The rise of regulatory RNA

Abstract

Discoveries over the last decade portend a paradigm shift in molecular biology. Evidence suggests that RNA is not only functional as a messenger between DNA and protein but also in the regulation of genome organization and gene expression, which is increasingly elaborated in complex organisms. Regulatory RNAs appear to operate at many levels, but in particular to play an important role in the epigenetic processes that control differentiation and development. These discoveries suggest a central role for RNA in human evolution and ontogeny. Here we survey the emergence of the previously unsuspected world of regulatory RNAs from an historical perspective.

Introduction

RNA has long been at the centre of molecular biology and was likely the primordial molecule of life, encompassing both informational and catalytic functions. It is thought that its informational functions were subsequently devolved to the more stable and easily replicable DNA, and its catalytic functions to the more chemically versatile polypeptides¹. The idea that the contemporary role of RNA is to function as the intermediary between the two had its roots in the early 1940s with the entry of chemists into biology, notably Beadle and Tatum², whose work underpinned the “one gene-one enzyme” hypothesis, which later morphed into the more familiar phrase “one gene-one protein”, and gained currency despite the prescient misgivings of experienced geneticists, notably McClintock³. The concept that genes encoded (solely) the functional components of cells (the ‘enzymes’) itself had deeper roots in the mechanical zeitgeist of the era, being decades before the widespread understanding of the use of digital information for systems control.

Although the one gene-one protein hypothesis has long been abandoned, due to the discovery of alternative splicing in the 1970s, the protein-centric view of molecular biology has persisted, aided by phenotypic and ascertainment bias towards protein-coding mutations in genetic studies and by the assumption that regulatory mutations affected cis-acting regulatory protein binding sites⁴. However, this view is challenged by the discovery of nuclear introns and the phenomenon of RNA interference (RNAi), as well as by the advent of high throughput sequencing, which led to the identification of large numbers and different types of large and small RNAs, whose functions are still under exploration.

Here we examine the history and chart the shift in thinking that is still underway about the role of RNA in cell and developmental biology, especially in animals. The emerging evidence suggests that there may be more genes encoding regulatory RNAs than encoding proteins in the human genome, and that the amount and type of gene regulation in complex organisms has been substantially misunderstood for most of the past 50 years.

Early ideas for the role of RNA

RNA, the central dogma and gene regulation

After the elucidation of the double-helical structure of DNA in 1953⁵, the following years were preoccupied with deciphering the ‘genetic code’ and establishing the mechanistic pathway between gene and protein: the identification of a transitory template (mRNA), an adaptor (tRNA) and the ‘ribosome’ factory comprised of ribosomal proteins and RNA (rRNA) for the translation of the code into a polypeptide. In 1958, Crick published the celebrated ‘central dogma’ to describe the flow of genetic information (DNA → RNA → protein), which has proved remarkably accurate and durable, including the prediction of reverse transcription⁶. Nonetheless, in conceptual terms, RNA was tacitly consigned to be the template (and infrastructural platform – ribosomal and transfer RNAs) for protein synthesis or, at least, has been interpreted in this way by most people since that time.

The link between rRNA (which is highly expressed in virtually all cells) and the structures termed ribosomes as the platform for protein synthesis was established in the mid 1950s⁷. The roles of tRNA and mRNA were experimentally confirmed in 1958⁸ and 1961⁹, respectively, the latter the same year that Jacob and Monod published their classic paper on the lac operon of Escherichia coli¹⁰, the first locus to be characterized at the molecular genetic level. These studies confirmed that (at least some, but presumed most) genes encoded proteins, and supported the emerging idea that gene expression was controlled by regulating the transcription of the gene, as indicated by the locus encoding the lac repressor – the repressor-operator model. At the time Jacob and Monod did not know the chemical identity of the repressor, speculating en passant that it “may be a polyribonucleotide” (i.e., RNA)¹⁰. However, Gilbert later showed that the repressor was a polypeptide that allosterically bound the substrate lactose, and the brief idea faded¹¹.

These studies reinforced and extended the conception that proteins are not only ‘enzymes’ but also the primary analogue components and control factors that comprise the cellular machinery. This in turn has led to the prevailing ‘transcription factor’ paradigm of gene regulation, including the derived assumption that combinatoric interactions would provide factorially ‘explosive’ regulatory combinations¹², more than enough to supervise human ontogeny. However, this assumption has not been substantiated theoretically or mechanistically, and the observed scaling of regulatory genes and the extent of the regulatory challenge in programming human developmental architecture appears to be quite different from these expectations¹³. In this context it is noteworthy that the genome-wide association studies have shown that most haplotype blocks influencing complex diseases fall outside the known boundaries of protein-coding genes¹⁴.

Small nuclear / spliceosomal RNAs and small nucleolar RNAs

Following the discovery and functional description of tRNAs and rRNAs, other new classes of common small nuclear RNAs were identified by biochemical fractionation¹⁵. Many of the small RNAs were found to be part of ribonucleoprotein complexes (RNPs) (reviewed in ¹⁶). One class, the small nuclear RNAs (snRNAs, Figure 1), were later found to be central co-factors in RNA splicing (see below)¹⁷, hence their newer designation as ‘spliceosomal’ RNAs. The snRNAs U1, U2, U4, U5 and U6 participate in a number of RNA-RNA and RNA-protein interactions in the assembly and function of canonical spliceosomes, recognizing the 5′-end splice site (U1) and the branch point (U2), followed by the recruitment of U4, U5 and U6, which displace U1 and interact with U2 (via U6) as well as the 5′ and 3′ splice sites (via U5)¹⁸. A set of less abundant snRNAs (U11, U12, u4atac and U6atac) along with U5 are found in a variant ‘minor’ spliceosome, termed U12-type¹⁹.

Figure 1. Complex expression of the genome and examples of non-coding RNA expression.

Graphical representation of the mammalian transcriptional landscape with genes expressing rRNAs, tRNAs, snRNAs, snoRNAs, various protein coding and non-coding transcripts (mRNAs and lncRNAs), as well as small regulatory RNAs including miRNAs, piRNAs, tiRNAs and spliRNAs, snoRNA-derived small RNAs, and tRNA-derived small RNAs.

Other small RNAs were found to be localized to nucleoli and to guide the methylation (the box CD subclass) and pseudouridylation (the box H/ACA subclass) of ribosomal RNAs (rRNAs), transfer RNAs (tRNAs), and snRNAs^20–22 (Figure 1). The chemical modifications of rRNAs, tRNAs and snRNAs proved to be essential in ribosomal and cellular function, in particular tRNA and mRNA maturation and pre-mRNA splicing (U2). Notably, the disruption of snoRNAs was found to lead to a loss of processing of the 18S, 5.8S, and 25S/28S rRNAs²⁰. Although some snoRNAs are subject to parental imprinting and/or differentially expressed, for example in the brain^23,24, and appear to target a wider range of RNAs including mRNAs²⁵, suggesting a regulatory role, and there are related small RNAs (scaRNAs) in other subnuclear structures called Cajal bodies (which also process telomerase RNA)²⁶, none of these early studies suggested anything other than that the role of RNAs was limited to protein synthesis.

The emergence of heterogeneous nuclear RNAs

The first hint that RNA may have additional roles in complex organisms was the discovery of ‘heterogenous’ nuclear RNA (hnRNA)²⁷ and the observation that the complexity of this population, as determined by denaturation-renaturation hybridization kinetics, was much greater in the nucleus than the cytoplasm. The existence of hnRNA and the concomitant discovery of the large amount of ‘repetitive’ sequences (different classes of retrotransposed sequences with similar composition that occupy large fractions of plant and animal genomes) led Britten & Davidson to speculate in 1969 that animal cells might contain extensive RNA-based regulatory networks^28–30. While this hypothesis attracted a great deal of interest at the time, it also quickly lapsed, with the proponents not re-visiting it even after the subsequent discovery of introns (see below), instead focussing on regulatory networks controlled by transcription factors^31,32 or the significance of transposons in protein evolution³³.

The discovery of introns

The discovery of introns in 1977^34,35 was perhaps the biggest surprise in the history of molecular biology³⁶, as no one expected that the genes of higher organisms would be mosaics of coding and noncoding sequences, all of which are transcribed. However, the prevailing conception of the flow of genetic information was not overly disturbed as the removal of the intervening sequences (‘introns’) and the reconstruction of a mature mRNA by RNA splicing preserved the conceptual status quo. That is, genes still made proteins. In parallel, it was assumed that the excised intronic RNAs were simply degraded, although the technology of the time was too primitive to confirm this. In any case, introns were immediately and universally dismissed as genomic debris, and their presence rationalized as evolutionary remnants involved in the prebiotic modular assembly of protein-coding RNAs that have lingered (and been expanded by transposition) in complex organisms³⁷. This was consistent, superficially at least, with the implications of the C-value enigma that eukaryotes contained varying amounts of DNA baggage, and the accompanying conclusion that retrotransposon sequences, often pejoratively referred to ‘repeats’, which occupy much of the genomic real estate in plants and animals, are largely selfish DNA that are parasitic co-travellers^38,39.

RNA as a catalyst

A few years later, Cech, Altman and colleagues showed that RNA itself was capable of enzymatic catalysis (‘ribozymes’)^40,41, which provided evidence in support of the ‘RNA early’ hypothesis, and showed that RNA catalysis exists and has persisted in particular contexts, notably at the core of RNA splicing⁴² and mRNA translation⁴³. This reinforced the mechanical conception of molecular biology, and the role of RNA as the platform for protein synthesis, but did not give any hint of RNA as a widespread regulatory factor, although that possibility is perfectly feasible. Indeed there is increasing evidence that catalytic RNA exists in animal and plant cells, in introns, UTRs and elsewhere, and may play a variety of roles including, for example, in regulating post-transcriptional cleavage reactions^44,45.

The small RNA revolution

The discovery of microRNAs

In 1993 Ambros and colleagues showed the first evidence for small (~22 nt) regulatory RNAs, by the discovery of the genetic loci lin-4 and let-7, which regulate the timing of Caenorhabditis elegans development^46,47. Although let-7 is highly conserved from nematodes to humans⁴⁸, very few miRNAs have been discovered genetically^49,50, and these RNAs remained interesting idiosyncrasies until the discovery of RNAi (see below), which led to the targeted cloning after size selection of many more^51–53 and the demonstration that these ‘microRNAs’ (miRNAs) act, at least in part, by imperfect base pairing with (usually) the 3′UTRs of target mRNAs to inhibit their translation and accelerate their degradation⁵⁴.

Currently, there are large numbers of evolutionarily widespread miRNAs in the databases⁵⁵, almost all of which had evaded prior detection by genetic screens (but many subsequently validated by reverse genetics). While many miRNAs can be identified by conservation, it is also evident that many are tissue- and lineage-specific^56,57, and that there may be many more to be discovered.

It has also become evident that many if not most protein-coding transcripts are targets for miRNA regulation^58,59, that miRNAs can, in some cases, regulate large numbers of target mRNAs⁶⁰, and reciprocally that many mRNAs contain target sites for many miRNAs⁶¹, although the implied regulatory logic of this complex multiplex arrangement has not been explained. The targets of miRNAs are usually thought to be mRNAs, but may also include other RNAs⁶². Biologically, miRNAs have been shown to regulate many physiological, developmental and disease processes, including, for example, pluripotency⁶³, epithelial-mesenchymal transition and metastasis⁶⁴, testis differentiation⁶⁵, diabetes⁶⁶, and neural plasticity and memory⁶⁷, among others⁶⁸.

The RNA interference pathway

MicroRNAs are just one facet of the phenomenon of ‘RNA interference’ (RNAi), which causes silencing of gene expression after the introduction of sense-antisense RNA pairs, discovered in 1998 in plants⁶⁹ and C. elegans⁷⁰. These discoveries were presaged by the curious phenomenon of transgene silencing, mainly in plants^71,72, linked to antisense RNA and small RNA-directed DNA methylation of transgenes, indicating transcriptional as well as post-transcriptional silencing^73,74. Mechanistic analysis of these silencing mechanisms showed that exogenous double-stranded RNA was processed into short fragments (short interfering RNAs or siRNAs) with a similar size to miRNAs, suggesting that miRNAs may represent a similar endogenous system.

This was confirmed and led to the elucidation of natural double-stranded precursors in stem-loop structures⁷⁵ and the identification of key genes and enzymes involved in their biogenesis and function, notably Drosha⁷⁶, Dicer⁷⁷ and multiple Argonaute (Ago) proteins⁷⁸, the latter of which were already known to play central roles in differentiation and development,⁷⁹ but are now known to also be involved in defence against RNA viruses in many organisms⁸⁰. Drosha and Exportin 5 are involved in the cleavage and export of double-stranded RNA (dsRNA) precursors from the nucleus to the cytoplasm⁷⁶, where they are further processed by Dicer to small (21–24 nt) dsRNA moieties, one strand of which is loaded into Ago component of the RNA-induced silencing complex (RISC), which also contains other proteins⁷⁷. The RISC is guided by the small RNA strand to complementary RNA targets, which are subsequently silenced by translational repression and/or RNA destabilization^81,82 (Figure 2).

Figure 2. Functional pathways of small regulatory RNAs.

(A) miRNA precursors are expressed as stem-loop structures⁷⁵, which (B) interact with Drosha⁷⁶ and DGR8, where they are processed then exported from the nucleus by Exportin 5. (C) These transcripts are further processed by Dicer to small (21–23 nt) dsRNAs, one strand of which is loaded into AGO component of the RNA-induced silencing complex (RISC). Exogenously introduced siRNAs can also be processed by RISC. Either the endogenous miRNA or exogenously added siRNAs can then to target (D) the repression of translation and/or (E) cleavage of homology containing transcripts^81,82. Some small RNAs are functional in the nucleus. (F) Exogenously introduced small antisense RNAs (asRNAs) can target epigenetic silencing of targeted loci^88,344,345, a pathway that miRNAs may also utilize in the nucleus⁹². (G) tiRNAs and spliRNAs^121,122 are also expressed through an unknown pathway that may involve RNAPII backtracking and TFIIS cleavage¹²³, with the tiRNAs shown to modulate CTCF chromatin localization and to be associated with nucleosome position¹²⁴.

While still under discussion, the current view is that siRNAs (and ‘short-hairpin RNAs’, shRNAs), which seem to naturally occur more commonly in plants, act primarily by perfect base-pairing and by Ago-mediated cleavage of complementary target RNAs (and hence are used widely as experimental tools and potential therapeutic agents⁸³), whereas miRNAs have incomplete homology with their targets and act primarily at the translational level^81,82 (Figure 2).

MiRNAs and siRNAs are thought to act post-transcriptionally and cytoplasmically, but the existence of Ago in the nucleus^84–87 and the role of the RNAi pathway in epigenetic modulation⁸⁸ suggests that the system is more complex and multifaceted than expected, with (for example) demonstrations that miRNA isoforms are developmentally regulated⁸⁹ that the target ‘seed’ sequence is only one factor in target recognition^90,91, and that miRNAs can also act to impose transcriptional silencing⁹² (Figure 2). There is also increasing evidence of intersecting pathways, such as RNA editing and modification, in these networks^93–96.

Piwi-associated small RNAs

While most Ago proteins are expressed ubiquitously and associate with miRNAs and siRNAs, there is a subclade of Argonaute proteins, termed Piwi, that is required for germ cell development^97–100. Piwi and Piwi-like proteins associate with a distinctive class of small RNAs (26–30 nt; piRNAs), which act to epigenetically and post-transcriptionally silence transposons in germ cells^101–110. Piwi is found predominantly in the nucleus¹¹¹, co-localizes in an RNA-dependent manner with Polycomb group proteins¹¹², and appears to be expressed in other tissues, including the brain¹¹³, suggesting a role beyond genome protection in epigenetic processes^114,115.

Other classes of small RNAs in eukaryotes

The molecular genetics, biochemistry and structural biology of the RNAi system are still being unravelled, but indicate an ancient, widespread and multilaterally adapted system that controls many cellular processes, whose dimensions are still being explored. These include potentially lineage-specific variations, such as the ‘21U’ RNAs in C. elegans¹¹⁶. Surprisingly, it appears that all snoRNAs from fission yeast to human produce at least 3 different subclasses of small RNAs¹¹⁷, one of which has the same size and functions as a miRNA¹¹⁸, and another that is similar in size to piRNA¹¹⁷. There are also intriguing and recurring reports of fragments of tRNAs produced in tissue-specific patterns¹¹⁹ and associated with Ago proteins¹²⁰.

More recently, deep sequencing of small RNA populations has revealed the existence of another class of small RNAs in animals but not plants, which are 17–18 nt in length and associated with transcription initiation (‘tiRNAs’)¹²¹ and splice sites (‘spliRNAs’)¹²² (Figure 2). The origin and function of these RNAs is uncertain, but preliminary evidence suggests that they may play a role in nucleosome positioning¹²³ and/or be involved in other levels of chromatin organization¹²⁴. There are also other reports of less distinct classes of promoter-associated RNAs called PASRs¹²⁵, TSSa-RNAs¹²⁶ and PROMPTS¹²⁷, some of which may play a role in RNA-directed transcriptional gene silencing¹²⁸.

Regulatory RNAs in prokaryotes

Many small regulatory RNAs (sRNAs) have been identified in bacteria, which regulate a wide variety of adaptive responses. Bacterial sRNAs generally function by simple antisense mechanisms to regulate translation or stability of target mRNAs by altering their secondary structure to expose or sequester cis-acting sites^129,130. Studies in bacteria have also identified cis-acting regulatory RNA sequences (‘riboswitches’), which act allosterically by binding metabolites to regulate gene expression^131,132, and which almost certainly exist as part of the RNA regulatory landscape in all kingdoms of life.

Very recently, the prokaryotic kingdom has once again surprised us with the sophistication of its molecular machinery. Many bacterial and most archaeal genomes possess loci comprised of regularly spaced repeats interspersed by other DNA sequences derived from viruses^133–136. These loci, now termed CRISPRs (Clustered Regularly Interspaced Short Palindromic Repeats) act as an innate immune system by incorporating fragments of viral DNA between the repeats, which are then transcribed and processed to produce small guide RNAs (linked to their effector complexes via the repeat sequence) that target and destroy viral DNA^137–140 or RNA¹⁴¹). This system has recently been adapted for RNA programmable sequence-specific genome manipulation in eukaryotes, including mammals^142–145, with extraordinary versatility including targeted gene excision and fusion, and modified CRISPRs capable of recruiting silencing and activating proteins to target loci^146–150. Moreover, the biological arms race continues, with bacteriophages encoding their own CRISPR system to evade host innate immunity¹⁵¹.

Long noncoding RNAs

The eukaryotic transcriptome and long non-coding RNAs

Noting that the density and size of introns (and, as it turned out later, intergenic sequences) expanded with developmental complexity, Mattick posited in 1994 that introns had evolved to express an expanding repertoire of trans-acting regulatory RNAs, that some genes subsequently evolved to express only (intronic or exonic) regulatory RNAs, and that this RNA-based regulatory system was the essential prerequisite for the emergence of developmentally complex organisms¹⁵². Subsequently, the application of genome tiling array technology and deep sequencing to the characterization of the transcriptome showed that there are tens of thousands of loci in mammals that express large transcripts that do not encode proteins, located intergenic, intronic and antisense to protein-coding genes. The initial findings^153–155 were confirmed in 2005^156–159 and extended by the ENCODE project^160–162, all of which showed that the vast majority (at least 80%) of the human and mouse genomes are differentially transcribed in one context or another, with other studies reporting similar findings in all organisms examined. Indeed, it seems most intergenic (and by definition intronic) sequences are differentially transcribed, and therefore that the extent of the transcriptome expands with developmental complexity¹⁶³.

Using more focussed deep sequencing methodologies, it has become evident that the full repertoire of the protein-coding and non-protein-coding transcriptome is still vastly under-sampled¹⁶⁴. In addition, many transcripts are not polyadenylated, and represent a largely different sequence class^156,165, some of which appear relevant to development (e.g., OCT4^166,167). Moreover, 95% of human transcription initiation sites are not associated with mRNAs, but rather mainly with non-polyadenylated noncoding transcription¹⁶⁸. These non-polyadenylated transcripts are as yet largely uncharacterized because of the historical use of polyA tails to remove the overwhelming rRNA contamination in RNA preparations, which is being alleviated by more sophisticated approaches, including cap trapping¹⁶⁹, oligonucleotide subtraction¹⁷⁰ and array capture^164,171.

Defining long noncoding RNAs

Long noncoding RNAs (lncRNAs) are operationally defined as any non-protein-coding RNA >200 nt in length, which corresponds to a convenient cut-off in biochemical fractionation and excludes all known classes of small RNAs¹⁷². Transcripts are judged to be ‘noncoding’ if they lack a long open reading frame (traditionally >100 codons) and/or do not show codon conservation, although with initially limited genomic and transcriptomic data for comparison, there was considerable uncertainty. However, recent analyses provide strong evidence that most annotated lncRNAs do not encode proteins, although some specify small proteins that had previously fallen under the bioinformatic radar^173–175.

These noncoding transcripts can be parsed into intronic, antisense or intergenic (‘lincRNA’) subsets in experimental studies and databases^159,176,177, partly because of mechanistic expectations¹⁷⁸ coupled with a desire to reduce ambiguity and overlap with protein-coding loci in functional analyses^179–181. However, there is no evidence of any intrinsic difference between RNAs that are intronic, intergenic, antisense or overlap with protein-coding genes, for example in their interaction with chromatin-activating or -repressive complexes (Figure 1 and below), although subclasses will inevitably exist and be defined, some of which may have be biased in relation to genomic origin.

Exploration of lncRNA functions

The unexpected discovery of large numbers of noncoding transcripts in eukaryotes, some of which span tens or hundreds of kilobases¹⁸², has led to debates about their functionality (see e.g. ^183,184), particularly since many have relatively low evolutionary conservation and low levels of expression, leading some to posit that they represent ‘transcriptional noise’ and/or redundant transcripts with no biological significance. This is a possibility, at least in part. However, it is clear that lncRNAs actually show a wide range of evolutionary conservation, from those that are ultraconserved¹⁸⁵ to those that are primate-specific^186–188, which can be explained as the result of different structure-function constraints and lineage-specific adaptive radiation¹⁸⁹. Indeed there is now considerable evidence that lack of primary sequence conservation in lncRNAs does not indicate lack of function^190,191, and that many lncRNAs show evidence of structural conservation^192,193.

Moreover, those loci expressing lncRNAs show all of the hallmarks of bona fide genes⁴, including conservation of promoters¹⁶⁹, indicative chromatin structure¹⁹⁴ and regulation by conventional morphogens and transcription factors¹⁹⁵. LncRNAs have been found to have a similar range of cellular half-lives as mRNAs¹⁹⁶ and to be differentially expressed in a tissue-specific manner^158,197, with higher resolution in the brain¹⁹⁸. The latter showed that while the level of expression of many lncRNAs superficially appears to be lower than mRNAs in whole tissues, lncRNAs are highly expressed and easily detectable in particular cells¹⁹⁸, and it appears that they have, on average, higher cell-specificity than proteins^165,199, consistent with their proposed role in architectural (as opposed to ‘cell-type’) regulation, where each cell has a unique positional identity in precisely sculpted organs, bones and muscles²⁰⁰.

Many lncRNAs are alternatively spliced²⁰¹, further evidence of the precision of their expression and hard to reconcile with the suggestion that they are simply transcriptional noise. It should also be noted that some functionally validated lncRNAs can have isoforms that encode proteins²⁰² and reciprocally that some (perhaps many) mRNAs may also have intrinsic functions as trans-acting regulatory RNAs^203–205, that in some contexts 3′UTRs can be separately expressed and convey genetic functions in trans²⁰⁴, and that both may be further processed to produce subsidiary species²⁰⁶.

LncRNAs have been shown to be dynamically expressed in a range of differentiating systems, including embryonal stem cell²⁰⁷, muscle²⁰⁸ T-cell²⁰⁹, breast^210,211, erythroid²¹¹ and neuronal differentiation^212–214, as well as in cancer and other diseases (see e.g.^{210,215–222}), at least partly controlled by conventional transcription factors^195,213.

The validation of lncRNA function has to date mainly relied on knockdown of candidate lncRNAs. It has proven surprisingly easy to knockdown lncRNA expression by si/shRNA-mediated approaches, and thereby to detect phenotypic changes in cultured cells, where most analyses have been carried out. By 2009, ~50 lncRNAs had been shown to be functional⁴ and hundreds more are now published or en route to publication, a large enough sample to draw the conclusion that these transcripts are generally functional.

Roles in development and differentiation

Some lncRNAs play a role in general or differentiation-specific cell biological processes. These include: Template RNAs that guide chromosomal rearrangements in ciliates²²³; Terra RNAs, involved in telomere biology²²⁴; 7S RNA, an essential component of the signal recognition particle involved in protein export²²⁵; 7SK, a highly expressed structured RNA that is the scaffold to assemble a multimeric protein complex containing SR splicing proteins and P-TEFb, a cyclin-dependent kinase required for transcriptional elongation by RNA polymerase II and other factors ²²⁶; Neat1, an essential component of paraspeckles, enigmatic subnuclear organelles that appear in differentiated but not stem cells in mammals^227,228; the nuclear-localized MALAT-1, which regulates alternative splicing²²⁹ and cell cycle progression²³⁰; Gomafu, which is expressed in an unknown subnuclear structure, possibly a specialized spliceosome, in a subset of neurons²³¹, and has recently been implicated in schizophrenia²³²; and others of unknown function associated with bipolar structures in the nuclei of Purkinje cells¹⁹⁸.

Not surprisingly, given their expression patterns, most functionally analyzed lncRNAs appear to play roles in the regulation of differentiation and development²³³. These include, based on studies in cell culture, the regulation of apoptosis and metastatic processes^{211,218,220,221,234}, retinal and erythroid development^211,235, breast development^210,236, and epidermal differentiation²³⁷, among many others.

Antisense knockdown of lncRNAs in zebrafish and deletion of sequences specifying lncRNAs in mouse have shown that some confer visible developmental defects^{181,191,238,239}, although others do not, including knockouts of the widely expressed Neat1 required for paraspeckle function²⁴⁰ and some of the most highly conserved sequences in the mammalian genome²⁴¹. This suggests that more sophisticated phenotypic screens may be required, especially in relation to cognitive function, since most mammalian lncRNAs are expressed in the brain¹⁹⁸ and many are mammal- or primate-specific^242,243. A good example is the retrotransposon-derived lncRNA BC1, which is widely expressed in the brain but whose knockout causes no visible anatomical abnormality, but leads to behavioural changes that would be lethal in the wild²⁴⁴.

Epigenetic roles of noncoding RNAs

Consistent with their roles in differentiation and development, a range of genetic and biochemical evidence suggests that a major function of lncRNAs and many small RNAs is the regulation of epigenetic processes^245,246, likely by guiding chromatin-modifying enzymes to their sites of action and/or acting as scaffolding for chromosomal organization^{179,246–249} (Figure 3).

Figure 3. Various roles for lncRNAs in cellular regulation.

(A) Long non-coding RNAs are expressed from many loci in the genome, sense and antisense, intronic, overlapping and intergenic with respect to nearby protein-coding loci, and function both in cis and trans. (B–E) Some lncRNAs interact with proteins to control the access of chromatin to cellular components and/or guide epigenetic regulatory complexes to target loci resulting in both (B) transcriptional suppression²⁰¹ and (C) activation or suppression (bimodal control)¹⁹⁴. Proteins involved in chromatin modification such as DNMT3a, EZH2 and PRC2 complexes have been associated with epigenetic targeted lncRNA regulation^194,201,277. (D) Some lncRNAs function to tether distal enhancer elements with their promoters^346,347. (E) LncRNAs can also function by binding proteins to sequester them away from their sites of action (decoy lncRNAs)²⁷⁴ while other lncRNAs can interact with each other and/or function to sequester small regulatory RNAs such as miRNAs and therefore RISC targeting complexes away from protein-coding mRNAs^201,339,340. (G) LncRNAs can also act as translational inhibitors by binding and sequestering mRNAs away from the translational machinery³⁴⁸ while other lncRNAs (H) appear to regulate splicing²³².

RNAs were first shown to induce transcriptional gene silencing in plants^74,250, fungi²⁵¹ and human cells⁸⁸, with an intimate involvement of small RNAs and the RNA interference pathway in the epigenetic processes involved^251–253, consistent with the observation that small RNAs interact with Polycomb²⁵⁴ and that Ago proteins occur in the nucleus^86,87 (Figure 3).

In parallel, dating back to 1990, antisense RNAs have also been shown to affect gene expression, again initially in plants⁷³ and later in animals^{159,166,255–257}. Some lncRNAs, similar to small ncRNAs²⁵⁸, have been shown to control alternative splicing^259,260. Other naturally occurring lncRNAs had been shown to control epigenetic processes in vivo, notably in X chromosome dosage compensation^261–265 and parental imprinting in mammals^266–268 and vernalization in plants²⁶⁹, with subsequent studies showing that intergenic and antisense RNAs bind to polycomb chromatin repressive complexes^{194,270–272}, to trithorax chromatin activating complexes and activated forms of histones²⁰⁷, and to DNA methyltransferases^201,273,274. These observations were writ large in 2009 when it was shown that approximately 20% of ~3,300 lncRNAs examined were bound by PRC2, with others bound by other chromatin-modifying complexes, that siRNA-mediated knockdown of lncRNAs associated with PRC2 led to changes in gene expression, with the up-regulated genes being enriched for those normally silenced by PRC2¹⁷⁹, and that Polycomb binds RNA with high affinity but low specificity²⁷⁵, consistent with the idea that many RNAs are Polycomb interactors.

One of the notable lncRNAs to emerge from the studies of Rinn, Chang and colleagues, HOTAIR, is derived from the HOXC locus and regulates HOXD in trans¹⁹⁴, is involved in cancer metastasis²²⁰ and, when inactivated, results in homeotic transformation in vivo²⁷⁶, Chang and colleagues also showed that lncRNAs could act as scaffolds for the assembly of histone modification complexes²⁷⁷, with the widespread alternative splicing of these RNAs suggesting that the cargo and/or target specificity can be varied in a context-dependent and differentiation-specific manner.

LncRNAs may also be involved in orchestrating the highly dynamic spatial structure of chromatin during differentiation and development^164,278, which would explain their often highly cell-specific expression patterns²⁰⁰. Developmental enhancers, as well as polycomb- and trithorax-response elements, are transcribed in the cells in which they are active^{203,279–284}, and are likely not only scaffolds for the recruitment of epigenetic regulators²⁸⁵ but also the physical mediators of the complex genetic phenomena of transvection and transinduction²⁴⁵. Moreover, many lncRNAs display the properties of enhancers¹⁸⁰. These RNAs may well guide the physical looping that occurs between enhancers, target promoters and exons, with precise positioning of nucleosomes^286–290, to control transcription and alternative splicing^248,291,292, again modulated by alternative splicing. Indeed the emerging picture is of a chromatin and transcriptional landscape that is exquisitely and precisely controlled in 4 dimensions by a suite of regulatory RNAs that assemble relatively generic (albeit often cell- or differentiation state-specific) enzyme complexes and isoforms to their sites of action in a context-dependent manner²⁴⁹.

A substantial proportion of lncRNAs reside within, or are dynamically shuttled, to the cytoplasm indicating roles in other cellular processes, including the regulation of protein localization²⁹³, mRNA translation²⁹⁴ and mRNA stability²⁹⁵.

RNA editing, modification, retrotransposition and inheritance

Regulatory RNAs may also be influenced by environmental signals and be transmitted between cells and generations, which has important implications for understanding gene-environment interactions and evolution. There is evidence that plasticity has been superimposed on RNA-directed epigenetic networks by the expansion of RNA editing, especially during cognitive evolution^296,297, and by retrotransposon utilization and mobility^{114,298–301}, which harks back to the insights of McClintock and Britten & Davidson. The raw material for evolution is gene duplication and transposition, the latter having the advantage of being able to mobilize functional cassettes in regulatory networks³⁰², which appears to be the main driver of adaptive radiation^245,303. Indeed many lncRNAs may have originated from retrotransposons and the evolution of mRNAs and lncRNAs may have been accelerated by retrotransposition of functional modules^304–308.

Moreover, apart from snoRNA-directed modifications, there are well over 100 other documented modifications of RNA^309,310, including cytosine and adenosine methylation which have known physiological and cognitive effects^311–314, indicating a new additional layer of RNA informational code and epitranscriptomics, an exciting field that is just beginning to emerge^315,316.

There is evidence for systemic transmission of RNA^317,318 and RNA-mediated epigenetic inheritance in plants and animals^319–323. There is also the intriguing possibility of RNA-directed DNA recoding, which may place RNA at the centre not only of gene regulation in the developmental ontogeny of higher organisms, but also of both hard- and soft-wired somatic and germline evolution^324–326.

Conclusions and outlook

The past two decades have seen an explosion in our understanding of the previously hidden and unanticipated world of RNA regulation. Indeed, in retrospect, it appears that we may have fundamentally misunderstood the nature of the genetic programming in complex organisms because of the assumption that most genetic information is transacted by proteins. This maybe largely true in simpler organisms, but is turning out not to be the case in more complex organisms, whose genomes appear to be progressively dominated by regulatory RNAs that orchestrate the epigenetic trajectories of differentiation and development.

The picture that emerges is of an extraordinarily complex transcriptional landscape in mammals and other multicellular organisms, comprised of overlapping, intergenic and intronic sense and antisense small and large RNAs with interlaced exons^327,328, whose promoters, splicing patterns, polyadenylation sites and regional repertoire varies in different cells and developmental contexts (see below). Since there appear to be few distinct boundaries to genes in humans, it seems better to change the focus of analysis to the transcript, with genetic loci redefined as fuzzy transcription clusters^165,328,329 albeit semantically anchored or related to an enclosed or nearby protein-coding locus. However, this can only be stretched so far, and non-protein-coding loci raise problems for existing schema of human genome nomenclature.

Indeed even the notion of a (simple) protein-coding sequence needs to be reassessed. It is becoming evident not only that mRNAs can have multiple functions²⁰⁵, but also that protein-coding sequences themselves can have other embedded functions, as suggested by constraints on synonymous codon usage^330,331, including regulatory functions as epigenetic modulators²⁰³, tissue-specific enhancers^331,332 and transcription factor binding sites³³³. The possibility, if not likelihood, is that there is a very complex functional and evolutionary interplay between the protein-coding and regulatory functions of RNAs²⁰⁰, and that some lncRNAs may have evolved, at least in part, from protein-coding genes, as appears to have occurred with Xist, by duplication or pseudogenization followed by the emergence of paralogous regulatory and/or coding functions^201,334. Conversely, it appears that new protein-coding capacity may also appear in lncRNAs¹⁷⁴.

The sheer number and diversity of RNAs juxtaposed with their extraordinarily complex molecular functions (Figure 3) in regulating epigenetic processes, subcellular organelles, protein-coding and non-coding gene transcription, translation, RNA turnover, chromosomal organization and integrity, and genome defence, among others, suggests that we have a long way to go to understand the structure and functions of what is surely a highly interconnected system. There are literally tens of thousands, if not more, of individual noncoding RNAs whose roles in cell and developmental biology, as well as brain function, remain to be determined. Moreover, many if not most regulatory RNAs, especially in complex organisms, remain to be identified, including new classes such as the circular RNAs and others that may function as miRNA ‘sponges’^62,335–340, which will require targeted deep sequencing of small and large RNAs that are derived from different genomic locations in different cells, using targeted techniques such as RNA CaptureSeq^164,171.

RNA is not a linear molecule, but can rather fold into complex and allosterically responsive 3-dimensional structures that can both recruit generic effector proteins and guide the resulting complexes sequence-specifically to other RNAs and DNA, via duplex or triplex formation. There are many important questions. These include the identification of functional domains in RNA and their interacting partners, so that we can predict and parse RNA functional interactions in the same way that is already done by recognition of well-characterized motifs and domains in proteins. One way to do this, already underway in many laboratories, is to combine immunoprecipitation of different types of RNA binding proteins (chromatin-modifying proteins, transcription factors, and RNA transport proteins, among others) with deep sequencing of the associated RNAs (RIP-Seq) followed by analysis of primary and predicted secondary structures, and ultimately by biochemical validation and characterization.

Determination of the structure of RNAs, RNA-RNA, RNA-DNA and ribonucleoprotein complexes will be a rapidly growing field, requiring the development of new technologies, such as RNA footprinting with high-throughput sequencing³⁴¹, as well in vivo studies using RNA-based genetic techniques like CRISPR-mediated mutation¹⁴³. Other objectives include determination of whether small RNA pathways are used in viral defence in humans⁸⁰, the functions of ti/spliRNAs and snoRNA-derived small RNAs, the roles of piRNAs in retrotransposon dynamics and the remodelling of the genome by retrotransposons in the brain¹¹⁴, the mechanisms and extent of RNA-mediated trans-generational epigenetic inheritance³⁴², the locations of RNA binding sites (RNA-DNA duplexes and RNA-DNA:DNA triplexes) in DNA, the cross-talk between different types of regulatory RNAs, the logic and hierarchy of RNA- and protein-mediated regulation of gene expression, and the extent, mechanisms and information content of RNA-mediated communication between cells, both within³¹⁸ and between organisms (‘social RNA’)³⁴³. Indeed it appears that RNA is the computational engine of cell biology, developmental biology, brain function and perhaps even evolution itself³²⁵. The complexity and interconnectedness of these systems should not be cause for concern but rather the motivation for exploring the vast unknown universe of RNA regulation, without which we will not understand biology.

Table 1.

Timeline Table:

1941	One gene-one enzyme2
1953	Double helical structure of DNA described349
1958	Central dogma proposed by Francis Crick6
1961	mRNA confirmed as intermediate between protein and DNA9
1961	Jacob and Monod speculate that the lac repressor is an RNA10
1966	Discovery of heterogenous nuclear RNA27
1969	Model proposed for RNA acting in intermediate fashion in gene regulation28
1972	hnRNAs, chromosomal RNAs shown to be functional without making protein350
1977	Intron ncRNA elements defined34,35
1982–3	Self Splicing catalytic RNAs40,41
1989	Transgene silencing observed in plants71,72
1990	Transgene silencing linked to antisense RNA73
1990	H19 ncRNA discovered351
1992	Xist ncRNA discovered261, 262
1993	Lin-4 miRNA discovered46
1994	Regulatory RNAs proposed to be central to animal evolution and development152
1994	RNA directed DNA methylation observed in plants74
1998	RNA interference described in plants69 and animals352
1999	Tsix, antisense transcript to Xist described264
1999	Small RNA required for PTGS in plants353
2000	Let-7 miRNA discovered47
2001	Dicer described involved in RNAi77
2001	RNAi (PTGS) found functional in human cells354
2001	Regulatory RNA networks proposed to control epigenetic processes245,355
2002	First reports of large numbers of noncoding RNAs in animals153–155
2002	AIR antisense RNA involved in imprinting267
2003	Drosha described in miRNA processing76
2004	Small RNA shown to epigenetically control transcription (TGS) in human cells88
2004	Argonuate 2 directs catalysis in RNAi in mammals356
2005	piRNAs described100
2005	Confirmation of large numbers of long noncoding RNAs in mammals156,158,159
2005	~70% of sense transcripts have antisense counterparts, some show function159
2005	Discovery of the CRISPR system of bacterial RNA-based defence134–136
2006	Antisense RNA TGS shown to require DNMT3a, EZH2, HDAC-1344
2006	Argonautes 1 and 2 found involved in RNA-directed TGS in human cells86,87
2006	ncRNAs involved in trithorax regulation285.
2007	HOTAIR shown to play a role in development and associate with polycomb194
2008	Long antisense RNAs found to epigenetically regulate sense counterparts256,257
2008	LncRNAs shown to interact with trithorax and activated chromatin207
2008	Hundreds of lncRNAs shown to have specific expression in brain198
2009	tiRNAs reported at transcription start sites in mammals121
2009	PRC2 found to interact with a large number of lncRNAs179
2009	Long antisense RNA shown to direct vernalization in plants269
2010	Pseudogene lncRNAs found to regulate protein-coding genes166,340
2012	ENCODE reports ~80% of the genome is transcribing ncRNAs162
2013	Enhancer RNAs shown in oestrogen-dependent transcriptional activation357

Glossary

Antisense RNA

A single stranded RNA that is complimentary to a messenger RNA or a gene

ENCODE

Encyclopaedia of DNA elements is a consortium of international collaborators involved in building a comprehensive list of functional elements in the human genome^160,161

hnRNA

Heterogenous nuclear RNA. Similar to messenger RNA, or pre-mRNAs, but retained predominantly in the nucleus

Intron

A term first coined by Walter Gilbert to describe those nucleotide regions in RNA that are removed, by being spliced out, to produce messenger RNAs^34,35

lncRNA

Long non-coding RNAs, first used to differentiate between smaller forms of non-coding RNA, e.g. greater than 200bp in size³⁵⁸

piRNA

Piwi associated RNAs. Small RNAs associated with the Piwi protein complex and emanating from transposable like elements¹⁰⁰

Pseudogene

Relics of genes that have lost their protein coding potential but remain transcribed and integrated within the genome³⁵⁹

PTGS

Post-transcriptional gene silencing. Silencing a gene at the messenger RNA or translational level, after transcription has occurred²¹⁰

RNA-directed DNA methylation

An epigenetic process whereby processed double stranded small RNAs (21–24bp) guide the methylation of homologous DNA loci⁵²

siRNA

Small interfering RNAs, double stranded RNAs that can be used to suppress homology containing transcripts in a transcriptional and post-transcriptional manner³⁵²

Transposons

Mobile genetic elements²⁶², with evolutionary links to retroviruses

tiRNAs

Transcription initiation RNAs are small RNAs associated with promoters with peak density ~10–30 nucleotides downstream of transcriptional start sites¹²¹. Similar RNAs are derived from splice sites (spliRNAs)¹²²

TGS

Transcriptional gene silencing. The regulation of a gene at the transcriptional level

UTRs

Untranslated regions, referring to either side, 5′ (leader sequence) or 3′ (trailer sequence) of a coding sequence on a strand of messenger RNA