Molecular basis of PRC1 targeting to Polycomb response elements by PhoRC

Here, Frey et al. report the structural basis by which the Drosophila Pho-repressive complex (PhoRC), the only Polycomb group protein complex with sequence-specific DNA-binding activity, binds to Polycomb-repressive complex 1 (PRC1) and thereby recruits it to Polycomb response elements in target genes.

Abstract

Polycomb group (PcG) protein complexes repress transcription by modifying target gene chromatin. In Drosophila, this repression requires association of PcG protein complexes with cis-regulatory Polycomb response elements (PREs), but the interactions permitting formation of these assemblies are poorly understood. We show that the Sfmbt subunit of the DNA-binding Pho-repressive complex (PhoRC) and the Scm subunit of the canonical Polycomb-repressive complex 1 (PRC1) directly bind each other through their SAM domains. The 1.9 Å crystal structure of the Scm-SAM:Sfmbt-SAM complex reveals the recognition mechanism and shows that Sfmbt-SAM lacks the polymerization capacity of the SAM domains of Scm and its PRC1 partner subunit, Ph. Functional analyses in Drosophila demonstrate that Sfmbt-SAM and Scm-SAM are essential for repression and that PhoRC DNA binding is critical to initiate PRC1 association with PREs. Together, this suggests that PRE-tethered Sfmbt-SAM nucleates PRC1 recruitment and that Scm-SAM/Ph-SAM-mediated polymerization then results in the formation of PRC1-compacted chromatin.

Polycomb group (PcG) proteins are transcriptional regulators that maintain cell fate decisions in animals and plants by repressing transcription of developmental regulator genes in cells where these genes should remain inactive. In Drosophila, 18 different proteins are classified as PcG members because animals lacking any of these proteins show widespread misexpression of Hox and other developmental regulator genes. Biochemical purification of these proteins revealed that these proteins are the subunits of four distinct protein complexes: Polycomb-repressive complex 1 (PRC1), PRC2, Pho-repressive complex (PhoRC), and Polycomb-repressive deubiquitinase (PR-DUB) (Shao et al. 1999; Czermin et al. 2002; Müller et al. 2002; Klymenko et al. 2006; Scheuermann et al. 2010). Protein assemblies identical or related to PRC1, PRC2, and PR-DUB have also been purified from mammalian cells (Cao et al. 2002; Kuzmichev et al. 2002; Levine et al. 2002; Machida et al. 2009; Sowa et al. 2009; Yu et al. 2010; Gao et al. 2012).

PcG protein complexes repress transcription of target genes by modifying their chromatin (for review, see Simon and Kingston 2013). Biochemical and genetic studies have provided compelling evidence that the trimethylation of Lys27 in histone H3 by PRC2 (Cao et al. 2002; Czermin et al. 2002; Kuzmichev et al. 2002; Müller et al. 2002) and the capacity of PRC1 to compact chromatin (Francis et al. 2001, 2004; Grau et al. 2011) are critical for repression of PcG target genes in vivo (King et al. 2005; Eskeland et al. 2010; Isono et al. 2013; Pengelly et al. 2013; McKay et al. 2015). In contrast, the monoubiquitylation of histone H2A by PRC1 (Wang et al. 2004a), long considered to be a critical step for gene silencing by the PcG machinery (e.g., Stock et al. 2007), was recently shown to be dispensable for repression of canonical PRC1 target genes in both Drosophila (Pengelly et al. 2015) and mice (Illingworth et al. 2015).

The molecular interactions that permit PRC1 or PRC2 to associate with specific genomic locations are currently only poorly understood. Studies in mammalian cells have suggested that targeting of these complexes to DNA may entail binding to transcription factors, long noncoding RNAs, nonmethylated CpG dinucleotides in CpG islands, or methylated lysines in histone proteins (for review, see Klose et al. 2013; Simon and Kingston 2013). Despite this wealth of reported interactions, structural information about these is currently available only for the binding of PRC1 and PRC2 subunits to methylated lysine residues in histone proteins (Fischle et al. 2003; Min et al. 2003; Grimm et al. 2007, 2009; Santiveri et al. 2008; Guo et al. 2009; Margueron et al. 2009; Jiao and Liu 2015). Progress toward elucidating the molecular basis of PcG protein complex targeting has come from studies in Drosophila, where PcG protein complexes assemble at Polycomb response elements (PREs). PREs typically comprise a few hundred base pairs and contain binding sites for the PhoRC subunit Pho, the only PcG protein with sequence-specific DNA-binding activity (Brown et al. 1998; Kwong et al. 2008; Oktaba et al. 2008; Schuettengruber et al. 2009). The molecular basis of how Pho binds to PRE DNA is known from the cocrystal structure of the human Pho homolog YY1 bound to its cognate DNA-binding site (Houbaviy et al. 1996); the YY1 residues contacting DNA are 100% conserved in Pho (Brown et al. 1998). Early studies proposed that the Pho spacer, a region of 30 amino acids that is highly conserved in mammalian YY1, directly interacts with different subunits of PRC1 or PRC2 (Mohd-Sarip et al. 2002, 2005; Wang et al. 2004b). However, attempts to reconstitute stable Pho:PRC2 or Pho:PRC1 assemblies with recombinant proteins have failed (Mohd-Sarip et al. 2005; Klymenko et al. 2006), and the molecular basis of the proposed interactions of Pho with the diverse PRC2 or PRC1 subunits has remained elusive. Recent structural studies revealed that the Pho spacer forms the interaction domain by which Pho binds its PhoRC partner subunit, Sfmbt (Alfieri et al. 2013). This finding, together with the observations that Sfmbt is co-bound with Pho at PREs genome-wide (Oktaba et al. 2008) and that Sfmbt association with PREs depends on intact Pho protein-binding sites in vitro and in vivo (Klymenko et al. 2006; Alfieri et al. 2013), suggest that Sfmbt rather than Pho itself may represent the docking platform for interaction with PRC1 and/or PRC2. In support of such a scenario, the PRC1 accessory subunit Scm interacts with recombinant Sfmbt protein in vitro (Grimm et al. 2009), but the molecular basis of this interaction is not known.

Here, we set out to purify proteins that associate with canonical PRC1 in Drosophila embryos and identified PhoRC as a major interactor of this complex. We found that the SAM domains of Sfmbt and Scm mediate the interaction between the two complexes and determined the crystal structure of the Sfmbt-SAM:Scm-SAM complex to reveal the recognition mechanism. Functional tests in Drosophila show that the Sfmbt-SAM:Scm-SAM interaction is critical for long-term PcG repression of target genes and that PRC1 association with PREs requires DNA-bound PhoRC. Together, these studies thus reveal the molecular basis of how PhoRC targets PRC1 to PREs.

Results

Biochemical purification identifies PhoRC as an interaction partner of PRC1

To identify proteins that associate with the PRC1 subunit polyhomeotic-proximal (Ph-p) or its paralog, polyhomeotic-distal (Ph-d), we performed tandem affinity purification (TAP) (Rigaut et al. 1999) on nuclear extracts from transgenic Drosophila embryos expressing TAP-tagged Ph-p or Ph-d, respectively (Supplemental Fig. S1A). The TAP-Ph-p and TAP-Ph-d fusion proteins were expressed at levels comparable with endogenous Ph-p and Ph-d proteins (Supplemental Fig. S1B), and either protein was able to rescue the severe phenotype of ph⁰ mutant embryos (Supplemental Fig. S1C), demonstrating that the fusion proteins could functionally substitute Ph-p and Ph-d. Mass spectrometric analyses of the purified material identified the PRC1 core subunits Psc, Su(z)2, Sce, and Pc and, intriguingly, also the PhoRC subunit Sfmbt as possible interaction partners of either Ph-p or Ph-d (Supplemental Fig. S1D; Supplemental Table S1). Western blot analysis confirmed that PRC1 and Sfmbt are enriched in TAP-Ph-p and TAP-Ph-d purifications and revealed that the purified assembly also contains the PRC1 accessory subunit Scm (Fig. 1A). The finding that PhoRC subunits are associated with PRC1 in Drosophila nuclear extracts is consistent with earlier studies that identified PhoRC subunits in Pc protein assemblies purified from Drosophila embryonic nuclear extracts (Strübbe et al. 2011) and in Pc and Scm chromatin assemblies purified from embryos after cross-linking (Kang et al. 2015). Moreover, this association also appears to be conserved in vertebrates where canonical PRC1 subunits were identified in purifications of Sfmbt1 and Sfmbt2 from human cells (Zhang et al. 2013).

Figure 1.

PRC1 and PhoRC interact in Drosophila via direct binding of Scm to Sfmbt. (A) Western blot analysis of material isolated by TAP from wild-type (wt) and α-tubulin1-TAP-Ph-d transgenic embryos. Total nuclear extract input (lanes 1,2) and material eluted from calmodulin affinity resin after purification (lane 4) and mock purification (lane 3) were probed with the indicated antibodies. Only PRC1 subunits and Sfmbt are clearly enriched in lane 4 and undetectable in lane 3. See the text for details. (B) Domain architecture of Sfmbt, Scm, and Ph. (C) Coomassie-stained SDS-PAGE gels of Flag affinity-purified material from Sf9 cells coexpressing the indicated proteins. See the text for details.

Sfmbt and Scm interact through the C-terminal SAM domains

We next wanted to identify the molecular basis of the PRC1–PhoRC interaction. We previously found that, upon coexpression in insect cells, Scm and Sfmbt can be isolated as a stable dimeric complex (Grimm et al. 2009). Scm itself associates with PRC1 by binding to Ph, an interaction that is mediated by the SAM domains that are present in the C termini of both proteins (Fig. 1B; Peterson et al. 1997, 2004; Kim et al. 2005). To test whether Sfmbt may also interact with Ph, we coexpressed Sfmbt with Flag-tagged Ph-p_1298–1589, a C-terminal fragment of Ph-p that has high sequence identity with Ph-d and contains the three structural domains in Ph proteins (Fig. 1B; Supplemental Fig. S1A). However, Flag affinity purification resulted in the isolation of Flag-Ph-p_1298–1589 alone (Fig. 1C, lane 1), suggesting that Sfmbt and Ph-p do not interact directly. In contrast, coexpression of Scm together with Sfmbt and Flag-Ph-p_1298–1589 protein permitted isolation of a stable complex containing all three proteins (Fig. 1C, lane 2). This suggested that Scm, binding to both Ph and Sfmbt, may act as the physical link that mediates the interaction between PRC1 and PhoRC.

These results prompted us to examine how Sfmbt and Scm bind to each other. Initial studies using C-terminally truncated Sfmbt and Scm proteins had suggested that the two proteins interact through poorly defined regions in their N termini (Grimm et al. 2009). However, we found that a Flag-Sfmbt_530–1220 protein that lacks the previously described N-terminal Scm-interacting region is also able to form a stable complex with Scm (Fig. 1C, lane 4), suggesting an additional Scm-interacting region in the C terminus of Sfmbt. Deletion of the C-terminal SAM domain in Sfmbt substantially reduced the ability of this truncated Flag-Sfmbt_530–1136 protein to bind to Scm (Fig. 1C, lane 3). This suggested that the Sfmbt SAM domain is important for binding to Scm.

Previous structural studies showed that the SAM domains of Scm and Ph both contain a mid-loop (ML) and an end helix (EH) surface (Fig. 2A) and have the propensity to form homopolymers or heteropolymers in a head-to-tail fashion where the EH surface of one SAM domain interacts with the ML surface of the SAM domain-binding partner (Kim et al. 2002, 2005). Work on Scm-SAM:Ph-SAM heteropolymers revealed that this assembly occurs in a defined orientation with preferential binding of the Scm-SAM EH surface to the Ph-SAM ML surface (Kim et al. 2005). Intriguingly, we found that recombinant Sfmbt-SAM_1137–1220 (referred to as Sfmbt-SAM), if expressed alone, eluted in a single peak corresponding to a monomer when purified on a gel filtration column (Supplemental Fig. S2A). The Sfmbt-SAM domain thus lacks the capacity to form homopolymers. We next tested whether Sfmbt-SAM might bind to the Scm-SAM domain. Considering that Sfmbt, Scm, and Ph can form a trimeric complex and that the EH surface of Scm-SAM engages in binding to Ph-SAM, we hypothesized that Sfmbt might bind to the ML surface of Scm-SAM. In a first set of pull-down experiments, we found that Scm-SAM_803–877 in which the EH surface had been mutated (Scm-SAM^L855E/L859E) to prevent the formation of Scm-SAM homopolymers bound to GST-Sfmbt-SAM (Supplemental Fig. S2B). In contrast, Ph-SAM^{L1561E/L1565E} containing the corresponding EH surface mutations to prevent Ph-SAM homopolymer formation interacted only very poorly with GST-Sfmbt-SAM in the same assay (Supplemental Fig. S2B). We then coexpressed Sfmbt-SAM and Scm-SAM^L855E/L859E and found that the Sfmbt-SAM:Scm-SAM^L855E/L859E complex (referred to here as Sfmbt-SAM:Scm-SAM) eluted as a single peak corresponding to a dimer on a gel filtration column (Fig. 2B; Supplemental Fig. S2A), suggesting that these two SAM domains form a stable complex.

Figure 2.

Structural analysis of the Sfmbt-SAM:Scm-SAM interaction. (A) Sequence alignment of the SAM domains of Sfmbt, Scm, Ph-p, and Ph-d. The ML (light orange) and EH (purple) surfaces are highlighted. (B) Elution profile of the Sfmbt-SAM:Scm-SAM complex and Coomassie-stained SDS-PAGE of peak fractions. (C) Crystal structure of the Sfmbt-SAM:Scm-SAM complex in ribbon diagram presentation. Close-up view of the Sfmbt-SAM:Scm-SAM complex interface. Hydrophobic residues (top panel) and polar residues (bottom panel) that form the interaction surface are depicted. Hydrogen bonds and salt bridges are indicated (dashed lines). See the text. (D) Structural superposition of the Sfmbt-SAM:Scm-SAM complex with the Scm-SAM:Ph-SAM complex (Protein Data Bank [PDB] ID 1PK1) to model a trimeric complex containing the SAM domains of Sfmbt, Scm, and Ph.

Structure of the Sfmbt-SAM:Scm-SAM complex

We obtained diffracting crystals of the Sfmbt-SAM:Scm-SAM dimer. We solved the structure by molecular replacement with Ph-SAM:Scm-SAM as a search model and refined it to 1.9 Å resolution (R_free of 27.7% and R_work of 25.5%). The asymmetric unit contained two Sfmbt-SAM:Scm-SAM dimers that interacted laterally. This lateral interaction between the two dimers is likely imposed by crystal packing because Sfmbt-SAM:Scm-SAM behaves as a heterodimer in gel filtration chromatography even at high concentrations (Fig. 2B), and, moreover, one of the two interactions between the dimers is a salt bridge formed by the mutated Scm^L859E residue (data not shown). In the following, we thus consider the Sfmbt-SAM:Scm-SAM complex to exist as a dimer.

The structure revealed that the Sfmbt-SAM domain is a helical bundle with the canonical SAM domain fold (Fig. 2C). The major interaction in the dimer is between the EH surface of Sfmbt-SAM and the ML surface of Scm-SAM and covers an area of ∼550 Ų (calculated with PDBePISA [proteins, interfaces, structures, and assemblies]) (Fig. 2C). The interface comprises a hydrophobic region and an adjacent polar region. The hydrophobic interactions involve the side chains of A838, L841, L842, M846, and Y850 on the Scm ML surface and M1180, V1187, G1188, and L1191 on the Sfmbt EH surface (Fig. 2C right, top panel). The adjacent polar interactions are formed by three salt bridges (E833^Scm–K1186^Sfmbt, D835^Scm–K1192^Sfmbt, and K849^Scm–D1177^Sfmbt) and hydrogen bonds between backbone [H832^Scm(O)–G1188^Sfmbt(N)] and side chains (Y850^Scm–D1177^Sfmbt) (Fig. 2C, right, bottom panel). Two water molecules bridging Y850^Scm with D1177^Sfmbt and E833^Scm with K1192^Sfmbt further stabilize the interface (data not shown).

The ML surface of Sfmbt-SAM lacks the capacity for forming SAM–SAM interactions

Inspection of the ML surface of the Sfmbt-SAM domain explains why this domain lacks polymerization capacity. In Sfmbt, a number of residues at positions corresponding to the apolar residues in the Scm ML surface are substituted with polar residues. In particular, R1170 in the ML surface of Sfmbt-SAM, corresponding to a conserved Ala in Scm-SAM and Ph-SAM (Fig. 2A), would block interaction with the EH surface of another SAM domain. This provides a likely explanation of why Sfmbt-SAM can engage in interaction with only a single SAM domain partner via the EH surface.

Molecular basis for the formation of extended Sfmbt–Scm–Ph assemblies

Because Scm-SAM adopts the same conformation in Sfmbt-SAM:Scm-SAM and Scm-SAM:Ph-SAM (Protein Data Bank [PDB] ID 1PK1), the two complexes can easily be superimposed via their common Scm-SAM domains (root mean square deviation [RMSD] of 0.42 over 67 Cα). This superposition permitted us to generate a model of the trimeric Sfmbt-SAM:Scm-SAM:Ph-SAM complex (Fig. 2D). In support of this, we reconstituted a complex containing all three SAM domains by coexpressing Sfmbt-SAM with an Scm-SAM–Ph-SAM^{L1561E/L1565E} fusion protein (Supplemental Fig. S2C). On a gel filtration column, the major peak contained stoichiometric amounts of the two polypeptides (Supplemental Fig. S2C). We presume that this defined complex containing all three SAM domains is formed because the fusion protein promotes the intramolecular interaction of Scm-SAM with Ph-SAM and thus effectively limits the oligomerization capacity of Scm-SAM, leaving its ML surface available for interaction with Sfmbt-SAM.

In each of the three proteins, the SAM domain is located at the very C terminus and is separated from adjacent domains by linker regions that are predicted to be mostly disordered (Fig. 1B). Therefore, even though the SAM polymer assumes a rigid conformation, these adjacent linker sequences likely allow flexibility for orientation of the rest of each protein. It is important to recall that both Scm-SAM and Ph-SAM have the capacity to form homopolymers or heteropolymers. PRE-tethered Sfmbt, lacking this polymerization capacity, may therefore act as an assembly platform to initiate formation of longer polymers containing Scm and/or Ph proteins that emanate from PREs.

The SAM domains of Sfmbt and Scm are essential for target gene repression

We next investigated the requirement of the Sfmbt-SAM and Scm-SAM domains for transcriptional repression of PcG target genes in Drosophila. In a first set of experiments, we tested whether Sfmbt^ΔSAM lacking the SAM domain can replace the endogenous Sfmbt protein in a genetic rescue assay in larvae, as follows: In the wing imaginal disc, clones of cells that are homozygous for the Sfmbt¹-null mutation fail to maintain PcG repression of the Hox gene Ultrabithorax (Ubx), and the Ubx protein becomes strongly misexpressed in the mutant clones (Fig. 3A, left). When such Sfmbt¹ mutant clones were induced in animals carrying a transgene with a genomic fragment expressing wild-type Sfmbt protein, the transgene-encoded Sfmbt protein fully rescued repression, and no Ubx protein was detected in the Sfmbt¹ homozygous cells (Fig. 3A, middle). In contrast, the Sfmbt^ΔSAM protein expressed from the same genomic fragment was inefficient in rescuing repression, and the Ubx protein was strongly misexpressed in a large fraction of Sfmbt¹ homozygous cells (Fig. 3A, right). Importantly, the truncated Sfmbt^ΔSAM protein was stable and present at levels comparable with that of the wild-type Sfmbt protein (Fig. 3B). Together, these results show that the SAM domain is critical for Sfmbt to function in PcG repression. The comparison of the repression defects in Sfmbt^ΔSAM and Sfmbt-null mutant clones nonetheless suggests that the Sfmbt^ΔSAM protein retains partial repressor function (Fig. 3A), suggesting that deletion of the SAM domain does not completely incapacitate the protein.

Figure 3.

SAM domains of Sfmbt and Scm are required for PcG repression. (A) Imaginal wing discs with clones of Sfmbt¹ homozygous cells from animals that carried no transgene (no TG) or the indicated genomic transgenes expressing wild-type Sfmbt (Sfmbt^wt) or Sfmbt^ΔSAM stained with antibody against Ubx protein (red). Clones of Sfmbt¹ homozygous cells are marked by the lack of GFP (green) and were induced 72 h before analysis. Only Sfmbt¹ homozygous cells (no TG) in the wing pouch (arrowheads) but not in the notum or hinge show strong misexpression of Ubx (Klymenko et al. 2006), and rescue of Ubx repression by Sfmbt^wt or Sfmbt^ΔSAM was therefore only analyzed in clones in the wing pouch area. In no TG animals, 98,4% of Sfmbt¹ homozygous clones (n = 63 clones) showed misexpression of Ubx (arrowheads). In animals carrying Sfmbt^wt, repression of Ubx was rescued in most Sfmbt¹ homozygous clones (empty arrowheads), and only 1% of clones (n = 104 clones) showed misexpression of Ubx. In animals carrying Sfmbt^ΔSAM, 55% of Sfmbt¹ homozygous clones (n = 159 clones) showed misexpression of Ubx (arrowheads). Thus, even though the Sfmbt^ΔSAM protein largely fails to rescue, the repression of Ubx in 45% of clones (empty arrowheads) suggests that the Sfmbt^ΔSAM protein retains some repressor activity. An asterisk marks normal Ubx expression in a trachea attached to the disc. (B) Western blots on serial dilutions (9:3:1) of extracts from whole nuclei of 0- to 12-h-old embryos of the indicated genotypes probed with antibodies against Sfmbt and, as a loading control on the same membrane, Lamin. The asterisk marks the band of wild-type Sfmbt in Sfmbt¹ heterozygotes. Note that in lanes 10–12, the levels of Sfmbt^ΔSAM and endogenous wild-type Sfmbt (bottom and top arrows, respectively) are comparable. (C) Analysis of Ubx repression as in A but in clones of Scm^H1 homozygous cells in animals carrying no transgene (no TG) or transgenes expressing Flag-tagged wild-type Scm (F-Scm^wt) or F-Scm^ΔSAM. Scm^H1 homozygous clones marked by a lack of GFP (green) were induced 72 h before analysis. Only clones in the pouch were analyzed for statistics. In no TG animals, 100% of Scm^H1 homozygous clones (n = 94 clones) showed misexpression of Ubx (arrowheads). F-Scm^wt rescued repression of Ubx in all clones (empty arrowheads), and none of the clones (n = 42 clones) showed misexpression of Ubx, consistent with the lack of a phenotype in Scm-null mutant adults that are rescued by this transgene (Peterson et al. 2004). In animals carrying F-Scm^ΔSAM, 100% of Scm^H1 homozygous clones (n = 130 clones) showed misexpression of Ubx (arrowheads). See also Supplemental Figure S3. (D) Western blots on serial dilutions (9:3:1) of total extracts from 14- to 18-h-old embryos of the indicated genotypes probed with antibodies against Scm and, as loading control on the same membrane, Caf1. An asterisk marks maternally deposited wild-type Scm protein that persists in Scm^H1 homozygous embryos; in lanes 7–12, this band is partially obscured ([*]) by the bands from the transgene-encoded F-Scm^wt and F-Scm^ΔSAM proteins (top and bottom arrows, respectively).

Next, we assessed the requirement of the Scm-SAM domain. Previous studies showed that a transgene containing a genomic Scm fragment rescues Scm-null mutant animals into viable and fertile adults but that an Scm^ΔSAM protein expressed from the same genomic fragment fails to rescue viability of these animals (Peterson et al. 2004). To extend this finding, we tested the capacity of the Scm^ΔSAM protein to repress PcG target genes in clones of Scm-null mutant cells. Only the transgene expressing wild-type Scm protein was able to rescue repression of Ubx in clones of cells that were homozygous for the Scm^H1-null mutation (Fig. 3C). The Scm^ΔSAM protein completely lacked Scm repressor activity, and the Ubx protein was as widely misexpressed as in Scm^H1 mutant clones in animals carrying no transgene (Fig. 3C). In parallel, we also assayed the rescue capacity of the Scm^ΔSAM protein in embryos that were homozygous for Scm^H1 (Supplemental Fig. S3). Only the transgene-encoded wild-type Scm protein, but not the Scm^ΔSAM protein, was able to rescue repression of the Hox gene Abdominal-B (Abd-B) in Scm^H1 homozygous embryos (Supplemental Fig. S3). As previously reported (Peterson et al. 2004), we found that the Scm^ΔSAM protein is stable and expressed at levels comparable with that of the wild-type Scm protein (Fig. 3D). Together, these results show that the Scm-SAM domain is essential for Scm to function in PcG repression.

Finally, we note that similar structure/function analyses of the Ph protein in Drosophila recently demonstrated that the Ph SAM domain is essential for all functions of Ph (Gambetta and Müller 2014). Taken together, these data thus show that deletion of the SAM domain in Sfmbt, Scm, or Ph in each case ablates protein function.

PhoRC DNA-binding is critical for PRC1 recruitment to a PRE

We next investigated the requirement of PhoRC for recruitment of PRC1 to PREs in developing Drosophila. A number of limitations precluded us from monitoring PRC1 binding in Sfmbt or pho phol mutant animals by chromatin immunoprecipitation (ChIP) assays. First, it is not possible to generate embryos lacking maternally deposited PhoRC because the complex is essential for germ cell development (Breen and Duncan 1986; Girton and Jeon 1994; Brown et al. 2003; Klymenko et al. 2006). Consequently, in Sfmbt or pho phol zygotic mutant embryos, maternally deposited wild-type PhoRC permits almost normal establishment of PcG repression (Simon et al. 1992; Brown et al. 2003; Klymenko et al. 2006). Second, as the Sfmbt and pho phol zygotic mutant animals that develop into larvae are becoming depleted of maternally supplied PhoRC, the diploid larval cells begin to show defects in proliferation and form only poorly developed central nervous system (CNS) and rudimentary imaginal disc tissues (Brown et al. 2003; Klymenko et al. 2006), making these tissues unreliable material for ChIP analysis.

As an alternative approach to analyze PRC1 recruitment to PREs in the absence of PhoRC, we analyzed binding at a PRE transgene with mutated Pho/Phol-binding sites. PRE_D, a 570-base-pair (bp)-long fragment containing the core of the bxd PRE in Ubx, contains six Pho protein-binding sites (Chan et al. 1994; Fritsch et al. 1999). Previous studies in embryos showed that Pho, Sfmbt, and Ph bind to the wild-type PRE_D in a transgene but that this binding was strongly reduced at a PRE_{D pho mut} transgene in which the Pho-binding sites had been mutated (Fritsch et al. 1999; Klymenko et al. 2006). Here, we performed ChIP assays in imaginal wing discs from PRE_D or PRE_{D pho mut} transgenic larvae to monitor binding of the Pho, Sfmbt, Scm, and Ph proteins in the same tissues where we investigated the function of the Sfmbt-SAM and Scm-SAM domains (Fig. 3). In addition, we also analyzed binding of the Trithorax (Trx) protein, the PRC2 core subunit E(z), and the levels of the H3K27me3 mark. Binding of each of these proteins at the native bxd PRE in Ubx, at the iab-7 PRE in Abd-B, and at other genomic regions not bound by PcG proteins served as internal controls in the two transgenic lines (Fig. 4A). The Pho, Sfmbt, Scm, and Ph proteins were all bound at the wild-type PRE_D transgene, but, for each protein, only very low-level binding was detected at the PRE_{D pho mut} transgene (Fig. 4A). In contrast, binding of the Trx protein was comparable at the PRE_D and PRE_{D pho mut} transgenes (Fig. 4A). E(z) and H3K27me3 were strongly enriched at the PRE_D transgene and drastically reduced at the PRE_{D pho mut} transgene, consistent with earlier findings that suggested that Pho is required for association of PRC2 with PREs (Wang et al. 2004b). In summary, these analyses show that recruitment of PRC1 and also PRC2 to the bxd PRE critically depends on the ability of PhoRC to bind to Pho DNA-binding sites.

Figure 4.

PRC1 tethering to PREs by the PhoRC complex. (A) Binding of PRC1 to the bxd PRE depends on Pho protein-binding sites. ChIP analysis monitoring binding of Ph, Scm, Sfmbt, Pho, Trx, and E(z) and the levels of H3K27me3 in PRE_D (white bars) and PRE_{D pho mut} (black bars) transgenic animals. In PRE_D transgenic animals, all proteins are specifically bound at PRE_D and at the native bxd and iab-7 PREs, and no or only a low level of binding is detected in regions flanking these PREs and at two control regions 1 and 2 elsewhere in the genome (kilobase coordinates indicate location of bxd and iab-7 PREs with respect to Ubx and Abd-B transcription start sites, respectively). (Left) In PRE_{D pho mut} transgenic animals, binding of Pho, Sfmbt, Scm, Ph, E(z), and H3K27me3 at PRE_{D pho mut} is strongly reduced due to mutation of all six Pho protein-binding sites in this PRE fragment, whereas binding of Trx at PRE_{D pho mut} is unchanged or even increased. Binding of all proteins and H3K27me3 levels at the other regions in PRE_{D Pho mut} animals are comparable with that in PRE_D animals. Graphs represent results from three independent ChIP reactions on three separately prepared batches of chromatin; quantitative PCR signals are represented as percentage of input material precipitated in each immunoprecipitation reaction, and error bars show standard deviation. (B) Molecular model of interactions with which PhoRC tethers canonical PRC1 to PREs. The Pho/YY1 zinc finger domain (ZnF; purple) recognizes the Pho/YY1-binding motif GCCAT in a PRE DNA (PDB ID 1ubd) (Houbaviy et al. 1996). The Pho spacer region (purple) binds to the 4MBT domain of Sfmbt (orange) to form PhoRC (PDB ID 4C5I) (Alfieri et al. 2013). The ML surface of Scm-SAM (slate) binds to the EH surface of Sfmbt-SAM (orange) (this study), while the EH surface of Scm-SAM binds to the ML surface of Ph-SAM (green) (PDB ID 1PK1) (Kim et al. 2005)), providing a physical link between PhoRC and canonical PRC1. Binding of the 2MBT domain of Scm and the 4MBT domain of Sfmbt to monomethylated or dimethylated lysines in histone N termini (PDB ID 2R57 and 2R5A; 3h6z) (methylated histone peptides are not shown here) (Grimm et al. 2007, 2009) possibly provide additional binding interactions with nucleosomes in target gene chromatin.

Discussion

Atomic-level information on the molecular interactions by which PcG protein complexes bind to the genes that they regulate is essential for understanding how the PcG system works. In Drosophila, the different PcG protein complexes assemble at PRE sequences in target genes. Previous structural studies showed how YY1, the mammalian ortholog of the PhoRC subunit Pho, recognizes its cognate DNA-binding site, providing an atomic model of how PhoRC binds to PRE DNA (Fig. 4B). More recent studies then revealed how Sfmbt interacts with the Pho spacer to form PhoRC (Fig. 4B). Here, we present the structural basis of how PhoRC binds to canonical PRC1 and provide functional evidence that this PhoRC–PRC1 interaction targets PRC1 to PRE DNA (Fig. 4).

The following main conclusions can be drawn from the work presented in this study. First, biochemical purification of canonical PRC1 from nuclear extracts of Drosophila identifies PhoRC but no other DNA-binding proteins among the most highly enriched PRC1 interactors. This observation is consistent with previous studies (Strübbe et al. 2011) and suggests that, in Drosophila, PhoRC is the main PRC1 interaction partner with sequence-specific DNA-binding activity. Second, biochemical reconstitution shows that the PRC1 subunit Scm and the PhoRC subunit Sfmbt mediate this interaction by binding to each other through their C-terminal SAM domains. Third, the crystal structure of the Sfmbt-SAM:Scm-SAM complex reveals the recognition mechanism and uncovers that, unlike Scm-SAM and Ph-SAM, the Sfmbt-SAM domain lacks the capacity to oligomerize. Fourth, genetic analyses in Drosophila demonstrate that the SAM domains of Sfmbt and Scm are both essential for repression of PcG target genes, providing functional evidence for the importance of the Sfmbt-SAM:Scm-SAM interaction. Fifth, binding of both PhoRC and PRC1 to a Hox gene PRE in Drosophila critically depends on intact DNA-binding sites for Pho, suggesting that PhoRC constitutes the binding platform for PRC1 association with this PRE.

The Sfmbt–Scm interaction links PhoRC and PRC1

The PRC1–PhoRC assemblies isolated in this study and by Strübbe et al. (2011) were identified in affinity purifications of PRC1 from soluble nuclear extracts that had been prepared by high-salt extraction from embryonic nuclei of Drosophila. Although it is possible that PRC1 and PhoRC bind to each other in the nucleoplasm independently of DNA, we consider it more likely that the PRC1-associated PhoRC in the purified material represents PRC1–PhoRC assemblies that had formed at target gene DNA and were solubilized from chromatin during extract preparation. This scenario is also supported by ChIP profiling and proteomic analyses on cross-linked chromatin that showed that PhoRC and PRC1 colocalize at a large fraction of genomic target sites (Kwong et al. 2008; Oktaba et al. 2008; Schuettengruber et al. 2009; Kang et al. 2015).

The requirement of the Sfmbt-SAM and Scm-SAM domains for target gene repression supports the importance of their interaction. Nevertheless, we found that the Sfmbt^ΔSAM protein is not completely incapacitated for repression (Fig. 3A). It is possible that interaction of Sfmbt^ΔSAM and Scm through a second interaction site in their N termini (Grimm et al. 2009) accounts for the repression in a fraction of Sfmbt^ΔSAM mutant cells. The observation that the Scm^ΔSAM protein is nonetheless completely incapacitated for repression (Fig. 3C; Supplemental Fig. S3) is not at odds with this interpretation because deletion of the SAM domain in Scm not only compromises Scm binding to Sfmbt but also prevents Scm binding to the Ph SAM domain and thus association with PRC1 (Fig. 4B; Peterson et al. 2004; Kim et al. 2005). In this scenario, the Scm-SAM domain provides the critical physical link that mediates the interaction between PhoRC and PRC1 (Fig. 4B). It is possible that other, unidentified DNA-binding proteins are also able to recruit Scm to PREs (Wang et al. 2010). However, at least at the bxd PRE, the PhoRC–PRC1 link reported here appears to be critical for recruitment of PRC1 because binding of both Scm and Ph is strongly reduced if PhoRC binding to PRE DNA is compromised (Fig. 4A).

PRE-tethered Sfmbt-SAM:Scm-SAM as a nucleation site for short- and long-range chromatin organization

Current models suggest that PRC1 represses target gene transcription by altering chromatin organization at two different levels. First, Psc and CBX2 in Drosophila and vertebrate PRC1, respectively, act at the level of nucleosomes by compacting arrays of nucleosomes and inhibiting their remodeling in vitro (Francis et al. 2001, 2004; Grau et al. 2011). This activity relies on low-complexity regions in the Psc and CBX2 proteins that are rich in basic amino acids, and, in the case of Psc, these regions were shown to be essential for target gene repression in Drosophila (King et al. 2005). The PRC1 subunit Ph, on the other hand, appears to alter chromatin at a higher level of organization by changing chromatin topology in part by forming long-range contacts between distant Ph-bound chromosomal sites (Cheutin and Cavalli 2012; Isono et al. 2013; Boettiger et al. 2016; Wani et al. 2016). This activity requires the Ph-SAM domain and its ability to form homo-oligomers or hetero-oligomers (Isono et al. 2013; Wani et al. 2016). The capacity of Ph-SAM to form oligomers is also strictly required for repression of PcG target genes in Drosophila (Gambetta and Müller 2014). The Sfmbt-SAM:Scm-SAM interaction reported here may thus provide the molecular basis toward understanding not only how PRE-tethered PRC1 locally compacts arrays of nucleosomes in the chromatin flanking PREs but also how it interacts with other PRC1 assemblies bound at distant chromosomal sites to form topologic domains of PcG-repressed chromatin.

Materials and methods

TAP of Ph complexes

The α-tubulin1-TAP-Ph-p and α-tubulin1-TAP-Ph-d transgenes in the previously described Drosophila transformation vector CaSpeR-NTAP (Nekrasov et al. 2007) contained a 2.6-kb fragment of the α-tubulin1 gene, including promoter and 5′ untranslated region sequences, followed by the TAP tag and either the Ph-p_1–1589 or Ph-d_1–1537 coding regions (plasmid maps are available on request). TAP from 0- to 14-h-old embryonic nuclear extracts was performed as described (Klymenko et al. 2006).

Mass spectrometric analysis of protein isolated by TAP

We performed two independent purifications from extracts of TAP-Ph-d transgenic embryos and one from TAP-Ph-p transgenic embryos; in each case, a mock purification from wild-type embryos was performed in parallel. Calmodulin eluates were analyzed by SDS-PAGE, and proteins were visualized by Coomassie staining. For each sample, lanes were cut into five equal gel slices, and peptides were generated by tryptic in-gel digestion (Shevchenko et al. 2006).

Desalted peptide samples were analyzed using liquid chromatography-tandem mass spectrometry (LC-MS/MS) on a LTQ Orbitrap Velos instrument online coupled to an Easy-nLC system (both Thermo Fisher Scientific). Samples were analyzed using 65- and 90-min reversed-phase chromatography, respectively. Settings were set to 200 nL/min flow and gradient from 5% to 30% solvent B (80% ACN, 0.5% AcOH) for 43 and 65 min, respectively. Mass spectrometric settings were as follows: full scan: 10⁶ ions, 100-msec maximum filling times; MS2 scan: 4 × 10⁴ ions, 150-msec maximum filling times; isolation width: 4; minimal signal required: 10³; and NCE: 40. The 10 most intense ions were fragmented using HCD mode and fixed at m/z 100. Unassigned charges and singly charged ions were rejected from MS2 fragmentation. Dynamic exclusion was enabled.

All .RAW files were analyzed together using the MaxQuant software suite, including the Andromeda search engine (version 1.5.0.0; (Cox and Mann 2008; Tyanova et al. 2014). The analysis was performed on tryptic peptides, including fixed modifications of carbamidomethylation (Cys) and variable modifications of N-terminal acetylation and oxidation of methionine. Files were searched against the UniProt database of Drosophila (18,796 entries) as well as the self-defined FASTA sequences of Ph-d and Ph-p. Peptide, protein, and site false discovery rates were set to 0.01. Label-free algorithm (Cox et al. 2011) was included, and the “match between runs” function was enabled.

Data were filtered and further analyzed using the in-house software tool Perseus. Intensity values were filtered for contaminant and reverse entries. The three experiments were split into subfiles, and each experiment was filtered for one valid value. Missing values of the LFQ intensities were imputed using a normal distribution at the lower-intensity range. Values are represented using the free software environment R (http://www.r-project.org).

A detailed list of peptide sequences obtained from MS analysis of the proteins is presented in Supplemental Table S1.

Protein expression and purification

Baculoviruses expressing Sfmbt, Scm, and Flag-Ph_1298–1589 have been described previously (Klymenko et al. 2006; Grimm et al. 2009; Gambetta and Müller 2014). For this study, new viruses for Flag-Sfmbt_530–1220 and Flag-Sfmbt_530–1136 expression were generated after cloning the appropriate coding fragments into pFastBac. Flag affinity purification of protein (complexes) was performed as described (Grimm et al. 2009) with the following modifications: Instead of whole-cell extracts, cytosolic, nuclear, and chromatin extract were prepared in extraction buffer (25 mM HEPES at pH 7.9, 150 mM NaCl, 20 µM ZnCl₂, 10% glycerol, 0.5 mM DTT, 0.1% pefabloc-SC, 1× Complete protease inhibitor [Roche]) and subsequently combined to get the maximal amount of recombinant protein for purification. Insect cell extracts were precleared with mouse IgG-agarose beads (Sigma) for 45 min at 4°C on a rotating wheel prior to incubation with anti-Flag M2 agarose beads (Sigma) overnight at 4°C. For 12 mL of extracts, 0.3 mL of anti-Flag beads was used.

For GST pull-down assays, GST-His-Sfmbt-SAM_1137–1220, His-Scm-SAM_803–877^L855E/L859E, and His-Ph-SAM_1499–1589^{L1561E/L1565E} were cloned into pET-derived vectors, individually expressed in Escherichia coli (Rosetta [DE], Novagen), and purified by nickel affinity purification. The purified GST-His-Sfmbt-SAM_1137–1220 domain was mixed with His-Scm-SAM_803–877^L855E/L859E or His-Ph-SAM_1499–1589^{L1561E/L1565E}, coupled to glutathione beads for 2 h in binding buffer (50 mM phosphate buffer at pH 7.5, 250 mM NaCl, 10 mM MgCl₂, 10% glycerol, 0.01% IGEPAL CA-630, 1 mM pefablok-CS, 4 mM dithiothreitol [DTT], 1× Complete protease inhibitor), and then washed three times with the same buffer but containing 200 mM instead of 250 mM NaCl. Beads were then resuspended in 150 µL of 1× NuPAGE LDS sample buffer and incubated for 5 min at 95°C prior to SDS-PAGE. Binding assays with purified GST-His-GFP served as control.

The Sfmbt-SAM:Scm-SAM complex was obtained by cloning Sfmbt-SAM_1137–1220 and Scm-SAM_803–877^L855E/L859E into pET-derived vectors and coexpression in E. coli (Rosetta [DE], Novagen) as GST-His fusion and His fusion proteins, respectively. One liter of expression culture was resuspended in 25 mL of lysis buffer (50 mM phosphate buffer at pH 7.5, 250 mM NaCl, 10 mM MgCl₂, 25 mM immidazole, 10% glycerol, 0.1% Triton-X100, 1 mM pefablok-CS, 2 mM β-mercaptoethanol, 1× Complete protease inhibitor), and cells were disrupted by sonication. SAM domains were purified by nickel affinity, GST affinity, ion exchange, and size exclusion chromatography. After the last purification step, samples were concentrated with Amicon Ultra centrifugal filters (3-kDa cutoff; Millipore) to 7–15 mg/mL.

The same strategy was used to purify the Sfmbt-SAM_1137–1220:Scm-SAM_803–869–Ph-SAM_1499–1577^L1561E/L1565 complex (Supplemental Fig. S2C).

Crystallization and X-ray structure determination

All diffraction data were collected at the Swiss Light Source (SLS) synchrotron facility at beamline PXII. The data were processed with XDS (Kabsch 2010). All crystal structures were solved using Phaser from the PHENIX suite (Adams et al. 2010). The atomic models were built with Coot (Emsley et al. 2010) and refined using either the PHENIX suite (Adams et al. 2010) or the CCP4 suite (Winn et al. 2011). Validation was performed with MolProbity (Davis et al. 2007). Figures were made with PyMOL (version 1.2).

Crystals of the Sfmbt-SAM:Scm-SAM heterodimeric complex were grown at 4°C in 0.05 M Tris/HCl (pH 7.5), 4% MPD, 0.2 M ammonium acetate, and 32.5% PEG3350. They contained two copies of the complex per asymmetric unit (space group P1 2₁ 1). A complete data set of an Sfmbt-SAM:Scm-SAM crystal, flash-frozen in liquid nitrogen, was collected to a resolution of 1.975 Å. The structure was solved by molecular replacement with the Ph-SAM:Scm-SAM heterodimer (PDB ID 1PK1) as a search model. Iterative model building and refinement were done with Coot and PHENIX-refine until the R factors converged. Interaction surfaces in the final structure were analyzed by visual inspection and by making use of the PDBePISA (proteins, interfaces, structures, and assemblies) service at the European Bioinformatics Institute (http://www.ebi.ac.uk/pdbe/prot_int/pstart.html; Krissinel and Henrick 2007).

Data collection and refinement statistics for the Drosophila Sfmbt-SAM:Scm-SAM complex structure are summarized in Supplemental Table S2.

Drosophila strains, transgenes, and antibodies

Drosophila strains generated for and used in this study are listed in Supplemental Table S3.

Transgenic lines expressing Flag-Scm and Flag-Scm^ΔSAM under the control of a genomic promoter have been described (Peterson et al. 2004). For Sfmbt transgenes, a 10.5-kb fragment comprising the Sfmbt promoter, coding, and 3′ flanking regions (Berkeley Drosophila Genome Project R6 chr2L: 13,166,649–13,177,149) was cloned into a modified attB vector and integrated into the VK33 attP site. For the Sfmbt^ΔSAM transgene, the TTG codon for Leu1133 in the same genomic fragment was mutated into TAG to create a premature termination codon.

The antibodies used are listed in Supplemental Table S4.

Total embryo extracts and small-scale embryo nuclear extracts

Embryos of the appropriate genotype were identified using GFP-marked balancer chromosomes. For total extracts, dechorionated embryos were briefly sonicated in 1× NuPage LDS sample buffer (Life Technologies) and heated for 5 min at 95°C. After centrifugation at 16,000g for 5 min, the supernatant was directly analyzed by SDS-PAGE. Small-scale nuclear extracts were prepared using the subcellular fractionation kit for tissues (Thermo Scientific, no. 87790).

Clonal analysis, immunostaining, and cuticle preparations of Drosophila

Immunostaining of embryos and imaginal discs and generation of clones in discs and adults were performed in animals of the appropriate genotypes using previously described protocols (Beuchle et al. 2001).

ChIP analysis

Chromatin preparation was performed as previously described (Gambetta et al. 2009) with the following modifications: Approximately 50 wing discs were sonicated for 30 min in 130 μL of sonication buffer using Covaris S220 and the following settings: duty cycle: 2%; peak incident power: 105 W; cycles per burst: 200. N-lauroylsarcosine was then added to a final concentration of 0.5 % for 10 min at 4°C, and debris was pelleted by centrifugation. The chromatin was dialyzed overnight and then used directly or stored at −80°C.

Primer sequences used for quantitative PCR analysis are listed in Supplemental Table S5 (Supplemental Table S5).

Accession Numbers

The accession number for the coordinates and structure factors reported in this study is PDB ID 5J8Y.