RNA-induced PRC2 inhibition depends on the sequence of bound RNA

Abstract

Methyltransferase PRC2 (Polycomb Repressive Complex 2) deposits histone H3K27 trimethylation to establish and maintain epigenetic gene silencing. PRC2 is precisely regulated by accessory proteins, histone post-translational modifications, and, particularly, RNA. Research on PRC2-associated RNA has mostly focused on the tight-binding G-quadruplex (G4) RNAs, which inhibit PRC2 enzymatic activity in vitro and in cells, a mechanism explained by our recent cryo-EM structure showing G4 RNA-mediated PRC2 dimerization. However, PRC2 binds a wide variety of RNA sequences, and it remained unclear how diverse RNAs beyond G4 associate with and regulate PRC2. Here, we show that variations in RNA sequence elicit disparate effects on PRC2 function. A G-rich RNA lacking consecutive G’s and an atypical G4 structure called a pUG-fold mediate PRC2 dimerization nearly identical to that induced by G4 RNA. In contrast, pyrimidine-rich RNAs, including a motif identified by CLIP-seq in cells, do not induce PRC2 dimerization and instead bind PRC2 monomers with retention of methyltransferase activity. Only RNAs that dimerize PRC2 compete with nucleosome binding and inhibit PRC2 methyltransferase activity. Thus, PRC2 binds many different RNAs with similar affinity; however, the functional effect on enzymatic activity depends entirely on the sequence of the bound RNA, a conclusion potentially applicable to any RNA-binding protein with a large transcriptome.

PRC2 binds many RNAs, but how RNA sequence determines PRC2 regulation has remained unclear. Here, the authors show that G-rich RNAs promote PRC2 dimerization and inhibition, whereas pyrimidine-rich RNAs bind a monomeric complex without disrupting enzymatic activity.

Introduction

In the last decade, non-canonical RNA-binding proteins (RBPs) that do not contain conventional RNA-recognition or RNA-binding motifs have been of increasing interest. Examples include histone modifiers^1,2, chromatin architecture remodelers^3–5, DNA methylases^6,7, transcription factors^8–10 and metabolic proteins^11,12, many of which participate in tuning differential expression of genes. The binding of RNA has been reported to have various functions, including promoting complex recruitment and enhancing target recognition specificity as a positive regulator and, on the other hand, inhibiting enzymatic activity and limiting target accessibility as a negative regulator, thereby emphasizing the under-characterized yet critical role of RNA in epigenetic regulation. Non-canonical RBPs lack a classical RNA-recognition motif, and they often engage RNA through flexible, multivalent, and electrostatic interfaces rather than a single defined binding pocket^13–15. As a result, the same protein can accommodate RNAs with different sequences and structures in distinct binding modes, potentially leading to different functional consequences^16–18. Understanding how such RNA-protein interactions encode regulatory specificity remains a critical challenge in RNA biology.

The histone methyltransferase Polycomb Repressive Complex 2 (PRC2) is a prevailing model system for studying the mechanism and significance of RNA binding to non-canonical RBPs^19–21. PRC2 trimethylates lysine 27 of histone H3 (H3K27me3), which is a repressive mark for gene expression and essential for normal development and cell differentiation^22,23. Studies have shown that RNAs capable of folding into G-quadruplex structures (G4 RNAs) bind PRC2, inhibiting its activity in cells^24–27. Our recent cryo-EM structure of a G4 RNA-bound PRC2 complex revealed the molecular mechanism of such inhibition²⁸. Instead of blocking PRC2-chromatin binding by simple steric inhibition, the G4 RNA bridges two PRC2 protomers to form a dimer that specifically buries PRC2 amino acids required for docking the histone H3 tail and binding the nucleosomal DNA.

However, the PRC2 interactome has been reported to contain thousands of pre-mRNAs and lncRNAs, including pyrimidine-rich RNAs and G-rich RNAs with fewer than four G-tracts, which cannot fold into a G4 structure^26,29–36. These other RNAs differ from canonical G4 structures in nucleotide composition, shape, charge distribution, and the availability of H-bond donors/acceptors, raising questions about how PRC2 recognizes various RNAs and, more importantly, what the functional consequences are for PRC2 associating with these different RNAs.

Here, we use biochemical methods and electron microscopy (EM) to investigate PRC2 binding to diverse RNAs. Notably, RNA-mediated PRC2 dimerization is sequence-dependent. G-rich RNAs, including sequences without consecutive Gs, can induce PRC2 dimerization. Flexible loops of PRC2 accommodate different RNA structures, showing how G-rich RNA beyond G4 can inhibit PRC2. In contrast, PRC2 binds pyrimidine-rich RNAs as a monomer, which does not prevent PRC2 activity. Overall, our study expands RNA-mediated PRC2 regulation to multiple RNA elements, enhancing the understanding of how diverse RNAs can mediate different functional consequences for this key regulator of epigenetic gene silencing.

Results

PRC2 complexes bind multiple RNA sequences

Although PRC2-RNA binding has been studied previously^35,37–40, we sought to extend these analyses by examining a series of PRC2 complexes with a more comprehensive set of RNAs encompassing different sequences and structural features. We first prepared the six-subunit PRC2 complex with accessory proteins AEBP2 (embryonic isoform, amino acids 210–503 corresponding to isoform 1) and truncated JARID2 (amino acids 119–450), as these accessory proteins have previously been implicated in RNA binding^24,41. To this end, seven 50-nucleotide RNA oligos were synthesized (Fig. 1a). TERRA contains four repeats of the human telomeric sequence and can form a single G4 structure, mimicking the noncoding RNA transcribed from telomeres⁴², which has been shown to recruit PRC2 to telomeres⁴³. TERRA_mut is a G-rich RNA that has the same base composition as TERRA, but the sequence is mutated to eliminate consecutive guanines, preventing G4 formation. The pUG-fold RNA has 20 repeats of GU, which form an atypical RNA quadruplex and are distinct from the canonical G4 by having three G quartets and one additional U quartet⁴⁴ (Fig. 1b). Poly(A) does not bind to the five-subunit PRC2³⁵ and serves as a negative control in our characterization of the six-subunit complex. Finally, Poly(U), Poly(C), and P10 are pyrimidine-rich RNAs. In particular, P10 has a C and U repetitive sequence, which was initially identified as a PRC2 binding consensus in cells by crosslinking and immunoprecipitation (CLIP)³⁶.

Fig. 1. PRC2 complexes bind RNAs of various sequences and structures.

a The seven RNA oligonucleotides used in this study. Colors highlight the functional sequences, while the 5’ (A)₁₀ or (U)₁₀ linkers provide flexibility for binding to streptavidin EM grids. b Schematic representations of canonical G4 (TERRA) and atypical G4 (pUG-fold) with G in gold, U in blue, and A in green. c Top: schematic representation of six-, five- and four-subunit PRC2 used in this study. Bottom: quantification of independent EMSA replicates and the corresponding K_d values of PRC2 binding RNAs. The exact number of independent replicates (n) is indicated in panel d. Error bars are mean ± standard deviation. d Summary of K_d values. Black dots are K_d values of individual experiments. n is the number of independent replicates. Error bars are mean ± standard deviation. P-values were calculated using an unpaired two-sided t test.

Native gel electrophoresis confirmed the formation of compact RNA structures for TERRA and pUG-fold RNAs but not for TERRA_mut or pyrimidine-rich RNAs (Supplementary Fig. 1b). Circular Dichroism (CD) spectra and melt assays confirmed the expected structure of the pUG-fold RNA⁴⁴ and indicated that TERRA_mut formed neither a pUG-fold nor a canonical G4 structure (Supplementary Fig. 1c–f).

Electrophoretic mobility shift assays (EMSA) were performed with six-subunit PRC2 and various RNAs in 100 mM KCl to approximate nuclear ionic conditions, stabilizing G4 structures (Fig. 1c and Supplementary Fig. 2a). PRC2 bound TERRA with high affinity (K_d= 18 ± 9 nM). Surprisingly, it bound Poly(U) (K_d= 28 ± 4 nM) and Poly(C) (K_d= 28 ± 8 nM) with similarly high affinities. Binding to these RNAs was approximately 2-fold, 3-fold, and 4-fold stronger than to pUG-fold, TERRA_mut, and P10, respectively. When binding was instead performed in 100 mM LiCl to preclude G4 formation, the three G-rich RNAs bound PRC2 with similar affinities (Supplementary Fig. 2d), indicating that the six-subunit PRC2 complex retained the preference for binding the folded G4 RNA but could also bind unfolded G-rich RNAs.

Our initial hypothesis for PRC2 preferring G4 was that the G4 RNA has a more condensed electrostatic charge distribution in its compact structure, but the high binding affinities of single-stranded pyrimidine-rich RNAs seen here do not support this idea. We therefore tested whether the accessory proteins (JARID2 and AEBP2) in the six-subunit complex aided binding to these additional RNA structures and sequences. The five-subunit PRC2 (missing JARID2) had near identical binding affinities for the three G-rich RNAs as the six-subunit, but it had substantially reduced binding affinities for all pyrimidine-rich RNAs (Fig. 1d and Supplementary Fig. 2b). Interestingly, the four-subunit holoenzyme (missing AEBP2 and JARID2) showed weak binding to TERRA_mut but not TERRA or pUG-fold (Fig. 1d and Supplementary Fig. 2c). The K_d(TERRA_mut)/K_d(TERRA) ratio was 12 for the four-subunit PRC2, significantly larger than the ratio of 4 obtained for the six-subunit and five-subunit complexes. Additional reduction of binding affinities to the Poly(C) and P10 RNAs was also found for the four-subunit complex compared to the five-subunit complex.

Thus, our systematic characterizations indicate that the four-subunit holoenzyme contains all essential protein elements required to recognize G4 as the most preferred RNA, consistent with our cryo-EM structure and other studies, which show that the catalytic subunit EZH2 is predominantly responsible for G4 RNA recognition^28,40,45,46. Conversely, the six-subunit complex binds RNA sequences and structures more broadly. Accessory proteins significantly enhance the association of PRC2 with pyrimidine-rich RNAs (via JARID2) and most single-stranded RNAs (TERRA_mut, Poly(C), and P10 via AEBP2). More generally, these biochemical results help explain the in vivo results, which show a broad PRC2 transcriptome containing many lncRNAs and pre-mRNAs^{26,30,32,33,47,48}.

PRC2 dimerization is RNA sequence-dependent

We recently adapted the streptavidin-affinity EM grid method⁴⁹ to allow specific selection and enrichment of ribonucleoprotein (RNP) complexes²⁸. Here, we used 5’-biotinylated RNAs and negative-staining EM to determine the architectures of six-subunit PRC2 bound to different RNA sequences (Fig. 2a). Reference-free 2D-class averages of PRC2-TERRA, PRC2-TERRA_mut, and PRC2-pUG-fold RNPs exhibited PRC2 dimerization with a similar protomer arrangement as shown in the cryo-EM structure of the (GGGAA)₄ G4 RNA-mediated PRC2 dimer²⁸. The finding that TERRA_mut could mediate PRC2 dimerization was interesting, given that it is unlikely to resemble a folded G4 RNA in terms of shape, hydrogen bond donor/acceptor potential, or electrostatic charge distribution. In contrast, the majority of PRC2 RNP particles remained monomeric when bound to pyrimidine-rich RNAs. We validated that the affinity grid method captured only RNA-bound complexes, as Poly(A) RNA did not capture any recognizable particle (Supplementary Fig. 3). Quantifying two negative-staining EM replicates of each PRC2-RNA complex revealed that dimerized PRC2 consistently comprised 70% to 90% of all recognizable particles when incubated with G-rich RNAs and less than 20% with pyrimidine-rich RNAs (Fig. 2b).

Fig. 2. RNA sequence-dependent dimerization of PRC2 and evidence that G4 and Poly(C) RNAs bind to different sites on PRC2.

a Negative-staining EM provided 2D-class averages of six PRC2-RNA complexes, three RNAs inducing PRC2 dimers and three RNAs binding to PRC2 monomers. The 2D-class averages are shown in order of prevalence from left to right. b Quantification of two negative-staining EM replicates (two independent EM grids, each grid utilizing PRC2 from a separate preparation). c Top: size-exclusion chromatography of PRC2 preincubated with RNAs and mock (protein only). Abs, absorbance; mAU, milli-absorbance unit. Bottom: standard curve used to estimate the molecular weights of complexes. d FRET assay indicates the formation of TERRA-Poly(C)-PRC2 ternary complex. Top left: PRC2-TERRA binding measured by fluorescence polarization (FP) of Cy3 signal. Top right: PRC2-Poly(C) binding measured by FP of Cy5 signal. Binding curves for each ligand were fit with Eq. (1) (see “Methods”). Bottom left: Predicted FRET behavior as a function of protein concentration is calculated from the binding curve fits via Eq. (2), which assumes the two ligands bind independently. Bottom right: the observed FRET of Cy3-TERRA + Cy5-Poly(C). The Cy3-Poly(A) + Cy5-Poly(C), Cy3-TERRA + Cy5-Poly(A) and Cy3-Poly(A) + Cy5-Poly(A) combinations serve as negative controls. All data were collected from the same reaction containing fixed, equal concentrations of both [Cy3]-labeled and [Cy5]-labeled RNAs and increasing concentrations of PRC2. Three independent replicates gave the same conclusion (n = 3). e Representative competitive EMSA assays. PRC2 was preincubated with ³²P-labeled TERRA (left) or Poly(C) (right) before adding competitor RNAs at different concentrations. Successful competition removed ³²P-labeled RNA from PRC2. Three independent replicates gave the same conclusion (n = 3).

Analytical size-exclusion chromatography was performed to validate our EM observations in solution (Fig. 2c). Six-subunit PRC2 alone and PRC2-Poly(A) chromatographed at a molecular weight of approximately 350 kDa, equal to the sum of the individual PRC2 subunits. PRC2-Poly(U), PRC2-Poly(C), and PRC2-P10 eluted at the same retention volume with the 280 nm/260 nm absorbance ratio significantly lower than 2, consistent with PRC2 binding pyrimidine-rich RNAs as a monomer. This chromatography trace differed from the profiles of PRC2 and G-rich RNAs, which formed RNPs with an approximate molecular weight of 700 kDa. Our results suggest that RNA nucleotide composition predicts the RNA-mediated dimerization of six-subunit PRC2.

The two distinct RNA sequence-dependent binding modes of PRC2 prompted us to hypothesize that the six-subunit PRC2 could utilize distinct protein regions to engage different RNAs. If so, then the PRC2 binding of G-rich and pyrimidine-rich RNAs might not be mutually exclusive. We tested this idea by fluorescence resonance energy transfer (FRET) (Fig. 2d) and RNA-competition EMSA (Fig. 2e). The clear concentration-dependent FRET signal at low PRC2 concentrations indicated the formation of a PRC2-TERRA-Poly(C) tertiary complex. In RNA-competition EMSA, Poly(C) could not remove the pre-bound radiolabeled TERRA from PRC2 (Fig. 2e, left panel). In the other direction, it required considerably more TERRA than Poly(C) itself to effectively compete with pre-bound Poly(C) (Fig. 2e, right panel). Both results support the conclusion that competition between different RNA ligands is always less efficient than self-competition. In summary, six-subunit PRC2 can simultaneously engage G-rich and pyrimidine-rich RNAs, and the sequence of the RNA determines its binding mode with PRC2.

Cryo-EM structure of TERRA_mut-bound PRC2 dimer

To understand the interaction of PRC2 with an RNA that does not fold into a canonical G4 structure, we solved a 3.3 Å cryo-EM structure of the TERRA_mut-bound six-subunit PRC2 (Fig. 3, Supplementary Figs. 4 and 5 and Supplementary Table 1). In this structure, the architecture of the PRC2 dimer was identical to that of the G4 RNA-mediated complex²⁸, including the protein-protein interface assembled by the CXC domain of the catalytic subunit EZH2. This dimer interface is key to explaining RNA inhibition of PRC2 activity, because it buries amino acids required for histone H3 tail and nucleosome binding in the active PRC2 monomer⁵⁰.

Fig. 3. Cryo-EM structure of TERRA_mut RNA-bound PRC2 dimer.

a Top: Cryo-EM density map of TERRA_mut-associated PRC2 dimer. EZH2 (CXC-SET) of protomer 1 in blue, EZH2 (CXC-SET) of protomer 2 in light blue, and TERRA_mut RNA in orange. Bottom: The corresponding atomic model. b Direct comparison of G4 RNA- and TERRA_mut RNA-bound PRC2. Model is from G4 RNA-bound PRC2 (PDB:8fyh). Transparent density is the local resolution-filtered cryo-EM map of TERRA_mut RNA-PRC2 complex. Dashed boxes are zoomed in to present details at right. Subunits of dimerized PRC2 are at the same arrangement in both structures, while the TERRA_mut-PRC2 complex has the RNA density partially overlapping with the G4 model and additional density of RBAP48 or AEBP2 visualized (arrow).

We could confidently identify one TERRA_mut RNA density, suggesting a single TERRA_mut is sufficient to induce PRC2 dimerization (Fig. 3a). Because PRC2 utilizes flexible loops to engage RNA, allowing multiple RNA orientations and conformations, we could not obtain high-resolution details to model TERRA_mut de novo. Thus, the data do not indicate whether this RNA was completely unfolded or not. The TERRA_mut location partially overlapped with that of the G4 RNA in our previous cryo-EM reconstruction (Fig. 3b). This can be explained if both PRC2 protomers contribute to RNA binding in a pincer-like arrangement. The G4 and TERRA_mut RNAs are suspended at similar positions in the middle of the two PRC2 protomers without a strict RNA structure requirement. Also, the TERRA_mut density is attached to the bottom lobe of one PRC2 protomer in a region of RBAP48 and AEBP2, which was less detectable in the G4 RNA-bound map (Fig. 3b). This difference may indicate that the TERRA_mut, expected to be more extended than a G4, interacts with additional protein surfaces. Overall, the use of flexible loops to bind RNA, the formation of a dimer with RNA suspended between the protomers, and the additional protein surfaces available for extended RNA elements appear to enable PRC2 to accommodate disparate RNA structures.

We also attempted to solve the cryo-EM structure of the pyrimidine-rich RNA-bound PRC2. However, a technical limitation of the affinity grid method is the requirement to apply three support layers on the surface of the EM grid: streptavidin crystal, biotinylated lipid, and thin carbon (≈ 2 nm thick)^49,51. Thus, most small particles (e.g., monomeric PRC2) do not exhibit strong intensity contrast after computational streptavidin-lattice subtraction, and therefore are insufficient for reliable particle picking and further alignment. Also, monomeric PRC2 is susceptible to damage during sample preparation, due to the absence of crosslinking known to stabilize monomeric PRC2 from our previous cryo-EM studies⁵². For these reasons, we were unable to obtain a cryo-EM map of monomeric PRC2 bound to a pyrimidine-rich RNA.

AEBP2 is implicated in contributing to single-stranded RNA binding

Because the details of RNA-protein interaction were not resolved by cryo-EM, we used AlphaFold 3⁵³ to generate models of TERRA_mut, Poly(U), and Poly(C) bound to a single six-subunit PRC2 (Supplementary Figs. 6 and 7). We did not model the PRC2 dimer because two copies of PRC2 exceed the input limit of 5000 residues in the AlphaFold 3 Server. In parallel, we performed multibody refinement of our cryo-EM consensus map to focus on the single PRC2 protomer that interacted with TERRA_mut, which could be directly compared to the AlphaFold 3 models. Across all PRC2-TERRA_mut predictions, the overall organization of PRC2 subunits remained consistent and agreed well with published PRC2 structures^52,54–59, while TERRA_mut RNA bound to the expected region of PRC2 but had variable conformations and orientations with respect to PRC2 in each predicted model (Supplementary Figs. 6a–d). AlphaFold 3 predictions had high confidence for PRC2 subunits (chain iPTMs = 0.4–0.6) and low confidence for TERRA_mut (chain iPTM = 0.1). Predictions of PRC2-TERRA_mut, PRC2-Poly(U), and PRC2-Poly(C) revealed no significant differences in terms of the overall PRC2 and RNA arrangement and the details of protein-RNA interactions.

Despite the low confidence in RNA modeling by AlphaFold 3, it was gratifying to see that the predicted TERRA_mut always overlapped with the density we designated as the RNA in our cryo-EM multibody reconstruction. To better present this feature, we generated a surface map that includes only regions overlapping between all top AlphaFold 3 models (Fig. 4a and Supplementary Fig. 6e). This map eliminates flexible regions of the RNA, similar to the principle used in cryo-EM reconstruction that averages several hundred thousand particles. This surface map had identical architecture to the actual cryo-EM map, particularly with only part of TERRA_mut being visualized, explaining how the intrinsic flexibility of PRC2-RNA binding affects our EM map.

Fig. 4. PRC2 accessory protein AEBP2 is implicated in contributing to RNA binding.

a Left: Density map generated by keeping only the overlapped regions of the top four AlphaFold 3 models (see “Methods”). Right: Local resolution filtered cryo-EM density from multibody refinement of TERRA_mut RNA-associated PRC2. Dashed boxes are zoomed in. The remaining density of TERRA_mut after averaging AlphaFold 3 models has a similar size and position compared to the observed cryo-EM density. b Quantification of three independent EMSA replicates of the six-subunit PRC2_{AEBP2Δ261-389} binding different RNAs (n = 3). Error bars are mean ± standard deviation. c Left: Representative images of the DNA-RNA competition assay. Two independent experiments yielded the same result (n = 2). Right: Quantification of DNA-RNA competition. Symbols indicate the average of two replicates with individual values shown as bars. Concentrations in nM represent the concentration of unlabeled RNA or DNA that removes 50% of the prebound radiolabeled dsDNA from PRC2.

AlphaFold 3 indicated the importance of PRC2 accessory protein AEBP2, utilizing zinc-finger motifs and an arginine-rich segment (379-KRRKLKNKRRR-389), for binding TERRA_mut, Poly(U) and Poly(C) RNAs (Supplementary Fig. 7). The second and third C2H2-type zinc-finger motifs within AEBP2 are in the periphery of the RNA, engaging RNA via electrostatic interactions with the phosphate backbone and hydrogen bonds with both backbone and bases. The residues responsible for this interaction differ across models, indicating that, unlike typical zinc-finger motifs that recognize specific DNA sequences, AEBP2 zinc-finger motifs exhibit less stringent RNA sequence specificity. In addition, the arginine-rich segment of AEBP2 contains nine positively charged amino acids, showing a strong electrostatic potential to bind many RNAs.

To test these predictions, we prepared the six-subunit PRC2 complex with truncated AEBP2 (PRC2_{AEBP2Δ261-389}), which lacks both zinc-finger motifs and the arginine-rich segment. Compared to WT six-subunit PRC2, PRC2_{AEBP2Δ261-389} had a modest reduction of TERRA_mut binding (K_d(PRC2_WT) = 66 ± 15 nM and K_d(PRC2_{AEBP2Δ261-389}) = 112 ± 24 nM, P = 0.05) and more dramatic decrease of Poly(C) binding (K_d(PRC2_WT) = 28 ± 8 nM and K_d(PRC2_{AEBP2Δ261-389}) = 195 ± 90 nM, P = 0.03) (Fig. 4b and Supplementary Fig. 8). The binding of all other RNAs was not statistically different between WT PRC2 and PRC2_{AEBP2Δ261-389}. This result is largely consistent with the reduced RNA binding affinities we observed for the PRC2 four-subunit holoenzyme (Fig. 1d), in which AEBP2 is absent, indicating that the zinc-finger motifs and arginine-rich segment are primarily responsible for enabling AEBP2 of PRC2 to accommodate diverse RNAs.

Because the arginine-rich segment of AEBP2 has been previously characterized as a nucleosomal DNA binding interface^54,60, we tested the hypothesis that RNA competes with DNA by binding the same region of AEBP2. Although both TERRA_mut RNA and Poly(C) RNA bind AEBP2, our DNA-RNA competition EMSA with six-subunit PRC2 revealed robust competition between a 60-bp double-stranded (ds) DNA and the G-rich TERRA_mut RNA, whereas no convincing competition was seen with the Poly(C) RNA (Fig. 4c). These results suggest that RNA-induced PRC2 dimerization, rather than direct competition at AEBP2, is the major mechanism driving dsDNA eviction from PRC2. Notably, in the context of a six-subunit PRC2 complex lacking only the AEBP2 arginine-rich segment (PRC2_{AEBP2Δ379-389}), we observed the same overall trend, with 4.5-fold higher concentrations of TERRA_mut being required to displace dsDNA from the PRC2_{AEBP2∆379-389} than from WT PRC2 (Fig. 4c). This weaker competition is consistent with the reduced TERRA_mut binding affinity upon deletion of this segment of AEBP2.

RNAs that induce dimerization inhibit PRC2 activity

Most biochemical characterizations of RNA-mediated PRC2 inhibition have focused on short G4-forming sequences^{24,27,28,40,61} or long RNAs that likely fold into complex 3D architectures containing multiple structural elements (e.g., HOTAIR and Xist lncRNAs)^24,26,38,62. The effect of simple G-tract sequences other than canonical G4 and especially pyrimidine-rich RNAs on PRC2 activity has not been measured.

Our structures of RNA-induced PRC2 dimers have demonstrated that the EZH2 dimer interface prevents nucleosomal DNA and H3 tail accessibility to the PRC2 catalytic groove. As expected, all G-rich RNAs that can induce PRC2 dimerization inhibited radiolabeled tri-nucleosomes from binding PRC2 in nucleosome-RNA competition assays (Fig. 5a). Pyrimidine-rich RNAs, instead, did not interfere with nucleosome binding (Fig. 5a), indicating that six-subunit PRC2 has separate nucleic acid-binding segments specialized for nucleosomal DNA, pyrimidine-rich RNA, and G-rich RNA. Consistent with this model, methyltransferase activity assays of PRC2 with co-incubation of RNAs showed a progressive reduction of H3K27 methylation when we included TERRA, TERRA_mut, and pUG-fold in trans, but exhibited no change with pyrimidine-rich RNAs and Poly(A) negative control (Fig. 5b and c). IC₅₀ of TERRA, TERRA_mut, and pUG-fold were all around 500 nM, and complete inhibition was achieved at higher RNA concentrations. The observed IC₅₀ is higher than the K_d because the inhibition of PRC2 is determined not only by the K_d of the inhibitor but by the RNA-PRC2 stoichiometry⁶³, given that 600 nM PRC2 was used in the assay.

Fig. 5. Only RNAs that induce dimerization inhibit PRC2 binding to nucleosomes and activity.

a Representative EMSA gels of nucleosome-RNA competition assays. PRC2 was preincubated with ³²P-labeled trinucleosomes before adding competitor RNAs at different concentrations. Successful competitions removed ³²P-labeled trinucleosomes from PRC2, represented by a reduction of PRC2-nucleosome signal and appearance of free trinucleosomes. This experiment was performed three times with equivalent results (n = 3). Incomplete PRC2-trinucleosome complexes are indicated by * and **. We assume two of three nucleosomes were occupied by PRC2 in *, and one of three nucleosomes in **. b Representative histone methyltransferase activity assays with ¹⁴C-labeled S-adenosylmethionine analyzed by SDS-PAGE, with gels imaged for ¹⁴C signals (top) or stained with Coomassie blue to confirm equal loading of PRC2 and nucleosomes (bottom). Two repeats of TERRA RNA are shown at left and right. Three independent replicates gave the same conclusion (n = 3). c Quantification of three replicates (n = 3). Error bars are mean ± standard deviation. IC₅₀ is the concentration of RNA that inhibits 50% of PRC2 activity relative to no RNA. d Table summarizes the four categories of RNA characterized in this study. Affinity (K_d): Representative K_d values derived from EMSA. CLIP in cells: RNA motifs identified by CLIP-seq in cells³⁶.

Most PRC2-associated physiological RNAs in cells are long and likely contain multiple sequence elements in a single RNA. Because G-rich and pyrimidine-rich RNA elements are not mutually exclusive in PRC2 binding (Fig. 2d, e), we hypothesized that dimer-inducing sequences represent the dominant and functional regulatory element within longer RNAs, and that pyrimidine-rich segments within the same transcript would not impede the activity of the regulatory sequence. To test this idea, we designed a chimeric RNA containing a pyrimidine-rich P10 sequence linked to a TERRA sequence via an (A)₁₀ spacer (Supplementary Fig. 9a). In methyltransferase activity assays, this chimeric RNA inhibited PRC2 almost as well as the G-rich RNA alone, indicating that inclusion of a pyrimidine-rich segment does not abrogate G-rich RNA-dependent inhibition (Supplementary Fig. 9b, c).

Overall, our biochemical characterizations clearly illustrate that different RNA sequences mediate distinct functional consequences in regulating PRC2 (Fig. 5d). Only RNAs that induce dimerization prevent PRC2 binding to nucleosomes and subsequently inhibit PRC2 activity.

Limitations of UV crosslinking for capturing PRC2-RNA complexes

A recent study failed to capture RNAs bound to PRC2 in living cells⁶⁴, which challenges the widespread reports of PRC2-RNA interaction and, in particular, is inconsistent with our model and published in vivo results. In this article, Guo et al. modified the conventional CLIP assay to a method called CLAP (covalent linkage and affinity purification), which allows denaturing washes before the identification of protein-associated RNA. Like CLIP, CLAP is completely dependent on 254 nm UV crosslinking, whose underlying biophysical and chemical principles are incompletely understood. However, it is unambiguous that protein-RNA photochemical crosslinking is biased towards uracil^65–67 and towards certain amino acids^68–70, and is highly sensitive to the precise distance and orientation of the RNA base relative to the protein amino acid^68,69,71,72. For example, RNA recognition motifs (RRMs), which rely on the stacking interaction between aromatic amino acids and RNA bases, are UV-crosslinked with unusually high efficiency, thereby dominating CLIP-seq databases⁷⁰. Therefore, we suspected that the efficiency of UV crosslinking might not be sufficient to covalently link PRC2 to G-rich RNAs bound to it in cells.

We therefore performed in vitro UV crosslinking assays of PRC2 simultaneously with the positive control used in Guo et al. namely PTBP1 (Polypyrimidine Tract-Binding Protein 1), which harbors four canonical RRMs and interacts with a defined RNA consensus UCUUUCU⁷³ (Supplementary Fig. 10a, b). We incubated PTBP1 or six-subunit PRC2, each with its cognate RNA, at concentrations 10 times greater than the K_d values to ensure saturated binding in all reactions. The PTBP1-(UCUUUCU)₆ interaction was efficiently crosslinked at multiple UV energies, including 0.25 J/cm², which is frequently applied to intact cells and was used in Guo et al. In the PRC2 reactions, we selected the RNA termed 2G4, which was used in the initial cryo-EM determination of a PRC2-RNA complex and does not contain any uracil in its RNA sequence. We also examined TERRA and TERRA_mut RNAs that contain a few U bases. We found that PRC2 did crosslink to these three RNAs, but with very low efficiency. The RRM-harboring protein PTBP1 crosslinked to associated RNA more than an order of magnitude better than PRC2 (Supplementary Fig. 10c, d). The intrinsic inefficiency of PRC2-RNA UV-crosslinking may contribute to the failure of the CLAP technique in capturing PRC2-RNA complexes. Note that these data do not show that PRC2 binds RNA in cells, but only that the UV-crosslinking CLAP method used to argue against such binding has severe limitations.

Discussion

Several previous studies have validated the biological importance of RNA binding in the precise control of PRC2 function in cells^24–27,47. Inhibition of PRC2 activity by nascent RNA is thought to help maintain epigenetic gene silencing, ensuring that PRC2 adds its repressive histone mark only to those genes that are already silenced at the transcriptional level. In contrast, specific lncRNAs have been reported to recruit PRC2 to target genes, serving as positive regulators of PRC2 chromatin occupancy^19,74. The sequence and structural diversity of these RNAs could potentially provide the specificity needed by PRC2 to respond differentially, but it also raises the critical question of how PRC2 can recognize different RNAs without a strict sequence consensus.

Here, we demonstrate that two categories of RNA have distinct binding modes with PRC2, which, importantly, lead to separate functional consequences. We previously found that the folded structure of G4 RNA mediated PRC2 dimerization and inhibition of histone methylation. This category is now expanded to include other G-rich RNAs, including TERRA_mut and pUG-fold, which act similarly to G4 RNA even though their structures are quite different. Our structural analysis indicates that the flexible, positively charged loops of PRC2 suspend each of these RNAs between two PRC2 protomers, thereby accommodating different RNA structures. In contrast, the second category includes pyrimidine-rich RNAs that exhibit a strong binding affinity for PRC2 without forming a PRC2 dimer. This monomeric RNA-binding mode allows PRC2 to simultaneously engage nucleosomes without limiting histone H3 tail accessibility or inhibiting PRC2 activity. AlphaFold 3 prediction and mutagenesis verification revealed the importance of accessory protein AEBP2 in monomeric PRC2 binding RNA. We also found that pyrimidine-rich RNA binding was enhanced by PRC2 accessory subunits in combination. Consistent with a contribution from JARID2, the six-subunit complex bound pyrimidine-rich RNAs with higher affinity than the five-subunit complex lacking JARID2. At the same time, truncation of AEBP2 reduced pyrimidine-rich RNA binding, indicating that both JARID2 and AEBP2 provide binding contributions, but neither is the sole determinant of pyrimidine-rich RNA recognition. In summary, this study determines the nucleotide compositions of RNA recognized by PRC2 in two distinct binding modes and elucidates the molecular basis of RNA sequence-dependent inhibition of PRC2, thereby reconciling the differential regulation of PRC2 by RNA.

Despite G4 RNA being predominantly studied for PRC2-RNA interaction, many independent research groups have investigated other RNA sequences, including HOTAIR lncRNA^45,75,76, Repeat A motif of Xist lncRNA^38,77, RNA transcribed from B2 SINE retrotransposon^78,79, and artificial sequences without continuous G-tracts^35,40. RIP-seq and CLIP-seq analysis of RNA bound by PRC2 identified C and U repetitive motifs and shorter G-tract motifs incapable of folding into G4 in addition to the G4-forming sequences^36,48. Our results support the conclusion that G4 structure is not necessary for RNA to be recognized by PRC2. Multiple RNA sequences bind PRC2 with relatively strong affinity.

Our previous structural characterization of G4 RNA-bound PRC2 applied the streptavidin-affinity EM grid method to an RNP complex²⁸, providing a technical advance that enriched RNA-bound complexes and improved particle quality by keeping particles away from the denaturing water-air interface^49,51. In this work, we showed the utility of this technique for more challenging RNPs composed of less tightly bound RNAs. However, the technique has the intrinsic disadvantage of applying multiple layers of support background, significantly decreasing signal intensity. This makes the method less useful for small-sized particles (i.e., monomeric PRC2).

In addition, our analysis integrated AlphaFold 3 predictions to illustrate the flexibility of RNA conformations and orientations in our protein-RNA complexes. The close agreement between the AlphaFold 3-predicted model and our cryo-EM density-derived experimental model supports the accuracy of the prediction and highlights how RNA flexibility can limit cryo-EM map quality. Furthermore, it provided clues for testing the role of the PRC2 accessory protein AEBP2 in accommodating RNA, which could not be directly inferred from our cryo-EM map. However, AlphaFold 3 could not predict RNA interactions with PRC2 subunits with high confidence (chain iPTM and chain-pair iPTM < 0.2), so our data do not provide an atomic model to illustrate how zinc-finger domains of AEBP2 bind RNA. Nevertheless, combining cryo-EM and AlphaFold is likely to facilitate many future studies of other non-canonical protein-RNA interactions.

Like PRC2, many other RBPs utilize intrinsically disordered arginine- or lysine-rich patches to engage with various RNAs, exhibiting limited sequence specificity (see reviews^13,80,81). In more than 20% of RBPs, intrinsically disordered regions have been found to comprise a significant portion of the protein sequence. Therefore, it is essential to recognize that the identification of multiple RNA motifs for a given RBP does not necessarily imply that all resulting RNPs perform similar functions. Although conventional methods, including CLIP-seq⁸² and RIP-seq³², are powerful tools for mapping RNA-protein interactions, they cannot resolve the functional heterogeneity among distinct RNA motifs. As demonstrated in this study, the functional outcomes of RNA binding to PRC2 can depend strongly on the sequence and structural properties of the RNA itself. This highlights the importance of detailed biochemical and structural analyses in determining and elucidating how specific RNA sequences modulate the activity or behavior of their RBP partners.

Finally, a limitation of this study is that our conclusions are based on in vitro biochemical and structural observations and are not yet directly validated in cells. Although our data define RNA sequences and structural features that modulate PRC2 dimerization and inhibition, how these mechanisms operate in vivo will depend on additional variables, including local RNA and PRC2 concentrations, RNA modifications, chromatin context, and the presence of PRC2 accessory factors that can tune PRC2 activity and chromatin occupancy. Future work combining quantitative perturbations of endogenous RNAs with measurements of PRC2 recruitment and H3K27 methylation will be required to establish the physiological regulatory impact of the RNA-mediated mechanisms described here. Also, our results do not address the possibility that, in some cases of sparse transcription, RNA may be able to recruit PRC2 to chromatin to facilitate its action (reviewed in refs. ^19,74). This activity is biophysically possible⁸³ but difficult to test in cells.

Methods

Protein expression and purification

For the preparation of the six-subunit PRC2, genes encoding full-length EED, SUZ12, RBAP48, His-tagged EZH2 isoform 2 (UniProt Q15910-2), Strep-GFP-tagged embryonic isoform of AEBP2 (amino acids 210–503 corresponding to isoform 1), and Strep-GFP-tagged truncated JARID2 (amino acids 119–450) were cloned into a single multi-bac plasmid⁵². Each expression cassette had an independent promoter and terminator. This multi-bac plasmid was used to make infectious baculovirus stock in Sf9 (Spodoptera frugiperda, IPLB-Sf-21-AE) cells using the Bac-to-Bac system (Invitrogen). Then, HighFive (Trichoplusia ni, BTI-Tn-5B1-4) cells were transfected with baculovirus at 28 °C for 66 h to express the recombinant complex. Cells were washed with cold PBS buffer and frozen in liquid nitrogen until use.

All purification steps were performed in a 4 °C cold room. Cells were lysed in lysis buffer (25 mM HEPES pH 7.9 at 4 °C, 250 mM NaCl, 2 mM MgCl₂, 1 mM TCEP, 10 mM imidazole, 0.5% NP-40, 10% glycerol, DNase I and protease inhibitor cocktail) for 1 hour and sonicated. Debris was then removed by centrifugation at 28,000 × g for 35 min. The supernatant was incubated with Ni-NTA agarose resin (Qiagen) for 1 hour, and resin was washed with 10 column volumes (CV) of lysis buffer, 10 CV of high-salt wash buffer (25 mM HEPES pH 7.9 at 4 °C, 1 M NaCl, 2 mM MgCl₂, 1 mM TCEP, 0.01% NP-40, and 10% glycerol), and 20 CV of low-salt wash buffer (25 mM HEPES pH 7.9 at 4 °C, 150 mM NaCl, 2 mM MgCl₂, 1 mM TCEP, 30 mM imidazole, and 10% glycerol). Proteins were then eluted in elution buffer (25 mM HEPES pH 7.9 at 4 °C, 150 mM NaCl, 2 mM MgCl₂, 1 mM TCEP, 300 mM imidazole, and 10% glycerol) and dialyzed twice (1 h each) in buffer (25 mM HEPES pH 7.9 at 4 °C, 150 mM NaCl, 2 mM MgCl₂, 1 mM TCEP, and 10% glycerol) to remove imidazole. Proteins were incubated with TEV protease overnight after concentrating to 3–5 mg/ml. The AKTA-FPLC system was used for subsequent purification with a HiTrap Heparin HP column (Cytiva) and a Superose 6 increase 10/300 column (Cytiva). The heparin column was equilibrated with buffer I (20 mM HEPES, pH 7.9 at 4 °C, 150 mM NaCl, 2 mM MgCl₂, 1 mM TCEP, and 10% glycerol), and the sample was eluted with a linear gradient of buffer II (20 mM HEPES, pH 7.9 at 4 °C, 2 M NaCl, 2 mM MgCl₂, 1 mM TCEP, and 10% glycerol). The Superose 6 increase 10/300 column was equilibrated and run with the final storage buffer (25 mM HEPES, pH 7.9 at 4 °C, 150 mM KCl, 2 mM MgCl₂, 10% glycerol, and 1 mM TCEP). Protein complex was flash frozen in liquid nitrogen as single-use aliquots and stored at − 80 °C.

Preparations of five-subunit PRC2 and four-subunit PRC2 were similar to six-subunit PRC2 with several modifications. Four-subunit PRC2 contained full-length EED, SUZ12, RBAP48, and EZH2 that were all MBP-tagged. Five-subunit PRC2 had the subunits included in the four-subunit with the addition of MBP-tagged short isoform of AEBP2. Amylose agarose resin was used to replace the Ni-NTA resin in the initial affinity purification of the complexes. After washing with lysis buffer (same lysis buffer without imidazole), high-salt wash buffer, and low-salt wash buffer (same buffer without imidazole), PRC2 was eluted with elution buffer (low-salt wash buffer with 10 mM maltose). Without dialysis, the eluate was directly concentrated and then digested by Prescission protease at 4 °C overnight to remove MBP tags. Further purification involving the HiTrap Heparin HP column and Superose 6 increase 10/300 column was identical to six-subunit PRC2. The protein complex was flash-frozen in liquid nitrogen as single-use aliquots and stored at − 80 °C.

RNP complex assembly

All RNA oligos were purchased from Dharmacon Custom RNA Synthesis (Horizon), including HPLC purification service. TERRA, TERRA_mut, pUG-fold, Poly(A), Poly(U), Poly(C) and P10 RNAs were synthesized in two versions. RNA oligos without modification had 5’-hydroxyl ends for ³²P radiolabeling, while 5’-biotinylated RNAs were used for structural studies compatible with streptavidin-affinity EM grids. To promote RNA folding, all RNAs, including single-stranded RNAs, were heated at 95 °C for 2-3 min, snap-cooled on ice for 5 min, then refolded in RNP complex buffer (25 mM HEPES, pH 7.9 at 4 °C, 50 mM KCl, 2 mM MgCl₂, 10% glycerol, and 1 mM TCEP) at 37 °C for 20 min. PRC2 and S-adenosylhomocysteine (SAH) were added into the reaction at final concentrations of 600 nM and 40 µM, respectively, and the reaction was incubated at 30 °C for 30 min to assemble the RNP complex.

EM sample preparation

Quantifoil Au 1.2/1.3 grids were converted to streptavidin-affinity grids in-house using procedures described in refs. ^49,51. Grids were re-hydrated in EM preparation buffer I (25 mM HEPES pH 7.9 at 4 °C, 50 mM KCl, 2.5% glycerol, and 1 mM TCEP) at room temperature (RT) for 1 hour. After removing the remaining buffer, 4 µl of the assembled RNP complex was applied to the surface of the streptavidin-affinity grid. The grid was incubated for 5–10 min in a humidified chamber, washed with 40 µl of EM preparation buffer I, and then washed with 40 µl of EM preparation buffer II (25 mM HEPES pH 7.9 at 4 °C, 50 mM KCl, 2.5% glycerol, 0.01% NP-40, and 1 mM TCEP). After the washes, the buffer was wicked away using Whatman filter paper, and 4 µl of the EM preparation buffer II was added immediately. The grid was then transferred to the Leica EM GP2 plunge freezer, blotted for 2-3 s at 10 °C and 90% humidity, and then plunged into liquid ethane. Negative staining of the streptavidin-affinity grid followed the same protocol, but instead of a plunge freezer, five droplets of 40 µl uranyl formate (30 mg/mL) stain were used.

EM data collection and processing

Cryo-EM data were collected using a Titan Krios G3i equipped with a Thermo Fisher Falcon 4 direct-electron detector (DED) camera and a Selectris energy filter set with a 10-eV slit width. Data acquisition was performed using Thermo Fisher EPU at 130,000x magnification (0.97 Å/pixel) with a defocus range of − 1.9 to − 0.5 µm. Movies were collected in EER format with a total dose of 50 electrons per square angstrom (e^–/Ų) and an exposure time of 5.49 s corresponding to 1323 frames. Gain correction was applied during motion correction using Relion’s own implementation of the UCSF motioncor2 program. The same parameters were used for ± 20° tilted stage data collection.

Negative staining datasets were collected on a Tecnai F20 microscope operated at 200 kV, with a Gatan K3 direct detector, at a nominal magnification of 25,000x, corresponding to 1.449 Å per pixel. Datasets were collected using a dose of 40–60 e^–/Ų on streptavidin-affinity grids with 3 nm carbon supports on the back of the grids.

Data were processed in RELION 5⁸⁴. The movie frames were aligned using RELION’s own (CPU-based) implementation of the UCSF MotionCor2 program⁸⁵, and CTF parameters were fit using CTFFIND4⁸⁶. The background streptavidin lattice of each micrograph was subtracted using in-house scripts⁴⁹. TOPAZ automatic picking was trained on selected particles and applied to pick all micrographs. Initial models were generated within RELION from negative staining data and used as a reference for the first round of 3D classification. Later classifications used references from previous good classes. Subsequent processing steps included several runs of regular 3D classification with Blush regularization, particle subtraction built in RELION 5, and 3D classification without alignment (regularization parameter T = 24). The selected 120,658 particles were then re-extracted and subjected to per-particle defocus refinement, beam-tilt refinement, and 3D refinement with Blush regularization to generate the consensus map. Soft-edged masks of individual PRC2 protomers were applied in the multibody refinement^87,88 to improve map qualities. Local resolution estimation was performed in RELION 5 using the same soft, spherical masks used during refinement. Local resolution filtered maps were generated in CryoSPARC⁸⁹.

Model building

Individual PRC2 protomers were built using cryo-EM maps from the multibody refinement. The coordinates of the G-quadruplex RNA-bound PRC2 six-subunit complex (PDB: 8FYH) provided a starting model from which all the coordinates were adjusted and rebuilt in the new map using COOT⁹⁰. The model of each PRC2 protomer was subjected to global refinement and minimization in real space using PHENIX⁹¹. These were then subjected to manual inspection and adjustment in COOT, followed by refinement again in PHENIX. TERRA_mut RNA model was generated using a 10-nucleotide fragment (UGAGUGUGAG) from AlphaFold 3 prediction model 1 and then docked into our map for the position we designated as the RNA density. The cryo-EM density maps and the molecular graphics were prepared with Chimera and ChimeraX⁹².

Generating representative EM map from AlphaFold 3 predicted models

PRC2 subunit sequences and TERRA_mut RNA sequence were provided to the AlphaFold Server for structure prediction. The top four AlphaFold 3 models were modified in Chimera by ‘molmap’ command to generate corresponding maps with an 8 Å low-pass filter. All four maps were added to a combined map by the ‘vop add’ command. In parallel, the ‘Segger’ function was used to create surface models with one segmentation region for each individual map. Then, the combined map was subtracted four times, and only the density within every surface model was kept. Therefore, the final map retained overlapping regions without flexible areas that differed between predictions.

Electrophoretic mobility shift assay (EMSA)

RNA oligos were radiolabeled at 37 °C for 30 min using T4 polynucleotide kinase (NEB, M0201L). After labeling, the oligos were purified on a denaturing polyacrylamide gel. The bands were excised from the gel and eluted with TE, then precipitated with NaCl, glycogen, and EtOH. Pellets were resuspended in TE, and the counts of the oligos were determined by liquid scintillation counting. Radiolabeled oligos were heated, snap-cooled, and refolded in EMSA binding buffer (50 mM Tris-HCl, pH 7.5 at 25 °C, 100 mM KCl, 2.5 mM MgCl₂, 0.1 mM ZnCl₂, 2 mM 2-mercaptoethanol, 0.05 mg/ml BSA, 0.05 mg/mL yeast tRNA, and 5% glycerol). Next, stock PRC2 was diluted with EMSA binding buffer to a series of concentrations and mixed with the radiolabeled oligos. Binding was carried out at 30 °C for 30 min, and then samples were loaded onto a non-denaturing 1.0% agarose gel (Lonza SeaKem GTG agarose) buffered with 1X TBE. Gel electrophoresis was for 90 min at 66 V in a 4 °C cold room. A Hybond N⁺ membrane (Amersham, Fisher Scientific 45-000-927) and two sheets of Whatman 3 mm chromatography paper were put underneath the gel, then the assembled gel was vacuum dried for 60 min at 80 °C. Dried gels were exposed to phosphorimaging plates and scanned using a Amersham Typhoon (Cytiva) for signal acquisition. Gel analysis was carried out with ImageQuantTL v10.2 (Cytiva), and data were fitted to the nonlinear binding curve (specific binding with Hill slope) using Prism 10 software (GraphPad).

Native gel electrophoresis

8% polyacrylamide (29:1 ratio of acrylamide to bisacrylamide), 0.5X TBE, and 100 mM KCl were mixed before pouring into a 0.75 mm thickness gel cassette. RNA samples were radiolabeled and refolded as described above. After mixing with loading dye (0.5X TBE, 10% glycerol, and a trace amount of bromophenol blue and xylene cyanol), 1000 cpm of each RNA sample was loaded. Native gel was run at room temperature in running buffer (0.5X TBE and 100 mM KCl) at 50–60 volts to prevent any substantial increase in temperature. After electrophoresis, two sheets of Whatman 3-mm chromatography paper were put underneath the gel, and then the assembled gel was vacuum-dried for 45 min at 80 °C. Dried gels were exposed to phosphorimaging plates and scanned using a Amersham Typhoon (Cytiva) for signal acquisition.

Circular Dichroism (CD) melt assay

RNAs were dissolved in nuclease-free water at a concentration of 0.3 mg/mL and heated to 95 °C for 3 min, snap-cooled on ice for 5 min, then refolded in one of the CD buffers as indicated (10 mM Tris pH 7.5 at 25 °C, 100 mM KCl or 100 mM LiCl, and 2 mM MgCl₂) at 37 °C for 20 min. RNA fold and thermal stability were analyzed by circular dichroism spectroscopy using a Chirascan-Plus system (Applied Photophysics) equipped with a Quantum Northwest AP/CD250 Peltier temperature controller and an in-sample temperature probe (Applied Photophysics) using a 0.5-mm path-length cell. Spectral data were collected from 360 to 200 nm at 0.5-nm steps with 0.5-s integration time per point. Following buffer correction, spectra were smoothed using a 5-point Savitzky-Golay filter and are reported as mean residue molar ellipticity. Thermal unfolding was assessed by incrementally increasing the sample temperature from 15 °C to 95 °C in 5 °C steps, with a 5-min equilibration period preceding data collection at each temperature point.

Analytical size-exclusion chromatography

In a 50 μL reaction, 2 μM PRC2 (final concentration) and 2 μM refolded RNA (final concentration) were mixed with RNP complex buffer (25 mM HEPES pH 7.9 at 4 °C, 50 mM KCl, 2 mM MgCl₂, 10% glycerol, and 1 mM TCEP). The reaction was incubated at 30 °C for 30 min to complete RNP assembly and then injected into a Superose 6 increase 3.2/300 column (Cytiva) pre-equilibrated with the RNP complex buffer. The column was run at a flow rate of 0.02 ml/min, monitored by UV260 and UV280 detectors. Gel filtration standard (Bio-Rad, #1511901) was injected and run using the same protocol to estimate the molecular weight of monomeric and dimeric PRC2 complexes.

Fluorescence resonance energy transfer (FRET)

40 µL reactions were prepared in FRET Binding Buffer (50 mM Tris-Cl pH 7.5 at 25 °C, 100 mM KCl, 2.5 mM MgCl₂, 0.1 mM ZnCl₂, 0.1 mg/mL nonacetylated BSA, 5% glycerol, 2 mM 2-mercaptoethanol), with 200 nM each of Cy3-labeled TERRA or Poly(A) and Cy5-labeled Poly(C) or Poly(A) RNAs and 0-2.4 µM PRC2 protein. Control reactions were also performed with individual RNAs and with no ligand to rule out FRET-independent fluorescence effects. All reactions were incubated for 1 h at room temperature in a black, 384-well microplate (Corning, #3575), then analyzed with a TECAN Spark microplate reader. Fluorescence properties measured included Cy3 fluorescence and anisotropy (Ex = 512 ± 30 nm, Em = 567 ± 20 nm), Cy5 fluorescence and anisotropy (Ex = 20 ± 30 nm, Em = 675 ± 20 nm), and FRET (Ex = 512 ± 30 nm, Em = 675 ± 20 nm).

Anisotropy binding curve data were fit with Eq. 1, where A is anisotropy, [E_T] is total PRC2 concentration, n is Hill coefficient, and K_D is PRC2-ligand equilibrium dissociation constant.

$$A=\left(A_{max}-A_{min}\right)\frac{{\left[E_T\right]}^n}{{K_D}^n+{\left[E_T\right]}^n}+A_{min}$$ 1

FRET behavior under an independent binding model was predicted from binding curve data via Eq. 2, where F is relative FRET signal, [L_T] is ligand concentration (each, assuming Cy3- and Cy5-labeled ligands are equal), [E_T] is protein concentration, Kd₁ and n₁ are the Eq. 1 coefficients from the Cy3-labeled ligand binding curve data, and Kd₂ and n₂ are the Eq. 1 coefficients from the Cy5-labeled ligand binding curve data.

$$F=\frac{{\left[L_T\right]\left[E_T\right]}^{\left(n_1+n_2-1\right)}}{\left({{Kd}_1}^{n_1}+{\left[E_T\right]}^{n_1}\right)\left({{Kd}_2}^{n_2}+{\left[E_T\right]}^{n_2}\right)}$$ 2

RNA-competition EMSA

RNA oligos were ³²P-radiolabeled as described in the conventional EMSA method. Radiolabeled oligos were heated, snap-cooled, and refolded in EMSA binding buffer (50 mM Tris-HCl, pH 7.5 at 25 °C, 100 mM KCl, 2.5 mM MgCl₂, 0.1 mM ZnCl₂, 2 mM 2-mercaptoethanol, 0.05 mg/ml BSA, 0.05 mg/mL yeast tRNA, and 5% glycerol). Six-subunit PRC2 (50 nM final concentration) was added and incubated at 30 °C for 20 min to achieve RNP assembly. Then, non-radiolabeled competitor RNA was refolded and diluted with EMSA binding buffer to corresponding concentrations and mixed with the previous reactions. Competition was carried out at 30 °C for 30 min, and then samples were loaded onto a non-denaturing 1.0% agarose gel (Lonza SeaKem GTG agarose) buffered with 1X TBE. Gel electrophoresis, gel drying process and signal acquisition were the same as described for conventional EMSA.

DNA-RNA competition EMSA

60-bp double-stranded DNA was ³²P-radiolabeled as described above and cleaned through Micro Bio-Spin Columns with Bio-Gel P6 (Bio-Rad). Radiolabeled DNA was incubated with six-subunit PRC2 (200 nM final concentration) at 30 °C for 20 min in DNA binding buffer (25 mM HEPES, pH 7.5 at 25 °C, 25 mM KCl, 1 mM TCEP, 0.05 mg/ml BSA, 0.05 mg/mL yeast tRNA, and 5% glycerol). Then, non-radiolabeled competitor RNAs were refolded and diluted with DNA-binding buffer to corresponding concentrations and mixed with the previous reactions. Competition was carried out at 30 °C for 30 min, and then samples were loaded onto a non-denaturing 1.0% agarose gel (Lonza SeaKem GTG agarose) buffered with 1X TBE. Gel electrophoresis, gel drying process and signal acquisition were the same as described for conventional EMSA. Signal intensities were quantified by ImageQuantTL v10.2 (Cytiva), and 50% competition values were calculated by the equation (log(inhibitor) vs. response -- variable slope) built in the Prism 10 software (GraphPad).

Nucleosome-RNA competition assay

We prepared ³²P-radiolabeled trinucleosomes in-house by adding a small fraction of radiolabeled DNA into non-labeled DNA during trinucleosome assembly. DNA was 621 bp in total, including a 33 bp 5’ tail, three 143 bp Widom sequences, two 64 bp linkers between Widom sequences, and a 31 bp 3’ tail. An unmodified human histone octamer was purchased from The Histone Source (SKU: HOCT). For nucleosome-RNA competition, 600 nM PRC2, 100 nM labeled trinucleosome, and serial dilutions of refolded RNA were combined in RNP complex buffer (25 mM HEPES pH 7.9 at 4 °C, 50 mM KCl, 2 mM MgCl₂, 10% glycerol, and 1 mM TCEP) and incubated at 30 °C for 30 min. In this assay, unlike RNA-competition EMSA, PRC2 was not pre-incubated with labeled nucleosomes. All three components in the reactions were combined simultaneously. After incubation, samples were loaded onto 1% agarose gel (SeaKem GTG Agarose) buffered with 1X TBE and resolved at 66 V for 110 min in a 4 °C cold room. Gels were vacuum dried for 60 min at 80 °C. Dried gels were exposed to phosphorimaging plates, and signal acquisition was performed using a Amersham Typhoon (Cytiva).

Methyltransferase activity assay

Mononucleosomes were prepared in house by assembling 192 bp DNA (21 bp tail + 143 bp Widom sequence +28 bp tail) and unmodified human octamer (The Histone Source, SKU: HOCT). 600 nM PRC2 and serial dilutions of refolded RNAs were pre-incubated at 30 °C for 20 min to reach binding equilibrium and then assembled into methyltransferase reaction mix including 1X methyltransferase buffer (25 mM HEPES pH 7.9 at 4 °C, 50 mM KCl, 2 mM MgCl₂, 10% glycerol, and 1 mM TCEP), 0.1 mg/ml BSA, 1X protease inhibitor, 1 µl RNase inhibitor/20 µl solution, 10 µM ¹⁴C SAM (PerkinElmer), and 300 nM mononucleosome. The volume of each reaction was 20 µl. Methyltransferase reactions were incubated at 30 °C for 20 min prior to boiling with SDS loading buffer to inactivate PRC2. Proteins were separated through NuPAGE 4–12% gel (Invitrogen) by running at 180 V for 52 min. The gel was vacuum dried at 80 °C for 30 min and then exposed to phosphorimaging plates. Signal intensities were quantified by ImageQuantTL v10.2 (Cytiva), and IC₅₀ values were calculated by the equation (log(inhibitor) vs. response -- variable slope) built in the Prism 10 software (GraphPad).

UV crosslinking assay

PTBP1-binding RNA and PRC2-binding RNAs were radiolabeled as described above. All RNAs were heated to 95 °C for 2-3 min, snap-cooled on ice for 5 min, diluted to 20,000 cpm with RNP complex buffer (25 mM HEPES pH 7.9 at 4 °C, 50 mM KCl, 2 mM MgCl₂, 10% glycerol, and 1 mM TCEP), and refolded at 37 °C for 20 min. Then, 600 nM PTBP1 or 600 nM PRC2 was added to reactions with corresponding RNAs at a final volume of 110 µl and incubated at 30 °C for 30 min to reach binding equilibrium. Then, 20 µl droplets of each reaction were loaded onto a glass cover slide on top of a cooling metal in an ice basket, exposed directly inside a UV crosslinker (UVP, CL1000) with various crosslinking energies. The four different reactions with the same crosslinking energy were exposed simultaneously and then transferred to tubes containing an SDS loading buffer. The glass cover slide was replaced with a new one prior to the next round of crosslinking to avoid sample contamination. Proteins were separated through a NuPAGE 4-12% gel (Invitrogen) by running at 150 V for 60 min. Gel was vacuum dried at 80 °C for 30 min, and then exposed to phosphorimaging plates. Signal intensities were quantified by ImageQuantTL v10.2 (Cytiva) and plotted by Microsoft Excel.

Statistical analysis

All statistical analyses and definitions of replicates are described in the corresponding figure legends. Statistical tests were performed by an unpaired two-sided t test using the GraphPad t test calculator.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

J.S., V.K., and T.R.C. conceived the study, analyzed data and wrote the manuscript. J.S. and L.Y. carried out EM sample preparation, data collection and processing. J.S., L.Y., A.R.G., V.T., W.O.H., K.J.G., and A.H.E. performed biochemical experiments.

Peer review

Peer review information

Nature Communications thanks Sungchul Kim and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

Cryo-EM density maps and fitted models have been deposited in the Electron Microscopy Data Bank (EMD-46751, consensus map; EMD-46722, Body1 from multibody refinement; and EMD-46726, Body2 from multibody refinement) and the Protein Data Bank (PDB: 9DCH). Source data are deposited in Figshare (DOI: 10.6084/m9.figshare.30456503). Requests for reagents, plasmids and cell lines used in this study should be directed to the corresponding authors. Source data are provided in this paper.

Competing interests

T.R.C. is a scientific advisor for Eikon Therapeutics. The other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72294-y.