A Common Mode of Recognition for Methylated CpG

Three methylated DNA binding domains

The control of gene expression in mammals relies significantly on the modification status of DNA cytosine residues. DNA cytosine modification is a dynamic process catalyzed by specific DNA methyltransferases that convert cytosine (C) to 5-methylcytosine (abbreviated 5mC or M) [1,2]. The 5mC may then be oxidized or demethylated via either an enzyme-catalyzed active pathway [3–5] or a DNA replication-dependent passive process [6,7]. The best-known modified DNA-recognition domains are two that recognize methylated cytosine: methyl-binding domains (MBDs) recognize fully-methylated CpG dinucleotides [8,9], and “SET and RING finger-associated” (SRA) domains bind hemi-methylated CpG sites generated transiently by DNA replication [10,11]. Both MBD and SRA domains have been structurally characterized in complexes with 5mC [12–18]. A third class of mammalian proteins that recognize methylated DNA is the C2H2 zinc finger (ZnF) proteins, which preferentially bind to methylated CpG within a longer specific sequence [19]. Recently, ZnF DNA binding domains from two proteins, Kaiso and Zfp57, have been structurally analyzed in complex with their respective methylated DNA elements [20,21], allowing comparison to the other 5mCpG-binding proteins.

Kaiso is a ZnF protein of the BTB/POZ family (Broad-complex, Tramtrack, and Bric-a-brac, also known as POZ for “poxvirus and zinc finger”) [22], whereas Zfp57 belongs to the family of KRAB (Krüppel-associated box) domain-containing ZnF transcription factors [23,24]. Among the C2H2 ZnF proteins, KRAB ZnF transcription factors (KRAB-ZnFs) act mostly as chromatin-modulating transcription repressors [25]. This family has expanded greatly during vertebrate evolution: humans have about twice as many KRAB-ZnF genes as mice do [26]. There are many reviews on the variation of ZnF gene regulation between primate species [24,27–29], and on the general ZnF architecture associated with DNA binding [30]. Here we survey the two most recent structural and biochemical studies of ZnF proteins, and discuss the broader implications for recognition of DNA methylation.

Both MBD and the ZnF DNA-binding domains occupy the DNA major groove, which is generally in an ordered B-DNA conformation, and recognize specific bases through classical hydrogen bonds and specific interactions unique to each protein [12–14,20,21]. By contrast, the mammalian SRA domain flips 5mC from a hemi-methylated CpG out of the double helix and into a hydrophobic binding cage [15–18]. We do not consider the SRA domain in our analysis, due to its vastly different DNA recognition mode from MBD and ZnF. Here, we compare the structural modes of 5mC recognition by three MBD proteins (MeCP2, MBD1 and MBD2) and two ZnF proteins (Kaiso and Zfp57). While each protein has some idiosyncrasies in its interaction with DNA, we discovered that the ZnF and MBD proteins each form a conserved triad (5mC-Arg-G) for recognizing 5mC, despite their unrelated amino acid sequences and structures. Guided by the common structures of Kaiso and Zfp57, we find a conserved sequence motif, RH (an arginine prior to the first zinc-binding histidine), that may be responsible for recognition of methyl-CpG dinucleotides.

A conserved methyl-Arg-G interaction for 5mCpG and TpG dinucleotides

Zfp57 recognizes the sequence TGCMGCR (M = 5mC, R = A or G) [31]. Structural analysis of the complex between fully-methylated DNA and the DNA-binding domain of mouse Zfp57, which has two ZnF motifs in tandem, revealed that the 5mCs of the two DNA strands engage in very different interactions. In the top strand a layer of ordered water molecules enclose the 5mC methyl group (Figure 1a). By contrast, the methyl group of the bottom strand 5mC makes van der Waals contacts with the guanidino moiety of Arg178, which in turn makes bifurcated hydrogen-bonding interactions (via the O6 and N7 atoms) with the 3′ guanine (Figure 1b). We refer to this interaction as the 5mC-Arg-G triad.

Kaiso recognizes either two consecutive methylated CpG dinucleotides [22] or a specific intrinsically-methylated DNA element TCCTGCNA (T contains a methyl group at C5) [32]. Structures of Kaiso’s three-ZnF DNA binding domain, in complex with its methylated CGCG or TpG-containing cognate sequences, have recently been analyzed by X-ray crystallography and NMR spectrometry [20]. Strikingly, Arg511 of Kaiso interacts with the methyl-CpG and TpG dinucleotides in a similar fashion to Arg178 of Zfp57, forming a methyl-Arg-G triad [20] (Figures 1c–d). The triad is maintained for TpG because the thymine methyl group is in the equivalent position (5-carbon) to that in 5mC.

Moving from ZnFs to the unrelated MBD domains, these bind the symmetrical, fully-methylated CpG site using two arginines, each of which hydrogen-bonds one guanine in a bifurcated pattern (Figures 1e–f). Like Arg178 of Zfp57 and Arg511 of Kaiso, the two MBD arginines are also engaged in van der Waals contacts with the methyl group of the neighboring 5′ 5mC of the same DNA strand. Interestingly, Zfp57 and Kaiso each form one 5mC-Arg-G triad to bind the fully-methylated DNA in an asymmetric way, while the MBD domain forms two 5mC-Arg-G triads to symmetrically bind the palindromic fully-methylated CpG duplex (Figures 1e–f). For each triad in MBD1, the calculated binding energy difference between methylated and non-methylated CpG is approximately −0.8 ± 0.4 kcal/mol [33]. As an aside, the report by Zou et al. [33] on the molecular dynamics simulations of MBD-DNA complexes also suggested that cation-pi interactions play an important role in the binding, with Arg providing the cationic moiety and 5mC the pi electrons; however the high-resolution structure of Zfp57-DNA complex indicates that cation-pi interactions are, at the least, not universal in 5mC-Arg-G triads.

In addition, both families can use water-mediated bonding: Zfp57 (though not Kaiso) uses a layer of hydration for one of the two methyl groups, whereas MBD uses a tyrosine to mediate water-associated interaction with one of the two methyl groups that further enhances the methyl DNA binding [12,13] (compare Figure 1, panels a and e). In the latter case, the hydroxyl oxygen atom of the tyrosine mimics a water molecule in the network surrounding the methyl group. In summary, Zfp57, Kaiso and MBD proteins appear to use a conserved mechanism (5mC-Arg-G triad) to recognize 5mC, despite their unrelated protein sequences and structures.

That Kaiso uses the same mechanism to recognize both methyl-CpG and TpG dinucleotides implies that Kaiso does not distinguish T:A from 5mC:G at the corresponding base pair (Figure 1c, d). The sequence element TGCMGCR, which is recognized by Zfp57, further illustrates this 5mC/T equivalence. Specifically, this sequence contains a TpG dinucleotide at base pair positions 1 and 2 in addition to the MpG and, interestingly, Arg157 forms a thymine methyl-Arg-G triad at that location (Figure 1g). Furthermore, a third methyl-Arg-G triad can be involved in Zfp57-DNA binding: the seventh position of the recognized sequence can be either an A or G (or less often C) [31], and an A at that position would generate a TpG dinucleotide on the opposite strand. It happens that an A was used at position 7 in a structural study [21] and, strikingly, an identical thymine-Arg-G triad was found for the TpG on the bottom strand (Figure 1h). We next sought to assess the relatedness of Kaiso and Zfp57 use of the 5mC/T-Arg-G triad by determining how well their triad structures could be superimposed.

Structural comparison of ZnF DNA binding domains of Kaiso and Zfp57

The DNA sequences recognized by Zfp57 and Kaiso have some similar elements, facilitating structural comparison of their complexes. The Kaiso recognition sequence TGCC resembles the first half of the Zfp57 recognition sequence (TGCM of the top strand), and MGMG, the alternative Kaiso recognition sequence, resembles the second half of the Zfp57 cognate sequence (TGMG of the bottom strand) (Figure 2a). Thus, the DNA components of the Kaiso-DNA complex structures were superimposed onto the first- and second-half sequence elements recognized by Zfp57. Only the superimposition of the second-half 5′-TGMG-3′ resulted in alignment of ZnFs, with the second ZnF of Zfp57 superimposed on the first ZnF of Kaiso (Figure 2a), but this was sufficient to allow comparison of the DNA-methyl recognition.

As in previously-characterized ZnF structures [30], the base contacts are made by the side chains in the N-terminal portion of the helix, particularly those prior to the first zinc-binding histidine (Figure 2a). The amino acids at positions −1 to −8 (relative to the first zinc-binding histidine) are well positioned to make base contacts. Interestingly, pairwise comparison of the aligned zinc fingers (the second ZnF of Zfp57 bound to TpG and the first ZnF of Kaiso bound to MpG; Figure 2b) revealed only a single conserved arginine at the −1 position (RH). The RH dipeptide is also conserved in the first ZnF of Zfp57, which recognizes a TpG dinucleotide on the opposite strand (Figure 2a). Bidentate arginine recognition of the guanine in a TpG dinucleotide has been noted previously in the yeast transcription factor Ndt80 [34,35]. Thus the functionally and structurally conserved methyl-Arg-G triads are associated with a sequence motif (RH) that, while small, allows a bioinformatic search of ZnF motifs in specific protein families for candidate methyl-binding regions.

Predicting RH-ZnF proteins that recognize methyl-CpG dinucleotides

The limited sequence conservation between Kaiso and Zfp57 makes the prediction of methyl-CpG binding ZnF proteins more challenging. However, all structurally characterized methyl-CpG binding proteins (except the base-flipping SRA domain proteins) involve a 5mC-Arg-G triad (Figure 1), and the zinc-binding C2H2 residues can be located accurately, therefore ZnF proteins could be surveyed for those that contain the RH dipeptide at the appropriate position.

KRAB-ZnF proteins were the target of this analysis for the following reasons: first, humans have about twice as many KRAB-ZnF genes as mice [26], so these transcription factors seem likely to play significant roles in human biology. (This twofold human/mouse difference begs for an explanation, but none is currently in the literature and the topic is beyond the scope of this analysis.) Second, a substantial number of these genes are preferentially expressed in brain and embryonic cells, where epigenetic reprogramming of DNA methylation takes place [36,37]. Third, KRAB-ZnP proteins are in part responsible for the establishment during early embryogenesis of site-specific DNA methylation patterns that are maintained through development [38].

The C2H2 Zinc Finger Gene database (SysZNF) was searched for ZnFs containing RH motifs at the proper position [39]. Among the 330 human KRAB-ZnF proteins examined, 224 (about 2/3) of them have at least one RH-ZnF (Figure 3a, left panel). These KRAB-ZnF proteins contain tandem ZnFs ranging from 3 to 35 fingers (ZNF91 contains 34–35 fingers [40]), with a mode of around 11–13 fingers (Figure 3a, right panel). The same trend is observed for mouse KRAB-ZnF proteins (Figure 3b).

A few examples of RH-containing mammalian KRAB-ZnF proteins with known biological roles are shown (Figure 3c). ZFP57 mutations - either deletions or missense mutations - have been found in patients with transient neonatal diabetes [23]. Two missense mutations, which change base-interacting R157 to H and zinc ligand H186 to N, each abolishes DNA-binding activity [21]. The absence of Zfp57 at its binding site might provide an opportunity for other proteins to compete and further modify 5mC (such as Tet-catalyzed oxidation, see below), resulting in the hypomethylation phenomenon associated with these human ZFP57 mutations [23]. Like Zfp57 [31,41], ZNF274 recruits the histone H3 lysine 9 methyltransferase SETDB1 (SET domain, bifurcated 1) [42] via the co-repressor TRIM28 (tripartite motif-containing 28; also known as KAP1 for Krüppel-associated protein 1), an essential regulator of genomic imprinting [43]. PRDM9 (PR domain zinc finger protein 9), a major determinant of meiotic recombination hotspots, contains a SET domain responsible for histone H3 lysine 4 methyltransferase activity [44]. These examples further illustrate the coordinated chromatin controls involving DNA methylation and the lysine methylation status of histone H3 (at residues 4 and 9) [45]. Zfp568 (also known as Chato) regulates extraembryonic tissue morphogenesis [46]. Zfp809 restricts retroviral transposition in embryonic stem cells [47], and retroviral silencing has been suggested to be the ancestral role of KRAB-ZnFs [48]. The 18-base pair DNA sequence recognized by Zfp809 contains two CpG sites, and the single base pair change of C-to-G at the corresponding cytosine residue disrupted a reconstituted protein-DNA complex [47]. Regulator of sex limitation (Rsl1) regulates sex- and tissue-specific promoter methylation [49]. In summary, where the functions of the KRAB-ZnFs containing appropriate RH motifs are known, they are consistent with a role that is modulated by DNA methylation. This makes it more likely that many of the numerous RH-containing KRAB proteins of unknown function are also sensitive to CpG methylation (or specific for TpG dinucleotides). Furthermore, some of these RH-containing KRAB proteins may respond to the oxidative derivatives of 5mC.

Can KRAB-ZnF proteins recognize the oxidative products of 5mC?

The ten–eleven translocation (Tet) proteins can oxidize 5mC to 5-hydroxymethylcytosine (5hmC) [3], and then further oxidize 5hmC to 5-formylcytosine (5fC) and 5-carboxylcytosine (5caC) [4,5]. Both 5mC and 5hmC occur in almost all human tissues and cell types examined [50], but 5hmC is relatively enriched in embryonic stem cells [3] and brain [50–52]. The genomic levels of 5fC and 5caC may also be similarly enriched, as both are derived from 5hmC oxidation. Interestingly KRAB-ZnF genes are also expressed at higher levels in embryonic cells and brain [26], so it seems entirely justifiable to suspect that some of the KRAB-ZnF proteins might have gained the ability to bind the oxidative products of 5mC. Some support for this suspicion comes from in vitro binding study of Zfp57 itself.

On the one hand, Zfp57 binds most strongly to methylated (5mC) DNA, with decreasing affinity in the order 5mC > 5hmC > 5fC > 5caC [21]. The effects of 5hmC and 5fC on binding by Zfp57 are similar to that of demethylation (loss of methyl groups altogether), whereas 5caC is similar to non-specific binding (i.e., changing one or more base pairs from the recognition sequence). On the other hand, a single spatially-conserved Glu residue in Zfp57 and Kaiso may explain the discrimination against oxidative products of 5mC, and even provide a possible means to predict which RH-containing KRAB proteins might recognize these oxidative products. Zfp57 has a Glu at position 182, which spatially corresponds to Glu535 of Kaiso (Figures 2c–d). Like Kaiso Glu535, Zfp57 Glu182 forms a van der Waals contact with the methyl group of 5mC and one of its carboxylate oxygen atoms interacts with the N4 atom of the same 5mC base. Substitution of Glu182 with its corresponding amide (E182Q) had no effect on methylated DNA binding [21] because the mutant could maintain the same interactions with the 5mC. Surprisingly, this E-to-Q mutant loses the ability to discriminate against unmethylated DNA as well as 5mC oxidative products and, most dramatically, has ~100-fold higher binding affinity to 5caC DNA (in comparison to wild-type protein; Figure 2e). It is possible that the uncharged amide group of E182Q can interact more favorably with the carboxylate group of 5caC. This observation raises the possibility that some KRAB-ZnF proteins might preferentially recognize the oxidative products of 5mC.

Concluding remarks

Despite the large number of KRAB-ZnF proteins, few of them have known biological roles (Figure 3c), and even fewer have known target DNA sequences. The discovery that Zfp57 recognizes a TGCMGCR element provided one of the first examples, among KRAB-ZnF proteins, of DNA recognition sensitive to sequence as well as methylation status [31]. It should be a very high priority to experimentally determine the target DNA sequence for each KRAB-ZnF protein.

Here, we propose that the two-thirds of human KRAB-ZnF proteins containing RH-ZnF motifs might provide specificity for 5mCpG or its oxidized derivatives, within a sequence context determined by the other associated tandem ZnF fingers. Asymmetric binding by one arginine (Kaiso and Zfp57) might be related to the asymmetric nature of the TpG sequence (CpA on the opposite strand). By contrast, MBDs use two arginines for the symmetrical 5mCpG dinucleotide. This prediction provides a testable model of how methylated DNA is recognized by various DNA binding proteins, once the specific target sequence is known, for example, by ChIP-Seq analyses. Structural analyses will clearly continue to play a central role, synergistically with biochemical and genetic studies, in addressing the recognition code of 5mCpG or its oxidized derivatives. Importantly, with ~200 human KRAB proteins having appropriately-located RH motifs, if even a fraction of them turn out to be truly responsive to DNA methylation status then this would represent a large increase in the number of such known proteins.

Finally, use of the methyl-Arg-G triad to recognize both 5mCpG and TpG raises an intriguing possibility. Perhaps TpG dinucleotides are selected for when it is advantageous for a particular DNA sequence to be treated as if it is permanently methylated; the dual binding (5mC/T) triad mechanism provides that option.