Variant PRC1 complex-dependent H2A ubiquitylation drives PRC2 recruitment and polycomb domain formation

Abstract

Chromatin modifying activities inherent to polycomb repressive complexes PRC1 and PRC2 play an essential role in gene regulation, cellular differentiation, and development. However, the mechanisms by which these complexes recognize their target sites and function together to form repressive chromatin domains remain poorly understood. Recruitment of PRC1 to target sites has been proposed to occur through a hierarchical process, dependent on prior nucleation of PRC2 and placement of H3K27me3. Here, using a de novo targeting assay in mouse embryonic stem cells we unexpectedly discover that PRC1-dependent H2AK119ub1 leads to recruitment of PRC2 and H3K27me3 to effectively initiate a polycomb domain. This activity is restricted to variant PRC1 complexes, and genetic ablation experiments reveal that targeting of the variant PCGF1/PRC1 complex by KDM2B to CpG islands is required for normal polycomb domain formation and mouse development. These observations provide a surprising PRC1-dependent logic for PRC2 occupancy at target sites in vivo.

Graphical abstract

Figure 1

PCGF 1, 3, and 5 Variant PRC1 Complexes Catalyze H2AK119ub1 and Create A Polycomb Domain Containing PRC2 and H3K27me3 (A) A schematic illustrating the core components of canonical and variant PRC1 complexes. (B) The TetO array at its integration site on mouse chromosome 8. (C) Targeting of factors to the TetO via the TetR DNA-binding domain. Numbers represent qPCR primer positions (kb) with respect to TetO array. (D) ChIP analysis for fusion protein occupancy (TetR), RING1B, H2AK119ub1, and histone H3 across the TetO containing locus. Fusion protein identity is indicated above each panel. (E) As in (D) ChIP analysis for PRC2 components and H3K27me3. All ChIP experiments were performed at least in biological duplicate with error bars showing standard error of the mean (SEM). See also Figure S1.

Figure 2

Hierarchical Recruitment of Canonical PRC1 Fails to Result in H2AK119ub1 (A) ChIP analysis for fusion protein occupancy (TetR), PRC2 components, and H3K27me3 across the TetO-containing locus in lines expressing TetR alone and a TetR-EED fusion. (B) ChIP analysis for PRC1 components and H2AK119ub1 performed as described in (A). All ChIP experiments in (A) and (B) were performed at least in biological duplicate with error bars showing SEM.

Figure 3

H2AK119ub1 Is Required for PRC1-Dependent Recruitment of PRC2 and H3K27me3 (A) Mass spectrometry analysis of purified TetR-PCGF4 and TetR-RPCD (minimal RING1B/PCGF4 catalytic domain) proteins. The mascot score and percentage coverage is indicated for polycomb group proteins in each sample. (B) ChIP analysis for PRC1, PRC2 and their respective modifications in cell lines expressing TetR, TetR-RPCD, and TetR-RPCDmut. All ChIP experiments were performed at least in biological duplicate with error bars showing SEM.

Figure 4

PRC1 Has a Genome-wide Role in PRC2 Recruitment and Polycomb Domain Formation at Target Sites in Mouse ESCs (A) A schematic of the Ring1a^−/−Ring1b^fl/fl system in which addition of tamoxifen (OHT) leads to deletion of RING1A/B. (B) ChIP analysis for RING1B and H2AK119ub1 at polycomb target sites before (−OHT) and after 48 hr (+OHT) tamoxifen treatment. ChIP experiments were performed at least in biological duplicate with error bars showing SEM. (C) Western blot analysis of polycomb factors following deletion of RING1A/B. (D) Snapshots of ChIP-seq traces for RING1B, SUZ12, EZH2, and H3K27me3 in the Ring1a^−/−Ring1b^fl/fl cells prior to (−OHT) and following 48 hr +OHT treatment at the Grik3 and Tmem163 genes. Bio-CAP indicates nonmethylated DNA and CpG islands (CGI) are shown as green bars. (E) Heat map analysis of SUZ12 peaks (n = 3,819), showing ChIP-seq data for RING1B, SUZ12, EZH2, and H3K27me3 at a 10 kb region centered over the SUZ12 peaks prior to (−OHT) and after 48 hr +OHT treatment. Bio-CAP is included to indicate nonmethylated DNA at these sites. (F) A box and whisker plot showing log2 fold changes in normalized read counts comparing the ChIP-seq signal at SUZ12 sites −OHT and after 48 hr +OHT treatment. (G) A Venn diagram showing the overlap of RING1B (light blue) and SUZ12 (light green) peaks including a further segregation of SUZ12-bound regions that show a greater than 1.5-fold change in SUZ12 occupancy (ΔSUZ12, dark green) after 48 hr tamoxifen treatment (+OHT). (H) A scatter plot comparing the fold change of RING1B and SUZ12 at SUZ12 peaks −OHT and after 48 hr +OHT treatment. See also Figures S2 and S3.

Figure 5

KDM2B Targets the PCGF1/PRC1 Variant Complex Leading to PRC2 Recruitment and Formation of a Polycomb Domain and a Novel System to Ablate KDM2B-Mediated Targeting of PCGF1/PRC1 to Chromatin (A) ChIP analysis for TetR, RING1B, PCGF1, H2AK119ub1, EZH2, and H3K27me3 across the TetO containing locus in lines expressing TetR, TetR-KDM2B, and TetR-KDM2A. ChIP experiments were performed at least in biological duplicate with error bars showing SEM. (B) Western blot analysis of PRC1 and PRC2 protein levels after knockdown of PCGF1 in the TetR-KDM2B fusion line. (C) ChIP analysis for TetR, PCGF1, RING1B, and EZH2 following PCGF1 knockdown in the TetR-KDM2B fusion line. ChIP experiments were performed at least in biological duplicate with error bars showing SEM. (D) A schematic illustrating tamoxifen (OHT)-dependent removal of the ZF-CxxC domain from both KDM2B isoforms. (E) Western blot analysis of the Kdm2b^fl/fl cell line following an OHT treatment time course. (^∗) is a nonspecific cross reactive band. (F) A heat map covering a 10 kb region centered over CpG islands showing KDM2B ChIP-seq in the Kdm2b^fl/fl cells prior to (−) and after 72 hr (+) OHT treatment. (G) ChIP analysis of KDM2B and PCGF1 at gene-associated CpG islands and gene body regions. ChIP experiments were performed in biological triplicate with error bars showing SEM. See also Figures S4 and S5.

Figure 6

Failure to Target the PCGF1/PRC1 Complex Leads to a Loss of H2AK119ub1, PRC2, and H3K27me3 (A) Snapshots of ChIP-seq traces for KDM2B, RING1B, and SUZ12 in the Kdm2b^fl/fl cells prior to (−OHT) and after 72 hr (+OHT) of tamoxifen treatment at the Epha7 and Klrg2 genes. (B) Heat map analysis of RING1B peaks (n = 3,488), showing ChIP-seq data for KDM2B, RING1B, and SUZ12 covering a 10 kb region centered over the RING1B peak −OHT and after 72 hr +OHT. (C) Log2 fold changes in normalized read counts comparing the ChIP-seq and RNA-seq signal −OHT and after 72 hr +OHT. (D) A Venn diagram showing all RING1B peaks (light blue), intersected with RING1B or SUZ12 peaks that have a greater than 1.5-fold reduction in RING1B/SUZ12 occupancy after 72 hr +OHT treatment (ΔRING1B [dark blue] and ΔSUZ12 [green]). (E) A box and whisker plot indicating the log2 fold change in RING1B and SUZ12 occupancy at sites that have RING1B changes of greater than 1.5-fold (ΔRING1B) or less than 1.5-fold (No Change) following 72 hr +OHT treatment. The significance of the changes in SUZ12 occupancy at these sites is indicated above the plot. (F) A scatter plot comparing the log2 fold change of RING1B and SUZ12 at RING1B sites in the Kdm2b^fl/fl cells −OHT and after 72 hr +OHT. (G) A box and whisker plot indicating log2 fold change in gene expression at sites described in (E). (H) A scatter plot comparing the log2 fold change of gene expression to the fold change in RING1B occupancy at sites that show RING1B alterations. (I) ChIP analysis for polycomb factors and modifications at regions showing loss of RING1B, no significant loss of RING1B, and a nontarget site (NTS). All ChIP experiments were performed in biological triplicate with error bars showing SEM. (J) Gene expression analysis for the target genes analyzed by ChIP in (I). RT-PCR was performed in biological triplicate. Error bars show SEM.

Figure 7

Failure to Target PCGF1/PRC1 Results in Homeotic Phenotypes and Embryonic Lethality (A) Kdm2b^wt/ΔCxxC mice were mated with wild-type mice and at 11 day postnatal (dpn) offspring were genotyped. Results are summarized in a table. (B) A schematic summarizing the homeotic phenotypes observed in newborn Kdm2b^wt/ΔCxxC mice (n = 10) with the normal wild-type vertebrae organization shown for comparison. The numbers in parentheses indicate the frequency of individual transformations. (C) Examples of the vertebral column in wild-type and Kdm2b^wt/ΔCxxC mice showing homeotic phenotypes. Top: 11 dpn heterozygotes that have additional ossification at the C7 position indicating partial posteriorization (white triangles). Center: comparison of newborn wild-type and newborn Kdm2b^wt/ΔCxxC mice missing dorsal processes or that have the position of the process repositioned to the anterior (black arrowhead). Bottom: 6th lumbar vertebral column transformed to sacrum in Kdm2b^wt/ΔCxxC mice (red arrowhead). (D) Kdm2b homozygous null embryos recovered at 9 dpc exhibited severe developmental delay (n = 5) or in some cases only extraembryonic development (n = 2) (representative examples are shown in center and right, respectively). A heterozygote sibling is shown to indicate expected development at this stage (left).

Figure S1

Variant PRC1 Complex-Dependent PRC2 Targeting Occurs on an Inert TetO Array Template and at a Single Natural TetO Site in Mouse ESCs, Related to Figure 1 (A) Bisulfite sequencing reveals that CpG dinucleotides flanking the engineered TetO array in the Human BAC DNA remain methylated, as they are in human tissue, when stably inserted into the mouse genome. This indicates that normal DNA methylation features are recapitulated on the integrated DNA sequence. (B) ChIP-qPCR analysis of the human BAC/TetO array stably integrated into mouse ESCs. ChIP was performed with antibodies specific for permissive chromatin marks and features of active transcription (left: H3K4me3, H3K9ac, H3K36me3, and RNAPII) and repressive chromatin features (right: H3K27me3, H2AK119ub1, H4K20me3, and H3K9me3). For each antibody, qPCR enrichment at a known target site is shown as a positive control. For gene-associated chromatin modifications this includes promoter (prom) and body amplicons for the indicated genes, while H3K9me3 and H4K20me3 were analyzed at sites in the repetitive major satellite regions. Together these observations indicate that the human BAC DNA, when inserted into mouse ESCs, retains its inert features and remains suitable for studying polycomb domain formation in tethering assays. (C) A position on mouse chromosome 1, distant from surrounding genes, contains a single site that has a high degree of homology to a consensus bacterial TetO-binding site. This provides a “natural” site in the genome at which to analyze TetR fusion protein binding and function. (D) ChIP-qPCR analysis of TetR, TetR-PCGF1 (variant PRC1), and TetR-PCGF4 (canonical PRC1) binding at the single natural TetO and flanking regions (Top). In each case, occupancy of PRC1 leads to RING1B recruitment (second panel). However, as observed at the engineered TetO array EZH2, SUZ12, and H3K27me3 are only recruited to the endogenous TetO site when a variant PRC1 complex occupies the site (bottom three panels). Together these observations demonstrate that variant PRC1 complex-dependent PRC2 recruitment and de novo polycomb domain formation is observed at a “natural” sequence in the mouse genome, and is not the result of copy number or site specific features inherent to the engineered TetO array. All ChIP experiments in (B and D) were performed in biological duplicate with error bars showing SEM.

Figure S2

Conditional Deletion of RING1A/B Results in Depletion of H2AK119ub1, Loss of PRC2 Occupancy, and H3K27me3 at Target Genes Independently of the Magnitude of Associated Gene Expression Change, Related to Figure 4 (A) Snapshots of ChIP-seq traces for RING1B, SUZ12, EZH2 and H3K27me3 in the Ring1a^−/−Ring1b^fl/fl cells prior to (−OHT) and following 48 hr (+OHT) of tamoxifen treatment. Three representative genes are depicted illustrating the reduction in RING1B, SUZ12, EZH2, and H3K27me3 following removal of the RING1A/B. A Bio-CAP sequencing trace is shown to indicate the location of nonmethylated DNA and a CpG Island (CGI) prediction annotation track is show as green bars under the traces. This complements examples already shown in main Figure 4. (B) ChIP-qPCR analysis for PRC2 components SUZ12, EZH2, and H3K27me3 at a series of sites showing loss of these factors in the ChIP-seq analysis in main Figure 4 after tamoxifen treatment. All ChIP experiments in (C-D) were performed in biological triplicate with error bars showing SEM. (C) Gene expression analysis for the polycomb target genes analyzed by ChIP in (B). There is no correlation between gene expression change and scale of polycomb group protein loss, suggesting that altered gene expression is not driving PRC2 loss. RT-PCR was performed in biological triplicate and is normalized to Hprt1 expression. Error bars show SEM.

Figure S3

PRC1 Has a Genome-wide-Role in PRC2 Recruitment and Polycomb Domain Formation at Target Sites in Mouse ESCs, Related to Figure 4 (A) A Venn diagram showing the overlap of RING1B (light blue) and EZH2 (light pink) peaks including a further segregation of EZH2-bound regions that show a greater than 1.5-fold change in EZH2 occupancy (ΔEZH2, dark pink) after RING1A/B deletion. The majority of EZH2-bound locations show changes in EZH2 signal following removal of RING1A/B, and most of the changes are restricted to sites associated with RING1B peaks. (B) A metaplot of RING1B and SUZ12 ChIP-seq data at RING1B peaks that do not overlap with SUZ12 sites identified by peak calling. This indicates that these regions exhibit SUZ12 enrichment but are likely below the enrichment level required for peak detection. (C) A metaplot of RING1B and SUZ12 ChIP-seq data at SUZ12 peaks that do not overlap with RING1B sites identified by peak calling. This indicates that these regions exhibit RING1B enrichment but are likely below the enrichment level required for peak detection. (D) A metaplot of SUZ12 ChIP-seq data in the Ring1a^−/−Ring1b^fl/flb cells at SUZ12 peaks exhibiting a less than 1.5-fold reduction in SUZ12 signal following tamoxifen treatment. Importantly these sites still show reduction in SUZ12 signal suggesting that loss of RING1A/B affects PRC2 occupancy at most sites. (E) A box and whisker plot indicating the Log2 fold change in polycomb factors at SUZ12-bound sites with and without REST following loss of RING1A/B. This indicates that loss of PRC2 occurs at REST-bound sites in the absence of PRC1. (F) Snapshots of ChIP-seq traces for RING1B, SUZ12, EZH2 and H3K27me3 in the Ring1a^−/−Ring1b^fl/fl cells prior to (−OHT) and following 48 hr (+OHT) of tamoxifen treatment at sites previously reported to rely on the Meg3 long noncoding RNA for PRC2 targeting. In all cases we observe appreciable loss of PRC2 following RING1A/B deletion indicating that Meg3-dependent targeting is not sufficient to maintain normal levels of PRC2 at these sites. (G) A box and whisker plot indicating the Log2 fold change in PRC2 factors and H3K27me3 at sites considered to be bivalent. Bivalent sites appear to have slightly larger fold changes in PRC2 occupancy following RING1A/B deletion. (H) A scatter plot comparing the fold change of RING1B and EZH2 at EZH2 peaks. This indicates a clear correlation between the magnitude in RING1B and EZH2 change suggesting these changes may be mechanistically linked.

Figure S4

Both the Short and Long Form of KDM2B Mediate Polycomb Domain Formation in a Histone Demethylase Activity-Independent Manner, Related to Figure 5 (A) KDM2B long form (LF) and short form (SF) with their domain organization indicated. Additional, TetR fusion constructs which have had domains removed (gray boxes) or mutated are indicated. (B) Western blot analysis of the TetR-KDM2B fusion cell lines indicating roughly equal protein expression. (C) ChIP-qPCR analysis for the TetR fusion protein, RING1B, H2AK119ub1, SUZ12, EZH2, and H3K27me3 across the TetO containing region in the TetR only, TetR-KDM2B LF, TetR-KDM2B SF, and TetR-KDM2B LF demethylase mutant (JmjC mutant). All three versions of KDM2B lead to efficient RING1B recruitment, H2AK119ub1, and formation of a polycomb domain containing SUZ12, EZH2, and H3K27me3. This indicates that both forms of KDM2B can form polycomb domains independent of their demethylase activity. (D) An epitope tagged version of the KDM2B-SF was stably expressed in mouse ESCs, affinity purified, and associated proteins identified by tandem mass spectrometry. This revealed that the short form of KDM2B forms the same variant PRC1 complex as the long form of the protein, consistent with its capacity to recruit RING1B and form polycomb domains in tethering assays. (E) Based on the capacity of KDM2B-SF to associate with the PCGF1/PRC1 complex (D) the domain(s) mediating this were further mapped in tethering assays. The C-terminal Fbox and LRR domains are required for RING1B recruitment whereas the PHD domain is dispensable.

Figure S5

A Model Cell System to Inducibly Disrupt Targeting of the PCGF1/PRC1 Complex, Related to Figure 5 (A) A schematic of the Kdm2b gene showing the long form (Kdm2b-LF) and short form (Kdm2b-SF) transcription start sites. The positions of LoxP sites are highlighted flanking the exon which encodes the ZF-CxxC domain. (B) PCR with primers spanning the floxed exon was performed on genomic DNA from the Kdm2b^fl/fl cells (UNT), the Kdm2b^fl/fl cells treated for 72 hr with tamoxifen (72 hr), and wild-type cells (WT). 72hrs of tamoxifen treatment leads to a clear deletion of the floxed exon. (C) Western blot analysis indicates that loss of the KDM2B ZF-CxxC domain does not lead to destabilization of the PCGF1 and RING1B components of the KDM2B variant PRC1 complex or upregulation of its paralogue KDM2A. Furthermore, PRC2 remains present as indicated by normal levels of SUZ12. (D) Affinity purification of full-length epitope tagged wild-type (WT) KDM2B and KDM2BΔCxxC followed by tandem mass spectrometry-based analysis of associated proteins. The mascot score and percentage coverage is indicated for the KDM2B/PRC1 complex components. Importantly, removal of the ZF-CxxC exon generates a product that still associates with the PCGF1/PRC1 variant complex but lacks its capacity to bind nonmethylated DNA.