Polymerase-IV occupancy at RNA-directed DNA methylation sites requires SHH1

Abstract

DNA methylation is an epigenetic modification that plays critical roles in gene silencing, development, and genome integrity. In Arabidopsis, DNA methylation is established by DOMAINS REARRANGED METHYLTRANSFERASE 2 (DRM2) and targeted by 24 nt small interfering RNAs (siRNAs) through a pathway termed RNA-directed DNA methylation (RdDM)¹. This pathway requires two plant-specific RNA polymerases: Pol-IV, which functions to initiate siRNA biogenesis and Pol-V, which functions to generate scaffold transcripts that recruit downstream RdDM factors^1,2. To understand the mechanisms controlling Pol-IV targeting we investigated the function of SAWADEE HOMEODOMAIN HOMOLOG 1 (SHH1)^3,4, a Pol-IV interacting protein³. Here we show that SHH1 acts upstream in the RdDM pathway to enable siRNA production from a large subset of the most active RdDM targets and that SHH1 is required for Pol-IV occupancy at these same loci. We also show that the SHH1 SAWADEE domain is a novel chromatin binding module that adopts a unique tandem Tudor-like fold and functions as a dual lysine reader, probing for both unmethylated K4 and methylated K9 modifications on the histone 3 (H3) tail. Finally, we show that key residues within both lysine binding pockets of SHH1 are required in vivo to maintain siRNA and DNA methylation levels as well as Pol-IV occupancy at RdDM targets, demonstrating a central role for methylated H3K9 binding in SHH1 function and providing the first insights into the mechanism of Pol-IV targeting. Given the parallels between methylation systems in plants and mammals^1,5, a further understanding of this early targeting step may aid in our ability to control the expression of endogenous and newly introduced genes, which has broad implications for agriculture and gene therapy.

SHH1 was recently identified as a Pol-IV interacting protein and shown to affect de novo DNA methylation³. To investigate the role of SHH1 in the RdDM pathway genome-wide, we generated siRNA profiles in wild-type Col plants, shh1 mutant plants, and several other RdDM mutants for comparison. In wild-type plants ~12,500 siRNA clusters were defined, representing 84.2% of all uniquely mapping 24 nt siRNAs. Consistent with previous findings, 81.4% of these siRNAs were Pol-IV-dependent^6,7 (Fig. 1a). Analysis of the siRNA clusters reduced in shh1 mutants demonstrated that SHH1 is a major regulator of siRNA levels, affecting 44% of Pol-IV-dependent clusters (Fig. 1b and Supplementary Fig. 1a). These shh1-affected clusters represent the majority of all 24 nt siRNAs, as well as a majority of clusters reduced in two downstream RdDM mutants (drm2 and pol-v) (Fig. 1b and Supplementary Fig. 1a). The overlap of the reduced siRNA clusters in these mutants formed four main subclasses (termed pol-iv only, shh1, shh1/drm2/pol-v, and drm2/pol-v; Fig. 1b), which were used for subsequent analyses. Interestingly, the clusters that depend solely on Pol-IV were more enriched in pericentromeric heterochromatin than those that also depend on SHH1, DRM2, and Pol-V (Fig. 1c and Supplementary Fig. 1b, c), suggesting that different mechanisms may be controlling siRNA production in the euchromatic arms versus pericentromeric heterochromatin.

Figure 1. Epigenetic profile of siRNA clusters affected in RdDM mutants.

a, Pie chart showing the abundance of 24 nt siRNA reads in wild-type (ecotype Col) sequencing libraries (5,967,213 uniquely mapping reads total). b, Schematic Venn diagram showing approximate relationships of 24 nt siRNA clusters reduced in each genotype and the subclasses used for downstream analysis. c, Pie charts showing the chromosomal distribution (based on previously described definitions of pericentromeric heterochromatin and euchromatin¹⁶) of affected siRNA clusters in the indicated subclasses. d and e, Boxplots of siRNA and CHH methylation levels at the subclasses shown in (b) for various RdDM mutants (* indicates significant reduction; P<1e-10 Mann-Whitney U test). f, Metaplots showing CMT3 and Pol-V enrichment at affected siRNA clusters (+/− 5000 bp from the siRNA cluster midpoint).

In shh1 mutants, siRNA levels at SHH1-dependent clusters (shh1 and shh1/drm2/pol-v subclasses) are reduced to nearly zero, while siRNA levels at SHH1-independent clusters experienced little to no change (Fig. 1d). These results demonstrate that SHH1 is a locus-specific RdDM component that has strong affects at a large subset of RdDM loci. Notably, the two downstream RdDM mutants (drm2 and pol-v) have the strongest affect on siRNAs levels at clusters that also require SHH1 (shh1/drm2/pol-v subclass), and these same clusters are amongst the highest siRNA-producing clusters in the genome (Fig. 1d, e and Supplementary Fig. 1d, e). Together, these findings suggest that SHH1, and the downstream RdDM mutants, converge to control siRNA levels at the most active sites of RdDM.

Using whole-genome bisulfite sequencing (BS-seq), we assessed DNA methylation levels at the loci showing reduced siRNA levels and found that, consistent with its interaction with Pol-IV, SHH1 is an upstream RdDM component—shh1 mutants only affect DNA methylation at sites where siRNA levels are reduced (Fig. 1e and Supplementary Fig. 2a). Furthermore, the residual siRNAs present in shh1 mutants appear to target some methylation (Supplementary Fig. 2b), as predicted for an upstream RdDM component. This is in contrast to the downstream mutants, drm2 and pol-v, which reduced DNA methylation to nearly pol-iv levels even at sites that retain siRNAs (Fig. 1e), presumably due to an inability of these mutants to utilize siRNAs to target DNA methylation.

At loci corresponding to the shh1/drm2/pol-v and drm2/pol-v subclasses of siRNA clusters, the observed losses of siRNAs were accompanied with a correspondingly large loss of DNA methylation (Fig. 1e and Supplementary Fig. 2a). However, at the pol-iv only and shh1 subclasses, large losses of siRNAs were accompanied by relatively little DNA methylation loss. A likely explanation for this finding is that other DNA methylation pathways are active at sites corresponding to the pol-iv only and shh1 siRNA clusters. In addition to the RdDM pathway, DNA methylation in Arabidopsis is controlled by two maintenance methyltransferase pathways¹: the DNA METHYLTRANSFERASE 1 (MET1) pathway, which acts to maintain CG methylation, and the CHROMOMETHYLTRANSFERASE 3 (CMT3) pathway, which acts along with several H3K9 histone methyltransferases to maintain CHG and some CHH methylation⁸. Consistent with this explanation we found, using a previously published CMT3 ChIP-seq dataset⁹, that the pol-iv only and shh1 subclasses of reduced siRNA clusters displayed the highest levels of CMT3 occupancy (Fig. 1f), suggesting that CMT3 is able to maintain DNA methylation at nearly wild-type levels at these loci. In contrast, the shh1/drm2/pol-v and drm2/pol-v subclasses, which show dramatic DNA methylation losses in RdDM mutants, display lower levels of CMT3 enrichment (Fig. 1f) and are more highly and precisely enriched for the Pol-V polymerase¹⁰ (Fig. 1f and Supplementary Fig. 2c), suggesting they are primarily targeted by the RdDM pathway.

To test the hypothesis that the siRNA losses observed in shh1 mutants are due to a lack of Pol-IV targeting, we determined the genome-wide profile of Pol-IV occupancy in wild-type and shh1 mutant backgrounds via chromatin immunoprecipitation coupled with high throughput sequencing (ChIP-seq) using a Flag-tagged version of the largest Pol-IV subunit, NRPD1³. Consistent with our profile of Pol-IV-dependent siRNA clusters (Supplementary Fig. 1b), Pol-IV was broadly enriched at pericentromeric heterochromatin (Supplementary Fig. 3a) and at the defined subclasses of siRNA clusters (Fig. 2a and Supplementary Fig. 3b). In the shh1 mutant background, Pol-IV levels were drastically reduced or eliminated specifically at shh1-dependent siRNA clusters (Fig. 2a and Supplementary Fig. 3c), further supporting the biological relevance of our ChIP-seq profile and confirming that the reduced-siRNA phenotype of shh1 mutants is due to altered Pol-IV chromatin association. At shh1-independent siRNA clusters, Pol-IV levels, like siRNA levels, were not reduced in shh1 mutants (Fig. 2a and Supplementary Fig. 3c), suggesting that Pol-IV targeting to these loci requires an alternative mechanism.

Figure 2. Pol-IV levels at defined siRNA clusters.

Metaplots of Pol-IV enrichment over the defined siRNA clusters in the indicated genetic backgrounds. Metaplots extend +/− 5000 bp from the midpiont of the siRNA cluster.

In addition to assessing the levels of Pol-IV enrichment over the affected siRNA cluster subclasses, we also defined 928 reproducible, high confidence Pol-IV peaks using multiple ChIP-seq datasets. These peaks were enriched for siRNAs and DNA methylation (Supplementary Fig. 4a) and preferentially overlapped with the high siRNA-producing shh1/drm2/pol-v or drm2/pol-v clusters as compared to the pol-iv only and shh1 clusters (P<2.2e-16, Fisher's Exact Test), suggesting the ChIP procedure is preferentially identifying sites where Pol-IV is most active. At the 928 defined Pol-IV peaks, we observed a variable level of SHH1-dependency and divided the peaks into three categories, SHH1-independent, SHH1-dependent, and SHH1-enhanced (Supplementary Fig. 4b). In shh1 mutants, DNA methylation and siRNA levels were reduced at the SHH1-dependent sites and, to a lesser extent, at sites defined as SHH1-independent (Supplementary Fig. 4c, d). However, siRNA and Pol-IV levels were increased at SHH1-enhanced sites in shh1 mutants, suggesting a redistribution of Pol-IV to these sites in shh1 mutants (Supplementary Fig. 4b, c). Notably, these SHH1-enhanced sites are unique amongst the Pol-IV peaks as they display very low levels of Pol-V enrichment (Supplementary Fig. 4b), which could explain the correspondingly low levels of CHH methylation observed at these sites in wild-type plants (Supplementary Fig. 4d). Together with our analysis of SHH1-dependent siRNA clusters, these findings demonstrate that SHH1 plays a critical role in facilitating Pol-IV chromatin association at a subset of the most active sites of RdDM.

To gain insight into the mechanism through which SHH1 facilitates Pol-IV targeting, we investigated the function of its previously uncharacterized SAWADEE domain¹¹. Since there are precedents for cross talk between DNA methylation and histone modifications^1,12, we tested the ability of the SAWADEE domain to bind modified histone tails using an Active Motif modified peptide array. This assay revealed that the SAWADEE domain has a preference for H3K9 methylation, but is also influenced by the methylation status of the H3K4 residue, with only unmodified or H3K4me1 modifications being tolerated (Supplementary Fig. 5a). To confirm these results, isothermal calorimetry (ITC) experiments were conducted using modified histone tail peptides (Fig. 3a, b and Supplementary Table 1). These analyses revealed that the SAWADEE domain is quite unique in its ability to bind all three H3K9 methylation states (me1, me2, and me3) with very similar affinity, K_d ≈ 2 µM, which is approximately 17 fold stronger than observed using unmodified H3 peptides (Fig. 3a and Supplementary Table 1). ITC experiments also confirmed that while the SAWADEE domain will bind H3K9me2 peptides that contain H3K4me1 modifications, the presence of H3K4me2 or H3K4me3 modifications resulted in reduced binding affinity (Supplementary Table 1). Finally, ITC experiments using modified peptides corresponding to other known methylated lysine residues on the N-terminal tails of the core histone proteins confirmed the specificity of the SHH1 SAWADEE domain for H3K9 methylation (Fig. 3b and Supplementary Table 1).

Figure 3. The SHH1 SAWADEE domain recognizes H3K9 methylation and adopts a unique tandem Tudor domain-like fold.

a and b, ITC-based measurements of the SAWADEE domain binding to the modified or unmodified histone peptides as indicated. K_d values are listed. NDB means no detectable binding. c, The overall structure of the SHH1 SAWADEE domain in the free form. The zinc-binding motif is shown as an enlarged ball-and-stick model, highlighting the details of the metal coordination. A bound detergent molecule 4-Cyclohexyl-1-Butyl-β-D-Maltoside moiety from the crystallization condition is shown in a stick representation.

The anti-correlated effects of H3K9 and H3K4 methylation on SHH1 binding are reflective of genome profiling studies in Arabidopsis showing that the distribution of H3K9 methylation is anti-correlated with H3K4 methylation¹³. Consistent with these studies and the observed in vitro binding specificity of the SHH1 SAWADEE domain to H3K9 methylation, SHH1-dependent Pol-IV ChIP-seq peaks are enriched for H3K9me2 (Supplementary Fig. 5b) and depleted for H3K4 methylation (Supplementary Fig. 5c). Together, these binding studies demonstrate that the SAWADEE domain is a novel chromatin binding module that probes both the K4 and K9 positions of the H3 tail and specifically binds repressive H3K9 methyl-modifications.

To determine the mode of methyl-lysine recognition by the SHH1 SAWADEE domain, crystal structures of this domain either in the free-state or in complex with modified H3 tails were solved (Supplementary Tables 2, 3, Fig. 3c, and Supplementary Fig. 6a). In the free-state, the SHH1 SAWADEE domain adopts a tandem Tudor domain-like fold that contains a unique zinc-binding motif located within the Tudor 2 subdomain (Fig. 3c). The overall structure of the SAWADEE domain resembles the UHRF1 tandem Tudor domain with an r.m.s.d. of 2.3 Å (Supplementary Fig. 6b) despite only sharing 11.8% sequence identity (Supplementary Fig. 7)^14,15. This finding demonstrates that although the sequence of the SAWADEE domain is plant specific its fold is highly conserved in eukaryotic organisms.

The structures of the SHH1 SAWADEE domain in complexes with H3K9me1, H3K9me2 and H3K9me3 peptides were also solved and all three peptides were bound in a similar manner (Supplementary Table 3). Given the known role of the H3K9me2 modification in gene silencing genome wide in plants¹⁶, we focused on the 2.70 Å structure solved with an H3(1–15)K9me2 peptide (Fig. 4a and Supplementary Fig. 8a). This peptide binds in a groove between the two Tudor subdomains, forming contacts with both subdomains (Fig. 4a–b and Supplementary Fig. 8b,c). Interestingly, there is no significant conformational change in the SAWADEE domain upon ligand binding (Supplementary Fig. 9a), which differs from the situation for UHRF1¹⁵.

Figure 4. Structural basis for recognition of H3(1–15)K9me2 peptide by the SHH1 SAWADEE domain and the functional impact of mutations of residues lining the K4 and K9me2 pockets.

a, Overall structure of the H3(1–15)K9me2-SAWADEE complex with the SAWADEE domain as a ribbon diagram and the peptide as a stick representation. The simulated annealing composite omit map at 1σ level of the bound peptide is also shown. b, Stereo view highlighting the intermolecular interactions between the SAWADEE domain and the bound peptide. Intermolecular hydrogen-bonding interactions are designated by dashed red lines. c, d and e, Close-up views of H3 lysine residues in their respective binding pockets. f and g, Boxplots of genome-wide % CHH methylation and siRNA levels in wild-type, shh1 mutants, and shh1 mutants transformed with SHH1 constructs (shh1 + SHH1) that encode wild-type SHH1 or K9 (F162AF165A and Y140A) or K4 (D141A and Y212A) binding pocket mutants. h, qPCR of Pol-IV enrichment in the backgrounds described in f at a defined Pol IV binding site. Bars are the average of two biological replicates normalized to input and actin levels (+/− SE).

Within the SHH1 SAWADEE domain, there are two pockets that form key intermolecular interactions with the unmodified K4 and the K9me2 side chains of the bound peptide (Fig. 4c, d). The unmodified H3K4 side chain inserts into an interfacial pocket formed by residues from both Tudor subdomains. In this pocket, the K4 side chain is stabilized via intermolecular hydrogen bonds and electrostatic interactions with the side chains of Glu130 and Asp141 (Fig. 4c). The H3K9me2 side chain inserts into a hydrophobic aromatic cage in the Tudor 1 subdomain (Fig. 4d) where it is stabilized by cation-π interactions in a manner similar to those reported previously for methylated lysine-binding modules¹⁷. The SAWADEE complexes with H3K9me3 and H3K9me1 peptides also position the methylated lysines within the same aromatic cage (Supplementary Fig. 10). The ability of the SAWADEE domain to bind equally against all three H3K9 methylation states can be well explained by structural observations: The methylated lysine recognition aromatic cage can accommodate both H3K9me2 and H3K9me3 side chains through common hydrophobic interactions, resulting in a lack of discrimination between these two methylation states. In the H3K9me1 complex, although the lower lysine methylation state has a decreased hydrophobic interaction with the aromatic cage, the side chain of His169 undergoes a small but significant conformational change in order to hydrogen bond with the K9me1 ammonium proton, thereby contributing to the recovery of the binding affinity (Supplementary Fig. 10). This lack of specificity for the state of K9 methylation is in contrast to the higher level of methylation specificity observed for the tandem Tudor domain of UHRF1, which has a slightly wider aromatic cage binding pocket (Supplementary Fig. 9b). Thus our structural analysis indicates how very subtle changes in the tandem Tudor domain fold can result in a fine tuning of methyl-lysine specificity.

Consistent with our peptide binding studies (Supplementary Table 1), we were also able to solve a structure of the SAWADEE domain in a complex with an H3(1–15)K4me1K9me1 peptide (Supplementary Table 3). Overall, this structure resembles the structure with the H3K9me2 peptide, with the K4me1 accommodated within the same K4 binding pocket. However, the methyl group forms a stabilizing hydrophobic interaction with Leu201 in place of the hydrogen bond that is formed between the unmethylated K4 and the Glu130 side chain (Fig. 4e). Since this K4 binding pocket is relatively closed and narrow, higher methylation states of K4 would likely introduce steric conflicts and/or disrupt all the hydrogen bonding interactions, explaining the observed decreases in binding affinity (Supplementary Table 1).

To test the biological significance of methyl-H3K9 binding activity observed for the SHH1 SAWADEE domain, we generated point mutations within the two lysine binding pockets as well as the zinc binding motif and tested their affect on DNA methylation, siRNA levels, and Pol-IV recruitment in vivo. These point mutations were engineered into an SHH1-3xMyc-BLRP-construct and transformed into an shh1 mutant background. DNA methylation levels were assessed at a well characterized locus, MEA-ISR, by southern blotting (Supplementary Fig. 11a) and genome-wide by BS-seq experiments (Fig. 4f and Supplementary Fig. 11c–e). Addition of a wild-type SHH1-3xMyc-BLRP transgene restored DNA methylation, but constructs harboring mutations within the H3K9 or the H3K4 pockets were unable to fully complement the methylation defect observed in the shh1 mutant (Fig. 4f and Supplementary Fig. 11c–e) despite being expressed at levels comparable to the wild-type SHH1-3xMyc-BLRP protein (Supplementary Fig. 11a). In line with a canonical role for the zinc-binding motif in protein structure and/or stability, mutations in the zinc coordinating residues resulted in nearly undetectable levels of protein (Supplementary Fig. 11b) and thus were not characterized further.

Similar to the shh1 null mutant, the DNA methylation defects in the SHH1 lysine binding pocket mutants were most pronounced in the shh1/drm2/pol-v subclass of affected siRNA clusters (Fig. 4f and Supplementary Fig. 11c–e). Consistent with their positions and predicted contributions to the binding affinity of the SHH1 SAWADEE domain, the F162AF165A and the D141A mutants display stronger DNA methylation defects (Fig. 4f). Assessment of siRNA levels in these lysine binding pocket mutants via siRNA-seq experiments revealed a similar pattern of defects (Fig. 4g and Supplementary Fig. 11f). Finally, to determine whether the observed losses of siRNAs and DNA methylation reflect a defect in Pol-IV activity at chromatin, Pol-IV ChIP experiments were conducted in the SAWADEE domain point mutant backgrounds. All four point mutants displayed reduced levels of Pol-IV occupancy in two biological replicates (Fig. 4h). In addition, co-immunoprecipitation experiments revealed that the SAWADEE domain point mutants were still able to interact with Pol IV (Supplementary Fig. 11g), demonstrating the interaction between SHH1 with the Pol IV complex is not dependent on its H3K9me binding activity. Together, these findings show that residues within both the K4 and K9 binding pockets are critical for SHH1 function in vivo and demonstrate a central role for methyl-H3K9 binding by SHH1 at the level of Pol IV association with chromatin.

The finding that the H3K4 binding pocket is critical for SHH1 function in vivo was unexpected considering that the SHH1 SAWADEE does not bind H3K4 methylation in the absence of H3K9 methylation and that the addition of a methyl group to K4 does not impart additional binding affinity (Supplemental Table 1). One hypothesis to explain these in vivo findings is that the mere presence of a lysine at the position five residues back from the methylated H3K9 residue is necessary for SAWADEE domain binding. Indeed, such dual lysine reading could serve to help ensure that the SAWADEE domain only binds lysine methylation when it is present at the K9 position of the H3 tail as opposed to a methylated lysine at a different position on the H3 tail, especially the H3K27 position which has similar ARKS sequence context as H3K9 but a Thr22 at five residues back. To test this hypothesis, ITC experiments were conducted using H3 tails harboring an H3K4A mutation with or without the presence of the H3K9me2 modification. Indeed, the SAWADEE domain binds the H3K4AK9me2 peptide with approximately 30-fold weaker affinity than the H3K9me2 peptide (Supplemental Table 1). Furthermore, the SHH1 SAWADEE domain binds the H3K4A peptide with weaker affinity than the wild type H3 tail (Supplemental Table 1) demonstrating that the K4 residue is contributing to binding independent of the methylation status of the K9 residue.

Together, these in vivo and in vitro analyses demonstrate that the SHH1 SAWADEE domain is probing the H3 tail at both the K4 and K9 positions and is quite selective for the combination of histone modifications present at transposons and other repetitive DNA elements, namely unmodified H3K4 and methylated H3K9. Although H3K9 methylation is anti-correlated with H3K4 methylation genome-wide¹³, the aversion of the SAWADEE domain to higher order H3K4 methylation could serve to allow transcription, which is correlated with H3K4 methylation, to overcome DNA methylation and associated repressive H3K9 methyl modifications in a developmental or locus specific manner. Likewise, the specificity of the SAWADEE domain could inhibit siRNA generation at body methylated genes which contain CG methylation and H3K4 methyl-modifications, but lack CHG and CHH methylation as well as siRNAs^13,18,19.

In summary, we demonstrate that SHH1 is a novel chromatin binding protein that functions to enable Pol-IV recruitment and/or stability at the most actively targeted genomic loci in order to promote siRNA biogenesis. The finding that SHH1 binds to repressive histone modifications, together with the observation that SHH1 is required for Pol-IV chromatin association at a similar set of loci as downstream RdDM mutants, could explain the previously observed self-reinforcing loop in which downstream RdDM mutants are required for the production of full levels of siRNAs from a subset of genomic loci^20–23. Indeed, it has been shown that downstream RdDM mutants can cause a reduction of both DNA methylation and H3K9 methylation at RdDM loci²⁴, suggesting that the loss of siRNAs in these mutants may be due to the associated loss of the appropriate chromatin marks necessary for SHH1 binding.

METHODS

ChIP-seq, BS-seq and siRNA-seq library construction and sequencing

The first replicate of ChIP-seq libraries (NRPD1-Flag and Col) was generated using the Ovation Ultralow IL Multiplex System (NuGEN) while the second replicate (NRPD1-Flag, NRPD1-Flag ; shh1, and Col) was generated using the Ovation Ultralow DR Multiplex System (NuGEN). Both sets of ChIP-seq libraries used 18 cycles for the library amplification step. BS-seq libraries were generated as previously reported¹⁹. siRNA-seq libraries were generated using the small RNA TruSeq kit (Illumina) following the manufacturer instructions with the exception that 15 cycles were used during the amplification step. The wild-type (Col) and nrpe1 BS-seq libraries used in this study were previously published¹⁰ and were subsequently reanalyzed for this study. All libraries were sequenced using the HiSeq 2000 platform following manufacturer instructions (Illumina) at a length of 50bp. Read statistics are listed in Supplementary Table 4.

Mapping and processing of reads

Sequenced reads were base-called using the standard Illumina pipeline. For ChIP-seq and BS-seq libraries, only full 50 nt reads were retained, whereas for siRNA-seq libraries, reads had adapter sequence trimmed and were retained if they were between 18 nt and 28 nt in length. For ChIP-seq and siRNA-seq libraries, reads were mapped to the Arabidopsis genome (TAIR8 – www.arabidopsis.org) with Bowtie²⁵ and only perfect matches that mapped uniquely to the genome were retained for further analysis although the total number of mapping reads, unique and non-unique, were used when normalizing the siRNA-seq libraries to total number of reads per library. For BS-seq libraries, reads were mapped using the BSseeker wrapper for Bowtie²⁶. For ChIP-seq and BS-seq, identical reads were collapsed into one read, whereas for siRNA-seq identical reads were retained. For methylation analysis, percent methylation was calculated as previously reported¹⁹ with the unmethylated chloroplast genome serving as the measure of non-bisulfite converted background methylation. For the second replicate of ChIP-seq, there was a large disparity of resultant reads for the NRPD1-Flag and NRPD1-Flag ; shh1 libraries, so the NRPD1-Flag and Col libraries were sampled down to match the read total of the smaller library (the NRPD1-Flag ; shh1 library).

DNA methylation analysis

For assessment of DNA methylation at siRNA clusters, only those clusters with at least one cytosine in the respective class being assayed (CG, CHG, or CHH), were considered. For calculating significance levels of methylation change via the Mann-Whitney U test of methylation levels for clusters within the different subclasses (Fig. 1e) the number of clusters within each subclass was down sampled to the smallest subclass (the drm2/nrpe1 subclass) to allow for comparable significance values between subclasses.

Identification of siRNA clusters

Small RNA clusters (Supplementary Table 5) in the Arabidopsis genome were defined in a manner similar to a previously published approach²⁷. In brief, the genome was divided into 200 bp bins, and the average coverage per bin of non-identical siRNA reads was calculated in two technical replicates of our wild-type (Col) library. This average was used assay the significance of the number of non-identical reads at a given bin in wild-type plants, assuming a Poisson distribution of such counts. In the R environment a Poisson exact test was carried out for each bin, and bins with a P-value less than 1e-5 in each wild-type technical replicate were considered as clusters.

Once clusters were defined, comparisons between read counts, including identical reads, were carried out for each mutant and the wild-type (Col) library using a Fisher’s Exact Test. Resultant P-values were Benjamini-Hochberg adjusted to estimate FDRs, and clusters reduced in a mutant background at a FDR<1e-10 were then considered to be dependent on the wild-type function the mutant protein (Supplementary Table 5). For boxplot analysis of siRNA levels, the first technical replicate of the Col library was used as representative of Col siRNA levels. For calculating significance levels of siRNA change via the Mann-Whitney U test of siRNA levels for clusters within the different genotypic subclassess (Fig. 1d) the number of clusters within each subclass was down sampled to the smallest subclass (the drm2/nrpe1 subclass) to allow for comparable significance values between subclasses.

Identification of NRPD1 peaks

The R package BayesPeak^28,29 was used to identify regions of Pol-IV enrichment in a NRPD1-Flag ChIP-seq library as compared to a paired Col ChIP-seq control library done in parallel. Only high scoring peaks (PP>0.999) identified in both NRPD1-Flag ChIP-seq replicates (928 peaks) were retained for further analysis (Supplementary Table 6). For the purposes of assaying overlap of Pol IV peaks with siRNA clusters, “overlap” is called when >=1bp of a peak overlaps with a locus.

To classify peaks as SHH1-dependent, -independent, or –enhanced, read counts over Pol IV peaks were compared between the NRPD1-Flag and NRPD1-Flag ; shh1 ChIP-seq libraries, and significance was assessed using Fisher’s Exact Test. Resultant P-values were Benjamini-Hochberg adjusted to estimate FDRs. Peaks with a loss of NRPD1 signal in the shh1 library at a FDR<0.001 were considered SHH1-dependent. Similarly, peaks that gained signal in shh1 at a FDR<0.001 were considered SHH1-enhanced. Peaks that fell into neither of these categories were considered SHH1-independent.

Protein preparation

The gene encoding the SAWADEE domain of atSHH1 (residues 125–258) was cloned into a self-modified vector, which fuses a hexa-histidine tag plus a yeast sumo tag onto the N terminus of the target gene. The plasmid was transformed into the E. coli strain BL21 (DE3) RIL (Stratagene). The cells were cultured at 37 °C until the OD₆₀₀ reached 0.8 and then the media was cooled to 20 °C and 0.2 mM IPTG was added to induce protein expression overnight. The recombinant expressed protein was first purified using a HisTrap FF column (GE Healthcare). The hexa-histidine-sumo tag was cleavage by the Ulp1 protease and removed by passing through a second HisTrap FF column. The pooled target protein was further purified using a Q FastFlow column and a Hiload Superdex G200 16/60 column (GE Healthcare) with buffer (150 mM NaCl, 20 mM Tris pH 8.0, and 5 mM DTT). In order to prepare the Se-methionine substituted protein, Leu200 and Leu218 of the SAWADEE domain were mutated to methionine using a QuikChange Site Directed Mutagenesis Kit (Stratagene). The Se-methionine substituted SAWADEE protein was expressed in M9 medium supplemented with amino acids Lys, Thr, Phe, Leu, Ile, Val, Se-Met, and purified using the same protocol as the wild-type protein. Peptides were synthesized by the Tufts University peptide synthesis facility or by Krzysztof Krajewski.

Crystallization

Crystallization of the SAWADEE domain was conducted at 4 °C using the sitting drop vapor diffusion method by mixing 1 µl of protein sample at a concentration of 5 mg/ml and 1 µl of reservoir solution (0.2 M NH₄F and 20% PEG 3350), which was equilibrated against a 0.4 ml reservoir. 4-Cyclohexyl-1-Butyl-β-D-Maltoside (CYMAL®-4, Hampton Research) was added in the drop with a final concentration of 7.6 mM as an additive, which resulted in considerable improvement in crystal quality. Thin plate-shaped crystals appeared within 2 days. To generate crystals of complexes of SAWADEE domain with modified H3 peptides [H3(1–15)K9me3, H3(1–15)K9me2, H3(1–15)K9me1, and H3(1–15)K4me1K9me1], the SAWADEE domain was mixed with peptides at a molar ratio of 1:2 at 4 °C for 1 hour. The crystals of the different complexes were grown under the same conditions as described for free SAWADEE protein. All the crystals were soaked into a reservoir solution supplemented with 20% glycerol for 2 minutes. The crystals were then mounted on a nylon loop for diffraction data collection. The diffraction data from the native SAWADEE protein and its Se-methionine substituted counterpart were collected at the NE-CAT beamline 24ID-C, Advanced Photon Source (APS), Argonne National Laboratory, Chicago, at the zinc peak and selenium peak, respectively. The data of the complex of the H3K9me3 peptide bound to the SAWADEE domain were collected at beamline X29A, National Synchrotron Light Source (NSLS) at Brookhaven National Laboratory, New York. The data on the SAWADEE domain in complex with H3K9me2, H3K9me1 and H3K4me1K9me1 peptides were collected at APS 24ID-E. All the crystallographic data were processed with the HKL2000 program³⁰. The statistics of the diffraction data are summarized in Supplementary Tables 2 and 3.

Structure determination and refinement

The structure of the selenomethionine-substituted SAWADEE domain was solved using the single-wavelength anomalous dispersion (SAD) method as implemented in the Phenix program³¹. The model building was carried out using the Coot program³² and structural refinement using the Phenix program³¹. The structure of the wild type SAWADEE domain in the free state was solved using the molecular replacement method using the Phenix program³¹. Zn²⁺ ions were identified and further confirmed by anomalous signal scattering. All the structures of SAWADEE domain in complexes with different modified H3 peptides were solved using the molecular replacement method with the same protocol as the native protein. The peptides showed clear electron density and were properly built with residues from Thr3 to Ser10 for H3(1–15)K9me3/2/1 and from Thr3 to Thr11 for H3(1–15)K4me1K9me1. Throughout the refinement, a free R factor was calculated using 5% random chosen reflections. The stereochemistry of the structural models were analyzed using the Procheck program³³. The refinement and structure statistics are shown in Supplementary Tables 2 and 3. All the molecular graphics were generated with the Pymol program (DeLano Scientific LLC).

Isothermal titration calorimetry

The protein samples were not stable at room temperature. Thus, all the binding experiments were performed on a Microcal calorimeter ITC 200 instrument at 6 °C. First, protein samples were dialyzed overnight against a buffer of 100 mM NaCl, 2 mM β-mercaptoethanol and 20 mM HEPES, pH 7.5, at 4 °C. Then the protein samples were diluted and the lyophilized peptides were dissolved with the same buffer. The titration was performed according to standard protocol and the data were fit using the Origin 7.0 program with a 1:1 binding model. Themodynamic parameters for complex formation are listed in Supplementary Table 1.

Modified peptide array binding

A GST-SHH1 SAWADEE domain (125–258aa) construct was generated in the pENTR/TEV/D plasmid (Invitrogen), recombined into the pDEST 15 plasmid (Invitrogen) and transformed into the Rosetta 2 (DE3) bacterial cell line (Novagen). Protein expression was induced by the addition of 500 µL of 1M IPTG per 500 mL at an OD of 0.6 and cultures were grown at 16 °C overnight. At the time of induction the media was supplemented with 500 µL of 500 mM ZnSO₄. The GST fusion protein was then purified as described in Johnson et. al³⁴ and dialyzed into storage buffer (50 mM Tris pH 6.8, 300 mM NaCl, 40% glycerol, 2mM DTT, 0.1% triton X-100). The purified GST-SHH1 (125–258aa) protein was used to probe a MODified™ Histone Peptide Array (Active Motif) under the following conditions: The array was blocked at 25 °C for 45 min in a 5% milk 1× TBS solution, washed three times in a 1× TBS-T solution at 25 °C for 5 minutes, and then probed overnight at 4 °C with the GST-SHH1 SAWADEE domain protein at a concentration of 6.5 µg/mL in Binding Buffer (50 mM HEPES pH7.5, 50 mM NaCl, 5% glycerol, 0.4 mg/mL BSA, 2 mM DTT). The array was then washed three times as above, and probed an HRP conjugated GST antibody at a 1:5000 dilution at 25 °C for 1 hour. The array then washed as detailed above and developed using an ECL Plus kit (GE healthcare).

Plant lines, site-directed mutagenesis, southern and western blotting

The various previously characterized Arabidopsis RdDM mutant alleles, the complementing SHH1-3xMyc-BLRP transgenic plant line, and the pSHH1::SHH1-3xMyc-BLRP construct used are as described in Law et. al³. The pol-iv and pol-v mutants correspond to mutations in the nrpd1 and nrpe1 subunits of these polymerases, respectively. The structure-based mutations were generated in the pSHH1::SHH1-3xMyc-BLRP construct using a QuikChange Site Directed Mutagenesis Kit (Stratagene) and were transformed into the shh1-1 mutant background via the floral dip method. siRNA-seq and ChIP-seq experiments in the Col and RdDM mutant lines were conducted using floral tissue and BS-seq experiments were conducted using 10 day old seedlings. Southern and western blotting experiments were conducted using tissue from the same individual plant lines in the T₁ generation and using previously described probes³⁴ and antibodies³⁵. The siRNA-seq and BS-seq experiments in the SAWADEE domain point mutant lines were conducted using floral tissue or 10 day old seedlings, respectively, from T₃ plants homozygous for the various pSHH1::SHH1-3xMyc-BLRP transgenes. The Pol IV ChIP experiments and co-immunoprecipition experiments in the various SAWADEE domain point mutant backgrounds were conducted using floral tissue from F₁ plants that were homozygous for the shh1 mutant allele.