Introduction

Epigenetic mechanisms play diverse roles in development and mediate the heritable repression or expression of genes in specific cell lineages. Most autosomal genes are expressed at comparable levels from each of the two copies. However, in different cell types, it has been observed that a few percent of genes are expressed from only one of the two alleles in individual cells and that this differential expression can be maintained through cell divisions. The silenced copy can be either the maternal or the paternal allele and the choice is made randomly during development (Gimelbrant et al. 2007; Singh et al. 2003). Mono-allelic repression of genes occurs also on the X chromosome, which in female cells is silenced on one of the two copies by a process called X-chromosome inactivation (Heard and Disteche 2006). The choice of which X chromosome will be inactivated after implantation is random in the embryo. In the extra-embryonic tissues, however, gene silencing occurs specifically on the paternally inherited X chromosome in the mouse (Takagi and Sasaki 1975).

Imprinted genes also show mono-allelic expression in somatic cells and tissues. This subset of genes is rather exceptional, however, because the allelic expression is entirely dependent on the parental origin of the gene, as it is for X inactivation in mouse extra-embryonic tissues. To date, about a hundred autosomal imprinted genes have been identified (Morison et al. 2005; Williamson et al. 2009). Functional studies have shown that imprinted genes play key roles in embryonic growth and development, whereas others influence behavior after birth (Smith et al. 2006; Wilkinson et al. 2007). Because of the important developmental roles of imprinted genes, and since they are expressed from only one of the two parental alleles, both parental genomes are required for mammalian development. Experimentally derived mouse conceptuses that have only maternal or paternal genomes do not develop beyond midgestation and show gross developmental abnormalities (Barton et al. 1984; McGrath and Solter 1984).

DNA methylation is essential for establishment and maintenance of imprinting. Methylation marks become established in the male and female germlines and bring about the mono-allelic expression of imprinted genes during development. Recent studies provide new insights into how these methylation “imprints” are established in the germ cells at “imprinting control regions” (ICRs) and how, after fertilization, they are maintained throughout development. The precise mechanism is poorly understood and besides the DNA methylation machinery, histone modifications and nonhistone proteins could contribute to this process. Although differential DNA methylation at ICRs is present in all somatic lineages of the embryo, at some domains it brings about imprinted expression in only specific tissues and cells. Our understanding of this intriguing developmental process has greatly improved during the last years, mostly through studies on two imprinted domains in the mouse, the IGF2-receptor (Igf2r) domain on proximal chromosome 17 and the Kcnq1 domain on distal chromosome 7. A major discovery was that long noncoding RNAs (ncRNAs) play essential roles at these and other imprinted loci. Furthermore, work from different laboratories has shown that histone methylation is involved in imprinted gene expression in specific tissues. These novel insights enhance our understanding of imprinting and of epigenetic regulation of gene expression in general. They also provide new tools to explore the mechanisms involved in the pathological deregulation of imprinted genes in epigenetic diseases and cancer. In this review we describe some of the best characterized imprinting clusters and discuss recent findings about the chromatin mechanisms that mediate allelic gene expression.

Imprinted gene clusters and ICRs

The mouse genome comprises 15 autosomal chromosomes on which imprinted domains were identified, largely through phenotypic studies on mice carrying uniparental disomies (Cattanach and Kirk 1985; Cattanach et al. 2006). Virtually all known imprinted genes in the mouse are localized in these chromosomal subregions (Williamson et al. 2009). Most imprinted genes are organized in gene clusters, which are similarly structured in mice and humans (Morison et al. 2005). Some imprinted gene clusters are small in size, less than 100 kb, whereas others comprise a megabase or more of DNA (Edwards and Ferguson-Smith 2007). The mouse has been the model of choice to explore imprinting, although many key insights have emerged from genetic studies in humans as well.

Key to the understanding of genomic imprinting was the discovery of sequence elements that are methylated on one of the two parental alleles only. These differentially methylated regions (DMRs) can be up to several kilobases in size. They are rich in CpG dinucleotides and many correspond to CpG islands. Importantly, the mono-allelic DNA methylation at several DMRs was found to originate from either the female or the male germline. Subsequent targeting studies in the mouse revealed that the regions with germline-derived DNA methylation are particularly important for imprinted expression. These essential regions have therefore been called imprinting control regions (ICRs) and are also referred to in the literature as imprinting centers or imprinting control elements (Wutz and Barlow 1998). Although all imprinting clusters comprise an ICR that controls the allelic expression at that domain, the DNA sequences of individual ICRs show little similarity with that of the others.

The regulation of DNA methylation at ICRs in somatic cells and during gametogenesis defines a developmental cycle of imprinting (Fig. 1). Before acquisition of the methylation imprint during oogenesis or spermatogenesis, the pre-existing methylation is erased in the primordial germ cells, probably by active DNA demethylation (Hajkova et al. 2002). The establishment of methylation imprints occurs later during gametogenesis and requires the de novo methyltransferase DNMT3A. At most ICRs, the imprint is established during oogenesis and all known maternal imprints are located at promoter regions. At only three ICRs in the mouse, the methylation imprint is established during spermatogenesis and all three concern intergenic regions. After fertilization, both the maternal and paternal methylation imprints are somatically maintained throughout development (Delaval and Feil 2004).

Fig. 1
figure 1

Developmental cycle of imprinting in mammals. Parental imprints are established during oogenesis or spermatogenesis at essential sequence elements called “imprinting control regions” (ICRs, black hexagons). DNA methylation (lollipops) is a consistent hallmark of these imprints. After fertilization, imprints are somatically maintained throughout development. In the primordial germ cells (PGCs) of the developing embryos, DNA methylation is erased to allow the subsequent establishment of new imprints. Two ICRs are shown, one on the paternal (dark color) and one on the maternal (light color) genome

Full size image

Factors involved in the acquisition of imprints

Methylation imprints are established in the germ cells according to the sex of the developing embryos. In male mouse embryos, methylation establishment in the germ cells initiates at 14.5-17.5 days post coitum (dpc), and the paternal ICRs are fully methylated several days after when final sperm cell differentiation ensues. Establishment of imprints at the maternally methylated ICRs in female germ cells, on the other hand, does not take place until later in postnatal life, during the final stages of oocyte growth. The de novo methyltransferase DNMT3A is required for the establishment of methylation imprints in both male and female germ cells (Kaneda et al. 2004). A related protein, DNMT3-like (DNMT3L), is also essential for imprint establishment (Bourc’his et al. 2001; Hata et al. 2002). It is thought that DNMT3L interacts with DNMT3A and stimulates its activity, most likely by recruiting DNMT3A to chromatin (Chedin et al. 2002; Jia et al. 2007). How the DNMT3A/DNMT3L complex becomes recruited to ICR sequences is not known. Several studies tested the hypothesis that close-by tandem repeat sequences could provide specificity to this process (Neumann and Barlow 1996). At the mouse Rasgrf1 locus, a tandem repeat sequence neighboring the ICR was shown to be essential for imprint establishment (Yoon et al. 2002). Interestingly, comparative sequence analyses have shown that the sequences of many ICRs comprise short imperfect repeats (Hutter et al. 2006; Paoloni-Giacobino et al. 2007).

Not all ICRs have imperfect tandem repeats, however, and such repeats are also detected at some nonimprinted CpG islands (Yamada et al. 2004). Transgenic experiments on the ICRs of the Igf2r and Kcnq1 domains indicate that tandem repeats, if present in multiple copies, contribute to imprint establishment as well as to somatic maintenance of methylation during early embryogenesis (Reinhart et al. 2006). Multicopy transgenes integrated into ectopic positions may not faithfully replicate imprinting mechanisms, however, and are not easy to interpret. Deletion of the repeated element at the Kcnq1 ICR was found to not affect imprint acquisition (Mancini-Dinardo et al. 2006). Also, repeats close to the Igf2-H19 ICR are not essential for imprint establishment (Lewis et al. 2004b; Reed et al. 2001). Rather, at the Igf2-H19 ICR many of the required sequences seem to be present in this approximately 3-kb differentially methylated region itself (Cranston et al. 2001). A 2.9-kb fragment of this region recapitulated part of the imprinting process when inserted in a single copy into the β-globin locus, with acquisition of DNA methylation occurring upon paternal transmission (Tanimoto et al. 2005). The same result was obtained for single-copy transgene insertions of this fragment into the α-fetoprotein locus and downstream of the H19 gene (Park et al. 2004).

Histone modifications potentially could influence the establishment of imprints in germ cells as well. DNMT3L, the protein thought to recruit DNMT3A to its target sequences, binds to histone H3 but not when this core histone is methylated on lysine-4 (H3K4 methylation) (Jia et al. 2007; Ooi et al. 2007). H3K4 methylation of chromatin, therefore, could prevent the acquisition of de novo DNA methylation by DNMT3A. Conversely, other histone modifications could facilitate the recruitment of DNMT3A to its targets and, here, symmetrical dimethylation of arginine-3 on histone H4 (H4R3me2s) is one candidate modulator. Using the β-globin locus as a model, it was shown recently that PRMT5-mediated H4R3me2s is recognized by DNMT3A through its PHD motif and thereby stimulates de novo DNA methylation at this locus (Zhao et al. 2009), but its role in vivo remains to be explored. From structural studies on the DNMT3A-DNMT3L complex (Jia et al. 2007), it follows that two CpG dinucleotides can be methylated at each binding site and methylation is most efficient when the CpGs are positioned 8-10 bases apart. This may explain why only certain DNA sequences are methylated by DNMT3a in germ cells. Indeed, at ICRs, CpG dinucleotides show a periodicity of about 8-10 bp, which is not a general feature of CpG islands (Jia et al. 2007).

Another indication that chromatin organization in germ cells could potentially play a role in imprint establishment comes from the Prader-Willi syndrome (PWS) imprinted domain on human chromosome 15q11. Its ICR, at the SNRPN gene, has a maternal methylation imprint. In some studies, this ICR was reported to become methylated after fertilization of the oocyte only (El-Maarri et al. 2001; Geuns et al. 2003; Kantor et al. 2004). In case this is a true finding, it indicates that the de novo DNA methylation machinery somehow detects the right parental allele in the early embryo. Based on recent studies using a transgenic mouse model, it has been proposed that this could involve a maternally derived chromatin signature (Kaufman et al. 2009). In the mouse, the Snrpn ICR acquires its methylation imprint earlier during the maturation of the oocyte, and both DNMT3A and DNMT3L are essential in this process. However, when DNMT3L is absent during oogenesis (in Dnmt3L-/- females), the Snrpn methylation imprint is nonetheless present in some of the offspring (Arnaud et al. 2006; Henckel et al. 2009). This observation suggests that there could be a predefined chromatin state also in the mouse which is inherited from the oocyte and which is able to attract DNA methylation during early development.

Support for the idea that specific chromatin features could be inherited from the germline independently of DNA methylation comes from the different transgenic studies on the H19-Igf2 ICR. When inserted as a single copy at specific exogenous positions in the genome (Park et al. 2004; Tanimoto et al. 2005), acquisition of paternal DNA methylation at the ICR did not happen during spermatogenesis but only after fertilization, during early development.

Could ICRs indeed present a specific chromatin signature preceding the establishment of imprints during gametogenesis? Although micromethodologies have been developed to perform chromatin immunoprecipitation (ChIP) on small batches of cells, it remains difficult to study chromatin in oocytes. For the male germline, the question is easier to address given the higher number of cells that can be obtained when the paternal imprints become established (i.e., at 14.5–17.5 dpc). In a preliminary study on early spermatogonia, high levels of H3K4 methylation were detected at maternal ICRs, supporting the hypothesis that this mark could protect against DNA methylation (Delaval et al. 2007). Besides the idea that ICRs could acquire a specific chromatin signature during early spermatogenesis, the question of whether these key regions are subject to the global histone-to-protamine exchange that happens during spermatogenesis is also crucial. In human sperm, more than a few percent of the genome is estimated to be associated with histone proteins rather than protamines (Bench et al. 1996; Gatewood et al. 1990). A recent study shows that this fraction includes developmental genes and imprinted gene domains (Hammoud et al. 2009). Significantly, large regions comprising ICRs are found to be nucleosome-enriched. Whereas paternally methylated ICRs show a lack of H3K4me3, several ICR regions coding for ncRNAs (Kcnq1ot1, Airn) are enriched in H3K4me3, suggesting that their chromatin is primed to facilitate transcription in the early embryo (Hammoud et al. 2009). In the mouse, less than 1% of the genome is associated with histones in mature sperm (Bench et al. 1996). It is important to determine whether also in this species this includes imprinted gene domains.

Regions that remain nucleosomally organized in spermatozoa could potentially transmit paternal epigenetic information to the zygote (Boussouar et al. 2008). To test this hypothesis, however, it is important to know whether the remaining nucleosomes are removed directly after fertilization, like protamines, or whether they contribute to the zygotic paternal chromatin. Recent work shows that at least some of the histone proteins remain associated with the paternal genome after fertilization. Specifically, core histones H3.1 and H3.2 were shown to be present in the male pronucleus directly after fertilization of the oocyte and to remain associated with the DNA until replication occurs (van der Heijden et al. 2008). Conversely, histone H2A variants, which are sperm-specific and are incorporated into heterochromatic regions, are lost from the paternal genome right after fertilization (Wu et al. 2008).

Nonhistone proteins contribute to the specificity of imprint establishment as well. A recent conditional targeting study in the mouse showed that the zinc finger protein ZFP57 contributes to imprint establishment at the Snrpn ICR in oocytes (Li et al. 2008). Furthermore, a screen for genetic mutations in familial cases of the imprinting disorder Beckwith-Wiedemann syndrome (BWS) led to discovery that the gene NLRP2 (NALP2) contributes to the establishment of maternal methylation imprints at several ICRs (Meyer et al. 2009). Genetic mutations in a related Nucleotide-binding oligomerization domain, Leucine rich Repeat and Pyrin domain gene (NLRP7/NALP7) had been associated previously with familial complete hydatidiform moles, which present loss of methylation at maternal ICRs (Murdoch et al. 2006). The precise mechanistic role of these nuclear proteins is still unclear.

Imprint establishment and transcription

Transcription could potentially play a role in the acquisition and/or maintenance of maternal imprints. Maternally methylated ICRs comprise promoters, some of which transcribe long ncRNAs involved in chromatin repression (see below). Many of these transcripts overlap in the antisense orientation with other transcripts at the imprinted locus in which they are located (Peters and Robson 2008). Therefore, double-stranded RNA molecules (dsRNAs) that might affect gene activity via the formation of small interfering RNAs (siRNAs) could be formed. In fission yeast, siRNAs, which are generated via the action of the Dicer protein, induce the formation of the RITS complex, which is required for heterochromatin silencing (Noma et al. 2004). Although some studies in mammals have suggested that artificially produced siRNAs can induce DNA methylation at promoters (Hawkins et al. 2009; Morris et al. 2004), endogenous genes of such a transcriptional silencing mechanism have not been identified. The DNA methylation at retrotransposon elements in mammals, however, is controlled by small RNAs during gametogenesis (Aravin and Bourc’his, 2008). To explore whether gene silencing through RNA interference (RNAi) could be involved in the control of imprinting, several groups have studied mouse mutants with reduced expression or absence of Dicer. Their findings suggest that Dicer and RNA interference are not involved in the control of imprinted genes in somatic cells (Fukasawa et al. 2006; Morita et al. 2007; Redrup et al. 2009). RNAi mechanisms could, however, still be involved in the establishment of imprints in germ cells. In that case, one would expect to find small RNAs that specifically overlap with the DNA sequences of ICRs at the time of imprint establishment. Such specific RNAs were not found in a systematic characterization of small RNAs purified from mature and metaphase II oocytes (Watanabe et al. 2006).

Evidence for the involvement of transcription in imprint establishment has come from the Gnas locus on mouse distal chromosome 2 (Fig. 2). A recent study (Chotalia et al. 2009) has shown that transcription across the ICRs of the imprinted Gnas domain is important for the establishment of the maternal methylation imprints at this locus. The imprinted Gnas domain spans about 80 kb and comprises multiple promoters that produce transcripts that encode signaling proteins involved in growth, differentiation, and homeostasis. The locus produces several imprinted noncoding transcripts as well. Two CpG islands in the domain mediate the allelic expression of the different transcripts and are both methylated on the maternal allele (Coombes et al. 2003; Liu et al. 2005; Williamson et al. 2004, 2006). Both these ICRs acquire their allelic DNA methylation during oogenesis. The locus’s furthest upstream promoter is the one of a gene called Nesp (Fig. 2). This transcription unit covers the entire domain. Chotalia et al. (2009) found that truncation of this long transcript disrupted the oocyte-specific acquisition of methylation at the two ICRs in the central portion of the domain. Thus, transcription across these CpG islands in oocytes seems essential for the establishment of their maternal DNA methylation imprints in oocytes. This interesting finding extends an earlier study in humans in which a deletion in NESP resulted in loss of methylation at the two ICRs of the GNAS locus (Bastepe et al. 2005). Whether this newly discovered mechanism is relevant to other imprinted domains is unknown, but the authors did detect transcription through several other maternal ICRs in oocytes.

Fig. 2
figure 2

Imprint establishment and transcription. At the Gnas locus, transcription from the Nesp promoter is required for the establishment of the DNA methylation imprints (lollipops) at the two ICRs of this imprinted domain during oogenesis (Bastepe et al. 2005; Chotalia et al. 2009). After fertilization of the oocyte, the maternal imprints mediate imprinted gene expression at the locus. Interestingly, each of the two ICRs produces a noncoding RNA on the paternal chromosome. One of these is antisense to the Nesp gene (Nespas ncRNA) and its expression could be essential to silence Nesp on the paternal chromosome (Williamson et al. 2006)

Full size image

It seems too early to conceptually link the involvement of transcription and the role of chromatin in imprint establishment in female germ cells. One possibility, however, could be that transcription through (regions including) CpG islands brings about a chromatin devoid of H3K4 methylation, which would facilitate recruitment of the DNMT3A-DNMT3L complex and acquisition of de novo methylation, if there is an appropriate spacing of CpG dinucleotides. Relative to such a scenario, it is interesting to note that in somatic cells H3K4 methylation is confined to promoter regions and is not found in the transcribed body of genes, which are enriched in H3 lysine-36 methylation instead (Bernstein et al. 2006; Mikkelsen et al. 2007). In human cells and in plants, DNA methylation is found in the bodies of highly expressed genes, indicating that transcription could indeed mediate DNA methylation (Ball et al. 2009; Zilberman et al. 2008). Chromatin- and transcription-mediated mechanisms may thus be linked together in the establishment of imprints in female germ cells. It should be interesting to explore whether similar mechanisms exist in the male germline.

Somatic maintenance of imprints through development

Proper somatic maintenance of imprints is essential for normal development. Perturbation of this process is causally involved in different human diseases and is thought to be an early contributing factor in cancer as well (Delaval et al. 2006). It remains poorly understood how allele-specific methylation at ICR regions is maintained throughout development. Remarkably, on the one hand, ICRs are fully resistant to the genome-wide waves of DNA demethylation that occur following fertilization and during the preimplantation stage (Reik et al. 2001). On the other hand, they are also protected against the widespread acquisition of de novo methylation that occurs after implantation of the embryo (Lepikhov et al. 2008; Mayer et al. 2000). The latter is not unique to ICRs, however, and is observed at the majority of nonimprinted CpG islands as well (Weber et al. 2007).

Continuous expression of DNMT1 is essential for the maintenance of methylation imprints; its level of expression also is critical (Biniszkiewicz et al. 2002). Conditional targeting studies have shown that during preimplantation, maternal and zygotic DNMT1 are sufficient to maintain methylation imprints (Hirasawa et al. 2008). The de novo methyltransferases DNMT3A and DNMT3B are dispensable for this process in vivo.

We have only just started to comprehend the intricate links between DNA methylation, histone methylation, and nonhistone proteins and how this cross-talk influences the somatic maintenance of patterns of CpG methylation in mammals (Cedar and Bergman 2009). Different groups have explored the question of whether chromatin features such as histone methylation could contribute to the maintenance of DNA methylation at ICRs (Delaval et al. 2006; reviewed in Feil and Khosla 1999). On the methylated allele of ICRs, chromatin is enriched in trimethylation at lysine-9 of histone H3 (H3K9me3) and at lysine-20 of histone H4 (H4K20me3), has symmetrical dimethylation on arginine-3 of histone H4/H2A, and is devoid of H3 and H4 acetylation (Delaval et al. 2007; Fournier et al. 2002; Henckel et al. 2009; Jelinic et al. 2006; Mikkelsen et al. 2007; Regha et al. 2007; Verona et al. 2008; Yang et al. 2003). It is also associated with heterochromatin protein 1 gamma (HP1γ) (Pannetier et al. 2008; Regha et al. 2007). These findings reveal interesting similarities between the methylated alleles of ICRs and pericentric heterochromatin (Fig. 3), where constitutive maintenance of DNA methylation is also observed (Lehnertz et al. 2003). Recent work on embryos in which maternal methylation imprints were not present shows that the occurrence of repressive histone methylation is tightly linked to the presence of the DNA methylation imprints (Henckel et al. 2009). H3K9me3 should be particularly interesting to explore in this context given its apparent contribution to the somatic maintenance of DNA methylation at pericentric heterochromatin (Lehnertz et al. 2003). It remains unclear, however, which histone methyltransferase (HMT) regulates H3K9me3 at ICRs. Unlike at pericentric heterochromatin, SUV39H1 and -H2 (called KMT1A and KMT1B in a new nomenclature; Allis et al. 2007) are not involved (Pannetier et al. 2008; Regha et al. 2007). Neither seems to be the HMTG9A (KMT1C) (Wagschal et al. 2008). One of the remaining candidates is SETDB1 (also called ESET, and KMT1E in the new nomenclature), a HMT which plays essential roles in cellular proliferation and development (Dodge et al. 2004). H4K20me3 at ICRs is regulated by SUV4-20H1 and -H2 (KMT5B and KMT5C), like at heterochromatin (Pannetier et al. 2008).

Fig. 3
figure 3

Differential chromatin organization at ICRs. In somatic cells, chromatin on the DNA-methylated allele is hypoacetylated on H3 and H4, is enriched in H3K9me3, H4K40me3, and H4/H2AR3me2s, and is bound by HP1. On the opposite parental allele, there is absence of DNA methylation, H3 and H4 acetylation, and enrichment in H3K4me2/3

Full size image

On the DNA-methylated allele, ICRs are devoid of H3K4 methylation (Delaval et al. 2007; Fournier et al. 2002; Gregory et al. 2001). In vitro, H3K4 methylation prevents the acquisition of DNA methylation (Ooi et al. 2007). Absence of this histone mark, therefore, could contribute to keeping the DNA methylated. Chromatin on the parental allele without DNA methylation, in contrast, has high levels of H3K4me2 in somatic cells (Delaval et al. 2007; Pannetier et al. 2008). It seems plausible that its continuous presence prevents aberrant acquisition of CpG methylation in vivo, particularly during early development (Ooi et al. 2007). All the maternally methylated ICRs comprise promoter sequences. Promoter activity is thought to promote H3K4 methylation, particularly H3K4me3, and to prevent DNA methylation (Bernstein et al. 2006; Mikkelsen et al. 2007; Verona et al. 2008). In agreement with this general idea, it was found that at the Igf2r ICR, a strong promoter activity is required to maintain the paternal allele unmethylated throughout development (Stricker et al. 2008).

There is growing evidence for the involvement of nonhistone proteins in the maintenance of genomic imprinting. For instance, the zinc finger protein ZFP57 contributes to the embryonic maintenance of maternal imprints. Mouse embryos lacking ZFP57 show loss of methylation at the Snrpn ICR and partial losses of methylation at several other ICRs (Li et al. 2008). Evidence for ZFP57 involvement in humans came from studies on patients with transient neonatal diabetes mellitus (TNDM) (Mackay et al. 2008). In about one fifth of these patients the disease is caused by loss of DNA methylation at the (putative) ICR controlling the transcription factor gene ZAC/PLAGL1 on chromosome 6q24. Intriguingly, this specific subgroup of patients showed methylation losses at other ICRs as well. The ZFP57 locus showed homozygosity only in the patients, not in unaffected family members, and the identification of specific genetic mutations confirmed its involvement in TNDM. However, there are TNDM patients with multiple methylation defects who do not show mutations in the ZFP57 gene. Through its KRüppel-Associated Box (KRAB) domain, ZFP57 potentially recruits the KAP1 (KRAB-associated protein 1, also called TIF1B) corepressor complex. This repressive protein complex comprises histone deacetylases and histone methyltransferases that could contribute to the maintenance of DNA methylation. Preliminary support for this hypothesis comes from a recent study on human cells in which disruption of KAP1 led to loss of DNA methylation and reduced H3K9me3 at a putative ICR region (Riclet et al. 2009). The early stages of development are critical for the maintenance of imprints and, besides ZFP57, other factors could be involved. One other key protein is PGC7/STELLA, a maternal factor that is essential for early development and protects against loss of DNA methylation at several imprinted loci. Precisely how this protein mediates its effects during the very early stages of development remains to be discovered (Nakamura et al. 2007).

It seems likely that yet other nuclear proteins contribute to the somatic maintenance of imprints at ICRs. Pocket protein(s) of the retinoblastoma RB1 family, for instance, could be good candidates since they interact with other proteins that repress chromatin, including SUV4-20H1/H2 and DNMT1 (Gonzalo and Blasco 2005). RB deficiency led to a strong reduction of H4K20me3 and decreased DNA methylation at pericentric heterochromatin (Gonzalo et al. 2005). Interestingly, two RB-binding proteins (RBBP1 and RBBP1-like1) were recently found to be important for the maintenance not only of H4K20me3 and H3K9me3, but also of DNA methylation at the Snrpn ICR (Wu et al. 2006). This finding suggests that DNA methylation, histone methylation, and nonhistone proteins are intricately linked, at least at this imprinting control region.

The balance between mechanisms that favor and maintain DNA methylation on the methylated allele and those that protect the opposite parental allele against DNA methylation has also to be taken into consideration. Besides the involvement of H3K4 methylation, which protects against the acquisition of de novo methylation in vitro (Ooi et al. 2007), there are only a few hints. One concerns CTCF, a zinc finger domain protein that binds to several ICRs, including that of the Igf2/H19 imprinted domain. CTCF binding is methylation-sensitive and, as a consequence, at the Igf2/H19 ICR it is detected on the unmethylated maternal allele only. A targeting experiment in the mouse provided evidence that binding of CTCF to the Igf2/H19 ICR is involved in protecting against DNA methylation (Engel et al. 2006). Deletion of CTCF binding sites on the maternal allele led to aberrant acquisition of DNA methylation in post-implantation embryos on this normally unmethylated parental allele. This indicates that in the absence of CTCF binding, the ICR region is susceptible to ectopic acquisition of methylation during the wave of de novo methylation that occurs after embryo implantation. Several other ICRs comprise canonical CTCF binding sites, but it is not known whether CTCF binding could be involved in protecting against DNA methylation at these sites.

Long noncoding RNAs mediate allelic silencing at imprinted domains

In silico analyses have suggested that as much as half of the mammalian genome could be transcribed. Many putative transcription units overlap with transcripts from the opposite strand and the majority of these are noncoding (Katayama et al. 2005; Brosnan and Voinnet 2009). Chromatin at transcribed regions of the genome is marked by H3 lysine-36 methylation (Mikkelsen et al. 2007). This unique signature was used recently to map long ncRNAs located between protein-encoding genes in the mouse. Some 1600 intervening long ncRNAs were identified in different cell types (Guttman et al. 2009). Although most long ncRNAs are evolutionarily conserved, their roles in development remain largely unknown. The situation is different for the long ncRNAs at imprinted chromosomal domains. Several of these have been characterized in great detail and have been shown to play essential roles in the control of imprinted gene expression (Peters and Robson 2008; Umlauf et al. 2008).

The Igf2r domain on mouse chromosome 17 was the first locus where expression of a long ncRNA was demonstrated to be essential (Sleutels et al. 2002). Igf2r is expressed from the maternal allele and comprises an ICR in its second intron which presents maternally derived DNA methylation. Deletion of this ICR on the paternal chromosome led to disruption of imprinted expression of Igf2r and flanking genes (Zwart et al. 2001). On the paternal chromosome, transcription initiates from the ICR and is in the antisense orientation compared to the Igf2r transcript. This noncoding antisense RNA is very large, going across the Igf2r promoter, through an intergenic region, and up to the neighboring gene (Lyle et al. 2000). This noncoding RNA was called Airn (previously, Air) and is nonspliced and escapes export from the nucleus (Seidl et al. 2006). It is essential for the paternal silencing of Igf2r and that of two close genes, Slc22a2 and Slc22a3, both expressed and imprinted in the placenta (Sleutels et al. 2002). Although Airn transcription clearly mediates differential gene expression from the maternal allele, imprinted expression of Igf2r seems not to be regulated by Airn itself but possibly via a transcriptional interference mechanism (Pauler et al. 2007). At the neighboring Slc22a2 and Slc22a3 genes, in contrast, Airn seems important for their imprinted expression in the placenta, in which the ncRNA was shown to accumulate on the chromatin in cis (Nagano et al. 2008; Terranova et al. 2008).

The Kcnq1 domain on distal chromosome 7 is structurally similar to the Igf2r locus. An intronic ICR of the Kcnq1 domain has maternal DNA methylation and produces a long ncRNA (called Kcnq1ot1, or Lit1) from the unmethylated paternal allele (Fig. 4). This ICR, called KvDMR1, is essential for silencing genes on the paternal allele of this 900-kb domain (Fitzpatrick et al. 2002). Targeted experiments in which Kcnq1ot1 was truncated by introduction of a polyadenylation site established that Kcnq1ot1 is essential for gene silencing at the domain (Mancini-Dinardo et al. 2006; Shin et al. 2008). Imprinting at the domain is most extensive in the placenta, where it concerns more than ten genes (Lewis et al. 2004a; Umlauf et al. 2004). In the embryo itself, only the genes located at the central portion of the domain are subject to imprinting, including the Kcnq1 gene itself. These tissue-specific differences raise the question of whether the KvDMR1 ICR mediates imprinted expression differently in the central portion compared with the proximal and distal parts of the domain (Shin et al. 2008). In the placenta, chromatin repression at the proximal and distal genes requires the Kcnq1ot1 ncRNA (Pandey et al. 2008). At the central portion of the domain the ICR could lead to imprinted expression through a boundary function as well (Fitzpatrick et al. 2007; Kanduri et al. 2002; Shin et al. 2008). Imprinting of the central genes is evolutionarily conserved in humans, whereas the distal and proximal genes are not imprinted in human placenta (Monk et al. 2006).

Fig. 4
figure 4

Long noncoding RNAs mediate chromatin silencing. At several imprinted domains, long noncoding RNAs (ncRNAs) mediate gene repression in cis. Shown is the Kcnq1 domain in the placenta. The ICR is located in one of the introns of the Kcnq1 gene. It comprises the promoter of an ncRNA called Kcnq1ot1, which is expressed from the paternal allele from early development onward. In the trophoblast, the ncRNA accumulates close to the locus and induces silencing of genes in cis. At the proximal and distal parts of the domain, silencing involves recruitment of the PRC2 complex and of the HMT G9A, which bring about H3 lysine-27 trimethylation (H3K27me3) and H3 lysine-9 di- and trimethylation (H3K9me2/3), respectively

Full size image

Several other imprinted domains express long ncRNAs, including the Gnas locus on chromosome 2 described above (Fig. 2). Transcription of one of its ncRNAs, the Nesp antisense (Nespas) (Williamson et al. 2006) seems important for repression of Nesp on the paternal allele (Fig. 2). Whether besides DNA methylation this involves repressive histone modifications is yet to be determined (Williamson et al. 2006).

Histone lysine methylation in tissue-specific imprinting

Recent studies in the mouse show that H3 methylation is important for restricted imprinted gene expression in specific tissues. At the Kcnq1 domain, placenta-specific repression on the paternal chromosome involves histone H3 lysine-9 dimethylation (H3K9me2) and lysine-27 trimethylation (H3K27me3) (Pandey et al. 2008; Umlauf et al. 2004). In mouse conceptuses deficient for EED, a component of the Polycomb repressive complex PRC2, there is partial derepression of Cdkn1c and Ascl2, two genes of the Kcnq1 domain (Mager et al. 2003). PRC2 complexes mediate H3K27me3 at the domain and contribute to the allelic silencing, at least at some of the genes. PRC2 proteins were found to be physically associated with the silenced allele at Ascl2 and other genes of the domain (Terranova et al. 2008; Umlauf et al. 2004). PRC2 recruitment to the domain requires expression of the long ncRNA Kcnq1ot1 (Pandey et al. 2008; Terranova et al. 2008). Precisely how the Kcnq1ot1 ncRNA is locally retained (Redrup et al. 2009) and recruits Polycomb complexes to the chromatin is yet unclear. The H3 lysine-9 dimethylation on the silenced paternal chromosome is controlled in part by the HMT G9A. In the absence of G9A, placenta-specific imprinting is perturbed and several of the genes along the domain show biallelic expression (Wagschal et al. 2008). Thus, besides the H3K27me3, the H3K9me2 contributes to the ncRNA-mediated silencing as well. Recent work indicates that the repressed chromatin along the Kcnq1 domain is marked by PRC1-mediated histone H2A lysine-119 ubiquitination as well (Terranova et al. 2008), and other modifications could also contribute to chromatin silencing. DNA methylation, however, is not involved in the allelic repression of the placental genes. There is no detectable DNA methylation at the promoters of these genes and allelic gene repression is maintained in the absence of DNMT1 (Lewis et al. 2004a; Tanaka et al. 1999).

In the trophoblast, imprinting regulation at the Igf2r locus could be similar to that of the Kcnq1ot1 domain. In trophoblast cells, Airn accumulation is observed physically close to the promoter of Slc22a3, at a developmental stage at which this gene is imprinted (Nagano et al. 2008). G9A binding was detected on the Slc22a3 repressed paternal allele but not in conceptuses that expressed a truncated form of the Airn ncRNA. Recruitment of G9A to the chromatin is functionally important since in G9A-deficient placentas Slc22a3 is no longer imprinted and shows biallelic expression (Nagano et al. 2008). These findings suggest a model in which allelic expression of ncRNA recruits HMTs to specific genes in the imprinted domain, leading to the induction of repressive chromatin and gene silencing. A similar chromatin-mediated mechanism could be active at other imprinted domains as well (Umlauf et al. 2008; Wagschal and Feil 2006). For instance, a recent study of mouse proximal chromosome 6 (and the syntenic chromosomal region in humans) identified a placentally imprinted gene, Tfpi2, that requires both G9A and EED for its allelic silencing (Monk et al. 2008). It is, however, unknown whether an ncRNA is involved in locally recruiting histone methyltransferase complexes to the chromatin also at this imprinted domain.

H3K27me3 plays a role also in the imprinted expression of the Grb10 gene on proximal chromosome 11 (Sanz et al. 2008). This gene encodes a growth factor receptor binding protein important for fetal growth. It comprises an intragenic ICR which is DNA-methylated on the maternal allele. The promoters that drive Grb10 expression in the brain colocalize with the ICR region and are active on the paternal allele only. They are fully repressed in other tissues and this is linked to the presence of H3K27me3. Unexpectedly, however, the paternal allele of the ICR region is also enriched in H3K4 methylation in the very same tissues. This type of chromatin, marked by both H3K4 methylation and H3K27me3, has been named bivalent chromatin and is detected at many developmentally regulated genes. Whereas H3K27me3 is proposed to keep chromatin repressed during early development, the presence of H3K4 methylation allows gene expression in tissues later in development when it is required (Azuara et al. 2006; Bernstein 2006). The bivalent chromatin at Grb10 is indeed developmentally regulated. H3K27me3 is detected in embryonic cells and tissues but not in adult brain. A gradual loss of H3K27me3 (but not of H3K4 methylation) occurs upon differentiation of neuronal progenitor cells into neurons and glial cells. In embryos and ES cells deficient in the PRC2 protein EED, ectopic expression from the brain-specific promoters is detected, suggesting that H3K27me3 at the ICR represses these promoters during development until neuronal differentiation occurs (Sanz et al. 2008). The human GRB10 gene also shows imprinted paternal gene expression in the brain and, as in the mouse, this is linked to the presence of bivalent chromatin (Monk et al. 2009). Several other domains show imprinted gene expression specifically in the brain. At one of these, the Rasgrf1 locus on mouse chromosome 11, the copresence of H3K4 and H3K27 methylation was detected on the unmethylated maternal allele in ES cells and liver (Delaval et al. 2007; Lindroth et al. 2008), but it is not known whether these histone methylations play a similar role as at the Grb10 domain.

Conclusions

In this review, we discussed some of the recent research progress on the epigenetic control of allelic gene expression. Many discoveries were made, particularly on chromatin modifications, nuclear proteins, and noncoding RNAs, that further unravel the complex intricacies of genomic imprinting and contribute to our understanding of epigenetic regulation in general. New ideas have emerged about the mechanisms involved in the establishment of imprints in female and male germ cells. The recent work on ZFP57 indicates that at least at the Snrpn ICR, this zinc finger protein contributes to imprint establishment in the oocyte. Several NALP proteins are also candidates for involvement in imprint establishment. Furthermore, specific histone modifications could also facilitate the establishment of DNA methylation imprints. Recent in vitro work on DNMT3A/DNMT3L indicates that this de novo DNA methylation machinery cannot access genomic DNA in the presence of H3K4 methylation. Does H3K4 methylation need to be removed by a lysine demethylase before DNA methylation can become established in vivo? It remains to be confirmed in vivo whether this covalent histone mark contributes to the choice of whether an ICR becomes methylated in one germline but not in the other. Together with a putative role of transcription through the ICR in female germ cells (Chotalia et al. 2009), these mechanisms provide new avenues for understanding of how diverse pathologies may be linked to perturbed imprint establishment and maintenance. For instance, in sperm of men with oligozoospermia, a condition characterized by strongly reduced sperm counts and often associated with infertility, abnormal DNA methylation patterns have been observed. Intriguingly, the DNA methylation changes appeared to be specific to imprinting control regions. This novel finding suggests the possibility that defects in spermatogenesis could be mechanistically linked to imprint establishment and maintenance in male germ cells (reviewed in Filipponi and Feil 2009).

DNMT1 is the DNA methyltransferase involved in maintenance of methylation imprints (Hirasawa et al. 2008). However, nonhistone proteins contribute to this process as well. Furthermore, in somatic cells, the parental alleles of ICRs are associated with specific patterns of histone methylation and these could also contribute to the faithful maintenance of the differential DNA methylation at the ICR regions. The challenge for the future will be to explore in an integrated manner the different mechanisms involved and to determine the relative importance of each of these.

A third theme for which many questions remain concerns how ICRs bring about imprinted gene expression during development, a process that often occurs in a tissue-specific manner. How do these regions mediate chromatin repression? At some loci, this occurs across a megabase or more of DNA and involves repressive histone methylation, brought about by the expression of ncRNAs. Another mechanism, observed at several ICRs, is the formation of an allele-specific chromatin insulator that prevents interaction between distantly positioned regulatory sequence elements (reviewed in Edwards and Ferguson-Smith 2007; Ideraabdullah et al. 2008). It remains largely unknown what brings about the specificity of these “reading” mechanisms and why they act tissue-specifically and, in some cases, in a developmental stage-specific manner. Future answers may come from large-scale chromatin studies that include imprinted domains. One recent genome-wide ChIP study, for instance, confirmed that the Kcnq1 domain is enriched in H3 lysine-9 dimethylation in placenta but not in liver (Wen et al. 2009). Such epigenomic studies highlight similarities between different imprinted domains. There are exciting times ahead, with novel insights on imprinting and other epigenetic mechanisms instigating genome-wide studies and, inversely, epigenomic explorations telling us about the control of genes and chromosomal domains.