Genomic Imprinting and Epigenetic Control of Development

Abstract

Epigenetic mechanisms are extensively utilized during mammalian development. Specific patterns of gene expression are established during cell fate decisions, maintained as differentiation progresses, and often augmented as more specialized cell types are required. Much of what is known about these mechanisms comes from the study of two distinct epigenetic phenomena: genomic imprinting and X-chromosome inactivation. In the case of genomic imprinting, alleles are expressed in a parent-of-origin-dependent manner, whereas X-chromosome inactivation in females requires that only one X chromosome is active in each somatic nucleus. As model systems for epigenetic regulation, genomic imprinting and X-chromosome inactivation have identified and elucidated the numerous regulatory mechanisms that function throughout the genome during development.

Genomic imprinting (e.g., at the Igf2/H19 locus) and X-chromosome inactivation have illuminated regulatory mechanisms that function throughout the genome during development.

1. INTRODUCTION

A striking example of epigenetic mechanisms in mammalian development comes from landmark studies on the developmental potential of uniparental mouse embryos (Mann and Lovell-Badge 1984; McGrath and Solter 1984; Surani et al. 1984). Zygotes reconstituted with either two maternal or two paternal pronuclei (gynogenotes and androgenotes, respectively) fail to develop beyond midgestation, suggesting that additional layers of information beyond DNA sequence alone are transmitted to the offspring. The functional inequivalence of parental genomes can be attributed to a limited subset of genes whose allele-specific transcriptional activity is dependent on the parent of origin. These genes undergo a process termed genomic imprinting, in which differential parent-specific epigenetic modifications are established during gametogenesis, transmitted to the zygote, and stably maintained throughout somatic development and in tissues of the adult offspring. Conversely, parental genomic imprints are erased in gametes of the offspring, where new epigenetic marks are established, dependent on the sex of the offspring. As a paradigm for epigenetic gene regulation, studies of imprinted loci have been instrumental in identifying the mechanisms and gene products required for establishing, maintaining, interpreting, and erasing epigenetic information. We now know that the factors involved in the regulation of imprinted loci function throughout the genome, and have essential roles in normal development. Moreover, the misregulation of these mechanisms has been implicated in numerous human diseases, further underscoring their importance in mammalian biology.

2. PARADIGMS OF EPIGENETIC REGULATION

2.1. Characteristics of Autosomal Imprinted Genes

Nearly 100 genes displaying imprinted expression patterns in the majority of somatic tissues have been identified in the mouse, most of which also show imprinted expression in humans. In addition, many more genes are imprinted in a tissue-specific manner, for example in certain cell types of the placenta (Hudson et al. 2010, 2011) and brain (Gregg et al. 2010a,b). Studies of imprinted genes have identified several common characteristics. First, most imprinted genes are found in clusters, often spanning hundreds of kilobases. Second, almost all imprinted loci contain at least one noncoding RNA transcript. Third, the alleles of imprinted genes are differentially associated with covalent modifications to DNA and histones, and display significant differences in chromatin structure. Of these, genomic regions of differential DNA methylation, called differentially methylated domains (DMDs) or differentially methylated regions (DMRs), have been identified at almost all imprinted gene clusters and appear to play a central role in defining parental identity. Finally, cis-acting sequences, termed imprinting control regions (ICRs), have been identified and are required to both establish parental identity of imprinted loci and maintain differential epigenetic patterns throughout development. Many ICRs have been experimentally defined and often overlap with DMRs. Despite these similarities, imprinted loci use a diverse number of mechanisms to establish and maintain their characteristic expression patterns.

2.2. X-Chromosome Inactivation: A Model for the Establishment of Epigenetic Information

In contrast to imprinted gene regulation, which occurs at the level of a few genes within an imprinted locus, X-chromosome inactivation (XCI) acts on an entire chromosome. Mammalian female cells contain two X chromosomes and must therefore inherently solve a problem of X-linked gene dosage. As such, XCI serves to balance sex-chromosome-linked gene dosage in female somatic cells, by rendering one of the two X chromosomes largely transcriptionally silent (Wutz and Gribnau 2007; Payer and Lee 2008).

Two different forms of XCI have been observed: imprinted XCI and random XCI. Imprinted XCI is characterized by the invariable inactivation of the paternally derived X chromosome (Xp), and occurs in the extraembryonic lineages of mice, as well as in all tissues of marsupials. In contrast, random XCI is found in the embryonic lineages of mice and all primate somatic cells (both extraembryonic and embryonic), and is characterized by the random inactivation of either the maternal X chromosome (Xm) or the Xp. Unlike imprinted XCI and autosomal imprinting, in which parent-specific epigenetic information is established in the germline, random XCI is accomplished by establishing contrasting epigenetic information on each homologous chromosome within the same somatic nucleus. Analogous to ICRs, an experimentally delimited region on the X chromosome, the X-inactivation center (XIC), is required for both imprinted and random XCI to define the epigenetic state of the X chromosome in cis (Brown 1991; Brown et al. 1991b). As such, XCI has also served as a vital system in understanding how epigenetic states are established during somatic development.

3. EPIGENETIC MECHANISMS OF GENE REGULATION

3.1. H19/Igf2: Imprinted Expression through Chromatin Structure

The best-known example of imprinted expression is the H19/Igf2 locus. These genes are closely linked, yet oppositely expressed: The H19 noncoding RNA is expressed exclusively from the maternal allele, whereas insulin-like growth factor 2 (Igf2) is expressed from the paternal allele in most tissues (Fig. 1). Misregulation of Igf2 has profound phenotypic consequences for development. Deletion of the paternal Igf2 allele results in smaller mice (DeChiara et al. 1990), whereas biallelic expression results in larger animals (Leighton et al. 1995; Thorvaldsen et al. 1998; Constancia et al. 2000). The regulation of IGF2 in humans is of clinical interest: A genetic disease characterized by somatic overgrowth, Beckwith–Weidemann syndrome, has been associated with biallelic IGF2 expression (Weksberg et al. 2003).

Figure 1.

The H19/Igf2 imprinted locus. (A) H19 and Igf2 are separated by approximately 80 kb on the distal end of mouse chromosome 7. Downstream of H19 lie shared endodermal enhancers. (B) On the paternal allele, the DMD is hypermethylated and the enhancers activate Igf2. (C) On the maternal allele, CTCF binds the hypomethylated sites within the DMD. The presence of CTCF forms a structure sequestering the Igf2 promoter and allowing H19 to be activated by the enhancers.

In between the two genes, upstream of the H19 promoter, lies a CpG-rich DMD. The maternal DMD allele contains very low levels of DNA methylation, whereas on the paternal allele the DMD is hypermethylated (Tremblay et al. 1995). These contrasting methylation patterns are established in the germline of the parents and maintained throughout the development of the soma in the embryo and adult. Engineered mutations have demonstrated the DMD to be required for imprinting of both H19 and Igf2 (Thorvaldsen et al. 1998). Inheritance of an allele with a DMD deletion from either parent results in transcription of both H19 and Igf2 from the mutant allele, thus defining the DMD as the ICR for this locus.

The DMD mediates the correct imprinted expression patterns by controlling the access of the H19 and Igf2 promoters to enhancers located downstream of H19 (Fig. 1). On the maternal allele, the DMD functions as an enhancer-blocking element: a sequence that can interfere with promoter–enhancer communication. This activity is attributable to the presence of binding sites within the DMD for the vertebrate insulator protein, CTCF. As inferred by chromatin conformation capture assays, the CTCF-bound maternal DMD causes the locus to adopt a three-dimensional conformation, which occludes the activation of Igf2 by the enhancers, leading to the activation of H19 expression (Engel et al. 2006; Kurukuti et al. 2006; Zhang et al. 2011). Importantly, association of CTCF with its binding sites is inhibited by the presence of DNA methylation. Therefore, because CTCF cannot associate with the paternally hypermethylated DMD, the enhancer-blocking structure cannot be formed, leading to Igf2 activation by the enhancers. Collectively, these findings demonstrate that the DMD acts as a methylation-sensitive switch to guide allele-specific expression through chromatin conformation.

The H19/Igf2 locus has been an important system for appreciating the role of chromatin structural components in gene regulation during development. Garnering the most interest is CTCF. Genome-wide analysis has revealed thousands of CTCF-binding sites in mammalian genomes (Kim et al. 2007), and other studies have implicated CTCF-driven chromatin structure in diverse developmental processes, including random XCI (Chao et al. 2002; Navarro et al. 2006; Donohoe et al. 2007) and somatic recombination in the immune system (Guo et al. 2011). Intriguingly, DNA-bound CTCF appears to be regulated through additional means. Recently, components of the cohesin complex have been found to occupy many of the same sites as CTCF during interphase in both human and mouse genomes (Parelho et al. 2008; Stedman et al. 2008; Wendt et al. 2008).

At the H19/Igf2 locus, depletion of cohesin results in loss of insulator activity of the DMD without affecting CTCF binding. Thus, the insulator activity of CTCF is regulated not only by DNA methylation, but by assembly into larger protein complexes. Interactions between promoters and distant regulatory elements, facilitated by cohesin, have been found to occur throughout the genome in a variety of cell types (Cuylen and Haering 2010; Kagey et al. 2010). In many cases the formation of these loops is directly linked to gene expression patterns, suggesting a broad, structural mechanism for transcriptional regulation. It is therefore not surprising that mutation of CTCF and cohesin, as well as misregulation of nuclear architecture, have been observed in cancers and number of human diseases (Zaidi et al. 2007). For example, mutation of a structural component of the nuclear membrane, LMNA, is associated with numerous diseases including a form of progeria (premature aging). Strikingly, nuclei from these patients show drastic changes in heterochromatic histone modifications and aberrant nuclear structure (Scaffidi and Misteli 2006). Together, these findings illustrate the importance of chromatin structure in development and human disease.

3.2. Noncoding RNAs: A Common Mechanism for Imprinted Regulation and Dosage Compensation

At first believed to be a hallmark of imprinted regions, noncoding RNAs (ncRNAs) have been identified throughout mammalian genomes (Kapranov et al. 2007a,b; Guttman et al. 2009; Khalil et al. 2009; Latos and Barlow 2009; Santoro and Barlow 2011). Conserved transcription units among disparate species such as zebrafish and humans, along with functional analysis in mice, suggest that ncRNA species may contribute significantly to gene regulation. Almost all known imprinted loci contain at least one ncRNA, and these transcripts are often closely linked with regulation of imprinted expression in cis. Perhaps the best-studied of such molecules are the imprinted Airn and Kcnq1ot1 ncRNAs, and the Xist ncRNA, which is required for XCI in mammals.

Spanning ∼500 kb on mouse chromosome 17, the imprinted Igf2r locus contains several imprinted protein-coding genes and the Airn ncRNA (Fig. 2). Similarly, the ∼900-kb imprinted Kcnq1 locus on mouse chromosome 7 encompasses 10 imprinted genes and the Kcnq1ot1 ncRNA (Fig. 2). Hypomethylated DMRs on the paternal chromosomes drive monoallelic paternal expression of both Airn and Kcnq1ot1. Although the protein-coding genes within these clusters are functionally unrelated, these large imprinted domains share a number of regulatory features. At both loci the DMRs lie within introns of protein-coding genes and serve as promoters for ncRNAs that are transcribed in an antisense orientation to their protein-coding counterparts. Transcription of these ncRNAs occurs exclusively from the hypomethylated paternal DMR (Fig. 2B); ncRNA transcription is suppressed from maternally hypermethylated DMRs (Fig. 2A). Additionally, the ncRNAs are responsible for silencing an additional set of genes in the placenta (Fig. 2C).

Figure 2.

The Igf2r and Kcnq1 imprinted loci. (A) On the maternal alleles, hypermethylation of the DMR silences ncRNA expression, resulting in locus-wide transcription. (B) Conversely, the repressive ncRNAs are transcribed from paternally hypomethylated DMRs. In several instances, these ncRNAs have been shown to associate with the chromatin at the promoter of the repressed alleles. (C) Both loci contain a number of genes that are only repressed by the ncRNA in a subset of extraembryonic lineages. Positions of genes, DMRs, and ncRNA transcripts are relative and not to scale.

Paternal inheritance of DMR deletions at either locus results in the absence of ncRNA expression and is accompanied by reactivation of the normally silent alleles in the region (Wutz et al. 1997; Fitzpatrick et al. 2002; Mancini-DiNardo et al. 2003). Similar effects on imprinted expression are observed when the paternal transcript of either ncRNA is terminated prematurely; however, DMRs remain hypomethylated (Sleutels and Barlow 2002; Mancini-Dinardo et al. 2006). Therefore, the DMRs function as ICRs and reflect parental identity in levels of DNA methylation, whereas the ncRNAs are required for silencing of imprinted genes in cis.

Perhaps the best-known model of ncRNA-mediated gene silencing is demonstrated by XCI in female mammals. In eutherians, XCI is in part regulated through a ncRNA, Xist, that coats the silent X chromosome (Xi) in cis (Fig. 3) (Borsani et al. 1991; Brockdorff et al. 1991; Brown 1991; Brown et al. 1991a; Clemson et al. 1996). The Xist gene is indispensible for both imprinted and random XCI: X chromosomes with deletions of the Xist gene are never inactivated (Kay et al. 1994; Penny et al. 1996; Marahrens et al. 1997). In addition, multicopy Xist transgenes located on autosomes are sufficient to locally silence adjacent genes (Lee and Jaenisch 1997). What makes XCI a particularly fascinating model for epigenetic regulation is the scope of silencing. Induction of Xist on the future Xi results in heterochromatinization throughout the entire chromosome.

Figure 3.

Xist induction results in chromosome-wide changes in chromatin and gene expression. After one X chromosome is chosen for inactivation in both embryos and differentiating ES cells, the Xist transcript can be seen coating the chromosome in cis. This event coincides with exclusion of RNA Pol II. Shortly thereafter, histone-modifying complexes such as polycomb repressive complex 1 and 2 (PRC1, PRC2) become enriched on the inactive X, along with the heterochromatic histone modifications they catalyze. As differentiation proceeds, the histone variant macroH2A becomes associated with the inactive X, and later CpG islands become methylated.

In the mouse, inheritance of a paternally derived deletion of Xist leads to failed imprinted XCI in the extraembryonic tissues of female embryos (Marahrens et al. 1997). Although unknown, the imprint required for imprinted XCI is established during oocyte maturation (Tada et al. 2000). Failure of imprinted XCI leads to two active X chromosomes in the extraembryonic lineages, as the imprint on the Xm prevents it from being silenced (Tada et al. 2000). Although these embryos implant, the lack of dosage compensation is incompatible with proper placental development, leading to early embryonic lethality. Conversely, maternally inherited deletions of Xist are compatible with normal embryonic development, as the Xp normally undergoes imprinted XCI in female extraembryonic cells and undergoes biased random XCI in the embryonic tissues. In this latter case, the Xm is always the Xa. Female embryos completely lacking Xist function are not viable and die before implantation. Males do not require Xist function, as dosage compensation is unnecessary.

Thus Xist, as well as Airn and Kcnq1ot1, has provided a robust system for identifying and understanding large-scale ncRNA regulatory mechanisms that are required for normal embryonic development.

3.3. Gene Silencing through Transcriptional Interference

Functional characterization of imprinted ncRNAs has suggested that they may involve numerous mechanisms to mediate transcriptional silencing of genes within their domains. The phenomenon of transcriptional interference occurs where sense–antisense gene pairs overlap at their transcriptional start sites. Antisense transcription of one gene by RNA polymerase II (Pol II) through the promoter of the other may displace required transcription factors at that promoter. Additionally, elongating RNA Pol II on one strand is known to promote changes in local histone modifications and chromatin structure, possibly precluding antisense Pol II transcriptional initiation events (Lee and Shilatifard 2007). Both Airn and Kcnq1ot1 have transcriptional start sites within introns of protein-coding genes (Igf2r and Kcnq1, respectively) and are transcribed in the antisense direction. The premature transcriptional termination of Airn ncRNA before the Igf2r promoter results in paternal Igf2r expression (Sleutels et al. 2002), suggesting that Airn functions, at least partially, through transcriptional interference (Shearwin et al. 2005). However, Airn and Kcnq1ot1 transcription traverses only a subset of genes, which they repress. Therefore, transcriptional interference can only explain the repression of Igf2r, implying additional mechanisms of ncRNA-based transcriptional repression for neighboring genes.

3.4. ncRNA-Mediated Formation of Repressive Nuclear Compartments

ncRNA-dependent silencing at the Igf2r and Kcnq1 imprinted loci extends hundreds of kilobases away from the site of transcription. In the case of XCI, Xist transcripts influence hundreds of megabases, making XCI a particularly attractive model for epigenetic phenomena. Although data suggest that Airn and Kcnq1ot1 may mediate the formation of a localized, transcriptionally silent nuclear compartment (Terranova et al. 2008; Redrup et al. 2009), Xist function during XCI has provided the best evidence of ncRNA-mediated formation of silent nuclear domains.

A critical facet of XCI is the localization of the Xist RNA. Using RNA fluorescence in situ hybridization (FISH), a spliced Xist transcript can be seen coating the entire Xi. At the onset of XCI, the future Xi up-regulates Xist expression, facilitating coating of the Xi by Xist transcripts. Once Xist coating is established, the chromatin of the Xi is dramatically altered (Fig. 3). The Xi becomes increasingly enriched for histone modifications similar to those found in transcriptionally silent heterochromatic regions of the genome. Like Xist coating, the enrichment of these modifications and other heterochromatic proteins can be readily visualized with specific antibodies. Together, these data suggest that formation of large heterochromatin domains by ncRNAs may be a core mechanism of large-scale epigenetic silencing.

In addition to Xist coating, the onset of XCI is also characterized by the cytologically visible exclusion of RNA Pol II and associated general transcription factors from the Xi chromosome territory (Fig. 3) (Chaumeil et al. 2006). Following the subsequent depletion of active chromatin modifications such as di- and trimethylated lysine 4 of histone H3 (H3K4me2/3), repressive histone modifications begin to accumulate on the Xi. The first such modification enriched on the Xi is the trimethylation of lysine 27 of histone H3 (H3K27me3), catalyzed by the Ezh2 histone methyltransferase of polycomb repressive complex 2 (PRC2) (Plath et al. 2003; Okamoto et al. 2004). Xist is believed to directly recruit the PRC2 complex to the Xi via sequences located in the 5′ end of the Xist RNA, termed the A repeat (Zhao et al. 2008). In both the early embryo and differentiating embryonic stem (ES) cells, chromosomes lacking Xist fail to accumulate H3K27me3 and are incapable of undergoing XCI. As XCI proceeds, the Xi accumulates a repertoire of histone modifications associated with silent chromatin, including H3K9me2 (dimethylated lysine 9, histone H3), H3K9me3, H4K20me1 (monomethylated lysine 4, histone H4), and H2A119Ub (ubiquitinated histone H2A) (Fig. 3).

The Xi eventually becomes enriched for the histone variant macroH2A, which is believed to be a stabilizing mechanism for the silent chromatin of the Xi (Costanzi and Pehrson 1998). At this point, Xist is no longer required for XCI maintenance. Finally, de novo DNA methylation of CpG islands on the Xi provide a mechanism for long-term, X-linked gene silencing. In agreement with a critical role in this late phase, mouse embryos deficient in Dnmt1 fail to maintain proper XCI patterns (Sado et al. 2000). Collectively, these observations demonstrate an orchestrated, precise chain of events that initiates with Xist RNA coating of the future Xi, and involves multiple epigenetic modifications and likely numerous trans-acting epigenetic regulators. Ultimately, these events produce a chromosome-wide heterochromatic domain refractory to efficient Pol II transcription.

4. TISSUE-SPECIFIC IMPRINTING

In addition to genes that are imprinted in all tissues, many more show tissue-specific imprinting, especially in the mouse placenta and brain (Gregg et al. 2010a,b; Hudson et al. 2010, 2011). Engineered mutations that cause both loss of activity (through targeted deletions) and overexpression (through loss of imprinting) have revealed significant effects on placental function and development, as well as on postnatal and adult behavior (Davies et al. 2007; Frost and Moore 2010). This is particularly evident at the Kcnq1 locus, where many of the genes within this region contribute to placental development. For example, Ascl2 is required for the proliferation of spongiotrophoblast progenitor cells and Cdkn1c for the development of trophoblast giant cells (Guillemot et al. 1995; Tanaka et al. 1997; Caspary et al. 1999; Takahashi et al. 2000). Moreover, deletion of the maternal allele of Phlda2 results in placental overgrowth (Frank et al. 2002). Conversely, reactivation of the paternal allele caused by genetic disruption of Kcnq1ot1 expression results in the development of a smaller placenta (Salas et al. 2004).

As in the embryo, Kcnq1ot1 is responsible for silencing of the placental genes; however, the locus has been observed to undergo a “compaction” specific to extraembryonic lineages (Terranova et al. 2008; Redrup et al. 2009). Reminiscent of the mechanism of Xist, the Kcnq1ot1 ncRNA forms a repressive nuclear compartment, devoid of Pol II and other marks of active chromatin (Terranova et al. 2008). Moreover, loss-of-function studies in mice have demonstrated that the PRC1 and PRC2 histone modification complexes, as well as the G9A histone methyltransferase (responsible for catalyzing the H3K9me2 modification), have essential roles in repressing placentally imprinted genes at several imprinted loci (Mager et al. 2003; Nagano et al. 2008; Wagschal et al. 2008).

It is not yet known how genes are designated for imprinting only in a subset of tissues. In the case of the Kcnq1 locus, where placentally and ubiquitously imprinted genes are in close proximity to one another, structural changes brought about by the Kcnq1ot1 ncRNA may recruit the placental genes into a silent compartment. Consistent with this hypothesis, the locus has been observed to undergo a compaction specific to extraembryonic lineages (Terranova et al. 2008; Redrup et al. 2009).

However, although most imprinted loci are associated with a ncRNA, function of the ncRNA is not necessarily conserved, implying additional mechanisms to designate imprinted alleles for tissue-specific regulation. The differential use of alternative promoters appears to be a common theme. For example, the Grb10 gene is maternally expressed in most tissues, yet predominantly paternal in the brain (Arnaud et al. 2003; Hikichi et al. 2003). This is accomplished through the use of a hypomethylated paternal DMR that is capable of serving as a promoter in this state (Arnaud et al. 2003; Hikichi et al. 2003; Yamasaki-Ishizaki et al. 2007; Sanz et al. 2008). The activity of this promoter is restricted to neuronal tissues in part through the action of PRC2 (Sanz et al. 2008). Collectively, these examples from tissue-specific imprinting illustrate a number of ways that genes may be subject to epigenetic control during development. The ordered and timely recruitment of epigenetic modifiers to promoters and other regulatory elements is the key to producing the correct patterns of expression during the developmental program.

5. ESTABLISHING EPIGENETIC STATES AT IMPRINTED LOCI: LESSONS FROM THE GERMLINE

During development, changes in gene expression bring about the differentiation of pluripotent cells into transient progenitor stages, and ultimately into specialized, terminally differentiated cells. Accompanying these transitions, specific gene expression patterns are established and maintained, then augmented as new developmental stimuli are received. Numerous epigenetic mechanisms participate in this process, including covalent histone modifications and DNA methylation. For example, apart from regulating imprinted genes, polycomb proteins are required for managing genome-wide expression patterns. Further, loss-of-function experiments have demonstrated that Dnmt3b, a member of the de novo DNA methyltransferase gene family, is required for development (Okano et al. 1999). Moreover, human immunodeficiency, centromere instability, and facial anomalies (ICF) syndrome is associated with mutations in DNMT3B (Bestor 2000). These observations illustrate the importance of establishing epigenetic information to development and human disease. Imprinted genes have been invaluable in elucidating how these patterns are generated at specific sites.

Differential DNA methylation has been observed at nearly all known imprinted gene clusters. These patterns are established in the gametes, and inherited by the embryo. The functional relevance of DNA methylation to genomic imprinting has been demonstrated by studies from mice with null mutations in the maintenance DNA methyltransferase, Dnmt1. In addition to significant developmental defects, Dnmt1-null embryos show loss of DNA methylation patterns and imprinted expression at most imprinted loci (Li et al. 1992, 1993). Therefore, the establishment and maintenance of this modification is central to imprinted gene regulation.

Unlike the maintenance of DNA methylation that occurs in the developing soma, de novo methylation of ICRs occurs in a sex- and locus-specific manner. Before these processes begin, DNA methylation patterns are erased during the development of primordial germ cells. These cells undergo genome-wide epigenetic changes relative to their somatic counterparts, such that by the time the sexually dimorphic stages of germ cell development begin, imprinted alleles contain equivalent, low levels of DNA methylation (Davis et al. 2000; Hajkova et al. 2002). De novo methylation of paternally methylated ICRs, such as the H19 DMD, begins during the later stages of embryogenesis in the quiescent prospermatogonia and is complete by the entry into meiosis (Davis et al. 1999, 2000; Ueda et al. 2000). In contrast, de novo methylation of ICRs in the female germline does not begin until after birth (Lucifero et al. 2002). This occurs during the period of oocyte growth, when oocytes are arrested in the first prophase of meiosis. Notably, the kinetics of de novo methylation at DMRs differs among imprinted loci, indicating that locus-specific characteristics, and perhaps even diverse mechanisms, regulate this event.

5.1. The CTCF-DMD Interaction Is Essential for the Establishment and Maintenance of the Maternal Epigenotype

The H19/Igf2 DMD is one of the few paternally hypermethylated ICRs; it remains free of significant levels of DNA methylation in the oocyte. Apart from other ICRs, the genome is undergoing large-scale reorganization and reprogramming of chromatin during oocyte growth, including de novo methylation of repetitive and nongenic sequences. Because of the apparent accessibility of similar sequences to de novo methyltransferases, a mechanism must exist to distinguish the H19 DMD and protect it from becoming hypermethylated.

The simplest mechanism to prevent de novo methylation is through a protective mechanism whereby the binding of a factor to DNA effectively masks the underlying CpG residues from the methyltransferase machinery. At the H19 DMD, CTCF serves this vital role. In oocytes depleted for CTCF, the H19 DMD acquires an abnormally high level of DNA methylation (Fedoriw et al. 2004). As the expression of the de novo methyltransferases is not limited to gametogenesis, and can occur in somatic tissues of both the embryo and the adult, the protective role of CTCF in maintaining a hypomethylated maternal H19 DMD extends beyond the female germline. Genetically engineered mutations in the DMD that alter or completely delete CTCF-binding sites,acquire DNA methylation during development (Schoenherr et al. 2003; Pant et al. 2004; Engel et al. 2006). These results demonstrate a role for CTCF in establishing and maintaining maternal identity of the H19 DMD. Importantly, a mutually exclusive relationship for CTCF and DNA methylation has been shown for a number of nonimprinted loci, including human C-MYC (Gombert and Krumm 2009) and mouse p16 loci (Witcher and Emerson 2009). Similarly, protective roles have been described for other DNA-binding factors, including Sp1 (Brandeis et al. 1994) and YY1 (Kim et al. 2009). Therefore, association of trans-factors with CpG-rich DNA sequences may be a common mechanism in preventing de novo DNA methylation at regulatory loci throughout the genome and important in guiding developmental decisions.

5.2. Targeting of DNA Methyltransferases to DMRs during Oogenesis

Unlike the H19 DMD and most other CpG-rich sequences, the majority of known ICRs are maternally hypermethylated. These ICRs are believed to possess some quality to attract de novo DNA methyltransferase machinery during the correct phase of oocyte growth. Recent data have suggested the signals that bring about DNA methylation may be both genetic and epigenetic in nature. Analysis of the primary DNA sequence of maternally methylated ICRs has suggested a specific periodicity to CpG dinucleotides (Jia et al. 2007; Glass et al. 2009). Additionally, DNMT3L, the noncatalytic member of the DNMT3 family of de novo DNA methyltransferases, interacts with H3K4 but only when this residue is unmethylated (Ooi et al. 2007). Although these qualities together produce some specificity, their broad genomic distribution may not fully explain discrete methylation events required for the establishment of specific methylation patterns at imprinted loci.

In many cases maternally hypermethylated ICRs serve as promoters for regulatory ncRNAs that lie antisense to protein-coding transcripts. Although the functionality of noncoding transcripts in regulating activity of overlapping or nearby genes has been explored, it appears that protein-coding transcripts may have an analogous role to their antisense counterparts. At the imprinted Gnas locus (Fig. 4), such an antagonistic relationship exists between the maternally expressed protein-coding Nesp gene and its paternally expressed antisense counterpart, Nespas. Like the Airn and Kcnq1ot1 ncRNAs, Nespas is expressed from a paternally hypomethylated DMR; the maternal DMR is hypermethylated in oocytes, preventing Nespas transcription from this allele in the soma of the embryo. Kelsey and coworkers found transcripts emanating from an alternate Nesp promoter transcribed during the time of de novo methylation of the DMR, whereas no other RNA species from elsewhere in the Gnas locus were detectable (Chotalia et al. 2009). Importantly, an engineered mutation that prevented transcription through the DMR was not appropriately methylated during oogenesis, demonstrating a role for transcription of the Nesp gene in hypermethylation of the DMR.

Figure 4.

Imprint establishment at the Gnas locus. This region contains two DMRs with germline methylation (gDMR), each acting as a promoter for an ncRNA, as well as a DMR at the Nesp promoter that acquires methylation specifically on the paternal allele during early development. (A) Both gDMRs are hypermethylated during oogenesis, allowing for high levels of Nesp and Gnas transcription from the maternal allele in the embryo. (B) On the paternal Gnas allele, ncRNAs are transcribed from promoters within each gDMR. Gnas transcripts are detectable from the paternal allele, albeit at lower levels than from the maternal allele. Thus, repression of Gnas is incomplete on the paternal allele.

These results implicate an RNA as a “guide molecule” for the establishment of DNA methylation, possibly by a direct interaction of the RNA with DNMT3 proteins and/or other epigenetic regulators. Intriguingly, interactions between ncRNAs and histone methyltransferase complexes have been detected in vivo (Nagano et al. 2008; Pandey et al. 2008; Zhao et al. 2010), suggesting that ncRNAs are capable of directing epigenetic modifications in mammalian nuclei. Alternatively, the epigenetic changes brought about by transcription itself may be responsible for producing a chromatin composition conducive to the action of the de novo DNA methyltransferases. A number of histone modification enzymes and demethylases are known to associate with elongating RNA Pol II (Fuchs et al. 2011), which may produce a pattern that would serve as an optimal chromatin template for the de novo DNA methylation. In this model, transcription, rather than a transcript, plays the pivotal role in the establishment of epigenetic information.

6. ESTABLISHING DIFFERENTIAL EPIGENETIC STATES IN THE SOMA: RANDOM XCI

In contrast to the reprogramming events in the germline that erase somatic imprinting patterns to establish parental identity on both alleles, asymmetric epigenetic states are established on each X chromosome within the same nucleus during early embryonic development in female mammals. In mice, initiation of random XCI is directly linked with exit from the pluripotent state. The factors known to be required for ES cell self-renewal, Oct4, Nanog, and Sox2, all act to repress Xist transcription (Navarro et al. 2008); therefore, ES cells contain two active X chromosomes. Upon their differentiation in vitro, the events of random XCI can be recapitulated, comparable to what has been observed in vivo (Chow and Heard 2009). As such, ES cells have served as a surrogate to the developing mouse epiblast, resulting in the rapid identification of genomic sequences required for Xist regulation.

Despite exhaustive genetic manipulations of the XIC, analysis of mutations of epigenetic modifying factors, as well as observations from human pathologies, the initiation of XCI remains enigmatic. Notably, gene silencing and Xist coating appear uncoupled during mammalian preimplantation development and during the erasure of imprinted XCI in the developing mouse epiblast. In mice, gene silencing during imprinted XCI precedes Xist coating, whereas in humans and rabbits, Xist coating occurs in the absence of gene silencing (Kalantry et al. 2009; Okamoto et al. 2011; Williams et al. 2011). However, Xist is an essential, if not central, component to the XCI process (Penny et al. 1996). Thus, although additional events may participate in XCI during the earliest developmental stages, establishing monoallelic Xist expression during the onset of random XCI is a critical aspect in generating a heritable, chromosome-wide, transcriptionally inert state.

Work from female mice and ES cells has identified a complex network of ncRNAs, proteins, and cytological events that may participate to establish differential Xist expression. Among them, an antisense RNA to Xist, Tsix, has a central role in regulating Xist (Fig. 5A) (Lee et al. 1999). Like imprinted sense–antisense gene pairs, the expression of Xist and Tsix from a single allele is mutually exclusive. During the onset of XCI, Tsix expression becomes limited to a single allele on what will become the future active X chromosome (Xa) (Fig. 5B), whereas up-regulation of Xist on the homologous allele leads to a chromosome-wide repressive state (Xi) (Fig. 5C) (Lee and Lu 1999; Lee et al. 1999). Tsix transcription traverses the Xist promoter and results in the accumulation of repressive histone modifications (Sado et al. 2005). As is the case with Airn, truncation of Tsix transcription before the Xist promoter results in a mutant allele that up-regulates Xist expression before and during XCI. Importantly, this mutation cannot be rescued by expression of a Tsix gene repositioned such that it does not overlap the Xist promoter (Shibata and Lee 2003, 2004; Ohhata et al. 2008). Therefore, the repression of Xist by Tsix represents another example of transcription playing a central role in establishing an epigenetic state.

Figure 5.

The X-inactivation center controls chromosome-wide silencing. (A) The antagonistic relationship between Xist and its antisense partner, Tsix, controls the inactivation of the entire X chromosome. During differentiation from a pluripotent state, female cells will choose one X chromosome for silencing. This mechanism occurs through Xist and Tsix regulation. (B) Tsix up-regulation represses Xist transcription, leading to an active X chromosome. (C) When Xist is up-regulated, either through down-regulation of Tsix or its genetic ablation, XCI occurs in cis.

Despite the known role of Tsix in down-regulating Xist transcription, it is not yet clear how this asymmetric pattern of expression is established. Several lines of evidence suggest two separate phases of the initiation process, counting and choice, which have been difficult to experimentally separate and appear very closely interconnected. Counting ensures that only one X chromosome is active per diploid set of autosomes. In mice, Xist itself is not required for counting, as heterozygous deletions of Xist preferentially inactivate the wild-type chromosome, and thus must be able to sense the existence of two X chromosomes for XCI to occur (Marahrens et al. 1998; Gribnau et al. 2005).

Counting is followed by choice, wherein a difference must be established between homologous X chromosomes, which will eventually lead to differences in transcriptional activity. For choice to occur, a difference must be recognized between two homologs, yet in mouse embryos and ES cells, the X chromosomes are often genetically identical. Not surprisingly, heterozygous deletions of Tsix and Xist favor inactivation and activation of the mutant chromosome, respectively (Marahrens et al. 1997; Lee and Lu 1999). Apart from these engineered effects, choice is skewed by cis-elements within the XIC. Alleles of X-controlling element (Xce) differ between mouse strains, and affect the probability of inactivation in F₁ hybrid females. However, the mechanism behind choice remains largely unknown.

Although choice is traditionally believed to take place after counting, recent data suggest an alternative possibility. Transient, stochastic differences between the alleles before differentiation may dictate which X chromosome is chosen for Xist up-regulation (Fig. 6) (Mlynarczyk-Evans et al. 2006). This effect is manifest in differences in cytological appearance of replicated alleles. During S phase, regions that have undergone DNA replication appear as two juxtaposed DNA FISH signals (a doublet), whereas regions that have not yet replicated appear as a single FISH signal (a singlet). Despite coordinated replication timing of homologous alleles among X-linked loci in undifferentiated female ES cells, one allele appears as an expected doublet, whereas the other mysteriously appears as a singlet. This phenomenon is coordinated across this chromosome, except the XIC, which shows an opposite pattern. For example, on the X chromosome on which genes appear as singlets, the XIC will appear as a doublet. These patterns do not appear fixed, as chromosomes oscillate between the two states in an undifferentiated population. Importantly, it is predictive of which allele will be inactivated after differentiation: The singlet-to-doublet ratio is skewed between X chromosomes carrying different Xce alleles, reflecting the probability of silencing dictated by the Xce locus. Therefore, it may be plastic states in chromatin structure that serve to differentiate between identical alleles.

Figure 6.

Is the future epigenetic state predetermined? Most models of counting and choice presuppose that these processes occur only after differentiation begins. However, recent data suggest that X chromosomes—and indeed other monoallelically expressed loci—fluctuate between states that will determine their activity after differentiation. (A) In undifferentiated ES cells, X chromosomes switch between a state that will predispose them to the active (green) or inactive (red) fate. The pattern observed at the XIC is opposite that of the rest of the chromosome. (B) After differentiation, the outcome of choice would reflect the state of the chromosome before XCI began.

Apart from imprinted genes, whose identity is set forth in the germlines, random monoallelic expression is pervasive in mammalian genomes (Gimelbrant et al. 2007). The selection of a single allele is important in numerous developmental processes in mammals. For example, in olfactory neurons only one allele of hundreds of olfactory receptor genes scattered in clusters throughout the genome is chosen for activity. Similar to that of X-linked loci before XCI, olfactory receptor loci display a singlet–doublet pattern preceding differentiation (Alexander et al. 2007). Although olfactory receptor gene clusters do not possess known regulatory ncRNAs like Tsix or Xist, transient differences in chromatin structure in undifferentiated cells may be a prevalent mechanism for establishing differential epigenetic modifications at genetically identical alleles during the course of development.

7. CONCLUSIONS

Genomic imprinting is perhaps the most striking display of epigenetic phenomenon in mammalian development. Importantly, it has provided a fruitful system for understanding how epigenetic states are established, interpreted, and maintained by the transcriptional machinery to produce very different outcomes on each genetically identical allele. XCI has also become a popular model for epigenetic gene regulation, especially for the distinct regulatory events that lead to chromosome-wide gene silencing. In contrast to how genomic imprints are established during gametogenesis, differential epigenetic states on each X homolog are established within the same nucleus. Each example of monoallelic gene expression has important ramifications not only for mammalian development, but for human disease. For example, loss of imprinted expression at numerous loci is associated with human pathologies including birth defects and cancer. Therefore, understanding the mechanisms that initiate and maintain differential epigenetic states has broad implications that extend into the clinical setting.