Detection of DNA modifications by sequence-specific transcription factors

Abstract

The establishment, detection, and alteration or elimination of epigenetic DNA modifications are essential to controlling gene expression ranging from bacteria to mammals. The DNA methylations occurring at cytosine and adenine are carried out by SAM-dependent methyltransferases. Successive oxidations of 5-methylcytosine (5mC) by Tet dioxygenases generates 5-hydroxymethyl (5hmC), 5-formyl (5fC), and 5-carboxyl (5caC) derivatives; thus DNA elements with multiple methylation sites can have a wide range of modification states. In contrast, oxidation of N6-methyladenine by homologs of E. coli AlkB removes the methyl group directly. Both Tet and AlkB enzymes are 2-oxoglutarate- and Fe(II)-dependent dioxygenases. DNA binding proteins decode the modification status of specific genomic regions. This article centers on two families of sequence-specific transcription factors: bZIP (basic leucine-zipper) proteins, exemplified by the AP-1 and CEBPβ recognition of 5mC; and bHLH (basic helix-loop-helix) proteins, exemplified by MAX and TCF4 recognition of 5caC. We discuss the impact of template strand DNA modification on the activities of DNA and RNA polymerases, and the varied tendencies of modifications to alter base pairing and their interactions with DNA repair enzymes.

Graphical Abstract

Introduction

Distribution of DNA methylation.

DNA methylation in bacteria and archaea is common, occurring at ring carbon C5 of the cytosine, the exocyclic amino groups of cytosine (at N4) and adenine (at N6) [1, 2] (Fig. 1a–Fig. 1c). The great majority of bacterial and archaeal DNA methyltransferases (MTases) are associated with restriction-modification systems, where the MTase protects a cell’s own DNA from digestion by its restriction endonuclease [3, 4]. Restriction-modification systems are important for defense against bacteriophage predation [5, 6], though they play other roles as well [7, 8]. Bacterial ‘orphan’ MTases - which are not paired with a restriction endonuclease as part of a restriction-modification system [9] - are sometimes involved in chromosome replication, DNA repair, and epigenetic gene regulation [10]. Examples of such regulatory orphan MTases include the DNA adenine MTase (Dam) in Escherichia coli (Gammaproteobacteria), and cell cycle-regulated DNA MTase (CcrM) in Caulobacter crescentus (Alphaproteobacteria) which are, respectively, responsible for maintenance adenine methylation of GATC or GAnTC sequences (N=any nucleotide) immediately after their replication [11, 12]. Deletion of ccrM in C. crescentus is lethal [13], while deletion of dam in Escherichia coli compromises methylation-directed mismatch repair and affects initiation of chromosome replication [10, 14, 15]. Dam also modulates the binding of RNA polymerase and transcription factors [16–18]. In contrast, deletion of the orphan DNA cytosine MTase (Dcm) in E. coli causes no obvious phenotypic change [19], but may affect G:T mismatch repair through short-patch repair [20]. In addition, some MTases that are associated with restriction-modification systems can affect gene expression, for example through phasevarions [21–23].

Fig. 1.

(a-c) Three types of DNA methylation occurring at carbon C5 of the cytosine ring, or the exocyclic amino groups of cytosine (at N4) or adenine (at N6). Note that thymine (T), like 5mC, contains a methyl group at pyrimidine ring carbon C5. (d-g) Enzymatic reactions carried out by Fe(II)- and 2-oxoglutarate-dependent dioxygenases: E. coli AlkB (d) and its mammalian homologs (e), Jumonji protein lysine demethylases (f), and Tet enzymes (g). (h) Examples of proteins involved in binding of modified cytosine bases. For additional examples, see [141] in this issue.

The focus of this review is on mammals, but it is worth briefly noting that DNA methylation is widespread in plants [24], and variable in fungi [25]. In the largest family of Animalia, the arthropods, DNA methylation is also variable. For example, DNA 5-methylcytosine (5mC) plays important roles in honeybees [26], but is present at barely-detectable levels in Diptera (flies and mosquitoes) [27]. Drosophila was found to contain both 5mC [28] and N6-methyladenine (N6mA) [29]; however, both observations are controversial [30–32].

In mammals the epigenetic DNA methylation marks are generated and maintained by DNA cytosine-C5 MTases Dnmt1 and Dnmt3 family at CpG and/or CpA dinucleotides [33, 34]. 5mC was first reported in mammalian DNA nearly 70 years ago [35], whereas DNA adenine methylation was reported only recently. Low levels of N6mA in DNA have been reported in mouse [36, 37], human [38, 39], and human malignant brain tumor glioblastoma [40]. Other studies have failed to detect N6mA in mammalian genomes [41–43]. There is also uncertainty over identification of the mammalian enzyme(s) responsible for generating the methyl marks – DNA adenine MTase(s) have not yet been convincingly established.

Demethylation of DNA.

Mouse Alkbhl and Alkbh4 – homologs of E. coli AlkB, which removes certain alkyl adducts from both DNA and RNA [44] – are demethylases for N6mA [36, 37]. Dam MTases are commonly present among enteric bacteria and some of their bacteriophages [45]. In the 1960s, the DNA of T-even bacteriophages was found to contain glucosylated 5hmC in position of normal cytosine [46]. Nearly 40-years later, 5hmC was found in the genome of higher organisms [47, 48], and the source turned out to be oxidation of 5mC. Ten-eleven translocation (Tet) dioxygenases can, in three consecutive oxidation reactions, successively oxidize 5mC to 5hmC, then 5fC, and finally 5caC [48–50]. Significantly, the Tet-mediated 5mC oxidations use the same enzymatic 2-oxoglutarate- and Fe(II)-dependent reactions as do N-methyl nucleic acid demethylases including E. coli AlkB, its mammalian homologs, and the Jumonji histone lysine demethylases, and many other enzymes [51] (Fig. 1d–1g). However, the mechanisms of these dioxygenase demethylases of N-methylation involve the formation of a N-hydroxymethyl intermediate, followed by the spontaneous release of formaldehyde. However, formaldehyde is not generated during Tet-mediated 5mC hydroxylation, presumably due to the fact that cytosine ring carbon atom C5 is a relatively inert carbon. On the other hand, the carbon-carbon bond in 5hmC can be broken in vitro more readily than in 5mC, via nucleophilic attack at the ring C6 carbon by DNA cytosine-C5 MTases [52] or by exogenous thiols [53]. However, 5hmC stays as a stable modification in the cell, or is further modified to 5fC and 5caC in Tet-mediated oxidation reactions, producing other stable modifications that are increasingly chemically distinct from 5mC.

Using synthetic isotope- and (R)-2′-fluorine-labeled deoxyC and deoxy5fC derivatives, incorporated into the genomic DNA of cultured mammalian cells, 5fC conversion to cytosine occurs in vivo by direct cleavage of the C-C bond [54], though the putative catalytic factor(s) is unknown. DNA 5fC has been reported to function in nucleosome organization by forming a reversible covalent Schiff base bond to lysines of histone proteins [55]. This is potentially very important, since lysine residues are often involved in DNA binding, though such covalent lysine-5fC adducts have not been reported for any other DNA binding proteins in vivo or in vitro. However, indirect support for such covalent linkages comes from observation of DNA-histone crosslinks between lysine residues in histone tails and the 5-position methyl group of thymine, following oxidatively-induced rearrangement of a phenyl selenide derivative of thymine [56].

Modification-specific transcription factors

Several classes of mammalian transcription regulators have been characterized for methylation-responsive interactions with DNA [57–59] (Fig. 1h). These DNA binding proteins, for instance MBD proteins (see article by Bird in this issue), C2H2 zinc finger proteins (see article by Buck-Koehntop), SRA domain proteins (see article by Wong), and homeodomain proteins [57], act in response to diverse modifications states of cytosine, functioning as epigenetic sensors to direct downstream outcomes. At least two classes of DNA binding proteins (C2H2 zinc finger proteins and MBD) operate the same approach – a methyl-Arg-Gua triad – to contact methylated CpG dinucleotides (5mCpG) [59, 60]. This same triad also serves to mediate recognition of TpG, because, like 5mC, thymine has a methyl group attached to the pyrimidine ring carbon C5 (Fig. 1c). The human tumor suppressor protein p53 utilizes a methyl–Arg–Gua triad to interact with the TpG dinucleotide within its recognition sequence TGCC(C/T). Replacing TpG with 5mCpG increases p53 binding affinity; whereas exchange with unmodified CpG lowers p53 binding [58]. This fits a general tendency for TpG to be relatively well recognized by 5mCpG-specific transcription factors. In fact, targeted 5mC-to-T modification by APOBEC proteins [61] might serve during differentiation as means of “permanently methylating” given nucleotides [62, 63]. Here, we focus on two transcription factor families: basic helix-loop-helix (bHLH) proteins MAX and TCF4 in recognition of 5caC and basic leucine-zipper (bZIP) proteins AP-1 and CEBPβ in recognition of 5mC.

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

Enhancer-box (E-box) sequences have the consensus motif of 5’-CAnnTG-3’ (where N=any nucleotide). The bHLH transcription factors bind to E-boxes and recruit coactivator or corepressor complexes to regulate gene expression. A subset of these E-proteins recognizes sequence motifs having a central CpG dinucleotide (5’-CACGTG-3’); examples include the oncogenic MYC and its binding partner MAX [64]. The central CpG and the two outer CpA dinucleotides (on opposite strands) are conventional DNA methylation sites. MAX binds an unmodified E-box element, and methylation of the central CpG prevents its binding [65]; whereas oxidation to 5caC restores its binding to the level of unmodified cytosine [66]. Arg36 of MAX uses a 5caC-Arg-Gua triad to interact 5caC (Fig. 2a–2b). The sequence conservation between the bHLH domains of MAX and its binding partners, particularly the invariant corresponding Arg36 (Fig. 2c), implies that the ability to bind the two carboxylate groups of symmetrically-modified 5caC:G base pairs in the central CpG also be valid to most (if not all) MAX heterodimers.

Fig. 2.

Recognition of E-box sequences by bHLH transcription factors. (a-b) MAX Arg36 recognizes the central 5caCpG using two 5caC-Arg-Gua triads. (c-d) Sequence alignments of basic regions of (c) MAX-like and (d) TCF4-like transcription factors. (e-f) TCF4 uses Arg569 and Arg576 to recognize, respectively, two negatively-charged 5caCpG motifs immediately outside of an E-box.

Like MAX, the bHLH protein TCF4 can heterodimerize with other family members. Unlike MAX, the TCF4 binding sites are enriched with E-box motifs having variable central sequences (CAnnTG; N=any nucleotide) [67–69]. However, 15-25% of TCF4 binding sequences having CpG or CpA at locations outside of but adjacent to the E-box [63] (Fig. 2e). The rate of recurrence of Cp(G/A) at these locations is greater than expected (6.25%), suggesting a preference of TCF4 binding for these sites. Indeed, 5caC in a CpG dinucleotide immediately outside of an E-box (5’-CG-CAGGTG-3’) increased TCF4 binding [70]. The acquired interactions include two positively-charged residues of TCF4 (Arg569 and Arg576) that interact, respectively, with two negatively-charged 5caC bases of fully-modified CpG sites (Fig. 2f). The gained electrostatic interactions and hydrogen bonds lead to favorable binding of 5caC DNA.

Interestingly, the corresponding Arg569 and Arg576 of TCF4, particularly Arg576, are conserved in the TCF4 related subfamily members, but not in the MAX-related subfamily (Fig. 2d). The corresponding residues in MAX are a lysine (which would at least preserve appropriate charge-charge interactions) and a leucine. Similarly, the corresponding MAX arginine Arg36 in TCF4 is a hydrophobic valine. Thus, two different bHLH proteins utilize unique positively-charged arginine residues for interaction with 5caC residing at different locations of E-box sequences, either the central CpG or the CpG immediately next to the E-box. It is interesting that, among the TCF4-like transcription factors, the residue interacting with the cytosine immediately adjacent to the E-box (Arg569) is less conserved than the residue interacting with the cytosine one base farther away (Arg576; Fig. 2d–2e), which might allow the protein to recognize the modified CpA/TpG site at the corresponding positions. At least one of these proteins (TWIST1, having Gln in place of Arg576) does still recognize the bases flanking the E-box [71]; however, the effects of Arg569 substitution on binding effects of DNA modifications have not to our knowledge been determined for TWIST1 or any other TCF4-like proteins.

Basic leucine-zipper (bZIP) transcription factors.

Activator protein 1 (AP-1) is a classic bZIP family of DNA binding proteins involved in gene regulation and controls of cell proliferation, oncogenesis, and apoptosis [72, 73]. Like TCF4 and MAX, AP-1 is a dimeric complex involving homodimers and heterodimers of the Fos, Jun, MAF (musculoaponeurotic fibrosarcoma), and ATF (activating transcription factor) proteins [73]. AP-1 (Fos or Jun) binds a 7-bp sequence of TPA-response element (TRE: 5’-TGAGTCA-3’) or a methylated response element (meTRE: 5’-MGAGTCA-3’ where M=5mC) with a 5’ TpG dinucleotide being substituted by a methylated CpG [74, 75]. Both TRE and meTRE motifs contain two methyl groups (from the 5’ end) at nucleotide positions 1 and 5, generating four methyl groups symmetrically located at base pair positions 1, 3, 5, and 7 (Fig. 3a). The conserved di-alanine in AP-1 dimer make van der Waals contact with these methyl groups (Ala265 and Ala266 in Jun; Fig. 3b) [62].

Fig. 3.

Methyl-dependent and spatially-specific DNA recognition by bZIP transcription factors. (a) Aligned DNA response elements with spatially-equivalent methyl groups from T or M (5mC). (b-c) Recognition of the four spatially-constrained methyl groups of meTRE by the AA dipeptide of a Jun dimer (b), or of meZRE by the AS dipeptide of a Zta dimer (c). (d) Recognition of the six spatially-constrained methyl groups by the AV dipeptide of a C/EBPβ dimer. (e) The V285A mutant of C/EBPβ permits Arg289 to interact with T₂-Arg-G₃ triad. (f) Sequence alignments of basic regions AA dipeptide (Jun-like) or AV dipeptide (C/EBPβ-like).

A bZIP transcription factor (Zta) of Epstein–Barr Virus, a human-specific B cell-infecting gamma-herpesvirus [76, 77], was the first example of switching epigenetic silencing and activating gene transcription by a sequence-specific transcription factor that preferentially binds methylated cytosine bases within a specific sequence [78]. The unmethylated virion genome becomes heavily methylated during the latent stage of the virus cycle [79–81]. The Zta homodimer preferentially recognizing methylated promoters (known as meZRE) activates early lytic cycle, and examples of which are 5’-TGAGMCA-3’ and 5’-TGAGMGA-3’ [78, 82–84]. Like the AP-1 binding site, the meZRE elements contain two methyl groups at nucleotide positions 1 and 5 (Fig. 3a), with a methylated cytosine replacing one of the inner thymine residues of the AP-1 element, generating the four spatially-equivalent methyl groups. As in the case of AP-1, the corresponding Zta residues alanine 185 and serine 186 make van der Waals contact with these methyl groups (Fig. 3c) [62]. Serine at that position (Ser186) is unique to Zta among proteins in this family. The side chain hydroxyl oxygen atom of Ser186 contacts differently the two meZRE half sites, allowing the Zta dimer to bind asymmetric sequences, whereas the corresponding Ala266 of AP-1 Jun protein has no such flexibility.

Besides binding the 7-bp TRE motif (5’-TGA-G-TCA-3’), Fos or Jun proteins form heterodimers with ATF to interact with 8-bp sequence of cAMP-response element (CRE: 5’-TGA-CG-TCA-3’) [73]. In addition, CRE-binding proteins (CREB) also recognize the CRE elements [85, 86]. Between TRE and CRE elements, the difference is a one base-pair expansion next to the central C:G base-pair to generate a CpG site (Fig. 3a). The extended central CpG potentially permits the CpG modification to have a modulating role (e.g., see MAX above). This was tested using an artificial reporter construct, in which the promoter included one CRE element incorporating modifications of the central cytosine in a hemi-methylated CpG (5’-TGA-XG-TCA-3’/3’-ACT-GM-AGT-5’, where M=5mC) [87]. The results showed that modification of the top strand (X=5mC) and its oxidized forms (X=5hmC, 5fC and 5caC) all mildly impaired CREB binding and decreased gene expression [87]. The fact that all of these modification states inhibited binding suggests that steric effects, rather than charge or polar interactions, are dominant in this case.

Examination of chromatin immunoprecipitation sequencing data (by the ENCODE consortium [88]) from RNA polymerase II associated transcription factors revealed that many bZIP transcription factors (and TCF4-like bHLH proteins) have variable binding sites corresponding to the central CpG or reduced to a single G:C base pair, but maintain a conserved outer CpA/TpG sequence [89]. This observation suggests an essential modification-dependent CpA recognition on the part of bZIP transcription factors.

Another group of bZIP proteins, C/EBP (CCAAT/enhancer binding proteins)-related transcription factors have a consensus binding sequence of 5’-TTG-CG-CAA-3’, with a central CpG dinucleotide flanked by two CpA sites. Methylation of the Cp(G/A) sites produces a DNA motif with every pyrimidine having a methyl group at the ring carbon C5 position of thymine or 5mC. However, the modifications at CpA and CpG sites do not contribute equally to the binding affinity, measured with C/EBPβ, with the CpA modifications having a dominant effect [89]. Significantly, as with meTRE and meZRE elements, CpA methylation would restore the four spatially equivalent methyl groups at base-pair positions 1, 3, 5 and 7 (Fig. 3a). Further, the C/EBP binding sequence has additional methyl groups at base-pair positions 2 and 6 (T:A). In the corresponding Ala-Ala position, C/EBP family has a unique alanine-valine dipeptide (Fig. 3f). The larger side chain of valine allows C/EBPβ V285 interaction with both methyl groups at positions 2 and 3 (Fig. 3d). To mimic the conserved Ala-Ala in many members of the bZIP family (Fig. 3f), mutation of valine 285 with alanine (V285A) in the alanine-valine dipeptide unexpectedly showed binding preference for CpA methylation over the unmodified sequence by a factor of 90. The smaller side chain of V285A variant allows arginine 289 to interact with the TpG dinucleotide at bp positions 2 and 3 [89] (Fig. 3e). This T₂-Arg-G₃ triad for C/EBP recognition sequence is unique among the DNA elements shown in Fig. 3a.

Recognition of DNA modifications by polymerases

The impact of cytosine modification of the DNA template has been examined for both RNA polymerase and DNA polymerase [90, 91]. Human DNA polymerase β is not detectably affected by 5mC, 5hmC, or 5fC in the DNA template; though the enzymatic efficiency of dGTP incorporation opposite 5caC in the template position is decreased ~20-fold relative to unmodified cytosine [91]. A positively-charged lysine 280 and polar residue asparagine 37 in the polymerase active site directly interact with the carboxylate moiety of 5caC (Fig. 4a). The lysine-5caC interaction is similar to that observed by the DNA binding CXXC domain of mouse Tet3 (Fig. 4b) [92, 93], the enzyme whose dioxygenase activity generates 5caC modification. In contrast, both 5caC and 5fC in the DNA template slow down transcriptional elongation by RNA polymerase Pol II, which uses a glutamine to interact with the carboxylate group of 5caC (Fig. 4c) [90]. The polar but uncharged glutamine interaction with 5caC is similar to that observed by Wilms tumor protein WT1 (Fig. 4d) [94]. These observations suggest that sequence-specific transcription factors, Tet dioxygenase itself and polymerases, function as epigenetic sensors for DNA modifications.

Fig. 4.

Recognition of 5caC by (a) DNA polymerase β, (b) Tet3, (c) RNA polymerase II, and (d) Wilms tumor protein WT1.

The altered interactions between 5caC and DNA/RNA polymerases might be related to the mutagenic potential of 5caC and 5fC [95, 96]. Both 5fC and 5caC display an intra-base hydrogen bonding interaction between the exocyclic nitrogen atom at N4 and their formyl or carboxyl oxygen atoms, respectively. This intra-base hydrogen bonding is seen both in the protein-bound state [94] and in the protein-free state [97], and and may alter the hydrogen bonding strength of C:G base pair [97–99]. It is worth noting that a DNA repair enzyme TDG, named for its thymine DNA glycosylase activity, cleaves 5caC and 5fC when either is paired with a guanine, but not 5hmC or 5mC [100]. The tendency for mismatches to appear opposite 5caC and 5fC (but not 5hmC or 5mC) is possibly associated with different excision rate by TDG, though TDG has much faster rate on G:U mismatches [101, 102].

A potential link between 5hmC and DNA repair involves a protein named for its activity as an embryonic stem cell-specific 5hmC binding protein (HMCES). It was initially discovered as a 5hmC binder in a proteomics experiment, using double-stranded oligonucleotides of (GAT)₂•(GAC)₄•GAT sequence as a bait to pull down proteins from embryonic cell lysates [103]. HMCES represents the mammalian ortholog of the SOS response-assisted peptidase (SRAP) family, which exists in organisms from bacteria and yeast to human [104]. The HMCES SRAP domain was also reported to incise 5hmC-containing duplexes as an autopeptidase-coupled nuclease [105]. However, the SRAP-associated nuclease activity was not observed in a recent independent study [106]. Instead, the SRAP domain proteins of both E. coli and human were seen to form covalent bonds to abasic sites in single-strand DNA [106–108]. We note that a 5hmC DNA glycosylase activity in mammalian tissue was reported over 30 years ago [109]. If such glycosylase activity exists, it could excise a 5hmC base and generate an abasic site for subsequent binding by HMCES.

To detect genomic 5fC, in vitro chemical labeling with malononitrile or 1,3-indanedione has been used to generate two unnatural cytosine bases [110]. The resulting 5fC adducts are recognized by DNA polymerases as thymine instead of cytosine, causing C-to-T mutations during DNA replication [111].

N6-methyladenine in mammalian DNA

As described above, N6-methyladenine (N6mA) has recently been observed in mammalian DNA, and its functions remain elusive. Unlike cytosine modification at the ring carbon atom C5, N6mA is on the exocyclic amino group that is involved in Watson-Crick base pairing hydrogen bonds (Fig. 1a). Indeed, incorporation of N6mA into a DNA template caused RNA polymerase Pol II pausing due to lower stability and slower kinetics of base pairing, and the stability of the UTP base paring with N6mA is compromised [112] (Fig. 5a). The site-specific pausing at N6mA in the kinetics of incorporation of thymine into newly synthesized DNA strand by a modified phage polymerase underlies the ability of SMRT sequencing to detect methylated adenines [113]. Not surprisingly, like TDG, human and mouse homologs of E. coli repair enzyme AlkB (Alkbh1 and Alkbh4) are sequence-independent DNA N6mA demethylases [36, 37, 40].

Fig. 5.

Recognition of N6mA by (a) RNA polymerase II, and (b-d) YTH domains.

Currently, it is unknown how N6mA DNA methyl marks in mammals are generated, read or altered. In addition to its known activity of glutamine methylation of eukaryotic release factor eRF1 [114–116], mammalian HemK2 has been reported to be a DNA adenine-N6 MTase [38] (renamed as N6AMT1), as well as a histone H4 lysine-12 MTase [117] (renamed as KMT9). We recently analyzed the in vitro activities of HemK2/KMT9/N6AMT1, in complex with Trm112 (named after tRNA methylation protein), on the three substrates side-by-side and found that human HemK2-Trm112 is only active on protein glutamine and lysine, but not active on DNA [118].

The other potential candidate N6mA DNA MTases include members of a functionally diverse MT-A70 family of SAM-dependent MTases [119]. A DNA adenine MTase complex in cillates (a group of protozoans), consists of two MT-A70 proteins and methylates double-strand DNA [120]. [Both 5-methylcytosine and N6-methyladenine are found in unicellular eukaryotes [121]]. Mouse Mettl4 was recently reported to be responsible for N6-methyladenine deposition in genic elements, corresponding with transcriptional silencing [37] (though the in vitro enzymatic activity of Mettl4 was not reported). In addition, the human Mettl3-Mettl14 heterodimer possesses methyl transfer activity on adenine of single-stranded mRNA ([122] and references therein), whereas human Mettl5-Trm112 might be responsible for 18S rRNA adenine methylation [123]. We emphasize that many nucleic acid modifying enzymes are able to act on both DNA and RNA, including members of the AlkB family that directly dealkylate DNA and RNA [124], and family members of the Apobec cytidine deaminases [125]. Tet2, one of the ten-eleven translocation proteins initially discovered as DNA 5mC dioxygenases [48, 126], mediates oxidation of 5mC in mRNA [127, 128]. We expect the characterization of potential mammalian DNA N6mA MTase(s) to be an active research area for the coming years.

Regarding N6mA-specific binding, the better-characterized recognition of N6mA in RNA provided a possible readout mechanism. N6mA-specific RNA-binding YTH domain, a conserved 100–150-residue polypeptide [129], binds the N6mA methyl group in an aromatic ‘cage’ [130–134] (Fig. 5b). In addition, YTH-ssRNA complexes could form dimers through RNA:RNA base stacking [134] (Fig. 5c) or base paring (Fig. 5d). This unusual dimer assembly via the nucleic acids could have relevance for RNA-protein structures in general.

Conclusions and Future Directions

In the era of epigenomics, many of the challenges of interpreting DNA methylations states of both cytosines and adenines themselves, in addition to effects on transcription factor binding or polymerase activity, have been summarized [135, 136]. One of the greatest such challenges is deconvoluting tissue-specific and disease-specific methylation patterns, and varied degrees of methylation even within a specific tissue. In addition, new and more sensitive detection techniques, genomic sample preparation and potential contamination from methylated bacterial DNA, are prone to inaccuracies in measurement of rare noncanonical DNA modifications in metazoan genomes [32, 137]. These challenges are further complicated by subtle differences intrinsic to transcription factors themselves in their interactions with DNA modifications within a particular sequence [59, 138]. Many transcription factors bind to DNA with high affinity while recognizing sequences with high variability within sequence motifs [139, 140]. This property can be traced to the ability of specific protein residues to adopt alternative conformations to establish versatile hydrogen-bonds with some DNA bases, but not with others [139, 140].

We can add to the above the effects of modification states. A motif such as E-box sequences, with four methylatable cytosines on two different strands could, counting Tet effects, have 20 possible states (all combinations of C, 5mC, 5hmC, 5fC, and 5caC at each of the four positions). On top of that, there can be flanking methylatable sites, each one adding fivefold diversity in possible states. Finally, different transcription heteromers can respond to many of these states in distinct ways. Such a diversity of states complicates the role of scientists in understanding the functional consequences of protein interactions with modified DNA elements, but such complexity is likely to be necessary for the complex processes of differentiation and homeostasis in mammals, and other organisms as well.

Research Highlights:

• Basic leucine-zipper (bZIP) proteins, exemplified by AP-1 and CEBPβ recognition of 5mC.

• Basic helix-loop-helix (bHLH) proteins, exemplified by MAX and TCF4 recognition of 5caC.

• 5caC and N6mA of template strand DNA modification have impact on the activities of DNA and RNA polymerases.

• Modifications that alter base pairing strength can affect interaction with DNA repair enzymes.

• DNA elements with multiple methylation sites can have a wide range of modification states.