The Polycomb group protein MEDEA and the DNA methyltransferase MET1 interact to repress autonomous endosperm development in Arabidopsis

Authors


For correspondence (e-mail grossnik@botinst.uzh.ch).

Summary

In flowering plants, double fertilization of the female gametes, the egg and the central cell, initiates seed development to give rise to a diploid embryo and the triploid endosperm. In the absence of fertilization, the FERTILIZATION-INDEPENDENT SEED Polycomb Repressive Complex 2 (FIS-PRC2) represses this developmental process by histone methylation of certain target genes. The FERTILIZATION-INDEPENDENT SEED (FIS) class genes MEDEA (MEA) and FERTILIZATION-INDEPENDENT ENDOSPERM (FIE) encode two of the core components of this complex. In addition, DNA methylation establishes and maintains the repression of gene activity, for instance via DNA METHYLTRANSFERASE1 (MET1), which maintains methylation of symmetric CpG residues. Here, we demonstrate that Arabidopsis MET1 interacts with MEA in vitro and in a yeast two-hybrid assay, similar to the previously identified interaction of the mammalian homologues DNMT1 and EZH2. MET1 and MEA share overlapping expression patterns in reproductive tissues before and after fertilization, a prerequisite for an interaction in vivo. Importantly, a much higher percentage of central cells initiate endosperm development in the absence of fertilization in mea-1/MEA; met1-3/MET1 as compared to mea-1/MEA mutant plants. In addition, DNA methylation at the PHERES1 and MEA loci, imprinted target genes of the FIS-PRC2, was affected in the mea-1 mutant compared with wild-type embryos. In conclusion, our data suggest a mechanistic link between two major epigenetic pathways involved in histone and DNA methylation in plants by physical interaction of MET1 with the FIS-PRC2 core component MEA. This concerted action is relevant for the repression of seed development in the absence of fertilization.

Introduction

In the life cycle of sexually reproducing plants, seed development is initiated by double fertilization: one sperm fuses with the egg cell giving rise to the diploid embryo, while a second sperm fuses with the central cell to form the triploid endosperm. The development of embryo and endosperm are actively repressed in the absence of fertilization (Rodrigues et al., 2010). After fertilization, the development of embryo, endosperm and seed coat are highly coordinated.

In plants as well as in animals, two major epigenetic pathways play important roles in the regulation of cell-fate decisions by modifying chromatin, namely DNA methylation and histone methylation. They are involved in establishing a heritable repression of gene activity. In animals, there is increasing evidence for a complex regulatory interplay between pathways regulating histone modifications and DNA methylation (Cedar and Bergman, 2009; Jin et al., 2011). A direct mechanistic link between the two major epigenetic pathways of DNA methylation and histone methylation has been described based on physical interactions of histone-modifying proteins and DNA methyltransferases (Viré et al., 2006). A similar mechanism has not yet been demonstrated in plants.

In the model plant Arabidopsis, DNA methylation is mediated by at least three classes of methyltransferases: the METHYLTRANSFERASE1 (MET1) class, the CHROMOMETHYLASE3 (CMT3) class and the DOMAINS REARRANGED METHYLASE (DRM) class (Finnegan and Kovac, 2000). MET1 is a CpG maintenance methyltransferase that is homologous to the mammalian DNA methyltransferase DNMT1, CMT3 proteins form a small family of chromomethylases, and the DRM class proteins are de novo methyltransferases with homology to the mammalian DNMT3 proteins (Finnegan and Kovac, 2000; Cao et al., 2003). While DNA methylation in the CpG context is predominant in animals, DNA methylation is abundant in CpG, CpNpG and CpNpNp sequence contexts in plants (Cokus et al., 2008; Lister et al., 2008).

Polycomb group (PcG) proteins maintain the inactive state of their target genes by establishing repressive histone modifications, e.g. by histone methylation mediated by the SET (SUPPRESSOR OF VARIEGATION 3–9, ENHANCER OF ZESTE, TRITHORAX) domain of some PcG proteins (Mahmoudi and Verrijzer, 2001; Simon and Tamkun, 2002; Reyes and Grossniklaus, 2003). During the Arabidopsis lifecycle, distinct variants of the Polycomb Repressive Complex 2 (PRC2) are involved in the regulation of important developmental processes, such as gametophyte and seed development, the transition from embryo to seedling growth, the induction of flowering, and floral organ development (Hennig and Derkacheva, 2009; Bouyer et al., 2011). The FIS-PRC2 is a well-studied PcG complex that has been shown to be involved in early seed development (Rodrigues et al., 2010). Plant variants of PRC2 catalyse trimethylation of lysine 27 at histone H3 (H3K27me3) (Makarevich et al., 2006), similar to the evolutionarily conserved PRC2 in animals (Köhler and Grossniklaus, 2002; Pien and Grossniklaus, 2007; Hennig and Derkacheva, 2009). The FERTILIZATION INDEPENDENT SEED (FIS) class proteins MEA, FIS2, FERTILIZATION INDEPENDENT ENDOSPERM (FIE) and MULTICOPY SUPPRESSOR OF IRA1 (MSI1), homologues of EZH2, SUZ12, EED and p/RbAp48 in mouse, respectively, are the core components of the FIS-PRC2 (Grossniklaus et al., 1998; Luo et al., 1999; Ohad et al., 1999; Köhler et al., 2003a). Two additional histone methyltransferases homologous to EZH2, CURLY LEAF (CLF) and SWINGER (SWN), are encoded in the Arabidopsis genome and participate in distinct plant PRC2s (Hennig and Derkacheva, 2009).

Identified target genes of the FIS-PRC2 include the imprinted genes PHERES1 (PHE1), a MADS box transcription factor, and MEA itself (Köhler et al., 2003b; Baroux et al., 2006; Gehring et al., 2006; Jullien et al., 2006a; Makarevich et al., 2006; Makaverich et al., 2008; Villar et al., 2009). In contrast to MEA, which is maternally active and paternally silent, PHE1 is paternally active and maternally repressed (Grossniklaus et al., 1998; Vielle-Calzada et al., 1999; Köhler et al., 2005). Regulation of PHE1 and MEA expression requires both H3K27me3 methylation introduced by the activity of the FIS-PRC2 and DNA methylation mediated by MET1, in addition to the activity of DECREASE IN DNA METHYLATION1 (DDM1), belonging to the SWI/SNF family of chromatin remodelling factors (Vielle-Calzada et al., 1999; Köhler et al., 2003b; Xiao et al., 2003; Gehring et al., 2006; Jullien et al., 2006a; Makaverich et al., 2008; Dumbliauskas et al., 2011).

The FIS-PRC2 plays an important role in reproductive development in Arabidopsis by regulating cell proliferation before and after fertilization (Köhler and Grossniklaus, 2002; Guitton and Berger, 2005). Plants carrying mutations in any of the FIS class genes show maternal-effect seed abortion (Grossniklaus et al., 2001) and endosperm development in the absence of fertilization (Ohad et al., 1996; Chaudhury et al., 1997; Grossniklaus et al., 1998; Köhler et al., 2003a). Recently, a mutation in the ubiquitin ligase gene CUL4, whose product is proposed to regulate the FIS-PRC2, has been demonstrated to mediate autonomous endosperm development (Dumbliauskas et al., 2011). In addition, a mutant affecting the gene encoding UBIQUITIN SPECIFIC PROTEASE 26 has been shown to cause autonomous endosperm development at low frequencies (Luo et al., 2008). Endosperm development is not completed in the fis class mutants described so far, and autonomous endosperm development is not a fully penetrant trait in all mutants, i.e. endosperm development does not initiate in all ovules carrying a mutant embryo sac (female gametophyte). This suggests that additional factors may contribute to the repression of autonomous endosperm development, e.g. DNA methylation targeting genes required for endosperm development.

Interestingly, in animals, recent evidence suggests that the PcG protein EZH2 controls DNA methylation by direct protein–protein interaction. The PcG proteins EZH2 and EED directly interact with the maintenance methyltransferase DNMT1 and the de novo DNA methyltransferases DNMT3a and DNMT3b, presumably in the context of PRC2 and PRC3, which contain different EED isoforms (Martin et al., 2006; Viré et al., 2006). The expression level of EZH2 is relevant for the DNA methylation state of PRC2/3 target genes (Viré et al., 2006). In contrast, in plants, it is unknown whether the two major epigenetic pathways of histone methylation and DNA methylation interact in a similar manner. Nevertheless, in Arabidopsis plants carrying the fie-1 mutant allele, autonomous endosperm development was promoted by DNA hypomethylation due to a dominant MET1 antisense silencing construct, allowing endosperm development to progress further than is usually observed in fie-1 mutants (Vinkenoog et al., 2000). However, the molecular mechanism underlying this effect in the control of endosperm development is not understood and may be an indirect effect due to global DNA hypomethylation occurring in the inbred MET1 antisense background.

Here, we show that MET1 physically interacts with MEA and probably also with FIE, both being core components of the FIS-PRC2. In addition, we demonstrate that MET1 is involved in the repression of endosperm development in the absence of fertilization. Our results suggest a mechanistic linkage between MEA activity, probably in the context of the FIS-PRC2, and MET1 in the control of endosperm development. In addition, mea mutant embryos showed altered DNA methylation patterns in differentially methylated regions of the FIS-PRC2 target genes PHE1 and MEA. Our results indicate that a concerted action of the two major epigenetic pathways of DNA and histone methylation plays a role in the regulation of seed development, and indicate an ancient evolutionary origin of the interaction of these two major epigenetic regulatory pathways in mammals and plants.

Results

MET1 and MEA interact in in vitro pull-down assays

MET1 and MEA are important players in the two major epigenetic pathways repressing target gene activity. The mammalian homologues of MEA and MET1, EHZ2 and DNMT1, respectively, have been reported to directly interact (Viré et al., 2006). To test for protein–protein interactions of Arabidopsis MET1 and MEA, in vitro GST pull-down assays were performed (Figure 1). A GST–MEA fusion protein was expressed in Escherichia coli and incubated with 35S-methionine-labelled MET1 produced by in vitro transcription and translation. GST-tagged proteins were bound to glutathione beads, and co-purified interacting proteins were analysed (Figure 1b and Figure S1). The assay shows that 35S-labelled MET1 binds to MEA more efficiently than to GST alone (Figure 1b).

Figure 1.

MET1 and MEA interact in an in vitro GST pull-down assay. (a) Scheme of the primary structure of the Arabidopsis MET1 protein and fragments used for this study. (b) Extracts of bacteria containing GST–MEA fusion proteins were mixed with 35S-labelled MET1 produced by in vitro transcription/translation and bound to glutathione beads. The specific interaction signal was higher than the faint background obtained with 35S-labelled MET1 tested with GST alone. (c) Extracts of bacteria containing GST fusion proteins with the N-terminal, BAH or C-terminal domains of MET1 were mixed with 35S-labelled MEA produced by in vitro transcription/translation and bound to glutathione beads. No interaction signal was observed with GST alone mixed with 35S-labelled MEA as a control. In (b) and (c), bound proteins were extensively washed and separated on an SDS–PAGE gel. 35S-labelled proteins were detected by autoradiography as described in Experimental Procedures.

Detailed studies on the interaction between the mammalian methyltransferases DNMT1, DNMT3a, DNMT3b and EZH2 indicate that several protein domains mediate these interactions. DNMT1 interacts with EZH2 through both the N- and C-terminal domains, and, in addition, DNMT3a and DNMT3b associate with EZH2 via a conserved plant homeodomain (PHD)-like motif (Viré et al., 2006). To test whether different domains of the MET1 protein mediate interaction with MEA, we cloned fragments encoding the N-terminal, BAH (bromo-adjacent homology) and the C-terminal methyltransferase domains of MET1 (Figure 1a) into the pGEX vector for expression of GST fusion constructs. The three fragments of MET1 were expressed as GST fusion proteins in E. coli and tested in a pull-down assay with 35S-labelled MEA (Figure 1c and Figure S1). All three domains of MET1 interacted with MEA in vitro (Figure 1c). The results suggest that MET1 and MEA interact directly, and that various domains of the MET1 protein mediate the interaction or contribute to the interaction.

MET1 and MEA probably interact in the context of the FIS-PRC2

MEA is the histone methyltransferase of the FIS-PRC2. However, distinct roles for MEA in regulating its own expression have been demonstrated: (i) auto-repression of the maternal MEA allele around fertilization, which is independent of other PRC2 components, and (ii) repression of the paternal MEA allele in the context of the FIS-PRC2 (Baroux et al., 2006; Jullien et al., 2006a,b). To determine whether the interaction of MET1 and MEA plays a role in auto-repression of the maternal MEA allele, we analysed MEA expression in flowers of heterozygous met1-3/MET1 mutant and wild type plants before fertilization. Similar MEA expression levels were observed in the mutant and the wild type (Figure S2), indicating that MET1 is not involved in an auto-repressive activity independent of the FIS-PRC2. We thus investigated whether MET1 interacts with the FIS-PRC2, analogous to the interaction of the mammalian homologues EZH2 and DNMT1 in the context of PRC2/3 (Viré et al., 2006). In addition to the MEA homologue EZH2, the FIE homologue EED has been shown to interact with DNMT1 (Viré et al., 2006). Thus, it is possible that FIE may also interact with MET1, likely in the context of the FIS-PRC2. To test this hypothesis, we performed yeast two-hybrid assays.

The Arabidopsis MEA and FIE coding sequences (Spillane et al., 2000) and, as a control, the first 48 amino acids of the Pm3b coiled-coil domain from wheat, were expressed as fusion proteins with the E. coli LexA protein as the DNA-binding domain (DB, bait). DB-FIE was tested against MEA as a fusion to the B42 activation domain (AD) (Figure 2) (Ma and Ptashne, 1987). DB-MEA, DB-FIE and DB-Pm3b were tested against the MET1 N-terminal, BAH domain, and C-terminal fragments (MET1Nterm, MET1BAHlong and MET1Cterm, respectively) expressed as AD fusion proteins. To check for auto-activation, DB-MEA and DB-FIE expression constructs were tested together with the empty pB42AD vector (Figure 2). Yeast was tested for activation of the lacZ reporter gene, encoding β-galactosidase. No β-galactosidase activity was observed with the Pm3b protein as bait tested together with any of the MET1 fragments as prey (Figure 2). The DB-MEA and DB-FIE expression constructs transformed together with the pB42AD vector resulted in a slight activation of the lacZ reporter gene. However, a stronger activation was observed with either DB-MEA or DB-FIE together with the MET1Nterm or MET1BAHlong domains as AD fusion proteins. These yeast two-hybrid assays indicate interactions of MEA and FIE with MET1, mediated by the N-terminal and BAH domains of MET1. Thus, the assays support the evidence for interaction of MEA and MET1 observed in the GST pull-down assay for these two domains, but not for the interaction with the MET1 C-terminal domain. In conclusion, MET1 may potentially interact with both MEA and FIE.

Figure 2.

Yeast two-hybrid assays showing interaction of MET1 with MEA and FIE. MEA, FIE and, as a control, the Pm3b coiled-coil domain from wheat were expressed as fusion proteins to the E. coli LexA protein as the DNA-binding domain (DB, bait), and tested against the MET1 N-terminal, BAH and C-terminal domains as fusion proteins to the B42 activation domain (AD, prey). As a further control, DB-MEA and DB-FIE were tested together with the empty pB42AD construct. FIE was also tested as a bait with MEA as a prey. Two yeast colonies of each combination of bait and prey were tested for activation of the lacZ marker gene by growth on medium containing X-Gal. Yeast co-expressing DB-FIE and AD-MEA fusion proteins, or DB-MEA or DB-FIE fusion proteins together with the MET1 N-terminal or BAH domain as the AD fusion protein, showed stronger activation of β-galactosidase activity than any of the controls, in contrast to colonies expressing the MET1 C-terminal domain as prey.

MET1 shares overlapping expression patterns with MEA and FIE before and after fertilization

To allow protein interactions between MET1 and MEA in vivo, they need to share the same temporal and spatial localization in plant tissues. Expression of MEA and FIE in the embryo sac harbouring the egg and central cell, in the developing embryo and endosperm after fertilization, and, in the case of FIE, in sporophytic tissues, has been described previously (Vielle-Calzada et al., 1999; Luo et al., 2000; Spillane et al., 2000; Yadegari et al., 2000; Baroux et al., 2006). Thus, we performed in situ hybridization studies to test for MET1 expression in the female gametophyte before fertilization and at early stages of embryo and endosperm development (2–3 days after pollination). We obtained signals in the egg, synergids and central cell of mature embryo sacs, as well as in sporophytic floral tissues (Figure 3a,b). Signals were detected in the embryo, the endosperm and also the seed coat (Figure 3c,d). No signals were obtained when using sense control probes (Figure 3e,f). The expression of MET1 in the female gametes (egg cell and central cell) is in agreement with a recent study analysing cell-type specific expression using Affymetrix GeneChips (Figure S3) (Wuest et al., 2010). Thus, MET1, MEA and FIE share overlapping expression patterns in the mature female gametophyte and in the embryo and endosperm during early stages of development, which is a prerequisite for physical interaction.

Figure 3.

MET1 is expressed in reproductive tissues before and after fertilization. In situ hybridization of sections of the mature female gametophyte before fertilization (a, b, e) and at early embryonic stages approximately 2–3 days after pollination (c, d, f) show expression of MET1 in the cells of the mature female gametophyte before fertilization [egg cell (egg), central cell (cc) and synergids (syn)], and in the embryo (emb) and endosperm (end). Weak expression is also observed in sporophytic tissues of the developing seed before and after fertilization. Sections were hybridized with MET1 antisense probe (a–d) or sense probe (e, f). Scale bars = 40 μm.

Lack of maternal MET1 activity enhances the penetrance of the fis phenotype in mea-1/MEA plants

Loss of function of any core component of the FIS-PRC2 (MEA, FIE, FIS2 or MSI1) leads to autonomous endosperm development (fis phenotype), presumably as a result of de-repression of genes controlling proliferation and development. However, unlike gametophytic maternal-effect seed abortion, which is fully penetrant, the fis phenotype is only observed at low frequencies in mea mutants, ranging from 3 to 20%, depending on growth conditions and genetic background, instead of the expected 50% for a gametophytic mutation (Grossniklaus and Vielle-Calzada, 1998; Kiyosue et al., 1999; Wang et al., 2006). This is partially due to genetic redundancy with the MEA paralogue SWN, and thus the mea single mutant still has considerable FIS-PRC2 activity (Wang et al., 2006; Spillane et al., 2007). However, even in mea swn double mutants, the fis phenotype is not completely penetrant (Wang et al., 2006). Therefore, we hypothesized that additional interaction partners of the FIS-PRC2 would aid in repressing target genes and inhibit autonomous seed development. To investigate whether the protein–protein interactions analysed in vitro are biologically relevant, we studied autonomous endosperm development in mea-1/MEA; met1-3/MET1 double heterozygous plants. No autonomous endosperm development has yet been observed in plants carrying a met1 mutant allele alone. However, enhancement of the fis phenotype has been described in plants carrying an fie-1 mutant allele together with an MET1 antisense silencing construct (Vinkenoog et al., 2000). However, in inbred plant lines carrying mutant alleles of met1 or dominant constructs down-regulating MET1 expression, non-specific cumulative effects, e.g. through ectopic epigenetic mis-regulation of other loci, cannot be excluded (Grossniklaus and Vielle-Calzada, 1998; Kiyosue et al., 1999), making interpretation of this effect difficult.

To avoid such indirect effects, we selected plants carrying the met1-3 mutant allele for full methylation at the centromeric 180 bp repeat after out-crossing via the pollen parent (Wöhrmann et al., 2012). Thus, complete loss of MET1 activity first occurs in the gametophytes of these plants. In mea-1/MEA; met1-3/MET1 plants, we observed a significant increase in the frequency of autonomous endosperm development compared to mea-1/MEA heterozygous plants (Figure 4a). In plants carrying only the met1-3 mutant allele, approximately 5% of the developing seeds showed two nuclei in the central cell space. It could not be clearly determined whether these are due to the first autonomous division of the central cell nucleus or due to unfused polar nuclei in the met1-3 background (Figure 4b–d). All developmental stages with two central cell or endosperm nuclei were therefore counted separately. Taking only developmental stages with more than two nuclei into account results in a significant increase in autonomous endosperm development from approximately 4% in mea-1/MEA mutants to approximately 19% in the double heterozygous mutants (< 0.0001, chi-squared test with Yates correction). This strongly indicates a synergistic effect of MEA and MET1 in repression of endosperm development in the absence of fertilization. As the mea single mutant still has considerable FIS-PRC2 activity, this synergism suggests that MEA and MET1 work together in the same pathway.

Figure 4.

MET1 is involved in repression of autonomous endosperm development. Analysis of autonomous endosperm development of ovules 6 days after emasculation in met1-3/MET1, mea-1/MEA and mea-1/MEA; met1-3/MET1 plants. (a) Percentage of ovules with two nuclei in the central cell region (light grey) or at later stages of autonomous endosperm development (>2 nuclei, dark grey). Error bars represent the SD. Siliques of four, three and three plants were analysed for mea-1/MEA, mea-1/MEA; met1-3/MET1 and met1-3/MET1, respectively. N indciates the total number of ovules counted. A χ2- test with Yates correction applied to test for statistical significance of the differences in autonomous endosperm development between mea-1/MEA and mea-1/MEA; met1-3/MET1 gave a P value <0.0001. (b) For met1-3/MET1, only ovules with two nuclei in the central cell area, presumably representing unfused polar nuclei, were observed. (c, d) Two nuclei in the central cell space were observed in mea-1/MEA; met1-3/MET1 at a frequency at approximately 8% (a), representing either unfused polar nuclei or the first division of autonomous endosperm development. (e) Autonomous endosperm development at later stages was observed in approximately 19% (a) of the plants heterozygous for both the mea-1 and met1-3 mutant alleles, in comparison with approximately 4% (a) in mea-1/MEA plants. N, nuclei in central cell region. Scale bars = 40 μm.

DNA methylation in the CpG context mediated by MET1 is reduced at the PHE1 and MEA loci in mea embryos

Our results indicate an interaction between the two major epigenetic pathways of histone and DNA methylation in Arabidopsis. Thus, we tested whether DNA methylation was affected in mea-1 mutants at two MEA target gene loci, PHE1 and MEA itself. Expression of PHE1 during reproductive development was shown to be regulated by both H3K27me3 histone methylation and DNA methylation mediated by MET1 (Makarevich et al., 2006; Makaverich et al., 2008; Dumbliauskas et al., 2011). In addition, de-repression of PHE1 expression was demonstrated previously in both the embryo and endosperm of mea mutants (Köhler et al., 2003b). Thus, we isolated mea-1 mutant and wild-type embryos (late globular to heart stage), which were morphologically clearly distinguishable in heterozygous mea-1/MEA mutant plants. For analysis of the methylation profile, we amplified the region between 2362 and 2661 bp downstream of the PHE1 stop codon, previously described to be relevant for regulation of imprinted PHE1 expression (Villar et al., 2009). The level of DNA methylation in the CpG context maintained by MET1 was slightly reduced at all CpGs analysed (mean of 9.5% for the six CpG residues). In addition, we analysed DNA methylation at the MEA locus in the region between −4374 and −4041 bp upstream of the start codon, a region previously identified as densely methylated (Xiao et al., 2003). Significant changes in DNA methylation were observed at the MEA locus in mea-1 mutant embryos in comparison to the wild type. On four of eight CpG residues analysed, the methylation level was reduced by a mean of 8.4%, but was unchanged at four CpG residues (Figure 5 and Table S1). The decrease in CpG methylation at two MEA target genes is consistent with an interaction between MET1 and MEA (Figure 5 and Table S1). In contrast to CpG methylation, DNA methylation in non-CpG contexts (CpNpG and CpNpN), which is not maintained by MET1, was slightly increased at both loci (Figure 5 and Table S1). To determine the efficiency of bisulfite conversion on DNA samples isolated from mea and wild-type embryos, we amplified 394 bp of the genomic FUSCA3 region, including exonic and intronic sequences. We observed DNA methylation levels below 2% at all cytosine residues, demonstrating a high efficiency of bisulfite conversion (Figure S4).

Figure 5.

DNA methylation of two MEA target genes is affected in mea-1 mutant embryos. Percentage DNA methylation at the 3′ regulatory region of PHE1 and at the MEA locus approximately -4 kb upstream of MEA analysed by bisulfite sequencing on wild-type embryos (top graph) and mea-1 embryos (bottom graph) (late globular to heart stage). Between 21 and 28 embryos were isolated per sample. After DNA isolation and bisulfite conversion, the region between 2362 and 2661 bp downstream of the stop codon and the region from -4374 to -4041 bp upstream of the start codon of PHE1 and MEA, respectively, were PCR-amplified and analysed by 454 sequencing. From two biological replicates each, 417 and 780 reads for the wild-type and 2809 and 343 reads for mea were analysed for the PHE1 locus, and 729 and 946 reads for the wild-type and 711 and 919 reads for mea were analysed for the MEA locus. P values were determined by Fisher's exact test on n the sum of methylated and unmethylated C residues from both replicates.

To test whether the increase in DNA methylation in non-CpG contexts is caused by elevated expression of genes encoding de novo methyltransferases important in the RNA-directed DNA methylation pathway (Cao et al., 2003), we performed RT-PCR for CMT3 and DRM2 (Figure S5). However, no increase in transcript levels was detectable in mea-1 embryos compared to the wild type. In agreement, no significantly higher levels of CMT3 or DRM2 expression were detectable in mea-3 compared to wild type seeds in a microarray study (Tiwari et al., 2010).

A number of genes sharing H3K27me3 and CpG DNA methylation marks are de-repressed in both met1 and mea mutant backgrounds

The decrease in MET1-mediated CpG methylation at two MEA target gene loci, PHE1 and MEA itself, suggested that MET and MEA act synergistically to epigenetically mark and regulate certain genes. We used bioinformatics analysis to identify genes associated with both H3K27me3 histone and CpG DNA methylation from publicly available datasets (Lister et al., 2008; Weinhofer et al., 2010; Bouyer et al., 2011). We mapped CpGs using the region between adjacent genes as the mapping distance and a cut-off of a minimum of 20 methylated CpGs. Of 5634 H3K27me3 marked genes in wild-type seedlings, 2062 were identified as CpG-methylated before fertilization (Lister et al., 2008; Bouyer et al., 2011), and 611 of 1773 genes with H3K27me3 marks were CpG-methylated in the endosperm (Lister et al., 2008; Weinhofer et al., 2010). These genes were used as candidate genes potentially targeted by both epigenetic pathways around fertilization. Using gene set enrichment analysis (GSEA), we then tested which candidate genes are significantly up-regulated in the met1, mea (fis1) fie, or msi1 mutant backgrounds during reproduction in published transcription studies (Köhler et al., 2003b; Lister et al., 2008; Erilova et al., 2009; Tiwari et al., 2010; Weinhofer et al., 2010). Depending on the FIS-PRC2 component mutant background and the platform used for expression analysis, between 15 and 57 genes were identified (Table S2) that are probably targeted synergistically by MET1 and PcG proteins. These genes included the imprinted gene FLOWERING WAGENINGEN (FWA) and a gene encoding a LEA family protein (Table S2), members of which have previously been reported to be up-regulated in the fie mutant background (Bouyer et al., 2011).

Discussion

In sexually reproducing plant species, endosperm and embryo development usually require double fertilization. In fis class mutants, affecting components of the FIS-PRC2, central cell proliferation and endosperm development are initiated in the absence of fertilization in a certain number of ovules (Grossniklaus et al., 1998; Luo et al., 1999; Ohad et al., 1999; Köhler et al., 2003a,b; Wang et al., 2006). Although autonomous endosperm development is only observed in a low percentage of ovules in mea-1/MEA heterozygous mutant plants, this phenotype is significantly enhanced in mea-1/MEA; met1-3/MET1 double heterozygous plants. This indicates that both genes, MET1 and MEA, are involved in the repression of central cell proliferation and endosperm development in the absence of fertilization, suggesting that DNA methylation and histone modifications interact to establish or reinforce the silencing of common target genes. Future studies focused on an analysis of co-expression and transcriptional de-repression of genes in reproductive tissues of met1 and mea mutants, as well as analyses of epigenetic modifications on these genes, will provide more insights into this mechanism.

It has been shown previously that autonomous endosperm development is enhanced in fie-1 mutant plants carrying an MET1 antisense silencing construct (Vinkenoog et al., 2000). While the functional mechanism of the de-repression of genes involved in endosperm development remained unclear, it was hypothesized that it results from global DNA demethylation, which may affect many genes (Vinkenoog et al., 2000). However, recent studies that detected reduced DNA methylation levels in the endosperm as compared to the embryo and vegetative tissues have challenged this hypothesis (Gehring et al., 2009; Hsieh et al., 2009). Global DNA hypomethylation in the endosperm was proposed to be established by transcriptional repression of MET1 in the female gametophyte (Jullien et al., 2008), together with active demethylation by demeter (DME), a DNA glycosylase that removes DNA methylation marks specifically in the central cell (Choi et al., 2002; Jullien et al., 2008). In agreement, a recent study on H3K27me3 targets in the endosperm found that DNA methylation and H3K27me3 marks are mostly mutually exclusive (Weinhofer et al., 2010). However, for a subset of transposable element genes, a synergistic effect of DNA methylation and H3K27me3 was found to cause silencing in the endosperm (Weinhofer et al., 2010), suggesting complex regulation of repression of genes and transposable elements in the endosperm. Furthermore, recent reports also indicate a global down-regulation of PcG-dependent H3K27me3 euchromatic repressive marks in central cell nuclei (Pillot et al., 2010). Thus, as both repressive DNA and histone modifications are globally down-regulated in central cells, it is likely that, in the absence of fertilization, a distinct set of genes regulating central cell proliferation are specifically targeted by repressive pathways. In contrast to the global trend, such genes are likely to be specifically targeted by DNA and histone modifications. In agreement, we identified a number of genes sharing both H3K27me3 and CpG DNA methylation marks and showing higher transcription levels in both met1 and mea mutants during reproductive development.

In this respect, our pull-down assays and yeast two-hybrid experiments suggest that MET1 interacts directly with MEA and potentially also with FIE, probably in the context of the FIS-PRC2, either as a stable component or transiently. This indicates that the two major epigenetic pathways of histone modification and DNA methylation interact directly to repress a distinct set of common target genes involved in endosperm development. This synergistic effect, rather than a global loss of methylation marks, is probably responsible for the enhancement of autonomous endosperm development in met1-3/MET1; mea-1/MEA plants.

A similar interaction of the two major epigenetic pathways mediated by direct protein interactions was demonstrated previously for EZH, EED and DNMT1, the mammalian homologues of MEA, FIE and MET1 (Viré et al., 2006), but genetic evidence was lacking to fully unravel its functional relevance during development. However, recent studies have provided more insights into this question. A significant percentage (44%) of genes that are de novo methylated in colon cancer were found to be PRC2 targets, supporting a functional interaction between DNA de novo methylation and PRC2 activity. However, EZH2 was sufficient to recruit the de novo methyltransferase DNMT3a, but not to confer de novo methylation in a murine erythroleukaemia cell line, suggesting a complex regulation of the interplay of these two epigenetic pathways (Jin et al., 2009; Rush et al., 2009). In both mammals and plants, these two pathways regulate cell proliferation and imprinted gene expression, indicating convergent evolution because imprinting evolved independently in these lineages. However, the physical interaction between these conserved regulatory proteins may have had its origin in their last common ancestor, and they might have only later been independently recruited to silence target genes involved in similar processes in animals and plants. Interestingly, similar interactions between the maintenance DNA methyltransferases MET1/DNMT1 and histone deacetylases, additional players in epigenetic regulatory pathways, have also been reported (Fuks et al., 2000; Liu et al., 2012).

As described for the animal proteins (Viré et al., 2006), several protein domains of MET1 mediate the interaction with MEA FIE, suggesting the possibility that protein interactions play distinct roles in different protein complexes. However, we found no evidence for complex formation between MET1 and MEA independently of the FIS-PRC2, as MET1 does not have a major effect on repression of MEA before fertilization. Thus, it remains unclear how the distinct protein domains contribute to complex formation in planta and whether they are relevant to only one or different complexes regulating distinct developmental processes. There may be a more complex and dynamic regulatory network operating, involving MET1 and PcG proteins, to regulate reproductive development. Different and partially overlapping effects on de-repression of a number of genes targeted by both histone and DNA methylation have been observed for the met1 mutant and for mutants in various FIS-PRC2 components. It is likely that the interactions are dependent on the developmental stage, cell or tissue type, and the genetic background. Future biochemical studies focussing on the identification of different complexes that may be formed between MET1 and FIS-PRC2 components and the structure of the complex(es) are required to shed more light on the involvement of the distinct MET1 domains in the interaction with MEA. In this context, it remains unclear why only two of the three MET1 domains observed to interact with MEA in the GST pull-down assay were found to interact in the yeast two-hybrid assays. However, the possibility of a false-negative result in this assay cannot be ruled out.

MET1, MEA and FIE share overlapping expression patterns in the mature female gametophyte and during early stages of embryo and endosperm development, a prerequisite for a potential in vivo interaction. In a previous report, absence of MET1 in the mature gametophyte was suggested based on the absence of the H2B–RFP marker expressed under the control of the MET1 promoter (Jullien et al., 2008). However, the sensitivity of this assay may depend on the stability and protein turnover of the fusion protein and the genomic sequence used for driving expression of the construct. We detected MET1 expression in the female gametophyte both by in situ hybridization and an analysis of microarray data, showing that the MET1 transcript is not absent but is probably down-regulated to low levels, allowing synergistic effects of the two major epigenetic pathways in reproductive tissues around fertilization.

Interestingly, no fis phenotype was observed so far in met1 single mutants (FitzGerald et al., 2008), but the phenotype is significantly increased in plants with mutations in both MET1 and the FIS class gene MEA. This suggests that at least partially functional repressive FIS-PRC2 is formed in the absence of MET1. The same is possibly true for other components of the FIS-PRC2, as several fis class mutants show autonomous endosperm development, but the trait is often not fully penetrant. Various components of the FIS-PRC2 (PcG proteins and MET1) are probably required to reinforce the silencing of common target genes. Alternatively, the cytosine methyltransferase activity of MET1 may be redundant with other methyltransferases in the regulation of central cell development, or PcG-dependent histone methylation alone may be sufficient to repress autonomous central cell proliferation.

MEA and MET1 both play important roles in either of the two major epigenetic pathways, mediating histone modifications and gene silencing by DNA methylation. It is possible that histone methylation introduced by the FIS-PRC2 is sufficient to repress genes that are important for the initiation of endosperm development. It is likely that the FIS-PRC2 recruits MET1 to introduce stable DNA methylation silencing marks, reinforcing this repression. Interestingly, CpG DNA methylation at the 3′ regulatory region of the PHE1 locus as well as a densely methylated region approximately -4 kb upstream of MEA was decreased in mea mutants. The significant but relatively weak effect may be due to the fact that the FIS-PRC2 is thought to specifically target only the maternal PHE1 and the paternal MEA allele. Since the DNA methylation status of the other allele at this stage of embryogenesis is not known, the effects on the targeted alleles might have been diluted. Unlike the MEA locus, where the same allele is targeted by both MET1 and the FIS-PRC2, the paternal PHE1 allele is marked by DNA methylation while the maternal one is targeted by histone methylation. Thus, trans-interactions between the two alleles have to be postulated, which may be weak and/or transient. Alternatively, the weak effects on DNA methylation may correlate with a gradual loss of CpG methlyation during embryo development in mea-1 mutants, or may indicate that other PcG proteins with homology to MEA act redundantly during embryo development. In this regard, CURLY LEAF and SWN have already been shown to act redundantly on the regulation of PHE1 at later stages of development, and were shown to be expressed in Arabidopsis embryos (Makarevich et al., 2006; Spillane et al., 2007). In contrast to CpG methylation, DNA methylation in non-CpG contexts was slightly increased at both loci, which may indicate a more complex regulation, possibly also involving de novo methyltransferases. Potentially, this may contribute to the de-repression of PHE1 in mea mutant embryos as previously described, because expression of the paternally active PHE1 allele is dependent on DNA methylation (Köhler et al., 2003b; Makaverich et al., 2008). This increase in DNA methylation in a non-CpG context does not appear to involve a general increase in expression of the de novo methyltransferases DRM2 and CMT3 in mea embryos, but may depend on their activity at specific target loci only. Alternatively, CpG and non-CpG methylation may be interconnected in more complex ways. For instance, it has already been proposed that reduction of CpG DNA methylation through DME in the endosperm is required for the enhanced activity of the small interfering RNA (siRNA) pathway in seeds (Hsieh et al., 2009). Global CpG methylation was increased in dme mutant compared to wild-type endosperm, but non-CpG methylation was decreased (Hsieh et al., 2009). In particular, an increase in CpG and a decrease in non-CpG methylation was observed in dme endosperm at the MEA 3′ repeats (Hsieh et al., 2009). This behaviour is opposite to the decrease in CpG methylation and increase in non-CpG methylation observed in the mea mutant background for two target loci, in agreement with the role of DME as an antagonist of MET1. Our findings thus support the notion that pathways mediating non-CpG methylation in seeds are dependent on a decrease in CpG methylation. In summary, our study indicates that MET1 physically interacts with MEA, a core component of the FIS-PRC2 of Arabidopsis, to repress central cell proliferation and endosperm development in the absence of fertilization.

Experimental Procedures

Plant material and growth conditions

Arabidopsis thaliana (L.) Heynh. accessions Landsberg erecta (Ler) and Columbia (Col-0) were used in this study. Seeds were sterilized in 70% EtOH and subsequently in 1.25% sodium hypochlorite. Plants were grown in a growth chamber under 16 h light/8 h dark cycles at 21°C/18°C, respectively, and 70% humidity. Mutant lines used in this study are described in Data S1.

Histological analysis

To test for autonomous endosperm development, flowers were emasculated and analysed 6 days after emasculation. Tissues were cleared in chloral hydrate/glycerol/water (8:1:2 w/v/v), and subjected to microscopic analysis after fixation using ice-cold 80% acetone for at least 24 h and rehydration in water twice for 10 min.

In situ hybridization

In situ hybridization was performed as described previously (Vielle-Calzada et al., 1999) with the following modifications. A 183 bp fragment of the MET1 transcribed region was amplified from a cDNA reverse-transcribed from RNA isolated from Col-0 inflorescences using primers P1 and P2 (Table S3). The fragment was cloned into pDrive (Qiagen, http://www.qiagen.com) in the sense and antisense orientations and used for transcription with T7 reverse transcriptase.

Microscopy

For cleared and in situ hybridized specimens, the slides were viewed under a Leica DMR microscope (Leica Microsystems, http://www.leica-microsystems.com), and photographs were taken using a digital camera (Magnafire model S99802, Optronics, http://www.optronics.com). Pictures were cropped and processed using Adobe Photoshop version 8.0.1 (Adobe Systems Inc., http://www.adobe.com).

In vitro pull-down assay

The cloning of constructs is described in Data S1. E. coli BL21 DE3 harbouring the pGEX-4T, pGEX_MEA, pGEX_MET1Nterm, pGEX_MET1BAH or pGEX_MET1Cterm plasmids were grown overnight in 2x YT (16 g Bacto-tryptone, 10 g Bacto-yeast extract, and 5 g NaCl in 1 l of water, adjusted to pH 7.2 with NaOH and autoclaved) medium at 37°C. Cultures were diluted 1:100 in 2x YT medium containing 0.2% glucose, and grown to an OD600 of 0.7–1.0, followed by addition of isopropyl thio-β-d-galactoside to 1 mm. Cells were pelleted for 10 min at 5000 g at 4°C after induction for 4–5 h at 28°C, and resuspended in binding buffer {20 mm Tris/HCl pH 7.5, 130 mm NaCl, 1 μm ZnSO4, 10% glycerol, 1x PI [complete Mini protease inhibitor cocktail tablet (Roche, http://www.roche.com)]}. Lysozyme to a final concentration of 0.5 mg ml−1 and phenylmethanesulfonyl fluoride to a concentration of 1 mm were added to the resuspended cells. The cells were incubated for 20 min on ice followed by brief sonication. Triton X-100 or IPEGAL (C-630) was added to 0.1%, and the samples were shaken for 30 min prior to centrifugation twice for 10 min at 12 000 g and 4°C. 35S-methionine-labelled MET1 and MEA proteins were generated using the TNT® Quick Coupled Transcription/Translation System (Promega, http://www.promega.com) according to the manufacturer's instructions. GST pull-down reactions were performed in a total volume of 5 ml binding buffer supplemented with 1 mg ml−1 BSA combining protein extracts and 45 μl of the in vitro transcription/translation reaction. Glutathione bead slurry (40 μl) was added to the reaction mixture, and incubation was continued for 2 h 30 min at 4°C. After six washes with binding buffer containing 0.1% detergent (no prolonged incubation was applied during the washes), proteins were eluted with SDS sample buffer or 25 μl of 50 mm Tris/HCl (pH 8.1), 75 mm reduced glutathione, separated on a 12% acrylamide gel. The gel was fixed for 30 min in 50% methanol, 10% acetic acid and 40% water before treatment with NAMP 100 Amplify solution (Amersham, http://www.ge.com). The gel was dried for 1 h under vacuum at 75°C. For detection a Hyperfilm (Amersham, http://www.gelifesciences.com) was exposed for up to 12 h at -70°C.

Yeast two-hybrid assay

The cloning of constructs is described in Data S1. The yeast strain EGY48[p8op-lacZ] was used (Estojak et al., 1995), and transfection of yeast and yeast two-hybrid assays were performed following the manufacturer's instructions (Clontech, http://www.chlontech.com) using SD induction medium containing X-Gal (5-bromo-4-chloro-indolyl-β-d-galactopyranoside) to test activation of the lacZ reporter gene.

Analysis of DNA methylation at the PHE1 and MEA loci of isolated mea-1 and wild-type embryos

Two biological replicates each of 25 and 28 mea-1 embryos and 21 and 22 wild-type embryos were isolated from mea-1/MEA heterozygous plants and wild type segregants of the same population, respectively. Isolated embryos were at the late globular to heart stage, a stage at which mea-1 seeds are easily morphologically distinguished from the wild type. Embryo isolation was performed as described previously (Autran et al., 2011), except that embryos were released in standard TE buffer (10 mM Tris, 1 mM EDTA, adjusted to pH 8.0 with HCl) and afterwards subjected to bisulfite conversion (see Data S1). Bisulfite DNA was eluted in 50 μl TE, and 5 μl DNA were used for a 30 μl PCR reaction to amplify the PHE1 3′ region 2362–2661 bp downstream of the stop codon, the region -4374 to -4041 bp upstream of the MEA start codon, and the genomic region 948–1342 bp downstream of the FUSCA3 start codon using HotStart-IT Taq DNA polymerase (USB, http://www.usb.affymetrix.com) and primers P17 and P18, P19 and P20, or P21 and P22, respectively (Table S3). Purified PCR products were sequenced using a 454 sequencer according to the standard protocol (Roche, http://www.454.com). The sequenced data were processed using BiQ AnalyzerHT software (http://biq-analyzer-ht.bioinf.mpi-inf.mpg.de, Lutsik et al., 2011).

Quantitative PCR and RT-PCR

Quantitative RT-PCR to measure MEA or CMT3 and DRM2 transcription in wild-type and met1-3/MET1 plants or mea-3 embryos, respectively, was performed as described previously (Baroux et al., 2006). Details for the RT-PCR are provided in Data S1.

Heatmap

The heatmap was generated using the Bioconductor package ‘gplots’ (www.bioconductor.org).

Analysis of genes co-repressed in met1 and mutants of FIS-PRC2 core components

A detailed description of the bioinformatics analysis is provided in Data S1.

Acknowledgements

We thank Justin Goodrich (University of Edinburgh, Institute of Molecular Plant Sciences, UK), Claudia Köhler (University of Agricultural Sciences, Department of Plant Biology and Forest Genetics, Uppsala, Sweden), Mark Curtis, Umut Akinci and Tina Jordan (University of Zürich, Institute of Plant Biology, Switzerland) for providing plasmids, E. coli and yeast strains, Daniel Bouyer and Arp Schnittger (Institut de Biologie Moléculaire des Plantes du Centre National de la Recherche Scientifique, Strasbourg, France) for plant material, Célia Baroux (University of Zürich, Institute of Plant Biology, Switzerland) for critical comments on the manuscript, Marian Bemer (University of Zürich, Institute of Plant Biology, Switzerland) for helpful discussions, and Samuel E. Wuest and Marc W. Schmid (University of Zürich, Institute of Plant Biology, Switzerland) for help with bioinformatics analyses. This work was supported by the University of Zürich, and grants from the Swiss National Science Foundation (31003A-12006) and the European Research Council (250358) to U.G.

Ancillary