Long antisense non-coding RNAs and their role in transcription and oncogenesis

Abstract

Long non-coding RNAs are estimated to qualitatively represent ~98% of expressed transcripts in human cells, a large proportion of which is antisense to protein-coding and non-coding transcripts. Here we review evidence from several experimental systems that suggests long antisense non-coding RNAs are involved in the transcriptional regulation of gene expression by altering epigenetic states at both adjacent and distal loci. We also review the initial evidence for a role of endogenous long antisense non-coding RNAs in oncogenic cellular transformation.

Several recent studies have demonstrated pervasive antisense transcription in human cells.1–8 Many of these antisense RNAs are unspliced, lack a poly-adenylation signal and are generally more than 200 bp long.2,9 The transcriptional landscape becomes even more complex when intergenic and size-selected RNAs are assessed.10–12 Clearly, much more of the human genome is transcribed than we have previously appreciated. Estimates for human and murine cells suggest that ∼61–72% of all transcribed regions express long non-coding RNAs that are in the antisense orientation relative to adjacent protein-coding genes.5,13 There appears to be little conservation of non-coding RNAs across species.14 Most of these transcripts are not annotated; they might play a role in determining species-specific traits.15 Here we consider evidence that suggests a role of long, antisense noncoding RNAs in the regulation of transcription.

Antisense non-coding RNAs as regulators of gene transcription have emerged from early observations in prokaryotes where antisense RNAs were shown to regulate transcription.16 Several examples of human genes regulated by antisense RNAs had been documented by the mid-1990s.17 In human cells, antisense RNAs were found to suppress the expression of their sense counterparts.18 This mode of regulation was thought to result from secondary structures generated by the sense and antisense RNAs or from a block in translation due to sense-antisense pairing.16 While there were several plausible explanations for the regulation of sense RNAs by their antisense counterparts, there was little mechanistic information supporting these suppositions.16 Data generated over the last year have begun to provide insights into the mechanisms of transcriptional regulation mediated by long antisense non-coding RNAs. These mechanisms involve antisense RNA-directed remodeling of chromatin at target loci.

The importance of such epigenetic changes and their role in the regulation of gene expression has long been recognized.19,20 Transcriptional activation and transcriptional silencing are mediated and accompanied by changes in the chromatin.21 Common transcriptional silencing marks include methylation of genomic DNA at CpG sites and methylation of histones. The principal marks for actively transcribed chromatin are acetylation of histone 3 at specific sites. At gene promoters, these epigenetic changes have profound effects on gene expression, presumably because they affect the ability of RNA polymerase II and of transcription factors to localize to the promoter. Epigenetic marks can be passed on to daughter cells, and a particular epigenetic change can profoundly affect generations of daughter cells.22 However, it is not known how the molecular machinery that induces epigenetic changes is recruited to a specific genetic locus. In plants, small non-coding double stranded RNAs direct DNA methylation to loci showing sequence homology.23 Similar observations have been made in human cells with small interfering RNAs (siRNAs) designed to target gene promoters.24 Small non-coding RNAs, either siRNAs or antisense RNAs, can function in human cells to direct the enzymatic machineries for methylation of histone 3 at lysines 9 and 27 (H3K9 and H3K27) and for methylation of DNA to genomic loci showing sequence homology to the particular non-coding RNA. These activities are correlated with transcriptional gene silencing (TGS) of the targeted locus.25,26 The observations suggest that it is possible to affect gene transcription in a targeted manner. The effector molecules governing such epigenetic control of transcription remained unidentified.

Two key observations with small RNA-directed TGS in human cells shed some light on possible endogenous mechanisms: (i) Only the antisense strand of a promoter-targeted siRNA was required to direct TGS; and (ii) DNA methyltransferase 3a (DNMT3a) was essential for TGS and was localized to the targeted locus.27 These findings were strikingly similar to previous observations showing that, in human cells, long antisense non-coding RNAs direct epigenetic marks to imprinted loci.28,29 Furthermore, X chromosome inactivation in females, which involves Xist and its counterpart antisense RNA, Tsix, requires the localization of DNMT3a to targeted loci.30

The similarities between imprinted genes and TGS suggested a common endogenous pathway, possibly involving long antisense non-coding RNAs in the recruitment of chromatin-modifying machinery.26,27 The role of small non-coding RNAs as mediators of TGS might then be to usurp an endogenous pathway that includes long non-coding RNAs. Recent evidence supports this supposition and suggests that long non-coding RNAs mediate both locus- and allele-specific regulation of non-imprinted genes.31,32

Observations on tumor suppressor genes in human cells have identified long antisense non-coding RNAs as regulators of transcription. Overexpression of the particular non-coding RNA resulted in stable, epigenetic silencing of the sense mRNA counterpart at the promoter,31 whereas the loss of the long antisense non-coding RNA induced a significant increase in the expression of the sense mRNA counterpart and hence gene activation.32 These non-coding RNAs may function by recruiting epigenetic modulatory proteins to their homologous target loci.33 By localizing to the target gene, they can modify the local chromatin architecture in a manner that would affect transcription (Fig. 1). Recent observations also document transcriptional regulation by long antisense non-coding RNAs in S. cerevisiae,34,35 bacteria36 and plants.37 Thus yeast, bacteria, plants and humans share a siRNA-independent mechanism involving long non-coding RNAs in the control of transcription. These observations imply the widespread occurrence of an RNA-based transcriptional regulation that operates through long non-coding transcripts.

Among cancer-relevant genes, those encoding p21, E-cadherin,32 p15,31 p53,38 Myc,39 epidermal growth factor homology domain-1 (tie-1)40 and PU.141 have all been shown to be regulated to some extent by long non-coding RNAs. Several of the genes listed in Table 1 can be epigenetically silenced in cancer.42 It is tempting to speculate that such epigenetic silencing might involve the dysregulated production of long antisense non-coding RNA. This supposition leads to the question of what regulates the expression of long antisense non-coding RNAs. One possible explanation is that microRNAs (miRNAs) control the output of bi-directionally transcribed loci.43 In the case of E-cadherin, the miRNA miR373 was found to bind to the E-cadherin antisense non-coding RNA and to affect the expression of E-cadherin.32 Another possible explanation assigns a regulatory role to tiny RNAs (tiRNAs). These are predominantly sense-oriented and centered just downstream of transcriptional start sites44 apparently related to the position of the first intragenic nucleosome,45 and could conceivably obstruct or influence gene-specific effects of long antisense non-coding RNAs.

The observations reviewed in this article set the stage for more extensive investigations of the role of antisense non-coding RNAs in the control of gene expression. The abundance of non-coding transcription continues to pose a huge challenge to our understanding of the genome and its regulation.46 Determining the functions of long antisense non-coding RNAs is part of that challenge. Whatever the exact natural mechanisms involved, this system of endogenous RNA-mediated transcriptional regulation could be made subject to control by exogenous small RNAs. With appropriate constructs, targeted to the antisense non-coding RNAs or designed to mimic the antisense non-coding RNAs, it should be possible to either activate or suppress gene expression transcriptionally,26 with enormous practical implications.

Figures and Tables

Figure 1 Model for long antisense non-coding RNA-mediated transcriptional regulation. (A) Long antisense non-coding RNAs are generated at bidirectionally transcribed loci (antisense strand in red, sense strand in black). (B) The long antisense non-coding RNA is assumed to interact with regions of homology in a sense promoter-associated transcript.50 Alternatively, the long antisense non-coding RNA could interact with the plus strand of the unwound DNA. Such interactions are postulated to trigger the recruitment of chromatin-remodeling proteins DNA methyltransferase 3a (DNMT3a), Argonaute 1 (Ago-1), Enhancer of Zeste (Ezh2), Suv39 h1, histone deacetylase 1 (HDAC-1) and G9a. (C) These chromatin-remodeling complexes would induce changes of the local chromatin architecture, including methylation of DNA, specifically at loci targeted by the long antisense non-coding RNAs.

Table 1 Cancer-relevant genes that generate long antisense non-coding RNAs1

Gene	Protein	Reference
CDKN2B	Cyclin-dependent kinase inhibitor p15INK4b	31
CDKN2A	Cyclin-dependent kinase inhibitor p16INK4a	31
CDKN1A	Cyclin-dependent kinase inhibitor p21/WAF1	31, 32
p27KIP1	Cyclin-dependent kinase inhibitor p27	31
ARF	p14ARF MDM2 inhibitor	31
TP53	p53 tumor suppressor protein	31, 38
TP73	p73 tumor suppressor protein	31
p57KIP2	p57 tumor suppressor protein	31
TP63	p63 tumor suppressor protein	31
APC	adenomatous polyposis coli tumor suppressor protein	31
WT1	Wilms tumor protein1	31
BRCA1	breast cancer type 1 susceptibility protein	31
BRCA2	breast cancer type 2 susceptibility protein	31
MLH1	mutL homolog 1, mismatch repair protein	31
VHL	von Hippel-Lindau tumor suppressor	31
RB1	retinoblastoma 1 protein	31
DCC	deleted in colorectal carcinoma transmembrane receptor	31
DCL1	Dicer-like 1 protein	31
NF1	neurofibromin 1	31
PTEN	phosphatase and tensin homolog	31, 47, 48
CDH1	E-cadherin protein	31, 32
MYC	c-Myc oncoprotein	49

1 Several genes that play important roles in cancer produce long antisense transcripts. Some tumor suppressors on this list are transcriptionally silenced in cancer. This silencing may be mediated by long antisense non-coding RNAs.

Acknowledgements

Work of the authors is supported by National Institutes of Health grants R01 HL083473-02 and RO1AI084406-02 (K.V.M.), R01 CA078230-12 (P.K.V.) and P01 CA078045-10 (P.K.V.). This is manuscript 20611 of The Scripps Research Institute. We thank Art Riggs, John S. Mattick, Jonathan R. Hart and Minghao Sun for valuable discussions and suggestions.