Combgap contributes to recruitment of Polycomb group proteins in Drosophila

Significance

The ability of organisms to regulate gene expression spatially and temporally is a crucial aspect of development and differentiation. Polycomb group proteins (PcG) are a group of transcriptional repressors that mediate silencing of developmental genes in places where they should not be expressed. Mutations in PcG proteins have been implicated in cancer. We aim to understand mechanisms of PcG-mediated repression, and in this study we report the involvement of Combgap, a DNA-binding protein, in PcG recruitment. Combgap binds to GTGT motifs, which are present within the regulatory regions of PcG target genes. Genome-wide analyses suggest that Combgap may directly recruit Polyhomeotic, a PcG protein. Overall, our data help provide a mechanism for PcG recruitment to target genes.

Abstract

Polycomb group (PcG) proteins are responsible for maintaining the silenced transcriptional state of many developmentally regulated genes. PcG proteins are organized into multiprotein complexes that are recruited to DNA via cis-acting elements known as “Polycomb response elements” (PREs). In Drosophila, PREs consist of binding sites for many different DNA-binding proteins, some known and others unknown. Identification of these DNA-binding proteins is crucial to understanding the mechanism of PcG recruitment to PREs. We report here the identification of Combgap (Cg), a sequence-specific DNA-binding protein that is involved in recruitment of PcG proteins. Cg can bind directly to PREs via GTGT motifs and colocalizes with the PcG proteins Pleiohomeotic (Pho) and Polyhomeotic (Ph) at the majority of PREs in the genome. In addition, Cg colocalizes with Ph at a number of targets independent of Pho. Loss of Cg leads to decreased recruitment of Ph at only a subset of sites; some of these sites are binding sites for other Polycomb repressive complex 1 (PRC1) components, others are not. Our data suggest that Cg can recruit Ph in the absence of PRC1 and illustrate the diversity and redundancy of PcG protein recruitment mechanisms.

Polycomb group (PcG) proteins are a class of transcriptional regulators that maintain the silenced state of many genes involved in cellular growth, differentiation, and development (1, 2). Members of the PcG protein family form multiprotein complexes that interact with chromatin and mediate the heritable repression of gene expression by posttranslational modification of histone proteins and inhibition of chromatin remodeling (3, 4). Two of these complexes, Polycomb repressive complex 1 and 2 (PRC1 and PRC2) carry the enzymatic activities of ubiquitination and methylation, respectively. PRC1 and PRC2 do not contain DNA-binding proteins, and the mechanism of recruitment of PcG complexes to chromatin is an active area of research. In Drosophila, PRCs interact with chromatin via cis-elements known as “Polycomb response elements” (PREs) (5–7). PREs were discovered in transgenic studies by their ability to act as silencers and to maintain the repression of reporter genes throughout development (8–11). In genome-wide studies, PREs are recognized as binding sites for PRC1 and PRC2 components; there are hundreds to thousands of PREs in the Drosophila genome (12–14). Despite the many studies conducted to predict PREs computationally (15–18), it still is not possible to predict which fragments of DNA have PRE activity. Recent results suggest that PREs are a diverse group and that this diversity has functional consequences (19).

PREs are made up of binding sites for many different proteins (5). The PRE DNA-binding protein Pleiohomeotic (Pho), present in the PhoRC complex with the methyl-lysine–binding protein Sfmbt (20), plays a key role in the recruitment of PRC1 and PRC2 to PREs. Pho-binding sites have been shown to be important for PRE activity in transgenes and at the endogenous Ubx (Ultrabithorax) gene (5, 21, 22). However, studies have shown that PcG proteins can remain bound to Polycomb target genes in the absence of Pho (and its closely related gene pho-like) (23, 24), thereby suggesting the involvement of additional DNA-binding proteins that operate at PREs. Further, Pho-binding sites alone are not sufficient for PcG recruitment; rather, PREs consist of a collection of binding sites for many proteins (5). Other DNA-binding proteins implicated in PRE activity include Spps (Sp1 factor for pairing-sensitive silencing), GAF/psq (GAGA factor/pipsqueak), Zeste/fs(1)h [female sterile (1) homeotic], Grh (Grainyhead), Dsp1 (dorsal switch protein 1), and Adf1 (alcohol dehydrogenase transcription factor 1) (25–29). Genome-wide studies have shown that many of these proteins are present at only a subset of PREs, suggesting diversity in PcG protein-complex recruitment.

Our laboratory (J.A.K.) is trying to identify all the DNA-binding proteins and binding sites required within a single PRE. For this purpose we have focused on a 181-bp region (−576 to −395 bp) upstream of the engrailed (en) transcription unit where Pho-, GAF/Psq-, Dsp1-, Spps-, and Zeste/fs(1)h-binding sites are present along with sites for two unknown factors (19, 30). In an experiment designed to identify one of these two unknown factors, we identified Combgap (Cg) as a candidate PRE-binding protein. Our data show that Cg binds to the sequence GTGT, a motif that is enriched at PREs (14, 15); GTGT sequences are important for the function of a vestigial (vg) PRE in a transgene (31). Genome-wide ChIP studies show that Cg colocalizes with the PcG protein Pho and the PRC1 component Ph (Polyhomeotic) at PREs. Our data also suggest that Cg may recruit Ph independently of other PRC1 components. Finally, our data show that even though Cg is present at most PREs, PcG protein binding is decreased at only a small fraction of PREs in a cg mutant. These findings illustrate the diversity and redundancy of PcG recruitment in Drosophila.

Results

Identification of Cg as a PRE DNA-Binding Protein.

Our group has identified all the DNA-binding sites within the 181-bp fragment (enPRE2) required for pairing-sensitive silencing (PSS) of the miniwhite gene (an assay for PRE activity); however, the identity of the proteins that bind two of these sites is unknown (Fig. S1A) (30). Site A, a 17-bp region, is bound by a protein(s) in Drosophila nuclear extracts and is required for PRE activity (19, 30). To identify the protein(s) that binds site A, we performed a DNA affinity pull-down coupled to an MS assay (Materials and Methods) using a multimerized 17-bp site A sequence (Fig. S1B). As a control we conducted a similar experiment using a multimerized 17-bp oligonucleotide with a 5-bp mutation that abrogated protein binding to the 17-bp oligonucleotide and reduced the PRE activity of enPRE2 (Fig. S1B) (19). Among the long list of candidate proteins (Dataset S1), we focused on the protein Cg, which bound the WT but not the mutant probe (Table S1).

Fig. S1.

(A) The en regulatory region. Shown is a schematic diagram of the en transcription unit and upstream regulatory region showing the two PREs, PRE1, and PRE2. Also shown are the defined (Pho, GAGA, Spps, Dsp1, Grh, and Zeste) and predicted (sites A and B) binding sites within PRE2. (B) WT and mutant oligomer sequences from the en PRE2 used for pulldown experiment. The underlined region shows the suspected Protein A binding site and the red font shows the mutations that were shown to reduce pairing-sensitive silencing and the site A EMSA band shift in previous studies.

Table S1.

MS data for identification of cg

Accession no.	Coverage, %	Number of unique peptides	Peptide sequence	Mascot score	Gene name
A1Z9M3	4	2	TGQTVLTAG SAAAK; VIQGLEDNED SQGEAPNLK	120	Combgap isoform

This table shows the details of the Combgap peptides identified by LC-MS/MS studies.

The hypomorphic cg¹ allele was isolated in 1925 by Bridges and got its name from two phenotypes, an increase in the number of sex comb teeth on the first legs in males (comb) and a gap in the fourth wing vein (gap). The gene encoding cg was isolated by three groups in 2000 and encodes a protein with 11 zinc fingers and a polyQ stretch, strongly suggesting that cg has a role in DNA binding as well as in transcriptional regulation (32–34). These groups also identified lethal alleles of cg and identified cg as a genetic repressor of the GLI-family protein Cubitus interruptus (Ci) in the posterior compartment of the wing imaginal disc. Further, other studies had shown that the PcG protein Ph directly represses ci, suggesting that ci is a PcG target (35). These data suggested to us that Cg was a good candidate for a PRE DNA-binding protein. To test whether Cg could bind to enPRE2 in vivo, we performed ChIP followed by quantitative PCR (qPCR) with anti-Cg antibody on third-instar larval extracts (Fig. S2). Cg showed enriched binding to enPRE2, suggesting we had identified a PRE-binding protein in our experiment.

Fig. S2.

Cg localizes to PREs. (Upper) ChIP analysis with anti-Cg in third-instar CNS and imaginal disc tissues. Cg binds to enPRE2. Results (mean ± SD from three independent experiments) are shown as a percentage of input DNA. **P < 0.05 indicates statistically significant data (t test). (Lower) Black bars above the schematic show regions amplified by qPCR. The en PRE amplified by qPCR is also shown in the magenta box in Fig. S4A.

Cg Localizes to PREs Within the Drosophila Genome.

As an initial screen to determine whether Cg is a likely regulator of PcG target genes genome-wide, we examined whether Cg colocalizes with the PRE-binding protein Spps on polytene chromosomes. Spps completely colocalizes with the PRC1 component Psc on salivary chromosomes (29) and thus is a good marker for PcG target genes. Cg colocalized with Spps at large number of sites on polytene chromosomes, suggesting that it may act at a subset of PcG target genes (Fig. S3). To examine this possibility in more detail, we performed ChIP-sequencing (ChIP-seq) analysis for Cg, along with anti-Ph and anti-Pho antibodies in parallel, in chromatin samples from CNS and imaginal discs from WT third-instar larvae. Genome-wide, Cg peaks were present at a total of 10,088 sites, and Ph and Pho were present at 8,900 sites and 6,653 sites, respectively (peaks scored at P ≤ 0.05). All three proteins overlap at 5,135 sites, which constitute ∼50% of the Cg sites and ∼80% of the Pho sites (Fig. 1A). Outside the regions where all three proteins bind, Cg and Ph colocalize at 3,006 sites, and Cg and Pho colocalize at 214 sites in the genome. Remarkably, nearly 90% of the Ph peaks overlap with Cg peaks, and these peaks make up ∼80% of all Cg-binding sites. These data strongly suggest a role of Cg at PREs.

Fig. S3.

Cg colocalizes with Spps in polytene chromosomes. Polytene chromosomes were costained with anti-Cg (red) and anti-Spps (green). Red and green arrowheads in the merged image indicate some bands where only Cg or Spps is bound. The yellow box in the merged image highlights a region in which the Cg and Spps bands overlap completely.

Fig. 1.

Cg colocalizes with Ph and Pho genome-wide. (A) Venn diagram showing the overlap of Cg, Pho, and Ph peaks. (B) Motif specificity for Cg as identified by MEME analysis of the top 500 ChIP-seq peaks bound by Cg, Pho, and Ph.

We looked in detail at the distribution of Pho, Cg, and Ph on the well-characterized PcG targets, the Bithorax complex and the invected–engrailed (inv-en) genes. Within the inv-en domain, Cg peaks overlap with Pho and Ph peaks at all characterized PREs, including PRE2 (Fig. S4A). In the Bithorax complex, Cg colocalizes with Pho and Ph at many sites, including all known PREs (Fig. S5A). We validated Cg binding at Ubx PRE_D by ChIP-qPCR (Fig. S5C). These data show that Cg is present at PREs and suggest it may play a role in PcG protein recruitment.

Fig. S4.

Cg-binding pattern at the inv-en domain. (A) ChIP-seq profiles for Cg, Ph, and Pho at the inv-en domain. Characterized inv (PRE_A, PRE_B) and en (PRE1, PRE2) PREs within the domains are identified. Cg binding to the en PRE2 is highlighted by a magenta box. The black box shows the PRE where Pho is weakly enriched. For en PRE2, Cg binding was validated by ChIP-qPCR (Fig. S2). (B) A 10-kb region up- and downstream of en. Characterized en PREs (PRE1, PRE2) within the domains are identified. enPRE2 is highlighted by a black box. (C) A 12-kb region including the inv upstream region. Characterized inv PREs (PRE_A, PRE_B) within the domains are identified.

Fig. S5.

Cg binding at Ubx PRE_D (A) and vg PRE (B). ChIP-seq profiles for Cg, Ph, and Pho at the Bithorax complex (A) and vg locus (B). Characterized PREs within the Bithorax complex are identified. Magenta boxes highlight Cg binding to the Ubx PRE_D (A) and to the vg PRE (B). (C, Upper) Validation of Cg binding at Ubx PRE_D and vg PRE in third-instar CNS and disc tissues by ChIP analysis. Results (mean ± SD from three independent experiments) are shown as a percentage of input DNA. **P < 0.05 indicates statistically significant data (t test). (Lower) Schematics show the regions amplified in. (D) Second- and third-highest-scoring motifs found associated with Cg.

A close examination of Cg, Pho, and Ph binding in the inv-en domain reveals many differences in the binding profiles of these three proteins. Although Cg, Pho, and Ph all associate with PRE2, Cg and Ph binding extends into the en promoter region independent of Pho (Fig. S4B). The peak heights of Cg, Pho, and Ph are all approximately equal at en PRE1 but vary greatly at the inv PREs PRE_A and PRE_B (Fig. S4C). Although it is hard to say exactly what these differences in peak height mean, we suggest that these data reflect diversity in PREs.

cg behaves as a genetic repressor of ci (32, 33, 35, 36); therefore we examined Cg binding at the ci locus. Cg binds near the ci promoter, colocalizing with weak Ph and Pho peaks, and also with three additional Ph peaks located further upstream, where there are no Pho peaks (see Fig. S11A). Note that the level of H3K27me3 is insignificant throughout the ci region; therefore, ci does not appear to be a target for PRC2 in larvae. Interestingly, the upstream Ph peaks were partially contained within a reporter construct that has been shown to recruit Ph to a ci-reporter transgene (35). These data suggest that Cg may participate in the recruitment of Ph in the absence of PRC2.

Fig. S11.

Cg recruitment of Ph is independent of PRC1 and PRC2. ChIP-Seq profile for Cg, Ph, and Pho in WT and Ph and H3K27me3 in cg^A22 mutants at the ci (A) and cg4562 (B) loci. Black boxes indicate primer positions used for qPCR in Fig. 5.

Cg Binds to the GTGT Motif Present in PREs.

Cg contains multiple zinc fingers of the DNA-binding types (32–34), suggesting it might bind DNA directly. To identify the sequence motifs where Cg binding takes place, we performed a motif search within the top 500 Cg peaks that also bound Pho and Ph, using the MEME program (37). Our analysis identified a GTGT motif as the highest-scoring motif with an E-value of 3.5e-194 (Fig. 1B). An analysis of the top 500 peaks that bound only Cg also identified the same motif. A GTGT motif was initially identified in predicted PREs (15) and later was found independently to be enriched in genome-wide ChIP-on-chip datasets for PRC1 components and PcG recruiter proteins (14). The two next-highest-scoring motifs were GAGAG, another motif found associated with PREs, and an A-rich sequence (Fig. S5D).

A 1.6-kb PRE from vg, a known PcG target essential for wing development (38–40), is highly enriched in this GTGT motif (31). This PRE was bound by Cg in our ChIP-seq experiment (Fig. S5B) and also in ChIP-qPCR experiments (Fig. S5C). Next, we tested Cg binding to the vg PRE by a gel-shift assay. A 106-bp region from the 1.6-kb vg PRE containing four GTGT motifs is known to be essential for mediating PRE function (31). A 44-bp region containing the GTGT motifs from this vg PRE (Fig. 2A) shifted a band in S2 nuclear extracts (Fig. 2B, Left, lane1). The shifted band is made up of different bands (and appears as a smear), likely because more than one protein binds this probe. Notably, at least one band is supershifted on the addition of anti-Cg antibody (Fig. 2B, Left, lane 2) but not by addition of anti-Pho antibody (Fig. 2B, Left, lane 3), suggesting that at least one of the proteins bound to the probe is Cg. To test directly whether the Cg protein binds to the vg PRE probe, we performed gel shift experiments with recombinant Cg protein (Fig. 2B, Right). The region encoding the zinc fingers (amino acids 810–2100) of the Cg protein was in vitro transcribed and translated and used in a gel shift assay. As a control, we included a sample that contained the vector alone (T7 control), in a parallel in vitro transcription/translation reaction. The recombinant protein specifically shifted a band (Fig. 2B, Right, lane 2) that was competed by 200-fold molar excess of the WT oligonucleotide but not by an oligonucleotide in which the GTGT motifs had been mutated (Fig. 2B, Right, compare lanes 3 and 4). This result suggests that the Cg protein binds to the vg PRE via the GTGT motifs. PRE1 of en (∼1 kb upstream of enPRE2) (Fig. S4 A and B) (19) contains a region with three GTGT motifs (Fig. S6A). A 111-bp oligonucleotide containing these GTGT motifs specifically competed with the vg PRE probe in a band shift experiment (Fig. S6B). Mutations in the GTGT motifs in this probe abrogated this competition (Fig. S6B, compare lanes 2 and 3).

Fig. 2.

Cg binds to PREs with multiple GTGT motifs. (A) Sequence of the 44-bp oligonucleotide containing four GTGT motifs (boxed) from the vg PRE used as probe for the gel shift assay. Arrows indicate the residues changed in the mutant oligomer. (B, Left) Gel supershift assay without antibody (lane 1), with anti-Cg antibody (lane 2), and with anti-Pho antibody (lane 3). (Right) Gel shift assay with reticulocyte lysate with vector alone (T7 control, lane 1) and with in vitro-translated Cg protein (lanes 2–4). Lane 3 contains a 200× molar excess of vg PRE cold competition, and lane 4 contains a 200× molar excess of mutated vg PRE cold competition.

Fig. S6.

Cg binds to PREs with multiple GTGT motifs. (A) The 111-bp oligonucleotide containing three GTGT motifs (boxed) from the en PRE1 that was used for competition in the EMSA. Arrows indicate the residues mutated in the mutant oligomer. (B) Gel shift assay with vg PRE only (lane 1) and using a 200× molar excess cold WT (lane 2) and mutated en PRE1 oligomers (lane 3) as competitors to vg PRE.

It is worth mentioning here that the oligonucleotide from enPRE2 that was used for the DNA affinity pulldown did not contain GTGT motifs (Fig. S1B). In addition, in vitro-transcribed/translated Cg did not bind to the probe used in the pull-down experiment (Fig. S7). This result raises the question, how was Cg precipitated in the pulldown? Because Cg is detected by ChIP at en PRE2 (Fig. S2), we suggest that Cg interacts with enPRE2 via protein–protein interactions with DNA-binding proteins that directly bind the probe. Taken together our data indicate that Cg binds to the GTGT motif in PREs of PcG target genes but also may be recruited via protein–protein interactions.

Fig. S7.

Interaction of in vitro-translated Combgap protein with the oligonucleotide used in the affinity pulldown. (Left) The interaction of the vg probe (as shown in Fig. 2), with in vitro-translated Cg protein. (Right) The gel shift using the pulldown oligonucleotide as a probe (MS WT probe in lane 4, sequence as shown in Fig. S1B). Lanes 1 and 3 are control reactions containing the probe and TNT reticulocyte lysate (which is used for in vitro transcription/translation reactions), respectively.

Cg Is Required for Recruitment of PcG Proteins to a Subset of PREs.

Cg is ubiquitously expressed in many tissues, and 0- to 2-h embryos contain a large amount of Cg RNA, showing that Cg is deposited maternally in the egg (41). Females homozygous for the cg¹ allele are sterile, suggesting that cg may be required for oogenesis. To test this notion further, we made germline clones in female with the hypomorphic cg¹ allele and cg^A22 allele (32) and found that very few eggs were laid by females of either genotype (SI Materials and Methods); this finding suggests that Cg plays an important role in oogenesis. In contrast, cg^A22 zygotic mutants die late in development, often surviving to pupal stages. cg^A22 third-instar larvae are smaller than WT larvae and emerge from the food 3 d later than WT larvae (8 d versus 5 d), showing that they develop more slowly (Fig. S8A). The imaginal discs of cg^A22 larvae also are smaller than WT discs but look normal in morphology and, as previously reported, correctly express En (Fig. S8B). A Western blot of lysates from the CNS and imaginal discs of WT and cg^A22-mutant third-instar larvae revealed a single band of ∼85 kDa (Fig. S9A). Note that a small amount of Cg (∼5–10% of the WT level) is present in cg^A22 homozygous third-instar larvae (Fig. S9A); we suggest that this protein results from perdurance of maternally deposited Cg.

Fig. S8.

cg mutants show developmental defects. (A) Heterozygous cg^A22/CyO (Left) and homozygous (Right) cg^A22 third-instar larvae. cg^A22 homozygotes are developmentally delayed and die as small pupae. AEL, after egg laying. (B) DAPI, Abd-B, and En staining of wing imaginal discs from heterozygous (Upper) and homozygous (Lower) cg^A22 third-instar larvae. Note that the wing disc of the homozygous cg^A22 is much smaller than that of its heterozygous counterpart. There is no expression of Abd-B in these discs.

Fig. S9.

Examination of Cg and Ph levels in cg^A22 mutants. (A) Total extracts from CNS and imaginal disc tissues from WT and cg^A22 third-instar larvae probed with anti-Cg antibody. Each lane contains lysate from approximately two third-instar larvae. β-Actin was used as an internal loading control. Note that a light band of Cg is visible in the cg^A22 lane. We suspect this band represents maternally deposited Cg protein. (B) Ph protein level does not differ in WT and cg^A22 third-instar larvae. (C) ph-p (ph-proximal) and ph-d (ph-distal) mRNA levels from WT and cg^A22-mutant larvae determined by qRT-PCR. Results (mean ± SD from three independent experiments) are shown as normalized to Rpl32 mRNA levels. n.s., nonsignificant difference as estimated by t test.

We reasoned that if Cg plays a role in PcG recruitment to PREs, then PcG protein binding to PREs might be decreased in a cg mutant. We examined the distribution of the PRC1 component, Ph, in cg^A22-mutant larvae. Importantly, Ph protein and RNA levels do not differ between cg^A22 and WT third-instar larvae (Fig. S9 B and C). A genome-wide Ph ChIP-seq analysis was performed on CNS and imaginal discs from third-instar cg^A22 mutants and was compared with Ph binding in a similar sample from WT larvae. Ph binding was altered at 434 sites [at a false-discovery rate (FDR) <0.2] (Dataset S2), a relatively small proportion of the total number of Ph-binding regions. We suggest that the relatively low number of Ph peaks that changed in cg-mutant larvae results from two factors: (i) the persistence of a low level of Cg protein in cg^A22 larvae and (ii) the combinatorial nature of recruitment of PcG proteins in Drosophila.

Turning our focus to the 434 peaks that changed in the cg-mutant larvae, we found that Cg peaks were present at 418 of these sites in WT larvae, suggesting that Cg plays a direct role in Ph binding. Further, 92 of the top 109 differentially bound sites (FDR <0.02) showed a decrease in Ph binding (90 were cobound by Cg), and 35 were in H3K27me3 domains, a mark indicative of PcG-target genes. We examined the decrease in Ph binding at two sites within PcG target genes, abdominal-B (Abd-B) and pox meso (poxm) in more detail (Fig. 3 and Fig. S10A) and validated it in three additional biological samples using ChIP-qPCR (Fig. S10 B and C). Note that Ph binding to only one PRE of the Abd-B transcription unit is significantly depleted (Fig. 3). Abd-B was not misexpressed in cg^A22-mutant wing discs (Fig. S8B), suggesting that the activity of the other PREs in the Abd-B domain were sufficient for PcG repression of Abd-B. Consistent with this notion, H3K27me3 was present over the Abd-B domain in cg^A22-mutant larvae (Fig. 3), although the H3K27me3 level flanking the affected PRE was decreased in cg mutants (Fig. 3, boxed region). Similarly, H3K27me3 also was present over the poxm locus in cg^A22 larvae; however, the height and pattern of H3K27me3 in the region surrounding the affected Ph peaks were altered (Fig. S10A). This trend of local depletion of H3K27me3 levels surrounding depleted Ph peaks was observed consistently in the cg-mutant data. These data suggest that recruitment of PRC2, which deposits on the H3K27me3 mark, is also depleted in cg mutants.

Fig. 3.

Decreased Ph binding at a subset of sites in the Abd-B gene in cg^A22-mutant larvae. Cg, Ph, Pho, and H3K27me3 in WT and Ph and H3K27me3 in cg^A22 at the Abd-B gene in CNS and imaginal discs from third-instar larvae. Red boxes indicate the reduced local H3K27me3 domains under the Ph peak that is significantly depleted in the cg mutant. The black box indicates the Abd-B PRE peak amplified in Fig. 5A and Fig. S8B.

Fig. S10.

Ph levels are reduced at Cg targets in cg mutants. (A) Cg, Ph, Pho, and H3K27me3 in WT and Ph and H3K27me3 ChIP-Seq profiles in cg^A22 mutants at the poxm gene in CNS and imaginal discs from third-instar larvae. Red boxes indicate the reduced local H3K27me3 domains under the Ph peak that is significantly depleted in the cg mutant. poxm PRE is highlighted by a black box. (B and C, Upper) ChIP-qPCR analysis was performed with anti-Ph antibody in CNS and imaginal disc tissues from WT and cg^A22 third-instar larvae at Abd-B (B) and poxm (C) loci. Results (mean ± SD from three independent experiments) are shown as a percentage of input DNA. **P < 0.05 indicates statistically significant data (t test). (Lower) Black bars above the schematic indicate regions amplified by qPCR. The Abd-B PRE amplified by qPCR is also shown by a black box in Fig. 3.

We next examined if the binding of Psc, another PRC1 component, also was reduced at selected targets in cg^A22 third-instar larvae. Similar to Ph, Psc binding at Abd-B and poxm was reduced, whereas binding to the en PRE was similar to that in WT (Fig. 4A). Coimmunoprecipitation experiments using S2 nuclear extracts showed an interaction of Cg with Psc, Pc, and Ph (Fig. 4B), suggesting that Cg may interact with the PRC1 complex. Interestingly, Kang and coworkers (42) recovered Cg as an enriched protein in cross-linked affinity purifications of BioTAP-tagged Pc and E(z) proteins in embryos. Based on these results we propose that Cg, in combination with other PRE DNA-binding proteins, is involved in the recruitment of PRC1 and PRC2 throughout the Drosophila genome.

Fig. 4.

Cg interacts with PRC1 components. (A) ChIP-qPCR with anti-Psc antibody on en, Abd-B, and poxm PREs in CNS and imaginal discs from WT and cg^A22-mutant third-instar larvae. Bars above the schematic show the region amplified by qPCR. The location of en, Abd-B, and poxm PREs amplified by qPCR are also shown in the magenta box in Fig. S4A and in the black boxes in Fig. 3 and Fig. S10A, respectively. Results (mean ± SD from three independent experiments) are shown as a percentage of input DNA. **P < 0.05 and *P < 0.1 indicate statistically significant data (t test). (B, Upper) Western blot analysis of anti-Pc, -Psc, and -Ph antibodies and nonspecific IgG immunoprecipitates derived from S2 nuclear extracts for the presence of Cg. (Lower) Results of Western blot analysis of anti-Cg antibody immunoprecipitate for the presence of Pc, Psc, or Ph in nuclear embryo extracts.

Cg Recruits Ph to Some Sites in the Absence of Psc and Pc.

Cg and Ph colocalize to many sites in the genome independent of Pho, and many of these sites are not in H3K27me3 domains. Similarly, Schaaf et al. (43) found that a large number of Ph sites occur outside H3K27me3 domains. Our data suggest that Cg may play a role in recruiting Ph to these sites. Of the top 90 Ph peaks that are reduced in a cg mutant, 55 are not in H3K27me3 domains. Two of the top Ph peaks that changed were at the promoters of the genes protein phosphatase 2B at 14D (Pp2b-14D) and CG4562 (Fig. 5A, Fig. S11B, and Dataset S2). Reduction of Ph at these sites in cg mutants is confirmed further by ChIP-qPCR (Fig. 5B). Similar to the results from another study (43) ChIP-qPCR showed that no Psc was present at these sites (Fig. 5C). In addition, another PRC1 component, Pc, is also not present at these sites (Fig. 5D). These data suggest that Cg is an important player in Ph recruitment independent of PRC1 and PRC2 complexes.

Fig. 5.

Cg recruits Ph independently of PRC1 and PRC2. (A) ChIP-seq profiles of Cg, Ph, Pho, and H3K27me3 in WT and of Ph and H3K27me3 in cg^A22 mutants at Pp2b-14D genes in CNS and imaginal discs from third-instar larvae. (B) ChIP-qPCR with anti-Ph antibody in CNS and imaginal disc tissues from third-instar WT lavae and cg^A22 mutants on fragments from ci, Pp2b-14D, and CG4562 loci. Black boxes indicate the fragments used for qPCR analysis here and in Fig. S11. The fragments from the en gene shown in Fig. S2 were used as controls. Results (mean ± SD from two independent experiments) are shown as a percentage of input DNA. (C) ChIP-qPCR using Psc antibody on WT larvae using the probes in B. Psc does not bind to these sites; en PRE2 and enCon (en control) provide the positive and negative controls, respectively, for these samples. Results (mean ± SD from two independent experiments) are shown as a percentage of input DNA. (D) ChIP-qPCR using antibody to Pc on WT larvae and the probes used in B. Pc does not bind to these sites; en PRE2 and enCon provide the positive and negative controls, respectively. Results (mean ± SD from two independent experiments) are shown as a percentage of input DNA.

We also examined Ph binding to the ci locus in the cg mutant. Ph-binding levels are very low at ci. However, examination of Ph binding in a cg mutant by genome browser in ChIP-Seq data (Fig. S11A) and by ChIP-qPCR (Fig. 5B) showed that Ph levels were reduced significantly in a cg mutant. We also note that no Psc or Ph was bound to the ci peaks (Fig. 5 C and D), suggesting that Ph acts independently of PRC1 at this gene. It has long been known that PcG mutants do not have identical phenotypes and must have independent functions (44). Cg may play an important role in mediating PRC1-independent Ph functions.

Genetic Interaction Studies with Cg.

PcG genes were discovered by their homeotic phenotypes (45), but it is now known that PcG proteins regulate hundreds to thousands of genes in the fly. The phenotype of cg¹ is not a typical PcG phenotype but reflects the action of PcG proteins at other targets, including ci. Because Cg-binding sites overlap extensively with Pho-binding sites genome-wide, we reasoned that cg and pho might interact genetically. Loss of one copy of pho did not enhance the phenotype of cg^A22-mutant larvae; similarly, loss of one copy of cg did not enhance the pho¹ phenotype. However, cg^A22; pho¹ third-instar larvae never go on to pupate, dying at an earlier stage than either single mutant (see SI Materials and Methods for details).

We also looked at the interaction between Spps and cg. Spps¹ homozygotes die as pharate adults or just after hatching and have no homeotic phenotypes. Loss of one copy of cg did not alter the phenotype of Spps homozygotes; however, cg^A22; Spps¹ third-instar larvae do not survive (SI Materials and Methods), dying earlier than either single mutant. Thus, cg interacts genetically with both pho and Spps.

SI Materials and Methods

DNA Affinity Pulldown and Mass Spectrometry.

For isolation of PRE DNA-binding proteins, complementary oligonucleotides containing a 3× binding site (WT, AGAGGGAGTGAACAGTGC, and mutated, AGAGGGATACCTCAGTGC) were synthesized with 5′ Biotin tags from Integrated DNA Technologies. Oligonucleotides were annealed to make double-stranded bait oligomers and were incubated with Dynabeads M280 (Invitrogen) at a concentration of 0.125 μg/μL in buffer A [10 mM Tris (pH 7.5), 1 mM EDTA, 1 M NaCl] for 20 min at room temperature. Following this step, beads were washed with buffer B [20 mM Hepes (pH 7.6), 100 mM KCl, 20% glycerol, 1.5 mM MgCl2, 0.5 mM DTT, 0.2 mM EDTA] and were added to 800 μg of precleared nuclear extract from 0- to 24-h embryos along with 40× poly(dI-dC) (by weight) over the oligonucleotide, 5 μg/μL BSA, 2 mM spermidine, and 10 μg tRNA, and were incubated at 4 °C for 2 h. Reaction mixes were clarified using magnetic stands, washed three times with buffer B to remove nonspecific binding, and treated with DNase (Turbo DNase; Ambion) for 30 min at 37 °C to elute the proteins. Then the beads were boiled in SDS sample buffer, and the supernatant was collected. The DNase elutions and SDS buffer elutions were loaded on gel, stained with Coomassie, and processed for MS analysis.

Gel Shift Assay.

For gel shift reactions, 15 μg of S2 nuclear extract (Active Motif) were incubated in a total volume of 15 μL with 0.5 pmol ³²P-labeled oligonucleotide, 10 μg tRNA, 4 μg poly(dI-dC), 100 mm KCl, 35 mm Hepes (pH 7.9), 1 mm DTT, 50 μm ZnCl₂, 12% glycerol, 2 mm spermidine, and 5 mg/mL BSA. Competing oligonucleotides were added at a 200× molar excess. Samples were incubated for 20 min at 25 °C, supplemented with 2.5 μL 4% BSA and 2 μL gel-loading buffer [2.5% Ficoll 400, 0.5× Tris/borate/EDTA (TBE), and tracking dyes], and then electrophoresed on a 1% agarose, 0.5× TBE gel (50 mA) until the lower dye (bromophenol blue) ran three-fourths of the gel. The gel was dried on DE81 paper (Whatman) and autoradiographed. Recombinant proteins were synthesized in vitro using the TNT-coupled transcription/translation system (Promega) according to the manufacturer’s instructions. Briefly, 1–2 μg of T7 vector with the respective gene (region) cloned in was incubated in the TNT reticulocyte lysate for 90 min at 30 °C. Gel mobility shift assays were performed using 3 μL of the in vitro translation reaction or TNT reticulocyte lysate (in control reactions).

Coimmunoprecipitation.

S2 nuclear extracts were incubated with respective antibodies in coimmunoprecipitation buffer (PBS, 2.5 mM MgCl₂, 3 mM KCl, 0.01% Nonidet P-40) for 3 h at 4 °C, followed by the addition of Protein A Sepharose beads for 1 h. The beads were washed three times with the coimmunoprecipitation buffer to remove nonspecific interactions and then were precipitated and eluted with 1 M NaCl in coimmunoprecipitation buffer. Samples were loaded onto Tris⋅glycine gels (Bio-Rad) and were processed for immunoblotting using the standard protocol.

Germline Clones.

Three independent w*; P[FRT(w^hs)]G13 cg^A22 recombinant females were crossed to P[hsFLP12], y¹ w*, P[FRT(w^hs)}G13 P[ovoD1-18]2R/CyO males; every 24 h flies were transferred to a new vial. Progeny were heat-shocked at 37 °C for 60 min 24 and 48 h after parents were transferred. Virgin females of the genotype w*/ y¹ w*; P[FRT(w^hs)]G13 cg^A22/P[FRT(w^hs)}G13 P[ovoD1-18]2R were crossed to w* P[neoFRT]42D cg^A22/CyO males. Very few eggs (10 eggs from 20 females over a 10-d period) were laid, suggesting that cg is required for oogenesis. None of the eggs hatched. Similar results were obtained with one w*; P[FRT(w^hs)]G13 cg¹ recombinant. As a positive control for the generation of germline clones, sibling P[hsFLP12], y¹ w*, P[FRT(w^hs)}G13 P[ovoD1-18]2R/CyO males were crossed to w*; P[FRT(w^hs)]G13 virgin females and subjected to the same heat-shock protocol. The P[FRT(w^hs)]G13/P[FRT(w^hs)}G13 P[ovoD1-18]2R females laid viable eggs.

ChIP-Seq and Data Analysis.

ChIP samples were processed, ligated to linkers, and PCR amplified, and DNA was purified using a Qiagen kit to synthesize the library as described previously (https://ethanomics.wordpress.com/chip-seq-library-construction-using-the-illumina-truseq-adapters). Briefly, single-end sequence reads of 50-bp length were aligned to the Drosophila genome version dm5 using ELAND. Peaks for Pho, Ph, and Cg were detected using MACS v1.4 with the –nomodel and –nolambda flags turned on, a fragment size of 200 bp, and a P value cutoff of 0.05. Peak calling was performed separately for each replicate, and peak calls for two replicates were subsequently merged (union) to produce a consensus peak list. For motif identification by MEME analysis, 39-bp sequences flanking the peak summit (peak summit ± 19 bp) of the top 500 Cg peaks that also bound Ph and Pho and the top 500 peaks that bound Cg alone were used as input for MEME motif finder. For analysis of Ph binding in cg^A22 mutants, differential binding analysis was performed using the DESeq2 package in R, and statistical significance was defined at an FDR of 0.2. For visualization, ChIP-seq data were normalized into reads per million using Integrated Genomics Viewer tools. Reads were deposited in the National Center for Biotechnology Information Gene Expression Omnibus under accession numbers GSE76892 and GSE77582.

Fly Stocks and Genetic Interaction Studies.

The full genotype of the larvae used in the ChIP-seq experiment was w-; FRT42D cg^A22 (a kind gift of Gerard Campbell, University of Pittsburgh, Pittsburgh). The stock used for the cg and pho genetic interaction experiment was yw; FRT42D cg^A22/CyO, GFP; pho¹/Dp(1;4)1021 y+ sv^spa-pol. From this stock, pho¹ homozygotes lack the yellow (y) gene and are a different color than pho¹/Dp(1;4)1021 y+ sv^spa-pol or Dp(1;4)1021 y+ sv^spa-pol homozygotes. cg^A22 homozygotes could be recognized by the lack of GFP.

The Spps¹ mutation was generated by homologous recombination, replacing the Spps coding region with armadillo-GFP; therefore GFP could not be used as a marker for the second chromosome for this cross. We made two balanced stocks that could produce cg^A22; Spps¹ double-mutant larvae: (i) w-; FRT42D cg^A22/CyO, P{w[+mC]=sChFP}2; Spps¹/TM6BTb,Hu (in this case, cg^A22; Spps¹ homozygotes could be recognized as non-Tb larvae which were negative for the fluorescence marker Cherry; no such wandering third-instar larvae were found) and (ii) w-; FRT42D cg^A22/P{w[+mC]=2xTb (1)-RFP}CyO; Spps¹/TM6BTb, Hu (in this case, cg^A22; Spps¹ homozygotes could be recognized as non-Tb larvae; no such third-instar larvae were found).

Discussion

Here we report the identification of Cg as a PRE-binding protein. In our gel shift analysis, in vitro-translated Cg bound to GTGT motifs in the PREs of vg and en. The GTGT motif was identified as an enriched motif in a PRE/TRE prediction study and also in genome-wide ChIP-on-chip experiments identifying PcG protein peaks in embryos (14, 15); further, deletion of GTGT motifs from a vg PRE led to the loss of PSS (31). Our studies show that Cg binds to GTGT motifs present in PREs.

In vitro-transcribed/translated Cg does not bind to the oligonucleotide used in the DNA-affinity column that identified Cg (Fig. S7), nor is a GTGT motif present at enPRE2, although ChIP experiments show that Cg binds to that PRE in vivo. These data suggest that Cg can be recruited to PREs via direct sequence binding or via protein–protein interactions.

Cg colocalizes with Ph extensively at both classical PcG-target genes (H3K27me3 domains) and throughout the genome. In cg-mutant larvae, we found a decrease in Ph binding levels at some peaks within H3K27me3 domains and a concomitant decrease in H3K27me3 levels in the vicinity of the decreased Ph peaks. However, many Cg/Ph peaks are not in H3K27me3 domains, and one of these is the Cg target gene ci. In genetic experiments both cg and ph act as repressors of ci. Our data suggest that this repression is direct and occurs in the absence of PRC1 and PRC2. We suggest that Cg plays a role in Ph recruitment independent of PRC1 and PRC2 complexes at many locations in the genome.

Recruitment of PcG complexes by PRE DNA-binding proteins is a complex process dependent on many DNA-binding proteins. Interactions between PRC1 and PRC2 proteins and PRE DNA-binding proteins stabilize each other and facilitate recruitment. The H3K27me3 and H2AK118ub marks also play a role in stabilizing and recruiting PRC1 and PRC2 components to the DNA. Increasing evidence indicates that PcG recruitment may be a result of cooperative and combinatorial binding of factors (24, 25, 46). The number of proteins implicated in PcG protein recruitment is continually growing. As in mammals, our results suggest that PcG proteins in Drosophila can interact with a diverse array of DNA-binding proteins and can be recruited by multiple mechanisms.

Materials and Methods

DNA Affinity Pulldown.

DNA affinity precipitation assays using biotinylated oligonucleotides were performed essentially as described by Franza et al. (47) with minor modifications. See SI Materials and Methods for detailed methods.

Polytene Squash.

Squashes and immunofluorescent staining of polytene chromosomes were performed as described previously (23) using anti-Cg (1:200) (33) and anti-Spps (1:200). Anti-guinea pig Alexa Fluor 488 (1:400) and anti-rabbit Alexa Fluor 555 (1:500) (Invitrogen) were used as secondary antibodies.

ChIP.

ChIP reactions were performed using CNS and disc tissues from third-instar larvae as described previously (48), with anti-Cg (1:100), anti-Pho (1:100), and anti-Ph antibodies (1:100) and the Millipore Chromatin Immunoprecipitation Assay Kit. Each sample consisted of imaginal discs and CNS from 10 larvae. The primers used are shown in Table S2.

Table S2.

Primers used in this study

Name	Sequence
enCon forward	CGCCTTAAGGTGAGATTCAGTT
enCon reverse	GGCGGTGTCAATATTTTGGT
enPRE forward	GCTTATGAAAAGTGTCTGTG
enPRE reverse	GGGGCTTGTTAGGCAGCAAT
UbxCon forward	CCAGCATAAAACCGAAAGGA
UbxCon reverse	CGCCAAACATTCAGAGGATAG
PRED forward	CGAAATGCTACTGCTCTCTA
PRED reverse	GCGTAGTCTTATCT GTATCT
vgCon forward	ACTGCTTGGCAGCAATGT
vgCon reverse	CCTCTTCTTTGGTTTGATGC
vgPRE forward	TTTGGCAAAAGTCAAGGAAG
vgPRE reverse	TTGAGCTCGCTCTCTCTCAT
Abd-B-PRE forward	CACCAACTGCCATCAACATA
Abd-B-PRE reverse	CATAACATCAACCACCACCA
PoxmPRE forward	CCTGATCGCTGAAAGACATT
PoxmPRE reverse	ATCCCAAGATCTCGCTCTCT
Pp2b-14D forward	CTTCCACTAGCTCGCTCTCC
Pp2b-14D reverse	CAGCCAGGGAAAGTACCAGA
cg4562 forward	CGGCAGCTTCTCGTTAAAACT
cg4562 reverse	CTCTCTTTGGCGATCTCTCG
Ci-Prom forward	ATGCCATTGTCCAAACACAG
Ci-Prom reverse	CGTGATCCGTAGGTACACACA
Ci-PromU forward1	CAAAACTCTATGAAGAAGCGGACA
Ci-PromU reverse1	CATAAAGGGTCTGGAAAATCAC
Ci-PromU forward2	AGCCTTACTGAGGTTAGCCTTT
Ci-PromU reverse2	ATGTATATTGACGCTTGTGGAA

Gel Shift Assay.

Probes were prepared by end labeling with T4 kinase (New England Biolabs) according to the manufacturer’s instructions. Gel shift reactions were performed as described previously (30) with S2 nuclear extract (Active Motif) or in vitro-translated protein. See SI Materials and Methods for the detailed protocol. For gel supershift assays 2 μL of the respective antibodies were added to the reaction mixture along with the nuclear extract and were incubated on ice for 1 h; after this incubation the probe was added, and the gel shift assay was performed.

Coimmunoprecipitation and Western Blotting.

Reactions were performed essentially as described by Mohd-Sarip et al. (49). Antibodies were used for Western blotting at the following dilutions: anti-Cg (1:1,000), anti–β-actin (1:1,000), anti-Ph (1:1,000), anti-Psc (1:40), and anti-Pc (1:200). Reactions were performed with S2 nuclear extracts (Active Motif) or embryonic nuclear extracts essentially as described by Mohd-Sarip et al. (49). For detecting Cg, third-instar CNS and imaginal disc tissues were ground in lysis buffer (PBS, 2.5 mM MgCl₂, 3 mM KCl, 0.01% Tx-100, and protease inhibitor) and processed as described above.