RNA-directed DNA methylation regulates parental genomic imprinting at several loci in Arabidopsis

In mammals and plants, parental genomic imprinting restricts expression of certain loci to one parental allele (Feil and Berger, 2007). Imprinting in mammals relies on sex-dependent de novo deposition of DNA methylation by specific methyltransferases Dnmt3a and Dnmt3b (Ferguson-Smith, 2011; Bartolomei, 2009). The asymmetric patterns of DNA methylation on each parental allele are further maintained in the embryo by the maintenance DNA methyltransferase Dnmt1. By contrast, in flowering plants, MET1 the homolog of mammalian Dnmt1 is involved in the mechanism of parent-specific gene activation and also in maintenance of imprinted expression (Jullien and Berger, 2009). Histone methylation by Polycomb group also assists imprinting both in plants and mammals (Feil, 2009).

During plant sexual reproduction, two sperm cells fertilize two distinct female gametes, the egg cell and the central cell, producing the embryo and the endosperm, respectively (Berger and Twell, 2011). The endosperm and the embryo are surrounded and protected by the seed coat, which is of maternal sporophytic origin. Genome-wide surveys of parental allele-specific expression in Arabidopsis, rice and maize have identified new series of candidate imprinted genes and shown that imprinting is largely confined to genes expressed only in endosperm (Luo et al., 2011; Wolff et al., 2011; Zhang et al., 2011; Gehring et al., 2011; Hsieh et al., 2011; Waters et al., 2011). These studies show that imprinting causes maternal expression for the majority of genes but some genes are rather expressed by the paternal allele. Not only genes but also non coding RNAs might be expressed maternally (Mosher et al., 2009; Luo et al., 2011), although whether these maternal RNAs originate from seed coat or endosperm remains unknown.

MET1 maintains methylation of cytosine residues in CG contexts (Feng et al., 2010), which results in a silenced status of many imprinted genes in diploid sporophytic vegetative cells and in haploid male gametes (Jullien and Berger, 2009). The endosperm-specific expression of maternally expressed imprinted genes originates from the mechanism that activates the expression of the maternal allele. In the central cell, the transcriptional repression of MET1 (Jullien et al., 2008), together with the active demethylation by the DNA glycosylase DEMETER (Gehring et al., 2009) cause the loss of methylated CGs, resulting in transcriptional activation in the endosperm precursor: the central cell. Although a few genes imprinted in the embryo have been reported (Luo et al., 2011; Jahnke and Scholten, 2009), the origin of the mechanisms involved remains unknown.

The study of two imprinted loci in maize showed that imprinted alleles acquire DNA methylation after fertilization (Gutiérrez-Marcos et al., 2006; Jahnke and Scholten, 2009), but the origin of this methylation remained unknown. Because de novo methylation is unable to distinguish between two equally unmethylated parental alleles, this mechanism alone is unlikely to account for these observations. Recent studies have also suggested that mechanisms involving controls of DNA methylation other than DME and MET1 might be involved in the control of imprinting (Luo et al., 2011; Wolff et al., 2011; Zhang et al., 2011; Gehring et al., 2011; Hsieh et al., 2011; Waters et al., 2011). In addition to methylation in the CG context, plant genomes have substantial amounts of CHG and CHH methylation where H represents any base other than G (Feng et al., 2010). In plants, the DOMAINS REARRANGED METHYLTRANSFERASES (DRM) methylate cytosine residues in the CHH context (Cao and Jacobsen, 2002). CHH methylation by DRM is associated with the RNA-directed DNA methylation pathway (RdDM). RdDM is initiated by the production of the 24-nucleotide small interfering RNAs (siRNAs) by a pathway involving the plant-specific RNA polymerases PolIV and PolV (Cao et al., 2003; Pontier et al., 2005; Daxinger et al., 2009). The major subunits of RNA POLIV and POLV, are NRPD1a and NRPD1b, respectively, and NRPD2A is the second largest subunit of RNA POLIV and POLV. Thus far, there has been no direct evidence that the RdDM pathway controls imprinting by DNA de novo methylation in plants. Here, we dissect the expression of various members of the RdDM pathway in Arabidopsis gametes and demonstrate that the RdDM pathway participates in parental genomic imprinting.

Arabidopsis thaliana ecotypes Columbia-0 (Col-0), Wassilewskija (Ws), Landsberg erecta (Ler-0), Cape Verde Islands (Cvi-0), RLD and C24 were obtained from the Arabidopsis Stock Centre (http://www.arabidopsis.org). The nrpd2a-1 (SALK_095689) was received from C. Pikaard’s Lab (Onodera et al., 2005). The mutant seeds nrpd1b-12 (SALK_033852), nrpd1a-4 (SALK_083051), drm1-2 (SALK_031705), drm2-2 (SALK_150863), double homozygous drm1, drm2, cmt3-11t (SALK_148381C), sdc (SALK_017593) and mop9.5 (FLAG_462F03, FLAG_508H08) were obtained from ABRC. Primers for genotyping are described in supplementary material Table S1. MET1 antisense (MET1a/s) was characterized by J. Finnegan (Finnegan et al., 1996). Reduction of 70% CpG methylation was reported in this line and the effect is dominant in vegetative tissues. The MET1/met1-3 (met1-3/+) line was provided by J. Paszkowski (Saze et al., 2003). The FIE co-suppressor (FieCoS) line has previously been characterized by N. Ohad’s group (Katz et al., 2004). The mutant dme11 has been characterized previously (Guitton et al., 2004).

The origin of parental transcripts of imprinted genes was analyzed by RT-PCR followed by enzymatic digestion or sequencing. Seeds or siliques from the reciprocal crosses between two selected ecotypes harboring the nucleotide polymorphism were used for RNA extraction (RNeasy Plant Minikit, QIAGEN) and reverse transcription. Transcripts were amplified by Illustra Taq polymerase (GE) in 20 μl PCR reaction for 35 cycles with specific primers that span the polymorphism sites. Different parental alleles were distinguished by using restriction enzymes (NEB) that recognize the restriction sites introduced by the polymorphism. For allele-specific RT-PCR analysis by sequencing, PCR was performed using HOTSTART Taq polymerase (QIAGEN) in 40 μl reaction. Prior to sequencing (BigDye Terminator Kit, ABI), PCR products were purified by MinElute column (QIAGEN) or gel purification (if required) using the Gel DNA Recovery Kit (Zymo Research). Sequencing results were analyzed by SeqMan (Lasergene).

Unfertilized ovules obtained from ten emasculated pistils were used for RNA isolation using the Arcturus PicoPure RNA Isolation Kit (Applied Biosystems, Foster City, CA). The RNeasy Micro Kit (QIAGEN) was used to isolate total RNA from seeds. RNA isolation from siliques and leaves was carried out using the RNeasy Mini Kit (QIAGEN). The same amount of total RNAs from different tissues was used in reverse transcriptase reaction reactions using Stratagene transcriptase (Agilent Technologies). Quantitative PCR was performed with the ABI Prism 7900 HT Fast System using SYBR Green PCR Master Mix (Applied Biosystems) according to supplier’s recommendations. Relative quantitative values were calculated by comparative C_T method (Schmittgen and Livak, 2008). The s.d. was calculated based on at least three technical replicates and two biological replicates.

Isolation of female gametophytic cells was performed as described previously (Ikeda et al., 2011). In detail, the total RNA was isolated from 10 egg cells, five central cells, 10 synergid cells and one ovary using following kits: Dynabeads mRNA DIREC Micro Kit for female gametes and Ambion RNAqueous-Micro for the ovary. The RT-PCR results were reproduced on two biological replicates using Ex Taq (TaKaRa) in 20 μl reaction with 45 cycle amplification.

Cloning was performed using GATEWAY technology (Invitrogen). The pSDC-H2B-RFP contains 2731 bp promoter sequences of SDC (At2g17690) locus in order to drive the expression of HISTONE2B-RFP in the plasmid pGnk-GW-H2B-RFP (Ingouff et al., 2009). Fifteen T1 transgenic lines displayed the consistent fluorescent expression pattern. The reporter transgenic line pFWA-GFP has been described previously (Kinoshita et al., 2004).

Endosperm and embryos dissected for bisulfite sequencing were described previously (Ikeda et al., 2011). In detail, embryos and endosperm were isolated from seeds produced by reciprocal crosses between Cvi and RLD carrying SNPs, which were used to differentiate the parental alleles (supplementary material Fig. S4). Sperm cells were sorted through via FACS as published previously (Borges et al., 2008) and the extracted DNA was fragmented, treated with sodium bisulfite and sequenced using next generation sequencing (Lister et al., 2008; Calarco et al., 2012). Levels of methylation were calculated at each cytosine as the ratio C/C+T, and only residues with at least four reads mapping to them were considered. Methylation profiles were displayed on separate tracks within each cytosine context (CG, CHG and CHH) using the Integrated Genome Viewer (IGV).

From the databases of candidate imprinted genes in Arabidopsis (Hsieh et al., 2011; Gehring et al., 2011; Wolff et al., 2011), we selected five maternally expressed genes that could be silenced by the RdDM pathway in vegetative tissues and expressed in endosperm according to databases available online (Le et al., 2010). Using single nucleotide polymorphisms we confirmed that SUPPRESSOR OF drm1 drm2 cmt3 (SDC), MOP9.5 (At5g24240), At3g21830, At1g61090 and At2g34880 are maternally expressed (Fig. 1A-C) and that their expression is repressed by RdDM and MET1 in vegetative tissues (Fig. 1D-G), suggesting that these five genes are maternally expressed imprinted genes controlled by RdDM. However, inheritance of maternal transcripts from the female gametes also lead to maternal expression, although no active transcription of the maternal allele takes place after fertilization. We thus investigated further the spatial and temporal paternal of expression of the five candidate imprinted genes.

In order to visualize SDC expression, we obtained a transcriptional reporter line that expresses the fluorescent protein fusion HISTONE2B-RFP (H2B-RFP) under the control of the SDC promoter. According to the pattern of expression of H2B-RFP, SDC was expressed in endosperm but not in seed coat in wild-type background (supplementary material Fig. S1B,E). By contrast, SDC was expressed in the seed coat in nrpd2a-1 homozygous mutant (supplementary material Fig. S1F), confirming the transcriptional repression of SDC by RdDM in diploid vegetative tissues. In wild-type plants carrying pSDC::H2B-RFP, we observed that SDC is expressed in the central cell but not in the egg cell (Fig. 2A,B). We did not observe SDC expression in sperm cells (Fig. 2E,F). After fertilization, we detected expression of SDC and pSDC::H2B-RFP in endosperm but not in the embryo (Fig. 2C,D,G). We observed an increase of SDC transcripts levels by twofold in 2 DAP (days after pollination) seeds compared with flower buds, and a 5-fold increase in 4 DAP seeds compared with flower buds (supplementary material Fig. S1B,C). These results indicated that SDC transcripts in endosperm did not originate from the central cell and that SDC was expressed in endosperm. The maternal expression of SDC was observed in crosses between different natural accessions (Fig. 2H; supplementary material Fig. S1A). These results confirmed that SDC is a maternally expressed imprinted gene expressed in endosperm.

MOP9.5 was also expressed in endosperm (Fig. 3A). Levels of MOP9.5 expression were too low to allow detection using a transcriptional reporter similar to that used for SDC (not shown). Allele-specific RT-PCR showed that only the maternal allele of MOP9.5 was expressed in seeds (Fig. 3B). Real-time PCR showed that levels of MOP9.5 maternal transcripts increased after fertilization (Fig. 3C), indicating that MOP9.5 was expressed after fertilization and that MOP9.5 is a maternally expressed imprinted gene.

Similar increasing levels of expression after fertilization were observed for At1g16090, At2g34880 and At3g21830 (supplementary material Fig. S2), indicating that these genes are expressed actively after fertilization and that their maternal expression does not reflect inheritance of maternal transcripts from the female gamete. However, the expression levels of the maternally expressed imprinted genes At3g21830, At1g61090 and At2g34880 were much lower than SDC and MOP9.5. We thus focused further work on SDC and MOP9.5 as models to study whether RdDM plays a role in imprinting.

SDC was not expressed in pollen (Fig. 2E,F) and, accordingly, we reported that in sperm cells DNA methylation is found in all contexts, including CHH in the region containing the SDC promoter (Calarco et al., 2012). This observation suggested that SDC expression is repressed by de novo DNA methylation in male gametes, but the effect of RdDM on the expression of the paternal allele of SDC in endosperm remained unknown. To address this issue we used Col and Cvi accessions to distinguish the parental alleles of SDC and crossed wild-type ovules with pollen from mutants for RdDM. In seeds from such crosses, we observed ectopic transcriptional activity of the endogenous SDC paternal allele (Fig. 4A). We also observed expression of the paternal allele of the pSDC::H2B-RFP reporter when paternally inherited from nrpd2a (Fig. 4B). These data show that the RdDM pathway is involved in silencing the SDC paternal allele that is inherited in endosperm.

In leaves and in sperm cells, a domain in the 5′ region of MOP9.5 that overlaps with the 3′ of the coding sequence of the neighbor gene At5g24250 was methylated in all contexts (supplementary material Fig. S3A), also suggesting that de novo methylation participated in the silencing of MOP9.5 in male gametes. In seeds produced by crosses between wild-type ovules and pollen of mutants affected for the RdDM pathway, we observed ectopic activation of the paternal MOP9.5 allele (Fig. 4C). We concluded that the imprinted expression of MOP9.5 also depends on the RdDM pathway that silences the paternal allele of this gene in endosperm.

However, the levels of CHH methylation at SDC and MOP9.5 promoters were lower in male gametes than in leaves (Calarco et al., 2012) (supplementary material Fig. S3A), questioning whether RdDM activity in male gametes silenced SDC and MOP9.5, effectively. In order to address this question, we crossed wild-type ovules with pollen from nrpd2a/+ plants that experience RdDM activity during vegetative development and lose RdDM activity only during gametogenesis. In endosperm from these crosses, we observed that both SDC and MOP9.5 remained fully imprinted with no detectable expression from the paternal allele (supplementary material Fig. S3B,C). In summary, we did not observed activation of SDC and MOP9.5 paternal alleles from nrpd2a pollen from nrpd2a/+ heterozygous plants, but we did observe activation of SDC and MOP9.5 paternal alleles from nrpd2a homozygous plants that inherit a genome from somatic sporophytic tissues deprived of NRPD2a. We thus concluded that RdDM was required in sporophytic diploid paternal tissues but not in the gametophytic male haploid germline to silence SDC and MOP9.5 paternal alleles. Hence, our results suggest that effective silencing of imprinted genes by RdDM takes place primarily prior to gametogenesis.

Moreover, CG methylation by MET1 participates in the extension of DNA methylation from the tandem repeats towards the transcription start site (TSS) and in silencing of SDC paternal allele (Henderson and Jacobsen, 2008). Accordingly, the paternal SDC allele contributed by met1/+ plants was ectopically expressed at low levels (Fig. 4D; supplementary material Fig. S3D). We further showed a stronger ectopic expression of the SDC paternal allele from crosses between wild-type ovules and pollen from nrpd2a/nrpd2a; met1/+ plants (Fig. 4D; supplementary material Fig. S3D). These results indicate that silencing of the paternal allele of SDC silenced by RdDM requires the joint activities of maintenance methylation by MET1 during male gametogenesis and de novo methylation by RdDM during sporophytic development.

In order to gain insight into the mechanism responsible for maternal activation of SDC and MOP9.5, we tested whether DEMETER, the DNA glycosylase that participates in maternal activation of several imprinted genes was also involved in SDC imprinting. We detected a marked decrease of SDC expression in seeds deficient for DME (Fig. 5A). By contrast, MOP9.5 expression in seeds did not depend on DME (Fig. 5A). These data suggested that another mechanism in female gametes was sufficient to activate the expression of MOP9.5. Among the DNA methyltransferases, only low levels of DRM2 were detected in the central cell (Jullien et al., 2012). DRM2 activity relies on RNA polymerases PolIV and PolV, which are essential for the production and use of siRNAs, respectively (Herr et al., 2005; Onodera et al., 2005; Mosher et al., 2008; Daxinger et al., 2009). We isolated central cells and the associated synergid cells from Arabidopsis ovules (Ikeda et al., 2011; Gebert et al., 2008), and analyzed the presence of transcripts for genes controlling POLIV and POLV activity. We could not detect NRPD1a, NRPD1b or NRPD2a transcripts by RT-PCR in these cells (Fig. 5B). Our results suggest that in addition to the low levels of DRM2, the absence of PolIV and POLV major subunits prevents active RdDM in the central cell. The joint repression of the DNA methyltransferases activity is sufficient to activate expression of RdDM targets in sporophytic tissues (Fig. 1) and thus is also likely to be sufficient to activate the maternal allele of MOP9.5 and SDC in the central cell. The moderate reduction of the levels of SDC transcripts by dme loss of function (Fig. 5A) also suggests that low RdDM and MET1 activities in the central cell assist DME to express SDC in the central cell, leading to inheritance of an active maternal allele in endosperm.

To determine whether the pattern of DNA methylation at the imprinted SDC locus paralleled the expression of the two parental alleles of SDC after fertilization, we performed bisulfite sequencing of DNA extracted from isolated endosperm where SDC is expressed and from isolated embryos that do not express SDC. The promoter of SDC contains seven 32 bp tandem repeats, which produce siRNAs and are methylated by the RdDM pathway in vegetative tissues (supplementary material Fig. S1A) (Henderson and Jacobsen, 2008; Numa et al., 2010). We analyzed DNA methylation at the transcription start site (TSS) and in the 5′ upstream region containing 32 bp repeats (Fig. 6A; supplementary material Fig. S1A, Figs S4, S5). In embryos, both SDC parental alleles were methylated with a comparable profile. By contrast in endosperm, we observed a marked reduction of CG but not CHH methylation on the maternal allele of SDC in repeats and at the TSS, compared with the paternal allele (Fig. 6B; supplementary material Fig. S4B), which correlated with the maternal-specific expression of SDC in the endosperm. Overall, our data suggest that, in endosperm, MET1 maintains the difference of methylation in CG contexts between the parental alleles, leading to the maintenance of maternal-specific expression of imprinted genes SDC and MOP9.5.

Recent studies reported the expression of genes involved in RdDM in developing seeds. During the early syncytial phase of endosperm development, which encompasses the period when SDC and MOP9.5 showed high levels of imprinted expression, endosperm showed low levels of expression of NRPD2a and other major components of the RdDM pathway: NRPD1b, the RNA DEPENDENT RNA POLYMERASE 2 (RDR2), DICERLIKE 3 (DCL3), ARGONAUTE 4 (AGO4) (Belmonte et al., 2013) and of DMR2 (Jullien et al., 2012). The low levels of expression of main components of the RdDM pathway in endosperm are compatible with the fact that both parental alleles of SDC do not become silenced, which would be the case if RdDM activity were high. The expression of RdDM related genes increases markedly after endosperm cellularization (Belmonte et al., 2013), which probably explains why we observed similar CHH methylation profiles on both parental alleles of SDC in cellular endosperm (Fig. 6A,B).

We hypothesized that the dedicated complex transcriptional mechanisms that lead to maternal expression of SDC and MOP9.5 in endosperm was selected because these genes play important functions, as shown for other imprinted genes (Berger et al., 2013). In order to study SDC and MOP9.5 function in endosperm, we isolated two null sdc and mop9.5 alleles (Fig. 6C,D). In absence of SDC or MOP9.5 expression, endosperm development was not affected (Fig. 6E), and null sdc and mop9.5 plants produced viable seeds similar to wild type (not shown). Hence, SDC and MOP9.5 functions are not essential for endosperm development.

Our study shows that several loci that are repressed by RdDM in somatic vegetative tissues are maternally expressed in endosperm. The repression of these loci also involves MET1. MET1 maintains DNA methylation in somatic tissues and male gametes, whereas we show that RdDM does not participate to the maintenance of the DNA methylation in the male gametes. In the central cell, the low expression of several key genes involved in RdDM together with the low activity of MET1 and other DNA methyltransferases (Jullien et al., 2008; Jullien et al., 2012) is expected to contribute to the activation of the maternal allele of the imprinted genes studied. In the case of some imprinted genes but not all of them, DME is also required for maternal activation. After fertilization, RdDM activity is limited, if not absent, during the early phase of endosperm development and, during this period, MET1 appears to be the main pathway responsible for maintenance of the different levels of DNA methylation between the parental alleles. We thus conclude that de novo methylation assists MET1 in the maternal expression of imprinted genes in plants.

De novo methylation relies on the production of 24-nucleotide non-coding siRNAs. The likely involvement of non-coding piRNAs in de novo DNA methylation in mammals (Law and Jacobsen, 2010) and the demonstration that small RNAs regulate one imprinted locus in mice (Watanabe et al., 2011) further support the idea of convergent evolution of imprinting mechanisms in plants and mammals (Feil and Berger, 2007). Yet in Arabidopsis, removal of DNA methylation creates the parent-specific expression, whereas in mammals imprinting is created by deposition of DNA methylation in a parent-specific manner (Ferguson-Smith, 2011; Bartolomei, 2009). In plants, the opposition of DNA methylation patterns between male and female gametes appears to be sufficient to create a blueprint that causes imprinted expression at specific loci. Hence, the maternal expression in the endosperm becomes a mere consequence of the differential expression in the central cell and sperm cells. By contrast, in mammals, imprinted expression is not reflected by differential expression in gametes and becomes apparent only after fertilization.

Non-coding RNAs of maternal origin have been detected from fractions containing endosperm and seed coat, and it has been proposed that these non-coding RNAs might be expressed only from the maternal allele (Mosher et al., 2009). Collectively, our study and others (Jullien et al., 2012; Belmonte et al., 2013) indicate that early endosperm development (before 7 DAP in Arabidopsis) is marked by low activity of RdDM. This is in agreement with the fact that the maternal allele of imprinted genes is not targeted and silenced by RdDM activity during early endosperm development. Expression of RdDM components appears to peak after 7-10 DAP in endosperm (Belmonte et al., 2013), which corresponds to the developmental stage of seeds used by Mosher et al. (Mosher et al., 2009) and could explain the maternal origin of the non-coding RNAs detected from materials containing endosperm and seed coat used in this study. Yet the mechanism of production of endosperm non-coding RNAs and their roles remain unclear (Mosher et al., 2011).

The association between transposable elements (TEs) and several imprinted loci in plants and in mammals suggests that TEs play a role in establishing imprinting expression (Kaneko-Ishino and Ishino, 2010; McCole et al., 2011; Wolff et al., 2011; Walter and Paulsen, 2003; Wilkins and Haig, 2003; Okamura et al., 2004; Fujimoto et al., 2008; Hsieh et al., 2011). This hypothesis is further supported by the fact that DME might have a preference for demethylating TEs (Gehring et al., 2009). Although it is now apparent that DME is also expressed in cell types other than the central cell (Schoft et al., 2011) and is dispensable for imprinting of certain loci (this study) (Jullien et al., 2008; Tiwari et al., 2008; Hsieh et al., 2011; Gehring et al., 2011). TEs are rather silenced by RdDM (Saze and Kakutani, 2007; Martienssen et al., 2008; Mirouze et al., 2009), which makes MET1 unlikely to be involved in the initial silencing event required in creating a new imprinted locus. Our findings suggests that any event (not only TE insertion) that leads to production of siRNA and is sufficient to target de novo DNA methylation and silencing, may also be sufficient to prime imprinted expression in endosperm. This would explain the origin of many imprinted genes, which are not associated with a TE (Gehring et al., 2011; Hsieh et al., 2011). The absence of impact of the loss of SDC and MOP9.5 on endosperm suggests that the primary selection of imprinting genes rather relies on advantages conferred by their silencing in sporophytic vegetative tissues. A recent survey shows that ectopic expression of imprinted genes during vegetative development is deleterious (Berger et al., 2013). We propose that the evolution of imprinted genes is initiated by silencing of a locus by RdDM, which is beneficial for vegetative development. As a result of the absence of RdDM activity, such a locus is activated in the central cell, while the silenced state in maintained in sperm cells. The differential epigenetic status of the locus is inherited in endosperm after fertilization, as long as the site methylated by RdDM can be maintained by MET1. We thus propose that the peculiar control of RdDM activity in plant gametes provides a mechanism that primes evolution of imprinted loci.

We dedicate this manuscript in memory of Simon Chan who provided critical reading of this manuscript, and we thank Professor David Baulcombe for providing seeds of nrpd2a, nrpd2b mutants.