Interaction of Polycomb-group proteins controlling flowering in Arabidopsis

A general feature of developmental patterning is that it occurs progressively so that patterns are gradually refined based on information from earlier, cruder patterns (Coen,1999; Stern,1968). This presents a mechanistic problem in growing embryos or organ primordia, because most patterning events are thought to involve gradients of morphogens that operate across small numbers of cells and cannot persist over larger distances (Lawrence and Struhl, 1996). A typical resolution to this problem is for the transcriptional output of early patterning events to become fixed, so that information from early events is inherited through cell division during somatic development. This often involves epigenetic changes in gene activity,i.e. changes that are heritable through mitotic (and sometimes meiotic)division but are not caused by alterations in DNA sequence(Russo et al., 1996). The advantage of epigenetic changes in a development context is that although stable they are also reversible, particularly during meiosis, so that changes that accrue during somatic development can be erased at the onset of each new generation.

In Drosophila and other animals, the epigenetic control of developmental patterning is mediated by members of the Polycomb group (Pc-G)and trithorax group (trx-G) of genes (for a review, see Francis and Kingston, 2001). A general feature of these genes is that they are required not for pattern initiation, but rather to ensure that the transcriptional output of early patterning events is stably inherited through somatic development. The two groups act antagonistically, so that the Pc-G genes are required for maintenance of transcriptional repression and the trx-G genes for maintenance of transcriptional activation. Recently, biochemical characterisation of their protein products has provided some mechanistic insight. The Pc-G products,which are structurally disparate from one another, are found in at least two distinct complexes, termed the Polycomb Repressive Complex 1 and 2 (PRC1 and PRC2) (Cao et al., 2002; Czermin et al., 2002; Francis et al., 2001; Kuzmichev et al., 2002; Muller et al., 2002; Saurin et al., 2001). Consistent with the epigenetic function of Pc-G proteins, the PRC2 was recently shown to modify chromatin. Thus, several groups have shown that the PRC2 has a histone methyltransferase (HMTase) activity, methylating specific residues (lysine 9 and lysine 27) on the N-tail of histone H3(Cao et al., 2002; Czermin et al., 2002; Kuzmichev et al., 2002; Muller et al., 2002). The precise biochemical function of the different PRC2 members is not well defined, with the exception of one member, the Enhancer of zeste [E(z)]protein, which has been shown to confer HMTase activity via a conserved motif,the SET domain. Unlike most other SET domain proteins, E(z) itself does not show HMTase activity in vitro unless associated with other members of the PRC2 complex (Czermin et al.,2002). The mK9 H3 and mK27 H3 modifications catalysed by the PRC2 are associated with the repression of transcription, although how they are interpreted and inherited through mitosis is not well understood. The Polycomb protein appears to recognise and bind mK27 H3, and may recruit other members of the PRC1 complex, which probably have roles in mediating transcriptional silencing and its propagation through mitosis(Cao et al., 2002; Czermin et al., 2002; Francis et al., 2001).

The PRC2 components are also found in plants, and were identified independently through genetic screens in Arabidopsis aimed at dissecting various developmental pathways. Thus, the FERTILISATION INDEPENDENT SEED (FIS) genes were mostly identified through screens for mutants that showed some aspects of seed development in the absence of fertilisation (Chaudhury et al.,1997; Grossniklaus et al.,1998; Guitton et al.,2004; Ohad et al.,1996). Currently four FIS genes have been identified: MEDEA (MEA), FERTILISATION-INDEPENDENT SEED2 (FIS2),FERTILISATION-INDEPENDENT ENDOSPERM (FIE) and MULTICOPY SUPPRESSOR OF IRA 1 (MSI1). These encode products with homology to the Drosophila PRC2 proteins E(z), Suppressor of zeste 12 [Su(z)12],Extra sex combs (Esc) and P55, respectively(Grossniklaus et al., 1998; Kiyosue et al., 1999; Kohler et al., 2003a; Luo et al., 1999; Ohad et al., 1999). The FIS genes repress expression of the MADS box gene PHERES1(PHE1) during early seed development, and presumably affect many other as yet unidentified target genes(Kohler et al., 2003b). A second group have been identified based on a common function in repressing floral homeotic gene expression. Mutants in this class are early flowering and exhibit mild homeotic transformations in flowers. The first two members identified were CURLY LEAF (CLF) and EMBRYONIC FLOWER2 (EMF2), which encode proteins with homology to E(z) and Su(z)12, respectively (Goodrich et al.,1997; Yoshida et al.,2001). Recently, the FIS genes MSI1 and FIE have also been implicated in repressing flowering homeotic genes during vegetative development. Because mutant alleles of the FISgenes all cause early embryo lethality when inherited maternally, this obstructed the phenotypic analysis of fis homozygotes during later developmental stages. However, studies of transgenic lines that confer a partial loss of FIS gene activity have revealed roles for FIE and MSI1 beyond seed development(Hennig et al., 2003; Katz et al., 2004; Kinoshita et al., 2001). A third class of Arabidopsis Pc-G genes was identified on the basis of the function of the genes in the epigenetic memory of vernalisation. In Arabidopsis, as with many other plant species originating from temperate latitudes, flowering is accelerated if plants are first vernalised by growing for 3-6 weeks at low temperatures (4-10°C). The vernalisation response has several epigenetic features, including stability during somatic development and resetting from generation to generation. Recent studies indicate that the underlying basis for the response involves transcriptional repression of FLC, a gene that itself represses flowering(Michaels and Amasino, 1999; Sheldon et al., 1999). The VERNALIZATION2 (VRN2) gene is required so that the cold-induced repression of FLC is mitotically stable during later periods of growth at warm temperatures(Gendall et al., 2001). VRN2, like FIS2 and EMF2, encodes a protein with homology to Drosophila Su(z)12(Gendall et al., 2001). The completion of the Arabidopsis genome sequence revealed that these comprised most of the Arabidopsis homologues of the core members of the PRC2. One exception was a third E(z) homologue, GenBank accession At4g02020, that had not been characterised genetically. In addition, there are four genes with weak similarity to MSI1, MSI2-5, with poorly defined functions (Ach et al., 1997; Hennig et al., 2003; Kenzior and Folk, 1998). Thus,unlike Drosophila, in which the PRC2 members are single copy genes,in Arabidopsis the different members are mostly small gene families. The duplicated members appear to have acquired distinct functions; thus CLF and MEA function in repressing flowering and repressing endosperm proliferation, respectively. It is not clear how this has occurred;for example, whether it simply reflects different expression patterns of MEA and CLF, or whether their protein products have also diverged in function.

In addition to the conservation of PRC2 members in plants, there is also evidence that their proteins may act together. Thus, several studies have shown that the FIE protein can interact with the E(z) homologues MEA, CLF and At4g02020 (Katz et al., 2004; Kohler et al., 2003a; Luo et al., 2000; Spillane et al., 2000; Yadegari et al., 2000). Also,compelling evidence for interaction of FIE and MSI1 proteins was recently presented (Kohler et al.,2003a). However, the role of the plant Su(z)12 homologues has remained obscure. Despite the similarities in the phenotype of fis2 and the other fis mutants, no interaction between FIS2(or any other Su(z)12 homologue) with the other Pc-G proteins has been found.

Here, we show that the plant E(z) and Su(z)12 homologues interact both genetically and physically through their protein products. We localise the interactions to motifs that are conserved between the plant and animal proteins. We show that the third Arabidopsis E(z) homologue functions largely redundantly with CLF, so that CLF has a more general role in control of plant development than was apparent from its single mutant phenotype. Characterisation of the misexpression phenotypes for the three E(z) homologues indicates that they have diverged not only in expression, but also at the protein level. We suggest that in plants an evolutionarily ancient complex (the PRC2) has been conserved, but gene duplication and divergence has given rise to several complexes with partially discrete functions.

The clf-2 and clf-9 alleles arose in Lerbackground and were described previously(Goodrich et al., 1997). The null clf-50 allele (Ws background) was provided by E. Huala and harbours a deletion spanning the CLF locus (J.G., unpublished). The weak emf2-10 allele arose in Ws background during a T-DNA mutagenesis experiment and was provided by M. Running. The weak swn-1 allele (Ws background) was identified in seed pool 5887 in the University of Wisconsin Arabidopsis knockout collection(Krysan et al., 1999). The swn-1 line also carried an unlinked, recessive mutation conferring late flowering. The line was backcrossed twice to the Ws progenitor and swn-1 lines with and without the late flowering mutation were generated. The severe swn-2 and swn-3 alleles (Columbia background) were obtained from the SIGNAL collection of T-DNA insertion lines(Alonso et al., 2003) and correspond to accessions SALK 010213 and SALK 050195, respectively. The position of the T-DNA inserts was confirmed by PCR amplification and sequencing of genomic DNA flanking the inserts.

Constructs for yeast two-hybrid analysis were generated using the vectors pGBT9 and pGAD424 (Clontech) that express protein fusions to the GAL4 DNA-binding domain or transcriptional-activation domain, respectively. cDNA inserts encoding plant Pc-G proteins were introduced as EcoRI/SalI fragments. The Quik Change site-directed mutagenesis system (Stratagene) was used to introduce in-frame EcoRI and SalI restriction sites within cDNA clones, with the exception of EMF2 clones, which were generated by PCR amplification using mutagenic primers. PCR-generated clones were validated by sequence analysis. The methods for two-hybrid analysis were as described in the yeast protocols handbook (Clontech). The analysis was performed in yeast strain Hf7c(Feilotter et al., 1994),which carries HIS3 and LacZ reporters for reconstituted GAL4 activity, or in strain AH109 (James et al., 1996), which carries HIS3 and ADE2 reporters.

Vectors were used as described in Kim et al.(Kim et al., 2002). CLF-C5 was cloned into pENTRY 3c (Invitrogen) and recombined into the bait vector using the Gateway system (Invitrogen), resulting in a CLF-C5-Cub-URA3 gene fusion. EMF2-VEFS was fused to a gene encoding the N-terminal part of ubiquitin in the vector pCGK. The plasmids were transformed into the Saccharomyces cerevisiae strain JD53 and interaction of the fusion proteins was monitored as ability to grow on 5-fluoroorotic acid (5-FOA) plates, containing yeast nitrogen base without amino acids (Difco) and glucose, supplemented with lysine, leucine, uracil, and 1 mg/ml 5-FOA.

A similar protocol to that described in Kohler et al.(Kohler et al., 2003a) was applied. The coding region for the CLF C5 domain (amino acids 257-331) was cloned into the pGEX-4T expression vector (Amersham) as a GST-fusion, whereas the EMF2 VEFS domain (amino acids 427-631) was cloned in the pET30a expression vector (Novagen) as a HIS₆-fusion. Escherichia coli strain BL21 DE3 Codon-plus (Stratagene) was freshly transformed with the pGEX-CLF-C5,pGEX or pET-EMF2-VEFS plasmids and grown in LB medium at 37°C overnight. After diluting the cultures 1:100 in 250 ml LB, they were grown at 37°C(pGEX-CLF-C5 and pGEX) or 18°C (pET-EMF2-VEFS) until OD₆₀₀=0.7. Production of recombinant protein was induced by adding isopropyl-β-D-thiogalactopyranoside (IPTG) to 0.2 mM and after growing the cells for 3 hours at 18°C, they were harvested and resuspended in 4 ml binding buffer [BB; 20 mM Tris pH 7.5, 150 mM NaCl, 0.1% Triton X100, 1 μM ZnSO4, 1 mM Pefabloc (Roche)]. The cells were lysed by the addition of Lysozyme to 2 mg/ml and incubated for 20 minutes on ice. The solution was centrifuged (20,000 g) for 10 minutes, the pellets discarded,centrifuged again and a 100 μl sample of supernatant was mixed with SDS sample buffer and frozen in liquid nitrogen (input control sample). Equal volumes of extract containing HIS₆-EMF2-VEFS were mixed with extracts containing GST-CLF-C5 or GST and 150 μl of pre-equilibrated glutathione-sepharose 4B beads (Pharmacia) and incubated with shaking for 2 hours at 4°C. The beads were washed four times with BB and then mixed with SDS sample buffer, analysed on protein blots and the HIS₆-EMF2-VEFS fusion detected with anti-HIS₆-antibodies (New England Biolabs).

Constructs for expression of CLF, SWN and MEA cDNAs under control of the cauliflower mosaic virus 35S promoter were assembled using the pART7 and pART27 vector systems (Gleave,1992). A full-length SWN cDNA clone (pda05864) was obtained from the Riken Bioresource centre, Japan (Seki et al., 2002), CLF and MEA cDNA clones were isolated previously(Goodrich et al., 1997; Spillane et al., 2000). The Quik Change site-directed mutagenesis system (Stratagene) was used to engineer restriction sites within the cDNA clones that facilitated subcloning the coding sequences into pART7. The constructs were introduced into Agrobacterium strain GV3101 pMP90(Koncz and Schell, 1986) and used to transform clf-50/+ heterozygotes by floral dip transformation(Clough and Bent, 1998). At least 23 primary transformants were identified for each construct. Selected plants in the T1 and T2 generations were genotyped for presence of a transgene and for clf-50 and CLF+ alleles by Southern blot analysis.

The methods for in-situ hybridisation analysis using digoxigenin-labelled mRNA probes were described previously(Narita et al., 2004). SWN probes were generated from a poorly conserved 700 bp region at the 5′ end of the SWN coding region. WUS probes were generated using the clone pMH WUS 16, generously provided by R. Simon.

Scanning electron microscopy (SEM) was performed on a Hitachi 4700 with a Gatan Alto cryo-stage. The methods for cryo-SEM were as described previously(Jeffree and Read, 1991). For measuring cell sizes, fully expanded rosette leaves were fractured in transverse section and photographed using the cryo-SEM. The cell outlines were traced onto transparencies, scanned, and quantified using image analysis software (image tool, University of Texas, available at http://ddsdx.uthscsa.edu/dig).

To identify genes acting like CLF to repress floral homeotic gene expression during vegetative development, we screened existing mutant collections for plants with a similar leaf curling and early flowering phenotype. We identified a single recessive mutation, designated moe leaf, which conferred a phenotype resembling, but more severe than, clf mutations. Subsequent genetic analysis (see next section)revealed that it was an unusual, weak emf2 allele. We hereafter refer to it as the emf2-10 allele but retain the name `moe leaf' to describe the phenotype. Both emf2-10 and clf mutants flowered early under both long days and short days(Table 1). However, emf2-10 plants were significantly earlier flowering than clf-50 plants, which carried a null clf allele isolated in the same genetic background as emf2-10. Both mutants gave small,dwarfed plants that had short slender inflorescence stems and narrow leaves that curled upwards along the leaf margin(Fig. 1A,B). emf2-10plants were smaller than clf-50 plants(Fig. 1A) and, unlike clf, also had cotyledons that were smaller than in wild-type plants(Fig. 1C). Comparison of emf2-10 and wild-type leaf epidermal surfaces by SEM showed that both had pavements of large cells with irregular outline(Fig. 2A,B). Unlike wild-type leaves, which had flat surfaces, the ventral (abaxial, lower) epidermis of emf2-10 leaves was uneven and corrugated(Fig. 2B). One possibility,which is also consistent with the upward curling of the leaves, is that growth of the ventral leaf surface was constrained by the dorsal surface during leaf development, leading to the observed corrugation. When leaves were frozen and fractured, so that they could be viewed in transverse section by SEM, emf2-10 leaves had a similar arrangement of cell types as in wild type (Fig. 2C,D), but the cells were smaller than wild type and there were also many fewer cells in the leaf length and leaf width axes (data not shown). The emf2-10 mutation therefore affects cell proliferation as well as cell size. Similarly,morphometric analysis has shown that clf mutant leaves also show reductions in both cell number and cell size(Kim et al., 1998).

The inflorescences of emf2-10 plants produced few flowers. In wild type, the primary inflorescence produced 25-30 flowers before arresting development. In emf2-10, after 6-12 flowers had opened, the remaining flower buds arrested development so that the inflorescence subsequently appeared determinate. The flowers of emf2-10 plants mostly had normal organ identity but were smaller than wild type, and petals and sepals were narrower than in wild type (Fig. 1D). The flower buds often opened later than normal, after fertilisation had occurred, so that the elongation of the developing silique(fruit) was constrained and the siliques became bent or folded over(Fig. 1E, Fig. 2E). This suggested that the sepals were impeding bud opening. Wild-type sepals have a hyaline margin,distinguishable under SEM as a region of regularly sized cells that lack the extremely elongated cells found elsewhere on the sepal(Fig. 2F). In emf2-10flowers the margin was less well defined, so that the elongated cells often extended to the margin (Fig. 2G). In addition, the sepals were more concave or boat shaped than wild type. Both features may have contributed to restricting bud opening. As with clf mutants, emf2-10 flowers produced late in development showed weak homeotic transformation of sepals to carpels(Fig. 1F, Fig. 2H) and petals to stamens(Fig. 2I,J).

The similarities in phenotypes suggested that the moe leaf phenotype, like that of clf mutants, could be caused by misexpression of floral homeotic genes during vegetative and floral development. Previous studies have shown that the AG and AP3 genes, whose expression is normally confined to flowers, are misexpressed in leaves of clfmutants (Finnegan et al.,1996; Goodrich et al.,1997; Serrano-Cartagena et al., 2000). We therefore used RT PCR to compare AG and AP3 expression in leaves of wild-type and emf2-10 plants. This indicated that both genes were expressed in emf2-10 leaves (data not shown). To confirm this, we introduced reporter constructs for AGand AP3 expression into the emf2-10 background. The pAG-I::GUS construct contains AG upstream promoter sequences and intragenic sequences fused to the GUS reporter and has been shown to contain the cis-acting sequences necessary for repression by CLF (Sieburth and Meyerowitz,1997). This construct was strongly misexpressed in seedlings of both emf2-10 and clf-2 mutants(Fig. 1G,H). In addition, both mutants showed misexpression in inflorescence stems(Fig. 1J,K) and occasional misexpression in the outer floral whorls. We also tested reporter constructs containing the AG second intron upstream of a GUS reporter gene (KB9)(Busch et al., 1999). This construct also confers the wild-type AG expression pattern in flowers, presumably because the second intron contains many AGregulatory elements (Busch et al.,1999). However, when the KB9 construct was introduced into clf or emf2-10 mutant backgrounds, no expression was seen in seedlings (data not shown). This suggested that the AG promoter contains additional enhancers that are required for misexpression of AG in clf and emf2-10 mutant backgrounds. An AP3 reporter construct containing 3.7 kb of upstream regulatory sequences (Jack et al., 1994)also showed weak expression in emf2-10 and clf seedlings but not wild-type (Fig. 1L,M).

Genetic data have indicated that the phenotype of clf mutants is chiefly caused by ectopic AG expression. Thus in clf agdouble mutant plants, in which AG activity is eliminated, leaf morphology is restored to near wild-type(Goodrich et al., 1997; Serrano-Cartagena et al.,2000). To test whether the moe leaf phenotype was also a result of ectopic AG activity, we made emf2-10 ag-2 double mutants. Although the double mutants had larger, less curled leaves than emf2-10 single mutants, there was less restitution of wild-type morphology than in the case of clf ag double mutants. Thus emf2-10 ag-2 plants were still much smaller than wild type, their leaves retained some curling, and they flowered earlier(Fig. 1N). This indicated that although AG⁺ activity contributes to the moe leaf phenotype, misexpression of other genes is also probably involved. Taken together, these results indicated that EMF2 and CLF shared common functions in repressing floral homeotic gene expression, with EMF2 required to repress a broader range of targets than CLF. These similarities suggested that EMF2 might act in a common pathway with CLF.

To determine the molecular basis for the moe leaf phenotype, we employed a map-based cloning strategy and initially localised the mutation responsible to a 10 cm interval between markers nga129 and ATTED2 on the lower arm of chromosome 5. It was striking that the plant Pc-G member EMF2 had also been located within this interval. All nine emf2 mutant alleles previously described have much more severe phenotypes than moe leaf, producing minute plants that appear to flower soon after germination without undergoing a prior phase of vegetative development(Sung et al., 1992; Sung et al., 2003; Yang et al., 1995). Instead, a few flowers and sessile cauline leaves are produced on an inflorescence with a severely shortened bolt (Fig. 1O, Fig. 3C). However, several features made EMF2 a promising candidate. Firstly,it was known to repress floral homeotic gene expression during vegetative development (Chen et al., 1997; Moon et al., 2003). Secondly,transgenic plants that expressed an antisense EMF2 construct had a phenotype resembling moe leaf, which probably reflected a partial loss of EMF2 function (Yoshida et al.,2001). To test whether the moe leaf phenotype could be caused by an unusual, weak allele of EMF2, we performed genetic complementation tests. Because emf2 mutants are sterile, and moe leaf plants have low fertility, we crossed heterozygotes for the two mutations. The resulting F1 population of 254 plants contained 71 mutants, consistent with the two mutations being allelic (1/4 mutants expected, χ²=1.2 P>0.1). We designated the new mutation responsible for the moe leaf phenotype as the emf2-10 allele. The phenotype of emf2-10/emf2-3 heterozygotes was intermediate between that of the two parental alleles, consistent with emf2-10 being a weaker allele than emf2-3 (Fig. 1O). To identify the lesion causing the emf2-10 mutation, we compared the sequence of the EMF2 locus from emf2-10 and the wild-type progenitor. This revealed that the emf2-10 allele carried a 17 bp deletion extending from the 3′ end of the second exon (9 bp) into the 5′ end of intron 2 (8 bp) followed by a cytosine to guanine substitution(see Fig. S1A,C in the supplementary material). Because this deletion was predicted to affect splicing of the EMF2 pre-mRNA, we used RT-PCR to amplify EMF2 cDNA from emf2-10 and wild-type seedlings. Whereas a single message corresponding to the spliced EMF2 transcript was detected in wild type cDNA, five novel transcripts were identified in emf2-10 cDNA (see Fig. S1B in the supplementary material). Molecular cloning and sequencing of these aberrant transcripts indicated that four contained frameshift mutations likely to abolish EMF2 activity. However, one transcript was predicted to produce a variant EMF2 protein that was truncated by 17 amino acids at the N-terminus (see Fig. S1C in the supplementary material). The region deleted does not correspond to a conserved region or to a known functional domain, so the variant protein is likely to retain EMF2⁺ activity. The weak emf2 phenotype may arise because only a small fraction (about 20%) of the various emf2-10 transcripts are likely to produce a functional protein. In addition, the resulting truncation of the protein may also reduce its activity.

The similarity in phenotypes of severe alleles of CLF and weak alleles of EMF2 suggested that the two genes might act in a common genetic pathway. To test for a genetic interaction we therefore combined weak alleles of CLF and EMF2 by constructing the clf-9 emf2-10 double mutant. The weak clf-9 allele was derived from the severe clf-2 allele by an imprecise excision of a transposon from the CLF locus (P. Puangsomlee, Phd thesis, University of East Anglia,1997) (Goodrich et al., 1997). clf-9 plants are very similar to wild type but are slightly smaller,show earlier flowering under short days and very weak leaf curling(Fig. 3A). A synergistic interaction was observed, so that the double mutant had a much more extreme phenotype than either parent, producing extremely small plants with few,sessile leaves and very short inflorescences(Fig. 3A,B). The double mutant phenotype therefore resembled that of severe emf2 alleles such as emf2-3 (Fig. 3C). A similar phenotype was also observed in double mutant combinations of the null clf-2 allele and weak emf2-10(Fig. 3D). In double mutant combinations of severe emf2-3 and severe clf alleles, emf2 was epistatic to clf (not shown). In general these observations were consistent with the two genes acting in a common pathway.

The severity of the clf emf2 double mutant phenotype suggested that it was unlikely to result simply from misexpression of AG. To confirm this, we constructed ag-1 clf-2 emf2-10 triple mutants. The triple mutants were minute plants, similar to severe emf2 mutants in size, and had 1-3 flowers with normal petals and ag– phenotype(Fig. 3E). This indicated that AG misexpression was not responsible for the severe effects of clf emf mutants on overall plant size, but did account for the poor development of petals in clf emf mutant flowers(Fig. 3D).

To test whether the genetic interaction of CLF and EMF2might reflect a direct interaction between their protein products, we performed yeast two-hybrid assays. We expressed full-length EMF2 protein, and a series of EMF2 truncations, as `prey' fusions with the GAL4 transcriptional activation (TA) domain. We tested these fusion proteins for interaction with a`bait' comprising a fusion of a truncated CLF protein (lacking the C-terminal SET domain) with the GAL4 DNA-binding domain. We did not observe an interaction between full-length EMF2 proteins with CLF in yeast. However,yeast strains expressing both CLF and a C-terminal portion of EMF2 expressed both two-hybrid reporter genes, consistent with the two proteins interacting(Fig. 4A). The C-terminal portion of EMF2 contained the VEFS domain, a motif originally defined on the basis of its conservation between plant and animal homologues of the Su(z)12 protein (Gendall et al.,2001). It was not clear why the full-length EMF2 protein, which includes the VEFS box, did not also interact with CLF.

To define the region of CLF that is required for interaction with EMF2, we tested a series of CLF truncations as baits with the EMF2 VEFS box prey construct (Fig. 4B). We thus mapped the interaction to a short 74 amino acid region of CLF that contains the C5 domain. The C5 domain contains five cysteine residues whose presence and spacing is conserved between plant, Drosophila, nematode and vertebrate E(z) homologues (Goodrich et al., 1997; Grossniklaus et al., 1998; Holdeman et al.,1998). No function has previously been ascribed to this domain. To verify the interaction between CLF and EMF2 in an independent system, we first used the yeast split-ubiquitin assay, which differs from the two-hybrid assay in that candidate proteins are fused to portions of the ubiquitin protein and the fusions are expressed in the cytoplasm rather than the nucleus(Johnsson and Varshavsky,1994; Kim et al.,2002; Stagljar et al.,1998). Again, we observed an interaction of CLF with the VEFS box domain of EMF2 (Fig. 4C). Secondly, to confirm that CLF and EMF2 interact directly, we performed in-vitro binding assays (Fig. 4D). Both proteins were expressed in E. coli, the C5 domain of CLF as a glutathione-S-transferase (GST) fusion and the EMF2 VEFS domain as a HIS₆-tagged fusion. As shown in Fig. 4D, the HIS₆-EMF2 VEFS protein bound to GST–CLF C5 (lane B) but not to GST alone (lane C), suggesting a direct physical interaction between the proteins. Thus, the CLF C5 domain and EMF2 VEFS domain bound to each other in vitro as well as in yeast.

The Arabidopsis FIS genes FIS2 and MEA encode homologues of EMF2 and CLF, respectively. Although the FIS2 and MEA genes share extremely similar mutant phenotypes, suggesting that their products may interact, we were previously unable to demonstrate any interaction between the full-length proteins using the two-hybrid assay(Spillane et al., 2000). However, the observation that the interaction of EMF2 with CLF was mediated by the VEFS box domain suggested that FIS2 and MEA might also interact via the VEFS box. We therefore specifically tested the VEFS box of FIS2 against MEA in two-hybrid assay and in this case were able to demonstrate an interaction(Fig. 5A).

The genetic and molecular interactions between CLF and EMF2 suggested that their protein products probably acted in a common complex. However, two observations were at odds with this scenario: firstly,null clf alleles had much less severe phenotypes than null emf2 alleles; secondly, the phenotype of null clf alleles was enhanced by emf2 mutant alleles. One possible explanation was that the CLF gene showed redundancy, so that even in a null clf background, a similar activity was provided by other genes. This seemed a possibility because the Arabidopsis genome contains two other genes with strong similarity to CLF. The first, MEA,shows expression confined to the female gametophyte and early seed development and is therefore unlikely to overlap significantly with CLF, which is expressed predominantly during vegetative and inflorescence development(Goodrich et al., 1997; Vielle-Calzada et al., 1999). The second gene, accession At4g02020 [referred to as EZA1 by Luo et al.(Luo et al., 2000)], had not been genetically characterised. We designated this gene SWINGER(SWN), because our subsequent analysis of its protein product and mutant phenotype indicated a potential to share partners with the CLF protein(see below). Phylogenetic analysis of plant E(z) homologues(Fig. 5B) indicated that SWN and CLF belonged to distinct clades that can be clearly distinguished even in species distantly related to Arabidopsis, for example maize and rice. The duplication event that gave rise to CLF and SWN was therefore an ancient one within the angiosperm lineage. In addition, the CLF and SWN clades were clearly much more similar to one another than either was to MEA, suggesting that the function of SWN was more likely to resemble that of CLF than MEA.

To determine the SWN expression pattern, we localised its mRNA by in-situ hybridisation to sections of seedlings and inflorescences. SWN was expressed throughout the apical meristem and leaf primordia of 8-day-old wild-type seedlings (Fig. 6A,B). Expression was also detected in the vasculature of hypocotyls and cotyledons (Fig. 6B). In inflorescences, SWN was expressed throughout the inflorescence meristem and young stage 1-3 floral meristems(Fig. 6C). In older flowers,expression was weak in the sepals and stronger in the inner whorls containing developing petals, stamens and carpels(Fig. 6D,E). In stage 12 flowers, strongest expression occurred in the ovules, particularly in the funiculus and maternal tissues of the ovule(Fig. 6F). Expression was also seen in the female gametophyte, but the tissues were too poorly preserved to distinguish the different cell types within the gametopyhte(Fig. 6F). Little signal was observed when seedlings and inflorescences were hybridised with a probe from the sense strand of the SWN cDNA(Fig. 6G,H), confirming that the signal was specific for the SWN antisense probe. As a positive control, we also hybridised seedlings with a probe for the WUSCHEL(WUS) gene and detected expression confined to the centre of the shoot meristem (Fig. 6I) as previously described (Mayer et al.,1998). The SWN expression pattern was therefore similar to that of CLF (Goodrich et al.,1997), with both genes being generally expressed during vegetative and reproductive development but with strongest expression in meristems and other regions of dividing cells.

To test whether the SWN protein had similar properties to those of CLF, we compared their interactions in yeast two-hybrid assays. We observed an interaction between the EMF2 VEFS domain and both of the CLF or the SWN C5 domains, indicating that SWN had similar potential to interact with EMF2 to that of CLF (Fig. 5C,D). We further tested whether CLF and SWN could interact with the related Arabidopsis VEFS domain proteins FIS2 and VRN2. Both were able to interact with FIS2 and VRN2 in yeast, indicating a potential for one or both to function in the FIS and vernalisation response pathways(Fig. 5C,D). In addition, we found that both CLF and SWN can interact with FIE through a 110 amino acid motif at their N-termini (see Fig. S2 in the supplementary material)(Luo et al., 2000). Thus SWN and CLF showed similar interactions with both EMF2 and FIE in yeast. Taken together, the similarities in expression pattern and protein–protein interactions confirmed the potential for the CLF and SWNgenes to act redundantly.

To identify the function of SWN, we exploited facilities for reverse genetics in Arabidopsis to identify a series of mutant alleles caused by T-DNA insertions within the locus. The swn-1 allele contained an insertion 3 bp upstream of the predicted ATG start codon. This allele is unlikely to be null, as RT-PCR analysis of swn-1 mRNA revealed chimeric transcripts that initiated within the T-DNA insertion and extended the full length of the SWN coding sequences (data not shown). The swn-2 insertion is within an intron and swn-3within an exon, but both are upstream of the catalytic SET domain and are therefore likely to represent null alleles. All three alleles were viable as homozygotes and had no obvious phenotype that we could discern from inspection of gross plant morphology, embryo or endosperm development (data not shown). However, all three alleles strongly enhanced the clf mutant phenotype in clf swn double mutant combinations, confirming that the two genes exhibit redundancy. The swn-1 allele gave a less severe enhancement than did swn-2 or swn-3, consistent with its being a weaker allele. The swn-1 clf-50 double mutant gave extremely small, early flowering plants with few flowers that resembled emf2 mutants(Fig. 3F,G). SEM analysis indicated that the floral organs showed weak homeotic conversion to carpelloid structures (Fig. 2K). In addition, filamentous organs were observed in place of stipules, a phenotype that has also been observed in plants that have a partial loss of FIE+ activity (Katz et al.,2004). Double mutants of the null clf-50 allele with either swn-2 or swn-3 were more extreme, and viable plants were recovered only when seedlings were grown in sterile tissue culture. The seed germinated and produced seedlings with narrow, but relatively normal,cotyledons, hypocotyl and roots. As the plants aged they became increasingly abnormal. The cotyledons developed finger-like projections on their margins. The shoot apex did not initiate leaves, but instead developed into a disorganised mass of green tissue on which poorly differentiated organs formed(Fig. 3H). In SEM analysis of the plants, the epidermi of these organs lacked trichomes and comprised small,isodiametric cells, which did not have the surface cuticular thickening or elongated cell shape that is characteristic of epidermal surfaces of most of the mature floral organs (Fig. 2M,N). In addition, colourless callus-like tissue formed and eventually gave rise to somatic embryos and roots(Fig. 3H,I). Unlike the single mutants, which had normal roots, the primary root of the double mutants became opaque, swollen and eventually produced green shoot-like tissue(Fig. 3J). A similar phenotype has been observed in seedlings of rescued fie homozygotes(Kinoshita et al., 2001). Together, these observations suggested that weak clf-50 swn-1 double mutants resembled emf2 mutants, whereas the null clf swndoubles were more extreme and resembled plants lacking FIE⁺ activity.

Although the above data suggest that the CLF and SWNgenes have very similar functions, the fact that clf mutants have a clear phenotype indicates that SWN is not identical in function to CLF, at least with respect to repression of AG. This might be due to subtle differences in level of expression between CLF and SWN, and/or changes in protein function. To clarify whether differences are solely due to changes in expression, we expressed full-length cDNA clones for each gene under control of a common promoter (the cauliflower mosaic virus 35S promoter) and introduced the two transgenes into the null clf-50 mutant background. Whereas the 35S::CLF construct fully complemented the clf-50 mutation, the 35S:SWNconstruct did not (Fig. 7). There are therefore subtle differences in function between the CLF and SWN proteins, as might be expected given the persistence of the CLF/SWN duplication within angiosperms. Expression of 35S::MEA failed to complement the clf-50 mutation,indicating that the MEA protein has also diverged from CLF(Fig. 7).

The plant and animal Pc-G proteins of the E(z) class share several motifs of which the CXC and SET domains towards the C-termini of the proteins are the most highly conserved. The proteins also share a less-well-conserved region towards their N-termini, termed the C5 domain, that contains five cysteine residues with conserved spacing in the arrangement CRRCX₂DCX₂HX_(22-27)CX₃CY. The arrangement of cysteines does not correspond with any previously defined cysteine cluster motif such as the C2H2 zinc-finger motif involved in binding DNA. However, the region is functionally important because at least one mutant allele maps within the C5 domain: the Drosophila E(z)²⁸allele is a mis-sense allele that swaps the fifth conserved cysteine for a tyrosine, and it gives a temperature-sensitive loss of function phenotype(Carrington and Jones, 1996). We show that this domain mediates the binding of plant E(z) homologues with the VEFS domain of plant Su(z)12 homologues. It is likely that the C5 and VEFS domains have a similar function in animals. Consistent with this, it was recently shown that mammalian Su(z)12 can interact through its VEFS domain with the mouse E(z) homologue EZH2(Yamamoto et al., 2004). Although the region of EZH2 required for the interaction was not mapped, we note that the region expressed in this study (residues 238-746) included the C5 domain. It is also noticeable the C5 domain is less well conserved in the Caenorhabditis elegans E(z) homologue, MES2, than in the other animal proteins. C. elegans, unlike insects and vertebrates, lacks a Su(z)12 homologue, so the relatively poor conservation of the MES2 C5 domain may be because it no longer functions in this protein–protein interaction.

In animals, the core members of the PRC2 complex comprise four proteins first identified in Drosophila as the Esc, P55, E(z) and Su(z)12 proteins (Cao et al., 2002; Czermin et al., 2002; Kuzmichev et al., 2002; Muller et al., 2002). There is now strong evidence that structurally and functionally equivalent complexes occur in Arabidopsis. Thus several previous studies have shown genetic and physical interactions of FIE with MEA and CLF, and also of FIE with MSI1 (for a review, see Reyes and Grossniklaus, 2003) (Fig. 8). In particular, Kohler et al.(Kohler et al., 2003a)partially purified an FIS complex and showed that it contained FIE, MEA, MSI1 and, based on molecular weights, probably several other unidentified components. However, the role of the FIS2, VRN2 and EMF2 proteins has remained enigmatic, although the strong similarity between the fis mutant phenotypes suggested that FIS2 might interact with one or more of the other FIS proteins. We have now shown that EMF2 interacts physically and genetically with CLF. We extend this to show a general potential for the Su(z)12 homologues FIS2 and VRN2 to interact with the E(z) homologues MEA, CLF, and SWN, at least in yeast two-hybrid assays. Taken together, these observations strongly suggest that there are Arabidopsis complex(es) that are structurally equivalent to at least the core members of the animal PRC2 members. It is also likely that they have an equivalent biochemical function in mK27 H3 histone methylation. Thus, the Arabidopsis VRN2 protein was recently shown to be required for vernalisation-induced mK27 H3 methylation at the FLC gene(Bastow et al., 2004; Sung and Amasino, 2004). However, biochemical purification of the plant PRC2 complexes will be necessary to confirm that they have a direct HMTase activity.

Whereas the FIE gene is single copy, the other Arabidopsis PRC2 members are represented by small gene families with three to four members. Within these families, the different members control at least three different processes: firstly, repression of endosperm proliferation during gametophyte and endosperm development (FIS2,MEA); secondly, repression of floral homeotic gene expression during embryo development and vegetative development (EMF2, CLF); thirdly,epigenetic control of vernalisation response (VRN2). We suggest that these reflect their participation in at least three complexes that differ in their target gene specificity (Fig. 7) (Reyes and Grossniklaus,2003). The distinct roles of the Arabidopsis PRC2 members may in part reflect differences in expression patterns. For example, several studies suggest that FIS2 and MEA expression is confined to female gametophyte and seed development, whereas EMF2 and CLF are also expressed more broadly during vegetative development,where they act to repress genes controlling flowering time or floral development (Goodrich et al.,1997; Luo et al.,2000; Vielle-Calzada et al.,1999; Yoshida et al.,2001). However, differences in expression are not sufficient to account for the altered roles. Thus, even when CLF, MEA and SWN cDNAs are expressed under control of a common promoter (CaMV 35S), only the CLF transgene is able to complement clfmutants. This suggests that differences between the CLF, MEA and SWN proteins are also important. Thus, following duplication, the plant Pc-G genes appear to have diverged in protein function as well as expression.

It is unclear how these complexes might acquire specificity for different target genes, as PRC2 members appear to lack intrinsic DNA-binding specificity and the recruitment of Pc-G members to specific targets is not yet well understood either in animals or plants(Birve et al., 2001; Carrington and Jones, 1996). One possibility is that PRC2 members are recruited to targets by interaction with sequence-specific DNA-binding proteins(Wang et al., 2004). A recent alternative model is that Pc-G members could achieve sequence specificity through interactions with small RNAs(Steimer et al., 2004). Because the FIE gene is a single copy, and its protein product is probably common to all complexes, it is unlikely that FIE could distinguish the activity of different complexes. However, small differences between the EMF2/VRN2/FIS2 and/or MEA/CLF/SWN proteins could change their affinities for protein partners that target the complex. It is striking that the FIS2/VRN2/EMF2 class of protein is the only one of the PRC2 members that is not also conserved in C. elegans. This implies that this protein is not absolutely required for the biochemical activity of the complex, and might therefore play a role in specifying its targets. In addition to differences between complexes in their target gene specificities, there must also be differences between the CLF family members and the EMF2 family members in their affinity for one another. For example, if CLF has equal affinity for FIS2, EMF2 and VRN2 and the FIS2 family members have equal affinity for CLF,MEA and SWN, then proteins such as MEA and CLF should be able to cross-complement one another when misexpressed. We did not observe such differences in yeast two-hybrid assays; for example, CLF and SWN showed a similar potential to interact with each of the EMF2, VRN2 and FIS2 proteins. However, the interactions in yeast may not accurately reflect subtle differences in affinity in plants. It will be interesting to test whether swapping the C5 domains between CLF and MEA proteins, or other regions, can alter their specificity in vivo.

The CLF and SWN genes show similar expression patterns and encode closely related proteins that display identical interactions in several yeast two-hybrid assays. We tested three independent swnmutant alleles and all three strongly enhance the clf single mutant phenotype, although they are without gross morphological phenotype by themselves. This suggests that there is substantial functional redundancy between the two genes, so that the roles of CLF are largely masked by SWN activity in clf single mutants. For example, a role for CLF and SWN in primary root development is not apparent from either single mutant phenotypes but is revealed in null clf swndoubles. The partial redundancy of CLF and SWN probably explains why null clf mutants have much less extreme phenotypes than null emf2 mutants, although the CLF and EMF2 proteins act together. Consistent with this, weak swn-1 clf-50 double mutants resembled emf2 mutants. However, null swn clf double mutants were more extreme than emf2 mutants and resembled plants lacking FIE⁺ activity. It is likely that the full function of EMF2is also masked by partial redundancy; for example, with the related VRN2 gene with which it shares overlapping expression.

Despite overlapping functions, CLF and SWN are not completely redundant with respect to one another: firstly, clfmutants have a phenotype, largely caused by ectopic AG expression,that is not complemented by SWN+ activity; secondly, 35S::SWN, unlike 35S::CLF, does not complement clfmutants; thirdly, phylogenetic comparisons indicate that CLF and SWN orthologues can be distinguished clearly in other plants,including monocotyledenous species such as rice and maize. This means that the CLF/SWN duplication is an ancient one within the angiosperm lineage. It is unlikely that the SWN gene would show such wide conservation if it did not have at least partially distinct functions from CLF. Although we did not identify gross morphological effects of null swnmutations, several of the phenotypes associated with other plant PRC2 members(for example, autonomous endosperm development or vernalisation response) are apparent only in specific phenotypic screens or genetic backgrounds. It is therefore likely that swn mutants do have a phenotype, but this was not manifest in our growth conditions or assays.

Recently, it was shown that VRN2 is required, after vernalisation treatments, for mK27 H3 methylation in chromatin of its target gene FLC (Bastow et al.,2004; Sung and Amasino,2004). In animals, mK27 H3 methylation by the PRC2 complex requires the E(z) protein, which contains a SET domain known to have HMTase activity (Cao et al., 2002; Czermin et al., 2002; Kuzmichev et al., 2002; Muller et al., 2002). Together these observations suggest that an E(z) homologue will be required for the vernalisation response. Consistent with this, we show that the VRN2 protein has potential to interact, through its VEFS domain, with the C5 domain of the E(z) homologues CLF and SWN. In preliminary experiments (data not shown), we did not observe gross effects of null clf or swnmutations on the vernalisation response comparable with those of vrn2or other vernalisation response mutants. It is possible that CLF and SWN act redundantly with respect to the vernalisation response, so that defects will be manifest only in double mutants. Unfortunately, the pleiotropic phenotype of clf swn double mutants makes it difficult to characterise their vernalisation response, at least by straightforward comparison of flowering times. One possibility will be to use chromatin immunoprecipitation (ChIP) to test whether FLC chromatin becomes enriched for CLF and/or SWN proteins following vernalisation treatments.

In summary, it is likely that the PRC2 complex is conserved between plant and animals, both structurally and also functionally in terms of its histone methylation activity. However, in plants there has been duplication of most components of the PRC2, and the duplicated members have diverged in protein function as well as in expression. This has given rise to several PRC2-like complexes in plants, with at least partially discrete functions in terms of target gene specificity. Expression of chimeric proteins that swap domains between duplicated components, such as CLF/SWN/MEA, may help identify how the changes in specificity are mediated. Despite the conservation of the PRC2 in plants, it is striking that there are no homologues of the animal Pc-G members that comprise the PRC1 complex. It is therefore possible that the mechanisms to interpret, maintain, and re-set epigenetic information conveyed by the PRC2 have evolved independently in plants. Alternatively, plants may employ similar protein motifs to those found in the animal PRC1 members, but in novel combinations.

J.G. was funded by a Royal Society University Research fellowship, Y.C. by a scholarship from the Government of Thailand, C.S. and A.B. by BBSRC PhD studentship awards, D.S. by a BBSRC postdoctoral fellowship and a fellowship from the German Academic Exchange Service and Z.R.S. by USDA99-35301-7984 and NSF IBN-0236399. We thank Chris Jeffree for help with SEM, Amelia Green for help with the yeast two-hybrid analysis, Magali Bic and Neil Haig for help with the in-vitro pull down assays, Mark Running and Elliot Meyerowitz for generously providing the emf2-10 allele and Caroline Dean for providing vernalisation requiring backgrounds and a VRN2 cDNA. We thank Laurent Deslandes and Imre Somssich for providing vectors for split-ubiquitin analysis and Nobamusa Yoshida for providing EMF2 cDNA clones.