Polycomblike PHD Fingers Mediate Conserved Interaction with Enhancer of Zeste Protein*

The products of Polycomb group (PcG) genes are required for the epigenetic repression of a number of important developmental regulatory genes, including homeotic genes. Enhancer of zeste (E(Z)) is a Drosophila PcG protein that previously has been shown to bind directly to another PcG protein, Extra Sex Combs (ESC), and is present along with ESC in a 600-kDa complex inDrosophila embryos. Using yeast two-hybrid and in vitro binding assays, we show that E(Z) binds directly to another PcG protein, Polycomblike (PCL). PCL·E(Z) interaction is shown to be mediated by the plant homeodomain (PHD) fingers domain of PCL, providing evidence that this motif can act as an independent protein interaction domain. An association was also observed between PHF1 and EZH2, human homologs of PCL and E(Z), respectively, demonstrating the evolutionary conservation of this interaction. E(Z) was found to not interact with the PHD domains of threeDrosophila trithorax group (trxG) proteins, which function to maintain the transcriptional activity of homeotic genes, providing evidence for the specificity of the interaction of E(Z) with the PCL PHD domain. Coimmunoprecipitation and gel filtration experiments demonstrate in vivo association of PCL with E(Z) and ESC inDrosophila embryos. We discuss the implications of PCL association with ESC·E(Z) complexes and the possibility that PCL may either be a subunit of a subset of ESC·E(Z) complexes or a subunit of a separate complex that interacts with ESC·E(Z) complexes.


gel filtration experiments demonstrate in vivo association of PCL with E(Z) and ESC in Drosophila embryos. We discuss the implications of PCL association with ESC⅐E(Z) complexes and the possibility that PCL may either be a subunit of a subset of ESC⅐E(Z) complexes or a subunit of a separate complex that interacts with ESC⅐E(Z) complexes.
Regulated homeotic gene expression is responsible for the specification of segmental identity along the anterior/posterior axis of developing metazoans. In Drosophila melanogaster this spatially restricted pattern of homeotic gene expression is ini-tiated early in embryogenesis by the products of the gap and pair-rule genes (see Ref. 1 for a review). The early patterns of gap and pair-rule gene expression cease at approximately stage 10, yet correct expression of homeotic genes needs to be maintained throughout development. Two groups of genes, the trithorax group (trxG) and Polycomb group (PcG), 1 act to maintain the correct expression of the homeotic genes after the degradation of the gap and pair-rule proteins. The trxG acts to maintain the expression of homeotic genes, whereas the PcG acts to maintain their repression (see Refs. 2 and 3 for reviews).
The PcG genes were initially identified in Drosophila on the basis of mutant phenotypes that resemble gain-of-function homeotic mutations. These phenotypes result from derepression of homeotic genes, outside of their normal expression domains, ϳ5-6 h after fertilization (see Ref. 4 for a review). Derepression occurred after normal initiation of expression, indicating that the PcG is required specifically for the maintenance of repression and not its initiation.
Fourteen PcG members have so far been identified from Drosophila, 10 of which have been characterized at the molecular level (5, 6; for review see Ref. 3). Mammalian homologs have also been identified for several members of the group, and mutant analysis of some of these members demonstrates a role in the regulation of HOX gene expression, analogous to that of the Drosophila PcG (see Ref. 7 for a review).
The mechanisms through which the PcG carries out its repressive function is unknown. Several models have been postulated, including the creation of a higher order chromatin structure that is inaccessible to transcription factors, or the sequestering of target genes into nuclear compartments that exclude the transcription machinery (3,8). Despite the limited understanding of the mechanism of action, there is considerable evidence that PcG proteins act in multimeric protein complexes to repress target gene expression. To date, two Drosophila PcG complexes have been isolated and several of their components have been identified. PRC1 is a 2-MDa complex that includes the PcG proteins Polycomb (PC), Polyhomeotic (PH), Posterior Sex Combs, and Sex Combs on Midlegs along with several other proteins (9). PRC1 has been shown to inhibit ATP-dependent chromatin remodeling by the human SWI/SNF protein complex in vitro (9). Several members of the Drosophila trxG are subunits of the Brahma complex, which is related to SWI/SNF (10 -12). Thus, PRC1 may contribute to PcG-depend-ent silencing by directly inhibiting chromatin remodeling that is necessary for transcriptional activation. ESC, E(Z), and PCL were found to not be part of the PRC1 complex. Recently, a separate 600-kDa complex containing ESC and E(Z) was identified (13,14). Analysis of the isolated ESC⅐E(Z) complex has identified two other subunits, the histone binding p55 protein and the histone deacetylase RPD3 (14), supporting the model that chromatin modification, including histone deacetylation, may play a central role in PcG-depending silencing. Composition of the ESC⅐E(Z) complex appears to be at least partially conserved in mammals (14,15). PCL is yet to be assigned to a protein complex.
In this study, we provide additional insight into the molecular activities of the PCL and E(Z) proteins. The PCL protein contains two PHD finger domains (16,17). These domains are characterized by a highly conserved Cys 4 -His-Cys 3 motif identified in a number of proteins involved in modulating transcription, including the trxG proteins TRX, Absent, small, or homeotic discs 1 (ASH1) and Absent, small, or homeotic discs 2 (ASH2) (17)(18)(19). It has been suggested that PHD fingers may mediate either protein⅐DNA interactions or protein⅐protein interactions (17). Two recent reports provide evidence that PHD fingers are capable of mediating protein⅐protein interactions. Adjacent PHD finger and bromodomains of the KAP-1 corepressor have been shown to work in concert to mediate interaction with a subunit of the NuRD complex (20). In addition, the PHD fingers of MLL, a human homolog of TRX, homodimerize and bind to Cyp33 (21). A PCL region encompassing the PHD fingers, the 40 amino acids separating them, and 40 flanking amino acids C-terminal to the second PHD finger shares 34% identity with two mammalian homologs, MTF2 (murine) and PHF1 (human) (22). Conservation of this region suggests that it is important in PCL function. Prior to this study, no interactions between PCL and other PcG members had been identified. E(Z) contains several domains that are highly conserved in mammalian homologs (23). However, with the exception of the Cys-rich domain, which is required for chromosome binding (24), the functions of these domains are not known. The ESCbinding domain has been mapped to a relatively unconserved 33-amino acid region near the N terminus (25).
Here we report yeast two-hybrid and in vitro binding data demonstrating direct interaction between PCL and E(Z). Significantly, this interaction is shown to be mediated by the PHD finger domain of PCL, providing supporting evidence that a PHD finger domain can function alone as a protein⅐protein interaction domain. Yeast two-hybrid analysis using human homologs also suggests that the PCL⅐E(Z) interaction is evolutionarily conserved. Coimmunoprecipitation and gel filtration chromatography experiments provide evidence for in vivo association of PCL with E(Z) and ESC.

EXPERIMENTAL PROCEDURES
Yeast Two-hybrid Constructs-All constructs, unless otherwise stated, were cloned using linker PCR to generate the appropriate restriction enzyme site at the 5Ј and 3Ј ends of fragments, which were then cloned into the pEG202 or pJG4-5 vectors (26). The template cDNA used for PCL constructs was the full-length Pcl cDNA AL15 (16). The template cDNA used for the LexA-PHF1 construct was pBS-PHF1 (22). The coding sequence for PHF1 residues 86 -240 were amplified and cloned as an EcoRI-XhoI fragment into pEG202. The template cDNA used for the LexA-hMTF2 and AD-hMTF2 constructs was human cDNA clone (ID number 589332; obtained from IMAGE, National Institutes of Health (27)). The PHD finger coding region (residues 102-256) was cloned as an EcoRI fragment into pJG4-5 and pEG202. The LexA-EZH1 and AD-EZH1 constructs were generated by dropping out a full-length EZH1 fragment (XhoI) from pGEX-EZH1 (25) and cloning it in-frame into pEG202 and pJG4-5, respectively. The full-length LexA-EZH2 and AD-EZH2 constructs were generated by PCR using pBS-EZH2 cDNA (28) as a template and cloned as an XhoI fragment. The LexA-TRX-PHDF construct was generated by PCR amplifying the coding sequence for TRX residues 1266 -1481 from the pTM15Ј-Trx PHD finger domains 1-3 template and ligating the PCR product into the EcoRI site of pEG202. LexA-ASH1-PHDF and AD-ASH1-PHDF constructs were generated by PCR-amplifying the coding sequence for ASH1 residues 1778 -1818 from a full-length ash1 cDNA and ligating the PCR product into the EcoRI-XhoI sites of pEG202 and pJG4-5, respectively. LexA-ASH2-PHDF and AD-ASH2-PHDF constructs were generated by ligating the PCR-amplified coding sequence for ASH2 residues 50 -131 from ash2 cDNA LD31680 into the EcoRI-XhoI sites of pEG202 and pJG4-5, respectively. Both ash1 and ash2 cDNAs were kindly provided by Allen Shearn. The AD-PcG constructs AD-PC, AD-⌬chrPC, AD-PC⌬3Ј, ADph, AD-phHD, AD-ph⌬N, AD-ph⌬S, AD-Psc, AD-Psc⌬B, AD-Su(z)2, AD-Su(z)2⌬B, AD-AsxA, AD-AsxZn, AD-AsxQ, AD-Scm2, AD-E(Pc)Ybox, and AD-E(Z) were generously provided by Hugh Brock and Michael Kyba (University of British Columbia) (29). 2 AD-PHO (full-length) was generated by PCR and cloned as an EcoRI fragment into pJG4-5 using pET-PHO (provided by Judy Kassis) as a template. AD-ESC (fulllength) was generated by PCR using pact-esc (provided by H. Brock) as a template and cloned into pJG4-5 as an XhoI fragment.
Site-directed Mutagenesis-Site-directed mutagenesis of Pcl was performed using the QuikChange site-directed mutagenesis kit (Stratagene), with the exception of the C430S-PCL CR mutation. A 2.54-kb open reading frame fragment of Pcl was cloned into pBSKSϩ and used as a template for the PCL CR mutagenesis. Specific oligonucleotides were used to generate C518S, C430A, C518A, PHDF1-QQQA, and PHDF2-QQQA mutations. The C430S-PCL CR mutation was generated by performing two high fidelity PCRs (using Pfu DNA polymerase, Stratagene), using the primer pairs Pcl5Ј Bam (5Ј-GGAGGATCCTGAT-GAACAACCATT-3Ј)/Nhe bottom (5Ј-AAAGCTAGCCACGCACATGGG-TCCACT-3Ј) and Nhe top (5Ј-AAAGCTAGCAAGCGATCGGATATCGA-A-3Ј)/Pcl3Ј (5Ј-GGAGGATCCTTATGATGCCATTTAC-3Ј). The products were restricted with NheI and then ligated to each other. All mutations were confirmed by sequence analysis and the PCL CR of each mutant (aa 424 -605) was PCR-amplified and cloned into pEG202 to generate LexA-mutant-PCL CR fusions. The stability of each LexA-PCL CR mutation was analyzed in yeast by performing Western blot analysis with the LexA antisera (provided by Roger Brent).
Site-directed mutagenesis of E(z) was performed using the Altered Sites mutagenesis kit (Promega) and the pAlter-e32-55.26 plasmid template, as previously described (25) to create the E(Z)EDE116AAA, E(Z)H122A, E(Z)L132A, G136A, E(Z)K144A, and E(Z)DGK147AAA mutations. All mutations were confirmed by sequence analysis, and the indicated coding regions were PCR-amplified using Pfu DNA polymerase and inserted into the pGEX-BgRP3i vector (25).
Yeast Two-hybrid Assays-The starting yeast strain used for the two-hybrid assay, EGY48 (MAT␣ his3 trp1 ura3 6lexAop-LEU2), was transformed with pEG202 and pJG4-5 plasmids, encoding the DNA binding domain and activator fusion proteins, respectively. Three individual transformant colonies were picked and streaked out on both glucose and galactose plates lacking leucine, uracil, histidine, and tryptophan. Growth was scored after 4 days at 30°C. Colonies of ϳ1 mm in diameter were scored as having strong interactions. Absence of growth on galactose medium indicated no interaction.
In Vitro Pull-down Assays-Radiolabeled PCL and E(Z) proteins were synthesized by in vitro transcription-translation with the TnTcoupled reticulocyte lysate system (Promega) and [ 35 S]methionine. 35 S-PCL was produced from Pcl cDNA AL15 (16) in the vector pBluescript-SK using T7 RNA polymerase. 35 S-E(Z) was produced from cDNA clone e32-55.26 in the vector pBC-SK using T7 RNA polymerase. Isolation of GST fusion proteins and GST pull-down assays were performed as previously described (25). pGEX-E(z) constructs were made by PCR-amplifying the indicated coding regions and ligating the PCR products into the pGEX-BgRP3i vector. His 6 fusion protein pull-down assays were performed in a manner similar to GST pull-downs, except that His 6 fusion proteins were purified from Escherichia coli according to Qiagen's protocol for purifying insoluble proteins under denaturing conditions, and immobilized on Ni-NTA-agarose instead of glutathione agarose. Binding and wash buffers were the same as those used for GST pull-downs, except that 20 mM imidazole was added to both buffers to reduce nonspecific sticking of radiolabeled proteins. Prior to using affinity-purified GST or His 6 fusion proteins in binding assays, aliquots of each protein sample were run on SDS-polyacrylamide gel electrophoresis gels, stained with Coomassie Blue to estimate protein concentra-2 H. Brock, personal communication.
tions, and sample volumes were adjusted to equalize the amounts of fusion proteins used in each assay. In GST and His 6 pull-downs, bound proteins were eluted in wash buffer supplemented with 100 mM free glutathione or 250 mM imidazole, respectively. pQE-PCL403-605 was constructed by inserting the PCR-amplified coding sequence for Pcl residues 403-605 into the pQE30 vector.
Antibody Production and Immunoprecipitations-Rabbit anti-PCL antibodies were generated against His 6 -PCL113-537 and affinity-purified essentially as previously described (24). The region encoding PCL residues 113-537 was amplified from Pcl cDNA AL15 and inserted into the pQE32 bacterial expression vector (Qiagen). His 6 -PCL113-537 fusion protein was affinity purified from E. coli using Ni-NTA-agarose and injected into rabbits (all injections and bleeds were performed at Covance Research Products, Inc.). Antibodies were affinity-purified by first mixing the antiserum with total E. coli extract that included His 6 -MDH (E. coli malate dehydrogenase; kindly provided by Bilal Armeneh and Steven Vik) that had been covalently attached to CNBractivated Sepharose. Unbound antiserum was then incubated with His 6 -PCL113-537 (attached to CNBr-activated Sepharose). Following thorough washing, bound antibodies were eluted in low pH buffer. These antibodies were found to cross-react with numerous proteins from Drosophila embryo extracts on Western blots. Therefore, a portion of these antibodies was re-affinity-purified against a smaller fusion protein, His 6 -PCL113-371. Both antibody fractions that bound to His 6 -PCL113-371 and those that were depleted of His 6 -PCL113-371-binding antibodies, which are assumed to recognize epitopes within PCL372-537, were tested on embryo Western blots. The anti-PCL372-537 antibody fraction displayed much greater specificity for PCL and was used for all immunoprecipitations and immunoblot assays. 10 l was used in immunoprecipitations, and 1:2000 dilutions were used for Western blots. Immunoprecipitations and Western blots were performed essentially as previously described (25). 20 and 15 l, respectively, of affinitypurified rabbit anti-E(Z) antibodies described before (24) and mouse HA.11 anti-HA monoclonal antibodies (Covance) were used for immunoprecipitations. For Western blots, anti-E(Z) and anti-HA antibodies were used at 1:800 and 1:10,000 dilutions, respectively.

The Conserved Region of PCL Interacts with E(Z) in the Yeast
Two-hybrid System-To determine whether PCL is able to interact with cloned members of the PcG, yeast two-hybrid assays were performed. Two constructs were generated in the LexA-yeast two-hybrid vector pEG202 (26), a full-length PCL construct, LexA-PCL, and a construct, LexA-PCL CR , containing the region of PCL that is conserved in its mammalian homologs, here denoted the PCL conserved region (PCL CR , aa 424 -605). This conserved region contains two PHD finger motifs and a region C-terminal to the second PHD finger (aa 566 -605, see Fig. 1A). Both the LexA-PCL and LexA-PCL CR constructs were tested for interaction with PcG members that had been cloned into the yeast two-hybrid activation domain vector, pJG4-5, to generate AD-PcG fusion proteins (see "Experimental Procedures" for a list of PcG members tested). Twohybrid interactions were assayed in the Saccharomyces cerevisiae strain EGY48, which contains six LexA binding sites upstream of an endogenous LEU2 reporter gene. Transformed colonies containing both a LexA and AD construct were tested for growth on leucine-deficient medium. Here, growth would only occur if the AD fusion interacted with the LexA fusion to activate the LEU2 reporter gene.
The LexA-PCL fusion protein did not interact with any PcG members in this assay (data not shown), but the LexA-PCL CR interacted with AD-E(Z) (Fig. 1B). In yeast containing both the LexA-PCL CR and AD-E(Z) constructs, growth was only observed on plates containing galactose, indicating that induction of the AD-E(Z) fusion protein (which is under the control of a galactose inducible promoter) is required for the interaction (Fig. 1B). Additional controls showed that growth required the presence of both the LexA-PCL CR and the AD-E(Z) fusion proteins (data not shown). The inability of the full-length PCL fusion protein to interact with E(Z) in this assay could reflect a limitation of using a transcriptional assay as a basis for identifying interactions between transcriptional repressors. Previous experiments showed that full-length Posterior Sex Combs and PH proteins appeared not to interact in the yeast twohybrid assay despite the fact that an interaction was observed between smaller fragments of the proteins (29).

FIG. 1. The conserved domain of PCL (PCL CR ) interacts with E(Z) in the yeast two-hybrid assay.
A, schematic representation of the PCL protein showing the conserved region that encodes the two PHD finger domains, depicted here by hatched boxes, the region between the PHD finger domains and the Polycomblike extended homology (EH) domain, which is represented by a shaded box. The amino acid position of each motif is stated above the respective boxes. B, growth assay with EGY48 yeast containing the following combinations of fusion proteins; left, LexA-PCL CR and AD alone; middle, LexA-PCL CR and AD-E(Z); right, LexA-alone and AD-E(Z). Yeast strains were streaked in parallel on medium lacking leucine but containing galactose (GAL, top) or glucose (GLU, bottom). Expression of the AD fusion proteins were induced on GAL but not GLU medium. Yeast strains were grown for 4 days at 30°C. Only yeast containing both LexA-PCL CR and AD-E(Z) were able to grow.

The Interaction between PCL and E(Z) Is
Direct-To test whether the interaction between PCL and E(Z) is direct, in vitro binding assays were performed. His 6 -PCL403-605 fusion protein, which includes the PCL conserved region, and His 6 -MDH (E. coli-malate dehydrogenase, which served as a negative control in these assays) were purified from E. coli and tested for binding to full-length radiolabeled E(Z) protein produced by in vitro translation. Equal amounts of His 6 -PCL-(403-605) and His 6 -MDH were immobilized on Ni-NTA-agarose and incubated with 35  Reciprocal pull-down experiments confirmed direct PCL⅐E(Z) interaction and were used to map the PCL-interacting domains of E(Z) protein. The assays were performed as described above, except that full-length radiolabeled PCL protein was tested for binding to GST alone or to GST-E(Z) fusion proteins that were purified from E. coli and immobilized on glutathione-agarose. PCL does not bind to the highly conserved E(Z) Cys-rich or SET domains, which are included in GST-E(Z)-(512-760) (Fig. 2B, lane 3), but rather binds to E(Z) domains included in GST-E(Z)-(1-511) (Fig. 2B, lane 2). Attempts to further delimit the region of E(Z) responsible for PCL binding revealed two separate PCL-interacting domains, E(Z)-(86 -171) (Fig. 2B, lane 7) and E(Z)-(377-446) (Fig. 2C, lane 2). All GST-E(Z) fusion proteins that were able to bind to PCL included one or both of these regions. Essentially identical results were obtained in yeast two-hybrid assays in which the same E(Z) regions were tested as AD-E(Z) or LexA-E(Z) fusions for interaction with LexA-PCL CR or AD-PCL CR , respectively (data not shown). Therefore, two separate E(Z) regions are able to bind independently to PCL in vitro and in yeast cells. E(Z)-(86 -171) includes the highly conserved Domain I (aa 95-170), which is 70.9 and 60.5% identical to the homologous regions of the human EZH2 and EZH1 proteins, respectively (23), but E(Z)-(377-446) is much less well conserved. On the basis of their newly defined functions, we will refer to the 86 -171 region as PBD1 and the 377-446 region as PBD2 (for PCL Binding Domains 1 and 2). Alignments of these E(Z), EZH2, and EZH1 sequences are shown in Fig. 3 (B and C). The positions of these and other E(Z) domains, which were identified on the basis of evolutionary sequence conservation (23,30) or defined by function (25), are shown in Fig. 3A.
Effects of E(Z) Point Mutations on PCL Binding-To identify E(Z) residues within the two PCL binding regions that are required for the interaction with PCL, and to provide additional evidence for the specificity of the interaction, selected E(Z) residues within the PCL binding regions were replaced by alanine by site-directed mutagenesis. Five mutations in PBD1 and two mutations in PBD2 were tested. The locations of these mutations are shown in Fig. 3 (B and C). Two of the PBD1 mutations, EDE116AAA and DGK147AAA, are substitutions of adjacent residues that are conserved in E(Z) and both human homologs, EZH1 and EZH2. Three PBD1 mutations, H122A, G136A, and K144A, are single-amino acid substitutions of residues that are conserved in E(Z) and EZH2, but not EZH1. These were selected because EZH2, but not EZH1, interacts with PHF1, a human homolog of PCL (described below). On the basis of this same reasoning, the two PBD2 mutations are substitutions of adjacent residues at which E(Z) and EZH2 are either identical or contain conservative substitutions. Initially, the effect of each mutation was tested within the context of the isolated respective regions, 86 -171 or 377-446. Of the five PBD1 mutants, only EDE116AAA exhibited strongly reduced PCL binding (Fig. 3D, lane 3). Of the two PBD2 mutants, EIN429AAA is essentially unable to bind PCL (Fig. 3E, lane 4), whereas the HEN412AAA mutation has no effect on PCL binding (Fig. 3E, lane 3). Either mutation alone has little, if any, effect on PCL binding when expressed as a larger fusion protein that includes a wild type version of the other PCL-interacting domain (Fig. 3F, lanes 5 and 6). However, GST-E(Z)-(1-511) fusion protein containing both mutations is unable to bind to PCL (Fig. 3F, lane 4). Thus, either PBD1 or the less conserved PBD2 is capable of binding to PCL in vitro.
The Highly Conserved PHD Finger Domains of PCL Are Responsible for the Interaction with E(Z)-To ascertain which sequences within the PCL conserved region are responsible for the interaction between PCL and E(Z), a LexA-PCL CR⌬EH (aa 424 -565) construct was generated containing a truncated conserved region lacking the extended homology region (aa 566 -605). When LexA-PCL CR⌬EH was tested for its ability to interact with AD-E(Z), growth was observed on leucine-deficient medium, indicating that the region of extended homology is not required for the interaction between E(Z) and PCL (Fig. 4B).
This refined the interaction domain to the two PHD finger domains and the conserved 40 amino acids separating them. To determine if the PHD finger domains mediated the interaction between PCL and E(Z), PHD finger mutants were generated using in vitro site-directed mutagenesis (Fig. 4A). As previously discussed, the PHD finger contains a Cys 4 -His-Cys 3 motif, which is thought to coordinate Zn 2ϩ (17). No structural studies have been performed on this recently defined domain, and it is not known whether the conserved spacing of the Cys and His residues is important. It is also not known whether the Cys residues form hydrogen bonds to coordinate Zn 2ϩ binding or covalently bind to Zn 2ϩ . Presumably, these Cys and His residues have a structural role. Mutating one of these highly conserved residues would therefore be expected to destroy the ability to coordinate Zn 2ϩ .
Three mutagenesis experiments were designed to test the role of either PHD finger in the PCL⅐E(Z) interaction. The first mutagenesis converted the second conserved Cys in each PHD finger (residue 430 in PHDF1 and residue 518 in PHDF2) to a Ser, which is likely to destroy the ability of that cysteine to covalently coordinate Zn 2ϩ but may not destroy the ability to form a hydrogen bond (Fig. 4A). These constructs were named LexA-C430S-PCL CR and LexA-C518S-PCL CR for mutations in PHDF1 and PHDF2, respectively. The second mutagenesis converted the second conserved Cys to an Ala in both PHDF1 and PHDF2 (residues 430 and 518, respectively). These constructs were called LexA-C430A-PCL CR and LexA-C518A-PCL CR , respectively (Fig. 4A). The conversion of a conserved Cys to an Ala would be expected to destroy the ability of the PHD finger to chelate Zn 2ϩ and the ability to form hydrogen bonds.
Apart from the conserved Cys and His residues, Aasland et

FIG. 2. In vitro binding of the PCL and E(Z) proteins and localization of the E(Z) domains that bind to PCL.
Depicted are autoradiographs of SDS gels. The input lanes contain 20% of the amount of radiolabeled protein used in each binding assay. A, radiolabeled E(Z) protein was tested for binding to His 6 -PCL-(403-605) protein, which includes PCL CR . B and C, radiolabeled PCL was tested for binding to GST or GST-E(Z) fusion proteins containing the indicated E(Z) residues.
al. (17) also noticed a region of highly conserved hydrophobicity adjacent to the third conserved Cys. This region corresponds to residues 439 -442 in PHDF1 and residues 527-530 in PHDF2 (Fig. 4A). The significance of this stretch of residues is not known, and, in the PHD finger domains of PCL, only two of the three residues are hydrophobic. But the conservation of this stretch of amino acids in many PHD finger-containing proteins suggests that it is of importance. The third mutagenesis was therefore designed to convert this stretch of hydrophobic residues to hydrophilic Gln, residues that would be expected to destroy the function of this stretch of amino acids. In addition to altering the hydrophobic residues to Gln, the adjacent Cys was changed to an Ala to generate LexA-PHDF1-QQQA-PCL CR and LexA-PHDF2-QQQA-PCL CR , respectively.
All six constructs, LexA-C430S-PCL CR , LexA-C518S-PCL CR , LexA-C430A-PCL CR , LexA-C518A-PCL CR , LexA-PHDF1-QQQA-PCL CR , and LexA-PHDF2-QQQA-PCL CR were shown by Western analysis using anti-LexA antisera to be expressed at a level similar to LexA-PCL CR (data not shown), and all six were tested individually for their ability to interact with AD-E(Z). All six PHD finger mutants failed to interact with AD-E(Z), indicating that the PHD finger domains are required to mediate the interaction between PCL and E(Z) and that both PHDF1 and PHDF2 are required for this interaction (Fig. 4C).

PCL Is Associated with E(Z) and ESC in Drosophila
Embryos-To test whether the interactions observed in the yeast two-hybrid and in vitro binding assays reflect in vivo association of PCL and E(Z), coimmunoprecipitation experiments were performed. Anti-E(Z) and anti-PCL antibodies separately were used to immunoprecipitate proteins from extracts prepared from Drosophila embryos expressing epitope-tagged ESC protein, HA-ESC, which provides wild type esc function in vivo (25). Precipitated proteins were electrophoresed and immunoblotted. Duplicate Western blots were probed with either anti-E(Z) (Fig. 5, top panel) or anti-PCL (Fig. 5, middle panel) antibodies. Lane 3 shows that the E(Z) protein was immunoprecipitated (top panel) and that the PCL protein was coimmunoprecipitated (middle panel) in the same sample. Reciprocal immunoprecipitations were performed using anti-PCL antibodies. Lane 4 shows that PCL was immunoprecipitated by anti-PCL antibodies (middle panel) and that E(Z) was coimmunoprecipitated (top panel). Neither protein was precipitated by protein A-Sepharose alone (lane 2, top and middle panels). From these results, we conclude that PCL and E(Z) are associated in Drosophila embryo extracts. E(Z) previously has been shown to bind directly to another PcG protein, ESC (25,31). Recently, E(Z) and ESC have been shown to be subunits of a 600-kDa complex in Drosophila embryos, which is distinct from the 2-MDa PRC1 complex (9,13,14). If PCL is associated with the ESC⅐E(Z) complex, then PCL would be expected to also coimmunoprecipitate with ESC. To test this, additional aliquots of the same immunoprecipitations described above were immunoblotted and probed with anti-HA antibodies (Fig. 5, bottom panel). Lanes 3 and 4 show HA-ESC was coimmunoprecipitated by both anti-E(Z) and anti-PCL antibodies, respectively. HA-ESC was not precipitated by Protein A alone (lane 2, bottom panel). Reciprocal coimmunoprecipitations were performed using anti-HA antibodies. Comigration of ESC with immunoglobulins precludes detection of HA-ESC in these samples (25). Nevertheless, E(Z) and PCL were coimmunoprecipitated by anti-HA (lane 6, top and middle panels, respectively). Neither protein was precipitated by Protein G alone (lane 5). Although the anti-HA coimmunoprecipitated E(Z) and PCL signals appear to be approximately equal, it should be noted that lanes 5 and 6 of the middle panel were derived from x-ray film exposed to the Western blot for 2 h, whereas a 5-min exposure of the same Western blot produced the signals in lanes 1-4 of that panel. All lanes in the top and bottom panels were derived from single exposures of the respective Western blots. This suggests that a relatively lower percentage of total PCL protein (compared with E(Z)) is coimmu-noprecipitated by anti-HA. To rule out possible artifactual explanations for these results, such as gel loading errors or unequal transfer of proteins to the immunoblots, the blots in the top and middle panels were stripped and re-probed with the reciprocal antibodies. When the blot originally probed with anti-PCL (middle panel) was re-probed with anti-E(Z), results identical to those in the top panel were obtained (data not shown). Likewise, when the blot originally probed with anti-E(Z) (top panel) was re-probed with anti-PCL, the PCL signal from the anti-HA immunoprecipitated sample was again weak relative to the signals in other lanes (data not shown).
The size(s) of native PCL-containing complexes were determined by gel filtration chromatography. Nuclear extract from 0-to 24-h HA-esc embryos was size-fractionated on a Superose 6 column, 0.5-ml fractions were collected, and aliquots from even-numbered fractions were subjected to Western blot analysis. As previously described, E(Z) and HA-ESC cofractionate within a complex of ϳ600 kDa, which is distinct from the higher molecular mass (2 MDa) PRC1 complex (13, 14; and 5. Coimmunoprecipitation of E(Z), PCL, and HA-ESC proteins from Drosophila embryo extracts. Proteins were immunoprecipitated from 300 g of HA-esc embryo extracts with anti-E(Z), anti-PCL, or anti-HA antibodies, as indicated. Equal amounts of immunoprecipitated proteins (one-third of each sample) were run separately on 8% SDS gels, transferred to nitrocellulose filters, and incubated with (top) anti-E(Z), (middle) anti-PCL, or (bottom) anti-HA antibodies. Lanes: 1, 30 g total embryo extract; 2, mock immunoprecipitation by protein A-Sepharose without antibody; 3, immunoprecipitation with anti-E(Z) antibody; 4, immunoprecipitation with anti-PCL antibody; 5, mock immunoprecipitation by protein G-Sepharose without antibody; 6, immunoprecipitation with anti-HA antibody. of the PRC1 complex and are consistent with inclusion of PCL in at least a subset of ESC⅐E(Z) complexes.
The Interaction between PCL and E(Z) Is Conserved in Their Human Homologs-Many interactions that have been identified between Drosophila PcG members have also been observed with their mammalian counterparts. For example, two human homologs of E(Z), EZH1 and EZH2, both interact with a human ESC homolog, EED (25,32). It was therefore of interest to determine whether the interaction between PCL and E(Z) was conserved in their human counterparts. Both EZH1 and EZH2 show a high level of sequence similarity in four domains spanning the open reading frame (23). One potential PCL homolog, PHF1, shows 34% identity over the conserved region (22). A second human homolog was identified using the BLAST program (AJ010014, available at www.ncbi.nlm.nih.gov (33)), here termed hMTF2, because it displays a high level of sequence identity (96%) with MTF2, the murine homolog of PCL, over the central region of the protein (22). hMTF2 displays 37% identity (57% similarity) to PCL over the conserved region (Fig.  7A).
To determine whether the interaction between PCL and E(Z) is conserved in their mammalian counterparts, the region spanning the two PHD finger domains of PHF1 and hMTF2 were cloned into pEG202 to generate LexA-PHF1 and LexA-hMTF2, respectively. Full-length AD-EZH1 and AD-EZH2 were also generated. LexA-hMTF2 alone was able to strongly activate reporter gene expression in yeast strains containing as few as two LexA binding sites. To overcome the problem of auto-activation, the reciprocal constructs, LexA-EZH1, LexA-EZH2, and AD-hMTF2, were generated. Both LexA-EZH1 and LexA-EZH2 were tested for auto-activation in EGY48. LexA-EZH1 did not self-activate the reporter gene, whereas LexA-EZH2 did. However, when transformed into EGY191 (which carries only 2 LexA binding sites), activation by LexA-EZH2 did not occur.
Because LexA-PHF1 alone did not activate, it was tested for an interaction with AD-EZH1 and AD-EZH2, whereas AD-hMTF2 was tested for an interaction with LexA-EZH1 and LexA-EZH2. An interaction was observed between LexA-PHF1 and AD-EZH2 (Fig. 7B). No interaction was observed with any other combination (Fig. 7B). Western analysis using anti-LexA antiserum and anti-HA antisera confirmed the expression of all fusion proteins (data not shown).
PHD Finger Domains of trxG Proteins Appear Not to Interact with E(Z)-There are several lines of evidence suggesting that E(Z) is a member of both the PcG and trxG (see "Discussion"). Three trxG proteins, TRX, Abnormal, Small or Homeotic discs 1 (ASH1) and Abnormal, Small or Homeotic discs 2 (ASH2), contain PHD finger motifs (17)(18)(19). The ability of E(Z) to bind to the PHD domain of PCL and the presence of PHD finger motifs in these trxG proteins, raised the possibility that the role of E(Z) in trxG-dependent transcriptional activation might involve interaction with the PHD finger domains of one or more of these trxG proteins. However, in reciprocal yeast two-hybrid assays, none of these trxG PHD finger domains were found to interact with E(Z) (data not shown). The absence of interaction between these trxG PHD finger domains and E(Z) further demonstrates the specificity of the interaction between E(Z) and the PHD finger domains of PCL. DISCUSSION There is now overwhelming evidence that PcG proteins act in multimeric complexes to repress transcription. Detailed elucidation of PcG-mediated repression therefore requires characterization of the interactions that establish and maintain these complexes. Using both yeast two-hybrid and in vitro binding assays, direct interaction was observed between the conserved PHD finger-containing region of PCL and E(Z). Deletion analysis and site-directed mutagenesis demonstrated that both PCL PHD fingers, PHDF1 and PHDF2, are important in mediating the interaction of PCL and E(Z), because mutations in conserved residues of either finger abolishes the interaction. Recently, adjacent PHD fingers and the bromodomain of KAP-1 were shown to function in a cooperative manner to mediate direct interaction with an isoform of Mi-2␣ (20). The PHD fingers domain of myeloid lymphoid leukemia protein has been shown to both homodimerize and bind to Cyp33 (21). Our data demonstrate that a PHD finger domain can function as an independent protein interaction domain. Two separate E(Z) regions bind to the PCL PHD fingers in vitro. One, PBD1, is highly conserved in mammalian homologs; the other, PBD2, is not. Each region is capable of independently binding to PCL in vitro. However, it is not known whether this is also true in vivo.
Coimmunoprecipitation of PCL with E(Z) from embryonic extracts demonstrates in vivo association of PCL and E(Z). PCL also coimmunoprecipitates with HA-ESC, another component of E(Z) complexes. Gel filtration analysis of native protein complexes from embryo nuclear extracts show partial cofractionation of PCL with E(Z) and ESC. Cumulatively, our data are consistent with inclusion of PCL in at least a subset of ESC⅐E(Z) complexes. However, it is also possible that PCL may be a component of a distinct ϳ600-kDa complex and that PCL may mediate interaction between this complex and the ESC⅐E(Z) complex.
Physical interaction of PCL with, and possible inclusion in, ESC⅐E(Z) complexes in Drosophila embryos is consistent with genetic interactions of Pcl and E(z) with esc. A 50% reduction in zygotic dosage of either Pcl ϩ or E(z) ϩ significantly enhances the mutant phenotypes of embryos produced by esc Ϫ females. Among PcG loci, Pcl is second only to E(z) in the strength of enhancement of the esc maternal effect embryonic phenotype (34). However, if ESC⅐E(Z) complexes required PCL activity for all embryonic functions, one would expect mutant alleles of Pcl, E(z), and esc to produce very similar, if not identical, phenotypes. This is not the case. Embryos lacking maternally and zygotically produced E(Z) or ESC do exhibit essentially identical extreme homeotic phenotypes in which all thoracic and abdominal segments are transformed into copies of the eighth abdominal segment (35)(36)(37). In contrast, embryos lacking both maternal and zygotic Pcl activity display relatively weak posteriorly directed cuticular transformations; abdominal segments are slightly more posterior-like in appearance, but thoracic segments are essentially unaffected (38). Indeed, the Pcl mutant embryonic cuticular transformations are relatively weak with respect to most other PcG mutations. For example, embryos that lack both maternal and zygotic Pc activity also display a severe homeotic phenotype similar to the E(z) and esc FIG. 6. Gel filtration analysis of E(Z), HA-ESC, and PCL proteins in Drosophila nuclear extracts. Nuclear extract from HA-esc embryos was fractionated by Superose 6 chromatography. 20-l aliquots of even-numbered 500-l fractions were run on separate 8% SDS gels, transferred to nitrocellulose filters, and incubated with anti-E(Z) antibodies (top), anti-HA antibodies (middle), or anti-PCL antibodies (bottom). Elution positions of molecular mass standards are indicated with arrows above the numbers of the appropriate fractions.
phenotypes described above (39). Therefore, lack of individual subunits of either the ESC⅐E(Z) complex or the PRC1 complex (e.g. PC) can effectively disrupt PcG-dependent repression of homeotic genes in embryos.
Why is it, then, that the lack of PCL in Drosophila embryos produces such mild homeotic phenotypes? It should be remembered that the cuticular phenotypes described above result from misexpression of homeotic genes in epidermal cells. Soto et al. (40) presented evidence for tissue-specific regulation of homeotic genes by some members of the PcG. Embryos lacking maternal and zygotic Pcl show extensive ectopic expression of the homeotic gene AbdB in the central nervous system and visceral mesoderm but only moderate ectopic expression in a subset of epidermal cells. Embryos that are homozygous for a null Pcl allele, but which were derived from heterozygous mothers, exhibit similar tissue-specific derepression of the homeotic gene abdA (41). In striking contrast, embryos lacking maternal and zygotic esc or E(z) activities show extensive ectopic expression of abdA and AbdB in all tissues (40,41). Moderate misexpression of abdA and AbdB in epidermal cells is consistent with the relatively weak posteriorly directed transformations of embryonic cuticles in Pcl mutants.
Collectively, the embryonic phenotypes produced by Pcl, E(z), and esc mutant alleles, the effects of these mutations on misregulation of homeotic genes, and our coimmunoprecipitation and gel filtration data suggest that PCL may be necessary for ESC⅐E(Z) complex activity only in a subset of embryonic tissues (i.e. central nervous system and visceral mesoderm cells and, possibly, a small subset of epidermal cells in Drosophila embryos, but not in the vast majority of epidermal cells). Alternatively, PCL may be required in a temporal-specific manner for ESC⅐E(Z) complex activity, and the tissue-specific misregulation of homeotic genes in Pcl mutants may be due to differences in the times when genes in different tissues require PcG-dependent repression. These two models are not necessarily mutually exclusive.
The interaction between Drosophila PCL and E(Z) proteins was found to be conserved through evolution, as PHF1 and EZH2, human homologs of PCL and E(Z), respectively, interact in yeast two-hybrid assays. Failure to detect an interaction between other combinations of human PCL and E(Z) homologs, PHF1 and EZH1, hMTF2 and EZH1, or hMTF2 and EZH2, could indicate an in vivo specificity between PHF1 and EZH2 or improper folding of some of the LexA and AD fusion proteins. If folding is not the explanation, hMTF2 may interact in vivo with an as yet unidentified E(Z) homolog. Another possibility is that some PCL PcG functions may not require interaction with E(Z). If so, hMTF2 may be specific for those functions. Alternatively, hMTF2 may have a PcG-independent function that does not involve interaction with an E(Z) homolog or other PcG proteins. The specificity of the interaction of PHF1 with EZH2, but not with EZH1, also suggests distinct functions for these E(Z) homologs.
The conservation of the PCL⅐E(Z) interaction in humans strongly suggests that this interaction plays an important role in PcG-dependent silencing in both mammals and Drosophila. Recent experiments have demonstrated that mammalian PcG proteins function in manners analogous to that of their Drosophila counterparts. Mice mutant for mammalian PcG homologs eed (the esc homolog (42)) or bmi-I or mel-18 (the Psc and Su(z)2 homologs (43, 44)) display posterior homeotic transformations of axial skeletal structures (42,45,46). This phenotype is analogous to the phenotype of the corresponding Drosophila mutants and reflects derepression of homeotic genes along the anterior/posterior axis. More striking is the observation that M33, the murine homolog of Drosophila PC, partially rescues Pc mutant phenotypes (47). Similarly, overexpression of either Drosophila E(Z) or human EZH2 in Drosophila enhances the phenomenon of position effect variegation (23), again demonstrating a conservation of function between the mammalian and Drosophila genes. Given the functional similarity of several components of the PcG in both mammals and Drosophila, it is likely that similar biochemical mechanisms exist for the heritable repression of homeotic genes throughout development.
Ample evidence classifies E(z) as a member of the PcG. Male flies homozygous for a hypomorphic E(z) allele display extra sex combs on the second and third legs, a characteristic phenotype of PcG mutants (48). This phenotype is caused by ectopic expression of the homeotic gene Sex combs reduced (Scr) in second and third thoracic leg imaginal discs (37). Larvae with reduced E(z) activity also exhibit ectopic expression of the FIG. 7. Interactions between the human homologs of PCL and E(Z). A, protein alignment of the region of similarity between PHF1 (GenBank TM accession number AF029678), hMTF2 (GenBank TM accession number AJ010014), and PCL (GenBank TM accession number Q24459). Over this region PHF1 and hMTF2 share 58% identity (71% similarity), PHF1 and PCL share 36% identity (48% similarity), and hMTF2 and PCL share 37% identity (57% similarity). Conserved residues are black, and similar residues are gray. B, growth assay with yeast containing human homologs of PCL and E(Z). The left panel is a schematic of the arrangement of yeast on the middle and right panels. Yeast were streaked onto leucine-deficient medium, which contained either galactose (GAL, right panel) or glucose (GLU, middle panel).
homeotic gene Ultrabithorax (Ubx) in wing imaginal discs, and ectopic expression of Scr and Ubx in the larval central nervous system (37). Embryos lacking maternally contributed E(z) activity display posteriorly directed homeotic transformations (36,37) and show derepression of homeotic genes (37,41). E(z) mutant alleles enhance the homeotic phenotypes produced by mutant alleles of some other PcG genes. For example, flies heterozygous for mutations in both E(z) and ph display a more severe extra sex combs phenotype when compared with single mutants alone (49). Both decreased and increased E(z) ϩ dosage enhance the homeotic phenotypes of embryos derived from homozygous esc females. Furthermore, E(Z) protein directly interacts with two other PcG proteins, ESC and PCL, and exists as a component of a multimeric complex in Drosophila embryos that includes ESC, p55, RPD3, and possibly PCL (13,14,25,31;Figs. 5 and 6).
However, LaJeunesse and Shearn (50) provided evidence that E(Z) also is involved in trxG-dependent activation of homeotic genes. Double-heterozygous combinations of recessive loss of function E(z) and ash1 alleles result in homeotic transformation phenotypes similar to that observed in double-heterozygous combinations of recessive loss of function alleles in trx and ash1 (50,51). Loss of expression of the homeotic genes Scr, Ubx, and Antennapedia (Antp) was also observed in the thoracic imaginal discs of larvae hemizygous for a null E(z) allele. This phenotype is also seen in the imaginal discs of larvae mutant for null ash1 alleles (52). Gildea et al. (53) have provided evidence that several additional PcG and trxG genes may also contribute to both repression and activation. The presence of PHD finger domains in PCL as well as the trxG proteins TRX, ASH1, and ASH2 and the observation that the PHD finger domains of PCL mediate binding to E(Z), raised the possibility that the trxG function of E(Z) might involve binding to the PHD finger domains of TRX, ASH1, and/or ASH2, perhaps in competition with binding to the PHD finger domains of PCL. However, no interaction was observed between the PHD finger domains of these trxG proteins and E(Z) in yeast twohybrid assays. An as yet unidentified interaction between E(Z) and a trxG protein may account for its role in trxG-dependent activation. It is also likely that the PHD domains of these trxG proteins mediate interactions with other proteins, such as Cyp33 (21), and others that have not yet been identified.