Mechanisms of long noncoding RNA function in development and disease

Molecular and genetic structure of LncRNAs

Like mRNAs, lncRNAs are transcribed by Polymerase II, mostly 5′-capped, polyadenylated and spliced, though on average they contain a lower number of exons than mRNAs and their expression level assessed across different tissues is lower [3, 4]. There are various algorithms calculating the coding probability based on the length of a potential open reading frame (ORF), the similarity of such an ORF to known protein-coding genes, frequency of in-frame nucleotide hexamers or other empirical sequence features [5]. In general, RNA transcripts containing short (<100 nt) non-conserved ORFs, which have no homology to known peptide sequences and do not match to peptides identified in mass spectrometry screens are considered noncoding [6]. Interestingly, the majority of lncRNAs are associated with ribosomes [7, 8], though they do not show the characteristic release of ribosomes [9] or the typical 3-nucleotide phasing corresponding to codons of an ORF [10, 11]. However, in rare instances, functional oligopeptides have been found to be translated from putative lncRNAs [12–14].

Deep-sequencing experiments revealed many examples of genes producing both protein-coding and noncoding transcripts by alternative splicing. However, so far, only very few reports demonstrate a functional role for both the noncoding RNA(s) and the protein(s) encoded by transcripts derived from the same gene [15, 16]. It is tempting to speculate that such dual usage of transcripts is more frequent than anticipated.

RNA molecules have the potential to form highly structured macromolecules by folding into double-stranded stems, single-stranded loops and bulges, which again can fold further into three-dimensional structures, allowing for the potential formation of complex shapes. So far, the structure of only a few RNAs has been experimentally determined using a combination of chemical assays, and by determining the accessibility of base-paired or single-stranded RNA by various RNases [17, 18]. However, these methods still have the limitation that they can only reveal the secondary, but not the tertiary (3D) structure. In addition, computational predictions are only beginning to provide reliable results, but program learning from experimental data might improve the predictions and more closely mirror experimental observations. This process is accelerated by recently developed techniques combined with high-throughput sequencing, such as SHAPE-MaP and icSHAPE [19, 20].

The similarity between mRNA-encoding and lncRNA genes is furthermore reflected by the chromatin signatures at the genomic regions from where they are transcribed. Their transcriptional start site is, in most cases, marked by the H3K4me3 histone modification and the transcribed region by the H3K36me3 mark, no matter if the lncRNA originates from its own promoter or from an enhancer. However, these histone marks are less prominent than observed at mRNA coding genes, whereas H3K4me1, a characteristic mark of enhancers, tends to be more prominent at genomic regions encoding lncRNAs [2, 21–23]. As expected for transcribed regions, lncRNA promoters correspond with DNaseI hypersensitive sites, which indicates accessible chromatin [24, 25].

So far, the most commonly used categorization of the highly heterogeneous class of lncRNAs is their position in the genome relative to protein-coding genes (Fig. 1). LncRNAs can be intergenic (lincRNAs) or divergently transcribed from the same promoter as a protein-coding gene (pancRNAs). LincRNAs expressed from a promoter (with high H3K4me3 to H3K4me1 ratio) are classified as promoter-associated lncRNAs (plncRNAs). However, lncRNAs can also be transcribed from enhancers and are then termed eRNAs. They are mostly transcribed in both directions, in contrast to enhancer-associated lncRNAs (elncRNAs), which are unidirectionally transcribed from a promoter with low H3K4me3 to H3K4me1 ratio [23]. Transcription of lncRNAs can also originate from within introns, or overlap with other transcripts in sense or antisense orientation [3]. Some lncRNAs are generated by backsplicing from introns of mRNAs or other lncRNAs and are thus circular (circRNAs) [26, 27].

Genome-wide sequencing of RNA species isolated from cytosolic or nuclear fractions of cells has shown that the majority of lncRNAs tend to be localized in the nucleus or are associated with chromatin, while a considerable fraction localizes to the cytoplasm [3, 28, 29]. The subcellular localization is a good indication of the putative function of a lncRNA, since in contrast to protein-coding mRNAs, lncRNAs can already function while transcription is occurring. In fact, many nuclear and, in particular, chromatin-retained lncRNAs co-regulate transcription and/or chromatin structure at or close to their site of transcription, i.e. in cis.

Conservation and evolutionary aspects of LncRNA

The genomic sequences of lncRNAs are, in general, less conserved than exons, but more than introns of protein-coding genes, pointing towards a rapid evolution with moderate constraints [21, 30]. In contrast to the actual sequence, splice sites of lncRNAs have been found to be more stable during evolution [31, 32]. Despite poor RNA sequence conservation, lncRNAs have frequently been identified across species in syntenic genomic regions [33, 34]. It is therefore quite likely that evolutionary conservation of lncRNAs is embodied by a conserved 3D structure, although this is difficult to assess with current methods. A further layer of conservation is functional conservation between lncRNAs playing equivalent roles in particular biological settings [35].

A different explanation for low conservation of lncRNAs is the increasing number of such transcripts in increasingly complex species. This finding, in combination with the observed highly tissue-specific expression of many lncRNAs, suggests that lncRNAs might be key molecules promoting species-specific features and organ complexity [36–38]. Thus, lncRNAs very likely have played important roles in the evolution of complex organisms.

Histone modifications regulated by LncRNAs

Prominent examples of histone-modifying complexes interacting with lncRNAs are the two polycomb repressive complexes, PRC1 and, in particular PRC2, which mediates methylation of lysine 27 on histone 3 (H3K27me), a histone mark associated with repressed or poised genetic loci. The first report of an interaction of PRC2 with a noncoding RNA came from studies on X-chromosome inactivation in mammals, which involves X-inactive specific transcript (Xist), a lncRNA that is highly expressed from the inactive X-chromosomes in females (Xi) and recruits PRC2 to the Xi to silence gene expression [39–41].

Another prominent example is Hotair, a conserved lncRNA that is transcribed from within the HoxC gene cluster. Hotair was shown to play a repressive role at the HoxD locus by interacting with PRC2. In addition, Hotair also interacts with the histone H3K4me1/2 demethylase LSD1 (KDM1), which removes a histone mark of active chromatin and thus reinforces the establishment of a repressive chromatin environment on target loci [42] (Fig. 2d, e). Targeted inactivation of Hotair in mice leads to de-repression of imprinted genes and of HoxD genes, which lose H3K27-methylation and gain H3K4-methylation marks, as expected from loss of the enzymatic activity of PRC2 and LSD1 at these loci. Furthermore, Hotair knockout mice show skeletal homeotic transformation phenotypes, which are typical for mutations affecting Polycomb mediated repression [43]. Interestingly, mice lacking most of the HoxC cluster including Hotair do not display these phenotypes, suggesting a complex interplay of multiple genomic regions in regulating HoxC cluster gene expression [44].

Braveheart (Bvht) is a lncRNA that is activated during early cardiac differentiation and acts upstream of MESP1, a basic Helix-loop-helix transcription factor involved in early heart development. Bvht also exerts its function via an interaction with PRC2 [45]. The lncRNA Fetal-lethal developmental regulatory RNA (Fendrr; or Foxf1 adjacent noncoding developmental regulatory RNA) is likewise involved in cardiac development and heart function and interacts with PRC2. Fendrr also interacts with WDR5, which is well known for its presence in the MLL complexes that mediate H3K4 methylation, a mark that is thought to oppose H3K27me. Fendrr might play a role at poised genes, or is involved in tuning the balance of active and repressive marks at its target gene promoters, thus adjusting the correct expression level of its targets [46, 47]. These examples are only a small selection and many more lncRNAs seem to function in concert with the PRC2 complex; for more examples, see [48–52].

An intriguing mechanism as to how lncRNAs might recruit their protein interaction partners to specific genomic loci is DNA-RNA triplex formation [53], which has been proposed for several lncRNAs forming a complex with PRC2 [46, 54, 55] (Fig. 2d). In this manner, lncRNAs might direct chromatin or transcriptional modulators to specific genomic sites. This could explain, to some extent, the site-specific action of chromatin-modifying complexes, which do not themselves bind to DNA in a sequence specific manner [38, 56].

An example of a lncRNA interacting with the PRC1 complex is Focally amplified lncRNA on Chromosome 1 (FAL1). The interaction of FAL1 with BMI1, an essential subunit of PRC1, regulates its protein stability [57]. The knockdown of FAL1 in an ovarian cancer cell line led to gene expression changes, slower cell cycle progression and induction of senescence, similar to the effects of the BMI1 knockdown. This latter phenotype could be partially rescued by knockdown of the senescence-promoting factor CDKN1A (p21). Additional ChIP data suggested that BMI1 binds to the promoter of CDKN1A and, together with FAL1, suppresses the expression of this target and of many other genes.

Whereas FAL1 acts in trans on PRC1, ANRIL, a lncRNA transcribed from the INK4B-ARF-INK4A tumor suppressor locus, acts in cis. The ANRIL transcript was discovered in a family with inherited melanoma-neural system tumors and assumed to play a role in tumorigenesis [58]. ANRIL recruits CBX7 (a PRC1 component) via a POLII-dependent mechanism to its locus in order to repress the neighboring INK4B-ARF-INK4A genes, antagonize cellular senescence and indirectly promote cell cycle activity [58, 59].

The repressive H3K27-methylation, catalyzed by EZH2 in the PRC2 complex, is opposed by MLL complexes, which mediate methylation of H3K4me2/3, a mark associated with loci that are actively transcribed or primed for activation. WDR5 is an integral part of all MLL complexes, but also interacts with several other protein complexes [60]. WDR5 has been found to interact with more than 200 lncRNAs in mouse embryonic stem cells (mESCs) [61]. These interactions are important for WDR5 binding to chromatin since a point mutant that cannot bind RNA also fails to stably associate with chromatin. More specifically, two lncRNAs, HOTTIP and NeST have been described to recruit WDR5 to their neighboring genes and thus enhance their transcription [62, 63].

LncRNAs modulate DNA methylation

Whereas many lncRNAs associating with Polycomb complexes act by promoting PRC occupancy at genomic target sites, lncRNAs described in the context of DNA methylation have mostly been found to oppose this epigenetic mark. Frequently, transcription of noncoding RNAs, which interact with DNMTs, keeps a locus free of DNA methylation [54, 64–66] (Fig. 2a). Initially, it was observed that knockdown of extracoding CEBPA (ecCEBPA), a transcript starting upstream of the CEBPA gene, leads to down-regulation of CEBPA and increased DNA methylation of the locus. A subsequent global analysis of all lncRNAs bound to DNMT1 [64] revealed that loci whose RNA products interact with DNMT1 show lower levels of DNA methylation than other loci. A similar observation was made in oocytes and two-cell stage embryos [66], in which the down-regulation of several divergently transcribed promoter-associated ncRNA (pancRNA) led to lower expression of the adjacent protein-coding gene and higher DNA methylation. In the case of one particular gene, IL17b, the pancRNA knockdown resulted in cell death of the developing blastocyst. Supplementing the embryos with IL17b protein rescued the phenotype.

In the case of Tcf21 antisense RNA reducing DNA methylation (TARID), keeping the protein-coding gene (Tcf21) on the opposite DNA strand free of DNA methylation required GADD45A and the TET proteins [65] pointing towards a mechanism involving active DNA demethylation.

LncRNAs regulate chromatin remodeling

In addition to interaction with enzymatic complexes, which covalently modify chromatin, lncRNAs have also been shown to control chromatin remodeling complexes that can alter the nucleosome spacing in an energy dependent manner. In human prostate cancer, gene expression analyses showed that SWI/SNF and the lncRNA SChLAP1 have opposing roles. SChlAP1 interacts with the SNF5 subunit of the chromatin remodeling complex SWI/SNF, and globally inhibits binding of SWI/SNF to chromatin, subsequently leading to genome-wide de-repression of gene activity [67]. Similarly, the lncRNA Myheart (Mhrt), a lncRNA transcribed divergently to Myh6 and overlapping with Myh7 in antisense direction, binds to BRG1, the ATPase subunit of the SWI/SNF complex, and excludes it from the Myh6/7 locus, thus preventing chromatin remodeling [68]. BRG1, in turn, down-regulates Mhrt. Cardiac stress leads to upregulation of BRG1 to a level allowing it to overcome the repulsion of the SWI/SNF complex from chromatin by Mhrt and to bind to the Myh6 locus.

The Evf2 lncRNA, transcribed from the genomic region between Dlx5 and Dlx6, promotes SWI/SNF binding to the enhancers of these genes. However, Evf2 inhibits the remodeling activity of SWI/SNF and thus interferes with upregulation of Dlx5/Dlx6. Accordingly, in Evf2 mutant mice, Dlx6 and, to a lesser extent, Dlx5 are upregulated [69, 70]. In a more recent study, it was observed that the binding of Evf2 to BRG1 can be out-competed by other RNAs of similar length and results in reduced remodeling activity, suggesting that the binding of lncRNAs to the SWI/SNF complex is promiscuous. Xist, on the other hand, binds to components of the SWI/SNF complex, but excludes the complex from the Xi, rather than recruiting it [71]. The SWI/SNF complex can also be recruited by lncRNAs and mediate gene activation, as has been found in hepatocellular carcinoma cancer stem cells. A lncRNA, lncTCF7 or WNT signaling pathway activating non-coding RNA (WSPAR), transcribed 200 kb upstream of TCF7, can recruit SWI/SNF to the TCF7 promoter and thus activate TCF7 expression and WNT signaling [72].

Is the observed binding of numerous lncRNAs to chromatin modifiers specific or promiscuous?

The specificity of lncRNA interactions with chromatin-modifying complexes has been questioned. For instance, a large number of lncRNAs have been shown to co-purify with PRC2, possibly suggesting non-specific binding [50, 52, 73]. Similarly, 15 % of all lncRNAs present in a particular cellular context were shown to be bound by DNMT1 [64, 74]. Likewise, WDR5 binds to more than a thousand RNAs in mESCs, of which 20 % are lncRNAs [61]. This issue of promiscuity was assessed exemplarily for PRC2 in a systematic biochemical study [75], where the authors demonstrate that PRC2 binds RNA in a promiscuous manner, but has remarkably higher affinity for specific lncRNAs than for e.g. bacterial mRNA. Furthermore, they provide evidence that the binding strength correlates with the length of the RNA. Though these studies were performed in vitro at non-physiological conditions, they suggest that the chromatin-modifying complexes can sufficiently differentiate between specific and non-specific interactors. Moreover, since most lncRNA transcripts are expressed at few copies per cell, it is likely that the number of PCR2 molecules per cell is in excess to the number of lncRNA transcripts, which contain a high affinity site for PRC2. Overall, it is not unexpected that a chromatin-modifying complex acting at numerous genomic loci binds many different individual lncRNAs, which stabilize it or recruit it to distinct genomic loci.

Interactions of lncRNAs with transcription factors

Besides chromatin modifiers, transcription factors have also been found to interact with lncRNAs. The lncRNA definitive endoderm-associated lncRNA1 (DEANR1) (also known as ALIEN), for instance, is transcribed downstream of FOXA2 and is highly induced during differentiation of ESCs towards definitive endoderm [76, 77]. Knockdown of DEANR1 inhibits endodermal differentiation due to the reduction of FOXA2 expression. This interpretation is strongly supported by the fact that endoderm differentiation is rescued in DEANR1 knockdown cells upon forced FOXA2 expression. Mechanistically, DEANR1 was proposed to interact with SMAD2/3 and to recruit it to the FOXA2 promoter.

The lncRNA rhabdomyosarcoma associated transcript (RMST) is required for neuronal differentiation by mediating binding of the transcription factor SOX2 to approximately half of its binding sites [78]. Certain lncRNAs have been linked in different ways to transcription factors of the NF-κB pathway. Lethe (after the ‘river of forgetfulness’ in Greek mythology) interacts with RelA and prevents it from binding to DNA [79]. p50-associated cyclooxygenase-2 (COX-2) extragenic RNA (PACER), on the other hand, sequesters p50, which forms homodimers and is repressive at high concentration, thereby lowering its concentration and allowing it to form activating heterodimers with p65 [80]. Another lncRNA, lnc dendritic cells (lnc-DC) binds to the transcription factor STAT3 in the cytoplasm and prevents its dephosphorylation by SHP1, thereby activating STAT3 and thus dendritic cell differentiation [81].

An elaborate mechanism linking transcription factor binding and chromatin modifications has been found for breast cancer anti-estrogen resistance 4 (BCAR4). This lncRNA binds to the SMAD nuclear interacting protein 1 (SNIP1) and a phosphatase (PNUTS) involved in RNA polymerase II regulation. In response to cytokine stimulation, BCAR4 lifts the inhibitory effect of SNIP1 on p300, a histone acetyl transferase and transcriptional activator. Acetylated histones are necessary for BCAR4 mediated recruitment of PNUTS, which in turn leads to active polymerase II at GLI2 controlled genes [82].

Also CTCF, a protein with the dual function of a transcriptional regulator and a boundary factor, has been shown to interact with numerous noncoding RNAs [83]. An example is the interplay between CTCF and the lncRNAs Jpx, Tsix and Xite, which is important for X-chromosome inactivation [83, 84]. Prior to inactivation, Tsix and Xite recruit CTCF in cis to the X-inactivation center promoting homologous X-chromosome pairing at the Xic [83]. During X-chromosome inactivation, Jpx evicts CTCF in trans from the Xist locus and thereby prevents promoter blockage by CTCF [83]. These examples illustrate the dynamic interaction between lncRNAs and transcription factors in regulating chromatin-mediated cellular functions.

Regulation of genome organization by looping of enhancers

Previously, transcripts found at enhancers had been interpreted as byproducts of transcription from promoters contacted by enhancers. Now, it is well established that RNA transcripts produced at active enhancers are functionally important [85–87]. An example of enhancer-linked transcripts, already identified two decades ago, has been described for the locus control region (LCR) of the β-globin cluster [88, 89]. To date, numerous studies have assigned functions to enhancer-associated RNAs [90] (Fig. 2b). There are different classes of RNAs observed at enhancers. The classical enhancer RNAs (eRNAs) are derived from bidirectional transcription, are unspliced and rather short transcripts [85, 91]. Additionally, there are also unidirectional, often spliced and processed transcripts, which are indistinguishable from canonical lncRNAs, but happen to be transcribed from an enhancer [23, 92]. These transcripts often function in cis. Finally, there are lncRNAs with enhancer-like functions (also called ncRNA-a) [93]. Whether this classification reflects different functional classes of RNAs remains to be investigated.

In general, eRNAs are co-regulated with their neighboring gene(s) [91] and presumed to be important for gene expression, possibly by supporting cohesin binding [94]. However, experimental support for the latter is rather mild. While eRNAs in some studies were shown to stabilize chromosome loops [94, 95], they appeared to be dispensable for promoter-enhancer contact in others, raising doubts about the role of eRNAs in chromosome looping [96, 97].

LncRNAs transcribed from enhancer regions can also have inhibitory effects on their target gene. The promoter deletion of the lncRNA Haunt (also known as linc-Hoxa1), for instance, has been shown to lead to upregulation of several genes of the neighboring HoxA gene cluster. However, larger deletions in the Haunt locus impair HoxA expression [98], demonstrating that the genomic locus can have opposing functions to the RNA product encoded by it. The repressive effect of Haunt was furthermore observed at the single cell level confirming that the expression of HoxA1 and Haunt are anti-correlated [99].

A similar mechanism was found for Playrr (D030025E07Rik), a lncRNA encoded upstream of the homeodomain transcription factor Pitx2 [100]. Pitx2 expression is constrained to the left side of the gut dorsal mesentery, whereas expression of Playrr is detected on the right, but spreads to both sides in Pitx2 deletion mutants. A CRISPR/Cas9 generated mutation resulting in Playrr RNA decay caused upregulation of Pitx2 expression [100]. Both experiments reveal a mutual inhibitory interaction between Pitx2 and Playrr. An enhancer involved in Pitx2 regulation in the gut overlaps the TSS of Playrr. In the cells on the left-hand side, where Pitx2 is expressed, the Pitx2 locus was found to be in closer proximity to the Playrr locus than in cells on the right side and to cause down-regulation of Playrr by an unknown mechanism. In cells on the right, on the other hand, Playrr RNA is thought to cause a separation of the Pitx2 locus from the enhancer at the Playrr locus suggesting that Playrr expression interferes with the looping of the Pitx2 promoter to its enhancer at the Playrr locus and thus with Pitx2 gene activation (Fig. 2c).

Furthermore, the lncRNAs PRNCR1 and PCGEM1 promote enhancer-promoter looping in prostate cancer cells [101]. Other lncRNAs have been suggested to assist chromosome conformation more globally. A particular situation can be observed at the inactive X-chromosome (Xi). The 3D-structure of Xi is dependent on Xist, supposedly because Xist excludes cohesin binding from Xi [71]. Deletion of Xist led to loss of the compact chromosome structure of the Xi and reorganization into chromosome domains as they are found on the active X-chromosome. Recently, another lncRNA, functional intergenic repeating RNA element (Firre), which escapes X-chromosome inactivation was suggested to stabilize the perinucleolar location of the Xi [102]. Additionally, Firre was proposed to mediate focal trans-chromosomal contacts [103]. This finding was questioned by a different study, which reported a dispersed nuclear distribution of Firre and a role in the control of factors involved in RNA processing [104]. These examples show that lncRNAs can play important roles in controlling local intra- and inter-chromosomal genomic structure and interactions.

Regulation at the posttranscriptional level

LncRNAs primarily found in the cytosol are thought to be involved in gene regulation on the posttranscriptional level (Fig. 2e–h). Accordingly, several lncRNAs have been shown to influence splicing patterns of either specific genes or globally by interacting with splicing factors. The lncRNA Pnky (divergent to Pou3f1) binds to the splicing factor PTBP1 and thereby regulates the splicing patterns of a subset of genes involved in neurogenesis [105]. Whether TUNA (megamind), which interacts with PTBP1 as well, functions via the same mechanism remains to be investigated [33, 106]. In addition, myocardial infarction associated transcript (MIAT) has been found to influence splicing in different systems. In embryonic neurogenesis, MIAT is involved in controlling the splicing pattern of Wnt7b [107]. In iPS cells and mouse primary neurons, MIAT binds to splicing factors, and in post-mortem brains of schizophrenia patients, it was found to be down-regulated, suggesting that it might contribute to this pathological phenotype [108]. MALAT1, a highly expressed lncRNA enriched in nuclear speckles, has been suggested to be a general component of the splicing machinery [109]. However, this view has been challenged since the MALAT1 knockout did not show disintegration of splicing speckles [110]. A very specific role in the cytoplasm was found for the lncRNA non-coding RNA activated by DNA damage (NORAD) [111]. NORAD sequesters PUMILIO proteins and thereby promotes genomic stability (Fig. 2f). Ablation of NORAD leads to hyperactivity of PUMILIO causing chromosome instability and aneuploidy due to repression of PUMILIO target mRNAs that function in DNA replication, mitosis and DNA damage repair.

The most recently described class of lncRNAs that came into focus are the circRNAs. Due to their circular structure, they are resistant to degradation by exonucleases and are therefore thought to be highly stable within a cell. The brain specific circRNA CDR1as was shown to act as a sponge for the miRNA let-7 [26, 27] (Fig. 2f). Accordingly, overexpression of CDR1as in zebrafish caused impaired midbrain development, similarly to the effect of a let-7 inhibitor. It was also shown that small polypeptides can be derived from some circRNAs, extending the functional repertoire of these stable RNA molecules [112]. Most likely, circRNAs do not only interact with other transcripts, but also with proteins and thus may function in a wide range of different processes.

Cancer

Approximately one third of publications currently listed in PubMed linked with the keyword “lncRNA” also contain the keyword “cancer”. Even before the era of next generation sequencing, a few studies found lncRNAs to be dysregulated in tumors [35, 143, 144]. To date, numerous primary tumor samples, metastatic biopsies, the adjacent normal tissue and cancer cell lines have been profiled for their RNA expression and the results recapitulate the high degree of specificity found for noncoding RNAs in normal tissue compared to protein-coding genes [25, 67, 145–147]. Clustering of lncRNA expression patterns identifies subclasses of tumor types with high significance, opening the possibility to define tumor subtypes with a high chance to respond to targeted therapy. A study that considered samples from more than 5000 tumors from the cancer genome atlas (TCGA) project, demonstrated that, while approximately the same number of protein coding and lncRNA genes were dysregulated, 60 % of lncRNAs showed specificity for only one tumor type [146].

In addition, somatic copy number alterations (SCNA) found in cancer tissue often include lncRNA loci [57, 146, 147]. For example, the lncRNA FAL1, a lncRNA that was found amplified in a number of different tumor types, was shown to be important for the tumorigenic effects of different cell lines via a BMI1-mediated mechanism (see above), and its expression level and copy number significantly correlated with patient survival [57]. Furthermore, several lncRNAs have been shown to be prognostic for overall or disease-free survival of patients [67, 82, 148–150], information that is integrated into therapeutic decisions. Depletion of selected lncRNAs led to reduced malignancy, assayed as tumor size or metastatic potential in xenograft experiments [67, 72, 82, 101, 148, 149, 151–154], although these experiments were not able to distinguish between effects on tumor formation versus tumor progression. Further refinement was achieved when RNA knockdown was induced after tumor formation, mimicking therapeutic conditions. Reduced tumor growth was, for instance, observed upon FAL1 depletion [57]. In a different study, the inhibition of MALAT1 after tumor induction reduced metastasis formation in a xenograft lung cancer model and in a metastasizing breast cancer model [149, 155]. MALAT1 knockout mice do not show overt phenotypes [110, 156, 157] suggesting that down-regulation of MALAT1 in humans might also not be harmful for normal human cells, which makes it an attractive target for systemic application in patients.

A special case is the noncoding pseudo gene BRAFP1, which has been shown to be either mutated or aberrantly expressed in cancers, like its coding homologue BRAF. A mouse model revealed that induced overexpression of Braf-rs1 (the murine homologue of BRAFP1) leads to DLBCL (diffuse large B cell lymphomas)-like tumors. However, when the transgene was first induced to form tumors and then shut down again, the tumors regressed demonstrating the functional significance of the pseudo gene [158]. Braf-rs1 was shown to cause upregulation of Braf by acting as a ceRNA (competing endogenous RNA) and by “sponging” miRNAs that would otherwise target Braf transcripts.

One mechanism that cells can employ to prevent tumor formation is oncogene-induced senescence (OIS) [159]. The INK4B-ARF-INK4A locus, which plays a major role in OIS, is silenced in proliferating cells by Polycomb group proteins. This mechanism is dependent on the lncRNA ANRIL, which is transcribed from within the locus and acts in cis [58, 59, 160]. Furthermore, the MIR31HG gene located 400 kb upstream of the INK4B-ARF-INK4A locus, also encodes a lncRNA involved in the regulation of INK4A [161]. A third lncRNA functioning in OIS, UCA1, stabilizes mRNAs of protein-coding genes involved in the senescence pathway, probably by sequestering hnRNPKA1 [162]. Also, the tumor suppressor p53 and its associated signaling pathways include lncRNAs. p53 binds and regulates numerous lncRNAs, such as the linc-p21, which in turn regulates gene expression in concert with p53 and p21 [163–166].

Cardiovascular diseases

LncRNAs have also been linked to cardiovascular diseases (CVD), which are a major cause of death in western societies, and have been studied both in patient data, as well as in mouse models. It was observed that several genomic regions significantly associated with myocardial infarction encode lncRNAs, for example MIAT and ANRIL [167, 168]. Deletion of a 70 kb region including a large proportion of the ANRIL locus led to increased mortality in mice, particularly in combination with a high fat diet [169].

In other studies, massive parallel RNA sequencing revealed numerous lncRNAs displaying altered expression following surgically-induced myocardial infarction in mice [170, 171]. In cellular differentiation models, the depletion of selected lncRNAs impaired cardiomyocyte formation making them promising candidates for in vivo assessment of their functions [172]. Analysis of the noncoding transcripts originating from the myosin heavy chain 7 (Myh7) locus, an important structural protein required for heart contractions, revealed that one, the lncRNA Mhrt, was down-regulated after induced myocardial infarction. Ectopic expression of Mhrt in mice, which had been subjected to this operation reduced the severity of symptoms observed in control mice during recovery [68]. When lncRNAs in plasma from myocardial infarction patients were profiled, a mitochondrial lncRNA named long intergenic noncoding RNA predicting cardiac remodeling (LIPCAR) was found to significantly correlate with mortality caused by CVD and could thus be considered as a biomarker [173]. Whether this lncRNA also plays a causative role in infarction or heart failure remains to be investigated.

LncRNAs have been widely suggested as biomarkers and therapeutic targets in certain pathological settings. One reason for this is the extremely specific expression of many lncRNAs, opening the potential of targeting cellular subgroups rather than affecting patients systemically as in conventional treatments. Even if a lncRNA itself is not the disease driver, it can serve as drug target in cases where it acts as co-factor for a driver gene, which is more broadly expressed. Also, since lncRNAs often play quantitative roles rather than acting as switches in genetic programs, side effects of therapeutic treatments might be easier to control.