Crystal Structure of Human DNA Methyltransferase 1

Abstract

DNA methyltransferase 1 (DNMT1) is responsible for propagating the DNA methylation patterns during DNA replication. DNMT1 contains, in addition to a C-terminal methyltransferase domain, a large N-terminal regulatory region that is comprised of an RFTS (replication foci targeting sequence) domain, a CXXC zinc finger domain and a pair of BAH (bromo adjacent homology) domains. The regulatory domains of DNMT1 mediate a network of protein-protein and protein-DNA interactions to control the recruitment and enzymatic activity of DNMT1. Here we report the crystal structure of human DNMT1 with all the structural domains (hDNMT1, residues 351–1600) in complex with S-adenosyl-L-homocysteine at 2.62 Å resolution. The RFTS domain directly associates with the methyltransferase domain, thereby inhibiting the substrate binding of hDNMT1. Through structural analysis, mutational, biochemical and enzymatic studies, we further identify that a linker sequence between the CXXC and BAH1 domains, aside from its role in the CXXC domain-mediated DNMT1 autoinhibition, serves as an important regulatory element in the RFTS domain-mediated autoinhibition. In comparison with the previously determined structure of mouse DNMT1, this study also reveals a number of distinct structural features that may underlie subtle functional diversity observed for the two orthologues. In addition, this structure provides a framework for understanding the functional consequence of disease-related hDNMT1 mutations.

Graphical abstract

Introduction

DNA methylation is one of the key epigenetic mechanisms that are essential for transcriptional silencing of retrotransposons (1–3), genomic imprinting (4), and X-chromosome inactivation (5). DNA methylation in mammals mainly occurs at the C-5 position of cytosine within the symmetric CpG dinucleotide and affects ~70–80% of the CpG dinucleotides throughout the genome (6). The DNA methylation patterns are established by the de novo DNA methyltransferases DNMT3A and DNMT3B during gametogenesis and early embryogenesis, and become stably inherited by the maintenance DNA methyltransferase DNMT1, in cooperation with DNMT3A and DNMT3B (7–10). DNMT1-mediated maintenance DNA methylation is supported by both its substrate preference toward hemimethylated CpG sites (11,12), and its recruitment to DNA replication foci (13,14), through its interactions with proliferating cell nuclear antigen (PCNA) (15) and histone H3 ubiquitinated at lysine 23 (16).

DNMT1 is a multi-modular protein that is comprised of ~1,620 amino acids. It contains a C-terminal methyltransferase (MTase) domain and a large N-terminal regulatory region, linked by a conserved (GK)n dipeptide repeat. The N-terminal region of DNMT1 is composed of an RFTS (replication foci targeting sequence) domain, a CXXC zinc finger domain, and a pair of BAH (bromo adjacent homology) domains. These N-terminal domains distinguish DNMT1 from its bacterial counterparts and impose a tight control on the recruitment and enzymatic activity of DNMT1 (7,17). To elucidate the regulatory mechanisms of DNMT1, we have previously determined the crystal structures of a C-terminal fragment of mouse DNMT1 (residues 650–1602, mDNMT1_650–1602) in complex with an unmethylated CpG DNA, and the equivalent human DNMT1 (hDNMT1_646–1600)-DNA complex, at 3.0 Å and 3.6 Å resolution, respectively (18). These structures reveal that the DNMT1 CXXC domain specifically binds to the CpG dinucleotide, which helps position the CXXC-BAH1 domain linker (a.k.a. autoinhibitory linker) into the catalytic cleft of the MTase domain, thereby forming an autoinhibitory conformation. This observation, together with mutational studies and enzymatic activity assays, suggests that the DNMT1 CXXC domain plays an inhibitory role in DNMT1-mediated de novo methylation. In a separate study, Takeshita et al (19) has determined the structure of a longer mDNMT1 fragment (residues 291–1620, mDNMT1_291–1620), free and in complex with cofactor S-adenosyl-L-methionine (AdoMet) or cofactor product S-adenosyl-L-homocysteine (AdoHcy). The structure of mDNMT1_291–1620 reveals that the RFTS domain forms a DNA competitive inhibitor through direct interaction with the MTase domain of mDNMT1. This observation suggests that the RFTS domain plays an inhibitory role in DNMT1-mediated methylation, and has subsequently been supported by a number of biochemical and cellular studies (19–22).

Human DNMT1 and its mouse orthologue share about 80% sequence identity, which raises a question about whether the structure of mDNMT1 can fully recapitulate the functional state of hDNMT1. Indeed, previous enzymatic analysis of full-length recombinant hDNMT1 and mDNMT1 proteins over the poly(dI-dC). poly(dI-dC) substrates have revealed a 30-fold difference in their k_cat values (23–25). Our recent studies have also shown that mDNMT1_650–1602 and its human counterpart, hDNMT1_646–1600, methylate a 14-mer hemimethylated CpG DNA with 2–3 fold different k_cat and K_m values (18). Together, these observations suggest that subtle functional diversity may exist between the two orthologues. In addition, the relative low resolution (2.75 – 3.25 Å) of the structure of mDNMT_291–1620 limited detailed analysis of protein interactions, such as the water-mediated hydrogen bonding. In this study, we report the crystal structure of a long C-terminal fragment of hDNMT1 (residues 351–1600, hDNMT1_351–1600) in complex with AdoHcy at 2.62 Å resolution. This structural study, combined with mutational, biochemical and enzymatic analysis, provides deep insights into the allosteric regulation of hDNMT1 and its structure-function diversity from mDNMT1. Importantly, this study reveals that the DNMT1 CXXC-BAH1 domain linker plays an important role in two different autoinhibitory mechanisms of DNMT1. It also provides a framework for understanding the functional effect of the disease-related mutations of hDNMT1.

Results

hDNMT1_351–1600 adopts an autoinhibitory conformation similar to that of mDNMT1_291–1620

The structure of hDNMT1_351–1600 reveals a closely-stacked three-layer architecture (Fig. 1A), with the RFTS and CXXC domains on the top, the MTase domain in the middle, and the two BAH domains situated at the bottom. In addition, it contains five zinc finger clusters, spreading over the RFTS domain, the CXXC domain, the BAH1 domain and the MTase domain, and an AdoHcy molecule in the active site (Fig. 1A). The N-terminal domains, except for the CXXC domain, make extensive contacts with the MTase domain, resulting in a compact fold (Fig. 1A). We have performed small X-ray scatting (SAXS) analysis to validate the observed structural fold (Fig. 1B). Our results revealed that the structure of hDNMT1_351–1600 fits well with the SAXS scatting curve, with a χ² of 1.4 (Fig. 1B), thereby confirming that this structure is consistent with the conformation of hDNMT1_351–1600 in solution state.

Figure 1. Structural overview of hDNMT1_351–1600 bound to AdoHcy.

(A) Ribbon representation of hDNMT1_351–1600 in complex with AdoHcy, with the RFTS, CXXC, BAH1, BAH2 and MTase domains colored in green, red, pink, orange, and light blue, respectively. The domain linkers and zinc ions are colored in silver and purple, respectively. The bound AdoHcy is in a space filling representation. (B) Measured scattering curve of free hDNMT1_351–1600 (black) is compared with the scattering curve calculated from the crystal structure (red). (C) Superposition of hDNMT1_351–1600 (light blue) and mDNMT1_291–1620 (magenta), with the N-terminal half of the RFTS domains highlighted in an expanded view. (D–E) Interactions between a loop from the RFTS domain and the α_RS-helix in hDNMT1_351–1600 (D) and mDNMT1_291–1620 (E), respectively.

The structure of hDNMT1_351–1600 reveals that its RFTS domain is anchored to the catalytic cleft of its MTase domain (the contacts are detailed below), potentially inhibiting the substrate binding (Fig. 1A). Such an autoinhibitory conformation is reminiscent of what was previously observed for mDNMT1_291–1620 (Fig. 1C) (19). In fact, the structures of hDNMT1_351–1600 and mDNMT1_291–1620 superimpose well, with a root-mean-square deviation (r.m.s.d.) of 1.0 Å over 989 aligned Cα atoms, consistent with the fact that these two orthologues share ~ 83% sequence identity in this region (Fig. S1). The main conformational difference between the two structures arises from the N-terminal half of the RFTS domain, which is associated with relatively lower (72%) sequence identity (Fig. S1). In particular, the α-helix (α_RS: residues 495–519 in hDNMT1) that connects the two halves of the RFTS domain adopts a straight conformation in hDNMT1_351–1600, but is kinked in the structure of mDNMT1_291–1620 (Fig. 1C). As a result, the orientation of the first half of the RFTS domain relative to the rest of the protein is deviated by 19° between hDNMT1_351–1600 and mDNMT1_291–1620 (Fig. 1C), despite the relatively conserved fold of this region. Such a conformational divergence between hDNMT1_351–1600 and mDNMT1_291–1620 is in attributed in part to their distinct molecular contacts within the RFTS domain (Fig. 1D,E). In the structure of hDNMT1_351–1600, loop residues F469, E473, K474, A475 and I477 from the RFTS domain make close contacts with residues F499, Q502, Y507 and K510 from the N-terminal part of the α_RS-helix (Fig. 1D). Whereas in the structure of mDNMT1_291–1620, the equivalent loop extends further toward the C-terminal end of the α_RS-helix, engaging in a distinct set of interactions that involve loop residues: L472, F475, D476, E479 and V481, and the α_RS-helix residues: I512, K516, E520 and Q523 (Fig. 1E). In agreement with these conformational differences, fitting the structure of mDNMT1 (a region equivalent to hDNMT1_351–1600) to the hDNMT1 SAXS curve gives a worse fit than hDNMT1, with χ² of 2.1 (Fig. S2).

The RFTS domain-mediated autoinhibition is supported by both direct domain-domain interactions and the CXXC-BAH1 domain linker

The previous structure of mDNMT1_291–1620 has revealed that four hydrogen bonds, formed by residues E531, D532, D554 and L593 from the RFTS domain and K537, R1576, S1495 and T1505 from the MTase domain, respectively, mediate the association of the RFTS domain with the MTase domain (19). Inspection of the structure of hDNMT1_351–1600 not only confirms the existence of the equivalent interactions in hDNMT1 (Fig. 2A), but also reveals an hDNMT1-unique interaction that supports the contact between the RFTS and MTase domains: a hydrogen bond between residues D583 and K1535 of hDNMT1_351–1600. Such an interaction is absent in mDNMT1 due to the fact that the residue equivalent to D583 is replaced by an alanine (19). In addition, the improved resolution of the structure of hDNMT1_351–1600 has allowed us to identify a number of inter-domain contacts that have failed to be detected in the previous structure of mDNMT1_291–1600 (19). For instance, we have identified that two water-mediated hydrogen bonds, formed by the side chain of D548 with the backbone amide nitrogen of M1535 and the side chain hydroxyl group of Y1515, respectively, help mediate the association of the RFTS domain with the MTase domain (Fig. 2A).

Figure 2. Structural details of hDNMT1_351–1600 highlighting the RFTS domain-mediated autoinhibition.

(A) Hydrogen bonding Interactions between the RFTS domain (green) and the MTase domain (light blue) in hDNMT1_351–1600. The hydrogen bonds are shown in dashed line. The water molecules are shown in sphere. (B) Hydrogen bonding interactions of the CXXC-BAH1 helical linker (grey) with the MTase and RFTS domains.

Analysis of the structure of hDNMT1_351–1600 reveals that the linker α-helix between the CXXC and BAH1 domains also plays a role in associating the RFTS domain with the MTase domain. On one side, this helix interacts with the RFTS domain, through hydrogen bonds formed by the side chain of residue D700 and the side chains of residues R582 and K586, and by the side chain of residue E703 and the side chain of residue R582 (Fig. 2B). On the other side, this helix makes contacts with the MTase domain through hydrogen bonds formed by the side chain of residue D702 and the backbone and side chain of residue N1233, and a main chain hydrogen bond between residues D702 and M1235 (Fig. 2B). Together, these interactions further strengthen the association between the RFTS domain and the MTase domain.

Inhibition of the enzymatic activity of hDNMT1_351–1600 by the CXXC-BAH1 domain linker

To test the role of the CXXC-BAH1 domain linker in the RFTS domain-mediated autoinhibition of hDNMT1, we have mutated residue R582 to glutamate to disrupt its interaction with the helical linker, and analyzed the hDNMT1_351–1600-DNA interaction using Electrophoretic Mobility Shift Assay (EMSA). Our results showed that the R582E mutation indeed modestly increases the binding affinity of hDNMT1_351–1600 to a 26-mer hemimethylated CpG DNA (Fig. S3A, B). We then compared the enzymatic activities of wild-type and R582E hDNMT1_351–1600 proteins using a mass spectrometry-based approach. The methylation reactions were carried out with a 12-mer hemimethylated DNA substrate containing one central CpG site (upper strand: 5′-GAGGC^mCGCCTGC-3′; lower strand: 5′-GCAGG^mCGGCCTC-3′). The activities of wild-type and R582E hDNMT1_351–1600 were measured using a previously published HPLC coupled with tandem mass spectrometry (LC-MS/MS) method (26,27). Quantitative analysis of the level of methylation was carried out based on the composite tandem mass spectrum of the [M – 3H]³⁻ ions (m/z 1218.9) of the two methylated strands (Fig. 3A, B). Along this line, collisional activation of deprotonated ions of short DNA gave rise to the cleavage of the N-glycosidic bond and the 3′ C-O bond within the same nucleoside to yield [a_n-base] and w_n series ions (28). As displayed in Fig. 3A, some [a_n-base] and w_n ions for the two DNA strands share the same m/z values (e.g. w₅, w₆ and w₇ ions), whereas others have unique m/z values. Considering that the upper strand is completely methylated, we estimated the level of methylation in the lower strand, expressed as %methylation, based on the relative signal intensities observed for selected fragment ions for the two methylated DNA strands that exhibit unique m/z values. In particular, we derived the %methylation by dividing the peak area found in the selected-ion chromatogram (SIC) for monitoring the four transitions corresponding to the formations of the [a₃-base], w₃, [a₄-base], and w₄ ions, i.e. m/z 1218.9 → m/z 715.1, 899.1, 1028.2, 1188.0, for the lower strand by that found in the SIC for the corresponding transitions for the upper strand, i.e. m/z 1218.9 → m/z 739.0, 939.0, 1068.2, 1228.1 (Fig. 3B). Despite the fact that the aforementioned peak area ratio may deviate to some extent from the %methylation of the lower stand due to the differences in ionization and fragmentation efficiencies of the two DNA strands, this method permits a direct comparison of the relative levels of methylation induced by wild-type or mutant hDNMT1 proteins. As shown in Fig. 3C, at 2 hr reaction time, we observed a ~69% increase in the methylation activity of hDNMT1_351–1600 by the R582E mutation. The methylation activity of hDNMT1_351–1600 becomes even higher when we further removed a patch of residues (residues 694–701; the resulting mutant is denoted as Δ694–701) in the helical linker (Fig. 3C). These data confirm that the CXXC-BAH1 linker-mediated interactions are inhibitory to the enzymatic activity of DNMT1. Note that the methylation activity of the R582E mutant remains lower than that of an RFTS-free DNMT1 fragment, hDNMT1_646–1600 (Fig. S3C), indicating that the direct interactions between the RFTS and MTase domains are also important for the RFTS domain-mediated autoinhibtion (Fig. 2A) (20–22). Next, we performed SAXS analysis to examine whether these mutations affect the conformational states of hDNMT1_351–1600 in solution (Fig. S4A–D). We observe that the radius of gyration (Rg) of hDNMT1_351–1600 increases from 40.5 Å for wild-type protein to 48.0 Å for the R582E mutant, and to 60.8 Å for the Δ694–701 mutant, while its maximum dimension (Dmax) increases from 98 Å for wild-type protein to 163 Å for the R582E mutant, and to 194 Å for the Δ694–701 mutant (Fig. S4E). The increased Rg and Dmax values by the R582E and Δ694–701 mutations suggest these two mutations shift the conformational states of hDNMT1_351–1600 to a more open and elongated form. Together, these data support that the CXXC-BAH1 helical linker constitutes a key regulatory element in the RFTS domain-mediated autoinhibition of DNMT1.

Figure 3. Enzymatic activities of wild type and mutant hDNMT1_351–1600.

(A) LC-MS/MS characterization of upper DNA strand (5′-GAGGCmCGCCTGC-3′, blue) and methylated lower DNA strand (5′-GCAGGmCGGCCTC-3′, red) in 12mer-DNA duplex after reaction with hDNMT1_351–1600. The product-ion spectrum (MS/MS) of the [M-3H]³⁻ ion (m/z 1218.9) is shown. (B) Quantification of the percentage of methylation by comparing the peak areas found in the selected-ion chromatograms (SICs) for monitoring the indicated transitions of upper DNA strand and newly methylated lower DNA strand. (C) Enzymatic kinetics of wild-type hDNMT1 and its R582E and Δ694–701 mutants.

Notably, the CXXC-BAH1 helical linker adopts an extended form in the structures of hDNMT1_646–1600 in complex with an unmethylated 19-mer CpG DNA (18) and free hDNMT1_601–1600 (PDB: 2SWR), suggesting that a helix-to-loop conformational transition was induced by the removal of the RFTS domain (Fig. S5). In both structures, this sequence is positioned along the catalytic cleft of hDNMT1 via its interaction with the MTase domain, thereby inhibiting the interaction between DNMT1 and DNA substrates (18). The inhibitory role of the CXXC-BAH1 linker in hDNMT1_646–1600 is further supported by our observation that the hDNMT1_351–1600 Δ694–701 mutant shows even higher methylation efficiency than hDNMT1_646–1600 (Fig. S3C). Taken together, these observations suggest that the CXXC-BAH1 domain linker inhibits the enzymatic activity of DNMT1 through a multi-layered mechanism.

The catalytic loops of AdoHcy-bound hDNMT1_351–1600 and mDNMT1_291–1620 assume different conformations

Previous studies on the DNA methyltransferase from Haemophilus haemolyticus (M.HhaI) or mDNMT1_291–1620 have demonstrated that the catalytic site cysteine in the PCQ loop undergoes a conformational rearrangement in response to cofactor or DNA binding (19,29,30). For instance, the catalytic residue C1229 of mDNMT1_291–1620 faces away from the DNA binding pocket in free state or AdoHcy-bound form; upon binding to AdoMet, it turns to approach the position where the flipped-out target cytosine is hosted (19). Based on this observation, it has been proposed that binding of the cofactor AdoMet provides a mechanism in switching the side chain conformation of the catalytic cysteine in mDNMT1 (19). Different from that of mDNMT1_291–1620, the structure of hDNMT1_351–1600 reveals that the active site cysteine, C1226, is poised to receive the target cytosine even in the presence of AdoHcy (Fig. 4A), suggesting that AdoMet-binding does not serve as a switch for the side chain conformation of C1226 in hDNMT1. Structural comparison of AdoHcy-bound hDNMT1_351–1600 and mDNMT1_291–1620 further indicates that such a conformational difference arises from the fact that residue C1226 in hDNMT1 experiences more steric repulsion from the phenyl group of the adjacent residue, F1229, than its counterpart in mDNMT1 (Fig. 4A). In fact, the conformation of the F1229 equivalent residue in mDNMT1 (F1232) is restrained by a hydrogen bond between the side chain of residue F1232 and the backbone of residue R1285; such a hydrogen bond is not formed in hDNMT1, since the residue equivalent to mDNMT1 R1285 is replaced by a lysine. These observations imply a different cofactor binding response in the catalytic site between hDNMT1 and mDNMT1. Interestingly, the conformation of the catalytic cysteine in hDNMT1_351–1600 is also different from that in the hDNMT1_646–1600-DNA complex (Fig. 4B). In the latter case, the CXXC-BAH1 domain linker occupies the DNA binding site, thereby pushing the side chain of residue C1226 away from the DNA binding pocket through steric hindrance.

Figure 4. The catalytic cysteine in hDNMT1_351–1600 is poised to receive the target cytosine.

(A) Superposition of the catalytic loop between hDNMT1_351–1600 (light blue) and mDNMT1_291–1620 (magenta). The distance between the backbone carbonyl oxygen of F1229 and K1285 in hDNMT1_351–1600 is 4.6 Å, whereas the equivalent groups in mDNMT1_291–1620 form a hydrogen bond, with a distance of 3.0 Å. (B) Superposition of the catalytic loop between hDNMT1_351–1600 (light blue) and hDNMT1_646–1600 – DNA complex (wheat, PDB 3PTA). (C) The DNMT1 CXXC domain undergoes a large repositioning from the free state (light blue) to the DNA bound form (wheat). The CXXC domain in free hDNMT1_351–1600 and in complex with CpG DNA is superimposed in a rectangle.

Dynamic positioning of the CXXC domain between different functional states of hDNMT1

The CXXC domain in the structure of mDNMT1_291–1620 was only partially determined (19), presumably due to the relatively low resolution and conformational dynamics. The improved resolution of the structure of hDNMT1_351–1600 in this study has permitted us to trace the entire CXXC domain (Fig. 1A and 4C). On this basis, we were able to compare the structure of the CXXC domain in free hDNMT1 (this study) and in its DNA-bound form (18), which reveals a surprisingly high conformational identity between the two forms of the CXXC domain (Fig. 4C). In particular, the CpG-recognition loop of the CXXC domain is well poised for DNA binding even in free state, suggestive of a preconfigured, functional conformation adopted by the CXXC domain. On the other hand, a large repositioning of the CXXC domain occurred between these two alternative states: The CXXC domain is spatially proximal to the MTase domain in free hDNMT1_351–1600, but shifts away by a distance of ~30 Å when a 19-mer unmethylated CpG DNA is bound (Fig. 4C). As described above, such repositioning of the CXXC domain is coupled with a helix-to-loop conformational transition for the linker sequence between the CXXC domain and the BAH1 domain, which then occupies the catalytic cleft to impose another layer of autoinhibition on DNMT1.

Structural analysis of the disease-related DNMT1 mutations

Recent evidence has revealed that mutations of DNMT1 are implicated in several neurological disorders (31–34), including hereditary sensory autonomic neurophathy with dementia and hearing loss (HSAN1E) and cerebella ataxia, deafness and narcolepsy (ADCA-DN). These mutations mainly fall into the RFTS domain of DNMT1 (Fig. 5A). For instance, mutations of D490E, P491Y, Y495C and Y495H were identified in a number of HSAN1E kindreds (31,32); whereas mutations of A554V, C580R, G589A and V590F were implicated in a cohort of ADCA-DN patients (33,34). Structural analysis of hDNMT1_351–1600 suggests that some of these mutations may result in altered conformation and stability of DNMT1. For instance, the side chain hydroxyl group of residue Y495 forms a hydrogen bond with the main chain carbonyl oxygen of residue C420, which helps stabilize the orientation of α_RS-helix in the RFTS domain (Fig. 5B). Mutation of Y495C or Y495H may disrupt such an interaction, leading to structural and dynamic change of the α_RS-helix. In addition, residues A554 and V590 both participate in formation of a hydrophobic cluster (Fig. 5C). Therefore, their respective mutations likely result in local structural perturbation of this region. Consistently, a previous study has shown that the Y495C mutation leads to reduced protein stability and enzymatic activity, and aberrant DNA methylation patterns (32). Further analysis of these disease-related mutations requires a detailed functional characterization of the RFTS domain of DNMT1.

Figure 5. Molecular interactions involving the disease-related mutation sites of hDNMT1.

(A) Surface and ribbon representation of hDNMT1_351–1600, with areas of the reported mutation sites in neurological disorders colored in red. (B) Hydrogen bond between the side chain of Y495 and the backbone carbonyl oxygen of C420 is shown in dashed line. (C) Residue A554 and V590 are involved in forming a hydrophobic cluster within the RFTS domain.

Discussion

Recent structural and functional studies on DNMT1 have provided important insights into the allosteric regulation and substrate specificity of DNMT1 (18,20–22,35–37). However, the fact that hDNMT1 and mDNMT1 show slightly different enzymatic properties (18,23–25) raises a question about whether structural analysis of mDNMT1 is sufficient to reveal the functional states of hDNMT1. This study presents the structure of an hDNMT1 construct that is comprised of all its structural domains, which therefore provides a framework for understanding of the structure-function relationship of hDNMT1, as well as the relationship between hDNMT1 mutations and human diseases.

The DNMT1 N-terminal domains are important in regulating the recruitment and enzymatic activity of DNMT1. The RFTS domain has been identified as a protein interaction module that mediates the recruitment of DNMT1 to replication foci (14,16), and the interaction between DNMT1 and UHRF1 (20,21,38,39). However, the molecular basis underlying these regulations remains unknown. Our structural study reveals detailed molecular interactions between the RFTS domain and the MTase domain, which explains the autoinhibitory role of the RFTS domain in the enzymatic activity of hDNMT1. In addition, we observed a conformational difference of the RFTS domain between hDNMT1_351–1600 and mDNMT1_291–1620, which, despite the possible contribution of different crystallization conditions, implies subtle structural diversity between the two orthologues.

Structural analysis and biochemical assays of hDNMT1_351–1600 reveal that the helical linker between the CXXC domain and the BAH1 domain constitutes an important regulatory element in the RFTS domain-mediated autoinhibition. It interacts with both the RFTS domain and the MTase domain, thereby strengthening the association of the RFTS domain with the MTase domain. Notably, this CXXC-BAH1 helical linker undergoes a helix-to-loop conformational transition when DNMT1 binds to unmethylated CpG DNA (18), positioned along the catalytic cleft of DNMT1 to inhibit the interaction between DNMT1 and DNA substrates (18). Together, these observations suggest that the CXXC-BAH1 domain linker, namely autoinhibitory linker, inhibits the enzymatic activity of DNMT1 through a multi-layered mechanism (Fig. 6). Such inhibition of DNMT1 will be conceivably released when DNMT1 binds to hemimethylated CpG DNA substrates, accompanied by a large conformational rearrangement of the RFTS and CXXC domains (Fig. 6), although the conformation of the autoinhibitory linker under this state is currently unknown.

Figure 6. The CXXC-BAH1 linker plays a role in the autoinhibitory mechanisms of DNMT1.

The CXXC-BAH1 helical linker (a.k.a. autoinhibitory linker) is colored green. The conformational transitions of hDNMT1 upon binding to different DNA substrates are proposed.

In comparison with the structure of mDNMT1_291–1620, this study reveals a number of distinct structural features for hDNMT1, including its catalytic site. In mDNMT1, the active site cysteine faces away from the DNA binding pocket in the presence of AdoHcy, but turns to approach the position where the target cytosine is hosted in the presence of AdoMet. Based on these observations, it has been proposed that the active site cysteine of mDNMT1 switches from an inactive conformation to an active conformation upon binding to AdoMet, but returns back to an inactive conformation after the methylation reaction occurs (19). By contrast, the side chain of the catalytic cysteine of hDNMT1 is poised to receive the target cytosine even in the presence of AdoHcy. Therefore, the AdoMet/AdoHcy cofactor binding does not appear to provide a switching mechanism for the active site of hDNMT1. Such divergence in functional regulation of hDNMT1 and mDNMT1 likely contributes to their slightly different enzymatic behaviors (23–25).

Materials and Methods

Protein expression and purification

The gene encoding hDNMT1_351–1600 was amplified by PCR and inserted into a modified pRSFDuet-1 vector, where it was preceded by an N-terminal His6-tag and a SUMO-tag. The sequence of hDNMT1_351–1600 was confirmed by DNA sequencing analysis. The hDNMT1_351–1600 construct used for crystallization contains an additional five-residue patch (LTRVW) introduced to the very C-terminus, due to cloning artifacts. The R582E and Δ694–701 mutants were obtained through site-directed mutagenesis. The plasmids of hDNMT1_351–1600 constructs were transformed into BL21 (DE3) RIL cell strain for overexpression. The bacterial cells were grown at 37 °C and induced by 0.4 mM isopropyl β-d-1-thiogalactopyranoside (IPTG) when the cells density reached an optical density (OD₆₀₀) of 0.8. After induction, the cells continued to grow at 16 °C for 16 hrs. The cells were then harvested, and lysed in a buffer containing 50 mM Tris-HCl (pH 8.0), 25 mM imidazole, 1 M NaCl, 0.5 mM DTT and 1 mM PMSF. The His-SUMO-DNMT1 fusion protein was first purified by a Ni-NTA column, followed by removal of the His6-SUMO tag by Ubiquitin-like protease 1 (ULP1) cleavage. The protein sample for hDNMT1_351–1600 was further purified through ion exchange (Hitrap Q, GE Health Sciences) and size-exclusion (Superdex 200 16/60, GE Health Sciences) chromatography methods. The final protein sample, with > 95% purity, was concentrated to ~15 mg/ml and stored in a buffer containing 25 mM Tris-HCl (pH 7.5), 100 mM NaCl and 5 mM DTT at −80 °C.

Crystallization and structure determination

For crystallization, hDNMT1_351–1600 was first mixed with AdoHcy in a molar ratio of 1:1.5. The complex was then crystallized using sitting-drop vapor diffusion method by mixing 1.5 μl of hDNMT1_351–1600 sample and 1.5 μl of reservoir solution containing 100 mM MES, pH6.0, 50 mM CaCl₂, and 45% PEG200 at 16°C. Crystals that grew into full size in three days were equilibrated in reservoir solution before flash-frozen in liquid nitrogen.

The X-ray diffraction data set for hDNMT1_351–1600 were collected on the BL 5.0.1 beamline at the Advanced Light Source (ALS), Lawrence Berkeley National Laboratory. The diffraction data were indexed, integrated and scaled using the HKL 2000 program. The structure of hDNMT1_351–1600 was solved using the molecular replacement method in PHASER (40) using the structure of mDNMT1_291–1620 (PDB 3AV4) as a search model. The resulting electron density revealed that there are two molecules of hDNMT1_351–1600 in each asymmetric unit (ASU). Further modeling of the hDNMT1 _351–1600 was carried out using COOT (41), and then subject to refinement using the PHENIX software package (42). The same R-free test set was used throughout the refinement. The B-factors were refined with individual B values. The statistics for data collection and structural refinement of hDNMT1_351–1600 is summarized in Table 1.

Table 1.

Crystallographic statistics for hDNMT1_351–1600.

Data collection
Wavelength (Å)	0.9774
Space group	P43
Cell dimensions
a, b, c (Å)	138.6, 138.6, 221.5
α, β, γ (°)	90, 104.4, 90
Resolution (Å)	50–2.62 (2.71–2.62) a
Rmergeb	11.8 (75.5)
I/σI	12.8 (2.1)
Completeness (%)	98.6 (99.0)
Redundancy	5.2 (4.9)
No. of unique reflections	122,755 (12,321)
Complex molecules/ asymmetric unit	2
Refinement
Resolution (Å)	48.22–2.62
No. reflections	122,708
Rwork/Rfree (%)c	19.3/24.0
No. atoms
Protein	18336
Ligand	52
Zinc	10
Water	637
B-factors
Protein	51.2
Ligand	35.9
Zinc	59.0
Water	44.8
R.m.s deviations
Bond lengths (Å)	0.009
Bond angles (º)	1.400
Ramachandran plot
Favoured regions (%)	91.7
Allowed regions (%)	6.7
Outliers (%)	1.6

^aHighest resolution shell is shown in parenthesis.

^bR_merge = Σ_hklΣ_i|I_i(hkl) <I(hkl)>|/Σ_hklΣ_iI_i(hkl), where I_i(hkl) is the intensity of the i^th observation of reflection hkl, and <I(hkl)> is the average over all observations of reflection hkl.

^cR_factor = Σ_hkl ||F_o| – |Fc||/Σ_hkl|F_o| for all data excluding the 5% that comprised the R_free used for cross-validation.

Small Angle X-ray Scatting experiment (SAXS)

To measure the X-ray scatting curve, 2–6 mg/ml hDNMT1_351–1600 wild-type, R582E or Δ694–701 protein samples were dissolved in a buffer containing 20 mM Tris-HCl, pH 7.5, 250 mM NaCl and 5 mM DTT. The SAXS data were collected at the SYBIL beamline at ALS, Lawrence Berkeley National Laboratory, and processed using the ScÅtter software (https://bl1231.als.lbl.gov/scatter/). The scattering curves calculated from the crystal structures were fit with the experimental curve using the fast open-source x-ray scatting (FoXS) method (43) through the FOXS server (http://modbase.compbio.ucsf.edu/foxs/).

Enzymatic methylation kinetics assay

The DNMT1 cytosine-5 methylation assays were carried out in triplicates, using protocols described previously (26). 30 pmol of 12-mer DNA duplex (upper strand: 5′-GAGGC^mCGCCTGC-3′; lower strand: 5′-GCAGGCGGCCTC-3′) was incubated with 15 nmol of S-adenosyl-L-methionine (New England Biolabs, Ipswich, WA) and 3 pmol of wild-type hDNMT1 (or its R582E mutant) at 37 °C in a reaction buffer (50 mM Tris-HCl, 1 mM DTT, 5% glycerol, 100 μg/mL BSA, pH 7.5). After incubation for 0.5, 1.0, and 2.0 hr, the enzyme was inactivated by heating at 65°C for 20 min and removed by chloroform extraction, and the aqueous layer was subjected to LC-MS/MS analysis.

LC-MS/MS Analysis and Data Processing

The LC-MS/MS experiments were conducted on an LTQ linear ion-trap mass spectrometer (Thermo Fisher Scientific, San Jose, CA) equipped with an Agilent 1200 capillary HPLC (Agilent Technologies, Santa Clara, CA). A 0.5 × 250 mm Zorbax SB-C18 column (5 μm in particle size, Agilent Technologies, Santa Clara, CA) was employed for the separation at a flow rate of 8.0 μL/min and with a gradient of 20 min of 5% methanol followed by 20 min of 65% methanol in 400 mM 1,1,1,3,3,3-hexafluoro-2-propanol (pH 7.0, adjusted with triethylamine). The spray, capillary and tube lens voltages were 4.5 kV, −31 V and −95 V, respectively. The temperature for the ion transport tube was maintained at 300°C.

EMSA (electrophoretic mobility shift assay)

A 26-bp hemimethylated DNA duplex (upper strand: 5′-ACACCAAGCCTGmCGGAGGCTCACGGA-3′, mC = 5-methyl-2′-deoxycytidine; lower strand: 5′-TCCGTGAGCCTCCGCAGGCTTGGTGT-3′) was used to characterize DNA binding of the wild-type hDNMT1_351–1600 and its R582E mutant. Binding reactions contained 0.1 μM dsDNA and various concentrations of protein, dissolved in 10 μl of binding buffer (20 mM Tris-HCl, pH 7.5, 50 mM NaCl, 1mM DTT). Binding reaction mixtures were electrophoresed in 6% DNA retardation gel (Invitrogen) in 0.25X TBE (89 mM Tris-borate, pH 8.4) buffer at 4 °C under 100 V for 50 min. DNA was visualized by SYBR green staining and quantitated with imageJ software (44).

PDB accession numbers

The atomic coordinates and structural factors have been deposited into the PDB under the accession code 4WXX.

Highlights.

Background

The function of DNMT1 is regulated by its N-terminal domains

Results

This study provides molecular basis for the allosteric regulation of DNMT1

Significance

This study reports the structure of human DNMT1 with all the domains