The Drosophila Sex Comb on Midleg (SCM) protein is a transcriptional repressor of the Polycomb group (PcG). Although genetic studies establish SCM as a crucial PcG member, its molecular role is not known. To investigate how SCM might link to PcG complexes, we analyzed the in vivo role of a conserved protein interaction module, the SPM domain. This domain is found in SCM and in another PcG protein, Polyhomeotic (PH), which is a core component of Polycomb repressive complex 1 (PRC1). SCM-PH interactions in vitro are mediated by their respective SPM domains. Yeast two-hybrid and in vitro binding assays were used to isolate and characterize >30 missense mutations in the SPM domain of SCM. Genetic rescue assays showed that SCM repressor function in vivo is disrupted by mutations that impair SPM domain interactions in vitro. Furthermore, overexpression of an isolated, wild-type SPM domain produced PcG loss-of-function phenotypes in flies. Coassembly of SCM with a reconstituted PRC1 core complex shows that SCM can partner with PRC1. However, gel filtration chromatography showed that the bulk of SCM is biochemically separable from PH in embryo nuclear extracts. These results suggest that SCM, although not a core component of PRC1, interacts and functions with PRC1 in gene silencing.
THE Polycomb group (PcG) proteins of Drosophila are transcriptional repressors that maintain gene silencing during development (Brock and van Lohuizen 2001; Francis and Kingston 2001; Simon and Tamkun 2002 for reviews). The most well-characterized targets of PcG repression are the fly Hox genes. Other developmental regulators such as engrailed and hedgehog are also under PcG control (Moazed and O'Farrell 1992; Maurange and Paro 2002) and many additional targets are inferred from the ∼100 sites of PcG protein localization on chromosomes (Franke et al. 1992; Rastelli et al. 1993) and from genome-wide prediction of Polycomb response elements (Ringrose et al. 2003). PcG proteins organize local chromatin to maintain repression after an initial decision to turn a gene off is made by other factors. This ensures that expression patterns of Hox genes and other target genes are maintained during development. PcG silencing can be propagated through many cell divisions and for long periods of developmental time, thereby providing a key model system for studying the memory of transcriptional states.
To address mechanisms of PcG repression, much recent effort has focused on defining subunit compositions and in vitro activities of PcG protein complexes. Two biochemically distinct PcG complexes from Drosophila embryos have been characterized in detail: the ESC-E(Z) complex and the Polycomb repressive complex 1 (PRC1; Shao et al. 1999; Ng et al. 2000; Francis et al. 2001; Saurin et al. 2001; Tie et al. 2001; Czermin et al. 2002; Muller et al. 2002). Mammalian complexes highly related to these fly PcG complexes in compositions and activities have also been described (Cao et al. 2002; Kuzmichev et al. 2002; Levine et al. 2002). The ESC-E(Z) complex contains the PcG proteins Extra Sex Combs (ESC), Enhancer of Zeste [E(Z)], and Suppressor of Zeste 12 [SU(Z)12] plus NURF-55 (Tie et al. 2001; Czermin et al. 2002; Muller et al. 2002), which is also found in other chromatin-modifying complexes (Tyler et al. 1996; Martinez-Balbas et al. 1998). A recombinant complex assembled from these four subunits has histone methyltransferase activity with primary specificity for lysine-27 of histone H3 (Muller et al. 2002).
PRC1 contains the PcG proteins Polycomb (PC), Polyhomeotic (PH), Posterior Sex Combs (PSC), and dRING1 plus additional polypeptides (Shao et al. 1999; Saurin et al. 2001). A recombinant complex containing just these four proteins, the Polycomb core complex (PCC), can block remodeling of nucleosome arrays by human SWI/SNF in vitro (Francis et al. 2001). Binding studies indicate that the block likely occurs by occluding subsequent binding of the SWI/SNF complex. PRC1 may also restrict template access of other factors required for transcription initiation (King et al. 2002).
Genetic loss of subunits belonging to either of these PcG complexes leads to severe defects in Hox gene repression (Struhl and Akam 1985; Jones and Gelbart 1990; Phillips and Shearn 1990; McKeon and Brock 1991; Simon et al. 1992; Soto et al. 1995; Birve et al. 2001). Thus, the repression mechanism requires key contributions from both complexes. In addition, there is evidence that ESC-E(Z) function is a prerequisite for PRC1 recruitment to target genes; chromosome immunostaining and chromatin immunoprecipitation experiments show that inactivation of E(Z) dislodges PRC1 components from target loci (Rastelli et al. 1993; Cao et al. 2002). Taken together, the in vitro and in vivo results suggest a stepwise model where ESC-E(Z) methylates local chromatin, which then helps to attract PRC1 through preferential binding to methylated H3 tails (Cao et al. 2002; Simon 2003). Although binding studies using an isolated PRC1 component, PC, and methylated tail peptides support this view (Cao et al. 2002; Czermin et al. 2002; Fischle et al. 2003; Min et al. 2003), the effect of site-specific H3 methylation upon binding of intact PRC1 to oligonucleosomes has yet to be tested.
The recent progress in studies on in vitro activities has focused attention on the ESC-E(Z) and PRC1 complexes. However, genetic data indicate that additional components, which are also essential for PcG repression, need to be incorporated into the molecular models. Of the ∼15 PcG proteins identified by mutant alleles in Drosophila, only 7 are accounted for as core members of the ESC-E(Z) or PRC1 complexes. The remaining PcG repressors could interact functionally with either of the known complexes or they could compose PcG complexes yet to be characterized. In either case, an understanding of PcG mechanisms requires deciphering functions of these additional group members.
The Sex Comb on Midleg (SCM) protein is one of the PcG repressors whose molecular role has not yet been defined. SCM is a critical PcG repressor since its complete loss from embryos produces Hox gene misregulation and phenotypes that are as severe as for loss of ESC, E(Z), PC, or PH (Struhl 1981; Breen and Duncan 1986; Jones and Gelbart 1990; Bornemann et al. 1998). The most revealing hints about SCM function so far come from in vitro data that link SCM to PRC1. Two-hybrid analyses and in vitro binding assays showed that SCM can associate with PH, a core component of PRC1 (Peterson et al. 1997). Furthermore, SCM is co-enriched with PRC1 isolated from fly embryos (Shao et al. 1999). However, since SCM association with purified PRC1 is substoichiometric (Saurin et al. 2001), its role appears distinct from the PRC1 core proteins PC, PH, PSC, and dRING1.
Although SCM lacks known DNA-binding or catalytic domains, three separate functional domains have been inferred from database comparisons (Bornemann et al. 1996). First, there is a pair of N-terminal zinc fingers that are distinct from classical DNA-binding zinc fingers. Second, there are two tandem 100-amino-acid repeats, called mbt repeats since they are also found in the fly tumor suppressor encoded by the l(3)mbt [lethal(3)malignant brain tumor] gene (Wismar et al. 1995). Finally, there is a C-terminal homology domain of 65 amino acids, called the SPM domain, that is 41% identical to a C-terminal domain in the fly PcG protein PH. Among these three SCM domains, biochemical function is known only for the SPM domain, which is a self-binding protein interaction module; the SPM domain from SCM can mediate SCM-PH and SCM-SCM contact in vitro (Peterson et al. 1997). The high conservation of mbt repeats (67% identical) and SPM domain (61% identical) in mammalian SCM homologs (Montini et al. 1999; Tomotsune et al. 1999) implies that they play key roles in SCM repressor function. In agreement with this, analysis of Scm mutant alleles revealed four loss-of-function mutations that map within the mbt repeats (Bornemann et al. 1998). However, lack of SPM domain missense alleles has hampered analysis of in vivo contributions of SPM domain interactions.
In this study, we used several approaches to address the in vivo role of the SPM domain of SCM in PcG repression. A collection of SPM domain missense mutations was isolated through a two-hybrid screen, tested for effects upon protein interactions in vitro, and then tested for SCM repressor function in vivo by transformation rescue. In addition, an isolated SPM domain was overexpressed in vivo to determine if this behaved as a dominant negative that compromises PcG repression. The results indicate that the SPM domain mediates protein interactions critical for PcG repression in vivo. On the basis of biochemical tests, we suggest that this reflects functional partnership of SCM with PRC1.
MATERIALS AND METHODS
Construction of SPM domain mutant libraries:
The SPM domain coding region of SCM was amplified with Taq polymerase (Promega, Madison, WI) to generate randomly mutated collections of products. pMinSPM (Peterson et al. 1997) or a derivative with an engineered BglII site at SCM codon 800 was used as PCR template. Amplification was performed using pBS primers flanking the SPM domain insert, and products were digested with EcoRI and NotI, inserted into yeast two-hybrid bait vectors pEG202 or pMW103 (Gyuris et al. 1993; Watson et al. 1996), and transformed into Escherichia coli. A library with high mutation frequency (multiple substitutions per SPM domain) was generated by error-prone PCR using the conditions described (Vogel and Das 1994), except that DMSO and β-mercaptoethanol were omitted and dATP concentration was varied (0.4–0.6 mm) with dCTP, dGTP, and dTTP each at 1 mm. Libraries with lower mutation frequencies (less than one nucleotide substitution per kilobase) were generated using standard PCR conditions (1× Promega Taq buffer, 1.5 mm MgCl2, and 0.2 mm dNTPs).
Isolation and identification of SPM domain mutations:
Mutant bait plasmid libraries were transformed into yeast strains carrying a lacZ reporter with LexA-binding sites (pMW108 or pMW110; Watson et al. 1996) plus pAD-SCM, which encodes full-length SCM fused to an activation domain (Peterson et al. 1997). In some cases, individual colonies were tested directly by streaking onto X-Gal indicator plates. In other cases, transformants were pooled and stored at −70°. The frozen cells were then replated on selective media at a density of 500 colonies/plate and tested by replica plating onto X-Gal indicator plates.
White and light blue yeast isolates were selected as potential loss-of-binding mutants. Dark blue colonies were selected to identify mutations that retain binding. Mutant SPM domain plasmids were recovered by isolating DNA from yeast cultures as described (Golemis et al. 1994) and electroporating into E. coli. pMW103-based plasmids were recovered in strain K802 by growth on rich media supplemented with 20 μg/ml kanamycin. For pEG202-based libraries, bait plasmids were recovered by selection for histidine prototrophy after transformation into strain KC8 (Golemis et al. 1994). In all cases, plasmid DNAs were purified and tested by PCR and/or restriction digestion to verify presence of the SPM domain insert.
Purified bait plasmid DNA for each isolated mutant was retransformed into a strain containing the pMW108 lacZ reporter plus pAD-SCM to verify that the two-hybrid interaction phenotype was plasmid borne. Mutant SPM domain plasmids were separately transformed into a strain with lacZ reporter plus pAD-PH (Peterson et al. 1997) to assess interaction with full-length PH (Figure 1A). The SPM domains of the isolated clones were sequenced using Scm and vector-based primers to identify the mutations.
In vitro protein interaction assays:
Mutant SPM domains were obtained on EcoRI-NotI fragments, which were inserted into pGEX4T-1 (Pharmacia) to create GST fusions and into pET-28a (Novagen) for expression by in vitro transcription/translation (TNT kit, Promega). Purification of GST fusion proteins, radiolabeling of proteins, and glutathione S-transferase (GST) pull-down assays were performed as described (Peterson et al. 1997). GST-fusion proteins were adjusted to ∼1 μg/μl prior to use in binding assays.
Scm rescue constructs:
A germline transformation construct, pCas-FSCM, that expresses FLAG-tagged SCM from its normal genomic promoter was generated as follows. A double-stranded oligonucleotide containing a consensus translation start site, initiator methionine codon, and the FLAG coding sequence was created by annealing the oligonucleotides 5′-AAT TTC GCG ACA ACA TGG ATT ACA AGG ATG ATG ACG ACA AGG-3′ and 5′-AAT TCC TTG TCG TCA TCA TCC TTG TAA TCC ATG TTG TCG CGA-3′. This oligonucleotide was inserted into the EcoRI site immediately preceding the start codon in the full-length Scm cDNA pRS3.1 (Peterson et al. 1997). The FLAG-SCM coding region was then isolated as a 2.9-kb NruI fragment and inserted downstream of the 2-kb Scm promoter in the genomic clone pProBX by ligation to an MscI site within the Scm 5′-untranslated region. A 5.0-kb NheI-NotI fragment containing 2 kb of Scm upstream DNA fused to the FLAG-SCM coding region was inserted into XbaI-NotI-cut pCas-poly(A) to create the pCas-FSCM rescue construct. pCas-poly(A) is the pCasper4 transformation vector with the SV40 poly(A) signal added on to a 0.5-kb fragment from pUAST (Brand and Perrimon 1993). Mutant SPM domains were shuttled into the rescue vector in place of the wild-type SPM domain using an engineered BglII site at Scm codon 800 and the 3′-flanking polylinker NotI site. pCas-FSCMΔSPM was constructed by inserting an XbaI-NotI Scm cDNA fragment that lacks codons 798–877 (Peterson et al. 1997) in place of the corresponding wild-type XbaI-NotI fragment.
Germline transformation and Scm genetic rescue tests:
Germline transformation was performed as described (Rubin and Spradling 1982) using y Df(1)w67c23 as a host strain. Transformants were recognized by w+ eye color, multiple independent lines per construct were established, and inserts were mapped to chromosomes X, 2, or 3. Lines used for Western analysis were homozygous for the insert chromosome. Lines used for Scm rescue tests were homozygous for the transgene insert or contained the transgene on a recessive lethal second chromosome balanced over CyO. Some second-chromosome FLAG-SCM inserts were obtained by mobilizing an X-linked transgene with a Δ2-3 P transposase source (Robertson et al. 1988).
Genetic tests for Scm transgene function were performed essentially as described (Bornemann et al. 1996) using the null ScmH1/ScmXF24 and hypomorphic ScmH1/ScmET50 allelic combinations. Genetic rescue was scored as viability of animals bearing either of these lethal Scm genotypes plus a single copy of an Scm transgene. For the wild-type FLAG-SCM transgene, six lines with inserts on either the X or the second chromosomes were tested for rescue. For each mutant FLAG-SCM transgene, two to six independent lines with second-chromosome insertions were tested for rescue (Table 2). Rescue by two copies of the FLAG-SCM-M43Δ trangene was also tested by mating w; P[w+, F-SCM-M43Δ]; Scm− e/TM3 Sb flies using all combinations of the H1, XF24, and ET50 alleles. Rescue was scored by the presence or absence of non-TM3 flies.
Protein extracts and Western blots:
Immunodetection of FLAG-SCM was performed using crude embryo lysates (Figure 3C) or nuclear extracts from 0- to 24-hr embryos (Figure 3B) prepared as described (Ng et al. 2000). Mouse monoclonal anti-FLAG M5 antibody (Sigma, St. Louis) and affinity-purified rabbit anti-SCM antibody (Bornemann et al. 1998) were used at 1:1000. Immunodetection of HA-SPM in crude embryo extracts (Figure 4B) was performed using mouse monoclonal HA.11 (CoVance) at 1:1000. Anti-SCM and anti-PH were affinity-purified rabbit polyclonal antibodies (Bornemann et al. 1998; Ng et al. 2000) used at 1:1000. Secondary antibodies were horseradish peroxidase (HRP)-conjugated goat anti-mouse antibody (Jackson Laboratories) used at 1:10,000, HRP-conjugated goat anti-rabbit antibody (Jackson Laboratories) used at 1:10,000, or the same antibody from Bio-Rad (Richmond, CA) used at 1:5000. Signals on Western blots were developed using an enhanced chemiluminescence detection kit (Amersham-Pharmacia).
SPM domain overexpression:
pUAST-HN was engineered by sequential insertion of oligonucleotides encoding an HA epitope tag and a nuclear localization signal (NLS) into the overexpression vector pUAST (Brand and Perrimon 1993). The upstream activator sequence (UAS)-SPM construct was made by inserting an EcoRI-NotI fragment encoding the C-terminal 77 amino acids of SCM into pUAST-HN. The UAS-SCM construct was made by inserting a full-length Scm cDNA as an EcoRI-NotI fragment into pUAST-HA, which is identical to pUAST-HN but lacks the NLS. Overexpression tests were performed by crossing GAL4 driver flies to flies bearing the UAS-SPM construct and scoring the GAL4 driver/+; UAS-SPM/+ progeny for survival and visible phenotypes. A UAS-SPM insertion on chromosome 2 (line 29-23) was used primarily for analysis and documentation of overexpression phenotypes (Figure 5), although independent insertions gave similar results.
Immunoprecipitations and gel filtration analysis:
Immunoprecipitations were performed on soluble extracts prepared from 0- to 20-hr embryos as described (Dingwall et al. 1995), except that the extraction buffer was at pH 7.5 and centrifugation was performed using either a Beckman Ti80 or a Beckman SW50 rotor at 30,000 rpm. A total of 1 mg of total soluble protein was incubated for 1 hr at 4° with 12.5 μl αSCM (Bornemann et al. 1998) or 25 μl αPH (Ng et al. 2000) antibodies in 150 μl IP buffer [10 mm HEPES (pH 7.5), 1 mm EDTA, 10% glycerol, 50 mm NaCl, plus protease inhibitors as for embryo extraction buffer (Dingwall et al. 1995)]. Immunoprecipitates were recovered with 50 μl of Protein A beads (Boehringer Mannheim, Indianapolis) and washed five times with IP buffer, and proteins were eluted with SDS sample buffer. A total of 15% of the bound material, and 3.3% of the unbound material, was loaded per lane for Western blot analysis. Superose 6 gel filtration fractionation was performed on nuclear extracts from 0- to 24-hr embryos as described (Ng et al. 2000).
Assembly of SCM with PCC:
PCC-SCM was prepared as described for PCC (Francis et al. 2001) except that baculovirus encoding SCM was coinfected with the other four subunits. Complexes were purified via a Flag epitope tag on either PSC or PH; as reported previously, no differences were observed between PCC prepared with Flag-PSC or Flag-PH.
For glycerol gradient sedimentation, 10 μg of PCC-SCM was loaded onto a 10–40% glycerol gradient in BC300 buffer (Francis et al. 2001) plus 0.1% NP40 and centrifuged for 6 hr at 35,000 rpm (80,000 × g) in a TLS-55 rotor in a TL-100 ultracentrifuge (Beckman). Fractions of ∼200 μl were collected from the bottom of the gradient and 10 μl of each fraction was analyzed by SDS-PAGE followed by silver staining.
Isolation of SPM domain missense mutations:
Error-prone PCR was used to introduce random mutations into a 200-bp fragment encoding the SPM domain from SCM. Amplified products were cloned into yeast two-hybrid bait vectors to create libraries of mutated SPM domains fused to LexA. Libraries were constructed that averaged either <1 amino acid change per 65-amino-acid domain or multiple substitutions per domain (see materials and methods). The bait libraries were introduced into yeast strains carrying a lacZ reporter and fusions of either SCM or PH to an activation domain (AD). We have previously shown that there is robust interaction of wild-type LexA-SPM with either full-length AD-SCM or AD-PH in the two-hybrid system (Peterson et al. 1997). Single yeast colonies were selected on the basis of lacZ expression levels and corresponding bait plasmids bearing mutant SPM domains were recovered. Plasmids were then retransformed separately into AD-SCM and AD-PH strains to determine the effects of mutations on interaction with either PcG protein in parallel (Table 1). Bait DNAs were sequenced to identify the SPM domain mutations.
To isolate mutations that disrupt SPM domain binding, we selected light blue and white colonies under conditions in which wild-type LexA-SPM produces dark blue colonies. Figure 1A shows qualitative comparisons of β-galactosidase levels in several of these mutants. The binding-defective mutants were isolated from libraries with a mutation frequency of less than one change per domain. Thus, in most cases, differences in binding behavior could be directly attributed to single amino acid substitutions. We identified 18 missense mutations that reduce or eliminate binding in the two-hybrid system (Figure 1B, changes in red). Multiple substitutions were independently recovered at residues F24, G31, and L48, indicating tight requirements for amino acids at these positions. Table 1 lists the sequenced mutations and their interaction capabilities. Collectively, these mutations identify amino acid positions that are critical for the binding properties or fold of the SPM domain.
Although the majority of the PCR-induced errors created amino acid substitutions, several deletions were also recovered (Figure 1B and Table 1). One deletion created a stop codon immediately after N58. Western blot analysis confirmed that this truncated fusion protein is expressed in yeast (data not shown). This truncation shows that the extreme C terminus of the SPM domain is required for protein interactions. Another deletion, M43Δ, removes a single ATG codon encoding one of three consecutive methionine residues. The M43Δ mutation was also notable for its differential effect upon binding to SCM vs. PH. Whereas the majority of the binding-defective mutations disrupted SCM and PH interactions to similar degrees (Figure 1A and Table 1), the M43Δ domain retained little SCM binding but its PH binding was only moderately reduced (see below).
A distinct set of mutations that do not perturb SPM domain binding was also identified. These amino acid changes (Figure 1B, shown in blue) were readily identified by sequencing plasmids from dark blue yeast colonies isolated from heavily mutated SPM domain libraries. In contrast to the binding-defective mutations, most of these neutral mutations were identified in domains containing multiple substitutions (Table 1). For example, the triple mutant E11G/M41V/K55R shows binding comparable to wild type (Figure 1A, “triple”). Since the chance of recovering random compensatory mutations is remote, these changes represent independent amino acid substitutions that are tolerated while preserving binding activity. In several cases, positions with neutral mutations are also represented by loss-of-binding mutations. Residue 29, for example, can tolerate an isoleucine-to-valine substitution, but not a threonine substitution, suggesting that a hydrophobic residue is required at this position. Similarly, the K55E substitution interferes with binding, but the conservative K55R substitution is neutral.
In vitro binding properties of mutant SPM domains:
We wished to characterize the mutant domains for binding activity in vitro to identify a subset to be tested for function in vivo. To more directly assess binding properties of the mutant domains, we employed GST pull-down assays (Figure 2). The triple mutant E11G/M41V/K55R shows binding comparable to wild type (Figure 1A; data not shown) and serves as a pseudowild-type positive control. This triple-mutant SPM domain binds equally well to the wild-type SCM and PH versions of the domain (Figure 2A). In contrast, the G31D, M46R, and K55E mutations effectively eliminate binding of the domain to either GST-SCM or GST-PH. The M43Δ and I29T domains show little binding to GST-SCM but retain reduced binding to GST-PH. Binding was also reduced, but not eliminated, when the reciprocal test of GST-M43Δ interaction with radiolabeled wild-type SCM domain was performed (Figure 2B). The mutated domains were also tested in the GST pull-down assays for direct self-association. Although the triple-mutant domain bound to itself as well as to wild-type domains, all other mutant domains tested showed negligible self-binding (Figure 2A, rightmost lane).
In general, the in vitro binding results correlated well with the yeast two-hybrid results. For example, the M43Δ domain exhibited preferred binding for PH vs. SCM in both assay systems. However, some minor differences between the two assays were observed. For example, the I29T domain displayed residual PH binding not predicted by the two-hybrid assay. In summary, the GST pull-down results identify three missense mutations (G31D, M46R, and K55E) that are essentially null for binding activity and two mutations (M43Δ and I29T) that show partial activity.
Binding-defective mutations disrupt SCM function in vivo:
Function of mutant SCM proteins was tested in vivo using a genetic rescue assay. A genomic Scm transgene containing 2 kb of 5′-flanking DNA has been shown to fully rescue lethality of Scm null alleles (Bornemann et al. 1996). A new rescue construct was designed to track transgenic protein and to provide convenient insertion of variant SPM domains to be tested (Figure 3A). This Scm transgene contains a full-length Scm cDNA, with an N-terminal FLAG tag, driven by the 2-kb Scm genomic promoter fragment. Figure 3B shows Western blot analysis of wild-type and FLAG-SCM transgenic embryo extracts. The anti-FLAG Westerns revealed a single reacting species migrating at ∼100 kD, indicating stable production of the expected transgenic product. Anti-SCM Westerns (Figure 3B, bottom) detected a tightly spaced doublet with the upper of the two species identified as FLAG-SCM. The similar signal intensities indicate that transgenic SCM is expressed at comparable levels to endogenous SCM. Genetic rescue tests were performed to assess function of the FLAG-SCM protein. Flies trans-heterozygous for two Scm null alleles, H1 and XF24, and carrying one copy of the transgene survived to adulthood and were phenotypically normal (Table 2). Thus, transgenic FLAG-SCM provides full SCM activity.
Five of the binding-defective SPM domain mutations and the pseudowild-type triple mutant were inserted into the FLAG-SCM construct and tested for function in vivo. In addition to the missense mutants, simple removal of the entire SPM domain was tested using FLAG-SCM truncated after amino acid 797 (ΔSPM construct). For each construct, multiple independent transformant lines were tested for transgene expression and genetic rescue. Western blots on embryo extracts showed that the triple-mutant protein and the binding-defective K55E and M43Δ mutants were expressed at levels comparable to wild type (Figure 3C). The ΔSCM protein, detected at the expected slightly smaller molecular weight, was also expressed at or above the level of wild-type SCM. In contrast, the M46R and I29T proteins were detected at diminished levels and the G31D protein was absent or barely detectable (Figure 3C and data not shown).
The results of genetic rescue tests for the mutant transgenes are shown in Table 2. In each case, the ability of a single transgene copy to rescue viability of either a null Scm mutant (XF24/H1) or a hypomorphic Scm mutant (ET50/H1) was assessed. The ET50 allele produces normal levels of a partially functional SCM protein bearing a single-amino-acid substitution in an mbt repeat (Bornemann et al. 1998). Although Scm null mutants die at the end of embryogenesis, the ET50/H1 mutants survive to pupal stages. Thus, rescue of ET50/H1 mutants could provide sensitivity to detect mutant SCM proteins that retain partial function.
We found that SCM bearing the triple-mutant SPM domain, which retains both PH- and SCM-binding activity (Figure 2), was the sole mutant to provide Scm activity in vivo (Table 2). These rescued flies were fertile and phenotypically normal, except that one line produced rescued males with a weak extra sex combs phenotype. The remaining mutant transgenes failed to rescue, even in the test of the ET50/H1 hypomorph. Since the M46R, I29T, and G31D proteins were expressed at reduced levels, their failure to rescue could not be attributed solely to defects in SPM domain function. However, the genetic inactivities of the ΔSPM, K55E, and M43Δ proteins (Table 2), which are expressed at or above the level of the positive-control FLAG-SCM protein, prove that the SPM domain is critical for SCM function as a repressor in vivo. Among these, the M43Δ protein is notable because it failed to rescue despite retaining some PH-binding activity (Figure 2). To further assess possible partial function of M43Δ protein in vivo, the M43Δ transgene was also tested for rescue when supplied in two copies. Again no rescue was observed, even with the ET50/H1 hypomorph. This result indicates that the residual interaction ability of the M43Δ domain is not sufficient for SCM function in vivo. In summary, the rescue tests show that binding activity of mutated SPM domains in vitro correlates with their ability to function in PcG repression in vivo.
Overexpression of an isolated SPM domain causes loss of PcG repression:
If the SPM domain mediates PcG protein interactions in vivo, then overabundance of isolated SPM domains might disrupt PcG repression by titrating out productive interactions. Such a dominant-negative effect would produce canonical PcG loss-of-function phenotypes, such as extra sex combs, due to ectopic Hox gene expression. To test this idea, we used the GAL4-UAS inducible expression system (Brand and Perrimon 1993) to overexpress an isolated SPM domain in vivo from the construct shown in Figure 4A. This construct expresses an HA-tagged wild-type version of the SPM domain from SCM, encompassing amino acids 800–877. The developmental times and tissues of expression were controlled by crossing transformants bearing this UAS-SPM construct to various “driver” lines, which express GAL4 in defined tissues. Figure 4B shows that the ∼10-kD HA-tagged SPM domain is produced stably in embryos containing the UAS-SPM construct and a GAL4 driver. This Western blot shows expression from a single copy of UAS-SPM inserted on the second chromosome, which is the same insert used to generate examples of phenotypes shown below. Similar phenotypic results were obtained with additional UAS-SPM insertion lines.
Combinations of the UAS-SPM construct with several constitutive and imaginal disc GAL4 drivers caused pupal lethality, indicating that the overexpressed domain is biologically active. Additional GAL4 drivers expressed in imaginal tissues yielded viable adult progeny with specific homeotic defects that are hallmarks of PcG loss of function. The 71B driver (Brand and Perrimon 1993) produced mild crumpling of the wing (not shown) whereas the A9 and apterous drivers (Haerry et al. 1998; LaJeunesse et al. 1998) produced more dramatically reduced wings that resemble halteres (Figure 5B, D and E). A dpp-disc driver (Staehling-Hampton et al. 1994) produced male progeny with the canonical extra sex combs phenotype (Figure 5F), indicating homeotic transformation of the legs. Extra sex combs were also found in dissected pharate adult males from crosses performed with a patched driver (Speicher et al. 1994; data not shown). The paired driver (Manseau et al. 1997) produced partial antenna-to-leg transformations in females (compare Figure 5, G and H). Male progeny from this cross did not survive to adulthood. As expected since the GAL4 system is temperature sensitive (Brand and Perrimon 1993), phenotypes were generally more severe when tested at 25°–29° than at 18° (compare Figures 5A and 5B). These various homeotic phenotypes are commonly seen with traditional PcG loss-of-function alleles. Thus, overexpression of the isolated SPM domain causes loss of PcG repression in a wide array of adult tissues.
We interpret these phenotypes to mean that the SPM interaction module can effectively compete with endogenous SCM for binding to PcG partners in vivo. Since additional functional domains found in full-length SCM are missing, PcG complexes containing the isolated domain would be rendered nonfunctional. If this type of dominant-negative mechanism is operating, then we would expect phenotype severity to depend upon levels of full-length SCM. To address this, we tested how either increasing or decreasing full-length SCM dosage influenced phenotypes produced by UAS-SPM. One set of tests used the A9 wing disc driver, which, in combination with UAS-SPM, yields crumpled wing phenotypes (Figure 5B). We found that the wing defect in A9-GAL4/+; UAS-SPM/+ animals was substantially suppressed if the flies also carried a transgene copy of full-length SCM under identical UAS control (Figure 5C). When SCM dosage was instead reduced using a genetic background heterozygous for the null allele, ScmH1 (genotype: A9-GAL4/+; UAS-SPM/+; ScmH1/+), we found that the penetrance of severe wing phenotypes was enhanced. Modified phenotypes were also seen in dosage tests with other drivers. A dramatic example was the synthetic lethality produced by adding ScmH1/+ to the otherwise viable dpp-disc driver/UAS-SPM combination (Figure 5F). Thus, loss of function due to SPM overexpression was sensitive to either increased or decreased dosage of full-length SCM.
Although pupal lethality and/or adult homeotic phenotypes were commonly observed in these experiments, we were puzzled that embryonic lethality was generally not seen. Since PcG repression is required throughout development, we suspected that at least some ubiquitous drivers would produce enough SPM domain poison to disrupt embryogenesis. Possible explanations could be that SPM domain interactions are critical only during postembryonic stages or that higher levels of the poison domain are required to compete effectively with maternally loaded PcG proteins in embryos. To address this, we tested if embryo lethality depended upon levels of overexpressed SPM domain by varying dosage of the UAS-SPM construct and increasing temperature, which enhances expression from the GAL4 system. At 25°, embryos carrying a single copy of the ubiquitous daughterless-GAL4 driver (Wodarz et al. 1995) and either one or two copies of UAS-SPM are fully viable. However, at 29°, one copy of UAS-SPM reduces viability by half and two copies produce embryonic lethality (Table 3). Immunostaining of embryos in the 29° test revealed ectopic expression of the HOX protein ABD-A (data not shown). Thus, efficient disruption of embryonic PcG repression is indeed achieved with conditions that favor high levels of SPM domain expression.
Specificity of SPM domain interactions in vivo:
The SPM domain is a specialized version of a larger set of homology domains used for protein interactions known as SAM domains (Ponting 1995). Twenty-one SAM domain proteins are encoded by the Drosophila genome (Rubin et al. 2000). On the basis of a minimum of 30% identity and the signature features described in Figure 6B, a total of six of these Drosophila proteins can be classified as SPM domain proteins (Figure 6A). SCM and PH are the only two members of this group known to function as PcG repressors. Among the remaining four, functional data have been reported only for MBT, which is implicated in control of proliferation in the larval nervous system and in early embryonic nuclear divisions (Gateff et al. 1993; Yohn et al. 2003).
The self-binding properties of SPM/SAM domains together with the large number of such proteins raise questions about interaction specificity. Do the SCM and PH versions of the domain have intrinsic ability to interact only with relevant partners in vivo? If interaction specificity was relatively loose, then we would expect overexpression of the isolated SPM domain to disrupt many cellular processes and to produce a wide array of phenotypes, possibly including general cell lethality. In fact, the phenotypes observed with many different driver/UAS-SPM combinations were remarkably limited. The imaginal drivers produced specific homeotic transformations in otherwise normal and fertile adults. Furthermore, targeted SPM domain overexpression in several specific tissues, including the eye, salivary glands, and embryonic mesoderm, did not visibly alter development. Thus, SPM domain overexpression disrupts PcG repression without causing general cell lethality or widespread defects in tissue development.
Tests for SCM-PH association in embryo extracts:
Previous reports of in vitro SCM-PH interactions and SCM enrichment in purified fractions of embryonic PRC1 (Peterson et al. 1997; Shao et al. 1999) suggested that SCM partners functionally with PRC1. To further investigate the nature of SCM interactions with PRC1 in vivo, we performed tests for SCM-PH interaction in fly embryo extracts. In particular, we wished to assess if SCM stably associates with PRC1.
Figure 7A shows co-immunoprecipitations (coIPs) performed on embryo extracts using previously characterized polyclonal antibodies against SCM and PH (Bornemann et al. 1998; Ng et al. 2000). Although both antibodies work in primary immunoprecipitations (lane 6, top; lane 4, bottom), we find little or no coprecipitation of SCM with PH in either direction tested. The PH coIP with SCM was routinely negative (lane 4, top) and the reciprocal SCM coIP with PH was negative in some tests and weakly positive at best (as exemplified in lane 6, bottom) in other repeats of the experiment. These results suggest that there is not a robust association between SCM and PH in these extracts. Alternatively, it is possible that SCM-PH interactions were disrupted in these experiments by antibody binding to the respective PcG proteins. To circumvent this technical possibility and to directly assess complexes containing SCM and PH, we performed Superose 6 gel filtration chromatography on embryo nuclear extracts. Figure 7B shows that the bulk of embryonic SCM fractionates in a peak at ∼500 kD, which is separable from the larger PH-containing complexes. This result indicates that most of the soluble pool of SCM in embryos is not stably associated with PRC1.
SCM assembly with a reconstituted PRC1 core complex:
The discrepancy between robust SCM-PH interactions in vitro vs. little or no SCM-PH association detected in embryo extracts could be explained in several ways. One possibility concerns the artificial nature of the pairwise SCM-PH in vitro binding assay. Numerous biochemical studies imply that PH functions in vivo as a component of PRC1 rather than as an isolated monomer (Shao et al. 1999; Saurin et al. 2001; Levine et al. 2002). Thus, a more relevant in vitro test would be to determine if SCM can associate with intact PRC1 complex. We have previously shown that a reconstituted PCC, containing PC, PH, PSC, and dRING1, can recapitulate PRC1 activities on chromatin templates in vitro (Francis et al. 2001). Thus, we tested whether SCM can efficiently assemble with PCC if all five components are coexpressed in a baculovirus system. PCC complexes were assembled and immunoaffinity purified using a FLAG epitope tag on either the PH or the PSC subunit. Figure 8A shows a silver-stained SDS gel displaying components of purified complexes obtained after coexpression of different PcG protein combinations. Comparison of lanes 1 and 2 shows that SCM (arrow) does assemble with the four PCC components purified via Flag-PH. Figure 8B shows that this PCC + SCM complex is stable on a glycerol gradient with peak fractions (5–8) containing similar amounts of all five components. Fractions 9–12 likely contain a stable Flag-PH + SCM subcomplex, which is observed when just these two PcG proteins are tested for binding in vitro (Peterson et al. 1997; N. J. Francis, unpublished results). Functional analysis showed that the PCC + SCM complex could inhibit remodeling of nucleosomal arrays in vitro (data not shown), with activity similar to that of PCC isolated in the absence of SCM (Francis et al. 2001). Thus, although SCM association with PRC1 does not appear robust in embryo extracts, SCM has the capacity to assemble efficiently with partner PcG proteins in PCC.
To address whether this SCM assembly requires the PH subunit, the experiment was repeated using purification via FLAG-PSC in either the presence or the absence of PH. Figure 8A shows that the level of copurified SCM is significantly reduced if PH is singly omitted from the coexpression mix (compare lanes 3 and 4). Thus, SCM-PH interactions, presumably mediated by their respective SPM domains, make key contributions to SCM coassembly with PCC.
Key role for SCM and the SPM domain in PcG repression:
Purifications of nuclear complexes and in vitro studies have identified eight proteins that are core components of two distinct fly PcG complexes: ESC, E(Z), SU(Z)12, and NURF-55 in the ESC-E(Z) complex plus PC, PH, PSC, and dRING1 in PRC1. One function of the ESC-E(Z) complex is histone H3 methylation on K27 (Czermin et al. 2002; Muller et al. 2002), which is implicated in recruiting PRC1 to local chromatin (Cao et al. 2002; Czermin et al. 2002; Fischle et al. 2003; Min et al. 2003). Further studies are needed to address whether the ESC-E(Z) complex has additional functions. The molecular mechanism of PRC1 is not yet known. Studies to date suggest that it represses transcription through a noncatalytic mechanism that restricts template access, but it is not yet clear how PRC1 molecularly affects nucleosome array organization and/or packaging of the chromatin fiber. Since genetic studies in Drosophila identify at least 15 genes involved in PcG repression, many additional components need to be fit into the framework of PcG complexes and functions. In addition to identifying the players, analyses of loss of function for individual PcG genes distinguishes those repressors with central PcG roles from those that are more peripheral. In good agreement with the biochemical studies, loss of function for core subunits of either PcG complex produces severe homeotic defects (Struhl and Akam 1985; Breen and Duncan 1986; Jones and Gelbart 1990; Phillips and Shearn 1990; Simon et al. 1992; Soto et al. 1995; Birve et al. 2001). These mutants show robust Hox misexpression and die as embryos with most segments transformed into copies of the eighth abdominal segment. By these criteria, SCM is clearly a central player in the PcG repression system (Breen and Duncan 1986; Bornemann et al. 1998). In contrast, other repressors such as ASX and PCL appear more peripheral since their complete loss from embryos yields significantly weaker homeotic defects (Breen and Duncan 1986; Soto et al. 1995).
In this work, we present a combination of in vivo and in vitro approaches to address SCM molecular function. Our mutational analysis shows that SCM function absolutely depends upon an intact SPM protein interaction domain (Figures 1 and 3). There is a strong correlation between disruption of protein interactions in vitro (Figure 2) and failure of SCM function in vivo (Table 2). These results agree with our previous finding that SCM repressor function in an in vivo tethering assay requires its SPM domain (Roseman et al. 2001). The importance of SPM domain interactions is also revealed by PcG loss-of-function phenotypes produced by overexpression of an isolated SPM domain (Figure 5). We suggest that this dominant negative reflects SPM domain interactions critical for PcG repression that are disrupted by this avidly binding but otherwise nonfunctional competitor. The embryonic lethality of SPM domain mutants (Table 2), together with embryonic and imaginal defects seen with SPM overexpression, indicate that SPM interactions contribute to PcG repression at both embryonic and postembryonic times. Thus, these interactions appear required for long-term maintenance of PcG silencing in vivo.
SCM function as a potential partner with PRC1:
The biochemical properties of the SPM domain suggest three potential types of SCM interactions in vivo: (1) binding to PRC1, (2) binding to other fly SPM domain proteins (Figure 6), or (3) binding to itself. Although our data do not rule out contributions from any of these, several lines of evidence favor SCM interaction and function with PRC1. First, in vivo evidence derives from studies showing that SCM can repress reporter genes when tethered by fusion to a DNA-binding domain (Roseman et al. 2001). Since this repression depends upon PH function, SCM cannot repress on its own but rather requires PRC1 to repress in this context. Second, substoichiometric quantities of SCM consistently copurify with tagged PRC1 complexes from both fly and mammalian extracts (Saurin et al. 2001; Levine et al. 2002). Although the majority of SCM appears to not be stably bound (Figure 7), the conserved association of some SCM with purified PRC1 likely reflects in vivo interactions. Finally, we have not detected stably associated partner proteins that copurify when FLAG-SCM is affinity purified from embryo extracts (D. R. Mallin and J. A. Simon, unpublished results). Thus, we do not have evidence for a heteromeric SCM-containing complex that could repress independently of PRC1.
If SCM does work with PRC1, then what might explain its substoichiometric association with purified PRC1? One possibility is that SCM assembles only into a subset of PRC1 complexes, perhaps restricted to certain tissues or times of development. Such a model has been proposed to explain how ASX contributes to PcG repression in the embryonic epidermis but not in the central nervous system (Soto et al. 1995). We do not favor this explanation for SCM, however, because its requirement in PcG repression is widespread in both embryonic and imaginal tissues (Breen and Duncan 1986; Simon et al. 1992; Bornemann et al. 1998; Beuchle et al. 2001). Another possibility is that SCM interaction with PRC1 is robust in chromatin but is not fully preserved during preparation of soluble nuclear extracts used in purification. In this view, nucleosome arrays might provide a platform that promotes SCM-PRC1 binding. Indeed, both PRC1 and SCM have affinity in vitro for nucleosome arrays (Shao et al. 1999; Francis et al. 2001; D. R. Mallin, N. J. Francis, R. E. Kingston and J. A. Simon, unpublished results). Additional in vitro studies will be needed to address the nature of SCM-PRC1 interactions in the context of chromatin templates. We note that the GAGA factor provides an example of a protein that is not stably associated with PRC1 in embryo extracts but can nevertheless help recruit PRC1 to nucleosomal templates in vitro (Mulholland et al. 2003).
At present, the evidence favors a noncatalytic role for PRC1 in PcG repression. How might SCM, which also lacks recognizable catalytic domains, contribute to PRC1 mechanism? Recent in vitro studies show that mouse PRC1 bound to a single nucleosome array can recruit a second chromatin template that then also becomes repressed (Lavigne et al. 2004). These bridging interactions between repressed templates in vitro may reflect the PcG-dependent chromosome-pairing and chromosome-chromosome interactions frequently observed in vivo (Chan et al. 1994; Kassis 1994; Muller et al. 1999; Bantignies et al. 2003). Thus, one role of PRC1 may be to promote higher-order chromosome interactions that spread or stabilize repression. Intriguingly, among the core PRC1 components, the mouse PH protein was found most critical for in vitro bridging activity (Lavigne et al. 2004). Since we find that PH is the key subunit that mediates SCM interaction with PRC1 (Figures 2 and 8), it raises the possibility that SCM could facilitate PRC1-mediated long-distance chromatin interactions. In this view, SCM might work by helping to anchor PRE-promoter and/or PRE-PRE interactions needed for PcG repression in vivo.
A second type of potential SCM-PRC1 partnership in chromatin has been proposed on the basis of structural properties of the SPM domain. The SPM domain of fly PH, determined by X-ray crystallography, is a five-helix bundle that has the special property of forming helical self-polymers in vitro (Kim et al. 2002). The possibility of an extended protein polymer that could bind alongside nucleosome arrays has prompted speculation that SPM proteins might organize higher-order chromatin arrangements. In such a model, SPM domain-containing proteins or complexes form a core helical polymer around which the chromatin fiber could be wrapped (Kim et al. 2001, 2002). This model, although speculative, is appealing since it brings structural data to bear upon the long-standing hypothesis that PcG proteins create extended tracts of repressed chromatin (Paro 1990). Intriguingly, when mixed together, the SPM domains of PH and SCM can also form copolymers in vitro (Kim et al. 2002). Thus, PH and SCM could collaborate in forming the proposed higher-order chromatin structures. In this context, the dominant-negative properties of overexpressed SPM domain (Figure 5) could reflect disruption of contacts needed to produce PH-SCM chromatin polymers. To evaluate this model, it will be necessary to test if full-length PcG proteins or their intact complexes can form polymers in vitro like those seen for their isolated SPM domains. If so, then further studies would need to address the existence and roles of such polymers in vivo.
Contributions of additional SCM functional domains:
Although the present study highlights the SPM interaction domain, additional domains are likely important for SCM function. Foremost among these is the pair of mbt repeats (Figure 6), which are 67% identical in mammalian SCM homologs (Montini et al. 1999; Tomotsune et al. 1999). Since several loss-of-function Scm alleles are missense mutations in the first repeat (Bornemann et al. 1998), this domain is clearly required for Scm function in vivo. However, all of these mbt alleles are hypomorphic: Scm nulls are embryonic lethal whereas the homozygous mbt mutants survive to pupal stages (Bornemann et al. 1998). Similarly, missense mutations that affect one of the three mbt repeats in the Drosophila MBT protein produce only partial loss of function (Yohn et al. 2003). In light of their evolutionary conservation, we favor a vital role for the mbt repeats in SCM. Mutations that cleanly disrupt both SCM repeats will be needed to settle this issue.
In addition to multiple mbt repeats, the fly SCM and MBT proteins also share an SPM domain and a specific type of zinc finger (Figure 6; see also Bornemann et al. 1998). Although this remarkable similarity in overall organization suggests a functional connection, studies to date define distinct developmental roles for SCM and MBT. There is little evidence that MBT protein contributes to PcG repression since l(3)mbt mutations do not show Hox phenotypes. Instead, l(3)mbt mutations cause overproliferation of larval neuroblasts (Gateff et al. 1993) and asynchrony of nuclear divisions in early embryos (Yohn et al. 2003). Perhaps the shared domains reflect a common role at the molecular level in chromatin, which, depending upon times and sites of action, yield distinct developmental functions. Since the human MBT homolog shows chromatin association restricted to mitotic chromosomes (Koga et al. 1999), a potential common role with SCM could be in chromatin condensation or packaging.
What molecular role might mbt repeats play in chromatin proteins? Some suggestions are provided by the recent structure determination of the mbt repeats in a human homolog of SCM (Sathyamurthy et al. 2003). Each mbt repeat consists of an N-terminal extended arm and a C-terminal core containing a five-stranded β-barrel. There is structural similarity within this β-barrel domain to the Tudor domain of survival motor neuron protein, which can bind methylated arginine residues (Selenko et al. 2001). This binding pocket is structurally superimposable upon the methylated histone H3-binding region of the chromodomain in heterochromatin protein 1 (Nielsen et al. 2002; Maurer-Stroh et al. 2003). Thus, it has been suggested that mbt repeats may also function as modules for binding methylated peptides. Given SCM function in PcG repression, a possible binding substrate could be methylated histone tails, similar to binding of the PC chromodomain to methylated H3-lysine 27 (Fischle et al. 2003; Min et al. 2003). Further in vitro studies will be needed to test whether SCM binds specifically to methylated histones and, if so, which methylated residues are recognized.
The functional requirement for the SCM Cys2-Cys2 zinc fingers (Figure 6) is also presently unresolved, as there are no alleles that alter this N-terminal domain. These fingers may be less central to the core SCM function in chromatin since they have apparently been lost in evolution of the mammalian SCM homologs (Montini et al. 1999; Tomotsune et al. 1999). Nevertheless, a functional role is supported by the shared occurrence of this precise zinc-finger subtype in the SCM, PH, and MBT proteins (Bornemann et al. 1996). Clearly, further mutational studies using both in vitro and in vivo assays will be needed to determine how the multidomain SCM protein contributes to PcG repression. On the basis of the recently solved domain structures, together with the functional studies described here, we suggest that SCM plays a noncatalytic role in binding and packaging chromatin in concert with PRC1.
We thank Theo Haerry, Guillermo Marques, Tom Neufeld, John Tamkun, and Mike O'Connor for providing GAL4 driver fly strains used in this work. We also thank Lesley Brown for sharing a baculovirus construct for SCM production. We are grateful to Ellen Miller, Dieter Roggy, and Richard Ha for help with plasmid constructions and processing some of the SPM domain mutants. This work was supported by National Institutes of Health (NIH) grants to J.A.S. and R.E.K. and by funding from Hoechst AG to R.E.K. D.R.M. was supported by a postdoctoral fellowship from the NIH (National Research Service Award GM20895). N.J.F. was supported by the Cancer Research Fund of the Damon Runyon-Walter Winchell Foundation (DRG-1589) and a fellowship from the Charles A. King Trust. A.J.P., C.S.K., and J.S. were supported in part by NIH training grant HD07480.
Communicating editor: J. Tamkun
- Received February 10, 2004.
- Accepted March 24, 2004.
- Genetics Society of America