RAD51 plays critical roles in DNMT1-mediated maintenance methylation of genomic DNA by dually regulating the ubiquitin ligase UHRF1 Authors: Guangxue Liu, Kaiyan Huang, Shiyao Liu, Yali Xie, Jinyan Huang, Tingbo Liang, Pumin Zhang Published: 2024 ABSTRACT SignificanceRad51, known for its evolutionarily conserved function in homologous recombination, has been found to play unexpected and crucial roles in the maintenance methylation of genomic DNA: inhibiting the ubiquitination and degradation of DNMT1 and promoting the ubiquitination of the histone H3 tail. Interestingly, these two unique roles of Rad51 are separable from its role in recombination and do not require additional domains or motifs to be invented in RAD51. In fact, the residues involved in DNA methylation, D231 and R254, are present in budding yeast RAD51, and the yeast RAD51 has the ability to replace its human homolog in human cells to carry out the aforementioned functions. This serves as another notable illustration of gene multifunctionalization in evolution. Methylation of genomic DNA is a major epigenetic mechanism that regulates chromatin architecture and gene expression. The overall genomic methylation profile is established in early embryonic development via de novo methylation, carried out by DNA methyltransferases DNMT3A and 3B (1). This global methylation profile is maintained, more or less, in all cells thereafter through maintenance methylation primarily by DNA methyltransferase DNMT1 (2, 3). DNMT1 cannot function alone. It requires a cofactor, UHRF1 (ubiquitin-like with PHD and ring finger domains 1), as its guide to find the CpG sites that need to be methylated after DNA replication (4, 5). During the maintenance methylation of newly synthesized DNA, UHRF1 modifies the histone H3 tail by adding monoubiquitin to K18 and K23 residues with an E3 ubiquitin ligase activity contained in its RING domain (6) as well as to K15 and K24 in PAF15 (7). The so-generated ubiquitin-modified H3 serves as the major signal for DNMT1 recruitment to hemimethylated sites for methylating the daughter strand DNA. The RAD51 recombinase is functionally related to the bacterial RecA protein and is highly conserved through evolution as evidenced by the fact that the budding yeast RAD51 is about 50% identical and over 60% similar in amino acid sequence to its human counterpart. RAD51 plays a critical role in the homologous recombination-mediated repair of DNA double strand breaks (8). Given its primary function in HR and genome stability, it is intriguing that RAD51 is a gene mutated in congenital mirror movements (CMM) in humans (9, 10). In CMM, there is usually a nonsense mutation in one allele of RAD51, and the inheritance is autosomal dominant (9). DCC (deleted in colorectal cancer) and DNAL4 (dynein light chain 4) are the other two genes when mutated monoallelically also cause CMM. Both DCC and DNAL4 are known to be involved in neurological development (10). Although it is not clear why RAD51 is haploinsufficient in CMM, the human disease genetics does suggest that it most likely plays roles outside DNA damage repair. Here, we report a completely unique and unexpected function of RAD51 in the maintenance methylation of genomic DNA, a function that is separable from its role in homologous recombination. First, RAD51 acts as an inhibitor of the E3 ubiquitin ligase UHRF1. Deficiency in RAD51 causes excessive ubiquitination and subsequent proteasomal degradation of DNMT1, leading to the loss of epigenetic methylation of genomic DNA. Second, RAD51 is a histone reader. It binds to histone H3 at residue R17, which promotes UHRF1 binding to and ubiquitinating H3 for the generation of DNMT1 recruitment signal. Disrupting the interaction between RAD51 and H3 diminishes DNMT1 recruitment and the failure of maintenance methylation of genomic DNA. These results shed fresh lights into the inner workings of the DNA maintenance methylation machinery and establish RAD51 as a guardian of the integrity of both the genome and the epigenome. We further showed that RAD51 suppresses neurite outgrowth when the PC12 cells were induced to undergo neuronal differentiation and this suppression relies on RAD51’s function in the maintenance methylation of genomic DNA. These latter results provide a plausible explanation for the involvement of RAD51 in CMM. Results RAD51 Interacts with UHRF1. To better understand how RAD51 functions, we performed an IP-mass analysis of RAD51 interacting proteins in unperturbed 293T cells. Not surprisingly, there were many DNA repair-related proteins found in the immunoprecipitates including UHRF1 (ubiquitin-like with PHD and RING finger domain 1) (SI Appendix, Fig. S1A). UHRF1 is best known for its role in maintenance methylation of the genome, but is also involved in DNA damage repair (11–13). To confirm this interaction, we expressed tagged UHRF1 and RAD51 in 293T cells and immunoprecipitated UHRF1 or RAD51. The two proteins could bring down each other (SI Appendix, Fig. S1B). Their interaction could also be detected with the endogenous proteins (SI Appendix, Fig. S1C) and with purified proteins in vitro (SI Appendix, Fig. S1D). We next wanted to determine the site in UHRF1 that is necessary for the interaction with RAD51. We first deleted the RING domain (SI Appendix, Fig. S1E) and found the ΔRING UHRF1 was no longer able to interact with RAD51, suggesting that the interaction site is contained within the RING domain (SI Appendix, Fig. S1F). With further deletions targeting the RING domain, we narrowed the RAD51-interacting site down to between residues 741 and 751 (SI Appendix, Fig. S1G) and finally down to between residues 745 and 750. Mutating 5 out of the 6 residues (UHRF15A, SI Appendix, Fig. S1E) destroyed the interaction between UHRF1 and RAD51 in vitro (SI Appendix, Fig. S1H) and in vivo (SI Appendix, Fig. S1I). Reciprocally, we sought to map the site in RAD51 that mediates the interaction with UHRF1. Serial deletions (SI Appendix, Fig. S1 J and K) indicated that the site is located between residues 251 and 257. Further point mutations identified R254 as a critical residue for the interaction (SI Appendix, Fig. S1J). It lies outside the Walker B motif and is not conserved in RAD51 paralogs. Changing R254 to Gln (R254Q) nearly abolished the interaction with UHRF1 (SI Appendix, Fig. S1 L and M). RAD51 Regulates DNMT1 Expression Via UHRF1. To begin to explore the biological significance of the interaction between UHRF1 and RAD51, we first looked at the possibility that UHRF1, as an E3 ubiquitin ligase, regulates the stability of RAD51. However, the depletion of UHRF1 expression did not result in an increase in the expression level of RAD51, but instead seemed to decrease it (Fig. 1A and SI Appendix, Fig. S2A), the reason behind and the significance of which will be described elsewhere. The expression of DNMT1, a known substrate of UHRF1 E3 ligase, was not affected much as reported previously (4). Depleting RAD51 expression had no effect on UHRF1’s abundance either, but unexpectedly, reduced that of DNMT1 to a large extent (Fig. 1B and SI Appendix, Fig. S2B). The reduction in DNMT1 expression was not seen at the transcriptional level (SI Appendix, Fig. S2C) and could be prevented by the proteasome inhibitor treatment (Fig. 1C), indicating that the methyltransferase was degraded through the proteasome when RAD51 was compromised. RAD51 regulates DNMT1 expression by suppressing UHRF1-mediated ubiquitination of the demethylase. (A) Western blotting analysis of RAD51 and DNMT1 protein levels following UHRF1 knockdown in HEK293T cells. (B) Western blotting analysis of UHRF1 and DNMT1 protein levels following RAD51 knockdown in HEK293T cells. (C) Western blotting analysis of DNMT1 protein level following RAD51 knockdown in HEK293T cells. The cells were transfected with scrambled or RAD51 siRNAs for 42 h, followed by 6 h treatment with 10 μM MG132. (D) Western blotting analysis of DNMT1 protein level after knocking down RAD51 and UHRF1, individually or simultaneously in HEK293T cells. (E) Western blotting analysis of DNMT1 protein levels following UHRF1 knockdown and transient reexpression of shUHRF1-resistant FLGA-tagged UHRF1 or UHRF15A in HCT116 cells. (F) Western blotting analysis of DNMT1 protein levels following RAD51 knockdown and transient reexpression of shRAD51-resistant HA-RAD51 or HA-RAD51R254Q in tet-on shRAD51-containing HCT116 cells. The cells were treated with doxycycline (1 μg/mL) for 48 h to induce shRAD51 expression. (G) Ubiquitination assay of DNMT1 in HEK293T cells transfected with siRAD51 (2 independent siRNAs). (H) Ubiquitination assay of DNMT1 in HEK293T cells with UHRF1 depleted and reexpression of UHRF1 or UHRF15A. (I) Ubiquitination assay of DNMT1 in HEK293T cells transfected with siRAD51, HA-RAD51-expressing plasmid, or HA-RAD51R254Q-expressing plasmid. (J and K) RAD51 inhibits DNMT1 ubiquitination in vitro. The in vitro assay of DNMT1 ubiquitination was performed using recombinant E1 (UBA1), E2 (UBC13), E3 (His-UHRF1 or His-UHRF15A), and His-DNMT1. Same amounts of His-RAD51 and His-RAD51R254Q were added to the reaction. The reduction of DNMT1 expression was not an off-target effect of siRAD51, as the reexpression of RNAi-resistant RAD51 could bring back DNMT1 levels in both 293 T (SI Appendix, Fig. S2D) and U2OS cells (SI Appendix, Fig. S2E). More importantly, simultaneous depletion of both RAD51 and UHRF1 resulted in an apparent restoration of DNMT1 expression levels in 293T cells (Fig. 1D) as well as in U2OS cells (SI Appendix, Fig. S2F), demonstrating that the degradation of DNMT1 is UHRF1-dependent. This point was further corroborated by the finding that in UHRF1-depleted cells, the reexpression of UHRF15A, a mutant identified above which cannot bind RAD51 anymore, caused downregulation of DNMT1 expression in a way similar to that seen with RAD51-depletion (Fig. 1E) and that in RAD51-depleted cells, the reexpression of RAD51R254Q, another mutant identified above which cannot bind to UHRF1, could not bring back DNMT1 levels as the wild-type RAD51 could (Fig. 1F). The most likely explanation for the downregulation of DNMT1 expression in RAD51-depleted cells is that RAD51 impedes UHRF1-mediated ubiquitination and subsequent proteasomal degradation of DNMT1. Indeed, the depletion of RAD51 led to large increases in the ubiquitination levels of DNMT1 (Fig. 1G). The ubiquitination is UHRF1-dependent since depletion of the E3 ligase almost abolished it (Fig. 1H). While the reexpression of UHRF1WT restored the ubiquitination of DNMT1 in the UHRF1-depleted cells, the reexpression of UHRF15A enhanced it, in agreement with the fact that this mutant is incapable of interacting with (and hence could not be interfered by) RAD51 (Fig. 1H). Moreover, in contrast to its depletion, the overexpression of RAD51 suppressed the ubiquitination of DNMT1 (Fig. 1I), whereas the overexpression of RAD51R254Q was ineffective in that regard (Fig. 1I). We next sought to reconstitute the ubiquitination of DNMT1 in vitro. To do that, we first determined that UBC13 was the E2 used by UHRF1 (SI Appendix, Fig. S2G), and then expressed and purified RAD51, DNMT1, and UHRF1 (SI Appendix, Fig. S2H). DNMT1 could indeed be ubiquitinated by UHRF1 in vitro (SI Appendix, Fig. S2I). There was no UHRF1 self-ubiquitination observed under the in vitro reaction conditions (SI Appendix, Fig. S2I). However, UHRF1 could self-ubiquitinate if UBE2D1 was supplied as an E2 and such self-ubiquitination was also preventable by RAD51 (SI Appendix, Fig. S2J). With the in vitro ubiquitination assay, we found that while both UHRF1WT and UHRF15A showed similar activities toward DNMT1, only the former could be inhibited by RAD51 (Fig. 1J). Further, RAD51WT could block the E3 activity of UHRF1, whereas RAD51R254Q could not (Fig. 1K). These in vivo and in vitro results firmly established that RAD51 is an inhibitor of UHRF1 E3 ligase. RAD51 Binding Prevents UHRF1 from Interacting with DNMT1. We next asked how RAD51 blocks UHRF1’s action against DNMT1. To that end, we first looked at the interaction between UHRF1 and DNMT1. In an immunoprecipitation assay, the depletion of RAD51 expression resulted in a significant increase in the association between DNMT1 and UHRF1 despite the considerably lower levels of DNMT1 observed in the depleted cells compared to the control cells (SI Appendix, Fig. S3A). The results were very similar if the immunoprecipitation was performed with the cell lysates treated with benzonase first to digest away chromatin DNA (SI Appendix, Fig. S3A). On the other hand, the overexpression of RAD51 suppressed the association of DNMT1 with UHRF1 (SI Appendix, Fig. S3B). The suppression could be recapitulated with purified RAD51, DNMT1, and UHRF1 (SI Appendix, Fig. S3C). Furthermore, UHRF15A could bring down more DNMT1 than wild-type UHRF1 could (SI Appendix, Fig. S3D), and its interaction with DNMT1 could not be suppressed by RAD51 (SI Appendix, Fig. S3B). Of note, there was no interaction between RAD51 and DNMT1 could be detected (SI Appendix, Fig. S3E). These data strongly support the notion that RAD51-bound UHRF1 is less likely to bind DNMT1, and the three of them do not form a complex. It was reported previously that the UBL domain of UHRF1 contacts the RING domain, resulting in more stable association of the RING with an E2 enzyme and therefore stimulating RING’s E3 activity (14, 15), The UHRF1 structure simulated by AlphaFold2 also shows that the UBL contacts the RING (SI Appendix, Fig. S3F). We therefore decided to determine whether RAD51 binding affects UBL–RING interaction in UHRF1. We expressed and purified His-tagged UBL and GST-RING and performed GST pulldown assays. As shown in SI Appendix, Fig. S3 G and H, it is clear that RAD51 competes with UBL to bind to RING, which was further corroborated by the immunoprecipitation experiment demonstrating that UBL-less UHRF1 (UHRF1Δ1-147) could bring down much more RAD51 than UHRF1WT could (SI Appendix, Fig. S3I). Since UBL–RING interaction is critical for UHRF1 to interact with an E2 (14), we reasoned that RAD51 might interfere with that. Indeed, as shown in SI Appendix, Fig. S3J, RAD51 could suppress the association between UHRF1 and UBC13. Furthermore, UHRF15A is much more efficient in interacting with UBC13 than UHRF1WT (SI Appendix, Fig. S3K). Taken together, we believe that RAD51 blocks UHRF1-mediated ubiquitination of DNMT1 by altering UHRF1’s conformation in a way such that it is now much less capable of binding DNMT1 and UBC13. RAD51 Is Required for the Maintenance Methylation of Genomic DNA. Given the essential role played by DNMT1 in the maintenance methylation of genomic DNA, we asked whether the downregulation of DNMT1 expression caused by RAD51 depletion would bear any consequences in that regard. To that end, we depleted the expression of RAD51, DNMT1, and UHRF1 with siRNA in HeLa cells (SI Appendix, Fig. S4A). DNMT1 expression plasmid was also included in one siRAD51 transfection to off-set the downregulation of DNMT1 (SI Appendix, Fig. S4A). The cells were collected 48 h after the transfection to assess the methylation levels in genomic DNA. First, we performed immunofluorescence staining of 5mC. As shown in Fig. 2A, the depletion of RAD51 resulted in a significant decrease in the intensity of 5mC staining, comparable to the decrease observed with UHRF1 or DNMT1 depletion (though less pronounced than the latter two). We next performed dot blot analysis of 5mC levels in these cells (SI Appendix, Fig. S4B). Again, the depletion of RAD51, UHRF1, or DNMT1 all resulted in a large reduction in 5mC levels. Surprisingly, however, exogenous expression of DNMT1 could not restore m5C levels in RAD51-depleted cells (Fig. 4A and SI Appendix, Fig. S4B), despite the restoration of DNMT1 protein levels (SI Appendix, Fig. S4A), strongly suggesting that RAD51 contributes to the maintenance DNA methylation more than just suppressing DNMT1 ubiquitination and degradation. RAD51 is required for the maintenance methylation of genomic DNA. (A) Immunofluorescent staining of 5 mC in HeLa cells. The average fluorescence strength of 500 cells from each sample was normalized and compared to that of control cells. Data are presented as mean ± SD. **** indicates P < 0.0001. (B) Whole-genome bisulfite sequencing (WGBS) analysis of global genomic DNA methylation in RAD51- or DNMT1-depleted HCT116 cells. The heatmap shows clustering of the 5,799 differentially methylated regions (DMRs, represented by horizontal lines). The color of the horizontal lines indicates methylation levels (1 represents 100%). (C) Boxplots of the distribution of the methylation levels of all DMRs. To definitively demonstrate the effect of RAD51 on genomic DNA methylation, we subjected the genomic DNA from RAD51- or DNMT1-depleted cells to whole-genome bisulfite sequencing (WGBS). A huge number of differentially methylated regions (DMRs, 1 kb in length) could be detected by WGBS in RAD51 or DNMT1 knockdown cells (Fig. 2B). Both of the depletion caused significant decreases in the overall methylation levels (Fig. 2C). We also took a look at the methylation levels in 3 regions of the genome: CpG islands, gene promoters, and gene bodies. In all three regions, the methylation levels were reduced in DNMT1 as well as in RAD51-depleted cells, although it seemed that depleting RAD51 caused more reduction of methylation in promoters (SI Appendix, Fig. S4C). The loss of genomic DNA methylation often leads to the expression of genes which are otherwise suppressed due to epigenetic controls. We examined a few such genes including HDAC9, RORA, FGFR2, and ITGA8. Indeed, the depletion of RAD51 caused them to be derepressed (SI Appendix, Fig. S4D). Taken together, these results clearly show that the recombinase RAD51 plays a critical role in genomic DNA methylation. RAD51 Promotes UHRF1-Mediated Histone H3 Ubiquitination. One of the key functions played by UHRF1 in the maintenance methylation of genomic DNA is to ubiquitinate histone H3’s tail to generate DNMT1 recruiting signals (6). We therefore were keen to know whether the inhibition of UHRF1’s E3 activity against DNMT1 by RAD51 also applies to the H3 ubiquitination. We measured H3 ubiquitination levels in RAD51- as well as in UHRF1-depleted cells. As expected, depleting UHRF1 led to significant decreases in H3 ubiquitination levels (Fig. 3A). Surprisingly, however, depleting RAD51 caused similar decreases instead of the expected increases (if RAD51 also inhibits UHRF1’s E3 activity toward H3) (Fig. 3A), suggesting that RAD51 might be required for UHRF1-mediated H3 ubiquitination instead. To that end, we performed an in vitro H3 ubiquitination assay. Indeed, bacteria-expressed and purified RAD51WT could stimulate UHRF1 to ubiquitinate H3 (Fig. 3B). Strikingly, RAD51R254Q could do so too (Fig. 3B), indicating that RAD51 promotes UHRF1-mediated H3 ubiquitination independent of its interaction with UHRF1. Apparently, RAD51 performs two exact opposite functions in regulating UHRF1-mediated ubiquitination: inhibiting DNMT1 ubiquitination while promoting H3’s, and the latter requires no interaction with UHRF1. RAD51 promotes UHRF1-mediated histone H3 ubiquitination. (A) Western blotting analysis of H3 ubiquitination levels. Histones were prepared with acid extraction. (B) In vitro H3 ubiquitination assay. (C) Immunoprecipitation analysis of the interaction between the ectopically expressed FLAG-H3 and HA-RAD51 or HA-RAD51D231A. H3 and UHRF1 binding sites in RAD51 are schematically illustrated. (D) Western blotting analysis of H3 ubiquitination levels in HCT116 cells containing the tet-on shRAD51-expressing cassettes transfected with HA-RAD51 (WT or mutants) plasmids. After the transfection, doxycycline (1 μg/mL) was added into the culture media for 48 h to induce shRAD51 expression. Histones were prepared by acid extraction. (E) In vitro H3 ubiquitination assay in the presence of RAD51 (WT and mutants). (F) Immunoprecipitation analysis of the interaction between HA-UHRF1 and H3-FLAG in U2OS cells with doxycycline-inducible expression of H3-FLAG. Various forms of RAD51 were transfected into these cells. (G) GST pulldown assay with GST-H3 and His-UHRF1 in the presence or absence of His-RAD51 or His-RAD51D231A. His-tagged proteins were detected with anti-his antibody by western blotting (the Upper panel). The Bottom panel shows Coomassie blue staining of the purified proteins used in the assay. (H) GST pulldown assay with GST-H3 or GST-H3R17A and His-RAD51. The proteins were separated in an SDS-PAGE gel and visualized with Coomassie blue staining. (I) Ubiquitination of H3 or H3R17A in vitro. (J) GST pulldown assay with GST-UHRF1 and His-RAD51 in the presence of increasing amounts of H3. His-RAD51 and H3 were detected with western blotting (the Upper 2 panel). The Bottom panel shows Coomassie blue staining of the purified proteins used in the assay. (K) GST pulldown assay with GST-RAD51 and His-UHRF1 in the presence of histone H3 peptide (AA1-23) or K9me2-modified H3 peptide. GST was used as a negative control. His-UHRF1 and H3 were detected with western blotting (the Upper 2 panels). Coomassie blue staining (the Bottom panel) shows the proteins used in the assay. (L) A model illustrating the relationship among RAD51, UHRF1, and H3. The results of the in vitro assay above with purified proteins could only be explained if RAD51 binds and presents H3 to UHRF1 for ubiquitination. We therefore decided to determine whether RAD51 is indeed a histone H3 binding protein. We purified histones from 293T cells (SI Appendix, Fig. S5A) and also expressed and purified H3 from Escherichia coli. In a GST pulldown assay, we could clearly see that RAD51 interacts with H3 either in histone complexes (SI Appendix, Fig. S5B) or in a purified form (SI Appendix, Fig. S5C). We then generated two C-terminal truncated RAD51 (RAD511–227 and RAD511–257, SI Appendix, Fig. S5D) fused with GST. As shown in SI Appendix, Fig. S5D, the full-length RAD51 and RAD511–257 strongly interacted with H3, whereas RAD511–227 was incapable of doing so, suggesting that the stretch of 30 amino acid residues between 227 and 257 contains the site that interacts with H3. Further deletion and mutation of this stretch of residues nailed down D231 as a critical residue required for the interaction with H3. This site is just 9 residues away from the Walker B domain and 23 residues upstream of the UHRF1-interacting site R254 (Fig. 3C). Mutating D231 to alanine (RAD51D231A) abolished the ability of RAD51 to bind to H3 (Fig. 3C), and more importantly, this mutant, unlike RAD51R254Q, is incapable of stimulating H3 ubiquitination by UHRF1 both in vivo (Fig. 3D) and in vitro (Fig. 3E). However, this site is not involved in RAD51–UHRF1 interaction, since RAD51D231A could still bind to UHRF1 (SI Appendix, Fig. S5E) and protect DNMT1 from UHRF1-mediated ubiquitination and degradation (SI Appendix, Fig. S5F). We next asked whether RAD51 presents H3 to UHRF1 or, in other words, helps H3 interact with UHRF1. For that, we expressed H3 with FLAG tag at its C terminus (H3-FLAG) in HCT116 cells containing inducible (tet-on) shRAD51 together with RAD51 and HA-tagged UHRF1, immunoprecipitated H3-FLAG, and looked at the amount of HA-UHRF1 in H3 immunoprecipitates. The expression of RAD51 or RAD51R254Qin RAD51-depleted cells allowed abundant UHRF1 to be brought down by H3 (Fig. 3F, compare lanes 4 and 5 to lane 2), but the expression of RAD51D231A could not do so (Fig. 3F, lane 6), strongly suggesting that the interaction between H3 and UHRF1 requires the presence of RAD51. We further examined the effect of RAD51 on UHRF1–H3 interaction in vitro with purified proteins in a GST-pulldown assay. As shown in Fig. 3G, with the addition of RAD51 to the pulldown assay, GST-H3 could bring down a lot more UHRF1 than without, but the addition of RAD51D231A had no such an enhancement effect. Taken together, these results strongly argue that RAD51 binds and presents histone H3 to UHRF1 for ubiquitination. Next, we wanted to know where in H3 RAD51 binds to. The most likely place is the N-terminal tail which protrudes out from the histone core and is available for binding by other proteins. Indeed, synthetic H3 N-terminal tail (aa. 1–23) could bind RAD51 (SI Appendix, Fig. S5G). K9me2, a H3 modification often seen in transcription silent regions of the chromatin, enhanced the binding, whereas K4me1, a modification associated with active gene expression, showed no effect on the binding (SI Appendix, Fig. S5G). Given that D231 in RAD51 is responsible for H3 interaction, we reasoned that this acidic residue might form a salt bond with a basic residue in the H3 tail. Through multiple rounds of mutagenesis, we determined that it was R17 in the H3 tail that mediates the interaction with RAD51. Mutating this site to Ala nearly abolished the interaction between H3 and RAD51 (Fig. 3H), and the ubiquitination of H3R17A by UHRF1 was much less stimulable than that of wild-type H3 by RAD51 (Fig. 3I, compare lane 2 to lane 3, and lane 3 to lane 5). The fact that RAD51 inhibits DNMT1 ubiquitination by UHRF1 (Fig. 1G) seems at odds with the observation here that the recombinase promotes H3 ubiquitination by UHRF1 (Fig. 3E). To reconcile these observations, we hypothesized that H3-bound UHRF1 will not interact with RAD51 anymore or, in other words, histone H3 prevents UHRF1 from interacting with RAD51, therefore avoiding the inhibition of UHRF1’s E3 function by RAD51. To test that hypothesis, we performed a GST pulldown assay in which increasing amounts of H3 were added to the GST-UHRF1 and His-RAD51 mix. As shown in Fig. 3J, with more and more H3 bound to UHRF1, there were less and less RAD51 associated with UHRF1. Similar results were obtained with GST-RAD51 pulldown of UHRF1 (SI Appendix, Fig. S5H). In addition, the synthetic H3 N-terminal tail (SI Appendix, Fig. S5G) was also effective in blocking RAD51–UHRF1 interaction as expected (Fig. 3K). These results strongly suggest that in the presence of H3, RAD51, and UHRF1 do not interact with each other anymore, but both do interact with H3, forming a complex much like two grapes on a vine (Fig. 3L). RAD51 Promotes DNMT1 Recruitment and the Maintenance Methylation of DNA. Since the ubiquitination of H3 by UHRF1 is required for the recruitment of DNMT1 and maintenance methylation, it is expected that weakening UHRF1–H3 interaction and H3 ubiquitination due to the lack of RAD51 function would interfere with genomic DNA methylation. If true, this would provide an explanation for the observation earlier that the restoration of DNMT1 expression could not rescue the DNA methylation defects caused by RAD51-depletion (Fig. 2 and SI Appendix, Fig. S4). To that end, we first examined the interaction between DNMT1 and H3. As expected, depleting the expression of UHRF1 blocked the interaction as detected with the anti-H3 immunoprecipitation assay (Fig. 4A). On the other hand, depleting the expression of RAD51 caused the downregulation of DNMT1, and hence less DNMT1–H3 interaction was detected (Fig. 4A). By reexpressing RAD51, DNMT1 expression as well as its interaction with H3 could be restored, and as expected, reexpressing RAD51R254Q could not restore DNMT1 expression nor its interaction with H3 (Fig. 4A). Further, the expression of RAD51D231A in RAD51-depleted cells could restore DNMT1 expression, but the interaction with H3 could not be restored (Fig. 4A). To look into the recruitment of DNMT1 directly, we performed immunofluorescence staining of the methyltransferase in S phase HeLa cells (EdU labeled). The cells were manipulated through RNA silencing and reexpression of relevant proteins (SI Appendix, Fig. S6A). As shown in Fig. 4 B and C, depleting the expression of DNMT1, UHRF1, or RAD51 all diminished DNMT1 staining in the nuclei, and the reexpression of neither RAD51D231A nor RAD51R254Q could bring back DNMT1 staining as wild-type RAD51. Thus, both RAD51R254Q and RAD51D231A do not support DNMT1 recruitment through H3, although for totally different reasons. RAD51 promotes DNMT1 recruitment. (A) Immunoprecipitation analysis of the interaction between DNMT1 and H3 in HCT116 cells with RAD51 depleted and supplemented with RAD51 mutants. (B) Immunofluorescent staining of endogenous DNMT1 in HCT116 cells similarly manipulated as in A. The cells were also pulse labeled with EdU. (C) The ratio of DNMT1 and EdU colocalization was determined and plotted. Data are presented as mean ± SD. * indicates P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. Consequently, while reexpressing RAD51 in the RAD51-depleted HeLa cells (SI Appendix, Fig. S6B) could restore DNA methylation as detected with the dot blot assay for 5mC (SI Appendix, Fig. S6C) or through immunofluorescent staining of 5mC (SI Appendix, Fig. S6D), reexpressing neither RAD51R254Q nor RAD51 R231A could do so (SI Appendix, Fig. S6 C and D). These results clearly demonstrate that RAD51 is critical in the maintenance methylation of genomic DNA by DNMT1 through protecting the methyltransferase from degradation and through promoting H3 ubiquitination and DNMT1 recruitment. DNMT1 Is Protected from UHRF1 during Maintenance Methylation. It is well established that the maintenance methylation of DNA starts with UHRF1 recognizing hemimethylated DNA, leading to interaction with and ubiquitination of histone H3 to generate DNMT1 recruitment signal. We wondered what happens to UHRF1–RAD51 interaction if the former is bound with hemimethylated DNA. As shown in Fig. 5A, the addition of hemimethylated DNA blocked the pulldown of UHRF1 by GST-RAD51. Thus, the interaction between UHRF1 and RAD51 can be broken up by either H3 (Fig. 5 J and K and SI Appendix, Fig. S5H) or hemimethylated DNA. We further wondered whether the binding of hemimethylated DNA helps UHRF1 bind to H3. To that end, we used the synthetic H3 tail to pulldown UHRF1 in the presence of hemimethylated DNA or RAD51. As shown in Fig. 5B, hemimethylated DNA indeed could enhance H3 binding by UHRF1, although a bit less potent than RAD51. The modification of the tail at K9 with bimethylation (K9me2) also enhanced the binding as this creates another binding site besides R2 at the N terminus for UHRF1. However, this modification was less potent than hemimethylated DNA. Further, the combinations of any two of the three factors all enhanced H3 binding by UHRF1 much more than any single one of them (Fig. 5B), suggesting synergistic effects among the three. DNMT1 is protected from UHRF1 during maintenance methylation. (A) GST pulldown assay with GST-RAD51 and His-UHRF1 in the presence of hemimethylated DNA. His-UHRF1 was detected with western blotting (the Upper panel). The Bottom panel shows Coomassie blue staining of the purified proteins used in the assay. (B) Biotin pulldown assay with the synthetic H3 tail (AA1-23-biotin) with or without K9me2 modification and His-UHRF1 in the presence of hemimethylated DNA or His-RAD51. His-tagged proteins were detected with western blotting. (C) GST pulldown assay with GST-UHRF1 and His-DNMT1 in the presence of hemimethylated DNA, H3, or histones. His-DNMT1 and H3 were detected with western blotting (the Upper 2 panels). The Bottom panel shows Coomassie blue staining of the purified proteins used in the assay. (D) In vitro ubiquitination of DNMT1 in the presence of various agents (H3, RAD51, or histones). (E) Same as D but in the presence of RAD51 or hemimethylated DNA. Apparently, once it is bound with H3 and/or hemimethylated DNA, UHRF1 no longer associates with RAD51. However, this raises a question: how does the incoming DNMT1 at the sites of maintenance DNA methylation then avoid being ubiquitinated by UHRF1 now free of RAD51 inhibition of its E3 ligase activity? The most likely answer is that UHRF1 now also no longer interacts with DNMT1 due to conformational changes induced by H3 and/or hemimethylated DNA binding. Indeed, GST-UHRF1 could not pulldown DNMT1 anymore in the presence of hemimethylated DNA, H3, or even histones (Fig. 5C). Consequently, H3 or histones could block polyubiquitination of DNMT1 by UHRF1 (Fig. 5D), so could hemimethylated DNA (Fig. 5E). As observed before (Fig. 1 J and K), RAD51 blocked the polyubiquitination as well (Fig. 5 D and E). The UHRF1-Regulating Function of RAD51 Is Separable from Its Function in DNA Damage Repair and Is Conserved in Budding Yeast Rad51. RAD51 is best known for its function in homologous recombination (HR) repair of DNA damage. We wondered whether that function is separable from its function in DNA maintenance methylation identified here. To that end, we depleted RAD51 expression in a HR repair-reporter cell line, U2OS/DR-GFP (16), reexpressed the recombinase (SI Appendix, Fig. S7A), and then assessed HR repair efficiency. As shown in SI Appendix, Fig. S7B, the reduction in HR repair efficiency caused by RAD51 depletion could be brought back by the reexpression of wild-type RAD51, RAD51R254Q, or RAD51D231A without noticeable differences in their efficiencies, indicating that neither the interaction with UHRF1 nor the interaction with histone H3, two functions of RAD51 implicated in the maintenance methylation of DNA, is required in HR repair. In addition to HR repair, RAD51 also plays a role in DNA replication through maintaining the stability of replication fork (8). Indeed, depleting RAD51 expression caused the appearance of spontaneous DNA damage (as detected by immunofluorescent staining of γH2AX) due to replication fork collapse. Again, supplementing these RAD51-depleted cells with either RAD51R254Q or RAD51D231A could suppress γH2AX focus formation as effectively as wild-type RAD51 (SI Appendix, Fig. S7C). Moreover, RAD51 is an essential gene for cell viability (likely due to its function in replication). In U2OS cells, depleting the expression of RAD51 caused the cells to die off 48 h after the siRNA transfection (SI Appendix, Fig. S7D). Exogenous expression of wild-type RAD51 could rescue the siRAD51-transfected cells from cell death, so could RAD51R254Q and RAD51D231A (SI Appendix, Fig. S7D). Taken together, these results indicate that the critical roles played by RAD51 in maintaining the genome and the epigenome are separable. Rad51 is highly conserved through evolution. The residues in human RAD51 critical for the maintenance methylation of genomic DNA, D231, and R254, are conserved in Rad51 from the budding yeast Saccharomyces cerevisiae (SI Appendix, Fig. S8A). We therefore wondered whether the yeast Rad51 could function as its human counterpart in regulating UHRF1. Indeed, yRad51 could interact with human UHRF1 (SI Appendix, Fig. S8B) and rescue the DNMT1 protein levels in the cells depleted of RAD51 expression (SI Appendix, Fig. S8C). In the in vitro ubiquitination assay, yRad51 could suppress UHRF1-mediated DNMT1 ubiquitination as good as human RAD51 (SI Appendix, Fig. S8D). Moreover, yRad51 could also stimulate histone H3 to bind to UHRF1 (SI Appendix, Fig. S8E) as well as the ubiquitination of H3 by UHRF1 (SI Appendix, Fig. S8F). These results demonstrate that RAD51’s ability to regulate UHRF1 predated the appearance of Uhrf1 in evolution. RAD51 Suppresses Neuronal Differentiation of PC12 Cells. RAD51 is a human genetic disease gene. It is haploinsufficient in congenital mirror movement (CMM), a condition characterized by involuntary movement of one side of the body when the intentional movement is the other side (mirror side) (9, 10). Since all the other genes mutated in CMM are related to neural development and given the role of epigenetic regulation of gene expression in development and differentiation, we suspected that RAD51 might be involved in neuronal differentiation through its function in the maintenance methylation of genomic DNA. To look into that, we employed the PC12 pheochromocytoma cells as a model for neuronal differentiation (17, 18). These cells will differentiate to neuron-like cells characterized by the outgrowth of neurites when treated with nerve growth factor (NGF). We first depleted Rad51 expression in PC12 cells (SI Appendix, Fig. S9A). It became obvious that those cells with Rad51 depleted started to grow neurites spontaneously (SI Appendix, Fig. S9B), and the addition of NGF to the media induced much stronger neurite outgrowth in the knockdown cells than in the control cells (SI Appendix, Fig. S9C), indicating that Rad51 does function to suppresses neuronal differentiation. Depleting the expression of Dnmt1 showed similar effects on neurite outgrowth in these cells (SI Appendix, Fig. S9 A–C). Further, DNMT1 as well as the maintenance methylation of genomic DNA is similarly regulated by RAD51 in PC12 cells (SI Appendix, Fig. S9 D and E). Taken together, these data support the notion that the epigenetic methylation of genomic DNA is critical in regulating neuronal differentiation. To further evaluate the function of Rad51 in the differentiation of PC12 cells, we established stable PC12 cell lines with inducible expression of GFP-RAD51 (or RAD51 mutants). These lines were transfected with siRad51 first and then induced to express GFP-RAD51 (SI Appendix, Fig. S9F). The lengths of the neurites in these cells were measured and compared (Fig. 6). As shown above, the depletion of Rad51 expression resulted in strong outgrowth of neurites, which was suppressed by the exogenous expression of RAD51WT. However, the expression of neither RAD51R254Q nor RAD51D231A, both defective in supporting the maintenance methylation of genomic DNA in PC12 cells (SI Appendix, Fig. S9E), could suppress the neurite outgrowth stimulated by the Rad51 depletion. These results strongly argue that RAD51 regulates neuronal differentiation through its function in epigenetic methylation of genomic DNA. RAD51 suppresses neuronal differentiation of PC12 cells. PC12 cells with inducible expression of GFP-tagged RAD51 (wild type or mutants) were transfected with siRad51 for 24 h and induced to express RAD51 doxycycline (1 μg/mL) and to differentiate with 50 ng/mL NGF for 48 h. The length of neurite outgrowth in 300 cells was measured and grouped. Data are presented as mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.001. Given that RAD51 is haploinsufficient in CMM, we wondered whether reducing its expression level by half would impact DNMT1 expression. Indeed, using inducible RAD51 knockdown to modulate the expression levels of RAD51 in HCT116 and PC12 cells, we found that down-regulating RAD51 by half did cause DNMT1 levels to decrease by more than 30% in both cell lines (SI Appendix, Fig. S10 A and B), indicating that DNMT1 strongly relies on RAD51 for stable expression. Discussion DNA methylation is a major form of epigenetic control of gene expression. We show here that RAD51, an evolutionarily conserved recombinase essential for DNA double-strand break repair, contributes critically to the maintenance methylation of genomic DNA through promoting the stability and recruitment of DNMT1 (Fig. 7). The recombinase functions by regulating UHRF1, a multidomain protein essential for DNMT1 recruitment to daughter DNA strand for methylation after replication, and the regulation goes in two ways. On the one hand, RAD51 directly interacts with UHRF1 and inhibits its E3 activity toward DNMT1. This interaction is mediated by the residue R254 in RAD51. On the other hand, RAD51 helps UHRF1 to ubiquitinate the H3 tail for the generation of DNMT1 recruitment signal. This function entails an interaction with H3 via the residue D231 in RAD51. A schematic illustration of RAD51’s function in maintenance DNA methylation. The recombinase interacts with (via R254) and inhibits UHRF1’s E3 ubiquitin ligase activity toward DNMT1, preserving the stability of the latter. It also helps UHRF1 in the generation of DNMT1 recruitment signal by binding to H3 (via D231). As an E3 ubiquitin ligase, UHRF1 not only can perform monoubiquitination on H3 tails to generate DNMT1 recruitment signal (6) but also polyubiquitination on a number of proteins that can lead to their proteasomal degradation when the ubiquitin chain is linked via K48 or can act as a modification for signaling when the ubiquitin chain is linked via K63 (13, 19, 20). Interestingly, DNMT1 can be polyubiquitinated by UHRF1 and subsequently degraded by the proteasome, which can be reversed by the deubiquitinase USP7 (21, 22). Although depletion or knockout of USP7 resulted in some degree of downregulation of DNMT1, there was little impact on global DNA methylation levels (21, 22), suggesting that a sufficient amount of DNMT1 was spared from polyubiquitination and degradation. Indeed, we observed almost no increases in DNMT1 protein levels when UHRF1 expression was depleted (Fig. 1A and SI Appendix, Fig. S2A), indicating that under normal conditions DNMT1 is not ubiquitinated and degraded extensively. We now know this is due to the presence of RAD51 which interacts with UHRF1 and inhibits its E3 activity toward DNMT1 outside the sites of maintenance methylation. Inside the sites of maintenance methylation, DNMT1 is protected by H3 and/or hemimethylated CpG dinucleotides both of which are capable of disrupting the interaction between UHRF1 and DNMT1 (Figs. 5 D and E and 7). It was suggested previously that the interaction between UHRF1 and DNMT1 probably helps recruit DNMT1 for maintenance methylation (4, 5). However, with the identification of H3 ubiquitination as the major recruitment signal (6), the importance of the UHRF1–DNMT1 interaction in that regard becomes questionable. We would argue that such an interaction might serve to limit the level of DNMT1 through proteasomal degradation to prevent unscheduled methylation of the genomic DNA as overexpressed DNMT1 could result in the loss of imprinting during development (23) or the repression of the expression of tumor suppressor genes (24). In addition, DNMT1 also regulates gene expression independent of its methyltransferase activity through complexing with other transcription-modulating proteins (either epigenetic or nonepigenetic) such as Rb and histone deacetylases (25, 26), which might need some restrain from UHRF1. Apparently, the restrain provided by UHRF1 must itself be restrained; otherwise, DNMT1 would be degraded so much that maintaining global DNA methylation becomes compromised, as seen in RAD51-depleted cells (Fig. 2 and SI Appendix, Fig. S6). RAD51 is not only an inhibitor of UHRF1 but ironically also a helper. It helps the latter to bind H3 (Fig. 3 F and G). We demonstrated that RAD51 itself is a H3 binding protein and identified D231 in RAD51 and R17 in H3 as two critical residues for the interaction (Fig. 3 C and H). These two residues very likely form a salt bond between them. Since the RAD51 binding site R17 is located in the H3 tail, one can foresee that the binding by RAD51 perhaps forces the flexible and unstructured H3 tail into a more rigid conformation that makes it easier for UHRF1 to contact its binding sites R2 and/or K9. In other words, RAD51 presents H3 to UHRF1, thus a helper. Homologous recombination is an ancient process that enables the exchange of genetic materials between and within species. It shapes the genome and drives evolution. Key to this process is the RAD51 recombinase which utilizes energy released from ATP hydrolysis to drive DNA strand invasion and pairing. There are five RAD51 paralogs in mammals including RAD51B, RAD51C, RAD51D, XRCC2, and XRCC3 that work together with RAD51 in HR repair of double strand breaks (27). These paralogs share limited homologies with RAD51 and form complexes among them to function. The key residues identified in RAD51 here that are required for the maintenance DNA methylation (D231 and R254) are not conserved among the paralogs, suggesting that the function in DNA methylation is unique to RAD51. Epigenetic modification of the genome was evolved to help control gene expression according to developmental plans and to help adjust gene expression in response to environmental cues. RAD51 is haploinsufficient in congenital mirror movement (9, 10). It is most likely the function in epigenetic DNA methylation, not its role in homologous recombination, that implicates RAD51 in this condition. We demonstrated in PC12 cells that Rad51 suppressed neuronal differentiation and such suppression relied on its function in DNA maintenance methylation (Fig. 6 and SI Appendix, Fig. S9). In CMM, perhaps the insufficient expression of RAD51 causes some downregulation of DNMT1 as we could see in PC12 and HCT116 cells (SI Appendix, Fig. S9), which might result in insufficient methylation of certain regions of the chromatin in the affected neurons and cause misregulation of genes involved in axon guidance such as DCC or DNAL4, two other genes that also cause CMM when mutated monoallelically. Methylating genomic DNA to control gene expression is a later event than homologous recombination in evolution. Thus, it is not surprising that both the budding yeast S. cerevisiae and the fission yeast Schizosaccharomyces pombe contain Rad51 and are proficient in HR, but lack 5-methylcytidine and do not contain Uhrf1-like genes in their genomes (28). As a later comer in evolution, UHRF1 apparently takes the advantage of an older already-present protein, RAD51, to help perform its function. Indeed, the budding yeast Rad51 is able to regulate UHRF1 (SI Appendix, Fig. S8). Thus, evolution did not invent a new protein to aid UHRF1 but instead made use of an existing one by multifunctionalizing it. As a result, RAD51 in higher eukaryotes now plays two separate and essential roles, one in homologous recombination and one in the maintenance methylation of genomic DNA, making the recombinase a guardian of both the genome and the epigenome. Materials and Methods In Vitro Ubiquitination Assay. 100 ng of E1 (UBA1, Abnova, USA), 300 ng of E2 (UBC13/MMS2, Abnova, USA), 1 μg of HA-ubiquitin (R&D Systems, USA), 1 μg of E3 (purified His-tagged UHRF1), 2 μg of His-DNMT1 or 3 μg H3 (or histones) were mixed in the reaction buffer (50 mM TrisHCl, pH 7.5/2 mM MgCl2/2 mM ATP/1 mM DTT) with a final volume of 40 μL and the mixture was incubated at 37 °C for 60 min (for DNMT1) or 90 min (for H3). The reaction was stopped by boiling in 5× SDS loading buffer and analyzed with western blotting. Other Methods. Cell culture and transfection, immunofluorescent staining, immunoprecipitation, mass spectrometry analysis, western blotting, recombinant protein purification, in vivo ubiquitination assay, dot blotting, WGBS sequencing and bioinformatic analysis, and statistical analysis can be found in SI Appendix, Supplemental Materials and Methods. Supplementary Material Appendix 01 (PDF)