Linking RNA biology to lncRNAs

Abstract

The regulatory potential of RNA has never ceased to amaze: from RNA catalysis, to RNA-mediated splicing, to RNA-based silencing of an entire chromosome during dosage compensation. More recently, thousands of long noncoding RNA (lncRNA) transcripts have been identified, the majority with unknown function. Thus, it is tempting to think that these lncRNAs represent a cadre of new factors that function through ribonucleic mechanisms. Some evidence points to several lncRNAs with tantalizing physiological contributions and thought-provoking molecular modalities. However, dissecting the RNA biology of lncRNAs has been difficult, and distinguishing the independent contributions of functional RNAs from underlying DNA elements, or the local act of transcription, is challenging. Here, we aim to survey the existing literature and highlight future approaches that will be needed to link the RNA-based biology and mechanisms of lncRNAs in vitro and in vivo.

Perhaps one of the biggest surprises of the post-genome era is the vast amount of transcription emanating from the noncoding regions of the genome. The human genome sequence has provided a unique opportunity to systematically survey genomic regions for biological activity. As a logical first proxy for activity, many groups began by mapping observed transcriptional events to the genome (Kapranov et al. 2002, 2005; Rinn et al. 2003; Carninci et al. 2005; Cheng et al. 2005; Maeda et al. 2006; Nordström et al. 2009). These early efforts identified thousands of long noncoding RNA (lncRNA) transcripts that did not appear to code for protein coding genes (Okazaki et al. 2002; Carninci et al. 2005; The ENCODE Project Consortium 2007; Kapranov et al. 2007; Guttman et al. 2009; Nordström et al. 2009; Cabili et al. 2011).

This approach quickly expanded the catalog of putative genomic regions in which RNA transcription may serve as a proxy for functional biological activities. Because RNA has a longstanding history of powerful regulatory responsibilities, with numerous classic examples of RNA-mediated roles in splicing, translation, and genomic imprinting, many have speculated that this vast catalog of transcribed noncoding genes represented a wealth of novel regulatory RNA elements. Indeed, lncRNAs are still routinely detected across a variety of cellular- and tissue-specific contexts with many contributing diverse functional roles (Harrow et al. 2012; Morris and Mattick 2014). However, although it is alluring to think that the sheer numbers of reproducibly expressed RNA species must signify an important biological contribution, an equivalent hypothesis is that these RNA molecules are by-products or transcriptional noise. In fact, current evidence suggests that the answer is somewhere in the middle. We define a functional lncRNA locus as being required for a specific cellular activity, but importantly, not constrained by any particular mechanism. In this regard, a functional locus may produce a gene product with a specific activity or serve to harbor regulatory sequences or properties that facilitate proper cellular physiology. Several lines of evidence support the claim that some lncRNAs are functional through RNA-mediated mechanisms, yet the extent and number of important lncRNAs remains highly contested. While this broader question remains unresolved, there is a clear path forward for identifying physiologically important lncRNA loci and their underlying RNA biology.

Identifying functional lncRNAs has been difficult, as it requires multiple lines of evidence that first identify candidate loci that contribute to a physiological function and then confirm that the observed effect is primarily mediated through an RNA-based mechanism. There is currently no universal experimental approach to characterizing lncRNA functional contributions, owing both to the complexities of transcriptional regulation and the diversity of functions that can be attributed to an RNA molecule. In the end, a thorough characterization of a functional lncRNA gene will involve the incorporation of an array of independent experiments, each designed to elucidate either the effect of the gene on cellular physiology or the precise molecular mechanism through which the gene achieves this effect.

An important first step in this process is to determine whether or not a lncRNA gene has any functional consequences within a particular cellular context. As a reasonable first approach to this question, many groups have performed loss- and gain-of-function studies and, as a result, have identified lncRNAs that reproducibly perturb in vitro physiology (Bond et al. 2009; Guttman et al. 2011; Nagano and Fraser 2011; Sun et al. 2013). With functional evidence accumulating, several groups began using mutant mouse models to test the functional roles of lncRNAs at an organismal scale. Taken together, the aggregate of dozens of studies has begun to show that several lncRNAs are critical regulators of cell and developmental biology (Rinn and Chang 2012; Ulitsky and Bartel 2013).

Yet with thousands more uncharacterized lncRNAs remaining, it is critical to take a step back from a serial model of functional lncRNA analysis. In contrast, efforts should be directed toward the identification and dissection of the RNA domains, sequences, structures, and characteristics that may be shared across multiple lncRNAs or other classes of RNAs more broadly. Here, we synthesize current lncRNA literature with a focus on the significance of their underlying RNA-mediated biological activities. For brevity, we have selected a small subset of functional lncRNAs (Table 1) to emphasize specific experimental strategies required to evaluate RNA, DNA, or transcriptional functionalities, or combinations thereof. Readers wishing to explore the current breadth of functional lncRNA genes are directed to http://lncRNAdb.org for an up-to-date catalog (Amaral et al. 2011).

Table 1.

Summary table of lncRNAs highlighted in this review

Transient methods for lncRNA functional determination

Early studies into lncRNA function primarily used cell-based assays as a means to dissect the contribution of a lncRNA to a particular cellular context (Guttman et al. 2011; Nagano and Fraser 2011; Wapinski and Chang 2011; Flynn et al. 2015). Numerous studies have found functional lncRNAs across a broad spectrum of cellular biology (Mattick 2009). Despite the diverse biological contexts, these studies used a common strategy of using RNAi-mediated LoF to screen for phenotypes (Fig. 1A). One advantage of such an approach is the ability to directly test for RNA-mediated function, as presumably, the RNA itself is targeted for degradation. Although there are many caveats with this experimental strategy that require independent confirmation (see below), this has been a powerful and scalable approach to screen for potentially functional lncRNAs. It is important to note that RNAi-mediated targeting is an acute exposure to an LoF context, often only measured for 72 h. In contrast, genetic approaches, as described in detail below, allow for longer-term compensatory mechanisms to arise and potentially recover acute phenotypes.

Figure 1.

Experimental approaches to manipulate the expression, perturb the activity, or evaluate the functions of long noncoding RNAs. Most commonly used are transient expression of exogenous oligonucleotides designed to exploit the endogenous RNAi machinery or RNase H activity to degrade an RNA (A) or occlude putative functional regions (B). Recently, strategies have used the flexible CRISPR/Cas9 system to positively or negatively affect the transcription of a lncRNA gene (C) or the incorporation of an auto-catalytic regulatory RNA element to destabilize the nascent lncRNA transcript (D). An alternative to manipulating lncRNA expression levels involves the recruitment of the RNA of interest to a particular genomic locus or reporter gene through fusion of DNA-binding proteins with RNA-binding elements such as MS2 stem–loops (E) or covalent tethering of a lncRNA transcript to the CRISPR/Cas9 guide RNA (F).

One of the first large-scale lncRNA screens used lentiviral shRNA-mediated LoF screening to survey the impact of 147 lincRNAs (Guttman et al. 2011) on the global transcriptional network of mouse embryonic stem cells. Microarray analysis of each knockdown indicated that the repression of the majority of tested lncRNAs resulted in a significant perturbation effect on the transcriptome. Interestingly, 26 (∼15%) of the lncRNA knockdown assays demonstrated significant reduction in expression of a panel of pluripotency marker genes, indicative of loss of the pluripotent state. Moreover, almost all of these 26 exhibited cellular phenotypes, such as morphology differences, indicative of a loss of pluripotency. Overall, this study demonstrates that lncRNAs can have a range of no-to-strong phenotypic effects in a given cellular context, somewhat on par with what would be expected for protein coding genes. Similarly, a study using RNAi-mediated LoF function to identify lncRNAs required for adipogenesis found 8/38 (∼20%) tested lncRNAs to have a cell based phenotype (Sun et al. 2013).

Transient exposure to targeting oligonucleotides can also be used to evaluate particular RNA-based mechanisms for lncRNAs. For example, Sarma et al. adopted transient exposure of “locked nucleic acid” (LNA) and Protein Nucleic Acids (PNA) oligomers, chemically modified to bind but not degrade the RNA transcript (Beletskii et al. 2001; Sarma et al. 2010). In these studies, LNAs and PNAs were designed against key regions of the XIST RNA thought to be responsible for the localization of XIST RNA at the future inactive X (Xi) Chromosome during dosage compensation. In this manner, XIST is displaced from Xi without the destruction of the RNA itself (Fig. 1B). Both studies found that loss of localization of otherwise intact XIST RNA resulted in loss of X-inactivation. Sarma and colleagues also systematically evaluated which regions of the XIST RNA were required for localization and the kinetics therein across a variety of regions of the XIST RNA to determine whether other “domains” also contributed to XIST localization (Sarma et al. 2010). Together, these and other studies have demonstrated the power of using transient oligonucleotides to target or block lncRNA functionality.

New technologies for the transient manipulation of RNA expression levels are also emerging, the most successful of which utilize the CRISPR/Cas9 system. In one particularly useful iteration, this system can provide a flexible control over RNA transcription rates. CRISPR-inhibition and activation (CRISPRi and CRISPRa, respectively) are two examples of Cas9-mediated activities to directly modulate gene transcription levels (Gilbert et al. 2013, 2014; Qi et al. 2013; Zalatan et al. 2015). These applications involve the coupling of a catalytically inert Cas9 variant with either a strong transcriptional activator (e.g., VP64, a synthetic concatemer of the Herpes Simplex Viral Protein 16, as in CRISPRa) (Maeder et al. 2013) or a transcriptional repressor (e.g., KRAB as in CRISPRi) (Gilbert et al. 2013) to directly modulate the expression levels of a target gene (Fig. 1C).

One major advantage of this approach is that higher abundance of targets is not a factor, as all targeted sites are present at two copies (relative to thousands of fold differences when targeting a lncRNA versus an mRNA directly). However, similarly to RNAi approaches, CRISPR-targeting of genomic loci is also prone to off-target effects. Another confounding property is that any local transcription may be affected, and therefore one may not be able to disambiguate the act of transcription from the function of a mature transcript.

One unique example of a successful alternative approach involved the engineering of a destabilization element at the end of a lncRNA to permit normal transcription while enabling the degradation of the mature transcript. This approach was applied to MALAT1 by adding a sequence into the lncRNA transcript that would be recognized, cleaved, and thus degraded by RNase P (Fig. 1D; Gutschner et al. 2011). As these studies indicate, there is no shortage of effective or creative ways to directly target and manipulate the expression level of a given lncRNA. Few approaches, however, come without significant caveats that can serve to confound the identification of a functional RNA molecule.

Considerations for oligo-based dissection of lncRNA function

The aforementioned studies illustrate the utility of transient LoF approaches to identify and dissect RNA-based mechanisms, and these experiments were key to identifying functional candidates and/or functional RNA domains. Generally speaking, these approaches have been hit-or-miss in their efficacy in targeting lncRNAs. More importantly however, a significant drawback to these approaches is the large number of off-target effects that may confound interpretations (Jackson et al. 2003; Adamson et al. 2012; Sigoillot et al. 2012). It is most important to consider several universally relevant stoichiometric caveats to any oligo-based methodology. One of the biggest concerns with any RNAi or RNase H-mediated depletion of lncRNAs is the significant differences in abundance that exist between mRNAs and lncRNAs; even if the most judicious controls are used (Jackson et al. 2003; Adamson et al. 2012; Sigoillot et al. 2012). For example, the lncRNA DALIR was recently detected at two molecules per cell (Chalei et al. 2014), and shRNA-mediated depletion of this RNA molecule resulted in the significant differential regulation of more than 200 genes in N2A cells. Given the low abundance of DALIR transcripts in this context and the excess of targeting shRNA molecules, the likelihood of off-target affects rises by orders of magnitude as has been previously suggested (Bassett et al. 2014).

Indeed, if targets are dramatically different in stoichiometric abundance with potential subcomplementary region, significant artifacts can arise. This was illustrated using single-molecule RNA FISH approaches. A single 8-nucleotide match in one of dozens of probes (25 bp in length) targeting one particular lncRNA also matched NEAT1 (at ∼1000-fold higher abundance) (Cabili et al. 2015). Strikingly, this resulted in a staining pattern similar to that of NEAT1. Thus, a single 8-mer off-target match of one oligo was able to out-compete all the other probe signals. More generally, oligo-based LoF approaches present even greater challenges in determining “seed”-based off-target effects (Gutschner et al. 2011; Cabili et al. 2015), especially when considering the relatively low abundance of their presumptive lncRNA targets.

With these caveats in mind, it is still possible that low abundance RNA molecules, such as the RNA component of telomerase (TERC), which can perform multiple turnover reactions, could accomplish super-stoichometric functionalities (Zappulla and Cech 2004; Mozdy and Cech 2006). In these instances, however, a brute-force LoF approach, such as oligo-based targeting, may not have the desired sensitivity or specificity to ascribe or dissect a precision function.

Ectopic RNA localization: a direct test for RNA function

One way to directly test the role of an RNA is to ectopically localize the RNA via RNA:Protein cognates—protein domains that recognize DNA and RNA domains that recognize fusions to the protein (Fig. 1E). An elegant study using a similar conceptual strategy was applied to the lncRNA NEAT1 (Mao et al. 2010) that plays an important functional role in paraspeckle formation (Chen and Carmichael 2009; Sunwoo et al. 2009). This is a nuclear “compartment” or “domain” associated with RNA editing and other functions. In this landmark study, the authors ectopically expressed inducible and visible NEAT1 RNA at a specific genomic location defined by an array of LacO DNA binding motifs. This approach allowed imaging of the DNA locus, paraspeckle protein localization and RNA accumulation relative to the site of transcription in living cells. The authors observed that the mature NEAT1 RNA was required to maintain the paraspeckle but only the act of transcription could establish it.

These studies have been informative but are often laborious, using genetic alterations and in ectopic locations. A recent advance has made localizing RNA to specific genomic locations much easier to test the sufficiency of RNA-based mechanisms. A technology termed CRISPR-Display allows lncRNAs to be covalently connected to short guide RNA (sgRNA) sequences to enable Cas9-mediated localization systems (Shechner et al. 2015). In this manner, a noncoding RNA can be localized to a specific genomic location, via sgRNA complementarity, to evaluate what, if any, biological activities might be attributed to the mature RNA itself (Fig. 1F). As a proof of principle, CRISPR-Display was developed to localize lncRNAs to reporter genes both endogenously and on transient reporters. This study demonstrated that when lncRNAs are directly localized to a particular genomic locus, some have modest but significant influences on activation and repression of targeted genes as predicted from previous studies (Shechner et al. 2015).

Although ectopic RNA localization can directly test an RNA's functional role, independent of the act of transcription, several caveats should still be considered, such as (1) the local chromatin environment of the targeted gene locus at which the RNA is presented (e.g., heterochromatin may prevent access to targeted site resulting in false negatives); (2) the stoichiometric consideration of the number of RNA molecules required at a locus to perform its function; and (3) the position of the presented RNA relative to the target.

It should be increasingly clear that it is important to consider multiple approaches to disentangle the precise mechanism of action of a putative lncRNA gene. We next explore this issue in more detail with respect to testing the functionality of a given lncRNA locus.

Assessing the biology of lncRNAs in vivo

After accumulating evidence for RNA-based mechanisms for lncRNAs using approaches described above, the next major challenge is determining the physiological relevance of these genes in vivo. Historically, this has been achieved through the generation of mutant animal models and investigation of organismal-scale phenotypes. Many of the first discovered lncRNAs (e.g., H19, XIST, AIRN) were investigated using a battery of genetically modified gain- and loss-of-function mouse models. Numerous additional lncRNA loci from more modern catalogs are currently being tested for in vivo functional roles as well. These studies have found that, much like mRNA genes, many lncRNAs appear dispensable, whereas others can have profound contributions to biology (Nakagawa et al. 2012; Sauvageau et al. 2013; Oliver et al. 2014). Despite their strength, genetic approaches can also be confounding, even in combination, and care must be taken in the interpretation of both positive and negative results.

In vivo lncRNA loss of function

One reasonable genetic approach to lncRNA characterization is whole-gene ablation (Fig. 2A). Indeed, this approach has been adopted to evaluate thousands of mRNAs (Austin et al. 2004; The International Mouse Knockout Consortium 2007) and in turn, has been the predominant means by which phenotypes have been attributed to a given gene. When combined with the introduction of a reporter gene at the target locus (Fig. 2A; Valenzuela et al. 2003), this approach serves to control for the act of transcription at that genomic locus and enables the identification of the spatial-temporal expression of the lncRNA locus in vivo, further reducing the space for phenotypic investigation (Sauvageau et al. 2013; Lai et al. 2015). Moreover, absence of a phenotype in whole gene ablation strategies is a strong indicator that the lncRNA locus is nonfunctional or dispensable (or highly context specific in function). As an example, for the well-conserved lncRNA C130071C03Rik (the mouse ortholog of human LINC00461), whole-gene ablation was found not to have a function (Oliver et al. 2014). Yet it is important to consider that the absence of evidence for an overt phenotype should not necessarily serve as evidence of absence.

Figure 2.

Genetic approaches to evaluate lncRNA functional contributions. (A) Genetic ablation of a lncRNA gene or a lncRNA promoter is a powerful technique that can be used to confirm any regulatory activity arising from a particular locus. The optional incorporation of a reporter gene at the deleted locus can be used to visualize the biological contexts in which the lncRNA gene is expressed and evaluate any contribution of the local act of transcription to an observed phenotype. (B) An alternative strategy is to introduce a premature transcriptional terminator sequence, which will prevent transcription of the full-length lncRNA transcript. With this approach, however, it is not possible to assess the contribution of the act of transcription. (C) With either approach, the rescue of any observed phenotype by transgene expression is currently still considered the gold standard to confirm a functional lncRNA molecule; however, a rescue may not be possible if the RNA works exclusively in cis.

With ablation of any large piece of DNA (e.g., mRNA, lncRNA, enhancer) it is important to consider the possible overlapping molecular contribution(s) of DNA, RNA, and transcription on the observed phenotype (Bassett et al. 2014). Thus, to properly disentangle these attributes requires multiple genetic models to discern between the individual contributions of each mechanism at a given lncRNA locus.

Another in vivo LoF strategy involves generating mutants with transcriptional terminators inserted into lncRNA gene bodies (Sleutels et al. 2002; Bond et al. 2009; Grote et al. 2013). Here, multiple copies of a polyadenylation signal are introduced at the lncRNA locus toward the 5′ end of the gene (Fig. 2B). In this manner, transcription through the gene is halted, and an abortive, presumably nonfunctional RNA product is produced. Due consideration must be given to both of these consequences. For example, Grote et al. (2013) used premature polyA termination to determine that the Foxf1a-adjacent lncRNA Fendrr is required for viability in mice. The authors further demonstrated a transgene rescue of the Fendrr RNA, strongly suggesting the RNA product is the functional element. Interestingly, different approaches targeting the same lncRNA locus may result in identifying similar phenotypic contributions. By way of example, the Fendrr locus has been subjected to both polyA termination (Grote et al. 2013) as well as whole gene ablation (Sauvageau et al. 2013). In both instances, a similar phenotype of decreased viability was observed.

This and other studies demonstrate that transcriptional termination is a viable approach to functional characterization of a noncoding RNA locus, but cannot be used in isolation to definitively determine the precise mechanisms of action. A positive result from a transcriptional termination assay may indicate a functional RNA product. However, an alternative hypothesis could be that transcription simply must occur at the locus for its activity. In either case, additional validations are needed to confirm the mechanism.

The AIRN locus is a primary example of how multiple approaches were required to determine the functional aspect of the locus. It has recently been revealed that the imprinting activity regulated by the lncRNA AIRN is a function of the local act of transcription across the locus and not mediated by the transcript itself (Latos et al. 2012; Santoro et al. 2013). Prior to this, it was thought to have an activity that was attributed to a functional RNA molecule. Sleutels et al. (2002) disrupted transcription through the AIRN locus and observed a loss of imprinting phenotype. Similarly, Nagano et al. (2008) conducted both a genetic ablation and a truncation assay and noted a significant reduction in G9A recruitment and also a loss of imprinting. These observations, combined with an earlier study demonstrating that transgene expression of the AIRN RNA was insufficient to induce imprinting (Sleutels and Barlow 2001) were initially used as evidence of a functional AIRN RNA acting in cis. More recent studies have demonstrated that antisense transcription of AIRN across the IGF2R promoter, regardless of the RNA product, is the specific functional activity that results in imprinting of this locus (Latos et al. 2012).

In vivo gain of function: transgene expression

lncRNA rescue or gain-of-function assay by transgene expression has in many ways been considered the “gold standard” to prove that a lncRNA molecule has a direct functional role (Fig. 2C). Examples of lncRNAs with demonstrated activity in trans include Fendrr (Grote et al. 2013), Evf2 (Bond et al. 2009), and Tmevpg1 (Gomez et al. 2013). In this last example, transgenic expression of Tmevpg1 RNA in both CD4⁺ and CD8⁺ T-cells was sufficient to confer a protective phenotype against Salmonella infection through activation of the TMEVP3 locus (Gomez et al. 2013). Yet again, there remain many possible interpretations. A lncRNA gene whose primary function is exerted locally in cis will not be able to rescue via trans expression, resulting in a false negative.

Conversely, a positive result in a trans rescue assay may also be falsely attributed to a functional RNA and rather due to novel small protein coding genes. Emerging evidence suggests that several lncRNAs may harbor cryptic micropeptides (Slavoff et al. 2013; Sun et al. 2013; Bazzini et al. 2014; Anderson et al. 2015). A recent study identified a conserved 46 amino acid micropeptide myoregulin (MLN) that was initially found within an annotated muscle-specific lncRNA gene and is an important regulator of Ca²⁺-mediated contractile machinery in skeletal muscle (Anderson et al. 2015). As in this particular case, the expression and translation of a cryptic, functional peptide within a lncRNA gene may in fact be the functional element resulting in an observed phenotype. With this in mind, many lncRNA catalogs have made concerted efforts to filter out lncRNAs with small peptides that are easily detectable by codon substitution frequency analysis (Lin et al. 2011), and genes potentially harboring such peptides are referred to as transcripts of unknown coding potential (TUCP) (Cabili et al. 2011; Guttman and Rinn 2012; Molyneaux et al. 2015).

The preceding examples demonstrate that no one approach to functionally characterize a lncRNA locus can uniquely satisfy all possible models. Rather, this effort should involve the coordinated use of multiple approaches to identify the often multiple overlapping functional modalities of a gene locus. One excellent example of a detailed, systematic approach is illustrated in a recent study that used multiple genetic models to dissect the HAUNT lncRNA locus. Intriguingly, they found that DNA, RNA, and transcription all play an important role in regulating HOX gene expression (Yin et al. 2015). Deletion of the HAUNT genomic locus resulted in a significant reduction in the expression and activation of HOXA genes. In striking contrast, however, reduction of HAUNT RNA by either RNAi or TSS inclusion resulted in an increase in HOXA gene expression. Furthermore, overexpression of HAUNT RNA in cis, but not in trans, was sufficient to down-regulate many of the HOXA genes as well. Using this suite of approaches, the authors are able to conclude that there are independent, and opposing, functional activities conferred by both the HAUNT genomic DNA, and the HAUNT RNA. Intriguingly, another study identified a functional role for the lncRNA in repression of HOXA1 in cis (Maamar et al. 2013). These studies underscore the potential complexity of any given genomic locus and highlight the multifaceted approach that should be used to thoroughly evaluate the functional activity of a lncRNA genomic locus.

A classic example bringing all these principles and approaches together is the XIST locus that controls X Chromosome dosage compensation. From this locus, a 17 kb transcript is sufficient to silence the majority of an entire X Chromosome. In many ways, XIST can be considered a model example of lncRNA biology, and the dissection of its mechanism of action can serve as a guide for future lncRNA studies. Principally, ablation of the DNA and promoter of XIST results in loss of X Chromosome inactivation (Penny et al. 1996), yet these models cannot rule out the contribution of the local act of transcription. It took additional experiments (Beletskii et al. 2001; Sarma et al. 2010) that blocked the localization region of XIST RNA (REPC) with a complimentary modified oligo that maintained transcription, preserved the XIST genomic DNA, and kept the mature transcript intact, but resulted in reactivation of the X Chromosome, thus demonstrating a mechanism that is dependent on the mature XIST RNA localizing to the inactive X Chromosome in cis. Thus, both genetic and molecular biology experiments were required to demonstrate, and specifically ascribe, XIST function to the activity of a functional RNA molecule.

A path forward toward common modalities of ncRNA biology

RNA has a long history of breaking the rules of biology through diverse and fascinating molecular modalities (Wapinski and Chang 2011; Cech and Steitz 2014; Rinn and Guttman 2014). Yet, RNA does almost always obey one rule, i.e., form ribonucleic protein complexes to carry out functional roles. Despite this property, we know very little about the common RNA sequences, structures, and domains of lncRNAs that facilitate these associations. In sharp contrast, primary sequences of many proteins are predictive of their functionality. For example, if a protein harbors a bHLH domain (Fig. 3), it is likely to be involved in DNA binding. Similarly, other protein motifs might suggest functional activities such as specific catalytic domains. Yet, almost no such common features have been identified for lncRNAs.

Figure 3.

Functionally equivalent RNA and DNA domains. Despite distinct mechanisms by which they achieve their equivalence, RNA domains are capable of affecting many biological activities that have traditionally been ascribed exclusively to proteins. Much work remains, however, in identifying common RNA primary, secondary, and tertiary structures that comprise orthologous functional domains and can mediate these and other subcellular activities.

One complication in identifying such domains may be the inherent flexibility of RNA and its unique ability (as compared to DNA) to form complex secondary structures or interact with a diversity of protein partners. Much of our understanding of motifs has arisen from analyses of primary sequence, and many of the tools used to define primary sequence motifs do not translate to more complex two- and three-dimensional structures. However, many current efforts are underway to develop techniques to resolve RNA 2D structures en masse (Wilkinson et al. 2008; Kertesz et al. 2010; Mortimer et al. 2012; Siegfried et al. 2014; Spitale et al. 2014; Wan et al. 2014). Additionally, classic biochemistry tools, such as RNA footprinting, are being adapted to genome-scale approaches (Silverman et al. 2014) to identify RNase-protected regions of the transcriptome that may represent sites of RNA:protein interactions. The data generated by these and other efforts will provide important insight into potential common structural and functional motifs for RNA.

Here, we summarize by discussing a few of the types of RNA domains that have been identified that have similar functions to proteins. By discussing just a few of these examples, we hope to highlight a logical framework toward unraveling the molecular grammar of lncRNAs.

Subcellular localization signals

Both RNA and proteins harbor endogenous sequences and structures that provide direction to their appropriate subcellular location. These localization queues (Fig. 3, first row), such as the signal peptide or nuclear localization signal for proteins, have recently been described in different noncoding RNA sequences. The lncRNA, BORG, harbors a pentamer RNA motif that is required for its proper nuclear retention (Zhang et al. 2014). A similar functional motif can be observed within the RRD repeat domain of the lncRNA, FIRRE (Hacisuleyman et al. 2014), which is also responsible for the nuclear localization of this lncRNA. In the future, it will be important to catalog the RNA localization sequences and structures within lncRNAs toward understanding shared modalities.

Molecular scaffold

Both proteins and RNAs exhibit the ability to serve as a scaffold for larger complexes (Fig. 3, second row). Protein scaffolding complexes such as the A-kinase anchor proteins (AKAP) (Diviani et al. 2011) use a specific EVH1 domain to tether and complex key neuronal proteins (Luo et al. 2012). Similarly, RNA is commonly used as a platform to scaffold many proteins for a related function. The ribosome is a dramatic example of RNA scaffolding structural organization of RNA-protein interactions, as well as catalytic activity to enable translation (Brimacombe and Stiege 1985). This and many other classic examples demonstrate the powerful roles of RNA scaffolds (Zappulla and Cech 2006). More recently, several long noncoding RNAs such as HOTAIR, XIST, FIRRE, and many others (Chaumeil et al. 2006; Tattermusch and Brockdorff 2011; Wang and Chang 2011; Yoon et al. 2013; Hacisuleyman et al. 2014; Quinn et al. 2014; Chu et al. 2015a,b; McHugh et al. 2015) have also been identified to function as RNA scaffolds that interface with and influence epigenetic regulatory complexes.

Nucleic acid interaction and targeting

Some of the earliest defined protein domains to bind directly to DNA were the basic helix-loop-helix (bHLH) domain (Murre et al. 1989) and the leucine zippers (Fig. 3, third row; Landschulz et al. 1988). Likewise, RNA possesses an apt property in that it can readily hybridize to nascent DNA via sequence complementarity, or DNA:DNA:RNA triplexes, to target specific genomic loci. Indeed, such examples have been observed for lncRNAs such as Fendrr (Grote et al. 2013), PRNA (Schmitz et al. 2010), and the recently described lncRNA PARTICLE (PARTICL) (O'Leary et al. 2015). In much the same manner, RNA–RNA interactions can occur through direct sequence complementarity and regulatory roles (Fig. 3, fourth row; Kretz et al. 2013; Engreitz et al. 2014; Tay et al. 2014). To this end, lncRNAs can also serve as host transcripts for various classes of small RNAs, including microRNAs (da Rocha et al. 2008; Royo and Cavaillé 2008; Augoff et al. 2012; Sabin et al. 2013).

RNA catalysis

Perhaps the most exciting yet elusive property RNA shares with protein is catalysis (Fig. 3, fifth row). Although the vast majority of catalytic activity appears to be mediated by proteins, there is increasing evidence for a number of mammalian catalytic RNAs. Most notably, this list includes the self-cleaving group I introns (Cech 1990); the RNase P M1 RNA, which catalyzes tRNA biosynthesis (Guerrier-Takada et al. 1983); the ribosome (Brimacombe and Stiege 1985); and others involved in RNA processing (Teixeira et al. 2004; Winkler et al. 2004; Strobel and Cochrane 2007).

These examples, and the methods used to discern their functional activities, serve as a conceptual framework for how to identify functional RNA sequences and domains. We propose that efforts within the lncRNA community should proceed with similar strategies (as have been applied extensively for protein-coding genes and the original noncoding RNAs classes) and in parallel, adopt and embrace new experimental designs that appropriately reflect the complexities of RNA biology. An important next step forward will be identifying the additional RNA-mediated localization signals, scaffolding motifs, protein-guidance cues, and catalytic domains that we already know RNA molecules are capable of harboring. It will take creative, diverse, and collaborative multifaceted approaches to determine the RNA-based functional importance of any single lncRNA. Here, we have summarized emerging approaches and their considerations toward the goal of unlinking the underlying ribonucleic logic of lncRNAs.