Crol contributes to PRE-mediated repression and Polycomb group proteins recruitment in Drosophila

Abstract The Polycomb group (PcG) proteins are fundamental epigenetic regulators that control the repressive state of target genes in multicellular organisms. One of the open questions is defining the mechanisms of PcG recruitment to chromatin. In Drosophila, the crucial role in PcG recruitment is thought to belong to DNA-binding proteins associated with Polycomb response elements (PREs). However, current data suggests that not all PRE-binding factors have been identified. Here, we report the identification of the transcription factor Crooked legs (Crol) as a novel PcG recruiter. Crol is a C2H2-type Zinc Finger protein that directly binds to poly(G)-rich DNA sequences. Mutation of Crol binding sites as well as crol CRISPR/Cas9 knockout diminish the repressive activity of PREs in transgenes. Like other PRE-DNA binding proteins, Crol co-localizes with PcG proteins inside and outside of H3K27me3 domains. Crol knockout impairs the recruitment of the PRC1 subunit Polyhomeotic and the PRE-binding protein Combgap at a subset of sites. The decreased binding of PcG proteins is accompanied by dysregulated transcription of target genes. Overall, our study identified Crol as a new important player in PcG recruitment and epigenetic regulation.


INTRODUCTION
Epigenetic control is r equir ed to establish and maintain correct patterns of gene expression that determine cellular identity and pre v ent the de v elopment of pathologies in m ulticellular organisms ( 1 ). Pol ycomb gr oup (PcG) pr oteins are transcriptional repressors that control chromatin state of facultati v e heter ochr omatin (2)(3)(4)(5). PcG genes were first identified genetically in Drosophila through studies on the Hox genes that specify segment identity during de v elopment ( 3 , 5-7 ). Mutations in PcG encoding genes led to homeotic transformations due to der epr ession of the Hox genes ( 3 , 5-7 ). After these initial discoveries, it was found that PcG proteins control transcription of many de v elopmental genes, and dysfunction of PcG genes is observed in many diseases, including cancer (8)(9)(10)(11)(12)(13).
Most of the known PcG proteins are organized into multiprotein complexes that interact with chromatin and mediate r epr ession of gene transcription ( 4 , 14 ). Polycomb

Yeast two-hybrid screening (Y2H)
The Cg (CG8367) PF 31-467 aa isoform was cloned into pGBT9 vector (Clontech) to make a fusion with the GAL4 DNA-binding domain. The cDNA fragments encoding the C2H2-type zinc finger (C2H2-ZFP) candidate proteins were cloned into pGAD24 vector (Clontech) to combine with the GAL4 activating domain. The full-length cDNA was used f or Crol (CG14938, isof orm A). Other proteins tested are listed in Supplementary File 1.
The Y2H assay was performed as previously described ( 53 ). Briefly, for the growth assays, plasmids were trans-formed into pJ69-4A strain by the lithium acetate method following the standard Clontech protocol and plated on the nonselecti v e media lacking tryptophan and leucine. After 3 days of growth a t 30 • C , pla tes wer e r eplicated on selecti v e media: 1 -lacking tryptophan, leucine, and histidine in the presence of 5 mM 3-aminotriazole ( −3 + 3AT); 2 -lacking tryptophan, leucine, histidine, and adenine (-4). Each assay was r epeated thr ee times and growth was compar ed after 2, 4 and 7 days. Based on the extent of growth the interactions ar e scor ed as strong (detected on day 2, '+++'), intermediate (detected on day 4, '++') or weak (detected on day 7, '+'). ' −' indicates no growth.

Antibodies
Antibodies against the following proteins were raised in rabbits: Cr olN (a.a. 1-188, isoform PA), Cr olC (a.a. 726-962, isoform PA), E(z) (a.a. 8-184, isoform PA). Antigens for antibody production were expressed as 6 × Histagged fusion proteins in Esc heric hia coli , affinity purified on Ni Sepharose 6 Fast Flow (GE Healthcare), according to the manufacturer's protocol, and injected into rabbits following standard immunization procedures. Antibodies were affinity-purified from serum on the same antigen as was used for the immunization and tested by immunoprecipitation / Western blotting (IP / WB) to confirm their specificity (Supplementary File 2). IP / WBs were pr epar ed as described pr eviously ( 54 ). Proteins wer e detected using the ECL Plus Western Blotting substrate (Pierce).
Other antibodies used in this study were created previously, with details provided in Supplementary File 3.

ChIP-qPCR
Chroma tin immunoprecipita tion (X-ChIP) was pr epar ed as described previously ( 55 ). For each experiment, 150-200 mg of third instar larvae or 0-16 h embryos were collected. The material was homogenized and crosslinked with 1.8% f ormaldehyde f or 15 min. Crosslinking was quenched with glycine and homogenate was cleared by passing through 100-m nylon cell strainer (BD Falcon), washed, and sonica ted to genera te 200-600 bp DNA fragments. Chroma tin was preincuba ted with A-Sepharose. One aliquot (1 / 10 volume) of chromatin extract after preincubation with Sepharose was kept as a control sample (Input). The resulting chromatin was used for 6-8 individual incubations with antibodies. Chromatin was incubated with specific antibodies-Sepharose at 4 • C overnight. A control incubation was made with IgG of nonimmunized animal (rabbit). All ChIP experiments were made in biological triplicate. After washing and eluting from beads, the crosslinking was re v ersed and the DNA was isolated.
The enrichment of specific DNA fragments was analyzed by real-time qPCR, using a C1000 ™ Thermal Cycler with CFX96 real-time PCR detection module (Bio-Rad) or a StepOne Plus Thermal Cycler (Applied Biosystems, United States). Detailed protocol for ChIP-qPCR is described in Supplementary File 4. Primers used in ChIP / real-time PCR analyses are listed in Supplementary File 5.

ChIP-seq
Brains and imaginal discs from third instar larvae (10 larvae per sample) were crosslinked with 2% formaldehyde for 15 min. Crosslinking was quenched with glycine. The samples were homogenized and sonicated to generate 200-500 bp DNA fragments. A percentage of the sample was removed as the input contr ol. Chr omatin was pr e-clear ed with protein A sepharose beads and then incubated overnight with antibod y a t 4 • C . Chroma tin was then precipita ted with Protein A Sepharose beads for 1 h at 4 • C. After washing and eluting from beads the crosslinking was re v ersed and DNA was isolated.
ChIP-seq libraries were obtained using the NEBNext Ultra ™ II DNA library preparation kit (New England Biolabs) or with the Thruplex DNA-seq and single index kits (Takara) following the manufacturer's instructions. Samples were sequenced by 50bp or 100bp single-end sequencing with HiSeq2500 (Illumina) or with NovaSeq6000 sequencer. Detailed procedure for ChIP-seq is described in Supplementary File 4.

Motif analysis
To predict the recognized DNA motif of Crol from its zinc finger arrays, the w e b server ( https://zf.princeton.edu ) developed by Persikov et al. was used ( 63 ). To perform motif discovery from ChIP-seq peaks, we ordered the peaks by qvalue and extracted sequences of 300 bp windows centered at the peak summits. MEME v5.4.1 ( 64 ) was executed with sequences from the top 500 peaks with default parameters, ChIPMunk ( 65 ) was executed with sequences from the top 500 and top 1000 peaks and provided with the peak summit (midpoint) location.

RNA-seq
For the extraction of RNA, the third instar larvae of wildtype (Oregon) and crol -KO mutants were collected in a PBS buffer (40 larvae per sample), in three biological repeats. Total RNA was extracted with the TRI reagent (Ambion). Pol yA comprising RN A fraction was isolated and pr epar ed for sequencing with the NEBNext Ultra ™ II Directional RNA Library Prep Kit. New generation sequencing was performed by Evrogen ( evrogen.ru ) with the Illumina No-vaSeq6000 sequencer. For each RNA-seq library approximately 20-25 millions of paired-end reads were obtained.
Aliquots of purified recombinant proteins (0.05-1.2 g) were incubated with fluorescently labeled DNA fragments (60-80 ng) in the presence of nonspecific binding competitor --1 ng of pol y(dI-dC). DN A fragments were labeled using PCR with primers containing the Cy5 fluorophore. Signals for Cy5 were detected at the Ex 630 nm / Em 700 nm. The sequences of 20xG, Control, 20 bpe v ePRE, 20 bp-e v ePREmut containing fragments obtained by PCR are gi v en in Supplementary File 4. Incubation was performed in 1 × PBS (pH 8.0) containing 5 mM MgCl 2 , 0.1 mM ZnSO 4 , 1 mM DTT, 0.1% NP-40 and 10% glycerol at room temperature for 30 min. The mixtures were then resolved by nondenaturing 5% PAGE (79 AA:1 BAA) in 0.5 × TBE buffer at 5 V / cm.

Generation of control, eve-wt and eve-mut transgenic lines
The details of cloning of the Control, e v e-wt and e v e-mut plasmid constructs are described in Supplementary File 4. The constructs were injected into embryos of attP2 line ( 71 ). The resulting flies were crossed with yacw 1118 flies, and the transgenic progeny was identified by their eye pigmentation. For phenotype analysis of white gene e xpression le v el, we visually determined the degree of pigmentation in the eyes of 3-to 5-day-old males, with r efer ence to standard color scales. Pigmentation of all flies was analyzed in homozygotes (P / P). All flies were maintained at 25 • C on the standard yeast medium.

Crol interacts with Polycomb group proteins
To identify novel proteins that participate in recruitment of PcG proteins, we employed the Y2H assay using the PREbinding protein Cg as a bait. Cg is tightly connected to PRC1 complex ( 39 , 72 ). Apart from PRC1 subunits, its interactome includes se v eral known PRE-binding proteins as well as DNA-binding factors with uncharacterized connection to PcG system ( 72 ). As candidate DNA-binding proteins, we tested a number of C2H2-ZFPs previously identified in the Cg protein complex ( 72 ) (Supplementary File 1). Cg was fused to the DNA binding domain of yeast GAL4 protein, while each C2H2-ZFP was fused to the activator domain of GAL4. As a result, we identified Crol ( S rooked legs, CG14938) as a direct Cg interactor (Figure 1 A). Interestingly, se v eral studies support the potential role of Crol in Polycomb-dependent transcriptional r epr ession. Crol was described as an ecdysone-induced r epr essor of wg transcription in wing imaginal discs ( 73 ). In addition, its knockdown was shown to disrupt the formation of Polycomb bodies Crol and Cg, Ph, E(z) and Pho. S2 Drosophila cell nuclear extracts were incubated with CrolN (IP CrolN), CrolC (IP CrolC) rabbit antibodies or IgG of a non-immunized rabbit (IP IgG). Lysates (Input 20% and 10%), precipitated fractions (IP CrolN, IP CrolC and IP IgC) are indicated above the blots. Antibodies used for Western blots are indicated on the right side from the blots. ( 74 ), and it has also been reported to be present in heter ochr omatin ( 75 ).
We next performed in vivo experiments to validate the interaction of Crol with Cg and other PcG proteins. For this purpose, we first pr epar ed polyclonal antibodies recognizing the N-or C-terminal parts of Crol, respecti v ely (see Materials and Methods), both antibodies were confirmed to be IP grade in IP / Western-blot assay ( Figure 1B; Supplementary File 2). To determine the interaction between Cg and Crol, we performed co-IP / Western blot assay for these two proteins. In support of our Y2H assay, Crol co-purified two major Cg bands ( 72 ) but not the IgG of non-immunized rabbit (Figure 1 B). We further performed western-blot assays with Crol co-IPs against the PRC1 subunit Ph, PRC2 subunit E(z), and PRE-binding protein Pho, part of the PhoRC complex. In case of Ph, antibodies recognized both paralogs, Ph-p and Ph-d ( 72 ). Our data showed that Ph isoforms as well as Pho protein, but not E(z) were immunoprecipitated by Crol antibodies (Figure 1 B). Crol's Nucleic Acids Research, 2023, Vol. 51, No. 12 6091 interactions with Cg and subunits of PRC1 and PhoRC complexes led us to hypothesize that Crol may play a role in PcG recruitment.

Crol binds to PRE-elements in vivo and colocalizes with PcG proteins genome-wide
To test whether Crol binds to PREs, we first performed ChIP-qPCR at third instar whole larvae (Figure 2 A) and embryo (Supplementary File 6, Figure S3). Crol showed enriched binding to se v er al char acterized PREs, including bxd PRE, bx PRE, Fab7 PRE, en PRE2 and e ve PRE, b ut not the coding sequences of the Ras64B and Tub56D genes. Ther efor e, Crol is a bona fide PRE-binding protein.
We applied ChIP-seq to determine the genome-wide occupancy of Crol in brains and imaginal discs of the third instar larvae. Since the two antibodies produced consistent results (Supplementary File 6, Figure S4), only data generated with the CrolN antibody were used for further analysis. We also performed ChIP-seq for E(z), Ph and H3K27me3, and collected public data for Pc, Psc, Pho, Spps and Cg ( 29 , 39 ). We first inspected se v er al well-char acterized PREs.
Within the e ven-skipped ( e ve ) gene domain, Crol co-binds with PcG proteins at the distant eve PRE adjacent to the TER94 promoter and the proximal PRE (i.e. PSEpro) adjacent to eve PRE promoter ( 76 ) (Figure 2 B). The co-binding of Crol with PcG proteins is also evident at the bx PRE and bxd PRE of the Ubx locus ( 77 ) (Figure 2 C), and all characterized PREs of the Abd-B and invected-engrailed loci (Supplementary File 6, Figure S5). Overall, Crol showed the highest degree of co-localization with Cg, Psc and Ph (Figure 2 D). For example, the majority (81%) of Crol peaks are bound by Cg, and vice versa . Although lower percentages of Crol peaks overlap other PcG proteins, they still constitute > 75% of the peaks for Pc, E(z), Pho and Spps. Together, these data confirmed the binding of Crol to canonical PREs, and uncovered the global co-localization between Crol and PcG proteins.
We next determined the genome-wide binding of Crol to PREs, which were defined as H3K27me3(+) regions cobound by E(z), Ph and Pc (Figure 2 E). Crol is present at 80% of PREs in brains and imaginal discs of the wild-type 3 rd instar larvae. Remar kab ly, this is comparab le to that estimated for Cg, Spps and Pho (Figure 2 E). According to previous studies ( 29 , 31 , 78 , 79 ), core PRC1 / 2 subunits and PcG recruiters (i.e. Pho, Spps and Cg) frequently bind outside of H3K27me3 domains. Consistently, we found that 84% of Crol peaks are outside of H3K27me3 domains (Supplementary File 6, Figure S6). PREs constitute about 10% of Crol peaks, which is comparable to other PcG recruiters including Pho, Spps and Cg (Figure 2 F).
Gi v en that larvae brains and discs are made of mixed cell popula tions, their da ta are insuf ficient to valida te the simultaneous binding of Crol and other PcG proteins at the same targets in the same cells. To overcome this limitation, we examined the overlap of Crol with PcG proteins and H3K27me3 modification using the S2 cell line as a model, which r epr esents a mor e homogeneous system. A similar degree of overlap between Crol and PcG proteins was observed in S2 cells (Supplementary File 6, Figure S7): 66%, 42% and 66% of Crol peaks overlapped with Cg, Pc, and E(z) and they constituted 95%, 75% and 74% of Cg, Pc and E(z) peaks, respecti v ely. Crol and Cg were present at 78% and 58% of PREs, respecti v ely. At the same time these corresponded to only about of 3% of each of the Crol and Cg proteins peaks. The lower percentage of Crol and Cg peaks corresponding to PREs is due to lower number of H3K27me3(+) loci in S2 cells, indicating that a lower percentage of acti v e PREs exist in S2 cells in comparison to third instar larvae brains and imaginal discs.
To assess the binding pr efer ence exhibited by Crol in vivo , we performed de novo motif discovery using the Crol ChIPseq data for third instar wild-type larval brains and discs (Figure 3 B, Supplementary File 6, Figure S8). As expected, a prominent yet unstructured G-rich motif is significantly enriched, which occurs in the majority of top-scoring Crol peaks (Figure 3 B, Supplementary File 6, Figure S8). A similar motif was identified using ChIP-seq data on S2 cells (Supplementary File 6, Figure S8). These data suggest that in vivo Crol might r ecognize mor e variable poly(G)-rich motifs than suggested by the in vitro studies.
Our EMSA assay further confirmed that Crol binds to a synthesized DNA fragment containing 20xG motif instead of the non-specific control (Figure 3 C, D). Moreover, cold 20xG competes with Cy5-20xG for Crol binding (Figure 3 E). As an example of a natural Crol-binding site, we selected the poly(G)-rich motif present in the wellcharacterized 344-bp eve PRE which is adjacent to the TER94 gene promoter ( 76 ) and is bound by Crol in vivo (Figure 2 B). This poly(G)-rich motif in eve PRE (

Crol is r equir ed f or eve PRE dependent silencing in tr ansgenes
We next tested the role of the Crol motif in ev e PRE acti vity in transgenes. The following transgene constructs were made (Figure 4 A): (i) the Control construct, carried the white reporter gene responsible for red-colored eye pigmentation and the attB site r equir ed for site-specific integration using the PhiC31 system ( 71 ); (ii) the e v e-wt construct, had an insertion of the wild-type 344-bp eve PRE; (iii) the e v emut construct, carried the 344-bp eve PRE with the same mutation of Crol site as was tested in EMSA (Figure 4 B). The transgene constructs were inserted in the attP2 site   in flies lacking a functional white gene ( 71 ). The silencing activity of the eve PRE was assessed by the reduction in eye pigmentation, which is known to be directly correlated with the le v el of white gene transcription ( 81 , 82 ). Figure 4 C shows the eyes of homozygous flies for each construct. The flies carrying the Control construct had r ed-color ed eyes indicating acti v e white gene tr anscription. In contr ast, flies with the wild-type ev e PRE hav e white-colored eyes indicating complete r epr ession of the white gene. Importantly, flies containing the eve PRE with mutation in poly(G)-rich motif ar e mor e pigmented, indicating partial r elease of silencing (Figure 4 C).
To further address the function of Crol at PREs, we crea ted cr ol knockout ( cr ol -KO) flies by CRISPR / Cas9 technology (Supplementary File 6, Figures S1, S2). The homozygotes of the crol-KO flies died mostly at the pharate adult stage with a 'crooked leg' phenotype (not shown) in accor dance with pre vious data ( 83 ). We e xamined the effect of crol-KO mutants on transgenic flies carrying e v e-wt construct. Figure 4 D shows eyes of pharate adults homozygous for the eve-wt construct with ( + / +; P / P) and without cr ol ( cr ol − / cr ol − ; P / P). Consistent with a role of Crol in PRE-mediated r epr ession of transcription, in the absence of crol , the silencing of the white reporter was suppressed. Similar r esults wer e obtained with another transgene carrying a 656-bp bxd PRE. Upon crol-KO r epr ession of the white gene by bxd PRE was reduced (Supplementary File 6, Figure S9). Thus, Crol is required for PRE-mediated repression in transgenes.
We next performed the ChIP-qPCR on the third instar whole larvae bearing e v e-wt or e v e-mut transgenes and examined the impact of Crol site mutation on Crol and PcG binding. ChIP was performed for Crol, Ph, Cg, E(z) and H3K27me3. As a negati v e control, the ChIP was performed against PRE-binding protein GAF which failed to directly interact with Crol (Supplementary File 1), and we expected that its binding to be independent of Crol. The results demonstra te tha t muta tion of the Crol site in the eve PRE affected the binding of Crol (Figure 4

Knock out of cr ol leads to decreased occupancy of Ph and Cg genome-wide
We next determined the genome-wide dependency of PcG binding on Crol by performing paralleled ChIP-seq in wildtype (WT) and crol -KO third instar larval brains and discs. Despite the apparent decrease of Crol protein upon crol -KO by Western-blot assay (Supplementary File 2), we only identified 1170 genomic loci with significantly decreased Crol binding (Supplementary File 6, Figure S11). This suggests tha t in cr ol -KO flies, part of ma ternally pr ovided Cr ol pr otein remains stably associated with its targets throughout de v elopment. Accor dingl y, onl y these 1170 Crol-decreased loci were examined for differential binding analysis of PcG proteins. Upon crol-KO, the binding of Ph and Cg was decreased at 346 and 72 of peaks respecti v ely, but no significant changes were seen for E(z) or H3K27me3 (Supple-mentary File 6, Figure S11). These data are consistent with the results obtained by mutating Crol binding sites in the ev e PRE-transgene (abov e). The coor dina tes of dif ferential binding peaks for Crol, Ph and Cg are gi v en in Supplementary File 7.
To ask whether cr ol-KO af fects transcription of ph or cg , we performed RNA-seq using crol-KO and wild-type third instar larvae. The results showed that crol-KO does not reduce the transcription of ph (ph-d and ph-p) or cg . Instead, we observed slightly increased transcription of both genes (Supplementary File 6, Figure S11), probably because they are under PcG-mediated repression. In agreement, this was previously demonstrated for ph ( 84 ).
Thus, Crol is involved in the genome-wide recruitment of Ph and Cg in both H3K27me3(+) and H3K27me3( −) domains.

Crol regulates transcription of the target genes
We next analyzed the impact of crol -KO on gene transcription by performing RN A-seq anal ysis. While a significant amount of Crol protein is still present in larvae, we identified 1683 genes with altered expression upon crol-KO, including 883 up-regulated and 880 down-regulated ( Figure  5 B, Supplementary File 8). GO analysis demonstrated that up-r egulated genes ar e associated with neuron-r elated functions, while down-regulated genes are associated with mannose metabolic process and infection response (Supplementary File 6, Figure S12). Importantly, previous studies have showed tha t up-regula ted genes in Spps and pho mutants are both enriched for neuro genesis, w hile down-regulated genes include those that are associated with infection response ( 29 ). Thus, the effect of crol-KO on the transcriptome is very similar to the effects observed upon mutation of pho and Spps .
To uncover the potential link between Crol and transcriptional r epr ession, we further examined the binding of Crol on each group of the differ entially expr essed genes. Inter estingly, 26.2% of up-regulated genes have Crol binding within + / -1kb of their TSSs, which is significantly higher than down-regulated genes of w hich onl y 12.6% were bound by Crol (Figure 5 C). We further compared the alter ed expr ession of different gene groups stratified by their distance to Ph-decreased loci after Crol -KO, and found that genes with decreased Ph-binding within 5 kb to their TSSs tend to have incr eased expr ession after crol -KO (Figur e 5 D). These r esults indica te tha t Crol has r epr essi v e effect ov er their target genes, likely via mediating the recruitment of Ph and probably some other PcG proteins.
Closer inspection identified loci at which crol -KO leads to decreased Crol / Ph binding and coincidently increased transcription. For example, the CG43402 , vn , and hth genes are within H3K27me3(+) domain (Figure 6 A), while Dl and tna are not (Figure 6 B). Of note, the CG43402 and Dl loci also demonstrate remar kab ly decreased Cg binding. In most cases, the significantly decreased Crol / Ph / Cg peaks ar e r egulatory r egions of upr egulated genes. We propose that the disbalance of Crol binding to the associated r egulatory r egions is sufficient for the der epr ession of the target genes. Of well-characterized PREs from ev e, inv ectedengrailed and Abd-B, we detected significantly decreased Crol binding only on Fab6 PRE, yet without significantly altered Ph / Cg binding or Abd-B transcription (not shown). Meanw hile, w hile the Crol / Ph binding wasn't affected at the bx-and bxd PREs, we detected significantly increased transcription of Ubx gene upon crol-KO (Figure 6 C). This correlated with significantly decreased binding of Crol and Ph upstream of the Ubx promoter (Figure 6 C). We suggest that this region might be an uncharacterized regulatory element responsible for observed Ubx gene activation upon crol -KO.
Thus, Cr ol contr ols the transcription of many genes and can r epr ess gene expr ession both inside and outside of H3K27me3 domains.

DISCUSSION
Here, we have identified Crol as a new PcG recruiter. We show that Crol is a direct Cg protein partner and using coimmunoprecipita tion experiments, we demonstra te tha t in addition to Cg, Crol interacts with Ph of the PRC1 complex and with the PRE-binding protein Pho. Using ChIPseq analysis, we demonstra te tha t Crol colocalizes with PcG proteins genome-wide, including the classical PREs of the eve , en , Ubx and Abd-B gene domains.
Crol is a C2H2-ZFP that binds to poly(G)-rich sequences both in vitro and in vivo ((80) and this paper). In a transgene assay, we demonstrate that Crol-bound poly(G)-rich motif is important for proper eve PRE-mediated silencing and for recruitment of Ph and Cg proteins to ev e PRE. Moreov er, crol -KO reduces the ability of the eve PRE and bxd PRE to silence the mini-white gene in the transgenes.
Due to maternally deposition of Crol protein and its stable association with many chromatin sites throughout development, w e w er e able to obtain only r educed le v els of Crol upon it's KO. Decreased le v els of Crol were sufficient to affect binding of Crol / Ph / Cg to transgenic e ve PRE, b ut not to the endogenous eve PRE. These can reflect the cooperati v e nature of PREs that support each other in the endogenous context and promotes the binding of Crol to PREs in its natural genome environment.
In total, Crol binding in Crol KO flies was decreased at 1170 of genome sites and of them Ph and Cg were decreased at 346 and 72 of sites, respecti v ely. Crol function extends beyond genes within H3K27me3 domains and Ph / Cg are af fected a t proportionally equal Crol decreased peaks in H3K27me3+ and H3K27me3-domains. The similar importance for Ph recruitment outside of H3K27me3 domains was previously reported for the Cg protein ( 39 ).
The lower sensitivity of Cg than Ph to decrease of Crol le v el suggests that Crol is not a bona fide Cg recruiter, and its role in Cg recruitment is highly dependent upon local content. This may reflect the combinatorial nature of Ph-associa ted regula tory elements with different degrees of participation of a particular DNA-binding factor in Ph recruitment to distinct targets. Importantl y, crol -KO leads to ina ppropriate expression of 1683 genes genome-wide, of which 883 genes are upregula ted, indica ting tha t partial loss of Cr ol pr otein is sufficient to disbalance gene transcription. The d ysregula tion of gene transcription correlates with loss of Crol / Ph peaks at corresponding loci but doesn't r equir e dissociation of Crol from all peaks in the loci.
For example, in the case of the Ubx gene, crol -KO doesn't affect the le v el of Crol / Ph protein at the well-defined bxand bxd PREs but leads to a decreased Crol / Ph binding at an upstream gene region and correlates with an increase in Ubx gene tr anscription. Similar ly, in the case of the vn and Dl genes, Crol / Ph are still bound to promoters of genes but are lost in regulatory regions of the corresponding genes.
Importantl y, GO anal ysis indica tes tha t cr ol-KO af fects the transcription of same set of genes as does the mutation of two other PRE-binding proteins, Pho and Spps. In crol , Spps and pho mutants, neuron-related functions genes are up-regula ted, and genes implica ted in infection r esponse ar e downregulated. This suggests that Crol, Pho, and Spps may co-regulate these genes.
The number of proteins implicated in PcG r epr essor r ecruitment is continually growing ( 22 ). Our results suggest that PcG factors can interact with a di v erse array of DNAbinding proteins and that interactions create a combinatorial platf orm f or the r ecruitment of PcG complex es at target sites in Drosophila genome.

ETHICS APPROVAL
Animal handling for the antibody production was carried out strictly according to the procedures outlined in the NIH (USA) Guide for the Care and Use of Laboratory Animals. The protocols used were approved by the Committee on Bioethics of the Institute of Gene Biology, Russian Academy of Sciences. All procedures were performed under the supervision of a licensed veterinarian, under conditions that minimize pain and distress.
Rabbits were purchased from a licensed specialized nursery, Manihino. Soviet chinchilla rabbits used in the study are not endangered or protected. Only healthy rabbits, certified by a licensed veterinarian were used. The rabbits were individually housed in standard size, stainless steel rabbit cages and provided an ad libitum access to alfalfa hay, commercial rabbit food pellets, and water. The appetite and behavior of each rabbit was monitored daily by a licensed veterinarian. Body weight and temperature of each rabbit were evaluated prior to and daily following the immunization. No animals became ill or died at any time prior to the experimental endpoint. At the end of the study period all rabbits were euthanized by intravenous injection of barbiturate anesthetics.

DA T A A V AILABILITY
All relevant data are within the paper and its Supporting Information. The ChIP-seq and RNA-seq data generated in this study have been deposited in NCBI GEO database with accession GSE202872. The following publicly available ChIP-seq was used for Psc, Pho and Spps (GSE102339), Pc (GSE102339), Cg (GSE77582) (third instar larval brains and imaginal discs); Pc (GSE24521), H3K27me3 (GSE41440) and E(z) (GSE101554) (S2 cells).

SUPPLEMENT ARY DA T A
Supplementary Data are available at NAR Online.