Macro lncRNAs

Abstract

In the past ten years, long non-protein-coding RNAs (lncRNAs) have been shown to comprise a major part of the mammalian transcriptome and are predicted from their highly specific expression patterns, to play a role in regulating protein-coding gene expression in development and disease. Many lncRNAs particularly those lying in imprinted clusters, are predominantly unspliced “macro” transcripts that can also show a low level of splicing, and both the unspliced and spliced transcript have the potential to be functional. Three known imprinted macro lncRNAs have been shown to silence from three to ten genes in cis in imprinted gene clusters. We review here the potential for functional macro lncRNAs, defined here as “inefficiently-spliced lncRNAs” to play a wider cis-regulatory role in the mammalian genome. This potential has been underestimated by the inability of current RNA-seq technology to annotate unspliced macro lncRNAs.

Keywords: :

• RNA-seq
• genomic imprinting
• linc
• lnc
• macro
• ncRNAs

Introduction

The goal of the Human Genome Project to understand the genome and improve treatment of major human diseases is made more difficult by the subsequent demonstration of the complexity of the transcriptome. This has shown that protein-coding transcripts comprise a minority of the transcriptome, despite the finding that abundant alternative splicing¹ and RNA-editing² increase their diversity. In contrast, non-protein-coding (nc) RNAs whose expression patterns indicate a regulatory role, contribute the majority of the transcriptome.^3-6 Currently non-coding RNA nomenclature is inconsistent and ncRNAs are named according to size (e.g., small, short, micro, long) but may also be annotated by their interaction partners (piRNAs), cellular localization (snRNAs) or by their position and orientation relative to the nearest protein-coding gene (asRNAs). However, initiatives are underway to coordinate nomenclature⁷ and to combine the growing number of specialized databases to create a common international database “RNAcentral” for all ncRNA sequences.⁸

How can non-coding RNAs function in gene regulation?

Non-coding RNAs have the potential, because of their highly tissue, developmental and cancer-specific expression to play a role in regulating the expression of protein-coding genes during these processes. Regulatory functions for ncRNAs can be broadly classified as occurring in cis or in trans. Trans-acting ncRNAs could act at many different levels to influence protein-coding gene activity either by regulating transcription itself, or, at a post-transcriptional level by regulating diverse processes including mRNA splicing, ribosomal function, mRNA translation or protein activity. In mammals three major classes of small gene-regulatory ncRNAs are known that mostly appear to regulate gene expression post-transcriptionally: micro (miRNAs ~22 nt) that appear to function only in somatic cells of the post-implantation embryo, endogenous small interfering (endo-siRNAs 21 nt) whose function appears to be restricted to oocytes and ES cells, and Piwi-interacting (piRNAs 26–31 nt) that function in spermatogenesis (reviewed in ref. 9) and may also induce epigenetic transcriptional silencing effects.¹⁰ The recent “competing endogenous RNA” hypothesis proposes however, that all transcribed RNAs have the potential to titrate the cells miRNA population.^11,12 Mammals also have numerous additional regulatory small nuclear ncRNAs that play general roles in gene expression such as: short nucleolar snoRNAs (60–300 nt) that guide rRNA/tRNA chemical modifications,¹³ small nuclear ribonucleoprotein snrpnRNAs (~150 nt) involved in RNA splicing,¹⁴ the 7Sk RNA (272 bp) that negatively regulates the PTEFβ RNA polymerase CTD-kinase,¹⁵ and B2 RNA (178 nt) that negatively regulates RNA polymerase in heat-shocked mouse cells.¹⁶ Recently, a new and abundant class of noncoding RNA has been distinguished that is generally referred to as lnc (long non-coding) RNAs and defined as being longer than 200nt. One of the most studied subclasses are lincRNAs (large intergenic non-coding) that are classified by their intergenic position. The HOTAIR ncRNA was the prototype lincRNA in humans and was shown to have a role in regulating repressive chromatin at the HOXD gene cluster and in modulating the cancer epigenome,^17,18 however this function may not be conserved in mice.¹⁹ Many thousands of lincRNAs have now been identified,^20-25 that mostly appear to have trans-regulatory functions acting at the level of transcription,²⁶ making this the largest lncRNA subclass in the mouse and human genome.

In contrast to the large number of possible functions for trans-acting lncRNAs, cis-acting lncRNAs whose regulatory functions are restricted to genes on the same chromosome are likely to operate only at the level of transcription. However, this needs to be experimentally verified as trans-acting lncRNA may also affect genes on the same chromosome (see comment ²⁶). To date, the known functional cis-acting lncRNAs are all in the (very) long size class. One example is the 19kb long Xist (X-inactive specific transcript) lncRNA that induces epigenetic silencing of one of the two X chromosomes in female mammalian cells.^27,28 Other examples that silence small clusters of protein-coding genes on autosomes are the imprinted lncRNAs and the ANRIL lncRNA that silences the p15/INK4B tumor suppressor gene (see below). Imprinted lncRNAs have been identified at several imprinted gene clusters and are unusual as they show inefficient co-transcriptional splicing and polyadenylation and have been termed 'macro' lncRNAs to reflect that their main product is unspliced, in contrast to the fully-spliced Xist lncRNA and all known lincRNAs.^21,29,30

In this review, we focus on functional cis-acting macro lncRNAs and their role in disease and review three main areas: (1) imprinted cis-acting macro lncRNAs, their unusual transcript biology and how they differ from spliced lncRNAs, (2) non-imprinted cis-acting macro lncRNAs linked to disease and (3), methods available to test the functions of large data sets of newly annotated but still uncharacterized lncRNAs.

Macro lncRNAs Act as cis-silencers in Genomic Imprinting

Genomic imprinting is a form of epigenetic gene regulation whereby genes are repressed in a parent-of-origin specific manner. This process affects ~140 genes in mice, many of which form clusters of 2–12 genes that with some exceptions, are conserved in humans.^31-34 In each of the six well-studied imprinted gene clusters, at least one gene is a macro lncRNA.²⁹ Expression of three imprinted macro lncRNAs: Airn (118 kb), Kcnq1ot1 (91 kb) and Nespas (> 14 kb), has been shown to be required for imprinted silencing of genes respectively in the Igf2r, Kcnq1 and Gnas clusters.^35-37 The single imprint control element (ICE) in each of these clusters has been genetically defined by in vivo deletions, to harbor the respective macro lncRNA promoter. On the maternal chromosome, the ICE in these three clusters is modified by a DNA methylation imprint established in the oocyte, that silences the macro lncRNA promoter. On the paternal chromosome, the ICE is unmethylated allowing expression of the macro lncRNA, which represses flanking protein-coding genes to generate imprinted expression (Fig. 1). The methylated ICE on the maternal chromosome prevents macro lncRNA expression thereby allowing expression of the flanking protein-coding genes. This regulatory system is unique as the DNA methylation imprint on the ICE correlates with protein-coding gene expression, thus its role is to repress the expression of the cis-acting silencer, i.e., the macro lncRNA, on the maternal allele.

Figure 1. Macro lncRNAs silence imprinted genes in cis. (A-C) Genomic organization of the mouse Igf2r, Kcnq1 and Gnas imprinted gene clusters (not drawn to scale). Red boxes: imprinted expression from the maternal chromosome. Blue boxes: imprinted expression from the paternal chromosome. Genes that are bi-allelically expressed are depicted as black boxes. Open colored boxes: genes with multi-lineage imprinted expression. Colored boxed with black stripes: genes whose imprinted expression is restricted to the extra-embryonic lineage. Genes above the line are expressed from the (+) strand, genes below the line are expressed from the (-) strand. Arrows show transcriptional direction. Solid arrows indicate strong transcription whereas dashed arrows indicate weak transcription. Macro lncRNAs are shown as wavy lines. See key for further details. ICE, imprint control element

The Airn macro lncRNA silences the Igf2r imprinted gene cluster

The Igf2r imprinted gene cluster on mouse chromosome 17 (human chromosome 6) spans a 490kb region that includes six protein-coding genes and the 118kb long Airn macro lncRNA (Fig. 1A). Airn is expressed from its unmethylated CpG island promoter contained in the defined ICE on the paternal chromosome and represses in cis three of the six protein-coding genes in the cluster, of which only Igf2r is overlapped in antisense orientation by Airn. The non-overlapped genes Slc22a2 and Slc22a3 in contrast to Igf2r, show restricted imprinted expression only in extra-embryonic tissues such as placenta and visceral yolk sac (VYS).^38,39 Deletion of the Airn promoter resulted in a loss of imprinted expression in this cluster and Igf2r, Slc22a2 and Slc22a3 show bi-allelic expression.⁴⁰ To discriminate whether the promoter or Airn expression leads to repression of these protein-coding genes, a polyadenylation cassette was inserted 3kb downstream from the Airn transcript start site.³⁶ This strategy left the promoter and transcription initiation intact, but shortened the Airn RNA to 4% of its original length and this truncated lncRNA failed to silence Igf2r, Slc22a2 and Slc22a3. This demonstrated that macro lncRNA expression was necessary for silencing but did not discriminate if the lncRNA product or the act of its transcription played the key silencing role (see review ⁴¹). In humans, IGF2R shows polymorphic imprinted expression in some fetal tissues but biallelic expression in adult tissues.⁴² The human AIRN promoter has been identified, however, a clear correlation between AIRN expression and IGF2R repression has not yet been established.⁴³

The Kcnq1ot1 macro lncRNA silences the Kcnq1 imprinted gene cluster

The Kcnq1 imprinted gene cluster on mouse chromosome 7 (human chromosome 11) spans a 780kb region that contains 15 protein-coding genes and the ~90 kb Kcnq1ot1 macro lncRNA (Fig. 1B). On the paternal chromosome Kcnq1ot1 is expressed in antisense orientation to Kcnq1 but only overlaps internal introns and exons. The Kcnq1ot1 CpG island promoter also serves as the ICE of the cluster that is modified and repressed by a maternal-specific methylation imprint acquired during oocyte development. Kcnq1ot1 expression is necessary to silence 10 protein-coding genes in this cluster.³⁵ Four genes: Kcnq1, Cdkn1c, Phlda2 and Slc22a18 are ubiquitously silenced (i.e, in embryonic and extra-embryonic tissues) on the paternal chromosome from which Kcnq1ot1 is expressed. Six genes: Ascl2, Tspan32, Cd81, Tssc4, Nap1l4 and Osbpl5 only show imprinted expression in placental tissues.^42,44 Recent studies however, challenge the imprinted status of Tssc4, Nap1l4 and Osbpl5, since their high expression in the maternal decidua invalidates their analysis in mouse placenta.^39,45,46

The Nespas macro lncRNA silences the Gnas imprinted gene cluster

The Gnas imprinted gene cluster on mouse chromosome 2 (human chromosome 20) spans a ~100 kb region that contains the two maternally expressed protein-coding Gnas isoforms Nesp and Gnas. The paternal allele expresses one protein-coding Gnas-xl isoform plus two paternally expressed macro lncRNAs Nespas and Exon1a⁴⁷ (Fig. 1C). On the paternal chromosome, the Nespas macro lncRNA is expressed in antisense orientation and overlaps and represses the Nesp protein-coding isoform.³⁷ The maternal Nespas lncRNA promoter is modified and repressed by a maternal-specific methylation imprint acquired during oocyte development. Whether the other Gnas protein-coding isoforms in the cluster are also regulated by Nespas is not yet known.

Cis-silencing by imprinted macro lncRNAs

Despite the fact that the Airn, Kcnq1ot1 and Nespas imprinted macro lncRNAs are known to be necessary for repression of flanking protein-coding genes, the mechanism by which they act and how their influence is restricted to a small cluster of genes, is unknown. In particular, for lncRNAs such as Airn that produce a majority of unspliced and a minority of spliced products, it is unknown which variant controls its repressor function. One of the key questions not yet settled is whether the macro lncRNA product or the act of its transcription causes repression, or if both features are involved.⁴¹ In the Igf2r and Kcnq1 clusters, the Airn and Kcnq1ot1 repressor macro lncRNAs overlap only one of the genes that they silence. In addition, silencing of most of the non-overlapped genes is restricted to extra-embryonic tissues (i.e., the placenta and membranes of the developing embryo).³⁸ This could indicate that different mechanisms act downstream of the macro lncRNA to repress overlapped and non-overlapped genes and that these mechanisms have different levels of activity in embryo vs. extra-embryonic tissues. In placenta, a role for the lncRNA product has been shown for Airn that accumulates at the distant Slc22a3 promoter and recruits the G9A/EHMT2 histone methyltransferase, which is necessary for Slc22a3 repression.⁴⁸ However, Igf2r repression is not dependent on G9A/EHMT2 in placenta indicating that at least two mechanisms act downstream of Airn in this tissue. Silencing models based on Airn transcription interference in which the activity of promoters or enhancers is reduced by overlapping macro lncRNA transcription have been proposed, but none have yet been demonstrated.^29,41 Expression of the Kcnq1ot1 macro lncRNA correlates with accumulation of repressive histone H3K9me3 and H3K27me3 on the silenced flanking genes in placental but not in embryonic tissues.^49,50 However despite the demonstration that the responsible G9A/EHMT2 and PRC2 histone methyltransferases physically interact with Kcnq1ot1,⁵¹ mice lacking G9A/EHMT2 and PRC2 only show stochastic loss of imprinted expression of some genes in the placenta.^52,53 A role for the Kcnq1ot1 macro lncRNA product was recently dismissed based on a siRNA-mediated knockdown experiment that did not change imprinted expression of these genes in undifferentiated embryonic stem cells.⁵⁴ However, this may not be relevant since RNAi activity is restricted to the cytoplasm and Kcnq1ot1 is nuclear-localized.^51,55,56 Thus the key questions of how macro lncRNAs silence flanking genes in cis and if they depend on more than one downstream mechanism, remains to be determined.

The Unusual Transcript Biology of Macro lncRNAs

When the first mammalian imprinted genes Igf2r⁵⁷ and Igf2⁵⁸ were discovered in 1991, repression by lncRNAs was not expected despite the simultaneous discovery of the Xist lncRNA responsible for X chromosome inactivation.^27,28,59 The respective lncRNAs in these two clusters (Airn and H19), although markedly different in transcript biology and function, served as model lncRNAs for many years,⁶⁰ see Figure 2. We have used the term 'macro' ncRNA in several publications describing Airn, to reflect its unique biology that produces mainly unspliced transcripts.²⁹ This term is used to contrast inefficiently processed transcripts with fully-spliced lncRNAs such as H19 and Xist.⁶¹ In the past ten years studies have identified many thousands of lncRNAs that are currently defined as 'longer than 200 nt'–a definition that relates to the standard method used to harvest RNA that collects transcripts greater than this size.⁶² LncRNAs include large intergenic 'lincRNAs' that occupy an intergenic position as well as those with a sense or antisense overlap of a protein-coding gene.^6,20 Many lincRNAs have been shown by knockdown of their spliced product, to play a trans-acting gene regulatory role (see review ²⁶). New categories of spliced or unspliced lncRNAs with potential cis-regulatory functions have also been identified as divergent transcripts at promoters or enhancers.^63-65 However it is not yet clear if these transcripts are a byproduct of open chromatin or if they play a functional cis-acting role in regulating transcription (see discussion in ⁶⁶).

Figure 2. Macro lncRNAs are unusual transcripts that differ from fully-spliced lincRNAs. Left panel: Macro lncRNAs are inefficiently spliced lncRNAs. In contrast to lncRNAs, macro lncRNAs are ultra-long, lack sequence conservation and have a short half-life. They are also inefficiently spliced and imprecisely terminated at their 3′ end. Some macro lncRNAs such as Kcnq1ot1 are only found as unspliced transcripts, others such as Airn produce both unspliced and spliced variants. Only spliced Airn transcripts are exported to the cytoplasm. Macro lncRNAs could repress in two ways: through their lncRNA product or through a form of transcriptional interference particularly if they overlap a silenced gene. Right panel: LncRNAs are ncRNAs of more than 200 nt, which are currently defined only by their spliced product. On average they contain two to three exons and are approximately 1–3 kb in length. LncRNAs are generally considered to act in trans e.g., by binding and recruiting chromatin-modifying enzymes to regulate gene expression. In contrast to macro lncRNAs they are highly spliced, relatively stable and likely to be exported to the cytoplasm. Only transcriptional effects are displayed, for details of post-transcriptional effects see text: How can non-coding RNAs function in gene regulation?

The majority of lncRNAs not associated with enhancers and promoters are only annotated as fully-spliced products estimated to have an average length of ~1 kb with ~2.9 exons⁶⁷ or ~3.2 kb with 3.7 exons.²¹ It is not yet known if these annotated lncRNA genes also produce a significant proportion of unspliced transcripts as seen for imprinted macro lncRNAs. However, strategies based on assembling overlapping cDNAs,⁶⁸ or analysis of RNA-seq data⁶⁹ indicate that macro lncRNAs can be found outside of imprinted clusters. This is an important distinction when considering function, because if the bulk of the lncRNA is unspliced it is possible that the unspliced but not the spliced product is functional. Until this issue is further investigated we provisionally retain the use of 'macro' to describe the subset of lncRNAs that show the inefficient splicing behavior of Airn and Kcnq1ot1. The Airn and Kcnq1ot1 macro lncRNAs are also distinguished by a relatively short half-life, a high interspersed repeat content, and low sequence conservation.^51,60,70 One additional feature visible from genome-tiling-array and RNA-seq analysis^30,60,71 is that the Airn and Kcnq1ot1 macro lncRNAs show reduced RNA abundance toward the 3′ end of the gene, indicating imprecise termination. Together these features indicate that the macro lncRNA product may be less important than its transcription in regulating imprinted expression. If the Airn and Kcnq1ot1 macro lncRNA products are not important, the alternative is that they exert repressive effects by the act of their transcription. Models have been proposed whereby macro lncRNAs particularly those that overlap silenced genes, may induce a form of transcriptional interference either of the promoter or of essential enhancers.^41,72,73 These models do not exclude that the macro lncRNA product may also have a functional role for example, to repress non-overlapping genes.

An explanation of the unusual transcriptional biology of the Airn and Kcnq1ot1 macro lncRNA is lacking. One genetic element that may be relevant is the promoter as it was shown in fission yeast that promoters of some genes specify splicing.⁷⁴ However, replacement of the Airn promoter with the mouse Pgk1 promoter showed that the endogenous promoter does not control the low splicing capacity of the Airn lncRNA.⁷⁵ Both the Airn and Kcnq1ot1 macro lncRNAs are unusual in that their CpG island lies immediately downstream of the promoter and although they lack sequence similarities, they each contain a series of tandem direct repeats that have been suggested to guide epigenetic modifications such as the acquisition of the maternal-specific methylation imprint.⁷⁶ In support of this, the Airn tandem direct repeats have recently been shown to be necessary for the oocyte-specific DNA methylation imprint.⁷¹ The Kcnq1ot1 tandem direct repeats have only been partially deleted as two separate overlapping deletions, and although they are necessary to block somatic methylation of the normally unmethylated paternal ICE, no other functional role has so far been identified.^35,77

Different strategies are needed to detect macro lncRNAs and fully-spliced lncRNAs

The main differences between macro lincRNAs and fully-spliced lncRNAs are shown in Table 1. Today much attention is centered on RNA-seq strategies to identify lncRNAs and this highlights another difference: their ability to be detected by this technology. LncRNAs are reliably detected due to their spliced nature, while detection of continuous very long unspliced macro lncRNAs is challenging for current bioinformatic approaches. The approach of ab initio reconstruction of the transcriptome based on identifying transcription islands and connecting them by split-mapped reads, has allowed the detection of many thousands of novel spliced lncRNAs.^22,25,67 Bioinformatic pipelines have been developed including programs such as Scripture,²² Cufflinks⁷⁸ and the split-mapper TopHat,⁷⁹ which greatly increased the numbers and reliability of detected lncRNAs. The spliced nature of lncRNAs thus increases confidence in identifying novel transcripts but a bioinformatic pipeline to detect macro lncRNAs in RNA-seq data are lacking. Traditionally, whole-transcriptome analysis to investigate the expression of mRNAs is based on polyA-enriched^25,67 or polyA-amplified²⁴ RNA as input. However, genome-tiling-array data sets show that more than 40% of transcripts are non-polyadenylated⁸⁰ and although imprinted macro lncRNAs are polyadenylated, their extreme length does not allow capture by polyA-enrichment due to RNA fragmentation before the polyA-selection step. Therefore, an effective RNA-seq strategy to discover novel macro lncRNAs should include an efficient rRNA removal step to allow an unbiased view of the complete non-ribosomal transcriptome.³⁰ In addition, strand-specific RNA-seq is necessary to identify overlapping lncRNAs (e.g., Airn) and those lying inside annotated genes (e.g., Kcnq1ot1). Some studies have shown that macro lncRNAs such as Airn and Kcnq1ot1 are better visible in RNA-seq data sets produced by a pipeline that replaces polyA+ selection with rRNA depletion.^30,69 One unsolved problem is how to determine if ultra-long macro lncRNAs are one continuous transcript or composed of transcripts from clusters of smaller genes. For the Airn and Kcnq1ot1 macro lncRNAs continuous transcription was demonstrated by laborious experiments that demonstrated their repressor function,^35,36 since insertion of a polyA signal downstream of the promoter removed all downstream transcripts. However, for the large number of novel lncRNAs now being identified, a bioinformatic approach needs to be developed.

Cis-silencing Macro lncRNAs in Human Disease

To date, three cases are known in which the expression of an aberrant RNA or a macro lncRNA is causative for transcriptional silencing of an overlapped gene leading to a disease phenotype in humans. Additionally, several human imprinting disorders such as the Beckwith-Wiedemann syndrome,⁸¹ Prader-Willi syndrome,⁸² Angelman syndrome⁸² and Russell-Silver syndrome⁸³ are known in which deregulation of an imprinted gene or gene cluster causes severe disease phenotypes, but as involvement of imprinted lncRNAs has not been directly demonstrated, they will not be discussed further here.

A LUC7L-HBA2 fusion transcript silences HBA2 in cis

α-Thalassemia is an autosomal recessive blood disease arising from reduced functional α-globin production from the clustered HBA1 and HBA2 genes on chromosome 16. This leads to anemia by unbalancing the mix of α, β and δ globin chains that produce hemoglobin. The promoters of all the genes in the α-globin cluster (Fig. 3A) are associated with unmethylated CpG islands irrespective of expression.⁸⁴ A thalassemia patient was described to carry a unique 18kb deletion that includes HBA1, HBQ1 and the 3′ end of LUC7L.⁸⁵ Although the HBA2 gene remained structurally intact and the deletion did not remove any known cis-activating sequences, HBA2 was silent and its CpG island was densely methylated (Fig. 3A). A follow-up study was able to explain this finding by identifying the expression of an aberrant RNA resulting from the deletion, which creates a gene fusion of the widely expressed upstream LUC7L gene with HBA2.⁸⁶ Transcription of the 3′ truncated LUC7L from the antisense strand overlaps the HBA2 gene on the sense strand in antisense orientation (Fig. 3A). The fusion locus was also analyzed in a transgenic mouse model and in differentiating embryonic stem cells that reproduced the correlation between cis-expression of the aberrant antisense transcript, silencing HBA2 and de novo methylation of its CpG island. This fusion antisense transcript shares some but not all features of macro lncRNAs as described here. It is more than 40kb long, originates from the LUC7L promoter but due to the fact that the last 3 exons and transcription termination site are deleted, transcription extends at least into the HBA2 CpG island.⁸⁶ The majority of fusion transcripts consist of correctly spliced and protein-coding LUC7L mRNA.⁸⁶ In contrast, the 3′ part antisense to HBA2 may be more like a macro lncRNA, as it acts in cis (as the wild type HBA2 allele on the other parental chromosome is unaffected), and is most probably not spliced and not likely to be protein-coding.

Figure 3. Aberrant and macro lncRNAs associated with human disease. (A-C) Genomic organization of three loci expressing disease-associated aberrant or macro lncRNAs (not drawn to scale). Each locus is depicted in the wild type (black boxes) and at least one disease situation (gray boxes). Genes above the line are expressed from the (-) strand, genes below the line from the (+) strand. Arrows show transcriptional direction. Aberrant or macro lncRNAs are shown as wavy lines, unmapped 5′ ends are indicated by dashed wavy lines. In (C), two disease conferring SNPs are shown. See key for further details.

EPCAM-MSH2 fusion transcripts silence MSH2 in cis

A second example of transcription-mediated promoter methylation and gene silencing was found in patient families suffering from Lynch syndrome. This genetic condition confers high susceptibility to mainly colorectal cancer due to inactivation of mismatch repair genes: MLH1, MSH2, MSH6 or PMS2. Affected Dutch and Chinese families were shown to carry a heterozygous germline deletion of the last two or four exons of EPCAM and as a consequence transcription of EPCAM overlaps the neighboring MSH2 mismatch repair gene in sense orientation (Fig. 3B).⁸⁷ The MSH2 promoter is silent and methylated in cis, but only in EPCAM expressing tissues that express the aberrant fusion transcript. Further studies confirmed the 3′ end EPCAM deletions as a recurrent cause of Lynch syndrome.^88,89 Similar to the LUC7L-HBA2 fusion transcript, the EPCAM-MSH2 transcripts are likely to consist mainly of co-transcriptionally spliced EPCAM mRNA and 30kb or 3kb of RNA transcribed from the intergenic region, depending on the size of the deletion (Fig. 3B). This gene silencing by transcriptional read-through in sense direction is an additional mechanism to the above-mentioned silencing of HBA2 by antisense read-through. How exactly transcriptional read-through leads to silencing of the overlapped promoter and a gain of DNA methylation, is unknown in both cases. Taken together, these examples indicate that aberrant coding as well as macro lncRNAs are implicated in epigenetic cis-silencing of overlapped genes.

ANRIL regulates a cluster of tumor suppressor genes

Strong support that macro lncRNAs are also involved in epigenetic silencing of tumor suppressor genes comes from a number of studies showing that the ANRIL macro lncRNA regulates the expression of the p16^INK4A/p14^ARF/p15^INK4B tumor suppressor gene cluster in physiology and disease. Initially, ANRIL was discovered as a spliced 3.8kb lncRNA spanning a genomic region of 126kb that was expressed from a bidirectional promoter shared with p14^ARF (Fig. 3C).⁹⁰ This study showed coordinated transcriptional regulation of ANRIL and the p16^INK4A/p14^ARF/p15^INK4B tumor suppressor genes, based on correlation analysis of expression levels in 22 normal tissues and 12 breast tumor tissues. Later the p15AS RNA was described as a 35kb unspliced nuclear lncRNA (the whole sequence of which is contained in the defined ANRIL locus) that inversely correlates with p15^INK4B expression in acute myeloid and lymphoblastic leukemia.⁹¹ Two acute myeloid cell lines with highly-expressed p15AS and lowly-expressed p15^INK4B exhibited DNA hypermethylation and heterochromatin formation at the p15^INK4B CpG island promoter. However, induction of p15AS expression using an episomal report system shows that heterochromatin formation in cis rather than DNA methylation correlates with gene silencing.⁹¹ More recently, ANRIL has been shown to directly interact with the Polycomb repressive complex (PRC1) and thereby target repressive histone modification to all three tumor suppressor genes in cis, resulting in their epigenetic silencing.⁹² Sequence-specific structural features of ANRIL were suggested to bind PRC1, but whether they arise from spliced or unspliced ANRIL is unknown. A role in disease was suggested as upregulation of ANRIL and PRC1 components in prostate cancer tissues correlates with significant downregulation of p16^INK4A. ANRIL has also been shown to bind to PRC2 in vivo and to facilitate the recruitment of H3K27me3⁹³ to silence p15^INK4B, but not p16^INK4A or p14^ARF. Interestingly, recent genome-wide association studies identified the ANRIL gene as a genetic susceptibility locus associated with coronary artery disease, type II diabetes,⁹⁴ atherosclerosis,⁹⁵ and multiple cancers such as breast cancer,⁹⁶ basal cell carcinoma,⁹⁷ and glioma.⁹⁸ How exactly ANRIL associates with these quite varying disease phenotypes is not yet clear but may involve anti-proliferative mechanisms such as senescence and apoptosis that are regulated by telomere shortening and the activities of p53 and the p16^INK4A/p14^ARF/p15^INK4B tumor suppressor genes.⁹⁹ These cancer-protective pathways may also interfere with the replicative function of stem cell as they age.⁹⁹ The resulting altered regenerative capacity could explain some ANRIL-mediated age-associated diseases such as type II diabetes, atherosclerosis and cancer.¹⁰⁰ More than eight ANRIL splice variants are known¹⁰¹ and it was proposed that SNPs in the ANRIL gene body could regulate alternative splicing of ANRIL and thus the expression of p16^INK4A and p15^INK4B in diseases that rely on a certain replication potential.⁹⁹ The SNPs linked to coronary artery disease are located in a 58kb interval on human chromosome 9p21 (Fig. 3C) and the deletion of the orthologous 70kb region on mouse chromosome 7 showed that expression levels of p16^INK4A and p15^INK4B are markedly reduced in cis.¹⁰² Despite the existence of at least eight non-coding splice variants,¹⁰¹ the functions of unspliced ANRIL in regulating p15^INK4B in cis cannot be dismissed.⁹¹

Assigning Functions to Large lncRNA Data Sets

A major challenge for the field is that for most newly annotated lncRNAs only a bioinformatic classification is available, and little is known with respect to their RNA biology and cellular distribution and function. Since lncRNAs are also generally lowly expressed relative to protein-coding genes,^21,30 it is also not known if some could arise from 'opportunistic' non-specific RNA polymerase activity that reflects an open chromatin state (see discussion in ⁶⁶). Thus, it is currently difficult to devise the correct nomenclature as well as to model possible lncRNA functions. Some features are relatively simple to investigate, for example to test where lncRNAs are sub-localized in the cell, if they are efficiently spliced, if they have a long or short half-life, all of which would allow a more accurate approach to study their function. How to determine the functionality of lncRNAs is not trivial. It could be assumed that all expressed lncRNAs are potentially functional, thus the current lncRNA data sets could be considered to contain RNA genes selected for expression.^17,18,20-25 This contrasts with the discovery of imprinted macro lncRNAs from the last decade where despite a negative correlation between expression of imprinted protein-coding genes and the macro lncRNA, functionality was not accepted until demonstrated in mice carrying deleted lncRNA alleles.^35-37 This approach is unfeasible for large lncRNA data sets. One possibility for large data sets that is starting to be applied is to test a series of more easily implemented features and use this to predict a possibility of function. This would include the features discussed above i.e., splicing efficiency, length, cellular sub-localization, short or long half-life. Publically available genome-wide chromatin-immuno-precipitation data sets (http://genome.ucsc.edu/ENCODE/) could be mined to predict promoters. In addition, tissue, developmental and cancer-specific expression status can be characterized, and if a sufficient number of different tissues are analyzed it would also be possible to characterize lncRNA expression relative to protein-coding genes to predict possible cis or trans regulatory effects.⁶⁷ The cumulative score for all these features would be used to generate a 'functionality index' for the lncRNA. For imprinted macro lncRNAs, the functionality index could also determine differential-methylation status of the promoter and SNP analysis of RNA-seq data could determine parental allelic expression. This approach will identify families of lncRNAs sharing standardized parameters and subsequently allows predictions of distinct functions for a particular class of lncRNAs. An obstacle is clearly that in-depth knowledge of the investigated transcripts is needed to define suitable parameters. Recently, a study determined a catalog of > 30 properties including sequence, structural, expression and orthology features, for more than 8000 human lincRNAs and revealed specific subclasses. This study also demonstrated that lincRNAs show remarkable tissue-specific expression compared with coding genes, and are typically co-expressed with neighboring genes to the same extent as seen for pairs of neighboring protein-coding genes.⁶⁷ These large-scale bioinformatic studies are valuable to predict functions for lncRNA classes but detailed studies on individual lncRNAs are still important to understand mechanisms of action.

How to directly experimentally test for lncRNA function?

The chosen experimental strategy would involve an approach to deplete the lncRNA and assess the effects on gene expression and on cell biology. However, the chosen strategy would clearly depend on knowing some degree of lncRNA biology. For example if the chosen lncRNA is nuclear localized, RNAi cannot be deployed, as its activity is restricted to the cytoplasm.^55,56 However, the nuclear localized Xist lncRNAs can be depleted by siRNA micro-injected into the zygote nucleus, but RNAi activity against a nuclear localized lncRNA is thought to be limited to the brief window during nuclear membrane breakdown in the cell cycle.¹⁰³ Alternative depletion methods based on antisense DNA oligos and endogenous nuclear RNaseH activity have been shown to deplete nuclear lncRNAs.^56,104 If the chosen lncRNA is predominantly unspliced and is predicted to be cis-acting, there is a possibility that its transcription and not its lncRNA product is functional as suggested for the Airn macro lncRNA.⁴¹ As a consequence lncRNA depletion that acts post-transcriptionally will not reveal a function, and these lncRNAs would require removal by homologous recombination in embryonic stem cells as described above for imprinted macro lncRNAs, or by zinc finger nucleases that can be used in any cell type.¹⁰⁵ Many studies have successfully used RNAi to knock-down spliced lncRNAs,^{17,18,21,24,25,106} however it was not known if these were nuclear localized or exported to the cytoplasm. For example, a recent large-scale RNAi knockdown study which showed that 137 out of 147 lincRNAs had a significant effect on the expression of 100–200 genes in mouse embryonic stem cells, did not determine their nuclear or cytoplasmic localization.²¹ The method of choice to investigate lncRNA function could depend on knowledge of which class and which mechanism of action would be expected.

Conclusions

The field of lncRNAs is rapidly expanding and will be further accelerated as the depth of RNA-seq transcriptome studies increases.¹⁰⁷ However, we argue here that omitting the analysis of unspliced macro lncRNAs leads to a gap in our knowledge about this potential new layer of cis-regulatory information in the mammalian genome. The remarkable tissue-specificity of lncRNAs⁶⁷ strongly argues that they represent a functional layer of the transcriptome, rather than transcriptional noise (see discussion in ⁶⁶). The recent emphasis to identify non-annotated transcripts will certainly be switched to focus on functional aspects of these transcripts. The fact that many lncRNAs have been found to be highly expressed in cancer, has also led to them being termed 'cancer-specific' lncRNAs.^25,106 However, it would be unfavorable for an organism to encode a gene whose expression leads to disease. Therefore, such lncRNAs are more likely to be involved at a time point during development of the organism or differentiation of the respective tissue and only become deregulated in cancer. Therefore, dual studies to determine the physiological function and the disease function of non-coding transcripts will be essential in the future.

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

Financial Disclosure

PG is a recipient of a DOC Fellowship of the Austrian Academy of Sciences; the laboratory is supported by GEN-AU III Epigenetic Control Of Cell Identity (GZ200.141/1-VI/12009), and by the Austrian Science Fund SFB RNA Regulation Of The Transcriptome (F4302-B09).

Acknowledgments

We thank Aleksandra Kornienko and Florian Pauler for comments on the manuscript.