Detecting and interpreting DNA methylation marks

Abstract

The generation, alteration, recognition, and erasure of epigenetic modifications of DNA are fundamental to controlling gene expression in mammals. These covalent DNA modifications include cytosine methylation by AdoMet-dependent methyltransferases and 5-methylcytosine oxidation by Fe(II)-dependent and α-ketoglutarate-dependent dioxygenases. Sequence-specific transcription factors are responsible for interpreting the modification status of specific regions of chromatin. This review focuses on recent developments in characterizing the functional and structural links between the modification status of two DNA bases: 5-methylcytosine and 5-methyluracil (thymine).

Introduction

Five chemical forms of cytosine have been found in the DNA of higher organisms, including unmodified cytosine (C), 5-methylcytosine (5mC), 5-hydroxymethylcytosine (5hmC), 5-formylcytosine (5fC) and 5-carboxylcytosine (5caC) (Figure 1a) [1–5]. While some of these modifications may affect the strength of base pairing C:G hydrogen (H) bonds [6–8], these chemical forms are equivalent in terms of base pairing specificity and protein coding. However, the various forms of cytosine can vary substantially in how they interact with transcription factors and other DNA-binding proteins, and in their influence on gene expression. There is great interest in the effects of these modifications in epigenetic regulation, development and differentiation, neuron function, and diseases [9,10]. In general, the post-synthetic modifications (or ‘epigenetic marks’) are added to cytosine in situ, following its incorporation into DNA in the unmodified form. DNA methyltransferases (Dnmts) modify specific cytosines to 5mC, usually within the sequence contexts CpG [11,12] or CpA [13,14•] (Figure 1b). A subset of these 5mC residues are then converted to 5hmC, 5fC, and 5caC in consecutive Fe(II)-dependent and α-ketoglutarate-dependent oxidation reactions by the ten-eleven translocation (Tet) dioxygenases [2–4].

Figure 1.

Methylation and oxidation in 5′-CpG-3′ and 5′-CpA-3′ dinucleotides. (a) DNA methyltransferases (DNMTs) convert C to 5-methylC (5mC). Tet dioxygenases then convert 5mC to 5-hydroxymethylC (5hmC), 5-formylC (5fC), and 5-carboxyC (5caC) in three consecutive Fe(II)-dependent and α-ketoglutarate-dependent oxidation reactions, without release of formaldehyde (Top). Tet dioxygenases can also convert thymine (5mU) to 5hmU, and potentially to 5fU and 5caU (Bottom). (b) CpG/CpG can be methylated on both strands; while TpG/CpA is intrinsically methylated on one strand (T, 5mU), and the C on the other strand may be methylated as well. (c) 5mC and 5mU (T) contain a methyl group (yellow sphere) in spatially equivalent positions at C5. (d) Dnmt3A and Dnmt3B can methylate cytosine in the context of a CpA/TpG dinucleotide. Tet dioxygenases can oxidize both 5mC and T (5mU).

Like 5mC, thymine contains a methyl group at C5 (Figure 1c). Thus CpA/TpG is an intrinsically hemimethylated DNA element. Cytosine methylation of CpA, by Dnmt3, generates a pseudo-symmetric fully-methylated 5mCpA/TpG dinucleotide. Further modification by Tet enzymes on both 5mC and T [15,16] can yield CpA/TpG sites having 5-hydroxymethyl (Figure 1d), 5-formyl or 5-carboxyl modifications on both 5mC and T.

In this review, we focus on the mechanisms for recognizing 5mCpG, 5mCpA, TpG, and their respective oxidative forms, in DNA. These modifications all protrude into the major groove of DNA, the primary recognition surface for proteins, and change its atomic shape and pattern of electrostatic charges. In principle, such changes can alter protein binding to specific recognition sequences in DNA, by strengthening interactions, weakening them, or abolishing them altogether. This, in turn, can modulate gene expression and thus affect cellular metabolism and differentiation and, on a broader scale, influence an organism’s development, aging and disease processes.

Several well-characterized classes of mammalian proteins interact with DNA in a methylation-responsive manner (Table 1). Two recent large-scale studies used SELEX-based approaches to reveal that DNA binding by proteins of the homeodomain, basic leucine zipper domain (bZIP), and tumor suppressor p53 families can be increased or decreased by cytosine methylation within the binding sequence [17••,18••]. There is thus a growing number of transcriptional regulators that have adapted to respond to different cytosine modification states, potentially acting as direct epigenetic sensors to instruct downstream events.

Table 1.

Structures of known DNA cytosine modification binding proteins

TF family	Protein name	PDB	DNA	Mode of interaction
MBD	MeCP2	3C2I	5mCpG	Methyl-Arg-G
	MBD1	1IG4	5mCpG
	MBD2	2KY8	5mCpG
	MBD4	2MOE/4LG7/3VXX	5mCpG
		3VYB	5hmCpG
C2H2 ZF	Kaiso	4F6M/4F6N	TpG/5mCpG
	Zfp57	4GZN	5mCpG
	Zfp57(E182Q)	4M9V	5caCpG
	Wf4	2WBU/4M9E/5KE6	CpG/5mCpG/TpG
	Wf4 (E446D)	5KEA	CpG
	Egr1	4R2A/4X9J	5mCpG
	WT1	4R2E	5mCpG
	CTCF	5T00	5mCpG and CpA
p53 bHLH-LZ (Other examples of 5caC binding)	p53	3Q05	TpG
	MAX	5EYO	5caCpG	Arginine
	WT1	4R2R	5caCpG	Glutamine
	Pol II	4Y7N	5caC	Glutamine
	Tet3 (CXXC)	5EXH	5caC	Lysine
bZIP	AP-1 (Jun)	5T01	T and 5mC	Van der Waals (vdw)
	Zta (Epstein-Barr virus)	5SZX	T and 5mC	vdw
Homeodomain	HOBX13	5EGO	5mCpG	vdw
	CDX1	5LUX	5mCpG	vdw
	CDX2	5LTY	5mCpG	vdw
	LHX4	5HOD	TAA7TA	vdw
Other candidates	MEIS1	5EGO/4XRM	TpG	Methyl-Arg-G
	HOXB1-PBX1	1B72	CpG and TpG	Vdw and methyl-R-G
	HOXA9-PBX1	1PUF	CpG and TpG	Vdw and methyl-R-G
	E2F8	4YO2	CCCGCC
SRA	UHRF1 (SRA)	2ZO0/3CLZ/2ZKD	5mC	Base-flipping
	UHRF2 (SRA)	4PW7/4PW5	5mC/5hmC
	SUVH5 (Arabidopsis)	3Q0B/4YGI	5mC/5hmC
TALE	dHax3 (X. campestris)	3V6P/4GJP	T and 5mC	vdw (Glysine)
Unmodified CpG recognition
	TLR9	3WPC	CpG	H-bond
	CFP1 (CXXC1)	3QMB	CpG
	MBD1 (CXXC3)	5W9Q	CpG
	ID AX (CXXC4)	5VC9	CpG
	RINF (CXXC5)	5W9S	CpG
	Tet1 (CXXC6)	6ASD	CpG
	Tet3 (CXXC)	4HP1/4HP3	CpC/CpG
	MLL1 (CXXC7)	2KKF/4NW3	CpG
	MLL2 (CXXC)	4PZI	CpG
	FBXL19(CXXC-PHD)	6ASB	CpG
RNA N6mA binding
YTH	YT521-B	2 MTV	N6mA	Aromatic cage
	YTH DC 1	4R3I
	YTHDF1	4RCJ
	MRB1	4U8T
	Pho92	4RCM

MBD proteins

Methyl-binding domain proteins (MBDs) recognize fully-methylated CpG sequences, in which both DNA strands contain 5mC (5mCpG/5mCpG) [19]. The MBD domain of MeCP2 binds the symmetrically methylated CpG site using two arginines, each of which H-bonds to a guanine [20]. These two arginines also each make van der Waals contacts with the methyl group of the neighboring 5mC of the same DNA strand and form what we termed ‘5mC–Arg–G triads’ [21] (Figure 2a).

Figure 2.

Examples of methyl-Arg-Gua recognition and 5caC recognition. (a) MBD domain of MeCP2. PDB codes are given, DNA recognition sequences are shown with M = 5mC, and key interactions of each DNA strand are shown above and below the recognition sequence. (b) Kaiso (ZF), (c) Zfp57 (ZF), (d) Klf4 (ZF), (e) Egr1 (ZF) and (f) WT1 (ZF). (g) Effects of binding of DNA with 5caC against Egr1 and WT1. (h) WT1 (Q369) recognition of DNA with 5caC. (i) The p53 proteins recognize TpG, which can be replaced by 5mCpG. (j) Klf4 mutant (D446) interacts with unmodified cytosine. (k) In CTCF (ZF), D451 and E362 recognize, respectively, unmodified cytosine at position 2 and 5mC at position 12. (l) Effects of methylation at specific sites on binding by CTCF. (m) MAX (bHLH) recognition of 5caC. (n) Pol II recognition of 5caC. (o) Tet3 CXXC domain recognition of 5caC.

MeCP2 also binds methylated 5mCpA/TpG sites [22–25,26•,27••], with similar affinity to that of fully-methylated 5mCpG sites [28•], suggesting that ‘methyl–Arg–G triads’ would be a more appropriate name. Likewise, the MBD domain of MBD4 binds dinucleotides containing fully-methylated 5mCpG, and even to G:T mismatches occurring in the CpG context (5mCpG/TpG) [29]. In all three cases (5mCpG/5mCpG, 5mCpA/TpG, and 5mCpG/TpG), there are two symmetrically-positioned methyl groups, suggesting that the MBD proteins simultaneously recognize the two methyl groups on opposite strands. The 5mCpA/TpG binding may allow MeCP2 to control transcription in a ‘rheostat-like’ manner, in response to Dnmt3a activity, fine-tuning the cell-type-specific transcription of genes that are critical for brain function [14•].

C2H2 zinc finger (ZF) proteins

Two decades ago, Holliday argued that ‘sequences longer than CpG would be necessary for the regulation of gene expression by methylation’ [30]. Indeed, the recognition of some sequence-specific transcription factors is blocked by cytosine methylation only within the context of a longer specific sequence (e.g., the proteins CTCF [31,32] and Myc [33]). By contrast, certain Cys2-His2 (C2H2) zinc finger (ZF) proteins bind preferentially to DNA when CpG sites embedded within their recognition sequences are methylated [34]. The structures of several ZF domains bound to 5mC-containing DNA have been solved, including the transcription factors Kaiso, Zfp57, Klf4 (Krüppel-like factor 4), Egr1 (growth response protein 1), WT1 (Wilms tumor protein 1), and CTCF [35–39,40••]. Kaiso recognizes either methylated CpG dinucleotides [41] or an unmodified sequence with a TpG in the place of 5mCpG [42]. Arg511 of the Kaiso ZF domain interacts with the 5mCpG and TpG dinucleotides in a similar fashion [35], forming a methyl–Arg–G triad [21] (Figure 2b) like the two arginines of MBD. Similarly, Zfp57 uses a pair of methyl–Arg–G triads to recognize 5mCpG and TpG dinucleotides on two DNA strands (Figure 2c).

Like Kaiso, the consensus-binding sequence element for Klf4 contains either CpG, which can be methylated, or TpG, which is intrinsically methylated on one strand (Figure 2d). Klf4 is one of the four Yamanaka reprogramming factors [43], and like the other three (Myc, Oct4, and Sox2), contains TpG/CpA in its consensus recognition sequence [44]. These two different categories of 5-methyl-bearing sites may play distinct roles in development: non-CpG (mainly CpA) methylation disappears upon induced differentiation of embryonic stem cells, and is restored in induced pluripotent stem cells by reprogramming factors [45] that recognize CpA (or 5mCpA)-containing sequences.

Both Egr1 and WT1 bind similar consensus DNA sequence containing two CpG sites, each having high affinity for the methylated form of the sequence (Figure 2e,f), but exhibiting much-reduced affinity when either 5hmC or 5fC is present. This indicates that both Egr1 and WT1 differentiate primarily between the oxidized and unoxidized forms of 5mC, rather than between methylated and unmethylated C [38]. By contrast, 5caC affects the two proteins differently, greatly reducing binding by Egr1 but not by WT1 (Figure 2g). This difference can be ascribed to electrostatic interactions. In Egr1, a glutamate (E354) repels the negatively-charged carboxylate of 5caC, while the corresponding glutamine of WT1 (Q369) interacts favorably with the carboxylate (Figure 2h). Analogous residues to Egr1 E354 are found in MBD proteins (D121/E137 of MeCP2), Kaiso (E535), Zfp57 (E182), Klf4 (E446) (Figure 2a–d), and p53 (D281) (Figure 2i, see below). Acidic residues, aspartate and glutamate, are used sparingly in protein–DNA interactions [46], in part because of unfavorable electrostatic interaction with DNA phosphate groups, but juxtaposing an acidic residue with cytosine can specifically repel the negatively-charged carboxylate of 5caC.

Aspartate for cytosine and glutamate for 5-methlycytosine in C2H2 ZFs

A phage display study of Egr1 (also known as Zif268) revealed that aspartate (D), rather than glutamate (E) in the wild type, distinctly prefers binding to unmodified cytosine [47]. This observation led Choo and Klug to comment that ‘The physical basis for the interaction of aspartate/glutamate and cytosine is not yet clear, since hydrogen bonding contacts between these groups have yet to be observed in zinc finger cocrystal structure’ [48]. In bacterial one-hybrid experiments, where only unmodified bases were present, D was again found to preferentially juxtapose to cytosine [49].

We tested D/E interaction with cytosine/5mC in two important, 5mC-responsive regulatory proteins: Klf4 and CTCF [40••,50•]. Klf4 is one member of the specificity protein/Krüppel-like factor (Sp/Klf) family of ZF transcription factors, and the 5mC-interacting E446 of Klf4 is invariant among family members [37]. In the corresponding position of mouse Klf1, the neonatal anemia (Nan) mouse carries an E-to-D mutation (E339D) [51]. As with Klf4, ChIP-seq data for wild type Klf1 has the consensus-binding sequence containing either C or T as the base recognized by the wild type (E339), while the mutant (D339) displayed increased specificity for (unmodified) cytosine [52•]. We generated a corresponding E446D mutant in Klf4. The wild type (E446) has roughly equal binding affinity for 5mC or T, while the mutant (D446) has a preference for unmodified cytosine, due to decreased affinity for 5mC [50•]. The carboxylate group of D446 forms H-bonds with unmodified cytosine’s exocyclic N4 amino group and with its ring carbon C5 atom (Figure 2j). Both of these interactions would favor cytosine over thymine, and the latter one could explain the D446 preference for unmodified cytosine. By contrast, the wild-type E446 interacts with neither the 5mC N4 nitrogen nor the thymine O4 oxygen (Figure 2d).

The CTCF protein influences global chromatin architecture by sequence-specific DNA binding, via a tandem array of eleven C2H2 ZFs, and is present at ~80 000 sites in mammalian genomes [53,54]. ChIP experiments uncovered a broad CTCF-binding motif that contains a 12–15 bp consensus sequence [55–57]. Comparison to bisulfite sequencing data of various human cell types indicated that ~40% of variable CTCF binding is linked to differential DNA methylation, concentrated at the two conserved Cyt positions (2 and 12) within the 15-bp recognition sequence [58]. CTCF binding to the H19 imprinting control region sequence was inhibited by DNA methylation at a single CpG site [31,32,59], corresponding to position 2. Two negatively charged residues, D451 and E362 of CTCF, recognize, respectively, the two invariant cytosines at positions 2 and 12 [40••] (Figure 2k). The binding affinity for the oligonucleotide methylated at C2 was drastically reduced (by a factor of 23), while affinity was slightly increased when methylated at C12 (by a factor of 1.5) [40••] (Figure 2l). The distinct effects of methylation at positions 2 and 12 on binding affinity are due to the amino acids used in the interaction, with D452 preferring C and E362 preferring 5mC (Figure 2k).

p53 protein

The human tumor suppressor protein p53 binds as a tetramer to two repeated DNA sites, with each monomer recognizing a pentamer repeat TGCC(C/T) [60]. Each p53 monomer uses a methyl–Arg–G triad to recognize the TpG dinucleotide (Figure 2i). Thus three very different structural classes of DNA binding domains, in MBD, C2H2 ZFs, and p53, use the same methyl–Arg–G triad [21]. Substituting TpG with 5mCpG results in the largest observed increase of binding affinity by p53; by contrast, substitution with unmodified CpG reduces binding by an order of magnitude [18••].

Basic-helix–loop–helix (bHLH) transcription factors

The basic-helix–loop–helix (bHLH) transcription factor MYC, together with its binding partner MAX, regulates gene expression by binding to enhancer-box (E-box) elements (5′-CACGTG-3′) that contain a central CpG dinucleotide. Methylation of the central CpG greatly inhibits DNA binding by MAX [33]. However, unexpectedly, MAX exhibits the greatest affinity for an E-box containing 5caC (which, as noted above, is negatively charged), and much reduced affinities for the corresponding 5mC, 5hmC or 5fC forms [61•]. For context, 5caC is found at low levels (1–10% that of 5hmC), and preferentially accumulates at enhancers and other distal regulatory regions [62–67]. Interestingly, MAX Arg36 recognizes 5caC using a 5caC–Arg–G triad (Figure 2m), and the nearest other MAX residues to the carboxylate group are Arg60, Arg33 and Lys40, which yield a basic environment ideal for electrostatic attraction to the two carboxylate groups of symmetrically-modified 5caC:G base pairs. Given the similarities between the bHLH domains of MAX and its binding partners, and the structural conservation of the critical amino acids whether partnered with MYC or other proteins, it is likely that the ability to discriminate among 5mC oxidation states applies to most (if not all) MAX heterodimers, and to the extended MYC family of bHLH transcription factors. Thus, 5caC has the potential to function epigenetically by repelling acidic residues of MBDs and ZFs, but not of neutral (uncharged Gln in WT1) or positive side chains (MAX). In addition, 5fC and 5caC in DNA retard transcriptional elongation by Pol II; we note that the polymerase II subunit Rpb2 also uses a glutamine to hydrogen bond with the carboxylate of 5caC (Figure 2n) [68]. These observations suggest that transcription factors, as well as Pol II itself, may act as direct epigenetic sensors for DNA modifications.

One of the mammalian 5mC dioxygenases, Tet3, has its binding sites enriched for a sequence motif (5′-TCACGTGA-3′) at the transcription start sites of genes, particularly those involved in lysosome function [69]. Interestingly, this 8-bp sequence motif contains an E-box, and perfectly matches the recognition sequence for TFEB, another bHLH transcription factor and the master regulator of lysosomal genes. It is not yet known if the TFEB recognition sequences are modified (methylated or 5-carboxylated) in cells, particularly those sites associated with Tet3. It is conceivable that the DNA-binding domain of TFEB, which resembles that of MAX, binds to 5caC generated by Tet3 within this sequence motif. Besides the dioxygenase domain, Tet3 contains a DNA-binding CXXC domain, which binds 5caC (Figure 2o), though in a different sequence context [69,70]. The relationship of the transcription factor TFEB with Tet3 echoes that of WT1 with Tet2, in that WT1 displays high affinity for 5mC or 5caC but much reduced affinity for 5hmC or 5fC [38]. WT1 and Tet2 physically interact with one another [71,72], so WT1 may recruit Tet2 to its targets (containing 5mC) and/or Tet2 — together with its product (5caC) — could recruit WT1.

Basic leucine-zipper (bZIP) transcription factor family

A classic basic leucine-zipper (bZIP) family transcription factor, activator protein 1 (AP-1), is substantially involved in gene regulation and controls such critical phenomena as oncogenesis, cell proliferation, and apoptosis [73,74]. Like MYC/MAX, AP-1 is a dimeric complex that comprises members of the Jun, Fos, ATF (activating transcription factor) and MAF (musculoaponeurotic fibrosarcoma) protein families [74]. AP-1 complexes include homodimers and heterodimers; for example, Jun/Jun and Jun/Fos activate a set of genes by binding a 7-bp element known as TRE (5′-TGAGTCA-3′), as well as a methylated response element known as meTRE (5′-MGAGTCA-3′ where M = 5mC) with a methylated CpG replacing the 5′ TpG dinucleotide [75–77]. Thus, the TRE and meTRE elements each contain two methyl groups at nucleotide positions 1 and 5 from the 5′ end, resulting in four methyl groups symmetrically positioned at base pair (bp) positions 1, 3, 5, and 7 (Figure 3a).

Figure 3.

Examples of methyl group recognition via van der Waals contacts. (a) Aligned DNA response elements with spatially equivalent methyl groups from T (5mU) or M (5mC). (b,c) Recognition of the four spatially constrained methyl groups of meTRE by Jun/Jun dimer (bZIP) or meZRE-2 by Zta dimer (bZIP). (d) Recognition of the two methyl groups of 5mCpG duplex by HOXB13 (homeodomain). (e) Homeodomain protein MEIS1 recognizes TpG via methyl–Arg–Gua triad. (f) HOXA9–PBX1 in complex with DNA. Both proteins could have preferential binding of methylated DNA. A methyl group (in yellow sphere) is modeled onto unmodified cytosine.

Epstein–Barr Virus (EBV) is a human-specific B cell-infecting gamma-herpesvirus [78,79]. An EBV transcription factor, Zta (also called EB1, BZLF1, or ZEBRA) was the first example of a sequence-specific transcription factor that preferentially binds methylated cytosine residues within a specific sequence, reverses epigenetic silencing and activates gene transcription [80]. The EBV virion genome is unmethylated, but becomes heavily methylated during the latent stage of the virus cycle [81–83]. Early lytic cycle activation depends upon the Zta homodimer preferentially recognizing methylated promoters containing meZREs, notable examples of which are meZRE1 (5′-TGAGMCA-3′) and meZRE2 (5′-TGAGMGA-3′) [80,84–86]. Significantly, this element contains two methyl groups at nucleotide positions 1 and 5 (Figure 3a), with a methylated cytosine in place of one of the inner thymine residues of the AP-1 element, resulting in the four spatially equivalent methyl groups. These methyl groups are in van der Waals contact with a conserved di-alanine in AP-1 dimer (Ala265 and Ala266 in Jun; Figure 3b), or with the corresponding Zta residues Ala185 and Ser186 (via its side chain carbon Cβ atom) (Figure 3c) [87•]. These analyses demonstrate a novel mechanism of 5mC/T recognition in a methylation-dependent, spatial and sequence-specific approach by bZIP transcription factors.

Besides binding 7-bp TRE elements (TGA-G-TCA), AP-1 Jun or Fos proteins can form heterodimers with ATF to recognize cAMP-response elements (CRE; TGA-CG-TCA) [74], which are also recognized by CRE-binding (CREB) transcription factors [88,89]. The difference between TRE and CRE elements is the one bp expansion in the central C:G bp to a CpG dinucleotide. Interestingly, CpG methylation decreased the affinity of ATF4, while replacing the outer T with 5mC led to greatly increased affinity [18••]. ChIP-seq data of CREB in hippocampal neurons identified a non-canonical CRE motif [90], where the inner A:T at bp-3 is replaced by G:C — methylation of the cytosine at bp-3 would restore the symmetrically positioned methyl groups. In addition, while methylation sensitivity was not seen for C/EBPβ binding to TTG-CG-CAA [18••], its binding was inhibited by 5hmC in the central CpG dinucleotide [91]. Indeed, the expanded central CpG dinucleotide potentially allows the CpG methylation/oxidation status to play a modulating role (e.g., see MAX above).

Homeodomain transcription factors

Homeodomain proteins are the family of transcription factors most-recently characterized as having preferential binding of methylated DNA [17••]. This class of proteins was enriched in factors playing central roles in embryonic and organismal development [92]. Crystal structures of the homeodomain proteins HOXB13, CDX1, CDX2, and LHX4 revealed a pair of hydrophobic residues making van der Waals contacts to the two methyl groups of duplex 5mCpG dinucleotides (Figure 3d) [17••]. HOXB13 also forms complexes with the TALE-class homeodomain protein MEIS1 [93], which recognizes a sequence motif containing two pairs of CpA/TpG sites (Figure 3e). MEIS1, like other proteins referred to above that recognize TpG, uses two neighboring arginines with each forming a methyl–Arg–G triad (Figure 3e). The observation made with HOXB13–MEIS1 complex could be extended to other TALE-mediated HOX complexes [18••], such as HOXB1–PBX1 [94] or HOXA9–PBX1 [95] (Figure 3f). Whether 5mCpG can replace the TpG, and the effects on MEIS1 or PBX1 binding, have yet to be tested.

SRA domains

The SET and RING finger associated (SRA) domains of UHRF1 and UHRF2 recognize hemimethylated CpG sequences, which contain 5mC in only one strand, such as arise during DNA replication [96–98]. UHRF1 (residues 124–628) binds to 5hmC DNA with >10-fold weaker affinity than to 5mC DNA [99], while UHRF2 (whether the SRA domain or full-length protein) showed small or no difference in binding affinity between the two modifications [100,131]. In addition, UHRF2 can directly and specifically bind A:5hmU in vitro (in contrast to A:5mU, which is A:T) [15]. The SRA domain uses a DNA recognition mode vastly different from that of other classes discussed above. Specifically, the SRA domain uses base flipping [101,102] to interact with DNA, similarly to DNA-modifying enzymes such as DNA methyltransferases, DNA glycosylases and Tet dioxygenases. Besides the SRA domain, UHRF1 contains a Tudor-PHD domain for binding one of three ligands: histone H3 methylated at Lys9 (H3K9, which is associated with methylated DNA) [103–107], an internal loop region of UHRF1 for allosteric regulation [108•], or DNA ligase 1 [109••] which is needed for the ligation of Okazaki fragments that are formed on the lagging template strand during DNA replication. UHRF1 also contains a C-terminal RING domain for E3 ligase activity that can ubiquitylate histone H3 [110,111], which is recognized by maintenance methyltransferase DNMT1 [112,113,114••]. Because UHRF1 is an important epigenetic regulator, maintaining DNA methylation and histone modifications in the cell, recent cancer research supports expression levels of UHRF1 serving as a universal diagnostic and prognostic biomarker [115], suggesting that better understanding of its activities is important.

N6-methyladenine in mammalian DNA and RNA

This section is brief due to space restrictions, but it has recently become clear that mammalian DNA can be methylated on adenines at the N6 position (N6mA) [116•,117•,118•]. This methylation is presumably subject to oxidative removal by Tet-like dioxygeneases [119]. Unlike the case for cytosine, however, Tet action on N6mA results in restoration of adenine (direct demethylation), involving formation of an N6-hydroxymethyl intermediate followed by spontaneous release of formal-dehyde. This differs from the series of oxidative states on 5mC, required because the cytosine C5 atom is an inert carbon. Very little is known about the ways in which N6mA DNA methyl marks are established, maintained, altered, or read in mammals, and which DNA-binding proteins are sensitive to those marks. We expect this to be a very active area of research over the next several years. However some hints as to a possible readout mechanism are provided by the better-studied detection of N6mA in RNA. This often involves an N6mA methyltransferase complex (METTL3–METTL14) [120,121], demethylases/dioxygenases (FTO and ALKBH5) [122,123], and N6mA-specific RNA-binding YTH domain proteins [124–127]. The YTH domain is a conserved 100–150-residue polypeptide [128], in which the N6mA methyl group is bound within a ‘cage’ of aromatic residues [129].

Summary

It is well established that methylation patterns are replicated, following semi-conservative DNA replication, via the selective recognition of hemi-methylated CpG dinucleotides at replication forks by DNMT1/UHRF1 complexes. This involves reading methyl marks on both DNA and associated histones, and mono-ubiquitination of histone H3. However, an unresolved fundamental question is how, and indeed whether, the pattern of oxidized 5mC derivatives is propagated at CpG and CpA sites. Many DNA binding proteins recognize consensus-binding elements, containing either CpG/CpG (which can be methylated on both strands), or TpG/CpA (which is intrinsically methylated on one strand, may be methylated on the other strand, and can be further modified on both strands). Because the methyl–Arg–G triad recognizes both 5mC and T, perhaps TpG dinucleotides are selected for when it is advantageous for a particular DNA sequence to be treated as if it is permanently methylated; this might even occur via targeted 5mC deamination [130] during development. On the other hand, the substitution of T by C (as in the cases of AP-1 and Zta) provides an opportunity for regulation by methylation and demethylation.

Transcription factors have adapted to respond to different states of cytosine modification, and gene activity is controlled on a finer ‘dimmer’ by these modifications than a simple ‘on’ or ‘off’ switch. For example, the DNA binding bHLH domain of MAX is responsive to all forms of modified CpG within its E-box recognition sequence, displaying the highest affinity for cognate sequences containing a central 5caC or unmodified C, with reduced affinity for 5fC and much lower affinity for 5mC or 5hmC. Progressive Tet-mediated oxidation of 5mC may thus be a way to titrate transcriptional activity in a graded and reversible fashion, with the 5mC form of such sites being the ‘off’ position and Tet-mediated oxidation steps a way of progressively increasing affinity for the binding site while moving towards 5caC. As many promoters are controlled by multiple transcription factors, several of which can have different patterns of methylation responsiveness, a very large number of combinatorial regulatory outputs is possible. In sum, a growing number of transcriptional regulators are being recognized as being responsive to different cytosine modification states, potentially acting as direct epigenetic sensors to instruct downstream events.