Transcription regulation by long non-coding RNAs: mechanisms and disease relevance

Abstract

Long non-coding RNAs (lncRNAs) outnumber protein-coding transcripts, but their functions remain largely unknown. In this Review, we discuss the emerging roles of lncRNAs in the control of gene transcription. Some of the best characterized lncRNAs have essential transcription cis-regulatory functions that cannot be easily accomplished by DNA-interacting transcription factors, such as XIST, which controls X-chromosome inactivation, or imprinted lncRNAs that direct allele-specific repression. A growing number of lncRNA transcription units, including CHASERR, PVT1 and HASTER (also known as HNF1A-AS1) act as transcription-stabilizing elements that fine-tune the activity of dosage-sensitive genes that encode transcription factors. Genetic experiments have shown that defects in such transcription stabilizers often cause severe phenotypes. Other lncRNAs, such as lincRNA-p21 (also known as Trp53cor1) and Maenli (Gm29348) contribute to local activation of gene transcription, whereas distinct lncRNAs influence gene transcription in trans. We discuss findings of lncRNAs that elicit a function through either activation of their transcription, transcript elongation and processing or the lncRNA molecule itself. We also discuss emerging evidence of lncRNA involvement in human diseases, and their potential as therapeutic targets.

Fig. 1 |. Mechanisms of transcription activation in cis by long non-coding RNAs.

a, The long non-coding RNA (lncRNA) Hottip is expressed from the HoxA locus and serves as a scaffold for the local recruitment of the histone methyltransferase complex comprising MLL1 (also known as KMT2A) and WDR5 to HoxA gene (A1–A13) sites of transcription. Consistent with local activity, RNAi-mediated depletion of Hottip preferentially affects Hottip proximal, compared with distal, HoxA genes (fading colour gradient). b, Local RNA abundance provides feedback on transcription initiation. Left: the Mediator complex, the histone-acetylation reader bromodomain-containing protein 4 (BRD4) and RNA polymerase II (Pol II) are present in low abundance at transcriptionally inactive promoter and enhancer elements. Middle: upon transcription initiation, nascent RNAs produced from promoter and enhancer regions nucleate the formation of a condensate, which increases the local concentration of transcription regulators, thereby causing a burst in transcription. Right: as transcription proceeds, the increase in local RNA abundance beyond a certain threshold generates electrostatic repulsive forces that disperse the transcriptional condensates, thereby ending the transcription burst. c, Many transcription factors (TF) have RNA-binding domains, which potentially interact with nascent transcripts, including of lncRNAs. These lncRNA–TF interactions could contribute to the targeting or the strength of association of the TFs to their genomic target sites by taking advantage of their pre-existing 3D proximity. d, lincRNA-p21 and its cis-activated target, the neighbouring gene Cdkn1a (encoding p21), are in 3D proximity, and are co-regulated by the TF p53. Transcription initiation of lincRNA-p21 is sufficient to enhance the expression of Cdkn1a by creating a scaffold for the recruitment of the Cdkn1a transcriptional co-activator heterogeneous nuclear ribonucleoprotein K (hnRNPK). e, Transcription elongation of the lncRNA Maenli increases local histone H3 trimethylated at lysine 4 (H3K4me3), which is a mark of transcriptionally active chromatin, and promotes the expression of the neighbouring gene En1. f, Transcription of a lncRNA in the Bcl11b locus named thymocyte differentiation factor (ThymoD) promotes the demethylation of CTCF-binding sites and, therefore, CTCF recruitment and chromatin reorganization. This process brings the ThymoD-associated enhancer region in proximity with their target, the promoter of Bcl11b, resulting in transcription activation. g, Pol II recruitment to enhancers and promoters blocks chromatin-loop extrusion and stabilizes the loops at a configuration that brings active enhancers in proximity of active promoters. Experimental degradation of Pol II leads to the formation of larger loops, extrusion of which is now limited by CTCF.

Fig. 2 |. Long non-coding RNAs as cis-acting transcription stabilizers.

a, Whereas enhancers promote cell type-specific gene activation and silencers prevent the expression of their target genes, transcription-stabilizing long non-coding RNAs (lncRNAs) act in cis to tune the transcription level of dosage-sensitive transcription-factor (TF) genes. b–g, Examples of lncRNAs (in red) that act as transcription stabilizers of adjacent genes, all of which encode transcription regulators (in blue). The lncRNAs are Halr1 (b), Hand2os1 (c), FENDRR (d), Flicr (e), HASTER (f) and CHASERR (g) — loss of function of all of these lncRNAs caused increased expression of the adjacent gene. HNF1 homeobox A (HNF1A) and chromodomain helicase DNA-binding protein 2 (CHD2) (blue circles) enhance the inhibitory effects of the adjacent lncRNAs, and therefore provide negative feedback. Homeobox A (HOXA1) and forkhead box F1 (FOXF1) proteins are positive regulators of the lncRNAs Halr1 and FENDRR, respectively, suggesting they could also form a negative feedback loop. Some transcription-stabilizing lncRNAs modulate their target genes in a signal-responsive manner; for example, interleukin-2 (IL-2) acts on Flicr (e) to reduce high Foxp3 expression levels in regulatory T cells. Two lncRNAs, Hand2os1 and Hdnr (c), restrict Hand2 expression. h,i, The promoters of transcription-stabilizing lncRNAs modulate interactions between their target TFs genes and local enhancers. h, Left: in pluripotent cells, Halr1 binds and sequesters proximal enhancers of Hoxa1, which dampens retinoic acid-induced expression of Hoxa1. Right: deletion of the Halr1 promoter increases enhancer–Hoxa1 interactions. HOXA1 (blue circles) binds to local enhancers and activates Halr1, which restrains Hoxa1 expression. Left and right in h depict retinoic acid-induced cells. i, The Haster promoter limits interactions between the Hnf1a promoter and intragenic enhancers. This effect is accentuated at high concentrations of HNF1A protein, thereby providing negative feedback on Hnf1a transcription. j, The active Pvt1 lncRNA promoter acts as a boundary element that associates with enhancers located within the Pvt1 gene body and limits access of the Myc promoter to these enhancers. Experimental inhibition of the transcription activity of the Pvt1 promoter through targeted promoter deletions or CRISPR inactivation (CRISPRi) leads to increased Myc promoter–enhancer engagement, high Myc transcription and increased cellular proliferation. The Pvt1 locus also harbours a p53-dependent isoform, Pvt1b, which downregulates Myc transcription during stress, decreases cell proliferation and increases cell senescence without apparent changes in Myc–enhancer contacts.

Fig. 3 |. Control of X-chromosome inactivation by long non-coding RNAs.

a, Transcription activation of mouse Xist. The X-inactivation centre (Xic) shows two topologically associating domains (TADs) in mouse cells. In one TAD (blue background), the long non-coding RNAs (lncRNAs) Linx and Tsix — antisense transcript of Xist — and the Tsix enhancers, termed Xite, are located. On the active X chromosome (Xa), Tsix transcription suppresses Xist transcription, whereas the Linx promoter acts across the TAD boundary to limit Xist expression in cis. In the other TAD (red background), the lncRNAs Xist, Jpx, Ftx and Xert are located. Xert enhancers, termed XertE, promote both Xert and Xist transcription on the inactive X chromosome (Xi). Following X-chromosome inactivation (XCI), Jpx and Ftx maintain Xist expression and accumulation at Xi. b, Similarities and differences between XIST regulation by JPX and FTX in human and mouse. In human, whereas FTX is not essential for XIST regulation, JPX transcription, but not the mature RNA, contributes to polymerase II (Pol II) loading and XIST transcription and accumulation. In mouse, Ftx transcription promotes Xist transcription, whereas mature Jpx transcript is responsible for Xist transcriptional activation and accumulation. c, XIST mediates transcriptional gene silencing at the X chromosome. XIST RNA highlighting its repeat regions A–F and showing the role of A-repeats in promoting the initial steps of gene silencing through SPEN-mediated and histone deacetylase (HDAC)-mediated histone deacetylation and RNA Pol II eviction; the role of B-repeats and C-repeats in heterochromatinization through recruitment of Polycomb repressive complex 1 (PRC1) and PRC2 downstream of heterogeneous nuclear ribonucleoprotein K (hnRNPK); and the role E-repeats in the CIP1-interacting zinc finger protein 1 (CIZ1)-dependent maintenance of XIST localization at Xi and in recruiting RNA-binding proteins to mediate the nuclear compartmentalization of Xi. H2AK119ub, histone H2A ubiquitylated at lysine 119; H3K27me3, histone H3 trimethylated at lysine 27; MATR3, matrin 3; PTBP1, polypyrimidine tract-binding protein 1; TDP-43, TAR DNA-binding protein 43.

Fig. 4 |. Long non-coding RNA-mediated allele-specific repression at imprinted loci.

The mouse imprinted Igf2r locus, which includes the long non-coding RNA (lncRNA) Airn. Maternally expressed (blue), paternally expressed (red), and biallelic-expressed (grey) genes are shown. The circle underneath Airn denotes a differentially methylated region, with a black filling indicating methylation of the maternal allele. Bottom right: occlusion and dislodgement models for transcription-mediated repression of Igf2r by Airn. Bottom left: Airn RNA-dependent recruitment of the gene-repressing chromatin modifier complexes Polycomb repressive complex 1 (PRC1), PRC2 and G9a, which repress the Airn-distal genes Slc22a2 and Slc22a3. Airn RNA is drawn as an unstructured molecule owing to lack of structural information. Pol II, polymerase II.

Fig. 5 |. Mechanisms of transcription regulation by trans-acting long non-coding RNAs.

a, Association of the long non-coding RNAs (lncRNAs) Charme, DIGIT and mammary tumour-associated RNA 25 (MaTAR25) with polypyrimidine tract-binding protein 1 (PTBP1)–matrin 3 (MATR3), bromodomain-containing protein 3 (BRD3) and purine-rich element-binding protein A (PURA)–PURB, respectively, drives condensate formation (blue background) and localization at target genes. This localization promotes the activation of broad developmental or disease-associated transcription programmes. b, The lncRNA Firre promotes inter-chromosomal contacts, which facilitates the co-regulation of a genes with shared functions in energy metabolism. c, Sequence complementarity between lncRNAs and one or many genomic regions enables targeting of the lncRNA to specific loci in trans through the formation of a DNA–DNA–(lnc)RNA triplex that involves DNA major groove Hoogstein base pairing. Various lncRNAs, including Fendrr, HOTAIR, HIF1A-AS1, Sarrah and others, have been shown to engage in triplex formation and exert positive or negative effects on target-gene expression. Pol II, polymerase II.

Fig. 6 |. Involvement of long non-coding RNAs in genetic diseases.

a, The human EN1 locus, which encodes a homeobox transcription factor (TF). EN1 harbours recessive coding mutations in an individual with limb and brain malformations, whereas far-upstream biallelic non-coding deletions (or a compound heterozygous deletion and insertion not shown here) cause dorsal-limb malformations. Maenli is a mouse long non-coding RNA (lncRNA) mapped to the orthologous minimal deleted region in humans. Deletion of the Maenli promoter, or insertion of polyadenylation signals of transcription termination, recapitulate limb malformations and lead to reduced En1 expression, whereas an inverted termination signal has no effect. b, Transcription stabilizers control dosage-sensitive TF genes. Small deviations in the expression levels of certain TFs can be caused by heterozygous loss-of-function mutations or duplications of TF genes, or by biallelic loss of function of the stabilizer lncRNA, causing abnormal cellular transcription with organismal phenotypes. In several examples, the defects in stabilizer lncRNAs and the dosage alterations of their target genes have the same phenotype (Table 1). c, An expression quantitative trait locus (eQTL), in which a single nucleotide variant (SNV) influences expression of a lncRNA (left). rs10000X is depicted as the identifier of a fictitious regulatory SNV that is causal for this eQTL. The two graphs on the right depict statistically significant P values of a group of adjacent SNVs for association with the presence of the autoimmune disease systemic lupus erythematosus (SLE) (top) or with a lncRNA eQTL detected in T cell lymphocytes, a cell type that is relevant to SLE (bottom). The co-localization of both sets of association P values means that the lncRNA is a plausible mediator of the disease association. Several hundred instances such as this have been identified, indicating that variation in lncRNA expression contributes to the susceptibility of common diseases. GWAS, genome-wide association studies.