DNMT1 reads heterochromatic H4K20me3 to reinforce LINE-1 DNA methylation

DNA methylation and trimethylated histone H4 Lysine 20 (H4K20me3) constitute two important heterochromatin-enriched marks that frequently cooperate in silencing repetitive elements of the mammalian genome. However, it remains elusive how these two chromatin modifications crosstalk. Here, we report that DNA methyltransferase 1 (DNMT1) specifically ‘recognizes’ H4K20me3 via its first bromo-adjacent-homology domain (DNMT1BAH1). Engagement of DNMT1BAH1-H4K20me3 ensures heterochromatin targeting of DNMT1 and DNA methylation at LINE-1 retrotransposons, and cooperates with the previously reported readout of histone H3 tail modifications (i.e., H3K9me3 and H3 ubiquitylation) by the RFTS domain to allosterically regulate DNMT1’s activity. Interplay between RFTS and BAH1 domains of DNMT1 profoundly impacts DNA methylation at both global and focal levels and genomic resistance to radiation-induced damage. Together, our study establishes a direct link between H4K20me3 and DNA methylation, providing a mechanism in which multivalent recognition of repressive histone modifications by DNMT1 ensures appropriate DNA methylation patterning and genomic stability.

T he eukaryotic genome is organized into different functional compartments, with the assembly of heterochromatin regulated by both DNA methylation and repressive histone modifications, such as H3 trimethylated at lysine 9 (H3K9me3) and H4 trimethylated at lysine 20 (H4K20me3) 1 . The signaling cascades invoked by these modifications coordinately underpin fundamental biological processes concerning genome compartmentalization, gene silencing, genomic stability, cell differentiation, and development 2,3 . In cancer, depletion of H4K20me3 is closely associated with DNA hypomethylation in repetitive sequences, such as long interspersed nuclear element-1 (LINE-1) 4 , resulting in genomic instability and/or aberrant gene expression [5][6][7] . Dysregulation of H4K20me3 and DNA methylation has also been associated with neurological and developmental disorders, such as fragile X syndrome 8,9 and Hutchinson-Gilford Progeria Syndrome (HGPS) 10 . However, it remains far from clear how these two gene-repressive epigenetic modifications cooperate in determining specific chromatin states during normal and pathological development.
In mammals, DNA methylation is stably propagated by DNA methyltransferase 1 (DNMT1) during mitotic division 11,12 . DNMT1-mediated DNA methylation maintenance is supported in part by its enzymatic preference for hemimethylated CpG DNA, the so-called maintenance methylation activity 11 . In addition, increasing evidence has suggested a role of de novo methylation activity of DNMT1 in maintaining DNA methylation patterns at H3K9me2/3-enriched 13 or paternal imprinting control regions 14 . DNMT1 contains a C-terminal methyltransferase (MTase) domain, preceded by several regulatory domains including a replication-foci-targeting sequence (RFTS), a CXXC zinc finger domain and a pair of bromo-adjacent homology (BAH) domains (Fig. 1a) [15][16][17][18] . Previous studies have demonstrated that the RFTS and CXXC domains serve to "sense' specific chromatin cues, which in turn influences DNA methylation maintenance via allosteric regulations 15,[17][18][19][20][21] . For instance, the DNMT1 RFTS domain (DNMT1 RFTS ) recognizes ubiquitinated histone H3 (H3Ub) [22][23][24][25] and PCNA-associated factor 15 (PAF15) 26,27 during S phase, which leads to cell cycle-specific chromatin targeting and enzymatic stimulation of DNMT1 23,26,28 . Our recent study further demonstrated that DNMT1 RFTS also directly recognizes H3K9me3, thereby reinforcing the DNMT1 RFTS -H3Ub readout for the DNA methylation maintenance at heterochromatic regions 29 . Likewise, the DNMT1 CXXC domain specifically recognizes unmodified CpG DNA to control DNMT1-mediated de novo DNA methylation 15,30 . It has been demonstrated that a~30-amino acid-long linker connecting the CXXC and BAH domains, termed autoinhibitory linker ( Fig. 1a) 15 , plays a pivotal role in the transition of DNMT1 between different conformational and functional states 15,18 . However, the mechanism by which these or other regulatory elements are coordinated in fine-tuning the activity of DNMT1 at discrete chromatin regions remains poorly understood.
The BAH domain belongs to an evolutionarily conserved class of reader modules for histone and non-histone proteins 31 . Our previous study identified that the BAH domain of origin recognition complex subunit 1 (ORC1 BAH ) specifically recognizes H4K20 dimethylation (H4K20me2), thereby regulating ORC1chromatin association and DNA replication initiation 32 . More recently, the BAH domains of two homologous plant proteins, EARLY BOLTING IN SHORT DAYS (EBS) and SHORT LIFE (SHL), effector of polycomb repression 1 (EPR-1) in Neurospora crassa, as well as BAH domain and coiled-coil containing 1 (BAHCC1) in human cells, have been shown to mediate Polycomb gene repression via specific readout of H3K27 trimethylation (H3K27me3) 4,[33][34][35] . In DNMT1, the two BAH domains are both associated with the MTase domain to form an integrated structural unit 15 . Although previous studies indicated that these domains play a role in targeting DNMT1 to replication foci 14,36 , their functions remain uncharacterized to date.
To elaborate how the BAH domains regulate DNMT1-mediated DNA methylation, we examined the histone binding activity of DNMT1 BAH domains and its relationship to the chromatin association and enzymatic activity of DNMT1. We identified the first BAH domain of DNMT1 (DNMT1 BAH1 ) as a reader for H4K20me3. Strikingly, the H4K20me3 binding by DNMT1 BAH1 causes displacement of the autoinhibitory linker, which frees DNMT1 from the linker-mediated autoinhibition and leads to allosteric stimulation of DNMT1, reminiscent of what was previously observed for the DNMT1 RFTS -H3K9me3Ub interaction 23,29 . Consistently, single-molecule Förster resonance energy transfer (smFRET) analysis revealed that the DNMT1 RFTS -H3K9me3Ub and DNMT1 BAH1 -H4K20me3 interactions both trigger fast conformational dynamics of DNMT1, promoting the transition of DNMT1 into an open conformation. In cells, the DNMT1 BAH1 mutation, which is defective in recognition of H4K20me3 and yet hyperactive due to the disrupted association between DNMT1 BAH1 and the autoinhibitory linker, exerts a dual effect on DNA methylation-it causes DNA hypomethylation within the H4K20me3positive LINE-1, but DNA methylation gains at genomic regions lacking H4K20me3, thereby demonstrating a role for DNMT1 BAH1 in shaping the landscape of DNA methylation. Finally, we also found that RFTS-and BAH1-mediated DNMT1 regulations cooperate to maintain proper levels of DNA methylation and genome stability of cells. Together, this work establishes a direct link between H4K20me3 and DNMT1-mediated DNA methylation, providing mechanistic insights into DNA methylation maintenance at LINE-1 and other heterochromatic regions.
H4K20me3 binding displaces the autoinhibitory linker for enzymatic stimulation. Previously, the autoinhibitory linker (residues 692-727; blue in Fig. 2a) was shown to reinforce the RFTS-mediated autoinhibition through both stabilization of the RFTS-MTase association and occlusion of the DNA substrate ( Fig. 2a and Supplementary Fig. 3a) 17,18,21 . Interestingly, structural comparison of the bDNMT1 BAH1 -H4K20me3 complex with the histone-free hDNMT1 351-1600 (PDB 4WXX) 18 revealed that the H4K20me3-binding site of DNMT1 BAH1 is shielded by the Cterminal half of the autoinhibitory linker in free hDNMT1 351-1600 , via extensive intramolecular interactions (Fig. 2a, b), implying that the H4K20me3 binding to BAH1 would lead to displacement of the autoinhibitory linker, thereby providing a potential mechanism for releasing DNMT1 from autoinhibition. To explore this possibility, we first tested whether the autoinhibitory linker and H4K20me3 compete against each other for BAH1 binding. Using ITC assays, we found that the hDNMT1 BAH1 domain binds to a peptide derived from the autoinhibitory linker with a K d of~70 µM (Fig. 2c); this binding was abolished in the presence of a twofold excess of H4K20me3 peptide (Fig. 2c). The ITC assay also reveals that, unlike the hDNMT1 BAH1 domain alone that binds to H4K20me3 peptide with a K d of 2.7 µM (Fig. 1d), the linkercontaining DNMT1 351-1600 only weakly binds to the H4K20me3 peptide, with a K d of over 300 µM ( Supplementary Fig. 3b). These data establish that the competition for BAH1 binding exists between the H4K20me3 peptide and the autoinhibitory linker in a histone-free form of DNMT1.
BAH1-H4K20me3 and RFTS-H3K9me3Ub2 bindings both drive to a conformationally dynamic state of DNMT1. The BAH1-H4K20me3 binding-mediated allosteric activation of DNMT1 is reminiscent of the previously reported RFTS-H3K9me3Ub2 readout 29 . We therefore asked how the BAH1-H4K20me3 and RFTS-H3K9me3Ub2 interactions crosstalk. Toward this direction, we performed smFRET experiments to interrogate how the RFTS-H3K9me3Ub and BAH1-H4K20me3 interactions impact the conformation of DNMT1. To measure the inter-domain movement between the RFTS and MTase domains, we introduced two mutation sites, one on the RFTS domain (S570C) and the other on the Cterminal end of the RFTS-CXXC linker (T616C), to a hDNMT1 fragment (residues 351-639 followed by a LPETG sequence) for statistical labeling with an equimolar mixture of FRET donor (Cy3) and acceptor (Atto647N) (Fig. 3a). Such a labeling strategy permitted close proximity between the two fluorophores in histone-free hDNMT1, taking advantage of the fact that the RFTS-CXXC linker is C-terminally anchored to the MTase domain through helical packing (Fig. 3b). This hDNMT1 fragment was then sortase-ligated 38 with a Cterminally biotinylated hDNMT1 fragment, creating a Cy3, Atto647N-labeled hDNMT1 fragment encompassing residues 351-1606 (denoted hDNMT1 CY ) for smFRET observation ( Fig. 3c-e).
SmFRET between the two fluorophore labels was observed for wild-type hDNMT1 CY , typically characterized by stable FRET observed from the beginning of the movie until acceptor photobleaching ( Fig. 3c), which is consistent with the closed, autoinhibitory structure of the enzyme. Addition of the H3K9me3Ub2 peptide to surface-immobilized hDNMT1 CY led to a dramatic reduction in the duration of FRET events, with a population of molecules cycling between non-FRET and FRET states (Fig. 3d). In comparison with the peptide-free FRET events, which had a single-exponential distribution of event durations, addition of the peptide therefore led to appearance of an additional, faster, exponential phase (Fig. 3f). These data confirm a previous MD analysis that showed enhanced mobility of the RFTS domain upon the H3Ub2 binding 23 , and suggest that the H3K9me3Ub2 binding promotes DNMT1 transition from a stable, autoinhibitory state to an activated state permitting fast "closed"-"open" conformational interconversion. Intriguingly, the presence of increasing concentrations of H4K20me3 peptide, but not H3K4me3 peptide, leads to a similar reduction in the duration of FRET events (Fig. 3e, g and Supplementary Fig. 3d-f), supporting the model that H4K20me3 binding influences the conformational dynamics of DNMT1. The W796A mutant showed an intrinsically reduced duration of FRET events ( Fig. 3h and Supplementary Fig. 3g-i), regardless of the addition of H4K20me3 peptide (Fig. 3h), which reinforces the notion that the hyperactivity of this mutant arises from the detachment of the autoinhibitory linker from the BAH1 domain of DNMT1. Together, these smFRET data support a model where specific interactions, modulated by peptide binding to the RFTS domain or the BAH1 domain, gate access to a conformationally dynamic, activated state of DNMT1.
BAH1-H4K20me3 binding potentiates CpG methylation deposition at H4K20me3-demarcated genomic regions, with the most striking effect seen at LINE-1 elements. Given that Dnmt1 deletion is lethal to dividing somatic cells, but not mouse ES cells 39 , we turned to mouse ES cells to examine the role of DNMT1 BAH1 in regulating DNA methylation maintenance. Using our recently reported gene complementation system 29 , we introduced comparable levels of exogenous DNMT1 WT or a BAH1-defective mutant (DNMT1 W796A ) into Dnmt1-knockout mouse embryonic stem cells (1KO-ESCs) (Fig. 4a). In agreement with the above structural and biophysical assays of DNMT1 BAH1 , we found that DNMT1 WT significantly co-precipitated with H4K20me3 ( Fig. 4b), and as previously observed 25 , co-localizes with punctate DAPI-dense heterochromatin foci marked with H4K20me3 ( Fig. 4c and Supplementary Fig. 4; upper panels). In contrast, DNMT1 W796A shows much reduced co-precipitation with H4K20me3 (Fig. 4b) and the reduced H4K20me3 costaining ( Fig. 4c and Supplementary Fig. 4; lower panels).
Further, we carried out genome-wide methylation profiling with enhanced reduced representation bisulfite sequencing (eRRBS), which showed nearly complete bisulfite conversion  Fig. 5d). Close inspection of eRRBS data, however, revealed that there was a slight but significant decrease in overall CpG methylation at the H4K20me3-demarcated genomic regions in 1KO-ESCs expressing DNMT1 W796A , relative to the DNMT1 WT controls, with the observed decreases being more apparent and significant at regions with the highest H4K20me3 (Supplementary Fig. 5e). In contrast, there was a slight but significant increase of overall CpG methylation at randomized control regions that lack H4K20me3 (Supplementary Fig. 5f, g). We have further defined differentially methylated regions (DMRs), either hypomethylated (hypo-DMR) or hyper-methylated (hyper-DMR), by comparing eRRBS profiles of 1KO-ESCs with DNMT1 W796A vs. DNMT1 WT . With two independent DMR calling methods (Supplementary Data 2 and 3), we consistently observed a significant enrichment of the DNMT1 W796A -associated hypo-DMRs at H4K20me3-marked regions, relative to genome background or DNMT1 W796A -associated hyper-DMRs ( Fig. 4d and Supplementary Fig. 5h). Meanwhile, compared to genome background, the DNMT1 W796A -associated hypo-DMRs were found significantly depleted from the H3K9me3-marked regions ( Fig. 4e and Supplementary Fig. 5i), indicating that different stimulating mechanisms exist to ensure optimal DNA methylation among H3K9me3-and H4K20me3-demarcated heterochromatin. As a control, CUT&RUN of H4K20me3 demonstrated no change of overall H4K20me3 levels in these cells, as exemplified by retrotransposon elements known to be targeted by H4K20me3 40,41 (Supplementary Fig. 5j, k). These genomic profiling results are in agreement with a notion revealed by our in vitro studies that the W796A mutation of BAH1 has dual effects, with one causing the impaired CpG methylations within H4K20me3-marked regions (due to disrupted association of DNMT1 with H4K20me3) and the other leading to a generally  enhanced methylation (due to the reduced association between the mutated BAH1 and the autoinhibitory linker of DNMT1 and hence, hyperactivity of this mutant). Interestingly, a majority of the DNMT1 W796A -related hypo-DMRs were found localized within the LINE-1 class of retrotransposon elements (also known as L1; Fig. 4f and Supplementary Fig. 6a), especially the 5′ cis-regulatory region of a newly evolved, non-truncated L1MdA subfamily (Fig. 4g, h and Supplementary Fig. 6b and Supplementary Data 4). LINE elements are known to be targeted by H4K20me3 and DNA methylation 40,41 , and in our eRRBS data, had comparable coverage and sequencing depth relative to the Alu/SINE family of repetitive elements and gene-coding regions ( Supplementary  Fig 6a, right panels). By individual bisulfite sequencing, we verified decreased CpG methylation at the examined regulatory region of L1 in cells with DNMT1 W796A relative to DNMT1 WT (Fig. 4i) (Fig. 4j), which was concurrent with CpG hypomethylation of the examined L1 as well (Fig. 4k). Altogether, these results support that DNMT1 BAH1 binding to H4K20me3 potentiates DNMT1-mediated CpG methylation of H4K20me3demarcated regions, notably those newly evolved non-truncated LINE-1 elements.
BAH1 and RFTS regulations cooperate in fine-tuning DNMT1 activity. Our recent study has identified that W464 and W465 within the DNMT1 RFTS module are essential for H3K9me3Ub2 recognition (Fig. 5a); introduction of the W464A and W464/W465A mutation leads to severely impaired RFTS-H3K9me3Ub2 binding, and consequently, a large reduction of DNA methylation in cells 29 . To further examine the interplay between the RFTS-and BAH1-mediated regulations of DNMT1, we combined the W796A mutation of DNMT1 BAH1 together with the RFTS-defective mutation, either W465A or W464A/W465A, and subsequently used such RFTS/BAH1 compound mutants for DNMT1 reconstitution in 1KO-ESCs (Supplementary Fig. 7a). Intriguingly, both LC-MS analysis of cytosine methylation (Supplementary Fig. 7b) and eRRBS (Fig. 5b, c and Supplementary Fig. 7c-e) revealed that, relative to the single mutant of RFTS (DNMT1 W465A ) 29 , the compound mutant of RFTS/BAH1 (DNMT1 W465A/W796A ) displayed a markedly enhanced methylation activity in cells, leading to partial but significant rescue of the global CpG hypomethylation phenotype caused by DNMT1 W465A , including the RFTS-related defects seen at the H3K9me3-marked regions as exemplified by sub-telomeric regions located at the chromosome 1 (Fig. 5d). Given the intrinsic link between the hyperactivity of DNMT1 W796A and the BAH1-mediated DNMT1 activation, these data therefore show that the BAH1-mediated DNMT1 activation may partially rescue the CpG methylation defects caused by the RFTS dysfunction. Note that such a rescue effect was not observed for the RFTS W464A/W465A mutation in the resultant triple mutation of DNMT1 W464A/W465A/W796A (referred to as "TM"; Fig. 5b-d and Supplementary Fig. 7b-d), presumably due to the more severe impairment of chromatin targeting capability associated with this mutant 29 .
RFTS and BAH1 coordinately regulate DNMT1-mediated genomic stabilization. We further challenged 1KO-ESC cells, reconstituted with WT or mutant DNMT1, with ionizing radiation (IR) followed by the neutral comet assay (Fig. 5e). First, compared to 1KO-ESC cells rescued with WT DNMT1, those with vector control or a catalytically inactive DNMT1 mutant (C1226S) showed similar hyper-sensitivity to IR ( Supplementary Fig. 7f-h), thus confirming a direct link between DNMT1-mediated DNA methylation and genomic stability. We next examined whether the BAH1 and RFTS mutations crosstalk during the IR response. As previously demonstrated 29 , cells with the RFTS W464A/W465A and W465A mutants displayed severe impairment of IR resistance (Fig. 5e, f). The compound mutation of DNMT1 W464A/W465A/ W796A (Fig. 5e, f) led to a stronger phenotype of DSB persistence following IR treatment, in line with severe hypomethylation seen with this TM mutation ( Fig. 5b-d). In contrast, while cells with the RFTS single mutant (DNMT1 W465A ) exhibited the impaired genome integrity post-treatment of IR, introduction of an additional BAH1 mutation (DNMT1 W465A/W796A ) significantly rescued this defect (Fig. 5e, f), in line with the rescuing effect of the BAH1 W796A mutation manifested in LC-MS-and eRRBS-based methylome analysis (Fig. 5b-d and Supplementary  Fig. 7b). Consistently, cell survival analysis revealed that the DNMT1 W465A/W796A mutation led to a significant reduction in IR sensitivity in comparison with the DNMT1 W465A mutation ( Supplementary Fig. 7i). Thus, it is evident that DNMT1-mediated DNA methylation is involved in maintenance of genomic stability, the molecular detail of which awaits further investigation.

Discussion
The spatio-temporal regulation of DNMT1-mediated DNA methylation is essential for the faithful inheritance of DNA methylation patterns during mitotic division. DNA methylation, in cooperation with other gene silencing mechanisms, such as histone modifications, provides a mechanism for the long-term stability of the heterochromatic state. However, how DNA methylation crosstalks with histone modifications remains elusive. Through a set of structural, biochemical, computational, and Fig. 4 The BAH1-H4K20me3 binding regulates DNMT1-mediated methylation at the H4K20me3-enriched genomic regions. a Immunoblots of the indicated Flag-tagged DNMT1 after its stable reconstitution into 1KO-ESCs among the independently derived cell lines. cellular analyses, this study addresses two critical questions regarding the functional regulation of DNMT1. First, this study links the repressive histone modification H4K20me3 directly to the reader activity of DNMT1. Here, we show that DNMT1 directly recognizes H4K20me3 through the regulatory domain BAH1, which in turn mediates the chromatin association and enzymatic activation of DNMT1. The BAH1-H4K20me3 interaction clarifies a putative role of the BAH domains of DNMT1 in genomic targeting 14,36 , and provides a crucial safeguard mechanism for efficient DNA methylation at H4K20me3-enriched regions, in particular the repetitive element LINE-1, thereby providing further insights into the mechanism underlying DNA methylation maintenance and heterochromatin formation. Distinct from the typical Kme3 readout that depends on aromatic or other hydrophobic residues only 43 , the BAH1 domain recognizes H4K20me3 via a pocket formed by mixed aromatic and acidic residues. In the DNA-free state of DNMT1, the H4K20me3-binding site is occluded by the C-terminal portion of the autoinhibitory linker, which likely reinforces the intramolecular RFTS-linker-MTase association ( Supplementary Fig. 8a),  (c)). White dots are the median and box lines are the first and third quartile of the data. d Representative IGV view of CpG methylation at an H3K9me3-marked genomic region located in the chromosome 1 among three replicated 1KO-ESC lines with stable expression of the indicated DNMT1. Cytosines covered by at least ten reads according to eRRBS data (with three replications merged for each group) are shown, with each site designated by a vertical line. e Neutral comet assays, experimental scheme (top) and DNA breaks (bottom). f Quantified DNA breaks after ionizing radiation (IR) treatment of 1KO-ESC cells reconstituted with vector control or the indicated DNMT1. Box-and-whisker plots in panel depict 25-75% in the box, whiskers are 10-90%, and median is indicated. Data are mean ± s.d. from >100 cells (n = 3 biologically independent replicates). One-way ANOVA with post Tukey analysis was used. ****p < 0.0001. The data for vector, WT, W465A, and W464A/W465A were adopted from those published previously 29 . ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22665-4 resulting in a direct link between the H4K20me3 binding and enzymatic activation of DNMT1. It is worth noting that de novo P767T mutation of DNMT1, located next to the H4K20me3binding pocket of the BAH1 domain ( Supplementary Fig. 8b), is reportedly linked to schizophrenia 44 . How the disease-associated DNMT1 BAH1 mutations affect genomic methylation and disease progression awaits further investigation.
While epigenetic silencing of repeated elements such as endogenous retroviruses (ERVs) and LINE-1 elements is crucial for maintaining integrity of the mammalian genome, the underlying silencing mechanisms are rather complex. Previous studies have pointed to a set of repressive complexes involved in H3K9me3 [45][46][47] or DNA methylation 48,49 pathways. In this work, we uncovered that the interaction between the BAH1 domain of DNMT1 and H4K20me3 is especially important for optimal DNA methylation at the 5′ cis-regulatory region of the newly evolved, non-truncated LINE-1 subfamily. A recent study also reported the repression of evolutionarily young LINE-1 by N 6methyladenine 50 . It is conceivable that complicated interplays exist among H3K9me3, H4K20me3, and DNA methylation by DNMT1 along with cofactors 29 and DNMT3A/3B (in complex with DNMT3L 51,52 ). Further investigation is warranted to determine the relative contribution of these epigenetic pathways to repression of different subgroups of ERVs and LINE-1 elements.
The specific interaction between DNMT1 BAH1 and H4K20me3 adds to the growing list of BAH domain-mediated histone readouts, which include the H4K20me2 readout by the BAH domain of ORC1 32 and the H3K27me3 readout by the BAH domains of effector proteins in Neurospora crassa, plants and mammals 4,33-35 . Intriguingly, these BAH domains harbor a similar histone-binding surface, but with distinct binding modes ( Supplementary Figs. 2b and 8c, d), highlighting the evolutionary divergence of the histone modification-binding mechanisms of this reader module family.
Second, this study delineates how DNMT1 transduces a multitude of environmental cues into its targeting and enzymatic activity. It has been established that Ubiquitin-like, containingPHD and RING Finger domains, 1 (UHRF1) plays a critical role in activating and targeting DNMT1 during S phase: UHRF1-mediated ubiquitylation of PAF15 26,27 and histone H3 22-25 serves to recruit DNMT1 to the replication foci during early and late S phase, respectively. This study demonstrates that the BAH1-H4K20me3 readout cooperates with the RFTS-H3K9me3/H3Ub readout in allosterically stimulating DNMT1, thereby uncovering another regulatory axis in DNA methylation maintenance in heterochromatin domains (Fig. 6). Disruption of these histone-directed regulations in cells impairs the CpG methylation by DNMT1, leading to an aberrant landscape of DNA methylation and defects in the maintenance of genome stability, a phenotype known to be associated with cancer 53 and developmental disorders 54 . Note that neither RFTS-nor BAH1-mediated DNMT1 activation involves the discrimination of the methylation state of DNA substrates, providing a potential mechanism for compensating the "imprecise" maintenance methylation activity of DNMT1, thereby strengthening the region-specific methylation maintenance 13,55 . It is worth mentioning that maintenance of histone lysine (e.g., H3K9 and H4K20) methylation has been shown to be gradually established following S phase and may differ between parental and newly incorporated histones [56][57][58] . For instance, the H4K20 of recycled histones is dominantly methylated during replication, whereas the new histones only become methylated in the G2/M phase 56 . In this regard, it is likely that H4K20me3-and H3K9me3-mediated DNMT1 targeting and activation may be initiated by recycled parental nucleosomes and persist beyond S phase, providing a mechanism that cooperates with the S phase-specific, UHRF1mediated regulation for efficient DNA methylation maintenance. Indeed, recent studies have indicated that the de novo methyltransferase activity of DNMT1, partially mediated by the BAH domains, is important in reinforcing DNA methylation maintenance in imprinting control or H3K9me3-marked regions 13,14 .
Together, this study uncovers a signaling pathway in which a BAH domain harbored within DNMT1 acts to transform H4K20me3-marked chromatin domains, in particular LINE-1 elements, into a more stable, DNA damage resistant, heterochromatic state with optimal levels of DNA methylation.

Methods
Plasmids. The DNMT1 cDNA was purchased from Addgene (cat # 24952). For cellular assays, full-length DNMT1 with an N-terminal 3×Flag tag was inserted into the pPyCAGIP vector 59 , a kind gift of I. Chambers. All of the DNMT1 point mutations were generated by a QuikChange II XL Site-Directed Mutagenesis Kit (Agilent). For all the in vitro assays, DNA encoding the human DNMT1 BAH1 domain (residues 728-900, hDNMT1 BAH1 ), the bovine DNMT1 BAH1 domain (residues 725-897, bDNMT1 BAH1 ), or the JMJD2A-Tudor domain (amino acids 894-1011, JMJD2A TD ) was cloned into a modified pRSF-Duet vector, which introduced an N-terminal His 6 -SUMO tag and ULP1 (ubiquitin-like protease 1) cleavage site. To facilitate crystallization, residues 838-858 of bDNMT1 BAH1 were replaced by a GAGSA sequence. For analysis of hDNMT1 351-1600 methylation activity and H4K20me3 binding, the interaction between hDNMT1 BAH1 and the autoinhibitory linker and the interaction between the hDNMT1 BAH2 domain (901-1107, hDNMT1 BAH2 ) and H4K20me3, the hDNMT1 constructs were inserted into an in-house expression vector as a His 6 -MBP-tagged form. For peptide binding assay, hDNMT1 BAH1 , hDNMT1 fragment spanning the BAH1, BAH2, and MTase domains (residues 728-1600, hDNMT1 728-1600 ) or JMJD2A TD (amino acids 895-1011) was cloned into pGEX6P-1 vector. For fluorescence labeling, an hDNMT1 fragment (residues 351-639), containing C409S/C420S/S570C/C580S/ T616C mutations and a C-terminal Lys-Pro-Glu-Thr-Gly tail, referred herein to as hDNMT1 RFTS-linker , was also cloned into the modified pRSF-Duet vector. For sortase ligation with hDNMT1 RFTS-linker , an hDNMT1 fragment (residues 646-1606) containing a C-terminal Lys-Pro-Glu-Thr-Gly tail was cloned into the in-house expression vector preceded by a His 6 -MBP-tag. All plasmid sequences were verified by sequencing before use, with the residue numerations based on the isoform 1 of DNMT1 that contains 1616 amino acids. The primers used for cloning are listed in Supplementary Table 2.
Protein purification. The E. coli BL21(DE3) RIL cells (Novagen Inc) were used for protein expression. The cells harboring the expression plasmids were first cultured in LB medium at 37°C until the OD 600 (optical density at 600 nm) reached 0.6. Totally, 0.1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) was then added to induce the protein expression at 16°C overnight. After harvesting, the cells were resuspended and lysed in a buffer containing 50 mM Tris-HCl (pH 7.5), 25 mM imidazole, 1 M NaCl, 0.5 mM DTT and 1 mM PMSF. To purify the His 6 -SUMOtagged hDNMT1 BAH1 , bDNMT1 BAH1 or JMJD2A TD protein, the clarified supernatant was applied to a nickel column and the fusion protein was eluted with buffer containing 25 mM Tris-HCl (pH 8.0), 100 mM NaCl and 300 mM imidazole. The eluted protein was then subjected to tag removal by ULP1 cleavage, anion exchange chromatography on a HiTrap Q XL column (GE Healthcare), nickel affinity chromatography, and size-exclusion chromatography on a HiLoad 16/600 Superdex 75 pg column (GE Healthcare) that was pre-equilibrated with buffer containing  Structural elements of DNMT1, including RFTS, CXXC, autoinhibitory linker and BAH1 are colored in corn, gray, blue and pink, respectively. The BAH2 and MTase domains are colored in cyan. Histone marks (H3K9me3 and H4K20me3) cooperate with UHRF1-generated H3Ub or PAF15Ub in activation and targeting of DNMT1 for optimal DNA methylation at specific genomic loci, which in turn leads to gene repression and genomic stabilization. Note that the H3K9me3Ub and H4K20me3 marks likely regulate the activity of DNMT1 in a synergistic manner. 20 mM Tris-HCl (pH 7.5), 50 mM NaCl and 5 mM DTT. For ITC binding assay, human ORC1 BAH (residues 1-185) was purified as described previously 32 . The His 6 -MBP tagged DNMT1 proteins were sequentially purified via Ni 2+ chromatography, ion-exchange chromatography on a Heparin HP (GE Healthcare) or Q HP column (GE Healthcare), TEV protease cleavage for tag removal, a second round of nickel affinity chromatography, and size-exclusion chromatography on a Superdex 200 16/600 column (GE Healthcare). The GST-tagged hDNMT1 BAH1 and JMJD2A TD for tri-methylated histone peptides-binding assay was purified through GST-affinity chromatography and eluted in 100 mM Tris-HCl (pH 8.0), 10 mg/mL reduced glutathione (Sigma-Aldrich). Purification of GST-hDNMT1 728-1600 fusion protein also involved ion-exchange chromatography on a Heparin HP column and size exclusion chromatography on a HiLoad 16/600 Superdex 200 pg column (GE Healthcare). The final protein sample was stored in a buffer containing 20 mM Tris-HCl (pH 7.5), 250 mM NaCl, 5% Glycerol, and 5 mM DTT. DNMT1 mutants were introduced by site-directed mutagenesis and purified as that described for wild-type protein. For preparation of H3K9me3Ub2 peptide, the His 6 -SUMOtagged Ub(G76C) protein was purified through Ni 2+ chromatography, followed by tag removal via ULP1 cleavage and a second step of Ni 2+ chromatography. All purified protein samples were stored at −80°C before use.
Chemical modification of histone peptide. The H3K9me3Ub2 peptide was generated and purified as previously described 29 , following a published protocol 37 . In essence, the H3 1-24 K9me3K18CK23C peptide was synthesized from LifeTein LLG, with an additional C-terminal tyrosine for spectroscopic quantification. To prepare the H3K9me3Ub2 peptide, the Ub(G76C) protein was mixed with the H3 1-24 K9me3K18CK23C peptide in a 4:1 molar ratio in buffer containing 250 mM Tris-HCl (pH 8.6), 8 M urea and 5 mM TCEP, and incubated at room temperature for 30 min. The cross-linker 1,3-dichloroacetone was then added to the reaction mixture with the amount equal to one-half of the total sulfhydryl groups. After 2 hincubation on ice, the reaction was stopped by 5 mM β-Mercaptoethanol. The H3 1-24 K9me3Ub2 peptide was further purified using a mono S column (GE Healthcare). The caveat of this chemical modification method is that the crosslinking to be introduced is only an analog of the isopeptide bond between Ub and lysine.
smFRET data analysis. Data processing were performed with in-house MATLAB scripts 70 . Individual single-molecule fluorescence traces were extracted from raw movie data, filtering for those that showed colocalized green and red fluorescence. A substantial fraction of the molecules exhibited significant fluorophore photophysical instability, with significant excursions in the donor intensity on a seconds time scale, including transient periods of complete donor quenching. To minimize the possibility of artifacts, traces were therefore curated manually and their event timings were assigned manually. Traces for assignment were thus selected with the following characteristics: (1) The beginning of all assigned FRET events was either the beginning of the movie or a frame in which there was an abrupt increase in the red signal with an anticorrelated decrease in the green signal, and (2) the end of all assigned events was an abrupt decrease in the red signal with an anticorrelated increase in the green signal. The resulting FRET event time durations for each condition were then converted to an empirical cumulative distribution function and fit to a single-exponential (P(t) = 1 − e -λt ) or double-exponential ðpðtÞ ¼ Að1 À e Àλ 1 t Þ þ ð1 À AÞð1 À e Àλ 2 t ÞÞ model as required to maximize the R 2 value from nonlinear least-squares regression.
Cell lines and tissue culture. The murine Dnmt1-knockout (1KO) ESCs were used as we previously described 29 . In brief, 1KO-ESCs were transfected with the pPyCAGIP empty vector or that carrying WT or mutant DNMT1, followed by drug selection with 1 μg/mL puromycin for over two weeks. Both the pooled stableexpression lines and independent single-cell-derived clonal lines were established and verified (such as immunoblotting of DNMT1) before use.
Antibodies and Western blotting. The cultured cells were collected, rinsed in cold phosphate-buffered saline (PBS), and suspended in buffer containing 50 mM Tris-HCl (pH 8.0), 150 mM NaCl and 1% NP-40. After brief sonication and centrifugation, the soluble fractions of cell lysates were mixed with 2× SDS-PAGE loading buffer and boiled for 5 min, followed by loading onto a SDS-PAGE gel for immunoblotting analysis, as previously described 51,71 . Information for the antibodies used in this work is included in the Supplementary Table 3.
Quantification of 5-methyl-2′-deoxycytidine (5-mdC) in gDNA. Measurements of the global levels of 5-mdC in cellular DNA were carried out as described previously 29,51 . Briefly, gDNA was digested into mononucleosides by nuclease P1 and alkaline phosphatase. The enzymes in the digestion mixtures were removed by chloroform extraction, and the resulting aqueous layer was dried, reconstituted in water, and subjected to LC-MS/MS/MS analyses on an LTQ XL linear ion trap mass spectrometer for quantifications of 5-mdC and dG. The amounts of 5-mdC and dG (in moles) in the nucleoside mixtures were calculated based on comparisons of their signal intensities with their corresponding stable isotope-labeled standards and calibration curves. The final levels of 5-mdC were calculated as molar ratios of 5-mdC over dG.
Confocal immunofluorescence (IF). G1/S phase synchronization of ES cells follows a previously established protocol 29 . In essence, ES cells were treated with thymidine (Sigma T9250) at a final concentration of 2 mM for 16 h, washed twice with prewarmed PBS, and then grown in fresh ES cell medium. After a 6 h release, thymidine was added to the medium again. Cells were incubated with thymidine for another 16 h, washed, released for 5 h from the thymidine block with the addition of fresh medium and then proceed for IF. The IF was carried out as described before 29,72 .
Co-immunoprecipitation (CoIP). Cells were lysed as described 29,72 . Anti-Flag M2conjugated agarose beads (Sigma) were incubated with the lysates overnight at 4°C. The beads were then extensively washed and the bound proteins analyzed by western blotting.
Enhanced reduced representation bisulfite sequencing (eRRBS). The eRRBS experiment, including construction of libraries and processing and analysis of eRRBS dataset, was performed as described previously 29 . Briefly, genomic DNA (gDNA) mixed with 0.1% of unmethylated lambda DNA (Promega) was digested with MspI, MseI, and BfaI, and subjected to end repair, A-tailing and ligation to NEBNext Methylated Adapters (NEBNext DNA Library Prep Kit). Next, the DNA product was purified using AMPure beads and subject to bisulfite conversion and library construction using the EpiMark Bisulfite Conversion Kit (NEB cat# E3318), followed by deep sequencing in an Illumina HiSeq 4000 platform with a paired end PE150 cycle (carried out by UNC HTSF Genomic Core). The eRRBS data were then analyzed with FastQC v0.11.2 [http://www.bioinformatics.babraham.ac.uk/ projects/fastqc/] and Bismark v0.18.1 73 . Analysis of the sequences corresponding to phase λ indicated a bisulfite conversion rate of >99%.
DMR detection. Counts of methylated and unmethylated bases for cytosines in the CpG context were analyzed with the DSS (v.2.30.1) R/Bioconductor package 74 for the calling of differential methylation loci or regions (DML/DMRs) and their quantification. The q values under 0.05 and mean differences of methylation greater than 0.15 were used as the cutoff line to define DMRs with DSS. Alternative identification of DML/DMRs was also performed using the R-package methylKit 75 (v.1.8.1) with tiling windows of 500 bp. In the setting of this tool, q values under 0.01 and mean differences of methylation greater than 0.25 were used as the cutoff line. In both analyses, at least five covered cytosine sites were utilized. After the calling of differential methylation, each locus or region was annotated using Homer 76 (v4.10.3).
CUT&RUN followed by deep sequencing. CUT&RUN were performed according to manufacturer's instructions (EpiCypher CUTANA ™ pAG-MNase for ChIC/ CUT&RUN, Cat# 15-1116). In brief, after washing with CUT&RUN wash buffer (20 mM HEPES pH 7.5, 150 mM NaCl, 0.5 mM spermidine, 1× Roche Complete Protease Inhibitor), a million of cells were first bound to activated ConA beads (Bangs Laboratories, cat# BP531), followed by addition of anti-H4K20me3 antibody (Abcam, ab9053; 1:100 dilution) and cell permeabilization with the digitonin buffer (CUT&RUN wash buffer plus 0.01% digitonin). After washing in the digitonin buffer, samples were incubated with pAG-MNase, followed by additional washes with digitonin buffer. After the final wash, pAG-MNase activation was induced for DNA digestion by suspending cell samples in the pAG-MNase digestion buffer (digitonin buffer plus 2 mM CaCl 2 ) and incubation on nutator at 4°C for 2 h. Solubilized chromatin was released using the stop buffer (340 mM NaCl, 20 mM EDTA, 4 mM EGTA, 50 µg/ml RNase A, 50 µg/ml glycogen) and collected using a PCR cleanup kit (New England BioLabs [NEB] Monarch PCR & DNA Cleanup Kit, cat# T1030). Ten nanogram of the purified CUT&RUN-enriched DNA was used for preparation of multiplexed Illumina libraries using the NEB Ultra II DNA Library Prep Kit according to manufacturer's instructions (NEB cat#E7103).
CRISPR/Cas9-based genome editing for site-specifically mutating Dnmt1 BAH1 . As we recently described 4 , the Alt-R ® CRISPR-Cas9 System, together with a singlestranded oligodeoxynucleotide donor (ssODN), was employed to introduce a W799A point mutation to mouse Dnmt1 gene (equivalent to W796A in the human DNMT1 BAH1 domain) in E14 ESCs. In brief, trans-activating CRISPR RNA (tracrRNA) with ATTO ™ 550, the S.p. HiFi Cas9 Nuclease V3, Electroporation Enhancer, and crRNA targeting the Dnmt1 genomic site to be mutated (with the sequence information of crRNA listed in Supplementary Table 2) were all ordered through IDT Inc. The ssODN was custom-designed with a tool of benchling ([https://benchling.com/crispr], the sequence information of ssODN listed in Supplementary Table 2) and synthesized as PAGE-purified Ultramer ® DNA oligonucleotides from IDT Inc. Three phosphorothioate bonds were added at both ends of ssODN to stabilize the donor oligo and make homology-directed repair more efficient. For introducing a Dnmt1 W799A mutation, the codon was mutated from TGG to GCG, which generates a site of MluI enzyme and thus facilitates subsequent cell screening and genotyping by enzyme digestion of PCR products. To make the crRNA:tracrRNA duplex, the crRNA and tracrRNA were mixed in equimolar concentration, followed by heating at 95°C for 5 min and cooling down to room temperature. The CRISPR-Cas9 ribonucleoprotein (RNP) complex was then prepared by diluting the crRNA:tracrRNA duplex and Cas9 enzyme components in PBS, followed by incubation at room temperature for 15 min. The RNP complex was next mixed with Electroporation Enhancer and ssODN, followed by electroporation-based delivery to E14 cells using the Mouse Embryonic Stem Cell Nucleofector Kit (Lonza). ATTO ™ 550 positive cells were sorted out by FACS (at UNC flow core) after 36 h and split into 96-well plates for genotyping. After genotyping, lines with homozygous mutation were further validated at DNA levels by direct sequencing of PCR products (the sequence information of genotyping primers listed in Supplementary Table 2).
Chemical compound. The chemical inhibitor selective for the H4K20 methyltransferases SUV420H1 and SUV420H2, A-196 42 (Sigma, SML1565), was dissolved in dimethylsulfoxide (DMSO) as 5 mM stock solution. A 3-day treatment with 10 µM A-196, compared to DMSO, was used in this study.
Neutral comet assay. To perform neutral comet assays for DNMT1 C1226S mutants ( Supplementary Fig. 7f-h), mouse DNMT1 knockout cells (7 × 10 4 cells) complemented by pPyCAGIP empty vector (vector), WT DNMT1, and two different clones of DNMT1 C1226S mutant (clones #1 and #2) were plated on gelatin-coated culture dishes without feeder cells. After 2 h post ionizing irradiation (5 Gy) cells were harvested and mixed with LMAgarose (Trevigen). The LMAgarose mixed samples were placed onto comet assay slides and immerged into comet assay lysis solution (Trevigen) at 4°C for 1 h. Subsequently, the slides were incubated with TBE buffer (90 mM Tris borate) for 1 h and subjected to electrophoresis at 40 V for 40 min. After electrophoresis, the samples were fixed with 70% ethanol at RT for 30 min and dried at 37°C for 30 min. DNAs were visualized using SYBR-green (Invitrogen) and imaged using Fluoview FV3000 confocal microscope (Olympus). Images were subsequently analyzed using ImageJ (v.1.53). Statistics and graph were calculated using Prism software (Graphpad v6). Experiments were performed with at least two independent replicates. Neutral comet assays for cells with DNMT1 W465A, W465A/W796A, W464A/W465A, or TM (Fig. 5e, f), were performed as described previously 29 .
Clonogenic cell survival assay. Cells were seeded into six-well plates and treated with different dosages of IR using a Faxitron X-ray irradiator. Following IR treatment, cells were incubated for 12 days in tissue culture incubator (37°C, 5% CO 2 ). Cells were washed with PBS and the colonies were stained with 0.5% (w/v) crystal violet and 20% (v/v) ethanol for 30 min at RT. Results were normalized to plating efficiencies of untreated cells for each group.
Statistics. The comet assays were performed using one-way ANOVA with post Tukey analysis. The p value lower than 0.05 was considered to be statistically significant. For all the other analyses, the two-tailed Student t tests were performed to compare distributions between different groups. And the p value lower than 0.01 was considered to be statistically significant.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.