Global changes of H3K27me3 domains and Polycomb group protein distribution in the absence of recruiters Spps or Pho

Significance

How Polycomb group (PcG) proteins are precisely recruited to their target genes remains poorly understood. In Drosophila, PcG proteins are recruited to Polycomb response elements (PREs), composed of binding sites for multiple DNA-binding proteins. To understand how PcG proteins are recruited to and maintained at PREs, we systematically investigated PcG binding, associated H3K27me3, and transcriptome in wild type and mutants for three PRE-binding proteins. We show two factors are essential for high levels of H3K27me3 in PcG domains. Loss of H3K27me3 does not automatically result in gene expression. Different PREs respond differently to the loss of one factor. Many Polycomb domains contain different types of PREs. The diverse and combinatorial nature of PREs contributes to the remarkable resiliency of PcG repression.

Abstract

Polycomb group (PcG) proteins maintain the silenced state of key developmental genes in animals, but how these proteins are recruited to specific regions of the genome is still poorly understood. In Drosophila, PcG proteins are recruited to Polycomb response elements (PREs) that include combinations of sites for sequence specific DNA binding “PcG recruiters,” including Pho, Cg, and Spps. To understand their roles in PcG recruitment, we compared Pho-, Cg-, and Spps-binding sites against H3K27me3 and key PcG proteins by ChIP-seq in wild-type and mutant third instar larvae. H3K27me3 in canonical Polycomb domains is decreased after the reduction of any recruiter. Reduction of Spps and Pho, but not Cg, causes the redistribution of H3K27me3 to heterochromatin. Regions with dramatically depleted H3K27me3 after Spps knockout are usually accompanied by decreased Pho binding, suggesting their cooperative binding. PcG recruiters, the PRC2 component E(z), and the PRC1 components Psc and Ph cobind thousands of active genes outside of H3K27me3 domains. This study demonstrates the importance of distinct PcG recruiters for the establishment of unique Polycomb domains. Different PcG recruiters can act both cooperatively and independently at specific PcG target genes, highlighting the complexity and diversity of PcG recruitment mechanisms.

Polycomb group (PcG) proteins are important regulators of developmental gene expression in metazoans. In Drosophila, about 15 genes are classified as PcG genes and are known to act as repressors of homeotic genes and other key developmental genes (1). Early biochemical studies identified two stable protein complexes, PRC1 and PRC2, that act together to stably repress gene expression through modifications of chromatin. PRC2 contains the proteins Caf1–55, Su(z)12, Esc (or Escl), and the histone methyltransferase E(z) that trimethylates histone H3 on lysine 27 (H3K27me3). PRC1 contains the proteins Psc or Su(z)2, Ph, Sce (also known as dRing), and Pc. Pc binds to the H3K27me3 mark, which often extends many tens and even hundreds of kilobases to create silenced “Polycomb domains.” Genetic and biochemical evidence suggest that Psc [and Su(z)2], Ph, and Sce/dRing have additional roles outside of the PRC1 protein complex (2–4). In mammals, most PcG proteins have numerous paralogs, which result in multiple variants for PcG complexes. Variants for PRC1 exist both in Drosophila and mammals (5).

How are Polycomb domains established? In Drosophila, specific DNA fragments, Polycomb group response elements (PREs), serve as anchors for the recruitment and continued association of PRC1 and PRC2 with chromatin. PREs contain binding sites for the zinc finger protein Pho and many other DNA-binding proteins (6). pho zygotic mutants die as pharate adults with homeotic phenotypes, making pho a bona fide member of the PcG group of genes (7). Pho forms a protein complex with dSfmbt (PhoRC) (8). Pho binds directly to binding sites within PREs (9), and dSfmbt can interact directly with Ph and the PRC1- and PRC2-associated protein Scm through SAM domains, suggesting that PhoRC plays a direct role in Scm and PRC1 recruitment (10). Other studies suggest a direct role of Pho in PRC2 recruitment (11). In fact, mutation of Pho-binding sites in a PRE in a transgene in vivo completely abrogates both PRC1 and PRC2 binding (10). PRC1 also stabilizes PhoRC binding, suggesting there is no clear hierarchy of PcG complex recruitment to many PREs (12, 13).

Pho-binding sites alone are not sufficient for PcG recruitment. PREs are made up of binding sites of a complex array of DNA-binding proteins, including Pho/Phol, Spps, Cg, GAF/Psq, Adf-1, Grh, Dsp1, and Zeste/Fsh-S (9, 14–23). The role of these proteins in PcG protein recruitment is not well understood. Spps is a member of the Sp1/KLF zinc finger protein family that binds PREs. Sp1-binding sites are required for the activity of a PRE from the engrailed gene in two different transgene assays (19, 24) and Spps is required for transgene silencing late in development (19). Maternal, zygotic Spps mutants develop into adults that die just before or after they emerge from the pupal case and look phenotypically normal (19). This allows the examination of PcG protein distribution in third instar larvae that have never had Spps. In contrast, pho maternal mutants die as embryos, and maternal cg mutants do not lay eggs. However, zygotic mutants of both develop into pharate adults and form healthy third instar larvae (7, 21). We compare the role of Spps, Pho, and Cg in PcG protein recruitment in vivo. Our goal is to understand how these three proteins might work together, and separately, to recruit and maintain PcG proteins binding to PREs. Our results indicate a key role for Spps in PcG recruitment or stability. Like pho mutants, Spps mutants show a global reduction in H3K27me3 levels within the canonical H3K27me3 domains with a drastic reduction at a small subset of targets. Coincident with these large changes in H3K27me3 are loss of recruitment of Pho, E(z), and PRC1 components. In addition, pho and Spps, but not cg mutants show a redistribution of H3K27me3 to heterochromatin, suggesting they are global regulators of chromatin structure in Drosophila.

Results

Using ChIP-seq, we compare the genome-wide binding of Spps to other recruiters (Pho and Cg) and PcG proteins in wild-type (WT) larval tissue to Spps maternal-zygotic mutants, and zygotic pho and cg mutants. In these mutants, pho and cg have been deposited in the egg by the mother, but gradually lose RNA and protein throughout development. Matched RNA-seq data for WT, Spps maternal-zygotic mutants, and zygotic pho mutants were also generated. All of the material used in these studies was generated from third instar larval imaginal discs and brains. The newly generated and public data used in this study are summarized in SI Appendix, Table S1. To evaluate the quality of ChIP-seq data, we calculated the Pearson’s correlation of genome-wide binding intensity among biological replicates, which ranges between 0.74 and 0.99 (Dataset S1). Peak calling was performed after pooling the reads from biological replicates. Finally, we obtained 203 broad H3K27me3 domains (Dataset S2), and 4,786 (for Spps) to 18,226 (for Cg) peaks for other factors in wild-type Drosophila (GSE102339).

First we analyzed the genome-wide binding of Spps in comparison with other recruiters and PcG proteins in WT larvae. As expected, binding of Spps, Cg, and Pho is highly correlated genome-wide (SI Appendix, Fig. S1A), and they are all enriched at promoter regions (SI Appendix, Fig. S1B). Motif analyses confirmed the enrichment of known Pho (25) and Cg (21) motifs in their peaks (Dataset S3). For Spps, the top enriched motif is similar to the GAGA motif, followed by a GT-rich motif similar to the Cg-binding site, and a motif similar to the Pho-binding site. The enrichment of these motifs suggests Spps localizes to PREs. Consistent with the high homology of Spps to mammalian Sp1, Sp3, and Sp4, and with our in vitro and in vivo binding studies with Spps (19), a GC-rich motif [CCAC(G/C)CC] similar to the human Sp3 motif (accession MA0746.1 from the JASPER database) is also significantly enriched (Dataset S3).

We next examined the binding of the recruiters and PcG proteins with respect to H3K27me3 domains. Consistent with recent studies (26–28), the binding of PcG proteins is not restricted to H3K27me3 domains. In fact, for most analyzed PcG proteins except Pc, more than 80% of peaks are outside of H3K27me3 domains (SI Appendix, Fig. S2A), indicating broader function for PcG proteins than traditionally thought. As PcG bound regions outside of H3K27me3 domains may behave differently than those in H3K27me3 domains, we analyzed regions inside and outside of H3K27me3 domains (hereafter denoted as H3K27me3+ and H3K27me3−) separately.

Cobinding of Multiple PcG Recruiters Within H3K27me3 Domains Marks Strong PREs.

We examined the cobinding of different recruiters inside of H3K27me3 domains and their association with PRC1 and PRC2 components (Fig. 1A). The binding of these three recruiters is highly overlapped. In particular, almost all (98.6%, 706 of 716) of H3K27me3+ Spps peaks are cobound by either Pho or Cg. However, a substantial number of Pho- (n = 587) or Cg-specific (n = 743) peaks exist. In an effort to understand how different recruiters contribute to the binding of each PcG protein, we grouped peaks based on the combinatorial binding of different recruiters and examined the occupancy of each PcG protein for each group (Fig. 1B). Peaks bound by all three recruiters (G1) are associated with the strongest binding of not only recruiters but also E(z), Ph, Psc, and Pc. Peaks cobound by two recruiters (G2, G3, and G4) also show strong PcG binding, while peaks bound by only one recruiter usually have weaker binding by other PcG proteins. However, regions bound by Pho (G6) are exceptional in that they usually still have strong PcG binding. Also evident in Fig. 1B is that weak recruiter peaks (especially evident in G7, the Cg-only peaks), can have weak or no E(z) binding. How then are these in H3K27me3 domains? These weak peaks are present in H3K27me3 domains that also have strong PcG peaks; in this case, the H3K27me3 may be deposited by E(z) bound to flanking PREs. Closer inspection of the AbdB locus (Fig. 1C) shows the association between multiple recruiter binding and strong PcG peaks and weak peaks with only Pho (green box) and Pho/Cg (yellow box) recruiters.

Fig. 1.

Differential binding of three PcG recruiters in H3K27me3 domains. (A) Venn diagram shows the overlap of the peaks for three PcG recruiters in H3K27me3 domains. (B) Heatmaps show the binding of three PcG recruiters in H3K27me3 domains. The peaks are organized as seven groups (G1–G7) based on the combinatorial binding of Spps, Pho, and Cg. For example, G1 are loci bound by all three recruiters, and G2 are loci bound only by Spps and Pho. G5–G7 show loci bound by only one recruiter protein. The group names are indicated on the Left. The occupancy of other factors including H3K27me3, E(z), Ph, Psc, and Pc are also shown. The color bar shows the log₂(ChIP/input) intensity. (C) IGV tracks show the binding of Spps, Pho, and other factors in the abd-B locus. The yellow box indicates a region bound by Pho and Cg recruiters, and green boxes highlight regions bound by only Pho recruiter.

We note that while Spps binding almost always occurs with other recruiters, numerous peaks bound only by Pho or Cg are observed. The existence of peaks bound by some but not all three recruiters implies different roles for recruiters in the recognition of different sets of PREs. Furthermore, the observation that peaks bound by all recruiters are usually among the strongest ones suggests a collaboration of multiple proteins in PcG recruitment.

Binding of PRC1 Proteins to Transcribed Genes Out of H3K27me3 Domains Is Accompanied by Recruiters and PRC2.

Several recent studies demonstrated extensive PRC1 binding outside of H3K27me3 domains in both Drosophila (27, 29) and mammals (27, 30, 31). In addition, in agreement with a few sample genes tested by Loubière et al. (27), our result shows that the PRC2 component E(z) is also present in the absence of H3K27me3 (SI Appendix, Fig. S2A). Similar to PcG proteins, the majority of peaks for the recruiters are outside of H3K27me3 domains (SI Appendix, Fig. S2A). Further examination showed that PRC1-bound regions (with at least one of Ph, Psc, or Pc) outside of H3K27me3 domains are also bound by different recruiters and E(z) (SI Appendix, Fig. S2B).

To examine the transcription of PcG-associated genes, we generated RNA-seq data from wild-type third instar imaginal discs and brains. The quality of RNA-seq data was evaluated by correlation of normalized read counts among biological replicates (Pearson’s r > 0.9995). Our results confirmed that genes in H3K27me3 domains are heavily repressed (SI Appendix, Fig. S2C). Further analysis of H3K27me3− genes showed that the recruiters and PcG proteins bind actively expressed genes (SI Appendix, Fig. S2D).

Spps and Pho Bind Distinct Regions Marked with Differential Ph and Psc Occupancy Outside of H3K27me3 Domains.

We examined the binding of different PcG recruiters out of H3K27me3 domains and correlated their occurrence with PcG binding. Outside of H3K27me3 domains, Cg binds more loci and usually with much stronger intensity compared with Spps and Pho. Almost half of Cg peaks are not bound by any other PcG recruiter (SI Appendix, Fig. S3A). This is consistent with the fact that cg functions in many biological processes (32, 33). Since most of H3K27me3− Spps (92.3%) and Pho (86.2%) peaks are cobound by Cg, we focused on the comparison between Spps and Pho.

Unlike in H3K27me3+ domains where Spps peaks are almost always cooccupied by Pho (Fig. 1A), 32.1% of H3K27me3− Spps peaks have no Pho binding (Fig. 2A). Not surprisingly, there are also 66.7% of H3K27me3− Pho peaks without Spps binding, indicating that Spps and Pho frequently bind distinct regions outside of H3K27me3 domains. We further examined the binding of each PcG protein in three groups of peaks (Spps+Pho, Spps-only, and Pho-only) categorized by Spps and Pho occupancy (Fig. 2B). Similar to the results in H3K27me3 domains, peaks bound by both recruiters have stronger PcG binding. Surprisingly, closer inspection of the binding of each PRC1 component showed that Spps-only peaks are associated with strong Ph yet weak Psc, while Pho-only peaks are associated with strong Psc yet weak Ph (Fig. 2 B and C). This is further confirmed by manual inspection of Integrative Genomics Viewer (IGV) tracks for Pho-only and Spps-only peaks (Fig. 2 D and E). In contrast, E(z) shows binding in all three classes of recruiter occupancy; however, E(z) binding is weaker outside H3K27me3 domains than within them (compare Figs. 1B to 2B). Pc binding outside of H3K27me3 domains is very weak. Further analysis revealed no obvious difference in genomic distribution among different groups of peaks out of H3K27me3 domains (SI Appendix, Fig. S4A). However, analysis based on RNA-seq showed that for genes out of H3K27me3 domains, those bound by Spps-only have significantly lower expression compared with those bound by both Spps+Pho or Pho-only genes (SI Appendix, Fig. S4B) (P < 0.001, Wilcoxon rank sum test).

Fig. 2.

Spps and Pho bind distinct H3K27me3− regions marked with differential Ph and Psc occupancy. (A) Venn diagram shows the overlap of Spps and Pho peaks outside of H3K27me3 domains. (B) Heatmaps show the binding of three PcG recruiters out of H3K27me3 domains. The peaks are organized as three groups based on the combinatorial binding of Spps and Pho. The occupancy of other factors including H3K27me3, E(z), Ph, Psc, and Pc are also shown. The color bar shows the log₂(ChIP/input) intensity. (C) Averaged curves show the differential binding of Ph and Psc in Pho- and Spps-only peaks. (D) IGV tracks show the occupancy of different factors in Pho-specific peaks in Scm promoter. (E) IGV tracks show the occupancy of different factors at Spps-specific peaks near the stv promoter. The red squares in D and E highlight the regions with differential binding of Spps and Pho.

The finding that Spps and Pho bind distinct regions marked with differential Ph and Psc occupancy outside of H3K27me3 domains suggests independent roles of these two recruiters at these loci. It also suggests recruitment of variant forms of PRC1 to these sites.

Reduction of PcG Recruiters Is Accompanied by Decreased H3K27me3 Occupancy and PcG Protein Binding.

Spps mutants lacking both maternal and zygotic Spps form phenotypically normal adults that die either during or shortly after eclosion. Thus, we were able to examine PcG binding in third instar larvae that never had Spps protein at any stage of their development. In contrast, maternal pho and cg are essential for embryogenesis and oogenesis, respectively. Therefore, we examined PcG binding and H3K27me3 in third instar larvae from pho or cg zygotic mutants that contained maternally supplied Pho or Cg. By third instar larval stages, maternal Pho is no longer detectable by Western blots (Fig. 3A); a small amount of Cg protein persists in the mutants at this stage (21) but that amount is very low to undetectable in some mutant samples (Fig. 3A).

Fig. 3.

Knockout of PcG recruiters strongly affects canonical H3K27me3 domains and the binding of some PcG factors. (A) Western blot shows unchanged protein abundance of H3K27me3, Pho, Ph, and Psc in pho and Spps mutant third instar larval disks and brains. Actin Westerns are shown as loading controls. A dilution series of Pho in a pho mutant and Cg in a cg mutant are shown with WT and actin controls. Pho levels are undetectable in the pho mutants (ns shows a nonspecific band); a barely detectable amount of Cg persists in the cg mutant larvae. (B) H3K27me3 intensity as measured by ChIP-RPKM for canonical H3K27me3 domains in wild type and mutants. The P values from Wilcoxon rank sum tests are indicated. (C) Density plots show the differential binding of Pho, E(z), Ph, and Psc in Spps mutants. The binding in and out of H3K27me3 domains is shown separately as solid and dashed lines, respectively.

To examine the changes of H3K27me3 occupancy after knockout of each PcG recruiter, we focused on the 203 canonical H3K27me3 domains identified in wild-type Drosophila (Dataset S2). By examining the fraction of reads present in these domains, we found that the fraction reduced significantly from the 34% of total reads in wild type to around 20% in Spps and pho mutants and to 28% in cg mutants (SI Appendix, Fig. S5). Further inspection of the calculated ChIP-reads per kilobase million (ChIP-RPKM) for each H3K27me3 domain confirmed the global reduction of H3K27me3 intensity from canonical H3K27me3 domains (Fig. 3B). Spps and Pho have a stronger impact on H3K27me3 levels in canonical Polycomb domains than Cg. Importantly, total H3K27me3 levels as measured by Western blotting are not detectably altered by the loss of Pho or Spps (Fig. 3A). The levels of the E(z) protein are also unchanged (Fig. 3A). Together, these data strongly suggest that H3K27me3 levels are specifically reduced within canonical Polycomb domains. We next compared the H3K27me3 changes in each H3K27me3 domain among different mutants. The result showed high correlation in the change of H3K27me3 after disruption of Spps and Pho, but both with a much lower correlation against cg mutant (SI Appendix, Fig. S6).

We examined the binding of PcG proteins in the absence of Spps (Dataset S4). The results showed that Pho, E(z), Ph, and Psc binding is markedly decreased in Spps mutants; the degree of changes differs for peaks within and out of H3K27me3 domains (Fig. 3 C and SI Appendix, Fig. S7). Surprisingly, Ph seems to behave differently from other PcG factors. Comparison of the differential binding of these factors after disruption of Spps showed that while the binding of Pho, E(z), and Psc is affected more in H3K27me3 domains, Ph binding is affected more outside of H3K27me3 domains (Fig. 3 C and SI Appendix, Fig. S7). We further examined the changes of Ph in pho and cg mutants, and also observed that its binding decreased more out of H3K27me3 domains (SI Appendix, Fig. S8). We validated our ChIP-seq results using ChIP-qPCR on two independent biological samples at specific targets (SI Appendix, Fig. S9).

A close examination of some of the targets highlights some of the resiliency and combinatorial features of PcG recruitment and/or stability. Fig. 4A shows H3K27me3 and PcG protein binding in a wild type vs. Spps mutant at the exex locus, one of the targets that shows drastically reduced H3K27me3 in the Spps mutant. H3K27me3 levels are also reduced but to a lesser degree in the pho mutant and only slightly reduced in the cg mutant. There is one strong and two weak PcG binding regions in this locus. In the absence of Spps, binding of Pho, Psc, Ph, and E(z) to all sites is greatly reduced. Thus, H3K27me3 and recruitment of Pho, E(z), Psc, and Ph at this locus is dependent on Spps; there is no efficient redundant recruitment mechanism for these factors in this domain.

Fig. 4.

IGV tracks over the exex and HGTX loci in wild-type and mutant larvae. This figure shows normalized ChIP-seq data over the exex (A) and HGTX (B) loci comparing H3K27me3 levels and Ph binding in WT and Spps, pho, and cg mutants and levels of Pho, Psc, and E(z) binding in WT vs. Spps mutants. Input controls are shown for WT and Spps mutants. WT Pc, Spps, and Cg binding profiles are shown for comparison. All tracks are scaled at 0–10. The peaks at the HGTX locus (B) are highlighted by the black boxes and numbered below. Peak 1 shows few changes. Peak 2 shows reduction in Pho, Psc, Ph, and E(z) binding and peak 3 shows more drastic reduction in Pho, Psc, Ph, and E(z) binding in Spps mutants.

Other loci show more modest effects on H3K27me3 levels, and different PREs within the same locus respond differently to the loss of Spps. At the HGT-X locus, H3K27me3 levels are reduced but not gone in the Spps mutant (Fig. 4B). Without Spps, at peak 3 there is no E(z) or Psc binding and only a very small peak of Pho and Ph. Thus, at this peak, PRC1, PRC2, and Pho recruitment almost entirely depends on Spps. In contrast, redundant recruitment mechanisms exist at peaks 1 and 2. Binding of Pho, Psc, Ph, and E(z) at peak 1 is largely unaffected, and peak 2 shows an intermediate effect with reduced levels of binding in Spps mutants.

The larger domains appear to be more resilient to the loss of one factor. Many of the loci containing well-studied PREs and canonical PcG targets do not show drastic changes in the binding of Pho, PRC1, and PRC2 components in Spps mutants, but still show significantly reduced H3K27me3 levels throughout the domains. The abd-B locus is shown in Fig. 5. The boxes outlined in black highlight peaks with reduced, but not abolished, Pho, Ph, Psc, and E(z) binding in Spps mutants. The boxes outlined in red highlight Ph peaks that show reduced binding in a cg mutant but not in the Spps or pho mutants. The Ubx and inv-en loci are shown in SI Appendix, Fig. S10 A and B, respectively. Like the abd-B locus, H3K27me3 levels are reduced and PcG-binding peaks are slightly reduced at some PREs in Spps mutants.

Fig. 5.

IGV tracks over the abdB locus in wild-type and mutant larvae. This figure shows the normalized ChIP-seq data over the abdB locus comparing H3K27me3 levels and Ph binding in WT and Spps, pho, and cg mutants and levels of Pho, Psc, and E(z) binding in WT vs. Spps mutants. Input controls are shown for WT and Spps mutants. WT Pc, Spps, and Cg binding profiles are shown for comparison. All tracks are scaled to 0–10. Changes in the Spps mutant are highlighted by the boxes outlined in black. Change of Ph binding in the Cg mutants are highlighted by the boxes outlined in red. Numerous peaks show reduced Pho, Ph, Psc, and E(z) binding in the Spps mutant but binding is not completely abolished at any sites. Loss of Cg and Spps affect different Ph peaks.

Inside of H3K27me3 domains, changes in PcG binding are clustered in Spps mutants and always occur at Spps-binding sites. In contrast, outside of H3K27me3 domains, the changes in binding of individual PcG proteins in WT vs. Spps mutant do not cluster at the same Spps-binding site, and there is no discernible pattern to the changes. The different PcG proteins seem to respond independently to the loss of Spps. Many of the changes also appear to be indirect since they occur in the absence of a clear Spps-binding peak. Some example tracks of changes outside of H3K27me3 domains are shown in SI Appendix, Fig. S11.

Disruption of PcG Recruiters Does Not Lead to Significant Derepression of Target Genes.

We next asked how gene expression is altered after disruption of PcG recruiters. In particular, how is the differential expression related to the alteration of H3K27me3? We performed matched RNA-seq for wild-type, Spps, and pho mutant third instar larvae. The decreased Spps and pho expression in the corresponding mutants was confirmed (SI Appendix, Fig. S12). Differential expression analysis [false discovery rate (FDR) < 0.05] identified around 1,000 significant differential genes after knockout of either Spps or Pho, with up- and down-regulated ones of similar number (SI Appendix, Fig. S13 A and B and Dataset S5). However, the degree of changes is usually small, with only several dozens of genes (50 for pho and 64 for Spps) changed by more than twofold. Further analysis showed that the alteration of gene expression after disruption of Spps and pho, as reflected by the overlapping of differentially expressed genes and correlation of log₂ fold changes (SI Appendix, Fig. S13 C and D), are positively correlated. Interestingly, gene ontology analyses showed up-regulated genes in Spps and pho mutants are both enriched for neurogenesis, glial cell development, and fundamental functions like protein folding, splicing, and DNA replication, while down-regulated genes are associated with immune response and defense to bacterium (Dataset S6). Given that our samples contain the larval brains, the high proportion of genes involved in neurogenesis is not surprising.

To clarify how differential expression is related to the altered binding of PcG proteins and the intensity of H3K27me3, we first classified genes into four groups based on whether they are within H3K27me3 domains and/or bound by recruiters in their promoters, and then examined how gene expression was altered for each group. No significant difference was observed among different groups regarding log₂ fold changes of expression (SI Appendix, Fig. S14). However, when further scrutinized, for the top differential genes (FDR < 0.05, fold change > 2) within H3K27me3 domains, the degree of differential expression usually could be well explained by the degree of H3K27me3 loss (SI Appendix, Fig. S15). Together, these results indicate that the altered expression for many genes, in particular for down-regulated ones, are likely due to indirect effects.

Redistribution of H3K27me3 to Constitutive Heterochromatin After Disruption of PcG Recruiters Is Independent of Other PcG Factors.

Intrigued by the loss of H3K27me3 from canonical Polycomb regions after disruption of PcG recruiters, we further asked: Could H3K27me3 be preferentially redistributed to specific regions originally without H3K27me3? We visualized the H3K27me3 domains by folding the chromosome using the Hilbert curve (34), which maintains the spatial proximity of linearly adjacent regions (Fig. 6). We observed the occurrence of H3K27me3 as broad blocks in wild-type Drosophila (Fig. 6B). Further inspection of the log₂ fold changes of H3K27me3 after disruption of PcG recruiters confirmed loss of H3K27me3 from canonical Polycomb domains. Surprisingly, we found that after disruption of Spps and Pho, but not Cg, H3K27me3 is accumulated in several ultra broad domains (Fig. 6C), corresponding to the annotated heterochromatin by the Drosophila Heterochromatin Genome Project (DHGP) (35) as in Drosophila reference genome [dm3, the Berkeley Drosophila Genome Project (BDGP) Release 5]. Repeated analysis using smoothed curves along the genome also showed H3K27me3 tends to accumulate in heterochromatin and chromosome ends after disruption of Spps and pho, but not cg (Fig. 6D). Analysis based on H3K27me3 ChIP-RPKM changes gave similar results (SI Appendix, Fig. S16), and closer inspection of biological replicates showed the accumulation of H3K27me3 to heterochromatin could be observed consistently and specifically only after disruption of Spps and pho (SI Appendix, Fig. S17).

Fig. 6.

Redistribution of H3K27me3 after knockout of PcG recruiters. (A) Chromosome boundaries for the Hilbert curves, with chromosome names indicated. (B) Hilbert curves show the H3K27me3 intensity as measured by ChIP-RPKM in wild type. (C) Hilbert curves show the log₂ fold changes of H3K27me3 in Spps, pho, and cg mutants against wild type. Log₂ fold change is indicated by color. (D) Smoothed curves show the log₂ fold changes of H3K27me3 after knockout of each recruiter. The smoothed curves are generated by fitting cubic smoothing spline to the log₂ fold values calculated by moving a 5-kb sliding window along the genome. The canonical H3K27me3 domains are marked as red bars along the chromosome.

Is the loss of H3K27me3 from canonical Polycomb domains after disruption of PcG recruiters accompanied by decreased binding of Pho, E(z), and Psc? For this purpose, we first calculated the changes of H3K27me3 by moving a 1-kb sliding window along the genome and defined regions with decreased and increased H3K27me3 (cutoff = 1.5-fold), respectively. Correlating H3K27me3 against PcG factors showed that regions with decreased H3K27me3 are usually accompanied by decreased binding of Pho, E(z), and Psc, but less frequently Ph (Fig. 7A). We confirmed that the regions with increased H3K27me3 in Spps mutants are highly enriched in heterochromatin (P = 1.8e-279, binomial test). However, we did not observe altered binding for any of the analyzed PcG factors in these regions (Fig. 7B).

Fig. 7.

Relationship between redistribution of H3K27me3 and changed binding of PcG factors after Spps knockout. (A) Scatterplots show the relationship between changes of H3K27me3 against Pho, Ph, and Psc in Spps mutant. (B) Density plots show the distribution for the changes of Pho, Ph, and Psc for regions with decreased and increased H3K27me3, respectively.

Discussion

Understanding how PcG complexes are recruited and maintained at PREs is essential to our understanding of PcG repression mechanisms. Our comparison of the genomic distribution of three recruiter proteins, Spps, Pho, and Cg, in wild-type and mutant larvae allows us to make the following conclusions: (i) Recruitment of PcG proteins is combinatorial and redundant at many sites. The strongest recruitment of PcG proteins occurs when all three recruiter proteins are present. (ii) Spps and pho mutants show a global reduction in H3K27me3 levels from canonical Polycomb domains and a redistribution of H3K27me3 to heterochromatin. (iii) Loci with drastically reduced H3K27me3 levels in Spps mutants have a concomitant loss of Pho-RC, PRC1, and PRC2 components. (iv) Individual PREs respond differently to the loss of one factor. (v) Spps and Pho play additional and semi-independent roles outside of H3K27me3 domains and show differential association with Psc and Ph, implying the recruitment of PRC1 variants rather than the canonical PRC1 at some of the sites. (vi) Global reduction in H3K27me3 levels from canonical Polycomb domains does not result in global derepression of associated genes. In summary, our results define a role for Spps in PcG recruitment and highlight the resiliency and diversity of PREs and recruitment mechanisms.

Pho is the most-studied and best-understood DNA-binding factor involved in PcG recruitment. However, Pho does not act alone. PREs are composed of a flexible array of binding sites for many different DNA-binding proteins, including Pho/Phol, Adf1, Cg, Spps, Dsp1, GAF/Psq, Zeste/Fsh-S, and Grh (6). Most PRE-binding proteins have been identified through studies of individual PREs from different loci in transgene assays. However, defining the precise role of these individual proteins in PcG recruitment has been quite intractable. Results are complicated by the fact that often more than one protein can bind to a particular binding site. For example, Pho-like (Phol) binds to the same sequence as Pho and can partially substitute for Pho in a pho mutant (14). Further, with the exception Pho and Phol, the other DNA-binding proteins do not appear to be stable components of characterized complexes. Here, we carried out a comparative analysis of the genome-wide distribution of three of these recruiter proteins, Pho, Spps, and Cg, in the same tissue type in wild-type and mutant animals to try to discern the roles of these factors in PRE function and PcG recruitment. We especially focused on Spps, since the phenotype of Spps mutants gave us the unique opportunity to look at PcG binding in the complete absence of Spps. While Phol can perform some functions of Pho in a pho mutant, this substitution is incomplete; pho mutants die as pharate adults with homeotic transformations. In fact, our data show that loss of zygotic pho has large effects on Ph binding and H3K27me3 levels in larvae; thus, Phol appears to be a poor substitute for Pho genome-wide.

Spps and pho mutants have reduced H3K27me3 levels within canonical Polycomb domains genome-wide. cg mutants showed a smaller reduction in H3K27me3 levels, suggesting cg may primarily act on a different part of the recruitment process. Ray et al. (21), proposed Cg may play a role in the recruitment of PRC1; cg mutants showed some changes in Ph recruitment and Cg protein coimmunoprecipitated with PRC1 components in S2 nuclear extracts. Loci that show drastic reduction of H3K27me3 levels in Spps mutants show a corresponding loss of recruitment of Pho (and presumably Pho-RC), PRC1, and PRC2 components. PcG recruitment requires Spps at these loci. Other PREs are relatively resistant to the loss of this one factor, while some others show intermediate effects. This highlights the presence of different, partially redundant recruitment mechanisms at different PREs, even within the same PcG target gene. Redundancy and multiple recruitment mechanisms make the system more resilient to the loss of one factor, especially in the larger domains with multiple PREs.

Pho binding is decreased in Spps mutants and our results argue for a combinatorial relationship between these two proteins. Spps may facilitate binding of Pho to some PREs and all of the effects of recruitment of PRC1 and PRC2 components and H3K27me3 levels may be a direct consequence of loss of Pho binding. The mammalian homolog of Pho, YY1, has been shown to interact with mammalian Sp1 (36); such interactions may be conserved in the Drosophila proteins. Spps may also contribute directly in the recruitment of some of these complexes or complex variants, rather than all of the effects being mediated only through Pho.

Recruitment of PcG proteins to PREs that show no or only modest changes in PcG protein binding in Spps mutant larvae presumably occurs via Pho, Cg, and the other DNA-binding factors that we have not studied here. Adf1 has been proposed to play a role in Pc and Pho recruitment (20). Dsp1, GAF, and Grh have similarly been proposed to influence Pho recruitment to PREs (16, 17, 37). Other factors play as yet unidentified roles. Other experiments show that not only are PcG complexes recruited to PREs by DNA-binding proteins but that the PcG complexes also influence recruitment of each other and of the DNA-binding proteins (12, 13, 38–40). Such reciprocal interactions lead to more robust and stable PcG binding.

One interesting result of our study was that the binding of Ph differs from that of other PRC1 components. Mutation of Spps, cg, and pho all affect recruitment or retention of Ph at a number of sites but sometimes at different PREs. Ph also behaves differently from other PRC1 components in that the differential binding of Ph after disruption of PcG recruiters is affected more out of, instead of within H3K27me3 domains. Our data suggest that Ph is being recruited independently of other PcG proteins to some sites in the genome; whether it is in a complex with other proteins or not is unknown.

Transgenes have been extremely useful in identifying individual DNA-binding sites and DNA-binding factors that play a role in PRE activity and PcG repression. However, PREs do not exist in isolation and in vivo the cooperative nature of PREs makes the system quite robust to the loss of one factor or binding site. Results from one isolated PRE do not predict what would happen at the locus and results with transgenes can be misleading if trying to infer recruitment mechanisms based on only one PRE. For example, repressive functions of transgene constructs carrying a PRE from the engrailed gene, enPRE2, are lost when either the Spps DNA-binding site or Spps protein is lost (19, 24). However, engrailed is not misexpressed in a Spps mutant and PcG proteins are still bound to enPRE2 in Spps mutant larvae (SI Appendix, Fig. S9). Interestingly the level of H3K27me3 is decreased within the invected-engrailed Polycomb domain, as it is in almost all canonical H3K27me3 domains. This reduction occurs with only minor decreases in PRC1 and PRC2 binding at multiple PREs (SI Appendix, Fig. S10). This is true in all large canonical H3K27me3 domains. Clearly the PREs within canonical H3K27me3 domains must work together to establish high H3K27me3 levels.

Spps and pho mutants show a larger reduction in H3K27me3 than cg mutants and Spps and pho mutants but not cg mutants show a redistribution of H3K27me3 to heterochromatin. There is no stable relocation of PRC2 components (or PRC1 or recruiters) to heterochromatin, consistent with other reports of E(z) roaming in the absence of recruitment (41). While it is generally believed that H3K27me3 binds facultative instead of constitutive heterochromatin, which is marked by another repressive histone mark, H3K9me3 (42), our results show that H3K27me3 can be redistributed to constitutive heterochromatin after disruption of Spps and Pho.

As shown by others (27, 29), PRC1 components occur outside of H3K27me3 domains in actively expressed regions in the neuronally derived BG3 cells, in embryos, and in larval tissues. In addition to PRC1, we find the PRC2 component E(z) and the three recruiter proteins binding outside of H3K27me3 domains. Loubière et al. (27) also report E(z) binding to selected active loci in larvae. Unlike in H3K27me3 domains where most Spps-binding sites coincide with Pho-binding sites, outside of H3K27me3 domains there are many sites where Spps and Pho bind independently and at these sites there is a differential association with Ph and Psc, respectively. This suggests there is a qualitative difference in the type of complex that is recruited to these binding sites. Binding at these sites cannot be due to binding of canonical PRC1. E(z) is recruited but not active in these sites, presumably due to the presence of active chromatin marks. The E(z) peaks are lower outside of the H3K27me3 domains and it may not be stably bound there. The presence of the recruiters at active genes is surprising and shows there is more we need to understand about what constitutes a PRE. Loubière et al. (27) have hypothesized that these active binding sites of PcG proteins play a role in tempering expression levels of active genes on differentiation. Schaaf et al. (29) have shown that cohesin plays a role in recruitment of PRC1 to actively expressed genes in wing imaginal discs in third instar larvae and in BG3 cells. Loss of Spps caused significant reductions of Pho and PcG binding at only a small number of binding sites in active genes, and the changes in Ph, Psc, and E(z) do not cluster to the same binding site the way they do inside of H3K27me3 domains. Thus, the role of PcG recruiters at actively transcribed genes is not known.

Decreases in H3K27me3 levels in Polycomb domains does not necessarily lead to global increases in RNA levels for associated genes (43). However, derepressed genes that are in H3K27me3 domains do show a corresponding large reduction in H3K27me3, linking H3K27me3 levels to repression in these domains. That not all targets show derepression is presumably due to the absence of required activators. However, we note that our RNA-seq and ChIP-seq data are taken from a mixed cell population and will not detect changes that occur in only a few cells. For example, Ubx is misexpressed in a few cells in wing discs in pho mutants (25) but we do not see a significant increase in Ubx RNA in our RNA-seq data.

Recent results on PcG binding in Su(z)12 (a PRC2 component) and Psc/Su(z)2 (PRC1 component) mutant transformed cell lines identified different classes of PREs (40). In one type of PRE, PRC2 could bind in the complete absence of PRC1, in another class, PRC2 binding was dependent on PRC1, clearly ruling out the role of H3K27me3 and H2AK118ub in recruitment and a hierarchical PRC2–PRC1 recruitment model. In our study we found that loss of the PRE-binding proteins Pho or Spps led to global losses of both PRC1 and PRC2 binding. Our data support the view that PRC1 and PRC2 act together to reinforce each other’s binding, and these in turn may stabilize the binding of Pho, Spps, and the other DNA-binding proteins.

The ability to examine PcG binding in the complete absence of the PRE-binding protein Spps allows us to make definitive statements of the nature of PREs. At some PcG-binding sites, Spps is required for PcG recruitment, while at others, it is completely dispensable. Recruitment of PcG proteins to most PREs is combinatorial. Most H3K27me3 domains have many PREs of different types. Large decreases in H3K27me3 levels are tolerated with only minor effects on gene expression. In summary, our results demonstrate the enormous complexity of PREs and PcG protein recruitment, the resiliency of multiple recruitment mechanisms, and how much we have yet to learn.

Materials and Methods

Generation of Mutant Drosophila Strain.

The Spps¹ mutant and maternal/zygotic larvae were generated as described in ref. 19. pho¹ were collected as GFP negative larvae from a pho¹/P{w[+mC] = ActGFP}unc-13[GJ] stock. The cg^A22 data are from ref. 21. Fly stocks were maintained at 25 °C.

ChIP-Seq.

ChIP was performed on CNS and imaginal discs from third instar larvae as described in ref. 44 with minor changes. Reads were aligned to reference genome (dm3, BDGP Release 5) using Bowtie v1.1.2 (45). After PCR duplicates were removed using the rmdup function of samtools (46), peak calling was performed using SICER v1.1 (47) for H3K27me3, and MACS v2.1.1 (48) for other datasets. To focus on canonical H3K27me3 domains, only those with FDR <0.1, length >5 kb, and with at least twofold enrichment were used for further analysis. We used DiffBind (47) to identify differential binding regions between wild type and mutants. Further details about ChIP-seq and analysis are provided in SI Appendix, Materials and Methods.

RNA-Seq.

Each RNA sample was derived from CNS and imaginal discs dissected from 20 third instar larvae. Total RNA was isolated with TRIzol (Invitrogen) followed by purification using a Qiagen RNeasy micro kit followed by PolyA extraction. Purified mRNAs were constructed to RNA-seq libraries with specific barcodes using the Illumina TruSeq Stranded mRNA Library Prep Kit and all of the 12 RNA-seq libraries were pooled together and sequenced using Illumina HiSEq. 2500 with one rapid run to generate about 30–50 million 2 × 100 paired-end reads for each sample. Reads were aligned to the reference genome (dm3, BDGP Release 5) using Tophat v2.1.1 (49). The FPKM for each gene was calculated using cufflinks and normalized between different datasets using cuffnorm (49). Differential expression analysis was performed using DESeq2 (50). Further details about RNA-seq analysis are provided in SI Appendix, Materials and Methods.

Western Blots.

Western blots were performed on CNS and imaginal discs isolated from third instar larvae. Details of sample preparation and antibody dilutions are given in SI Appendix, Materials and Methods.

Data Accession.

Data used in this study have been deposited in GEO database with accession GSE102339. All of the newly generated and public data used in this study are summarized in SI Appendix, Table S1.