The stochastic silencing phenotype of Arabidopsis morc6 mutants reveals a role in efficient RNA-directed DNA methylation

Authors


For correspondence (e-mail louise.jones@york.ac.uk).

Summary

The RNA-directed DNA methylation (RdDM) pathway is of central importance to the initiation and maintenance of transcriptional gene silencing in plants. DNA methylation is directed to target sequences by a mechanism that involves production of small RNAs by RNA polymerase IV and long non-coding RNAs by RNA polymerase V. DNA methylation then leads to recruitment of histone-modifying enzymes, followed by establishment of a silenced chromatin state. Recently MORC6, a member of the microrchidia (MORC) family of adenosine triphosphatases (ATPases), has been shown to be involved in transcriptional gene silencing. However, reports differ regarding whether MORC6 is involved in RdDM itself or acts downstream of DNA methylation to enable formation of higher-order chromatin structure. Here we demonstrate that MORC6 is required for efficient RdDM at some target loci, and, using a GFP reporter system, we found that morc6 mutants show a stochastic silencing phenotype. By using cell sorting to separate silenced and unsilenced cells, we show that release of silencing at this locus is associated with a loss of DNA methylation. Thus our data support a view that MORC6 influences RdDM and that it is not acting downstream of DNA methylation. For some loci, efficient initiation or maintenance of DNA methylation may depend on the ability to form higher-order chromatin structure.

Introduction

In eukaryotes, small RNA pathways are involved in directing a diversity of transcriptional and post-transcriptional regulatory events (Vaucheret, 2006; Ghildiyal and Zamore, 2009; Castel and Martienssen, 2013). In all of these pathways, small RNAs must be generated from endogenous or exogenous templates and guide effector complexes to complementary targets. In plants, one such well-studied pathway is RNA-directed DNA methylation (RdDM), which has a role in maintaining DNA methylation of transposons and repeats across the genome (Chan et al., 2006; Huettel et al., 2006; Zhong et al., 2012).

The mechanism of RdDM may be divided into three phases: generation of small interfering RNAs (siRNAs), recruitment of DNA methyltransferases to target sequences, and reinforcement of DNA methylation patterns (Law and Jacobsen, 2010). As RdDM results in transcriptional silencing, there are also likely to be changes in chromatin structure occurring as a downstream consequence. Generation of siRNAs involves the plant-specific RNA polymerase IV (Pol IV) (Herr et al., 2005; Onodera et al., 2005; Pontier et al., 2005; Pikaard et al., 2008). Pol IV produces RNA templates that are converted into double-stranded RNA by the RNA-dependent RNA polymerase RDR2, and processed into 24-nucleotide siRNAs by DICER-like 3 (DCL3) (Xie et al., 2004; Haag et al., 2012). A SNF2-like chromatin remodelling factor, CLASSY1 is also involved in this first phase, possibly by aiding transcription by Pol IV (Smith et al., 2007). The siRNA-generating pathway may also be supplied by RNA polymerase II transcripts at certain inverted repeat loci, and is influenced by the DNA methylation status of the genomic template (Zhang et al., 2007; Eamens et al., 2008; Mosher et al., 2008). The siRNAs are thought to act as sequence guides for the DNA methylation step, and are bound by Argonaute (AGO) effector proteins. Of the ten Arabidopsis AGO proteins, AGO4, AGO6 and AGO9 are involved in RdDM, with AGO4 having the major role (Havecker et al., 2010; Eun et al., 2011).

The DNA methylation phase of RdDM requires that the AGO–siRNA complex and the de novo methyltransferase DRM2 be recruited to target loci. A key step in this recruitment is production of long non-coding RNAs (lncRNAs), predominantly by plant-specific RNA polymerase V (Pol V) but also by Pol II at certain loci (Wierzbicki et al., 2008; Zheng et al., 2008; Wierzbicki, 2012). The production of Pol V-dependent lncRNAs requires the chromatin-remodelling protein DRD1, the chromosome hinge domain protein DMS3, and the protein RDM1 that binds single-stranded methylated DNA in vitro (Kanno et al., 2005, 2008; Law et al., 2010). These proteins form the so-called DDR complex that may facilitate Pol V transcription. The lncRNAs are proposed to act as scaffolds for recruitment of the RdDM machinery via RNA–RNA pairing between the DCL3-generated siRNAs and the Pol V transcripts, and by interaction between AGO4 and the ‘AGO hook’ domain of NRPE1, the largest subunit of Pol V. The ‘AGO hook’ domain is also found in KTF1/SPT5L, a transcription factor-like protein that is required for RdDM and that may act together with AGO4 in recruitment of chromatin-modifying enzymes (Bies-Etheve et al., 2009; He et al., 2009; Rowley et al., 2011). DRM2 is the de novo DNA methyltransferase that acts in RdDM and has been shown to co-localize with AGO4 and RDM1 in the nucleus, presumably as part of a complex that is active in the actual methylation step of RdDM (Li et al., 2008). The SRA domain proteins SUVH2 and SUVH9 act with partial redundancy at several RdDM loci, with mutants displaying reduced DNA methylation with little or no effect on siRNA accumulation (Johnson et al., 2008). The SRA domain binds methylated DNA and therefore may function to maintain the methylated status of loci targeted by RdDM. Several histone-modifying proteins are also involved in RdDM, including the histone deacetylase HDA6 and the Jumonji C protein JMJ14 (Aufsatz et al., 2002; Deleris et al., 2010; Searle et al., 2010). Both of these are likely to be involved in removing active marks from targeted loci; however, mutants in these genes also affect DNA methylation, indicating that their role is not simply downstream of RdDM. Clearly there is a feedback mechanism whereby removal of active histone marks and laying down of inactive marks is required to establish high levels of DNA methylation via RdDM.

In the present study, Arabidopsis mutants defective in RdDM that show an unusual stochastic silencing pattern were investigated. These mutants were found to all be defective in MORC6, a member of the microrchidia (MORC) family of ATPases. A role for MORC family members in RdDM-related silencing events has been demonstrated in two other recent publications (Lorković et al., 2012; Moissiard et al., 2012). However, whereas Moissiard et al. (2012) propose that MORCs act in formation of higher-order chromatin structure, Lorković et al. (2012) suggest that MORC6 may act in concert with the DDR complex and therefore play a role in the DNA methylation step itself. Here we examine the effects of mutations in the MORC6 ATPase on RdDM and silencing using a previously described dual reporter system that is based on targeting of a 35S:GFP reporter gene. This system has previously been characterized as requiring both Pol IV and Pol V for target silencing and DNA methylation (Eamens et al., 2008). We show that morc6 mutants give rise to a mixed silencing phenotype, with cells either showing release of silencing and loss of DNA methylation at the target locus or retaining DNA methylation and silencing. Thus we provide evidence that MORC6 influences the efficiency of DNA methylation at certain target loci in a stochastic manner, and that its role is not solely downstream of RdDM.

Results

Release of transcriptional gene silencing by mutation of MORC6

A forward genetic screen was performed using a dual transgene silencing system in which an inverted repeat of 35S promoter sequences directs DNA methylation and transcriptional silencing of a 35S-driven GFP reporter gene (Eamens et al., 2008). The line carrying the two transgenes was described previously as line 3–7 but is henceforth referred to as line 142S. The line carrying only the 35S:GFP transgene is referred to as 142 (Dalmay et al., 2000). Ethyl methanesulfonate mutagenesis resulted in a number of mutants in which GFP expression was re-activated, presumably either due to defects in RNA-directed DNA methylation itself or in formation of transcriptionally silent chromatin at the 35S promoter. Through allelism tests, three independently generated mutants (M1, M6 and M9) were identified as forming a complementation group, and were therefore assumed to carry mutations in the same gene (Figure 1a and Figure S1a). M6 was identified after M1 and M9; it originally had a linked stunted growth phenotype that was later removed by back-crossing, therefore the majority of the further characterization was performed on M1 and M9 only. GFP expression was confirmed by Northern blot analysis (Figure 1b and Figure S1b). For analysis of M1 and M9, two time points of 7 days post-germination (dpg) and 21 dpg were used. This was based on visual inspection of the fluorescence phenotype, suggesting that the mutants had reduced levels of GFP expression at 21 dpg compared to 7 dpg. At 7 dpg, the levels of GFP expression in the M1 and M9 mutants were comparable to those of line 142 that carries the 35S:GFP target transgene but lacks the silencer transgene. At 21 dpg, the levels of GFP expression in the mutants were reduced compared to line 142 but were still significantly higher than observed in line 142S (Figure 1b). The rmd3–1 mutant, which is defective in the NRPD1 subunit of Pol IV, was included as a control for reactivation of GFP expression in line 142S (Eamens et al., 2008).

Figure 1.

Release of gene silencing by mutations in MORC6. (a) Leaf images taken under UV light of representative first leaves of 142 and 142S wild-type plants and M1, M6 and M9 mutants at 21 dpg. (b) RNA gel-blot analysis of GFP and ACTIN (ACT) mRNA levels in non-transgenic (NT), 142, 142S, M1, M9 and rmd3–1 samples taken at 7 and 21 dpg. (c) Schematic representation of the location of the mutations within the At1 g19100 locus and the AtMORC6 protein. The motifs I–IV within the GHKL domain are indicated, as is the C–terminal coiled coil domain. (d) RT–PCR analysis of MORC6 and ACTIN transcripts in morc6–5, morc6–6 and morc6–7 mutants and wild-type line 142S. No amplification was observed in reactions that lacked reverse transcriptase.

Following two rounds of back-crossing of the mutant line M1 to the parental line 142S, deep sequencing was performed on a pooled sample of 100 F2 seedlings that exhibited green fluorescence and are therefore expected to carry the defective gene (Methods S1). The SHORE and SHOREmap analysis pipelines (Ossowski et al., 2008; Schneeberger et al., 2009) were then used as a basis for identifying potential causative mutations. A region on the left arm of chromosome 1 was enriched for homozygous single-nucleotide polymorphisms (SNPs) compared to the parental line 142S and the C24 reference genome (Table S1). This region corresponded to a rough map position that had been obtained previously for M1 and M9 using CAPS and SSLP markers, and included a C→T transition in codon 41 of At1 g19100 that results in a coding change from glutamine to a stop codon. At1 g19100 was sequenced in M6 and M9, and C→T transitions were identified in codons 392 and 267, respectively (Figure S2a). Both of these changes also convert glutamine to a premature stop codon. The identification of mutations in At1 g19100 in M1, M6 and M9 provides strong evidence to support these being the causative mutations responsible for reactivation of GFP expression. Allelism tests between M1 and an At1 g19100 T–DNA knockout line (atmorc6–3) confirmed that At1 g19100 was the defective gene (Figure S2b). Indeed, a requirement for At1 g19100 in transcriptional gene silencing has been recently demonstrated in two other forward genetics screens (Lorković et al., 2012; Moissiard et al., 2012). At1 g19100 encodes MORC6, a member of the MORC ATPase family, of which there are seven in the Arabidopsis genome. Following on from the numbering given to the previously described alleles, M1, M6 and M9 are referred to as morc6–5, morc6–6 and morc6–7, respectively. The MORC6 protein has a GHKL ATPase domain, and enzymatic activity has been confirmed for the wild-type protein (Dutta and Inouye, 2000; Lorković et al., 2012). It has four identifiable motifs within this domain that, based on the results of other studies, are involved in ATP binding (Bergerat et al., 1997). There is also a C–terminal coiled coil domain, suggestive of protein–protein interactions. From the position of the point mutations in morc6–5, morc6–6 and morc6–7, we suggest that morc6–5 is likely to result in full loss of function, whereas morc6–6 may retain some activity (Figure 1c). If translated, MORC6–5 would lack the entire ATPase domain, whereas MORC6–7 would retain three of the four ATPase motifs and MORC6–6 would retain all four. All mutants would lack the coiled-coil domain. Although protein expression has not been confirmed, morc6 mRNA is detectable in the mutants using RT–PCR, albeit at a lower level than in the wild-type parental line 142S (Figure 1d).

Stochastic and cell-autonomous silencing in the morc6 mutants

Based upon visualization using a UV-dissecting microscope, it appeared that, although GFP was expressed in the morc6 mutants, it was not ubiquitous over the leaf surface (Figure 1a). In order to investigate this in more detail, GFP expression in parental lines (142 and 142S) and morc6 mutants was visualized using confocal microscopy at various stages of vegetative development. In this analysis, the epidermal cells of the abaxial and adaxial leaf surfaces were imaged. Line 142, which lacks the 35S inverted repeat silencer transgene, exhibits ubiquitous GFP expression, whereas line 142S is almost entirely silenced, with only the occasional GFP-expressing cell (Figure 2 and Figure S3). This pattern of GFP expression was consistent in the parental lines throughout development, although a low level of GFP expression was observable in cotyledons of line 142S. Interestingly, over time, the leaves of the morc6 mutants developed a mosaic pattern of epidermal GFP expression, with some cells expressing GFP and others not. The number of GFP-expressing cells observed fell over time, indicating that the mutants were able to establish GFP silencing but did so in a delayed manner. The distribution of GFP-expressing cells appeared stochastic, with no obvious pattern, although the number of GFP-expressing cells on the adaxial surface was higher than on the abaxial surface (Figure 2 and Figure S3). Cells lacking GFP expression were more numerous in younger leaves; however, GFP-fluorescing cells remained present in leaves of all ages (Figure 2). From the confocal analysis, it appeared to us that GFP-expressing cells were more numerous in line morc6–5 compared to morc6–7, and that the onset of silencing occurred earlier in morc6–7. This suggests that morc6–5 is the more severe allele, and this is consistent with the position of the respective stop codons within the gene (Figure 1c). In order to quantify the degree of silencing in morc6–5 compared to morc6–7, protoplasts were isolated from leaves of the mutants and parent lines 142 and 142S, and flow cytometry used to identify the number of GFP-positive and GFP-negative cells (Methods S1). In order to validate the flow cytometry protocol, protoplasts were first counted in a 1:1 mixed sample from lines 142 (GFP-positive) and 142S (GFP-negative). Table 1 shows that the proportion of GFP-positive and GFP-negative protoplasts from the mixed sample was 48:52, therefore indicating that we correctly identified GFP-positive and GFP-negative protoplasts. For line 142, the majority of the protoplasts were identified as GFP-positive. For line 142S, 97% of protoplasts were GFP-negative and 3% were GFP-positive. This confirms our confocal studies in which the occasional GFP-positive cell is observed in the silenced line. For morc6–5, the majority (83%) of protoplasts were GFP-positive, whereas for morc6–7, the levels of GFP-positive and GFP-negative protoplasts were approximately equal (53:47) (Table 1). This analysis therefore supports the observations from confocal microscopy suggesting that morc6–5 results in a stronger mutant phenotype than morc6–7.

Table 1. Proportion of GFP-positive and GFP-negative protoplasts in samples prepared from wild-type lines 142 and 142S, a 1:1 mixture of 142 and 142S, and mutants morc6–5 and morc6–7
Sample% GFP-positive protoplasts% GFP-negative protoplastsNumber of protoplasts counted
142 + 142S (1:1)48.2751.733681
14299.920.0811 010
142S2.9197.09278 406
morc6–5 82.6417.36980 500
morc6–7 53.0346.97429 000
Figure 2.

Mosaic silencing phenotype of morc6 mutants. Confocal microscope images showing GFP expression in epidermal cells of representative leaves of 142, 142S, morc6–5 and morc6–7 plants. Leaf surface, leaf number and days post-germination are indicated. For clarity, only the GFP channel is shown. Figure S3 shows the same images but with both GFP and propidium iodide/chlorophyll channels visible. Scale bar = 100 μm.

Loss of DNA methylation associated with the morc6 mutation

From these observations, we concluded that release of silencing in the morc6 mutants is not complete, and that, given the lack of obvious patterning, stochastic processes are involved. GFP expression was cell-autonomous, with silenced cells identified that were surrounded by expressing cells and vice versa (Figure 2). We did not observe any significant reduction in 35S siRNA levels in morc6–5 and morc6–7 compared to line 142S (Figure S4), therefore release of transcriptional silencing was unrelated to siRNA biosynthesis and is likely to either be a consequence of a failure to methylate the DNA target or to form a chromatin structure that is inhibitory to transcription. In recent publications, Moissiard et al. (2012) suggest that AtMORC6 has a role in forming higher-order chromatin structures, whereas Lorković et al. (2012) provide evidence for a role in the DDR complex that is required for scaffold transcript production by Pol V. In order to shed light on these discrepancies and to understand the nature of the mosaic silencing phenotype observed, we used bisulfite sequencing to investigate whether DNA methylation of the targeted 35S promoter sequence was affected in the morc6 mutants. Bisulfite sequencing was performed on samples taken at 7 and 28 dpg. The region examined included the region targeted by the 35S inverted repeat (Figure 3) and a downstream non-targeted region. The downstream non-targeted region was unmethylated in all samples.

Figure 3.

Reduction in 35S target promoter DNA methylation in morc6 mutants. Results of bisulfite sequencing of the target 35S promoter region in samples taken from lines 142 (28 dpg), 142S, morc6–5 and morc6–7 (7 and 28 dpg). A minimum of ten clones per sample were sequenced and analysed using CyMATE (Hetzl et al., 2007). Green triangles, blue squares and red circles represent cytosines in CHH, CHG and CGN configurations, respectively, with filled shapes being indicative of DNA methylation at that site. The region shown is that targeted by the 35S inverted repeat. Sequencing reads in samples from 142S, morc6–5 and morc6–7 that are indicative of RNA-directed DNA methylation are indicated by asterisks. Ambiguities in the sequence data are indicated as gaps.

Compared to parental line 142S, a significant reduction in cytosine methylation in all sequence contexts (CG, CHG and CHH) was observed in morc6–5 and morc6–7 at both time points (Figure 3). However, examination of the sequence reads of the mutants indicated that, although methylation had been lost on the majority of strands, it had been retained on others. For morc6–5, three of ten reads at 7 dpg and five of 19 reads at 28 dpg were found to have levels of unconverted cytosines that were indicative of RdDM. For morc6–7 the values were four of ten reads at 7 dpg and two of ten reads at 28 dpg. Given the distribution of the methylated cytosines, and in comparison with the results obtained for lines 142 and 142S, this is unlikely to be due to a failure in the bisulfite conversion protocol. These results are therefore indicative of a mixed population of methylated and unmethylated DNA strands, and suggests that the mosaic silencing phenotype observed under the confocal microscope is a consequence of some cells having undergone RdDM of the 35S promoter and others not having done so. It should be noted that the percentage of methylated versus unmethylated strands in the morc6 mutants was lower than that predicted from the percentage of silenced versus unsilenced cells as determined by flow cytometry (Table 1). Clearly there was a large difference in sample size between the protoplast counts and sequencing reads. Additionally, although the primers used to generate the bisulfite sequencing data were designed to amplify both methylated and unmethylated templates, a 40-cycle PCR and plasmid cloning step were used, which may have introduced some bias. The pattern of DNA methylation observed in the wild-type 142S line showed significant variation between reads, and included the occasional apparently fully unmethylated read. In addition, the level of DNA methylation retained on some strands in the morc6–5 or morc6–7 mutants was considered to be less extensive than that observed for methylated strands in line 142S. From this observation, we conclude that the lower level of DNA methylation retained in morc6–5 and morc6–7 is still sufficient to enable silencing to take place in a percentage of cells.

These data strongly suggest that the morc6 mutants have a reduction in RdDM efficiency at this locus, and that it is loss of DNA methylation that is driving the regain of GFP expression rather than a downstream effect relating to chromatin structure. In order to clarify this issue, we investigated whether GFP expression correlates with loss of DNA methylation. Protoplasts were prepared from morc6–7 leaf tissue at 28 dpg, and cell sorting was used to separate GFP-positive and GFP-negative protoplasts (Methods S1). The morc6–7 mutant was chosen for this analysis as it had roughly equal numbers of GFP-positive and GFP-negative cells as previously determined by flow cytometry (Table 1). RNA samples were prepared from the sorted protoplasts, and quantitative reverse-transcriptase–PCR was used to assess GFP and MORC6 mRNA levels. As expected, GFP mRNA was considerably more abundant in the GFP-positive protoplasts compared to the GFP-negative protoplasts (Figure 4a). This indicated that the cell-sorting protocol successfully separated the protoplasts according to GFP expression. MORC6 mRNA levels were approximately equal between the two protoplast groups, thus ruling out variations in residual MORC6 activity as being responsible for the heterogeneous GFP expression (Figure 4b). Bisulfite sequencing was then performed on DNA samples extracted from the sorted protoplasts to assess the DNA methylation status of the 35S promoter. For the GFP-positive protoplasts, the bisulfite reads showed full or near-full conversion, indicative of unmethylated DNA, whereas, for the GFP-negative protoplasts, the majority of reads contained a proportion of unconverted cytosines, indicative of DNA methylation (Figure 4c). Consistent with the previous analysis with the morc6–5 and morc6–7 mutants in whole leaf samples (Figure 3), the level of methylation observed in the GFP-negative protoplasts was variable.

Figure 4.

Correlation between loss of DNA methylation and release of silencing in line morc6–7. (a,b) Relative GFP (a) and MORC6 (b) transcript abundance in protoplasts prepared from morc6–7 leaf tissue at 28 dpg, and sorted using a MoFlo Astrios cell sorter (Beckman Coulter) based on green fluorescence. ACTIN mRNA levels were used as a control for quantification, and data are presented relative to the levels in the GFP-negative protoplasts. Error bars represent 1 SE. Reactions were performed in triplicate. (c) DNA methylation profile of 35S target promoter sequence in samples prepared from GFP-positive and GFP-negative sorted protoplasts. A minimum of 15 clones per sample were sequenced and analysed using the CyMATE tool (Hetzl et al., 2007). Green triangles, blue squares and red circles represent cytosines in CHH, CHG and CG configurations, respectively, with filled shapes being indicative of DNA methylation at that site. The region shown is that targeted by the 35S inverted repeat.

The results suggest that, although MORC6 is not absolutely required for RdDM at this locus, it influences its occurrence, and loss of methylation is linked to release of silencing. This result is in contrast to the observations of Moissiard et al. (2012) who reported release of silencing without significant loss of DNA methylation at several loci in morc6 mutants. This discrepancy is suggestive of locus-specific effects and suggests that initiation or maintenance of RdDM at specific loci may be influenced by higher-order chromatin factors. Attempts to identify the genomic location of the 35S:GFP target transgene have unfortunately been unsuccessful. Data from deep sequencing suggest that the locus is complex and contains vector backbone sequence. It is currently unknown to what extent these features influence the requirement for MORC6 in RdDM at this locus.

Endogenous targets of RdDM were also examined in morc6–5, morc6–6 and morc6–7 mutant plants. No differences in the methylation status of the endogenous RdDM targets 5S rDNA, Mea–ISR and AtMuI were observed using methylation-sensitive Southern blot analysis (Figure S5). However, elevated levels of expression of soloLTR and AtMu1 elements were observed in the morc6 and rmd1–1 (nrpe1) mutants compared to parental lines 142 and 142S (Figure 5a). For the soloLTR locus, this was associated with a reduction in DNA methylation as assessed by MspI digestion and Southern blotting (Figure 5b). The reduction observed for the morc6 mutants was not as extensive as that observed for the nrpe1 mutant (rmd1–1). Overall, our results are consistent with MORC6 influencing DNA methylation at a subset of loci.

Figure 5.

Requirement for MORC6 in silencing and DNA methylation of endogenous RdDM target loci. (a) Southern blot analysis assessing DNA methylation at the soloLTR locus using MspI digestion and DNA samples from morc6–5, morc6–6, morc6–7 and rmd1–1 (nrpe1) mutants and 142 and 142S wild-type lines. A reduction in the ratio of the upper to lower bands is indicative of loss of DNA methylation. (b) RT–PCR analysis of soloLTR and AtMu1 transcripts in morc6–5, morc6–6, morc6–7 and rmd1–1 (nrpe1) mutants and 142 and 142S wild-type lines. ACTIN was used as a positive control. No amplification was observed in reactions that lacked reverse transcriptase. (c) RT–PCR analysis of soloLTR transcripts in morc1–4, morc2, morc4, morc5, morc6–3 and morc7 mutants compared to Col–0 wild-type. The arrow indicates the position of the specific soloLTR PCR product. ACTIN was used as a positive control. No amplification was observed in reactions that lacked reverse transcriptase.

To determine whether any of the other Arabidopsis MORC genes are involved in transcriptional gene silencing, we assessed soloLTR expression in homozygous T–DNA insertion mutants of MORC1, MORC2, MORC4, MORC5, MORC6 and MORC7, in the Col–0 background. Mutations in MORC3 have been reported to be embryo-lethal (Kang et al., 2010). The MORC1 and MORC6 mutants used for this analysis have been described previously as morc1–4 and morc6–3 (Moissiard et al., 2012). Elevation of soloLTR transcripts was only clearly observed in morc1–4 and morc6–3 samples (Figure 5c). A low level of expression was also observed in the morc4 mutant, but this was inconsistent over repeat experiments. Thus it appears that only MORC1 and MORC6 have a clear role in silencing of this element.

Discussion

In this study, we demonstrate that MORC6 is required for efficient RdDM in a locus-specific manner. At the 35S:GFP locus used in the genetic screen, a reduction of DNA methylation in all sequence contexts occurred but was variable, with some strands losing all DNA methylation and others retaining near wild-type levels. The mutations in MORC6 resulted in an apparently stochastic silencing phenotype, in which loss of silencing correlated with loss of DNA methylation. Evidence was also found for a reduction in DNA methylation at the soloLTR RdDM target, but other target loci tested were unaffected. Several questions arise from these observations: primarily, why do morc6 mutants give rise to a stochastic silencing pattern, what is the role of MORC6 in RdDM, and why are some loci affected and not others?

The stochastic silencing phenotype observed in the morc6 mutants is striking, and suggests that silencing occurs in a cell-autonomous manner (Figure 2). A stochastic phenotype was not observed at this locus for mutants in nrpd1 or nrpe1 (Eamens et al., 2008). In the morc6 mutants, the 35S promoter driving expression of the GFP transgene appears to be either methylated and silenced, or unmethylated and active. Silenced and non-silenced cells occur across the entire leaf, with no obvious patterning that would be suggestive of clonal events or reduced transmission of a silencing signal. Generally the abaxial leaf surface showed a higher percentage of silenced cells than the adaxial surface. Whether this is an intrinsic property of the different epidermal cells or is influenced by external factors, such as light levels, remains to be determined. The stochastic effect is consistent with a model in which MORC6 is required to promote efficient RdDM but the requirement is not absolute. morc6–7 mutants have a higher percentage of silenced cells than morc6–5, suggesting that morc6–7 encodes a protein that retains more function than morc6–5 but not to wild-type levels. This is in agreement with the position of the premature stop codon in morc6–7 being downstream of the ATPase domain; however, the C–terminal coiled-coil domain is absent (Figure 1c). This suggests that MORC6 has functions that do not rely upon interactions that may be mediated by the coiled-coil domain. This influence on a bi-stable epigenetic state is somewhat reminiscent of the influence of the duration of cold exposure on the polycomb-based switch acting at the FLC locus, with the duration of cold affecting the likelihood of establishing a stable silenced state (Angel et al., 2011). In our case, establishing the silenced state is influenced by the level of MORC6 activity, and, when MORC6 activity is disrupted by mutation, the likelihood of establishing a stable silenced state is reduced.

In terms of how MORC6 acts in RdDM, it is interesting that two previous reports provided strong evidence for release of transcriptional silencing in morc6 mutants but differed in their observations regarding whether DNA methylation was affected (Lorković et al., 2012; Moissiard et al., 2012). Using a transgene system that is based on silencing of a meristem-specific enhancer sequence, Lorković et al. (2012) demonstrated a partial loss of CHH methylation specifically associated with the targeted region, but a more extensive reduction in all sequence contexts in a downstream region that is normally methylated in a Pol IV-dependent manner. MORC6 was shown to have ATPase activity and to interact with DMS3, the structural maintenance of chromosomes (SMC) component of the DDR complex. SMC proteins normally have ATPase domains, and so it was proposed that MORC6 provides the ATPase component that is missing from DMS3 (Hirano, 2006; Lorković et al., 2012). Our data are consistent with a role for MORC6 that is downstream of siRNA production and are therefore indirectly supportive of the proposed role in the DDR complex. The DDR complex is believed to facilitate association of Pol V predominantly with evolutionarily young transposons and promoters that contain transposons (Zhong et al., 2012). How this association is facilitated by the DDR complex is unclear, and it is possible that, if MORC6 does play a role in the DDR complex, it may not be required at all loci.

In contrast to our data and those of Lorković et al. (2012), Moissiard et al. (2012) did not report significant changes in DNA methylation in mutants defective in either MORC6 or the closely related MORC1, despite performing whole-genome bisulfite sequencing and specifically analysing transposable elements that were up-regulated in the mutants. Rather they provide evidence that the MORC proteins act downstream of RdDM in formation or reinforcement of compact chromatin, predominantly in peri-centromeric regions. How may these differing observations regarding the role of MORCs in RdDM be rationalized? Clearly there are differences both between the transgene loci analysed in these three studies, the ecotypes of the various Arabidopsis lines, and the methods used to analyse the endogenous loci. A simple conclusion as stated above is that some loci require MORC6 (and possibly other MORCs such as MORC1) to achieve efficient RdDM, whereas others do not. This may be influenced by the position in the genome, the local chromatin environment and/or the transgenic nature of some of the target loci studied. For the 35S:GFP locus analysed in this work, MORC6 is required for efficient RdDM, and we did not find evidence to suggest that release of transcriptional silencing occurs at this locus in the absence of loss of DNA methylation (Figure 4b). However, it is also clear that release of transcriptional silencing may occur at other loci in the absence of both a loss of DNA methylation and repressive histone marks, and this appears to be the predominant effect of morc mutants (Moissiard et al., 2012). This suggests that either MORC6 has a dual role in silencing, or that, for some loci, the RdDM mechanism is mechanistically linked with formation of silenced chromatin, with a MORC-dependent process influencing initiation and/or maintenance of RdDM. It is well established that reinforcement of DNA methylation patterns requires covalent histone modifications, and it has been demonstrated recently that lncRNAs influence nucleosome positioning (Law and Jacobsen, 2010; Zhu et al., 2013). One possibility to be considered is that, for some loci, failure to form a higher-order structure promotes demethylation. Indeed, it is known that the chromatin environment influences the ability of demethylases, such as ROS1, to function (Qian et al., 2012). In this scenario, we predict that RdDM is taking place at the 35S promoter in the morc6 mutants, but methylation is removed because of the open state of the chromatin. In cells that are able to establish a silenced state, perhaps with the aid of another MORC or an alternative pathway, the DNA methylation is maintained. This model requires that silencing takes place at the locus in the absence of MORC6, albeit at reduced efficiency, and that the locus is a target for demethylase activity. It will therefore be interesting to test the involvement and influence of demethylases in our silencing system, and to undertake a detailed examination comparing the features of loci that require MORC6 for efficient RdDM with loci that do not.

There are seven related MORC genes in Arabidopsis thaliana, and MORC1 has also been implicated in transcriptional silencing (Moissiard et al., 2012). We were unable to investigate whether MORC1 is also acting at the targeted 35S:GFP locus studied here, as the morc1–4 mutant allele tested carries a T–DNA insert that interferes with expression of the reporter transgene. It is therefore possible that MORC1 acts in the cells that retain methylation. Consistent with the results obtained by Moissiard et al. (2012), we observed activation of a soloLTR element in morc1 and morc6 mutants (Figure 5b). In this analysis, we also assessed lines carrying T–DNA insertions in MORC2, MORC4, MORC5 and MORC7. A low level of soloLTR transcript was observed for a putative morc4 mutant, but this was not as strong or as consistent as that observed for morc1 and morc6. Thus it appears that the other MORC genes tested may not play a role in transcriptional silencing at this locus. Interestingly, it has been reported that mutants in morc1 and morc6 have a very similar profile in terms of sequences that are de-repressed, and that the single mutants do not differ significantly from the double mutants (Moissiard et al., 2012). This is suggestive of proteins that act in the same pathway but whose functions are not redundant, which is perhaps surprising given the strong sequence conservation between the two proteins.

Experimental Procedures

Plant material and growth conditions

Arabidopsis thaliana lines 142 and 142S were described previously as lines GFP142 and 3–7, respectively (Dalmay et al., 2000; Eamens et al., 2008). Ethyl methanesulfonate mutagenesis of line 142S was performed as described previously (Eamens et al., 2008). All wild-type and mutant lines used were of the C24 ecotype apart from the T–DNA insertion lines morc1–4 (SAIL_1239_C08), morc2 (SALK_072774C), morc4 (SALK_0534000C), morc5 (SALK_021871C), morc6–3 (GK_599B06), morc7 (SALK_151643C) and rdr2–1. rmd1–1 and rmd3–1 are mutants that are defective in NRPE1 and NRPD1, respectively (Eamens et al., 2008). The mutants M1, M9 and M6 generated in this study have been named morc6–5, morc6–6 and morc6–7. Plants were grown in growth rooms under long-day conditions (16 h light/8 h dark) at temperatures of 22°C/20°C.

Imaging

A Leica MZ–FLIII dissecting microscope (Leica, http://www.leicamicrosystems.com) and a CoolSNAP RS Photometrics camera (Photometrics, http://www.photomet.com) were used to visualize and photograph whole leaves and seedlings. Confocal imaging was performed using a Zeiss 510 microscope (Zeiss, http://www.zeiss.co.uk) with a 20 × 0.5 objective and an argon laser emitting at 488 nm and a helium–neon laser emitting at 543 nm. Leaves were stained using propidium iodide (20 mg ml−1) for up to 2 h at room temperature, followed by washing in dH2O for 10 min prior to imaging.

Nucleic acid preparation

Genomic DNA samples were prepared using the DNeasy plant maxi or mini kits (Qiagen, http://www.qiagen.com) according to the manufacturer's instructions. Total RNA was prepared from either leaf tissue, mixed-stage floral tissue or sorted protoplasts using TRIzol (Invitrogen, http://www.invitrogen.com) according to the manufacturer's instructions. For analysis of siRNAs, the small RNA fraction was enriched following poly(ethylene glycol) precipitation of high-molecular-weight RNAs as described by Lu et al. (2007).

Nucleic acid analyses

Southern and high-molecular-weight Northern blot analyses were performed as described previously (Eamens et al., 2008). PCR-generated DNA fragments corresponding to GFP, ACTIN and solo LTR were used to create 32P-labelled DNA probes. The primers used to generate the probes are listed in Methods S1. Northern blot detection of small RNAs was performed as described by Pal and Hamilton (2008) using 32P-labelled riboprobes for 35S and soloLTR siRNAs and an antisense oligoprobe for miR163. cDNA samples for RT–PCR and quantitative RT–PCR were generated using Superscript II (Invitrogen) according to the manufacturer's instructions. Quantitative PCR was performed as described previously (Vaistij et al., 2010).

DNA methylation analysis of leaf samples by bisulfite sequencing was performed using the EpiTect bisulfite kit (Qiagen), with EpiMark Taq polymerase (New England Biolabs, http://www.neb.com) being used for target amplification as described previously (Eamens et al., 2008). For DNA methylation analysis from protoplasts, approximately 100 000 sorted protoplasts were used per sample, and the EZ DNA Methylation-Direct™ kit (Zymo Research Corporation, http://zymoresearch.com) was used for bisulfite treatment. PCR products were cloned into pGEM–T Easy (Promega, http://www.promega.com), and a minimum of ten clones were sequenced per sample. The CyMATE tool was used to analyse the bisulfite sequencing data (Hetzl et al., 2007).

FACS cell sorting of protoplasts

Protoplasts were generated from approximately 30 rosette leaves from plants at 28 dpg using the method described by Yoo et al. (2007). Protoplasts were resuspended in 1 ml resuspension buffer (0.6 M mannitol, 3 mm CaCl2, 4 mm KCl, 4 mm MES, pH 5.7), and a MoFLo Astrios cell sorter (Beckman Coulter, http://www.beckmancoulter.com) was used to count and separate protoplasts. Intact protoplasts were identified from cell debris and free chloroplasts using chlorophyll fluorescence and then sorted according to GFP fluorescence. Full methods for the cell sorting and counting are described in Methods S1.

Acknowledgements

We thank Fabian Vaistij, Tom Smith and Eleanor Walton for valuable comments on this work. We thank Jo Marrison, Graeme Park and Karen Hodgkinson (Technology Facility, Department of Biology, University of York) for their assistance with imaging and flow cytometry. The horticulture staff at the University of York provided excellent plant care. This work was supported by a Biotechnology and Biological Sciences Research Council studentship (to T.R.B.) and a Gatsby Charitable Foundation award (to L.J.).

Ancillary