Long non-coding RNAs direct the SWI/SNF complex to cell type-specific enhancers

Abstract

The coordination of chromatin remodeling is essential for DNA accessibility and gene expression control. The highly conserved and ubiquitously expressed SWItch/Sucrose Non-Fermentable (SWI/SNF) chromatin remodeling complex plays a central role in cell type- and context-dependent gene expression. Despite the absence of a defined DNA recognition motif, SWI/SNF binds lineage specific enhancers genome-wide where it actively maintains open chromatin state. It does so while retaining the ability to respond dynamically to cellular signals. However, the mechanisms that guide SWI/SNF to specific genomic targets have remained elusive. Here we demonstrate that trans-acting long non-coding RNAs (lncRNAs) direct the SWI/SNF complex to cell type-specific enhancers. SWI/SNF preferentially binds lncRNAs and these predominantly bind DNA targets in trans. Together they localize to enhancers, many of which are cell type-specific. Knockdown of SWI/SNF- and enhancer-bound lncRNAs causes the genome-wide redistribution of SWI/SNF away from enhancers and a concomitant differential expression of spatially connected target genes. These lncRNA-SWI/SNF-enhancer networks support an enhancer hub model of SWI/SNF genomic targeting. Our findings reveal that lncRNAs competitively recruit SWI/SNF, providing a specific and dynamic layer of control over chromatin accessibility, and reinforcing their role in mediating enhancer activity and gene expression.

Fig. 1. Trans-acting lncRNAs are enriched with SWI/SNF at cell type-specific enhancers.

a Schematic representation of the Red-C, RedChIP and CUT&RUN experimental workflow. HUVEC underwent crosslinking, followed by bridging of DNA and RNA ends with an oligonucleotide-adapter. BRG1 was pulled down, followed by a biotin pulldown of the RNA-DNA-bridge adapter interactions. Red-C served as a control for the global RNA-DNA background. BRG1 CUT&RUN provided high-confidence BRG1 DNA binding sites. TSS (Transcription Start Site), seq (sequencing), pAG-Mnase (protein A and G Micrococcal Nuclease). b, c Quantitative assessment of unique RNAs (b) and DNAs (c) involved in the interactions identified by Red-C and BRG1 RedChIP. d Unique RNA-DNA interactions identified by Red-C and BRG1 RedChIP, and corresponding fold-change, in interaction frequency bins. e Comparison of the number of unique RNA-DNA interactions density across increasing interaction frequency bins. f Frequency of both Red-C and RedChIP RNA-DNA interactions within BRG1 CUT&RUN peaks. Box plots are defined as follows: “Red-C False” minima: 0.1791, maxima: 10.7481, center: 1.7913, lower bound (25^th percentile): 0.5374, upper bound (75^th percentile): 4.6575; “Red-C True” minima: 0.1791, maxima: 15.7638, center: 3.4036, lower bound (25^th percentile): 1.0748, upper bound (75^th percentile): 6.9862; “RedChIP False” minima: 0.0837, maxima: 13.2273, center: 1.0046, lower bound (25^th percentile): 0.0837, upper bound (75^th percentile): 5.3579; “RedChIP True” minima: 0.0837, maxima: 21.8502, center: 2.9301, lower bound (25^th percentile): 0.3349, upper bound (75^th percentile): 8.9577. Whiskers extend from lower bound to minima and from upper bound to maxima. A Wilcoxon signed-rank test was performed within each condition (n = 1, Red-C and RedChIP). g Classification of BRG1 CUT&RUN peaks based on the absence (None) or presence of trans-acting or cis-acting RNAs. h Enrichment of RNA classes (trans, cis, mixed or none) across different DNA elements (intergenic, intronic and promoter) bound by BRG1. i Absolute number of trans-acting RNA-DNA interaction sites within HUVEC-specific features (intergenic enhancers, intronic enhancers and promoters). j Heatmap of HUVEC-specific features where trans-acting RNAs bind, with the classification of each feature in different cell types. k Visualization of selected BRG1-associated lncRNAs (PVT1, MALAT1 and MIR100HG) from RedChIP and their proximity to BRG1 CUT&RUN peaks. The red window superimposed onto the BRG1 CUT&RUN trace represents a 5 kb RNA binding site from RedChIP. Source data are provided as a Source Data file.

Fig. 2. SWI/SNF preferentially binds trans-acting lncRNAs.

a Correlation between BRG1 iCLIP coverage and RNA-seq coverage, normalized to RNA length. Reads per kilobase million (RPKM). b Analysis of RNA classes based on normalized iCLIP coverage. Log₂(iCLIP coverage/RNA-seq coverage) values are plotted to compare RNA classes (lncRNAs, mRNAs, snoRNAs, snRNAs, and miRNAs). Reads per kilobase million (RPKM). Box plots are defined as follows: “miRNA” minima: -4.8725, maxima: 3.2751, center: -0.5896, lower bound (25th percentile): -1.8241, upper bound (75th percentile): 0.2226; “snRNA” minima: -4.4778, maxima: 3.1295, center: -0.4113, lower bound (25th percentile): -1.8856, upper bound (75th percentile): 0.4326; “Protein coding” minima: -3.9667, maxima: 4.2052, center: 0.1668, lower bound (25th percentile): −0.9103, upper bound (75th percentile): 1.1378; “snoRNA” minima: −4.9670, maxima: 6.9218, center: 0.2799, lower bound (25th percentile): −1.0903, upper bound (75th percentile): 2.1851; “lncRNA” minima: −2.5777, maxima: 5.2470, center: 1.5063, lower bound (25th percentile): 0.3676, upper bound (75th percentile): 2.3317. Whiskers extend from lower bound to minima and from upper bound to maxima. c Proportion of RNA classes enriched in BRG1 iCLIP (counts per million > 5). d Statistical analysis comparing the expected (based on relative basal expression) versus observed frequency of BRG1 binding sites within lncRNAs and mRNAs. Fisher’s exact test was performed within each condition, n = 4 technical replicates. e Stratification to identify the top bound lncRNAs. Initial screen for lncRNA presence in Red-C, BRG1 RedChIP, BRG1 iCLIP and RNA-seq. LncRNAs were then ranked compared to a mean scaled value for all lncRNAs for their degree of iCLIP coverage (RPKM), their expression level (RPKM), normalized enrichment score (iCLIP normalized to expression and RNA length) and number of iCLIP binding sites. f Scaled value plots for the top 16 lncRNA candidates based on selection criteria compared to the mean scaled value for all identified lncRNAs. g Comparative analysis of trans- and cis-acting lncRNAs identified in BRG1 RedChIP and iCLIP datasets. h Visualization of selected lncRNAs (PVT1 and LINC00607) and their BRG1 iCLIP strand-specific binding profiles. Called iCLIP binding sites are displayed. Reads per kilobase million (RPKM). Source data are provided as a Source Data file.

Fig. 3. Knockdown of SWI/SNF-bound lncRNAs causes a global genomic redistribution of SWI/SNF.

a BRG1 CUT&RUN lost (blue) and gained (red) peaks following lncRNA knockdowns. Negative control siRNA (siCTL) compared to untransfected control (top left). siRNA lncRNA knockdowns compared to siCTL. b ChromHMM chromatin state analysis of BRG1 binding patterns after lncRNA knockdown. c–f Visualization of specific lncRNA knockdowns illustrating BRG1 binding changes. Example traces for MIR100HG (c), JPX (d), and PVT1 (e) knockdowns highlight reduced BRG1 binding at enhancers (indicated by purple rectangles), while increased BRG1 binding at enhancers is observed in LINC00342 knockdown (f). g Browser traces of BRG1 CUT&RUN after siRNA-mediated knockdown of MIR100HG and knockdown followed by overexpression rescue with a MIR100HG transcript. 5 different genomic regions annotated with ChromHMM are shown. siRNA control (siCTL). h Browser traces of BRG1 CUT&RUN after siRNA-mediated knockdown of JPX and knockdown followed by overexpression rescue with a JPX transcript. 5 different genomic regions annotated with ChromHMM are shown. siRNA control (siCTL). Source data are provided as a Source Data file.

Fig. 4. LncRNAs retain BRG1 at functional SWI/SNF-dependent enhancers.

a Volcano plots of differential ATAC-seq peaks after treatment with a PROTAC against BRG1 and BRM ATPases. 30, 60 and 240 minutes of PROTAC treatment are displayed. b BRG1 binding (CUT&RUN) and corresponding chromatin accessibility (ATAC-seq) at enhancers and promoters after PROTAC treatment. c Upset plot of total lost enhancers for each lncRNA knockdown and their overlap. d Enhancers lost with any lncRNA siRNA, PROTAC and both. e Number of DEG (Differentially Expressed Gene) promoters associated with enhancers perturbed by both siRNA of lncRNAs and PROTAC. f Number of DEGs after PROTAC that loop to enhancers lost after siRNA knockdown of individual lncRNAs. g Number of BRG1-dependent enhancers lost after LINC00607 CRISPR knockout (KO), overlapped with the differential expression (DE) of the connected genes after LINC00607 siRNA. h Browser visualization of an enhancer (right-most) sensitive to PROTAC (lost BRG1 CUT&RUN binding and diminished ATAC peak) and siRNA against LINC00607. P-E loops (Promoter-Enhancer loops). Source data are provided as a Source Data file.