Methylation of DNA, along with modification of histones and positioning of nucleosomes, is one of the main epigenetic mechanisms operating across the genome. the nucleotide bases in RNA are subject to a wide range of chemical modifications. DNA, however, normally suffers only one major modification, methylation of cytosine to produce 5-methylcytosine (5-meC), although 6–7 adenines per million, averaged over the genome, are also methylated as N6-methyladenine. 5- methylcytosine base-pairs with guanine in exactly the same way as unmethylated cytosine, so any genetic information encoded in the nucleotide sequence is unchanged. However, the methyl group, located in the major groove of the double helix, can attract methyl-binding proteins that affect gene expression, either directly or through binding partners. Hence DNA methyltransferases and methyl DNA-binding proteins are important components of the regulatory landscape—for more detail, see the review by Schübeler (2015), PMID 25592537. Humans have three DNA methyltransferases, DNMT1, DNMT3A, and DNMT3B, plus two related proteins DNMT2 and DNMT3L (Table 1).
Table1. HUMAN DNA METHYLTRANSFERASES
There is no corresponding DNA demethylase to act as an eraser. Instead two mechanisms are available to demethylate DNA:
• Passive dilution—an original methylation signal can be diluted out if cells proliferate without methylating newly synthesized DNA;
• Oxidative demethylation—enzymes of the TET (ten-eleven translocation) family convert 5-meC into 5-hydroxymethylcytosine (5-hmC). The process can continue through 5-formylcytosine to 5-carboxycytosine, which is then excised from the DNA by thymine-DNA glycosylase or methyl-CpG-binding domain protein 4 (MBD4; Figure 1). Humans have three TET-family enzymes; they use molecular oxygen together with α-ketoglutarate and ferrous iron to oxidize 5-meC. There is ongoing debate about how far 5-hmC is simply an intermediate in this process and how far it constitutes a separate epigenetic signal in DNA.
Fig1. DNA demethylation. TET enzymes oxidize 5-methylcytosine (5-meC) in stages to 5-carboxycytosine (5-carboxy-C); this is excised from the DNA by thymine-DNA glycosylase to complete the removal of methylated cytosines. 5-hmC, 5-hydroxymethylcytosine; 5-formyl-C, 5-formylcytosine; α-KG, α-ketoglutarate.
Techniques for studying DNA methylation are outlined in Box 10.3.
The significance of CpG sequences
Methylation is largely restricted to cytosines that lie immediately upstream of guanines in so-called CpG sequences (the p represents the phosphate linking the two nucleosides). CpG dinucleotides occur symmetrically in the double helix. Opposite 5′-CpG-3′ in one strand there will be 3′-GpC-5′ in the complementary strand, so that reading in the usual 5′ → 3′ direction, both strands carry CpG (Figure 2).
Fig2. CpG sequences occur symmetrically in double-stranded DNA.
The significance of this is that it provides a mechanism for epigenetic memory. The DNMT1 methyltransferase specifically methylates cytosines in CpG sequences where the CpG in the complementary strand is already methylated (Figure 3). Thus when a cell divides, it reproduces the pattern of methylation from the mother cell in both daughter cells.
Fig3. Maintenance methylation. When DNA containing a methylated CpG sequence is replicated, the CpG in the newly synthesized strand is initially unmethylated. The DNMT1 DNA methyltransferase specifically methylates CpG sequences that are paired with methylated CpG, thus preserving the pattern of methylation that was present in the DNA before replication.
This cannot be the only mechanism that can allow epigenetic marks to be preserved from mother to daughter cell. Epigenetic memory functions in flies, worms, and yeast, which do not methylate their DNA. There is some evidence for transmission of specific histone marks through DNA replication and mitosis. The most likely candidates are pat terns of lysine methylation, which are known to be stable over hours or days in cells; acetylation marks, by contrast, often have half-lives of only minutes. It is known that Polycomb proteins (responsible for repressive histone marks) can remain bound to chromatin and DNA during DNA replication, at least in vitro. Histones from nucleosomes seem to be randomly distributed between parental and new DNA strands when the replication fork passes. Some proteins that recognize and bind to specifically modified histones also possess the ability to impose those same modifications on adjacent nucleosomes; this explains the tendency of some chromatin marks to spread, but could also suggest a mechanism for epigenetic memory independent of DNA methylation.
In some cell types, cytosines in other contexts may also be methylated (so-called CH, CHG, and CHH sequences, where H represents any base except G). This is particularly seen in embryonic stem cells and oocytes, but also in neurons, where CpA methylation builds up over time to equal the level of CpG methylation; it occurs at a lower level in other tissues. Note that non-CpG methylation is not symmetrical across the two strands of the double helix. Its significance is unclear, but non-CpG methylated sequences may bind methyl-DNA-binding proteins differently, and so carry a specific functional signal.
Methylation of DNA is normally a repressive signal. In plants and many invertebrates, CpG methylation is concentrated on repetitive sequences, including dispersed transposons and the satellite repeats characteristic of pericentric heterochromatin. DNA methylation at repeated sequences probably serves to repress transcription. This functions as a defense mechanism since unbridled transcription of transposons is a threat to the integrity of the genome. In mammals, CpG methylation is more pervasive, for example in the body of genes, especially in exons, and in intergenic sequences. Of the roughly 28 million CpGs in the human genome 60–90% are generally methylated. Less than 10% are located in CpG islands.
CpG islands are stretches of a few hundred base pairs of DNA where cytosines are unmethylated and, as a result, there has not been the depletion of CpG sequences over evolutionary time that is seen in the bulk of the genome. CpG islands are associated with genes. They are found at around 70% of promoters (high-CpG-density promoters), where they are normally unmethylated whether or not the associated gene is expressed. On occasion CpG islands can become methylated. This shuts down transcription, as observed on the inactive X chromosome or at imprinted genes (see below). Aberrant methylation of CpG islands at promoters, particularly those associated with tumor suppressor genes, is a general characteristic of cancer cells. In promoters without CpG islands, such CpGs as are present are normally methylated, without any apparent effect on transcription. Around 15% of CpG sequences show variable methylation that often correlates with tissue-specific functions. Figure 4 shows examples of the distributions of CpG methylation in different tissues across one 340 kb region.
Fig4. Patterns of CpG methylation may be person-specific, tissue-specific, or locus-specific. Each square represents a CpG sequence from a subregion (labeled A–D) of the major histocompatibility complex on chromosome 6, with the intervening nucleotides omitted. Squares are color coded to indicate the extent of methylation (see scale at bottom). Rows show the methylation pattern seen in different tissue samples. Region A shows diverse patterns of methylation both between different tissues and between samples of the same tissue from different individuals, whereas region B shows more tissue-specific methylation. Region C is largely unmethylated, and region D is almost completely methylated. (From Hermann A et al. [2004] Cell Mol Life Sci 61:2571–2587; PMID 15526163. With permission from Springer International Publishing AG.)
DNA methylation sometimes affects gene expression directly, in that some transcription factors such as YY1 fail to bind to methylated DNA. More generally the effect is mediated by proteins that contain a methyl-CpG-binding domain (MBD). One such protein, MeCP2, has been studied closely because loss of function causes Rett syndrome (OMIM #312750). This is a strange X-linked condition in which heterozygous girls develop normally for their first year but then regress in a very characteristic way. MECP2 mutations are normally lethal in males. DNA methylation proceeds normally in patients with Rett syndrome but the signals are not read correctly. MeCP2 protein is normally very abundant in neurons where it seems to be particularly involved in repressive regulation of unusually long protein-coding genes. Interestingly, MeCP2 recognizes methylated CpA but apparently not hydroxymethylcytosine, both of which accumulate in neurons over the timescale in which Rett symptoms appear. In a mouse model of Rett syndrome, restoring Mecp2 function caused even established Rett-like symptoms to regress, holding out hope that the human syndrome might be treatable.
DNA methylation shows striking changes during embryonic development
A recently fertilized oocyte has methylation differences at paternal and maternal alleles of many genes, reflecting the very different methylation profiles of the sperm and oocyte. The methylation patterns then show striking changes during embryonic development, summarized in Figure 5. The erasure of parental epigenetic settings that occurs shortly after fertilization is substantial, resetting the epigenetic information originating from the previous generation to achieve the relatively blank and versatile pluripotent state necessary at the onset of development. However, it cannot be complete because at imprinted loci, genes retain an epigenetic memory of their parental origin.
Fig5. Changes in DNA methylation during mammalian development. Drastic and often tissue specific changes in overall methylation accompany gametogenesis and early embryonic development. Note the breaks (slashes) in the developmental lines. PGCs, primordial germ cells.
The roles of RNA in regulation of gene expression
Table 2 detailed the many classes of noncoding RNA (ncRNA). Many of these have roles in gene regulation. Among the short ncRNAs, microRNAs are involved at the level of translation, as described below, while piRNAs and siRNAs have specialized roles in keeping transposons quiescent. Human cells contain many thousands of different long (>200 nt) ncRNAs. ENSEMBL (November 2017) listed 14,720 lncRNA genes. lncRNAs have many diverse roles in gene regulation; Table 10.3 shows some examples, and the paper by Ponting and colleagues (Ponting et al. [2009], PMID 19239885; see Further Reading) gives others. Some, like HOTTIP, HOTAIR, and FOXCUT, are cis-acting regulators of a nearby gene. They may activate or repress expression. lncRNA-COX2, and probably many others, forms trans-acting complexes that can affect expression of large numbers of genes. Some guide epigenetic regulators such as histone modifying enzymes or DNA methyltransferases to specific sites in the genome. Many are antisense transcripts that overlap a protein-coding gene and inhibit its expression. They may act by recruiting repressive protein complexes (ANRIL; see Table 3), but often the simple act of transcription prevents transcription of the target gene (KCNQ1OT1, SNHG14). XIST is an example of an RNA that acts in a spatially defined part of the nucleus, in this case on the future inactive X chromosome (see below). A frequent role of noncoding RNAs may be to act as scaffolds for assembling multiprotein complexes, and to define substructure within the cell nucleus.
Table2. THE ARBITRARY DIVISION OF NONCODING NONRIBOSOMAL RNAs INTO SHORT AND LONG CATEGORIES
Table3. EXAMPLES OF LONG NONCODING RNAS
Interactions among epigenetic mechanisms
As mentioned in the introduction to this chapter, the different processes involved in gene regulation are not independent linear systems, but form interacting networks. Methyl-DNA-binding proteins can recruit other proteins associated with repressive structures, such as histone methyltransferases or histone deacetylases, while methylated histones can recruit DNA methyltransferases (Figure 6). Transcription fac tor binding inhibits DNA methylation. Some long noncoding RNAs guide epigenetic regulators such as histone-modifying enzymes or DNA methyltransferases to specific sites in the genome.
Fig6. Mutually reinforcing histone and DNA methylation in inactive chromatin. Methylation of DNA attracts proteins that modify associated histone proteins. Histone deacetylation and methylation attract proteins that methylate associated DNA. In addition to the effects shown here, some histone methyltransferases and some long noncoding RNAs can interact directly with DNA methyltransferases. HDACs, histone deacetylases; MeCP2/MBD1, methylcytosine-binding proteins.
This raises the question of which epigenetic changes are the primary causes, and which are downstream effectors, or just consequences, of changes in gene expression. For example, the methylation changes shown in Figure 10.11 are very dramatic, but it is debatable how far they are causative of the events they accompany, rather than being downstream results of the mechanisms that cause differentiation. DNA methylation is certainly required for vertebrate development: animals deficient in DNA methyltransferase activity die at various stages of development, and mice that are specifically unable to methylate sperm DNA are infertile. Similarly, although pluripotent embryonic stem cells derived from blastocyst-stage embryos can grow normally even in the complete absence of DNA methylation, these cells immediately undergo apoptosis when stimulated to differentiate. However, although DNA methylation is a feature of vertebrates, flowering plants, and some fungi, as mentioned above it is virtually absent from many of the best-studied model organisms, including yeast, C. elegans, and Drosophila—yet the mechanisms of differentiation seem fundamentally similar in all these organisms.
This introduces something of a chicken-and-egg question—what is the primary determinant? The review by Smith & Meissner (2013) (PMID 23400093; see Further Reading) underscores the complexity of the interactions involved. Some evidence suggests that binding of transcription factors may be a primary event, particularly of so-called pioneer factors that seem able to bind to unmodified chromatin. This then triggers a self- reinforcing process. It has also been reported that housekeeping genes, which are relatively independent of external signals, rely more on histone modification, while signal-responsive genes depend more on transcription factor binding, which can act rapidly by relocating pre-existing transcription factors from the cytoplasm to the nucleus.
The new technologies of genome editing will allow answers to this question. The CRISPR/Cas system can be modified by fusing a defective Cas9 nuclease to one of the epigenetic writers or erasers. The guide RNA will then direct the modification to one specific location, allowing just the effect of that one change to be investigated. It would not be surprising if it turned out that a small minority of epigenetic modifications were causal (instructive), while the majority were downstream con sequences of the feedback mechanisms that reinforce the initial events to produce stable states.
