MainCytosine DNA methylation and histone modifications are two deeply conserved epigenetic marks that together maintain genome stability and regulate gene expression in eukaryotes. These marks are critical for silencing transposable elements and controlling gene activity, thereby shaping development, physiology and genome integrity in a wide range of organisms1,2,3. In plants, DNA methylation occurs in three different sequence contexts: CG, CHG or CHH (where H is A, T or C) and is established primarily through the plant-specific RNA-directed DNA methylation (RdDM) pathway4,5. This process begins with RNA polymerase IV (Pol IV)-mediated transcription of 25–40-nucleotide small interfering RNA (siRNA) precursors6,7,8. The recruitment of Pol IV is orchestrated by the CLASSY (CLSY) chromatin remodelling factors, CLSY1–4, which act in a tissue- and locus-specific manner to regulate 24-nt siRNA production9,10,11,12,13,14,15.In leaves, CLSY1 and CLSY2 dominate siRNA biogenesis9, while in ovules, CLSY3 and CLSY4 are highly expressed and direct Pol IV complexes to generate siRNAs at specific loci called siren (small-interfering RNA in the endosperm) loci11. These siren-derived siRNAs promote DNA methylation in cis at their originating loci and also in trans at certain protein-coding genes, modifying gene expression and influencing reproductive processes such as hybrid seed failure in Capsella and seed abortion in Brassica rapa12,16,17. A similar mechanism is observed in male germ lines, where the CLSY3 CLSY4-dependent hypermethylated transposable element (hyperTE) loci in tapetal nurse cells produce siRNAs that move into microspore mother cells, directing DNA methylation and gene regulation10,12.Repressive chromatin states are stabilized through self-reinforcing loops between DNA methylation and histone modifications, particularly H3K9 methylation18,19. For instance, the SU(VAR)3-9 HOMOLOG 4 (SUVH4), SUVH5 and SUVH6 methyltransferases bind methylated DNA and deposit H3K9 methylation in heterochromatin20,21,22,23,24. Conversely, CHROMOMETHYLTRANSFERASE 2 (CMT2) and CMT3 read H3K9 methylation to add CHH and CHG methylation to nearby DNA, reinforcing repressive chromatin25,26,27,28. In addition, the CLSY1 and CLSY2 remodellers facilitate the interaction between H3K9 methylation readers, such as SAWADEE DOMAIN HOMOLOG 1 (SHH1), and Pol IV, ensuring efficient siRNA production and DNA methylation29,30,31. The chromatin remodeller ZINC FINGER, MOUSE DOUBLE-MINUTE/SWITCHING COMPLEX B, PLUS-3 Protein (ZMP) supports siRNA biogenesis independently of SHH1 by confining Pol IV to chromatin regions devoid of H3K4me, preventing interference with actively expressed genes32. Together, these interconnected layers have supported a model in which epigenetic patterns are shaped predominantly by chromatin-based features, rather than by the underlying DNA sequence itself.However, recent observations suggest that this dogma may be incomplete. At siren loci, which rely on CLSY3 and CLSY4, siRNA production and DNA methylation occur independently of H3K9 methylation9,11. Instead, these processes are associated with a conserved DNA motif, indicating that sequence-specific cues can shape the epigenetic landscape11. Here, we report that reproductive meristem (REM) transcription factors recognize defined DNA sequences and, with the assistance of GENETICS DETERMINES EPIGENETICS1 (GDE1), recruit the Pol IV complex to produce siRNAs and direct DNA methylation in Arabidopsis ovule tissue. Our findings establish a direct link between genetic elements and epigenetic regulation, expanding our understanding of how DNA methylation patterns are determined.ResultsGDE1 encodes an uncharacterized protein at RdDM sitesTo identify additional players in the RdDM pathway, we used crosslinked immunoprecipitation–mass spectrometry (IP-MS) with the RdDM-associated protein MORC733,34, and discovered a previously uncharacterized protein, AT1G77270, hereafter named GDE1. To determine the genomic localization of GDE1, we performed chromatin immunoprecipitation sequencing (ChIP-seq) with flower tissues from pGDE1::GDE1-3FLAG (GDE1–3FLAG). GDE1 largely co-localized with MORC7 across the genome (Extended Data Fig. 1a,b), as well as with key components of the RdDM pathway, including Pol IV and Pol V (Fig. 1a–c and Extended Data Fig. 1b). Genome-wide correlation analysis indicates that GDE1 is more closely associated with the Pol IV arm than the Pol V arm of the RdDM pathway (Extended Data Fig. 1b). These results suggest that GDE1 is involved in RdDM function, with a stronger connection to the Pol IV arm.Fig. 1: GDE1 is a RdDM protein and co-localizes with Pol IV recruiter CLSY3/4.a,b, Metaplot and heatmap showing enrichment of GDE1 ChIP-seq signal over Pol IV (n = 5,077) (a) and Pol V (n = 16,327) (b). c, A screenshot of MORC7, Pol IV, Pol V and GDE1 ChIP-seq at a representative locus. Square brackets indicate the range on a bar graph. d, TomTom analysis showing the similarities of CLSY3 and GDE1 binding motifs. Three repeats are boxed in red. E value is the expected number of false positives in the matches up to this point. e,f, Metaplot and heatmap showing enrichment of GDE1 ChIP-seq signal over CLSY3 (n = 380) (e) and CLSY4 (n = 1,368) (f).Source dataFull size imageMotif enrichment analysis of GDE1 ChIP peaks with a fold enrichment greater than 5 revealed the presence of a highly conserved motif (hereafter named CLSY3 CLSY4 motif 1), which the RdDM component CLSY3 was previously shown to favour as well11 (Fig. 1d). We found that CLSY4 ChIP signals exhibited strong enrichment at CLSY3 binding sites, with over 92% of CLSY3 peaks overlapping with CLSY4 peaks (Extended Data Fig. 1c,d). Furthermore, GDE1 was highly enriched at both CLSY3 and CLSY4 binding sites (Fig. 1e,f). Similar to the unique expression pattern of CLSY3 and CLSY411, GDE1 was highly expressed in flower tissues (Extended Data Fig. 1e) based on the ePlant database35. Taken together, these results suggest that GDE1 might collaborate with CLSY3 and CLSY4 (CLSY3/4).GDE1 directs CLSY3/4-Pol IV to modulate siRNA productionCLSY3/4 have been reported to be required for Pol IV activity and siRNA generation in ovules9,11. Our analysis identified 753 siRNA sites that were reduced in the clsy3 clsy4 double mutant in ovule tissues (Fig. 2a). To determine whether GDE1 affect siRNA biogenesis, the levels of siRNA in gde1-1 (SALKseq_10069.1) ovule tissue were profiled genome wide. Strikingly, 57% of the CLSY3 CLSY4-dependent siRNA loci showed reduced siRNAs in gde1-1, while 35% exhibited increased siRNAs in gde1-1 (Fig. 2a–c and Extended Data Fig. 2a), suggestive of the overlapping and differential siRNA modulation function of GDE1 with CLSY3 and CLSY4. Within these CLSY3/4-dependent siRNA loci, we defined GDE1-dependent loci as group 1 sites, while loci showing increased siRNA in gde1-1 were defined as group 2 sites and the rest of the loci were defined as group 3 sites (Fig. 2a).Fig. 2: GDE1 localizes at a subset of CLSY3 CLSY4-dependent siRNA loci for siRNA production and the loss of GDE1 redistributes the CLSY3/4–Pol IV complex to other loci.a, A pie chart showing the percentage of three groups of CLSY3 CLSY4-dependent siRNA sites. b,c, Violin plots and heatmaps showing 24-nt siRNA levels in group 1 (b) (n = 429) and group 2 (c) (n = 262) of CLSY3 CLSY4-dependent sites. Pairwise t-tests showed P = 4.4805 × 10−123 (Col-0 versus gde1-1) and P = 1.4105 × 10−114 (Col-0 versus clsy3 clsy4) for group 1, and P = 2.71776 × 10−52 (Col-0 versus gde1-1) and P = 2.81529 × 10−48 (Col-0 versus clsy3 clsy4) for group 2. The line in the centre of each violin plot represents the median. The thick grey bar in the centre represents the interquartile range. The whiskers represent the rest of the distribution. d, Metaplots showing enrichment of CLSY3/4 and Pol IV ChIP-seq signal over three groups of CLSY3 CLSY4-dependent sites in WT and gde1-1 backgrounds. The region is −2 kb to 2 kb from the centre. Same regions in gde1-1 were used for all metaplots. e, A metaplot showing enrichment of GDE1 ChIP-seq signal over three groups of CLSY3 CLSY4-dependent sites. f, A circular genome view showing the enrichment of groups 1 and 2 of CLSY3 CLSY4-dependent siRNA across all five chromosomes, with the pericentromeric heterochromatin marked in red along the inner circle.Full size imageTo test whether the effect of siRNA production in gde1-1 mutant relies on CLSY3/4 and Pol IV distribution, ChIP-seq of CLSY3/4 and Pol IV in Col-0 (wild type, WT) and gde1-1 mutant was performed. CLSY3/4 and Pol IV displayed strong enrichment at group 1 in WT, but this enrichment was largely lost in the gde1-1 background (Fig. 2d), mirroring the changes in siRNA abundance. Consistent with GDE1 playing a role in recruiting siRNA biogenesis to Group 1 sites, we found that GDE1 was strongly localized to group 1 sites (Fig. 2e), and GDE1–3FLAG fully complemented the gde1-1 siRNA phenotype (Fig. 2b,c). By contrast, CLSY3/4 and Pol IV were barely enriched at Group 2 in WT, but more ChIP signals were observed in the gde1-1 background (Fig. 2d and Extended Data Fig. 2a). These findings suggest that GDE1 plays a pivotal role in recruiting the CLSY3/4–Pol IV complex to generate siRNAs at a subset of CLSY3 CLSY4-dependent siRNA loci and that, in its absence, these complexes redistribute to other loci. Interestingly, the majority of group 2 sites were found in heterochromatic regions, while group 1 sites were located predominantly in euchromatic regions (Fig. 2f). This suggests that GDE1 primarily regulates siRNA production in euchromatic regions and competes with Pol IV recruitment to heterochromatic regions.GDE1 colocalizes with REM transcription factorsTo understand how GDE1 influences siRNA production, we performed IP–MS with GDE1–3FLAG transgenic plants to uncover interacting proteins. Unexpectedly, none of the CLSY3/4–Pol IV components was identified in the IP–MS experiments (Fig. 3a and Supplementary Table 1). To further assess potential interactions using a more sensitive approach, we conducted co-immunoprecipitation (co-IP) assays using F2 transgenic plants expressing GDE1–3FLAG with CLSY3–9myc or Pol IV–9myc. GDE1 proteins were scarcely detectable in input samples, probably due to their restricted expression in flower tissues. However, in the immunoprecipitated samples, GDE1 successfully pulled down CLSY3 and Pol IV (Extended Data Fig. 2b), suggesting that GDE1 forms a complex with CLSY3 and Pol IV.Fig. 3: REM transcription factors associate with GDE1 and Pol IV complex at CLSY3 CLSY4-dependent siRNA loci.a, A volcano plot showing proteins that have significant interactions with GDE1 as detected by IP–MS, with REM type transcription factors labelled with green letters. The two-sided empirical Bayes test performed by LIMMA was used for statistical analysis. b, An upset plot displaying the comparative analysis among all tested transcription factors bound to CLSY3 CLSY4-dependent siRNA loci. VDD, VAL, REM13 and REM19 common targets are labelled with orange, REM8 unique target is labelled with green and REM22 unique target is labelled with magenta. c, A metaplot showing enrichment of VDD, VAL, REM13, REM19, REM22 and REM8 ChIP-seq signal over three groups of CLSY3 CLSY4-dependent sites. The region is −2 kb to 2 kb from the centre. Same regions were used for all metaplots. d, A heatmap showing normalized transcription factor ChIP-seq intensity at CLSY3 CLSY4-dependent siRNA loci. e, A screenshot of VDD, VAL, REM13, REM19, REM22, REM8, GDE1, CLSY3, CLSY4 and Pol IV ChIP-seq at a representative group 1 of CLSY3 CLSY4-dependent siRNA sites. Square brackets indicate the range on a bar graph. f, A volcano plot showing proteins that have significant interactions with VDD as detected by IP–MS, with REM type transcription factors labelled with green letters, GDE1 labelled with cyan letters and CLSY3, RDM4 and Pol IV subunits labelled with purple letters. The two-sided empirical Bayes test performed by LIMMA was used for statistical analysis. g, A Venn diagram showing the similarities between VDD and VAL ChIP-seq common targeted siRNA loci with REM12 DAP-seq targeted siRNA loci. h, A bar plot showing the percentage of each REM transcription factor binding to three distinct groups of CLSY3 CLSY4-dependent siRNA loci. i, A pie chart displaying the percentage of transcription factors binding to CLSY3 CLSY4-dependent siRNA loci.Full size imageInterestingly, multiple members from the REM transcription factor family were pulled down by GDE1 in the IP–MS assay (Fig. 3a and Supplementary Table 1). To examine whether REM transcription factors co-localize with GDE1–CLSY3–Pol IV complexes, we performed ChIP-seq with 9myc-tagged transgenic lines. Two of these REM proteins, VDD and VAL, showed strong ChIP signals over the same set of CLSY3 CLSY4-dependent siRNA loci in group 1 (Fig. 3b–e and Extended Data Fig. 3a). In addition, a similar DNA motif with CLSY3 CLSY4 motif 1 was found in VDD ChIP-seq data (Extended Data Fig. 3b). Another factor, REM22, was enriched at a distinct subset of CLSY3 CLSY4-dependent siRNA loci within group 1, but also showed some colocalization with VDD and VAL (Fig. 3b–e and Extended Data Fig. 3a,c).To further explore transcription factor–GDE1 complexes, IP–MS was performed by using VDD–3FLAG transgenic lines. VDD successfully pulled down GDE1, CLSY3, Pol IV subunits36, RDR2 and RDM4, an IWR-type transcription factor known to interact with Pol IV1 (Fig. 3f and Extended Data Fig. 2b), indicating that VDD forms a complex with GDE1 and Pol IV complex components. In addition, we detected VAL, REM46 and additional members of the REM transcription factor family (Fig. 3f), suggestive of the in vivo interactions among these transcription factors family. Consistent with this observation, VAL was previously reported to interact with itself and VDD by yeast two-hybrid and a bimolecular fluorescence complementation assay37. In addition, most of these REM transcription factors showed expression patterns similar to those of CLSY3/4 and GDE1, exhibiting high expression in ovules in comparison with other tissues (Extended Data Fig. 3d). Taken together, these results suggest that multiple REM transcription factors may be involved in the recognition of the CLSY3 CLSY4-dependent siRNA loci and their association with GDE1 probably recruits CLSY3/4–Pol IV to generate siRNAs for locus-specific DNA methylation.To search for additional REM transcription factors that might recognize the CLSY3 CLSY4-dependent siRNA loci, we performed ChIP-seq using 9myc-tagged REM transcription factors identified in the IP–MS data (Fig. 3a,f and Supplementary Tables 1 and 2) and/or co-expressed with GDE1 by ATTED-II RNA coexpression analysis38 and/or highly expressed in ovules (Extended Data Fig. 3d). We found that REM13 exhibited a localization and motif pattern similar to VDD and VAL, demonstrating strong signals over the same set of CLSY3 CLSY4-dependent siRNA loci in group 1 (Fig. 3b–e and Extended Data Fig. 3a,b). REM12 was identified from VDD–3FLAG IP–MS and highly expressed in ovule; however, we failed to observe ChIP signal using transgenic lines. Nevertheless, publicly available DNA affinity purification sequencing (DAP-seq) data39 showed that the binding sites of REM12 highly overlapped with the same set of CLSY3 CLSY4-dependent siRNA loci bound by VDD and VAL (Fig. 3g). In addition, a similar motif enrichment was found in REM12 DAP-seq as the CLSY3 CLSY4 motif 1 (Extended Data Fig. 3b). For REM19, 80% of its ChIP-seq signals were enriched at group 1 CLSY3 CLSY4-dependent siRNA loci, with a relatively lower intensity, while 20% extended to group 2 regions (Fig. 3b–e,h and Extended Data Fig. 3a). By contrast, 86% of REM8 ChIP-seq peaks localized to group 2 of CLSY3 CLSY4-dependent siRNA loci instead, showing a preference for gde1-1 upregulated siRNA regions (Fig. 3b–d,h and Extended Data Fig. 3a,e).In summary, REM transcription factors VDD, VAL, REM12, REM13 and REM19, along with REM22 and REM8 to a lesser extent, were enriched at group 1 CLSY3 CLSY4-dependent siRNA loci (Fig. 3b–e,g,h and Extended Data Fig. 3a,c,e), suggesting functional collaboration in targeting these loci. Meanwhile, REM8, together with REM19, localized to a subset of group 2 of CLSY3 CLSY4-dependent siRNA loci (Fig. 3b–d,h and Extended Data Fig. 3a,e), where the CLSY3/4–Pol IV complex became enriched in the absence of GDE1. Collectively, the transcription factors described here can bind over 28% of CLSY3 CLSY4-dependent siRNA loci (Fig. 3i).GDE1 controls siRNA production and DNA methylation at siren sitesCLSY3/4 are required for siRNA production at siren sites, the predominate siRNA loci in female reproductive tissues (ovules)11. Notably, 86 out of 133 siren loci contained the same binding motif as CLSY3 CLSY4 motif 1 (Extended Data Fig. 4a), which was also exhibited by the VDD, REM12 and REM13 (Extended Data Fig. 3b). These transcription factors were enriched at more than 77% of the siren loci (Fig. 4a and Extended Data Fig. 4b), and the strongly bound subset of group 1 sites primarily corresponds to the siren loci (Fig. 4b). Moreover, REM19, REM22 and REM8 were found at siren loci with somewhat lower enrichment (Extended Data Fig. 4b). Furthermore, GDE1 and the Pol IV recruiters, CLSY3 and CLSY4, displayed substantial enrichment at siren loci (Fig. 4b and Extended Data Fig. 4c). These results suggest that REM–GDE1 complexes may recruit CLSY3/4 to siren loci to regulate siRNA levels.Fig. 4: VDD/VAL/REM13–GDE1–CLSY3/4 complex localizes to siren loci for siRNA production.a, A Venn diagram showing the relationship between siren loci and VDD, VAL and REM13 ChIP-seq targets. b, Top: metaplots showing enrichment of VDD, VAL and REM13 ChIP-seq signal over siren and group 1 non-siren region. Bottom: metaplots showing enrichment of GDE1, and CLSY3/4 ChIP-seq signal over siren loci. The region is −2 kb to 2 kb from the centre. Same regions were used for all metaplots. c, A violin plot showing the 24-nt siRNA levels at siren loci (n = 133) in the indicated genotypes. Pairwise t-tests showed P = 2.64555 × 10−80 (Col-0 vs gde1-1) and P = 9.98545 × 10−70 (Col-0 versus clsy3 clsy4). The line in the centre of each violin plot represents the median. The thick black bar in the centre represents the interquartile range. The whiskers represent the rest of the distribution. d, A violin plot showing the CHH and CHG methylation levels at siren loci (n = 133) in the indicated genotypes. P values calculated by pairwise t-tests are indicated. Pairwise t-tests showed P = 7.02463 × 10−29 (Col-0 versus gde1-1) and P = 2.52025 × 10−20 (Col-0 versus clsy3 clsy4) for CHH methylation. The line in the centre of each violin plot represents the median. The thick black bar in the centre represents the interquartile range. The whiskers represent the rest of the distribution. e, A violin plot showing the 24-nt siRNA levels at siren loci (n = 133) in the indicated genotypes. P values calculated by pairwise t-tests are indicated. Pairwise t-tests showed P = 2.41619 × 10−56 (Col-0ov versus rem46 val rem12 triple 1 ov) and P = 3.18267 × 10−22 (Col-0 ov versus rem46 val rem12 triple 2 ov). The line in the centre of each violin plot represents the median. The thick black bar in the centre represents the interquartile range. The whiskers represent the rest of the distribution. f, A screenshot of 24-nt siRNA levels at a representative siren site. Square brackets indicate the range on a bar graph. g, A Venn diagram showing the relationship of downregulated siRNA loci in each background with siren loci. h, A violin plot showing the 24-nt siRNA levels at de novo ZF fusion targeted sites (n = 58) in the indicated genotypes. P values calculated by pairwise t-tests are indicated. The line in the centre of each violin plot represents the median. The thick black bar in the centre represents the interquartile range. The whiskers represent the rest of the distribution. i, A screenshot of 24-nt siRNA and VDD–ZF ChIP-seq at a de novo ZF fusion loci. Square brackets indicate the range on a bar graph. j, Metaplot and heatmap showing enrichment of VDD ChIP-seq signal over siren loci in Col-0, gde1-1 and clsy3 clsy4 double backgrounds. The region is −2 kb to 2 kb from the centre. Same regions were used for all metaplots.Full size imageTo test the function of GDE1 at siren loci, 24-nt siRNA were measured and we found that, in gde1-1 ovules, 24-nt siRNA exhibited a strong reduction, phenocopying the clsy3 clsy4 double mutant (Fig. 4c and Extended Data Fig. 4d). This dramatic loss of 24-nt siRNA is probably due to the significant loss of CLSY3/4 and Pol IV at siren loci in gde1-1, which did not occur at non-siren loci (Extended Data Fig. 4e–g). Consistent with the fact that CHH and CHG methylation levels are affected in siRNA biogenesis mutants40, we found that these types of methylation were significantly reduced at siren loci in gde1-1 ovules, resembling the pattern observed in the clsy3 clsy4 double mutant (Fig. 4d and Extended Data Fig. 4h). Taken together, these results indicate that GDE1 is required for CLSY3/4 and Pol IV recruitment, siRNA production and DNA methylation at siren loci in ovules.REM transcription factors promote the biogenesis of siRNAVDD has previously been implicated in controlling the death of the receptive synergid cell during fertilization37. Loss-of-function mutants exhibit strong female gametophytic defects37, and VDD–9myc fully restored the lethality phenotype (Extended Data Fig. 5a). The genetic lethality of knockout mutants and the multitude of these REM transcription factors make mutant analysis challenging. VAL, REM12 and a VDD interactor, REM46, form a gene cluster in the genome. To create partial loss-of-function mutants of the siren targeting REM transcription factors, we utilized the CRISPR–Cas9 editing system to simultaneously delete all three genes (Extended Data Fig. 5b). The 24-nt siRNA levels at siren loci were significantly reduced in two independent rem46 val rem12 triple mutants (Fig. 4e,f), indicating the involvement of these REM transcription factors in regulating siRNA levels at siren loci. Furthermore, the siRNA level at siren loci in rem46 val rem12 gde1-1 or rem46 val rem12 clsy3 clsy4 mutants resembled those of their respective parents gde1-1 or clsy3 clsy4, further suggesting that REM transcription factors function in the same pathway with GDE1 and CLSY3/4 at siren loci (Fig. 4e). In addition, around 30% of CLSY3 CLSY4-dependent siRNA loci showed reduced siRNA levels in the rem46 val rem12 mutants, although not to the same extent as in the clsy3 clsy4 mutant (Fig. 4g and Extended Data Fig. 5c). All of these reduced siRNA in the rem46 val rem12 mutants belonged to group 1 of CLSY3 CLSY4-dependent sites, implying broader effects of these REM transcription factors on regulating siRNA production than those revealed by ChIP-seq.We also utilized a gain-of-function approach by fusing the REM transcription factor VDD with the artificial zinc finger 108 (ZF), which target hundreds of binding sites throughout the Arabidopsis genome41,42. A VDD–ZF fusion was transformed into the WT background, and ChIP-seq was performed to identify the binding sites. We identified 397 clear peaks with ZF binding to the genome which were highly enriched for the known ZF binding motif (Extended Data Fig. 5d). Small RNA sequencing analysis of the VDD–ZF transgenic plants revealed that 24-nt siRNAs were produced de novo at many of the ZF fusion targeted regions in ovule tissues (Fig. 4h–i), indicating the sufficiency of REM transcription factors to promote siRNA biogenesis.To investigate whether the localization of REM transcription factors was affected by GDE1 or CLSY3/4. We performed ChIP-seq using flower tissues collected from VDD–9myc gde1-1 and VDD–3FLAG clsy3 clsy4 and found that the ChIP-seq signals of VDD were reduced significantly at siren loci in both gde1-1 and clsy3 clsy4 mutants (Fig. 4j). These results suggest that GDE1 and CLSY3/4 are both involved in the stabilization of REM transcription factors to siren loci.REM factors drive siRNA biogenesis independently of H3K9meH3K9 methylation has been reported to recruit Pol IV for siRNA biogenesis9,30. To test the relationship between REM transcription factors and H3K9 methylation-initiated RdDM pathways, siRNA levels were measured in ovule tissues of suvh4 suvh5 suvh6 triple mutant, where H3K9 methylation is largely lost27. H3K9 methylation-dependent siRNAs largely overlapped with CLSY1 CLSY2-dependent loci, while there was little overlap with REM transcription factors- dependent or CLSY3 CLSY4-dependent loci (Extended Data Fig. 6a,b). Furthermore, no significant reduction at siren loci was observed in either clsy1 clsy2 double or suvh4 suvh5 suvh6 triple mutants (Extended Data Fig. 6c), suggestive of independence of REM transcription factor-initiated siRNA production with H3K9 methylation. Collectively, in ovule tissue, 20.25% of Pol IV-dependent 24nt-siRNAs reside in clusters reduced in the clsy3 clsy4 double mutant and REM transcription factor-GDE1 recruitment appears to be the dominant mechanism operating at these sites (Extended Data Fig. 6d). By contrast, 9.68% depended on CLSY1 CLSY2, where H3K9 methylation plays a crucial role (Extended Data Fig. 6d).Essential role of the RBHG domain in GDE1 functionAll transcription factors (VDD, VAL, REM12 and REM13) capable of recognizing the CLSY3 CLSY4 motif 1 feature two B3 domains, located at both ends of the proteins (Extended Data Fig. 7a). AlphaFold343 confidently predicted dimerization of these factors through their N-terminal B3 domains, while their C-terminal B3 domains were predicted to be responsible for DNA recognition across almost all combinations (Fig. 5a,b and Extended Data Fig. 7a). Notably, an α-helix of GDE1 (amino acids 249–261, hereafter named RBHG (REM-binding helix of GDE1) domain) was predicted to fit into a pocket formed by the transcription factor dimers, establishing extensive electrostatic and hydrophobic interactions (Fig. 5a–c and Extended Data Fig. 7a). To experimentally assess the significance of the α-helix of GDE1’s predicted association with REM transcription factor dimers, a triple mutant (E251A/Y252A/Y255A) was generated and introduced into the gde1-1 mutant. Even though this mutant produced GDE1 transcript levels comparable to the WT GDE1, it did not fully rescue the siRNA deficiency of the gde1-1 mutant (Fig. 5d and Extended Data Fig. 7b). These results suggest that this α-helix of GDE1 is critical for function, consistent with its proposed role in interacting with REM transcription factors.Fig. 5: The RBHG domain of GDE1 is critical for siRNA production.a, A table showing the residues and types of interaction formed between RBHG domain of GDE1 and transcription factor dimers. AA, amino acid. b, AlphaFold3 predicted the structure of VDD–VAL–GDE1 bound to dsDNA containing TTTTGCTTATGTTTTGCTTAT (sequence below, high-affinity nucleotide acids red bolded). The DNA is shown as a ribbon representation. c, Interaction of the RBHG domain of GDE1 with VDD–VAL. Interacting residues are highlighted with sticks. Hydrogen bonds and pi bonds are highlighted with dashed lines. d, A violin plot showing the 24-nt siRNA levels at siren loci (n = 133) in the indicated genotypes. P values calculated by pairwise t-tests are indicated. Pairwise t-tests showed P = 2.6169 × 10−24 (WTGDE1 in gde1-1ov versus gde1-1ov), P = 9.5561 × 10−17 (WTGDE1 in gde1-1ov versus GDE1 E251A Y252A y255A in gde1-1ov), and P = 1.39057 × 10−19 (Col-0ov versus gde1-1ov). The line in the centre of each violin plot represents the median. The thick black bar in the centre represents the interquartile range. The whiskers represent the rest of the distribution.Full size imageREM–GDE1 recruit CLSY3/4–Pol IV for unidirectional transcriptionThe CLSY3 CLSY4 motif 1 exhibits two or three tandem TTTTGCTTAT sequences with a single nucleotide spaced between them (Fig. 1d and Extended Data Fig. 3b). The REM proteins belong to the B3 DNA-binding domain superfamily, and some family members, the auxin response factors, possess the ability to form dimers that recognize sequences with a spacer of a specific length44,45. Throughout the genome, the occurrence of the three repeats (n = 38) and two repeats (n = 48) is rare, with 37,456 loci having a single repeat. ChIP-seq signals for all components (VDD, VAL, REM13, GDE1, CLSY3, CLSY4 and Pol IV) showed strong enrichment at all triple-repeat sites and most double-repeat sites, but very minor enrichment at the single-repeat sites (Fig. 6a and Extended Data Fig. 8a). These findings suggest that at least double repeats are required for robust recruitment of the REM transcription factors–GDE1–CLSY3/4–Pol IV transcription complex.Fig. 6: Pol IV transcription complex recognizes CLSY3 CLSY4 motif 1 and engages in a unidirectional transcription.a, Metaplot and heatmap showing enrichment of VDD, VAL, REM13, GDE1, CLSY3, CLSY4 and Pol IV ChIP-seq signal over triple-repeat sites (n = 38) and double-repeat sites (n = 48). b, Sequence logo assessment of DNA sequences identified by VDD–VAL–GDE1 DAP-seq. c, The position of VDD, VAL, REM13, GDE1, CLSY3, CLSY4 and Pol IV ChIP-seq summits to the closest CLSY3 CLSY4 motif 1 centre. d, Metaplot and heatmap showing enrichment of CLSY3, CLSY4, Pol IV ChIP-seq, MNase-seq signal and 24-nt siRNA signal over Pol IV left shift (n = 23) or Pol IV right shift (n = 27). The region is −1 kb to 1 kb from the centre. Same regions were used for all metaplots.Full size imageIt was notable that there was always a one-nucleotide space between repeats with strong binding, leading us to speculate that this space might allow proper spatial dimerization of transcription factors. To examine this further, we identified 27 loci in the genome that contained double repeats but with a two-nucleotide space. Interestingly, we found that no factors in the Pol IV transcription complex were enriched at these loci (Extended Data Fig. 8b). In addition, we performed DAP-seq46 by incubating Halo-tagged recombinant proteins with DNA extracted from WT unopened flower buds to test the ability of the VDD–VAL–GDE1 complex to bind Arabidopsis genomic DNA in vitro. All sequences that bound the VDD–VAL–GDE1 complex showed a one-nucleotide spacer (Fig. 6b). Overall, these results strongly support that the REM–GDE1 complex can directly bind the CLSY3 CLSY4 motif 1 and that the single-nucleotide spacer is important.We found that REM transcription factors and GDE1 ChIP-seq peak summits, but not CLSY3/4–Pol IV summits, localized to the centre of motifs (Fig. 6a,c). On average, CLSY3/4 and Pol IV were located approximately 200 or 300 base pairs away from the centre of the motif, respectively (Fig. 6c). In addition, CLSY3/4 and Pol IV summits were found on either one side or the other side of the centre of the motif, but not on both sides (Fig. 6a,c). To further characterize the direction of CLSY3/4 and Pol IV distribution relative to the sequence motif, we separated motifs into two groups based on the DNA strand that contained the TTTTGCTTATNTTTTGCTTAT motif sequence in the genome. We found that the distribution of CLSY3/4 and Pol IV signals was still localized to either one side or the other side of the motif centres (Extended Data Fig. 8c), indicating that the direction of the motif alone does not determine the direction of Pol IV transcription. An alternative explanation is that the directionality of Pol IV transcription may be influenced by the chromatin environment of the regions. To explore this, we divided the motifs into two additional groups based on the relative positions of Pol IV summits and observed that the direction of the Pol IV shift closely aligned with more nucleosome-occupied regions (Fig. 6d). It is not clear, however, if the higher nucleosome density is a cause or a consequence of Pol IV transcription. In any case, these results suggest that CLSL3/4 and Pol IV are recruited to the CLSY3 CLSY4 motif 1 by the REM transcription factor–GDE1 complex, after which they engage in unidirectional transcription (to one side of the motif or the other) to produce Pol IV transcripts required to produce siRNAs. Consistent with this model, the accumulation levels of 24-nt siRNA transcripts aligned with the localization pattern of the CLSY3/4–Pol IV complex at the CLSY3 CLSY4 motif 1 (Fig. 6d).REM8 recognizes CLSL3 CLSY4 motif 2 for siRNA productionAs opposed to the above-described REM transcription factors that appear to mainly drive CLSY3/4–Pol IV complex to group 1 loci of CLSY3 CLSY4-dependent siRNA, we found that REM8 exhibited unique binding sites within group 2 sites (Fig. 3b–d,h and Extended Data Fig. 3a,e). A REM8 deletion mutant was generated using CRISPR–Cas9 (Extended Data Fig. 9a), but this mutation did not affect siRNA production at REM8 binding sites (Extended Data Fig. 9b). Given the genetic redundancy of REM transcription factors at the CLSY3 CLSY4 motif 1, it is likely that additional factors contribute to the recognition of motif 2. In line with upregulation of 24-nt siRNA in gde1-1 at group 2 loci (Fig. 2a,c), we found that more 24-nt siRNAs were detectable at REM8-binding non-siren regions in gde1-1 (Fig. 7a), which is correlated with the enhanced enrichment of CLSY3/4 and Pol IV enrichment at these sites in the gde1-1 background (Extended Data Fig. 9c). This also shows that siRNAs accumulate at these sites in a GDE1-independent manner and that REM8 acts via a different mechanism to recruit Pol IV complexes.Fig. 7: REM8 redistributes Pol IV complex to CLSL3 CLSY4 motif 2 in the absence of GDE1, and GDE1 is required for siRNA in anther.a, A violin plot showing the 24-nt siRNA levels at REM8-binding non-siren loci (n = 162) in the indicated genotypes. P values calculated by pairwise t-tests are indicated. The line in the centre of each violin plot represents the median. The thick black bar in the centre represents the interquartile range. The whiskers represent the rest of the distribution. b, MEME analysis of REM8-binding non-siren loci. E value is the expected number of false positives in the matches up to this point. c, A violin plot showing the 24-nt siRNA levels at CLSL3 CLSY4 motif 2 (n = 22) in the indicated genotypes. P values calculated by pairwise t-tests are indicated. The line in the centre of each violin plot represents the median. The thick black bar in the centre represents the interquartile range. The whiskers represent the rest of the distribution. d, The total number of differentially expressed siRNA in the anther tissue of indicated genotypes. e, A Venn diagram showing the relationship between CLSY3 CLSY4-dependent and GDE1-dependent siRNA loci in anther. f, A violin plot showing the 24-nt siRNA levels at anther clsy3 clsy4 and gde1-1 downregulation non-siren loci (n = 223) in the indicated genotypes. P values calculated by pairwise t-tests are indicated. Pairwise t-tests showed P = 2.4074 × 10−120 (Col-0an versus clsy34an) and P = 2.0517 × 10−119 (Col-0an versus gde1-1an). The line in the centre of each violin plot represents the median. The thick black bar in the centre represents the interquartile range. The whiskers represent the rest of the distribution.Full size imageMotif analysis of REM8 peaks revealed three repeats of the AAGCGGAT sequence with one-nucleotide spaced between them (Fig. 7b) (hereafter named the CLSY3 CLSY4 motif 2). Fifty CLSY3 CLSY4 motif 2 sites were identified, matching to 22 24-nt siRNA loci throughout the genome, and REM8 and CLSY3/4 were enriched at most of them (Extended Data Fig. 9d). siRNA levels at CLSY3 CLSY4 motif 2 loci decreased in the clsy3 clsy4 mutant but increased in the gde1-1 mutant (Fig. 7c). These results suggest that REM8 may act together with other unidentified factors to recruit the CLSY3/4–Pol IV complex to CLSY3 CLSY4 motif 2 for 24-nt siRNA production and that this recruitment is more robust in the gde1-1 mutant, where CLSY3/4–Pol IV are released from the siren loci.GDE1 is required for anther siRNA productionCLSY3 is also required for the biogenesis of siRNAs at hyperTE loci in male meiocyte cells10. Notably, there is a limited overlap of only 12 shared loci between the maternal siren loci and paternal hyperTE loci12, suggesting distinct siRNA biogenesis mechanisms. In tapetum cells, most REM transcription factors, with the exception of REM46 and REM16, are very lowly expressed (Extended Data Fig. 3d). Intriguingly, GDE1 stands out as showing very high expression levels in tapetum cells (Extended Data Fig. 3d), suggestive of a potential role in meiocyte cells. 449 siRNA loci showed reduction in the anthers of the clsy3 clsy4 mutant (Fig. 7d), and over half of these loci exhibited similar levels of reduction in the anthers of the gde1-1 mutant (Fig. 7e,f). However, hardly any differentially expressed siRNAs were found in rem46 val rem12 triple mutants (Fig. 7d). Consistently, only GDE1, CLSY3/4 and Pol IV, but none of the tested REM transcription factors, showed enrichment at anther CLSY3 CLSY4-dependent siRNA loci (Extended Data Fig. 10). Together, these results suggest that GDE1 is also required for anther siRNA production, but probably via a mechanism different from that seen in ovules.DiscussionOur study uncovers distinct mechanisms by which CLSY proteins recruit Pol IV complexes to direct siRNA production, highlighting an alternative paradigm in the regulation of DNA methylation and siRNA biogenesis. While CLSY1 and CLSY2 operate in association with SHH1, an H3K9 methylation reader, to recruit Pol IV to chromatin via histone marks and other epigenetic modifications9,30,31,47,48, we demonstrate that CLSY3 and CLSY4 recruit Pol IV in a fundamentally different manner. In ovules, CLSY3 and CLSY4 rely on GDE1, which acts in concert with REM transcription factors to direct siRNA production to specific sequences. Our work suggests that CLSY proteins have evolved multiple recruitment strategies, some driven by epigenetic features and others by sequence-specific genetic information, allowing flexible and tissue-specific control of siRNA production.Our study identifies GDE1 as a critical mediator that enhances the recruitment of CLSY3/4 to siren loci. The strong enrichment of CLSY3/4 and Pol IV at siren loci is lost in the gde1-1 mutant, where these complexes are redistributed to group 2 loci. This redistribution underscores the specificity of GDE1 in guiding siRNA biogenesis at specific CLSY3 CLSY4-dependent loci, reinforcing its central role in the spatial and functional organization of Pol IV recruitment.VDD and VAL, two REM transcription factors that target siren loci, have been reported as critical for the degeneration of synergid cells, a vital step in ensuring successful fertilization37. vdd-1/+ and VAL RNAi mutants exhibit pronounced female gametophytic defects, underscoring the importance of these factors in reproductive success. However, the viability of our rem46 val rem12 triple mutant in this study suggests that the reduced fertility observed in the VAL RNAi mutant may result from off-target silencing of related homologues. The expression of GAMETOPHYTIC FACTOR 2 (GFA2), a key regulator of synergid cell death, can rescue the phenotypic effects observed in the vdd-1/+ mutant37. Notably, none of the gde1-1, clsy3 clsy4 and pol iv mutants display any obvious developmental phenotypes, implying that these factors may not directly contribute to developmental processes. Our results also suggest diverse regulatory roles of REM transcription factors using siRNA-dependent and siRNA-independent mechanisms.Although we identified only 86 CLSY3 CLSY4 motif 1 sites across the genome, over 400 loci were affected in the rem46 val rem12 triple mutants. This observation suggests that REM transcription factors do not exclusively recognize CLSY3 CLSY4 motif 1. Instead, additional transcription factors such as REM19, REM22 and REM8 are also enriched at siren loci. Given their in vitro association37, these factors may form distinct combinatory heterodimers, enabling broader recruitment and siRNA regulation at loci that lack the CLSY3 CLSY4 motif 1. This mechanism of cooperative binding and redundancy probably enhances the robustness of siRNA biogenesis, ensuring proper siRNA regulation even when individual motifs or factors are absent.The recruitment of Pol IV by REM transcription factors with the aid of GDE1 draws striking parallels to transcription factor-mediated DNA methylation and gene silencing in mammals. For example, KRAB-zinc finger proteins (KRAB-ZFPs), the largest transcription factor family in mammalian cells, are known to bind specific DNA sequences and recruit the repressor KRAB-associated protein 1 (KAP-1)49,50, which recruits diverse complexes to regulate DNA methylation, histone deacetylation, H3K9 trimethylation and transposon and gene silencing51,52. In an analogous fashion, our results suggest that REM transcription factors recruit CLSY3/4–Pol IV complexes via GDE1 to direct siRNA production and subsequent DNA methylation at target loci. These parallels suggest that sequence-specific DNA binding proteins are used widely throughout different eukaryotic species to properly pattern DNA methylation.In summary, our work demonstrates that CLSY proteins use distinct strategies to recruit Pol IV, integrating both epigenetic and genetic cues. REM transcription factors and GDE1 provide a sequence-specific recruitment mechanism for CLSY3/4–Pol IV complexes, adding an additional layer of regulation to siRNA biogenesis and DNA methylation. At other sites and in other tissues, Pol IV employs a primarily epigenetic mechanism for recruitment, using particular histone marks to recruit the machinery for siRNA biogenesis. These findings broaden our understanding of the diversity and complexity of epigenetic regulation and open additional avenues for exploring the interplay between genetic and epigenetic information in development and genome stability.MethodsPlant materials and growth conditionsAll Arabidopsis plants used in this paper are Col-0 ecotype, and plants are grown under standard condition with 16 h light/8 h dark at 22 °C. The T-DNA insertion lines used in this study included gde1-1 (SALKseq_10069.1), clsy3-1 (SALK_040366) and clsy4-1 (SALK_003876). Agrobacterium (AGL0 strain)-mediated floral dipping was used to generate all the transgenic plants.Plasmid constructionFor pGDE1:GDE1-3FLAG, pCLSY3:CLSY3-9myc, pCLSY4:CLSY4-9myc, pPol IV:Pol IV-9myc, pVDD:VDD-9myc, pVDD:VDD-3FLAG, pVAL:VAL-9myc, pREM13:REM13-9myc, pREM19:REM19-9myc, pREM22:REM22-9myc, pREM8:REM8-9myc, pVDD:VDD-FLAG-ZF108 and the genomic DNA sequences of each with promoter sequences (around 2 kb upstream from the start codon) were first cloned into pENTR/D-TOPO vectors (Invitrogen) and then to the destination vector pEG302-GW-3FLAG, pEG302-GW-9Myc or pEG302-GW-3FLAG-ZF108 by LR reaction (LR Clonase II, Invitrogen). For pIX-HALO-VDD, pIX-HALO-VAL and pIX-HALO-GDE1, VDD, VAL and GDE1 were cloned from cDNA into the pENTR/D-TOPO vector and then to the destination vector pIX-HALO by LR reaction. Primers used in this study are presented in Supplementary Table 3.ChIP-seqAround 2.0 g of floral tissues were used for each ChIP. The plant materials were ground into a fine powder with liquid nitrogen and resuspended with nucleus isolation buffer (50 mM HEPES, 1 M sucrose, 5 mM KCl, 5 mM MgCl2, 0.6% Triton X-100, 0.4 mM phenylmethylsulfonyl fluoride (PMSF), 5 mM benzamidine, 1% formaldehyde (Sigma) and 1× protease inhibitor (Roche)) for 10 min with rotation. Then, 1.7 ml of 2 M glycine solution was added immediately to stop the crosslinking. Lysates were filtered through Miracloth, and the nuclei were collected by centrifugation at 4 °C with 2,880g for 20 min. The pellet was resuspended in 1 ml of extraction buffer 2 (0.25 M sucrose, 10 mM Tris–HCl pH 8.0, 10 mM MgCl2, 1% Triton X-100, 5 mM beta-mercaptoethanol (BME), 0.1 mM PMSF, 5 mM benzamidine and 1× protease inhibitor (Roche)) and centrifuged at 12,000g at 4 °C for 10 min. The nuclei were then resuspended with extraction buffer 3 (1.7 M sucrose, 10 mM Tris–HCl pH 8.0, 2 mM MgCl2, 0.15% Triton X-100, 5 mM BME, 0.1 mM PMSF, 5 mM benzamidine, and 1× protease inhibitor (Roche)), at 4 °C with 12,000g for 60 min. The relative pure nuclei were lysed with 400 µl nucleic lysis buffer (50 mM Tris–HCl pH 8.0, 10 mM EDTA, 1% sodium dodecyl sulfate (SDS), 0.1 mM PMSF, 5 mM benzamidine and 1× protease inhibitor (Roche)) on ice for 10 min and a total of 1.7 ml of ChIP dilution buffer (1.1% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris pH 8.0, 167 mM NaCl, 0.1 mM PMSF, 5 mM benzamidine and 1× protease inhibitor (Roche)) was added to the lysed nuclei. Chromatin was sheared by Bioruptor Plus (Diagenode) for 30 cycles with 30 s on/30 s off per cycle. The lysate was centrifuged twice at 4 °C with 20,000g for 10 min, and the supernatant was incubated with either FLAG epitope (Sigma F1804 1:400 dilution) or myc epitope (Cell Signaling, 71D10 1:200 dilution) at 4 °C overnight. Next, the magnetic Protein A and Protein G Dynabeads (Invitrogen) were added and inoculated at 4 °C for 2 h with rotation. The beads were washed with low-salt solution twice (150 mM NaCl, 0.2% SDS, 0.5% Triton X-100, 2 mM EDTA and 20 mM Tris pH 8.0), high-salt solution (150 mM NaCl, 0.2% SDS, 0.5% Triton X-100, 2 mM EDTA and 20 mM Tris pH 8.0), LiCl solution (250 mM LiCl, 1% IGEPAL, 1% sodium deoxycholate, 1 mM EDTA and 10 mM Tris pH 8.0) and TE solution (1 mM EDTA and 10 mM Tris pH 8.0) for 5 min at 4 °C, respectively. The chromatin was eluted with elution buffer (1% SDS, 10 mM EDTA and 0.1 M NaHCO3) and subjected to reverse crosslinking by adding 20 µl 5 M NaCl and incubated at 65 °C overnight. Then, 1 μl of Protease K (20 mg ml−1, Invitrogen), 10 μl of 0.5 M EDTA pH 8.0 and 20 μl of 1 M Tris pH 6.5 were added to deactivate the protein for 4 h at 45 °C, and DNA was purified through phase lock gel (VWR) and precipitated with 1/10 volume of 3 M sodium acetate (Invitrogen), 2 μl GlycoBlue (Invitrogen) and 1 ml 100% ethanol at −20 °C overnight. The precipitated DNA was used for library construction following the manual of the Ovation Ultra Low System V2 kit (NuGEN), and the libraries were sequenced on Illumina NovaSeq 6000 or NovaSeq X Plus instruments.IP–MSAround 10 g of floral tissues from FLAG-epitope-tagged transgenic plants were used for each IP–MS experiment, and floral tissues of Col-0 plants were used as the control. Flower tissue was ground to a fine powder in liquid nitrogen with a homogenizer. Tissue powder was completely resuspended in 25 ml IP buffer (50 mM Tris–HCl pH 8.0, 150 mM NaCl, 5 mM EDTA, 10% glycerol, 0.1% Tergitol, 0.5 mM dithiothreitol, 1 mg ml−1 Pepstatin A, 1 mM PMSF, 50 µM MG132 and cOmplete EDTA-free protease inhibitor (Roche)) at 4 °C for 10 min with rotation. The issue was further disrupted with a Dounce homogenizer. The lysates were filtered with Miracloth and centrifuged at 20,000g for 10 min at 4 °C. The supernatant was incubated with 250 μl anti-FLAG M2 magnetic beads (Sigma) at 4 °C for 2 h with rotation. The magnetic beads were washed four times with IP buffer and eluted with 250 µg ml−1 3×FLAG peptides. Eluted proteins were used for trichloroacetic acid precipitation and mass spectrometric analysis.Quantitative proteomicsProtein pellets were resuspended in 8 M urea and 100 mM Tris pH 8.5, then reduced by adding Tris (2-carboxyethyl) phosphine to a final concentration of 5 mM and incubation for 30 min. Next, the proteins were alkylated by adding iodoacetamide to a final concentration of 10 mM for another 30 min at room temperature. Before protein digestion, the urea concentration was diluted to 2 M with 100 mM Tris pH 8.5. Then, the proteins were digested with LysC (BioLabs) at a 1:100 enzyme/protein ratio at 37 °C for 4 h, followed by the trypsin digestion at 1:100 (trypsin:protein) at 37 °C for 12 h. To stop the digestion, 5% formic acid was added to the samples. Next, the peptides were desalted using C18 pipette tips (Thermo Scientific) and reconstituted in 5% formic acid before being analysed by LC–MS/MS. Tryptic peptide mixtures were loaded onto a 25-cm-long, 75-μm-inner-diameter fused-silica capillary, packed in-house with bulk 1.9 μM ReproSil-Pur beads with 120 Å pores as described53. The peptides were delivered by a 140-min water–acetonitrile linear gradient in 6–28% buffer (acetonitrile solution, 0.1% formic acid and 3% dimethyl sulfoxide) using a Dionex Ultimate 3000 nanoflow UHPLC (Thermo Scientific), at a flow rate of 200 nl min−1, further increased to 35% and followed by a rapid ramp-up to 85%. The eluted peptides were ionized and the Orbitrap Fusion Lumos Tribrid Mass Spectrometer (Thermo Scientific) was used to acquire the mass spectrometric data. The data-dependent acquisition strategy consisted of a repeating cycle of a full MS1 spectrum (resolution 120,000) followed by sequential MS2 scan (resolution 15,000). Label-free quantification (LFQ) was performed using the MaxQuant software package (v1.6.17.0) with LFQ default setting54, and the Arabidopsis TAIR 10 proteome database was used for the database search. Trypsin digestion was applied, and a maximum of two missed cleavages were allowed in all searches for tryptic peptides of length 8–40 amino acids. In all, 1% false discovery rate (FDR) was used as a filter at both protein and peptide-spectrum match levels. IP–MS of Col-0 plant tissue was used as the control. The empirical Bayes test performed by LIMMA was used for statistical analysis.Co-IPTen millilitres of floral tissues was collected from GDE1–3FLAG × VDD–9myc, GDE1–3FLAG × CLSY3–9myc, GDE1–3FLAG × Pol IV–9myc, VDD–9myc, CLSY3–9myc and Pol IV–myc. Tissues were ground into a fine powder with liquid nitrogen, mixed with 10 ml IP buffer (50 mM Tris–HCl pH 7.5, 150 mM NaCl, 2 mM EDTA, 2 mm dithiothreitol, 0.8% Triton X-100 and 1× protease inhibitor (Roche)) and incubated at 4 °C for 20 min. The lysate was centrifuged at 18,000g for 10 min at 4 °C, and the supernatant was centrifuged one more time. The supernatant was incubated with 30 μl anti-FLAG M2 magnetic beads (Sigma) for 2 h at 4 °C. The beads were washed with IP buffer five times, and proteins were eluted with 40 μl elution buffer (IP buffer containing 100 μg ml−1 3×FLAG peptide as final concentration) by vigorously shaking at 37 °C for 15 min. The elutions were mixed with 2× SDS loading buffer for western blot. Anti-Myc/c-Myc antibody (9E10) HRP (Santa Cruz Biotechnology sc-40 HRP, 1:3,000 dilution) and monoclonal ANTI-FLAG M2 HRP (Sigma-Aldrich A8592, 1:7,500 dilution) were used for western blot.Small RNA-seqPistils and anther (stage 9 or younger) of each genotype were collected and ground into a fine powder with liquid nitrogen. Total RNA was extracted using the Direct-zol RNA Miniprep kit (Zymo) according to the manufacturer’s instructions. Two micrograms of total RNA were mixed with equal volume of the 2× RNA loading dye43, denatured at 65 °C for 10 min and immediately chilled on ice. Denatured total RNA was separated on 15% TBE urea gel (Invitrogen), and small RNAs from 15 to 30 nucleotides were excised. The excised gel pieces were mashed by pestles. Small RNA was eluted with 400 μl nuclease-free water at 70 °C for 10 min and then precipitated with ethanol. Small RNA libraries were made using NEBNext Small RNA Library Prep Set for Illumina (Multiplex Compatible)43 according to the manufacturer’s instructions. The libraries were sequenced on Illumina NovaSeq 6000 or NovaSeq X Plus instruments.RNA-seqSeedlings, pistils and pollens of Col-0 were collected and ground into a fine powder with liquid nitrogen. Total RNA was extracted using the Direct-zol RNA Miniprep kit (Zymo) according to the manufacturer’s instructions. One microgram of total RNA was used to prepare the libraries for RNA sequencing (RNA-seq) following the TruSeq Stranded mRNA kit (Illumina), and the libraries were sequenced on Illumina NovaSeq 6000 or NovaSeq X Plus instruments.Whole-genome bisulfite sequencingDNA from pistils was extracted with Qiagen DNeasy plant mini kit (Qiagen 69106). RNA was removed with PureLink RNase A (Invitrogen). A total of 100 ng of DNA was sheared to 200 bp (duty cycle 10%, intensity 5, cycles per burst 200, treatment time 120 s) with a Covaris S2 (Covaris). Libraries were prepared with the Epitect Bisulfite Conversion kit (Qiagen) and the Ovation Ultralow Methyl-seq kit (NuGEN) following the manufacturer’s instructions. The libraries were sequenced on Illumina NovaSeq 6000 or NovaSeq X Plus instruments.DAP-seqThe DAP-seq experiment was performed as previously46. First, for genomic DNA library preparation, 5 µg of genomic DNA was purified with the DNeasy Plant Mini Kit (Qiagen, 69106) from Col-0 unopened flower buds and the DNA was diluted with water into 55-µl aliquots each containing 1,000 µg of DNA and sheared with the Covaris S2 instrument to an average fragment size of 200 bp (duty cycle 10%, intensity 5, cycles per burst 200, treatment time 120 s). Fifty microlitres of sheared DNA underwent steps 1–3 of library construction using the KAPA Hyper Prep kit (KR0961) (end repair and A-tailing, adapter ligation and post-ligation cleanup). Second, 1 µg of pIX-HALO-VDD, pIX-HALO-VAL and pIX-HALO-GDE1 was used to express their corresponding proteins in TNA SP6 Coupled Wheat Germ Extract system (Promega). Magnetic HALO beads were used to enrich proteins and inoculated with genomic DNA library at room temperature for 1 hr. The final PCR amplification of the pulled-down DNA was performed using Fusion High-Fidelity PCR Master Mix. Each sample was individually run on agarose gel, and the smear between 200 bp and 400 bp was cut out and purified. The libraries were sequenced on Illumina NovaSeq 6000 or NovaSeq X Plus instruments.Alphafold3 structure predictionThe three-dimensional models of VDD-VAL-GDE1-motif 1, REM12-REM13-GDE1-motif 1, VDD-REM13-GDE1-motif 1, VDD-REM12-GDE1-motif 1, VAL-REM13-GDE1-motif 1, VAL-REM12-GDE1-motif 1, VDD-VDD-GDE1-motif 1, REM13-REM13-GDE1-motif 1, REM12-REM12-GDE1-motif 1 and VAL-VAL-GDE1-motif 1 were generated by Alphafold343 and visualized and analysed by PyMOL 3.0.2.Bioinformatic analysisFor ChIP-seq and DAP-seq analysis, raw reads were aligned to the Arabidopsis reference genome (TAIR10) with Bowtie2 (v2.3.4.3)55, allowing only uniquely mapped reads with perfect matches. The Samtools version 1.9 was used to remove duplicated reads56. The deeptools version 3.1.3 was used to generate Bigwig tracks57. Peaks were called using MACS2 (v2.1.1)58.For differential ChIP-seq localization analysis, ChIP-seq levels at the CLSY3 CLSY4-dependent siRNA regions were quantified with the HOMER (v4.11.1) annotatePeaks.pl script using the ‘-noadj, -size given and -len 1’ options. Differentially expressed 24-nt siRNA compared with the WT controls were then identified using DESeq (version 1.42.1) (log2FC ≥1 and FDR ≤0.05). FC, fold change. The data were plotted using the R package ggplot (v3.5.1).For binding motif analysis, MEME 5.5.0 was used to discover the motifs of the ChIP-seq datasets59. FIMO (v5.5.7) was used to scan genome-wide distributions of CLSY3 CLSY4 motif 1 TTTTGCTTAT (single repeat) with one mismatch allowed, TTTTGCTTATNTTTTGCTTAT (double repeats) with one mismatch allowed in each repeat, TTTTGCTTATNTTTTGCTTATNTTTTGCTTAT (triple repeats) with one mismatch allowed in each repeat, and TTTTGCTTATNNTTTTGCTTAT (double repeats with a two-nucleotide space) with one mismatch allowed in each repeat60; CLSY3 CLSY4 motif 2 AAGCGGATNAAGCGGATNAAGCGGAT had a P value less than 5 × 10−9 and q value less than 0.025. TomTom (v5.5.7) was used to analyse the similarities between motifs61.For whole-genome bisulfite sequencing (WGBS) analysis, WGBS raw reads were aligned to both strands of reference genome TAIR10 using BSMAP (v.2.74)62 allowing up to two mismatches and one best hit (-v 2 -w 1). Reads with more than three consecutively methylated CHH sites were considered as non-converted reads and removed. Methylation levels were calculated with the ratio of C/(C + T).For RNA-seq analysis, Col-0 leaf, meiocyte and tapetum RNA seq data were downloaded from NCBI Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/) as accession codes GSM2306324; GSM2306325; GSM2306326 ref. 63, GSM2306313; GSM2306314; GSM2306315 ref. 63 and GSM4911399; GSM4911400; GSM4911401 ref. 10, respectively. All raw reads of RNA-seq data were aligned to reference genome TAIR10 by Bowtie2 (v2.3.4.3)55, and expression abundance was calculated by RSEM (v1.3.1) with default settings64. The bamCoverage of deeptools (version 3.1.351) was used to normalize the data with RPKM (reads per kilobase per million mapped reads)57.For small RNA-seq analysis, the adaptor sequence (TGGAATTCTCGG) of small RNA-seq reads was trimmed with trim_galore, and trimmed reads were mapped to the reference genome TAIR10 using Bowtie2 (v2.3.4.3) with only one unique hit and zero mismatches55. Small RNA reads that mapped to chloroplast, mitochondrial DNA, tRNA, rRNA, small nucleolar RNAs and small nuclear RNAs were removed using bedtools (v2.26.0)65. The deeptools (version 3.1.3) was used to generate Bigwig tracks57. The bamCoverage of deeptools (version 3.1.3) was used to normalize the data with RPKM57.For differentially expressed 24-nt siRNA clusters analysis, Pol-IV-dependent master siRNA was defined from a previous publication11. The 24-nt siRNA levels at the master 24-nt siRNA were quantified with the HOMER (v4.11.1) annotatePeaks.pl script using the ‘-noadj, -size given and -len 1’ options. The 24-nt siRNA expression levels were normalized by the total miRNA amount, which was defined previously66. Comparison of the differentially expressed 24-nt siRNA with the WT controls was then performed using DESeq (version 1.42.1) (log2FC ≤1 and FDR ≤0.05).Statistics and reproducibilityPairwise t-tests were used for the siRNA and DNA methylation levels analysis. The two-sided empirical Bayes test performed by LIMMA was used for statistical analysis on IP–MS. Three biological replicates were included for all siRNA analyses, whereas two biological replicates were included for DNA methylation analysis and IP–MS analysis. No data were excluded from all analysis. The western blot experiments shown in Extended Data Fig. 2b were independently repeated three times with similar results.Reporting summaryFurther information on research design is available in the Nature Portfolio Reporting Summary linked to this article.Data availabilityAll the high-throughput sequencing data generated in this study are accessible at NCBI’s GEO via GEO series accession number GSE269181. The mass spectrometry proteomics data generated in this study have been deposited in the ProteomeXchange Consortium via the MassIVE partner repository under accession code MSV000097625. The TAIR10 genome is available at https://www.arabidopsis.org/index.jsp. The Col-0 leaf, meiocyte and tapetum RNA seq data used in this study are available via the National Center for Biotechnology information Gene Expression Omnibus database under accession codes GSM2306324; GSM2306325; GSM2306326, GSM2306313; GSM2306314; GSM2306315 and GSM4911399; GSM4911400; GSM4911401, respectively. All custom scripts are available from the corresponding author upon request. Source data are provided with this paper.ReferencesHe, X. J. et al. A conserved transcriptional regulator is required for RNA-directed DNA methylation and plant development. Gene Dev. 23, 2717–2722 (2009).Article CAS PubMed PubMed Central Google Scholar Slotkin, R. K. & Martienssen, R. Transposable elements and the epigenetic regulation of the genome. Nat. Rev. Genet. 8, 272–285 (2007).Article CAS PubMed Google Scholar Law, J. A. & Jacobsen, S. E. Establishing, maintaining and modifying DNA methylation patterns in plants and animals. Nat. Rev. Genet. 11, 204–220 (2010).Article CAS PubMed PubMed Central Google Scholar Chan, S. W. L. et al. RNA silencing genes control de novo DNA methylation. Science 303, 1336–1336 (2004).Article CAS PubMed Google Scholar Matzke, M. A. & Mosher, R. A. RNA-directed DNA methylation: an epigenetic pathway of increasing complexity. Nat. Rev. Genet. 15, 394–408 (2014).Article CAS PubMed Google Scholar Zhai, J. X. et al. A one precursor one siRNA model for Pol IV-dependent siRNA biogenesis. Cell 163, 445–455 (2015).Article CAS PubMed PubMed Central Google Scholar Kuo, H. Y., Jacobsen, E. L., Long, Y. P., Chen, X. Y. & Zhai, J. X. Characteristics and processing of Pol IV-dependent transcripts in Arabidopsis. J. Genet. Genomics 44, 3–6 (2017).Article CAS PubMed Google Scholar Herr, A. J., Jensen, M. B., Dalmay, T. & Baulcombe, D. C. RNA polymerase IV directs silencing of endogenous DNA. Science 308, 118–120 (2005).Article CAS PubMed Google Scholar Zhou, M., Palanca, A. M. S. & Law, J. A. Locus-specific control of the de novo DNA methylation pathway in Arabidopsis by the CLASSY family. Nat. Genet. 50, 865–873 (2018).Article PubMed PubMed Central Google Scholar Long, J. C. et al. Nurse cell-derived small RNAs define paternal epigenetic inheritance in Arabidopsis. Science 373, eabh0556 (2021).Article CAS PubMed Google Scholar Zhou, M. et al. The CLASSY family controls tissue-specific DNA methylation patterns in Arabidopsis. Nat. Commun. 13, 244 (2022).Article CAS PubMed PubMed Central Google Scholar Chow, H. T. & Mosher, R. A. Small RNA-mediated DNA methylation during plant reproduction. Plant Cell 35, 1787–1800 (2023).Article CAS PubMed PubMed Central Google Scholar Grover, J. W. et al. Abundant expression of maternal siRNAs is a conserved feature of seed development. Proc. Natl Acad. Sci. USA 117, 15305–15315 (2020).Article PubMed PubMed Central Google Scholar Kawakatsu, T. et al. Unique cell-type-specific patterns of DNA methylation in the root meristem. Nat. Plants 2, 16058 (2016).Article CAS PubMed PubMed Central Google Scholar Mosher, R. A. et al. Uniparental expression of PolIV-dependent siRNAs in developing endosperm of Arabidopsis. Nature 460, 283–U151 (2009).Article CAS PubMed Google Scholar Burgess, D., Chow, H. T., Grover, J. W., Freeling, M. & Mosher, R. A. Ovule siRNAs methylate protein-coding genes in trans. Plant Cell 34, 3647–3664 (2022).Article PubMed PubMed Central Google Scholar Dziasek, K. et al. Dosage-sensitive maternal siRNAs determine hybridization success in Capsella. Nat. Plants 10, 1969–1983 (2024).Article CAS PubMed PubMed Central Google Scholar Xu, G. & Law, J. A. Loops, crosstalk, and compartmentalization: it takes many layers to regulate DNA methylation. Curr. Opin. Genet. Dev. 84, 102147 (2024).Article CAS PubMed PubMed Central Google Scholar Du, J., Johnson, L. M., Jacobsen, S. E. & Patel, D. J. DNA methylation pathways and their crosstalk with histone methylation. Nat. Rev. Mol. Cell Biol. 16, 519–532 (2015).Article CAS PubMed PubMed Central Google Scholar Johnson, L. M. et al. The SRA methyl-cytosine-binding domain links DNA and histone methylation. Curr. Biol. 17, 379–384 (2007).Article CAS PubMed PubMed Central Google Scholar Rajakumara, E. et al. A dual flip-out mechanism for 5mC recognition by the Arabidopsis SUVH5 SRA domain and its impact on DNA methylation and H3K9 dimethylation in vivo. Genes Dev. 25, 137–152 (2011).Article CAS PubMed PubMed Central Google Scholar Ebbs, M. L. & Bender, J. Locus-specific control of DNA methylation by the Arabidopsis SUVH5 histone methyltransferase. Plant Cell 18, 1166–1176 (2006).Article CAS PubMed PubMed Central Google Scholar Jackson, J. P. et al. Dimethylation of histone H3 lysine 9 is a critical mark for DNA methylation and gene silencing in Arabidopsis thaliana. Chromosoma 112, 308–315 (2004).Article CAS PubMed Google Scholar Jackson, J. P., Lindroth, A. M., Cao, X. & Jacobsen, S. E. Control of CpNpG DNA methylation by the KRYPTONITE histone H3 methyltransferase. Nature 416, 556–560 (2002).Article CAS PubMed Google Scholar Bartee, L., Malagnac, F. & Bender, J. Arabidopsis cmt3 chromomethylase mutations block non-CG methylation and silencing of an endogenous gene. Genes Dev. 15, 1753–1758 (2001).Article CAS PubMed PubMed Central Google Scholar Lindroth, A. M. et al. Requirement of CHROMOMETHYLASE3 for maintenance of CpXpG methylation. Science 292, 2077–2080 (2001).Article CAS PubMed Google Scholar Stroud, H. et al. Non-CG methylation patterns shape the epigenetic landscape in Arabidopsis. Nat. Struct. Mol. Biol. 21, 64-+ (2014).Article CAS PubMed Google Scholar Du, J. et al. Dual binding of chromomethylase domains to H3K9me2-containing nucleosomes directs DNA methylation in plants. Cell 151, 167–180 (2012).Article CAS PubMed PubMed Central Google Scholar Yang, D. L. et al. Four putative SWI2/SNF2 chromatin remodelers have dual roles in regulating DNA methylation in Arabidopsis. Cell Discov. 4, 55 (2018).Article PubMed PubMed Central Google Scholar Law, J. A. et al. Polymerase IV occupancy at RNA-directed DNA methylation sites requires SHH1. Nature 498, 385–389 (2013).Article PubMed PubMed Central Google Scholar Zhang, H. et al. DTF1 is a core component of RNA-directed DNA methylation and may assist in the recruitment of Pol IV. Proc. Natl Acad. Sci. USA 110, 8290–8295 (2013).Article CAS PubMed PubMed Central Google Scholar Wang, Y. et al. ZMP recruits and excludes Pol IV-mediated DNA methylation in a site-specific manner. Sci. Adv. 8, eadc9454 (2022).Article CAS PubMed PubMed Central Google Scholar Xue, Y. et al. Arabidopsis MORC proteins function in the efficient establishment of RNA directed DNA methylation. Nat. Commun. 12, 4292 (2021).Article CAS PubMed PubMed Central Google Scholar Zhong, Z. H. et al. MORC proteins regulate transcription factor binding by mediating chromatin compaction in active chromatin regions. Genome Biol. 24, 96 (2023).Article CAS PubMed PubMed Central Google Scholar Waese, J. et al. ePlant: visualizing and exploring multiple levels of data for hypothesis generation in plant biology. Plant Cell 29, 1806–1821 (2017).Article CAS PubMed PubMed Central Google Scholar Rymen, B., Ferrafiat, L. & Blevins, T. Non-coding RNA polymerases that silence transposable elements and reprogram gene expression in plants. Transcription 11, 172–191 (2020).Article PubMed PubMed Central Google Scholar Mendes, M. A. et al. Live and let die: a REM complex promotes fertilization through synergid cell death in Arabidopsis. Development 143, 2780–2790 (2016).Article CAS PubMed Google Scholar Obayashi, T. et al. ATTED-II: a database of co-expressed genes and cis elements for identifying co-regulated gene groups in Arabidopsis. Nucleic Acids Res. 35, D863–D869 (2007).Article CAS PubMed Google Scholar O’Malley, R. C. et al. Cistrome and epicistrome features shape the regulatory DNA landscape. Cell 165, 1280–1292 (2016).Article PubMed PubMed Central Google Scholar Zhang, X. Y., Henderson, I. R., Lu, C., Green, P. J. & Jacobsen, S. E. Role of RNA polymerase IV in plant small RNA metabolism. Proc. Natl Acad. Sci. USA 104, 4536–4541 (2007).Article CAS PubMed PubMed Central Google Scholar Gallego-Bartolomé, J. et al. Co-targeting RNA polymerases IV and V promotes efficient de novo DNA methylation in Arabidopsis. Cell 176, 1068–1082 (2019).Article Google Scholar Wang, M. et al. A gene silencing screen uncovers diverse tools for targeted gene repression in Arabidopsis. Nat. Plants 9, 460–472 (2023).Article PubMed PubMed Central Google Scholar Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493–500 (2024).Article CAS PubMed PubMed Central Google Scholar Boer, D. R. et al. Structural basis for DNA binding specificity by the auxin-dependent ARF transcription factors. Cell 156, 577–589 (2014).Article CAS PubMed Google Scholar Freire-Rios, A. et al. Architecture of DNA elements mediating ARF transcription factor binding and auxin-responsive gene expression in Arabidopsis. Proc. Natl Acad. Sci. USA 117, 24557–24566 (2020).Article CAS PubMed PubMed Central Google Scholar Desprez, T. et al. Resistance against herbicide isoxaben and cellulose deficiency caused by distinct mutations in same cellulose synthase isoform CESA6. Plant Physiol. 128, 482–490 (2002).Article CAS PubMed PubMed Central Google Scholar Liu, J. et al. An atypical component of RNA-directed DNA methylation machinery has both DNA methylation-dependent and -independent roles in locus-specific transcriptional gene silencing. Cell Res. 21, 1691–1700 (2011).Article CAS PubMed PubMed Central Google Scholar Law, J. A., Vashisht, A. A., Wohlschlegel, J. A. & Jacobsen, S. E. SHH1, a homeodomain protein required for DNA methylation, as well as RDR2, RDM4, and chromatin remodeling factors, associate with RNA polymerase IV. PLoS Genet. 7, e1002195 (2011).Article CAS PubMed PubMed Central Google Scholar Groner, A. C. et al. KRAB-zinc finger proteins and KAP1 can mediate long-range transcriptional repression through heterochromatin spreading. PLoS Genet. 6, e1000869 (2010).Article PubMed PubMed Central Google Scholar Stoll, G. A. et al. Structure of KAP1 tripartite motif identifies molecular interfaces required for retroelement silencing. Proc. Natl Acad. Sci. USA 116, 15042–15051 (2019).Article CAS PubMed PubMed Central Google Scholar Yang, P., Wang, Y. X. & Macfarlan, T. S. The role of KRAB-ZFPs in transposable element repression and mammalian evolution. Trends Genet. 33, 871–881 (2017).Article CAS PubMed PubMed Central Google Scholar Sun, M. Y., Ju, J. Y., Ding, Y., Zhao, C. L. & Tian, C. Y. The signaling pathways regulated by KRAB zinc-finger proteins in cancer. Biochim. Biophys. Acta Rev. Cancer 1877, 188731 (2022).Article CAS PubMed Google Scholar Jami-Alahmadi, Y., Pandey, V., Mayank, A. K. & Wohlschlegel, J. A. A robust method for packing high resolution C18 RP-nano-HPLC columns. J. Vis. Exp. 10, e62380 (2021).Google Scholar Cox, J. & Mann, M. MaxQuant enables high peptide identification rates, individualized p.p.b.-range mass accuracies and proteome-wide protein quantification. Nat. Biotechnol. 26, 1367–1372 (2008).Article CAS PubMed Google Scholar Langmead, B. & Salzberg, S. L. Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357–U354 (2012).Article CAS PubMed PubMed Central Google Scholar Li, H. et al. The Sequence Alignment/Map format and SAMtools. Bioinformatics 25, 2078–2079 (2009).Article PubMed PubMed Central Google Scholar Ramírez, F. et al. deepTools2: a next generation web server for deep-sequencing data analysis. Nucleic Acids Res. 44, W160–W165 (2016).Article PubMed PubMed Central Google Scholar Zhang, Y. et al. Model-based Analysis of ChIP-Seq (MACS). Genome Biol. 9, R137 (2008).Article PubMed PubMed Central Google Scholar Bailey, T. L. & Elkan, C. Fitting a mixture model by expectation maximization to discover motifs in biopolymers. Proc. Int. Conf. Intell. Syst. Mol. Biol. 2, 28–36 (1994).CAS PubMed Google Scholar Grant, C. E., Bailey, T. L. & Noble, W. S. FIMO: scanning for occurrences of a given motif. Bioinformatics 27, 1017–1018 (2011).Article CAS PubMed PubMed Central Google Scholar Gupta, S., Stamatoyannopoulos, J. A., Bailey, T. L. & Noble, W. S. Quantifying similarity between motifs. Genome Biol. 8, R24 (2007).Article PubMed PubMed Central Google Scholar Li, Y. X. & Li, W. BSMAP: whole genome bisulfite sequence MAPping program. BMC Bioinf. 10, 232 (2009).Article Google Scholar Walker, J. et al. Sexual-lineage-specific DNA methylation regulates meiosis in Arabidopsis. Nat. Genet. 50, 130-+ (2018).Article CAS PubMed Google Scholar Li, B. & Dewey, C. N. RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinf. 12, 323 (2011).Article CAS Google Scholar Quinlan, A. R. & Hall, I. M. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26, 841–842 (2010).Article CAS PubMed PubMed Central Google Scholar Kozomara, A., Birgaoanu, M. & Griffiths-Jones, S. miRBase: from microRNA sequences to function. Nucleic Acids Res. 47, D155–D162 (2019).Article CAS PubMed Google Scholar Download referencesAcknowledgementsWe thank J. Law and G. Xu at Salk Institute for discussion and sharing the clsy mutants; J. Du at Southern University of Science and Technology for discussion of the Alphafold3 results; C. Picard, M. Wang and Y. He for advice on the manuscript; and J. Deng, Y. Yamamoto, J. Um and Z. Guo for technical support. We also thank M. Akhavan and the UCLA BSCRC BioSequencing Core for sequencing support. This work was supported by NIH grant R35 GM130272 to S.E.J. and a Life Sciences Research Foundation Postdoctoral Fellowship and the Hundred Talents Program at Zhejiang University to Z.W. S.E.J. is an Investigator of the Howard Hughes Medical Institute.Author informationAuthors and AffiliationsState Key Laboratory of Rice Biological Breeding, Key Laboratory of Molecular Biology of Crop Pathogens and Insects, Institute of Biotechnology, Zhejiang University, Hangzhou, ChinaZhongshou WuDepartment of Molecular, Cell and Developmental Biology, University of California at Los Angeles, Los Angeles, CA, USAZhongshou Wu, Yan Xue, Shuya Wang, Yuan-Hsin Shih, Zhenhui Zhong, Suhua Feng, Jonathan Draper, Allen Lu, Carsten A. Hoeke & Steven E. JacobsenHoward Hughes Medical Institute, University of California at Los Angeles, Los Angeles, CA, USAZhongshou Wu & Steven E. JacobsenShandong Laboratory of Advanced Agricultural Sciences at Weifang, Peking University Institute of Advanced Agricultural Sciences, Weifang, ChinaYan XueInstitute of Plant Biology, National Taiwan University, Taipei, TaiwanYuan-Hsin Shih & Keqiang WuMinistry of Education Key Laboratory for Bio-Resource and Eco-Environment, College of Life Science, State Key Laboratory of Hydraulics and Mountain River Engineering, Sichuan University, Chengdu, ChinaZhenhui ZhongEli and Edythe Broad Center of Regenerative Medicine and Stem Cell Research, University of California at Los Angeles, Los Angeles, CA, USASuhua Feng & Steven E. JacobsenDepartment of Biological Chemistry, University of California at Los Angeles, Los Angeles, CA, USAJihui Sha, Lu Li, James Wohlschlegel & Steven E. JacobsenAuthorsZhongshou WuView author publicationsSearch author on:PubMed Google ScholarYan XueView author publicationsSearch author on:PubMed Google ScholarShuya WangView author publicationsSearch author on:PubMed Google ScholarYuan-Hsin ShihView author publicationsSearch author on:PubMed Google ScholarZhenhui ZhongView author publicationsSearch author on:PubMed Google ScholarSuhua FengView author publicationsSearch author on:PubMed Google ScholarJonathan DraperView author publicationsSearch author on:PubMed Google ScholarAllen LuView author publicationsSearch author on:PubMed Google ScholarCarsten A. HoekeView author publicationsSearch author on:PubMed Google ScholarJihui ShaView author publicationsSearch author on:PubMed Google ScholarLu LiView author publicationsSearch author on:PubMed Google ScholarJames WohlschlegelView author publicationsSearch author on:PubMed Google ScholarKeqiang WuView author publicationsSearch author on:PubMed Google ScholarSteven E. JacobsenView author publicationsSearch author on:PubMed Google ScholarContributionsZ.W. and S.E.J. designed the research, interpreted data and wrote the manuscript; Z.W. performed most of the experiments and bioinformatic data analysis; Y.X. initiated the project and generated GDE1-tagged transgenic lines; S.W. performed siRNA differentially expression analysis and transcription factor binding analysis; Y.-H.S., K.W. and C.A.H. generated transcription factors tagged lines and performed ChIP-seq; Z.Z. performed WGBS bioinformatic data analysis and initial bioinformatic data analysis of small RNA-seq, RNA-seq and ChIP-seq; S.F. performed high-throughput sequencing; Z.W., J.S., L.L. and J.W. performed IP–MS and interpreted the data. J.D., A.L. and C.A.H. provided technical help.Corresponding authorCorrespondence to Steven E. Jacobsen.Ethics declarationsCompeting interestsThe authors declare no competing interests.Peer reviewPeer review informationNature Cell Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available.Additional informationPublisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.Extended dataExtended Data Fig. 1 GDE1 is a previously uncharacterized protein localizing to RdDM sites.a, Metaplot and heatmap showing enrichment of GDE1 ChIP-seq signal over MORC7 (n = 6,102). b, Genome-wide Spearman correlation analysis between GDE1 ChIP-seq signals and those of MORC7, Pol IV, and Pol V at co-targeted regions. Correlations were calculated in deepTools (the multiBamSummary was followed with plotCorrelation tools). c, Venn diagram showing the relationship between CLSY3 and CLSY4 binding sites. d, Metaplot and heatmap showing enrichment of CLSY4 ChIP-seq signal over CLSY3 (n = 380). e, Relative expression levels of the GDE1 genes in select tissues from ePlant expression viewers.Source data.Extended Data Fig. 2 GDE1 co-localizes and interacts with Pol IV complex for siRNA production.a, Screenshot of CLSY3-9myc, CLSY3-9myc gde1-1, CLSY4-9myc, CLSY4-9myc gde1-1, Pol IV-9myc, Pol IV-9myc gde1-1 and GDE1-3FLAG ChIP-seq signals with control ChIP-seq signals subtracted over a Group 1 (CLSY3 CLSY4-dependent, GDE1-dependent) representative locus (left) or three Group 2 (CLSY3 CLSY4-dependent, GDE1-upregulated) representative loci (right). Square brackets indicate the range on a bar graph. b, Western blot showing a Co-IP assay in GDE1-3FLAG F2 crossed lines with CLSY3-9myc (left), Pol IV-9myc (middle) or VDD-9myc (right), respectively. Similar results were observed in two biological replicates. Source numerical data and unprocessed blots are available in source data.Extended Data Fig. 3 Differential localization of REM transcription factors at CLSY3 CLSY4-dependent siRNA sites.a, Heatmaps showing enrichment of VDD, VAL, REM13, REM19, REM22 and REM8 ChIP-seq signal over three groups of CLSY3 CLSY4-dependent sites. The region is −2Kb to 2Kb from the center. Same regions were used for all metaplots. b, TOMTOM analysis showing the similarities of VDD, REM12 and REM13 binding sites with CLSY3 CLSY4 motif 1. E value is the expected number of false positives in the matches up to this point. c, Screenshot of REM22, VDD, VAL, REM13, REM19, REM8, GDE1, CLSY3, CLSY4 and Pol IV ChIP-seq at a distinct representative Group 1 of CLSY3 CLSY4-dependent siRNA sites, where REM22, but not other REM proteins exhibit enrichment. Square brackets indicate the range on a bar graph. d, RNA-seq expression analysis of REM transcription factors, GDE1, CLSY3 and CLSY4 in different tissues. e, Screenshot of REM8, REM19, VDD, VAL, REM13, REM22, GDE1, CLSY3, CLSY4 and Pol IV ChIP-seq at two representative Group 2 of CLSY3 CLSY4-dependent siRNA loci. Square brackets indicate the range on a bar graph.Extended Data Fig. 4 Transcription factors-driven Pol IV transcription complex localizes to siren loci for siRNA production and DNA methylation.a, TOMTOM analysis showing the similarities of siren loci and CLSY3 CLSY4 motif 1. E value is the expected number of false positives in the matches up to this point. b, Metaplot and heatmap showing enrichment of VDD, VAL, REM13, REM19, REM22 and REM8 ChIP-seq signal over siren loci (n = 133). c, Heatmap showing enrichment of GDE1, CLSY3 and CLSY4 ChIP-seq signal over siren loci (n = 133). The region is −2Kb to 2Kb from the center. d, Heatmap showing the 24nt-siRNA levels at siren loci (n = 133) in the indicated genotypes. e, Screenshots of VDD, VAL, REM13, GDE1, CLSY3, CLSY3 gde1-1, CLSY4, CLSY4 gde1-1, Pol IV and Pol IV gde1-1 ChIP-seq signals with control ChIP-seq signals subtracted and 24nt-siRNA by sRNA-seq and CHG/CHH DNA methylation level by WGBS over a representative siren site in the indicated genotypes. Square brackets indicate the range on a bar graph. f, Metaplot and heatmap showing enrichment of CLSY3/4 ChIP-seq signal over CLSY3/4-bound siren loci (n = 133) and CLSY3/4 non-siren regions in WT and gde1-1 backgrounds. The region is −2Kb to 2Kb from the center. g, Metaplot and heatmap showing enrichment of Pol IV ChIP-seq signal over Pol IV-bound siren loci (n = 133) and Pol IV non-siren regions in WT and gde1-1 backgrounds. The region is −2Kb to 2Kb from the center. h, Heatmap showing the CHH and CHG methylation levels at siren loci (n = 133) in the indicated genotypes.Extended Data Fig. 5 REM transcription factors are required for siRNA production at CLSY3 CLSY4-dependent siRNA loci.a, Genotyping results from the T2 generation confirm the identification of vdd null mutants following the transformation of VDD-9myc into heterozygous vdd knockout mutants. Among the T2 generation, five T1 plants with homozygous vdd mutations were validated, all of which exhibit vdd null mutations. b, Schematic description of CRISPR/Cas9 construct design for knocking out the tandemly duplicated REM46, VAL, and REM12. Deletion detected in genomic DNA of triple mutants are shown with chromatographs from Sanger sequencing. c, Violin plot showing the 24nt-siRNA levels at siRNA down regulation loci at both triple and clsy3 clsy4 mutants (n = 219) in the indicated genotypes. p values calculated by pairwise t-tests are indicated. Pairwise t-tests showed p = 1.0995E-126 (colov vs rem46 val rem12 triple 1 ov), p = 1.82024E-60 (colov vs rem46 val rem12 triple 2 ov) and p = 9.05326E-20 (gde1-1ov vs rem46 val rem12 triple 1 ov). The line in the center of each violin plot represents the median. The thick black bar in the center represents the interquartile range. The whiskers represent the rest of the distribution. d, Metaplot and heatmap showing enrichment of VDD-ZF ChIP-seq signal over ZF fusion targeted sites (n = 397) and motif identified by MEME motif analysis for VDD-ZF. Source numerical data are available in source data.Extended Data Fig. 6 REM transcription factors-initiated siRNA biogenesis is independent of H3K9 methylation.a, Venn diagrams showing the relationships between reduced 24nt-siRNA clusters in suvh4 suvh5 suvh6 triple and the clsy1 clsy2 double mutants. b, Venn diagrams showing the relationships between reduced 24nt-siRNA clusters in suvh4 suvh5 suvh6 triple, clsy3 clsy4 double and the rem46 val rem12 triple mutants. c, Violin plot showing the 24nt-siRNA levels at siren loci (n = 133) in the indicated genotypes. p values calculated by pairwise t-tests are indicated. The line in the center of each violin plot represents the median. The thick black bar in the center represents the interquartile range. The whiskers represent the rest of the distribution. d, Pie charts showing the proportions of 24nt-siRNAs from all Pol IV-dependent clusters in ovule tissue that are reduced in clsy1 clsy2 or clsy3 clsy4 double mutants and the proportions of REM46 VAL REM12-dependent loci in GDE-dependent loci, GDE1-dependent siRNA among CLSY3 CLSY4-dependent group or SUVH456-dependent siRNA among CLSY1 CLSY2-dependent group. Source numerical data are available in source data.Extended Data Fig. 7 AlphaFold3 predicted interactions within the REM transcription factor–GDE1– CLSY3 CLSY4 motif 1 complex.a, Predicted aligned error (PAE) of VDD-VAL-GDE1-motif 1, REM12-REM13-GDE1-motif 1, VDD-REM13-GDE1-motif 1, VDD-REM12-GDE1-motif 1, VAL-REM13-GDE1-motif 1, VDD-VDD-GDE1-motif 1, VAL-REM12-GDE1-motif 1, REM13-REM13-GDE1-motif 1, REM12-REM12-GDE1-motif 1, VAL-VAL-GDE1-motif 1. b, GDE1 gene expression in ovule tissues of indicated genotypes as determined by qRT-PCR. Error bars represent means ± standard deviation (sd) (n = 3 from technical replicates). Individual data dots represent technical replicates. Three biological replicates were performed with similar results. Source numerical data are available in source data.Extended Data Fig. 8 Pol IV transcription complex recognizes CLSY3 CLSY4 motif 1.a, Metaplot and heatmap showing enrichment of VDD, VAL, REM13, GDE1, CLSY3, CLSY4 and Pol IV ChIP-seq signal over single-repeat sites (n = 37,456). b, Metaplot and heatmap showing enrichment of VDD, VAL, REM13, GDE1, CLSY3, CLSY4 and Pol IV ChIP-seq signal over double-repeats with two gaps (n = 27). c, Metaplot and heatmap showing enrichment of CLSY3, CLSY4 and Pol IV ChIP-seq signal over all double-repeats sites at positive strand (n = 56) or negative strand (n = 53).Extended Data Fig. 9 Pol IV transcription complex recognizes CLSY3 CLSY4 motif 2.a, Schematic description of CRISPR/Cas9 construct design for knocking out REM8. Deletion detected in genomic DNA of rem8-cr mutants are shown with chromatographs from Sanger sequencing. b, Violin plot showing the 24nt-siRNA levels at REM8 binding non-siren loci (n = 162) in the indicated genotypes. p values calculated by pairwise t-tests are indicated. The line in the center of each violin plot represents the median. The thick black bar in the center represents the interquartile range. The whiskers represent the rest of the distribution. c, Metaplot and heatmap showing enrichment of CLSY3, CLSY4 and Pol IV ChIP-seq signal over REM8 binding non-siren sites and shuffle regions in WT or gde1-1 backgrounds. The region is −2Kb to 2Kb from the center. d, Metaplot and heatmap showing enrichment of REM8, CLSY3, CLSY4, and Pol IV ChIP-seq signal over CLSY3 CLSY4 motif 2 sites (n = 50). Source numerical data are available in source data.Extended Data Fig. 10 GDE1, CLSY3/4 and Pol IV, but not REM transcription factors enriches at clsy3 clsy4 and gde1-1 down-regulated siRNA loci in anther.Metaplot and heatmap showing enrichment of VDD, VAL, REM13, GDE1, CLSY3, CLSY4 and Pol IV ChIP-seq signal over anther clsy3 clsy4 and gde1-1 down regulation non-siren loci (n = 223).Supplementary informationReporting SummaryPeer Review FileSupplementary Table 1Supplementary Table 1. GDE1–3FLAG IP–MS data. Supplementary Table 2. VDD–3FLAG IP–MS data. Supplementary Table 3. Primers used in this studySource dataSource Data Fig. 1Uncropped western blots.Source Data Extended Data Fig. 1Statistical source data.Rights and permissionsOpen Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.Reprints and permissionsAbout this article