Lysine acetylation of histone proteins usually correlates with active transcription, whereas lysine methylation can mark either gene activation or repression depending on positions of the lysine residue and methylation multiplicity (86, 87). H3K27 acetylation is mediated by p300/CBP histone acetyltransferase (HAT), and H3K27ac is preferentially enriched at both promoters and active enhancers (88, 89) ( Figure 2A ). Studies in Drosophila embryos and mESCs have shown that H3K27ac and H3K27me3 are antagonistic, suggesting a direct competition between PRC2 and p300/CBP in posttranslational modification of H3K27 ( Figure 2B ) (90, 91). Furthermore, PC, a component of Drosophila PRC1, impedes H3K27 acetylation by directly inhibiting the HAT activity of CBP (92). In line with the competition model, the widespread deposition of H3K27me2 is believed to protect target loci from HAT activity, ensuring proper control of cell type-specific enhancers (29). Additionally, EZH1 or a catalytically impaired EZH2 mutant is unable to fully restore H3K27me2 or H3K27me3 levels in an EZH2 knockout background, leading to invasion of H3K27ac into the otherwise H3K27me3-marked promoters and causing defects in mESC differentiation (43).

PRC2 is mechanistically linked to pol II transcription elongation through Elongin and SPT6 ( Figure 2C ). Elongin positively regulates pol II elongation by suppressing transcriptional pausing; the heterotrimeric Elongin complex consists of the largest active subunit Elongin A and two smaller regulatory subunits, Elongin B and C (93–95). PRC2 methylates Elongin A at residue K754, and loss of this methylation results in de-repression of PRC2 target genes and defective EB formation ( Figure 2C ) (96). In addition, the accessory subunit EPOP from PRC2.1 has been proposed to connect Elongin B and C to PRC2 in mESCs, maintaining low expression of PRC2 target genes through a mechanism that is not understood ( Figure 2C ) (97, 98). Direct crosstalk also exists between PRC2 and the elongation factor SPT6: PRC2 recruitment and H3K27me3 deposition are impeded by SPT6 due to the competitive binding of SPT6 and EZH2 to SUZ12, which blocks the assembly of functional PRC2 ( Figure 2C ) (99).

RNAs provide another link between PRC2 and transcription machinery ( Figure 2D ). PRC2 mediates promiscuous binding to RNAs with a preference for G-tract sequences (100, 101). RNAs associate with dispersed sites on PRC2 surface (102, 103), antagonizing direct chromatin binding of PRC2 but linking PRC2 to chromatin physically (104, 105). In particular, PRC2 interacts with short RNAs transcribed from target genes, where pol II and H3K4me3 are observed, to repress gene expression in cis in primary T cells ( Figure 2D ) (106). Congruently, PRC2 is also found to bind nascent RNAs from active promoters devoid of H3K27me3 in mESCs ( Figure 2D ) (107). RNAs can inhibit PRC2 enzymatic activity both in vitro and in vivo, possibly by preventing nucleosomal substrate binding (108–110), which may serve as a mechanism to sustain the active state of genes while keeping PRC2 poised for chromatin surveillance (100, 107, 109). Additional data suggest PRC2 interaction with nascent transcripts can tune the transition between promoter-proximal pausing and productive elongation ( Figures 2A, D ) (111).

PRC2 has been linked to at least two types of distinct high-order chromatin structures: compacted chromatin and long-range chromatin loops or contacts. Chromatin compaction causes transcriptional inactivation by restricting promoter accessibility ( Figure 2E ). In a hierarchical recruitment model, PRC2 is targeted to chromatin decorated by H3K27me3, which recruits canonical PRC1 via the chromodomain (CD) of CBX proteins (112). This allows PHC and CBX2 subunits of PRC1 to compact target chromatin, independent of PRC1 enzymatic activity (113–116). EZH1-containing PRC2 also exhibits chromatin compaction activity, which appears to depend on histone tails but is independent of SAM (67). Unlike the EZH2-containing counterpart, EZH1-containing PRC2 displays a higher binding affinity towards nucleosomes and, correspondingly, represses in vitro transcription from chromatinized templates (67, 117).

Long-range chromatin interactions between genes and regulatory elements, like enhancers, repressors, and insulators, are widely implicated in gene regulation in the context of a three-dimensional genome, for which PRC2 and H3K27me3 are also known as important players (118–123). H3K27me3-rich genomic regions can act as gene silencers through PRC2-dependent long-range chromatin interactions (124, 125). Differential chromatin binding of PRC2 along the HoxA cluster impacts long-range interactions of HoxA in mouse embryos, which unexpectedly also promotes enhancer-promoter contacts (126). Likewise, PRC2 facilitates contact between poised enhancers and target genes during mESC differentiation (127). Hox clusters are organized into multiple connected chromatin loops, which involve PcG proteins and other architectural factors, like CTCF ( Figure 2F ) (128, 129). Relatedly, the SUZ12 subunit of PRC2 has been found to interact with CTCF at a long-range intrachromosomal loop between IGF2/H19 imprinting control region (ICR) and IGF2 promoters, maintaining monoallelic expression of IGF2 ( Figure 2F ) (130, 131). In line with these data, PRC2 directly mediates DNA bending and looping in reconstituted systems, as shown by atomic force microscopy (AFM) and molecular dynamics simulations (132, 133).

Proper assembly of core and accessory subunits is a prerequisite for PRC2 enzymatic activity and chromatin recruitment. Crystal structures of a minimal mouse EZH2–EED complex and of NURF55 (homolog of RBBP4 in Drosophila) bound to a minimal fragment of SU(Z)12 (homolog of SUZ12 in Drosophila) represent the initial attempts to tackle the puzzle of complex assembly (134, 135). The negative stain EM structure of a five-member PRC2.2, PRC2-AEBP2, further reveals a bipartite structure architecture (136). Recent X-ray crystallographic and cryo-EM studies have together built a near-complete atomic model of the core complex and started to add pieces of various accessory subunits into the picture of holo complex structures ( Figure 3 ). Overall, the PRC2 core complex consists of two structurally and functionally distinct modules, including the catalytic module and the accessory subunit-binding module ( Figure 3A ) (137): whereas the former contains the minimally active ternary complex of EZH2, EED, and the C-terminal VEFS (VRN2, EMF2, FIS2, and SU(Z)12) domain of SUZ12 (SUZ12(VEFS)) ( Figure 3B ) (138–140), the latter is composed of the N-terminal portion of SUZ12 (SUZ12(N)) and RBBP4, which together form docking sites for various accessory subunits ( Figure 3C ) (140, 141). EZH2 is folded into a series of functional domains dispersed across the catalytic module (138, 139) ( Figure 3B ). EED and SUZ12(VEFS) are bound to distinct subsets of EZH2 domains, juxtaposing with each other ( Figure 3B ) (138, 139). In the accessory subunit-binding module, functional domains of SUZ12(N) are scattered on the surface of the RBBP4 WD40 repeat, forming both intramolecular and intermolecular complexes ( Figure 3C ) (141, 142). The accessory subunit-binding module also mediates PRC2 dimerization in the context of PRC2.1, which facilitates chromatin binding of PRC2 (see below) (142). Key structural features of the assembly of these two modules are described in sections below, with the disease connection or clinical relevance highlighted.

EZH2–EED interaction and stapled peptide inhibitor of PRC2. In the catalytic module of PRC2, the EBD (EED-binding domain) of EZH2 associates with the bottom face of the EED WD40 repeat, dominating the interaction of these two proteins ( Figure 4A ) (134, 138, 139). Stapled α-helical peptide inhibitors have opened a new window for drug discovery by targeting protein-protein interactions, which has been difficult with small molecule inhibitors (143). Stapled peptides designed based on the EBD structure compete with EZH2 for EED binding in vivo, leading to disruption of PRC2 assembly ( Figure 4A ) (144). These peptides inhibit PRC2-dependent growth and induce differentiation of MLL-AF9 leukemia cells (144).

Intramolecular complex and gene activation function of EZH2 in cancer cells. In the catalytic module of PRC2, the SANT1 (SWI3, ADA2, N-CoR, and TFIIIB 1) domain and the SBD (SANT1-binding domain) of EZH2 mediate intramolecular interactions ( Figure 4B ) (138, 139). The SANT1 harbors a partially disordered acidic transactivation domain (TAD), which is normally sequestered by SBD binding (145). The TAD can be released by EZH2 phosphorylation at residue S21 by AKT kinase or at residue Y244 by JAK3 kinase ( Figure 4B ) (145). These two phosphorylation events are activated in prostate cancer and natural killer/T-cell lymphoma cells, respectively, converting EZH2 from a gene repressor to a gene activator (80, 81, 146). In line with this model, the EZH2 TAD forms stable complexes with components of the active transcription machinery, such as p300/CBP (145). Intriguingly, the EZH2 TAD interacts with the MYC oncoprotein in a PRC2-independent manner to promote the growth of leukemia cells (147). Indeed, EZH2 also physically associates with MYC or MYCN in other cellular contexts, including neuroblastoma and prostate cancer cells, to drive cancer development, making EZH2 degradation an attractive strategy for cancer treatment (147–150). In contrast, it remains to be shown whether enzymatic inhibitors of PRC2 can specifically target the ‘solo’ EZH2, as the drug-binding site may not exist in the absence of EED and SUZ12 (see 6.2 below).

C2 domain, a mediator of mutual exclusivity and PRC2 dimerization. In the accessory subunit-binding module of PRC2, SUZ12(N) contains a mobile non-canonical C2 domain, which becomes ordered only when bound to the accessory subunit PHF19 or AEBP2 ( Figure 5A ) (141, 142). The C2 domain is named based on the observation that it adopts an eight-strand β-sandwich structure, mimicking the classical C2 domain that mediates phospholipid binding ( Figure 5A ) (141). It is not unprecedented that phospholipids regulate the function of chromatin complexes, and it remains to be explored if such a mechanism also exists for PRC2 (141, 151). PHF19 and AEBP2 interact with overlapping surfaces of the C2 domain, which accounts for the mutual exclusivity in complex assembly between PRC2.1 and PRC2.2 (141, 142). Formation of a PRC2 dimer has long been implicated (67, 117, 152–155), and recent structural work reveals that the PRC2 core complex forms an intrinsic dimer via the C2 domain of SUZ12: a basic loop region on the C2 domain from one protomer is swapped to occupy the acidic central cavity of RBBP4 from the other protomer ( Figures 5A, B ) (142). This dimeric structural scaffold has a profound impact on the ability of PRC2 to bind chromatin (see below).

PHF19 binding and stabilization of the intrinsic PRC2 dimer. The RC (reversed chromo) domain of PHF19 (PHF19(RC)) mediates concurrent binding to the C2 domain of SUZ12 from one protomer and the SUZ12(N)–RBBP4 protein body from the other protomer (142) ( Figure 5B ). In this way, PHF19 is not involved in PRC2 dimerization per se; instead, it stabilizes the intrinsic dimer of PRC2 ( Figure 5B ). Structure-guided functional studies indicate that PRC2 dimerization promotes chromatin binding of MTF2 or PHF19-containing PRC2.1, possibly through an avidity effect (142). There is no evidence on whether PRC2 plays a direct architectural role in forming long-range chromatin contacts. However, it is not impossible that the dimeric structural scaffold may allow PRC2 to bring two distal chromatin regions close to each other via direct chromatin binding. In this regard, a set of specific structure-based dimer-disrupting mutations can serve as a valuable tool to test this possibility (142).

AEBP2 binding and the dimer-to-monomer transition. There are long and short isoforms of AEBP2 in mammals (156). AEBP2 used in the published structural studies corresponds to the isoform 2 (UniProtKB Q6ZN18-2), lacking the N-terminal D/E-rich, S-rich, and G-rich patches (140, 141, 157). The C2B (C2-binding) domain of AEBP2 (AEBP2(C2B)) is bound to the C2 domain of SUZ12 ( Figure 5C ) (141). A part of the AEBP2(C2B) also associates with the ZnB (zinc finger binding) helix of SUZ12 (SUZ12(ZnB)), where disease mutations are found to specifically disrupt PRC2.2 assembly ( Figure 5C ) (57, 141). Notably, AEBP2 binding induces a dimer-to-monomer transition of the PRC2 core complex by relocating the swapped C2 domain of SUZ12, suggestive of drastically different structural architectures of PRC2.1 and PRC2.2 (142).

JARID2 binding, mutual exclusivity, and oncogenic chromosomal translocation. The Zn (zinc finger) domain and ZnB helix of SUZ12 assemble into a binding platform for the TR domain of JARID2 (JARID2(TR)) ( Figure 5D ) (141). An overlapping surface may also mediate EPOP binding according to biochemical data, underlying the mutual exclusivity between JARID2 and EPOP in the assembly of PRC2.1 and PRC2.2 holo complexes (141). The N-terminal half of the SUZ12(ZnB) helix is replaced by the zinc finger protein JAZF1 in a recurrent oncogenic chromosomal translocation found in endometrial stromal tumors ( Figure 5D ) (158). In line with the structural finding that the SUZ12(ZnB) helix is essential for JARID2 binding, PRC2 containing the JAZF1-SUZ12 fusion protein displays greatly impaired JARID2 association, which may at least in part account for the global loss of the H3K27me3 histone mark in patient samples (141, 159).

EZH1 and EZH2 represent a distinct lysine methyltransferase family (160). The SET domains of EZH1 and EZH2 differ from many other SET domains by requiring two binding partners, EED and SUZ12, for the enzymatic activity (49). Structural data indicate that EZH2 contains an unusual split catalytic domain: the SAL (SET activation loop) from the N-terminal portion and the SET domain at the C-terminus are folded together to enable catalysis ( Figure 6A ) (138, 139). Structural comparison to the inactive conformation of an isolated SET domain of EZH2 indicates that the SET domain undergoes structural reorganization from the inactive to the active state, reshaping the H3-binding groove and SAM-binding pocket for effective substrate binding (138, 139, 161, 162). H3K27me3 is a major mediator of PRC2 function in gene regulation in cells; correspondingly, the enzymatic activity is a focal point of the cellular regulation of PRC2 function. The catalytic mechanism, enzyme regulation in normal cells, and enzyme dysregulation in diseased states are summarized in the following paragraphs.

The catalytic mechanism, cancer mutations, and SAM-competitive inhibitors. Histone substrate recognition and methyl transfer reaction are two key steps for PRC2-mediated H3K27 methylation. The A25-R26-K27-S28 core sequence within the histone H3 tail and in particular residue R26 are read by the SET domain directly; residue A25 also contributes to the substrate specificity by the mechanism of steric exclusion (139). Enzyme reactivity data on in vitro methylation of peptide arrays have expanded the substrate specificity of PRC2 to a preferred sequence of [A/C/V/P]24-[A/V/L]25-[R/K]26-[K]27-[F/Y/H]28-[A/V/C/T/S]29 (96). Phosphorylation of residue S28 of histone H3 by cell signaling-dependent MSK kinases diminishes H3K27 methylation, displaces PRC2 from chromatin, and promotes H3K27 acetylation (96, 163). Based on the structure of PRC2 bound to H3K27M ( Figure 6A ), the substrate residue K27 is predicted to be hosted in the lysine-binding channel lined by several aromatic residues such that its nitrogen atom is placed close to SAM for the methyl transfer reaction (138, 139). Compared to H3K27me1 and H3K27me2, the production of H3K27me3 is naturally disfavored because of spatial restriction at the active site (139). Residue Y641 of EZH2 at the enzyme active site is mutated to smaller residues, like F, N, S, H, or C, in follicular and diffuse large B-cell lymphomas, which results in an H3K27 hypertrimethylation disease phenotype, possibly due to a relatively expanded active site ( Figure 6B ) (61–63). Another gain-of-function active site mutation, A677G, found in human B-cell lymphoma, causes hypertrimethylation likely via a similar structural mechanism ( Figure 6B ) (64). A family of SAM-competitive pyridone inhibitors of EZH2, like GSK126, EPZ-6438, and CPI-1205, have been developed to treat cancers with EZH2 gain-of-function mutations as well as other EZH2-dependent cancers (164). These inhibitors exhibit high potency and specificity and are mostly bound to a binding pocket within a unique interface between the SAL and SET of EZH2, with only a small set of atoms protruding into the SAM binding pocket to block SAM binding ( Figure 6B ) (165–167).

Allosteric stimulation, Weaver syndrome mutations, and H3K27me3-competitive inhibitors. H3K27me3 stimulates PRC2 through an allosteric mechanism, which involves H3K27me3 binding to EED and transmission of the stimulating signal to the enzyme active site ( Figure 6A ). H3K27me3 is recognized by an aromatic cage on the top face of the EED WD40 repeat ( Figure 6A ) (52). EED provides limited sequence specificity beyond the trimethylated lysine; in vitro binding assays indicate the A-R-Kme3-S sequence content found for H3K27me3, H3K9me3, and H1K26me3 is favored (52, 168). Similar to H3K27me3, di- and trimethylation products of a lysine residue from two accessory subunits of PRC2, JARID2 and PALI1, can also bind EED and stimulate PRC2 enzymatic activity. PRC2 stimulation by JARID2K116me3 and PALI1K1241me3 may serve as a cellular mechanism promoting H3K27me3 deposition at loci devoid of existing H3K27me3 (169, 170). The SRM (stimulation-responsive motif) of EZH2 bridges the stimulating signal to the enzyme active site through simultaneous binding to the Kme3-containing sequence positioned on the EED surface and to an otherwise partially ordered helix from the SET domain ( Figure 6A ) (138, 139). The SRM itself is a mobile structural element, adopting an ordered α-helix structure in the presence of H3K27me3 or JARID2K116me3 and becoming disordered in their absence (138, 139). Structural stabilization of the SET domain and change of structural dynamics of active site residues may both contribute to the observed allosteric enzyme stimulation (138, 139). Mutational disruption of the EED aromatic cage in mESCs compromises not only the formation of H3K27me3 chromatin domains at nucleation sites, but also the spreading of H3K27me3 from nucleation sites (171). A class of chemical inhibitors, such as A-395 and EED226, specifically targets the EED aromatic cage, which blocks H3K27me3 binding and thereby abolishes the allosteric stimulation of PRC2 ( Figure 6C ) (172, 173). These inhibitors induce tumor regression in a xenograft model and complement the anti-tumor activity of SAM-competitive inhibitors of PRC2 when acquired resistance arises (172, 173).

Enzyme inhibition by H3K27M and EZHIP in pediatric brain tumors. Expression of H3K27M oncohistone in diffuse midline gliomas causes global loss of the H3K27me3 histone mark (71, 72). A crystal structure captures H3K27M bound to a minimally active PRC2 ( Figure 6A ) (139). H3K27M is found to occupy the lysine binding channel at the active site of PRC2, conferring competitive enzyme inhibition by excluding the H3K27 substrate ( Figure 6A ) (139). Compared to H3K27, H3K27M displays an over 10-fold tighter binding to PRC2 (139), which also depends on the SAM concentration (174, 175). H3K27M preferentially associates with the H3K27me3-stimulated state of PRC2, leading to defects in H3K27me3 spreading while retaining H3K27me3 at PRC2 recruitment sites on selected CGIs (174–177). Remarkably, EZHIP, a protein normally found in gonads, is abnormally expressed in posterior fossa ependymoma, where it inhibits PRC2 enzymatic activity through an H3K27M-like protein sequence (73–77). Nucleosome binding to PRC2 surface helps guide residue K27 of the histone H3 tail to the enzyme active site (178, 179); likewise, additional binding sites may be needed to provide similar structural guidance to the H3K27M-like sequence of EZHIP (76).

Whereas the core subunits of PRC2 provide critical binding surfaces to nucleosome core particles, functional domains of some accessory subunits of PRC2 are known to interact with linker DNAs and histone marks, which together dictate chromatin targeting of PRC2. Importantly, recent cryo-EM work reveals how AEBP2 and JARID2-containing PRC2.2 engages with a mononucleosome harboring the H2AK119ub histone mark (157), representing a relatively complete structural view of PRC2.2 at a functional state. In comparison, PRC2.1 in the nucleosome-bound state remains less structurally understood. The current knowledge on structural features conferring chromatin binding by PRC2 is discussed below.

H3K36me3 and linker DNA binding by PHF1, MTF2, and PHF19. H3K36me3 is specifically recognized by PHF1, MTF2, and PHF19, with the trimethylated lysine accommodated in an aromatic cage of the Tudor domain, which links PRC2 to chromatin to possibly mediate de novo silencing of active genes ( Figure 7A ) (154, 180–182). Disruption of the H3K36me3-Tudor interaction in PHF19 and PHF1 hampers PRC2 recruitment during mESC differentiation and PHF1-mediated DNA damage response, respectively (154, 180–182). Residue H3K36 is found close to the binding interface between EZH2 and nucleosomal DNAs in a recent cryo-EM structure, where H3K36 methylation is predicted to impair PRC2 enzymatic activity allosterically (179, 183). How the Tudor domain of PHF1 may contribute to H3K36me3-mediated PRC2 inhibition remains to be elucidated (181). The EH/WH (extended homology/winged-helix) domain of PHF1, MTF2, and PHF19 bind DNAs containing CpG dinucleotides, which directly facilitates PRC2 targeting to CGI chromatin in mESCs ( Figure 7A ) (184). Specifically, two lysine residues from the conserved W1 loop of the EH/WH domain of PHF1 and MTF2 are inserted into the major groove of DNA, reading the unmethylated CpG sequence content ( Figure 7A ) (184). In contrast, the EH/WH domain of Drosophila PCL does not seem to display sequence specificity towards DNA binding (185).

PRC2 binding to dinucleosomes and mononucleosome binding to the PRC2 dimer. Bifunctional dinucleosomes, in which a ‘substrate nucleosome’ is connected to a trimethyllysine analog-containing ‘allosteric nucleosome’ via a 30 base pair (bp), 35 bp, or 40 bp linker DNA, are captured mostly engaged with the catalytic module of PRC2 ( Figure 7B ) (178, 185). Different linker DNA lengths are accommodated by variations of the allosteric nucleosome conformation and linker DNA path, whereas substrate nucleosomes stay relatively static (178). In addition to EED aromatic cage, a major binding surface for the allosteric nucleosome is the SBD–SANT1 intramolecular complex; the substrate nucleosome contacts the CXC domain of EZH2 such that H3K27 is brought to the substrate-binding groove at the active site ( Figure 7B ) (178, 185). The accessory subunit-binding module of PRC2 also appears to loosely associate with nucleosomes (178), although its structure is difficult to resolve (178, 185). In another cryo-EM structure, an EZH1-containing PRC2 dimer is found to engage with opposite sides of a mononucleosome via similar interacting surfaces used by the substrate nucleosome mentioned above ( Figure 7C ) (186). The linker DNA exiting the mononucleosome also contacts the accessory subunit-binding module, through which the PRC2 dimer is formed ( Figure 7C ) (142, 186). In line with its important role in CGI DNA binding (142), PRC2 dimerization is also required for chromatin compaction by EZH1-containing PRC2 (186). Interestingly, AEBP2 and JARID2 present in the reconstituted holo complex are visible only in the structure of a monomeric form of PRC2.2 but not in the structure of the PRC2 dimer (186), which is consistent with the previous observation that AEBP2 binding disrupts the intrinsic PRC2 dimer (142).

PRC2.2 binding to nucleosomes bearing H2AK119ub. In contrast to the traditional hierarchical recruitment model, in which H3K27me3 added by PRC2 helps recruit PRC1 (112), recent studies indicate that H2AK119ub deposited by PRC1 can be critical for PRC2 recruitment by associating with AEBP2 and JARID2 (187–190). This model is directly supported by the cryo-EM structure of a PRC2.2 holo complex bound to a mononucleosome with chemically installed H2AK119ub ( Figure 7D ) (157). Notably, the ubiquitin interaction (UI) motif of JARID2 [JARID2(UI)] and zinc finger (Zn) domain of AEBP2 [AEBP2(Zn)] are shown to independently interact with the two copies of ubiquitin on the nucleosome ( Figure 7D ) (157). A basic bridge (Br) helix of EZH2 [EZH2(Br)] is induced to form in this context, bridging nucleosomal DNA and histone H3 tail to the SET domain ( Figure 7D ) (157). The bridge helix is also known to be disordered in the absence of the nucleosome and become automethylated in the presence of SAM, leading to PRC2 activation (191, 192).