Epigenetic regulation of gene programs by EMF1 and EMF2 in Arabidopsis.

The EMBRYONIC FLOWER (EMF) genes are required to maintain vegetative development in Arabidopsis (Arabidopsis thaliana). Loss-of-function emf mutants skip the vegetative phase, flower upon germination, and display pleiotropic phenotypes. EMF1 encodes a putative transcriptional regulator, while EMF2 encodes a Polycomb group (PcG) protein. PcG proteins form protein complexes that maintain gene silencing via histone modification. They are known to function as master regulators repressing multiple gene programs. Both EMF1 and EMF2 participate in PcG-mediated silencing of the flower homeotic genes AGAMOUS, PISTILLATA, and APETALA3. Full-genome expression pattern analysis of emf mutants showed that both EMF proteins regulate additional gene programs, including photosynthesis, seed development, hormone, stress, and cold signaling. Chromatin immunoprecipitation was carried out to investigate whether EMF regulates these genes directly. It was determined that EMF1 and EMF2 interact with genes encoding the transcription factors ABSCISIC ACID INSENSITIVE3, LONG VEGETATIVE PHASE1, and FLOWERING LOCUS C, which control seed development, stress and cold signaling, and flowering, respectively. Our results suggest that the two EMFs repress the regulatory genes of individual gene programs to effectively silence the genetic pathways necessary for vegetative development and stress response. A model of the regulatory network mediated by EMF is proposed.

Molecular genetic analysis found that the three Arabidopsis PRC2s regulate diverse functions Chanvivattana et al., 2004). The PRC2 composed of MEA, FIE, FIS2, and MSI1 represses PHERES1 (PHE1) gene expression during seed development (Kohler et al., 2003b;Makarevich et al., 2006). The complex composed of CLF/SWN, FIE, VRN2, and MSI1 represses FLOWERING LOCUS C (FLC), mediating the vernalization response (Chanvivattana et al., 2004;Wood et al., 2006;De Lucia et al., 2008). FLC is a floral repressor that suppresses the floral activators FLOWERING LOCUS T (FT) and SUPPRESSOR OF OVEREXPRESSION OF CO1 (SOC1; Michaels et al., 2005). FLC repression would activate FT and SOC1, leading to flowering. The third complex, composed of CLF/SWN, FIE, EMF2, and MSI1, represses the flower MADS box genes [e.g. AGAMOUS (AG)], enabling vegetative growth (Yang et al., 1995;Chen et al., 1997;Goodrich et al., 1997;Calonje and Sung, 2006;Schubert et al., 2006). Previous gene expression analysis using an 8,000 Arabidopsis gene GeneChip array (8 K GeneChip) identified eight categories of genetic pathways modulated by EMF activities . Here, we report a genome-wide analysis of gene expression in emf mutants that identified additional pathways regulated by the two EMF genes, the abscisic acid (ABA), stress, cold, and heat signaling pathways. We employed chromatin immunoprecipitation (ChIP) to determine whether EMF regulates these pathways directly. Analysis showed that EMF1 and EMF2 are recruited to the transcription factor genes of several genetic pathways. A model of EMF regulation of seed maturation, stress/cold signaling, and flowering is presented.

Genome-Wide Gene Expression Pattern in emf1 and emf2 Mutants
Previous 8 K GeneChip analyses showed that EMF1 and EMF2 negatively regulate the flower organ identity and seed maturation genes in Arabidopsis . To study genome-wide regulation of EMF, we performed a global gene expression analysis using a custom-designed Affymetrix GeneChip containing probe sets representing 25,996 unique Arabidopsis genes (26 K GeneChip; Zhu, 2003).
The expression level of each gene was determined by the difference in hybridization signal intensity measured by perfect match and mismatch probes (Zhu, 2003). Gene expression patterns of 7-and 15-dold emf mutants (emf1-1, emf1-2, and emf2-1) were compared with that of the wild type at the same age to identify genes showing expression change. Expression patterns of genes in wild-type flower buds and flowers were also examined to investigate the causal relationship of the gene expression change to the emf mutation or its reproductive state.
The high-density Arabidopsis oligonucleotide arrays were designed based on the unigene set assembled at the time. The expression level of each gene was determined by hybridization signal intensity measured by perfect match probes using the MOID algorithm (Zhou and Abagyan, 2002). Despite some differences in probe sets, a high correlation coefficient (0.85) between the two arrays among 7,582 commonly presented genes was reported (Zhu, 2003). This supports the strategy of cross-comparison for commonly presented genes between the previous analysis  and this study. Our microarray study is a quick survey of the transcriptome to identify additional putative components of EMF-regulated pathways. The previous study  showed that a simple, stringent cutoff can effectively produce a global reference profile of the EMF-regulated transcriptome. Transcript changes of key genes in these pathways were confirmed by reverse transcription (RT)-PCR (see below).
Genes with hybridization signals greater than 50 in at least one of the four lines (the wild type and the three emf alleles) were selected for further analysis. Of the 25,996 genes analyzed, 11,447 from 7-d-old and 10,204 from 15-d-old samples satisfied this criterion (Table I). Fold change of gene expression was determined by the hybridization signal ratio between the wild type and the mutants. Genes showing more than a 2-fold change in the hybridization signal were designated up-regulated if those signals were higher in the mutants and down-regulated if the signals were lower in the mutants, as described in Table I. For example, in the 7-d-old emf1-1 sample, 3,336 (29.1%) genes were up-regulated and 1,033 (9%) were downregulated; in total, 38% of the genes changed expression. All three mutants showed extensive changes from the wild-type gene expression pattern. In general, more genes were up-regulated than downregulated in the mutants. We compared gene expression patterns among the three emf mutants to determine whether the same genes were similarly up-or downregulated. Among the 11,447 genes selected from the 7-d-old samples, 3,957 genes were up-regulated and 1,629 were down-regulated in at least one of the three emf alleles (Fig. 1, A and B). A total of 1,654 of the 3,957 genes (41.8%) were up-regulated and 303 of the 1,629 genes (18.6%) were down-regulated in all three mutants. The overlap in up-regulated genes between the two emf1 alleles, emf1-1 and emf1-2, was higher than between the two mutants of similar phenotypes, emf1-2 and emf2-1. These similar phenotype mutants are both weak alleles compared with emf1-2. The overlap between the two emf1 alleles for the 7-d-old samples was 69.9% (2,773 of 3,957) in up-regulated genes (Fig. 1A) and 48.6% (803 of 1,629) in down-regulated genes (Fig. 1B). The overlaps between emf1 and emf2 were fewer, 44.2% (emf1-2 versus emf2-1; Fig. 1A) and 47.3% (emf1-1 versus emf2-1) for up-regulated genes (Fig. 1A). Similar results were found for the down-regulated genes.
In 15-d-old samples, among 10,204 selected genes, 2,522 were up-regulated and 1,909 were downregulated in at least one of the three emf alleles (Fig. 1, C and D). A total of 556 of the 2,522 genes (22%) were up-regulated and 457 of the 1,909 genes (23.9%) were down-regulated in all three emf alleles. Again, there was a higher percentage of overlap between the two emf1 alleles than between emf2-1 and emf1-1.
In summary, our 8 K and 26 K GeneChip results were similar in that the emf mutants display major gene expression pattern changes from the wild type at the same number of days after germination (DAG). Our results also confirmed that, despite the fact that emf1-1 is morphologically more similar to emf2-1 than to emf1-2, its gene expression pattern is more similar to that of emf1-2. That the gene expression patterns correlate with alleles rather than phenotype suggests possible EMF1-and/or EMF2specific gene programs.

Functional Analysis of EMF-Regulated Genes
Genes showing expression changes in emf mutants were grouped into 15 functional categories, including the eight previously described (i.e. flowering, seed maturation, photoreceptors, photosynthesis, auxin, GA, ethylene, and expansin; Moon et al., 2003; Table  II). In Moon et al. (2003), the flowering category included both flowering time genes and flower organ identity genes, which are separated into two different categories for the 26 K GeneChip data analysis. As 26 K GeneChip data showed expression changes in many more genes, we created additional categories to cover these genes and their homologs: ABA synthesis and signaling, stress, cold, heat, and transcription factors. The histone category is also included because histone proteins are major components of the chromatin and their modification by PcG proteins affects gene activity. Genes with hybridization signals greater than 50 in at least one of the eight RNA samples were grouped into functional categories based on annotation of the probe set in the 26 K GeneChip and their relations to biological functions, processes, and cellular locations (Supplemental Tables S1 and S2).

Flower Organ and Flowering Time Genes
Consistent with the 8 K GeneChip data analysis, the 26 K GeneChip data showed increased expression of most flower organ genes that include the flower MADS box or organ identity genes, namely AG, PISTILLATA (PI), APETALA1 (AP1), AP3, SEPALLATA1 (SEP1), SEP2, and SEP3, in the emf mutants. Table II shows that all but one of 23 flower organ genes examined were up-regulated in the emf mutants (Table II; Supplemental Table S1A). Most flower MADS box genes are also up-regulated in wild-type flowers, consistent with the notion that emf mutants are in the reproductive state. However, no flower organ primordia have yet developed in the 7-d-old emf mutants, suggesting that up-regulation of the MADS box genes results from the loss of EMF activity rather than as a consequence of flower organ development.
In contrast to the flower organ genes, some genes in the flowering time category were up-regulated and others were down-regulated (Table II; Supplemental Table S1B). In addition, flowering time genes did not show corresponding expression changes from wild-type seedlings between wild-type flower and emf mutants. For example, GIGANTEA (GI) and FLC were up-regulated in emf mutants but not in wild-type flowers (Table III; Supplemental Table  S1B). Hence, their expression change in emf mutants is not related to the reproductive state of the emf mutants.
Differential gene expression of major flowering time genes was observed in the emf1 and emf2 mutants. For example, both 8 K and 26 K GeneChip data as well as RT-PCR analysis showed that FT is upregulated in emf2 mutants but not in emf1 mutants (Table III; Fig. 2; Yoshida et al., 2001;Moon et al., 2003). The 26 K GeneChip data showed up-regulation of GI in emf seedlings (Table III) that is confirmed by RT-PCR analysis in 7-d-old emf mutants (Fig. 2). As GI and FT are activators of CONSTANS (CO) and SOC1, respectively (Kim et al., 2008), we investigated CO and SOC1 RNA levels (Table III). Unexpectedly, CO and SOC1 were not up-regulated in emf seedlings. As transcript levels were low, we carried out RT-PCR analysis to confirm expression levels and found no significant differences between the wild type and emf mutants (Fig. 2). These results suggest additional factors controlling CO and SOC1 expression. Several negative regulators of CO and SOC1 were investi- gated. Interestingly, the transcript levels of LONG VEGETATIVE PHASE1 (LOV1), a known CO repressor (Yoo et al., 2007), and FLC, a known SOC1 repressor, were increased in all three emf mutants on the 26 K GeneChip array (Table III). RT-PCR results confirmed increased LOV1 and FLC expression in the emf mutants ( Fig. 2). Since CO activates FT, it is curious that the FT transcript is elevated in emf2-1 while the CO level remains constant. To understand this, we examined the expression patterns of FT repressors such as the TARGET OF EAT1 (TOE1) and TEMPRANILLO (TEM) genes (Jung et al., 2007;Castillejo and Pelaz, 2008). Table III shows that TOE1, TEM1, TEM2, and RELATED TO ABI3/VP1 (RAV1) transcript levels were all reduced in emf mutants. However, it is not clear if the down-regulation of these genes is related to the lack of EMF repressors or to the reproductive state, as these genes were also down-regulated in wild-type flowers.

Genes Involved in Seed Development
The 32 seed maturation genes investigated in the 8 K GeneChip analysis were all up-regulated in the two emf1 mutants but not in emf2-1 . We examined 120 genes involved in seed development in the 26 K GeneChip data set (Supplemental Table  S1C). Many were up-regulated in emf1 or emf2 mutants, but some did not change or showed reduced expression (Table II). Of the 120 seed category genes examined, 61 were up-regulated in at least one of the three emf mutants at 7 DAG and 56 were up-regulated at 15 DAG. Figure 3 shows that 26.3% were upregulated in all three emf mutants at 7 DAG and 26.7% at 15 DAG. The two emf1 allele samples have more genes with overlapping expression patterns than the emf1 and emf2 mutant samples. There is only 31.2% and 30.2% overlap between emf1-1 and emf2-1 but 37.7% and 55.2% overlap between the two emf1 mutants at 7 and 15 DAG, respectively (Fig. 3). For example, the AtEm1, ABSCISIC ACID INSENSITIVE3 (ABI3), and four Arabidopsis 2S Albumin3 (At2S3) genes were up-regulated in the two emf1 mutants but not in emf2-1 (Supplemental Table S1C).
ABI3, a master regulator of the seed maturation program that activates these seed storage protein genes (Krebbers et al., 1988), has low transcript level and thus may be precluded by GeneChip detection as a gene changing expression in the emf mutants, Thus, we performed RT-PCR analysis that showed that transcripts of ABI3, as well as At2S3 and AtEm1, are increased in emf1-1 and emf1-2 but not in emf2-1 (Fig. 4). The differential expression pattern of ABI3 in the mutants is likely to account for the expression pattern of the downstream seed storage protein genes.

Genes Involved in Hormone Synthesis and Signaling
In the auxin and GA categories, the 26 K GeneChip data showed a similar number of genes up-and downregulated in all three emf mutants (Table II; Supplemental Table S1, D and E).
In the ethylene category, more genes were upregulated than down-regulated in 7-d-old mutants, but more genes were down-regulated in 15-d-old emf mutants (Table II). This is in part a result of agedependent expression changes in the wild type, where transcript levels of many genes were increased from 7 to 15 d old. Many genes in the mutants also Figure 2. RT-PCR analysis of flowering time genes in emf mutants. RNA samples extracted from 7-and 14-d-old wild-type (WT) and emf mutant plants were subjected to RT-PCR analysis. ACTIN8 was used as a loading control. Numbers presented below the gels represent relative expression levels and were calculated using the ACTIN8 signal as a reference. showed increased expression from 7 to 15 d, but to a lesser extent than genes in the wild type (Supplemental Table S1F). Therefore, down-regulation in 15d-old mutants is not necessarily a direct result of the loss of EMF function; rather, it may result from developmental change in the wild type. However, the results from 7-d-old samples do show an upregulation of gene expression in emf mutants, indicating that EMF negatively regulates genes in the ethylene category.
In contrast, almost all of the ABA-related genes investigated were up-regulated in emf mutants (Table  II; Supplemental Table S1G). There is no obvious agedependent change in gene expression in wild-type samples. Thus, ABA signaling is clearly negatively impacted by EMF. EMF regulation is consistent with ABA's role in repressing seed maturation genes after germination.

Genes Involved in Cold, Stress, and Heat Signaling
As ABA also regulates stress gene expression (Zhang et al., 2008;Kempa et al., 2009), we examined the expression of stress-related genes (i.e. genes in the cold, stress, and heat categories). Many were upregulated in emf mutants (Table II; Supplemental Table  S1, H-J).
In the cold category, more genes were up-regulated than down-regulated in the three emf mutants compared with the wild type in both the 7-and 15-d-old samples (Table II). C-REPEAT-BINDING FACTOR1 (CBF1) encodes the primary cold response regulator that binds to cis-elements in the promoter of the COLD REGULATED (COR) genes to activate their expression (Chinnusamy et al., 2007). The 26 K GeneChip data showed up-regulation of COR15A and CBF1 in emf mutants (Table IV). RT-PCR analysis also showed a dramatic increase of transcript levels of CBF1, COR15A, and KIN1 in all three emf mutants (Fig. 4). Since LOV1, which activates KIN1 and COR15A (Yoo et al., 2007), was up-regulated in all three emf mutants at 7 DAG (Table III; Fig. 2), it is possible that increased LOV1 and CBF1 in emf synergistically activates the downstream genes KIN1 and COR15A in cold signaling. KIN1 is also regulated by stresses, such as dehydration and osmotic pressure (Wang et al., 1995).
In the stress category, which represents salt tolerance, dehydration-induced, and cation efflux family protein/metal tolerance (Park et al., 2009;Supplemental Table S1I), transcript levels of many genes were distinctly elevated in 7-d-old but not in 15-d-old mutants (Table II). This is because the wild type showed an age-dependent increase in transcript level from 7 to 15 d (Supplemental Table S1I). As a result, it is difficult to determine whether the mutation affected transcription for 15-d-old samples. However, the upregulation of stress genes in all 7-d samples of all three mutants is consistent with EMF repression of stress genes.
In the heat category, most wild-type genes showed an age-dependent increase in transcript level that is not consistently observed in the three mutants (Supplemental Table S1J). This has caused an apparent down-regulation of heat genes in the 15-d-old emf Figure 4. RT-PCR analysis of seed and stress genes in emf mutants. RNA samples extracted from 7-and 14-d-old wild-type (WT) and emf mutant plants were subjected to RT-PCR analysis. ACTIN8 was used as a loading control. Numbers presented below the gels represent relative expression levels and were calculated using the ACTIN8 signal as a reference.  (Table II). Thus, like the ethylene and stress categories, the 7-d-old samples showed negative regulation of heat genes by EMF. Most genes in the heat category encode Heat Shock Proteins (HSPs; Supplemental Table S1J), such as HSP70, which is involved in a variety of cellular processes (Su and Li, 2008). Our data suggest EMF regulation of diverse cellular processes via the modulation of the heat shock gene expression.

Genes in Other Categories
The photosynthesis category genes investigated in the 8 K array displayed the same expression pattern on the 26 K array. All but one of 54 photosynthesis genes examined on the 26 K array were down-regulated in one of the three mutants (Table II). The RNA level of 31 of the 54 genes was similar in wild-type seedlings (7 and 15 d old) and flowers (Supplemental Table S1K). However, the expression of these 31 genes was significantly down-regulated in emf mutants. For example, the transcript level of At2g47450, a chlorophyll a/bbinding protein gene, was similar between wild-type seedlings and flowers but was reduced by at least 3-fold in emf mutants. This result suggests that the expression change in photosynthesis genes is caused by the lack of EMF proteins, not the reproductive state of the emf mutants.
In contrast, in the photoreceptor category, most genes examined were up-regulated in the emf mutant samples (Table II; Supplemental Table S1L).
In the histone category, H1, H2A, H2B, H3, and H4, most genes in the wild type showed age-dependent decreased expression from 7-to 15-d-old plants, while many emf mutants showed age-dependent increased expression (Supplemental Table S1M). This opposite expression pattern resulted in the upregulation of all histone genes in 15-d-old mutants that cannot be attributed directly to the emf mutation. The comparative data of 7-d-old samples do not show an apparent expression change resulting from the emf mutations.
In the expansin category, despite the age-dependent decreased expression in the wild type from 7 to 15 d (Supplemental Table S1N), there is nevertheless an apparent down-regulation of expansin genes in emf mutants relative to the wild type, most notably in the emf1-2 samples (Table II). These findings suggest that EMF normally promotes the expression of expansin genes.
The transcription factor category (Qu and Zhu, 2006), which includes homeodomain, zinc finger, WRKY, SET domain, MYB, MADS, AP2, bHLH, and others (Supplemental Table S1O), is by far the largest category of genes showing expression changes in emf mutants (Table II). More genes are up-regulated than down-regulated in five of the six mutant samples. In the two emf1 mutants, more than 30% of the genes were up-regulated and less than 10% of the genes were down-regulated in 7-d-old emf mutants. In 15-dold mutants, only 15% of the genes were upregulated. EMF regulation of transcription factor genes is consistent with the PcG's role of master regulator of gene programs (Schwartz et al., 2006;Tolhuis et al., 2006).

Target Genes That Interact Directly with EMF1 and EMF2
Functional grouping of the genes that showed expression changes in emf mutants indicated EMF regulation of multiple pathways. However, it is not clear whether genes showing expression changes are directly or indirectly regulated by the two EMFs. Using protein tagging and ChIP technologies, we showed that AG, AP3, and PI are direct targets of the EMF1 protein (Calonje et al., 2008). EMF1 proteins are recruited to the promoter region of these genes and also to the second intron of AG. To investigate whether EMF2 also interacts with these genes directly, we employed the same strategy to tag EMF2 and study EMF2 interaction with potential target genes.

EMF2 Tagging and Protein Expression Pattern
EMF2 cDNA was tagged with three repeats of a FLAG sequence and expressed under the control of the EMF2 promoter (Fig. 5A). The EMF2promoter::EMF2 cDNA::3XFLAG construct was introduced into an emf2-1 mutant background as described in "Materials and Methods." It rescued the mutant; transgenic plants grew like wild-type plants (Fig. 5B). Nuclear proteins were extracted from wild-type and E2-3F transgenic plants. EMF2-FLAG proteins were detected on western blots in 7-and 14-d-old seedlings and in three different organs, rosette leaf, inflorescence apex, and silique, of plants grown in soil (Fig. 5C). This result is similar to the EMF1 protein expression pattern (Calonje et al., 2008). Finding EMF2 in all organs examined is consistent with constitutive EMF2 gene expression (Yoshida et al., 2001).

EMF1 and EMF2 Recruitment to Target Genes
To determine whether the EMF1 target genes recruit EMF2 proteins, chromatin from E2-3F transgenic seedlings was immunoprecipitated by anti-FLAG antibody as described (Calonje et al., 2008). Primers from the regulatory region of a flower organ identity gene, PI, were able to amplify DNA in the immunoprecipitated chromatin, indicating that EMF2 was recruited to the PI promoter region (Fig. 6).
To investigate whether the two EMF proteins directly regulate pathways other than those involved in flower organ development, we performed ChIP analysis on genes that encode transcription factors involved in seed development, stress, cold signaling, and flowering: ABI3, LOV1, FLC, and FT, respectively. PI serves as a positive control of EMF target genes in the ChIP experiments. PHE1 is a MADS box gene expressed in seed. PHE1 gene expression is regulated by FIS2-containing but not the EMF2-containing PRC2 (Kohler et al., 2003a). ChIP results showed that PHE1 promoter sequence did not recruit EMF1 or EMF2 protein (Fig. 6). Thus, PHE1 serves as a negative control in experiments aimed at identifying EMF target genes.
Using primers amplifying a promoter region of FT (FT-P in Fig. 6), we found that the FT promoter sequences did not recruit EMF1 or EMF2, suggesting that FT is not a target gene of EMF. However, using primers from the second exon of FT, Jiang et al. (2008) showed that FT chromatin is marked with K27me3 and bound to CLF. To examine possible EMF interaction with the second exon of FT, we tested the same primer sequences reported by Jiang et al. (2008;FT-I in Fig. 6). Neither EMF1 nor EMF2 was bound by FT-I. Thus, FT is not likely to be controlled by EMF directly. FT up-regulation in emf2 mutants may be a secondary effect. For example, FT may be negatively controlled by TOE1 and TEM, which, in turn, are regulated by EMF.
Both EMF1 and EMF2 were recruited to ABI3, LOV1, and FLC genes ( Fig. 6; Supplemental Figure  S1). Recruitment of EMF proteins to these genes is consistent with EMF repression of the master regulatory genes of seed development, cold signaling, and FLC-mediated flowering pathway during vegetative development.

DISCUSSION
emf mutants are characterized by their prominent precocious flowering phenotype (Sung et al., 1992;Bai and Sung, 1995); however, the mutation is pleiotropic, consistent with expression changes in genes in multiple pathways . Our genome-wide GeneChip study confirms and extends the previous findings by identifying additional gene programs coordinately regulated by EMF1 and EMF2, notably, stress-related gene programs.
The full-genome analysis identified 15 gene programs regulated by the two EMF genes. In most functional categories (i.e. flower organ, flowering time, photosynthesis, photoreceptor, seed, auxin, GA, ABA, and cold), gene expression levels remain relatively unchanged in 7-and 15-d-old wild-type seedlings, enabling the investigation of the effect of emf mutation on gene expression. Wild-type transcript levels changed from 7 to 15 d in several of the functional categories. For example, genes in the ethylene, stress, and heat categories had higher transcript levels in 15-d-old than in 7-d-old wild-type samples. Thus, in these categories, the 15-d-old data often showed an apparent gene expression change from mutant to the wild type that cannot be attributed to the mutation directly but rather relates to the developmental change in the wild type. Therefore, we ascertained the effect of mutation from the 7-d-old data. At this age, mutants and the wild type possess the same organs (e.g. cotyledons and hypocotyl), differing only in size. Figure 5. EMF2promoter::EMF2::3XFLAG construct and the protein expression pattern. A, The construct that harbors tagged EMF2 used in plant transformation is depicted. EMF2 cDNA was tagged with the 3XFLAG sequence and expressed under the EMF2 promoter. This construct was introduced into emf2-1, and it rescued the mutant phenotype. B, emf2-1 mutant and rescued emf2-1 (E2-3F) harboring EMF2promoter::EMF2::3XFLAG and E2-3F grown on agar plates (left) as well as E2-3F and the wild type (WT) grown in soil (right). C, Western-blot analysis of nuclear protein from various tissues of wild-type and E2-3F seedlings at 7 and 14 DAG, rosette leaves, inflorescence apex, and siliques of plants grown in soil. Anti-FLAG antibody was used to detect the EMF2-3XFLAG fusion protein. A portion of the Coomassie Brilliant Blue-stained gels in the range of 50 to 100 kD is shown as a loading control for western-blot analysis.
In most functional categories, with the obvious exception of the photosynthesis and expansin categories, more genes are up-regulated than downregulated in all three emf allele samples (Table II). EMF gene activity exerts a repressive role on flower organ, seed, photoreceptor, ethylene, ABA, stress, cold, heat, and transcription factor genes.
There is more overlap in genes with changed expression between the two emf1 mutants than between emf2-1 and emf1-1. Although this may indicate differential gene regulation by EMF1 and EMF2, gene redundancy between EMF2 and its homolog VRN2 (Schubert et al., 2005) may allow VRN2 to substitute EMF2 function in emf2-1 mutants. This is evident from the fact that, while EMF2 can interact with ABI3 (Fig.  6), presumably to repress its expression, ABI3 transcripts remained low in emf2 mutants. This can be explained by the redundant function of VRN2 , which would repress ABI3 in the absence of EMF2.

EMF Regulation of Flowering Time Genes
We have previously proposed that EMF proteins, bypassing flowering time genes, repress flower organ genes directly to prevent flowering. Indeed, EMF1 interaction with AG, AP3, and PI confirms them as direct EMF1 target genes (Calonje et al., 2008). Nevertheless, some flowering time genes in the photoperiod, autonomous, and vernalization pathways (e.g. GI, FT, and FLC) were up-regulated in some samples of emf mutants. In the photoperiod flowering pathway, GI activates CO, which then activates both FT and SOC1 (Michaels et al., 2005). It is curious that the increased GI expression did not cause up-regulation of CO. Perhaps, positive regulation of CO by GI is offset by enhanced negative regulation. To date, four genes, SUPPRESSOR OF PHYA1, CYCLING DOF FACTOR1, RED FAR-RED INSENSITIVE1, and LOV1, have been identified as CO negative regulators (Imaizumi et al., 2005;Chen and Ni, 2006;Laubinger et al., 2006;Yoo et al., 2007), but only LOV1 is up-regulated in emf mutants (Table III; Fig. 2). Thus, in emf mutants, reduced LOV1 repression may antagonize increased GI activation of CO.
The elevation of FT transcripts in emf2-1 without an increase of CO transcripts (Fig. 2) might be explained by the reduced level of FT repressors. The increased level of GI function in emf mutants would reduce the level of TOE1, which is negatively regulated by GI (Jung et al., 2007), allowing FT level to rise. Expression of the FT repressors TEM1 and TEM2 was also reduced in emf mutants, resulting in further elevation of FT transcripts (Table III).
The CLF-containing PRC2 is thought to mediate flowering by interacting with FLC and FT, as CLF binds to intron regions of FLC and FT and mediates the deposition of repressive histone H3K27me3 on FLC and FT chromatin (Jiang et al., 2008). In this study, we found EMF1 and EMF2 recruited to FLC but not FT (Fig. 6). As CLF interacts with EMF2 as well as VRN2 in vitro through the VEF domain shared between EMF2 and VRN2 (Chanvivattana et al., 2004;Wood et al., 2006;Chen et al., 2009), FLC may interact with the PRC2 composed of CLF/SWN-EMF2-FIE-MSI as well as that of CLF/SWN-VRN2-FIE-MSI1. Since FT is not bound by EMF2, it may interact only with the PRC2 composed of CLF/SWN-VRN2-FIE-MSI1. VRN2 and EMF2, the two Su(z)12 homologs, are partially redundant in that double mutants display a more severe phenotype than emf2 single mutants (Schubert et al., 2005). VRN2-containing PRC2 represses FLC to mediate the vernalization process (Gendall et al., 2001). It is conceivable that EMF2 also serves a redundant function to VRN2 by repressing FLC-mediated vernalization. In addition, EMF may participate in the endogenous mechanism of flowering time determination via epigenetic regulation of FLC.
The up-regulation of FT or FLC may not affect flowering time in emf mutants, as the downstream flower MADS box genes are already derepressed. Constitutive expression of multiple flower MADS box genes is sufficient to convert leaves into flower organs upon germination (Honma and Goto, 2001), precluding the influence of upstream flowering Figure 6. EMF1 and EMF2 proteins interaction with target genes. A, Schematic representation of gene structures of PI, ABI3, LOV1, FLC, FT, and PHE1. Black boxes depict the first exon including the 5# untranslated region, and arrows indicate the region of the gene amplified in ChIP analysis, 2320 to 274 bp for PI, 2746 to 2586 bp for ABI3, +267 to +393 bp for LOV1, 294 to +232 bp for FLC, 2330 to 270 bp for FT-P, +1,050 to +1,325 bp for FT-I, and 259 to +141 bp for PHE1. P and I denote promoter and intron regions in FT, respectively. B, ChIP results showing EMF1 and EMF2 interaction with target genes. Anti-FLAG antibody was used to immunoprecipitate nuclear proteins from 14-d-old transgenic plants, and primers from A were used to amplify the DNA in the immunoprecipitated chromatin (IP). Input is preimmunoprecipitated DNA after sonication. Thirty cycles of PCR were performed using the primers covering the DNA regions shown in A. PHE1 was used as an internal negative control. E1-3F represents emf1-2 rescued by EMF1::EMF1::3XFLAG (Calonje et al., 2008). genes on flowering time in these plants. However, reducing without eliminating EMF or CLF activities results in a weak phenotype (i.e. early-flowering, rather than emf mutant, phenotype; Goodrich et al., 1997, Aubert et al., 2001Yoshida et al., 2001;Sanchez et al., 2009). In these early-flowering plants, it remains possible that their flowering time may be influenced by the altered expression of flowering time genes.

EMF1 and EMF2 Target Transcription Factor Genes to Mediate Vegetative Development and Stress Response
In each pathway regulated by EMF, some genes encode transcription factors and others are downstream genes regulated by these transcription factors. ChIP analysis showed that EMF1 and EMF2 interact with genes encoding transcription factors (e.g. ABI3 and LOV1) that regulate the seed maturation program and stress-induced genes. Loss of repression of these genes in emf mutants would explain the up-regulation of their downstream genes, AtEm1, KIN1, and COR15A.
Functional analysis showed that the two EMFs promote the photosynthetic program and regulate hormone signaling as well as cold-, heat-, drought-, and other stress-induced gene programs. Among the genes related to the four major plant hormones, those in the ABA signaling pathways are most clearly repressed by EMF. ABA is essential in many aspects of vegetative development, such as stomata regulation and plant adaptation to stress conditions, like drought and low temperature (Seki et al., 2007). A number of cold-related genes were up-regulated in all three emf alleles (Table IV; Fig. 4). EMF1 and EMF2 were directly bound to ABI3 and LOV1, which are genetically located upstream in cold signaling (Fig.  7). EMF regulation of all categories of stress genes indicates its involvement in protection mechanisms against abiotic environmental stresses. Stress response requires rapid activation of gene expression, recently shown to involve ATP-dependent chromatin remodeling (Mlynarova et al., 2007;Walley et al., 2008). Our finding of EMF interaction with the key stress regulatory gene LOV1 implies PcG involvement in epigenetic regulation of the stress response program, probably to maintain stress genes in a silent state until stress is encountered. Future investigation of stress regulation of EMF expression may clarify the biological function of EMF recruitment to stress genes.
Vegetative development requires not only the repression of seed and flower programs but also of the extensive growth and differentiation needed to elaborate the shoot architecture. These processes are coordinated by the hormone signaling pathways. Expansins are key regulators of cell wall extension during growth (Li et al., 2003). Down-regulation of expansin genes in emf mutants corresponds with reduced cell size in emf mutants (Sung et al., 1992). Thus, EMF is apparently required for cell wall extension in growing seedling cells (e.g. hypocotyl and root cells). Robust vegetative development also depends on vigorous photosynthesis, and EMF is required for transcription of genes in the photosynthesis category. The two EMF genes apparently are involved in the regulation of all of these processes needed for successful vegetative development.
In summary, our results indicate that EMF1 and EMF2 repress both developmental and stress response pathways, primarily through direct repression of the transcription factor genes (Fig. 7). Coordinated EMF regulation of these diverse programs is crucial to plant survival. Future studies on genome-wide targets of EMF in a cell-, tissue-, and organ-specific manner would further elucidate the EMF-mediated epigenetic mechanisms.

Mechanism of EMF-Mediated Gene Silencing
The transcriptome of emf1 and emf2 mutants shows global regulation of gene expression patterns by EMF, consistent with PcG's role as master regulator of gene programs (Schwartz et al., 2006;Tolhuis et al., 2006). PcG functions to maintain gene silencing via histone modifications, primarily through H3K27 trimethylation . A genome-wide study showed that 4,400 Arabidopsis genes were marked with H3K27me3 (Zhang et al., 2007a). PcG proteins (i.e. EMF1, EMF2, and CLF) are recruited to the regions of the flower organ identity genes (AG, AP3, and PI) in seedling nuclei that are also marked with H3K27me3. Furthermore, study of mutants showed that H3K27 trimethylation depends on EMF2, CLF, and partially on EMF1 Calonje et al., 2008).
This study identified additional EMF targets that encode transcription factors, FLC, ABI3, and LOV1, which are marked by H3K27me3 (Zhang et al., 2007b). These findings suggest a close association of H3K27me3 with EMF-targeted silencing. How- ever, not all genes marked by H3K27me3 are bound by EMF: PHE1 and FT are marked with H3K27me3 in seedling nuclei (Zhang et al., 2007b;Jiang et al., 2008) but are not bound by EMF1 or EMF2 (Fig. 6). It is possible that another PRC2 component, such as VRN2, is involved in their methylation. Whole genome analysis of the EMF-binding pattern in the wild type and the H3K27me3 pattern in emf mutants will shed light on the relationship of EMF and H3K27me3 in determining target gene regulation in Arabidopsis.

Plant Materials and Growth Conditions
Wild-type and emf mutant plants of Arabidopsis (Arabidopsis thaliana) used in this study are from the Columbia ecotype background and have been described . Seeds were surface sterilized and plated on agar plates containing two-fifths-strength Murashige and Skoog medium (Murashige and Skoog, 1962). The plates were placed for 2 d in a refrigerator and then transferred to a short-day growth room (8 h of light/16 h of dark) at 21°C. Seedlings were harvested after growth for 7 or 15 d at 21°C, called 7or 15-d-old seedlings or seedlings at 7 or 15 DAG, respectively. Ten-day-old seedlings grown in short days were transplanted to soil and grown in a long-day (16 h of light/8 h of dark) greenhouse maintained at 21°C. Rosette leaves and inflorescence apex tissues were harvested about 3 weeks after transplantation.

Microarray Analysis
RNA isolation, cDNA synthesis, biotinylated cDNA probe synthesis, and hybridization were performed according to standard protocols with minor modification (Zhu et al., 2001). The images were acquired and quantified by MAS 5.0 (Affymetrix). Hybridization signals of perfectly matched probes were condensed into a gene-level expression index by a custom algorithm (Zhou and Abagyan, 2002). The average gene expression indices of all chips were scaled to an arbitrary target of 100.

Transgenic Plant Construction and Protein Expression Analysis
EMF2 promoter of 1,020 bp and 3# 3XFLAG-tagged EMF2 cDNA were cloned into binary vector pCAMBIA1380. This construct was introduced into Agrobacterium tumefaciens strain GV3101 by the freeze-thaw method and subsequently transformed into emf2-1 heterozygous plants by floral dipping (Clough and Bent, 1998). The homozygous emf2-1 mutants rescued by EMF2promoter::EMF2::3XFLAG transgenic plants were confirmed by genotyping. Genomic DNA extracted from wild-type-like hygromycin-resistant trans-genic plants was used to amplify the region that covers the emf2-1 mutation in the EMF2 locus using the following primers: 5#-AACAACAAATTGCAGAA-GACTGAAG-3# and 5#-CTTGGATATCATTGTCTCAGTCTTG-3#. The resulting fragments (460 bp) were digested with Hyp188III, which would cut the emf2-1 allele into two fragments, 180 and 280 bp, but would not cut the wildtype EMF2 allele.
For EMF2 protein expression analysis, nuclei from different ages and tissues of wild-type and transgenic plants were obtained as described (Bowler et al., 2004). Nuclei were then resuspended in SDS-PAGE loading buffer, heated, and loaded on an SDS-PAGE gel. Proteins from the gel were transferred onto a membrane, and EMF2-3XFLAG protein was detected using anti-FLAG antibody (Sigma-Aldrich).

Supplemental Data
The following materials are available in the online version of this article.
Supplemental Figure S1. ChIP-PCR analysis of EMF interaction with target genes at three PCR cycles: 27, 30, and 33.
Supplemental Table S1. Transcript levels of genes in the 15 functional categories.
Supplemental Table S2. Selection criteria for the 15 functional categories.