Twists and turns of DNA methylation

DNA methylation, the postreplicative transfer of a methyl group to the C5 position of cytosine bases, was the first epigenetic modification identified and has been intensively studied for more than half a century. By now it is clear that Dnmt1, the major eukaryotic DNA methyltransferase, faithfully maintains genome-wide methylation patterns and plays an essential role in the epigenetic network controlling gene expression and genome stability during development. However, the molecular mechanisms that ultimately control DNA methylation still remain elusive. This is, in part, attributable to the remarkable complexity of the DNA methylation reaction, the apparent involvement of several inter- and intramolecular protein interactions, and the limited structural information. The crystal structure of Dnmt1 presented in PNAS (1) now provides detailed insights into the inner workings and possible regulation of one of the most intriguing enzymes.

The mammalian Dnmt1 likely evolved by fusion of several ancestral genes joining a C-terminal catalytic domain with strong similarity to prokaryotic C5 DNA methyltransferases and several additional domains making up a large N-terminal region with regulatory functions (2, 3) (Fig. 1A). The multiple steps of the methyl transfer reaction have been worked out for the related prokaryotic enzymes in molecular detail (4, 5). First, the target cytosine is flipped out of the DNA double helix. Then, a conserved cysteine residue in motif IV (Fig. 1A) forms a covalent complex with the C6 cytosine position to activate the C5 position for transfer of the methyl group from S-adenosyl-l-methionine (SAM). Finally, the enzyme is released by β-elimination, and the methylated base is flipped back into the DNA double helix. Surprisingly, the catalytic domain of Dnmt1 alone, although it contains all conserved motifs of an active C5 DNA methyltransferase, was found to be inactive and to require the N-terminal region for activation (3, 6, 7). In vivo, Dnmt1 associates with the replication machinery via a PCNA-binding domain (PBD) and a targeting sequence [TS or replication focus TS (RFTS)] mediates association with heterochromatin (8–10) (Fig. 1A). This brief outline of the enzymatic mechanism as well as the intra- and intermolecular interactions already gives a taste of the rather convoluted working and regulation of Dnmt1.

Fig. 1.

Structural insights into the regulation of Dnmt1 function. (A) Dnmt1 domain structure and interacting factors. Both the TS domain and the CXXC domain, plus adjacent linker, have been suggested to have an autoinhibitory function by interfering with substrate DNA binding. PBD, proliferating cell nuclear antigen (PCNA) binding domain; TS, targeting sequence; CXXC, zinc finger domain; BAH1/2, bromo-adjacent homology domain 1/2; TRD, target recognizing domain. (B) Structure of Dnmt1 solved by Takeshita et al. (1) (PDB ID code 3AV4). The single domains are color-coded as in A. (C) Superposition of Dnmt1 structures illustrating the expected sterical clash between the TS domain, CXXC domain plus adjacent linker, and DNA. Note that the structure from Takeshita et al. (1) starts with the TS domain (pink; TS domain in surface representation), whereas the shorter structure from Song et al. (12) starts with the CXXC domain and is solved in complex with unmethylated DNA (DNA in blue, CXXC domain/autoinhibitory linker in black, and remaining structure in cyan; PDB ID code 3PT6).

The crystal structure of Dnmt1 that Takeshita et al. (1) report comprises the complete catalytic domain and most of the N-terminal regulatory region (amino acids 291–1,600), and it reveals the spatial arrangement and possible functional interactions of Dnmt1 domains (Fig. 1B). Interestingly, the N-terminal TS was found inserted into the DNA-binding pocket of the catalytic domain. This anchoring of the TS domain in the catalytic DNA-binding pocket seems to be based on complementary electrostatic surface potentials of the TS domain and the catalytic domain, and it is further stabilized by specific hydrogen bonds. In addition, hydrophobic interactions between the peptide stretch, connecting the TS and zinc finger (CXXC) domains, and the PCQ-loop at the catalytic center are suggested to stabilize the position of the TS domain by narrowing the entrance of the DNA-binding pocket. The authors conclude that the methylation reaction requires the release of the TS domain from the catalytic center to allow DNA substrate binding and, indeed, deletion of the TS domain lowered the activation energy for the methylation reaction. Recently, competitive inhibition of Dnmt1 activity by an isolated TS domain was also reported (11). Thus, both studies assign an inhibitory role to the TS domain.

Interestingly, the very recently reported structure of a shorter Dnmt1 fragment lacking the TS domain indicated an autoinhibitory role for the linker connecting the CXXC and bromoadjacent homology 1 (BAH1) domains (12). Based on this crystal structure and on biochemical analyses, the authors suggested that the autoinhibitory function of the linker is activated on binding of the adjacent CXXC domain to unmethylated DNA. Interestingly, this autoinhibitory linker was also positioned within the catalytic pocket, similar to the TS domain in the structure of the longer Dnmt1 fragment. Superposition of both Dnmt1 structures (Fig. 1C) clearly indicates that the TS domain, as well as the linker, collides with substrate DNA binding of the catalytic domain, suggesting that Dnmt1 undergoes several conformational changes during the entire methylation reaction. Which of these autoinhibitory interactions occur in vivo remains to be determined. Syeda et al. (11) pointed out that the TS binding might be stronger and likely to prevail over the linker interaction. Notably, precise deletion of the CXXC domain in the context of the full-length protein did not alter the maintenance activity of Dnmt1 in vitro and in living cells (13). In any case, for activation of DNA methylation activity, all inhibitory domains, including the TS domain and/or the linker between the CXXC and BAH1 domains, have to be displaced to allow substrate DNA binding by the catalytic domain.

Both autoinhibitory mechanisms would fit well with the maintenance function of Dnmt1. After DNA replication, Dnmt1 methylates cytosines on the newly synthesized DNA strand at hemimethylated palindromic CpG sites. For the maintenance of a given DNA methylation pattern, it is essential for Dnmt1 to recognize and methylate all hemimethylated sites efficiently, but it is equally important not to methylate any unmethylated sites. On the one hand, binding of the CXXC domain to unmethylated DNA seems to inhibit the catalytic domain, and might thus prevent the erroneous methylation of previously unmethylated CpG sites (12). On the other hand, the TS domain seems to inhibit the catalytic domain constitutively, suggesting the need for specific activating factors (1, 11). One good candidate is Uhrf1 (also known as Np95 or ICBP90) because it interacts with the TS domain, preferentially binds hemimethylated DNA, and is indispensible for Dnmt1 activity in vivo (14–16). Uhrf1 binding might induce conformational changes displacing the inhibitory TS domain, and thereby activate Dnmt1 at hemimethylated target sites.

The new Dnmt1 structure (1) also hints at an additional mechanism for the specific recognition of hemimethylated CpG sites. Based on superposition with the structure of the prokaryotic methyltransferase M.HhaI in complex with hemimethylated DNA (17), Takeshita et al. (1) fitted DNA into the catalytic pocket of Dnmt1. This model suggests that the large target recognition domain of Dnmt1 might be in close proximity to the putative DNA-binding pocket and specifically interact with the target cytosine base of hemimethylated CpG sites. Finally, Takeshita et al. (1) also report the structures of two Dnmt1 complexes with SAM and S-adenosyl-l-homocysteine (SAH). SAM serves as a methyl group donor in the methylation reaction and gets turned over to SAH on methyl group transfer. All three structures are very similar, only differing in the orientation of residue C1229, which is involved in covalent complex formation with the target cytosine. In the complex with SAM, residue C1229 faces the putative DNA-binding pocket, optimally positioned for interaction with hemimethylated DNA, whereas in the free and SAH-bound states, residue C1229 faces away from the putative DNA-binding pocket.

In summary, the new Dnmt1 structures (1, 12) show similar domain structures but very different spatial arrangements most likely representing alternative states and molecular snapshots of the different steps of the methylation reaction. The comparison of these Dnmt1 structures suggests that multiple conformational changes occur during inhibition and activation of the catalytic domain. Given the above outlined complexity of Dnmt1 function in mammalian cells, it is clear that additional crystal structures of alternative complexes and reaction intermediates combined with targeted biochemical and cell-based studies are necessary to unravel the molecular mechanisms of DNA methylation and its regulation in mammalian cells. In particular, interactions with regulatory factors, cross-talk with other epigenetic pathways, and chromatin dynamics are expected to play important roles in the regulation of DNA methylation in normal cell proliferation and differentiation as well as in development and disease.