Mammalian DNA methyltransferases: new discoveries and open questions

Structure and mechanism of DNMT1

Similar to other mammalian DNMTs, DNMT1 comprises a catalytic C-terminal and regulatory N-terminal part (Figure 2A). The catalytic domain is conserved among all bacterial and eukaryotic DNA-(cytosine C5)-methyltransferases and all of them follow a common catalytic mechanism for transferring a methyl group from AdoMet to the target cytosine base (reviews: [5,6,12]) (Figure 2B). This mechanism involves target base flipping and binding of the cytosine into a catalytic pocket, hydrogen bonded to the side chain of a glutamate from a conserved ENV motif. The methylation reaction is initiated by the attack of a Cys residue from a conserved PCQ motif (PCN in the case of DNMT3 enzymes) on the C6-position of the aromatic ring, which is supported by acid/base catalysis of the glutamate. This step is followed by methyl group transfer to the C5-position and resolution of the covalent enzyme–DNA complex assisted by acid–base catalysis by an undefined base, perhaps a water molecule (reviews: [5,12]). The amino acid motifs containing these catalytic residues were identified in early sequence alignments and found to be conserved in all enzyme of this type (reviews: [12,13]). This general mechanism was recently confirmed in a kinetic isotope effect study with DNMT1, revealing that methyl transfer indeed occurs after the attack of the active site Cys residue and the methyl transfer reaction is the chemically rate-limiting step of the enzyme [14].

Different structural analyses of DNMT1 have provided strong evidence for an allosteric regulation of DNMT1 activity by autoinhibition and its interaction with chromatin modifications (review: [6]). The structure of DNMT1 in the absence of DNA showed that the RFTS domain binds to the catalytic domain occluding the DNA-binding site (Figure 3A). This indicated that DNMT1 must exist in at least two different conformations, a catalytically inactive state, in which the RFTS domain blocks the active site, and a catalytically active state, in which DNA is bound in the active site [15]. Dynamic autoinhibition by the RFTS domain was also observed in biochemical studies [16,17], further showing that weakening of RFTS domain binding to the catalytic domain increased the activity of DNMT1 [18].

The structure of a truncated DNMT1 containing the CXXC, BAH and catalytic domains showed that an unmethylated DNA specifically bound to the CXXC domain, but not to the catalytic domain [19]. Based on this observation, an additional autoinhibition model was proposed, in which binding of unmethylated CG sites to the CXXC domain would prevent their methylation, which was supported by kinetic data. However, the role of the CXXC domain in the specificity of DNMT1 is not yet clear, because another study showed that mutations in the CXXC domain that block its DNA binding did not change the preference of full-length DNMT1 for hemimethylated CpG sites [20].

The molecular interactions that confer specificity for hemimethylated CpG sites were revealed in the structure of a DNMT1 C-terminal fragment lacking the CXXC domain with hemimethylated DNA bound in a catalytically competent conformation [21] (Figure 3B). In this structure, DNMT1 forms several base-specific contacts with its target sequence at the methylated cytosine–guanine base pair, which were all shown to be essential for catalysis in a biochemical study [22]. The target cytosine is flipped out of the DNA helix and bound into a protein pocket, which presents the active site residues conserved among all DNA-(cytosine C5)-methyltransferases described above (Figure 3C). Side chains from the catalytic and recognition loops are inserted into the DNA through both grooves filling the space emptied by base flipping. The 5mC methyl group of the hemimethylated CpG site approaches a hydrophobic pocket formed by residues from the catalytic domain, which explains the preference of DNMT1 for hemimethylated sites. This study supports previous bio-chemical observations, which suggested that DNMT1 catalytic domain has an intrinsic specificity for hemimethylated DNA [20]. Additional conformational changes of the DNA were observed including flipping of another base, though these observations were not supported by biochemical studies [22].

Interestingly, the various structures of DNMT1 differ in the precise conformation of an α-helix following the active site loop, which either adopts a kinked (in inactive complexes) or straight conformation (in the complex of DNMT1 with hemimethylated DNA). A mutational study combined with MD simulations could identify two residues, mutations of which affected these conformational transitions and led to reduced catalytic activity [23] showing how dynamic conformational changes of DNMT1 are essential for catalysis.

UHRF1 interaction of DNMT1

UHRF1 belongs to the family of RING-finger type E3 ubiquitin ligases. The observations that UHFR1 co-localizes with DNMT1 and PCNA at replicating heterochromatic regions and that DNMT1 association with chromatin was lost in UHFR1 knock-out (KO) cells [24,25] were key discoveries, which stimulated many follow-up studies on the cross-talk between chromatin modifications, other chromatin factors and the activity of DNMT1. UHRF1 KO is embryonically lethal in mice and UHRF1-deficient embryos showed strongly reduced levels of genome-wide DNA methylation, indicating that UHRF1 has an essential role in the maintenance of DNA methylation [24,25]. These observations led to a model that UHFR1 recruits DNMT1 to replicated hemimethylated DNA to facilitate its efficient remethylation. The cooperation of PCNA and UHRF1 in DNMT1 targeting has been documented by showing that expression of a DNMT1–PCNA fusion protein can rescue DNA methylation in the absence of UHRF1 [26].

Like DNMT1, UHRF1 is a huge multidomain protein (review: [6]). The structural, mechanistic and functional details of the DNMT1–UHRF1 interaction have been subject of intense investigations showing a complex and multifaceted mechanism. UHRF1 stimulates the catalytic activity of DNMT1 by an interaction with the DNMT1 RFTS domain, which appears to open the autoinhibited conformation [17,27]. UHRF1 binds to hemimethylated DNA with its SET and RING-associated domain (review: [28]). Moreover, its tandem Tudor domain (TTD) and the plant homeodomain (PHD) bind H3K9me3 in a cooperative reaction [29,30]. The binding of UHRF1 to H3K9me3 is required for the localization of UHRF1 to heterochromatin and for the maintenance of DNA methylation, since a mutation in TTD, which prevents binding of UHRF1 to H3K9me3, abolished both functions [31,32]. However, most recent work showed that the knock-in of a UHRF1 gene with mutated H3K9me3-binding site into uhfr1 deletion cells led to an almost complete recovery of DNA methylation, indicating that the H3K9me3-binding site of UHRF1 alone is not essential for DNA methylation [37]. Disruption of an H3R2-binding pocket in the PHD domain of UHRF1 (which is needed for H3 tail binding) abolished DNA methylation by DNMT1 in cells as well [33]. Recently, methylation of H3R2 by PRMT6, an enzyme that is often up-regulated in cancer, was also shown to disrupt UHRF1 binding and maintenance DNA methylation [34]. Reductions in H3K9me2 and in expression of UHRF1 are observed during global DNA demethylation when ESCs are transited from the serum-to-2i medium, suggesting a connection between both processes [35]. Regulation of global DNA methylation via the UHRF1–DNMT1 axis by Zscan4 has also been implicated in the control of telomere elongation [36].

The E3 ligase activity of the UHRF1 RING domain ubiquitinates H3 at K14, K18 and K23 [33,38–40]. A ubiquitin interaction motif was identified in the N-terminal part of DNMT1 and shown to be essential for DNA methylation in vivo [33]. Recently, two papers showed structures of the DNMT1 RFTS domain sand-wiched by two ubiquitin moieties from a dual ubiquitinated H3 tail (at K18 and K23) [39,40] (Figure 4A). In the context of full-length DNMT1, this binding must lead to a rearrangement of the RFTS domain, which could trigger the stimulation of DNMT1’s activity, highlighting once more the central role of the RFTS domain-mediated autoinhibition in the regulation of DNMT1. Moreover, direct binding of the ubiquitin-like domain of UHRF1 to the same site in DNMT1 was also reported [40].

Regulation of DNMT1 by PTMs

In addition to ubiquitination, several other PTMs have been found on DNMT1 (review: [6]). For example, DNMT1 is monomethylated at K142 by SET7/9 mainly during late S-phase, which promotes proteasomal degradation of the methyltransferase in a cell cycle-dependent manner [41]. Very recently, it has been shown that the L3MBTL3 protein binds K142 methylated DNMT1 and triggers its ubiquitination and proteolysis [42].

The abundance of DNMT1 during cell cycle is also regulated by acetylation of the lysine residues in the GK linker of DNMT1 catalyzed by the acetyltransferase Tip60 [43,44]. GK acetylation leads to ubiquitination of DNMT1 by UHRF1, resulting in its proteasomal degradation at the end of DNA replication. In turn, histone deacetylase 1 increases the stability of DNMT1 [43,44]. This effect is meditated by the deacetylation of the GK linker, followed by binding of the USP7 deubiquitinase to DNMT1, which prevents its ubiquitination [45]. DNMT1 and USP7 form a stable, soluble complex that associates with UHRF1 as a trimeric complex on chromatin [46]. The crystal structure of a human DNMT1 fragment starting with the CXXC domain in complex with USP7 revealed that the interaction of both proteins occurs through the GK linker of DNMT1 and an acidic pocket near the C-terminus of USP7 [45]. Mutations of these acidic residues in USP7 and the absence of acetylation of the GK linker in DNMT1 disrupted the interaction between DNMT1 and USP7, leading to increased proteasomal degradation of DNMT1 [45]. Interestingly, binding of USP7 was also shown to increase the activity of DNMT1 in vitro [46]. However, recent data showed no change in DNMT1 levels in USP7 KO cells [47]. Furthermore, mutation of the GK repeat to GQ, which cannot be acetylated, did not have an effect on the stability and activity of DNMT1 in cells [47], thus refuting an essential role of USP7 and GK repeat acetylation in the regulation of DNMT1 activity and stability. It is unclear if these differences are due to the use of different cell types in these experiments, which is particularly important, given differences in the rate of proliferation and cell cycle control.

Structures and mechanism of DNMT3 enzymes

The DNMT3 family comprises three orthologs, DNMT3A, DNMT3B and DNMT3L, all of which share considerable sequence similarity (Figure 2A). DNMT3L lacks catalytic activity, but it stimulates DNMT3A and DNMT3B and regulates their multimerization and nuclear localization (review: [6]). A DNMT3A/3L C-terminal domain structure published in 2007 showed the formation of a linear 3L–3A–3A–3L heterotetramer, which is mediated through two distinct interfaces, a 3A–3A central ‘RD’ interface and two peripheral 3A– 3L ‘FF’ interfaces [52]. Additional work showed that all interfaces also mediate the self-interaction of DNMT3A (review: [6]). The linear heterotetramer contains two active sites (one in each DNMT3A subunit) in a distance corresponding to 8–10 bp of DNA. Methylation studies confirmed the preference of DNMT3A to co-methylate two CpG sites placed at this distance [52,53]. Recently, 8–10 bp methylation patterns were also observed for non-CpG methylation catalyzed by DNMT3A in ES cells and neurons [54].

Different structures of a longer DNMT3A fragment containing the ADD chromatin interaction domain (see below) led to the discovery of an allosteric regulation of the catalytic activity of this enzyme [55] (Figure 4B). One structure revealed a conformation, in which the ADD domain interacts with the DNA-binding region of the catalytic domain, thus autoinhibiting its activity (inhibitory binding site). A second structure was generated in the presence of an H3 tail peptide bound to the ADD domain. It showed the ADD domain bound to the catalytic domain at another site, which should not interfere with enzyme activity (allosteric binding site). Interestingly, binding of the H3 peptide to the ADD domain occurs by interaction with residues, which are involved in the autoinhibitory-binding interface. Therefore, peptide binding is only possible in the active conformation and this conformation is stabilized in the presence of the H3 peptide. Enzyme kinetics confirmed that binding of the H3 tail peptide to the ADD domain leads to the release of autoinhibiton and an activation of DNMT3A [55], in agreement with biochemical studies published earlier [56,57] (review: [6]).

Recently, the first structures of DNMT3A/3L heterotetramers with DNA were published [58] (Figure 5A). In one of them, two short oligonucleotides were bound to the two active sites, in the second, one DNA containing two CpG sites in a distance of 14 bp was used. The structures were obtained after the formation of covalent bonds between the active site Cys residues and modified bases introduced at the positions of the target cytosines. It is, therefore, unclear how the two site structure resembles a complex containing two CpG sites in 8–10 bp distance as found to be preferentially methylated in biochemical studies (reviews: [5,6]). As in other DNMTs, the target cytosine is rotated out of the DNA helix and placed into the active site pocket, where it interacts with the catalytic resides described above (Figure 5B). The cavity after base flipping is filled by V716 from the target recognition loop. The guanine base of the CpG site is recognized by R836, the loss of which resulted in a reduced preference for CpG site methylation in vitro and in cells [58]. While the DNA bound to the two active sites was relaxed, the part of the DNA connecting them in the double-site substrate revealed a sharp bend accompanied by a strong compression of the major groove. Several residues make contacts with the bases flanking the CpG dinucleotide. This explains the strong flanking sequence preferences of DNMT3A, which have been previously observed (reviews: [5,6]). Moreover, the structure demonstrates that the R882 residue in human DNMT3A, which is frequently mutated in AML cancers (review: [59]), is involved in a backbone DNA contact. This observation can explain the strong shift in flanking sequence preference of the R882H cancer mutation that was reported recently [60] and may contribute to the cancer-promoting effect of this mutation.

Recently, another unexpected activity of DNMTs was discovered by showing that different DNMTs (bacterial M.SssI and mouse DNMT3A catalytic domain) generate low levels of 3mC as a byproduct [61]. This modified base represents an alkylation damage of DNA, which in cells is repaired by ALKB2 family enzymes in an oxidative process. Indeed, it has been found that ALKB2 enzymes are evolutionarily strongly connected with active DNMTs in many species [61], highlighting the functional connection between them.

Multimerization and processivity of DNMT3 enzymes

As mentioned before, DNMT3A has two interfaces for hetero- and homomultimerization, and it can form dimers, tetramers and higher oligomers (review: [6]). Higher protein oligomers of DNMT3A can bind to more than one DNA strand oriented roughly in parallel [62], which could support binding of adjacent linker DNA regions in chromatin. Indeed, different studies recently showed a periodicity of co-methylation of CpG sites during de novo methylation in the range of ∼180 nucleotides [63,64], which roughly fits to the distance of two linker regions separated by a nucleosome, suggesting that a DNMT3 multimer could bind to and methylate adjacent linker DNA regions. A similar pattern was also observed in DNMT3 introduced non-CpG methylation [54]. Most residues in the two interfaces are conserved in DNMT3B, which is likely to form similar multimers, but this property has not yet been demonstrated. DNMT3L contains only one interface that restricts the DNMT3A/3L complex to a tetramer, which cannot oligomerize further [62]. Additionally, DNMT3A and DNMT3A/3L complexes multimerize on DNA next to each other [65]. In recent years, this multimerization has been shown to mediate cooperative DNA binding by DNMT3A [66] and to contribute to the efficient DNA methylation by DNMT3A in cells [67].

Interestingly, while DNMT3A binds cooperatively to DNA and formation of protein–DNA complexes prevents processive DNA methylation [66], recent work has demonstrated that DNMT3B does not bind to DNA cooperatively and it has a processive methylation mechanism [68], in agreement with an earlier study [69]. It was also shown that mutation of the arginine residue in DNMT3B analogous to R882 of DNMT3A had no effect on the catalytic activity or processivity of the DNMT3B enzyme. This study also suggests that the differences in the interface residues between DNMT3A and DNMT3B influence the distinct catalytic mechanisms of the two enzymes [68]. Future biochemical studies characterizing the interface residues of DNMT3B and/or its crystal structure analysis will be critical for understanding the distinct targets and biological functions of DNMT3A and DNMT3B. Moreover, both enzymes are known to interact and form mixed complexes [70], which adds another level of complexity to this question.

Chromatin interaction of DNMT3 enzymes

The N-terminal parts of DNMT3 enzymes contain two important chromatin interaction domains, the PWWP domain specific for binding of H3K36me3 and the ADD domain binding to the N-terminus of the H3 tail (review: [6]). ADD binding to H3 is disrupted by methylation or acetylation of K4 or phosphorylation of T3 and T6 [56]. The functional role of the readout of these chromatin marks was studied experimentally by designing of DNMT3A ADD variants that were no longer sensitive towards K4 methylation or T6 phosphorylation [71]. Indeed, expression of these DNMT3A mutants in cells perturbed the differentiation program by causing defects in lineage commitment during ESC differentiation. However, the pluripotency and self-renewal genes in these cells were decreased similar to that in wild-type cells. Interestingly, another study showed a critical role of DNMT3A in methylation of pluripotency gene enhancers that is required for pluripotency gene repression during ESC differentiation. This study revealed a prominent role of the ADD domain in the targeting of DNMT3 enzymes at enhancers of pluripotency genes, where LSD1-dependent H3K4 demethylation was necessary for localized DNMT3A binding and activation, which led to enhancer methylation [72]. Taken together, these two studies demonstrate the biological relevance of epigenetic cross-talk mechanisms and their importance during development.

Recently, the structure the PWWP domain ( present in DNMT3A and DNMT3B) bound to an H3 peptide containing H3K36me3 was solved [73] and confirmed the binding pocket previously identified in biochemical studies [74]. Using DNMT3B KO mouse ES cell lines, intragenic DNA methylation was shown to be deposited by DNMT3B [63,75]. This activity of DNMT3B was dependent on SETD2 deposited H3K36me3 through the interaction of PWWP with H3K36 methylation. Gene body methylation by DNMT3B was shown to prevent spurious transcription initiation in line with the general function of gene body H3K36 methylation [76]. Another important role of H3K36me3 binding to the PWWP domain was described in a study showing that DNMT3A and DNMT3B associate with most active enhancers in epidermal stem cells in an H3K36me3-dependent manner, suggesting that this binding is mediated by the PWWP domain [77]. Both proteins differ in their effects on enhancer DNA modification. DNMT3B was involved in enhancer body methylation, while DNMT3A co-operates with TET2 promoting enhancer DNA hydroxymethylation. Interestingly, both DNMT3A and DNMT3B were required for enhancer activity and enhancer RNA production, illustrating a dual regulatory potential of DNA methylation for gene repression and activation.

New genes and function of isoforms

Recent studies have shown that the Dnmt3b gene in mice underwent a duplication forming Dnmt3c, another active DNMT gene, which has a specific role in the silencing of retrotransposons in the development of male germ cells [78,79]. While it is clear that the human genome does not encode additional DNMTs, it will be interesting to see if similar careful inspections will also reveal more DNMT3 homologs in other mammalian subbranches.

In addition to gene duplications, there is potential for more functional specialization of DNMTs through the flexible use of alternative start sites and splicing isoforms. There are two known isoforms of DNMT3A, a longer DNMT3A1 and a shorter DNMT3A2, which is expressed from an internal start site [80]. It has been shown that DNMT3A2 is the dominant form in embryonic tissues, while DNMT3A1 is expressed in all tissues. Recent work revealed that the DNMT3A1 enzyme localizes to the shores of bivalent CpG island promoters, suggesting that the very N-terminal segment absent in DNMT3A2 interacts with other proteins and may have a role in targeting of DNMT3A1 [81].

For DNMT3B, several splicing isoforms have been identified, many of which are catalytically inactive due to deletions in the catalytic domain but some also contain deletions in the N-terminal part. Interestingly, one catalytically inactive splice variant (DNMT3B3) is the dominant isoform of DNMT3B expressed in somatic cells [82]. It was shown that inactive variants still interact with DNMT3A and DNMT3B, and they could regulate their catalytic activity [83]. Recently, it was shown that catalytically inactive DNMT3B isoforms could also bind to DNMT3A and stimulate its gene body methylation in somatic cells [84].

Non-CpG methylation by DNMT3 enzymes

Quantification and distribution of non-CpG methylation from genome-wide studies showed its enrichment in gene bodies and transposons at >2% of cytosines in neurons, >1% in frontal cortex and embryonic cells, while it is lower in most other cell types (review: [10]). Interestingly, non-CpG methylation in ES cells and in early development mainly occurs in a CAG context, while in differentiated neurons, CAC methylation is mainly observed (review: [10]).

Biochemical studies showed that non-CpG methylation can be introduced by DNMT3 enzymes, due to their relaxed specificity (reviews: [5,6]). However, DNMT1, the most active and important DNMT in mammals, shows an exclusive CpG specificity and is almost unable to methylate non-CpG sites [20]. Therefore, while CpG methylation is efficiently duplicated after each cell division, this is not true for non-CpG methylation, which must be generated de novo each time after DNA replication. Interestingly, although CNG sites are palindromic, DNA methylation at these sites is also not ‘maintained’ in the sense that in mammals no DNMT exists, which has specificity for hemimethylated CNG sites. Methylation analyses in DNMT KO cell lines clearly demonstrated that DNMT3 enzymes are responsible for non-CpG methylation [85]. Consequently, non-CpG methylation is mainly found in tissues with high expression of DNMT3 enzymes like embryonic tissues and neurons. In neurons particularly, the lack of cell division also favors the stable presence of non-CpG methylation. Studies in mouse showed that after re-expression of DNMT3A or DNMT3B in TKO ES cells, DNMT3A preferred introduction of CAC methylation, while DNMT3B was more active in CAG methylation [54]. This finding is in agreement with the general observation that DNMT3B has a role in early development and DNMT3A is highly expressed in brain tissues. In the brain, CpG and CpA methylation is observed in the gene bodies of long MeCP2 repressed genes [86]. Disruption of CpA methylation by conditional DNMT3A KO revealed that CpA methylation is critical for binding of MeCP2 and repression of long genes [87]. In another study, binding of MeCP2 to non-CpG methylated DNA was found to regulate gene expression in both directions (up and down) in adult mouse brain [88].