bp, 1448 bp and 737 bp for the Mll2F, Mll2+ and Mll2FC alleles,respectively. For double-targeted ES cells, the targeting construct was changed from a lacZ-neo fusion to a lacZ-hygro (hygromycin resistance) fusion by recombineering(Testa et al., 2004). Double-targeted cells were selected for G418 (350 μg/ml) and hygromycin(180 μg/ml) resistance.

Total RNA from individual embryos or cells was extracted using TriReagent(Sigma) according to the manufacturer's instructions. First-strand cDNA was synthesized from 1 μg of total RNA using M-MLV reverse transcriptase(Promega). The primer pairs that produced a 1228 bp product were: mll2se,5′-GCAGCAGAGGAGAACCAGACC-3′; mll2as,5′-GGAGGAACCTCCCCTGCCATC. LacZse (5′-AAGTTCAGATGTGCGGCGAGTT) and LacZas (5′-GGCTTCATCCACCACATACAGG) produced a 511 bp product. Actin primers were described previously (Testa et al., 2004). Q-PCR was performed using a Stratagene MX4000 according to the manufacturer's instructions. Cells were homogenized in buffer E (20 mM HEPES, 350 mM NaCl, 10% Glycerol, 0.1% Tween, 1 μg/ml pepstatin A,0.5 μg/ml leupeptin, 2 μg/ml aprotinin, 1 mM PMSF) and snap-frozen three times for protein extraction. Protein extracts were fractionated by 5%SDS-PAGE, transferred to nitrocellulose membranes and revealed with a rabbit IgG polyclonal antibody raised against the Mll2 amino acids between SNL1 and SNL2 (see Fig. S1 in the supplementary material), and a polyclonal CBP antibody (Santa Cruz). The primer pairs for Q-PCR amplification of Hox cluster genes were: Hoxa2, TTCCCAGTTTCGCCTTTAACC and CAGTTCTGGCCCATTGTTGAC; Hoxa3,CCTTTCCCTTTTCTCCTCTGC and ACTGACAGCCTTTCCAGCAAC; Hoxa5, TATAGACGCACAAACGACCGC and CATTTGGATAGCGACCGCA; Hoxb2, CCCGCTGTCTTGGAGACATTT and TTTTGGCTCCCTGGTCTCTGA; Hoxb4, CGGAAACAGGAAAACGAGTCA and TGTGAATACTCCTCGCACGGA;Hoxb5, CCCCAAGTTGCCAGTGTTTCT and AACCTCAACTGCTGCCCCTTA; Pbx1,GCGCCGGGAGCCCATTTCTGC and GGTCCCTCCGGCCCCATCCTG.

Embryos were dissected in PBS containing 0.4% BSA. Fixation was carried out at 4°C in 4% PFA for 20 minutes for E7.5 embryos and 1 hour for later stages. Embryos were washed three times for 5 minutes in PBS and stained overnight at 37°C in PBS containing 0.8 mg/ml X-Gal, 0.2 mg/ml sodium deoxycholate, 5 mM K₃FE(CN)₆, 5 mM K₄Fe(CN)₆, 2 mM MgCl₂ and 0.02% NP-40. Stained embryos were washed twice in PBS and fixed for 1 hour at room temperature in 4% PFA. Embryos were fixed in 4% paraformaldehyde, dehydrated,embedded in paraffin and sectioned at 6 μm. Sections were processed for immunohistochemistry using the monoclonal Ki67 antigen antibody (Novo Castra)and a universal detection kit (Novostain) according to the manufacturer's instructions. Apoptotic cells on paraffin-embedded sections of three mutant embryos and three heterozygous littermates were detected by using an in situ Cell Death Detection Kit (Roche). Total cells numbers were determined on adjacent sections by DAPI staining. Embryos were collected and fixed in 4% PFA in PBS for 1 hour on ice. After three washes in PBS plus 0.1% Tween 20,embryos were stored at –20°C in methanol. Mutant embryos were identified by yolk-sac PCR. Whole-mount in situ hybridization was performed as described (Wilkinson and Nieto,1993). An Mll2 probe was cloned with the following primers: 5-TAGAAGCAGCAGAGGAGAACC-3′ and 5′-GGAGGAACCTCCCCTGCCATC-3′.

Blastocysts were isolated at embryonic day E4.5. After injection of 20-25 targeted ES cells per blastocyst, 14 blastocysts were reintroduced into the uterus of pseudopregnant foster mothers. Injected Mll2^+/– and Mll2^–/–cells were detected in embryos by β-galactosidase staining as whole mount(E8.5 and E9.5) or in sections (E10.5, E18.5). ES cells were differentiated on mass by plating onto bacterial plates in the absence of LIF or retinoic acid.

The murine SET domain proteins were identified by searching the murine proteome (ENSEMBL release 30.33f) using a set of 11 profile HMMs(Eddy, 1998) based on a classification of the human SET domain proteins. All proteins included had E-values lower than 10^–9. Each murine SET protein was classified by the highest scoring profile. The SET domain profile HMMs are available at http://www.uib.no/aasland/chrab/. In brief, they were obtained as follows: initially, a non-redundant set of human SET proteins was obtained with SMART(http://smart.embl-heidelberg.de/). After alignment and clustering the SET domains using Clustal X (Thomspon et al., 1997), nine groups were identified. Clusters of related sequences for each group were obtained by Blastp searches in all of SwissProt/Trembl. For each group, six to eight of the closest sequences were used to build profile HMMs that were subsequently used in iterative searches in SwissProt/Trembl. During this process, the groups SET7/9 and SET8 were separated from the larger Suv3-9 group, resulting in a total of 11 groups and corresponding profile HMMs. An alignment of SET domain sequences from mouse, human, zebrafish and Drosophila corresponding to the 11 groups described above was obtained by progressive profile-profile alignments using Clustal_X guided by a structure mask based on information from the structures of SET7/9, DIM-5, Clr4 and LSMT (pdb: 1O9S, 1PEG, 1MVH, 1P0Y). The mouse subset of this alignment(available at http://www.uib.no/aasland/chrab/)was used to generate a phylogenetic tree using the MrBayes software(Huelsenbeck and Ronquist,2001).

To explore the function of Mll2, we created a multi-purpose allele(Fig. 2A), which can be converted from one state to another using FLPe and/or Cre recombination(Testa et al., 2004). An FRT flanked, genetrap-type, stop cassette(Friedrich and Soriano, 1991)was inserted into the first intron. As determined by RT-PCR(Fig. 2B), western blotting using an Mll2 specific antibody (Fig. 2C), in situ hybridization(Fig. 2D) andβ-galactosidase expression (Fig. 3), this cassette captures and truncates the transcript before the second exon. Because the Mll2 initiating methionine is in the first exon, a transcript encoding the first 121 amino acids of Mll2 fused to 40 frameshifted amino acids encoded by the second exon of the engrailed 2 gene could be expressed. The retained first 121 amino acids of Mll2 do not appear to contain any conserved residues or motifs (see Fig. S1 in the supplementary material).

Removal of one or both FRT cassettes either in homozygously mutated ES cells or in the germline fully restored wild-type function (data not shown). The targeted allele also included loxP sites flanking the small 73 bp second exon (Fig. 2A). Removal of exon 2 by Cre recombination invokes a frameshift in the mRNA. Embryos homozygous for deletions of both the FRT cassette and exon 2 displayed a phenotype indistinguishable from embryos homozygous for the targeted allele(Fig. 2E, Table 1). This phenotype was also observed in embryos homozygous for the removal of the second exon while leaving the FRT cassette in the gene (data not shown). Therefore we conclude that the mutant alleles are most probably nulls. For clarity we term the targeted allele `-', the FLP recombined allele `F', and the FLP and Cre recombined allele `FC' (Fig. 2A). As determined by β-galactosidase expression, Mll2 is expressed widely from both alleles, except in the extra-embryonic tissues (Fig. 3).

Mll2^–/– and FC/FC embryos die before E11.5(Table 1). The Mll2^–/– mutation was originally created on a mixed 129Sv/C57BL/6 background. Careful inspections, particularly of skeletal preparations, failed to identify any heterozygous phenotype on this background. Increasing the C57BL/6 content resulted in a slight reduction in the severity of the homozygous phenotype (data not shown). We were unable to identify any cell-type specific defect in mutant embryos before E9.5. Mutant embryos developed all structures and cell types that we looked for. On the mixed 129Sv/C57BL/6 background, most embryos, but not all, showed incomplete closure of the neural tube. This phenotype diminished with increasing C57BL/6 content. Regardless of background, growth was increasingly slowed in all embryos from E6.5 (Fig. 3) and development was retarded from E7.5, as assessed later by counting somite numbers (Table 2) and observing other morphological markers, such as embryo turning.

The lack of specificity in the phenotype suggested that a general explanation, like a nutritional problem caused by a placental defect, could be involved (Guillemot et al.,1994; Nagy and Rossant,2001). Therefore we used Mll2^–/–ES cells in chimera experiments. The Mll2^–/–cells were made by exchanging the selectable gene in the targeting construct from neomycin resistance to hygromycin resistance by recombineering(Testa et al., 2004), followed by targeting in the Mll2^+/– ES cells and double selection for G418 and hygromycin resistance. Targeting of the second allele occurred at a frequency compatible with the first targeting frequency (3%versus 8%, respectively), hence Mll2 is not required for ES cell survival.

Chimeric embryos were analyzed at E8.5, E9.5, E10.5 and E18.5(Fig. 4, Table 3). As determined by staining for β-galactosidase, highly (>50%) chimeric Mll2^–/– embryos recapitulated the Mll2^–/– phenotype up to E10.5. No heavily blue stained embryos were found at E18.5. Therefore, in concordance with the observed lack of Mll2 expression in extra-embryonic cells(Fig. 3), placental defects do not explain the Mll2^–/– phenotype. Embryos that had a low percentage (<50%) null cells developed normally. However Mll2-null cells were steadily eliminated from these chimeras with increasing developmental age. By E18.5, only a very few Mll2^–/– cells, which appeared to be macrophages or chondrocytes, were found in any embryo(Fig. 4K,L; Table 3). Therefore, Mll2 is cell-autonomously required.

We looked for other causes of the general catastrophe in embryogenesis that were consistent with a cell-autonomous requirement. Embryos were examined for the proliferation marker Ki-67 but showed no significant difference between mutant and wild type (data not shown), as well as for apoptosis by TUNEL staining. Widespread apoptosis was obvious(Fig. 5). Consequently, the simplest explanation for the phenotype is an increased rate of apoptosis,which slows growth and in turn retards development.

A sign of cell type specificity was observed in the Mll2^–/– embryos after E9.5 because the neural tube became kinky. In situ hybridization to Mox1, a marker for the somites (Candia et al., 1992),was used to examine this issue more closely(Fig. 6). Up to about the eight-somite stage, which is E9.5 for the mutant embryos(Table 2), Mox1expression was initiated normally. Thereafter it was lost from the anteriormost somites (Fig. 6B,C) before decaying entirely(Fig. 6D). Loss of Mox1 expression appears to precede the disappearance of the somites for two reasons. First, the observed generalized apoptosis does not explain why Mox1 expression is lost first from the anteriormost somites. Second, inspection of sections showed that Mox1 expression is lost even though some somitic cells remain (Fig. 6F). Later, cells in this region show intense apoptosis(Fig. 6G).

By E9.5, Mll2^–/– embryos are moribund, so the loss of Mox1 expression could be due to a generalized catastrophe rather than to a cell type-specific effect. Evidence in favor of cell type specificity was found because of the persistence of Wnt1 expression along the neural tube (Fig. 7B). As can be seen from this staining, the neural tube becomes highly buckled after the disappearance of the somites.

Trithorax in flies is a known regulator of HOM-C expression, and Mll regulates at least several Hox genes in mouse. Therefore, we looked at Hox complex gene expression. The developmental retardation and embryonic lethality prevented any meaningful analysis of middle to late Hox complex genes. However, a collapse of expression of Hoxb1 occurred sufficiently early in the mutant embryos to permit some confidence(Fig. 8). The correct expression domains of Hoxb1 were established in mutant embryos;however, expression decayed after establishment. Notably, expression decayed at the same time in all three main expression areas. By contrast, expression of Otx2, brachyury (Fig. 7C-E) and Hoxa1 (not shown) appeared normal.

Using Mll^–/– ES cells in embryoid body differentiation experiments, Ernst et al.(Ernst et al., 2004b) showed that the induction of several Hoxa genes was severely impaired; however, Hoxb gene induction was not. Therefore, we performed similar experiments with Mll2^–/– ES cells, using parental E14 cells and the doubly targeted Mll2^–/– cells rescued by FLPe transfection to restore Mll2 expression, as controls(Fig. 8). Loss of Mll2 had no significant effect on expression of several Hoxa genes; however it did have a strong effect on expression of Hoxb2 and Hoxb5. These data strengthen the conclusion that Mll and Mll2 regulate different target genes.

The trxG was originally identified in flies as supressors of PcG mutations. Recent evidence demonstrates that the PcG has biochemical coherence and is linked to histone lysine methylation at H3 K27 and possibly H3 K9(Cao et al., 2002; Czermin et al., 2002; Kuzmichev et al., 2002; Muller et al., 2002; Levine et al., 2004). Whether the trxG has biochemical coherence is not clear. Based on finding a linkage between the yeast homolog of trxG member Ash2 and the first identified H3 K4 methyltransferase Set1, we proposed that a part of trxG action is linked to histone lysine methylation at H3 K4(Roguev et al., 2001). Because active chromatin is characterized by methylation at H3 K4, and inactive chromatin by methylations at H3 K9 and/or H3 K27, the opposition between PcG and trxG lies, in part, at the level of histone methylation.

Mammals appear to have at least two, and up to seven or more, enzymes for each of the histone lysine methylation sites(Fig. 1). The yeast S. pombe has only one enzyme per site and these enzymes appear to play general roles in chromatin status, rather than specific roles in gene regulation. The added histone methyltransferase capacity in mammals suggests that gene-specific regulation in development, based upon epigenetic mechanisms of histone lysine methylation, is possible. To explore this possibility, the function of candidate histone lysine methyltransferases in development must be established. To date, Mll is the only H3 K4 methyltransferase to be analyzed in mouse development. Here, we have analyzed the roles of its sister gene, Mll2.

According to several criteria, the Mll2 alleles studied here show complete loss of function and we report the following observations. Mll2 is expressed in ES cells but is not required for ES cell viability. Mll2^–/– ES cells or embryos show no detectable decrease in global H3 K4 dimethylation level, indicating that Mll2 is not a major source of this modification (data not shown). The first manifestation of loss of Mll2 in development is widespread growth retardation apparent by E7.5. The growth retarded embryos display widespread apoptosis and an increasing delay of developmental progress. By E9.5 the developmental delay is ∼1 day. At this time, the first evidence for cell type specificity, involving specific loss of expression of Mox1 and Hoxb1, was observed.

Previous studies with H3 K4 methyltransferases in development have focused upon gene-specific regulation, including regulation of HOM-C and Hox complex genes, as well as other transcription factors such as the ecdysone receptor(Breen, 1999; Yu et al., 1998; Sedkov et al., 2003). Although we uncovered evidence for gene specificity, the loss of expression of Hoxb1 and Mox1 occurred well after other, widespread,cell-autonomous aspects of the phenotype. Growth retardation, apoptosis and developmental retardation all preceded noticeable cell type-specific effects. These defects were not due to inadequate function of extra-embryonic cells. Because both Hoxb1 and Mox1 expression patterns are established properly in the absence of Mll2, and then decay shortly before death, it is possible that the collapse of expression is due to secondary effects in the dying embryo. However, we found that expression of Wnt1 in cells neighboring those affected by the loss of Mll2 is maintained well after the loss of Mox1 expression. This observation lends some reason to conclude that Mll2 is a specific maintenance factor for Mox1 expression. It is also concordant with the existing propositions regarding its homolog in flies, Trithorax, and sister, Mll, which are both believed to act as maintenance factors for specific gene expression(Sedkov et al., 1994; Yu et al., 1998; Klymenko and Muller,2004).

The roles for H3 K4 methyltransferases in gene expression remain enigmatic. In yeast, there is no evidence so far that the H3 K4 methyltransferase Set1 regulates specific gene expression(Santos-Rosa et al., 2002). Rather, Set1 and H3 K4 methylation play general roles in the maintenance of chromatin status, which includes an association with transcriptional activity(Ng et al., 2003; Krogan et al., 2003) and an opposition to H3 K9 methylation (Noma and Grewal, 2002). In development, the action of H3 K4 methyltransferases as maintenance factors is related to the opposition of PcG repression. That is, specific genes require trxG factors to maintain expression because otherwise PcG action will extinguish expression(Klymenko and Muller, 2004). This opposition may be due to the control of opposing nucleosomal methylations at H3 K4 and H3 K27. However, Trithorax and homologs appear to be transcriptional co-factors, both biochemically, because of associations with CBP (Ernst et al., 2001; Petruk et al., 2001) and the elongating RNAP II (Ng et al.,2003; Krogan et al.,2003; Smith et al.,2004; Guenther et al.,2005), and functionally (Milne et al., 2002; Sedkov et al.,2003; Smith et al.,2004; Milne et al.,2005). Because Trithorax and homologs are invariably large multi-domain proteins, it is likely that they act in both roles: as epigenetic factors to maintain active chromatin and as transcriptional co-factors that interact with the transcriptional machinery.

The Mll2^–/– phenotype displays characteristics consistent with a general role implicit to all cells of the embryo, as well as gene-specific regulation in some cell types. The general role becomes evident after gastrulation and relates to widespread growth retardation, which is probably due to elevated apoptosis. If so, Mll2 may regulate an apoptotic component. Alternatively, increased apoptosis may be a consequence of complications caused by disorganized gene expression resulting from the loss of Mll2. If so, this may reflect a general role for Mll2 in gene expression, such as an association with RNA polymerase, as has been suggested for Mll (Guenther et al.,2005).

Some doubt exists as to the exact nature of the Mllloss-of-function phenotype in development because the three published alleles differ and each could express a part of the protein containing highly conserved regions. Nevertheless, both the Yu(Yu et al., 1995) and Yagi(Yagi et al., 1998) alleles share specific homozygous phenotypic features that are distinct from those reported here for Mll2. When homozygous, neither of those alleles provokes obvious growth and developmental retardation, and somitic development appears to proceed normally at least until E10.5. Thereafter, apoptosis in the somites, but not the neural tube, was observed(Yu et al., 1998). Similarly we observed apoptosis confined to the somites(Fig. 6), but at an earlier developmental stage (Table 2). This represents another difference between the developmental roles of these sister genes. Furthermore, we suggest that the Yagi allele is the null phenotype because of the lack of polyadenylation, nuclear export of truncated Mll transcripts and protein expression. By contrast, it is known that a conserved part of Mll is expressed as a fusion protein withβ-galactosidase in the more aggressive Yu allele, which also shows a heterozygous phenotype as expected for a dominant-negative effect. If this suggestion is correct, then the differences between the Mll and Mll2 loss-of-function phenotypes in mouse development are considerable. The conclusion that Mll and Mll2 regulate different processes in development, including differences in Hox gene regulation, is strengthened by the observation that Mll2^–/– ES cells show severely impaired Hoxb gene inductions(Fig. 9), whereas Mll^–/– ES cells do not(Ernst et al., 2004b). Whether similar conclusions regarding specificities will also apply to the other four members of the H3 K4 methyltransferase group, or indeed to the other classes of epigenetic maintenance or silencing factors(Fig. 1), remains to be determined. Our findings indicate that mammalian development not only relies on networks of transcriptional regulation mediated by sequence specific transcription factors, but also on networks of epigenetic specificities.

We thank Giuseppe Testa and Michelle Meredyth for discussions, Matthew Betts (CBU, Bergen) for advice on using the MrBayes software, and Jussi Helppi for services at the animal facility of the Max-Planck-Institute for Cell Biology and Genetics, Dresden. K.R.T. held a fellowship from Research Council of Norway (146652/431). This work was supported by funding from the VW Foundation, Program on Conditional Mutagenesis and the Sixth Research Framework Programme of the European Union, Project FunGenES(LSHG-CT-2003-503494).