Splicing noncoding RNAs from the inside out.

Eukaryotic precursor-messenger RNAs (pre-mRNAs) undergo splicing to remove intragenic regions (introns) and ligate expressed regions (exons) together. Unlike exons in the mature messenger RNAs (mRNAs) that are used for translation, introns that are spliced out of pre-mRNAs were generally believed to lack function and to be degraded. However, recent studies have revealed that a large group of spliced introns can escape complete degradation and are processed to generate noncoding RNAs (ncRNAs), including different types of small RNAs, long-noncoding RNAs, and circular RNAs. Strikingly, exonic sequences can be also back-spliced from pre-mRNAs to form stable circular RNAs. Together, the findings that ncRNAs can be spliced out of mRNA precursors not only expand the ever-growing repertoire of ncRNAs that originate from different genomic regions, but also reveal the unexpected transcriptomic complexity and functional capacity of eukaryotic genomes.

INTRODUCTION

A fundamental feature of eukaryotic protein‐coding genes is that they are in pieces.1 It is crucial that intragenic regions (introns1) are spliced out of the precursor‐messenger RNA (pre‐mRNA) and expressed regions (exons1) are ligated together to form a final mature messenger RNA (mRNA) that encodes for a protein. The pre‐mRNA splicing offers flexibility in regulating gene expression,1 and in higher eukaryotes, alternative splicing of a single pre‐mRNA yields multiple mature mRNAs and therefore multiple protein products.2 Genome‐wide studies have suggested that nearly all human multiexonic protein‐coding genes undergo alternative splicing3, 4 to significantly increase the transcriptomic/proteomic complexity and hence their functional diversity.2, 5 Additionally, large/long‐intergenic/intervening noncoding RNAs (lincRNAs) were recently demonstrated to be alternatively spliced as well,6 although their splicing efficiency is relatively low.7 (For simplicity, this review will use pre‐RNA to refer to an unspliced precursor RNA irrespective of whether the primary transcript is used to generate an mRNA or lincRNA.)

Unlike the ligated exons in the mature RNAs, the intron lariats are generally debranched and ultimately degraded after splicing (Figure 1(a)). From a coding perspective, introns were generally regarded as ‘junk’ as they do not influence the sequence of the end product.8 However, it is now widely recognized that introns are not just passively removed during splicing, but play important roles in regulation of gene expression. Introns harbor diverse cis‐regulatory elements that affect pre‐RNA splicing,5 and have various fates that affect gene expression. For instance, introns can be retained in the final RNA product. A recent study reported that intron retention can tune mammalian transcriptomes by suppression of inappropriately expressed transcripts.9 Self‐splicing group I and group II introns are catalytically active as ribozyme to guide their own excision.10, 11, 12 These self‐splicing group I and group II intronic sequences, after spliced out, can be further processed to yield circular molecules10, 13; however, such circular transcripts are unstable and their functions are at best limited.14, 15

Figure 1

MicroRNAs (MiRNAs) are processed from spliced introns. (a) Eukaryotic precursor RNAs (pre‐RNAs) undergo splicing (dash lines) to remove introns (lines) and ligate exons (bars) together to form either mature mRNAs that are subsequently translated or noncoding RNAs. After splicing, intron lariats are generally debranched and ultimately degraded. ss, splice site. BP, branchpoint. (b) Drosha/DGCR8‐dependent model of canonical mirtron processing. (c) Splicing‐dependent model of mirtron processing. Notably, some mirtrons that derive from small introns have the hairpin exactly ending at the splice sites to resemble pre‐miRNAs, thus do not need to be trimmed by exonucleases.40, 49

In addition to generating RNA circles from self‐splicing introns10, 13 and tRNA introns,16 a variety of noncoding RNAs (ncRNAs) are processed from nuclear pre‐RNA through the spliceosomal pathway. These intragenic ncRNAs include many, but not all, microRNAs (miRNAs),17, 18 small nucleolar RNAs (snoRNAs),19, 20 RNase P RNA subunit,21 new type of long‐noncoding RNAs (lncRNAs)22 and circular RNAs from either excised introns23 or excised exons.24, 25 Different from ncRNAs independently transcribed from intergenic loci by RNA polymerase II (RNA Pol II),6, 26 the expression of these intragenic ncRNAs is dependent on the transcription and splicing of their host pre‐RNAs. Importantly, such ncRNAs play important roles in altering gene expression both in cis 23 and in trans.22, 27, 28 This review focuses on the biogenesis of intragenic ncRNA species excised from the inside of nuclear pre‐RNAs by splicing.

A LARGE NUMBER OF miRNAs ARE PROCESSED FROM SPLICED INTRONS

The miRNAs are endogenous ncRNA species of ~22 nts that function as guide molecules in post‐transcriptional gene silencing.28, 29, 30 The miRNAs play a key role in both physiological and pathological processes, such as self‐renewal of embryonic stem cells (ESCs), development, and cancers.31 In the canonical mammalian miRNA biogenesis pathway, RNA Pol II transcribes a primary miRNA (pri‐miRNA) transcript, which is 5′ capped and 3′ polyadenylated.26, 32 This pri‐miRNA is processed by Drosha/DGCR8 microprocessor to produce a miRNA precursor (pre‐miRNA).33, 34 Through an association with Exportin‐5, the pre‐miRNA is subsequently exported from nucleus to cytoplasm,35 where this hairpin intermediate is cleaved by Dicer to yield a miRNA/miRNA* duplex.36, 37 Finally, the guide (but not the passenger) strand of the miRNA/miRNA* duplex is incorporated into the RNA‐induced silencing complex (RISC) to repress gene expression based on miRNA–mRNA sequence complementarity.38, 39, 40

In addition to the canonical biogenesis pathway, miRNAs can be also produced from introns of protein‐coding genes (termed as mirtrons) in both invertebrate41, 42 and mammals.43, 44 Indeed, a significant population of human and murine miRNAs originate from mirtrons.45, 46, 47 A Drosha/DGCR8‐dependent and splicing‐independent model has been proposed to yield mirtrons. In this model, the intronic pre‐miRNA hairpin is cleaved from the pre‐mRNA by Drosha/DGCR8 prior to the splicing catalysis,18 and then enters the miRNA biogenesis pathway (Figure 1(b)). Interestingly, the split intron by Drosha/DGCR8 showed little effect on the following exon linkage or mRNA maturation.18 As located within host genes, the expression of some mirtrons is coregulated by transcription and splicing of their host pre‐RNAs.48 In another splicing‐dependent pathway (Figure 1(c)), after cotranscribed with the host gene, the pre‐miRNA hairpin is excised out of host pre‐RNA with spliceosome, trimmed by exonucleases (for tailed mirtrons only) and exported to the cytoplasm where it can be further processed by Dicer.40, 49 Notably, some mirtrons that derive from small introns have the hairpin exactly ending at the splice sites to resemble pre‐miRNAs, thus do not need to be trimmed by exonucleases.40, 49

Besides miRNAs, some of Piwi‐interacting RNAs (piRNAs) and endogenous small interfering RNAs (siRNAs) are also likely to be generated from introns and exons,46 while the detailed mechanisms require further investigation. Another type of house‐keeping ncRNA, the catalytic RNA subunit of RNase P (RPR), has been reported to be processed from the last intron of an RNA Pol II transcript of the gene ATPsynC in insects/crustaceans,21 while other animal RPR genes are independently transcribed from RNA polymerase III. The evolutionary driving force for this divergence over 500 million years ago is unknown.21

THE MAJORITY OF HUMAN snoRNAs ARE PROCESSED FROM SPLICED INTRONS

SnoRNAs are a family of conserved nuclear ncRNAs (~70–200 nts in length) that are usually located in nucleoli and participate in the modification of small nuclear RNAs (snRNAs)/ribosomal RNAs (rRNAs) or in the processing of rRNAs during ribosomal maturation.27, 50, 51 Two types of snoRNAs, box C/D and box H/ACA snoRNAs, are defined by their conserved sequence motifs.27 Hundreds of human cellular sno and scaRNAs (snoRNA variants that localize to Cajal bodies) have been annotated by snoRNA‐LBME‐db.52 In yeast, most snoRNAs are produced from independent transcripts by RNA Pol II.20 While in human, only a small portion of annotated snoRNAs is likely produced as independent RNA Pol II transcripts. Instead, the vast majority of human snoRNAs reside within introns of their host (coding or noncoding) genes.52, 53 During splicing and exonucleolytic trimming from debranched introns, the assembly with the snoRNA‐associated proteins (snoRNPs) protects the snoRNA sequences from further exonucleolytic degradation27, 54, 55 (Figure 2 (a)). The processing of intronic snoRNA is coupled to splicing; indeed, snoRNAs positioned about 70 nts upstream to the 3′ splice site is critical for efficient expression.56 In addition, the expression of individual snoRNAs from multi‐snoRNA host genes is coordinated with alternative splicing and nonsense‐mediated RNA decay (NMD), resulting in unbalanced expressions of intronic snoRNAs and their cognate spliced RNA from the same host gene locus.57

Figure 2

Small nucleolar RNAs (SnoRNAs) and snoRNA‐ended long‐noncoding RNAs (sno‐lncRNAs) are processed from spliced introns. (a) SnoRNAs are processed from spliced introns. During splicing and exonucleolytic trimming from debranched introns, the assembly of snoRNA with the snoRNA‐associated proteins (snoRNPs, blue spheres) protects it from further exonucleolytic degradation and leads to the formation of mature snoRNPs. (b) Sno‐lncRNAs are processed from spliced introns and flanked with snoRNAs at both ends. Introns containing two snoRNAs are processed from their ends by the snoRNP machinery (blue spheres) and the intronic sequences between these two snoRNAs are protected, thus leading to the formation of lncRNAs with snoRNA ends.

The aberrant expression of snoRNAs and their associated proteins is linked to human diseases. An extreme example of noncoding genes with snoRNAs is located at the human imprinted 15q11‐q13 locus, which has been implicated in Prader‐Willi/Angelman syndrome (PWS).19, 20 Within this imprinted region, two clusters of tandemly repeated snoRNAs (29 SNORD116s and 42 SNORD115s) and several single snoRNA genes (such as SNORD109A) are processed from downstream introns of a gigantic, 470‐knt long‐paternal transcript.19, 58 The minimal paternal deletion region associated with PWS (108 kb) removes SNORD109A, the SNORD116 cluster of 29 similar snoRNAs and Imprinted in Prader‐Willi syndrome ncRNA (IPW), and the most current published model suggests that the deficiency of SNORD116s is associated with PWS.59, 60, 61 However, although most snoRNAs guide rRNA or snRNA modifications by a base pairing mechanism, SNORD116s show minimal complementarity to rRNAs or snRNAs,58 and thus are unlikely to function in guiding rRNA/snRNA modification. In this case, the molecular mechanism of how SNORD116 snoRNAs are possibly involved in the PWS remains elusive.

PAIRED snoRNAs STABILIZE A NEW TYPE OF lncRNAs FROM SPLICED INTRONS

While most introns are unstable after being spliced out of pre‐RNAs, a large number of lncRNA candidates have been predicted by computational analysis to originate from postspliced introns.46 In addition, profiling of the nonpolyadenylated (poly(A)−) RNAs have revealed mature RNA transcripts from excised introns,62 such as sno‐lncRNAs.22 Unlike the majority of lncRNAs that contain 5′ cap structures and 3′ poly(A) tails, sno‐lncRNA is a new type of lncRNAs that are derived from spliced introns and are flanked by snoRNAs at both termini.22 As they do not contain poly(A) tails at their 3′ ends, sno‐lncRNAs have been missed by most polyadenylated (poly(A)+) RNA‐seq.62 Mechanically, after splicing, introns containing two snoRNAs are processed from their ends by the snoRNP machinery and the internal intronic sequences between the two snoRNAs are protected, leading to the formation of lncRNAs with snoRNA ends22 (Figure 2 (b)).

Sno‐lncRNAs are widely expressed in cells and tissues and can be produced by either box C/D or box H/ACA snoRNAs in human genome.22, 53 Strikingly, the most abundant sno‐lncRNAs in human embryonic stem cells (hESCs) reside in the PWS deletion region.22 There are five sno‐lncRNAs that are produced from the SNORD116 cluster in hESCs. Rather than localizing to nucleoli or Cajal bodies, PWS‐region sno‐lncRNAs strongly accumulate near to their sites of synthesis, suggesting that they are functionally different from snoRNAs. Importantly, these PWS‐region sno‐lncRNAs regulate alternative splicing by interacting with splicing factor Fox2.22 For example, knocking down these sno‐lncRNAs resulted in the aberrant splicing regulation of known Fox2‐targeted cassette exons, many of which are from genes with a clear connection to neuronal function.22 Likely, in PWS patients where the PWS region sno‐lncRNAs are not expressed due to the paternal deletion, altered patterns of Fox2‐regulated splicing may happen along development, possibly causing neurogenetic disorder in PWS patients. Thus, the finding of PWS region sno‐lncRNAs and their potential role in altering Fox2‐regulated alternative splicing lead to a possible association between a new class of lncRNAs and PWS pathogenesis.

Although the primary sequences are highly conserved from mouse to human, mouse SNORD116s are scattered in individual introns.53 As one intron containing two snoRNAs is a prerequisite for the biogenesis of a sno‐lncRNA,53, 63 the lack of PWS region snoRNA pairs within single introns in the mouse genome may result in undetectable PWS region sno‐lncRNAs in mouse transcriptomes.53 Finally, genome‐wide analysis suggests that only a small portion of paired human snoRNAs are identified in single introns based on the current splicing annotations. Considering the widespread tissue‐/cell‐specific alternative splicing,64, 65 it is reasonable to expect identification of more sno‐lncRNAs when additional RNA‐seq datasets become available.

CIRCULAR RNAs FROM SPLICED INTRONS

In addition to sno‐lncRNAs, another type of intron‐derived ncRNAs, circular intronic RNAs (ciRNAs), has been also identified from poly(A)− RNA‐seq datasets.23 The ciRNAs are produced from excised intron lariats that fail to be debranched after splicing, leading to a covalent circle with 2′,5′‐phosphodiester bond between 5′ splice donor site and the branchpoint site (Figure 3(a)). Different from lariat RNAs containing a variety length of 3′‐tails,66 The ciRNAs are derived from partially processed lariats that do not likely contain 3′ linear appendage,23 as only one sharp band on the native high resolution PAGE (polyacrylamide gel electrophoresis) could be detected with or without RNase R (an enzyme that can degrade linear and Y‐structure RNAs, while preserving the loop portion of a lariat RNA67) treatment.23 Moreover, evidence at both bioinformatic and experimental levels has suggested that the formation of ciRNAs depends on a consensus RNA motif containing a 7‐nt GU rich element near 5′ splice site and an 11‐nt C‐rich element near the branchpoint (Figure 3(a)). However, it is still unclear how these cis‐elements function to resist debranching and what other trans‐factors are involved in this process.

Figure 3

Two types of circular RNAs are processed from excised introns or exons. (a) Circular intronic RNAs (CiRNAs) are processed from excised introns. CiRNAs fail to be debranched after splicing, leading to a covalent circle with 2′,5′‐phosphodiester bond between 5′ splice donor site and the branchpoint site. The formation of ciRNAs depends on a consensus RNA motif containing a 7‐nt GU rich element near 5′ splice site (magenta bar) and an 11‐nt C‐rich element near branchpoint (yellow bar). (b) Back‐spliced circular RNAs (CircRNAs) are processed from excised exons. Different from canonical splicing (dashed lines in black), which ligates an upstream 5′ splice site (5′ ss) with a downstream 3′ splice site (3′ ss) to form a linear RNA (top), back‐splicing (dashed line in red) connects downstream a 5′ ss reversely with an upstream 3′ ss to yield a circular RNA with normal 3′,5′‐phosphodiester bond and an alternatively spliced linear RNA with exon exclusion (bottom). Both complementary sequences and protein factors can facilitate back‐splicing by bridging downstream 5′ ss close to upstream 3′ ss. See text for details.

Intron‐lariat‐derived human ciRNAs are abundantly localized in the nucleus and are largely associated with the nuclear insoluble fractionation.23 Some abundant ciRNAs play a cis‐regulatory role in promoting the transcription of their host genes by associating with the elongation RNA Pol II machinery.23 Additionally, a recent study on some back‐spliced circular RNAs that contain both exons and introns has suggested a similar function on transcription regulation.68 Finally, stable intronic sequence RNAs (sisRNAs) were also revealed from both oocyte nucleus69 and cytoplasm70 of Xenopus tropicalis. However, whether these sisRNAs can form similar circle structures as ciRNAs remains to be further investigated.

CIRCULAR RNAs FROM BACK‐SPLICED EXONS

Profiling of poly(A)− RNAs has surprisingly revealed signals from not only excised introns but also excised exons,62 which were further proven as circular RNAs.24 Genome‐wide analyses with specific computational approaches, which identify junction reads with reversed genomic orientation, successfully identified thousands of circRNAs from back‐spliced exons (circRNAs) in various cell lines and from different species.25, 71, 72, 73, 74, 75 Most circRNA exons are located in the middle of annotated genes25 and excised from pre‐RNA by back‐splicing. Different from canonical splicing that ligates an upstream 5′ splice site (5′ ss) with a downstream 3′ ss to form a linear RNA, back‐splicing connects a downstream 5′ ss with an upstream 3′ ss to yield a circular RNA with 3′,5′‐phosphodiester bond24, 25, 71, 72, 76 (Figure 3(b)). Although catalyzed by the canonical spliceosomal machinery,77 the efficiency for circRNAs formation is often very low, possibly due to the unfavorable spliceosome assembly for back‐splicing.25, 73, 76

Back‐splicing competes with canonical splicing for circRNA biogenesis,76 leading to the ‘lariat intermediate’ or ‘direct back‐splicing’ models.24, 71, 73, 78 The main difference between these two models relates to the question of timing: which takes place first: canonical splicing or back‐splicing?76 In the ‘lariat intermediate’ model, the transcribed pre‐RNA first undergo canonical splicing to generate a linear RNA with skipped exon(s) and a long intron‐lariat intermediate containing these skipped exon(s). This long intron‐lariat intermediate is further processed by back‐splicing to generate a circRNA. In contrast, pre‐RNA might be ‘directly back‐spliced’ to first generate a circRNA and an unusual exon‐intron(s)‐exon intermediate, which can be further processed to linear RNAs with skipped exon(s) or degraded. In fact, both mechanistic possibilities might be used in a context (organism)‐dependent fashion. In lower eukaryotes, such as Schizosaccharomyces pombe, circRNA are suggested to favorably generate through the ‘lariat intermediate’ mechanism with short flanking introns.79 While, in human and mouse, complementary sequences25, 80 (mostly repetitive Alu elements in human) across long flanking introns can facilitate ‘direct back‐splicing’ by bridging downstream 5′ ss close to upstream 3′ ss to generate circRNAs76 (Figure 3(b)). In addition to cis‐elements, RNA‐binding proteins were also reported to regulate circRNA biogenesis.75, 81, 82, 83 It is possible that cis‐elements and trans‐factors might work together to synergistically alter back‐spliced circularization, which requires further investigation.

Despite lowly expressed in general, some circRNAs are more abundant than their linear counterparts.73 It has been recently reported that circRNAs are highly enriched in brain (from fly to mammals) with a potential to regulate synaptic function and to be used as biomarkers74, 83, 84; however, the underlying mechanism for enhanced expression in the brain is largely undetermined. Such a differential expression might reflect an array of possible functions for this new class of RNAs. First, some circRNAs can function as miRNA72, 85 or protein sponges,81 but a large scale analysis revealed that only a limited number of circRNAs can potentially act as sponges for miRNAs.86 Second, with the competition between splicing and back‐splicing,76, 81, 87, 88 The circRNA biogenesis might also regulate the alternative splicing of linear RNAs.15 Third, the potential of circRNAs on translation might further expand the diversity of proteome. Artificial circRNAs with internal ribosome entry sites (IRESs) generated from expression vectors are translatable.89, 90 However, endogenous circRNAs have not yet been reported to associate with ribosomes for translation.71, 86 Finally, similar to the intron‐derived ciRNAs,23 some circRNAs with retained introns can promote transcription of their host genes by interacting with U1 snRNP and RNA Pol II.68 Despite recent studies have revealed some biological roles of certain circRNAs, further investigation is required to gain a comprehensive understanding of what most other circRNAs really do in cells.

CONCLUSION

Although generally believed that intragenic sequences (usually introns) are degraded after splicing and therefore functionally inconsequential, accumulated lines of evidence have shown that some spliced introns can be further processed to produce a variety of ncRNAs, including new types of lncRNAs. In addition, recent studies have shown that intragenic exons can be back‐spliced from inside of pre‐RNAs to form RNAs in circle.25, 71, 72, 73, 74, 81 Apparently, the production of these intragenic ncRNAs (short or long) are largely dependent on splicing to occur, but many questions remain to be addressed. How is the processing of intragenic ncRNAs linked with other RNA processing pathways, including transcription,91 NMD,57 and canonical splicing?25, 77, 81 How are these different pathways coregulated and crosstalked? How are different protein cofactors involved in the entire life cycle of these intragenic ncRNAs?92 Moreover, it appears that the expression of many intragenic ncRNAs is not conserved across species; it will be of particular significance to study how and when such sequences were embedded into or removed from their host genes in evolution. Finally, as many of these intragenic ncRNAs were identified from a combination of high‐throughput sequencing and newly developed computational methods, it will be not surprising to find other types of ncRNAs by applying novel genome‐wide approaches. Collectively, the finding that a ncRNA gene is embedded inside of another gene and can be activated to function by splicing sheds new light on the unanticipated complexity of transcriptome and the multifaceted regulation by splicing.