Non‐coding RNAs: key players in cardiac disease

Molecular mechanisms underlying heart failure (HF) are only partly understood. Non‐coding RNAs (ncRNAs) have been reported to control function and signalling routes in the myocardium. As ncRNAs such as microRNAs (miRNAs), long non‐coding RNAs (lncRNAs) or circular RNAs (circRNAs) can be selectively targeted via pharmacological approaches, this opens new avenues for diagnostic and therapeutic approaches. Here, we review the main ncRNA classes and how they influence cardiac biology. In addition we provide insight into the role of ncRNAs in chemotherapy‐induced cardiac dysfunction. To provide a better understanding of ncRNAs in cardiovascular biology we present an outlook on specialized functions such as chromatin remodelling, biomarker potential and the recently discovered ncRNA‐derived micropeptides.


Introduction
For a long time, one of the central dogmata of biology was simply 'DNA makes RNA makes protein' , but in the last several years it has become clear that this hypothesis may need to be revised. As indicated in Fig. 1A about three-quarters of the genome is actively transcribed, and not-transcribed areas of the genome mainly locate to the telomere ends and the centromere regions of the chromosomes (greyed out chromosome regions) (Frith et al. 2005;Djebali et al. 2012). The actively transcribed part spans both repeated (grey) and unique sequences. Since sequencing technologies have significantly advanced, it became clear that only 2-3% of the genome encodes proteins containing exons (E, orange). Instead, other unique genomic regions containing regulatory sequences such as promotor and enhancer regions (blue) as well as intronic regions (purple) present the overwhelming portion of arising transcripts. No longer being just an intermediate step between DNA and proteins these non-coding RNAs (ncRNAs), that may contain exons of protein-coding genes, span a wide range, arbitrarily classified by their size (below or above 200 nt, denoted "short" and "long" in Fig. 1B) and basic molecular features. In this review we summarize key findings on ncRNAs and their diagnostic and therapeutic application in the context of heart failure as presented recently by Prof. Thomas Thum at the Gordon Research Conference on Cardiac Regulatory Mechanisms, June 3-8, 2018.
In 1993, Lee and colleagues identified the first short non-coding transcript which interacted with a messenger RNA (mRNA) (Lee et al. 1993). By binding to complementary sequences in the 3 -untranslated region (UTR) of mRNAs, these microRNAs (miRNAs, also miR-#), short and highly conserved oligonucleotides of ß22 nt, led to mRNA degradation or translational repression by directing a silencing complex (Bartel, 2009). The modus operandi is directly dependent on sequence complementarity -and as miRNAs have a rather short binding site of just 6-8 bp, they usually exhibit rather broad specificity. As such, >60% of all mRNAs possess such binding sites, potentially being targets of miRNA regulation (Friedman et al. 2009). As of now 1917 human miRNA sequences are known (included in the "short ncRNA" category, Fig. 1B), allowing for a high degree of tissue-or disease-specific regulation despite their broad specificity (Griffiths-Jones, 2004;Kozomara & Griffiths-Jones, 2014). Some miRNAs, e.g. miR-146a or miR-21 * (asterisk indicates the usually degraded miRNA passenger strand), can even act on other cell types in a paracrine fashion through exosome transport (Halkein et al. 2013;Bang et al. 2014).
In contrast, long non-coding RNAs (lncRNAs) are single-stranded RNA molecules of >200 nt, often up to several kilobases (included in the "long ncRNA" category, Fig. 1B). As lncRNAs described a very heterogeneous class of molecules, they are often further subdivided by their genomic location. As such, lncRNAs can span several exonic and intronic regions of protein-coding genes (sense), solely reside within introns (intronic), be transcribed from the opposite strand (antisense and bidirectional) or be located within the former "junk DNA" in-between protein-coding genes (intergenic). As lncRNAs contain palindromic stretches within their sequence they can form simple or complex loops and hairpin structures. By arranging the single loops into three-dimensional structures, lncRNAs can bridge the gap between nucleic acids and proteins, being an integral scaffold for protein complexes in transcription and translation, e.g. the lncRNA NEAT1 maintaining structural integrity of nuclear paraspeckles (Sasaki et al. 2009). Importantly, and in contrast to miRNAs, lncRNA sequences are not well conserved across species but rather lncRNA structure and function, presenting a hurdle to in silico prediction of homologues.
The first lncRNA identified, already in the early 1990s, was the X-inactive-specific transcript (XIST), a 20 kb molecule mediating transcriptional silencing of the inactive X chromosome through direct protein interactions, leading to recruitment of histone deacetylases and polycomb repressive complex 2 (Brown et al. 1991(Brown et al. , 1992McHugh et al. 2015). Overall lncRNAs are enriched in the nucleus, e.g. XIST locates strongly to the nucleus and is therefore often used as quality control in RNA fractionation experiments . In contrast, the endothelial-cell enriched lncRNA MIR503HG localizes to the cytosol, where it acts as scaffold to promote angiogenic signalling (Fiedler et al. 2015). As already implied by the name of MIR503HG, some lncRNAs can act as host genes for miRNAs, e.g. miR-503 for MIR503HG or miR-675 in the muscle-enriched lncRNA H19 (Cai & Cullen, 2007). Moreover, a number of functions have been proposed based on the location of lncRNAs: nuclear-enriched lncRNAs seem to regulate gene expression by interacting with enhancer regions and transcription factors or serving as guide for histone-modifying enzymes. Cytoplasmic lncRNAs instead regulate mRNA translation by (de-)stabilization of ribonucleoprotein complexes, mRNA stability or miRNA sponging. Biogenesis and potential functions of lncRNAs have already been reviewed in more detail by Beermann and colleagues (Beermann et al. 2016).
Another member of the ncRNA family receiving increasing attention is the class of circular lncRNAs (circRNAs), which usually form during the processing of mRNA transcripts. Several years ago circRNAs were regarded as sequencing artifacts and only ongoing improvement of bioinformatics tools led to their recent (re-)discovery. The hallmark of circRNAs is the formation of a covalently closed loop structure through backsplicing or exon-skipping events (Nigro et al. 1991;Cocquerelle et al. 1993;Jakobi et al. 2016). CircRNAs often possess higher stability than their linear counterparts as circRNAs lack a poly(A) tail and are reasonably resistant to exonucleases. Initial studies suggested mainly a role as miRNA sponges, thereby fine-tuning RNA interference mechanisms (Hansen et al. 2013;Memczak et al. 2013). In strong contrast Boeckel et al. were able to show recently that the majority of circRNAs do not possess any miRNA binding sites, or only a single one, limiting the potential cross-regulation with miRNAs (Boeckel et al. 2015). Endogenous circRNA expression is often tissue specific, indicating an important role of circRNAs in transcriptional regulation of biological processes. For instance, Tan et al. recently observed that of the 100 highest expressed genes in cardiac tissue (protein-coding and non-coding), about one-third is also transcribed into circular isoforms (Tan et al. 2017). While a certain cross-regulation with miRNAs seems to be present, the main function of those circRNAs, and how their biogenesis is controlled, still remains elusive.

ncRNAs regulate the cardiac phenotype
In the last years several reports have underlined that inhibition of miRNAs has been associated with therapeutic effects during heart failure (HF) or ischaemic injury J Physiol 598.14 (Montgomery et al. 2011;Hullinger et al. 2012). In line with this, Ucar et al. identified the miR-132/212 cluster consisting of two miRNAs driving cardiomyocyte hypertrophy. Whereas cardiomyocyte-specific overexpression led to cardiac hypertrophy, heart failure and death in mice, the silencing of miR-132 protected against heart failure induced by pressure overload (Ucar et al. 2012). Another factor promoting cardiac hypertrophy is the lncRNA CHAST, which was found to be upregulated in cardiomyocytes of transverse aortic constriction (TAC)-operated mice and hypertrophic heart tissue from aortic stenosis patients. Therapeutic gain-of-function led to induction of cardiomyocyte hypertrophy indicating a pivotal role for cardiac performance . While these ncRNAs actively promote hypertrophy, negative regulation of hypertrophy by ncRNAs has also been described. An outstanding example of such a preventive lncRNA is Mhrt (or Myheart), which prevents the helicase domain of the Brg1-Hdac-Parp chromatin repressor complex from recognizing its target sequence by competitive binding and thereby protecting the heart from cardiomyocyte hypertrophy (Han et al. 2014). In contrast to those heart failure-associated ncRNAs Eulalio and colleagues deciphered the role of miRNAs in early cardiac development and showed that administration of miR-199a and miR-590 in neonatal rodents could enhance cardiomyocyte proliferation as another layer of cardiac therapeutics (Eulalio et al. 2012).
While these ncRNAs are acting on cardiomyocytes, the first proof-of-concept of treating cardiac disease with miRNA-antagonists focused on miR-21, a micro-RNA selectively upregulated in fibroblasts of the failing heart. Silencing of miR-21 induced repression of ERK1/2-dependent growth factor signalling leading to regression of interstitial cardiac hypertrophy and fibrosis in a murine pressure overload model (Thum et al. 2008). Similarly, inhibition of the lncRNA MEG3 decreased cardiac fibrosis and improved cardiac parameters by targeting the production of matrix metalloproteinase-2 (MMP-2) (Piccoli et al. 2017).
A driving force of the controlled revascularization response after cardiac ischaemia is GATA2, a major transcription factor highly associated to miRNA biology. As such, GATA2 is both controlled by miR-24, but also induces miR-126 and miR-221 expression (Fiedler et al. 2011;Hartmann et al. 2016). Moreover, GATA2 was also repressed after silencing of the lncRNA MIR503HG, highlighting the importance of lncRNAs towards maintaining balanced gene expression as well (Fiedler et al. 2015). Interestingly, upregulation of miR-503 (the miRNA transcribed from the same locus as MIR503HG) was shown to inhibit angiogenesis independent of GATA2 Wen et al. 2018). Aside from interacting with transcription factors, members of the miR-17-92 cluster regulate vascular signalling in endothelial cells by targeting the mRNAs of several proangiogenic genes (Doebele et al. 2010). Another central regulator of endothelial cell biology is miR-146a, a microRNA induced upon peripartum cardiomyopathy in patients or in pressure overload models. Interestingly, miR-146a is released from those endothelial cells via exosomes and then acts on cardiomyocytes where it promotes cardiac hypertrophy and left ventricular dysfunction (Halkein et al. 2013;Heggermont et al. 2017). In the context of atherosclerosis, miR-146a was described to restrict pro-inflammatory signalling in endothelial cells, highlighting an atheroprotective role of this microRNA (Cheng et al. 2017).
Taking research from linearized to the previously mentioned circularized RNA structures, Boeckel et al. reported the induction of a circRNA, cZNF292, in a model of endothelial hypoxia. Specific silencing of the circRNA, but not the host gene, significantly inhibited sprouting and tube formation (Boeckel et al. 2015). Interestingly, cZNF292 does not contain miRNA binding sites, strongly implying a miRNA-independent mechanism for circRNA-dependent regulation. The diverse roles of ncRNAs in the cardiac phenotype (hypertrophy, vascularization, protein signalling and potential use as biomarker) are summarized in the Abstract figure.

Non-coding RNAs in chemotherapy-induced cardiac dysfunction
State-of-the-art for the treatment of cancer is mainly based on the application of chemotherapeutics. While targeted delivery approaches have increased over the past years, these compounds, mostly potent cytostatica or cytotoxins, still also reach healthy tissue. Depending on mechanism of action and condition of the patient, the high doses needed for successful treatment of the cancer (to manage the risk of secondary malignancies) often also result in secondary damage to whole organ systems. This is especially true for cardiac tissue with its extraordinary situation of constant, high workload and low regenerative capabilities, eventually leading to chemotherapy-related cardiac dysfunctions (CRCDs). Efforts were made to distinguish between permanent myocardial damage (type I CRCD) and reversible dysfunction (type II CRCD) based on mechanism and dose dependency (Ewer & Lippman, 2005). Over several decades studies reported type I CRCD after treatment with anthracyclines such as doxorubicin (Doxo), visibly e.g. in the high rate of human cancer patients developing congestive heart failure (Hoff et al. 1979;Lipshultz et al. 1991Lipshultz et al. , 2008Seidman et al. 2002;Swain et al. 2003). This is contrasted to type II CRCD generally reported by upcoming targeted therapies such as the antibody Trastuzumab (Ewer & Lippman, 2005;Ewer & Ewer, 2010).
In this setting, Gupta and colleagues recently highlighted the importance of circRNA formation during chemotherapy-induced cardiotoxicity (Gupta et al. 2018). In murine myocardium, Doxo treatment strongly repressed expression of the RNA-binding protein Qki5 (Quaking). In line with this, therapeutic Qki5 overexpression was successful to counteract detrimental cardiac effects of Doxo treatment in mice. Interestingly, repression of cardiac Qki5 efficiently lowered expression of circRNAs derived from important cardiomyocyte genes, e.g. Ttn (Titin), thereby implying circRNAs play an important anti-apoptotic role in cardiac muscle (Gupta et al. 2018). The effects of Doxo-based chemotherapy are not entirely understood, as several recent long-term follow-up studies of Doxo treatment could not confirm worsening of cardiac function (Ganz et al. 2017) or identified valvular abnormalities, but not congestive heart failure (Materazzo et al. 2017). Nevertheless, the study of Gupta and colleagues linked circRNA biogenesis and function with a promising new route for potential adjuvant therapy of (circular) ncRNAs during chemotherapy. Importantly, most data on potential ncRNA therapies are derived from in vitro studies or animal models with restricted translatability, given the limited conservation of lncRNAs across species (compare 'Introduction'). Therefore current clinical trials of non-coding RNA-based therapies focus on miRNAs: antifibrotic properties of anti-miR-21 intervention are currently being studied in a phase II trial for the treatment of Alport syndrome (NCT02855268) and Miravirsen represents a miR-122 antagonist against hepatitis C virus infection (Lindow & Kauppinen, 2012;Janssen et al. 2013;Ottosen et al. 2015). Although these examples are not about cardiac diseases, ncRNA-based therapies are already in clinical trials and will explore the general potential and safety of this molecule class. Nevertheless, more research is needed to determine the optimal formulation and delivery of ncRNAs, to investigate their pharmacokinetics, and to standardize ncRNA therapy. To achieve this goal, intensive research on delivery routes will initiate a path towards therapeutic application of miRNAs, lncRNAs or circRNAs in cardiac disease.

Specialized ncRNA function
The functions and mechanisms described above are now established, leading to the development of ncRNA-based drugs. In addition, we want to offer insight into recent developments which we believe will become relevant over the coming years.
One subclass of intergenic ncRNAs are enhancerderived lncRNAs (eRNAs, included in the "long ncRNA" category, Fig. 1B), which are synthesized specifically at enhancers, thereby regulating mRNA transcription (by this means being in close proximity with the transcription complex) (Kim et al. 2010). Another important subclass, originally termed ncRNA-a, which are characterized by another set of histone marks, describes ncRNAs able to activate the expression of neighbouring genes, both in cis and in trans (Ørom et al. 2010). Whether or not these reports describe truly distinct classes of ncRNAs, both variants establish chromatin accessibility by recruiting chromatin remodelling to enable polymerase II assembly, e.g. by directing chromatin looping through interaction with the Mediator complex (Lai et al. 2013), the Long intervening/intergenic noncoding RNA (linc RNA) Yam-1 enabling transcription of miR-715 in myoblasts (Lu et al. 2013), or eRNAs enabling MyoD access (Mousavi et al. 2013). Ounzain and colleagues identified CARMEN, a lncRNA derived from a super enhancer (SE). By directing components of polycomb repressive complex 2 (PRC2), CARMEN regulates cardiomyocyte differentiation and maintains this functional state (Ounzain et al. 2015). Similarly, lncRNA Wisper is SE-derived, but regulates cardiac fibroblast proliferation, migration and survival. Pharmacological intervention effectively reduced fibrotic scar formation supporting cardiac healing (Micheletti et al. 2017).
The complexity of lncRNAs also offers possible interactions with ribosomes suggesting that a certain set of lncRNAs is able to encode for short peptide sequences, so-called micropeptides. By way of example, Anderson et al. identified myoregulin, a 46 amino acid micropeptide constraining Ca 2+ handling in mice (Anderson et al. 2015). Recently, the same group identified DWORF, a 34 amino acid micropeptide activating the Ca 2+ pump SERCA (Nelson et al. 2016). It is under debate whether these micropeptides are actually functional, e.g. as inter-or intracellular messengers, or only derive due to pioneering rounds of ribosomes before nonsense-mediated decay can take place, as experimental identification and validation is challenging and calls for integration of carefully conducted proteome-based approaches. Similarly, it is unclear how to correctly distinguish between true ncRNAs encoding micropeptides and misannotations of protein-coding genes, which harbour short open reading frames (Bánfai et al. 2012).
Given the sheer number of different ncRNAs, tissue-specific expression could open their potential use as biomarkers. One candidate microRNA could be miR-122, a miRNA highly liver specific and associated with risk of developing metabolic syndrome (Willeit et al. 2016(Willeit et al. , 2017. Also type 2 diabetes has been associated with a certain miRNA expression pattern, especially loss of miR-126 in endothelial cells, thereby linking diabetes and impaired angiogenesis (Zampetaki et al. 2010). Zampetaki et al. further investigated this pattern and found platelets as a major miRNA reservoir (Zampetaki et al. 2012). Also lncRNAs have been found to be possible biomarkers, e.g. Kumarswamy et al. identified the mitochondrial lncRNA F. P. Kreutzer and others J Physiol 598.14 LIPCAR to be a predictive factor in plasma of myocardial infarction patients (Kumarswamy et al. 2014).
Based on their resistant structure, circRNAs naturally lend themselves as biomarker candidates. Indeed, Memczak et al. could identify >4000 distinct circRNAs in human blood (Memczak et al. 2015). Which ncRNA subtype is successful to become a superior biomarker compared to troponin-based evaluation needs to be determined in future research.

Outlook
In the last 20-25 years, it has become increasingly obvious that central concepts of biology are insufficient to describe the larger picture, e.g. gene number and genome size are not predictive of an organism's perceived complexity (Gregory 2004).
Cook and colleagues noticed the enrichment of active RNA polymerase II in discrete foci of the nucleus, which they termed "transcription factories" (Iborra et al. 1996). These foci are able to enrich and concentrate proteins, and possibly other components as well (Eskiw et al. 2008). Although the precise components are unknown, the foci seem to be held together by some kind of underlying scaffold which counteracts diffusion and aids forming new foci after mitosis. Given the multi-layered model of ncRNA interactions, it would be no surprise if ncRNAs would also be part of these foci, possibly as part of the scaffold directing chromatin loops and transcriptional compartments into proximity (similar to the function of lncRNA NEAT1 in paraspeckles, Sasaki et al. 2009). The direct involvement of lncRNAs in the transcriptional machinery is providing first evidence that the assembly and regulation of multiple-molecule complexes are not the sole domain of proteins, but rather proteins binding to DNA and RNA can profit from the guidance of stretches from lncRNA molecules. Next to these speculative intracellular functions of lncRNAs in particular, validation of miR-based therapeutic strategies to improve cardiac output will be in demand as a field of cardiac drug research in the near future. This also implies that multiple efforts have to be taken in identifying true target structures to promote pharmacological evaluation at the pre-clinical and clinical level.