RDM15 is required for transcriptional silencing at RdDM target loci
To clone the RDM15 gene, we used TAIL-PCR and found a T-DNA insertion in the 7th exon of AT4G31880 (Fig. 1c). A complementation test showed that a transgene containing the native promoter-driven genomic sequence of AT4G31880 can rescue the pRD29A-LUC silencing phenotype of the rdm15-1/ros1-1 double mutant (Supplementary Fig. 1c). To confirm the silencing function of RDM15, we ordered two T-DNA insertion alleles of RDM15 (Col-0 background): Salk_013481 and Salk_024055 (hereafter referred to as rdm15-2 and rdm15-3). Both of these mutants contain T-DNA insertions in the 7th exon of RDM15 (Fig. 1c and Supplementary Fig. 1d). The silencing of AtGP1 and AtMEA-ISR, two known RdDM targets 23, is released by both rdm15-2 and rdm15-3 mutations (Fig. 1d), supporting the notion that RDM15 is required for the silencing of RdDM target loci.
RdDM controls the expression of the DNA demethylase ROS1 via methylation of the MEMS sequence located in the promoter region of ROS1 24, 25. ROS1 expression is dramatically decreased in many RdDM mutants, including nrpd1a, nrpd1b, and ago4 24, 25. We found that rdm15 mutants have lower ROS1 transcript levels (Fig. 1e and Supplementary Fig. 1e), although the levels are not as low as in nrpd or ago4 mutants. These data suggest that RDM15 functions in RdDM-dependent transcriptional gene silencing (TGS).
RDM15 is required for RdDM-dependent DNA methylation
To characterize the effect of RDM15 on DNA methylation, we generated single-base resolution maps of DNA methylation for Col-0, rdm15-2, and rdm15-3, with two biological replicates for each mutant. Principal components analysis showed very good consistency between the replicates (Supplementary Fig. 2a). Using the R package methylKit 26, which considers variations among replicates of two mutant alleles, we identified 1,390 differentially methylated regions (DMRs) in rdm15 mutants compared with the WT; most of the DMRs (1,354 out of 1,390) were hypomethylated, suggesting that RDM15 is required for DNA methylation at more than 1,300 endogenous genomic regions (Supplementary Table 1). We compared the DNA methylation patterns of rdm15 mutants with those of known RdDM mutants. NRPD1 and NRPE1 are the largest subunits exclusive to Pol IV and Pol V, respectively. Using the same methylKit method and compared to the WT, we identified 4,293 hypo DMRs in nrpd1-3 (Col-0 ecotype) and 4,629 hypo DMRs in nrpe1-11 (Col-0 ecotype) (Supplementary Table 1). We found that the Pol IV, Pol V, and RDM15 targets have comparable genomic compositions. Among the hypo DMRs, 82% in nrpd1-3, 81% in nrpe1-11, and 79% in rdm15 are in TE regions (Fig. 2a). Similar to known RdDM targets, rdm15 DMRs show DNA hypomethylation in all three sequence contexts (mCG, mCHG, and mCHH; Fig. 2b), although the methylation level in rdm15 mutants is not as low as in nrpd1-3. We further analyzed the overlap between RDM15 targets and known RdDM targets. As shown in Fig. 2c, 94% (1,269/1,354) and 95% (1,281/1,354) of rdm15 hypo DMRs overlap with the nrpd1 and nrpe1 hypo DMRs, respectively. DNA methylation profiles of several representative rdm15 hypo DMRs are shown in Fig. 2d and Supplementary Fig. 2b. These results revealed that RDM15 is required for methylation at a subset of RdDM target loci, suggesting that RDM15 may be a novel RdDM component.
RDM15 regulates RdDM-dependent siRNA accumulation
We generated genome-wide siRNA profiles for WT (Col-0), rdm15-2, and rdm15-3 with two biological replicates for each genotype (Supplementary Fig. 3a). In Arabidopsis, RdDM has two main steps, i.e., biogenesis of 24-nt siRNAs and guidance of DRM2 to RdDM target by the siRNAs. Genomic regions showing RdDM-dependent DNA methylation always have siRNA enrichment15. To determine whether RDM15 is involved in siRNA biogenesis, we compared rdm15 mutants with Col-0, and identified 2,808 RDM15-dependent siRNA clusters (hereafter referred to as RDM15 siRNAs) (Supplementary Table 2) that showed decreased 24-nt siRNA accumulation in rdm15 mutants relative to Col-0. This result suggested that RDM15 is required for 24nt siRNA accumulation in vivo (Fig. 3a).
RdDM-dependent siRNAs can be classified into upstream siRNAs (siRNAs dependent on Pol IV only) and downstream siRNAs (siRNAs dependent on both Pol IV and Pol V) 15.The upstream siRNAs are affected only in mutants defective in upstream RdDM components, such as nrpd1, whereas the downstream siRNAs are not only affected in these mutants but also in mutants defective in downstream RdDM components, such as nrpe1 and drm2 (Fig. 3b). To position RDM15 in the RdDM pathway, we examined the enrichment of RDM15-dependent siRNAs, and found that RDM15 siRNA levels are lower in nrpd1-4, nrpe1-12, and drm2-2 compared to the WT (Fig. 3b). In addition, the Pol IV-only siRNAs are not significantly affected in rdm15 mutants (Fig. 3a). These results showed that RDM15 is mainly required for the accumulation of downstream siRNAs in the RdDM pathway, suggesting that RDM15 functions in a downstream step of RdDM.
To further understand the relationship between the siRNAs and DNA methylation in rdm15 mutants, we monitored changes in siRNA enrichment at hypo DMRs in rdm15 mutants. The siRNA levels were clearly lower at hypo-DMRs in rdm15 mutants than in the WT (Fig. 3c). Consistent with this result, the DNA methylation level, including mCG, mCHG, and mCHH, was decreased in regions of RDM15-dependent siRNA clusters (Fig. 3d and Supplementary Fig. 3b). The siRNA levels at several representative hypo-DMRs are shown in Fig. 3e and Supplementary Fig.3c. These results suggest that RDM15 influences DNA methylation at RdDM target regions by regulating siRNA accumulation.
RDM15 physically interacts with NRPE3B, a subunit exclusive to Pol V
To further investigate how RDM15 affects RdDM, we identified RDM15-interacting proteins by performing immunoprecipitation (IP) followed by liquid chromatography-tandem mass spectrometry (LC-MS/MS) with RDM15-3xMYC/rdm15-3 and RDM15-3xFLAG/rdm15-3 plants (Supplementary Fig. 4). WT plants that do not express RDM15-3xMYC or RDM15-3xFLAG were used as controls. We found that NRPE3B, a subunit exclusive to Pol V, was co-immunoprecipitated with anti-MYC antibodies and anti-FLAG antibodies in the RDM15-3xMYC and RDM15-3xFLAG transgenic plants, respectively, but not in the control plants (Fig. 4a and Supplementary Table 3). We validated the interaction between RDM15 and NRPE3B using a split luciferase complementation assay and a BiFC assay in tobacco leaves (Fig. 4b and 4c). These data indicated that RDM15 may affect RdDM by interacting with Pol V.
The RDM15 tudor domain specifically recognizes H3K4me1
The results above show that RDM15 functions in the regulation of siRNA accumulation and DNA methylation at a subset of RdDM target loci. To further investigate how RDM15 functions, we analyzed its protein sequence and found two domains with known functions: a N-terminal ARM repeat, which is known to mediate protein–protein interactions, and a C-terminal tudor domain, which functions as a histone mark reader (Fig. 5a). The tudor domain has been found in many proteins related to epigenetic regulation and acts as a histone mark reader module that recognizes methylated lysine or arginine marks 27, 28. In plants, histone mark readers sometimes have plant-specific binding targets and functions 12. This prompted us to explore the histone mark binding property of the RDM15 tudor domain. In our assay using histone peptide arrays, which contained several hundred combinations of histone marks, the RDM15 tudor domain showed a significant binding to H3K4me1 and H4K20me1 marks (Supplementary Fig. 5a and Supplementary Table 4). We used isothermal titration calorimetry (ITC) to further confirm the binding and to measure the binding affinity between the RDM15 tudor domain and different methyllysine-modified histone peptides. The H3K4me1 peptide yielded the strongest binding (1.47 mM) (Fig. 5b), while the H4K20me1 yielded a 15-fold lower binding affinity (22.9 mM) (Supplementary Fig. 5b), indicating that H4K20me1 is not the optimal binding partner for RDM15. In addition, researchers previously reported that the existence of H4K20 methylation in plants is controversial and that no in vivo H4K20 methylation could be detected by mass spectrometry 29. We therefore focused on the interaction between the H3K4me1 mark and RDM15. We assessed the binding between the RDM15 tudor domain and peptides with different H3K4 methylation states. ITC measurements clearly showed that RDM15 tudor domain bound much more strongly to the monomethylation state of H3K4 than to the unmethylated H3K4 (H3K4me0) or to the higher methylation states of H3K4 (H3K4me2 and H3K4me3) (Fig. 5b).
The structure of RDM15 tudor domain in complex with H3K4me1 peptide
To further investigate the molecular mechanism underlying the specific recognition of the H3K4me1 mark by RDM15, we carried out structural studies. The crystal structure of the RDM15 tudor domain in complex with the H3(1-15) K4me1 peptide was determined using the SAD method and was refined to 1.7-Å resolution (Supplementary Table 5, Fig. 5c). Although a 15-residue H3K4me1 peptide was used in the crystallization, only the H3A1 to H3Q5 segment of the peptide exhibited well-defined electron density and was built into the final model (Fig. 5c). The tudor domain of RDM15 exhibits a canonical tudor domain fold with five b-strands forming a twisted b-barrel structure that resembles the structure of other tudor domains (Fig. 5c) 27.
Our results show that the RDM15 tudor domain can specifically recognize H3K4me1. The main chain of the H3K4me1 peptide has a ‘U’-shaped conformation such that the N-terminal H3A1 and the C-terminal H3Q5 extend out away from the peptide-protein binding interface (Fig. 5c). The side chains of H3R2 and H3K4me1 form a pincer-like conformation, thereby anchoring on two adjacent negatively charged surface pockets of the tudor domain and highlighting a significant structural and chemical complementarity (Fig. 5d). In detail, the H3R2 inserts its side chain into a negatively charged pocket and forms hydrogen-bonding interactions with Glu647 and Gln654 of RDM15, as well as a salt bridge interaction with Glu647 (Fig. 6a). For H3K4me1, the interactions can be divided into two parts. Three aromatic residues, Trp616, Tyr623, and Tyr641, form an aromatic cage to accommodate the methyl group of the monomethyllysine from one side, which is similar to other canonical methyllysine readers 28, 30 (Fig. 6b). On the other side, the two free protons of the monomethylammonium group are fully coordinated with Asp643 and Asp645 by hydrogen bonding and salt bridge interactions (Fig. 6b), representing a new monomethyllysine recognition mechanism (Supplementary Fig. 6 and see detailed analysis in the Discussion). The H3T3 positioned between H3R2 and H3K4me1 stretches its side chain against the protein and is not involved in the interaction with the RDM15 tudor domain (Fig. 5c). In general, both H3R2 and H3K4me1 are specifically recognized by a hydrogen bonding network and electrostatic interactions involving their charged side chains, as well as by an aromatic cage that captures the methylation modification (Fig. 6c). Mutations of residues of either the aromatic cage or of those involved in hydrogen bonding interactions with the peptide result in a significant decrease in the binding affinity (Fig. 6d and 6e), revealing that these residues are critical for the recognition of H3K4me1 by RDM15.
In the structure of the RDM15-H3K4me1 complex, the recognition of the H3 tail by the RDM15 tudor domain highlights the specific recognition of H3R2 and H3K4me1, while other surrounding residues and the intervening H3T3 do not contribute to the recognition. The RDM15 tudor domain therefore recognizes an RXKme1 (X here stands for any residue) motif on histone tails. Among all four types of Arabidopsis histone proteins, there are two additional sites that contain the same motif and that might be recognized by the RDM15 tudor domain: H2A R3TK5 and H4 R3GK5. A previous mass spectrometry study showed that the lysine residues of both of these two sites can be acetylated, but no methylation modification was identified in vivo 29. It follows that although the RDM15 tudor domain only recognizes two residues on the histone tail, the specific RXKme1 pattern only occurs in the H3K4 region, which ensures that recognition only occurs at H3K4me1, thereby explaining the sequence specificity.
Characterization of chromatin targets of RDM15 binding
We performed chromatin immunoprecipitation using native promoter-driven RDM15-3xFLAG/rdm15-3 followed by high throughput sequencing (ChIP-seq). We observed higher RDM15 protein enrichment in RDM15-dependent siRNA regions than in Pol IV-only siRNA regions (Fig. 7a), which is consistent with our above finding that RDM15 is required for the accumulation of RDM15 siRNAs but does not affect the accumulation of Pol IV-only siRNAs (Fig. 3). When we ranked RDM15 siRNA regions by the difference in siRNA accumulation between rdm15 and the WT, we found that the change in siRNA accumulation (rdm15 vs. the WT) was positively correlated with RDM15 protein enrichment; the change in siRNA accumulation in nrpd1 vs. the WT served as the negative control (Fig. 7b and Supplementary Fig. 7a). These results suggest that the influence of RDM15 on siRNA accumulation is correlated with its protein enrichment. In addition, the distribution pattern of H3K4me1 but not of H3K4me2 and H3K4me3 was similar to that of RDM15 (Fig. 7c and Supplementary Fig. 7b), which is consistent with our finding of high affinity binding of RDM15 to H3K4me1. The enrichment of RDM15 and H3K4me1 is shown for several representative RDM15 target regions in Fig. 7d and Supplementary Fig. 7c.