Introduction: the arrival on the scene of the non-coding RNAs

With protein-coding genes representing only 2 % of the human genome, the other 98 % have been considered for a long time as inactive material, regions of several mega-bases without any function, so-called ‘junk DNA’ (Taft et al. 2007). However, the overwhelming development of high-density micro-arrays and high-throughput sequencing technologies, as well as of bioinformatics analyses, has enabled to go deeper into transcriptomes, revealing that eukaryotic genomes are pervasively transcribed (Berretta and Morillon 2009). For instance, the ENCODE project revealed that up to 75 % of the human genome is transcribed in at least one cell line or condition (Djebali et al. 2012).

The pervasive transcription of eukaryotic genomes produces thousands of non-coding transcripts, that are commonly classified into small and long (l)ncRNAs, and that are now recognized as major regulators involved in multiple cellular processes, including cell differentiation and development, chromosome dosage compensation, imprinting, regulation of gene expression, cell cycle control and adaptation to environment changes (Rinn and Chang 2012; Sole et al. 2015; Wery et al. 2011). In addition, lncRNAs show tissue-specificity (Djebali et al. 2012), indicating that their expression is tightly regulated. Furthermore, dysregulation of ncRNAs has been associated to human diseases, such as cancer or neurodegenerative disorders (Taft et al. 2010).

In most eukaryotes, small and long regulatory ncRNAs coexist and even cooperate. However, the budding yeast Saccharomyces cerevisiae constitutes an exception to this paradigm as it has lost the RNA interference system and is, therefore, devoid of small-interfering (si)RNAs and micro (mi)RNAs (Drinnenberg et al. 2009). In this respect, S. cerevisiae is a unique model to specifically study the regulatory effects of lncRNAs, which in other organisms might be partially hidden by the effects of the small RNAs (such as siRNAs and miRNAs).

Cytoplasmic lncRNAs are targeted by a translation-dependent surveillance pathway

Over the last years, thousands of lncRNAs have been annotated in S. cerevisiae (Tisseur et al. 2011). Strikingly, most of them were found to be cryptic due to their rapid and extensive degradation in the nucleus or in the cytoplasm (Tisseur et al. 2011; Tudek et al. 2015). Notably, the nuclear and cytoplasmic RNA degradation machineries target distinct types of lncRNAs. For instance, the nuclear exosome-dependent 3′–5′ decay pathway degrades the so-called Cryptic Unstable Transcripts (CUTs), a class of lncRNAs that are mainly transcribed from divergent bidirectional promoters (Neil et al. 2009; Xu et al. 2009). On the other hand, the cytoplasmic Xrn1-dependent 5′–3′ RNA decay pathway is specialized into the degradation of another class of lncRNAs referred to as Xrn1-sensitive unstable transcripts (XUTs), most of which are antisense to protein-coding genes (Van Dijk et al. 2011).

XUTs are synthesized by RNA Polymerase II as capped and poly-adenylated transcripts, similar to mRNAs (Van Dijk et al. 2011; Wery et al. 2016). But what determines their instability has remained unclear until the recent finding that the majority of them are specifically targeted by the Nonsense-Mediated Decay (NMD) pathway (Malabat et al. 2015; Wery et al. 2016). NMD is a translation-dependent RNA decay pathway that targets mRNAs with aberrant translation termination, such as mRNAs with premature stop codon (Muhlrad and Parker 1994) and long 3′-UTR (Muhlrad and Parker 1999). The sensitivity of XUTs to NMD suggests that once in the cytoplasm, they associate to the translation machinery and undergo at least a pioneer round of translation. Supporting this hypothesis, ribosome profiling data revealed the presence of small open reading frames (smORFs) on yeast lncRNAs (Smith et al. 2014), including a large set of XUTs (Malabat et al. 2015; Wery et al. 2016), and some of them were shown to be translated into peptides in vivo (Smith et al. 2014). Strikingly, ribosomes on NMD-sensitive XUTs are restricted to a short 5′-proximal region, followed by a long ribosome-free 3′-UTR, which probably constitutes the NMD-activating signal (Wery et al. 2016).

Notably, ribosome-profiling approaches also identified smORFs on transcripts annotated as non-coding in other Eukaryotes, including Drosophila (Aspden et al. 2014), zebrafish (Bazzini et al. 2014) and mouse (Ingolia et al. 2011), although there is debate on the extent to which ribosome-footprints detection reflects genuine on-going translation (Chew et al. 2013; Guttman et al. 2013).

The observation of ribosome binding to smORFs on transcripts annotated as lncRNAs challenges the initial assumption that these transcripts are really noncoding and raises the fundamental question of the function of the peptides produced upon translation of such smORFs. In this regard, recent works described lncRNAs producing smORFs peptides that control heart activity in Drosophila (Magny et al. 2013) and mammals (Nelson et al. 2016), or cell movement during embryogenesis in zebrafish (Pauli et al. 2014). In yeast, the evolutionary conservation of a subset of lncRNAs smORFs within yeast species indicates that the encoded peptides might have biological importance (Smith et al. 2014).

Thus, many transcripts initially thought to lack coding potential are likely to bear smORFs, that can be translated and give rise to functional peptides (Fig. 1). On the other hand, we speculate that a fraction of smORFs-bearing transcripts, reminiscent of the yeast NMD-sensitive XUTs, will also be targeted by the NMD in other eukaryotic cells (Fig. 1). In this respect, NMD inhibition in mouse embryonic stem cells has been shown to result in stabilization of a subset of annotated lncRNAs (Smith et al. 2014). Conceptually, one can also imagine that a cryptic transcript targeted to the NMD in one condition could escape the NMD and be stabilized in another condition (see below), possibly giving rise to a functional smORF peptide. Additional work will be required to define the comprehensive landscape of NMD-sensitive lncRNAs in different eukaryote models, but given the extent of ribosome association to lncRNAs, we anticipate the NMD to be recognized in the future as a major regulator of cytoplasmic lncRNAs.

Fig. 1
figure 1

Translation of lncRNAs smORFs can lead to RNA degradation or functional peptide production. Translation of smORF (red box) on a transcript annotated as lncRNA might target this transcript to the degradation via the NMD pathway, or alternatively give rise to the production of a small peptide, possibly functional (see examples in main text)

Full size image

NMD as an additional layer in lncRNA-mediated regulation of gene expression

In yeast, antisense XUTs can regulate paired-sense gene expression, at the transcriptional level, through histone modifications (Berretta et al. 2008; Van Dijk et al. 2011) and constitute to date the only class of lncRNAs for which the associated gene-regulation is thought to depend on the lncRNA per se, rather than its transcription. Interestingly, NMD specifically and exclusively targets this class of regulatory lncRNAs (Wery et al. 2016) and might be considered in that way as a novel player in the lncRNA-dependent buffering of genome expression.

NMD not only acts as a surveillance pathway targeting aberrant mRNAs and cryptic lncRNAs to degradation, but it also directly regulates physiological mRNAs in yeast, Drosophila and human (Peccarelli and Kebaara 2014). NMD itself is tightly regulated, and its activity is modulated in response to multiple stresses, including hypoxia, amino-acid or nutrient deprivation (Karam et al. 2013; Lykke-Andersen and Jensen 2015). Interestingly, many stress-related mRNAs are targeted by the NMD under normal physiological conditions but are stabilized upon stress, due to NMD activity inhibition (Lykke-Andersen and Jensen 2015). Note that under stress conditions, global translation also decreases, selectively preserving translation of stress-related mRNAs to the detriment of “housekeeping” mRNAs (Yamasaki and Anderson 2008). Another consequence of such a stress-mediated reduction of translation is that transcripts (aberrant mRNAs or lncRNAs) that are normally targeted by the NMD will be stabilized, since transcripts evading translation escape NMD.

On this basis, we propose a model where a stress that results in translation inhibition and/or NMD inhibition will lead to stabilization of regulatory antisense NMD-sensitive lncRNAs (such as yeast XUTs), which could in turn repress the transcription of their paired-sense genes (Fig. 2). This would prevent the synthesis of mRNAs that could probably not be translated, avoiding the cell to waste an energy that could be crucial to survive the stress. Alternatively but not exclusively, the regulatory antisense lncRNAs could also regulate the paired-sense mRNAs at the post-transcriptional level, potentially through the formation of double-stranded (ds) RNA structures. In yeast, sense mRNAs and antisense XUTs have been shown to form dsRNA in vivo, and this protects XUTs from NMD (Wery et al. 2016). Reciprocally, formation of dsRNA with a stabilized antisense lncRNA might affect sense mRNA stability, as suggested by a recent study of mRNA isoforms half-lives (Geisberg et al. 2014). Besides RNA stability, regulatory antisense lncRNAs might also interfere with mRNA splicing, localization, or translation. Future work will be needed to decipher these potential lncRNA-mediated regulatory mechanisms and determine whether they can be integrated within larger stress-activated signalling networks (Ho and Gasch 2015).

Fig. 2
figure 2

NMD buffers genome expression by controlling levels of regulatory antisense lncRNAs. NMD contributes to genome expression buffering by restricting the levels of a class of cytoplasmic antisense regulatory lncRNAs (red) that can attenuate the transcription of their paired-sense protein-coding genes (blue). In this model, any stress leading to translation and/or NMD inhibition would result in the stabilization of the lncRNA and transcriptional attenuation of the paired-sense gene

Full size image

Concluding remarks

Over the years, lncRNAs have been recognized as major regulators of multiple cellular processes. However, the initial assumption that they are devoid of coding potential is now challenged. Conceptually, coding a peptide/protein in specific circumstances and functioning as a regulatory RNA molecule in others are not exclusive possibilities for a transcript, whatever it has been primarily annotated as a lncRNA or mRNA. Examples of bifunctional RNAs with dual coding and regulatory functions have been reported (Ulveling et al. 2011). In this respect, in yeast, convergent mRNAs can regulate each other at the RNA level providing additional intriguing cases where mRNAs can switch their initial coding function into regulatory RNAs (Sinturel et al. 2015). But how many among the thousands of lncRNAs annotated in the different eukaryotic models correspond to such bifunctional RNAs remains to be determined. The classical distinction between coding and non-coding RNAs might, therefore, become less strict in the future, if the possibility to switch between regulatory and coding functions in response to specific stimuli appears to be a common feature of “lncRNAs”.