Cryo-EM structures of DNMT3A2-DNMT3L reveal nucleosome binding and oligomeric organization

To reveal the molecular mechanism of DNMT3A-3L binding to nucleosomes, we reconstituted NCPs containing the 147 bp 'Widom 601' sequence and 10 bp linker on each side as previously described (NCP167), and then assembled them with the full-length human DNMT3A2-3L complex (Extended Data Fig. 1a). Interaction was first confirmed by AlphaScreen luminescence proximity assays and native PAGE (Extended Data Fig. 1b,c). The cryo-EM structure had well-resolved densities for the nucleosome core and the DNMT-bound flanking DNA. To a lesser degree, the structure showed clear density for secondary structure elements of the DNMT3A2 catalytic domains, the proximal DNMT3A2 ADD domain and the proximal DNMT3L (Fig. 1c and Extended Data Fig. 5a,b). The distal DNMT3A2 ADD domain density only demonstrates the overall position (Extended Data Fig. 5b inlet), while the PWWP domains and histone tails were unresolved, probably because of the conformational flexibility. Thus, the cryo-EM structure of this complex shows a clear heterotrimeric DNMT3L-3A2-3A2 arrangement with clear density for the F-F interface and the distinct C-terminal helix (which we named 'switching helix') from DNMT3L at the distal interface (Fig. 1c and Extended Data Fig. 5c), indicating the presence of a second DNMT3L (distal) to form a more stable and active heterotetramer. Similar to the DNMT3A2-3B3 complex, DNMT3A2-3L also binds to nucleosomes asymmetrically: the proximal DNMT3L engages the nucleosome core, while the distal DNMT3A2 catalytic domain interacts with the linker DNA (Fig. 1c). DNMT3 complexes are known to adopt two conformations in equilibrium between active and autoinhibitory states (Extended Data Fig. 5d). Releasing the autoinhibition state requires the ADD domain to bind to an unmodified histone H3 N terminus. Despite the presence of unmodified H3 tails in our NCPs, the DNMT3A2 ADD domains remained in an autoinhibitory conformation in our cryo-EM structure (Extended Data Fig. 5d). Interestingly, the proximal DNMT3L ADD domain adopted an active conformation (Extended Data Fig. 5e), suggesting that engagement with the nucleosome core selectively modulates DNMT3L ADD domain conformations to regulate its binding and activity.

To assess the conformational states of DNMT3A2-3L in solution, we performed mass photometry analysis. The DNMT3A2-3L complexes mainly equilibrate between heterodimers and heterotetramers at low concentrations (<800 nM, based on heterotetramers), which is consistent with the stoichiometry calculation based on the size-exclusion chromatography (SEC) fractions and SDS-PAGE analysis (Extended Data Fig. 1a). With increasing concentration, the complex becomes more oligomerized (hexamer, octamer and dodecamer). DNMT3A2 shows a tendency to oligomerize through its F-F and R-D interfaces, whereas DNMT3L is primarily in monomer form at low concentrations and dimer form at high concentrations. (Extended Data Fig. 1d). To gain further insights into the DNMT3A assembly, we determined the nucleosome-free structures of DNMT3A2-3L in two oligomeric states. The heterotetramer of the DNMT3A2-3L catalytic and catalytic-like domain revealed two distinct interaction interfaces: a 'flat', non-polar, heterodimeric interface (3A-3L F-F interface) and a 'V-shaped' polar, homodimeric interface (3A-3A R-D interface) (Extended Data Fig. 5f). These structural features facilitated the identification of four R-D interfaces in the dodecamer (Fig. 2a) and three in the octamer (Fig. 2b), corresponding to four and three DNMT3A2 dimers, respectively. In the dodecamer, regions beyond DNMT3L-3A2-3A2-3A2-3A2 could not be confidently assigned because of resolution limits (Extended Data Fig. 4e); therefore, they were omitted from the final model. The fact that DNMT3L can only form the F-F interface and prevents higher-order oligomerization of DNMT3A2 (ref. ), and the molecular mass (~800 kDa) measured by mass photometry indicates that the distal regions correspond to two additional DNMT3L molecules, resulting in the DNMT3A2-3L dodecamer complex (Extended Data Fig. 1d). In the octamer, regions beyond the three DNMT3A2 dimers were too weak to confidently assign (Extended Data Fig. 4f); therefore, they were omitted from the final model. The molecular mass (~510 kDa) indicates that the distal regions correspond to two additional DNMT3Ls, resulting in octamer formation (Extended Data Fig. 1d). Although DNMT3B homo-oligomerization has been structurally characterized, our cryo-EM analysis provides the first evidence of DNMT3A2-3L forming hetero higher-order oligomers, which may increase the local concentration of DNMT3A2-3L or change chromatin binding profiles that may contribute to abnormal DNA methylation in tumorigenesis, given that DNMT3A2 is highly upregulated in multiple cancers. Importantly, this unique, novel arrangement enables dynamic assembly and allosteric regulation of the DNMT3A2-3L complex.

These findings indicate that nucleosomes function not only as substrates for the DNMT3A complex but also as regulators for DNMT3A assembly. When bound to nucleosomes, the heterotetrameric arrangement probably stabilizes DNMT3A2-3L while preventing higher-order oligomerization and enabling the complex to adopt a more stable and active conformation. Despite this stabilization, we still did not observe the flipped DNA base in the DNA-engaged DNMT3A2 catalytic center (Extended Data Fig. 5g), indicating that nucleosome binding does not require CpG motifs and is not sequence-specific. This sequence-independent binding may provide a reasonable explanation for the absence of strict sequence patterns in genome-wide DNA methylation.

Structural comparison of nucleosome-bound DNMT3A2-3L and nucleosome-bound DNMT3A2-3B3 reveals a striking ~180° conformational reorientation of the DNMT3L C-terminal 'switching helix' relative to its DNMT3B3 counterpart. This rotated conformation prevents interactions between the accessory protein DNMT3L and the nucleosome acidic patch (Fig. 3a and Extended Data Fig. 6a,b), a key binding site formed by histones H2A and H2B for various nucleosome-interacting proteins, including DNMT3A2-3B3 (ref. ).

To validate this observation, we performed nucleosome-binding competition assays using peptides that specifically interact with the acidic patch. Consistent with structural data, acidic patch-interacting peptides reduced the nucleosome binding of the DNMT3A2-3B3 complex, but had no effect on DNMT3A2-3L (Fig. 3b), suggesting distinct chromatin recruitment mechanisms for the different DNMT3 accessory proteins. Interestingly, the histone H3 peptides selectively competed with DNMT3A2-3L for nucleosome binding (Fig. 3b), indicating that histone tails may have a greater role in DNMT3A2-3L recruitment owing to its lack of direct interaction with the nucleosome acidic patch. Sequence alignment of DNMT3 family C-terminal regions highlights high conservation between DNMT3A and DNMT3B but not DNMT3L (Extended Data Fig. 6c), indicating functional divergence. Closer examination identified residues Q348 and Q351 in DNMT3L as equivalents of the DNMT3B3 'R finger' residues (R740 and R743). However, owing to the unique confirmation of DNMT3L's C-terminal 'switching helix', these residues, along with nearby K350 and K354, do not directly interact with the acidic patch (Fig. 3a). Nucleosome-binding assays with DNMT3L-mutant complexes further confirmed that unlike DNMT3B3, DNMT3L does not directly engage the acidic patch (Fig. 3c). Given the proximity of the DNMT3L C-terminal 'switching helix' and adjacent β-sheets to the nucleosome core (Fig. 3a), we evaluated their roles in nucleosome recruitment. Notably, although residues Q348, Q351, K350 and K354 had a subtle impact on nucleosome binding, deletion of the entire C-terminal 'switching helix' (E340 to K354; Δα-helix) modestly reduced nucleosome binding by approximately 25%, suggesting that this region mediates moderate-range charge and polar interactions. However, these interactions alone appear insufficient to engage nucleosomes, implying the requirement for additional contacts. Conversely, deleting the β-sheet loop region (D311 to N319; Δβ-sheet), which contains negatively charged and polar residues (Extended Data Fig. 6c), led to a slight increase in nucleosome binding (Fig. 3c). Together, these data strongly support the finding that DNMT3A2-3L binds to nucleosomes through a mechanism distinct from acidic patch engagement, instead relying on moderate, potentially electrostatic interactions mediated by the C-terminal 'switching helix' and other nucleosome-proximal regions.

To investigate how DNMT3L contributes to the recruitment of the DNMT3A2-3L complex to nucleosomes, we overlaid the DNMT3A2-3L-nucleosome map with the DNMT3A2-3B3-nucleosome map (EMD-20281). This comparison revealed an additional well-defined density corresponding to the DNMT3L ADD domain, emphasizing its critical role in nucleosome binding (Fig. 4a). Structural analysis showed that the DNMT3L ADD domain was anchored by negatively charged residues, including E103, which interacts electrostatically with positively charged histone H4 tails, particularly K20 (Fig. 4b). Although eliminating this charge interaction significantly weakened nucleosome binding, the deletion of the DNMT3L ADD domain almost completely abolished it (Fig. 4c). This data reinforces the conclusion that DNMT3L ADD domain has a critical role in nucleosome binding.

Although H4K20 methylation, particularly H4K20me3, is commonly associated with heterochromatin and often coincides with DNA methylation, there is limited evidence that H4K20 methylation directly recruits DNMT3 enzymes. Therefore, we performed the AlphaScreen assay to quantitatively assess the interaction between DNMT3A2-3L and nucleosomes harboring different H4K20 methylation states. Unexpectedly, although H4K20me3 marks constitutive heterochromatin regions that showed enriched DNA methylation, it reduced DNMT3A2-3L binding to nucleosomes (Fig. 4d). This indicates that DNA methylation in these regions may also involve other DNMTs or regulators, including DNMT1 (ref. ). Interestingly, H4K20me1 showed very similar binding strength to the unmodified nucleosomes, whereas H4K20me2 enhanced DNMT3A2-3L binding (Fig. 4d), despite its less well-characterized relationship to DNA methylation. We speculate that the di-methylation may serve as a precursor for tri-methylation and facilitate DNA methylation. In conclusion, methylation states of H4K20 showed different nucleosome binding strengths for DNMT3A2-3L, highlighting the importance of these key site interactions in DNA methylation establishment and histone modification crosstalk.

Surprisingly, deletion of the DNMT3A2 PWWP domain led to a fivefold increase in nucleosome binding signal (Fig. 4c) despite its known affinity for H3K36me2/3 and DNA, which was confirmed by the AlphaScreen titration assay (Fig. 4e). In the absence of preferred histone modifications such as H3K36me3, the PWWP domain probably disengages with histone H3 tails, leading to reduced nucleosome binding. By contrast, in the presence of these modifications, the PWWP domain significantly enhanced DNMT3A2-3L binding to nucleosomes (Fig. 4e). This observation indicates that the PWWP domain has a key role in multilayered regulations: communicating with H3K36 methylation and/or modulating DNA binding dynamics. The precise regulatory role and mechanism of the PWWP domain await further investigation.

Moreover, by overlaying the recently resolved DNMT3A1 ubiquitin-dependent recruitment (UDR) maps onto our DNMT3A2-3L-nucleosome map, we found that the unoccupied nucleosome acidic patch caused by the 180° rotated conformation of the DNMT3L C-terminal 'switching helix' could potentially accommodate the DNMT3A1 UDR motif (Fig. 4f). This spatial compatibility suggests that the full-length DNMT3A1, when complexed with DNMT3L, may adopt a similar conformation when binding to nucleosomes and function in a similar dynamic DNA engagement model.

To assess the role of the DNMT3L C-terminal 'switching helix' and ADD domain in restoring DNA methylation in cells, we analyzed DNA methylation in DKO8 cells expressing either wild-type or mutant DNMT3L 56 days post infection using the Infinium MethylationEPIC BeadChip array. Consistent with previous reports, wild-type DNMT3L effectively restored DNA methylation to a relatively high level (from a β-value of 0.43 in the empty vector control to 0.71) (Fig. 5a). The E103R and Δα-helix complexes, which showed attenuated nucleosome binding, exhibited slightly reduced DNA methylation compared to the wild-type DNMT3L complex, with β-values of 0.69 and 0.67, respectively. In addition, the DNMT3L ADD deletion complex, which retained only ~10% of nucleosome binding capacity, was much less efficient (with a β-value of 0.64) despite its elevated protein levels (Fig. 5b,c). Although some additional sites were methylated, the decreased DNA methylation compared to wild-type DNMT3L was primarily caused by a failure to methylate specific target regions (Fig. 5d). Both E103R and Δα-helix complexes showed decreased methylation, yet the retained regions shared similar chromatin features with the wild-type DNMT3L, being more enriched in intergenic regions, gene bodies, open seas and solo CpGs, and more associated with H3K27me3 (Fig. 5d,e). By contrast, deletion of the DNMT3L ADD domain resulted in loss of enrichment for intergenic regions, open seas, WCGW sequence context and LINE elements, while gaining enrichment for shores near CpG islands, and more CpG-dense regions (Fig. 5e), indicating that the ADD domain is important for 3L to target the late replication region of the genome to maintain genome integrity. Notably, the Δβ-sheet complex, while maintaining nucleosome binding capacity, showed the lowest levels of DNA methylation in cells (with a β-value of 0.49), probably because of insufficient protein expression (Fig. 5a), and was therefore excluded from chromatin feature analysis. These results emphasize that both the 'switching helix' and the ADD domain of DNMT3L are critical for efficient restoration of DNA methylation in cells, with the ADD domain having a particularly important role in targeting specific genomic contexts.

Although the histone tails, in addition to the nucleosome core, were unresolved in the cryo-EM map, prior studies have shown that DNMT3A complexes interact with histone tails to target methylation to specific genome regions. To explore this possibility, we next used the AlphaScreen assay to assess the binding profiles for the whole panel of DNMT3A2-related complexes. Both DNMT3A2-3B3 and DNMT3A2-3L complexes showed strong nucleosome binding (Fig. 6a,b). Surprisingly, these two complexes displayed distinct behaviors depending on different histone modifications. DNMT3A2-3B3 bound uniformly to nucleosomes regardless of histone modifications, while DNMT3A2-3L nucleosome binding was modulated by histone modifications. Specifically, nucleosomes with H3K4me3, typically associated with active promoters and low DNA methylation, inhibited the binding, whereas nucleosomes carrying H3K36me2/3, which are associated with gene body DNA methylation, exhibited enhanced binding. Further analysis revealed that DNMT3A2 alone bound moderately to H3K36-methylated nucleosomes, while DNMT3L alone showed only very weak binding, suggesting a regulatory role for DNMT3L (Fig. 6b). Additionally, the binding profile of DNMT3B3 mirrored that of DNMT3A2-3B3 (Fig. 6b and Extended Data Fig. 7), indicating that DNMT3B3 might facilitate DNMT3A2-3B3 binding to a broader range of nucleosomes through acidic patch interactions to ensure continuous de novo DNA methylation in somatic cells. By contrast, DNMT3L, predominantly expressed in embryonic cells where DNA methylation patterns are established, probably modulates DNMT3A2-3L complexes to sense histone modifications and co-regulate DNA methylation. Interestingly, the methylation status of linker DNA selectively affected the binding of DNMT3A2-3L to nucleosomes, further suggesting special regulatory roles for the DNMT3L accessory protein (Fig. 6c).

To assess nucleosome binding in a more physiological context, we used biotinylated DNMT3A2-3B3 or DNMT3A2-3L complexes to pull down nucleosomes isolated from living cells. Although DNMT3A2-3B3 showed a higher overall pull-down efficiency (more H3), the trend remained consistent: DNMT3A2-3B3 exhibited higher efficiency (twofold to sixfold) in pulling down nucleosomes with H3K4me3 compared to DNMT3A2-3L, whereas DNMT3A2-3L showed slightly higher enrichment for nucleosomes with H3K36me3 (Fig. 6d). Notably, these effects were not as significant as those observed in the in vitro AlphaScreen assays, indicating a more complex regulation in living cells.

We next used the MTase-Glo methyltransferase assay to evaluate the DNA methylation activity of DNMT3A2-related complexes. As expected, DNMT3A2 alone showed minimal activity, while the accessory protein DNMT3L enhanced DNMT3A2 catalytic activity by approximately twofold, and DNMT3B3 increased it by an additional threefold, consistent with prior findings that heteromeric DNMT3A complexes are more stable and active. Remarkably, histone H3 peptide (1-44) increased DNMT3A2 catalytic activity by approximately 17-fold, far exceeding the stimulation observed with any accessory protein alone (2.3-fold and 7.0-fold). When combined with H3 peptides, accessory proteins further increased activity by another twofold (Fig. 6e and Extended Data Fig. 8a,b). In agreement with previous findings, our data indicated that the gain in DNMT3A2 catalytic efficiency by accessory proteins and histone H3 peptides is primarily driven by an improved catalytic rate (K): in the absence of H3 peptide, DNMT3L enhanced K of DNMT3A2 by tenfold (0.32 h vs 0.03 h) and DNMT3B3 by 30-fold (0.96 h vs 0.03 h); H3 peptide itself increased the catalytic rate more than 50-fold (1.71 h vs 0.03 h), with further stimulation by accessory proteins (Fig. 6e and Extended Data Fig. 8c). By contrast, DNA substrate binding (K) remained relatively stable, ranging from 0.37 µM to 0.83 µM in the absence of H3 peptide and 0.47 µM to 1.03 µM in the presence of H3 peptide (Extended Data Fig. 8c), suggesting that the binding and catalysis are not strictly correlated. Importantly, DNMT3A2-3L mutants that affect nucleosome binding did not reduce the methylation of naked DNA in vitro (Extended Data Fig. 8d), supporting our hypothesis that DNMT3 binding to nucleosomes and methylating DNA are two relatively separate biological processes, potentially regulated by different factors. Notably, our in-cell DNA methylation data and free DNA catalytic analysis suggest that the observed defects in CpG methylation in cells were a result of defects in nucleosome targeting, not DNA binding or catalysis.

In cells, DNA methylation is regulated by the chromatin environment, including multiple post-translational modifications of histones, such as methylation at H3K4 and H3K36. To investigate this process, we tested the stimulation capacity of histone H3 peptides harboring K4me3 and/or K36me3, as these modifications significantly altered the nucleosome binding behavior of the DNMT3A2-3L complex. As expected, H3K4me3-modified peptides diminished the stimulation capacity, even in the presence of accessory proteins. Notably, the DNMT3A2-3B3 complex still retained relatively high activity in the presence of H3K4me3, probably because of the high basal activity of this complex (Fig. 6f). Interestingly, H3K36me3 slightly enhanced the activity of DNMT3A2 and DNMT3A2-3L, and could partially attenuate the inhibitory effect of H3K4me3 (K from 0.83 h to 1.63 h), but only in the DNMT3A2-3L complex, not in DNMT3A2-3B3 (Fig. 6f and Extended Data Fig. 8a,b,e), indicating the histone modification regulation is complicated and environment-dependent. By further investigating the H3 peptide, we identified that the N-terminal region (residues 1-21), which includes K4, retained stimulatory potential (Extended Data Fig. 8e), which is consistent with the previous experiments focused on DNMT3A2 (ref. ).

Based on these findings, we propose that although accessory proteins are important for DNMT3A2 activity, histone H3 may have a more direct stimulatory role in vivo. Meanwhile, accessory proteins also have regulatory functions on DNA methylation in addition to activity stimulation. Further studies will be needed to map functional regions of histone H3 and further elucidate the regulatory roles of histone modifications in DNA methylation in living cells.

The distal DNMT3A2 catalytic domain exhibits weak DNA binding (Extended Data Fig. 5g), suggesting a flexible interaction between DNMT3A2-3L and DNA. To explore the dynamics of the DNMT3A2-3L-NCP complex, we performed 3D classification without alignment in RELION (v.5), which revealed a swing motion in the distal region of the DNMT3A2-3L complex (Extended Data Fig. 2c). This finding was confirmed by the recently developed CryoROLE (CryoEM Relative Orientation LandscapE). Four representative centered points were selected, and superimposition of the reconstructed maps from each point within 5 Å revealed 'up' and 'down' movements of the distal DNMT3A2-3L (Fig. 7a,b). For a more detailed analysis of the relative orientations between DNMT3A2-3L and nucleosomes, 3D reconstructions were calculated from locations at every 10° along each direction. The landscape analysis identified swiveling motion in the α direction and swinging motion in the β and γ directions (Extended Data Fig. 9a and Supplementary Video 1). These movements were also captured by 3D multi-body refinement in RELION (v.5) within the same dataset (Extended Data Fig. 9b and Supplementary Video 2). These findings underscore the dynamics of the DNMT3A2-3L-NCP complex, especially the distal DNMT3A2-3L region and the linker DNA. With an additional cryo-EM dataset, we identified three different conformational states of the nucleosome-bound DNMT3A2-3L complex that recaptured the swing motion: a DNA-engaged state, an intermediate state and an 'up' state (Fig. 7c and Extended Data Fig. 3).

Based on these observations and biochemical data, we have identified DNMT3L as a critical histone modification sensor and propose a model for DNMT3A2-3L-mediated de novo DNA methylation on chromatin in which DNMT3A2-3L dynamically assembles, senses histone tails and interacts with nucleosomes in a flexible manner to achieve optimal DNA engagement for methylation (Fig. 7d). Even in the presence of the accessory protein DNMT3L, DNMT3A2 retains the ability to form higher-order oligomers such as dodecamers and octamers. However, the DNMT3A2-3L heterotetramer is selectively recruited to nucleosomes, stabilizing a high catalytically active form of the complex that can sense histone modifications, cooperate with other epigenetic regulators and dynamically search for DNA engagement through interactions between DNMT3L and nucleosome cores. These flexible interactions, potentially involving charge and polar interactions, could be regulated by histone modifications. Such flexibility allows the distal part of DNMT3A2-3L to move 'up' and 'down', facilitating engagement with linker DNA for methylation.