Research ArticleBIOCHEMISTRY

Establishing RNA-RNA interactions remodels lncRNA structure and promotes PRC2 activity

See allHide authors and affiliations

Science Advances  14 Apr 2021:
Vol. 7, no. 16, eabc9191
DOI: 10.1126/sciadv.abc9191


Human Polycomb Repressive Complex 2 (PRC2) catalysis of histone H3 lysine 27 methylation at certain loci depends on long noncoding RNAs (lncRNAs). Yet, in apparent contradiction, RNA is a potent catalytic inhibitor of PRC2. Here, we show that intermolecular RNA-RNA interactions between the lncRNA HOTAIR and its targets can relieve RNA inhibition of PRC2. RNA bridging is promoted by heterogeneous nuclear ribonucleoprotein B1, which uses multiple protein domains to bind HOTAIR regions via multivalent protein-RNA interactions. Chemical probing demonstrates that establishing RNA-RNA interactions changes HOTAIR structure. Genome-wide HOTAIR/PRC2 activity occurs at genes whose transcripts can make favorable RNA-RNA interactions with HOTAIR. We demonstrate that RNA-RNA matches of HOTAIR with target gene RNAs can relieve the inhibitory effect of a single lncRNA for PRC2 activity after B1 dissociation. Our work highlights an intrinsic switch that allows PRC2 activity in specific RNA contexts, which could explain how many lncRNAs work with PRC2.


Chromatin regulation can depend on long noncoding RNA (lncRNA) transcripts (1, 2), although in many cases, the exact mechanism of action for the RNA is unclear. The most well-established example of lncRNA regulation of chromatin is via the Xist RNA that is required for silencing of one copy of the X chromosome in female mammals (3). Another more recent example of an lncRNA associated with chromatin-based silencing is the HOX antisense intergenic RNA (HOTAIR) transcript (4). Both of these lncRNAs are associated with the activity of the histone methyltransferase (HMTase) complex Polycomb Repressive Complex 2 (PRC2), which is involved in facultative heterochromatin formation (5). The chromatin of the inactive X chromosome and of HOTAIR-repressed genes is marked in an lncRNA-dependent manner by histone H3 lysine 27 trimethylation (H3K27me3), the product of PRC2. There is active debate in the field as to how these lncRNAs regulate PRC2 (6). Recent evidence favors a model where Xist “tunes” PRC2 activity only after PRC2 recruitment, while recruitment itself is mediated by other mechanisms (7). Transcription is also reduced by Xist and HOTAIR independent of PRC2 (8, 9), consistent with low levels of transcription promoting PRC2 activity (10). PRC2 binds to single-stranded RNA (ssRNA) with a preference of G-tracts and G-quadruplexes (11). The prevalence of G-tracts, especially at the 5′ end of genes (12), explains why PRC2 interacts with many pre-mRNAs (13), perhaps sampling nascent transcripts on chromatin (14) due to relatively fast on and off rates for RNA binding (15). When RNA binds to PRC2, the methyltransferase activity for nucleosomes is inhibited (16, 17), suggesting that some specific additional context is required for PRC2 activity when RNAs such as Xist or HOTAIR are associated at chromatin that is methylated by PRC2.

Specificity of PRC2 activity is multifaceted and differs depending on the organism and whether it initiates or maintains heterochromatin. In Drosophila, PRC2 is largely dependent on a small set of specific DNA binding proteins to recruit the complex for initiation of heterochromatin (de novo methylation) (5). Maintenance of H3K27me3 is facilitated by the ability of the EED subunit of PRC2 to recognize H3K27me3, recruiting and stimulating PRC2 at previously established heterochromatin regions (5). This mechanism is thought to play a role in spreading and maintenance of methylation, although DNA binding proteins are still essential in this maintenance mechanism as well (5). In humans, PRC2 recruitment can occur by multiple, sometimes overlapping, mechanisms including DNA binding proteins with lower specificity that work in combination with each other or other cofactors such as chromatin-associated lncRNA (5). The complex nature of PRC2 recruitment in mammals, often necessitating multiple molecular mechanisms for action, has made it difficult to establish basic rules for PRC2 catalytic activity, especially in de novo initiation of H3K27me3-triggered heterochromatin.

The finding that PRC2 binds with substantial affinity to nearly every naturally occurring lncRNA and mRNA tested (13, 16, 17), which then inhibits nucleosome methylation, has called into question the specificity of lncRNAs that were suggested to work with PRC2 to silence chromatin. However, these findings did help establish that the RNA being produced by an RNA polymerase on chromatin, nascent RNA, would be an intrinsic inhibitor of PRC2 activity (16), preventing initiation of heterochromatin at active genes. Tethering a high-affinity RNA substrate for PRC2 directly to chromatin in the nucleus actively antagonizes PRC2 activity at normal target genes (12), because of the higher affinity of PRC2 for single RNAs with G-tracts than for chromatin (15, 18). It has been proposed that the binding of PRC2 to nascent RNA may allow the complex to sample the landscape of the genome, searching for a context where PRC2 activity is promoted (13, 14, 19). This may occur through encounter of H3K27me3, guiding PRC2 to preestablished heterochromatin and allowing PRC2 activity (20). For de novo heterochromatin formation, it is less clear how the nascent RNA–inhibited PRC2 is activated. One model suggests that transcription shutdown is required for PRC2 activity (10). This model is supported by the demonstrated antagonism of PRC2 by histone modifications of highly active genes and nascent RNA (16, 21). While both Xist and HOTAIR can turn off transcription upstream of PRC2 activity (8, 9), the association of both of these lncRNAs with chromatin promotes H3K27me3 at lncRNA target loci. Therefore, even if the gene is repressed and little nascent RNA is present, the lncRNA is at the chromatin locus. Regardless of how PRC2 is recruited, Xist and HOTAIR RNAs inhibit PRC2. Therefore, lncRNA-induced transcriptional repression does not resolve the issue that an inhibitory chromatin-associated RNA is present at the site of PRC2 activity.

RNA and chromatin compete for PRC2 binding (15, 18). Specifically, the linker DNA between nucleosomes competes with RNA (15). At equimolar nucleotide concentrations, ssRNA wins the competition for PRC2 binding over chromatin (15), which explains the inhibitory effect of RNA for PRC2 activity. The affinity of PRC2 for a ssRNA is much higher than for a hairpin version of the same nucleotide content (11), suggesting a more favorable context for chromatin to compete. Within the PRC2 mechanism of sampling chromatin via nascent RNAs and associating with lncRNAs on chromatin, we hypothesize that there may be a context for double-stranded RNA (dsRNA) that may tip the balance from RNA to chromatin for productive PRC2 interaction.

dsRNA occurs in the nucleus, resulting from multiple mechanisms including intra- and intermolecular RNA-RNA interactions. LncRNAs can make intermolecular RNA-RNA interactions with different types of transcripts. For example, Xist pairs with its antisense RNA, TsiX, during X inactivation (22). We have previously shown that HOTAIR can interact directly with target gene RNA, such as Junctional Adhesion Molecule 2 (JAM2), through an imperfect RNA-RNA base-pairing match sequence (23). This specific matching region within the sequence of HOTAIR is predicted to have a propensity for stable RNA-RNA interactions with known target transcripts (24). The region is a “hotspot” for RNA-RNA interactions as identified by computational prediction querying HOTAIR against the entire mRNA transcriptome (25). This hotspot overlaps a region within HOTAIR that was found to be conserved across vertebrates (except teleosts) with potential for RNA-RNA interaction with a Homeobox D Cluster (HOXD) transcript (26). HOTAIR RNA-RNA interaction was first identified because of its association with the RNA binding protein (RBP), heterogeneous nuclear ribonucleoprotein B1 (hnRNP B1). hnRNP A2/B1 was found to regulate HOTAIR-dependent PRC2 activity in cells. Furthermore, the B1 isoform was found to bind preferentially to HOTAIR and its target transcripts over A2 (23). This protein has two tandem RNA recognition motif (RRM) domains that can associate in a head-to-tail dimer, binding two RNAs in an antiparallel nature (27), suggesting a mechanism to promote RNA-RNA base-pairing interactions. We have previously shown that B1 can promote RNA-RNA interactions between HOTAIR and an RNA from a target gene, JAM2, which suggested that RNA-RNA interactions have a role in chromatin regulation by PRC2 (23) (Fig. 1A), although the underlying mechanism behind this has remained unclear. LncRNAs such as Xist and HOTAIR can adopt favorable structured states (28, 29), presenting an additional challenge for any proposed intermolecular RNA matching where intramolecular interactions occur.

Fig. 1 HOTAIR intermolecular interaction is promoted by hnRNP B1.

(A) Model of B1-mediated HOTAIR RNA-RNA interactions with nascent target RNA leading to PRC2 activity and gene silencing. (B) In vitro RNA pulldown with MS2-tagged HOTAIR or Anti-luc with recombinant B1 or A2. “A2+” is 3× the concentration of A2. Minus MS2-MBP fusion protein pulldown was included to account for background bead binding. Western blot analysis was performed using A2/B1 antibody. RNA recovery was quantified by quantitative polymerase chain reaction (qPCR; n = 3). (C) Crystal structure of RRM domains of A2/B1 in complex with 10-mer RNA (yellow) (27). Two molecules of the tandem RRMs are shown in purple/green. Blue circles highlight the N terminus. Adapted from (27) ( (D) Schematic of RNA-RNA interaction assay: RAT-tagged JAM2 and HOTAIR in vitro transcriptions (IVTs) incubated ± recombinant hnRNP B1. JAM2 tethered by PP7 coat protein on magnetic beads. Recovery of HOTAIR by JAM2 pulldown was quantified by RT-qPCR and protein by Western blot. (E) Assays from (D) with HOTAIR or Anti-luc RNA ± hnRNP B1 or A2 (n = 6). (F) As in (E) with full-length HOTAIR, HOTAIR with the JAM2 interaction site deleted or mutated ± hnRNP B1 (n = 3). Error bars in (E) and (F) represent SDs. P values were determined using two-way analysis of variance (ANOVA) and two-tailed Student’s t tests. n.s., not significant; WT, wild type; Pol II, RNA Polymerase II.

In the current study, we gain mechanistic insight into how B1-mediated RNA-RNA interactions can modulate HOTAIR structure and function to promote PRC2 activity. We determine the necessary features of hnRNP B1 for HOTAIR interaction and identify the specific regions of HOTAIR that interact with B1. We use chemical probing of HOTAIR secondary structure to highlight the structural changes that occur when B1 and an RNA-RNA match engage with HOTAIR. We find that genome-wide HOTAIR-dependent PRC2 activity occurs at loci whose transcripts make more favorable RNA-RNA interactions with HOTAIR. Last, we demonstrate that specific intermolecular RNA-RNA interaction relieves the inhibitory nature of HOTAIR RNA for PRC2 methyltransferase activity on nucleosomes. By dissecting these molecular changes to RNA structure, we highlight a switch that can result in PRC2 recruitment and activation by an lncRNA on chromatin. This may be a general mechanism that applies to many contexts where RNA plays a role in potentiating PRC2 activity.


RNA-RNA interaction is directly promoted by hnRNP B1 but not the A2 isoform

We have previously identified hnRNP A2/B1 as an important component of HOTAIR-dependent PRC2 activity in breast cancer. We found that B1 was enriched preferentially in RNA pull-down assays with HOTAIR using nuclear extracts, and we subsequently demonstrated direct in vivo binding of hnRNP B1 to HOTAIR, over the highly abundant isoform hnRNP A2. In addition, B1 also bound preferentially to HOTAIR target transcripts, over A2 (23, 24). B1 differs from A2 by the inclusion of exon 2, which encodes 12 unique amino acids on the N terminus (Fig. 1B). To further profile the recognition mechanism of HOTAIR by hnRNP B1, we first tested whether the isoform preference, B1 over A2, that was observed in the nuclear extract pulldown was recapitulated with purified protein. Using recombinant A2 or B1 proteins, expressed in Escherichia coli, we performed in vitro HOTAIR pull-down assays and found that B1 binds preferentially to HOTAIR compared to A2 (Fig. 1B). Even a threefold increase in A2 concentration did not recapitulate the same level of binding as B1 to HOTAIR. Little to no binding was observed for B1 to equal amounts of the control noncoding RNA, of similar size and GC content, that corresponds to the antisense sequence of the luciferase mRNA (Anti-luc), which we have used previously as a control (23). We conclude from this that the B1-specific N-terminal domain (NTD) directly confers preferential binding to HOTAIR (Fig. 1B). The presumed position of the NTD, based on the N-terminal position in the A2 isoform crystal structure (27), would place the NTD in proximity to bound RNA (Fig. 1C). This suggests that the B1 NTD itself directly interacts with RNA as an extension of the RRM, to increase specificity, affinity, or both. Next, we evaluated B1 versus A2 in the in vitro RNA-RNA interaction assay that we have previously used to characterize matching with the HOTAIR target mRNA JAM2. Briefly, the RAT (RNA Affinity in Tandem)–tagged JAM2 match RNA fragment was incubated with full-length HOTAIR or Anti-luc control RNA in the presence or absence of hnRNP A2 or B1 and then affinity-purified via the RAT tag. The association of HOTAIR or Anti-luc with JAM2 was quantified by reverse transcription quantitative polymerase chain reaction (RT-qPCR) (Fig. 1D). Consistent with previous results, B1 was able to stimulate significant HOTAIR interaction with the target gene JAM2 RNA, and we found that this was not the case for A2. Moreover, B1 was able to promote a minimal amount of RNA interaction between JAM2 and the Anti-luc control RNA, suggesting that B1 can indirectly bridge two RNA molecules while they remain separated, not base-paired, as seen in the crystal structure for A2 (Fig. 1C).

B1 bridges RNAs in the absence of strong base-pairing potential

The crystal structure of the A2 tandem RRMs bound to RNA revealed a head-to-tail dimer in complex with two RNAs (27). Each RNA crosses the RRM of one protomer to the next, and the two RNAs are aligned in an antiparallel nature (Fig. 1C). Based off of this RNA arrangement, where the dimer binds two molecules through single-strand engagement by the RRMs, we asked whether the ability of B1 to promote interaction of HOTAIR with its targets is dependent on base pairing between the RNAs. We used the HOTAIR RNA-RNA interaction assay described above to test this. Matching of HOTAIR with its target does not require B1 in vitro, as we have previously shown (23). Addition of B1 in our previously published RNA matching assays only modestly promoted the interaction between the RNAs. To test whether B1 can promote interaction with RNAs that do not have strong complementarity, we reduced the concentration of the RNAs to promote more stringency and B1 dependence. Under these conditions, B1 stimulates HOTAIR interaction with the JAM2 target RNA roughly threefold (Fig. 1F), compared to no significant background association (fig. S1A). When the fairly extensive imperfect base-pairing interaction between JAM2 and HOTAIR is disrupted by deleting or mutating (changing to the complement base) the 64–nucleotide (nt) interaction region on HOTAIR, B1 is able to recover nearly the same level of RNA-RNA interaction as with wild-type HOTAIR (Fig. 1F). This result suggests that B1 can bridge two RNAs because of the strength of binding to those individual RNAs and the ability of the dimer to interact with each RNA at the same time. Complementarity of two RNAs at, or proximal to, the B1 bridge would subsequently promote intermolecular base pairing. The HOTAIR-JAM2 match is stable in the absence of B1, suggesting that, once formed, it would persist after B1 dissociation.

Single-nucleotide mapping of B1 ultraviolet cross-links to HOTAIR identifies major direct interactions

We used the eCLIP (enhanced cross-linking and immunoprecipitation) method with recombinant B1 and in vitro transcribed HOTAIR (in vitro eCLIP) to profile the interactions between the two (24). Briefly, ultraviolet (UV) cross-linking was performed on preincubated B1 and HOTAIR, and the sample was digested with a low amount of ribonuclease (RNase) to generate RNA fragments that could be reverse-transcribed and made into a sequencing library. The eCLIP protocol involves size selection and downstream ligation of one sequencing adapter to the 3′ end of the complementary DNA (cDNA) as part of the sequencing library preparation (Fig. 2A and fig. S1C). Because a base that remains cross-linked to an amino acid “scar” can prematurely terminate RT, we mapped the termination sites of the in vitro eCLIP sequencing (eCLIP-seq) to better refine the specific site of direct B1 interaction on HOTAIR. The in vitro eCLIP results of B1 binding to HOTAIR revealed multiple regions with high RT stops. We observed six main peaks of RT stops from five locations on HOTAIR in the size of ~25 to 100 nt (Fig. 2B). Four of these peaks fell within domain 1 of HOTAIR (nucleotides 1 to 530), suggesting that HOTAIR domain 1 is important for B1-mediated function (Fig. 2B). Controls with non–cross-linked protein and a cross-linked nonbinding protein yielded background levels of recovered RNA with poor mapping capability (fig. S1D), and RT of HOTAIR alone demonstrated that the B1 RT stops are specific (fig. S1E). We conclude from this that B1 binds to multiple distinct locations within domain 1 of HOTAIR. On the basis of other lower-frequency RT stops that still produce a peak, we suspect that additional secondary interactions are made, potentially facilitated by the proximity of RNAs in tertiary structure near the primary interaction sites.

Fig. 2 Examining hnRNP B1–specific interactions with the lncRNA HOTAIR using in vitro eCLIP mapping.

(A) Schematic of in vitro eCLIP experiments: recombinant B1 incubated with IVT HOTAIR. HOTAIR-B1 complexes were UV–cross-linked. RNA was fragmented with limited RNase A treatment, followed by in vitro eCLIP protocol [see Materials and Methods and (24)]. (B) Top: Mapping of RT termination events from HOTAIR-B1 in vitro eCLIP as a measure of direct protein cross-linking (HOTAIR domains alternately shaded violet). Termination sites normalized to read count with significant peaks were determined by values greater than 5000. Bottom: Zoom-in on domain 1 (1 to 530) for titration experiments highlights multiple B1 interaction sites (shaded in gray). (C) Top: Diagram of constructs including construct with all RGGs mutated (“5XRGG MUT”) and B1 glycine-rich domain deletion (“ΔGR”). Bottom: Assays as in Fig. 1B with recombinant truncated versions of the constructs depicted above. Western blot for A2/B1. Intensities should be compared to input, since the antibody recognizes the constructs differentially. Equal protein loading for samples is demonstrated by Coomassie gel with equal amounts of protein loaded as in each pulldown. Bar graph of HOTAIR recovery as percent input from each pulldown was quantified by qPCR (n = 2).

B1 C-terminal glycine-rich domain participates in HOTAIR engagement

The crystal structure of the A2 RRMs demonstrates a minimal complex for bridging of RNAs (Fig. 1C). We have also demonstrated that the B1 NTD is required for efficient HOTAIR engagement and promotion of RNA-RNA interactions (Fig. 1B). Although these interactions are important features in establishment of RNA-RNA interactions, we wished to also test other features of B1 that may participate in this activity. In addition to the NTD and tandem RRMs, B1 has a run of “RGG” motifs proximal to the second RRM domain, as well as a low-complexity glycine-rich C-terminal domain. We generated minimal RRM constructs for B1 and A2, an alanine substitution of all five RGGs (arginine-glycine-glycine motifs) in the full-length construct, and a construct with the C-terminal glycine-rich domain deleted. Equal amounts of protein loading and HOTAIR recovery are demonstrated by Coomassie gel and qPCR quantification. RNA pull-down experiments demonstrated a clear requirement for the C-terminal portion of B1; however, the RGG motifs were unnecessary, suggesting that the unstructured glycine-rich domain is required for tight binding to HOTAIR (Fig. 2C). The additional amino acids within glycine-rich domains have been proposed to influence the interplay of this region with other RBPs or RNA (30, 31).

Chemical probing highlights B1 interaction with HOTAIR

To further investigate mechanistically how RNA-RNA interactions are facilitated, we performed chemical probing experiments on HOTAIR domain 1 with the JAM2 RNA match (62 nt) and/or B1 (Fig. 3A). RNA was chemically modified using 1-methyl-7-nitroisatoic anhydride (1M7), which reacts with the 2′-hydroxyl of the RNA backbone to form adducts on accessible or flexible nucleotides, primarily in single-stranded regions of the RNA (Fig. 3A). Adduct formation was quantified using primer extension via RT, where RT stops equate to reactivity to the modifier. Because of the strong RT stop introduced when HOTAIR and JAM2 interact, we had to include JAM2 removal steps to the protocol to generate complete reactivity data (fig. S2). To analyze the capillary electrophoresis data, HiTRACE RiboKit (3236) was used and subsequently normalized using published methods (37). The normalized reactivity values were evaluated for each experimental condition: HOTAIR only, HOTAIR + JAM2, HOTAIR + B1, and HOTAIR + B1 + JAM2; for control conditions: HOTAIR + poly(A) and HOTAIR + BSA (bovine serum albumin) (Fig. 3, A and B, and fig. S4). We observed reproducible chemical reactivity patterns across all conditions (see error trace plotted in fig. S3 and data file S1). We were able to detect subtle and larger changes in regions of HOTAIR upon the addition of JAM2 RNA and/or B1, consistent with our previous data and predictions for establishment of RNA-RNA interactions (Fig. 3, B and C). We found that the addition of B1 reduced chemical probing reactivity in many regions of HOTAIR (Fig. 3, B and C). This decreased reactivity was highlighted at the B1 interaction sites identified from our in vitro eCLIP analysis (highlighted in blue; 141 to 172, 304 to 314, and 460 to 523 nt) (Fig. 3, D and E). These results support a model of multiple prominent B1 interactions with HOTAIR and indicate that the eCLIP and chemical probing data are in agreement. The extent of reduced reactivity that we observed may be explained by either (i) a broader direct influence of the protein on reactivity, perhaps through proximity in three-dimensional space, or (ii) a significant change to RNA structure induced by B1, perhaps increased secondary structure, leading to decreased 1M7 reactivity in regions of HOTAIR that B1 does not bind directly.

Fig. 3 Chemical probing of HOTAIR highlights B1 interactions.

(A) Diagram of the IVT HOTAIR domain 1 construct for chemical probing experiments. Schematic representation of 1M7 chemical probing of HOTAIR domain 1 with B1 and/or the JAM2 fragment (54 nt). 5′ SL, 5′ stem-loop. (B) Heatmap of normalized reactivity for HOTAIR only (HA only), HOTAIR + JAM2 (HAJ2), HOTAIR + B1 (HAB1), and HOTAIR + B1 + JAM2 (HAB1J2). White values represent nonreactive nucleotides, and red values represent more reactive nucleotides. (C) Line graphs of normalized reactivity for each condition. Light blue shaded boxes highlight the in vitro B1 eCLIP sites identified. Light red shaded box highlights the RNA-RNA interaction site. (D) Boxplots of normalized reactivity for all nucleotide positions, the JAM2 interaction region (245 to 306 nt), and B1 eCLIP-derived binding sites (141 to 172, 304 to 314, and 460 to 523 nt). Error bars represent SDs. P values were determined using one-way ANOVA multiple comparison tests between HA only compared to each variable condition (*P < 0.01, **P < 0.005, and ***P < 0.0001). (E) Bar graphs for normalized reactivity of HOTAIR only and HOTAIR + B1 at specific eCLIP B1 binding sites, as well as a minimally changed control region, by nucleotide.

Establishment of RNA-RNA interactions alters HOTAIR structure

We next generated a difference map of chemical probing conditions, to quantify reactivity changes compared to HOTAIR RNA alone (Fig. 4A). To highlight regions of change, we used a sliding-window analysis, trained using control conditions (see data file S4 and Materials and Methods), to identify regions of change above a control threshold value. We mapped these regions of change, according to each experimental condition, onto the previously determined secondary structure model of HOTAIR domain 1 (Fig. 4, B to E) (28). Analysis of HOTAIR + JAM2 demonstrated that JAM2 pairing with HOTAIR changes the 1M7 reactivity of HOTAIR in the regions of the secondary structure that surround the JAM2 base-pairing site (nucleotides 245 to 306) (Fig. 4B). This includes reduced reactivity across the base-pairing site, consistent with the strong complementarity of the intermolecular RNA base-pairing match. As noted above, the addition of B1 caused significant changes in reactivity to a majority of HOTAIR nucleotides, as evidenced by the many highlighted regions on the secondary structure (Fig. 4C). Notably, although the general trend of nucleotide reactivity was downward, B1 caused higher reactivity at specific regions of HOTAIR, suggesting that these regions are more exposed and single-stranded.

Fig. 4 Establishment of RNA-RNA interactions alters HOTAIR structure.

(A) Heatmap of HOTAIR reactivity changes upon addition of JAM2 (ΔJ2), hnRNP B1 (ΔB1), or both (ΔB1J2), compared to HOTAIR only. Red values become more reactive, white values do not change, and blue values become less reactive when JAM2, B1, or both are present. (B) Prominent regions of change mapped to the HOTAIR domain 1 secondary structure (28) (with permission) with either JAM2 RNA, (C) hnRNP B1 (highlighted in gray), or (D and E) both JAM2 + B1 using either control RNA threshold analysis (highlighted in light green) or control protein threshold analysis (highlighted in light purple). Regions of change were determined as described in Materials and Methods. In vitro eCLIP B1 binding sites are highlighted in blue, and the JAM2 RNA-RNA interaction site is highlighted in green.

The combination of both JAM2 and B1 also generated regions of altered reactivity for HOTAIR RNA, especially surrounding the JAM2 interaction site (Fig. 4, D and E). Because B1 alone changes reactivity for so many nucleotides of HOTAIR, we subtracted those changes from the JAM2/B1 changes to mask these (fig. S5). By doing this, we can see that the JAM2/B1 condition is significantly different from B1 alone, despite the large extent of changes with B1 alone. The profile of these B1-subtracted data does not resemble the JAM2 condition. This suggests that JAM2 is able to either synergize with, or counteract, the effects of B1 alone. This is further emphasized by subtracting both JAM2- and B1-alone conditions from the JAM2/B1 reactivity (fig. S5, bottom). This double subtraction resulted in significant changes that persist, which cannot be accounted for by additive effects of the individual conditions. Together, the chemical probing data clearly show that the steps in establishing RNA-RNA interactions do remodel HOTAIR structure and suggest that the B1 and the intermolecular RNA base pairing have individual and potentially synergistic effects on changing HOTAIR structure.

Conversion from ssRNA to dsRNA promotes PRC2 activity

It is not clear how PRC2 goes from an RNA-mediated inhibition state to an enzymatically active state in situations where an lncRNA such as Xist or HOTAIR is known to be present and no preexisting H3K27 methylation has occurred. PRC2 has been shown to have a strong affinity for ssRNA, but has a weaker affinity for duplex RNA, like that found in perfectly base-paired stem-loop structures (11). We analyzed RNA-RNA interaction predictions between HOTAIR and the entire transcriptome (24, 25) and compared them to previously identified HOTAIR-dependent PRC2 targets in a model of HOTAIR-overexpressing triple-negative breast cancer cells (23, 38). Genes that acquired new PRC2 activity when HOTAIR was overexpressed were biased toward those with transcripts that have more favorable RNA-RNA interaction propensity with HOTAIR (Fig. 5A). This led us to hypothesize that duplex RNA might provide the correct context for PRC2 enzymatic activity by allowing PRC2 to transfer to chromatin via lower-affinity binding of dsRNA. To determine this, we performed HMTase assays with single-stranded or double-stranded HOTAIR RNA and evaluated levels of H3K27me3. We first assessed optimal H3K27me3 activity in these assays using dinucleosome templates composed of two 601 sequences surrounding 40 base pairs (bp) of linker DNA (Fig. 5B), recombinant PRC2 complex (Fig. 5C), and the cofactor JARID2 (Fig. 5D). We next annealed HOTAIR RNA with a titration of reverse complement RNA to form perfect duplex RNA and introduced these into the HMTase assays (Fig. 5E). We observed a distinct reduction in H3K27me3 in the presence of ssRNA versus no RNA. As we increased the amount of reverse complement RNA, H3K27me3 levels increased in a manner concurrent with the formation of dsRNA (Fig. 5E). These results suggest that while an ssRNA inhibits PRC2 activity, formation of a duplex between the inhibitory RNA and its match is able to relieve this inhibition and promote PRC2-mediated H3K27me3.

Fig. 5 Duplex RNA promotes PRC2 activity.

(A) Histograms comparing the predicted minimum free energy for RNA-RNA interactions between HOTAIR and 40,740 RNAs across the transcriptome (gray) or between HOTAIR and 885 transcripts from genes that gain PRC2 activity when HOTAIR is overexpressed in breast cancer cells (red). Data are from (23, 25, 38). (B) Native gel of dinucleosomes reconstituted via salt dialysis using a DNA template containing two 601 sequences surrounding 40 bp of linker DNA. DNA and nucleosome samples were run on a 5% native polyacrylamide gel and stained with SYBR Gold. (C) Recombinant human PRC2 complex includes SUZ12, EZH2, EED, RBBP4, and AEBP2, analyzed by SDS–polyacrylamide gel electrophoresis (SDS-PAGE), and stained with Coomassie blue. (D) HMTase assay was performed with recombinant PRC2 complex, hexanucleosomes, S-adenosylmethionine with and without the cofactor JARID2 (amino acids 119 to 574). PRC2 activity was determined by SDS-PAGE followed by H3K27me3 and total H3 Western blot analysis. (E) Native 0.5× tris-borate-EDTA gel of RNA annealing titration with HOTAIR forward and reverse fragments to show formation of dsRNA. HMTase assay with annealed HOTAIR dsRNA titration analyzed by Western blot.

HOTAIR target RNA-RNA matches promote PRC2 activity

Fully duplex dsRNA was used above, which may be relevant for some lncRNA contexts, but HOTAIR makes imperfect matches with RNA targets. To begin to address whether HOTAIR imperfect RNA-RNA matches with target mRNA could also promote PRC2 catalytic activity, we evaluated the HOTAIR matches with endogenous targets JAM2 and HOXD10 (Fig. 6A). Consistent with previous results, HOTAIR alone was able to inhibit PRC2 activity. When increasing ratios of HOTAIR-JAM2 duplex were present, the match relieved the inhibitory effect of the single-stranded HOTAIR fragment alone, thereby stimulating H3K27me3 (Fig. 6B). Equal molar ratios of HOTAIR to JAM2 resulted in the highest levels of H3K27me3 promotion (Fig. 6C). Similar results were observed with the HOTAIR-HOXD10 duplex RNA match (Fig. 6D), with equal molar ratios of match RNA relieving PRC2 inhibition the most (Fig. 6E). Titration of JAM2 or HOXD RNA alone had no effect on PRC2 (fig. S6, A to C), highlighting that the duplex formed with HOTAIR RNA is the effector. In contrast, a control poly(A) RNA of similar size did not significantly relieve inhibition (Fig. 6F and fig. S6D).

Fig. 6 HOTAIR matching with target RNA promotes PRC2 activity.

(A) Native PAGE of RNA annealing titration with HOTAIR fragments and either JAM2 or HOXD10 matching RNAs. (B) RNA-RNA interaction between HOTAIR and JAM2 predicted by IntaRNA (50). HMTase assay performed with recombinant PRC2, dinucleosomes, HOTAIR fragment (62 nt), and JAM2 match (54 nt) titration. PRC2 activity was determined by H3K27me3 and total H3 by Western blot. (C) Quantification of (B) (n = 3). (D) IntaRNA result of HOTAIR and HOXD10. HMTase as in (B) with HOTAIR fragment (31 nt) HOXD10 match (37 nt) titration. (E) Quantification of (D) (n = 5). (F) As in (D), with poly(A) instead of HOXD10 (n = 4). (G) HMTase as above with HOTAIR domain 1 (nucleotides 1 to 530) and JAM2 match (62 nt) titration. (H) Quantification of (G) (n = 3). (I) As in (G), except using HOTAIR with JAM2 match site deleted “HOTAIR D1_del.” (J) As in (G), except with the addition of hnRNP B1. (K) Quantification of (J) (n = 3). (C, E, F, H, I, and K) Percent relief of inhibition is normalized to H3 signal and relative to no RNA and described as percent relief from HOTAIR-only reaction. Error bars represent SDs. P values were determined using unpaired t tests with a 95% confidence interval.

The experiments above used RNA fragments surrounding the matching regions. These fragments were similar in size to the model RNA substrates used in previous biophysical studies of RNA-PRC2 interaction (11). Next, we tested a much larger portion of HOTAIR RNA, using the HOTAIR domain 1 sequence (1 to 530 nt), to evaluate whether the JAM2 match was sufficient to relieve inhibition of PRC2 activity. We found that HOTAIR domain 1 was a more potent inhibitor of PRC2 by molar ratio, consistent with longer RNAs binding with higher affinity to PRC2. Notably, JAM2 was able to relieve PRC2 inhibition by this longer RNA (Fig. 6, G and H). However, when we substitute HOTAIR domain 1 with a version in which the JAM2 interaction site is deleted, we observe no relief of inhibition (Fig. 6I and fig. S6E). Together, these results demonstrate that the formation of dsRNA via HOTAIR matches with known genomic RNA targets, like JAM2 and HOXD10, is sufficient to promote H3K27me3 in vitro. This suggests that endogenous RNA-RNA interactions between lncRNA and target nascent transcripts could facilitate PRC2 catalytic activity for gene silencing at specific genomic locations (Fig. 5A).

Last, we addressed the contribution of B1 to PRC2 activity that is modulated by RNA-RNA interactions. Under identical conditions to HOTAIR domain 1 and a JAM2 titration (Fig. 6D), we found that addition of B1 did not allow JAM2 to relieve PRC2 inhibition (Fig. 6, J and K). This may be due to how B1 normally functions in a step-wise RNA-RNA interaction mechanism, where the protein facilitates RNA bridges but must dissociate before PRC2 inhibition can be relieved. We cannot rule out alternative possible mechanisms, including that B1 may first disrupt the preassembled HOTAIR-JAM2 interaction and then sequester JAM2. However, we have not observed B1 unwinding activity. We note that B1 does increase HOTAIR 1M7 reactivity in multiple regions, which persist when JAM2 matches to HOTAIR (Figs. 3 and 4). These exposed regions of HOTAIR that are induced by B1 may contribute to PRC2 inhibition until B1 dissociates, leaving the matched RNA state that facilitates relief of PRC2 inhibition by the lncRNA.


RNA binding to PRC2 inhibits its catalytic activity (16, 17). It has remained unclear how this inhibition is relieved in contexts where RNA is present in a region of chromatin that has no previously deposited H3K27 methylation, including lncRNA-associated loci and genes bearing nascent transcripts. On the basis of our previous observation that the HOTAIR lncRNA makes preferential interactions with hnRNP B1 (23), a multivalent RBP that promotes RNA-RNA interactions, we have further profiled the molecular basis of this mechanism and how it relates to the observation that HOTAIR can somehow promote PRC2 activity when overexpressed in cancers (38). We find that hnRNP B1 uses multiple domains to engage HOTAIR in a manner that can bridge it to a target gene RNA (Figs. 1 and 2). When HOTAIR matches with a target RNA, the lncRNA structure is remodeled by B1 and the matching RNA (Figs. 3 and 4). We find that HOTAIR-mediated PRC2 targets make more favorable RNA-RNA interaction with HOTAIR (Fig. 5A). In turn, the formation of duplex RNA between HOTAIR and its targets limits the ability of the lncRNA to inhibit PRC2 activity (Fig. 6). When B1 is still present, RNA matching is not sufficient to relieve PRC2 inhibition, suggesting that B1 must dissociate from the RNA for PRC2 activity to be promoted. PRC2 binds to many individual RNAs in the nucleus (5), including nascent transcripts, presumably in an inactive state. Our results suggest a model where an RNA can be a positive effector of de novo PRC2 activity in a context where RNA-RNA interactions relieve PRC2 inhibition on chromatin (Fig. 7).

Fig. 7 Model for HOTAIR-mediated chromatin silencing via intermolecular RNA-RNA interactions.

Binding of PRC2 to single-stranded regions of HOTAIR inhibits enzymatic activity. B1 promotes RNA-RNA matching of HOTAIR with target nascent RNA via bridging of the RNAs and conformational changes in the lncRNA to promote intermolecular base pairing. B1 dissociates from the RNAs, promoting a conformation that may reduce PRC2 affinity for those RNAs through formation of the dsRNA match, thereby increasing PRC2 interaction with chromatin, leading to H3K27me3 and transcriptional repression.

HOTAIR RNA engagement and RNA-RNA bridging by hnRNP B1

The crystal structure of the tandem RRMs of hnRNP A2/B1 (27) demonstrates a potential for two RNAs to be engaged by an A2 or B1 dimer in an antiparallel orientation. While the RRMs likely only bind ssRNA, the adjacent RNA sequences are in a favorable orientation for base pairing. This engagement is intermolecular in the crystal structure and likely explains the ability of B1 to bring two RNAs together, even when the base-pairing potential between them is limited (Fig. 1F). There is potential for the same mode of engagement to work intramolecularly as well, and this may underlie the multiple sites of direct interaction that we observe for B1 on HOTAIR (Fig. 2B) and the ability of the protein to reduce chemical reactivity of multiple regions of HOTAIR (Fig. 3). A possible explanation for why the B1 NTD promotes HOTAIR binding is because it promotes B1 dimerization. Further work is required to determine whether this is the case. The C-terminal domain of B1 is necessary for HOTAIR binding and thus promotion of RNA-RNA interactions (Fig. 2C). The C terminus of the related hnRNP A1 can bind RNA (30, 31) and includes an intrinsically disordered domain that has been shown to self-associate and phase separate at high concentrations (39). Whether all of these properties are important for the mechanism that we have characterized remains to be determined; however, the ability to self-associate, at least into a dimeric state, would potentially promote the RRMs of two monomers forming the head-to-tail conformation that can promote RNA-RNA interactions (Fig. 1C).

We find that B1 cannot be present with the matched RNAs for relief of PRC2 inhibition (Fig. 6, J and K). We interpret this result by placing the function of B1 upstream in an RNA-RNA interaction mechanism model that ultimately leads to catalysis of H3K27 trimethylation and heterochromatin formation (Fig. 7). In this model, B1 must dissociate before PRC2 activity is promoted. The requirement for a matchmaker protein to “get out of the way” has been proposed in other molecular matchmaker models (40) and is consistent with the match itself being the ultimate effector, rather than the protein. We acknowledge that this is a proposed model and that there may be alternative interpretations, requiring further studies.

RNA-RNA interactions promote PRC2 activity

The ability of an RNA with base-pairing potential to relieve the inhibition of PRC2 that is imposed by a single RNA binding event may apply beyond the HOTAIR mechanism. There are multiple examples of lncRNAs with intermolecular RNA-RNA interaction capability involved in PRC2 activity (2). For example, some imprinted loci such as Kcnq1 have antisense transcripts that induce PRC2 activity and repression occurring coincident with sense transcripts, present in an RNA “cloud” at the locus that is methylated by PRC2 (41). Xist also has a perfect complement antisense transcript, TsiX (41). In mice, Tsix transcription promotes PRC2 activity at the Xist promoter, coincident with the formation of Xist:TsiX dsRNA, to repress Xist expression on the active X chromosome (22, 42). This Xist:TsiX dsRNA does not inhibit PRC2 (17). These RNA matching capabilities may underlie how PRC2 can methylate chromatin in a cloud of an lncRNA that would otherwise repress methyltransferase activity. Further work will be needed to directly capture these RNA-RNA interactions in cells, for HOTAIR and other lncRNAs, using approaches such as 4′-aminomethyltrioxsalen (AMT)–based cross-linking [reviewed in (2, 43)]. In addition, recapitulating these events at a genomic locus in the cell in a manipulatable system, similar to what was previously done to demonstrate G-tract RNA inhibition of PRC2 (12), will be necessary to understand this mechanism in even more detail.

LncRNA matching may occur with the nascent transcript at the locus where PRC2 activity is promoted. Although nascent transcription from highly active genes is likely to “win out” by inhibiting PRC2 (13, 16), there are multiple pieces of evidence suggesting that PRC2 is active in the presence of nascent transcripts at lowly expressed genes. Xist and Kcnq1ot1 lncRNAs are both present with low levels of their antisense transcript (TsiX and the protein-coding gene Kcnq1, respectively) while PRC2 methylates the chromatin at these loci (22, 44). More generally, H3K27me3 deposition has been shown to occur in genes with moderate transcription activity (4547). In addition, building up paused RNA polymerase II with the drug DRB (5,6-dichloro-1-β-D-ribofuranosylbenzimidazole) leads to more accumulation of promoter H3K27me3 than does complete inhibition of transcription initiation (10). A recent study found that endonucleolytic cleavage of nascent transcripts by a Polycomb-associated enzyme complex is important for maintaining low expression of Polycomb-repressed genes, suggesting that nascent transcription persists even after PRC2 activity occurs (48). These results suggest that PRC2 can deal with the inhibitory effects of a nascent transcript to achieve H3K27 methylation. Intermolecular RNA-RNA interactions are one potential mechanism to achieve this. In addition, recent work has highlighted that, once H3K27me3 has been established, this modification can help further relieve inhibition of methyltransferase activity by RNA (20). Where the mechanisms mentioned above do not act, other mechanisms must exist to prevent nascent RNAs from inhibiting PRC2, such as additional RBP interactions with nascent RNA to mask it. Our findings fit into a model where the inhibitory effects of RNA on PRC2 catalytic activity can be overcome by specific intermolecular RNA-RNA interactions to promote de novo H3K27 methylation. This mechanism may operate with multiple lncRNA pathways in normal contexts and when aberrantly high lncRNA expression occurs in disease, such as with HOTAIR, that may drive improper PRC2 activity.


In vitro transcription of RNAs

In vitro transcription (IVT) of RNA was performed using the MEGAscript T7 kit (Thermo Fisher Scientific or Ambion), and reactions were incubated for 4 hours at 37°C. RNA was treated with Turbo DNase for 15 min, and RNA was purified with the RNeasy kit (QIAGEN) and visualized by 6% urea polyacrylamide gel or by agarose bleach gel.

In vitro RNA pull-down experiments

IVT 10× MS2-tagged RNA (15 nM; full-length HOTAIR or Anti-Luc) was rotated (end over end) at room temperature for 15 min with 80 nM recombinant hnRNP B1 or A2 in EMB 300 Buffer [10 mM Hepes (pH 7.9), 300 mM NaCl, 3 mM MgCl2, 0.5% NP-40, 10% glycerol, 0.1 mM phenylmethylsulfonyl fluoride (PMSF), and 0.5 mM dithiothreitol (DTT)], RNase inhibitor (New England Biolabs), and 20 μg of competitor yeast transfer RNA (Roche) in a total volume of 300 μl per sample. At the same time, 300 nM MS2-MBP was prebound to 20 μl of amylose resin (New England Biolabs) in EMB 300 Buffer, RNase inhibitor (New England Biolabs), and 1% BSA and rotated at room temperature for 15 min. The MS2-MBP amylose resin was then added to each IVT-hnRNP sample and incubated for an additional 15 min rotating at room temperature. Resin was washed 4× in 800 μl of EMB 300 Buffer, and then protein association was analyzed by Western blot using antibody for hnRNP A2/B1 (Abcam, #ab6102). In addition, 10% of each sample was used for RNA analysis, where RNA was isolated by phenol/chloroform extraction, purified by ethanol precipitation, and quantified by RT-qPCR.

Purification of PRC2 complex using baculovirus expression system

Human PRC2 was purified essentially as described (13), using individual pFastBac1 constructs of EZH2, SUZ12, EED, RBBP4, and AEBP2 with HMBP-PrS (6XHis, MBP, and PreScission Protease sequences) N-terminal tags (courtesy of C. Davidovich and T. Cech). Briefly, equal multiplicity of infection of each viral construct was used to infect Sf9 cells for 72 hours at 27°C. Cells were harvested in phosphate-buffered saline (PBS) and snap-frozen. Cells were thawed and resuspended in lysis buffer [10 mM tris-HCl (pH 7.5), 250 mM NaCl, 0.5% NP-40, 1 mM TCEP [tris(2-cardoxyethyl)phosphine], and 1× cOmplete Protease Inhibitor (Roche)], 20 ml per 500 ml of culture, and slowly rocked for 30 min at 4°C. All further steps were done at 4°C but never on ice. Lysate was clarified at 29,000 rcf for 30 min. Supernatant was incubated with 0.75 ml of amylose resin (New England Biolabs) for 2 hours. Sample was poured into a column, and resin was washed with 8 ml of lysis buffer, 12 ml of wash buffer 1 [10 mM tris-HCl (pH 7.5), 500 mM NaCl, and 1 mM TCEP], and 12 ml of wash buffer 2 [10 mM tris-HCl (pH 7.5), 150 mM NaCl, and 1 mM TCEP]. Protein was eluted with wash buffer 2 + 10 mM maltose. Fractions (0.5 ml) were collected until minimal protein was detected. Protein was concentrated ~10-fold. Sample was incubated with PreScission Protease (GE Healthcare) overnight. Full PRC2 complex was separated by size exclusion chromatography on a 24-ml Superose 6 column (GE Healthcare) in 10 mM tris-HCl (pH 7.5), 250 mM NaCl, and 1 mM TCEP. Five subunit complex–containing fractions were concentrated, glycerol was added to 10%, and sample was aliquoted and flash-frozen.

Recombinant protein purification in E. coli

Human hnRNP B1 (~37.4 kDa), A2 (~36 kDa), and B1 truncations were performed as previously described (23). Human GST-JARID2 (amino acids 119 to 574) was expressed in BL21 (DE3) CodonPlus E. coli cells overnight at 18°C. Cells were lysed on ice using lysozyme (2 mg/ml) and sonication in PBS (500 mM NaCl total), 1 mM EDTA, 1 mM EGTA, 1 mM PMSF, 0.5% Triton X-100, and 15 mM DTT. Sample was spun at 29,000 rcf at 4°C, and supernatant was incubated with rocking with Pierce Glutathione Agarose Resin (Thermo Fisher Scientific) for 2 to 3 hours. Resin was washed with at least 20 times volume of PBS (350 mM NaCl total), 1 mM EDTA, 1 mM EGTA, 0.1 mM PMSF, 0.5% Triton X-100, and 0.1 mM DTT. Protein was eluted in 50 mM tris-HCl (pH 8.0), 350 mM NaCl, 1 mM DTT, and 10 mM glutathione. Sample was dialyzed in 50 mM tris-HCl (pH 8.0), 350 mM NaCl, 1 mM DTT, and 5% glycerol and then aliquoted and flash-frozen.

In vitro RNA-RNA interaction assays

In vitro RNA-RNA interaction assays were performed as previously described (23) using 125 nM RAT-tagged JAM2 RNA fragment (IVT from gBlock, Integrated DNA Technologies) incubated with 5 nM IVT versions of full-length HOTAIR, HOTAIR with the JAM2 site deleted, HOTAIR with the JAM2 site mutated (to its complement base), or control RNA anti-luciferase (antisense to the luciferase mRNA) at equal amounts by nanogram (UV quantification). Protein additions included 50 nM recombinant hnRNP B1, hnRNP A2, and 500 nM PP7–protein A fusion protein purified from E. coli. The PP7 fusion protein is required to pull down RAT-tagged RNA with immunoglobulin G–conjugated Dynabeads, with minimal RNA recovery of both tagged and untagged RNA in the absence of the PP7 fusion protein (fig. S1A) with B1 binding contingent on JAM2 (fig. S1B).

RNA quantification by qRT-PCR

RT was performed using the cDNA High-Capacity Kit (Life Technologies). Standard curves using IVT JAM2, HOTAIR, or anti-luciferase were used to calculate amounts of RNA recovered. RNA inputs for experiments and standard curve samples were tested for integrity by agarose or acrylamide gel. RT-qPCR was performed using SYBR Green MasterMix (Takyon), with two qPCR replicates performed for each sample. Technical replicates were averaged before analysis of biological replicates.

In vitro eCLIP-seq

In vitro eCLIP-seq was performed as previously described (24). A total of 1.78 pmol of HOTAIR RNA and 8.9 or 17.8 pmol of recombinant hnRNP B1 [see (23) for sequence details] were incubated in 100 μl of RNA refolding buffer [20 mM Hepes-KOH (pH 7.9), 100 mM KCl, 3 mM MgCl2, 0.2 mM EDTA (pH 8.0), 20% glycerol, 0.5 mM PMSF, and 0.5 mM DTT] for 20 min at room temperature. The mixture was diluted to 250 μl in refolding buffer and UV–cross-linked twice in one well of a 24-well plate at 250 mJ and 254-nm wavelength, with mixing by pipette in between. B1-RNA cross-linked samples were treated with 0.1 ng of RNase A for 4 min at 37°C and 1200 rpm mixing and then stopped with 200 U of murine RNase inhibitor (New England Biolabs). Following this, the in vitro samples were subjected to end repair, adaptor ligation, SDS–polyacrylamide gel electrophoresis (SDS-PAGE) and transfer to nitrocellulose, and the remainder of the eCLIP-seq protocol and then sequenced multiplexed with other eCLIP-seq libraries as previously described (24). PCR products from different cycle number were analyzed by gel to avoid overamplification (fig. S1C). Control samples HOTAIR-B1 non–cross-linked, B1-RRMs cross-linked, and HOTAIR only (without gel and transfer steps) were included (fig. S1, D and E).

Chemical probing and analysis

Chemical probing procedure was similar to (28, 36). Specifically, 20 pmol of in vitro transcribed and purified HOTAIR RNA was incubated in 500 μl of reactions with equimolar amounts of JAM2 RNA, hnRNP B1, or both in RNA refolding buffer [50 mM Hepes (pH 7.4), 200 mM KCl, 5 mM MgCl2, and 0.1 mM EDTA] at room temperature for 20 min and then divided into two tubes, each containing 245 μl. The reaction was started by addition of 544 nmol of 1M7 (+) or of an equal amount of pure dimethyl sulfoxide as a control (−). Samples were incubated for 5 min at 37°C, and reactions were quenched with 5 μl of 0.5 M MES-NaOH. Samples with the JAM2 RNA underwent a JAM2 removal step where they were incubated with a JAM2 DNA complement oligo, heated at 94°C for 3 min to denature RNA, and slow-cooled at room temperature for 10 min. All chemically modified RNA was purified using TRIzol extraction + isopropanol precipitation and reverse-transcribed using 0.2 μM RNA with SuperScript III reverse transcriptase (Thermo Fisher Scientific) at 48°C for 1 hour using a fluorescently labeled primer (Integrated DNA Technologies): 5′-/5-6FAM/. Labeled DNA products were eluted in Hi-Di formamide–spiked GeneScan ROX 1000 size standard (Thermo Fisher Scientific). Samples were run on an Applied Biosystems 3500 XL capillary electrophoresis system, and the data were analyzed using HiTRACE RiboKit (3236) with MATLAB (MathWorks).

The HiTRACE normalized data, error, and replicate experiments (n ≥3) are contained in data file S1. The HiTRACE normalized data for each condition were subsequently normalized using published methods to better compare each condition on the same scale to prevent extremely reactive positions from dominating the data analysis (37). Specifically, outlier positions within the data were temporarily excluded from the data, and the remaining top 10% of the data were used to calculate a mean reactivity value used to normalize the data. The highly reactive data points were included in the data but represent extremely reactive positions. The interquartile range for each experiment was then multiplied by a constant of 6 (empirically chosen) where six times the interquartile range is considered an extremely reactive position. The error generated from the original HiTRACE analysis was carried through this normalization by converting the error from the HiTRACE analysis to percent error. The percent error for each nucleotide position was then multiplied by the normalized value reproducing the error for all positions in each experiment. The normalized values and corresponding error used for further analysis can be found in data file S2. Difference mapping of the data was completed by subtracting the HOTAIR-only mean-normalized data from the experimental mean-normalized data. The mean-normalized and difference data are represented in heatmap form using Morpheus (, and the raw values are reported in data file S3.

To determine a region of change, the absolute value of the difference data was smoothed by using a sliding-window mean for every 3-nt position along the region of interest. A threshold value was determined by calculating the mean of the smoothed data from normalized poly(A) control data minus the HOTAIR-only data. The absolute mean value and SD were calculated to be 0.067 and 0.049, respectively. A threshold value of 0.17 would represent the mean reactivity value plus two times the SD and was subsequently applied to the HOTAIR + JAM2 experiments. A threshold value of 0.32 was generated using the same procedure for the BSA control experiments (mean = 0.12, SD = 0.1) and applied to HOTAIR + B1 and HOTAIR + B1 + JAM2 experiments. The regions of change were defined as five consecutive nucleotides (JAM2 alone) or four consecutive nucleotides (B1 and JAM2/B1) above the threshold value where two consecutive flanking positions had values below the threshold value. The regions of change are located in data file S4. The identified regions of change were then displayed back onto the secondary structural model for HOTAIR (Fig. 4, B to E). All mathematical treatments for the normalized data, data subtractions for difference data, and defining the regions of change were performed using python in Jupyter Notebook. Graphs for the correlation plots of HOTAIR with the control data [BSA and poly(A)] and experimental samples (JAM2, B1, and JAM2 + B1) in fig. S4 and graphs for the normalized reactivity with error shading (fig. S3) were generated using python in Jupyter Notebook. Replicate information for each experiment is as follows: HOTAIR only (four replicates); HOTAIR with poly(A) control (four replicates); HOTAIR with BSA control (three replicates); HOTAIR with JAM2 experiment (five replicates); HOTAIR with hnRNP B1 experiment (three replicates); HOTAIR with JAM2 and B1 experiment (four replicates).

RNA-RNA interaction analysis of genome-wide PRC2 targets

Using a previously published database of computationally predicted interactions between human lncRNAs and the entire transcriptome (25), a list of 40,740 annotated mature transcripts and computational predictions for RNA-RNA interaction with HOTAIR was analyzed. This list was sorted by the predicted minimum free energy found among the interactions contained within each pair of RNA sequences. We compared the HOTAIR RNA-RNA interaction list with previously published HOTAIR-dependent PRC2 targets in MDA-MB-231 breast cancer cells identified by chromatin immunoprecipitation sequencing (23, 38), a combined list of 885 genes from multiple groups. Histograms displayed as a fraction of the total identified for each list are plotted together relative to the predicted minimum free energy, and a t test was performed comparing the two distributions.

Nucleosome reconstitution

Dinucleosomes were assembled using salt dialysis, as previously described (49). To generate a DNA template for chromatin reconstitution, a 343-bp PCR product consisting of 2× 601 positioning sequences separated by 40 bp of linker sequence was cloned into pUC57 vector backbone. PCR product was purified using a NucleoSpin DNA purification kit (Macherey-Nagel). Chromatin was reconstituted by salt dialysis: DNA template and human core histones were dialyzed 18 to 24 hours at 4°C from Hi salt buffer [10 mM tris-HCl (pH 7.6), 2 M NaCl, 1 mM EDTA, 0.05% NP-40, and 5 mM 2-Mercaptoethanol (BME)] to Lo salt buffer [10 mM tris-HCl (pH 7.6), 50 mM NaCl, 1 mM EDTA, 0.05% NP-40, and 5 mM BME]. Chromatin was dialyzed for an additional 1 hour in Lo salt buffer and concentrated using an Ultra-4 10K centrifugal filter device (Amicon) and stored at 4°C for no longer than 1 month.

RNA annealing

RNA pre-annealing was performed by heating RNA at 94°C for 4 min in annealing buffer [6 mM Hepes (pH 7.5), 60 mM KCl, and 1 mM MgCl2] and then slow cooling on bench for 40 min followed by placing the samples on ice.

HMTase assays

HMTase assays were performed in a total volume of 15 μl containing HMTase buffer [10 mM Hepes (pH 7.5), 2.5 mM MgCl2, 0.25 mM EDTA, 4% glycerol, and 0.1 mM DTT] with 75 μM S-adenosylmethionine (New England Biolabs), varying amounts of ssRNA and duplex RNA (see above), 600 nM JARID2, 360 nM dinucleosomes, and 600 to 660 nM recombinant human PRC2 complexes under the following conditions. The reaction mixture was incubated for 30 min at 30°C and stopped by adding 12 μl of Laemmli sample buffer (Bio-Rad). After HMT reactions, samples were incubated for 5 min at 95°C and separated on SDS-PAGE gels. Gels were then subjected to wet transfer (30% MeOH transfer buffer) of histones to 0.22-μm polyvinylidene difluoride membranes (Millipore), and protein was detected by Western blot analysis using primary αRb H3K27me3 antibody (Millipore, #07-449), secondary antibody (Bio-Rad, #170-6515), and H3-HRP (horseradish peroxidase) (Abcam, #ab21054). Similar experiments were performed, except that the total ribonucleotide concentrations of all RNAs used were kept constant (fig. S2).


Supplementary material for this article is available at

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: We thank K. Ragunathan, S. Ramachandran, and members of the Johnson Laboratory for helpful comments on the manuscript. We thank C. Davidovich and T. Cech for PRC2 baculovirus constructs. Funding: This work was supported by NIH grants R35GM119575 (to A.M.J.), R35GM118070 (to J.S.K.), and T32GM008730 and F31CA247343 (to J.T.R.); RNA Bioscience Initiative Scholar Awards (to M.M.B. and E.W.H.); and an RNA Bioscience Academic Enrichment Award (to C.B.) from the University of Colorado School of Medicine RNA Bioscience Initiative. We thank the University of Colorado Cancer Center Genomics Core and the Protein Production, Monoclonal Antibody, and Tissue Culture Shared Resource, both supported by NIH grant P30-CA46934, for technical support. Author contributions: M.M.B. and A.M.J. conceived the study. M.M.B., E.W.H., C.B., and S.K.W. performed experiments. M.M.B., E.W.H., and J.T.R. performed bioinformatic analysis. R.B. and A.M.G. designed and purified protein constructs and provided experimental advice. M.M.B., E.W.H., J.S.K., and A.M.J. wrote the manuscript. All authors reviewed and approved of the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Specific materials generated during this study are available upon request.

Stay Connected to Science Advances

Navigate This Article