BAH domains and a histone-like motif in DNA methyltransferase 1 (DNMT1) regulate de novo and maintenance methylation in vivo

DNA methyltransferase 1 (DNMT1) is a multidomain protein believed to be involved only in the passive transmission of genomic methylation patterns via maintenance methylation. The mechanisms that regulate DNMT1 activity and targeting are complex and poorly understood. We used embryonic stem (ES) cells to investigate the function of the uncharacterized bromo-adjacent homology (BAH) domains and the glycine–lysine (GK) repeats that join the regulatory and catalytic domains of DNMT1. We removed the BAH domains by means of a CRISPR/Cas9-mediated deletion within the endogenous Dnmt1 locus. The internally deleted protein failed to associate with replication foci during S phase in vivo and lost the ability to mediate maintenance methylation. The data indicate that ablation of the BAH domains causes DNMT1 to be excluded from replication foci even in the presence of the replication focus–targeting sequence (RFTS). The GK repeats resemble the N-terminal tails of histones H2A and H4 and are normally acetylated. Substitution of lysines within the GK repeats with arginines to prevent acetylation did not alter the maintenance activity of DNMT1 but unexpectedly activated de novo methylation of paternal imprinting control regions (ICRs) in mouse ES cells; maternal ICRs remained unmethylated. We propose a model under which DNMT1 deposits paternal imprints in male germ cells in an acetylation-dependent manner. These data reveal that DNMT1 responds to multiple regulatory inputs that control its localization as well as its activity and is not purely a maintenance methyltransferase but can participate in the de novo methylation of a small but essential compartment of the genome.

DNA methyltransferase 1 (DNMT1) is a multidomain protein believed to be involved only in the passive transmission of genomic methylation patterns via maintenance methylation. The mechanisms that regulate DNMT1 activity and targeting are complex and poorly understood. We used embryonic stem (ES) cells to investigate the function of the uncharacterized bromo-adjacent homology (BAH) domains and the glycine-lysine (GK) repeats that join the regulatory and catalytic domains of DNMT1. We removed the BAH domains by means of a CRISPR/ Cas9-mediated deletion within the endogenous Dnmt1 locus. The internally deleted protein failed to associate with replication foci during S phase in vivo and lost the ability to mediate maintenance methylation. The data indicate that ablation of the BAH domains causes DNMT1 to be excluded from replication foci even in the presence of the replication focus-targeting sequence (RFTS). The GK repeats resemble the N-terminal tails of histones H2A and H4 and are normally acetylated. Substitution of lysines within the GK repeats with arginines to prevent acetylation did not alter the maintenance activity of DNMT1 but unexpectedly activated de novo methylation of paternal imprinting control regions (ICRs) in mouse ES cells; maternal ICRs remained unmethylated. We propose a model under which DNMT1 deposits paternal imprints in male germ cells in an acetylation-dependent manner. These data reveal that DNMT1 responds to multiple regulatory inputs that control its localization as well as its activity and is not purely a maintenance methyltransferase but can participate in the de novo methylation of a small but essential compartment of the genome. DNMT1 2 is required for the maintenance of genomic methylation patterns during S phase (Ref. 1); for reviews, see Refs. 2 and 3). In biochemical assays, DNMT1 has a modest (Ͻ10-fold) preference for hemimethylated DNA (4) and has been assigned a function limited to maintenance methylation, although a role for DNMT1 in de novo methylation has not been excluded. Multiple mechanisms increase the fidelity of maintenance methylation; these include autoinhibition of de novo methylation by the CXXC domain (5) and the replication focustargeting sequence (RFTS) (6), which can inhibit de novo methylation by occlusion of the active site of DNMT1 (7). The ubiquitin-like with PHD and Ring finger domains 1 (UHRF1) protein, which is required for faithful maintenance methylation in vivo (8,9), recognizes hemimethylated CpG sites generated during DNA replication via its SRA domain (10 -12). While bound to hemimethylated sites, UHRF1 recognizes nearby histone H3 methylated at lysine 9 with its TTD-PHD domains (13,14) and ubiquitylates H3 at lysines 18 and 23, which is then bound by the RFTS domain of DNMT1 (15,16). Binding of RFTS to ubiquitylated histone H3 brings about a conformational change in DNMT1, which increases its catalytic activity in vitro (17). UHRF1-dependent histone modifications have been proposed to affect maintenance methylation through effects on the interaction of DNMT1 with chromatin (15)(16)(17).
There has been a focus on the mechanism of interaction of UHRF1 and DNMT1, but little attention has been directed to the function of well-conserved domains located within the N-terminal regulatory region of DNMT1 that are not known to interact with UHRF1. Two bromo-adjacent homology (BAH) domains of unknown function are conserved among nearly all DNMT1 homologs. BAH domains are present in ϳ20 mammalian proteins of diverse function. Some BAH domains have been reported to bind to histone tails in a modification-dependent manner (for a review, see Ref. 18). BAH domains fall into two general groups: ORC1-like and SIR3-like. In DNMT1, BAH1 is a member of the ORC1-like group, and BAH2 is a member of the SIR3-like group. Just C-terminal of BAH2, a stretch of alternating lysine and glycine residues (the GK repeats) joins the N-terminal regulatory region and the C-terminal catalytic domain (19) and had been proposed to mediate the degradation of DNMT1 via an acetylation-dependent association with USP7 (20), but recent data indicate that the stability of DNMT1 is independent of the GK repeats and of USP7 (21).

cro ARTICLE
We have addressed the function of the BAH domains and the GK repeats. We found that the BAH domains are required for maintenance methylation by mediating the association of DNMT1 with replication foci during S phase. Substitution of the GK repeats with GR repeats did not affect DNMT1 protein stability and did not affect maintenance methylation in vivo. It was very surprising to find that the GR substitution caused de novo methylation specifically of imprinting control regions (ICRs) in mouse embryonic stem (ES) cells that are normally methylated in male germ cells; maternal ICRs remained unmethylated. These and other data suggest that DNMT1 is likely to be involved in the de novo methylation of paternal ICRs in male germ cells.

The BAH domains are necessary for targeting DNMT1 to replication foci during S phase
As shown in Fig. 1A, DNMT1 contains multiple functional domains. In the crystal structure of DNMT1 bound to unmethylated DNA (5) (Fig. 1B), the BAH domains are remote from the active site of the catalytic domain, but an acidic autoinhibitory linker extends from BAH1 to the CXXC domain, which binds to unmethylated CpG dinucleotides and prevents their entry into the active site of the catalytic domain (5). BAH2 extends a long solvent-exposed loop (the BAH2-TRD loop) that interacts with the target recognition domain (TRD) within the catalytic domain. The function of this interaction is unknown.
Both the BAH1 autoinhibitory linker and the BAH2-TRD loops bind to the BAH-binding pockets (Fig. 1C) that in other proteins interact with histone tails in a manner that depends on the modification status of lysine residues (for a review, see Ref. 18). In the case of DNMT1, BAH1 binds to a sequence centered on Leu-721, whereas BAH2 binds to Phe-952; lysines are not involved in either case. However, interaction with neighboring lysines cannot be excluded as the BAH1-CXXC linker shifts positions between different conformations of DNMT1 (5,22). The BAH1-CXXC linker is highly conserved among vertebrates (Fig. 1D).
To address the functions of the BAH domains, the endogenous Dnmt1 locus was modified by Cas9-mediated deletion of exons 11-16 ( Fig. 2A), which encode only the BAH domains and leave the remainder of the protein intact. The joining of exons 10 and 17 produces an in-frame splicing event. As shown in Fig. 2B, the DNMT1 protein lacking the BAH domains is stable and produced at levels very similar to that of the fulllength protein. Internally deleted protein does not aggregate in the cytoplasm and localizes to the nucleus as does full-length DNMT1 (Fig. 2C). Two ES cell clones that had undergone homozygous deletions were subjected to further study.
DNMT1 is present at constant levels in populations of dividing cells but is recruited to replication foci at S phase (6). As shown in Fig. 2C, deletion of the BAH domains caused a failure of DNMT1 to be incorporated into replication foci with Ͼ90% of S-phase cells displaying a diffuse nucleoplasmic localization

In vivo regulation of DNMT1
in independent clones. This was unexpected as the RFTS is intact in the internally deleted protein, and recruitment of DNMT1 to replication foci had been reported to involve binding of the RFTS domain to histone H3 ubiquitylated by UHRF1 (16,17). In addition, the isolated RFTS domain is capable of localizing to replication foci (6). The diffuse nucleoplasmic localization phenotype observed in cells with the BAH deletion is similar to that of Uhrf1-null ES cells (23). These data indicate that BAH domains are required for targeting of DNMT1 to replication foci during S phase even in the presence of the RFTS and UHRF1.
As shown in Fig. 2D, the BAH-deleted form of DNMT1 was unable to maintain DNA methylation genome-wide or at IAP retrotransposons and major satellite DNA, sequences that are normally densely methylated. Minor satellite DNA was partially methylated (Fig. 2D), which confirms that BAH-deleted DNMT1 has enzymatic activity in vivo. The inability of BAHdeleted DNMT1 to perform maintenance methylation is consistent with the failure of the protein to be incorporated into replication foci.

GK repeats are dispensable for maintenance DNA methylation
All known DNMT1 homologs contain at least one GK dipeptide between the N-terminal regulatory region and the C-terminal catalytic domain (Fig. 3A) (21). DIM-2, a DNMT1 homolog from the ascomycete Neurospora crassa, has a single GK dipeptide at this position (24). DNMT1 homologs of eutherian mammals contain 13 alternating Gly and Lys residues. The GK repeats are not resolved in any of the extant DNMT1 crystal structures and are likely to be very mobile; they are rendered in Fig. 1B to show the approximate spatial relationship of the GK repeats to the other domains of DNMT1. The GK repeats resemble the N-terminal tails of histones H2A and H4, which are also highly flexible. It had been previously shown that the GK repeats were not required for the stabilization of DNMT1 protein (21). The GK repeats are subject to acetylation at multiple lysine residues (25). We tested the function of the GK repeats by replacement of each of the lysines with arginine, which maintains the basic nature of the sequence but prevents acetylation.
Full-length DNMT1 bearing the GK 3 GR substitutions (Fig. 3B) was expressed in transfected Dnmt1 Ϫ/Ϫ ES cells (26) under the control of the Dnmt1 promoter, which causes DNMT1 to be expressed at close to the levels of endogenous DNMT1 (27). As shown in Fig. 3C, ES cell clones were obtained that expressed GR DNMT1 at levels very close to that of endogenous DNMT1. Levels of DNA methylation in cells expressing GR DNMT1 were compared with those of WT cells and Dnmt1 Ϫ/Ϫ cells expressing recombinant GK DNMT1 of WT sequence. In contrast to the BAH deletion mutants, the GR mutants were capable of rescuing and maintaining DNA methylation genome-wide as assessed by luminometric methylation assay (LUMA) (Fig. 3E) and at IAP retrotransposons and minor satellite DNA (Fig. 3D). These data indicate that the GK repeats are not essential for the genome-wide maintenance of DNA methylation. Flanking exons undergo an in-frame splicing event; internally deleted protein product is shown below. B, immunoblot analysis shows that BAH domain deletions are stable and expressed at close to WT levels. Two independent clones are shown. C, failure of BAH deletion proteins to associate with replication foci. The top row shows that WT DNMT1 colocalizes with PCNA, a marker of replication foci, whereas the bottom two rows show that BAH-deleted DNMT1 fails to associate with replication foci. D, BAH-deleted DNMT1 fails to maintain genomic methylation patterns. In all cases, resistance to methylation-sensitive restriction enzymes (HpaII and MaeII) is equivalent between Dnmt1 Ϫ/Ϫ ES cells and DNMT1 BAH-deleted ES cells. aa, amino acids; LTR, long terminal repeat.

In vivo regulation of DNMT1 De novo methylation of paternal ICRs in the presence of GR DNMT1 but not GK DNMT1
As was shown in Fig. 3D, both GR DNMT1 and GK DNMT1 were capable of restoring genome-wide DNA methylation when expressed in Dnmt1 Ϫ/Ϫ ES cells. Dnmt1 Ϫ/Ϫ ES cells expressing either GR DNMT1 or GK DNMT1 were examined by bisulfite genomic sequencing for methylation at ICRs where allele-specific methylation confers allele-specific expression to imprinted genes (28 -30). We analyzed methylation of all known paternally methylated ICRs, H19, Dlk1/Gtl2, and Ras-grf1, and the following maternal ICRs: Igf2, Snrpn, Peg3, Kcnq1ot1, and Peg1. It had previously been reported that Dnmt1 rescue resulted in global genome remethylation, but allele-specific methylation at imprinted genes was not reestablished without passage through the germ line (31,32). In accordance with these data, Dnmt1 Ϫ/Ϫ ES cells that expressed GK DNMT1 showed full rescue of genome-wide DNA methylation but did not remethylate maternal or paternal ICRs. Sur-

In vivo regulation of DNMT1
prisingly, in the presence of GR DNMT1, all paternally methylated ICRs (H19, Dlk1/Gtl2, and Rasgrf1) underwent de novo methylation, whereas maternal DMRs (Igf2, Snrpn, Peg3, Kcnq1ot1, and Peg1) remained unmethylated (Fig. 4). Paternal ICRs exhibited a statistically significant gain of methylation in GR as compared with GK cells (Fig. 4C). We performed bisulfite sequencing of ICRs in biological replicates, demonstrating reproducibility of de novo methylation at paternal ICRs. These data indicate that lysine-to-arginine substitution that prevents acetylation of the GK repeats activates de novo methylation specifically at paternal ICRs.

Discussion
The preference of DNMT1 for hemimethylated DNA could have been predicted to result simply from the selective recognition and methylation of hemimethylated DNA. However, there is no evidence that this occurs, and DNMT1 does not possess a domain that is able to recognize and bind hemimethylated DNA. DNMT1 has a CXXC domain that binds unmethylated DNA and is involved in autoinhibition (5). In our study, we discovered that the BAH domains and the GK repeats are involved in two distinct targeting mechanisms, one to maintain DNA methylation in the case of BAH domains and one to carry out a novel function, de novo methylation of paternal imprints, in the case of the GK repeats.
The function of the two BAH domains of DNMT1 had not been previously addressed. We used a novel Cas9-mediated approach to delete the exons that encode the BAH domains from the endogenous Dnmt1 locus. The internally deleted protein was stable and expressed at the same level as the full-length protein. However, it was unable to associate with replication foci during S phase and displayed a diffuse nucleoplasmic localization. The same localization phenotype has been observed in Uhrf1-null ES cells (23). Previous data had shown that the RFTS alone is capable of targeting heterologous proteins to replication foci (6). Based on these data, we conclude that in the absence of the BAH domains DNMT1 is actively excluded from replication foci by a mechanism that regulates conformational change within the molecule necessary for RFTS-and UHRF1dependent targeting. We propose that this conformational change could be brought about by the autoinhibitory (BAH1-CXXC) linker, which forms hydrogen bonds with RFTS (22) and BAH1.
The function of the GK repeats was also unknown; a role in an acetylation-dependent interaction with USP7 that protected DNMT1 from proteasomal degradation (20) is unlikely to occur as the stability of DNMT1 is independent both of USP7 and of acetylation of the GK repeats (21). We addressed the function of the GK repeats by in vivo expression of a form of DNMT1 in which the GK repeats had been replaced by GR repeats, which maintains the basic nature of the sequence but prevents acetylation. GR DNMT1 was capable of restoring genome-wide methylation to the same extent as WT GK DNMT1, but GR DNMT1 also induced the de novo methylation of paternally methylated ICRs as observed in male germ cells. This surprising finding is supported by additional lines of evidence. First, the establishment of paternal genomic imprints is largely independent of DNMT3L (33) and of DNMT3A (34), whereas the establishment of maternal genomic imprints is completely dependent on DNMT3L (29) and DNMT3A (34). Second, although DNMT1 is not normally present in nondividing cell types, it is present in nondividing prospermatogonia (35) at the time when paternal genomic imprints are established (36). These data indicate that DNMT1 may carry out the de novo methylation of paternal ICRs and that regulated deacetylation of the GK repeats may target DNMT1 to these sequences in prospermatogonia. As has been noted previously, there is profound sexual dimorphism in genomic imprinting in mammals (37). Paternal ICRs are characterized by sequence characteristics distinctly different from those of maternal ICRs, and maternal ICRs are usually located at promoter regions, whereas paternal ICRs are found in intergenic regions and are not associated with promoters or known enhancers (38).
A role for DNMT1 in de novo methylation has never been excluded, although the enzyme has commonly been attributed a role strictly limited to maintenance methylation. This is largely due to the influence of predictions of the existence of an enzyme having this activity made in 1975 by Holliday and Pugh (39) and by Riggs (40). In fact, DNMT1 has a much higher specific activity on unmethylated DNA than either DNMT3A or DNMT3B in biochemical assays (41). Furthermore, DNMT1 is the only DNA methyltransferase homolog present in several insect orders that have methylated genomes (42). The methylated genome of the lepidopteran Bombyx mori encodes only a DNMT1 protein that is very similar to that of mammals and that according to the Conserved Domain Database (43) contains no conserved domains not also present in mammalian DNMT1 (Fig. 5). This suggests that DNMT1 must be involved in both de novo and maintenance methylation in Bombyx. We speculate that in mammalian spermatogonia one or more factors selectively recognize and bind to unmethylated paternal ICRs; these factors recruit and deacetylate the GK repeats of DNMT1 to activate the de novo activity of DNMT1 specifically at paternal ICRs.
These results show that DNMT1 is not a simple enzyme that recognizes and methylates hemimethylated CpG dinucleotides immediately after passage of the replication fork. It is instead under multiple regulatory inputs and may play a much larger role in the dynamics of genomic methylation patterns than may have been formerly believed.

Cell culture
Mouse ES cells were cultured on gelatin at 37°C with 5% CO 2 in Dulbecco's modified Eagle's medium supplemented with 15% fetal bovine serum, 2 mM L-glutamine, minimum essential medium nonessential amino acids, 100 IU/ml penicillin, 100 g/ml streptomycin, 0.12 mM 2-mercaptoethanol, and leukemia inhibitory factor from the conditioned medium of leukemia inhibitory factor-secreting cells.

Generation of rescue ES cell lines
Stable ES cell lines were generated by nucleofection of 2 ϫ 10 6 Dnmt1-null ES cells (26) with 19 g of MT80 minigene (27) and 1 g of PGKPuro plasmid (for puromycin resistance) using

In vivo regulation of DNMT1
Amaxa P3 Primary Cell 4D-Nucleofector X Kit L. Nucleofected cells were seeded onto mouse embryonic fibroblast extracellular matrix in 15-cm dishes. MT80 minigene, carrying 12 kb of 5Ј Dnmt1 genomic sequence with endogenous promoter and 5.5 kb of Dnmt1 cDNA (27), was modified by the addition of an N-terminal FLAG-HA tag after the translation start site. Mutations were introduced using the QuikChange Site-Directed Mutagenesis kit (Agilent). Individual clones were selected with puromycin (2 g/ml) for 10 -14 days and picked into 96-well plates. Clones were genotyped using primers specific to the FLAG-HA tag and Dnmt1 (forward primer, GGACTA-CAAGGACGACG; reverse primer, GCTGACCAAAGAGG-GGAACC). Positive clones were propagated, and levels of DNMT1 expression were tested by immunoblotting. Clones expressing DNMT1 at WT levels were selected for further studies.

Generation of CRISPR/Cas9 ES cell lines
To delete BAH domains, Broad Institute software was used to design gRNAs targeting introns upstream of BAH1 and downstream of BAH2 of endogenous Dnmt1 locus (Table 1). gRNAs were cloned into pX330-PURO plasmid. 5-10 g of plasmid DNA was nucleofected into 2 million WT CCE mouse ES cells (derived from 129/Sv mouse strain) (44,45) using Amaxa P3 Primary Cell 4D-Nucleofector X Kit L. Nucleofected cells were seeded onto mouse embryonic fibroblast extracellular matrix in 15-cm dishes. Transient selection with puromycin (1 g/ml) was applied 48 h postnucleofection for a period of 24 h. Surviving clones were picked 10 -14 days after nucleofection. Remaining clones were pulled for genomic DNA and tested with T7 endonuclease to check the percentage of cuts in the population. Individual clones were genotyped using primers flanking the deletion site (forward primer, CAGGTAGCC-CATCCGCTTG; reverse primer, AATTCCTAGCACCCAC-ACGG; expected product size, 333 bp). Homozygous and heterozygous clones were distinguished by using one primer outside the deleted site and one primer inside (forward primer, CAGGTAGCCCATCCGCTTG; reverse primer, CAGTAA-GATCCCATCTCCAAACCA; expected product size, 291 bp for WT allele and no product for deleted allele).

Methylation analysis
Genomic DNA was extracted by digestion with proteinase K and RNase A followed by phenol/chloroform extraction and isopropanol precipitation. Genomic DNA was digested for two rounds with methylation-sensitive enzyme HpaII, its isoschizomer MspI as a control, or MaeII (for major satellite analysis). All enzymes were from New England Biolabs. DNA was quantified, fractionated on 0.8% agarose gels, and stained with ethidium bromide.
Global levels of DNA methylation were quantified using LUMA as described previously (46). 400 ng of genomic DNA was digested with MspI/EcoRI and HpaII/EcoRI in parallel. The overhangs from the enzymatic digestion were quantified by pyrosequencing (PyroMark Q24, Qiagen, Hilden, Germany) with the following dispensation order: GTCACAGTGT. Global DNA methylation levels were calculated from the peak heights at positions 3, 4, 7, and 8 using the following formula: Global methylation (%) ϭ (1 Ϫ (HpaII ͚G/͚T)/(MspI ͚G/͚T)) ϫ 100. Statistical analysis was performed on biological replicates using the two-tailed t test.

Bisulfite sequencing
Genomic DNA isolated from ES cells was bisulfite-converted using the EZ DNA Methylation-Gold kit. ICRs of three paternal and five maternal imprinted genes were amplified with two  (42) but does not contain any annotated domains not also present in mammalian DNMT1 as assessed by CDD/SPARCLE (43). This indicates that DNMT1 is very likely to be involved in both maintenance and de novo methylation in Bombyx and suggests that the same is likely to be true of mammalian DNMT1.

In vivo regulation of DNMT1
rounds of nested PCR using primers listed in Table 2. The PCR product resulting from the second round of PCR was sent out for Sanger sequencing (Genewiz). Peak heights were quantified using 4Peaks software.

Statistical analysis
Statistical data were calculated for groups with normal distributions and similar variances. Variations within each group of data are reported as standard deviation. n represents the number of biological replicates. p values for LUMA were calculated using the two-tailed t test. The R program was used to ensure that n ϭ 2 was sufficient to establish a power of 0.8. p values for bisulfite sequencing were calculated using the Mann-Whitney test.