Arabidopsis MORC proteins function in the efficient establishment of RNA directed DNA methylation

The Microrchidia (MORC) family of ATPases are required for transposable element (TE) silencing and heterochromatin condensation in plants and animals, and C. elegans MORC-1 has been shown to topologically entrap and condense DNA. In Arabidopsis thaliana, mutation of MORCs has been shown to reactivate silent methylated genes and transposons and to decondense heterochromatic chromocenters, despite only minor changes in the maintenance of DNA methylation. Here we provide the first evidence localizing Arabidopsis MORC proteins to specific regions of chromatin and find that MORC4 and MORC7 are closely co-localized with sites of RNA-directed DNA methylation (RdDM). We further show that MORC7, when tethered to DNA by an artificial zinc finger, can facilitate the establishment of RdDM. Finally, we show that MORCs are required for the efficient RdDM mediated establishment of DNA methylation and silencing of a newly integrated FWA transgene, even though morc mutations have no effect on the maintenance of preexisting methylation at the endogenous FWA gene. We propose that MORCs function as a molecular tether in RdDM complexes to reinforce RdDM activity for methylation establishment. These findings have implications for MORC protein function in a variety of other eukaryotic organisms.

E pigenetic chromatin modifications are fundamental to genome integrity, gene regulation, and transposon (TE) silencing. DNA methylation is a particularly stable epigenetic modification 1,2 , which in plants occurs in three sequence contexts, CG, CHG, and CHH (where H represents A, T, or C) 3 . Symmetric CG and CHG methylation are maintained by DNA METHYLTRANSFERASE1 (MET1) and CHROMOMETHY-LASE3 (CMT3), respectively 4,5 , while asymmetric CHH methylation is mediated by CHROMOMETHYLASE 2 (CMT2) and DOMAINS REARRANGED METHYLTRANSFERASE1 and 2 (DRM1/2) 6 . The plant-specific RNA-directed DNA methylation (RdDM) pathway is required for both de novo and maintenance DNA methylation.
Canonical RdDM is composed of two major arms driven by either RNA Polymerase IV (Pol IV) 7 or RNA Polymerase V (Pol V) 8 .
The Pol V arm of the RdDM pathway constitutes the DNA methylation step. SUV(VAR)3-9 homologs SUVH2 and SUVH9 recognize preexisting DNA methylation and recruit the DDR complex [15][16][17] 18 . The DDR complex is required for the recruitment of Pol V, which generates non-coding scaffold RNAs that serve as a platform to recruit AGO4-siRNA effector complexes through sequence complementarity 8,18 . This effector complex in turn recruits DRM1/2 for de novo DNA methylation in all sequence contexts, leading to transcriptional gene silencing 6,19 .
Non-canonical RdDM pathways have also been described and are essential for the de novo methylation and silencing of newly integrated TEs and transgenes [20][21][22][23] . Pol II transcripts of TEs or other sequences can be processed by RDR6 and DCL2/4 to generate 21-22-nt siRNAs, which function primarily in posttranscriptional silencing but can occasionally trigger transcriptional silencing 20 through the Pol V pathway, leading to the initial establishment of DNA methylation, after which canonical RdDM takes over to achieve full methylation and silencing of TEs.
The Arabidopsis thaliana genome encodes six functionally redundant MORC proteins, MORC1, 2, 4, 5, 6, and 7 25 . MORC6 forms stable heterodimers with either MORC1 or MORC2 28 , whereas MORC4 and MORC7 are closely related and redundant proteins that form homodimers 25,34 . MORC5 is understudied because its expression is limited to pollen 32 . MORC6 mutations have been found in several independent forward genetic screens for factors involved in the maintenance of transcriptional gene silencing 24,34,35 . However, despite transcriptional upregulation of DNA methylated genes and transposons, Arabidopsis morc mutations show little effect on the maintenance of DNA methylation patterns. For instance, Moissiard et al. found no methylation changes in morc mutants either at the reactivated SDC transgene locus used for the mutant screen, or genome wide 34 . Lorkovic et al. found a moderate reduction of methylation in morc6 at the 35S promoter in their screen 35 and Brabbs et al. also reported moderate methylation reduction in the transgene used in their screen, but no methylation changes at some other morc6 reactivated loci 24 . Similarly, Jing et al. identified morc6 as a methylation independent repressor in their genetic screen 36 . A genome wide analysis of methylation patterns in a morc hextuple mutant (morc1 morc2 morc4 morc5 morc6 morc7, hereafter termed hex) eliminating all functional MORC genes came to a similar conclusion, as only a few hundred loci showed DNA methylation losses whereas the majority of the genome showed no changes in methylation despite transcriptional reactivation at many loci 25 . Similar but weaker gene expression changes and methylation losses were observed in morc4 morc7 double mutants and morc6 single mutants, indicating functional redundancy of the different MORC genes in Arabidopsis. Together these studies all suggest a role for MORCs in gene silencing, acting at least in part downstream of DNA methylation, but the relationship between MORCs and the DNA methylation pathways remains unclear.
Here we report the first genome-wide localization of MORCs in A. thaliana and find they are primarily localized to RdDM sites. Remarkably, MORC localization to these sites was stably maintained even in the absence of a functional RdDM machinery. However, it was lost in met1 at sites that lost all DNA methylation, and redistributed to newly hypermethylated ectopic RdDM sites. We also found that ectopic recruitment of MORC7 to the FWA promoter was sufficient to cause de novo DNA methylation and transcriptional gene silencing, and this was dependent on functional RdDM. Surprisingly, despite no detectable loss of DNA methylation at the endogenous FWA locus in morc mutants, establishment of methylation on newly integrated FWA transgenes was greatly impaired in the absence of MORCs. This suggests that in addition to the role of MORCs acting downstream of DNA methylation, MORCs play a role in the efficient establishment of DNA methylation over newly integrated transgenes. We propose that MORCs act as chromatin tethers, loaded by the RdDM machinery, to facilitate the retention or re-recruitment of RdDM complexes at chromatin to promote efficient establishment of DNA methylation.

Results
MORC4 and MORC7 colocalize at RdDM sites. To determine the genomic localization of MORC proteins, we performed chromatin immunoprecipitation (ChIP)-seq with pMORC4::MORC4-3xFLAG (MORC4-FLAG) and pMORC7::MORC7-3xFLAG (MORC7-FLAG) expressed in the their corresponding mutant backgrounds as reported previously 25,34 . MORC4 and MORC7 largely co-localized genome wide (Fig. 1a, b, Supplementary Fig. 1a). We identified 3440 and 6119 clear peaks for MORC4 and MORC7, respectively (MACS2, q value < 0.01, fold enrichment >2) with 80% of MORC4 peaks overlapping with MORC7 peaks (Fig. 1a, b, Supplementary Fig. 1b). The remaining 20% of the MORC4 "specific" peaks were also enriched in MORC7 ( Supplementary Fig. 1b), but the level of enrichment did not pass our stringent peak calling threshold. MORC7 displayed stronger peak intensity at the majority of shared target regions. This is consistent with our previous report that MORC4 and MORC7 are close homologs that function largely redundantly to repress gene expression, with MORC7 playing a more dominant role 25 . Therefore, the absence of MORC4 ChIP signal over some MORC7 regions likely reflects the lower expression level of MORC4 and the sensitivity of ChIP. Given the stronger genetic effects as well as ChIP signal of MORC7, we focused on MORC7 for our subsequent analyses.
MORC7 was localized to regions that display strong characteristics of RdDM, including high levels of CHH methylation (Fig. 1a, Supplementary Fig. 1c, right panel) and enrichment over TEs, including the edges of long TEs and the entire length of short TEs (Fig. 1c, d). To further characterize the relationship with RdDM, we compared the binding patterns of MORC7 with that of the essential RdDM component, NRPE1 37 and found that 80% of MORC7 peaks overlap with NRPE1 sites (Fig. 1e, Supplementary Fig. 1a). We defined these overlapping regions as MORC7-NRPE1 common peaks. Interestingly, about 20% of MORC7 peaks were devoid of NRPE1 and 40% of all NRPE1 peaks were devoid of MORC7. We define these two groups as MORC7-unique peaks and NRPE1-unique peaks, respectively (Fig. 1e, Supplementary Fig. 1d). MORC7-unique peaks are devoid of DNA methylation, more proximal to the transcriptional start sites of genes, and displayed higher levels of RNA Pol II signal compared to MORC7-NRPE1 common peaks (Supplementary Fig. 1c, e, f). The functional relevance of MORC7-unique peaks remains unclear.
To reveal the relationship between transcriptional repression and DNA methylation mediated by MORCs, we divided the Log2 Flag/Input  Fig. 1g). Over these 128 MORC7 bound regions, we observed mild loss of DNA methylation in hex compared with nrpe1 ( Supplementary Fig. 1h). Interestingly, although loss of DNA methylation was much less in hex compared to nrpe1, transcriptional activation of these loci was much stronger in hex ( Supplementary Fig. 1i). This result is consistent with the known role of MORCs in silencing downstream of DNA methylation 24,34,36 .
MORC7 is associated with other MORCs and with RdDM components. To determine whether MORC7 associates with RdDM components, we performed immunoprecipitation followed by mass spectrometry (IP-MS) of MORC7-FLAG. Interestingly, we detected MORC4, MORC6, and MORC1 peptides, indicating that these MORC proteins interact in vivo (Table 1). This was in contrast to our previous report that MORC4 and MORC7 formed mutually exclusive homodimers and that MORC6 exclusively interacted with either MORC1 or MORC2 25,28 . This discrepancy likely reflects dramatic improvements in mass spectrometry sensitivity (Q-Executive instrument utilized in the previous study versus the Orbitrap Fusion Lumos used in the current study), and suggests that the interaction of MORC7 with MORC4, MORC6, and MORC1 may be weak or transient. In addition, we identified several RdDM components in the MORC7 IP-MS (Table 1). However, with the exception of DMS3, which was detected at higher abundance, relatively few peptides were detected for other RdDM components. This suggested that MORC7 interactions with RdDM components may be weak or transient. To test this, we adopted a crosslinked nuclear immunoprecipitation and mass spectrometry (CLNIP-MS) 38 assay which stabilizes weak or transient interactions and captures proteins in close proximity in addition to directly interacting protein complexes. As a result, we observed a significant enrichment of both the MORCs and RdDM components (Table 1). Interestingly, without crosslinking, the MORC7 only pulled down components belonging to the Pol V arm of the RdDM pathway, and the CLNIP-MS consistently detected more peptides of Pol V arm components (Table 1). In addition, MORC6 has been reported to interact with DMS3 by in vitro coimmunoprecipitation 35 , and with SUVH9 and SWI3D by IP-MS and yeast two hybrid 36,39 . Together with our observation that MORC7 pulled down peptides of MORC1, 4, 6, and RdDM components in vivo, these data suggest that MORC proteins associate with the Pol V arm of the RdDM pathway, likely through the interaction with RdDM components, for instance, SUVH2/9 and DMS3.
MORC7 can be recruited by the RdDM machinery. Because MORC7 localized to RdDM sites and was associated with RdDM components, we next asked whether MORC7 can be recruited via the RdDM machinery. To test this, we utilized a fusion of DMS3 to an artificial zinc finger (ZF108) that was previously shown to target DMS3 to thousands of ectopic sites throughout the genome 40 . This leads to NRPE1 recruitment to these ectopic sites largely without triggering DNA methylation establishment 15,40 . Thus, if the RdDM machinery is sufficient to recruit MORC7, we expect MORC7 to also follow ZF108-DMS3 to these ectopic loci. We transformed UBQ10::ZF108-DMS3-3XHA into the MORC7-3XFLAG line and performed ChIP-seq for both ZF108-DMS3 and MORC7. We identified 548 ZF108-DMS3 ectopic sites with greater than fivefold enrichment of MORC7 in plants expressing ZF108-DMS3 but displaying no MORC7 signal in control plants ( Fig. 2a, b), indicating that MORC7 can be efficiently recruited by the RdDM machinery to many ectopic sites in the genome.
MORC7 retention on chromatin is dependent on DNA methylation or factor associated with DNA methylation.
Although CHH methylation is eliminated in RdDM mutants at RdDM/MORC7 binding sites, CG and CHG methylation are largely retained 43 . We speculated that this remaining methylation, or another chromatin characteristic associated with methylation, might be sufficient to maintain MORC7. To test this, we performed MORC7 ChIP-seq in met1 which shows complete elimination of CG methylation genome-wide. Due to inter-dependencies of DNA methylation pathways 2 , complete loss of DNA methylation is also observed over many regions 43 . We observed a dramatic reduction of MORC7 ChIP signal throughout the genome in met1 (Fig. 3a). We examined the genome-wide correlation between changes in MORC7 occupancy and DNA methylation in met1 (met1/WT) and observed a clear positive correlation, indicating that regions that retained more DNA methylation in met1 also retained more MORC7 (Fig. 3b). Importantly, regions that lost DNA methylation completely in all sequence contexts in met1 also lost all MORC7 binding (Fig. 3c,  d). Together, these data indicate that MORC7 localization depends on DNA methylation or a factor associated with DNA methylation. MORC7 is unlikely to be recruited by DNA methylation directly because it is not present at other methylated sites, for instance at gene bodies, that are only methylated at CG sites. Instead it seems more likely that histone marks or variants associated with heterochromatin might be involved in the retention of MORC7.
MORC7 is recruited to ectopic RdDM sites in met1. Interestingly, we observed ectopic gain of MORC7 over certain regions in met1. Some of these sites corresponded to reactivated TEs (Fig. 4a). Expression of TEs may trigger non-canonical RdDM through the Pol II-RDR6 pathway, leading to the synthesis of siRNAs and the establishment of RdDM 22,44,45 . We observed 903 ectopically gained MORC7 peaks in met1, and found that these peaks were also associated with hyper-methylation in both CHG and CHH sequence contexts in met1 (Fig. 4b), suggesting that these ectopic MORC7 peaks are likely a consequence of ectopic RdDM activity. For example, SUPERMAN is a well-characterized non-TE locus that becomes targeted by RdDM ectopically in met1 46 . As expected, we observed an ectopic gain of MORC7 over the promoter of SUPERMAN along with non-CG hyper-methylation in met1 (Fig. 4a) 47 . Collectively, these observations further support our conclusion that MORC7 can be recruited by the RdDM machinery.
ZF108-MORC7 represses FWA expression through RdDM. FWA is normally methylated and silent 48,49 . However, the fwa-4 epiallele that has stably lost DNA methylation is highly transcribed, leading to a heritable late flowering phenotype. We previously found that targeting MORC6 or MORC1 to FWA in an fwa-4 background using ZF108 was able to trigger RdDMdependent de novo DNA methylation and silencing of FWA 40 .
To test whether MORC7 can also silence FWA, we fused ZF108 to MORC7 and transformed the fusion protein into fwa-4. Indeed, we observed an early flowering phenotype in T1 plants carrying UBQ10::ZF108-MORC7 indicating transcriptional silencing of FWA. We examined three early flowering T1 plants and confirmed partial repression of FWA by RT-qPCR and an increase in  DNA methylation over the FWA promoter (Fig. 5a), as well as ectopic establishment of DNA methylation over some of the ZF ectopic sites (Supplementary Fig. 3a). We also followed these three early flowering T1 plants to the next (T2) generation and observed even earlier flowering (Fig. 5b), stronger repression of FWA, and a higher level of DNA methylation over the FWA promoter (Fig. 5c).
To test whether MORC7 is able to repress gene expression in the absence of RdDM, we transformed ZF108-MORC7 into nrpe1 introgressed into the fwa-1 background (fwa-1 nrpe1-1) 40 . We did not observe early flowering plants in the T1 generation and failed to detect significant repression of FWA in five independent ZF108-MORC7 expressing T1 lines ( Supplementary Fig. 3b, c). Furthermore, we followed three T1 lines to the T2 generation and confirmed that FWA expression was not repressed (Supplementary Fig. 3d). These results show that MORC7-mediated repression of FWA expression is dependent on RdDM, which is similar to the reported behavior of ZF108-MORC6 40 .
Efficient transgene silencing is impaired in morc mutants. Because MORCs are able to repress FWA gene expression through the recruitment of RdDM, and RdDM is responsible for the establishment of DNA methylation 50 , we asked whether MORCs may be required for the establishment of DNA methylation. To test this, we transformed plants with an FWA transgene, which in wild type is immediately and efficiently targeted for de novo DNA methylation and transcriptional silencing, leading to an early flowering phenotype 50 . However, in strong RdDM mutants, like nrpe1, de novo silencing of the FWA transgene is abolished 16,50,51 and plants remain late flowering due to FWA overexpression. Considering the known functional redundancy of MORC proteins 25 , we transformed the FWA transgene into the hex mutant in which all functional MORC genes are mutated 25 .
In the T1 generation, we observed 59% (22/37) late flowering plants in hex, compared to 88% (29/33) and 3% (1/32) in the nrpe1 and wild-type controls, respectively (Fig. 6a). As expected, late flowering phenotype was associated with FWA transgene overexpression as well as reduced DNA methylation establishment over the FWA promoter (Fig. 6b, c). This indicates that the efficiency of FWA transgene silencing through RdDM is greatly reduced in hex.
To determine if FWA transgene silencing might be established in subsequent generations, we followed four late flowering T1 hex plants to the T2 generation. Two out of the four T2 lines converted to mostly early/medium flowering (T2-2, T2-3) while the other two remained medium/late flowering (T2-1, T2-4) (Fig. 6d). We chose one late flowering plant from each T2 line and one early flowering plant from T2-2 and observed a clear association between the flowering time and the expression and methylation status, with late flowering plants displaying lower levels of DNA methylation and higher levels of FWA expression (Fig. 6e, f). Therefore, although efficient establishment of transgene silencing is strongly impaired in the hex, stochastic intergenerational silencing still occurs within the population.
We also asked whether the stochastic silencing of some lines but not others might be due to the location of the transgene within the genome. To test this, we chose three late flowering plants from the hex T2-1 population, which carries the FWA transgene at a single insertion site as indicated by a 3:1 segregation (hygromycin resistant:sensitive ratio in the T2 was 67:21) and followed these plants to the T3 generation. We observed early flowering in one T3 line (T3-3), while the other two (T3-1 and T3-2) remained mostly late flowering (Fig. 6g). Since all three have the same insertion, the stochastic silencing is likely a property of the hex mutant background, rather than a reflection of the transgene location within the genome. We also performed methylation and expression analysis on one early and three late flowering plants from each of the T3 lines, and consistently found that late flowering plants displayed a lower level of DNA methylation and higher level of FWA expression (Fig. 6h, i). These results show that FWA methylation and silencing failed to establish efficiently in hex in the T3 generation. In contrast to hex, the strong RdDM mutant, nrpe1, showed late flowering T1 plants that remained late throughout all generations tested (Fig. 6a, d, g), displaying no stochastic silencing.
To examine the establishment of DNA methylation specifically over the transgene in more detail, we performed bisulfite PCR (BS PCR) over two regions of the FWA promoter with the same T2 and T3 plants used for  ARTICLE allowed us to distinguish endogenous-and transgene-derived reads. In the CG context, most transgene-derived reads were fully unmethylated in nrpe1, while few fully unmethylated reads were detected in wild type ( Supplementary Fig. 4a), demonstrating that the transgene was efficiently silenced in wild type but not in nrpe1 as expected. hex plants displayed intermediate levels of unmethylated transgene-derived reads, with late flowering hex plants having more fully CG-unmethylated reads than early flowering hex in both T2 and T3 ( Supplementary Fig. 4a, upper panels). We also detected more fully CHH-unmethylated reads in all hex lines over the transgene compared to the endogenous locus, consistent with hex inhibiting RdDM ( Supplementary Fig. 4a lower panels, 4b). We did not analyze CHG methylation given the paucity of CHG sites at FWA. We also found that maintenance of preexisting DNA methylation over the endogenous FWA promoter was completely unaffected in hex, whereas RdDM mutants such as nrpe1 and suvh2 suvh9 lost most non-CG methylation as expected ( Supplementary Figs. 4 and 5). Thus MORCs are required for efficient RdDM-mediated establishment of DNA methylation of newly integrated FWA transgenes, even though maintenance of preexisting DNA methylation at the same FWA sequences at the endogenous gene is completely unaffected in the hex mutant.

Discussion
In this study we show that Arabidopsis MORC7 is localized to sites of RdDM, and likely associates with the Pol V arm of the RdDM pathway. Tethering of MORC7 to the unmethylated FWA locus with an artificial zinc finger was sufficient to recruit RdDM, methylation, and silencing to FWA. Conversely, tethering DMS3 to ectopic sites in the genome using an artificial ZF was sufficient to recruit MORC7. Furthermore, MORC7 was recruited to ectopic RdDM sites in met1. Together, these data suggest a mutual interaction and recruitment between MORC7 and RdDM. Finally, we found that a morc hextuple mutant is defective in the    The MORC gene family in Arabidopsis consists of six potentially active genes, and several pieces of evidence suggest that the other MORCs have similar properties as we have observed here for MORC7. First, MORC4 was found to co-localize with MORC7 over RdDM sites. Second, our IP-MS and crosslink IP-MS show interactions between MORC7 and MORC4, MORC1 and MORC6, suggesting they likely function together. Third, consistent with our IP-MS data showing MORC7 association with Pol V arm RdDM components, MORC6 was previously shown to interact with two Pol V arm components, DMS3 and SUVH9 35,39 . Fourth, similar to ZF108-MORC7, ZF108-MORC1 and ZF108-MORC6 were previously shown to recruit RdDM to the FWA locus 40 , consistent with an interaction between MORC complex and the RdDM machinery. Finally, comparison of hex to morc4 morc7 and morc6 showed genetic redundancy between these family members 25 .
Given their association with RdDM components and their localization to RdDM sites, the MORCs are clearly acting in some capacity as RdDM factors. However, relative to known RdDM factors, the MORCs have unique genetic properties, in that they show strongly impaired efficiency of DNA methylation and silencing of transgenes, yet show very little effect on the maintenance of preexisting methylation at RdDM sites. All RdDM mutants that have been studied to date, including very weak mutants such as frg1 frg2, suvr2, and drm3 show significant losses of methylation at the majority of RdDM sites 52 . However, an analysis of the morc hex mutant showed only a very small number of sites with any loss of methylation. Notably, although the morc hextuple mutant shows greatly impaired DNA methylation establishment of FWA sequences when new transgenes are introduced into plants, the morc hextuple has no defect in the maintenance of preexisting methylation of similar sequences at the endogenous FWA gene, while all other known RdDM mutants do (Supplementary Fig. 5). Another interesting difference is that while other RdDM components studied dissociate from chromatin in the absence of a functional RdDM machinery (e.g., Pol V dissociates from chromatin in the absence of the DDR complex, Pol IV is lost in the absence of SHH1, and AGO4 is lost in the absence of DCL3 or RDR2), MORC7 maintenance on chromatin was not disrupted in RdDM mutants.
To account for these unique properties, we suggest a model (Fig. 7a) in which MORCs are loaded onto sites of RdDM via interaction with RdDM components, after which they stay stably associated with these sites. One possible mechanism is that the Arabidopsis MORC proteins may be able to topologically entrap the chromatin at these sites by clamping around the DNA, similar to what has been shown for C. elegans MORC-1 27 . In this way, MORCs may serve as "tethers", or a memory component, to facilitate the continuous recruitment of RdDM complexes back to sites after their initial targeting by RdDM. This tethering process may be mainly important during the early stages of the establishment of RdDM, which would explain why MORCs are required for efficient establishment of methylation at FWA, but not for maintenance of preexisting methylation at these same sequences. After initial establishment of methylation, and in the presence of strong maintenance of DNA methylation activity, MORCs may no longer be required for continued RdDM recruitment because of factors such SUVH2 and SUVH9 which bind to methylated DNA to continually recruit Pol V, and SHH1 which binds to H3K9 methylation to recruit Pol IV 9,10,15 . An RdDM tethering model could also explain the effects of morc mutants on maintenance DNA methylation at a small subset of sites. A previous analysis of the morc hextuple mutant identified 519 sites, representing around 5% of RdDM sites, showing losses of methylation 25 . These "fragile sites" display a loss of DNA methylation as well as transcriptional activation in the absence of MORCs, as well as in RdDM mutants 25 . These fragile sites were shown to possess lower levels of symmetrical CG and CHG sequences compared to the genome average and thus are likely to show weakened CG and CHG maintenance DNA methylation 25 . We suggest that these sites may be fragile to the loss of MORCs because of transcriptional reactivation in the morc mutants, leading to a need for continued recruitment of RdDM to these sites in the face of weaker than average reinforcing maintenance of DNA methylation at symmetrical sites. MORC tethering to these sites may provide the needed boost in RdDM recruitment to maintain methylation and silencing.
Our findings on the biology of the Arabidopsis MORCs could have implications for the study of MORCs in other organisms. For instance mouse MORC1 is required for de novo DNA methylation and silencing of evolutionarily young and expressed TEs that are more recalcitrant to de novo methylation over the course of epigenetic reprogramming in primordial germ cells 29 . A MORC tethering model could nicely explain the continued recruitment of silencing complexes to these TEs during the establishment process. Similarly, human MORC2 is recruited by the HUSH complex to H3K9me3 marked heterochromatin and loss of MORC2 results in a decrease in H3K9me3 levels and transcriptional activation at a subset of HUSH loci 30 . Here again, MORC2 could act as a chromatin tether to promote the retention of the HUSH complex at its targets. Thus while MORCs in different organisms are clearly targeted to chromatin by different binding partners, the underlying mechanism of MORCs acting as tethers to assist silencing complexes may be conserved.
Plasmid construction ZF-DMS3-3xHA. Modified pDT1H plasmid containing UBQ10 and C-terminal 3xHA 53 was linearized by AatI digestion. ZF108 was fused to the N-terminus of DMS3 (cDNA) through overlapping PCR and subsequently ligated into linearized modified pDT1H with In-Fusion cloning system (Takara). Cloning primers are listed in Supplementary Table 1.
All plants in this study were grown under long day conditions (16 h light/8 h dark). Transgenic plants were generated by Agrobacterium-mediated transformation through floral dipping. T1 transgenic plants were selected with hygromycin B on ½ MS agar medium or sprayed with Basta in soil.
Crosslinked IP was performed with 15 g of floral tissues and resuspended in Nuclei Isolation Buffer (50 mM HEPES, 1 M sucrose, 5 mM KCl, 5 mM MgCl 2 , 0.6% Triton X-100, 0.4 mM PMSF, 5 mM benzamidine, cOmplete EDTA-free Protease Inhibitor Cocktail (Roche)) to reach a final volume of 40 ml and supplemented with 1% formaldehyde for 12 min with rotation. Glycine was added immediately to stop the crosslinking. Clumps were broken by Dounce homogenizer and lysate was filtered through Miracloth and centrifuged at 1500 g for 10 min at 4 • C. Nuclei pellet was resuspended and washed with NRBT buffer (20 mM Tris-HCl pH 7.5, 2.5 mM MgCl 2 , 25% glycerol, 0.2% Triton X-100) twice and resuspended in 6 ml of RIPA buffer (1x PBS, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS). Resuspended nuclei were split into 3x 2 ml aliquots for sonication for 20 min (30 s on/30 s off) with Bioruptor Plus (Diagenode). Sheared lysate was centrifuged at 8000 g for 15 min at 4 • C and combined supernatant was incubated with 200 ul of FLAG-M2 magnetic beads (50% slurry, Sigma) for 2 h at 4 • C with rotation. Beads were washed, eluted, and precipitated as described in native IP.
The TCA precipitated samples were resuspended in 50 μl of digestion buffer (8 M urea, 100 mM Tris pH 8.5). Each sample was reduced and alkylated by adding TCEP and iodoacetamide to final concentrations of 5 mM and 10 mM, incubated at room temperature in the dark for 20 min, and then digested by 0.1 μg of Lys-C (Thermo Scientific, 90051) and 0.8 μg Trypsin (Thermo Scientific, 90057) proteases at 37°C overnight. The digested samples were quenched by the addition of formic acid to 5% (v./v.) final concentration. Finally, each sample was desalted via C18 tips (Thermo Scientific, 87784) and reconstituted in 15 μL of 5% formic acid before analyzed by LC-MS/MS.
Digested peptides were resuspended in 5% formic acid and fractionated online using a 25 cm long, 75 uM inner diameter fused silica capillary packed in-house with bulk C18 reversed phase resin (length, 25 cm; inner diameter, 75 uM; particle size, 1.9 μm; pore size, 100 Å; Dr. Maisch GmbH). The 140-minute water-acetonitrile gradient was delivered using a Dionex Ultimate 3000 UHPLC system (Thermo Fisher Scientific) at a flow rate of 300 nl/min (Buffer A: water with 3% DMSO and 0.1% formic acid and Buffer B: acetonitrile with 3% DMSO and 0.1% formic acid). Fractionated peptides were ionized and analyzed by tandem mass spectrometry (MS/MS) Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific). Data was acquired using a Data-Dependent Acquisition (DDA) method comprised of a full MS1 scan (Resolution = 120,000) followed by sequential MS2 scans (Resolution = 15,000) to utilize the remainder of the 3 second cycle time.
Data analysis including peptide and protein identification was performed using MS2 spectra were searched using the ProLuCID algorithm against Arabidopsis reference proteome followed by filtering of peptide-to-spectrum matches (PSMs) by DTASelect using a decoy database-estimated false discovery rate of <1%.

Bioinformatic analysis
ChIP-seq. All libraries were sequenced at a length of 50 bps with HiSeq 2500 or NovaSeq 6000 platforms following manufacturer's instructions (Illumina). Raw reads were aligned to the Arabidopsis reference genome (TAIR10) with Bowtie2 (v2.1.0) 54 , allowing only uniquely mapped reads with perfect matches. Duplicated reads were removed with Samtools (v1.9) 55 . Peaks were called using MACS2 (v2.1.1) 56 . To increase sequencing depth, three independent ChIPs of MORC7 in wild-type and two independent MORC7 ChIPs in RdDM mutants were pooled for peak calling.
Whole Genome Bisulfite Sequencing (BS-seq) and analysis. BS-seq reads were mapped to TAIR10 reference genome by bsmap (v2.90) with allowing 2 mismatches and 1 best hit (-v 2 -w 1) 57 . Reads with three or more consecutively methylated CHH sites were considered as non-converted reads and removed from the analyses. DNA methylation levels were calculated by #C/ (#C + #T). Differential Methylated Regions (DMRs) were called by methdiff function with every 100 bp bin for where the difference in CG, CHG, and CHH methylation is at least 0.4, 0.2, and 0.1, respectively. met1 RNA seq analysis. met1 RNA seq data were downloaded from NCBI Gene Expression Omnibus (GEO) as accession GSE93584 58 . Cleaned short reads were aligned to reference genome TAIR10 by Bowtie2 (v2.1.0) 54 , and expression abundance was calculated by RSEM (v1.3.1) with default parameters 59 .
Bisulfite PCR analysis. Only reads with sequences matching primers used for bisulfite PCR were used for analysis. Primer sequences were trimmed off reads and reads were aligned to a 'genome' consisting of the target amplicons (both wild-type and transgene sequences with polymorphisms) using bismark 60 . Fully unmethylated reads were extracted and quantified for each amplicon using a custom script (available upon request).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Data supporting the findings of this work are available within the paper and its Supplementary Information files. The datasets generated and analyzed during the current study are available from the corresponding author upon request. All high-throughput sequencing data generated in this study is accessible at NCBI's Gene Expression Omnibus (GEO) via GEO Series accession number GSE160285. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD026674. Source data underlying Figs. 5 and 6 are provided as a Source Data file. Souse data are provided with this paper. Source data are provided with this paper.