Retinoblastoma and Its Binding Partner MSI1 Control Imprinting in Arabidopsis

Parental genomic imprinting causes preferential expression of one of the two parental alleles. In mammals, differential sex-dependent deposition of silencing DNA methylation marks during gametogenesis initiates a new cycle of imprinting. Parental genomic imprinting has been detected in plants and relies on DNA methylation by the methyltransferase MET1. However, in contrast to mammals, plant imprints are created by differential removal of silencing marks during gametogenesis. In Arabidopsis, DNA demethylation is mediated by the DNA glycosylase DEMETER (DME) causing activation of imprinted genes at the end of female gametogenesis. On the basis of genetic interactions, we show that in addition to DME, the plant homologs of the human Retinoblastoma (Rb) and its binding partner RbAp48 are required for the activation of the imprinted genes FIS2 and FWA. This Rb-dependent activation is mediated by direct transcriptional repression of MET1 during female gametogenesis. We have thus identified a new mechanism required for imprinting establishment, outlining a new role for the Retinoblastoma pathway, which may be conserved in mammals.


Introduction
Genomic imprinting causes parental allele-specific expression during early development in mammals and plants [1]. In mammals, silencing marks are acquired by de novo DNA methylation during gametogenesis. Each imprinted locus receives a mark depending on its parental origin [2]. In embryonic cells, epigenetic marks are maintained on the silenced allele by a semiconservative mechanism involving the DNA methyltransferase Dnmt1 [3]. Following DNA replication, Dnmt1 methylates preferentially hemimethylated DNA, resulting in the maintenance of silencing marks on one parental allele. In the embryo, imprinting marks are read, leading to monoallelic gene expression. Certain genes remain imprinted and expressed in the adult [4]. A new imprinting cycle is initiated in the embryonic primordial germ cells where the epigenetic marks are lost following a global demethylation of DNA. The mechanism by which such demethylation originates is still unknown [5].
Embryogenesis in flowering plants and mammals depends on the function of embryo-nurturing annexes, the endosperm in plants and the placenta in mammals. In mammals, the placenta derives from the trophoblastic lineage separated from the embryonic lineage at the blastocyst stage. In flowering plants, the endosperm lineage separates from the embryonic lineage during female gametogenesis before fertilization [6]. The meiotic spore produces the haploid gametophyte in which two specialized female gametes differentiate: the egg cell and the central cell [7]. The fertilized egg cell produces the embryo, and the fertilized central cell generates the endosperm [8]. All imprinted genes known to date in plants are solely expressed in the endosperm, and some of these are essential for endosperm development [9].
Only a few imprinted genes have been identified in Arabidopsis [1]. MEDEA (MEA) [10], FWA [11], and FERTILIZA-TION INDEPENDENT SEED 2 (FIS2) [12] are only expressed from the maternal allele, whereas PHERES1 (PHE1) is preferentially expressed from the paternal allele [13,14]. PHE1 and FWA encode transcription factors of unknown function in the endosperm. MEA and FIS2 encode subunits of a Polycomb Group (Pc-G) complex [15]. Both alleles of MEA, FIS2, and FWA are silenced by distinct epigenetic mechanisms throughout the plant lifecycle until gametogenesis takes place. Silencing of MEA results from histone 3 lysine 27 (H3K27) trimethylation by Pc-G complexes [16,17], whereas silencing of FWA and FIS2 is mediated by the DNA METHYLTRANSFERASE 1 (MET1), which maintains DNA methylation of CpG sites [11,12]. Silencing marks are maintained in the sperm cells during male gametogenesis. During endosperm development, the inherited paternal copy remains silenced by MET1 or Pc-G activities, whereas the maternal copy is inherited as transcriptionally active, which results in monoparental expression [1].
During female gametogenesis, the expression of MEA, FIS2, and FWA is activated specifically in the central cell [18,19]. The activation of the above genes relies on DEMETER (DME), a DNA glycosylase protein [11,12,19]. DME removes methylated cytosines [16,20], causing the loss of the silencing marks from FIS2 and FWA promoters and allowing transcription in the central cell. The maternal allele inherited from the central cell remains active after fertilization in the endosperm while the paternal allele remains silenced. Hence the gene is imprinted in the endosperm while silenced in all other tissues. Loss of DNA methylation from the promoter of imprinted genes is conserved in the maize central cell [21,22]. Imprinting in plants thus results from the loss of silencing epigenetic marks during gametogenesis, which leads to gene expression. However, persisting expression of FIS2 in the dme mutant [12] suggested that mechanisms parallel to DME lead to removal of DNA methylation marks from FIS2 during female gametogenesis. In mammalian cell cultures, the expression of the DNA methyltransferase Dnmt1 is controlled by the Retinoblastoma pathway [23][24][25]. The Retinoblastoma protein (pRb) is known to repress the expression of S-phase genes during the G1 phase of the cell cycle through inhibition of E2F transcription factors. The pRb binding protein RbAp48 is critical for this function [26]. The Arabidopsis pRb homolog RETINOBLAS-TOMA RELATED 1 (RBR1) is highly expressed in the mature central cell, preventing uncontrolled syncytial proliferation [27]. MULTICOPYSUPPRESSOR OF IRA1 (MSI1) protein, a RbAp48 homolog, prevents production of endosperm from unfertilized central cells [28,29]. However, the interaction between RBR1 and MSI1 was not shown in Arabidopsis. We hypothesized that in Arabidopsis, the Retinoblastoma pathway may control the expression of MET1. In this study, we demonstrate that Arabidopsis MSI1 and RBR1 interact in vivo and down-regulate MET1 directly during female gametogenesis. Reduction of MET1 activity by the Retinoblastoma pathway in the central cell is essential for transcriptional activation of FIS2 and FWA. We thus provide evidence for a new mechanism essential for the activation of MET1-dependent imprinted genes in plants.

RBR1 Interacts with MSI1
Interaction between the tomato homolog of MSI1 and human or maize homologs of RBR1 has been shown in vitro [30]. Employing an in vitro glutathione S-transferase (GST) pull-down assay, we have detected direct interaction between Arabidopsis MSI1 and RBR1 proteins ( Figure 1A). In order to test whether the Arabidopsis MSI1 and RBR1 proteins interact in living cells, we have used the bimolecular fluorescence complementation (BiFC) assay [31]. MSI1 and RBR1 coding sequences were fused to the sequences encoding the Nterminal (YN) or the C-terminal (YC) part of the yellow fluorescent protein (YFP). We observed YFP fluorescence from nuclei in which YN-MSI1 and YC-RBR1 were transiently cotransformed into leaf epidermis and reconstituted a functional YFP protein ( Figure 1B). No fluorescence was observed using YN alone with YC-RBR1 or using YN-MSI1 with YC alone ( Figure 1F and 1G). We thus concluded that MSI1 and RBR1 interact in vitro and in vivo as shown previously between homologs of RBR1 and MSI1 in human, Drosophila, and maize [32].
In order to identify which part of the RBR1 protein is required for the interaction with MSI1, we performed a deletion analysis. The Arabidopsis RBR1 protein contains the two conserved domains RB-A and RB-B ( Figure 1H). Human Retinoblastoma RB-A and RB-B domains form a tridimensional structure called the A/B pocket. Human pRb interacts through the A/B pocket with the LxCxE domain of the histone deacetylase 1 (HDAC1) [33], which in turns binds RbAp48 [26,34]. Using truncations of the RBR1 protein fused to YC, we demonstrated that the interaction between MSI1 and RBR1 occurs through the RB-A domain ( Figure 1B-1E). We concluded that the interaction between MSI1 and RBR1 takes place in absence of the A/B pocket, suggesting that an HDAC1 is not involved in this interaction. In addition, the maize HDAC ZmRpd3I [35] and all Arabidopsis HDAC homologs do not contain the LxCxE domain, which is required for the interaction between Rb, HDAC1, and RbAp48 in mammals.
In conclusion, we suggest that in contrast to mammals, the plant homologs of Rb and RbAp48 likely interact directly and may not require an HDAC1 homolog.

MSI1 Represses the Expression of the DNA Methyltransferase MET1
To investigate the effect of MSI1 on MET1 expression, we used transgenic plants with a reduced level of MSI1 protein (MSI1cs) [36] since plants homozygote for the null msi1 alleles cannot be obtained due to embryo lethality [28]. Quantitative PCR (Q-PCR) analyses showed a 5-fold increase of MET1 expression in MSI1cs leaves in comparison to wild-type leaves ( Figure 2A). We concluded that MSI1 represses MET1 expression. We performed bisulfite sequencing of the known methylated regions of the FIS2 and FWA promoters to determine the level of CpG methylation in plants with reduced level of MSI1. Although most CpG sites in the two regions investigated are already methylated in the wild type, we could detect a modest increase of DNA methylation in leaves of MSI1cs plants ( Figure S1).
In addition to RBR1, MSI1 participates in the Pc-G complex comprising MEA, FIS2, and FIE [28,29,37]. MSI1 is also an essential component of the Arabidopsis chromatin assembly factor-1 (CAF-1) complex with the two other proteins FASCIATA1 (FAS1) and FASCIATA2 (FAS2) [38]. The CAF1 complex is conserved in yeast, Drosophila, and mammals and is essential for the deposition of the heterodimer H3-H4 at the replication fork [39].

Author Summary
Imprinting in plants and mammals involves a process whereby one of the two inherited gene variants (alleles) is inactivated. During imprinting, the transcriptional silencing of one allele is mediated by histone modifications or DNA methylation. The expressed parental allele is activated during gametogenesis by poorly understood mechanisms that remove silencing marks. In Arabidopsis, we studied genes expressed only from the maternal allele because the paternal allele is silenced by DNA methylation. We report that the expression of the maternal allele requires the repression of transcription of the major DNA methyltransferase by the sustained activity of the Arabidopsis homologs of the Retinoblastoma pathway. Repression is confined to the female gamete and is essential for the expression of imprinted genes in plants. The conserved transcriptional repression of DNA methyltransferases by the Retinoblastoma pathway suggests that this new regulation of imprinting might be also active in mammals.
In order to investigate which MSI1-dependent pathway is responsible for the transcriptional control of MET1, we tested whether MET1 expression was affected by loss of function of essential members of three distinct MSI1dependent pathways involving RBR1, FIE, or FAS1 [32]. We failed to obtain plantlets with a significant reduction of RBR1 expression from RBR1 RNA interference (RNAi) lines and thus have been unable to investigate the function of the Retinoblastoma pathway (unpublished data). Reduction of FAS1 activity in plants homozygote for the null fas1-1 allele [38] did not modify the level of MET1 expression ( Figure 2B). Similarly, the level of MET1 expression was not affected by reduction of FIE activity in FIE cosuppressed plants (FIEcs) [40] ( Figure 2B). We thus concluded that MSI1-mediated transcriptional repression of MET1 is independent of the Pc-G and CAF-1 pathways. We thus hypothesized that MET1 Figure 1. RBR1 Interacts with MSI1 (A) Pull-down assay testing for interaction between Arabidopsis MSI1 and RBR1. The full-length RBR1 protein, labeled with [ 35 S] methionine, was incubated with GST-MSI1 protein bound to agarose beads (GST MSI). As a negative control, the labeled protein was incubated with GST only, bound to agarose beads (GST). Input indicates the labeled protein used in the binding assays. Immunodetection of GST and GST-MSI1 were performed using anti-GST antibody.

MSI1 and RBR1 Bind to the MET1 Promoter
To test whether MSI1 and RBR1 control directly MET1 transcription by binding to the MET1 promoter, we performed chromatin immunoprecipitation (ChIP) using antibodies against MSI1 ( Figure S2) and RBR1 [41]. We focused our analysis on a 992-bp region of the MET1 promoter spanning À869 bp to þ123 bp relative to the predicted translation start site ( Figure 3A). Chromatin from wild-type Columbia buds was immunoprecipitated with antibodies against MSI1 or RBR1. A significant enrichment in MSI1 and RBR1 binding was observed for the DNA fragment 2 spanning À608 bp to À303 bp relative to the start codon ( Figure 3B and 3C). We conclude that MSI1 and RBR1 bind to the MET1 genomic locus on a domain containing a putative E2F binding site ( Figure 3A). These results strongly support the concerted action of MSI1 and RBR1 in regulating MET1 expression.

MSI1 and RBR1 Repress MET1 Expression during Female Gametogenesis
In plants with the genomic reporter construct pMSI1:MSI1-mRFP, we observed that MSI1-RFP was expressed throughout female gametogenesis ( Figure 4) similar to RBR1 expression [27]. Coexpression of MSI1 and RBR1 during female gametogenesis suggested that MET1 expression might be downregulated in a specific manner in the female gametes. We obtained transgenic plants expressing the HISTONE2B fused to red fluorescent protein (RFP) under the control of the MET1 promoter (pMET1-H2B-RFP). We determined the pattern of MET1 expression during female gametogenesis by confocal microscopic observations of developing ovules from four independent transgenic lines. The H2B-RFP signal was observed in nuclei of the ovule integuments and in nuclei of the syncytial embryo sac from the megaspore till the fournuclei stage ( Figure 5A-5C). At the eight-nuclei stage, the signal can no longer be detected in the embryo sac apart from the three antipodal nuclei ( Figure 5D). After cellularization of the eight nuclei, we could not observe any signal in the central cell, egg cell, and synergids, although some weak signal persisted in the three antipodals ( Figure 5E). In the mature gametophyte, we could no longer detect any fluorescence from H2B-RFP in the central cell marked by pFWA-GFP expression ( Figure 5F). As the result of our observations, we concluded that MET1 is transcriptionally repressed in the central cell. Taking into account the expected H2B-RFP retention in nuclei after nuclear division, we estimated that pMET1 activity is down-regulated at least from the fournuclei stage of the female gametophyte.
In order to test the action of MSI1 on MET1 during female gametogenesis, we introduced the MET1 transcriptional reporter in the msi1 mutant background. In msi1-1/þ plants hemizygote for pMET1-H2B-RFP, the RFP signal was detected in about 25% of female gametophytes, the expected proportion of ovules inheriting both msi1-1 and pMET1-MET1-RFP ( Figure 6C and 6D). This result suggested the down-regulation of MET1 expression by MSI1 in female gametophytes. We also observed an increased level of MET1 transcripts in ovules from msi1/þ plants ( Figure 2C), confirming that MSI1 inhibits MET1 expression in the female gametophyte. We did not detect any change of pMET1 activity in the female gametophyte of the fie/þ mutant ( Figure  6D), showing that MET1 down-regulation in the female gametophytes is independent of FIE. However, we did observed an increased level of MET1 transcripts in ovules from fie/þ plants ( Figure 2C). Increased cell division activity in fie ovule integuments [27] could be responsible for the global increased MET1 expression linked to its cell cycle dependence. To test whether RBR1 was also required for the downregulation of MET1 in the female gametophyte, we studied pMET1-H2B-RFP expression in rbr1 mutant. We observed ectopic pMET1-H2B-RFP expression in ovules from rbr1-1/þ plants ( Figure 6B). RFP signal was observed in about 25% of female gametophytes from rbr1/þ plants hemizygous for the pMET1-H2B-RFP reporter ( Figure 6D). We also directly observed increased MET1 expression in rbr1/þ ovules (Figure 2C) compared to wild type, which cannot result from integument cell proliferation because integument cells surrounding rbr1 embryo sacs do not overproliferate in absence of fertilization [27]. We concluded that the Retinoblastoma pathway involving RBR1 and MSI1 down-regulates MET1 expression during female gametogenesis.

MSI1 and RBR1 Are Required for the Expression of FIS2
In order to investigate the role of MSI1 and RBR1 on the transcription of imprinted genes silenced by MET1, we monitored the expression of the transcriptional reporter FIS2-GUS in a msi1-2/þ mutant background. Plants heterozygous for the null allele msi1-2 produce 50% of msi1-2 female gametes, which are deficient for MSI1 function [28,29]. We observed that only half of the ovules produced by msi1-2/þ; FIS2-GUS/FIS2-GUS plants expressed the FIS2-GUS marker as compared to FIS2-GUS expression in the wild-type background ( Figure 7A-7C), which correlated with the inheritance of msi1-2 in 50% of the central cell (v 2 ¼ 1.2, p . 0.27). The female gametogenesis is not perturbed by msi1-2 [28,29], and the defects observed in FIS2 expression are likely directly caused by the loss of MSI1 function.
FIS2-GUS was also expressed in half of the developing seeds after fertilization ( Figure 7A), Accordingly, FIS2 transcripts were not detected by reverse-transcription PCR (RT-PCR) in isolated msi1-2 mutant seeds selected on the basis of overexpression of the marker KS117 [28] ( Figure 7D). MSI1 is expressed from both parental alleles in the endosperm [28]. If the wild-type MSI1 allele paternally provided could rescue the inactive maternal allele of FIS2 delivered by msi1 ovules, we would expect 25% of ovules with FIS2-GUS expression. However, the proportion of seed without FIS2 expression was 50%. This result indicates that the inactivated status of FIS2 is inherited from the msi1 ovule in the endosperm, supporting further a role of MSI1 in the imprinted status of FIS2. Similar results were obtained using the msi1-1 allele ( Figure S5). We concluded that MSI1 is necessary for FIS2 expression in the central cell and that the loss of MSI1 in the central cell prevents the expression of FIS2 in the endosperm.
To test whether the two MSI1-containing complexes, Pc-G and CAF1, are responsible for the repression of FIS2 expression in absence of MSI1, we studied the effect of mutants in the genes MEA, FIE, and FAS1. We did not observe any loss of FIS2-GUS expression in ovules nor in endosperm of mea-6/þ, fie-11/þ, and fas1-1/þ mutants ( Figure S3). Thus, activation of FIS2 expression by MSI1 is independent from the Pc-G and CAF-1 activities.
Altogether, the in vivo interaction between RBR1 and MSI1, their coexpression in the central cell, and their concerted regulation of MET1 transcription, suggested a common requirement of MSI1 and RBR1 in the regulation of FIS2 expression. Plants heterozygous for rbr1-1/þ loss-offunction allele showed a 50% reduction of ovules expressing the FIS2-GUS reporter in comparison to the wild-type background ( Figure 7E and 7F). This reduction was correlated with the inheritance of rbr1 in half of the ovules (v 2 ¼ 0.27, p . 0.6). A dramatic reduction of FIS2 expression was also detected by RT-PCR in rbr1-1 mutant ovules ( Figure 7G). We thus concluded that RBR1 is necessary for FIS2 expression in the central cell.
Since FIS2 expression also depends on DME [12], we tested whether the loss of FIS2 expression could be attributed to the loss of DME expression in response to the MSI1/RBR1 pathway. DME and MSI1 were still expressed in rbr1 ovules, and conversely, the expression of MSI1 and RBR1 was not reduced in dme mutant buds ( Figure S4). Our results suggest that the transcriptional controls of DME and MSI1/RBR1 are mutually independent from each other.

RBR1 and MSI1 Are Required for FWA Expression
The expression pattern of FIS2 is similar to the expression patterns of the other MET1-dependent, maternally expressed imprinted gene, FWA [11,12]. We tested the effect of rbr1 and msi1 mutations on the expression of FWA using the tran- Similar results were obtained using the msi1-1 allele ( Figure  S5). Hence, both MSI1 and RBR1 are required for FWA and FIS2 expression.
Activation of FIS2 and FWA by MSI1 Is Mediated by the DNA Methyltransferase MET1 In order to investigate the genetic relationship between the activities of MET1 and MSI1, we constructed a double-mutant plant of msi1/þ; met1/þ carrying either FIS2-GUS or FWA-GFP reporters. In this experiment, we have used different msi1 null  Considering the recombination frequency between the two genes, we expected that msi1/þ; met1/þ plants would produce 41% of ovules carrying both the msi1 mutation and the met1 null mutation. If the effect of msi1 and rbr1 mutations on FIS2 and FWA expression was mediated by a repression of MET1, we would have expected to observe 45.5% of ovules expressing FIS2 or FWA in msi1/þ; met1/þ plants (Table S1).

RBR1 and MSI1 Are Not Required for MEA Expression
Silencing of MEA depends on H3K27 tri-methylation by Pc-G complexes [16,17], whereas silencing of FIS2 and FWA is caused by DNA methylation by the maintenance methyltransferase MET1 [12]. However, MEA activation in the central cell depends on DME action antagonized by MET1 [42]. Thus it remained to be tested whether MET1 repression by RBR1 and MSI1 could have a direct impact on MEA expression in the central cell. In contrast to the transcriptional repression of FIS2 and FWA mediated by msi1 and rbr1 mutants, we could not detect any impact of msi1-2 or rbr1-1 on the expression of the MEA-GUS reporter ( Figures 8A, 8E-8G, and S6). Similar results were obtained using the msi1-1 allele and the reporter MEA-YFP that encodes a fusion protein that complements mea ( Figure S5). We thus concluded that MEA expression in the central cell is not regulated by MSI1 and RBR1.

Conservation of the Transcriptional Repression of MET1 by the Retinoblastoma Pathway
We identify a site in the putative promoter of MET1 where MSI1 and RBR1 associate. In addition, we show that MET1 expression is repressed by MSI1. This repression does not depend on the MSI1 association to the complexes Pc-G and CAF1, but depends on RBR1. Our results thus support a repression of MET1 transcription by the MSI1/RBR1 complex. We have found one putative E2F binding site (ATTGCCGC) situated À387 bp from the predicted translation start site of the MET1 promoter. Accordingly, MET1 expression is strongly increased in plants overexpressing the Arabidopsis E2Fa and DPa proteins, suggesting that MET1 is an E2F target gene [43,44] and MET1 is expressed during the S phase of the cell cycle [45,46]. We thus propose that the activation of MET1 transcription during the S phase requires the release of the sequestration of E2F by RBR1 at the G1/S cell cycle checkpoint. The RBR1/MSI1 complex would prevent E2F from activating the transcription of MET1. In human cells, the transcription of the human homolog of MET1, Dnmt1 is also repressed by the Rb/E2F pathway [24,25]. Hence maintenance of DNA methylation would be coordinated with the S phase by an E2F control, and this mechanism is likely conserved in Eukaryotes.

Repression of MET1 Is Essential for Activation of Imprinted Genes during Female Gametogenesis
MSI1 and RBR1 are expressed throughout female gametogenesis, and close inspection of MET1 expression during female gametogenesis showed that MET1 is specifically repressed during female gametogenesis by the Retinoblastoma pathway. The reduced expression of MET1 in female gametes could explain why the inheritance of met1 by the female gamete has no effect on endosperm and seed development [47]. In contrast, MET1 is expressed in male gametes ( Figure S7), and the inheritance of met1 by the male gametes causes a strong reduction of endosperm and seed size [47,48] presumably as a result of the ectopic activation of the paternal allele of imprinted genes regulated like FWA and FIS2.
Our results indicate that the RBR1/MSI1 pathway activates the maternal expression of FIS2 and FWA via the transcriptional repression of MET1 in the central cell. The imprinted genes FIS2 and FWA are repressed throughout the vegetative stage by MET1-dependent DNA methylation on their promoters [11,12]. At the end of female gametogenesis, DNA methylation is removed in the central cell by DME [11,16,19] leading to transcriptional activation. The demethylated maternal allele remains active in the endosperm. DME has a high affinity for hemimethylated DNA and causes excision of methylated cytosine residues followed by repair of single-strand break and incorporation of a nonmethylated cytosine residue [16,20]. However, we showed that DME activity is not sufficient to account for the activation of FIS2 [12]. We propose that in addition to DME, the low level of MET1 activity during female gametogenesis causes a gradual dilution of the DNA methylation marks after each cycle of DNA replication (Figure 10). Following the third cycle of syncytial division, the central cell and the egg cell become isolated from each other. In the egg cell, the remaining marks are sufficient to maintain FIS2 and FWA silenced before fertilization and after fertilization in the embryo. In contrast, the DNA glycosylase DME is expressed in the central cell [19] and previous enrichment in hemimethylated DNA at the FIS2 and FWA loci would prime DME to remove the remaining marks leading to activation of FIS2 and FWA. If the Retinoblastoma pathway does not function during female gametogenesis, DNA methylation is retained causing the absence of activation of FIS2 and FWA, thus preventing their maternal expression in the endosperm.
Although FWA and FIS2 repression by MET1 appear similar, FWA expression is completely dependent on DME, whereas FIS2 expression is still expressed even in absence of DME activity. Similarly, loss of MET1 in vegetative tissues causes FWA ectopic expression, but FIS2 remains silenced. This suggests that at least another transcriptional activator control FIS2 expression in the endosperm lineage.

MEA Imprinting Is Not Primarily Controlled by MET1
Previous studies have shown that MEA expression is regulated by MET1 and DME [16,42]. However, it remained unclear whether MET1 and DME regulate MEA imprinting directly. Genome-wide array of DNA methylation and H3K27 trimethylation have shown that the MEA locus is covered with H3K27 trimethylation but devoid of DNA methylation. DNA methylation can be found in the MEA locus only at repeats at the 39 end of the gene, the MEA ISR domain [49][50][51]. However, a MEA reporter construct that does not contain the MEA ISR domain displays imprinted expression and complements the mea mutant phenotype [37] strongly suggesting that the MEA ISR is not involved in the regulation of MEA expression. Furthermore, the direct loss of DNA methylation in met1 pollen does not cause activation of the MEA paternal allele in endosperm [12], whereas MEA is activated directly by the loss of H3K27 trimethylation in mutants for Pc-G activity in both vegetative and reproductive tissues [16,17]. We do not observe any impact of the ectopic expression of MET1 by msi1 and rbr1 on MEA expression. Hence, we can conclude that MEA expression is not directly controlled by MET1. Rather, the direct silencing of MEA by H3K27 methylation implies that trimethylated H3K27 must be removed from the MEA locus during female gametogenesis to obtain transcriptional activation. It is thus possible that MET1 and DME are required directly or indirectly for the activation of a pathway that removes the H3K27 trimethylation mark from MEA leading to its activation. Alternatively, maternal MEA expression may require a transcriptional activator that is itself directly controlled by DNA methylation and DME activity.

A Genome-Wide Demethylation during Female Gametogenesis?
It is not clear whether DNA demethylation during female gametogenesis affects only a discrete number of loci as a result of still-unknown targeting mechanisms. Alternatively, the DNA demethylation regulated by the Retinoblastoma pathway could affect the entire genome leading to a reduction of the constitutive heterochromatin fraction in the central cell, which could be inherited in the endosperm. A global decrease of maternal DNA methylation in the endosperm was suggested to occur in maize based on the analysis of methylation-sensitive amplified polymorphism [52] and in Arabidopsis based on the reduction of the heterochromatin fraction [53]. A global demethylation is not expected to affect significantly DNA of the egg cell since it has been clearly established that transgenes silenced by DNA methylation remain silenced from one generation to the next [54].

Potential Conservation of Regulation of Imprinting by the Retinoblastoma Pathway
Our results provide evidence for a direct link between the Retinoblastoma pathway and genomic imprinting. The regulation of the CpG maintenance methyltransferase by the Retinoblastoma pathway is conserved between plants and mammals. Although there has been no direct evidence in mammals for a role of the Retinoblastoma protein itself in genomic imprinting, in a human cell line, the up-regulation of the expression of the DNA methyltransferase Dnmt1 caused by misregulation of pRb was correlated with an increase of CpG methylation of the paternally expressed imprinted gene Peg3 [25]. The loss of function of pRb during mouse embryonic development results in embryonic lethality. However, this lethality can be partially rescued by providing mutant embryo with a wild-type placenta, demonstrating a crucial role of pRb during placenta development. pRb deficiency in the placenta results in a decrease of nutrient transport from the mother to the embryo [55,56]. The identification of two Retinoblastoma binding proteins (Rbbp1/Arid4a and Rbbp1l1/Arid4b) that are required for the correct epigenetic marks and imprinting status at the PWS/AS domain provides further evidence for a potential role of pRb on imprinting during placenta development [57]. The data above indicate that the impact of the Retinoblastoma pathway on parental genomic imprinting in plants, that we have demonstrated may also exist in mammals and would thus provide further evidence for a convergence of imprinting between flowering plants and mammals.
RNA extraction, cDNA preparation, and RT-PCR. Sample tissues were collected from Arabidopsis plants and frozen in liquid nitrogen. As only msi1-2/þ plants survive, we selected msi1-2 mutant seeds characterized by the overexpression of the KS117 marker at 5 d after pollination [28]. As only rbr1-1/þ plants survive, we selected rbr1-1 mutant ovules using the size difference between undeveloped rbr1-1 ovules and wild-type seeds 1 DAP before rbr1 ovules degenerate (4 DAP) [27]. RNA extraction, DNase treatment, and reverse transcription were performed as described previously [12]. Primer sequences are listed in Table S2.
Quantitative real-time RT-PCR. Real-time PCR assays were performed using Power SYBR Green PCR Master Mix (Applied Biosystems). One microliter of RT product was used to perform each PCR reaction. Amplification reaction was carried out using specific primers at a concentration of 0.5 mM in a 10-ll reaction. The MET1specific primer pair was qPCR-MET1_Rev2 and qPCR-MET1_Fwd2 (Table S2).
The specificity of the amplification was determined by performing a dissociation curve analysis. The PCR reaction and quantitative measurements were achieved with 7900HT Fast Real-Time PCR System (Applied Biosystems). Thermal cycling parameters were 2 min at 50 8C, 10 min at 95 8C, and 50 cycles of 15 sec at 95 8C, 60 sec at 60 8C. Three technical replicates were done for each sample. The DCt was calculated using ACT11 gene as endogenous control (Table S2). Relative quantitation (RQ) values were calculated using the 2 ÀDCt method (RQ ¼ 2 ÀDCt ) [62]. Values given in Figure 2A represent the RQ average of five biological replicates for each point and three biological replicates ( Figure 2B and 2C).
Chromatin immunoprecipitation. The ChIP experiment was performed as described previously [63] with minor modifications. Buds were ground with a pestle in liquid nitrogen and fixed with 1% formaldehyde for 10 min. The chromatin was precleared by incubating with protein A beads (Upstate) and IgY-beads (Aves Labs). The protein-DNA complexes were incubated overnight with 40 ll of MSI1 antiserum ( Figure S2), 5 ll of purified RBR1 antibody [41], 10 ll of Rabbit Control IgG (Abcam, ab46540) or 20 ll of Histone H3 antibodies (Abcam, ab1791) and immunoprecipitated with protein A beads (Upstate) and IgY-beads (Aves Labs) respectively. After reverse cross-linking, proteinase K, and RNase treatment, the immunoprecipitated DNA was purified using a silica-gel membrane (Qiagen) and analyzed by PCR. PCR reactions were performed in 20 ll using HotStarTaq DNA Polymerase (Qiagen). Quantitative measurements of enrichment from the fragment 2 of MET1 were performed using the absolute quantitation method achieved with 7900HT Fast Real-Time PCR System (Applied Biosystems). Relative enrichments were calculated as the percentage of the obtained values in immunoprecipitated and input fractions. Primer sequences are listed in Table S1.
GST pull-down assay. A full-length MSI1 cDNA was cloned into the BamHI site of pGEX 4T-1 vector (Amersham Pharmacia) in frame to GST. RBR1 was cloned in to the pCITE 3a (Novagen) as described [64]. Pull-down assays were performed as described previously [65]. A GST-MSI1 fusion protein was expressed in Escherichia coli BL21 cells and immobilized on glutathione agarose beads. Beads were incubated with radioactively labeled RBR1 protein followed by six consecutive washes with NETN buffer (100 mM NaCl, 1 mM EDTA, 20 mM Tris [pH 8.0], and 0.5% NP-40). Protein labeling was performed using a coupled transcription-translation system in a reticulocyte lysate system (Promega) according to the manufacturers instructions using radiolabeled [ 35 S] methionine (Amersham-Pharmacia). Following SDS/PAGE, the translated RBR1 products appeared as a major band of 120 kDa, as expected. Beads were washed and resuspended in SDS-polyacrylamide gel sample buffer (60 mM Tris [pH 6.8], 2% SDS, 10% glycerol, 0.002% bromphenol blue, and 100 mM DTE). Sample were separated by electrophoresis on a 10% SDS-polyacrylamide gel, and transferred to a polyvinylidene difluoride (PVDF) Immobilion-P membrane (Millipore). Labeled RBR1 protein was detected by autoradiography. The GST protein was detected using anti-GST antibodies (Amersham Pharmacia Cat#27457701) diluted 1:1,000 (v:v).
Analysis of protein-protein interaction in plants by bimolecular fluorescence complementation assay. Protein-protein interactions in plants were examined by BiFC assay [31]. Equal concentrations of Agrobacterium tumefaciens strain GV3101/pMp90 containing plasmids of interest (Table S3) were transiently coexpressed in Nicotiana benthamiana leaves via leaf injection procedure. Following incubation at 25 8C for 24 h, samples were examined with Leica TCS-SL confocal laser scanning microscope with 203 and 633 water immersion objectives. Image analysis was performed with Zeiss AxioVision, Zeiss CLSM-5, and Adobe Photoshop 7.0 YFP was visualized by excitation with an argon laser at 514 nm. Emission was detected with spectral detector set between 525 nm and 570 nm. RBR1 and MSI1 full-length cDNA and three deletion constructs of RBR1 were cloned into the SpeI site of pSY 735 and pSY 736 containing the N-terminal (YN), or C-terminal (YC) fragments of the YFP protein, respectively [31]. Negative controls with vectors bearing only YN or YC alone were carried out in every experiment to verify the specificity of the interactions. Each confocal section detects only a fraction of the nuclei in a given field of view, and not every cell expresses both constructs. This explains why only a fraction of cells show fluorescent nuclei in a given field of view when interaction takes place.
Microscopy and statistics. FIS2-GUS, FWA-GFP, MEA-GUS, and MEA:YFP expression were analyzed as previously described [12,17]. Developing seeds or pistils cleared with derivative of Hoyer's medium were observed with differential interference contrast (DIC) optics and with 203 PlanApo objective (DM6000 B, Leica). Images were acquired with a DXM1200F digital camera (Nikon) and processed using Metamorph (version 6.2, Universal Imaging Corp). In Figures 7,8,9,S3, S5, and S6, the error bars represent the standard deviation calculated from the mean measured per silique containing, on average, 40 ovules. The n number is represented on the top of each bar and corresponds to the total number of ovules observed. pMSI1-MSI1-RFP and pMET1-H2B-RFP fluorescence was imaged using laser scanning confocal microscopy (Zeiss Exciter) for mRPF1 with selective settings for RFP detection (excitation, 543 nm; emission, band-pass 560 to 615 nm).
pMET1:H2B-mRFP1 plasmid construction and transformation. The MET1 promoter was amplified by PCR using the KOD-plus-PCR kit (TOYOBO) and primer pair: R2met1-F1td and MET1-int66R (Table S2). The PCR fragment was then cloned into pENTR-D-TOPO (Invitrogen). The promoter was then fused to H2B-RFP in frame by recombination using Gateway technology (Invitrogen) into the pGreenII-GW-H2B-RFP destination vector [66]. The final vector (pGreenII-pMET1-H2B-RFP) contains 947 bp upstream of the translation start site until the first 66 amino acids of MET1 in frame with H2B-RFP. Wild-type Columbia were transformed using the Agrobacterium-mediated floral dip method [67], and transgenic lines were selected on kanamycin. The presence of the transgenes was confirmed using PCR. Thirteen transgenic lines showed consistent patterns of transgene expression. Three transgenic lines were used for the detailed observation of the expression pattern using confocal microscopy. Figure S1. Comparison of CpG DNA Methylation between Wild-Type and MSI1cs Plants In order to know whether increased MET1 expression in MSI1cs leaves also leads to an increase in the amount of CpG methylation, we performed bisulfite sequencing of the methylated region of the FIS2 and FWA promoters. (A) Analysis of CpG methylation of FIS2 promoter region using bisulfite sequencing. Twelve clones were sequenced for wild-type Col, and 11 clones were sequenced for MSI1cs. (B) Analysis of CpG methylation of FWA promoter region using bisulfite sequencing. Eleven clones were sequenced for wild-type Col, and eight clones were sequenced for MSI1cs. In comparison to wild-type levels, we observed in MSI1cs plants a slight increase in the methylation of the FIS2 promoter, but did not see any significant modification of the methylation level in the FWA promoter. Several factors may explain this limited effect. Increased level of MET1 mRNA may not result in an increased MET1 activity or the increase of MET1 activity in vegetative tissues is not sufficient to be effective. Alternatively, the targets studies may not be responsive to increased MET1 activity in vegetative tissues. The most likely hypothesis is that FIS2 and FWA methylated regions are already highly methylated in wild-type leaves, which prevents the observation of an increase in MSI1cs. Methods: bisulfite-sequencing analysis was carried out with the EZ DNA Methylation-Gold Kit (Zymo Research). The treated DNA was cleaned up in accordance with the manufacturer's instructions and used for subsequent PCR. After PCR, the products were cloned in pGEM-Teasy (Promega), and individual clones were sequenced with SP6 and T7 primers. The sequences were then aligned with AlignX (Invitrogen), and the methylation level was analyzed using BiQ Analyzer (http://biq-analyzer.bioinf.mpi-sb.mpg.de). Primers used for bisulfite-sequencing analyses are listed in Table S1. The ASA1 gene was used as a positive control of bisulfite chemical reaction (Kinoshita et al. [11]). Found at doi:10.1371/journal.pbio.0060194.sg001 (928 KB EPS). Figure S2. Specificity of the Antiserum against MSI1 Antibodies against the C-terminal region of MSI1 were obtained from immunization of rabbits with the peptide MGKDEEEMRGEIEERLINE (Invitrogen). A western blot was performed using serum against protein extracts from 10-d-old seedlings from Arabidopsis, wild-type Columbia, and a transgenic Columbia line carrying the gene encoding the MSI1-RFP protein fusion under the control of the MSI1 promoter (Figure 4). A major band was detected by the antibody in the wild-type background around 55 kDa, i.e., slightly above the predicted molecular weight of MSI1 (48.2 kDa). A minor band was detected at 68 kDa. In the extract of the MSI1-RFP line, an additional band was detected at 80 kDa. The molecular weight of RFP is 27 kDa, and the predicted protein fusion weight of 82 kDa corresponds to the size observed, indicating that the major protein detected by the antibody is MSI1. The intensity of the major band in the wild type was reduced, indicated a potential cosuppression effect of the expression of an additional copy of MSI1.