Metastable Differentially Methylated Regions within Arabidopsis Inbred Populations Are Associated with Modified Expression of Non-Coding Transcripts

Abstract

Individual plants within a population may vary at both genetic and epigenetic levels. The rate of genetic divergence and its underlying mechanisms is well understood. Less is known about the factors contributing to epigenetic divergence among isogenic populations except that, despite the presence of mechanisms that faithfully maintain epigenetic marks, epigenetic differences are more frequent than genetic variation. Epigenetically divergent stretches of isogenic DNA sequence are called epialleles. Currently, it is not clear why certain regions exhibit variable epigenetic status. We identified and characterised two long RNA transcripts with altered expression and DNA methylation in an ago5 mutant. However, further investigation revealed that these changes were not dependent upon AGO5. Rather, the variable transcription of these loci in Arabidopsis mutant and wild-type populations corresponds to spontaneous differential methylated regions (DMRs) or epialleles. These two DMRs are delineated by RNAs which are highly expressed when the DMR is hypomethylated. Furthermore, they control the expression of 5′ transcriptional start site mRNA variants of nearby protein coding genes. Our data support the recent observations that meiotically stable DMRs exist within inbred populations. We further demonstrate that DMR boundaries can be defined by putative non-coding promoter-associated transcripts.

Introduction

Epigenetic information is stored through chemical modification of DNA and its associated histones [1], [2]. These modifications may be targeted to the appropriate genomic regions through an ancient mechanism in eukaryotes in which small RNA (sRNA) molecules are the specificity determinant. The sRNAs are associated with proteins that cause DNA or chromatin to be epigenetically modified and, in many instances, transcriptionally silent [3], [4]. Paradoxically, transcription of RNAs (likely through non-coding RNAs) is required for this transcriptional silencing process [5], [6], [7].

In Arabidopsis thaliana, the sRNAs associated with epigenetic modification and transcriptional gene silencing (TGS) are 24 nt long [8]. The 24 nt RNA silencing pathway involves a POL IV complex in which the largest subunit is NRPD1. POLIV transcribes a non-coding RNA that is made double-stranded (ds) by RNA-dependent RNA polymerase 2 (RDR2). This dsRNA is then cleaved into 23–24 nt sRNAs by Dicer-like 3 (DCL3). The 24 nt sRNA duplexes associate with an Argonaute (typically AGO4) and target it to non-coding RNA scaffold transcripts produced either by a POLV complex in which the largest subunit is NRPE1 [5], [6], [7], [9] or by POL II [10]. Targeting of the scaffold transcripts by AGO4 leads to recruitment of the DNA methyltransferase DRM2 and de novo cytosine methylation [9]. This process is referred to as RNA-directed DNA methylation (RdDM).

Once initiated, RdDM can be maintained through DNA replication independently of these RNA silencing components if the methylated cytosine is in a symmetrical context. The DNA methyltransferase MET1 maintains CpG methylation while CMT3 maintains CpHpG methylation [11]. However, for maintenance of methylated cytosine residues in an asymmetric context and, in some instances, at CpHpG motifs, the POLIV/POLV 24 nt sRNA silencing pathway is required [11].

The gain or loss of methylated cytosines [12] generates epialleles or differentially methylated regions (DMRs). Epialleles are phenotypically distinct if the change is in or close to a coding or regulatory locus [13], [14], [15], [16] and, together with changes in DNA sequence, they contribute to the heritable variation between members of populations. Stable epialleles can be generated by inbreeding of met1 or ddm1 mutants in which CpG methylation is defective [17], [18], [19]. Epialleles can also occur spontaneously [12]. Such epigenetic changes in A. thaliana are much more frequently than changes in DNA sequence [20], [21].

In this study, we identified two long RNAs that were mis-regulated in an ago5 mutant. Mis-regulation of these loci in the ago5 line was associated with differential DNA methylation. However, these RNAs were not mis-regulated because of a loss of AGO5. Instead they correspond to meiotically stable DMRs or epialleles that exist in random SALK populations and amongst natural A. thaliana ecotypes as well as in other RNA silencing mutants. Methylation maintenance at these loci is controlled by MET1 independently of RNA silencing. The epigenetic status of these DMRs affected their associated non-coding RNAs as well as specific mRNA variants of nearby protein coding genes.

Our results confirm two recent studies indicating that genetically identical individuals from a single population can be epigenetically divergent and that the epigenetic status of A. thaliana populations can change in relatively short time periods [20], [21]. Furthermore, our characterization of these two DMRs suggests that transcripts in promoter regions may promote epiallele formation. Our findings are relevant to understanding heritable variation in plant populations, and they also carry a cautionary message for those using mutants to investigate epigenetic effects: detailed controls should be used to differentiate spontaneous epigenetic changes from those due to any mutant in the machinery of epigenetics.

Results

Identification of Differentially Methylated Regions in the DNA of RNA Silencing Mutants

The initial aim of our study was to assign a biological role to AGO5 by analysis of long RNA transcripts that are mis-regulated in ago5 mutants. However, there was no overlap between the top 200 mis-regulated transcripts (Figure S1A) and AGO5 associated sRNAs (data not shown). To investigate this mis-regulation we analyzed in detail the two loci with the largest RNA abundance changes in ago5-2 compared with Col-0.

The RNA with the largest change was more abundant in ago5-2 than in Col-0 and was derived from an un-annotated region near Chr1∶24080015 (this transcript was renamed RITA for RNA In The Antisense orientation) (Figure S1B). The tag with the second largest abundance change was, in contrast, reduced in ago5-2, and matched Methionine Responding Down 1 (MRD1/At1g53480), a transcript described to be down-regulated in mutant mto1 [22] that over-accumulates soluble methionine (Figure S1B).

To find out whether the change in methylation and transcription at RITA and MRD1 is linked to RdDM we analyzed these loci in ago4-3 as well as in Col-0, ago5-2. An McrBC assay indicated a loss of DNA methylation in ago5-2 and ago4-3 at RITA and an increase in methylation at MRD1 relative to Col-0 (Figure S1C). In these assays, high levels of DNA methylation corresponded to low RNA abundance. We also assessed the transcriptional and methylation status of these loci in other RNA silencing mutants in: AGO1, AGO2, AGO3, AGO4, AGO5, AGO6, AGO7, AGO8, AGO9, AGO10, RDR1, RDR2, RDR6, SGS3, DCL4, AtTEX1, SDE5, DCL1, JMJ14, NRPD1, NRPE1, and DCL3.

At RITA there was loss of DNA methylation in plants lacking DCL4, DCL3, RDR2, AGO4, AGO7, AGO5, AGO8 and SDE5 (Figure 1A). As in ago5-2 and ago4-3, there was a negative correlation of long RNA abundance and DNA methylation phenotypes in the different mutants: lower levels of RITA DNA methylation corresponded to increased long RNA (Figures 1B). RITA (referred to as BT025685) was also previously identified as a non-coding RNA increased in hyl1, but not in dcl1 or serrate [23], suggesting other RNA silencing proteins might affect its epigenetic status.

Figure 1. RITA characterization.

(a) Methylation of RITA in seedlings as estimated by McrBC digestion in a panel of mutants. (b) RITA transcript abundance in seedling in a panel of mutants. (c) Bisulfite sequencing of RITA. Y-axis indicates percent of methylated cytosines. (d) sRNA northern blot using RITA probe. U6 is a loading control.

However the DNA methylation changes did not have either of two hallmarks of RdDM. First the bisulfite DNA sequencing of RITA in Col-0, ago5-2, ago4-3, nrpd1-4, nrpe1-1, rdr2-2, dcl3-1, dcl4-2, ago8-2, ago7-1 and sde5-2 confirmed hypomethylation but only at cytosine residues in CpG and CpHpG contexts (Figure 1C). Second there was no correlation of DNA methylation with 24 nt sRNAs. These RNAs were absent in nrpd1-4 in which there was no loss of DNA methylation and in rdr2-2 and dcl3-1 in which there was. These sRNAs were also reduced in other mutants but without correlation to the DNA methylation status at RITA (Figure 1D). Based on these analyses we conclude that the variable expression of RITA is due to spontaneous methylation changes at the DNA of these loci rather than the effects of the individual mutations.

A similar spontaneous change in DNA methylation is also likely to explain the variation at MRD1. Two recent studies identified the MRD1 locus as a variable DMR [20], [21] and in addition to the variation in expression and DNA methylation of this locus in Col-0 and ago5-2 (Table S1, Figures S1B and S1C) there were inversely correlated changes in other mutants. Thus the MRD1 DNA was hypomethylated and MRD1 RNA was abundant in Col-0 and plants mutant for DCL1, JMJ14, NRPD1, NRPE1, RDR2 and AGO3 whereas the DNA was hypermethylated relative to Col-0 (Figure 2A) and RNA was present at a low level in other lines (Figure 2B). Bisulfite sequencing of lines with different MRD1 states confirmed our McrBC assay and also that DNA methylation changes were at CpG and CpHpG contexts (Figure 2C). Thus, like RITA, the RNA abundance of MRD1 correlated with its epigenetic state (Figure 2B). This correlation, as with RITA, was not correlated with sRNA: the MRD1 24 nt sRNAs were reduced in plants that were mutant for NRPD1, RDR2, DCL3 and DCL4 (Figure 2D) in which there were variable levels of MRD1 DNA methylation and RNA.

Figure 2. MRD1 characterization.

(a) Methylation of MRD1 in seedlings as estimated by McrBC digestion in a panel of mutants. (b) MRD1 transcript abundance in seedling in a panel of mutants. (c) Bisulfite sequencing of MRD1. Y-axis indicates percent of methylated cytosines. (d) sRNA northern blot using MRD1 probe. U6 is a loading control.

RITA and MRD1 are DMRs Independent of RNA Silencing Proteins

The independence of DNA methylation and sRNA accumulation was further reinforced by the analysis of RITA in different alleles of AGO5, SDE5, DCL2, AGO8 and POLV. In all members of an ago5 allelic series (Figure S2A) there were reduced levels of AGO5 protein relative to Col-0 (Figure S2B) but RITA was hypomethylated only in ago5-2 (Figure 3A). Similarly with alleles of SDE5, an RNA silencing protein implicated in the trans-acting sRNA pathway rather than the RdDM pathway [24], [25] there was hypomethylation of RITA DNA in sde5-2, sde5-3, sde5-5 and sde5-7 (Figure 3B) but not sde5-4 and sde5-6. Of these mutants the strongest loss of function, manifested as accelerated juvenile-adult transition in leaves and loss of TAS1, TAS2 and TAS3 transacting sRNAs was in sde5-3 and sde5-5 in which there was hypomethylation of RITA DNA and in sde5-6 in which RITA DNA was fully methylated (Figure 3C–3D).

Figure 3. RITA and MRD1 epigenetic statuses in various allelic series.

(a) Methylation (estimated by McrBC digestion) of RITA and MRD1 in the AGO5 allelic series. (b) McrBC estimation of the methylation levels of RITA and MRD1 in an SDE5 allelic series. (c) sRNA northern blot for SDE5 allelic series. miR165 and U6 are loading controls. (d) SDE5 allelic series mutant phenotypes. White arrows highlight accelerated juvenile-adult transition phenotype in leaves. (e) Methylation of RITA and MRD1 in seedling RNA polymerase mutant alleles. The lines indicated with a * are also known to be mutants in the POLV complex (NRPE). (f) DNA methylation of RITA and MRD1 in Col-0 lines obtained from different lab (denoted by the principal investigator’s surname). #’s refer to individual plants; 4 individuals were assayed per line.

Similarly, we assayed three independent T-DNA insertion mutants of AGO8 exons of which only ago8-2 showed RITA hypomethylation (Figure S3A). All three mutants had inserts in exons and are likely to have lost AGO8 function. Additionally, in a DCL2 allelic series, a dcl2-1 line but not dcl2-2 or dcl2-3 had hypomethylated RITA (Figure S3B). Finally, we assayed two mutant alleles of the largest subunit of POLV (nrpe1-1 and nrpe1-11), the nrpd1-4 nrpe1-11 double mutant and multiple alleles of nrpd2a, the second largest subunit for POLIV and POLV (Figure 3E). As with the other mutants there was no correlation of strong loss of RNA silencing function and the hypomethylated RITA DNA.

In a similar analysis of DNA methylation at MRD1 we tested alleles of AGO5, SDE5, AGO8, DCL2 and NRPE1. Most of these mutants were hypermethylated at MRD1 relative to Col-0 (Figure 3A–D, Figure S3A-B). However, MRD1 DNA was hypomethylated in nrpe1-1 and hypermethylated relative to Col-0 in nrpe1-11 (Figure 3E). Thus the methylation status of MRD1 DNA is independent of the RNA silencing machinery.

Combined, these results strongly suggest that RITA and MRD1 are DMRs or epialleles and that the different epigenetic states in RNA silencing mutants are due to spontaneous variation. In support of this hypothesis, MRD1 (but not RITA) was identified as being a variable DMR over 30 generations of of individual lineages [20], [21] and its epigenetic state varied in Col-0 stocks from different labs. This locus is hypomethylated in the Col-0_Baulcombe line but hypermethylated in Col-0_Henderson and Col-0_Fisher lines (Figure 3F). Similarly there was epigenetic variation at both MRD1 and RITA in randomly selected SALK mutant populations and in A. thaliana ecotype populations. Fourteen randomly selected SALK lines along with Columbia based ecotypes (Col-0, Col-1, Col-2, Col-4, Col-6) demonstrated variability in the epigenetic status of both RITA and MRD1 (Figure S4A-B).

We did not detect variable methylation of RITA DNA in Col-0_Baulcombe and in five natural ecotype populations (C24, Kas, Koud, Sha, Ws-0). We were unable to amplify RITA with our standard primers in Bay-0 and Na-1 ecotypes suggesting sequence divergence in these lines. By comparison, MRD1 was hypomethylated in Col-0_Baulcombe, C24, Koud and Sha ecotypes hypermethylated in Kas, Ws-0, Bay-0 and Na-1 (Figure S4C-D). Thus the DMR at MRD1 is more variable in natural ecotypes than the DMR at RITA.

DNA Methylation at RITA and MRD1 is Controlled by MET1

Bisulfite sequencing indicated that the levels of CpG and CpHpG DNA methylation can approach 100% at RITA and MRD1 alleles but in met1-3 methylation is almost completely lost (Figure 4A–B). This result confirms previous data and suggests that, at these loci, MET1 maintains both CpG and CpHpG methylation [26]. Alternatively it could be that direct maintenance of CpG methylation by MET1 is indirectly required for CpHpG methylation.

Figure 4. Role of MET1 on DMRs.

(a) Bisulfite sequencing for RITA in ago5-2 and met1-3 mutants. (b) Bisulfite sequencing for MRD1 in ago5-2, met1-3 and cmt3-11 mutants. (c) McrBC analysis for RITA methylation in ago5-2, met1-3 heterozygotes and met1-3 heterozygotes and these met1 mutants combined with ago5-2. Each column represents a pool of individuals from separate populations. (d) McrBC analysis for MRD1 methylation in ago5-2, met1-3 heterozygotes and met1-3 heterozygotes and mutants combined with ago5-2. Each column represents a pool of individuals from separate populations.

In order to confirm that MET1 controls the epigenetic status of MRD1, we crossed a met1-3 line with hypomethylated MRD1 to a line with hypermethylated MRD1 (ago5-2). Three F4 populations were assayed for their MRD1 methylation status in various mutant genotypes. In all 3 populations, the met1-3/met1-3 plants were hypomethylated at MRD1 (Figure 4C). Thus, MET1 is likely to be required for maintaining high methylation at MRD1 epialleles. However, in the MET1/met1-3 genotypes, MRD1 and RITA are similarly hypomethylated, suggesting that MET1 cannot re-establish methylation at these loci by itself. We cannot rule out that a hypermethylated MRD1 epiallele segregated away in all 3 populations, but the most likely explanation is that MET1 is required for the maintenance of methylation at these loci and an additional factor is involved in remethylation (Figure 4C-D).

Stable Inheritance of RITA and MRD1

We assessed the meiotic stability of DNA methylation at RITA and MRD1 by using reciprocal backcrosses between Col-0 and ago5-2, and between Col-0 and ago4-3 in which the parents differed in their epigenetic status of RITA and MRD1. Without regard to direction of the cross, the McrBC assay indicated the F1 progeny were 50% methylated at RITA and MRD1 (Figure S5A-B). Thus, neither a functional copy of AGO5 nor AGO4 could remethylate MRD1 and RITA. This was in contrast to the canonical POLIV locus AtSN1, which was remethylated in F1 progeny of ago4-3 and Col-0 (Figure S5C).

We hypothesized that 50% methylation in the F1 progeny was an additive effect resulting from inheritance of one methylated and one unmethylated epi-allele from each parent. In order to differentiate the parental alleles for these DMRs, we crossed a line with hypomethylated RITA (Col-0-derived ago5-2) to a line with hypermethylated RITA (Ler with an insert at RITA relative to Col-0). The McrBC assay in the F1 progeny allowed us to visualize the methylation status of and differentiate between parental RITA alleles. In the F1 progeny of ago5-2 and Ler, the Ler allele retained methylation whereas RITA inherited from ago5-2 retained its hypomethylated state (Figure S5D). Thus, the epigenetic status of RITA is inherited through meiosis.

DMRs and POL II Transcripts from RITA and MRD1

The methylation status of RITA and MRD1 DNA may influence transcription of their nearby protein-coding genes as do the epialleles of FWA or SUPERMAN [14], [27] (Table S1, Figures S1B, 1, 2). To investigate this possibility we used 5′ and 3′ RACE to characterise RITA and MRD1 transcripts in lines with the different epialleles at these loci.

RITA generates a 687 base pair (bp) transcript that corresponds to a partial EST clone BT025685 and is divergently transcribed from ILA (Figure 5A). Northern analyses confirmed the size of the RITA RNA and its increased abundance in lines with hypomethylated RITA DNA (ago4-3 and ago5-2) relative to a line in which RITA DNA is hypermethylated (nrpd1-4) (Figure 5B). Two ILA transcripts were identified that differed in their 5′ transcriptional start site (Figure 5A). The longer ILA transcript (designated as β-ILA) begins within and overlaps RITA, while the shorter ILA transcript (designated as α-ILA) is separated from RITA by 165 bp.

Figure 5. Analysis of RITA and MRD1 locus transcripts.

(a) Schematic of RITA/ILA locus. Stars indicate the location of individual 5′ and 3′ RACE clones. A ruler present is present above to estimate sizes. Open boxes indicate coding regions, gray shading indicated untranslated regions (UTRs) and black shading indicates RITA or MRD1 RNAs. RM is the region assayed by bisulfite sequencing with oligos DBO489 and DBO490. RB is the region covered by McrBC assay using the oligos DBO447 and DBO448. MB was also used to probe the long RNA northern in (b). (b) Long RNA northern probed for RITA. 18S and 28S ribosomal RNAs indicate loading. (c) Schematic of MRD1/YAZ locus indicating amplicons (MB and AB) used for long RNA northern blot probes. MB was used for the McrBC assay with oligos DBO547 and DBO762. MM refers to the region assayed by bisulfite sequencing with the oligos DBO600 and DBO601. Stars indicate the location of individual 5′ and 3′ RACE clones. A ruler present is present above to estimate sizes. Open boxes indicate coding regions, gray shading indicated UTRs and black shading indicates non-coding RNAs. (d) RT-PCR for β-YAZ accumulation normalized to ACT2. (e) Long RNA northern blot with a probe (MB) that detects β -YAZ and MRD1, along with the non-specific hybridization of 18S and 28S ribosomal RNAs. (f) Long RNA northern blot with a probe (AB) that detects β -YAZ and α-YAZ. Loading is shown by 18S and 28S ribosomal RNAs in (e).

The RITA transcript has two open reading frames (ORFs) encoding short proteins of 12 and 78 amino acids that do not have homologues in the genomes of any sequenced organism. It is likely therefore that RITA is a non-coding RNA (Figure S6A). In contrast, ILA is a large protein-coding gene required for embryogenesis and plant innate immunity. The encoded protein is likely to be the Arabidopsis GCN1 homolog [28], [29].

To determine whether the status of RITA influenced the adjacent ILA locus, we used nested RT-PCR to assay the combined α+ β-ILA transcripts or the β-ILA transcript alone (Figure S7A). The abundance of the combined α+ β-ILA RNA did not differ whether RITA was hypo- or hyper-methylated (Figure S7B). However, the β-ILA RNA was more abundant than wild type in lines in which RITA DNA was hypomethylated (such as in ago5-2) (Figure S7B). It is likely therefore that the epigenetic state of RITA affects the expression of the overlapping β-ILA but not the non-overlapping a-ILA transcript.

Gene models for MRD1 predict an overlap with At1g53490 encoding an uncharacterized protein predicted to contain a DNA binding domain, which we renamed as Yet Another Zinc finger (YAZ). Previous reports identifying MRD1 as a DMR also noted a transcriptional affect on the At1g53490 gene [20], [21]. However, our RACE characterisation demonstrated that three transcripts are present at the MRD1/YAZ locus (Figure 5C). MRD1 is an intron-less transcript of approximately 1,082 bp [22], whereas YAZ is divergently transcribed and has two mRNA variants that arise from different transcriptional start sites. The longer YAZ transcript (β- YAZ) coincides with the full-length EST clone BX814259 and it overlaps MRD1 (Figure 5C). The transcriptional start site of the α-YAZ RNA is separated from that of MRD1 by 43 bp.

Three predicted ORFs for the MRD1 transcript encode peptides of 5 amino acids (ORF1), 32 amino acids (ORF2) and 120 amino acids (ORF3) (Figure S6B). BLAST searches using the protein sequence of ORF3 did not identify any conserved domains or similar proteins within plants and weak similarity with bacterial methylases (Figure S6B). Thus, the transcriptional configuration of the MRD1 and RITA loci both have an RNA that is divergently transcribed from two adjacent mRNAs and that it overlaps the longer transcript.

To determine whether the epigenetic state of MRD1 affected the adjacent YAZ locus we used northern analysis of lines with different MRD1 epialleles (Figure 5D–F). Using a probe (MB) that detected both the MRD1 transcript and β-YAZ it was clear that β-YAZ RNA accumulated at a lower level than MRD1 RNA. Both species are less abundant in lines with methylated MRD1 DNA (consisting of: ago4-3, ago5-2, ago7-1, ago8-2, dcl3-1, dcl4-2, nrpb2-3, rdr2-2 and sde5-2) (Figure 5E). A second probe (AB) confirmed the inverse correlation of β-YAZ transcripts and DNA methylation but it indicated that α-YAZ is unaffected by the epigenetic status of MRD1 (Figure 5F). The northern analysis of β-YAZ was confirmed by RT-PCR (Figure 5D).

Combined, these results indicate that RITA and MRD1 are both DMRs in which DNA methylation has a highly localised silencing effect. The DMRs affect divergently transcribed RNAs but only if they overlap the methylated regions.

Discussion

DMR Variability within Arabidopsis

Two recent studies have shown that DMRs in Arabidopsis can vary over 30 generations of inbreeding [21], [20] independent of DNA sequence or known environmental stimuli. These studies identified the DMR status of MRD1 but not of RITA. Other examples of DMRs or ‘pure’ epialleles in maize inbreds in the absence of associated genetic polymorphisms have also been recently documented [30]. Our findings suggest that the DMRs at RITA and MRD1 are likely similarly due to such spontaneous loss or gain of DNA methylation amongst SALK T-DNA or other mutant lines and between Columbia-accessions (Figures 3, S3 and S4).

DNA methylation at these loci is tightly associated with transcriptional repression of adjacent and overlapping genes (Figures 2, 5 and S7) so that changes in the status of a DMR could influence the plant phenotype. Thus, DMRs could affect the fitness of plants in wild populations as well as influence crop breeding programmes. For example, plant phenotypes resulting from DMRs include the peloric epimutation affecting flower morphology in Linaria [13] and genetic incompatibility between Arabidopsis accessions due to the epigenetic state of the gene AtFOLT [31].

As demonstrated here, DMRs complicate analysis of plant genomes because epigenetic differences amongst inbred laboratory lines may be the consequence of spontaneous changes in DNA methylation rather than the effect of mutations (Figures 1–3, S3, and S4S4). In this light, it may be necessary to re-evaluate the role of RNA silencing at some DMRs. For example, the reported effects of hyl1 on RITA [23] and of nrpd1-3 on MRD1 [32] could be the result of spontaneous epigenetic changes in those lines rather than mutant phenotypes. To differentiate whether epigenetic differences amongst lines are the result of spontaneous epialleles or a genetic mutation, the epigenetic lesion should be restored either a backcross or by complementation with a wild type gene.

As with other epialleles, the methylation status of RITA and MRD1 correlates with control of their nearby protein coding genes. Interestingly, the influence of the methylated MRD1 and RITA DMR had a strong effect on only one of the two detected nearby protein coding gene variants. Similarly at locus At5g43500 (ACT11) an upstream DMR controlled only one mRNA isoform [20]. These observations suggest that the effect of a DMR on nearby protein coding regions is highly localised. Furthermore and similar to other epialleles, the methylated status of MRD1 varied amongst natural populations of Arabidopsis [31], [33], [34].

Control of Epigenetic Changes at DMRs

Why are some loci prone to differential methylation whereas others are not? It is unlikely that any single feature is responsible but it is striking that, at RITA and MRD1, the DMR boundaries correlate with long and potentially non-coding RNAs (Figure 5). The presence of POL II transcription within gene promoters is widespread in eukaryotes and it cannot explain the differential methylation by itself [35], [36], [37], [38], [39]. However it could be that the length of the non-coding transcript and/or its overlap with the transcribed region of the protein-coding gene contribute(s) to the metastable status of the DMRs. Alternatively, perhaps the transcription of the RNA can mediate the erasure of methylation marks in the promoter region and render the RITA and MRD1 DMRs (more so than others) susceptible to hypo-methylation.

Methods

RNA Silencing Mutants & Growth Conditions

The following mutants are in the Col-0 genetic background. ago1-25 [40], ago2-1 (SALK_003380), ago3-1 (SM_3_31520), ago4-3 (WISC_338A06), ago5-2 (SALK_118422), ago5-1/5-3 (SALK_063806), ago5-5 (SALK_123506), ago5-6 (SALK_062912), ago6-2 (SALK_031553), ago7-1 (SALK_037458), ago8-1 (SALK_139894), ago8-2 (SALK_151983), ago8-3 (SALK_060402), ago9-1 (SALK_127358), nrpd1-4 (SALK_083051), nrpe1-1 [41], nrpe1-11 (SALK_029919), nrpd2a-2 (SALK_046208), nrpd2b-1 (SALK_008535), jmj14-3 [42], rdr1-1 (SAIL_672F11), rdr2-2 (SALK_059661), rdr6-15 (SAIL_617), attex1-4 (SALK_100012), sgs3-1 [43], dcl2-1 (SALK_064627), dcl2-2 (SALK_123586), dcl2-3 (SALK_095069), dcl3-1 (SALK_005512), dcl4-2 (GABI_160G05), sde5-2 (SALK_020726), sde5-3 (WISC_429G09), sde5-4 [25], sde5-5 (SALK_026595C), sde5-6 (SALK_115496C), sde5-7 (SALK_026511C), met1-3 [44], cmt3-11 (SALK_148381), and nrpb2-3 [10]. dcl1-7 [45] was introgressed into Col-0. ‘dcl234’ combines dcl2-1, dcl3-1 and dcl4-1. The nrpd2a-1/2b-1 mutant combines nrpd2a-1 (SALK_095689) with nrpd2b-1. The NRPD2a and NRPD2b mutations affect both the POLIV and POLV complexes (and so are also considered NRPE2a and NRPE2b, respectively). zll-3 [46] is in Ler.

The random collection of SALK lines and other Columbia based ecotype lines were obtained from the European Arabidopsis Stock Centre (NASC) [47]. Table S2 details where the insertion is predicted to be located. Zygosity of the T-DNA inserts was not confirmed. Some Col-0 lines were obtained from other laboratories and are abbreviated as Col-_0Henderson (Ian Henderson lab.) and Col-0_Fischer (Robert Fischer lab.). Col-0 from the David Baulcombe lab designated as Col-0_DCB or simply Col-0.

Genotyping primers for alleles characterised in this study are in Table S3. ago5-2 wild-type (DBO368/DBO369), ago5-2 mutant (DBO369/Lba1), ago5-1/ago5-3 wild-type (DBO372/DBO373), ago5-1/ago5-3 mutant (DBO373/Lba1), ago5-5 wild-type (DBO368/DBO371), ago5-5 mutant (DBO371/Lba1), ago5-6 wild-type (SALK_062912LP/SALK_062912RP), ago5-6 mutant (SALK_062912LP/Lbb1), ago8-1 wild-type (DBO80/DBO136), ago8-1 mutant (DBO80/LBa1), ago8-2 wild-type (DBO137/DBO119), ago8-2 mutant (DBO119/Lba1), ago8-3 wild-type (DBO132/DBO137), ago8-3 mutant (DBO137/LBa1), sde5-2 wild-type (DBO827/DBO828), sde5-2 mutant (DBO828/Lba1), sde5-5 wild-type (DBO829/DBO830), sde5-5 mutant (DBO830/Lba1), sde5-6 wild-type (DBO827/DBO828), sde5-6 mutant (DBO827/Lba1), sde5-7 wild-type (DBO831/DBO832), and sde5-7 mutant (DBO832/Lba1).

Mixed stage inflorescence tissue was collected from plants 6-7 weeks post germination grown at 20°C, 16 hr day lengths. Seedlings (collected 14 days post germination) were grown on ½MS, 21°C (light), 20°C (dark) in 16 hr days.

DNA Analyses

DNA samples were prepared using Qiagen Puregene reagents. 30 mg tissue was incubated in 300 µl Cell Lysis Solution (No. 158906) at 65°C (60 min), followed by incubation at 37°C (15 min) with 6 µg RNaseA. 100 µl Protein Precipitation Solution (No. 158910) was added followed by centrifugation at 13,000×g (3 min). DNA was precipitated with 100% isopropanol, followed by centrifugation at 13,000×g (1 min). DNA was washed and resuspended in DNA Hydration Solution (No. 158914).

For bisulfite analyses, 500 ng DNA was converted using EZ DNA Methylation-Gold Kit (Zymo Research), using the ‘Alternative 1’ conversion methods. PCR conditions (100 ng converted DNA) were: GoTaq PCR buffer (Promega), 2.5 mM MgCl₂, 0.25 mM dNTPs, 10% DMSO, 1 mM primers and 1.5 units Taq. Bisulfite primers were: RITA (DBO489/DBO490), MRD1 (DBO600/DBO601), AtSN1 (JP1821/JP1822), AtREP2 (DBO411/DBO412), Simplehat2 (DBO413/DBO414). Amplification conditions were 94°C, 3′, 40 cycles of 94°C, 30″, 55°C, 30″, 50°C, 30″ and 62°C, 1.5′. At least 12 colonies were sequenced per locus, except at MRD1 where at least 24 colonies were sequenced because approximately one-half correspond to At5g03090. At5g03090 has approximately 90% sequence homology to MRD1 and was methylated in all samples analyzed.

Bisulfite converted sequences were aligned to an unconverted parent using ClustalX. Non-cytosine SNPs were used to differentiate MRD1 from At5g03090. Evaluation of methylation was with Kismeth [48].

For the McrBC assay, 1000 ng of DNA were diluted in 12.5 µl of DNA Hydration Solution, 1X NEB Buffer 2, 10 µg BSA, 1 mM GTP, and 75.5 µl water. 500 ng was then digested at 37°C for 12 hrs with 20 units McrBC (New England Biolabs) or mock digested using 2 µl 50% glycerol. The reactions were inactivated at 65°C, 20 min. 20 ng digested DNA was used per quantitative PCR reaction. McrBC assay primers (Table S3) are: RITA (DBO447/DBO448), except for the random SALK assay in which the more robust pair of (DBO709/DBO448) was used for RITA, MRD1 (DBO547/DBO762), ACTIN2 (ACT2 FW/ACT2 REV), AtSN1 (AtS15, AtSN1-F4), At3g01345 (DBO986/DBO987), and At5g24250 (DBO1028/DBO978).

All quantitative PCR reactions were performed using 1X SYBR Green (Sigma Aldrich) and 0.15 µM primers (annealing temperature 56°C). Three technical replicates were done for all samples and the quantity of PCR product estimated as 1/1.8?C(t).

For the McrBC assay, the gene-specific quantity of McrBC digested DNA was normalized to the quantity of mock digested DNA. This value was then normalized to the normalized value (digest/mock) of the unmethylated gene ACTIN2. Error was calculated by standard propagation of error. Biological replicates were done at least twice for all experiments (with the exception of the random SALK lines which were only assayed once) but only 1 experiment is represented in the figures. There were no discrepancies observed amongst biological replicates.

RNA Analyses

Total RNA was isolated using Trizol (Invitrogen). cDNA synthesis was done using a polyT primer on 500 ng DNAse (Turbo DNAse system, Ambion) treated RNA using SuperScriptIII (Invitrogen). Primers (Table S3) for RT-PCR reactions are: RITA (DBO447/DBO448), MRD1 (DBO762/DBO547), ACTIN2 (ACT2FW/ACT2RV), AGO5 (DBO372/DBO373), β-ABE (DBO637/DBO758), α+β-ILA (DBO487/DBO488), β-ILA was detected with a primary PCR amplification (27 cycles) using (DBO920/DBO882) followed by 50X dilution and a nested PCR reaction (1 µl of the dilution, 28 cycles, DBO735/DBO834). A negative control was used for the nested β-ILA assay with DBO920/DBO708, where DBO708 lies outside the 5′ end of β-ILA. For the additionally characterised DMRs, transcripts were detected with these primers: α-At5g24240 (DBO1067/DBO1068), β-At5g24240 (DBO976/DBO1068), UBC9 (DMB175/DMB176), At3g01345 (DBO1069/DBO1070), At3g01340 (DBO1008/DBO1010).

5′ and 3′ RACE reactions were carried using Clontech SMART RACE system following manufacturer’s instruction with the exception of NEB Phusion used for PCR amplification. Transcript ends were mapped from Col-0 transcripts with these primers (Table S3): α-ILA (DBO712, 5′ RACE; DBO710, 3′ RACE), MRD1 (DBO714, 5′ RACE; DBO713, 3′ RACE), β-YAZ (DBO728, 5′ RACE), (DBO714 primary PCR followed by DBO715, 3′ RACE), α-YAZ (DBO757, 5′ RACE; DBO730, 3′ RACE). RITA and β-ILA were mapped in ago5-2 because they are expressed there; RITA (DBO708, 5′ RACE; DBO709, 3′ RACE) and β-ILA (DBO877 primary PCR followed by β-ILA nested, 5′ RACE). All PCR products were gel purified and cloned (pGEM Teasy) and at least 8 colonies were sequenced.

Long RNA northern blots were performed with polyA RNA (Micro PolyA Purist, Ambion). Long RNAs were separated in 1X MOPS as described (Northern Max Protocol, Ambion), followed by capillary transfer (Zeta-Probe Membrane, BIO-RAD). Probe MB was amplified using DBO762/DBO547, Probe AB with DBO765/DBO766 and RITA probe with DBO447/DBO448 from cDNA. PCR products were radioactively labelled (Rediprime II, Amersham). Hybridizations were at 68°C (PerfectHyb Plus, Sigma).

sRNA northerns were carried out as previously described [51]. Oligos (Table S3) were radioactively labelled and hybridized at 35°C (ULTRAhyb-Oligo, Ambion). MRD1 was detected with the oligo mix DBO810, DBO901, DBO843 and DBO900. RITA was detected with a DNA probe amplified with DBO447/448 (Rediprime II, Amersham), and hybridized at 45°C (ULTRAhyb, Ambion). All northern blots were washed in 2X SSC 0.1% SDS.

Antibody Generation, Immunoblotting and Immunoprecipitation

The AGO5 peptide H₂N-SKEESKNTEVSETMSC-CONH₂ was KLH conjugated and used to raise rabbit polyclonal antibodies and affinity purified (Eurogentec). AGO5 immunoblotting was done at 1∶1500 (α-AGO5) in 5% milk, followed by a 1∶10000 titer of HRP-conjugated goat-anti-rabbit secondary antibody (Santa Cruz) in 5% milk and detected using ECL PLUS (Amersham). AGO1 (VRKRRTDAPSEGGEGC) anti-peptide antibody and immunoblotting conditions were described previously [49], [50]. αACTIN (1∶1000, Affinity Bioreagents) and αGAPDH (1∶5000, Abcam) were used following manufacturer’s recommendations and appropriate HRP-conjugated secondary antibodies.

Proteins were extracted from mixed staged floral tissue by grinding in liquid nitrogen and homogenization in 3X (w:v) extraction buffer (20 mM Tris HCL pH 7.5, 5 mM MgCl₂, 5mM DTT, 300mM NaCl, 0.1% NP-40, 1% plant protease inhibitor (Sigma)). Extract was centrifuged and visualized by immunoblotting using 6% acrylamide SDS PAGE gels.

The Col-0 and ago5-2 EST Tag sequences were prepared using the Illumina Digital Gene Expression – Taq Profiling for Nla III following manufacturer’s instructions for polyA RNA. All libraries were sequenced on a Genome Analyzer I or II.

The sequence data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus and are accessible through the GEO Series accession number GSE30970 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE30970).

The accession numbers are: Col-0 floral sRNAs (GSM767710), ago5-2 floral sRNAs (GSM767711), AGO5 IP (GSM767709), Col-0 EST tag replicate 1 (GSM767714), Col-0 EST tag replicate 2 (GSM767712), ago5-2 EST tag replicate 1(GSM767715), ago5-2 EST tag replicate 2 (GSM767713).

Bioinformatic & Statistical Analyses

The Illumina sequence data were processed as described previously [51]. EST tags were ranked according to their likelihood of differential expression as evaluated by ‘baySeq’ Bioconductor package [51] between Col-0 and ago5-2. The top 200 differentially expressed tags were aligned to the Arabidopsis genome (TAIR 10) and annotated as overlapping protein-coding genes, non-coding RNAs, or transposons.

Supporting Information

Figure S1

ago5 -2 EST tag sequencing identifies mis-regulated transcripts. (a) Annotations for genome regions corresponding to the 200 EST tags most likely to be differentially represented between Col-0 and ago5-2. Combined percentages are greater than 100 because some tags correspond to more than 1 annotation. (b) Quantitative RT-PCR confirms EST tag data for some ago5-2 mis-regulated RNAs. Sample names appear below the bottom panel. Genes or loci tested appear above each individual panel. Chromosome coordinates are given for the locus named RITA, which did not correspond to a known gene. (c) McrBC digest of genomic DNA followed by PCR amplification of affected loci confirmed to be differentially expressed in ago5-2. Tall bars indicate low levels of methylation and low bars indicate high levels of methylation. Sample names appear below the bottom panel and individual loci are given above each panel. Error bars represent a propagation of errors.

(PDF)

Figure S2

AGO5 allelic series analysis. (a) Schematic diagram of the AGO5 gene (At2g27880). Boxes represent that AGO5 transcript; exons (white), UTRs (shaded). Vertical lines indicate the location of transfer-DNA insertions. Each allele is marked above its representative triangle, and the association of the SALK line and description with the mutant is under the schematic. No phenotypic changes were observed in any of the lines. Images were taken at the same magnification approximately two weeks after germination. (b) Western blot using αAGO5. αAGO1 was used as a loading control.

(PDF)

Figure S3

RITA and MRD1 methylation in the AGO8 and DCL2 allelic series (a) Three T-DNA insertions within the predicted AGO8 (At5g21030) gene were identified. A schematic shows the location of each of the insertions (open boxes represent putative exons). McrBC digestion followed by PCR estimates the methylated status for RITA and MRD1 in each of these lines. (b) McrBC digestion followed by PCR estimates the methylated status for RITA and MRD1 in 3 different DCL2 alleles along with the dcl234 triple mutant, which harbours the dcl2-1 mutation.

(PDF)

Figure S4

Naturally occurring RITA and MRD1 epialleles in A. thaliana mutant lines and ecotypes. (a) McrBC digestion estimating RITA methylation in various Columbia based ecotypes and randomly chosen SALK T-DNA insertion lines. N##### refers to the NASC identifier for each line. Each column represents a pool of approximately 6 individual seedlings. (b) McrBC digestion estimating MRD1 methylation in various Columbia based ecotypes and randomly chosen SALK T-DNA insertion lines. N##### refers to the NASC identifier for each line. Each column represents a pool of approximately 6 individual seedlings. (c) DNA methylation analysis of RITA epialleles in Col-0, C24, Kas, Koud, Sha and Ws-0 seedlings by McrBC digest. DNA methylation at MRD1 for Col-0_DCB is also shown. Four individuals were analysed for each ecotype. (e) DNA methylation analysis of MRD1 epialleles in Col-0, C24, Kas, Koud, Sha, Ws-0 Bay-0 and Na-1seedlings by McrBC digest. Four individuals were analysed for each ecotype.

(PDF)

Figure S5

Inheritance of RITA and MRD1 . (a) McrBC digestion followed by PCR estimating the DNA methylation at RITA of in F1 progeny of crosses between Col-0 and ago5-2 and Col-0 and ago4-3. (b) McrBC digestion followed by PCR estimating the DNA methylation at MRD1 of in F1 progeny of crosses between Col-0 and ago5-2 and Col-0 and ago4-3. (c) McrBC digestion followed by PCR estimating the DNA methylation at AtSN1 of in F1 progeny of crosses between Col-0 and ago5-2 and Col-0 and ago4-3. (d) McrBC digestion followed by PCR shows allele specific methylation in crosses between Ler and ago5-2 and Col-0 and ago5-2.

(PDF)

Figure S6

Coding predictions for RITA and MRD1 . (a) The RITA RNA transcript has two Open Reading Frames (ORFs). The predicted number of amino acids appears below the ORF. Predicted homologous proteins are listed for ORF2 with accession, description and E value. (b) The MRD1 transcript contains 3 ORFs. The number of amino acids predicted for each ORF is listed below the schematic. Blast P results including accession, description, and E value are listed for each of the hits.

(PDF)

Figure S7

Transcript abundance changes of ILA in mutants with various RITA epigenetic states. (a) Locus structure of ILA and RITA. Upper panel depicts the orientation of α - and β-ILA relative to RITA. Lower panel is a schematic of the RNA transcripts present at the locus. Open boxes represent coding regions; gray boxes represent UTRs; RITA is depicted in black. Labelled arrows correspond to the oligos used in RT-PCR reaction shown in (b). (b) RT-PCR of ILA transcripts. ACT2 and α+β-ILA (most likely to be composed of mainly α -ILA) as measured by DBO487/DBO488do not significantly differ amongst the RNA silencing mutants tested. A nested PCR reaction using DBO735/DBO844on a primary PCR reaction designed only to amplify β-ILA transcripts demonstrated an increase in β-ILA in the mutants where methylation is lost at RITA. No amplification was observed in a primary reaction (DBO920/DBO882) or when the oligos were present outside of the mapped β-ILA transcript region (DBO920/DBO708).

(PDF)

Table S1

(XLSX)

Table S2

(XLSX)

Table S3

(XLSX)

Funding Statement

PF was supported by a short term EMBO fellowship ASTF158-2008. This work was supported by the Gatsby Charitable Foundation and the European Union Sixth Framework Programme Integrated Project SIROCCO (LSHG-CT-2006-037900). DCB was supported as a Royal Society Research Professor. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.