Many human large intergenic noncoding RNAs associate with chromatin-modifying complexes and affect gene expression

Abstract

We recently showed that the mammalian genome encodes >1,000 large intergenic noncoding (linc)RNAs that are clearly conserved across mammals and, thus, functional. Gene expression patterns have implicated these lincRNAs in diverse biological processes, including cell-cycle regulation, immune surveillance, and embryonic stem cell pluripotency. However, the mechanism by which these lincRNAs function is unknown. Here, we expand the catalog of human lincRNAs to approximately 3,300 by analyzing chromatin-state maps of various human cell types. Inspired by the observation that the well-characterized lincRNA HOTAIR binds the polycomb repressive complex (PRC)2, we tested whether many lincRNAs are physically associated with PRC2. Remarkably, we observe that approximately 20% of lincRNAs expressed in various cell types are bound by PRC2, and that additional lincRNAs are bound by other chromatin-modifying complexes. Also, we show that siRNA-mediated depletion of certain lincRNAs associated with PRC2 leads to changes in gene expression, and that the up-regulated genes are enriched for those normally silenced by PRC2. We propose a model in which some lincRNAs guide chromatin-modifying complexes to specific genomic loci to regulate gene expression.

Fig. 1.

Intergenic K4-K36 domains in the human genome produce multiexonic noncoding RNAs. (A) Representative example of an intergenic K4-K36 domain for lincFOXF1. For each histone modification (K4me3, green; K36me3, blue), the results of ChIP-sequence experiments are plotted as the number of DNA fragments obtained by ChIP-sequence at each position divided by the average number across the genome. Intergenic K4-K36 domains were interrogated for presence of transcription by hybridizing RNA to DNA tiling arrays. The resulting RNA hybridization intensity (red) within each K4-K36 domain is plotted with respect to its genomic location. The start and stop of each exon, as determined by our RNA peak calling algorithm (SI Methods), is indicated by a black bar. Arrowheads indicate the orientation of transcription. (B) Sequence conservation scores (SI Methods) across 21 mammalian species indicates lincRNAs (blue) are much more conserved than neutrally evolving intronic sequences (red), although less so than protein-coding genes (green). For each lincRNA exon, protein-coding gene exon, and protein-coding gene intron, a conservation score was calculated and plotted along the x axis as a log-odd enrichment score (compared with random genomic regions of equivalent size). The cumulative number of exons with a given score or lower is represented on the y axis.

Fig. 2.

Numerous lincRNAs are physically associated with PRC2. Several examples of lincRNA exons (black box) that are enriched in RIP experiments relative to the IgG control in hFF (Left), hLF (Central), and HeLa (Right) cells; lincRNAs were enriched in RIP experiments performed with antibodies recognizing the chromatin-modifying complexes: PRC2 (blue), but not with antibodies recognizing the chromatin protein H3K27me3 (gray). Coprecipitated RNA for each antibody and for the respective control (IgG) was hybridized to the DNA tiling arrays. The hybridization values for each probe within a lincRNA exon are plotted as the log2 values for RIP hybirdization intensity divided by control (IgG) hybridization intensity.

Fig. 3.

Diversity and nuclear localization of chromatin associated lincRNAs. (A) Many lincRNAs, but not protein-coding mRNAs are physically associated with chromatin-modifying complexes. (Left) Pie chart representing the percentage of unique lincRNAs expressed in all 3 tested cell types (hFF, hLF, and HeLa) that are bound only to PRC2 (blue), only to CoREST (red), bound by both PRC2 and CoREST (yellow), and those not bound by either complex (black). (Right) Pie chart indicating the percentage of protein-coding genes (pink) reproducibly bound to PRC2 and or CoREST in all 3 cell types relative to the total number of expressed protein-coding genes (dark blue). (B) Subcellular localization analysis of lincRNAs by RNA FISH demonstrates localization of lincRNAs to the nucleus. Each panel represents the in situ hybridization of ≈40 fluorescently labeled DNA oligos with complementarity to the interrogated lincRNA. RNA FISH experiments were performed in male hFF for each represented lincRNA (XIST, HOTAIR, TUG-1, lincMKLN-1, lincFOXF1, and lincSFPQ), and also in female hLF for XIST (XX). White “speckles” indicate the subcellular localization of each lincRNA. The nuclear compartment is demarked by DAPI staining (purple).

Fig. 4.

Genes repressed by PRC2 associated lincRNA overlap with genes repressed by PRC2. (A) GSEA comparing the protein-coding genes that are up-regulated on depletion of a PRC2 bound lincRNA and those up-regulated on depletion of various components of PRC2. The black line represents the observed enrichment score profile of protein-coding genes in the lincRNA gene set to the PRC2 gene set. To represent the significance of the black line, we permuted the enrichment score profiles for 100 random (size matched) gene sets. The dark gray region indicates the 5th to the 95th percentile confidence region; thus, results above the dark gray region are significant at P < 0.05. The enrichment profiles for all lincRNAs tested were significant at P < 0.05, whereas as the enrichment profile for an unrelated protein depletion (YY-1) was not significant. The rank of each gene in the lincRNA gene set is indicated by tick marks (below each enrichment score plot) on a schematic color bar indicating levels of differential expression, up-regulation in red and down regulation in blue. (B) lincRNA TUG1 is transcriptionally regulated by p53 in response to DNA damage. The y axis indicates the log2 ratio of lincRNA TUG1 expression in p53 wild-type cells divided by the expression value in p53 knock-out cells. The x axis indicates time after induction of DNA damage. (C) Gene ontology (GO) enrichment analysis identified numerous cell-cycle regulation pathways that were specifically derepressed on knock down of lincRNA TUG1. The enrichment FDR is plotted as −log(FDR) on the x axis. Results are shown from knockdown experiments in hLFs (gray) and in hFFs (black). Dashed line denotes FDR < 0.05.