Nucleosome spacing regulates linker methylation by DNMT3A2/3B3

SUMMARY

De novo CpG methylation (mCpG) is deposited by DNMT3A and DNMT3B, which target DNA linkers between nucleosomes. Cells contain millions of unique linkers, but the rules dictating which linkers get targeted by DNMT3 enzymes are not understood. We show that nucleosome spacing controls linker DNA methylation and H3K36me2 recognition by human DNMT3A2/3B3, linking de novo methylation to chromatin architecture. We present structures of DNMT3A2/3B3 bound to di-nucleosomes, revealing that short linkers promote di-nucleosome bridging, blocking access to linker DNA and suppressing methylation, whereas long linkers allow DNMT3A2/3B3 to engage each nucleosome separately and methylate linker DNA. Finally, we show that DNMT3A2/3B3 position PWWP domains to scan for H3K36me2. However, H3K36me2 recognition is blocked when DNMT3A2/3B3 bridges di-nucleosomes with short linkers, imposing an additional structural constraint on DNMT3A2/3B3 function. Together, these findings uncover the mechanisms that govern de novo methylation in chromatin and explain how DNMT3 enzymes target linkers in cells.

Graphical Abstract

ETOC

Xie et al. show that the spacing between nucleosomes determines whether DNA can be methylated. By revealing how DNMT3A2/DNMT3B3 senses linker length and histone marks, this work explains how chromatin architecture guides where de novo DNA methylation occurs in the genome.

INTRODUCTION

Different patterns of methylated CpG dinucleotides (mCpG) play critical roles in establishing constitutive heterochromatin^1,2, directing various developmental programs^3,4 and regulating interactions between DNA and various transcription factors^5–7. Importantly, dysregulated deposition of mCpG underlies various human diseases including cancer⁸. Therefore, proper cell physiology requires highly regulated mechanisms to direct the accurate deposition of mCpG in the genome.

In humans, DNA methylation patterns are established by the highly conserved de novo DNA methyltransferases DNMT3A and DNMT3B⁴. Due to their critical role in establishing methylation patterns, the catalytic activities of DNMT3A and DNMT3B are tightly regulated to prevent off-target DNA methylation. Both enzymes have weak DNA methylation activity in isolation and are stimulated by the accessory proteins DNMT3L⁹ or DNMT3B3 (a splice variant of DNMT3B)^10,11, which lack catalytic activity. DNMT3A and DNMT3B form elongated hetero-tetrameric complexes with these accessory proteins¹² where the catalytic DNMT3A or DNMT3B subunits are positioned at the center and the non-catalytic DNMT3L or DNMT3B3 subunits are positioned at the ends^12,13. Expression of DNMT3A, DNMT3B, their isoforms and their accessory proteins is tightly controlled because each protein has specific developmental roles^14,15. For example, DNMT3A methylates major satellite repeats and determines allele-specific imprinting during gametogenesis¹⁶, while DNMT3B plays a dominant role in early embryonic development and silences minor satellite repeats¹⁶. In addition, DNMT3L is specifically expressed early in development⁹, while DNMT3B3 is expressed widely in somatic tissues¹⁷.

DNMT3A and DNMT3B also contain specific histone recognition domains that further modulate their activities. The ATRX–DNMT3–DNMT3L (ADD) domain inhibits DNA methylation by blocking DNA binding¹⁸. H3 tails containing unmodified H3K4 interact with the ADD domain and induce a conformational change that allows DNA to bind^18,19, while H3 tails containing H3K4me2-3 cannot bind the ADD domain and, therefore, inhibit DNA methylation²⁰. The Pro-Trp-Trp-Pro (PWWP) domain binds to H3K36me2-3^21–24 and recruits the DNMT3 hetero-tetramer to gene bodies²¹. The PWWP domain also inhibits DNMT3 activity^25,26 and this PWWP-dependent inhibition is relieved through recognition of H3K36me2 modifications in nucleosomes²⁶, but not by peptides²⁵. Therefore, the ADD and PWWP domains inhibit CpG methylation by DNMT3A and DNMT3B and this inhibition is relieved by recognition of H3K36me2-3 and unmodified H3K4 within the nucleosome.

Although the modification state of histone proteins is critical for deposition of mCpG, the histone octamer itself prevents methylation of nucleosome DNA²⁷. Multiple in vitro studies have shown that the DNA wrapped by the nucleosome is highly resistant to methylation by a variety of methyltransferases including DNMT3A and DNMT3B^27,28. Furthermore, in mammals²⁹, plants²⁸ and synthetic yeast models³⁰ DNA methylation is preferentially deposited within DNA linkers. However, nucleosome DNA can also be methylated in the cell, but at lower levels than in the linker DNA²⁸. To explain this discrepancy, current models suggest that DNA methylation is deposited within linkers first and then moves into nucleosomes through processes that disrupt nucleosome positioning or structure^28,31,32. Although it is clear that DNMT3A and DNMT3B methylate linker DNA, it is not understood how these enzymes target the DNA linker or how linker length in higher-order chromatin substrates might affect linker methylation. Our previous Cryo-EM structure of the DNMT3A2/3B3 hetero-tetramer bound to a mono-nucleosome showed that DNMT3B3 binds to the nucleosome acidic patch while DNMT3A2 binds to 10bp of DNA that protrudes from the nucleosome¹³. Furthermore, in vitro DNA methylation studies showed that DNMT3A and DNMT3B preferentially methylate mono-nucleosomes ~10bp from DNA entry site³³. However, these studies were conducted using mono nucleosomes with DNA overhangs, which do not contain natural linker DNA. Therefore, in this study we use di-nucleosome substrates to examine how DNA methylation occurs in the context of bona fide nucleosome linkers.

Here, we show that di-nucleosome linker length and the position of CpG dinucleotides relative to the nucleosome determines the extent of linker DNA methylation by DNMT3A2/3B3 (hereafter, 3A2/3B3). We find that di-nucleosomes with short linkers (5–15bp) potently inhibit DNA methylation by 3A2/3B3, while di-nucleosomes with long linkers (>20bp) permit methylation. We report cryo-EM structures of the 3A2/3B3 hetero-tetramer bound to short linker (5–8bp) and long linker (25bp) di-nucleosomes and to nucleosomes modified with H3K36me2. The 3A2/3B3 hetero-tetramer binds short linker di-nucleosomes by simultaneously bridging each nucleosome surfaces, which prevents the enzyme from binding linker DNA. 3A2/3B3 binding is accompanied by large-scale rearrangements of the di-nucleosome conformation, which are required to enable di-nucleosome bridging. However, in long-linker di-nucleosome substrates 3A2/3B3 recognizes each nucleosome independently, allowing linker methylation. When bound to nucleosome substrates, the ADD domains of 3A2 and 3B3 specifically position a PWWP domain adjacent to the H3 tail exit point to scan for H3K36me2-3 modifications. In short linker di-nucleosome substrates the 3A2 ADD domain blocks the PWWP domain from reading the H3K36me2 modification. However, in mono-nucleosome substrates H3K36me2 recognition by the PWWP domains alters the conformation of 3A2/3B3 on the nucleosome, suggesting a potential mechanism for catalytic activation of the 3A2/3B3 hetero-tetramer.

RESULTS

DNA methylation is regulated by the nucleosome linker length

To systematically dissect effects of linker length on CpG methylation by 3A2/3B3, we reconstituted a set of di-nucleosome substrates with 40bp and 38bp overhangs (comprising a 10bp overhang sequence with 30bp or 28bp sequencing primer binding sites) and linker DNA lengths ranging from 5bp-35bp (Fig. 1a). We hereafter refer to these as Y-X-Y di-nucleosomes, where Y indicates the overhang length and X indicates the linker length. We inserted specific CpG dinucleotides in the linker DNA, then incubated the di-nucleosomes with 3A2/3B3 and the SAM methyl donor to allow linker methylation. After the reaction, we collected endpoints to simultaneously assess the effects of linker length and CpG location on DNA methylation using bisulfite sequencing (Fig. 1b,c). Interestingly, CpG methylation was not equal between the sense and anti-sense DNA strands, suggesting that 3A2/3B3 produces hemi-methylated DNA, agreeing with previous work showing that DNMT3A preferentially forms hemi-methylated products³⁴. There was almost no CpG methylation within 5bp, 8bp or 10bp linkers, small amounts of methylation within the 15bp linker and increased methylation in the 20bp 25bp and 35bp linkers (Fig. 1b). This effect depends on the presence of nucleosomes because there was no correlation between linker length and CpG methylation for identical DNA substrates what were not reconstituted with histones (Fig. 1c). Methylation of the overhang DNA was insensitive to the linker length, showing a low level of methylation for all di-nucleosome substrates (Fig. S1a–b). This indicates that the effects of linker length on DNMT activity are restricted to the linker DNA and that the DNA overhangs used in the methylation experiment are poor substrates for 3A2. Low methylation within 5–15bp linkers was unexpected because the di-nucleosomes used in the reaction contain unmodified H3 tails, and an excess of unmodified H3(1–20) peptide was added to the reaction, both of which stimulate 3A2 activity. Indeed, even without added H3 peptide, H3 tails the di-nucleosome alone were sufficient to drive linker DNA methylation within the 20bp and 35bp linkers (Fig. S1c–d).

Figure 1: DNA linkers regulate methylation by DNMT3A2/3B3.

a Sequences of engineered di-nucleosome linkers. Positions of CpG dinucleotides in the sense and antisense strands are numbered based on their location relative to the upstream nucleosome and are colored red and blue, respectively. b-c Bisulfite sequencing assay showing the percentage of CpG methylation at each linker CpG position. Specific cytosines are indicated with colored arrows. Top: nucleosome substrates, bottom: naked DNA substrates. n=2 for 5bp, 8bp, 10bp, 15bp, 20bp, 25bp linkers. n=1 for 35bp linker. d Electromobility shift assays for DNMT3A2/3B3 binding to the di-nucleosome substrates used in a-c. e Electromobility shift assay for DNMT3A2/3B3 binding to 10-8-10 and 10-25-10 di-nucleosomes. f Manually built models of di-nucleosomes with 20bp and 35bp linkers. CpG positions that are accessible and oriented for methylation are indicated with solid blue and red arrows. CpG positions that are inaccessible and oriented between the di-nucleosome are shown with empty blue and red arrows. g Distribution of linker lengths in mouse ES cells ranging from 0–80bp. Distributions for all linkers in the cell (gray line) and at promoter regions +/− 1000bp from the transcription start site (colored lines). Linker length distributions at promotor regions are separated in quartiles based on gene expression level (blue = no expression, green = bottom 25–50%, yellow = top 25–50%, red = top 25%). Linker lengths that inhibit methylation are indicated with a transparent red box, and linker lengths that allow methylation are indicated with a transparent green box. See also Figure S1.

The reduced linker methylation in di-nucleosomes containing 5–15bp linkers could potentially be explained by impaired binding of 3A2/3B3. However, 3A2/3B3 bound to all di-nucleosome substrates with similar affinities, regardless of linker length (Fig. 1d). The long DNA overhangs used in the methylation and binding reactions could allow extra 3A2/3B3 hetero-tetramers to bind the overhangs in a way that is not sensitive to di-nucleosome linker length, potentially contributing to the similar binding affinity across all nucleosomes. To control for this, we generated long-linker (25bp) and short linker (8pb) di-nucleosomes with 10bp overhangs (lacking the sequencing primer binding sites) that can only interact with a single 3A2/3B3 hetero-tetramer in conjunction with the nucleosome¹³. However, the affinity of 3A2/3B3 for the 10-8-10 di-nucleosome is similar to its affinity for the 10-25-10 di-nucleosome (Fig. 1e). Therefore, closely spaced di-nucleosomes strongly inhibit linker methylation by 3A2/3B3 without reducing binding affinity as measured by EMSA.

Methylation in the 20bp and 35bp linkers showed a pronounced strand-, and position-specific methylation pattern (Fig. 1b). In the 20bp di-nucleosome substrate, the CpGs 12bp away from the upstream nucleosome in both the sense and anti-sense strands are highly methylated while the other CpG sites have low methylation levels (Fig. 1b). The 35bp di-nucleosome exhibits a distinct methylation profile, with high levels of CpG methylation ~12bp away from the upstream nucleosome in the sense strand and anti-sense strands, followed by decreased methylation farther away from the upstream nucleosome (Fig. 1b). Manually built models of di-nucleosomes with 20bp and 35bp linkers show that CpG dinucleotides at position 12 are exposed and oriented for methylation, based on the structure of 3A2/3B3 bound to a mono-nucleosome (Fig. 1f). In contrast, CpGs at other positions are turned into the space between the nucleosomes, potentially explaining why these CpG sites are not methylated efficiently. Therefore, patterns of methylation in the 20bp and 35bp di-nucleosomes are most likely caused by 3A2/3B3 tetramer binding to the upstream nucleosome and using one of its two catalytic domains to methylate exposed CpGs ~10–12bp away (Fig S1e). This agrees with biochemical and structural studies showing that 3A2/3B3 preferentially methylates mono-nucleosome DNA overhangs ~10–12bp from the DNA exit site^13,33. Therefore, linkers shorter than 15bp inhibit DNA methylation and linkers longer than 15bp allow DNA methylation with a preference for CpG dinucleotides 10–12bp away from the nucleosome (Fig. 1b). Together, these results show that closely spaced nucleosomes impose a barrier to DNA methylation without impairing the apparent affinity of 3A2/3B3 binding as measured by EMSA, and that the position of CpG sites relative to the nucleosome are critical for methylation efficiency.

Short nucleosome linkers are enriched at promoter regions

Because linker length regulates CpG methylation by 3A2/3B3 in vitro, we examined how linkers of different lengths are distributed at promoters and how they correlate with transcription, DNA methylation and 3A2 localization in the cell. We re-analyzed previously mapped nucleosome linkers³⁵, ChIP-seq profiles of 3A2, and genome-wide mCpG locations²¹ from mouse ES cells. The most abundant linker length in mammalian chromatin is approximately 35–45bp, showing peaks in linker length every 10 base pairs, with the most prominent peak at 35bp (Fig. 1g). The linker length distribution at promoters with no expression closely matches the genome wide linker length distribution (Fig. 1g, S1f–g). However, at promoters of actively transcribed genes the distribution is shifted toward shorter linkers, especially in the 0bp-15bp lengths (Fig. 1g). Linker length shortening at promoters is correlated with transcription and CpG methylation, as highly expressed promoters have the shortest nucleosome linker lengths (Fig. 1g, S1f–g) and the lowest levels of CpG methylation (Fig. S1h). This effect is specific for the promoter region and is much less pronounced at gene bodies (Fig S1f–g). Because linkers in the 0bp-15bp range cannot be methylated (Fig. 1b), enrichment of short linkers at activate promoters may be a mechanism used by the cell to restrain DNA methylation at expressed genes. Interestingly, nucleosome linkers (0–10bp) are specifically enriched ~1000bp upstream from the TSS at expressed genes, but not silent genes (Fig. S1i). These linker lengths co-localize with 3A2 (Fig. S1j) and sit at the transition point between the demethylated promoters and upstream methylated region (Fig. S1h), potentially indicating a role for short linkers in storing DNMT3A2 or in controlling the spread of CpG methylation from intergenic regions³⁶. Together, these results indicate that nucleosome linker lengths can regulate de novo DNA methylation in vitro and are correlated with methylation patterns in the cell.

Di-nucleosome bridging by DNMT3A2/3B3 blocks access to short linkers

To understand how linker length regulates 3A2/3B3 activity, we determined multiple Cryo-EM structures of 3A2/3B3 bound to unmodified di-nucleosomes containing 10bp overhangs and short DNA linkers (5bp, 6bp, 8bp) that resist methylation and a long linker (25bp) that allows methylation (Fig. 2, S2–3, Table 1). The 10-8-10 di-nucleosome in complex with 3A2/3B3 had the best overall Cryo-EM density and was used for all subsequent analysis (Fig. 2, S2). The consensus map of the complex has an overall resolution of ~4.1Å (FSC = 0.143) with higher local resolution in the nucleosomes (~3Å) and lower resolution in the 3A2/3B3 tetramers (~5–10Å, Fig. S2). When filtered to 10Å for clarity (Fig. 2b–c), the consensus map shows that the overall conformation of 3A2/3B3 on the 10-8-10 di-nucleosome is similar to our previously determined mono-nucleosome bound structure¹³ (Fig. S4a). However, in this structure two 3A2/3B3 tetramers bind symmetrically and bridge the di-nucleosome, allowing each 3B3 subunit to interact with the folded histone core (Fig. 2b–c). Because of the 2-fold symmetry of the structure, we refer to each nucleosome in the complex as either “upper” or “lower” and each subunit of the 3A2/3B3 tetramer as “proximal” or “distal”, depending on their position relative to the upper nucleosome. We generated a single, higher-resolution map (~3.5–5.5Å) of the complex by combining the consensus map and focused maps (Fig. 2d–e, S2) and used this composite map to build a model of the entire 3A2/3B3 di-nucleosome complex (Fig. 2f,g). The 3A2/3B3 tetramer binds to the DNA protruding from either side of the di-nucleosome (Fig. S4b) but does not interact with the di-nucleosome linker itself. In fact, the 3A2 active sites are far away from the 8bp linker (83Å and 66Å, Fig. 2f), which agrees with our DNA methylation assays and explains why the 8bp linker cannot be methylated by 3A2/3B3.

Figure 2: Structure of the DNMT3A2/DNMT3B3 tetramer bound to a di-nucleosome.

a Domain architecture of DNMT3A and DNMT3B isoforms. Catalytic (CD), catalytic-like (CLD), ADD and PWWP domains are indicated. b-c Cryo-EM map of the complex between the DNMT3A2/3B3 tetramer and the 10-8-10 di-nucleosome filtered to 10Å, displayed at a threshold of 0.0239 and colored as in a. d-e Combined focused maps of each DNMT3A2 and DNMT3B3 tetramer and each nucleosome displayed at a threshold of 0.06. Unresolved density for the PWWP domains due to focused refinement is indicated with shaded regions colored pink. f-g Atomic model of the DNMT3A2/DNMT3B3 tetramer bound to a di-nucleosome. Protein domains are colored as in a-e. The 8bp linker DNA separating the two nucleosomes is colored red. h-i Close up view of the proximal (h) and distal (i) acidic patch contact sites in the nucleosome focused maps. Density for the DNMT3B3 arginine anchor is shown as a semi-transparent surface. j Composite map showing the contact between distal DNMT3B3 and the lower nucleosome. k Close up view of the contact point between the distal DNMT3B3 ADD and the histone core. Basic resides of the DNMT3B3 ADD domain in proximity to acidic residues in H3 and H4 are shown in stick representation. See also Figures S2–S4.

Table 1:

Cryo-EM data collection, model refinement and validation statistics

Sample	DNMT3A2/3B3-10n8n10-dinucleosome						3A2/3B3 10n5n10	10n5n10 alone	3A2/3B3 10n6n10	3A2/3B3 10n8n10 H3 peptide	3A2/3B3 ΔPWWP 10n8n10	3A2/3B3 10n25n10	10n8n10 alone
Data Statistics
Microscope	Titan Krios						Talos Arctica	Talos Arctica	Titan Krios	Talos Arctica	Talos Arctica	Talos Arctica	Talos Arctica
Voltage (kV)	300						200	200	300	200	200	200	200
Electron exposure (e−/Å2)	60						50	50	60	50	50	50	50
Pixel size (Å)	0.828						0.96	0.96	0.828	0.96	0.96	0.96	0.96
Symmetry imposed	C1						C1	C1	C1	C1	C1	C1	C1
FSC threshold	0.143						0.143	0.143	0.143	0.143	0.143	0.143	0.143
Map resolution range (Å)	3-10						15-5	10-3	10-3	15-5	15-5	15-5	10-3
Initial particle projections	2584528						96074	367292	2887646	170287	430957	331789	430957
Focus Region	Composite	Upper DNMT	Lower DNMT	Upper Nuc.	Lower Nuc.	Consen-sus	Consen-sus	Consen-sus	Consen-sus	Consen-sus	Consen-sus	Consen-sus	Consen-sus
EMD code	47344	47354	47355	47367	47353	47349	47448	47450	47461	47462	47495	47496	47499
Final particle projections	478167						40908	61854	356421	44359	37292	59084	75131
Map resolution (Å)	3.75						6.08	5.93	3.14	6.08	6.92	4.85	4.24
Model Refinement
PDB code	9E00	9E0F	9E0G	9E0R	9E09	9E05	9E2D	9E2F	9E2Q	9E2R	9E3R	9E3U	9E49
Model composition:
Non-hydrogen atoms	43875	12187	12152	12305	12355	49286	48657	24620	46712	49043	46691	24818	24442
Protein residues	3865	1510	1543	749	752	4499	4444	1459	4188	4822	4183	2248	1458
DNA base pairs	321	0	0	155	156	321	316	318	319	319	319	164	314
Ligands	4	2	2	0	0	4	4	n/a	4	4	4	2	0
Zn ions	18	9	9	0	0	9	18	n/a	18	18	18	9	0
B factors (Å2):
Overall	133.9	156.7	187.6	139.3	147.8	165.7	708.5	494.0	456.5	770.6	883.2	478.7	322.8
Protein/DNA	133.9	156.6	187.6	139.3	147.8	165.5	708.4	494.0	456.4	770.6	883.2	478.5	322.8
Ligands and Zn	310.1	283.5	287.1	n/a	n/a	363.3	912.7	n/a	672.3	770.6	1000	879.7	n/a
Validation:
Bond lengths (RMSD Å)	0.0106	0.0134	0.0117	0.0054	0.0047	0.0054	0.0118	0.0083	0.0123	0.0113	0.0116	0.0116	0.0097
Bond angles (RMSD °)	1.27	1.42	1.44	0.89	0.86	1.04	1.37	1.02	1.33	1.32	1.35	1.36	1.00
MolProbity score	1.45	1.61	1.73	1.07	1.08	1.33	1.49	1.15	1.54	1.53	1.55	1.45	1.09
Clashscore	5.16	4.46	4.29	2.81	2.31	4.15	4.66	3.21	5.67	5.44	5.68	4.51	2.98
Rotamer outliers (%)	0.0	0.0	0.0	0.0	0.0	0.0	0.03	0.00	0.00	0.03	0.00	0.00	0.00
Ramachandran plot
Favored (%)	96.95	96.62	95.53	98.35	97.68	97.32	96.26	97.83	96.47	96.44	96.42	96.59	98.11
Allowed (%)	3.05	3.38	4.47	1.65	2.32	2.68	3.74	2.17	3.53	3.56	3.58	3.41	1.89
Disallowed (%)	0	0	0	0	0	0	0	0	0	0	0	0	0

The ADD and PWWP domains regulate the activities and genomic localization of DNMT3 enzymes^18,21. However, these domains have not been resolved on DNA or nucleosome substrates, so it is not clear how these domains might regulate DNMT3 substrate recognition in the context of chromatin. Our structure shows density for the ADD domains of proximal and distal 3A2 (Fig. 2b–c) and the ADD domain from distal 3B3, however the ADD domain from proximal 3B3 is not resolved (Fig. 2c). The consensus map also contains a globular density positioned at the H3 tail exit site that bridges the ADD domains of proximal 3A2 and distal 3B3 (Fig. 2b). This globular density is absent in composite map dues to high flexibility (Fig. 2d). We determined a structure where the PWWP domains of 3A2 and 3B3 are deleted (~6.9 overall, ~10–15Å 3A2/3B3 tetramer), showing that the globular density that contacts the proximal 3A2 ADD and the distal 3B3 ADD is a PWWP domain from 3A2 or 3B3 (Fig. S4c). Interestingly, the interfaces between the PWWP domain and each ADD domain in our structure are distinct from the PWWP/ADD interaction in the recently reported Cryo-EM structure of the DNMT3B homo-tetramer²⁵ (Fig. S4d). The 3A2 ADD domains in our model closely align with the ADD domains of DNMT3A in an inhibited conformation¹⁸ (Fig S4e–f) indicating that the 3A2/3B3 hetero-tetramer is inhibited in our structure. This also agrees with our DNA methylation assays showing that 3A2/3B3 is inactive on the 40-8-38 di-nucleosome (Fig. 1b). Excess unmodified H3 peptide did not cause the ADD domains to move into the active conformation (Fig. S4g), suggesting that the inhibited state is stabilized in our structure and agrees with our biochemical data showing that excess unmodified H3 peptide does not stimulate methylation within short linkers (Fig. 1b). Together, these data show that 3A2/3B3 does not methylate short linkers because it does not interact with the linker DNA and is stabilized in an inhibited conformation.

Each DNMT3B3 subunit forms distinct interactions with each nucleosome

Because 3A2/3B3 tetramer bridges both nucleosomes using 3B3, we examined the interactions between each 3B3 subunit and the histone octamer core. A previously reported¹³ contact between 3B3 and the acidic patch is recapitulated in the di-nucleosome bound structure for proximal and distal 3B3 (Fig 2g–i). Therefore, interactions between proximal 3B3, distal 3B3 and each acidic patch are the same for each nucleosome.

While residue-level interactions with each acidic patch are identical for each 3B3 subunit, proximal and distal 3B3 contact the nucleosome in distinct conformations. The contact between proximal 3B3 and the nucleosome is similar to the previously determined structure of the 3A2/3B3 tetramer bound to a mono-nucleosome¹³ except that proximal 3A2 moves ~25Å closer to the edge of the nucleosome (Fig. S5a). Proximal 3B3 binds to the upper nucleosome at an angle of ~46° but distal 3B3 binds the lower nucleosome at an angle of ~91° (Fig. 2g). This different contact angle allows the ADD domain from distal 3B3 to contact the nucleosome surface (Fig. 2j). The interaction between the distal 3B3 ADD domain and the histone core is mediated through a loop encompassing residues 498–521 that interacts with the surface of the H3/H4 dimer (Fig. 2k). Due to the resolution of the map, no specific side chain contacts between the ADD and the histone core can be directly determined, and the sidechain rotamers in the model are largely inferred from the crystal structure of the DNMT3B ADD domain that was used for modeling (PDB ID: 7O45). However, the 498–521 loop positions several positively charged residues (R511, R517, K520) near negatively charged residues in the H3/H4 dimer (H3 D77, H3 D73 and H4 D24) indicating that these residues may interact (Fig. 2k). We recently reported a structure of the DNMT3A2/3L tetramer bound to a mono-nucleosome, which shows that the ADD domain of proximal DNMT3L makes contacts with histone H4K20 that are critical for nucleosome binding and sensitive to H3K20me3 methylation³⁷. Therefore, the ADD domains of the DNMT3L and DNMT3B3 accessory subunits make important contacts with histone core that are critical for nucleosome binding. Together, these data show that distinct interactions between each 3B3 subunit and the di-nucleosome enable di-nucleosome bridging and allow the ADD domain from distal 3B3 to bind the lower nucleosome.

Linker length dependence of di-nucleosome binding by DNMT3A2/3B3

To understand how changes in linker length affect 3A2/3B3 interactions with the di-nucleosome, we compared Cryo-EM structures of 3A2/3B3 in complex with the 10-5-10 (~6.1Å overall, ~8–12Å 3A2/3B3 tetramer), 10-6-10 (~3.1Å overall, ~6–10Å 3A2/3B3 tetramer), 10-8-10 and 10-25-10 (4.9Å overall, ~8–12Å 3A2/3B3 tetramer) di-nucleosomes (Fig. 3a–f, S2–3). The DNMT-bound 10-5-10 and 10-6-10 di-nucleosomes adopt a similar conformation to the DNMT-bound 10-8-10 di-nucleosome (Fig. 3d). However, the lower nucleosome is rotated relative to the upper nucleosome by 22° for the 10-5-10 structure and by 8° for the 10-6-10 structure (Fig. 3e). This rotation also changes the conformation of the second 3A2/3B3 tetramer on the opposite face of the di-nucleosome which moves by 49Å in the 10-5-10 di-nucleosome and by 26Å in the 10-6-10 di-nucleosome (Fig. 3e). However, the 22° rotation of the lower nucleosome in the 10-5-10 structure (Fig. 3e) is less than the ~100° rotation expected for b-form DNA based on the 3bp difference in linker length. Therefore, 3A2/3B3 binding may restrain the conformation of the di-nucleosome leading to underwinding of the linker DNA.

Figure 3: Linker length dependence of DNMT3A2/3B3 di-nucleosome bridging.

a-c Gaussian filtered cryo-em maps of DNMT3A2/3B3 bound to the 10-5-10 (a) 10-6-10 (b) and 10-8-10 (c) di-nucleosomes. d Superimposed consensus cryo-EM maps from panel a-c. e Consensus cryo-EM maps of the 10-8-10 DNMT3A2/3B3 complex superimposed with the 10-5-10 (left) and 10-6-10 (right) DNMT3A2/3B3 complexes. Rotations in the lower nucleosome are indicated with the black angle indicator. Movement of the lower DNMT tetramer is indicated with an arrow. f Consensus Cryo-EM map of the 10-25-10 DNMT3A2/3B3 complex with missing density for the lower nucleosome. g Schematic of the experimental setup for binding. h A representative kymograph showing 3A2/3B3-GFP tetramer binding on the 8-bp linker di-nucleosome. i Schematic of the experimental setup for force manipulation. j Representative force–distance curve of a bare 8bp linker di-nucleosome-containing DNA tether showing force-induced transitions. The insert shows a zoom-in view of two example transitions for which the distance change (ΔL) and the transition force are recorded. k Fraction of trajectories on bare 8 or 25 bp linker di-nucleosome (N = 22 for 8bp linker, N=8 for 25bp linker) or 3A2/3B3-GFP tetramer bound 8 or 25 bp linker di-nucleosome (N = 12 for 8bp linker, N=8 for 25bp linker). l Representative force–distance curve of a 3A2/3B3-bound 8bp linker di-nucleosome as shown in j. m Low-force regime transition forces recorded from force-distance curves of bare (N = 10 from 10 independent tethers) or DNMT3A2/3B3-bound (N = 9 from 9 independent tethers) 8-bp di-nucleosomal DNA tethers. Significance was calculated using a two-tailed unpaired t-test (*P ≤ 0.05). See also Figure S5.

The structure of 3A2/3B3 bound to the 10-25-10 di-nucleosome shows clear density for the 3A2/3B3 tetramer in a conformation that matches its interaction with the “upper” nucleosome with no density for the “lower” nucleosome (Fig. 3f). This structure also closely matches the conformation of 3A2/3B3 when bound to a mono-nucleosome (Fig. S5b). The loss of lower nucleosome density is presumably due to high flexibility of the lower nucleosome caused by the longer DNA linker and missing contacts with distal 3B3 (Fig. 3i–k). Therefore, 3A2/3B3 does not bridge di-nucleosomes separated by long linkers, but binds each nucleosome individually.

Given that 3A2/3B3 makes extra contacts when bridging di-nucleosomes with short linkers, we next investigated whether 3A2/3B3 alters the mechanical properties of di-nucleosomes using single-molecule force spectroscopy. We reconstituted unmodified short linker (8-bp) and long-linker (25-bp) di-nucleosomes and tethered them between two biotinylated 6.5-kbp DNA handles which were held between streptavidin-coated beads in a C-Trap instrument that combines dual-trap optical tweezers and single-molecule confocal fluorescence microscopy (Fig. 3g)^38,39. When the tether was held at a constant low force (1 pN), 3A2/3B3-GFP was observed to stably reside at the di-nucleosome site, whereas it exhibited fast diffusive behavior on the flanking bare DNA and dissociated more quickly (Fig. 3h). We then conducted pulling experiments by gradually increasing the inter-bead distance and recording the force measurements on the tether (Fig. 3i). The resultant force-distance (f-d) curves report on the mechanical unwrapping signature of the tethered chromatin and its associated complexes^40,41. For bare di-nucleosomes (with a 8-bp or 25-bp linker), the f-d curves displayed two characteristic abrupt ruptures in the high-force regime (>10 pN) with a transition size of ~24 nm, corresponding to the release of the inner DNA wrap (~75 bp) of each nucleosome (Fig. 3j)⁴². In the low-force regime (<10 pN), we observed a transition in approximately half of the tethers containing a di-nucleosome with an 8-bp linker (Fig. 3j, k). This transition can be partially attributed to disruption of the inter-nucleosomal interaction, as it was not observed when we pulled on tethers containing a mono-nucleosome (Fig. S5c). When incubated with 3A2/3B3, a larger fraction of the 8bp-linker di-nucleosome tethers exhibited a transition in the low-force regime (Fig. 3k, l), which was accompanied by an increase in the average transition force (Fig. 3m). These changes indicate that 3A2/3B3 binding alters the di-nucleosome mechanics, likely by bridging the two nucleosomes. In comparison, pulling on tethers containing a di-nucleosome with a 25-bp linker showed less frequent low-force transitions compared to 8bp-linker di-nucleosome tethers, and the addition of 3A2/3B3 did not increase the frequency of low-force transitions (Fig. 3e, S5d–e). These observations indicate that 3A2/3B3 does not induce or stabilize inter-nucleosomal interaction for di-nucleosomes with a long linker, consistent with the lack of bridging from our structural data. Together, these single-molecule data support our structural results and highlight the plasticity of 3A2/3B3 interactions with di-nucleosomes where bridging is restricted to short-linker di-nucleosomes that inhibit DNA methylation.

The DNMT3A2/3B3 hetero-tetramer alters the di-nucleosome conformation

The similar conformations of the 10-5-10, 10-6-10 and 10-8-10 di-nucleosome 3A2/3B3 complexes (Fig. 3d), as well as changes in the mechanical profile of 8bp-linker di-nucleosomes when bound to 3A2/3B3 (Fig. 3j–m), prompted us to examine how 3A2/3B3 binding may alter the structure of the di-nucleosome substrate itself. Therefore, we determined cryo-EM structures of the 10-8-10 (~4.2Å overall) and 10-5-10 (~5.9Å overall) di-nucleosomes without the 3A2/3B3 hetero-tetramer (Fig. 4a,b S3). Compared to the 10-8-10 di-nucleosome, the 10-5-10 di-nucleosome is more compact, with the lower nucleosome positioned under upper nucleosome, partially occluding its acidic patch (Fig. 4a,b). 3A2/3B3 binding to the 10-5-10 di-nucleosome rotates the lower nucleosome along the DNA linker by 23° and straightens the DNA linker, moving the acidic patch by 61Å (Fig. 4c). 3A2/3B3 binding to the 10-8-10 di-nucleosome rotates the lower nucleosome along the DNA linker by 27° and moves the acidic patch of the lower nucleosome down by 26Å (Fig. 4d). In the DNMT-free states, the lower nucleosome in the 10-5-10 and 10-8-10 di-nucleosomes would overlap with distal 3B3 (Fig. 4c,d) suggesting that these DNMT-free conformations of the 10-5-10 and 10-8-10 di-nucleosomes are incompatible with 3A2/3B3 hetero-tetramer binding. Therefore, these movements allow distal 3B3 to engage the acidic patch of the lower nucleosome but also put strain on the di-nucleosome linker (Fig. 3d,e). Interestingly, the linker DNA is better resolved in the structure of the 10-8-10 di-nucleosome alone than in the structure of the 10-8-10 di-nucleosome in complex with 3A2/3B3 (Fig. S5f,g), even though the complex structure has a higher overall resolution (Fig. S2, S3e). This indicates that strain placed on the linker DNA may increase its flexibility. Therefore, conformational changes in the di-nucleosomes are required for the 3A2/3B3 hetero-tetramer to bridge the di-nucleosome.

Figure 4: DNMT3A2/3B3 manipulates the conformation of the di-nucleosome.

a-b Cryo-EM maps for the isolated 10-5-10 (a) and 10-8-10 (b) di-nucleosomes. The location of the acidic patch in each di-nucleosome is indicated with a red oval. c Consensus cryo-EM map 10-5-10 DNMT3A2/3B3 complex superimposed with the cryo-EM map of the 10-5-10 di-nucleosome alone (light blue). Movements of the lower nucleosome upon DNMT3A2/3B3 tetramer binding are indicated. d Composite cryo-EM map of the 10-8-10 DNMT3A2/3B3 complex superimposed with the 10-8-10 di-nucleosome alone depicted as a semitransparent blue surface. Movements of the lower nucleosome upon DNMT tetramer binding are indicated.

Context dependent recognition of H3K36me2 by PWWP domains

A PWWP domain binds between the ADD domains of proximal 3A2 and distal 3B3 at the H3 tail exit site (Fig. 2b, Fig. 5a). The low resolution of the PWWP domain density did not allow us to model the PWWP domain structure conclusively, so a crystal structure of the DNMT3A PWWP domain was docked into the map and fit into the density as a rigid body. Density for the H3 tail connects to the PWWP, but the PWWP domain does not bind to H3K36 (Fig. 5a) as it is unmodified in the structure. The proximity of the PWWP domain to the H3 tail suggests that it may “scan” the modification state of H3K36 and bind if H3K36 is methylated. Therefore, we determined cryo-EM structures of 3A2/3B3 bound to an H3K36me2 modified mono-nucleosome and 10-8-10 di-nucleosome (Figs. 5b,c, S6–7, Table 2). Overall resolutions for the H3K36me2 modified mono-nucleosome and 10-8-10 di-nucleosome were 3. 63Å and 3.84Å, respectively. However local resolutions for 3A2/3B3 tetramers were much lower (~10Å). Therefore, we combined several maps from focused refinements to generate a single, higher resolution composite map of the complexes (~3.3–5.5Å for the 10-8-10 complex, 4.4–6.4Å for the mono-nucleosome complex). We observed densities for two PWWP domains (PWWP#1 and PWWP #2) in structure of 3A2/3B3 bound to the H3K36me2 modified 10-8-10 di-nucleosome. Surprisingly, PWWP #1 does not bind to H3K36me2 and is positioned in the same “scanning” conformation observed for the 10-8-10 complex with unmodified H3K36 (Fig. 5a,b). However, PWWP #2 binds to H3K36me2 on the opposite copy of histone H3 (Fig. 5b) indicating that, in the context of a short linker di-nucleosome, only one of the H3K36me2 modifications can be recognized. In contrast, when 3A2/3B3 binds to the H3K36me2 modified mono-nucleosome PWWP #1 and #2 engage each H3K36me2 modification (Fig. 5c). The 3A2/3B3 hetero-tetramer binds to long-linker di-nucleosomes in an identical manner as mono-nucleosomes (Fig. S5b) so we expect that the interactions observed on the mono-nucleosome would match interactions on di-nucleosomes with long linkers. When the PWWP domains bind H3K36me2, the entire tetramer rotates on the nucleosome, moving each 3A2 subunit closer to each H3 exit site (Fig. 5d). This movement allows the ADD domain of distal 3A2 to contact PWWP #2. In addition, the large-scale rotation of the tetramer moves the proximal 3A2 ADD domain by 26° to accommodate PWWP #1 binding to H3K36me2 (Fig. 5d). Rotation of the DNMT tetramer is likely required for PWWP #1 to recognize H3K36me2 because the PWWP domain would clash with the ADD domain in the un-rotated conformation that is observed for the 10-8-10 di-nucleosome structures (Fig. S5h). This movement cannot occur in the short linker di-nucleosome because distal 3B3 is tethered to the lower nucleosome. Therefore, 3A2/3B3 bridges short linker di-nucleosomes in a conformation that does not allow PWWP #1 to recognize H3K36me2. Since recognition of H3K36 methylation stimulates 3A2 activity²⁶, blocking recognition H3K36me2 by PWWP #1 may further inhibit 3A2 activity in the context short-linker di-nucleosomes. These structures show that recognition of H3K36me2 by 3A2/3B3 depends on the context of the chromatin substrate. In particular, long linkers allow H3K36me2 recognition by both PWWP domains and short linkers impede H3K36me2 binding by PWWP #1.

Figure 5: The PWWP domain scans the modification state of H3K36.

a, b Cryo-EM maps of DNMT3A2/3B3 in complex with the unmodified (a) and H3K36me2 modified (b) 10-8-10 di-nucleosomes. Close-up views of each PWWP domain docked into the semi-transparent EM density are shown in the insets. H3 tail density connecting the PWWP domains to the nucleosome is outlined in black. c Composite Cryo-EM map of DNMT3A2/3B3 in complex with a H3K36me2 nucleosome with 10bp overhangs. Close-up views of each PWWP domain docked into the semi-transparent EM density are shown in the insets. d 10Å low pass filtered Cryo-EM map of DNMT3A2/3B3 bound to an unmodified nucleosome (gray, EMD-20281) superimposed with the composite Cryo-EM map of DNMT3A2/3B3 bound to a nucleosome modified with H3K36me2 (colored). Movements of DNMT3A2/3B3 that are triggered by PWWP binding to H3K36me2 are indicated using black arrows. e Working model for DNMT3A2/3B3 methylation on DNA linkers. See also Figure S6–S7.

Table 2:

Cryo-EM data collection, model refinement and validation statistics

Sample	DNMT3A2/3B3-10n8n10-H3K36me2-dinucleosome							DNMT3A2/3B3-H3K36me2-mononucleosome
Data Statistics
Microscope	Titan Krios							Titan Krios
Voltage (kV)	300							300
Electron exposure (e−/Å2)	50							50
Pixel size (Å)	0.828							0.828
Symmetry imposed	C1							C1
FSC threshold	0.143							0.143
Map resolution range (Å)	3–10							3–10
Initial particle projections	349410							1569478
Focus region	Consensus	Upper Nuc. PWWP	Upper Nuc. high res	Lower Nuc.	Upper DNMT	Lower DNMT	Composite	consensus	nucleosome	DNMT	PWWP1	PWWP2	composite
EMD code	71604	71609	71607	71608	71605	71606	72487	71301	71306	71302	71304	71305	72318
Final particle projections	176300	16793	176300	123813	176300	176300	NA	111367	25397	48074	21950	25397	NA
Map resolution (Å)	3.84	4.99	3.28	3.51	4.89	5.51	5.5–3.51	3.63	4.43	6.35	4.91	5.86	6.35–3.63
Model Refinement
PDB code	9Y4P							9Q7U
Model composition
Non-hydrogen atoms	91817							48157
Protein residues	4453							2362
DNA base pairs	321							167
Ligands	4							2
Zn ions	18							9
B factors (Å2)
protein	155.05							145.99
DNA	118.68							153.35
Ligands and Zn	330.17							231.54
Validation
Bond lengths (RMSD Å)	0.007							0.005
Bond angles (RMSD °)	0.627							0.879
MolProbity score	1.75							1.78
Clashscore	7.40							7.16
Rotamer outliers (%)	0.68							0.05
Ramachandran plot
Favored (%)	67.55							94.36
Allowed (%)	5.01							5.64
Disallowed (%)	0							0

DISCUSSION

This work sheds light on how 3A2/3B3 target DNA linkers in the context of chromatin and reveals a surprising relationship between DNA linker length and CpG methylation. While long linkers allow 3A2/3B3 to bind each nucleosome independently and hemi-methylate the linker DNA, short DNA linkers enforce di-nucleosome bridging by 3A2/3B3 and block access to the linker DNA to inhibit DNA methylation. The specific linker length preference for 3A2/3B3 is likely determined by the length of the 3A2/3B3 tetramer. Nucleosomes that are spaced farther apart than the distance between the arginine anchors of 3B3 may not allow di-nucleosome bridging and therefore enable linker methylation to occur. Therefore, the 3A2/3B3 tetramer may act as a molecular ruler to determine which linkers are in the correct chromatin context for methylation. This provides a rational basis to understand why heterochromatin and active promoters are methylated to different extents: long linkers within heterochromatin allow DNA methylation while enrichment of short linkers at promoters may impede methylation.

In the cell, it is likely that any effects of linker length on DNA methylation occur in conjunction with the presence or absence of H3K4me2-3 and H3K36me2-3. However, our methylation data indicate that short DNA linkers are not methylated even in the presence of marks that stimulate DNMT3 activity (Fig. 1b). Therefore, linker length may ultimately determine whether a linker can be methylated or not, with histone modifications controlling 3A2/3B3 activity within long linkers that allow methylation. Importantly, chromatin remodeling complexes determine the length of nucleosome linkers genome wide and are critical for the appropriate patterning of DNA methylation²⁸. Therefore, the action of nucleosome remodelers may restrict, or allow DNA methylation in different chromatin regions by adjusting the relative spacing of nucleosomes.

Remarkably, di-nucleosome binding by 3A2/3B3 stabilizes inter-nucleosome interactions, changes the di-nucleosome structure and alters linker DNA winding, suggesting that 3A2/3B3 may have a non-catalytic role in chromatin remodeling. These effects are completely dependent on the length of the DNA linker and only occur when 3A2/3B3 bridges the nucleosomes. Interestingly, higher order interactions between nucleosomes (such as nucleosome stacking) are also sensitive to linker length, and the relative positions between nucleosomes^43,44. In addition, compaction of chromatin domains is critical to regulate gene expression⁴⁵. Therefore, in the context of longer chromatin substrates, 3A2/3B3 binding could disrupt nucleosome stacking interactions that are essential to form higher order chromatin structures. While extended fiber structures have not been observed in tomograms of nuclei^46,47, short structures resembling fibers have⁴⁶. Therefore, 3A2/3B3 binding in either the mono-nucleosome state or bridging conformation may change higher order nucleosome structures which could play a role in regulating deposition of CpG methylation.

The effects of linker length on DNA methylation may also depend on the specific combinations of DNMT isoforms expressed in a cell. 3A2 is expressed in ES cells and in various cancers¹³, but 3A1 is expressed in differentiated tissues¹⁵ and its N-terminal extension (which is missing in 3A2) binds the nucleosome acidic patch in a way that would compete with 3B3^26,48,49. This suggests that the precise composition of 3A/3B isoforms and their accessory proteins may alter interactions with nucleosomes. This agrees with a recent cryo-EM structure of DNMT3A2/3L bound to the nucleosome where only the N-terminal extension of DNMT3A was visible and no folded domains of DNMT3A1 or DNMT3L were resolved²⁶. Because di-nucleosome bridging may be particularly important to restrict DNA methylation in short linkers, different DNMT3 isoforms could potentially allow methylation in shorter linkers or inhibit methylation in longer linkers, depending on the needs of the cell. Furthermore, since isoforms of DNMT3A/3B are expressed in a cell-type-specific manner, these differences could partially account for differences in DNA methylation observed in distinct cell types^13,50–52.

All our structures resolved the DNMT3A2/3B3 tetramer in an inhibited conformation where the DNMT3A2 ADD domains are flipped down. This was surprising because all the substrate nucleosomes used in our study contained unmodified H3K4, which binds the ADD domains and flips them up into an active conformation¹⁸. However, the slow rate of CpG methylation by DNMT3 enzymes^26,34 could indicate that the ADD domain flips into the active state with very low frequency, even in the presence of unmodified H3K4. If the rate of DNMT3A2 catalysis is limited by the ADD domains transiently sampling the active conformation, this would explain why we were not able to isolate any classes with the ADD domains in the active conformation, even with excess H3 peptide. Nevertheless, our comprehensive structural analysis of DNMT3A2/3B3 interactions with various nucleosome substrates details critically important regulation of CpG methylation in chromatin.

Our structures show that recognition of H3K36me2 is highly dependent on the linker length between neighboring nucleosomes, with short linkers blocking PWWP engagement of H3K36me2. Recognition of H3K36me2-3 by the PWWP domains recruits DNMT3 enzymes to gene bodies and intergenic regions^21,53 and also releases PWWP-dependent autoinhibition of DNA methylation^25,26. Therefore, inhibiting H3K36me2 recognition on short linker di-nucleosomes could inhibit DNMT3A2 recruitment or activation on these short linkers. Coupled with the intrinsic resistance of short linkers to DNA methylation, this suggests that short linkers cannot be methylated even if they are labeled with H3K36me2. However, one of the two H3K36me2 modifications can still engage PWWP #2, even in the context of a short linker di-nucleosome (Fig 5b). This suggests that that the specific geometry or multiplicity of H3K36me2-3 modifications within a single nucleosome may tune the activity of DNMT3 enzymes toward DNA linkers.

We now propose a model of DNMT3A2/3B3 methylation on DNA linkers (Fig 5e). DNA methylation is strongly inhibited on short DNA linkers because DNMT3A2/3B3 bridges the nucleosomes that flank the linker, making the linker DNA inaccessible. In this state, one PWWP domain can bind to the modified H3K36me2 modification while the other PWWP domain is positioned at the H3 exit site but cannot bind to H3K36me2 due to clashes with the ADD domain of DNMT3A2. In contrast, DNMT3A2/3B3 binds to nucleosomes with long DNA linkers by engaging each nucleosome individually. This allows DNMT3A2/3B3 to bind the linker DNA and scan the modification state of H3K36 using the PWWP domain positioned at the H3 exit site. On nucleosomes modified with H3K36me2, both PWWP domains bind both copies of H3K36 and the DNMT3A2/3B3 tetramer rotates on the nucleosome to place each DNMT3A2 ADD domain in contact with the H3K36me2-bound PWWP domains. This may make transitions of the ADD domain to the active state upon H3K4 binding more likely, thereby activating the enzyme for linker DNA methylation.

Limitations of the study

A limitation of this study is that we did not capture a structure of DNMT3A2/3B3 in a fully active, DNA-engaged state with cytosine positioned in the catalytic pocket. Although our structures define how nucleosome spacing and H3K36me2 recognition regulate DNMT3A2/3B3 positioning and inhibition on chromatin, they likely represent pre-catalytic or inhibited conformations. As a result, the structural transitions that accompany cytosine engagement and catalysis, including stabilization of the active ADD conformation, remain unresolved. In addition, this study focuses exclusively on DNMT3B3 as the accessory subunit of DNMT3A2. DNMT3L, which functions as the primary accessory protein in germ cells and early development, has distinct structural features and nucleosome interactions compared with DNMT3B3. Therefore, whether DNMT3L enforces similar linker-length–dependent regulation or adopts alternative chromatin-binding modes remains an open exciting question.

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Evan Worden (evan.worden@vai.org).

Materials availability

All unique/stable reagents generated in this study are available from the lead contact with a completed materials transfer agreement.

Data and code availability

• Raw EM data are available from the lead contact upon request. Cryo-EM maps and associated models are deposited in the Protein Data Bank with the following codes. DNMT3A2/3B3 in complex with 10-8-10 Di-nucleosome (Composite: PDB: 9E00 & EMD-47344; Consensus:PDB:9E005 & EMD-47349; Upper DNMT: PDB:9E0F, EMD-47345; Lower DNMT: PDB:9E0G, EMD-47355; Upper nucleosome: PDB: 9E0R, EMD-47367; Lower nucleosome: 9E09, EMD-47353). DNMT3A2/3B3 in complex with 10-5-10 Di-nucleosome (Consensus: PDB: 9E2D, EMD-47448). 10-5-10 Di-nucleosome (Consensus: PDB: 9E2F, EMD-47450). DNMT3A2/3B3 in complex with 10-8-10 Di-nucleosome with H3(1–20) peptide (Consensus: PDB: 9E2R, EMD-47462). DNMT3A2/3B3 in complex with 10-6-10 Di-nucleosome (Consensus: PDB: 9E2Q, EMD-47461). PWWP domains deletion DNMT3A2/3B3 in complex with 10-8-10 Di-nucleosome (Consensus: PDB: 9E3R, EMD-47495). 10-8-10 Di-nucleosome (Consensus: PDB: 9E49, EMD-47499). DNMT3A2/3B3 in complex with H3K36me2 Mono-nucleosome (Composite: PDB: 9Q7U, EMD-72318; Consensus: EMD-71301; DNMT: EMD-71302; PWWP-1 domain: EMD-71304; PWWP-2 domain: EMD-71305; Nucleosome: EMD-71306). DNMT3A2/3B3 in complex with H3K36me2 10-8-10 Di-nucleosome (Composite: PDB: 9Y4P, EMD-72487; Consensus: EMD-71604; Upper DNMT: EMD-71605; Lower DNMT: EMD-71606; Upper nucleosome: EMD-71607; Lower nucleosome: EMD-71608; Upper nucleosome_PWWP1: EMD-71609). Bisulfite sequencing data have been deposited in the GEO at GEO accession number GSE309819.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Sf9 insect cells

Sf9 cells were cultured in ESF 921^™ Insect Cell Culture Medium as a suspension culture at 27 °C.

METHOD DETAILS

DNMT3A2/DNMT3B protein preparation

The human proteins DNMT3A2 and DNMT3B3 were expressed and purified from SF9 insect cells as described previously¹³. Briefly, P1 viruses containing DNMT3A2 and 8xHis-GFP-TEV-DNMT3B3 were used to co-infect SF9 cells and express proteins. The proteins were purified by using a nickel-NTA affinity column. After the removal of the 8xHis-GFP tag by addition of TEV protease, the proteins were further purified using a Superdex 200 size exclusion column. The peak containing the DNMT3A2/3B3 tetramer was collected, concentrated, and flash-frozen in liquid nitrogen and stored at −80°C.

The DNMT3A2 /8xHis-GFP-DNMT3B3 complexes used in single molecule experiments were prepared by the same method, but without removing the 8xHis-GFP tag.

Histone octamer preparation

For nonmodified octamers, soluble X. laevis histone octamers containing 6xHis-SUMO-H2A, H2B, 6xHis-SUMO-H3, and H4 were purified as described previously¹³. Briefly, the histone proteins were co-expressed in E. coli and purified by nickel-NTA affinity chromatography. The 6xHis-SUMO tag was removed by ULP1 protease and the histone proteins were further purified using a Hitrap SP cation exchange chromatography column. The proteins were loaded and washed on the column using buffer A (20mM Tris PH8.0, 500 mM NaCl,10% glycerol, 5mM β-mercaptoethanol) and eluted with a 0% - 100% linear gradient of buffer B (20mM Tris PH8.0, 2M NaCl,10% glycerol, 5mM β-mercaptoethanol) over 6 column volumes (CV). The histones were further purified using a Superdex 200 size exclusion column (Cytiva), and peaks containing the desired octamer were pooled and concentrated. The purified octamer was flash-frozen in liquid nitrogen and stored at −80°C.

Full length histone H3 with the native K36me2 modification was purchased from KS-V peptide. Full length H3K36me2, H2A, H2B and H4 proteins were used to reconstitute histone octamer, and the histone octamer was purified as described previously⁵⁸. In brief, the four histone proteins were mixed based on molar ratio of H2A/H2B to H3/H4=1.1 in unfolding buffer (20mM Tris PH7.5; 7M Guanidine-HCl; 5mM DTT). The mixture was dialyzed against 2L histone refolding buffer (10mM Tris PH7.5; 2M NaCl; 1mM EDTA; 5mM β-mercaptoethanol) with 3 changes over 3 days. Then the mixture was further purified using a Superdex 200 size exclusion column (Cytiva), and peaks containing the desired octamer were pooled and concentrated. The purified octamer was flash-frozen in liquid nitrogen and stored at −80°C.

Preparation of 167bp DNA fragments and mono-nucleosome reconstitution

The 167bp DNA fragment is made and purified as described previously¹³. Briefly, the 167bp Widom 601 DNA fragment was amplified by PCR and purified by Hitrap Q column. The sequence of the 167bp DNA fragment is as follows: 5′ATCGGCCGCCCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTGGCGGCCGAT −3′.

The 10bp overhang DNA sequences are indicated in bold with underline.

167bp mono-nucleosome was reconstituted by the salt dialysis method as described previously⁵⁸.

Preparation of Di-nucleosome DNA fragments

The di-nucleosomal DNA fragments contain two copies of the Widom 601 nucleosome positioning sequence⁵⁹ separated by a variable linker sequence and are flanked by defined overhang DNAs and EcoRV cut sites to excise the DNA from the vector. All DNAs were synthesized and cloned into pBluescript II KS (−) using Genscript custom gene synthesis service. OmniMAX 2 T1R cells (Thermofisher) were transformed with the plasmid containing the di-nucleosome DNA and plated on ampicillin plates. 30 ml of LB medium containing ampicillin (100 mg/ml) was inoculated with single colony and cultured overnight at 37 °C while shaking at 220 rpm. For the DNA growth, 10 ml of preculture was transferred into to each 4L flask containing 1000 ml of terrific broth (TB) and ampicillin at 100 mg/ml. The cells were incubated under vigorous shaking for 18 h at 37°C and harvested by centrifugation in 500-ml centrifuge bottles. Plasmid was extracted as described previously⁶⁰. After RNase A digestion, the plasmid solution was centrifuged at 17,500 rpm for 30 min at 4 °C. The supernatant containing plasmid DNA was filtered and loaded onto a 5-ml HiTrap Q column pre-equilibrated with buffer A (10 mM Tris, pH 8.0, 1 mM Na-EDTA) and washed with 10 CVs of 30% buffer B (10 mM Tris, pH 8.0, 1 mM Na-EDTA). The plasmid was then eluted with a linear NaCl gradient from 300 mM to 1M over 6 CVs. Each fraction was analyzed by 1% (w/v) TAE agarose gel electrophoresis. Fractions with plasmid-DNA were pooled and precipitated by the addition of 1/10 volume of 3M sodium acetate PH 5.2 and 2.5 volumes of 100% ice-cold ethanol. The plasmid DNA pellet was air dried for 30 min at room temperature and dissolved in 10 ml TE 10/0.1 buffer (10 mM Tris-HCl, pH 8.0, 0.1 mM EDTA).

The di-nucleosome DNA insert was excised by addition of EcoRV at 37°C for 20h and complete digestion was confirmed by 6% PAGE analysis. The excised insert was separated from the linearized vector by PEG precipitation as described⁶⁰. 7.5% PEG 6000 effectively precipitated vector DNA, with most of the insertion fragments presented in the supernatant. After precipitation in 3M sodium acetate PH 5.2 and 2.5 volumes of 100% ice-cold ethanol the di-nucleosome DNA pellet was air-dried for 30 min at room temperature and dissolved in 10 ml TE 10/50 buffer (10 mM Tris-HCl, pH 8.0, 50 mM EDTA). The insertion fragment was further purified using a HiTrap Q column with the same elution procedure as the plasmid DNA purification. Fractions containing the di-nucleosome DNA were precipitated and the DNA pellet was air-dried for 10 minutes, dissolved with TE 10/0.1 buffer, and then stored at −80°C. The sequences of the prepared template DNA fragments are listed below. The linker and overhang DNA sequences are indicated in bold with underline. 10-X-10 sequences were used for structural studies and the 40-X-38 were used for DNA methylation assays and bisulfite sequencing.

10-5-10: 5’ATCGGCCGCCACAGGATGTATATATCTGACACGTGCCTGGAGACTAGGGAGTAATCCCCTTGGCGGTTAAAACGCGGGGGACAGCGCGTACGTGCGTTTAAGCGGTGCTAGAGCTGTCTACGACCAATTGAGCGGCCTCGGCACCGGGATTCTCCAGCAGCTACAGGATGTATATATCTGACACGTGCCTGGAGACTAGGGAGTAATCCCCTTGGCGGTTAAAACGCGGGGGACAGCGCGTACGTGCGTTTAAGCGGTGCTAGAGCTGTCTACGACCAATTGAGCGGCCTCGGCACCGGGAT TCTCCAGGGCGGCCGAT 3’

10-6-10: 5’ATCGGCCGCCACAGGATGTATATATCTGACACGTGCCTGGAGACTAGGGAGTAATCCCCTTGGCGGTTAAAACGCGGGGGACAGCGCGTACGTGCGTTTAAGCGGTGCTAGAGCTGTCTACGACCAATTGAGCGGCCTCGGCACCGGGATTCTCCAGCAGCTGACAGGATGTATATATCTGACACGTGCCTGGAGACTAGGGAGTAATCCCCTTGGCGGTTAAAACGCGGGGGACAGCGCGTACGTGCGTTTAAGCGGTGCTAGAGCTGTCTACGACCAATTGAGCGGCCTCGGCACCGGGATTCTCCAGGGCGGCCGAT 3’

10-8-10: 5’ATCGGCCGCCACAGGATGTATATATCTGACACGTGCCTGGAGACTAGGGAGTAATCCCCTTGGCGGTTAAAACGCGGGGGACAGCGCGTACGTGCGTTTAAGCGGTGCTAGAGCTGTCTACGACCAATTGAGCGGCCTCGGCACCGGGATTCTCCAGCAGCTGCAACAGGATGTATATATCTGACACGTGCCTGGAGACTAGGGAGTAATCCCCTTGGCGGTTAAAACGCGGGGGACAGCGCGTACGTGCGTTTAAGCGGTGCTAGAGCTGTCTACGACCAATTGAGCGGCCTCGGCACCGGGATTCTCCAGGGCGGCCGAT 3’

10-25-10: 5’ATCGGCCGCCACAGGATGTATATATCTGACACGTGCCTGGAGACTAGGGAGTAATCCCCTTGGCGGTTAAAACGCGGGGGACAGCGCGTACGTGCGTTTAAGCGGTGCTAGAGCTGTCTACGACCAATTGAGCGGCCTCGGCACCGGGATTCTCCAGGGCGGCCGCACGGGGTGCGGCCGCCACAGGATGTATATATCTGACACGTGCCTGGAGACTAGGGAGTAATCCCCTTGGCGGTTAAAACGCGGGGGACAGCGCGTACGTGCGTTTAAGCGGTGCTAGAGCTGTCTACGACCAATTGAGCGGCCTCGGCACCGGGATTCTCCAGGGCGGCCGAT3’

40-5-38: 5’ATCAGACAGACTGCGACCCACTGGGATGCGCGTACTTGATCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTACGTACTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTTAGAACTACGCGCAGCCCAACTGGCTGGCCAGATGGAT 3’

40-8-38: 5’ATCAGACAGACTGCGACCCACTGGGATGCGCGTACTTGATCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTGAACGTGACTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTTAGAACTACGCGCAGCCCAACTGGCTGGCCAGATGGAT 3’

40-10-38: 5’ATCAGACAGACTGCGACCCACTGGGATGCGCGTACTTGATCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTCGAACGTACGCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTTAGAACTACGCGCAGCCCAACTGGCTGGCCAGATGGAT 3’

40-15-38: 5’ATCAGACAGACTGCGACCCACTGGGATGCGCGTACTTGATCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTAACGCGATTCGCGATCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTTAGAACTACGCGCAGCCCAACTGGCTGGCCAGATGGAT 3’

40-20-38: 5’ATCAGACAGACTGCGACCCACTGGGATGCGCGTACTTGATCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTCAGTCAACGCGCGTCATGATCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTTAGAACTACGCGCAGCCCAACTGGCTGGCCAGATGGAT 3’

40-25-38: 5’ATCAGACAGACTGCGACCCACTGGGATGCGCGTACTTGATCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTCAGTCATAACGCGGCGTATTGACTACTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTTAGAACTACGCGCAGCCCAACTGGCTGGCCAGATGGAT 3’

40-35-38: 5’ATCAGACAGACTGCGACCCACTGGGATGCGCGTACTTGATCTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTCAGTCGCATAACGTACGACGTACGTATTGCGACTACTGGAGAATCCCGGTGCCGAGGCCGCTCAATTGGTCGTAGACAGCTCTAGCACCGCTTAAACGCACGTACGCGCTGTCCCCCGCGTTTTAACCGCCAAGGGGATTACTCCCTAGTCTCCAGGCACGTGTCAGATATATACATCCTGTTAGAACTACGCGCAGCCCAACTGGCTGGCCAGATGGAT 3’

Di-nucleosome reconstitution

Di-nucleosomes were reconstituted as described previously⁶⁰. Briefly, to reconstitute di-nucleosomes, histone octamers and purified di-nucleosome DNA were mixed at a 2.1:1 molar ratio under high salt. Then, the mixture was dialyzed in high-salt buffer for 4 hours followed by gradient dialysis into low salt buffer over 22 hours.

Electrophoretic mobility shift assays

20 nM 10-X-10 di-nucleosomes or 100 nM 40-X-38 di-nucleosomes were mixed with the DNMT3A2/3B3 tetramer (molar ratio ranging from 0 to 32 fold) in EMSA binding buffer (20 mM HEPES pH 7.8, 50 mM KCl, 1 mM MgCl2, 1 mM DTT, 0.1 μM S-adenosyl homocysteine (SAH), 0.1 mg/ml bovine serum albumin, and 5% glycerol). The final volume of the binding reaction was 30 μl. The binding reactions were incubated on ice for 30 minutes and then were examined on a 6% polyacrylamide native gel running in 0.5 x tris-borate EDTA on ice. The gels were stained with SYBR Safe DNA stain (Thermo Fisher) for 15 minutes before being imaged.

DNA methylation assay and bisulfite sequencing

To assess the activity of the DNMT3A2/3B3 complex on di-nucleosomes with variable linker lengths, 10 pmol of each di-nucleosome (100nM) or its corresponding naked DNA was incubated with 100 pmol of purified DNMT3A2/3B3 tetramer (1mM) in 100 ml binding buffer (20 mM HEPES pH7.8, 50 mM NaCl, 1% glycerol, 1 mM DTT, 0.1 mg/ml BSA, 100 mM SAM, and 1 mM histone H3 peptides (residues 1–20)) at 37 °C for 4 hours. To remove free DNA from the reaction with the di-nucleosomes, one unit of AluI (New England Biolabs) was added to the reaction and incubated at 37 °C for 15 minutes. The 601 DNA sequence contains a single AluI site that is blocked by histone binding, allowing for specific cleavage of free DNA not bound to the nucleosome. All reactions were then terminated by heating at 80°C for 20 minutes. The DNA was then immobilized on Agencourt AMPure XP Beads (Beckman Coulter), washed twice with 80% ethanol, eluted with water, and subjected to bisulfite conversion using the EZ DNA Methylation Kit (Zymo Research) following the manufacturer’s instructions. Bisulfite-converted DNA was then amplified by PCR using the ZymoTaq DNA Polymerase (Zymo Research) with the following primers: forward, 5′-GATAGATAGTTGTTGAATTAATGGGATT-3′, and reverse, 5′-CTTCATCTTACCAACCAATTAAACAA-3′. PCR was carried out with the following conditions: 95 °C for 3 minutes, followed by 15 cycles of 95 °C for 10 seconds, 55 °C for 30 seconds, and 72 °C for 1 minute, with a final incubation at 72 °C for 5 minutes. PCR products were then sent out for next-generation sequencing (Genewiz) using the Amplicon-EZ service. The sequencing data were analyzed using BISCUIT software (https://huishenlab.github.io/biscuit/).

Cryo-EM sample preparation for Mono-nucleosome and Di-nucleosome DNMT3A2/3B3 complexes

To prepare the cryo-EM sample for DNMT3A2/3B3 and mono-/ di-nucleosome complexes, mono-/ di-nucleosomes were mixed with DNMT3A2/3B3 at a ratio of 1:2.1. The final concentration of DNMT3A2/3B3 tetramer was typically between 1.6–2.0 mM. The mixture was incubated and dialyzed into binding buffer (25 mM HEPES, pH 7.8, 50 mM NaCl, 1 mM DTT, 100 mM SAH) for 2–4 hours, then loaded. The complex was subjected to GraFix^61,62 in GF buffer (25 mM HEPES, pH 7.8, 30 mM NaCl, 1 mM DTT, 100 mM SAH) with a glycerol gradient of 10–30% and glutaraldehyde gradient of 0–0.15%. The gradient was fractionated from bottom to top with 500 μl for each fraction. All fractions were checked using a 6% polyacrylamide native gel running in 0.5 x tris-borate EDTA on ice. Fractions containing the desired DNMT3A2/3B3 and mono-/ di-nucleosome complex were pooled and dialyzed into sample buffer (25 mM HEPES, pH 7.8, 50 mM NaCl, 1 mM DTT) overnight, concentrated and stored on ice. Cryo-EM grids were prepared by incubating 3 μl of the sample for 1 second on glow-discharged 1.2/1.3 or 2/1 Quantifoil holey carbon grids (SPI Supplies) in the humidity chamber of a Mark IV vitrobot (FEI) maintained at 6°C and 100% humidity. The sample was then blotted for 2.5 seconds before being plunged into liquid ethane.

Data acquisition and processing of the unmodified 10-8-10 Di-nucleosome DNMT3A2/3B3 complex

Cryo-EM movies for 10-8-10 di-nucleosome & DNMT3A2/3B3 complex were collected at 300-kV using a Titan Krios electron microscope equipped with a K3 direct electron detector (Gatan) in super-resolution mode. The defocus range was −1.1 to −1.5 μm, with a nominal magnification of ×105,000, resulting in a calibrated super-resolution pixel size of 0.414 Å. 50 frames were recorded for each movie with a total dose of 61 e/A². 19,437 micrographs were collected and processed. All data sets were processed with CryoSPARC⁵⁴.

CryoSPARC Live installed in a local workstation was used to monitor data collection and preprocess movies during collection. Image preprocessing included patch motion correction (crop factor of 2, pixel size 0.828) and contrast transfer function (CTF) correction. Blob-picking picked particles (200 – 350 Å diameter) were extracted with a binning factor of 4, subjected to 2D classification and good classes were used as templates for further picking. To remove poor-quality images, we set thresholds of CTF fit resolution as 15 Å and ice thickness as 1.16. A total of 18,565 raw micrographs were accepted using these thresholds. Template picking was performed with the accepted images with a reference generated from blob-picking particles. About 3.19 million particles were extracted with a binning factor of 4, and subjected to 2D image classification. Good classes were selected and pooled to result in 2.58 million particles. Multiple round of 3D Ab-initio and heterogeneous refinement ultimately yielded a cleaned particle stack 408,670 unbinned particles. A further heterogeneous refinement job with 2 outputs was performed with 408,670 particles. One of the classes (211,221 particles) showed slightly better density for the PWWP domain and was used as a consensus map for figure preparation.

For particle subtraction and focused refinement, all 408,670 particles were pooled and refined to a resolution of 3.75 Å using the non-uniform refinement job in cryoSPARC. Particle subtraction was performed using a mask around regions of the map that were not included in the focused refinement. After particle subtraction, four separate focused refinements were performed using masks encompassing each DNMT3A2/3B3 tetramer and each nucleosome. The local resolution of all reconstructions were assessed using CryoSPARC local resolution estimation.

The composite map of the 10-8-10 di-nucleosome DNMT3A2/3B3 structure was generated by combining the unsharpened, and DeepEMhancer⁶³ sharpened maps together using the Vop Maximum command in chimera. First, focused maps from each nucleosome were individually combined with a scale factor of 1 for the unsharpened maps and a scale factor of 0.85 for the DeepEMhancer map. Then the focused maps from each DNMT tetramer were individually combined with a scale factor of 1 for the unsharpened maps and a scale factor of 0.9 for the DeepEMhancer maps. The Combined maps for the nucleosomes and the DNMT tetramers were then combined using Vop Maximum with a scale factor of 1.1 for each nucleosome and 0.8 for each DNMT tetramer.

Data acquisition and processing of the H3K36me2 10-8-10 di-nucleosome DNMT3A2/3B3 complex

Cryo-EM movies for H3K36me2 modified 10-8-10 di-nucleosome DNMT3A2/3B3 complex were collected at 300-kV using a Titan Krios electron microscope equipped with a K3 direct electron detector (Gatan) in super-resolution mode. The defocus range was −1.1 to −1.5 μm, with a nominal magnification of ×105,000, resulting in a calibrated super-resolution pixel size of 0.414 Å. 50 frames were recorded for each movie with a total dose of 50 e/A². 19408 micrographs were collected and processed. All data sets were processed with CryoSPARC⁵⁴.

CryoSPARC Live installed in a local workstation was used to monitor data collection and preprocess movies during collection. Image preprocessing included patch motion correction (crop factor of 2, pixel size 0.828) and contrast transfer function (CTF) correction. Blob-picking picked particles (200 – 350 Å diameter) were extracted with a binning factor of 4, subjected to 2D classification and good classes were used as templates for further picking. To remove poor-quality images, we set thresholds of CTF fit resolution as 15 Å and ice thickness as 1.16. A total of 18779 raw micrographs were accepted using these thresholds. 1.71 million particles were picked and extracted with a binning factor of 4 from blob-picking job, and 1.56 million were pooled after 2D classification. Particles were split into 3 subsets and subjected to 3D Ab-initio reconstruction. Good classes were pooled and subjected to several round of heterogenous refinement, yielding a cleaned particle stack 211,621 particles. Template picking was performed with the accepted images with a reference generated from blob-picking particles. About 3.16 million particles were extracted with a binning factor of 4. Multiple rounds of 3D Ab-initio and heterogeneous refinement ultimately yielded a cleaned particle stack 217073 particles. Particles from blob-picking and template picking were combined and duplicated particles were removed yielding 349401 particles. These particles were reextracted at full resolution and classified using a heterogeneous refinement job with 3 outputs. One of the classes (176300 particles) showed complete and better overall density and was used as a consensus map for figure preparation.

For particle subtraction and focused refinement, all 176300 particles were pooled and refined to a resolution of 3.84 Å using the non-uniform refinement job in cryoSPARC. Particle subtraction was performed using a mask around regions of the map that were not included in the focused refinement. After particle subtraction, four separate focused refinements were performed using masks encompassing each DNMT3A2/3B3 tetramer and each nucleosome. The local resolution of all reconstructions were assessed using CryoSPARC local resolution estimation.

The composite map of the H3K36me2 10-8-10 di-nucleosome DNMT3A2/3B3 structure was generated by combining the focused maps sharpened with DeepEMhancer⁶³ and an 8Å low pass filtered consensus map (to bring out pow resolution information) using the vop maximum command in UCSF chimera⁶⁴. First, focused maps for the lower nucleosome, upper DNMT, upper nucleosome with clear PWWP density, lower DNMT, and the higher res upper nucleosome were docked into the consensus map and combined with scale factors of 1, 1, 0.7, 1 and 3, respectively. Focused refinement composite map was then combined with the 8 Å low pass filtered consensus map using vop maximum with a scale factor of 1.0 for merged composite map and 0.8 for lowpass filtered consensus map.

Data acquisition and processing of the H3K36me2 167bp mono-nucleosome DNMT3A2/3B3 complex

Cryo-EM movies for H3K36me2 modified 167 mono-nucleosome & DNMT3A2/3B3 complex were collected at 300-kV using a Titan Krios electron microscope equipped with a K3 direct electron detector (Gatan) in super-resolution mode. The defocus range was −1.1 to −1.5 μm, with a nominal magnification of ×105,000, resulting in a calibrated super-resolution pixel size of 0.414 Å. 50 frames were recorded for each movie with a total dose of 50 e/A². 27063 micrographs were collected and processed. All data sets were processed with CryoSPARC⁵⁴.

CryoSPARC Live installed in a local workstation was used to monitor data collection and preprocess movies during collection. Image preprocessing included patch motion correction (crop factor of 2, pixel size 0.828) and contrast transfer function (CTF) correction. Blob-picking picked particles (200 – 350 Å diameter) were extracted with a binning factor of 4, subjected to 2D classification and good classes were used as templates for further picking. To remove poor-quality images, we set thresholds of CTF fit resolution as 25 Å. A total of 26778 raw micrographs were accepted using these thresholds. 6.58 million particles were picked and extracted with a binning factor of 4 from blob-picking, and 5.54 million were pooled after 2D classification. All particles were split to 3 subsets and subjected to heterogeneous refinement with a reference map generated from an Arctica data set. Good classes were pooled and subjected to several round of heterogenous refinement to yield a cleaned particle stack of 458118 particles. Template picking was performed with the accepted images with a reference generated from blob-picking particles. About 7.15 million particles were extracted with a binning factor of 4, and subjected to 2D image classification. Good classes were selected and pooled to result in 6.03 million particles. Multiple rounds of heterogeneous refinement ultimately yielded a cleaned particle stack 1370595 particles. All particles from blob-picking and template picking were combined and after removing duplicated particles, 1569478 particles were left, which were extracted to at bin 2 and subjected to heterogeneous refinement with 4 outputs. A single class with clear nucleosome features was selected and subjected to an additional round of heterogeneous refinement with 8 outputs. This yielded 6 classes with reasonable densities for the nucleosome and PWWP domains. These particles were pooled and refined to yield a 3.4 Å map with 652144 particles. Class 04 (111367 particles) showed good densities for both PWWP domains and was used as consensus map. Particle subtraction was performed using a mask around regions of the map that were not included in the focused refinement. After particle subtraction, four separate focused refinements were performed using masks encompassing DNMT3A2/3B3 tetramer, nucleosome and two PWWP domains. The local resolution of all reconstructions were assessed using CryoSPARC local resolution estimation.

The composite map of the H3K36me2 167 mono-nucleosome DNMT3A2/3B3 structure was generated by combining DeepEMhancer⁶³ sharpened maps of the nucleosome and the DNMT tetramer with unsharpened maps for PWWP #1 and PWWP #2 using the vop maximum command in chimera⁶⁴ with scale factors of 5,6,1 and 2, respectively.

Data acquisition and data processing of Di-nucleosomes with different linker lengths

Movies of the 10-6-10 Di-nucleosome DNMT3A2/3B3 complex and the 10-8-10 Di-nucleosome (H2BK120R) DNMT3A2/3B3 complex were collected at 300 keV using a Titan Krios electron microscope as described for the 10-8-10 structure. All other samples were collected at 200 keV using a Talos Arctica electron microscope equipped with a K2 direct electron detector with a calibrated super resolution pixel size of 0.46Å.

The Di-nucleosome & DNMT3A2/3B3 complexes with different DNA linkers were processed with CryoSPARC. First, blob-picking was used to acquire the initial particle template. Then, template-based automated particle picking was used to select more particles. Following 2D classification, the desired particles were chosen and subjected to 3D ab initio modeling. The particle set was cleaned using several rounds of heterogeneous refinement. Further refinements, such as homogeneous refinement, CTF refinement, and non-uniform refinement, were utilized to obtain the final maps.

Model building for DNMT3A2/3B3 di-nucleosome complexes

Unmodified 10-8-10 di-nucleosome DNMT3A2/3B3 complex composite map

The model was built into the composite map that was generated from four focused maps. The starting models for the individual subunits were taken from PDB entries 6PA7¹³ (DNMT3A2/3B3, histones, and 601 nucleosome DNA sequence), 4U7P¹⁸ (DNMT3A ADD) and 3LLR⁶⁵ (PWWP). These models were docked into the EM maps using UCSF Chimera⁶⁴ and rigid body-refined using PHINEX⁵⁵, followed by manually adjustment and rebuilding in COOT⁵⁶. The DNA linker region was manually generated and fitted into the density in COOT. The DNMT3B3 catalytic-like domain and the arginine finger peptides that interact with the nucleosome acidic patch were initially adopted from the model 6PA7 and were refined against the composite map. However, the regions of the connecting loop between the CLD core domain and the arginine finger peptide was truncated due to the disordered density. The whole model was refined against the composite map using phenix.real_space_refine in PHENIX.

Unmodified 10-8-10 di-nucleosome DNMT3A2/3B3 complex consensus map

The DNMT3A2/3B3 10-8-10 di-nucleosome complex model generated from the composite map was docked into the consensus map. The PWWP and ADD domains that were not visible in the composite map were docked into the EM density, corrected with Coot and refined against the map using Phenix.

Unmodified 10-8-10 di-nucleosome focused maps for nucleosome 1, nucleosome 2, DNMT tetramer 1 and DNMT tetramer 2

Each portion of the DNMT3A2/3B3 10-8-10 di-nucleosome complex composite model was used to dock into nucleosome 1, nucleosome 2, DNMT tetramer 1 and DNMT tetramer 2 focus maps, respectively, and refined in PHENIX.

10-8-10 and 10-5-10 di-nucleosomes alone

The each nucleosome from the DNMT3A2/3B3 10-8-10 di-nucleosome complex model was docked into the EM map and the linker region was manually built in COOT. The final models were refined against the EM density in PHENIX.

DNMT3A2/3B3 10-8-10 di-nucleosome complexes with H3 peptide, or PWWP deletion

The DNMT3A2/3B3 10-8-10 di-nucleosome complex model from the composite map was docked into the respective density maps. Models were refined with PHENIX against the respective EM density maps.

Unmodified di-nucleosome DNMT3A2/3B3 complexes with 10-5-10, 10-6-10, or 10-25-10 linkers

Each nucleosome each DNMT from the10N8N10 di-nucleosome complex model docked into respective EM maps. The DNA linker regions were manually re-built to fit into each density map in COOT. The models were manually re-adjusted and refined in PHENIX.

H3K36me2 modified 10-8-10 di-nucleosome complex composite map

The structure of the unmodified 10-8-10 di-nucleosome complex was docked into the composite map and used as the initial model for the di-nucleosome and the DNMT3A2/3B3 tetramers. The DNA linker and DNA overhang was manually rebuilt in coot. PWWP #1 from the unmodified 10-8-10 di-nucleosome complex did not fit the density well, so the PWWP domain was docked into the low-resolution density for PWWP#1 and fit using ridged body refinement in Chimera⁶⁴ and Coot. EM density for PWWP#2 was sufficiently resolved to confidently place the DNMT3A PWWP domain structure (PDB: 3LLR) into the density. The H3 tail from the structure of the DNMT3B PWWP domain complex with H3K36me3 (PDB: 5CIU) was placed in the H3K36 binding site of DNMT3A PWWP domain, mutated to H3K36me2 and merged with the H3 protein in the nucleosome using Coot. After the initial model was constructed, the atomic coordinates were refined in PHENIX with Secondary structure restrains imposed on all chains.

DNMT3A2/3B3 H3K36me2 modified 167bp mono-nucleosome complex composite map

The initial model was constructed using the nucleosome from the DNMT3A2/3B3 mono-nucleosome complex structure (PDB:6PA7) and the DNMT3A2/3B3 tetramer from the unmodified 10-8-10 di-nucleosome complex. EM densities for PWWP #1 and PWWP #2 were sufficient to confidently place the DNMT3A PWWP domain structure (PDB: 3LLR) into the density for each PWWP domain. The H3 tail from the structure of the DNMT3B PWWP domain complex with H3K36me3 (PDB: 5CIU) was placed in the H3K36 binding site of for each PWWP domain, mutated to H3K36me2 and merged with each H3 protein in the nucleosome using Coot. After the initial model was constructed, the atomic coordinates were refined in PHENIX with secondary structure restrains imposed on all chains.

All the EM map statistics and refinements were listed in Tables 1&2. All figures were prepared in UCSF ChimeraX⁵⁷.

Single-molecule force spectroscopy

Experimental setup

Single-molecule experiments were performed on a LUMICKS C-Trap instrument at room temperature. This instrument combines three-color confocal fluorescence microscopy with dual-trap optical tweezers³⁸. Data were acquired using LUMICKS Bluelake software version 1.6.16. Rapid optical trap movement was achieved using a computer-controlled stage in a flow cell containing 5 separate channels. Channels 1–3 were separated by laminar flow and were used to tether di-nucleosome substrates between two 4-μm streptavidin-coated polystyrene beads (Spherotech) held in optical traps. In channel 1, a single bead was caught in each trap. The traps were next moved to channel 2, and biotinylated the di-nucleosome substrate was tethered between the two beads as indicated by an increase in the force reading. The tether was then moved into channel 3 containing incubation buffer (30 mM HEPES 7.6, 50 mM sodium chloride, 1 mM dithiothreitol) supplemented with 0.1 mM SAH, and the flow was stopped. Channel 4 was loaded with either 2 or 10 nM 3A2/3B3-GFP in incubation buffer with 0.1 mM SAH. Flow was turned off when collecting data and visualizing protein behavior.

Fluorescence detection

GFP-labeled 3A2/3B3 was excited by the 488 nm laser line. Kymographs were generated by confocal line scanning through the center of the two beads at 100 ms/line. To visualize the behavior of 3A2/3B3-GFP, the di-nucleosome tether was held at 1 pN of constant tension and moved into channel 4 for a specified duration. The tether was then moved into channel 3 containing incubation buffer only for removal of nonspecific binding events and continued imaging.

Force manipulation

Di-nucleosomal DNA tethers (bare or bound with 3A2/3B3-GFP) were first relaxed by lowering the distance between traps in channel 4 until ~0 pN of force was reached. The force was zeroed, and the tether was subjected to pulling by moving one trap relative to the other at a constant velocity of 0.1 μm/s until the DNA entered the overstretching regime (~65 pN) or the tether broke. For experiments with 3A2/3B3-GFP, tethers were recorded only when the 3A2/3B3-GFP fluorescence signal was detected at the di-nucleosome site for the duration of pulling.

QUANTIFICATION AND STATISTICAL ANALYSIS

Details of statistical tests and the exact number of replicates (n) are provided in the corresponding figure legends. All analyses were performed using R and GraphPad Prism 10. For the single-molecule experiments, the P value was determined from a two-tailed unpaired t-test using GraphPad Prism 10 (*P ≤ 0.05).

Supplementary Material

Document S1. Figures S1–S7.

KEY RESOURCES TABLE

REAGENT OR RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
BL21 pLysS	Thermo Fisher Scientific	Cat#: C606010
DH5alpha	New England Biolabs	Cat#: C2987H
Ominimax	Homemade	N/A
Rosetta (DE3)	Homemade	N/A
MAX Efficiency™ DH10Bac Competent Cells	Thermo Fisher Scientific	Cat#: 10361012
DH10Bac Competent Cells	Homemade	N/A
Chemicals, peptides, and recombinant proteins
Glutaraldehyde solution	Sigma-Aldrich	Cat#: G7651
BS3 (bis(sulfosuccinimidyl)suberate)	Thermo Fisher Scientific	Cat#: A39266
S-Adenosylhomocysteine (SAH)	Cayman Chemical	Item No. 13603
S-adenosylmethionine (SAM)	New England Biolabs	Cat#: B9003S
Histone H3 (1–20) - 1 mg	AnaSpec	Cat#: AS-62753
Human H3K36me2 Histone	KS-V Peptide	Cat#: H3111
Xenopus H3 K36me3 C110A (trimethylated K36)	The Histone Source at Colorado State University	Cat#: XH3_K36me3
Xenopus H2B	The Histone Source at Colorado State University	Cat#: XH2B
Xenopus H4	The Histone Source at Colorado State University	Cat#: XH4
MTase-Glo™ Methyltransferase Assay	Promega	Cat#: V7601
Deposited data
Composite structure of DNMT3A2/3B3-10n8n10-dinucleosome complex	This work	EMD-47344 PDB: 9E00
Upper DNMT tetramer structure of DNMT3A2/3B3-10n8n10-dinucleosome complex	This work	EMD-47354 PDB: 9E0F
Lower DNMT tetramer structure of DNMT3A2/3B3-10n8n10-dinucleosome complex	This work	EMD-47355 PDB: 9E0G
Upper nucleosome structure of DNMT3A2/3B3-10n8n10-dinucleosome complex	This work	EMD-47367 PDB: 9E0R
Lower nucleosome structure of DNMT3A2/3B3-10n8n10-dinucleosome complex	This work	EMD-47353 PDB: 9E09
Consensus structure of DNMT3A2/3B3-10n8n10dinucleosome complex	This work	EMD-47349 PDB: 9E05
Consensus structure of DNMT3A2/3B3-10n5n10-dinucleosome complex	This work	EMD-47448 PDB: 9E2D
Consensus structure of 10n5n10-dinucleosome alone	This work	EMD-47450 PDB: 9E2F
Consensus structure of DNMT3A2/3B3-10n6n10-dinucleosome complex	This work	EMD-47461 PDB: 9E2Q
Consensus structure of DNMT3A2/3B3-10n8n10-dinucleosome -H3(1–20) complex	This work	EMD-47462 PDB: 9E2R
Consensus structure of DNMT3A2/3B3DPWWP - 10n8n10-dinucleosome complex	This work	EMD-47495 PDB: 9E3R
Consensus structure of DNMT3A2/3B3-10n25n10-dinucleosome complex	This work	EMD-47496 PDB: 9E3U
Consensus structure of 10n8n10-dinucleosome alone	This work	EMD-47499 PDB: 9E49
Consensus structure of DNMT3A2/3B3-10n8n10-H3K36me2-dinucleosome complex	This work	EMD: 71604
Upper Nuc. and PWWP structure of DNMT3A2/3B3-10n8n10-H3K36me2-dinucleosome complex	This work	EMD: 71609
Upper Nucleosome structure of DNMT3A2/3B3-10n8n10-H3K36me2-dinucleosome complex	This work	EMD: 71607
Lower Nucleosome structure of DNMT3A2/3B3-10n8n10-H3K36me2-dinucleosome complex	This work	EMD: 71608
Upper DNMT tetramer structure of DNMT3A2/3B3-10n8n10-H3K36me2-dinucleosome complex	This work	EMD: 71605
Lower DNMT tetramer structure of DNMT3A2/3B3-10n8n10-H3K36me2-dinucleosome complex	This work	EMD: 71606
Composite structure of DNMT3A2/3B3-10n8n10-H3K36me2-dinucleosome complex	This work	EMD: 72487 PDB: 9Y4P
Consensus structure of DNMT3A2/3B3-H3K36me2-mononucleosome complex	This work	EMD: 71301
Nucleosome structure of DNMT3A2/3B3-H3K36me2-mononucleosome complex	This work	EMD: 71306
DNMT tetramer structure of DNMT3A2/3B3-H3K36me2-mononucleosome complex	This work	EMD: 71302
PWWP1 structure of DNMT3A2/3B3-H3K36me2-mononucleosome complex	This work	EMD: 71304
PWWP2 structure of DNMT3A2/3B3-H3K36me2-mononucleosome complex	This work	EMD: 71305
Composite structure of DNMT3A2/3B3-H3K36me2-mononucleosome complex	This work	EMD: 72318 PDB: 9Q7U
Unprocessed gel images	Mendeley data	DOI: 10.17632/hcj5zyc3kp.1
Experimental models: Cell lines
SF9 insect cells	Thermo Fisher	Cat#: 11496015
Oligonucleotides
GAAGACCTCAACGGCCGCCACAGGATGTATATATC	This work	167_BbsI F
GAAGACAGACCAGGCCGCCCTGGAGAATCCCGGTG	This work	167_BbsI R
GATAGATAGTTGTTGAATTAATGGGATT	This work	Bisulfite sequencing F
CTTCATCTTACCAACCAATTAAACAA	This work	Bisulfite sequencing R
ATCGGCCGCCCTGGAGAATCC	This work	167_F
ATCGGCCGCCACAGGATGTATATATC	This work	167_R
Recombinant DNA
pEWL1	This work; GenScript	N/A
pEWL2	This work; GenScript	N/A
pEWL3	This work; GenScript	N/A
pEWL4	This work; GenScript	N/A
pEWL5	This work; GenScript	N/A
pEWL6	This work; GenScript	N/A
pEWL7	This work; GenScript	N/A
pEWL8	This work; GenScript	N/A
pEWL9	This work; GenScript	N/A
pEWL10	This work; GenScript	N/A
pEWL11	This work; GenScript	N/A
pEWL12	This work; GenScript	N/A
pEWL13	This work; GenScript	N/A
pEWL14	Xu et al. (2020)13	N/A
pEWL15	This work	N/A
pEWL16	Xu et al. (2020)13	N/A
pEWL17	This work	N/A
pEWL18	Xu et al. (2020)13	N/A
Software and algorithms
CryoSPARC	Punjani et al. (2017)54	RRID:SCR_016501
PHENIX	Adams et al. (2010)55	RRID:SCR_014224
COOT	Emsley et al. (2004)56	RRID:SCR_014222
BISulfite-seq CUI Toolkit (BISCUIT)	https://huishenlab.github.io/biscuit/	N/A
UCSF ChimeraX	Goddard et. al. (2018)57	RRID:SCR_015872
GraphPad Prism 10	https://www.graphpad.com	RRID:SCR_000306
R (v4.3.1)	https://www.r-project.org/	RRID:SCR_001905
LUMICKS Bluelake software version 1.6.16	https://lumicks.github.io/bluelakeapi/1.6.16/index.html	N/A

Highlights.

Nucleosome spacing dictates whether linker DNA can be methylated

DNMT3A2/3B3 bridges closely spaced nucleosomes to block linker methylation

DNMT3A2/3B3 binding actively reshapes short-linker di-nucleosome structure

PWWP recognition of H3K36me2 is regulated by nucleosome spacing