Sustained PRC1.1 activity preserves gene repression independently of PRC2.2 and restrains leukemic cell differentiation

Abstract

Loss-of-function mutations in BCOR, a subunit of the non-canonical Polycomb Repressive Complex 1.1 (PRC1.1), are frequently observed in acute myeloid leukemia (AML) and associate with adverse risk. Paradoxically, leukemic stem cell viability in BCOR wild-type AMLs strongly depends on PRC1.1 activity. Here, we use BCOR and KDM2B degron models to study PRC1.1 dependency in leukemic cells and find that BCOR is a bridging factor tethering the catalytic and chromatin-binding moieties of the PRC1.1 complex. BCOR degradation induces a quick localized loss of H2AK119ub at PRC1.1 target genes, whereas PRC2-induced H3K27me3 remains unaffected. Degron-mediated depletion of BCOR or KDM2B induces a rapid but time-dependent transcriptional induction, whereby late-upregulated genes are more heavily decorated with H3K27me3 compared to early-upregulated genes. Combined PRC1.1 inactivation and PRC2 inhibition further amplifies gene induction, suggesting collaborative yet distinct control over target genes. Strikingly, both JARID2/AEBP2 and SUZ12 knockout cells, devoid of PRC2.2 or PRC2.1/PRC2.2 respectively, retain PRC1.1 loss-induced transcriptional activation, underscoring that PRC1.1 can repress target genes independently of a downstream PRC2.2-canonical PRC1 repressive axis. Finally, combined targeting of PRC1.1 and PRC2 induces differentiation of leukemic cells, emphasizing that co-targeting PRC1.1 and PRC2 represents a promising strategy to improve treatment of AML patients.

Graphical Abstract

Graphical Abstract.

Introduction

The Polycomb group (PcG) proteins are a family of epigenetic regulators that are essential for proper embryonic development and lineage fate commitment. PcG proteins promote stem cell maintenance by silencing lineage-specification genes, and their timely release from the chromatin allows controlled differentiation to a more committed state [1–3]. Interference with Polycomb-mediated silencing in metazoans leads to developmental defects, and their deregulation is frequently observed in various diseases, including cancer. Polycomb-mediated silencing classically depends on the Polycomb repressive complexes PRC1 and PRC2. PRC1 can ubiquitinate H2AK119 (H2AK119ub), and PRC2 trimethylates H3K27 (H3K27me3). The core PRC2 complex consists of the methyltransferase EZH1/2, EED, and SUZ12, whereas identification of distinct accessory proteins prompted the subclassification of PRC2.1 (including PCL1/2/3, EPOP, and PALI1/2) [4–6] and PRC2.2 (containing AEBP2 and JARID2) [7–10], which collaborate to reach appropriate H3K27me3 levels [11]. PRC1 complexes are comprised of a PCGF (PCGF2/4), PHC (PHC1/2/3), SCM (SCML1/2/SCMH1), RING1 (RING1A/B), and a CBX subunit (CBX2/4/6/7/8), which contains a chromodomain that is capable of binding H3K27me3 and has the capacity to compact chromatin [7, 12–14]. In Drosophila, PRC2 is tethered to Polycomb response elements and initiates Polycomb silencing, whereas mammalian PRC2 chromatin targeting is facilitated by PRC2.1-associated PCL proteins that can interact with H3K36me3 via their TUDOR domain [15–19]. Subsequent H3K27me3 deposition then leads to CBX chromodomain-mediated recruitment of canonical PRC1 [20–22]. The identification of non-canonical PRC1 complexes containing RYBP/YAF2 instead of a CBX subunit demonstrated that variant PRC1 complexes can target chromatin independently of H3K27me3 [23, 24]. For example, non-canonical PRC1.1, containing PCGF1, is targeted to non-methylated CpG islands by its KDM2B subunit [25–27], and can initiate Polycomb silencing by robust H2AK119ub deposition, enabling JARID2/AEBP2-mediated PRC2.2 recruitment [8, 28–32]. So, Polycomb silencing can be initiated by (i) PRC2.1 chromatin targeting followed by PRC1 tethering, or (ii) by PRC1.1 binding followed by subsequent PRC2.2 and PRC1 recruitment. However, it remains poorly understood whether these modes of Polycomb-mediated repression collaborate on the same target genes or function independently.

We have previously shown that the PRC1.1 complex is essential for leukemic stem cell viability, its downregulation impairs leukemogenesis in vivo, and that primary acute myeloid leukemia (AML) cells are sensitive to USP7 inhibition-induced PRC1.1 destabilization [33, 34]. The chromatin-binding subunit KDM2B is frequently overexpressed in various cancer types, and loss of KDM2B expression abolishes tumor formation [34–38]. Whereas all these observations suggest an oncogenic role for PRC1.1 in cancer, paradoxically, the PRC1.1 subunits BCOR and BCORL1 are frequently mutated in AML patients, and these mutations are associated with adverse risk [39, 40]. BCOR/BCORL1 mutations in AML are typically loss-of-function mutations leading to frameshifts/early stop codons, resulting in nonsense-mediated decay of messenger RNA (mRNA) or the expression of truncated proteins [41–43], which seems in apparent contrast with an oncogenic dependency on PRC1.1 in hematopoietic malignancies. Based on these data, we hypothesize that BCOR/PRC1.1 is required for viability of established AML cells, whereas it acts as a tumor suppressor in healthy hematopoietic stem and progenitor cells (HSPCs) prior to initiation of the transformation process. It is not well understood why BCOR loss induces such distinct phenotypes in different cellular backgrounds and how that impacts on PRC1.1 functionality.

Here, we investigated PRC1.1 dependency of leukemic cells using BCOR and KDM2B degron models. We show that BCOR is continuously required to connect the catalytic core and chromatin-binding part of the PRC1.1 complex, to maintain H2AK119ub at PRC1.1-bound sites, and that BCOR and KDM2B loss immediately impairs PRC1.1-mediated transcriptional control. We find that the epigenetic ground state of target genes determines the transcriptional kinetics downstream of PRC1.1 loss, and that PRC1.1 can silence target genes independent of PRC2.2. Finally, we find that PRC1.1 and PRC2 together maintain a differentiation block in leukemic cells.

Materials and methods

Cell culture and treatment

K562 cells were cultured in RPMI 1640 (Bio-Whittaker, Lonza, Verviers, Belgium), including 10% heat-inactivated fetal bovine serum (FCS, HyClone Laboratories, Logan, Utah, US) and 1% penicillin/streptomycin (p/s, PAA Laboratories). Cells were cultured at 37°C in 5% CO₂. Cells were treated with 1 µM 5-phenyl-indole-3-acetic acid (5-Ph-IAA; BioAcademia), GSK343 (MedChemExpress; various concentrations), or 5 µM UNC1999 (MedChemExpress).

Generation of constructs

The BCOR-mAID and KDM2B-mAID HDR template plasmids were generated by amplification of the genomic region surrounding the stop codon of both genes, ~800 bp, in the pJet1.2 vector. Next, using inverted polymerase chain reaction (PCR), a BamHI site was introduced at the center of this genomic fragment, and simultaneously the stop codon was removed and the PAM sequence mutated. Subsequently, we subcloned the mAID-mCherry-NeoR cassette from the pMK292 mAID-mCherry-NeoR plasmid into the BamHI site. Finally, we incorporated a Spot-P2A moiety in between the mAID and mCherry portion to allow immunoprecipitation of the fusion proteins and monomeric mCherry expression as a marker for proper integration downstream of BCOR or KDM2B. pRRL-SFFV-OsTIR1F74G-IRES-mBlueberry was generated by subcloning OsTIR1F74G from pMK381 AAVS1 CMV-OsTIR1F74G (Addgene #140536) and ligating this into the pRRL-SFFV-IRES-mBlueberry2 lentiviral vector. pRRL-SFFV-PCGF1-GFP, pRRL-SFFV-GFP-RING1B, pRRL-SFFV-KDM2Blf-GFP, pRRL-SFFV-KDM2Bsf-GFP, pRRL-SFFV-KDM2BΔ(CxxC)-GFP, pRRL, and pRRL-SFFV-KDM2BΔ(LRR)-GFP lentiviral were described previously [33, 34].

Generation of knock-in and knock-out cell lines

To generate K562 BCOR-mAID and KDM2B-mAID knock-in models, cells were electroporated with a CAS9-GFP expression plasmid (pMJ920; Addgene #42234), a gRNA expression plasmid, and an HDR template plasmid (pJet1.2 BCOR-mAID-Spot-P2A-mCherry or pJet1.2 KDM2B-mAID-Spot-P2A-mCherry) using an Amaxa electroporation device (Amaxa Nucleofector 1; program T-16). Single guide RNAs were designed to cut directly downstream to the stop codon of the BCOR or KDM2B genes. The HDR template contained 400-bp homology arms and contained a mutated PAM sequence to prevent recutting after proper integration. Seven days after electroporation, single cells expressing mCherry were sorted in 96-wells plates to generate isogenic clones. Next, clones were expanded and selected based on mCherry expression, proper mAID integration using PCRs on gDNA, and expression of mAID fusion mRNA (but not wild-type mRNA) using quantitative PCRs (qPCR) on complementary DNA (cDNA). Selected clones were transduced with a pRRL-SFFV-OsTIR1F74G-IRES-mBlueberry lentiviral vector, and cells double-positive for mCherry and mBlueberry were sorted 3 days after transduction. Next, for BCORL1 knockout, K562 BCOR-mAID-Spot-P2A-mCherry OsTIR1F74G-expressing cells were electroporated with an RNP complex generated from 10 µg of Cas9 and 6 µg of a BCORL1 single guide RNA (sgRNA), and mCherry/mBlueberry double-positive single cells were sorted into 96-well plates 5 days after electroporation. The BCORL1 sgRNA was designed to target the start of exon 4 of the BCORL1 gene and was generated by Sp6-mediated transcription. BCORL1 knockout of individual clones was done using PCR amplification of a region surrounding the gRNA targeting site, followed by Sanger sequencing and TIDE analysis (https://tide.nki.nl/). For pull out and rescue experiments BCOR-mAID or KDM2B-mAID cells were transduced with previously described pRRL-SFFV-PCGF1-GFP, pRRL-SFFV-GFP-RING1B, pRRL-SFFV-KDM2Blf-GFP, pRRL-SFFV-KDM2Bsf-GFP, pRRL-SFFV-KDM2B Δ(CxxC)-GFP, pRRL-SFFV-KDM2B Δ(LRRs)-GFP, or pRRL-SFFV-GFP lentiviral vectors [33, 34] and sorted for GFP, mCherry, and mBlueberry positive expression.

JARID2 knockout cells were generated using a sgRNA targeting exon 3. AEBP2 knockout cells were made by targeting exon 2, which is shared between the short and long isoform of AEBP2. To generate K562 KDM2B-mAID SUZ12 knock out cells exon 1 of SUZ12 was targeted, which is incorporated in all different splice forms of SUZ12. K562 KDM2B-mAID cells were washed and resuspended in K562 electroporation buffer, followed by addition of precomplexed Cas9 (12 µg) and sgRNA (6 µg), and the mixture was pulsed with program CM-137 and resuspended in growth medium. All sgRNAs used in this study can be found in Supplementary Table S9.

Lentiviral transductions

Lentiviral transductions were performed as described earlier [14]. Briefly, 5 × 10⁵ K562 cells were incubated with incubated with virus in the presence of 8 mg/ml polybrene (Merck). The next day, virus was washed away with three consecutive washes with phosphate buffered saline (PBS) containing 5% FCS, and cells were resuspended in RPMI, 10% FCS, P/S. For knockdown studies, we used previously validated short hairpin RNAs (shRNAs) that were expressed from a lentiviral pLKO.1 mCherry vector [14, 34]. All shRNAs used in this study can be found in Supplementary Table S9.

Flow cytometry analysis

In this study, cells were sorted using a MoFlo Astrios (Beckman Coulter), and flow cytometry analysis was performed using a NovoCyte Quanteon flow cytometer (Agilent). Resulting data were processed with FlowJo software (Tree Star Inc., Ashland, OR, USA). For antibody staining, anti-human FcR Blocking Reagent (Miltenyi Biotec) was used to prevent non-specific FC receptor binding. Subsequently, samples were incubated with anti-CD41-APC or anti-CD61-APC (BioLegend) for 20 min at 4°C and washed twice with PBS before analyzing. Annexin V-APC (BioLegend) staining was performed in Annexin V Binding Buffer (BD Biosciences), and samples were measured after 20-min incubation at 4°C. THP-1 BCOR-mAID cells were stained with anti-CD32-FITC (Biolegend) or anti-CD35-FITC (PharMingen).

Pull outs

For pull outs, nuclear extraction was carried out on 1 × 10⁷ cells for small-scale and 1–2 × 10⁸ cells for large-scale pull outs. Nuclear extract preparation was done as described previously [14]. For large-scale pull outs, nuclear extracts were precleared by incubating them with pre-equilibrated magnetic agarose beads (Chromotek) at 4°C on a rotating platform. Subsequently, the samples were incubated with pre-equilibrated GFP-Trap or Spot-Trap magnetic agarose beads (Chromotek), for 1h at 4°C on a rotating platform. After incubation, beads were washed six times with wash buffer (TBS, 0.3% IGEPAL CA-630, 1× CLAP, 0.1 mM PMSF), and subsequently boiled for 10 min in 2× Laemmli sample buffer. For endogenous pull outs, nuclear extracts were pre-cleared by incubation with pre-equilibrated Dynabeads Protein G (Invitrogen) at 4°C. Subsequently, pre-cleared nuclear extracts were incubated overnight with either 5 µl anti-RING1B antibody (D22F2; Cell Signaling Technology) or 5 µl anti-KDM2B antibody (09-864, Merck), supplemented together with 25 µl Dynabeads Protein G. The next day, beads were washed and processed as indicated earlier.

Proteomics sample preparation

Immunoprecipitated samples were loaded on 4%–20% Mini-PROTEAN® TGX™ Precast Protein Gels (Bio-Rad), followed by Coomassie Blue staining (Imperial™ Protein Stain; Thermo Scientific), destaining using Milli-Q H₂O, and excision of the band containing all proteins. The gel band was sliced into small pieces, washed subsequently with 30% and 50% v/v acetonitrile in 100 mM ammonium bicarbonate (dissolved in Milli-Q H₂O), each incubated at RT for 30 min while mixing (500 rpm), and lastly with 100% acetonitrile for 5 min, before drying the gel pieces at 55°C. The proteins were reduced with 10 mM dithiothreitol (in 100 mM ammonium bicarbonate dissolved in Milli-Q H₂O, 30 min, 55°C) and alkylated with 55 mM iodoacetamide (in 100 mM ammonium bicarbonate dissolved in Milli-Q H₂0, 30 min, in the dark at RT). The gel pieces were washed with 100 mM ammonium bicarbonate for 10 min, followed by a 100% acetonitrile wash for 30 min while mixing (500 rpm), and dried in an oven at 55°C before overnight digestion with using 5 ng/μl trypsin (sequencing-grade modified trypsin V5111, Promega; diluted in 100 mM ammonium bicarbonate) at 37°C. The next day, the peptides were first eluted from the gel pieces using 5% v/v formic acid in Milli-Q H₂O, followed by a second elution using 75% v/v acetonitrile plus 5% v/v formic acid. The combined eluted fractions were dried under vacuum and resuspended in 25 μl 5% v/v formic acid in Milli-Q H₂O.

Discovery-based proteomics analyses

Discovery mass spectrometric analyses were performed on a quadrupole Orbitrap mass spectrometer equipped with a nano-electrospray ion source (Orbitrap Exploris 480, Thermo Scientific). Chromatographic separation of the peptides was performed by liquid chromatography (LC) on a Evosep system (Evosep One, Evosep) using a nano-LC column (EV1137 Performance column 15 cm × 150 µm, 1.5 µm, Evosep; buffer A: 0.1% v/v formic acid, dissolved in Milli-Q H₂O; buffer B: 0.1% v/v formic acid, dissolved in acetonitrile). The digests were injected in the LC-MS with 20 µl of the peptide digests, and the peptides were separated using the 30SPD workflow (Evosep). The mass spectrometer was operated in positive ion mode and data-independent acquisition mode (DIA) using isolation windows of 16 m/z with a precursor mass range of 400–1000, switching the FAIMS between CV-45V and -60V with three scheduled MS1 scans during each screening of the precursor mass range. LC-MS raw data were processed with Spectronaut (version 16.0.220606) (Biognosys) with a Human SwissProt database (20 376 entries) using the standard settings of the directDIA workflow, except that quantification was performed on MS1. For the quantification, the Q-value filtering was set to the classic setting with local normalization and no imputation. For downstream processing LFQ values were processed using Perseus 1.5.8.5 and volcano plots were generated.

Western blot

Western blotting was performed as previously described [14]. The following primary antibodies were used in this study: anti-KDM2B (09-864, Merck, 1:500), anti-BCOR (sc-514576, Santa Cruz Biotechnology, 1:200), anti-BCORL1 (PA5-56855, Thermo Fisher, 1:1000), anti-PCGF1 (ab183499 or ab259943, Abcam, 1:1000), anti-USP7 (A300-033A, Bethyl Laboratories, 1:1000), anti-TRIM27 (18 791, IBL, 1:1000), anti-SKP1 (D3J4N, Cell Signaling Technology, 1:1000), anti-RING1A (ab180170, Abcam, 1:1000), anti-RING1B (ab181140, Abcam, 1:1000), anti-JARID2 (D6M9X, Cell Signaling Technology, 1:1000), anti-AEBP2 (D7C6X, Cell Signaling Technology, 1:1000), anti-SUZ12 (D39F6, Cell Signaling Technology, 1:1000), anti-GFP (ab290, Abcam, 1:1000), and anti-bActin (C4, Santa Cruz Biotechnology, 1:1000). Swine anti-rabbit immunoglobulins/HRP (Agilent-Dako) and rabbit anti-mouse immunoglobulins/HRP (Agilent-Dako) were used as secondary antibodies. For endogenous pull outs we used light-chain specific mouse anti-rabbit IgG HRP-conjugated monoclonal antibody (#93702; Cell Signaling Technology). Imaging was carried out using a ChemiDoc XRS + System (Biorad).

RNA sequencing and quantitative real-time PCR

For RNA sequencing, total RNA was extracted from these cells using the RNeasy Mini Kit from Qiagen (Venlo, The Netherlands) according to the manufacturer’s recommendations. RNA samples were qualified by automated gel electrophoresis on the 2200 TapeStation System (Agilent Technologies). Using KAPA RNA HyperPrep kit with riboErase (HMR) from Roche Sequencing and Life Sciences, cDNAs were synthesized from fragmented RNA, and the sequences library were generated according to the manufacturer’s recommendations. These libraries were then sequenced on an Illumina NextSeq500 using standard settings, and the single-end reads obtained from treated cells were analyzed with Strand Avadis NGS (v3.0) software (Strand Life Sciences Pvt. Ltd). The STAR aligner [44] was used in two-pass mode with default settings to map the obtained paired-end reads from treated cells to the hg38 reference genome. Alternatively, RNA-seq libraries were prepared using the Smart-3SEQ protocol from 100 ng of total RNA input as previously described [45] at IPsomics (https://ipsomics.com/). Next, libraries were pooled and sequenced using an Illumina NovaSeqX plus 150PE at Novogene (10B flow cell). Raw reads were trimmed using TrimGalore v.0.6.7 [46]. Using the DESeq2 package [47], the data were normalized and differentially expressed genes (DEGs) were found. Genes with an adjusted P-value < .05 and log₂FoldChanges > < 0.585 were found to be differentially expressed and were used in further analysis. Log-log correlations were done with a Pearson’s test. Quantitative RT-PCR was performed using the iScript cDNA Synthesis Kit (Bio-Rad). The resulting cDNA samples were then amplified using SsoAdvanced SYBR Green Supermix (Bio-Rad) on a CFX384 Touch Real-Time PCR Detection System (Bio-Rad).

Chromatin immune precipitation

Chromatin immune precipitation (ChIP) was essentially performed as described previously [34]. Briefly, cells were cross-linked using PBS including 1% formaldehyde for 10 min. Cross-linked samples were quenched using 125 mM glycine, followed by a PBS wash and snap freezing sodium dodecyl sulfate (SDS) buffer (100 mM NaCl, 0.5% SDS, 5 mM ethylenediaminetetraacetic acid (EDTA), 50 mM Tris–HCl (pH 8.0), protease inhibitors (CLAP and 0.1 mM PMSF), and 0.2% NaN₃. Next, cross-linked samples were IP buffer containing: 2× SDS buffer and 1× Triton Dilution buffer (5% Triton X-100, 5 mM EDTA, 100 mM Tris–HCl, pH 8.6, 100 mM NaCl, 0.2% NaN3, 1× CLAP, and 0.1 mM PMSF). Sonication was performed using a Bioruptor (Diagenode) or PIXUL® Multi-Sample Sonicator (Active Motif). Then, the samples were precleared using 1 mg pre-equilibrated Dynabeads Protein G (Invitrogen) for 20 min, and subsequently, ChIP reactions were conducted in 1 ml IP buffer and incubated overnight at 4°C with 5% mouse spike-in control chromatin and 2 µg of either of the following antibodies: anti-GFP (ab290, Abcam), IgG (C15410206, Diagenode), anti-KDM2B (09-846, Merck), anti-BCOR (12107–1-AP, Proteintech), anti-PCGF1 (ab259943, Abcam), anti-EZH2 (D2C9, Cell Signaling Technology), anti-H2AK119ub (D27C4, Cell Signaling Technology), anti-H3K4me3 (C15410003, Diagenode), anti-H3K27ac (C15410196, Diagenode), and anti-H3K27me3 (C15410069, Diagenode). Next, ChIP reactions were incubated for 3 h with 25 ml pre-equilibrated Dynabeads Protein G (invitrogen) at 4°C and subsequently washed with Mixed Micelle Wash buffer (150 mM NaCl, 20 mM Tris–HCl, pH 8.1, 5 mM EDTA, 5.2% sucrose, 1% Triton X-100, 0.2% SDS, 0.2% NaN₃, 3 washes), buffer 500 (50 mM Hepes-KOH, pH 7.5, 500 mM NaCl, 1% Triton X-100, 0.1% sodium deoxycholate, 0.2% NaN₃; 2 washes), LiCl Wash buffer (250 mM LiCl, 10 mM Tris–HCl, pH 8.0, 1 mM EDTA, 0.5% IGEPAL CA-630, 0.5% sodium deoxycholate, 0.2% NaN₃; 2 washes), and TE buffer (10 mM Tris–HCl, pH 8.0, 1 mM EDTA; 1 wash). Finally, chromatin was eluted from the beads using Elution Buffer (0.1 M NaHCO₃, 0.1% SDS) and reverse cross-linked overnight at 65°C. Finally, 2 µl RNase A (10 mg/ml; Thermo Fisher) was added and samples were incubated for 2 h at 37°C, followed by addition of 2 µl Proteinase K (20 mg/ml; Thermo Fisher) and incubation for 2 h at 55°C. Finally, the DNA was extracted using a PCR purification kit (Qiagen). The purified DNA samples were analyzed by qPCR of sequencing.

ChIP-seq analysis

ChIP-seq libraries were prepared with the KAPA Hyper Prep Kit (Roche Sequencing and Life Sciences) following the manufacturer’s instructions and sequenced on an Illumina NextSeq500 using default parameters. Subsequently, quality control was done on the obtained paired-end reads using FastQC tools. Next, the reads were aligned to the hg38 reference genome using Burrows–Wheeler Aligner [48] with default settings. Aligned reads were further processed using SAMtools. Then, the generated aligned reads were sorted, indexed, and prepared in two different files containing reads that aligned to human chromosomes and a file containing reads that aligned to mouse chromosomes using SAMtools [49]. Spike-in normalization factors were calculated per sample. Using BEDtools genomecov, the total number of overlapping fragments at each position in the genome was determined and scaled according to the spike-in normalization factor. The resulting BedGraph files were converted to BigWig files using UCSC bedGraphToBigWig. Ngs.plot [50] was used to generate average plots of read count per reference million mapped reads, and heatmaps, representing a −/+ 5 Kb region around the transcription start site (TSS) or gene body. Peak calling was performed on the BAM files using MACS3 [51]. Tracks were visualized using the Integrative Genomics Viewer [52].

Data analysis

All sequencing data are available in Gene Expression Omnibus under GEO numbers GSE286991 (RNA-seq) and GSE286993 (ChIP-seq). Proteomics data have been deposited in the Proteomics Identification Database under PRIDE number PXD065157. The accession number for previously published AML ChIP-seq data used for analysis in this study is GSE54580 [34], which were aligned to the hg38 reference genome. Gene Ontology (GO) analysis was performed using DAVID [53, 54]. Gene Set Enrichment Analysis (GSEA) was performed using GSEA 4.2.3.

Results

BCOR-mAID is functional and efficiently degraded

To study BCOR/PRC1.1 dependency in leukemic cells, we applied the mini-auxin-inducible degron 2 system, allowing rapid degradation of endogenous proteins [55, 56]. We used a CRISPR-mediated knock-in strategy to integrate an mAID-Spot-P2A-mCherry cassette in-frame at the C-terminus of BCOR in leukemic K562 cells, resulting in the expression of a BCOR-mAID-Spot (hereafter BCOR-mAID) fusion protein and a separate mCherry monomer (Fig. 1A). Single cells expressing the fusion protein were sorted on the basis of mCherry expression, and four clones were selected for further analysis on the basis of mCherry mean fluorescent intensity (Supplementary Fig. S1A). Quantitative PCR analysis showed that all selected clones expressed the BCOR-mAID fusion mRNA and lacked wild type BCOR expression (Fig. 1B). Next, OsTIR1(F74G) [56] was lentivirally expressed, transduced cells were sorted on the basis of mBlueberry2 expression (Supplementary Fig. S1B), and cells were treated with 1 µM 5-phenyl-indole-3-acetic acid (5-Ph-IAA; Fig. 1A,C). 5-Ph-IAA treatment led to a rapid degradation of BCOR-mAID, which was already apparent after 15 min. To verify that the fusion protein was functional and incorporated into PRC1.1, we performed pull outs using Spot-Trap beads on nuclear extracts from both wild-type and BCOR-mAID K562 cells (Fig. 1D). PRC1.1 subunits KDM2B, RING1A, RING1B, and PCGF1 efficiently coprecipitated with BCOR-mAID, indicating that the C-terminal mAID-Spot moiety does not affect BCOR functionality.

Figure 1.

BCOR is a bridging factor between the catalytic and chromatin-binding moiety of the PRC1.1 complex. (A) Overview of BCOR-mAID-Spot-P2A-mCherry (referred to as BCOR-mAID) model generation. (B) Quantitative PCR showing wild-type BCOR and BCOR-mAID expression in individual clones (error bars represent technical triplicates; n = 3). (C) Western blot showing 5-Ph-IAA-induced degradation of BCOR-mAID (asterisk marks non-specific band) in BCOR-mAID knock-in cells expressing TIR1F74G within 1 h. (D) Western blot analysis of Spot-Trap-mediated pull-outs in wild-type and BCOR-mAID-Spot expressing K562 cells, showing input (I), non-bound (NB) and bound (B) fractions. (E) Volcano plot of label-free quantification of triplicate measurements of KDM2B-GFP pull outs from BCOR-mAID;BCORL1 KO cells, treated for 2 h either with DMSO or 1 µM 5-Ph-IAA. Western blots of endogenous RING1B (F) and KDM2B (G) pull outs in K562 BCOR-mAID;BCORL1 KO cells showing input (I), NB and B fractions. (H) Volcano plot showing label-free quantification of triplicate measurements of PCGF1-GFP pullouts from KDM2B-mAID cells, treated for 2 h either with DMSO or 1 µM 5-Ph-IAA. (I) Western blots of endogenous RING1B pull out in K562 KDM2B-mAID cells showing input (I), NB and B fractions. (J) Endogenous ChIP-qPCR experiments in K562 BCOR-mAID;BCORL1 KO cells using antibodies directed against BCOR (n = 4), KDM2B (n = 4), or PCGF1 (n = 5). (K) Endogenous ChIP-qPCR experiments in K562 KDM2B-mAID cells using antibodies directed against BCOR (n = 3), KDM2B (n = 4), or PCGF1 (n = 3). Statistical analysis was performed using Student’s t test; ^∗P < .05 and ^∗∗P < .01.

BCOR is a bridging factor in the PRC1.1 complex

Structural studies have shown that the C-terminal PUFD domain of BCOR and BCORL1 can interact with the RAWUL domain of the PRC1.1 subunit PCGF1, resulting in an interaction surface for the PRC1.1 chromatin tethering factor KDM2B [57, 58]. To study the immediate effect of BCOR-mAID degradation on PRC1.1 complex integrity, KDM2B-GFP or PCGF1-GFP were lentivirally expressed in K562 BCOR-mAID cells and GFP-Trap pull outs were performed in cells treated for 2 h with 5-Ph-IAA or not (Supplementary Fig. 1C). Whereas KDM2B-GFP and PCGF1-GFP were efficiently precipitated in both conditions, BCOR-mAID, as expected, was undetectable in pull outs from 5-Ph-IAA-treated cells. A strong loss of PCGF1, RING1A, RING1B, USP7, and TRIM27 co-precipitation with KDM2B-GFP was observed, whereas SKP1 binding remained unaffected. PCGF1-GFP pull outs displayed an immediate loss of the KDM2B/SKP1 heterodimer upon BCOR-mAID degradation, whereas RING1A and RING1B binding were not affected. Taken together, these data underline a requirement for BCOR as a bridging factor to warrant PRC1.1 integrity, holding together the chromatin binding KDM2B-SKP1 heterodimer and the catalytic PRC1.1 core, in line with previous studies using constitutive knockout models [59].

Full incorporation of USP7 and TRIM27 in PRC1.1 was BCOR-mAID-dependent in both KDM2B-GFP and PCGF1-GPF pull outs, suggesting that integration of this heterodimer is dependent on both the chromatin-binding and catalytic core of PRC1.1. BCORL1 can theoretically take over the function of BCOR given their mutually exclusive PUFD domain-dependent incorporation into PRC1.1 and largely overlapping interaction domains. Indeed, we observed that both KDM2B-GFP and PCGF1-GFP precipitated BCORL1 in the absence of BCOR, and that BCOR-mAID cells upregulated BCORL1 upon BCOR-mAID degradation (Supplementary Fig. S1C and D). Therefore, BCORL1 was knocked out using a gRNA targeting exon 4 and a single cell-derived clone was isolated with a two base pair frame-shift leading to an early stop codon (Fig. 1A and Supplementary Fig. S1D and E). Next, we lentivirally expressed KDM2B-GFP and performed large scale pull outs on nuclear extracts from BCOR-mAID;BCORL1 KO cells treated for 2 h with 5-Ph-IAA or DMSO, followed by LC-MS/MS analyses (Fig. 1E and Supplementary Table S1). A strong loss of all KDM2B-interacting PRC1.1 subunits was observed, with the exception of SKP1. In line with these data, endogenous RING1B and KDM2B pull outs showed a strong loss of KDM2B and RING1B coprecipitation, respectively, upon BCOR-mAID degradation (Fig. 1F and G), in line with a bridging function of BCOR within the PRC1.1 complex. In addition, our LC-MS/MS analysis showed a decreased interaction of KDM2B with CBX8 and MLLT1/ENL (Fig. 1E). It has previously been shown that BCOR can directly interact with wild-type MLLT4/AF9 and both AF9 and ENL in the context of an MLL-AF9 and MLL-ENL fusion protein [60]. Our data suggest that in MLL wild-type cells, BCOR may form a molecular bridge between PRC1.1 and the super elongator complex (SEC) in an ENL-dependent manner, in line with our previous findings showing strong interactions between KDM2B and the SEC [33].

BCOR binding to the catalytic PRC1.1 core is independent of KDM2B

Next, we investigated whether BCOR integration in PRC1.1 would require the presence of the KDM2B-SKP1 heterodimer. Therefore, single cell-derived K562 KDM2B-mAID-Spot-P2A-mCherry (hereafter KDM2B-mAID) clones were generated expressing KDM2B-mAID but not wild type KDM2B (Supplementary Fig. S1F). OsTIR1(F74G) was expressed in these cells and 5-Ph-IAA treatment induced a rapid degradation of KDM2B-mAID (Supplementary Fig. S1G). Subsequently, we ectopically expressed GFP-RING1B, performed GFP-Trap pull outs, and observed a complete loss of KDM2B-mAID co-precipitation after 2 h of 5-Ph-IAA treatment (Supplementary Fig. S1H), showing that KDM2B-mAID is effectively degraded. KDM2B-mAID pull outs showed co-precipitation of RING1A, RING1B, and PCGF1 demonstrating its incorporation into PRC1.1 (Supplementary Fig. S1I). Next, we expressed PCGF1-GFP in these cells and performed large scale GFP-Trap pull outs after 2 h of treatment with DMSO or 5-Ph-IAA, followed by LC-MS/MS analysis (Fig. 1H and Supplementary Table S2). While KDM2B and SKP1 were lost from the PRC1.1 complex, this was not observed for BCOR and BCORL1, indicating that BCOR/BCORL1 incorporation into PRC1.1 is independent of the KDM2B/SKP1 heterodimer and mostly depends on the interaction with the RAWUL domain of PCGF1. In line with this observation, endogenous RING1B pull outs in KDM2B-mAID cells persistently showed BCOR co-precipitation after KDM2B-mAID degradation (Fig. 1I). To assess the impact of BCOR-mAID degradation on PRC1.1 chromatin tethering, we performed ChIP experiments using antibodies directed against endogenous BCOR, KDM2B, and PCGF1 (Fig. 1J). As expected, BCOR-mAID depletion resulted in a reduction of BCOR and PCGF1 binding, whereas KDM2B chromatin-binding was not affected. The latter observation underlines the notion that KDM2B is still capable of binding PRC1.1 target genes in the absence of the catalytic core of the complex. In contrast, ChIP experiments in KDM2B-mAID cells showed that KDM2B-mAID degradation induced loss of KDM2B, BCOR, and PCGF1 (Fig. 1K). The effects of BCOR-mAID depletion on BCOR and PCGF1 chromatin binding were somewhat moderate compared to the KDM2B-mAID model, which may be a consequence of different degradation kinetics of BCOR-mAID in its chromatin-bound form compared to KDM2B-mAID.

BCOR locally controls H2AK119ub levels at PRC1.1 target genes

Next, we evaluated the effect of BCOR-mAID degradation on the epigenetic landscape. ChIP-seq tracks were generated for various histone marks in BCOR-mAID;BCORL1 KO cells treated with DMSO or 5-Ph-IAA for 8 h. In line with a role of BCOR in bridging the chromatin binding and catalytic part of the PRC1.1 complex, BCOR-mAID depletion led to an overall reduction in H2AK119ub levels across TSSs genome wide after 8 h 5-Ph-IAA treatment, whereas PRC2-dependent H3K27me3 marking, H3K27ac and H3K4me3 levels were unchanged (Fig. 2A and Supplementary Fig. S2A) [34]. To evaluate epigenetic changes at genes targeted by PRC1.1, canonical PRC1/2, or both, we generated endogenous BCOR and EZH2 ChIP-seq tracks in K562 BCOR-mAID;BCORL1 KO cells. This led to the identification of 5575 shared target genes, but also 7360 targeted by BCOR alone, and 855 genes that solely displayed EZH2 tethering (Fig. 2B and Supplementary Table S3). Of note, whereas BCOR and EZH2 single-targeted genes clearly showed strong enrichment over NB genes, common genes showed slightly stronger enrichment for BCOR and EZH2 (Fig. 2C). H2AK119ub levels were strongly reduced at genes targeted by BCOR alone and genes targeted by both BCOR and EZH2, whereas EZH2-only genes were only mildly affected (Fig. 2D and E). Screenshots of representative examples of BCOR only, both and EZH2 only target genes can be found in Fig. 2F. The loss of H2AK119ub at BCOR-associated genes was most pronounced in the immediate vicinity of BCOR peaks. Based on these observations, we hypothesized that other (variant) PRC1 complexes are responsible for H2AK119ub deposition at genomic loci where PRC1.1 is not enriched. To test which PRC1 complex is dominantly controlling H2AK119ub marking, we expressed previously validated shRNAs targeting various PCGFs and RING1 proteins (Supplementary Fig. S2B) [14, 34]. Western blot analysis showed that knockdown of PCGF1, PCGF2, and PCGF4 impacted on H2AK119ub, whereby PCGF4 knockdown had the strongest impact on H2AK119ub levels (Supplementary Fig. S2C), indicating that the canonical PRC1.4 complex is likely the major contributing complex to genome-wide H2AK119ub deposition in these cells. Also RING1B knockdown significantly affected H2AK119ub, whereas milder effects were observed upon knockdown of PCGF6 and RING1A. H3K27me3, H3K27ac, and H3K4me3 levels were not changed upon BCOR-mAID depletion and BCOR only target genes are devoid of H3K27me3, which is not in line with earlier observations that PRC1.1-mediated H2AK119ub deposition precedes PRC2.2-dependent H3K27me3 deposition [8, 28–32].

Figure 2.

BCOR degradation leads to loss of H2AK119ub at target genes but does not affect H3K27me3 deposition. (A) Average ChIP-seq profile showing relative intensities in two duplicate samples of various epigenetic marks at TSSs throughout the genome in K562 BCOR-mAID;BCORL1 KO cells treated for 8 h with DMSO or 1 µM 5-Ph-IAA. (B) Venn diagram showing overlap of BCOR and EZH2 target genes in K562 BCOR-mAID;BCORL1 KO cells. (C) Average ChIP-seq profile showing relative intensities of BCOR and EZH in K562 BCOR-mAID;BCORL1 KO cells at “BCOR only,” “EZH2 only,” and “both” target genes. (D) Average ChIP-seq profile showing relative intensities in two duplicate samples of various epigenetic marks at TSSs in K562 BCOR-mAID;BCORL1 KO cells treated for 8 h with DMSO or 1 µM 5-Ph-IAA. (E) Heat maps showing the average density signal for 5 kb regions surrounding TSSs of indicated groups for BCOR, EZH2, H2AK119ub, H3K27me3, H3K27ac, and H3K4me3. For histone marks, duplicate tracks are shown for DMSO and 5-Ph-IAA treatments. (F) Representative examples of BCOR-only, both, and EZH2-only genes, showing enrichment for BCOR, EZH2, H2AK119ub, H3K27me3, H3K27ac, and H3K4me3.

BCOR/KDM2B degradation results in immediate transcriptional changes

Next, we focused on the direct effects of BCOR-mAID or KDM2B-mAID degradation on gene expression. In both BCOR-mAID;BCORL1 KO and KDM2B-mAID cells, RNA-seq was performed on control samples or after 2, 8, 24, and 72 h of 5-Ph-IAA treatment. Principal component analysis clearly showed that 5-Ph-IAA treated samples over time increasingly deviated from untreated controls in both models (Supplementary Fig. S3A). This was reflected in our DEGs analyses, where after 2 h of 5-Ph-IAA treatment, the BCOR-mAID;BCORL1 KO cells showed significant upregulation of 77 genes, which extended to 278 at 8 h, and eventually reached 543 significantly upregulated genes at 72 h (Fig. 3A and Supplementary Table S4). GO analysis showed that upregulated genes were enriched for classical Polycomb-related GO terms associated with development (Supplementary Fig. S3B). For KDM2B-mAID cells a similar pattern was observed, where 164 genes were already significantly upregulated 2 h after 5-Ph-IAA induction, increasing to 455 after 8 h and reaching a maximum 1206 genes after 24 h with similar GO terms (Fig. 3B, Supplementary Fig. S3B, and Supplementary Table S5). We also observed a small set of downregulated genes in both models, suggesting that PRC1.1 mostly has a gene-repressive role but is also important to maintain transcriptional activity of a confined set of genes, mostly related to nucleosome assembly and RNA processing (Supplementary Fig. S3B). Identified DEGs showed a considerable overlap between the various time points within both models (Fig. 3C). Whereas we consider early DEGs (2–24 h) as direct PRC1.1 targets, we anticipate that late DEGs (72 h) may be a consequence of secondary or even tertiary events. BCOR-mAID;BCORL1 KO and KDM2B-mAID DEGs showed a strong overlap at all time points (Fig. 3D), and we observed a significant correlation between gene expression changes in BCOR-mAID;BCORL1 KO and KDM2B-mAID models when assessing DEGs identified in either one of these cell lines (Fig. 3E). This suggests that BCOR and KDM2B degradation-induced gene expression changes are due to PRC1.1 loss of function, albeit that the number of DEGs was higher upon KDM2B loss. We speculate that remaining KDM2B monomer chromatin-binding upon BCOR-mAID depletion, as observed in our ChIP experiments in BCOR-mAID;BCORL1 KO cells (Fig. 1J), may partially prevent transcriptional changes of PRC1.1 target genes, suggesting that PRC1.1 target-gene activation is most apparent upon KDM2B depletion. GSEA in both BCOR-mAID;BCORL1 KO and KDM2B-mAID models showed that upregulated genes were strongly enriched for Polycomb-associated genes sets, like “PRC2 MODULE ARMSTRONG” and “BENPORATH_PRC2_TARGETS” (Fig. 3F and G), indicating that PRC1.1 controls Polycomb target genes. Downregulated genes in both models were enriched for gene signatures associated to RNAPI transcription, DNA replication, and DNA repair.

Figure 3.

BCOR/KDM2B loss directly impacts on target gene expression and regulates overlapping gene modules. (A, B) MA-plots of RNA-seq in BCOR-mAID;BCORL1 KO and KDM2B-mAID K562 cells (2, 8, 24, and 72 h, 1 µM 5-Ph-IAA, n = 3). DEGs are highlighted in red (log₂FC ←0.585, > 0.585, padj < 0.05). (C) Venn diagram of DEGs identified at indicated time points (2, 8, 24, and 72 h). (D) Venn diagram of DEGs of BCOR-mAID BCORL1 KO and KDM2B-mAID K562 cells (2, 8, 24, and 72 h). (E) Scatter plots depicting log₂ fold change (FC) of DEGs in KDM2B-mAID and BCOR-mAID;BCORL1 KO K562 cells at different timepoints (2–72 h). (F, G) GSEA of the transcriptomic data confirmed the role of PRC1.1 in maintaining repression of Polycomb target genes.

To test whether KDM2B can exert gene expression control independent of other PRC1.1 components, we investigated whether KDM2B-mAID depletion-induced gene expression changes could be rescued by overexpression of short- or long-form KDM2B-GFP (KDM2Bsf and KDM2Blf; two naturally occurring KDM2B variants), KDM2BΔ(CxxC)-GFP (lacking the DNA-binding domain) or KDM2BΔ(LRR)-GFP (Fig. 4A). We previously showed that KDM2Bsf-GFP, KDM2Blf-GFP, and KDM2BΔ(CxxC)-GFP are all incorporated in PRC1.1, whereas KDM2BΔ(LRR)-GFP only interacts with SKP1 [33]. Principal component analysis of RNA-seq data of K562 KDM2B-mAID cells showed that constitutive overexpression of either KDM2Bsf-GFP/KDM2Blf-GFP or KDM2BΔ(CxxC)-GFP readily induced opposite gene expression changes (Fig. 4B), suggesting a gain of function of wild-type KDM2B overexpression and a dominant-negative effect imposed by KDM2BΔ(CxxC)-GFP. We observed a strong overlap in DEGs between the different groups (Fig. 4C and Supplementary Table 6), and unsupervised clustering of these genes showed explicit opposing gene expression changes within the largest cluster (cluster 2) upon overexpression of KDM2Bsf-GFP or KDM2Blf-GFP, and to a lesser extend KDM2BΔ(LRR)-GFP, versus KDM2BΔ(CxxC)-GFP (Fig. 4D). Next, we analyzed the fold change of overlapping DEGs that we also found differentially expressed in our BCOR-mAID and KDM2B-mAID models (Fig. 3A), since these genes are likely directly regulated by PRC1.1. Clearly, KDM2Blf-GFP/KDM2Bsf-GFP expression versus KDM2BΔ(CxxC)-GFP expression resulted in mostly opposing fold changes in gene expression (Fig. 4E). In addition, 5-Ph-IAA treatment of KDM2B-mAID cells induced a transcriptional change that made these cells more similar to KDM2BΔ(CxxC)-GFP cells (Fig. 4B). Importantly, KDM2BΔ(CxxC)-GFP, KDM2Bsf-GFP, and KDM2Blf-GFP expressing cells were not sensitive to 5-Ph-IAA treatment, likely due to the fact that overexpression of KDM2Bsf or KDM2Blf overruled the effect of KDM2B-mAID loss on gene expression (Fig. 4F and Supplementary Table 7). In contrast, KDM2BΔ(LRR)-GFP expression did not override KDM2B-mAID loss-induced transcriptional changes, indicating that incorporation of KDM2B into PRC1.1 is a requirement to rescue the KDM2B-mAID degradation-induced gene expression changes. Of note, KDM2BΔ(LRR)-GFP expressing cells, although sensitive to KDM2B-mAID depletion, did show a slightly changed baseline gene expression which could be attributed to PRC1.1-independent KDM2B-mediated gene control (Fig. 4B). The presence of the JmjC domain in KDM2Blf-GFP did not change the transcriptional changes as compared to KDM2Bsf-GFP expression, suggesting that PRC1.1 transcription regulatory functions are independent of the KDM2B demethylase function. Indeed, KDM2Blf-GFP and KDM2Bsf-GFP had largely overlapping gene binding profiles, whereas KDM2BΔ(CxxC)-GFP did not significantly interact with chromatin (Fig. 4G and H). Finally, KDM2B-GFP long-form and short-form chromatin targeting strongly overlaps with endogenous BCOR binding in K562 cells (Fig. 4I and J), underlining that the GFP moiety does not change KDM2B target specificity. Taken together, we attribute the majority of KDM2B-mAID and BCOR-mAID DEGs to a PRC1.1 loss-of function phenotype.

Figure 4.

KDM2B overexpression rescue experiments are dependent upon PRC1.1 integration. (A) Schematic representation of the naturally occurring KDM2B short form (SF) and KDM2B long-form (LF; containing a Jumonji C [JmjC] domain), and two deletion mutants lacking either the chromatin binding CxxC domain or leucine-rich-repeats (LRRs). (B) Principal component analysis of RNA-seq data from KDM2B-mAID cells lentivirally expressing the indicated forms of KDM2B-GFP or empty vector, and treated for 24 h with DMSO or 5-Ph-IAA (n = 3). (C) Overlap of DEGs in KDM2B-mAID cells expressing indicated KDM2B proteins compared to empty vector controls. (D) Unsupervised clustering of overlapping DEGs in KDM2B-mAID cells expressing indicated KDM2B proteins compared to empty vector controls. (E) Scatter plot showing log₂ fold change (FC) of common DEGs in KDM2B LF versus KDM2BΔ(CxxC) and KDM2B SF versus KDM2BΔ(CxxC) cells. (F) MA-plots of RNA-seq data from KDM2B-mAID cells lentivirally expressing the indicated forms of KDM2B-GFP or empty vector, and treated for 24 h with DMSO or 5-Ph-IAA. Average plot (G) and heat map (H) of TSS-oriented KDM2B LF, KDM2B SF, and KDM2BΔ(CxxC) genome-wide chromatin binding. (I) Venn diagram showing overlap of endogenous BCOR, KDM2Blf-GFP, and KDM2Bsf-GFP target genes. (J) Representative examples of PRC1.1 target genes showing overlapping enrichment of endogenous BCOR and KDM2B LF, KDM2B SF, and KDM2BΔ(CxxC) GFP-fusions.

Early- and late-responding genes upon PRC1.1 loss display distinct epigenetic ground states

Next, we investigated whether gene sets could be identified with specific kinetic profiles among overlapping DEGs from both KDM2B-mAID and BCOR-mAID models. Therefore, we performed unsupervised clustering of all overlapping DEGs (n = 542) using the KDM2B-mAID RNA-seq data (Fig. 5A). Treatment time points clustered away from untreated cells depending on 5-Ph-IAA treatment time and three distinct clusters were identified: cluster one containing early upregulated genes (2 h), cluster two containing late upregulated genes (8–24 h), and cluster three containing downregulated genes. We hypothesized that a steady-state epigenetic imprint of genes from these different clusters may explain the response time of DEGs toward loss of KDM2B-mAID. Indeed, derepressed genes in cluster one and two showed an increased enrichment for H2AK119ub and H3K27me3 compared to cluster three, whereas the active mark H3K27ac was lower compared to cluster three genes (Fig. 5B). Late-responding genes (cluster two) displayed a much stronger enrichment for H2AK119ub and H3K27me3, whereas H3K27ac was lower compared to cluster one genes. H3K4me3 was high on all clusters but most prominent at cluster one and two, suggesting a bivalent epigenetic signature at these genes. Treating cells with 5-Ph-IAA strongly reduced H2AK119ub levels at early- and late-upregulated genes (cluster 1 and 2) but not downregulated genes (cluster three; Fig. 5B), other epimarks were unchanged. BCOR showed equal enrichment at early- and late-upregulated genes (cluster one and two), whereas downregulated genes (cluster three) only showed low BCOR binding. EZH2 binding was most prominent at late upregulated cluster two genes, in line with higher H3K27me3 levels at these genes compared to early upregulated genes.

Figure 5.

KDM2B-mAID degradation reveals early and late responding gene sets that display distinct epigenetic landscapes. (A) Unsupervised clustering analysis of RNA-seq normalized count data from KDM2B-mAID K562 cells (untreated or 2, 8, 24, and 72 h with 1 µM 5-Ph-IAA identifies three main clusters. (B) Average ChIP-seq profile of indicated histone marks of genes in cluster one, two, and three in BCOR-mAID;BCORL1 KO cells treated for 8 h with DMSO or 5-Ph-IAA. (C) Average ChIP-seq profile of BCOR and EZH2 of genes in cluster one, two, and three in BCOR-mAID;BCORL1 KO cells at baseline. (D) GO analysis (biological function and molecular function) of genes in cluster one and two.

We conclude that genes in cluster one are regulated by PRC1.1 alone and much more poised toward transcription compared to late-upregulated cluster 2 genes, which are regulated by PRC1.1 and PRC2 and displayed a bivalent epigenetic signature strongly enriched with repressive and active histone marks. Taken together, our data suggest that the transcriptional response toward loss of PRC1.1 is directly related to the steady-state epigenetic wiring of individual loci. GO analysis showed that genes in cluster one were associated with developmental processes and enriched for transcription factors, whereas cluster two contained many genes related to growth factor binding, signal transduction, and integrin binding (Fig. 5D).

PRC1.1 can exert gene-repressive functions independent of PRC2

The epigenetic signature of KDM2B-mAID degradation-induced late-upregulated genes displayed high H3K27me3 levels and strong enrichment of EZH2 (Fig. 5B and C), suggesting that PRC2 is involved in co-repressing this set of PRC1.1 target genes. Previously, it was shown that PRC1.1 can initiate Polycomb repression through strong H2AK119ub deposition, which is thought to induce JARID2/AEBP2-dependent PRC2.2 recruitment and H3K27me3 marking, followed by canonical PRC1 recruitment [8, 28–31]. To test whether PRC1.1 and PRC2 cooperatively contribute to control of PRC1.1 target genes we performed RNA-seq on KDM2B-mAID cells treated for 24 h with either 5-Ph-IAA, the EZH2 inhibitor GSK343, or both. Indeed, we observed a strong increase in DEGs and associated fold changes upon combined 5-Ph-IAA and GSK343 treatment compared to single treatments (Fig. 6A and Supplementary Table S8). Using unsupervised clustering we identified three gene clusters: (i) cluster A that contained genes that were co-dependent on PRC1.1 and PRC2, (ii) cluster B that was comprised of genes that were only upregulated upon GSK343 treatment, and (iii) cluster C that was enriched for genes that were downregulated upon treatment with both 5-Ph-IAA and GSK343 (Fig. 6B). Baseline epigenetic state analysis in KDM2B-mAID cells showed that cluster A, containing PRC1.1/PRC2 co-regulated genes, were highly enriched for H2AK119ub and H3K27me3 (compared to cluster B and C), whereas the active marks H3K27ac and H3K4me3 were lower compared to cluster C (Fig. 6C). Similarly, ChIP-seq data from primary AML patient samples [34] showed highest enrichment of KDM2B, H2AK119ub and H3K27me3 on cluster A genes in five individual AML samples (Supplementary Fig. S4A). In line with steady-state high H3K27me3 levels on late 5-Ph-IAA-induced genes (Fig. 5A and B) we observed that combined 5-Ph-IAA/GSK343 treatment resulted in a significantly higher fold change on late (cluster 2) versus early (cluster 1) upregulated genes (Fig. 6D), suggesting that PRC2 plays a more dominant role on late-upregulated genes.

Figure 6.

PRC1.1 can exert gene-repressive functions independent of PRC2. (A) MA-plots of RNA-seq in KDM2B-mAID K562 cells treated for 24 h with DMSO, 1 µM 5-Ph-IAA, 5 µM GSK343, or 1 µM 5-Ph-IAA and 5 µM GSK343 (n = 3). DEGs are highlighted in red (log₂FC ←0.585, > 0.585, padj < 0.05). (B) Unsupervised clustering analysis of RNA-seq normalized count data from KDM2B-mAID K562 cells treated with 1 µM 5-Ph-IAA, or 5 µM GSK343, or both, leading to the identification of clusters A, B, and C. (C) Average ChIP-seq profile of genes in clusters A, B, and C in untreated BCOR-mAID;BCORL1 KO cells. (D) Bar chart showing fold change of early and late upregulated genes (Fig. 5A) upon combined 5-Ph-IAA/GSK343 treatment over DMSO controls. Mann–Whitney test was used for statistical analysis. (E) Western blot analysis of K562 KDM2B-mAID cells treated with DMSO, 5-Ph-IAA, GSK343, or both. (F) Western blot analysis of K562 KDM2B-mAID cells with a double knockout for JARID2 and AEBP2. Asterisks indicate non-specific bands. (G) Western blot analysis of K562 KDM2B-mAID cells with a SUZ12 knockout. (H) Quantitative PCR analysis on PRC1.1-mediated genes in K562 KDM2B-mAID and K562 KDM2B-mAID JARID2/AEBP2 DKO cells (n = 3). (I) Log₂ fold change (FC) plot of DEGs detected in K562 KDM2B-mAID and K562 KDM2B-mAID JARID2/AEBP2 DKO cells. WT and DKO indicate DEGs that are uniquely identified in the respective group. Both indicates DEGs that are shared between both groups. (J) Quantitative PCR analysis on PRC1.1-mediated genes in K562 KDM2B-mAID and K562 KDM2B-mAID SUZ12 KO cells (n = 3).

The strong additive effect of EZH2 inhibition on top of KDM2B-mAID degradation-induced gene induction in cluster A genes suggested that PRC1.1 and PRC2 likely independently co-control target genes, not in line with a linear PRC1.1-PRC2.2-canonical PRC1 gene repressive model. Western blot analysis showed that, in contrast to 5-Ph-IAA, GSK343 treatment led to strongly reduced H3K27me3 levels after 24 h, which was not significantly further reduced by combined 5-PH-IAA and GSK343 treatment (Fig. 6E). This suggests that KDM2B-mAID degradation induced-transcriptional induction is PRC2-independent, although the lack of H3K27me3 reduction upon KDM2B-mAID depletion could also be explained by the reported weak PRC2.2-dependent catalysis of H3K27me3 downstream of PRC1.1 [61].

To directly investigate whether PRC1.1-mediated gene repression requires PRC2.2, we knocked out both JARID2 and AEBP2 in KDM2B-mAID cells. AEBP2 sgRNAs were designed to target both the short and long form of AEBP2, although expression of the short isoform of AEBP2 is restricted to early embryogenesis [62]. Western blot analysis of independent single cell-derived clones showed that both JARID2 and AEBP2 were not expressed in gene-edited clones, whereas EZH2 and H3K27me3 were not affected and H2AK119ub was slightly reduced (Fig. 6F). We attribute the latter to loss of PRC2.2-mediated canonical PRC1 recruitment. Importantly, KDM2B-mAID degradation-induced gene activation of a selected set of PRC1.1 target genes (all cluster A genes) was not impaired in JARID2/AEBP2 double knock out (DKO) cells and similar observations were made in single JARID2 and AEBP2 knock out KDM2B-mAID cells (Fig. 6H and Supplementary Fig. S4B and C). RNA-seq analysis showed that DEGs identified in both wild type and JARID2/AEBP2 DKO cells showed a strongly correlating fold change (Fig. 6I), suggesting that PRC1.1-mediated gene repression does not strictly require the presence of PRC2.2.

To investigate whether PRC1.1 can also maintain target-gene repression in the complete absence of PRC2, we knocked out SUZ12 in our KDM2B-mAID model. SUZ12 is part of the core PRC2 complex that is shared between PRC2.1 and PRC2.2, and therefore its depletion leads to loss of both PRC2.1 and PRC2.2. Multiple single-cell-derived clones were isolated, and SUZ12 knockout was confirmed using western blot (Fig. 6G). SUZ12 knockout clones displayed reduced expression of EZH2 and a near-complete loss of H3K27me3, in line with earlier observations [63, 64], whereas H2AK119ub was not affected. Next, we tested the effect of KDM2B-mAID degradation in all clones and found that even in the complete absence of PRC2.1 and PRC2.2, PRC1.1 loss-of-function still induced gene expression, albeit at slightly lower levels (Fig. 6J). Taken together, these data explicitly show that PRC1.1 and PRC2 co-repress target genes, but that PRC1.1 can also repress these genes in a PRC2-independent manner.

PRC1.1 and PRC2 cooperatively prevent differentiation of leukemic cells

Since our data suggested collaborative control of PRC1.1 and PRC2 over target genes, we investigated how loss of PRC1.1 complex function combined with PRC2 inhibition would impact on leukemic cell viability. Whereas 5-Ph-IAA treatment alone did not reduce cell proliferation in vitro of our K562 BCOR-mAID;BCORL1 KO and K562 KDM2B-mAID models, combined 5-Ph-IAA and GSK343 treatment strongly decreased cell counts (Fig. 7A). A comparable phenotype was observed when cells were treated with 5-Ph-IAA and the EZH1/2 inhibitor UNC1999, where 5-Ph-IAA/UNC1999 co-treatment significantly impaired cell proliferation compared to DMSO controls (Supplementary Fig. S5A). Annexin V staining showed that apoptosis levels were not increased upon 5-Ph-IAA/GSK343 or 5-Ph-IAA/UNC1999 co-treatment (Fig. 7B and Supplementary Fig. S5B), suggesting that reduced proliferation may be a consequence of cell cycle inhibition or cellular differentiation. It has been shown that K562 leukemic cells have the capacity to differentiate toward the megakaryocytic lineage [65, 66]. To investigate differentiation of cells upon 5-Ph-IAA/GSK343 treatment we evaluated the expression of CD61 and CD41 as markers of the megakaryocytic lineage. Both K562 KDM2B-mAID and K562 BCOR-mAID;BCORL1 KO cells showed a clear induction of CD61 expression upon combined 5-Ph-IAA/GSK343 treatment (Fig. 7C and D), which became more apparent after 6 days of treatment. Similarly, CD41 was also induced in both cell models upon treatment with 5-Ph-IAA and GSK343 (Fig. 7E). Increased differentiation was also observed when K562 KDM2B-mAID cells were treated with 5-Ph-IAA and UNC1999 (Supplementary Fig. S5C and D). Finally, we observed a striking change in cell morphology, where megakaryocytic cells were observed upon combined 5-Ph-IAA and GSK343 treatment, both in K562 BCOR-mAID;BCORL1 KO and KDM2B-mAID cells (Fig. 7F). To validate our findings in an independent leukemic cell model we generated endogenous BCOR-mAID knock-in THP-1 cells. Single cell-derived THP-1 BCOR-mAID clones expressing OsTIR1F74G showed no wild type BCOR expression and strong expression of BCOR-mAID (Fig. 7G). 5-Ph-IAA treatment for 2 h showed effective degradation of BCOR-mAID, loss of BCOR-mAID precipitation, and reduced co-precipitation of PRC1.1 subunits (Fig. 7H and I). Treatment of THP-1 BCOR-mAID cells with 5-Ph-IAA led to a significant growth disadvantage compared to DMSO controls that was further enhanced by co-treatment with 5-Ph-IAA and GSK343 (Fig. 7J). Morphological analysis after showed that combined 5-Ph-IAA and GSK343 treatment led to a loss of smaller dark blast-like cells and accumulation of monocytic lineage-committed cells, which was not apparent in single treatments (Fig. 7K). In line with this notion, combined 5-Ph-IAA and GSK343 treatment let to a strong induction of CD32 and CD35 expression (Fig. 7L). Taken together, these data underline a pivotal role for PRC1.1 and PRC2 in independently controlling repression of target genes, thereby maintaining a differentiation block in leukemic cells. These data highlight that combined targeting of these complexes may provide a novel therapeutic strategy in AML.

Figure 7.

PRC1.1 and PRC2 cooperatively prevent K562 and THP-1 leukemic cell differentiation. (A) Viable cell counts of BCOR-mAID;BCORL1 KO and KDM2B-mAID K562 cells treated for 3 and 6 days with 0, 0.04, 0.2, 1, or 5 µM EZH2 GSK343 and co-treated with DMSO or 1 µM 5-Ph-IAA (n = 3). (B) Annexin V-APC positive percentage, 6 days after treatment with GSK343 EZH2 inhibitor (0, 1, and 5 µM) co-treated with DMSO or 1 µM 5-Ph-IAA (n = 3). (C) FACS analysis of KDM2B-mAID cells treated for 6 days with DMSO, 1 µM 5-Ph-IAA, 5 µM GSK343, or 1 µM 5-Ph-IAA and 5 µM GSK343 and stained with anti-CD61-APC. Mean fluorescence of CD61-APC (D) and CD41 (E) positive cells of KDM2B-mAID and BCOR-mAID;BCORL1 KO cells treated for 6 days with DMSO, 1 µM 5-Ph-IAA, 5 µM GSK343, or 1 µM 5-Ph-IAA and 5 µM GSK343 (n = 3). (F) Representative images of May-Grünwald-Giemsa staining of KDM2B-mAID and BCOR-mAID;BCORL1 KO cells after 6 days treatment with DMSO or 1 µM 5-Ph-IAA and 5 µM GSK343. Arrowheads indicate cells with megakaryocytic differentiation characteristics. Scale bar represents 50 micrometers. (G) Quantitative PCR showing wild type BCOR and BCOR-mAID expression in individual THP-1 clones (error bars represent technical triplicates). (H) Western blot showing of THP-1 BCOR-mAID cells treated for 2 h with DMSO or 5-Ph-IAA (asterisk marks non-specific band). (I) Western blot analysis of Spot-Trap-mediated pull outs on nuclear extracts from THP-1 BCOR-mAID cells treated for 2 h with DMSO or 5-Ph-IAA, showing input (I), NB and B fractions. (J) Cumulative cell growth of THP-1 BCOR-mAID cells treated with DMSO, 1 µM 5-Ph-IAA, 5 µM GSK343, or 1 µM 5-Ph-IAA and 5 µM GSK343 (n = 3). (K) Representative images of May-Grünwald-Giemsa staining of THP-1 BCOR-mAID cells after 10 days treatment with DMSO, 1 µM 5-Ph-IAA, 5 µM GSK343, or both. Scale bar represents 50 micrometers. (L) FACS analysis of THP-1 BCOR-mAID cells after 11 days treatment with DMSO, 1 µM 5-Ph-IAA, 5 µM GSK343, or both and staining cells with CD32 and CD35 antibodies (n = 3). For statistical analyses a two-way analysis of variance analysis was performed, *padj < 0.05, **padj < 0.01, ***padj < 0.001.

Discussion

This work investigates PRC1.1 dependency in leukemic cells. Using degron models for BCOR and KDM2B, we showed that BCOR is a bridging factor in PRC1.1 tethering together the catalytic and chromatin-binding moiety of the complex, and its depletion induced a rapid localized loss of H2AK119ub and immediate transcriptional changes. Both BCOR-mAID or KDM2B-mAID degradation mostly led to gene activation and early- and late-upregulated genes showed distinct baseline epigenetic characteristics. Finally, we found that PRC1.1 and PRC2 co-repress shared target genes via independent pathways and that combined targeting of PRC1.1 and PRC2 resulted in differentiation of leukemic cells.

Our degron models showed that already 2 h after BCOR-mAID depletion the PRC1.1 complex fell apart in a chromatin-associated KDM2B/SKP1 heterodimer and a catalytic core (including amongst others PCGF1 and RING1B). Integration of USP7 and TRIM27 was co-dependent on both the chromatin- and catalytic moieties of PRC1.1, stressing a continuous need for BCOR to stabilize the PRC1.1 complex, which is in line with prior observations in constitutive BCOR knockout models [59]. Leukemic cells display a dependency on PRC1.1 function as previously observed [33, 34], in line with murine Ring1a-/-Ring1bfl/fl models, which lack all PRC1 complexes, interfering with MLL-AF9-dependent leukemogenesis [67]. Paradoxically, BCOR loss-of-function mutations in healthy hematopoietic stem and progenitor cells (HSPCs) predispose to the development of hematopoietic malignancies. Deletion of BCOR exons in murine HSPCs led to lymphoid leukemia development but also myeloid skewing, which was associated with increased HOXA9 and CEBPA expression [68, 69]. Other murine models confirmed these findings, showing that combined BCOR knockout and KRAS(G12D) overexpression led to a transplantable leukemia [70]. We assume that the structural role of BCOR in PRC1.1 is the same in healthy HSPCs and leukemic cells and therefore hypothesize that phenotypical differences upon PRC1.1 loss are due to differences in the pre-existing epigenetic and transcriptomic state at distinct target genes in these cell types. Our previous work showed a robust overlap of PRC1.1 target genes in primary AML cells and mobilized peripheral blood stem cells but also highlighted many specific target genes [34]. A more elaborate study will be required to explore potential differences between PRC1.1 functionality and/or chromatin targeting between human primary HSPCs versus primary AML samples.

Depletion of either BCOR-mAID or KDM2B-mAID led to immediate changes in the epigenome and transcriptome. BCOR-mAID degradation induces a strong localized loss of H2AK119ub at BCOR target genes. H2AK119ub at sites not bound by BCOR was not affected, suggesting that other (variant) PRC1 complexes are required for H2AK119ub deposition at these loci. Indeed, shRNA-mediated knockdown of various PCGF proteins resulted in a reduction of H2AK119ub levels, with the strongest effect upon PCGF4 knockdown, suggesting that the canonical PRC1.4 complex likely is responsible for the majority of deposited H2AK119ub in these cells. Already 2 h after starting 5-Ph-IAA treatment we observed DEGs, stressing that continuous PRC1.1 chromatin tethering is required for PRC1.1-mediated gene regulation. Most genes were upregulated upon PRC1.1 loss, in line with a predominant role in transcription repression. Nevertheless, also a small cluster with downregulated genes was observed (cluster three), mostly containing chromatin-related genes (i.e. histones), of which many are located to a histone gene cluster on chromosome six. These replication-dependent genes typically localize to the phase-separated histone locus body [71]. Overall, these genes showed only little BCOR binding and associated H2AK119ub and were devoid of PRC2. It remains to be understood how the activity of these genes is coupled to PRC1.1 activity. Among the upregulated genes we found early- and late-upregulated genes, where early-upregulated (cluster one) genes were derepressed after 2 h and late-upregulated (cluster two) genes were transcriptionally induced between 8 and 24 h after PRC1.1 loss. Baseline epigenetic states were distinct in these clusters, where late-upregulated genes were more enriched for repressive marks like H2AK119ub and H3K27me3 and showed less H3K27ac. We hypothesize that cluster one genes are more transcriptionally poised compared to cluster two genes that are more heavily repressed. Late-upregulated genes that were highly enriched for H2AK119ub and H3K27me3 also showed strong BCOR and EZH2 occupancy suggestive of PRC1.1 and PRC2 collaborative control at these gene loci. Indeed, late-upregulated genes showed a stronger transcriptional response to combined depletion of KDM2B and PRC2 inhibition.

This prompted the question to what extent PRC1.1-mediated repression is truly dependent on PRC2 activity. It was previously shown that variant PRC1 complexes are main contributors to H2AK119ub deposition and that PRC1.1 activity can also initiate Polycomb repression through recruitment of JARID2/AEBP2-containing PRC2.2 in an H2AK119ub-dependent manner [8, 28–32, 72]. These findings suggested a linear model where PRC1.1-dependent H2AK119ub deposition leads to PRC2.2 recruitment, followed by canonical PRC1 recruitment. However, the strong additive gene activation as a consequence of EZH2 inhibition on top of KDM2B-mAID degradation was not in line with such a model, where one would expect that targeting either PRC1.1 or PRC2 would have comparable consequences for target gene expression. To rule out additive effects due to non-complete EZH2 repression we generated K562 KDM2B-mAID JARID2/AEBP2 double knockout cells that are completely devoid of PRC2.2. Strikingly, these cells showed equal transcriptional induction upon KDM2B-mAID degradation compared to PRC2.2 proficient cells stressing that PRC1.1 can repress target genes independent of PRC2.2. Importantly, H3K27me3 levels were not affected in these cells, most likely as a consequence of remaining PRC2.1 and in line with the fact that PRC2.1 and PRC2.2 cooperatively maintain H3K27me3 levels [11, 73]. Furthermore, it was shown that PRC2.1 displays higher methyltransferase activity compared to PRC2.2 [61]. To investigate a potential compensatory effect where PRC2.1 rescues loss of PRC2.2, we generated K562 KDM2B-mAID SUZ12 KO cells. These cells showed a near complete loss of H3K27me3 marking and destabilization of EZH2, in line with previous observations in murine Suz12 null ES cells [63, 64]. Importantly, K562 KDM2B-mAID SUZ12 KO cells still displayed transcriptional induction of PRC1.1 target genes upon KDM2B-mAID degradation, albeit less compared to wild type cells. Based on these data we conclude that PRC1.1 can also repress target genes independent of PRC2. Previous work by Zepeda-Martinez et al. showed that variant PRC1 complexes and canonical PRC1 and PRC2 complexes can independently control H2AK119ub levels at target genes in ES cells [74]. Here, the authors used both knockout and degron models for RYBP, a common factor in all variant PRC1 complexes. In these experiments YAF2, a mutually exclusive protein with RYBP in variant PRC1, could potentially rescue the loss of RYBP. Our models specifically address PRC2 and PRC1.1 co-dependency and we speculate that PRC1.1 functionality does not only encompass H2AK119ub deposition but also involves interaction with transcription regulatory complexes. In previous studies we found that PRC1.1 strongly interacts with the various transcription regulatory complexes including the super elongation complex (SEC), Mediator and polymerase associated factor complex [33]. Interestingly, our own data showed that BCOR depletion also led to a reduction of MLLT1/ENL and CBX8 co-precipitation with KDM2B-GFP. Previous work has shown that CBX8 interacts with PRC1.1 in a RING1B-dependent manner in DLBCL cells [75], and we propose that a similar interaction takes place in AML cells. The reduction of MLLT1/ENL co-precipitation upon BCOR-mAID depletion was likely due to loss of BCOR-associated MLLT1/ENL. BCOR was previously shown to interact with the ANC1 homology domain (AHD) of MLLT3/AF9 [76]. Schmidt and colleagues showed that the MLLT3/AF9 AHD domain can interact with an LXVXIXLXXL motif present in both BCOR, CBX8, and DOT1L, and interacts with these proteins in a mutually exclusive manner [60]. The MLLT1/ENL AHD domain is 80% identical to that of MLLT3/AF9, showed a stronger association with BCOR, and also interacts with CBX8, in line with reports indicating that both CBX8 and DOT1L can interact with MLL-ENL [60, 77]. ENL is a subunit of SEC and previously we found that various SEC subunits coprecipitate with KDM2B, whereas KDM2BΔCxxC, a deletion mutant that does not bind chromatin, only interacts with ENL [33]. Based on these data we hypothesize that PRC1.1 and ENL interact in a BCOR-dependent manner and that ENL may form a molecular bridge toward SEC. This is in line with previous observations that non-canonical PRC1 complexes may control the binding of preinitiation complex components to Polycomb target genes [78].

Finally, our data showed that combined PRC1.1 depletion and PRC2 inhibition leads to a strong induction of leukemic cell differentiation, suggesting that both complexes collaborate to silence target genes and provide a differentiation block, reminiscent of previously observed increased ES cell differentiation in Rybp/Eed KO models [74]. Based on our data we speculate that co-targeting of both PRC1.1 and PRC2 is required to obstruct their respective independent gene repressive axes to induce differentiation of leukemic cells. Taken together, our data underlines an essential role for BCOR in AML cells to preserve PRC1.1 complex integrity and maintain permanent transcriptional control of target genes independent of PRC2. Simultaneous targeting of PRC1.1 and PRC2 drives leukemic cell differentiation and is an interesting novel strategy to treat AML patients.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work is supported by grants from the Dutch Cancer Foundation (KWF) for V.v.d.B., J.J.S., F.M., and R.K. (KWF 12725), V.v.d.B., J.J.S., A.D.M., and S.G.A. (KWF 15286), and J.J.S. and S.M.H. (KWF 11013). Funding to pay the Open Access publication charges for this article was provided by KWF Kankerbestrijding (Dutch Cancer Society).

Data availability

All sequencing data are available in Gene Expression Omnibus under GEO numbers GSE286991 (RNA-seq) and GSE286993 (ChIP-seq). Proteomics data have been deposited in the Proteomics Identification Database under PRIDE number PXD065157. The accession number for previously published AML ChIP-seq data used for analysis in this study is GSE54580 [34], which were realigned to the hg38 reference genome.