The Arabidopsis CUL4–DDB1 complex interacts with MSI1 and is required to maintain MEDEA parental imprinting

Recent work identified DDB1 and DCAFs as possible substrate receptors of CRL4 E3 ligases (reviewed in Lee and Zhou, 2007). The largest class of DCAFs are WD40 repeat proteins, which interact with DDB1 via one or several conserved WDxR motifs. The Arabidopsis genome encodes 237 WD40 repeat proteins; however, only a subset of them (∼80 proteins) carry one or more WDxR motif(s) (Lee et al, 2008 and our unpublished data). Among these proteins we identified MSI1 and four other Arabidopsis MSI1‐related proteins, named MSI2–MSI5 (reviewed in Hennig et al, 2005). When all MSI1‐like proteins from plant and non‐plant organisms were compared, it appeared that most of them share a highly conserved WDxR motif (Figure 1). In metazoans, MSI1‐like proteins exhibit also a second WDxR motif, which is less conserved in plants, but is also present in fungi. Therefore, most if not all MSI1‐like proteins are structurally related to DCAFs.

We first investigated whether MSI1 interacts with DDB1A in a yeast two‐hybrid assay. Similarly to DDB2 (Molinier et al, 2008), which served as a positive control, MSI1 and DDB1A interacted, although the interaction was weak as yeast growth was only detected on (‐LWH) medium (Figure 2A). We further confirmed this interaction by an in vitro pull‐down assay. In this experiment, a fusion protein between glutathionine‐S‐transferase (GST) and DDB1A, GST‐DDB1A, was incubated with in vitro translated, ³⁵S‐methionine‐labelled MSI1 or DDB2. Consistently, MSI1 and DDB2 co‐precipitated with GST‐DDB1A, but not with GST alone (Figure 2B). To provide evidence for a physical interaction between both proteins in plant cells, we carried out bimolecular fluorescence complementation (BiFC) experiments. Plasmids YC‐MSI1 and YN‐DDB1A were co‐bombarded into etiolated mustard hypocotyls. A strong YFP signal was observed in the nucleus of 81% examined cells (35/43; Figure 2C). These data are similar to those obtained with cells transformed with the positive control YN‐DDB1A+YC‐DDB2 (43/46). Only a weak fluorescence signal was observed after bombardment with the following plasmid combinations YN‐DDB1A+YC‐BPM3 (9/35) and YN‐BPM3+YC‐MSI1 (2/27), where BPM3 (BTB/POZ‐MATH3 protein encoded by At2g39760) is a nuclear cullin‐ring ubiquitin ligase3 (CLR3) receptor, used here as a negative control. Taken together, our data clearly demonstrate a physical interaction between DDB1A and MSI1.

Next, we tested whether MSI1 is also part of a protein complex containing Arabidopsis CUL4. Thus, we immunoprecipitated Arabidopsis CUL4 from plants expressing the MSI1–RFP fusion protein under the control of its own promoter (Chen et al, 2008). Hence, MSI1 was successfully co‐immunoprecipitated in this assay (Figure 2D). Since CUL4 interacts with DDB1A (Bernhardt et al, 2006) our results, collectively, support the existence of a CUL4–DDB1A–MSI1 protein complex in Arabidopsis.

In Arabidopsis, loss‐of‐function of MSI1 causes maternal effect embryo lethality leading to seed abortion early in development (Köhler et al, 2003). We have previously isolated a T‐DNA mutant, cul4‐1 (Bernhardt et al, 2006), in which CUL4 expression was severely downregulated. Although viable cul4‐1 homozygous mutants were obtained, these plants showed various developmental abnormalities (Bernhardt et al, 2006). When selfed, we noticed that cul4‐1 homozygous plants exhibited altered seed development leading eventually to seed abortion (Supplementary Figure S1). Thus, we examined cul4‐1 homozygous mutant seeds at different developmental stages (Figure 3). Already at the octant stage, we observed a lower proliferation of the endosperm (Figure 3B) while at later seed developmental stages we scored abnormally large endosperm nuclei and delayed embryo development (Figure 3D and F). Because of the pleiotropic and hypomorphic nature of the cul4‐1 allele, we aimed to identify amorphic CUL4 loss‐of‐function mutants. As no such mutants were available in public collections, we screened a collection of Arabidopsis T‐DNA insertion lines (Ríos et al, 2002). Two T‐DNA insertions were identified within the coding region of CUL4, called cul4‐2 and cul4‐3 (Supplementary Figure S2A). Both cul4‐2 and cul4‐3 mutants were backcrossed to the wild type and Southern blots confirmed single T‐DNA insertions. Although we genotyped 137 and 72 progeny from selfed cul4‐2 and cul4‐3 mutant plants, respectively, we were unable to identify homozygous mutants, suggesting that CUL4 is an essential gene in Arabidopsis.

As both lines contained single T‐DNA insertions with integral hygromycin selection markers, we self‐pollinated cul4‐2 and cul4‐3 heterozygous plants and analysed the segregation of this marker among their progeny (Table I). This genetic analysis revealed a segregation ratio close to 2:1 consistent with nearly fully penetrant zygotic embryo lethality. Because the segregation ratio of the marker was slightly below 2:1 for the cul4‐2 allele, suggesting a weak defect in gametophytic transmission, we performed reciprocal crosses with wild‐type plants. The transmission efficiency of the marker was slightly reduced through both male and female gametophytes (Table I).

Next, we examined mature siliques for the presence of aborted seeds. The number of aborted seeds was consistent with zygotic embryo lethality, where a segregation of aborted:normal seeds of 1:3 is expected (Table II; Supplementary Figure S1). To further investigate at which developmental stage embryogenesis is arrested, we analysed cleared seed specimens from siliques of selfed cul4‐2 mutant plants at different developmental stages. At the octant stage, the mutant seeds exhibited a low number of large nuclei in the endosperm (Figure 3H). At later stages, embryos arrested their development at the globular stage with abnormal shapes and cell division defects in both the suspensor and the embryo proper (Figure 3J and L). Moreover, in cul4‐2 homozygous mutant seeds, the endosperm was always severely underdeveloped with a dozen fewer enlarged, abnormal nuclei. When siliques were analysed at later stages, harbouring bent‐cotyledon stage or mature wild‐type sibling embryos, the arrested seeds had degenerated (not shown), indicating a strict arrest and not only a delay in seed development. Similar results were obtained with the cul4‐3 mutant allele (Supplementary Figure S3B).

Because CUL4 interacts with DDB1 to form CRL4 E3 complexes, we also investigated whether DDB1 is required for embryogenesis. The Arabidopsis genome encodes two expressed DDB1‐related proteins, named DDB1A (At4g05420) and DDB1B (At4g21100), exhibiting 89% sequence identity at the amino‐acid level (Schroeder et al, 2002). DDB1A loss‐of‐function mutants are viable (Molinier et al, 2008). Therefore, we searched for T‐DNA insertion mutants in the related DDB1B gene and identified one mutant, named ddb1b‐1, from the SALK collection (SALK 061944) (Alonso et al, 2003). In the dbb1b‐1 allele, the T‐DNA interrupts the coding sequence in the last exon (Supplementary Figure S2B). Homozygous ddb1b‐1 mutant plants developed normally and were fully fertile. To test whether DDB1A and DDB1B act redundantly during embryogenesis, the ddb1a‐2 mutant was used to pollinate a homozygous ddb1b‐1 mutant plant. Among the progeny of this cross, we selected F2 plants that were DDB1A/ddb1a‐2 ddb1b‐1/ddb1b‐1 (referred as DDB1A/ddb1a ddb1b) and ddb1a‐2/ddb1a‐2 DDB1B/ddb1b‐1 (referred as ddb1a DDB1B/ddb1b). Because both ddb1a‐2 and ddb1b‐1 mutants carry the same selection marker, we used PCR‐based genotyping for further genetic analyses. Among the progeny of self‐pollinated DDB1A/ddb1a ddb1b and ddb1a DDB1B/ddb1b plants, no double mutant were identified, despite the analysis of ∼60 plants for each genotype (Table III).

Next, we evaluated the effect of both DDB1‐related genes on male and female gametophytic transmission (Table III). Reciprocal crosses between the different genotypes revealed that the two genes do not contribute equally to gametophyte development and/or function as indicated by unequal transmission defects: while in the absence of DDB1B, DDB1A is required for normal transmission through the male, the converse is true for the female gametophyte.

The number of aborted seeds was consistent with zygotic embryo lethality in self‐pollinated ddb1a DDB1B/ddb1b plants (Table II). Light microscopic observations of cleared seeds revealed that double homozygous ddb1a dbb1b embryos derived from selfed DDB1A/ddb1a ddb1b (not shown) or ddb1a DDB1B/ddb1b (Supplementary Figure S3D)) mutants arrest at the globular stage, with a phenotype reminiscent of that of the cul4 mutants. Thus, both CUL4 and DDB1A/B functions are required for normal development of embryo and endosperm.

To determine the expression pattern of CUL4 in reproductive tissues and during embryogenesis, we performed mRNA in situ hybridization experiments on sections of flower buds and developing siliques using CUL4‐specific antisense and sense control probes. CUL4 transcripts were detected in the tissues of young flower buds, that is in petals, stamens, and carpels (Supplementary Figure S4A). A distinctive signal was observed in emerging ovules (Supplementary Figure S4B), but not in the developing embryo sac (Supplementary Figure S4E). After fertilization, the expression level of CUL4 was prominent in the developing embryo (Supplementary Figure S4C, D, G–L). The signal intensity decreased after the heart stage (Supplementary Figure S4I–G). This expression pattern correlates with the requirement of CUL4 for embryo development. In endosperm cells, the hybridization signal was low but detectable at all stages of seed development (Supplementary Figure S4C, D, G–L). Overall, the CUL4 expression pattern in developing seeds is consistent with the CUL4 loss‐of‐function phenotype.

MSI1 is a component of the FIS–PRC2 complex, which is required for seed development (Köhler et al, 2003). However, MSI1 was also found in other protein complexes potentially involved in chromatin functions (Hennig et al, 2005). Thus, we wondered whether loss of CUL4 and/or DDB1 activity affects PRC2‐like functions during plant reproduction. In contrast to the msi1 mutant, cul4‐1 siliques did not show a clear elongation after emasculation (Supplementary Figure S5A). Nevertheless, we observed 16% (n=229) and 3.6% (n=224) of autonomous endosperm division in homozygote cul4‐1 and heterozygote cul4‐2 mutants, respectively (Supplementary Figure S5B). As expected, no extra divisions were observed in wild‐type ovules (n=71) (Supplementary Figure S5B). Thus, although at a lower penetrance, cul4 mutants share the fis class phenotype of endosperm initiation in the absence of fertilization (Ohad et al, 1996; Chaudhury et al, 1997; Grossniklaus and Vielle‐Calzada, 1998; Kiyosue et al, 1999; Köhler et al, 2003; Guitton et al, 2004). Next, we investigated whether mutations of CUL4 affect parental imprinting of MEA and/or PHE1, two genes that are regulated by the FIS–PRC2 complex (Köhler et al, 2003, 2005; Baroux et al, 2006; Gehring et al, 2006; Jullien et al, 2006). We used sequence polymorphisms between different Arabidopsis accessions to distinguish parental alleles. Interestingly, we detected paternal MEA expression in the cul4‐1 mutant (Figure 4A). It is noteworthy that repression of the paternal MEA allele in the control experiment was not complete, as a weak but detectable expression was observed when Columbia (Col‐0) pollen was used. Indeed, previous genetic analyses suggested that Col‐0 carries a paternal modifier of mea seed abortion (Vielle‐Calzada et al, 1999), and it is possible that this leads to a weak de‐repression of the paternal MEA allele. However, full expression of paternal MEA allele was only observed 3 days after pollination (DAP) when cul4‐1 or cul4‐1 ddb1a pollen was used.

To further investigate the loss of repression of the paternal MEA allele in the cul4‐1 or cul4‐1 ddb1a mutants, we introgressed the pMEA::MEA‐YFP reporter gene (Wang et al, 2006a) into the cul4‐1 and cul4‐2 mutant backgrounds. When pMEA::MEA‐YFP line in a Col‐0 background was used to pollinate Col‐0 plants, we observed by confocal microscopy a detectable fluorescence signal in the endosperm of about half of the seeds (Figure 4B; Table IV). This result is inconsistent with a full repression of the paternal MEA allele in the Col‐0 background, but is in agreement with our results using sequence polymorphisms (see above). A similar result was observed when the cul4‐1 hypomorphic mutant was used as a female and pollinated with wild‐type pollen carrying the pMEA::MEA‐YFP reporter. However, when we used the cul4‐2 pMEA::MEA‐YFP pollen to fertilize Col‐0 plants, we observed a stronger fluorescence signal in some seeds (Figure 4B), supporting a reactivation of the paternal pMEA::MEA‐YFP reporter gene if derived from cul4 mutant pollen.

To better deplete CUL4 activity in such experiments, we combined the weak cul4‐1 with the strong cul4‐2 alleles. When homozygous cul4‐1 plants were used as a female and pollinated with pollen from cul4‐2 heterozygote plants, we observed a category of seeds (∼50%), which arrested at the late globular stage, thus corresponding to cul4‐1/cul4‐2 homozygous embryos. In these seeds, the underdeveloped endosperm contained large coenocytic cells (Supplementary Figure S6) and degenerated at 4 DAP. Interestingly, ∼70% of seeds were scored for a fluorescent signal when the cul4‐1 mutant was pollinated with pollen from cul4‐2 pMEA::MEA‐YFP plants (Table IV). The fluorescence signal was particularly strong in the aberrant endosperm of arrested cul4‐1/cul4‐2 seeds at 2 and 3 DAP (Figure 4B). This finding was further supported by the accumulation of paternally expressed MEA‐YFP protein in this cross, as detected by western blotting (Figure 4C). Thus, our results indicate that CUL4 is necessary to maintain repression of the paternal MEA allele.

Since the PRC2 complex mediates trimethylation of histone H3 at the lysine residue 27 (H3K27me3) (Gehring et al, 2006; Jullien et al, 2006; Makarevich et al, 2006), we investigated whether CUL4 knockdown affects histone methylation. Interestingly, chromatin immunoprecipitation (ChIP) analysis performed on young siliques revealed a decrease in H3K27 trimethylation at the MEA locus (Figure 5B). It is remarkable that a similar effect was not observed when ChIP assays were performed with young floral buds (Figure 5A), suggesting that CUL4 is involved in the maintenance rather than in the establishment of those histone repressive marks.

In vegetative tissues, both alleles of MEA are silenced by PRC2 complexes through the deposition of H3K27me3 marks on chromatin (Gehring et al, 2006; Jullien et al, 2006). Thus, we tested whether CUL4 is also required to repress MEA later during development. Indeed, MEA expression was detected in homozygous cul4‐1 knockdown plants, though to a lesser extend than in a mutant compromised in the SET‐domain protein CURLY LEAF (CLF) used here as control (Supplementary Figure S7). Moreover, ChIP analysis revealed that MEA reactivation was correlated with a decrease in H3K27 trimethylation (Supplementary Figure S7C).

An intriguing observation was that in contrast to MEA, we did not observe the activation of the maternal PHE1 allele in the cul4 knockdown mutant when using the different Arabidopsis accessions (Figure 6A). Nevertheless, when examining H3K27 trimethylation at the PHE1 locus (Figure 5), we observed a clear reduction in the repressive histone marks, to a similar level as at the MEA locus. Although parental imprinting of both genes requires the FIS–PRC2, it was however recently found that Polycomb‐dependent silencing is necessary but not sufficient to establish PHE1 imprinting. Indeed, a distantly located region downstream of the PHE1 locus, named DMR (differentially methylated region), has also an important function in silencing the maternal PHE1 allele (Makarevich et al, 2008). Thus, it is possible that the reduction in repressive histone methylation marks seen in cul4‐1 is not sufficient to impair maternal PHE1 silencing. To examine this possibility, we took advantage of the PHE1‐GUS reporter construct, which contains the ∼3 kbp promoter sequence but no 3′ regulatory elements (Köhler et al, 2003). The paternally derived PHE1‐GUS transgene was similarly expressed in wild‐type (n=55; Figure 6B) and cul4‐2 heterozygote mutant seeds (n=49; Figure 6C). Expression of the maternally derived PHE1‐GUS was mainly restricted to the chalazal endosperm when pollinated with wild‐type pollen (Figure 6B). In contrast, 3 DAP with heterozygote cul4‐2 pollen, the maternally derived PHE‐GUS transgene was expressed at a higher level in about half of the seeds (56 out of 118 seeds exhibited GUS staining in the endosperm; Figure 6C). A similar result was observed with pollen from heterozygote msi1 mutant plants (78 out of 148 seeds exhibited GUS staining in the endosperm; Figure 6D). Thus, paternally expressed CUL4 and MSI1 are both required to maintain the silencing of the maternal PHE1‐GUS reporter, suggesting haplo‐insufficiency. It is, therefore, likely that the silencing of endogenous maternal PHE1 is maintained in cul4 mutant seeds due to additional regulatory elements, most likely located in the 3′ region of the gene (Makarevich et al, 2008).

Next, we investigated the kinetics of both MEA transcripts and protein accumulation. The Col‐0 pMEA::MEA‐YFP line was self‐pollinated and MEA‐YFP mRNA and protein levels were determined before and during the first 3 DAP. While the MEA mRNA level gradually decreased after pollination, MEA protein quickly disappeared and was hardly detectable at 2 DAP (Figure 7A). Thus, the MEA protein level is highly dynamic and seems under both transcriptional and post‐transcriptional control. Therefore, we investigated MEA‐YFP protein accumulation in cul4 mutant plants. It is noteworthy that the pMEA::MEA‐YFP transgene was partially silenced in the cul4‐2 background. This phenomenon, at least in part, could explain the lower MEA‐YFP protein content detected in the mutant (Figure 7B). Since MEA transcript levels did not decay in the cul4‐2 mutant (Figure 7B), it seems that the MEA‐YFP protein accumulation mainly reflects transcriptional regulation by CUL4, though a post‐transcriptional regulation cannot be excluded. Nevertheless, our results clearly demonstrate that CUL4 activity is necessary to restrict MEA expression during seed development.

The Arabidopsis ddb1a‐2 mutant is described in Molinier et al (2008). The ddb1b‐1 (SALK 061944) mutant was identified using the web‐assisted program at http://signal.salk.edu/cgi-bin/tdnaexpress. The precise location of the T‐DNA in ddb1b‐1 mutant was determined by sequencing, showing an insertion after nucleotide 6211 in the last exon of DDB1B. The cul4‐1 (GABI‐KAT 600H03) mutant is described in Bernhardt et al (2006). The cul4‐2 (Koncz 42460) and cul4‐3 (Koncz 57891) mutants were identified by PCR screening of the Köln Arabidopsis T‐DNA mutant collection (Ríos et al, 2002). The precise location of the T‐DNA in cul4‐2 and cul4‐3 was determined by sequencing, showing an insertion after nucleotide 4478 in the 14th exon and after nucleotide 2718 in the 7th intron of CUL4, respectively.

For in vitro culture, seeds were surface sterilized using the ethanol method, plated on GM medium (MS salts (Duchefa, The Netherlands), 1% sucrose, 0.8% agar, pH 5.8) in the presence or absence of a selectable agent, stored 2 to 3 days at 4°C in the dark, and then transferred to a plant growth chamber under a 16‐h/8‐h photoperiod (22uC/20uC). For soil‐cultured plants, seeds were sown (20/pot) and put at 4°C in the dark during 3 days. Two weeks later, single plants were transferred to pots in the greenhouse and kept under a regime of 16 h/8 h photoperiod (20uC/16uC; 70% humidity).

The DDB2 and MSI1 cDNAs were cloned as fusions to the GAL4 activation domain and the DBB1A cDNA fused to the GAL4‐binding domain, respectively, in Gateway‐compatible pGAD424gate and pGBT9gate (Ghent plasmids collection, http://bccm.belspo.be/index.php) yeast two‐hybrid vectors. The yeast strain AH109 (Clontech) was transformed with the appropriate combinations of bait and prey vectors. Transformants were selected on synthetic (SD)/‐Leu/‐Trp (‐LW) media and interactions were tested on SD/‐Leu/‐Trp/‐His (‐LWH) or SD/‐Leu/‐Trp/‐Ade (‐LWA) media, allowing growth for 4 days at 28°C.

The full‐length DDB1A cDNA was cloned into Gateway vector pDEST15 (Invitrogen) by recombination for expression in Escherichia coli BL21AI (Invitrogen). In this construct GST is placed in frame at the N‐terminus of DDB1A protein. After 4 h of 0.2% Arabinose induction at 16°C, the fusion proteins were purified on bulk gluthatione‐sepharose following the manufacturer's instructions (GE Healthcare). For GST pull‐down assays, DDB2 and MSI1 proteins were translated in vitro, using the TNT7‐coupled wheat germ extract system (Promega) and radiolabelled with [³⁵S]‐methionine. Purified GST‐DDB1A or control GST proteins, immobilized on glutathione‐sepharose beads, were incubated for 2 h at 4°C with equal amounts of ³⁵S‐methionine‐labelled DDB2 and MSI1 protein following the manufacturer's instructions (GE Healthcare). Labelled DDB2 and MSI1 proteins were detected by autoradiography.

The DDB1A, DDB2 and MSI1 coding sequences were cloned into the split YFP destination vectors by recombination (Invitrogen) in order to obtain the YN‐DDB1A, YC‐DDB2 and YC‐MSI1 constructs. The YN‐, YC‐ and ³⁵S‐CPRF2‐CFP vectors and split YFP experiments were performed as described in Stolpe et al (2005). Vectors bearing YN or YC either alone or fused to BPM3 (At2g39760) were used as negative controls. Images were recorded 20 h after bombardment with a Nikon fluorescent stereomicroscope E800 equipped with a 40 × water immersion optic by using CFP‐ and YFP‐specific filters.

Total soluble proteins were extracted from pMSI1::MSI1–RFP plants (Chen et al, 2008) using buffer A (100 mM NaHPO₄ pH 8.0, 1% Triton × 100, protease inhibitor mix (Complete; Roche Molecular Biochemical)). Immunoprecipitation assay was performed using anti‐CUL4 polyclonal antibody coupled to ProteinA‐sepharose beads. The MSI1–RFP protein was detected using anti‐DsRed antibodies (Clontech Laboratories Cat^#632496) diluted 1:1000 (v:v).

Developing seeds were prepared from siliques of different developmental stages and directly mounted on microscope slides in a clearing solution of 8:2:1 chloral hydrate:distilled water:glycerol as described in Grini et al (2002). Observations were performed with a Zeiss Axiophot or a Leica DMR microscope under differential interference contrast (DIC) × 20 and × 40 optics. For YFP marker analysis, seeds from dissected siliques at different DAP were mounted in a 1:10 glycerol:distilled water. Specimens were observed under a Zeiss confocal laser‐scanning microscope. Histochemical assays to detect GUS activity were performed as described in Capron et al (2003).

In situ hybridization was performed on the 10‐μm Paraplast Plus (Sigma) sections as described in Brukhin et al (2005). For probe synthesis, the fragment spanning the region of the CUL4 (At5g46210) cDNA from 1501 to 2000 (500 bp) sequence was used. The fragment was inserted into the pGemT plasmid (Promega). Sense and antisense digoxygenin‐UTP‐labelled riboprobes were generated by run‐off transcription using T7 and Sp6 RNA polymerases, respectively. The probes were hydrolyzed into 120 bp fragments in carbonate buffer, pH=10.2 for 59 min at 60°C.

RNAs from siliques at 3 DAP and from 17‐day‐old plantlets were prepared with the Trizol reagent (Invitrogen). In all, 5 μg of total RNA was treated with Dnase1 kit (Fermentas) and reverse transcribed with Superscript III Reverse Transcriptase kit (Invitrogen). To detect PHE1 and MEA mRNA, allele‐specific RT–PCR was performed as described previously in Kinoshita et al (1999) and Köhler et al (2005). Primers used to amplify the control GAPDH (glyceraldehyde‐3‐phosphate dehydrogenase) RNA, were GAPDH3′ (5′‐GTAGCCCCACTCGTTGTCGTA‐3′) and GAPDH5′ (5′‐AGGGTGGTGCCAAGAAGGTTG‐3′). To detect DDB1B, specific primers DDB1FWD (5′GGAAAATGAACCAACTAAGGAAGG‐3′) and DDB1bREV (5′AGAGCTTGGATTTGCTTCAGTG‐3′) were used. To detect EF1α‐specific primers EF1Fwd (5′‐TTGCTCCACAGGATTGACCACTG‐3′) and EF1Rev 5′‐TCACTTCGCACCCTTCTTGACG‐3′) were used.

Total RNA for quantitative PCR (qPCR) was extracted from inflorescences and siliques at different DAP of pMEA::MEA‐YFP plants in Col‐0 and cul4‐2 heterozygous mutant backgrounds, respectively, using the kit RNeasy MINI PLUS (Qiagen). In all, 2 μg of total RNA was reverse transcribed with Superscript III Reverse Transcriptase kit (Invitrogen). PCR was performed using gene‐specific primers in a total volume of 10 μl SYBR Green Master mix (Roche) on a Lightcycler LC480 apparatus (Roche) according to the manufacturer's instructions.

The mean value of three replicates was normalized using the ACTIN2 (AT3G18780), GAPDH (AT3G26650) genes as internal controls.

Primer list:

MEA: GCAGGACTATGGTTTGGATG and CACCTTGAGGTAACAATGCTC

YFP: ATATCATGGCCGACAAGCA and GAACTCCAGCAGGACCATGT

ACTIN2: CTTGCACCAAGCAGCATGAA and CCGATCCAGACACTGTACTTCCTT

GAPDH: TTGGTGACAACAGGTCAAGCA and AAACTTGTCGCTCAATGCAATC

ChIP experiments were performed as described by Jiang et al (2008) using young flowers before fertilization and young siliques around 3 DAP as well as 17‐day‐old plants. A measure of 5 ml mixed tissue powder was prepared after cross‐linking DNA with proteins by formaldehyde. Preparation of chromatin, sonication, and immunoprecipitation using anti‐H3 (05‐499; Millipore), anti‐trimethyl‐histone H3K27 (07‐449; Millipore) and anti‐acetyl‐H3 (06‐599; Millipore) antibodies were carried out using Millipore ChIP kit according to the manufacturer's instructions. The immunoprecipitated DNA was analysed by real‐time qPCR using MEA primers (region −700–1500; Baroux et al, 2006) and PHE1 primers (Makarevich et al, 2006). FUSCA primers (Kwon et al, 2009) were used as internal standards for normalization. Data analysis was done as described in Mutskov and Felsenfeld (2004). Experiment was performed three times using independent biological samples.

Total proteins were extracted from Col‐0 and cul4 siliques at 1, 2 and 3 DAP using denaturating buffer as described in Büche et al (2000). A measure of 40 μg of total protein extracts were separated on SDS–PAGE gels and blotted onto Immobilon‐P membrane (Millipore). The MSI1–RFP protein was detected by using the anti‐DsRed antibody (Clontech Laboratories Cat^#632496) diluted 1:1000 (v:v), whereas the MEA‐YFP protein was identified by using a Rabbit anti‐GFP polyclonal antibody diluted 1:10 000 (v/v).

Supplementary data are available at The EMBO Journal Online (http://www.embojournal.org).