Advertisement

A mono‐allelic bivalent chromatin domain controls tissue‐specific imprinting at Grb10

Lionel A Sanz, Stormy Chamberlain, Jean‐Charles Sabourin, Amandine Henckel, Terry Magnuson, Jean‐Philippe Hugnot, Robert Feil, Philippe Arnaud

Author Affiliations

  1. Lionel A Sanz1,
  2. Stormy Chamberlain2,,
  3. Jean‐Charles Sabourin3,,
  4. Amandine Henckel1,
  5. Terry Magnuson2,
  6. Jean‐Philippe Hugnot3,
  7. Robert Feil*,1 and
  8. Philippe Arnaud*,1
  1. 1 Institute of Molecular Genetics, CNRS UMR‐5535 and University of Montpellier II, Montpellier, France
  2. 2 Department of Genetics, Carolina Center for the Genome Sciences, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA
  3. 3 INSERM U583, Physiopathologie et Thérapie des Déficits Sensoriels et Moteurs, Institut des Neurosciences de Montpellier, Hôpital St Eloi, Montpellier, France
  1. *Corresponding authors: Institute of Molecular Genetics, CNRS UMR‐5535 and University of Montpellier II, 1919 route de Mende, Montpellier 34293, France. Tel.: +33 4 6761 3661; Fax: +33 4 6704 0231; E-mail: philippe.arnaud{at}igmm.cnrs.fr or E-mail: robert.feil{at}igmm.cnrs.fr
  1. These authors contributed equally to this work

Abstract

Genomic imprinting is a developmental mechanism that mediates parent‐of‐origin‐specific expression in a subset of genes. How the tissue specificity of imprinted gene expression is controlled remains poorly understood. As a model to address this question, we studied Grb10, a gene that displays brain‐specific expression from the paternal chromosome. Here, we show in the mouse that the paternal promoter region is marked by allelic bivalent chromatin enriched in both H3K4me2 and H3K27me3, from early embryonic stages onwards. This is maintained in all somatic tissues, but brain. The bivalent domain is resolved upon neural commitment, during the developmental window in which paternal expression is activated. Our data indicate that bivalent chromatin, in combination with neuronal factors, controls the paternal expression of Grb10 in brain. This finding highlights a novel mechanism to control tissue‐specific imprinting.

Introduction

Genomic imprinting is a form of non‐Mendelian inheritance in mammals, whereby some genes are expressed from a single allele, depending on whether it is inherited from the mother or the father. Many of the 80 imprinted genes discovered so far in humans and mice are involved in the regulation of cellular proliferation and growth, whereas others have key functions in neurological processes and behaviour (Smith et al, 2006; Wilkinson et al, 2007). Not surprisingly, therefore, deregulation of imprinting gives rise to abnormal development and is causally involved in a number of growth and behavioural syndromes in humans (Arnaud and Feil, 2005).

To achieve allele‐specific expression, imprinted loci are regulated by epigenetic modifications that differentially mark the parental alleles as either active or repressed. This is manifest at key control elements known as imprinting control regions (ICRs), which mediate imprinted gene expression. All known ICRs are marked by DNA methylation on either their maternal or their paternal allele. This allelic methylation is established in the female or male germ line and is subsequently maintained throughout development (Delaval and Feil, 2004). In addition to DNA methylation, ICRs are also marked by differential histone modifications in somatic cells. The allele harbouring DNA methylation is associated with ‘repressive’ histone marks such as trimethylation of lysine 9 of histone H3 (H3K9me3), whereas the unmethylated allele is marked by ‘permissive’ marks such as H3K4me2 (Fournier et al, 2002; Yang et al, 2003; Delaval et al, 2007). DNA methylation at ICRs is essential for orchestrating mono‐allelic gene expression. However, the precise role of histone tail modifications in this process remains poorly understood.

One possibility is that histone modifications could be important for the ICR‐mediated repression of imprinted domains. It is well documented that these modifications can influence gene expression (Margueron et al, 2005; Li et al, 2007). Thus, their allelic distribution at imprinted loci could be involved in regulating mono‐allelic expression. Recent studies of tissue‐specific imprinting at the Igf2r, NDN and Gnas loci (Yang et al, 2003; Lau et al, 2004; Li et al, 2004) have correlated imprinted expression with tissue‐specific differences in histone modifications. A more direct relationship between histone modifications and tissue‐specific imprinting exists at the imprinted Kcnq1 domain in the mice, where a maternally methylated ICR controls a number of paternally repressed genes, some of which are imprinted in the placenta only. Across the Kcnq1 domain, the repressed paternal allele is enriched in repressive marks (H3K9me2 and H3K27me3), especially in the placenta (Umlauf et al, 2004). A functional role for these marks is suggested by studies showing that placenta‐specific imprinting at Kcnq1 domain is perturbed in mouse conceptuses deficient in G9A, a H3K9me2 methyltransferase (Wagschal et al, 2008), or in EED (embryonic ectoderm development), a component of the PRC2 complex that mediates H3K27 methylation (Mager et al, 2003).

The growth factor receptor‐binding protein 10, GRB10, is a potent growth inhibitor (Charalambous et al, 2003). We and others have shown that the mouse Grb10 gene displays a tissue‐specific reciprocal imprinting pattern based on the use of tissue‐specific promoters (Arnaud et al, 2003; Hikichi et al, 2003). In most tissues, Grb10 is expressed from the maternal allele, whereas in brain, its expression is mainly from the paternal allele. Strikingly, the maternal and paternal expressions initiate from different promoter regions.

DNA methylation has been shown to be important in the imprinting of Grb10 (Arnaud et al, 2006). This agrees with the current model to explain the maternal Grb10 expression that proposes the existence of a methylation‐sensitive boundary mechanism that prevents promoter–enhancer interactions (Arnaud et al, 2003; Hikichi et al, 2003). However, this model does not account for the brain‐specific paternal expression where other epigenetic modifications are likely to be involved. To investigate whether histone modifications could be involved, we compared their distribution across key chromosomal regions of Grb10 in brain and several other tissues. We focused our analysis on the CpG island (CGI) region that corresponds to the putative ICR and comprises brain‐specific paternal promoters. We identified a mono‐allelic bivalent chromatin domain that carries both active (H3K4me2) and repressive (H3K27me3) histone marks and that controls paternal expression of Grb10. The loss of this bivalent chromatin, together with neuronal‐specific factors, is important to de‐repress Grb10 expression in brain.

Bi‐allelic bivalent chromatin has been observed at the promoters of many genes in embryonic stem (ES) cells as well as in differentiated cells (Azuara et al, 2006; Bernstein et al, 2006; Roh et al, 2006; Pan et al, 2007; Mikkelsen et al, 2007; Zhao et al, 2007), and has been considered to be involved in maintaining key developmental genes in a ‘permissive’ chromatin configuration, while being repressed by H3K27me3 (Azuara et al, 2006; Bernstein et al, 2006). Our work shows that bivalent chromatin is also important in the control of imprinted gene expression. Furthermore, our data provide one of the first examples of the implication of bivalent chromatin in the control of gene expression in a differentiated tissue.

Results

Paternal enrichment for H3K27me3 on the brain‐specific paternal promoters is lost in brain

We previously established that in neonatal tissues, the three CGIs located in the 5′ untranslated region (UTR) of Grb10 display reciprocally imprinted and tissue‐specific promoter activities (Arnaud et al, 2003). Here, we extended and confirmed this analysis by using the material obtained from reciprocal crosses between C57BL/6J (B6) and Mus musculus molossinus JF1 mice. Maternal expression was detected in all tissues and developmental stages analysed, except in brain, and initiates from exon 1A located in CGI1. By contrast, the brain‐specific paternal expression arises from three independent regions: exon 1C in CGI3 and exons 1B1 and 1B2, immediately upstream and at the 3′ end of CGI2 (Figure 1A; Supplementary Figures 1 and 2A). Further analysis of RNA from glial cells and a neuron‐enriched fraction of cells derived from embryonic cerebral cortices showed that the paternal expression from exons 1B2 (Yamasaki‐Ishizaki et al, 2007) and 1B1 (this study) is restricted to neuronal cells (Supplementary Figure 1D).

Figure 1.

Paternal enrichment of H3K27me3 at CGI2 is lost in brain. (A) Overview of tissue‐specific imprinted expression of Grb10. The upper panel gives a schematic representation of the Grb10 gene. Untranslated and coding exons are shown in grey and black, respectively. The 5′ UTR region encompassing the three CpG islands (CGI, CGI3 and CGI2) is drawn to scale. CGI1 is located roughly 70 kb upstream of the first coding exon; CGI2 10 kb downstream of CGI1 and CGI3 8.5 kb downstream of CGI1. Maternal expression arises from exon 1A within CGI1, whereas brain‐specific paternal expression arises from exons 1C, 1B1 and 1B2 located in CGI3 and CGI2. (B) Representative results of allele‐specific analysis of histone marks at CGI2. Native ChIP followed by SSCP‐mediated allelic discrimination in the antibody‐bound (B) and unbound (U) fractions was performed on the material from (B6 × JF1)F1 mice (indicated on the left). Maternal (mat.) and paternal (pat.) alleles are indicated. The relative position of the SSCP polymorphism at CGI2 is shown (black star). (C) Relative enrichment of the paternal versus maternal allele in chromatin precipitated with anti‐H3K27me3 and anti‐H3K9ac antisera, respectively. For each tissue, the value is reported as the mean of the results obtained from at least two independent ChIP experiments (n) normalized to the allelic results obtained in the corresponding input chromatin. Brain (n=3); neonatal brain (n=3); 12.5 d.p.c. embryo (n=4); 12.5 d.p.c. placenta (n=2); liver (n=4); kidney (n=5).

We previously observed brain‐specific hypomethylation of CGI3 on the paternal allele (Arnaud et al, 2003), suggesting that DNA methylation is involved in controlling the expression from promoter 1C. However, the regulation of Grb10 paternal expression cannot be explained solely by DNA methylation. Indeed, DNA methylation pattern does not vary at the brain‐specific paternal promoters 1B1 and 1B2 between brain and other tissues. The putative ICR CGI2, which comprises 1B1 and 1B2, maintains its maternal methylation in all tissues and developmental stages analysed (Arnaud et al, 2003; Hikichi et al, 2003; Yamasaki‐Ishizaki et al, 2007). Therefore, we focused our analysis on histone tail modifications at CGI2. Particularly, we wished to determine whether allelic histone modification patterns could vary between brain and other tissues. To this purpose, we performed chromatin immunoprecipitation (ChIP), followed by electrophoretic detection of single strand conformational polymorphism (SSCP) to evaluate for each antibody‐bound and unbound fractions the relative abundance of the maternal and paternal alleles. ChIP was performed with antisera against permissive (H3K9ac and H3K4me2) and repressive (H3K27me3, H3K9me3 and H4K20me3) marks. In all tissues and developmental stages analysed, we observed constitutive allelic enrichment of specific histone modifications. H3K4me2 was always associated with the unmethylated paternal allele of CGI2, whereas H3K9me3 and H4K20me3 were consistently associated with the DNA‐methylated maternal allele (Figure 1B, not shown). By contrast H3K9ac, which marks active chromatin (Jenuwein and Allis, 2001), was found to be specifically associated with the paternal allele in brain and, to a lesser extent, in 12.5 days post‐coitum (d.p.c.) embryos in which paternal expression arising from the developing brain is detected (not shown) (Figure 1B and C). A similar distribution was observed for H3K27ac, with no allelic enrichment in adult liver and kidney and a clear paternal enrichment in brain (Supplementary Figure 3A).

Unexpectedly, we observed that the repressive mark H3K27me3 was enriched on the unmethylated paternal allele of CGI2 (Figure 1B). This observation suggested that both the repressive H3K27me3 and the permissive H3K4me2 coexisted on the paternal allele of CGI2. Moreover, paternal H3K27me3 enrichment was specifically lost in brain (Figure 1B and C, not shown). Indeed, relative H3K27me3 paternal enrichment in placenta and kidney (3.29±0.56 and 3.20±0.35, respectively) was much higher than in adult (1.03±0.05) and neonatal brain (1.26±0.04) (Figure 1C). These observations indicated that at CGI2, the paternal enrichment in H3K27me3 correlates with repression of this promoter region, suggesting a role for this modification in the regulation of the paternal Grb10 expression.

Allele‐specific bivalent ‘H3K4me2/H3K27me3’ chromatin marks CGI2

The finding that the permissive H3K4me2 and the repressive H3K27me3 marks are both enriched on the paternal allele of CGI2 is reminiscent of the bivalent chromatin domains recently described in ES cells (Azuara et al, 2006; Bernstein et al, 2006). Accordingly, we established that the allelic chromatin pattern across CGI2 in undifferentiated ES cells (line SF1‐G) was also characterized by paternal co‐enrichment of H3K4me2 and H3K27me3 (Figure 2A). This chromatin pattern was also observed in undifferentiated trophoblast stem (TS) cells (Supplementary Figure 3B). To firmly establish that CGI2 is mono‐allelically marked by bivalent chromatin in differentiated cells, we performed sequential ChIP on mouse embryonic fibroblasts (MEFs). In this approach, a first round of ChIP precipitated chromatin fragments enriched in H3K4me2. These fragments were subsequently precipitated with antibodies against H3K27me3. This sequential round of immunoprecipitation retains only chromatin fragments that carry both the histone modifications. To validate this approach, we included the Irx2 and Tcf4 genes as controls. The promoter region of Irx2 has a H3K4me3/H3K27me3 bivalent chromatin in ES cells (Bernstein et al, 2006; Azuara et al, 2006) that is maintained in MEFs (Mikkelsen et al, 2007; http://www.broad.mit.edu/seq_platform/chip/). The detection of the Irx2 promoter in the final bound fractions was consistent with these observations (Figure 2B). Conversely, the Tcf4 promoter is known to be enriched in H3K4me3 only in MEFs (Mikkelsen et al, 2007; http://www.broad.mit.edu/seq_platform/chip/), and could be PCR amplified only after the first round of ChIP with the anti‐H3K4me2 antibody (Figure 2B). Analysis performed on the CGI2 region demonstrated that both the modifications were present on the same chromatin fragments (Figure 2B), indicating the presence of a bivalent chromatin in this region in somatic cells.

Figure 2.

CGI2 is specifically marked by an H3K4me2/H3K27me3 bivalent domain. (A) H3K4me2 and H3K27me3 are enriched on the paternal allele of CGI2 in undifferentiated ES cells. Analysis was performed on (C57BL/6J × M. spretus) F1 undifferentiated ES cells. (B) H3K4me2 and H3K27me3 are enriched on the same chromatin fragments at the CGI2 region. Sequential ChIP was performed on MEFs to detect chromatin fragments enriched for both H3K4me2 and H3K27me3. Irx2 was used as a positive and Tcf4 as a negative control. (C) The H3K4me2/H3K27me3 bivalent domain is restricted to CGI2. Allelic analysis of histone modifications in kidney and brain at CGI1, CGI3 and a non‐CpG‐rich region located upstream of the transcriptional unit (region ‘UR’). Kidney data are shown as an example of representative results obtained with ‘non‐brain tissues’. At the three regions, H3K27me3, but not H3K4me2, was enriched on the paternal allele. This paternal enrichment is lost in brain tissue. Relative positions of polymorphisms used for the SSCP‐based allelic discrimination are shown (black stars). TSS: transcriptional start site.

To evaluate the extent of this bivalent chromatin domain, we analysed chromatin at CGI1, CGI3 and at a third region upstream of the transcriptional unit, which we called UR, for upstream region (Figure 2C). At all three regions, in ‘non‐brain‐tissues’ H3K27me3 was associated with the paternal allele. However, H3K4me2 was enriched only on the maternal CGI1 allele and displayed no allelic enrichment at CGI3. This finding shows that the bivalent domain observed in non‐brain tissues does not extend beyond the CGI2 region. We completed these observations by consulting an online ChIP database (Mikkelsen et al, 2007; http://www.broad.mit.edu/seq_platform/chip/). Chromatin patterns obtained from ES and MEF cells indicate that the entire Grb10 5′ UTR region is enriched in H3K27me3, whereas H3K4me3 is more specifically localized in CGI2 and CGI1 (Supplementary Figure 4). This information together with our allelic analysis establish that the paternal allele of the Grb10 5′ UTR region is enriched in H3K27me3 and that at CGI2, this mark is specifically associated with H3K4me2/me3, thus forming a bivalent chromatin domain.

During neural precursor cell differentiation the CGI2 bivalent chromatin domain is resolved and Grb10 paternal expression is induced

To evaluate how the CGI2 bivalent chromatin evolves during nervous system development and to explore its effects on Grb10 imprinted expression, we used cultures of highly enriched embryonic neural precursor cells called neurospheres (Reynolds and Weiss, 1992). Neurospheres are generated in suspension by clonal expansion of neural stem cells and are composed of a mixture of neural stem–progenitor cells with little or no differentiated cells (Deleyrolle et al, 2006; Supplementary Figure 5A). Upon growth factor withdrawal in adherent culture conditions, neurospheres generated about 70% astrocytic cells (GFAP+), 10–20% neurons (β‐3‐tubulin+) and approximately 1% oligodendrocytes (O4+) (Supplementary Figure 5B; Deleyrolle et al, 2006).

By exon‐specific RT–PCR amplification coupled with sequencing, we revealed that in undifferentiated embryonic neurospheres, Grb10 was maternally expressed from exon 1A (Figure 3A; Supplementary Figure 2B). This expression pattern correlated with the presence of bivalent H3K4me2/H3K27me3 chromatin on the paternal CGI2 allele (Figure 3B, upper panel). Upon differentiation, we detected both maternal expression from exon 1A and paternal expression from exon 1B1 (Figure 3A; Supplementary Figure 2B). The bivalent chromatin was almost resolved, as the H3K27me3 enrichment on the paternal allele was largely reduced, whereas H3K4me2 remained unchanged (Figure 3B). To evaluate whether this relative decrease in paternal H3K27me3 was due to the loss of H3K27me3, we determined its abundance by real‐time PCR amplification after ChIP. A decrease of nearly 75% was detected at CGI2 upon neurosphere differentiation (Figure 3C). A similar decrease was also observed at CGI1 (Supplementary Figure 5D). Paternal enrichment of H3K9ac was detected in both undifferentiated and differentiated neuronal cells. However, the amount of this modification at CGI2 clearly increased upon differentiation (Figure 3C), suggesting that the allelic enrichment observed in undifferentiated neurospheres arises from a limited subset of cells. We observed a similar pattern for H3K27ac (Supplementary Figure 5C). Thus, during neural stem/progenitor cell differentiation, the CGI2 bivalent domain is almost completely resolved through a decrease of H3K27me3 whereas acetylation increases. This chromatin change is concomitant with the induction of Grb10 paternal expression from exon 1B1 (Figure 3A).

Figure 3.

Resolution of the bivalent domain upon neural differentiation correlates with paternal Grb10 expression from CGI2. (A) The promoter usage in undifferentiated (undiff.) and differentiated (diff.) embryonic neurospheres was determined by exon‐specific RT–PCR amplification. Relative positions of the polymorphism (black star) and primers used are indicated. Paternal expression from the CGI2 region was detected only after differentiation. De‐repression appeared to be limited to exon1B1, as we failed to detect transcripts from exon1B2. (B) Representative allele‐specific analysis of histone modifications at CGI2 in undifferentiated (undiff.) and differentiated (diff.) neurospheres. The upper panel shows the results of the SSCP analysis. The lower panel shows the relative enrichment of the paternal versus the maternal allele in H3K27me3 precipitations. (C) The level of H3K27me3 decreases at CGI2 during neurosphere differentiation, whereas H3K9ac increases. Following ChIP, the abundance of H3K27me3 and H3K9ac was determined by real‐time PCR quantification of bound fractions and reported as a percentage of the value obtained with input chromatin.

Ectopic Grb10 paternal expression in Eed−/− ES cells and embryos

In ES cells, bivalent chromatin is thought to maintain key developmental genes in a ‘permissive’ chromatin configuration through H3K4me2, while repressing them by means of H3K27me3 (Azuara et al, 2006; Bernstein et al, 2006; [35]Mikkelsen et al, 2007). This idea is supported by the finding that some of these genes are inappropriately upregulated in ES cells deficient for EED (Azuara et al, 2006), a component of the PRC2 complex that mediates H3K27 tri‐methylation. Therefore, to investigate the functional significance of the developmental loss of H3K27me3, we analysed Grb10 imprinted expression in two Eed−/− ES cell lines (i.e. G8.1 and B1.3) (Azuara et al, 2006). As controls, undifferentiated wild‐type ES cell lines were analysed. Expression from the 1B1 promoter was detected in both mutant lines, but not in the wild‐type cells (Figure 4A). However, the absence of allelic polymorphisms in G8.1 and B1.3 cells did not allow us to draw formal conclusions about the parental origin of this transcript. Nevertheless, bisulphite treatment of DNA followed by sequencing suggested that the maternal DNA methylation at CGI2 was unaltered in the two mutant cell lines (data not shown). The detected transcript was thus likely to originate from the unmethylated paternal allele. These results indicate that, in ES cells, EED deficiency and the resulting loss of H3K27me3 leads to a partial de‐repression of brain‐specific promoters at CGI2.

Figure 4.

Effects of EED deficiency on Grb10 imprinted expression. (A) EED is required for repressing the GCI2 promoter 1B1 in ES cells. Eed‐deficient and wild‐type ES cells were studied by RT–PCR amplification; Gapdh is used as a semiquantitative reference. We failed to detect transcripts initiating from exons 1C and1B2 in both wild‐type and Eed‐deficient ES cell lines. (B) EED deficiency has a limited impact on Grb10 paternal expression in 6.5 d.p.c. Grb10XC302 conceptuses. Embryos from a cross between an Eed−/+ female and an Eed−/+; Grb10XC302 male were collected at 6.5 d.p.c., genotyped and stained with X‐Gal.

To extend our analysis, we studied mice bearing a novel gene‐trap insertion in intron 7 of Grb10, referred to as Grb10XC302. In these mice, LacZ staining can be used to visualize Grb10 expression. RNA in situ hybridization confirmed that LacZ staining faithfully recapitulated endogenous Grb10 expression (data not shown). Through paternal and maternal transmission of Grb10XC302, Grb10 imprinted expression was monitored in wild‐type and Eed−/− embryos. As expected, Grb10 was widely expressed from the maternal allele in both cases, as indicated by the ubiquitous blue staining (Supplementary Figure 6). Conversely, following paternal transmission, Grb10 expression was restricted to specific cell populations in wild‐type 6.5 d.p.c. embryos (Figure 4B). Particularly, Grb10 was expressed in the anterior visceral endoderm (AVE), an extra‐embryonic tissue required for specifying early anterior patterning in the mouse embryo. In Grb10XC302/Eed−/− conceptuses, ectopic paternal expression of Grb10 was not found to be ubiquitous, but rather, was limited to one part of the extra‐embryonic ectoderm (Figure 4B). This ectopic expression is likely to be a direct consequence of the absence of EED rather than an expansion of the AVE region in the mutant embryo. Previous lineage analysis showed that the AVE does not expand into the extra‐embryonic region (Faust et al, 1998; Rivera‐Pérez et al, 2003). Together these observations suggest that absence of EED (and the associated H3K27me3 mark), although involved in, is not sufficient on its own to promote the paternal expression of Grb10.

Discussion

In this study, we unravelled how the paternal expression of Grb10 in brain is controlled. Our main finding is that this process is regulated at least in part by a mono‐allelic bivalent chromatin domain that marks the brain‐specific paternal promoters of Grb10.

We confirmed and extended previous Grb10 imprinted expression studies (Miyoshi et al, 1998; Arnaud et al, 2003; Hikichi et al, 2003). Grb10 is likely to be imprinted at pre‐implantation stages as maternal expression is detected in both undifferentiated ES and trophoblast cells. Maternal expression is subsequently maintained in the developing embryo, placenta and in all adult tissues, except brain. The reciprocally imprinted, paternal expression observed in brain relies on three different promoters and is restricted to neuronal cells (this study; Yamasaki‐Ishizaki et al, 2007). This is consistent with the expression pattern we observed in neurospheres, where paternal expression from the promoter in exon 1B1 is initiated upon differentiation. Hence, the paternal expression is activated in the neural lineage, when the neural stem/progenitor cells are committed towards the neuronal fate. Nonetheless, the three promoters are not activated simultaneously, and paternal expression from exon 1C is detected only after birth (data not shown). Therefore, the discrete paternal expression we observed in 6.5 d.p.c. conceptuses is intriguing. This paternal expression could result from a ‘leakage’ of imprinting at this stage, or could somehow pre‐empt the forthcoming brain‐specific paternal expression.

The CGI2 region corresponds to the putative ICR of the Grb10 locus and displays chromatin features found at other ICR regions (Fournier et al, 2002; Yang et al, 2003; Delaval et al, 2007). Specifically, the DNA‐methylated maternal allele is associated with repressive marks (H3K9me3 and H4K20me3), whereas the unmethylated paternal allele is enriched for a permissive modification (namely H3K4me2). This constitutive chromatin pattern could be involved in the maintenance of the allelic DNA methylation observed at ICRs (Fournier et al, 2002; Ooi et al, 2007; Jia et al, 2007). Unlike these constitutive marks, H3K27me3 is enriched on the paternal unmethylated allele of the CGI2 region in all tissues but brain, suggesting that it could be involved in the control of the brain‐specific paternal expression of Grb10. This paternal enrichment was not restricted to CGI2 but was detected along the whole 5′ UTR. Interestingly, we observed a reciprocal enrichment for H3K27me1 that we detected on the maternal allele of the three CGI in different adult tissues, but brain (Supplementary Figure 3C, data not shown). H3K27me1 was shown in human T cells to be enriched downstream of active transcriptional start sites (Barski et al, 2007), and could thus be linked to the maternal expression of Grb10.

The most striking finding of our study is the colocalization of H3K27me3, a repressive mark, with the permissive mark H3K4me2 on the paternal allele of the putative ICR, constituting an allele‐specific bivalent chromatin domain. Bi‐allelic bivalent chromatin domains have been characterized for numerous genes in human and mouse ES cells, as well as in differentiated cells (Azuara et al, 2006; Bernstein et al, 2006; Roh et al, 2006; Pan et al, 2007; Mikkelsen et al, 2007; Zhao et al, 2007), but so far they have not been linked to imprinted gene expression. The Grb10 mono‐allelic bivalent chromatin domain is established at an early developmental stage and is maintained in all somatic tissues, but is absent in the brain. The mechanism involved in establishing and maintaining this peculiar chromatin configuration remains elusive. MLL and PRC2 histone methyltransferase complexes, involved in H3K4 and H3K27 methylation, respectively, could localize to the same nucleosomes or on different nucleosomes in close proximity to each other. A direct interaction or colocalization of homologues of these protein complexes has been described in plants and flies (Cavalli and Paro, 1998; Orlando et al, 1998; Klymenko and Müller, 2004; Saleh et al, 2007) and was suggested in mice (Xia et al, 2003). Such close proximity could account for the formation and maintenance of bivalent chromatin. Important information on the mechanisms involved could be obtained by investigating the allelic distribution of the components of these complexes at the putative ICR of Grb10.

Bivalent domains have been proposed to mark genes that are ‘poised’ for expression to occur in response to appropriate developmental cues (Azuara et al, 2006; Bernstein et al, 2006) and whose transcription is temporarily repressed by H3K27me3 (Boyer et al, 2006; Lee et al, 2006). At the Grb10 locus, the presence of bivalent chromatin is strictly correlated with paternal Grb10 repression. Furthermore, in neurospheres the bivalent domain is resolved upon neural commitment, in the same developmental window during which the paternal promoters become de‐repressed. Moreover, one of the paternal‐specific promoters is de‐repressed in ES cells deficient for EED, a component of the PRC2 complex. Nonetheless, removal of H3K27me3 on its own is probably not sufficient to trigger de‐repression. Indeed, analysis of Grb10XC302 conceptuses indicates that the absence of EED and the resulting loss of H3K27me3 fail to activate Grb10 paternal expression in the whole embryo. Similar results have been reported for Gata1, Eomes and Hoxa1, genes with bivalent domains that do not become de‐repressed in Eed−/− ES cells or following loss of H3K27me3 (Azuara et al, 2006; Jørgensen et al, 2006; Lee et al, 2007a). Analysis of Hoxa1 suggested that in addition to the loss of H3K27me3, increased histone acetylation could be important for gene reactivation (Lee et al, 2007a). Upon removal of H3K27me3, histone acetyltransferases could be recruited by specific transcription factors, or indirectly through H3K4 methylation (Pray‐Grant et al, 2005). If this pathway relies on tissue‐specific factors, de‐repression will be observed only in cells or tissues where such factors are expressed.

Our data lead to a model for the brain‐specific paternal expression of Grb10 from the CGI2 region (Figure 5). In non‐neural lineages, paternal expression is poised but silenced through the bivalent chromatin domain, whereas DNA methylation associated with repressive histone marks locks promoters on the maternal allele. In the developing neural lineage, the bivalent chromatin on the paternal allele is resolved, through the loss of H3K27me3. This leads to de‐repression of the promoters at CGI2. To account for the neuron‐specific paternal expression, the loss of the bivalent chromatin should occur only upon neural stem cell commitment towards neurons. However, the large decrease in H3K27me3 in neurospheres, when they are induced to differentiate, suggested that removal of this mark occurs in both glial and neuronal cells. Therefore, neuronal‐specific factors are probably required to translate the loss of H3K27me3 into active gene expression. Possibly, such factors could be involved in recruiting histone acetyltransferases, through H3K4me2 and/or by interacting with a putative neuronal‐specific enhancer.

Figure 5.

Model for the brain‐specific paternal expression of Grb10. From early developmental stages onwards, in all the non‐neural tissues, promoters at CGI2 are poised but silenced through the paternal allele‐specific bivalent chromatin. Upon neural lineage formation, the bivalent chromatin is resolved through the loss of H3K27me3. As a consequence, in neurons, the maintained H3K4me2 in association with neuronal‐specific factors allows transcriptional activation, possibly through recruitment of histone acetyltransferases. Repressive marks are in bold letters, permissive in italics. DNA methylation is represented by filled lollipops.

An intriguing aspect of our model is the removal of H3K27me3 from Grb10 specifically during lineage commitment in the developing brain. The loss of H3K27me3 in brain could be actively carried out by JMJD3 and the homologous UTX protein, which are functional H3K27me3 demethylases (Agger et al, 2007; De Santa et al, 2007; Lan et al, 2007; Lee et al, 2007b). A recent report by Jepsen et al (2007) suggests that JMJD3 has a key function in the differentiation of neural stem cells towards neurons. Removal of H3K27me3 from the Grb10 promoter upon commitment to the neural lineage could be part of a more general epigenetic programme that exerts an effect on several genes involved in neural development. Furthermore, it is tempting to speculate that the mechanism described here for Grb10 could apply to other imprinted genes as well. There is indeed increasing evidence that genomic imprinting influences brain function through effects on neuro‐developmental processes (Wilkinson et al, 2007). Among the roughly 80 imprinted genes characterized so far, many are expressed in the brain and some, such as Ubea3a and Commd1, show a brain‐specific imprinting pattern (Yamasaki et al, 2003; Wang et al, 2004; Kishino, 2006). The recently developed allelic large‐scale mapping of histone methylation (Mikkelsen et al, 2007) offers a promising tool to investigate whether bivalent chromatin could be a feature of other imprinted genes as well. If so, the biological significance of mono‐allelic bivalent chromatin in imprinted gene regulation, particularly in brain, could be assessed by analysis of cells and tissues deficient for Polycomb group proteins and specific histone demethylases. Beyond imprinting, our study represents one of the first examples of the implication of bivalent chromatin in tissue‐specific gene expression. We demonstrate for the Grb10 bivalent chromatin domain that it is faithfully maintained in somatic tissues and becomes resolved in a lineage‐specific manner, underscoring the potential role of bivalent chromatin in the developmental control of gene expression.

Materials and methods

Material collection

Material was obtained from reciprocal crosses between C57BL/6J (B6) with M. musculus molossinus JF1 mice, (B6xJF1) F1 and (JF1xB6)F1. Whole embryos, placentae and isolated tissues (brain, liver and kidney) were recovered at various developmental stages. Directly following dissection, tissues were frozen in liquid nitrogen and stored at −80°C.

ES cell lines

Allelic chromatin and expression analyses were performed on a (C57BL/6J × M. spretus)F1 ES cell line (SF1‐G). Cells were grown and maintained in an undifferentiated state as previously described (Dean et al, 1998). DNAs and RNAs from OS25 (WT) and G8.1 and B1.3 (Eed−/−) ES lines (Azuara et al, 2006) were a kind gift from Drs Helle Jorgensen, Veronique Azuara and Amanda Fisher (Lymphocyte Development Group, MRC, London, UK).

RNA extraction and allelic expression analysis

Total RNA extraction and first‐strand cDNA synthesis were performed as previously described (Arnaud et al, 2006). RT–PCR products from maternal and paternal alleles were distinguished by SSCPs, or by direct sequencing. Expression analyses were performed in ES and TS cells and in brain, liver and kidney obtained from 1‐ to 3‐day‐old and, 2‐, 4‐week, and 4‐ and 8‐month‐old animals. Similar results were obtained in (B6 × JF1)F1 and (JF1 × B6)F1 reciprocal crosses (not shown). Primers used and details on polymorphism are given in Supplementary Table 1. RNAs from glial and neuronal cells were a kind gift from Drs Takahiro Yamada, Layla Parker‐Katiraee, and Stephen W Scherer (The Hospital for Sick Children, Toronto, Ontario, Canada). RNAs were prepared by Dr Takahiro Yamada from the cerebral cortex of 15.5–16.5 d.p.c. (B6 × JF1)F1 embryos as described in Mnatzakanian et al (2004).

ChIP

ChIP on native chromatin was carried out as described in our protocol at http://www.epigenome‐noe.net/researchtools/protocol.php?protid=22. We used antisera against H3K9ac (06–942), H3K4me2 (07–030), H3K9me3 (07–442), H4K20me3 (07–463) and H3K27me1 (07–448) (Upstate Biotechnology) and H3K27ac (ab4729) (Abcam). For H3K27me3, we obtained similar results with two different antisera: 07‐449 from Upstate Biotechnology and Ab6002 from Abcam. As a negative control (mock precipitation), we used a rabbit antiserum against chicken IgG (Sigma; C2288). For each tissue, at least two independent chromatin immunoprecipitations were performed. The same allelic enrichments were observed in adult tissues obtained from (B6 × JF1)F1 and (JF1 × B6)F1 crosses (not shown). Sequential ChIP was conducted as described by Bernstein et al (2006), on cross‐linked chromatin obtained from semiconfluent MEF cells.

Analysis of immunoprecipitated chromatin

For each antiserum used, in the antibody‐bound and unbound fractions the parental alleles were distinguished by radioactive PCR amplification, performed in the presence of [α32P]dCTP (1% of total dCTP), followed by electrophoretic detection of SSCP polymorphisms. Relative intensities of the maternal and paternal bands were determined using the Aida 1D quantification software (Fujifilm). Primer sequences used and details on polymorphism are given in Supplementary Table 2. Precipitation levels were determined by real‐time PCR amplification, using a SYBR Green PCR Kit (Qiagen). Each PCR was run in triplicate and results are presented as a percentage of immunoprecipitation, calculated by dividing the average value of immunoprecipitated DNA by the average value of the corresponding input chromatin. For H3K27me3, precipitation levels were similar between two different antisera, 07‐449 from Upstate Biotechnology and Ab6002 from Abcam.

Neurosphere culture

Neurospheres were derived from spinal cord of 13.5 d.p.c. (B6 × JF1)F1 embryos and cultured with EGF and FGF2 growth factors as described (Deleyrolle et al, 2006). Neurospheres were passaged 4–5 times and then allowed to differentiate for 3 days. This was achieved by plating undissociated neurospheres at 300 000 cells/cm2 on poly‐d‐lysine‐coated (20 μg/ml) dishes in differentiation medium (neurobasal‐B27; Invitrogen) supplemented with 0.5% fetal calf serum, 10 ng/ml neurotrophin 3 (Peprotech) and 10 μg/ml gentamicin. Neurosphere differentiation into neurons and glial cells was checked by immunofluorescence using cell‐specific markers (GFAP, β‐3‐tubulin and O4) as described in Deleyrolle et al (2006). Accurate quantification of the number of astrocytic (GFAP), neuronal (β‐3‐tubulin) and oligodendrocytic (O4) cells was achieved in a separate experiment in which neurospheres were first dissociated with trypsin before differentiation and plated for 4 days without growth factor at a density of 250 000 cells/cm2 onto 20 μg/ml poly‐d‐lysine (Sigma)‐coated coverslips, as described (Deleyrolle et al, 2006). Around 200 cells were individually observed and counted at high magnification ( × 400) for expression of differentiated cell markers.

Grb10XC302 expression in WT and Eed−/− conceptuses

We used Eed17Rn5‐3354SB‐null 6.5 d.p.c. embryos, in which Eed deficiency results from a proline substitution (L196P) that leads to a null mutation (Rinchik and Carpenter, 1993). To assess allelic expression of Grb10, female mice heterozygous for the Eed17Rn5‐3354SB allele were mated with heterozygous males that also carried a gene trap insertion of a β‐geo cassette in intron 7 of the Grb10 gene. This Grb10gtTMag1 strain was generated from the XC302 ES cell line from BayGenomics, and is herein referred to as Grb10XC302. Gravid females were killed at 6.5 d.p.c. Embryos were dissected from the maternal uterus and decidual tissue, and a small tissue sample was used for genotyping. Embryos were fixed in a mixture of glutaraldehyde and formaldehyde and stained with X‐Gal, a substrate for galactosidase, as previously described (Morin‐Kensicki et al, 2001). Stained embryos were post‐fixed in 4% paraformaldehyde and cleared through a glycerol gradient before photographing. Blue staining indicates cells expressing the β‐geo gene trap. Eed genotyping was carried out as previously described (Kalantry and Magnuson, 2006). Grb10XC302 genotyping was performed using specific primers for the neomycin gene within the gene trap insertion. Primer sequences are available upon request.

Supplementary data

Supplementary data are available at The EMBO Journal Online (http://www.embojournal.org).

Supplementary Information

Supplementary Figures and Tables [emboj2008142-sup-0001.pdf]

Acknowledgements

We thank T Forné and E Julien for valuable advice and critical reading of the paper, A Fisher, H Jorgensen and V Azuara (for the generous gift of DNA and RNA from Eed(−/−) ES cell lines, and T Yamada, L Parker‐Katiraee and SW Scherer for the kind gift of RNA from glial and neuronal cells. We are grateful to M Pannetier for her precious help with ‘cross‐link’ ChIP and to K Chebli, C Jacquet and the Animal Facility for expert animal husbandry. This project was supported mainly by the ESF EUROCORES programme EuroSTELLS, the ARC programme ARECA and the Agence National de la Recherche (ANR, ‘EMPREINTE’) awarded to RF. Additional funding was derived from an NIH grant to TM, and European FP6 Rescue project and the Association Française contre les myopathie (AFM) grants awarded to J‐PH. We thank the European Framework‐6 network EPIGENOME for encouragement and support. SC is a recipient of an NIH fellowship. LAS, J‐CS and AH are recipients of a PhD fellowship from the French ministry of Education and Science.

References