Methylation of the cytosine base in DNA, DNA methylation, is an essential epigenetic mark in mammals that contributes to the regulation of transcription. Several advances have been made in this area in recent years, leading to a leap forward in our understanding of how this pathway contributes to gene regulation during embryonic development, and the functional consequences of its perturbation in human disease. Critical to these advances is a comprehension of the genomic distribution of modified cytosine bases in unprecedented detail, drawing attention to genomic regions beyond gene promoters. In addition, we have a more complete understanding of the multifactorial manner by which DNA methylation influences gene regulation at the molecular level, and which genes rely directly on the DNA methylome for their normal transcriptional regulation. It is becoming apparent that a major role of DNA modification is to act as a relatively stable, and mitotically heritable, template that contributes to the establishment and maintenance of chromatin states. In this regard, interplay is emerging between DNA methylation and the PcG (Polycomb group) proteins, which act as evolutionarily conserved mediators of cell identity. In the present paper we review these aspects of DNA methylation, and discuss how a multifunctional view of DNA modification as an integral part of chromatin organization is influencing our understanding of this epigenetic mark's contribution to transcriptional regulation.
Transcriptional regulation is achieved through the concerted action of regulatory networks of trans-acting factors and dynamic chromatin organization . A major challenge of molecular biology is to decipher this interplay, and to establish the role that chromatin structure plays in the co-ordination of gene expression. The structure of chromatin is influenced by several layers of information encoded at the genetic and epigenetic levels . One such layer of information is encoded by chemical modifications of DNA, referred to as DNA methylation, and is thought to play integral roles in many biological processes, including embryonic development and disease pathology . The three most abundant forms of cytosine in mammalian genomes are 5mC (5-methylcytosine or ‘DNA methylation’), unmodified cytosine and 5hmC (5-hydroxymethylcytosine) . Owing to the position of the modification site within the cytosine base, these three isoforms do not show differential base pairing properties and therefore do not alter the genetic sequence. Instead, the modification state of cytosine has long been linked to transcriptional regulation, although its precise function in this process remains unclear. In the present paper we review developments in this field, focusing on those that point to previously unappreciated functions of DNA methylation in genome regulation. We first discuss insights that have arisen through studies detailing the distribution of DNA modifications throughout the genome, and how this distribution varies between tissues and in disease states. Next, we review data suggesting that DNA modifications are important modifiers of chromatin structure, and focus on the emerging interplay between DNA methylation and the Polycomb system in this regard. We discuss how this interplay has the potential to enhance our understanding of the enigmatic targeting mechanism of the Polycomb system, and also to expand upon the way that we view the role of DNA methylation during development and in human disease.
NOVEL ASPECTS OF DNA METHYLATION: CLUES FROM THE DNA METHYLOME
A detailed knowledge of the global distribution of cytosine modifications, the DNA methylome, and how the methylome differs between cell types and individuals is a critical step towards understanding the role of DNA methylation in genome regulation . Various technical advances have facilitated the mapping of modified cytosine bases at unprecedented resolution and coverage, and consequently our understanding of the DNA methylome has improved greatly in recent years. In mammals, the 5mC form of cytosine is primarily found where a cytosine is followed by a guanosine in the 5′→3′ direction (a ‘CpG dinucleotide’), and it is in this context that 5mC is associated with transcriptional regulation [6–10]. Mammals, as do all vertebrates studied, have a ‘global’ methylome with the bulk of CpGs in the genome being found in the methylated state (5mC) [6,11]. This high level of genomic methylation is punctuated by short stretches of unmethylated DNA, many of which correspond to regions of high CpG content known as CGIs (CpG islands) [4,12]. Although the majority of CGIs are found in the unmethylated state in all tissues studied, a subset are found to be methylated in a tissue-specific manner [6,11,13]. DNA methylation patterns in general are variable between different tissues and are dynamic during cell differentiation [6,7,14]. It is this property, together with the mitotic heritability of CpG methylation patterns, and the ability of CpG methylation to modify transcription, that implicate DNA methylation as a process influencing maintenance of cell identity [15–19].
Canonical repression by DNA methylation: gene promoter regions and repetitive elements
The most heavily studied regions of the genome with respect to DNA methylation-mediated gene regulation are those directly proximal to transcription start site-promoter elements. It is well established that if a promoter element contains a sufficient density of CpG dinucleotides, the presence of high levels of DNA methylation (5mC) is strongly associated with gene inactivity . The function of DNA methylation at gene promoters has frequently been attributed to two mechanisms: (i) its negative effect on certain transcription factor–DNA interactions [21–23], and (ii) its attraction of methyl-CpG-binding proteins [4,24–26] (Figure 1). This canonical form of DNA methylation-mediated repression is essential for the parent-of-origin-specific expression of imprinted genes [27–29], and the process of Xchromosome inactivation, where many genes on one of the two X-chromosomes in female somatic cells become repressed [30,31]. Functional DNA methylation at other gene promoter regions is relatively infrequent, and is related to CpG density: CpG-poor promoters are often highly methylated irrespective of their activity status, whereas promoters with high CpG-density are commonly unmethylated, such as CGIs . At high-CpG-density promoters the methylated state is tightly associated with transcriptional inactivity, whereas the unmethylated state can be associated with either actively transcribing genes, or inactive genes subject to other forms of repression . Despite canonical repression now being recognized as relatively uncommon, promoter DNA methylation is an important mechanism in the control of tissue-specific gene regulation [28,32,33]. For example, a cohort of genes expressed in the mouse germline relies on promoter DNA methylation for repression in somatic cell lineages [20,34–38]. For these genes DNA methylation appears to be a primary factor regulating their expression patterns. However, in most cases methylation of promoter regions appears to represent a secondary mechanism contributing to maintenance of the repressed state [39–41]. This type of repression by DNA methylation is exemplified by its role in X-chromosome inactivation and appears to contribute mainly to long-term transcriptional repression, which in some cases needs to be maintained in a given tissue for the life of the organism. In addition to single-copy gene promoters, it is well established that DNA methylation contributes to the repression of certain repetitive elements throughout the genome, particularly IAP (intracisternal A-particle) elements .
Canonical functions of DNA methylation in gene regulation
Beyond gene promoters: non-canonical roles for DNA methylation?
New methods of DNA methylation mapping with improved resolution and coverage have revealed novel aspects of the methylome, while drawing attention to genomic regions outside of gene promoter elements. These observations have highlighted the possibility that DNA methylation plays important, and as yet unclear, roles in genome regulation in addition to its canonical function at promoter elements.
Gene body methylation
Vertebrate methylomes are classified as ‘globally methylated’, so it comes as no surprise that genes are also methylated internally. Provocative observations suggest that gene body methylation may represent a crucial function in gene regulation. Intragenic DNA methylation appears to be evolutionarily ancient and many organisms with so-called ‘mosaic’ methylomes also exhibit abundant gene body methylation. For example, in the model flowering plant Arabidopsis thaliana the vast majority of DNA methylation occurs within gene bodies and transposon sequences . Genes that are moderately transcribed in Arabidopsis contain the most 5mC in their gene body regions, whereas genes at either extreme of the expression range contain lower methylation levels . Experimental reduction of DNA methylation leads to up-regulation of genes with gene body methylation, suggesting a repressive role . DNA methylation has also been suggested to inhibit transcriptional elongation in the filamentous fungi Neurospora crassa . However, in contrast with Arabidopsis and Neurospora, a positive correlation between the abundance of gene body methylation and mRNA levels has been observed in mammals [7,44–47]. In support of this, the co-transcriptionally deposited H3K36me3 (indicating the histone number modified, and the level of methylation on the lysine residue indicated, e.g. this modification is histone H3 trimethylated on Lys36) histone modification has been proposed to play a direct role in the targeting of DNA methylation [48,49]. In addition, many tissue-specific differentially methylated regions have been found in gene bodies in mammals, including CGIs that are found in intragenic regions [50–52]. However, the relationship between gene body methylation and transcript abundance is not universal and may be cell-type-specific [6,46,53]. It should also be noted that studies of gene body methylation are complicated by the additional presence of the 5hmC base within these regions , and by the inability of many of the bisulfite-based 5mC mapping methods to distinguish between the 5mC and 5hmC bases .
The role of gene body methylation in mammals is currently unclear, but two potential functions have been proposed (Figure 2B): (i) the regulation of co-transcriptional RNA processing such as alternative splicing, and (ii) the prevention of spurious transcription initiation within gene bodies. It is now becoming apparent that chromatin structure may play important roles in the regulation of alternative splicing and mRNA processing . Associations between DNA methylation levels and alternative splicing have been noted in the honeybee, Apis mellifera, and also in humans, with increased DNA methylation of an exon being associated with its preferential inclusion in a mature transcript [57–59]. A direct link has been made through the multi-functional DNA-binding protein CTCF (CCCTC-binding factor) which is sensitive to CpG methylation in certain binding contexts . During alternative splicing, lower affinity sites becoming more favourable if transcription elongation occurs more slowly . Binding of CTCF protein downstream of a splice site was proposed to cause RNA polymerase II pausing, reduction in elongation speed and therefore an alteration in the choice of splice site in a methylation-modulated manner  (Figure 2Bi). This raises the possibility that tissue-specific differences in DNA methylation within gene bodies could contribute directly to differential RNA processing. A second potential function that has been speculated is the prevention of transcription initiation events within gene bodies. As the act of transcription may destabilize nucleosomes, leading to a more open chromatin conformation, this could promote new transcription initiation events producing cryptic transcripts (Figure 2Bii). Some evidence does exist that gene body methylation prevents transcriptional initiation. A tissue-specific methylated CGI was shown to act as an alternative promoter of the Shank3 gene, in both mouse and humans; its activity was negatively regulated by DNA methylation . However, a general role for gene body methylation in prevention of spurious transcription initiation within transcription units currently remains speculative.
Emerging non-canonical functions of the DNA methylome in gene regulation
Orphan CpG islands
Approximately half of CGIs are not associated with the 5′ end of annotated genes, and are instead found in intragenic or intergenic locations [13,51]. Interestingly, these ‘orphan’ CGIs are more frequently found in the methylated state than their promoter counterparts and they also show a greater incidence of tissue-specific methylation [13,51]. The function of orphan CGIs and their DNA methylation state is currently unclear . Unmethylated CGIs in general show many properties suited for function as regulatory elements, including an increased frequency of nucleosome-depleted regions, depleted linker histone occupancy, and histone marks that are associated with a more open chromatin configuration [4,60,61]. Therefore it is likely that many orphan CGIs represent promoters for unannotated transcripts (Figure 2Ci). In support of this idea, many unmethylated orphan CGIs show evidence of transcription initiation that varies in a tissue-specific manner . Orphan CGIs may act as developmentally regulated alternative promoters for known transcripts . Alternatively, they may represent promoters of functional non-coding RNAs, such as long non-coding RNAs, which may play important roles themselves in gene regulation in cis or trans. Methylation at orphan CGIs could conceivably play a role in the regulation of local chromatin configuration by modulating the binding of effector proteins that directly or indirectly modify chromatin structure [61,62] (Figure 2Cii). Intriguingly, genes with key roles in embryonic development, particularly transcription factors such as Hox genes, are often found in the proximity of CGI clusters and orphan CGIs that are methylated in a tissue-specific manner [13,63]. Future studies will delineate the function of orphan CGIs and the importance of their methylation status in transcriptional regulation.
A set of studies noted that tissue-specific methylation occurs less frequently within CGIs and gene promoters when compared with flanking regions [64–66]. The greatest variation in DNA methylation between cell types was observed at regions immediately adjacent to CGIs, termed ‘CpG shores’. These regions also showed the most variation between human induced pluripotent stem cells and their parental fibroblasts, and between normal colon and colon cancer [64,65]. CpG shores were defined as regions immediately flanking a CGI, within 2 kb on either side, and therefore have a lower CpG content relative to their neighbouring CGI . The functional consequence of CpG shore methylation is unclear, but it correlates with reduced transcription of the nearby gene [64,66]. A study of candidate genes showed that differential methylation at CpG shores is associated with the use of alternative promoters in some cases, suggesting that this process may influence the regulation of transcription and promoter choice . In this regard, it has been suggested that promoter-flanking methylation can act as a docking site for MeCPs (methyl-CpG-binding proteins) such as MeCP2, and can therefore modulate the transcriptional output of a non-methylated promoter .
Distal regulatory elements
DNA methylation mapping studies revealed that regulatory elements located distal to gene promoters are often differentially methylated between tissues [6,14,53,68–71]. Examples are enhancer elements, short stretches of DNA that are bound by trans-acting factors and act to stimulate the expression of distal target genes. DNA methylation at enhancers is often inversely correlated with their activity [6,53,68,69,71]. The function of DNA methylation at enhancers has not been extensively studied, and the cause and effect relationships are unclear (Figure 2D). Recent data from the ENCODE project shows that DNA methylation at the binding sites for a given transcription factor is negatively correlated with tissue-specific expression of that transcription factor in a panel of human cell lines, suggesting that the presence of a DNA-binding factor is sufficient to predict hypomethylation at its binding sites . This idea has been directly tested by inserting a fragment of DNA containing a transcription factor motif into the genome of mammalian cells. This fragment was protected from methylation only when the DNA-binding motif was intact, suggesting that binding precludes DNA methylation, and hence leads to the observed hypomethylation at enhancer regions [68,72] (Figure 2Di). In addition, insertion of the same fragment after completely methylating it in vitro did not preclude binding of this particular transcription factor, suggesting that in this context CpG methylation is compatible with transcription factor occupancy . However, intriguing evidence exists for the alternative hypothesis that DNA methylation can modulate enhancer activity (Figure 2Dii). The in vitro CpG methylation of certain tissue-specific enhancers can strongly attenuate their activity in reporter assays . In addition, a study of the glucocorticoid receptor showed that DNA methylation prevents binding of this factor to an enhancer element in vitro, and the formation of a more open chromatin structure in vivo . It is of course possible, and indeed likely, that any effect of DNA methylation on enhancer binding is context-dependent, varying for different genomic locations and for different DNA-binding factors.
Partially methylated domains
Global single-nucleotide resolution maps of DNA methylation have enabled the study of methylation over large genomic domains. In the first base-resolution methylome map of mammals it was noticed that large genomic regions of lower methylation stood out from the background of high DNA methylation . These PMDs (partially methylated domains) had a mean methylation of less than 70%, spanned hundreds of kilobases and covered a substantial proportion of each chromosome [6,53]. Subsequently, PMDs have been identified in other cell types and have been suggested to become further hypomethylated in cancer cell lines and tumour tissue [73–75]. However, it is currently unclear whether PMDs are a general feature of DNA methylomes or if they represent changes that occur upon adaption to culture and tumorigenesis . The mechanisms that control DNA methylation at this scale, and its function in gene regulation, are currently opaque. Interestingly, PMDs often coincide with other aspects of chromatin organization, such as large domains of histone marks that are generally associated with transcriptional repression, suggesting that they may reflect aspects of higher-order chromatin structure [6,53,74,75]. Gene expression is also associated with DNA methylation level at the domain scale, with genes within PMDs exhibiting a low level of expression [6,53,73]. Whether PMDs are a cause or consequence of higher-order chromatin organization is unknown, but as CpG methylation can modulate certain aspects of chromatin structure (see later section), it is possible that PMD formation contributes to this level of genome organization. These studies, together with gene body methylation, demonstrate that the association of DNA methylation with transcription is context- and location-dependent, and goes beyond promoter elements. Gene regulation is more frequently being linked to higher-order organization of chromatin, so it will be interesting to see what future studies tell us about the functional significance of PMDs.
Previously it was discovered that 5mC can be oxidized to 5hmC by the family of TET (ten-eleven translocation) enzymes, and that genomic 5hmC is abundant in certain mammalian tissues [76–78]. These discoveries have prompted intense research into the distribution and putative function of the 5hmC base, reviewed recently [79,80]. Mapping of the 5hmC base showed high tissue specificity, both in its global abundance and locus-specific distribution [54,81–87]. 5hmC is found throughout the genome, primarily in the sequence context of CpG; it is generally less abundant than 5mC, and is reportedly enriched at enhancer elements, promoters and intragenic regions [54,82]. The 5hmC base has been linked to the enigmatic process of DNA demethylation, where the 5mC base is converted into cytosine either through an active or passive mechanism, and this is where much of the more recent research has focused [88–97]. In this regard, 5hmC appears to be an intermediate in the various pathways that have been proposed to lead from 5mC to cytosine. It is unclear whether the 5hmC base performs functions independent of this process, and whether it is an epigenetic mark in its own right. Gene targeting in mice has shown that the TET1 and TET2 proteins are individually dispensable for embryonic development, but mutants show minor birth weight reduction and defects in the haemopoietic system respectively [98–100]. It will be interesting to see the effect of combined TET protein deficiency, where a greater decrease in 5hmC abundance would be expected. At the molecular level, 5hmC and 5mC are differentially bound by most methyl-CpG-binding proteins [96,101,102], and show differential effects on the stability of the DNA duplex , suggesting that conversion of 5mC into 5hmC may alleviate some of the activities of the 5mC base. Recently a mammalian protein that specifically recognizes 5hmC has been described, and corresponds to MeCP2 . This observation, together with the other aspects of 5hmC biology, is likely to be an active area of research in the near future.
DNA METHYLATION: MODULATOR OF CHROMATIN STRUCTURE
In addition to a detailed understanding of the DNA methylome, recent years have led to a more complete understanding of the molecular mechanisms by which this epigenetic mark influences transcriptional regulation. The function of DNA methylation has frequently been attributed to two mechanisms: (i) its negative effect on certain transcription factor–DNA interactions [21–23], and (ii) its attraction of methyl-CpG-binding proteins [4,24–26]. However, current evidence implies that different modification states of cytosine are interpreted by numerous mechanisms, suggesting that DNA methylation influences genome regulation in a multifactorial manner [61,62,104]. The list of proteins that show DNA-methylation-modulated binding to chromatin is growing and includes proteins with diverse functions . Many of these proteins, directly or indirectly, influence different aspects of chromatin structure, suggesting that this may be a major role of DNA methylation in genome regulation (Figure 3A). Indeed, DNA methylation has been suggested to contribute to diverse aspects of chromatin organization, including the genomic distribution of histone modifications such as histone acetylation [4,105], H3K4me3 , H3K27me3 , H3K36me3  and H3K9me3 [107,108]; histone variants such as H2A.Z [109,110], and the multi-functional CTCF  (Figure 3A).
DNA methylation shapes the chromatin landscape
Polycomb repressive chromatin complements DNA methylation
During X chromosome inactivation, Polycomb proteins are initially localized to ~150 strong sites concentrated within bivalent domains coinciding with CpG islands. PRC2 (Polycomb repressor complex 2) and histone H3 Lys27 methylation patterns are indicative of subsequent spreading to thousands of sites, most of which are intergenic, non-bivalent and lack CpG islands . Although gradient dispersal may be unique to X inactivation, it has also been proposed that PRC2 can sense the chromatin environment, and respond to local chromatin compaction upon transcription cessation. In this dense chromatin, regions of histone H3 activate PRC2 to regulate H3 Lys27 methylation . Microscopy studies have shown that PcG (Polycomb group) proteins concentrate into nuclear foci, called Pc bodies, whose number and size change upon cellular differentiation [113,114]. This suggests that PcG proteins may mediate the nuclear organization of their target genes. Several 3C (chromosome, conformation, capture)-based approaches have shown that PRC2 is involved in mediating long-range interactions in mammals, as well as Drosophila . Whether PcG proteins are directly responsible for mediating these physical interactions is not clear; CTCF and insulators have been shown to co-localize and contribute to the nuclear organization of Pc domains . Drosophila PcG proteins can remain associated with their DNA response elements through replication, possibly re-establishing the histone modifications on newly assembled unmethylated histones, as well as their nuclear organization .
DNA methylation and the Polycomb repression system
The local chromatin structures adopted by the epigenetic repression systems remain largely unknown, as functional models merely refer to their condensed nature to explain their gene repressive influence. The microscopically dense compartments in the nucleus were originally defined as heterochromatin, currently further categorized into constitutive (always condensed) heterochromatin and facultative heterochromatin, which is more dynamic and varies between tissues . Pericentromeric heterochromatin and inactive X chromatin are paradigms for constitutive and facultative heterochromatin respectively. These also contain archetypal examples for DNA methylation and Polycomb-repressed regions . The epigenetic signature of pericentromeric heterochromatin is not established until late during fetal development, even though it is present in ES (embryonic stem) cells derived from embryos, indicating developmental plasticity and a role in maintenance of epigenetic states . Nucleosome deposition and DNA methylation peaks at coding region boundaries suggest that DNA methylation can have effects at the local chromatin level [119,120]. PcG proteins were identified as developmental factors, but are now also known to be involved in controlling dynamics and plasticity of gene regulation, particularly during differentiation, by interacting with other components of the transcriptional apparatus .
DNA methylation and the Polycomb repressor system have been viewed as independent and parallel mechanisms of maintaining a cell-heritable memory of transcriptional repression . However, in recent years intriguing links have been drawn between the two systems, suggesting that these epigenetic mechanisms are in fact connected. This interaction promises to enhance our understanding of each process individually, while their inter-connectivity may influence the way we view their roles in development and disease. The Polycomb repressor system comprises chromatin-associated proteins that form distinct complexes with histone-modifying activity . Of central importance is the PRC2, a protein complex composed of four core components that include the histone H3 Lys27 methyltransferase EZH2 (enhancer of zeste homologue 2) . Both the PRC2 complex, and its signature histone mark H3K27me3, play an important and evolutionarily conserved role in the maintenance of transcriptional repression of a large set of genes, many of which are key regulators of embryonic development [123,125–127]. The first link between PRC2 and DNA methylation was the suggestion that the EZH2 protein may directly recruit DNA methyltransferases to establish aberrant DNA methylation in cancer . Indeed, genes aberrantly hypermethylated in cancer are frequently PRC2 targets in early development, suggesting that PRC2 may somehow pre-mark genes for cancer-associated DNA methylation . However, the generality of this mechanism is currently unclear as EZH2-mediated recruitment of DNA methyltransferases is not observed in all cancer cells, or in non-cancer tissues .
DNA methylation modulates Polycomb targeting
How mammalian Polycomb complexes are targeted to specific genomic regions is a long-standing enigma in the field and remains an active area of research. In the fruit fly Drosophila melanogaster, PcG protein complexes are recruited to chromatin by DNA elements called PREs (Polycomb-response elements) , which mediate the inheritance of silent chromatin states during development. However, examples of elements that function like fly PREs are rare in mammalian genomes [123,131,132]. Recently, a model has been proposed in which the PRC2 component PHF19 binds H3K36me3 via its Tudor domain, forming essential contact points that allow recruitment of PRC2 and H3K36me3 demethylase activity to expressed gene loci during their transition to a Polycomb-repressed state . Interestingly, it has emerged that DNA methylation influences the genomic targeting of the PRC2 complex, revealing it as an important piece of the mammalian Polycomb targeting puzzle (Figure 3B). The first clues came from studies that mapped PRC2 components in mammalian cells and revealed enriched binding at CGIs, regions that represent a large portion of the DNA methylation-free fraction of the genome [134–136] (Figure 3Bi). It was shown that de novo insertion of a CGI is sufficient to recruit PRC2 , as long as it is depleted of activating motifs , suggesting that Polycomb binding may be the default chromatin state for a CGI in ES cells. Importantly, endogenous CGIs containing high levels of DNA methylation are not bound by PRC2 . Other studies have also revealed a negative correlation between DNA methylation and the PRC2 signature histone mark H3K27me3 in mammalian genomes, both at single loci, including mono-allelic H3K27me3 at an imprinted region , and at the genome-wide level [53,75,138,139]. Specifically, it appears that high levels of DNA methylation and H3K27me3 do not co-exist at CpG-rich regions of the genome [138,140]. A negative correlation between the two marks has also been noted in plant tissues, implying deep conservation of this relationship . These observations led to the hypothesis that DNA methylation could be a determining factor in Polycomb targeting, by negatively modulating the binding of the PRC2 complex to chromatin. This idea was directly tested by mapping PRC2 binding and/or H3K27me3 upon perturbation of DNA methylation levels. Different laboratories have now shown in multiple contexts that experimentally reducing DNA methylation levels results in redistribution of PRC2 activity [106,136–138,142]. For example, deficiency in the de novo DNA methyltransferase Dnmt3a in mouse neural stem cells leads to loss of methylation from specific intergenic and intragenic regions . At certain expressed genes required for neurogenesis, loss of methylation from regions surrounding their proximal promoters is associated with increased binding of the PRC2 complex, increased H3K27me3 and PRC2-dependent transcriptional silencing that results in aberrant neurogenesis . This shows that an important role of DNA methylation at these genomic regions is to prevent PRC2-mediated repression of actively transcribed genes (Figure 3Bii). That study nicely demonstrated that the role of DNA methylation in transcriptional regulation is context-dependent and extends beyond its canonical role at gene promoter elements. Other studies have shown that global reduction in DNA methylation levels have a widespread effect on PRC2 targeting, with many regions showing increased PRC2 binding upon loss of DNA methylation [136,138]. The mechanistic basis for the inhibition of PRC2–chromatin binding by DNA methylation is currently unclear. PRC2 is proposed to associate with chromatin through multiple histone and DNA contacts  and it is feasible that one or more of these contacts are perturbed by DNA methylation. Indeed, in vitro experiments suggest that chromatin binding by PRC2 is directly attenuated by CpG methylation [104,106], a possibility that future experiments will need to investigate further.
This interplay between two epigenetic mechanisms raises interesting questions for future research. Is the difference in Polycomb targeting in CpG methylation-free organisms, such as Drosophila, due to the absence of restrictive activity of DNA methylation? Sequence-specific DNA-binding proteins have been more strongly linked to Polycomb targeting in flies than in mammals . What is the consequence of disease-associated changes in DNA methylation on PRC2 distribution and gene expression? In this regard, cancer-associated DNA hypomethylation and hypermethylation have been reciprocally linked to increased and decreased H3K27me3 respectively, raising the possibility that DNA methylation changes in cancer affect gene expression and pathology in as yet unappreciated ways [75,139].
Collectively, the observations described above have widened the perspective for how DNA methylation contributes to gene regulation and cell identity. They have demonstrated that DNA methylation has important regulatory functions in addition to its well-established repressive role at gene promoters, and therefore probably contributes to genome regulation in a multifactorial manner. Mechanistically, it is now clear that DNA modifications are an integral component of chromatin structure, and influence epigenetic landscapes through modulating the binding of a wide range of effector proteins to chromatin. Consequently, we are now a few steps closer to understanding the contribution that this epigenetic mark makes to the regulation of the genome during development, and the involvement of aberrant patterns of DNA modification in disease.
enhancer of zeste homologue 2
partially methylated domain
Polycomb repressor complex 2
We apologize to all colleagues whose work we have not cited due to space limitations. We thank Professor Javier Cáceres (MRC Human Genetics Unit, Edinburgh, U.K.) for encouragement in writing the present review.
Work in the laboratory of R.M. is supported by the Medical Research Council, the Biotechnology and Biological Sciences Research Council and by the Innovative Medicine Initiative Joint Undertaking [grant number 115001 (MARCAR project)]. The S.P. laboratory acknowledges support from the Biotechnology and Biological Sciences Research Council and the British Heart Foundation.