Elements and machinery of non-coding RNAs: toward their taxonomy

Abstract

Although recent transcriptome analyses have uncovered numerous non-coding RNAs (ncRNAs), their functions remain largely unknown. ncRNAs assemble with proteins and operate as ribonucleoprotein (RNP) machineries, formation of which is thought to be determined by specific fundamental elements embedded in the primary RNA transcripts. Knowledge about the relationships between RNA elements, RNP machinery, and molecular and physiological functions is critical for understanding the diverse roles of ncRNAs and may eventually allow their systematic classification or “taxonomy.” In this review, we catalog and discuss representative small and long non-coding RNA classes, focusing on their currently known (and unknown) RNA elements and RNP machineries.

Introduction

Recent studies have uncovered numerous ncRNAs, which account for a large proportion of eukaryotic transcriptomes ¹. ncRNAs are literally defined as RNAs that do not code proteins, that is, all RNAs except mRNAs. These include traditional classes of ncRNAs such as rRNAs, tRNAs, snRNAs, as well as more recently discovered classes such as microRNAs (miRNAs) and long non-coding RNAs (lncRNAs). Prokaryotes also have their own ncRNAs including Hfq-binding small RNAs and CRISPR RNAs (crRNAs). In eukaryotes, many ncRNAs are first transcribed by RNA polymerase II (RNAPII), and as such, their primary transcripts are thought to be capped and poly(A)-tailed just like mRNAs. Yet, some transcripts are selectively processed into miRNAs to exert post-transcriptional gene silencing, while others are retained in the nucleus to form specific foci and/or mediate epigenetic regulation. What then determines their diverse fates and functions? This question seems analogous to asking what determines diverse fates and functions of proteins, which are all synthesized by the ribosome. Like domains and motifs harbored in proteins, RNAs should also embed some fundamental features or “RNA elements”—such as RNA sequences, structures, and chemical modifications—determining their various roles. However, there is one big difference between proteins and RNAs; while domains and motifs of proteins can usually function by themselves, elements of ncRNAs almost always need their protein partners to form operating “RNP machineries.”

Roughly speaking, eukaryotic ncRNAs can be divided into small RNAs (~20–30 nucleotides (nt); e.g., miRNAs), intermediate-sized RNAs (~30–200 nt; e.g., snRNAs), and lncRNAs (> ~200 nt, e.g., Xist RNA). RNA elements in primary transcripts of small RNAs are usually bound by RNA-processing proteins, which cleave the primary transcripts into small RNA pieces that are finally assembled into the operating RNP machinery (Fig 1). This RNP machinery is called RNA-induced silencing complex (RISC), and at the heart of RISC generally lie Argonaute (Ago) family proteins. The Ago family can be divided into the AGO subfamily that binds to miRNAs and siRNAs and the PIWI subfamily that binds to piRNAs. Both subfamilies have the common function to silence target genes complementary to their small guiding RNAs. In contrast, RNA elements in lncRNAs are often, if not always, bound by RNA-binding proteins without processing activity, which remain associated with lncRNAs to form the operating RNP machinery (Fig 1). lncRNAs participate in diverse cellular functions including chromatin modification, transcription, splicing, mRNA decay, translation, protein transport, and assembly, and their RNA elements and RNP machineries are also thought to be extremely diverse.

Figure 1. Formation of operating RNP machineries for small RNAs and lncRNAs.

For both small RNAs and lncRNAs, specific RNA elements that determine their fate and functions are thought to be embedded in their primary transcripts. In the case of small RNAs, those RNA elements are recognized and cleaved by RNA-processing proteins into small RNA fragments, which are finally assembled into the operating RNP machinery. In contrast, RNA-binding proteins often remain associated with RNA elements in lncRNAs to form the RNP machinery.

Despite such differences between small RNAs and lncRNAs, systematic studies on the relationship between fundamental RNA elements, operating RNP machineries, and molecular and physiological functions are important for understanding how ncRNAs work. In this review, we catalog representative small and long ncRNA classes in eukaryotes and the CRISPR system in prokaryotes, focusing on their currently known (and unknown) RNA elements and RNP machineries. For ncRNA classes not covered here, we refer the reader to other comprehensive reviews ^{2, 3}.

microRNAs

Functions

First discovered as a peculiar form of gene regulation in Caenorhabditis elegans ^{4, 5}, miRNAs are now recognized as a major class of small RNAs encoded in the genome of animals, plants, some unicellular eukaryotes, and their viruses ⁶. As of March 2014, more than 24,000 miRNAs have been registered in the miRNA database, miRBase (http://www.mirbase.org), and the human genome alone is thought to encode more than 2,000 miRNAs. Mature miRNAs are ~22-nt single-stranded RNAs with 5′ monophosphate, which bind to complementary sequences in mRNAs and direct silencing of protein synthesis from them. Most animal miRNAs are only partially complementary to their targets; base pairing within a limited region of 6–7 nt in length at the 5′ end of the miRNA, termed the “seed,” makes a disproportionally large contribution to determine target selection ⁷. In addition, some non-seed target sites with base paring in the central and/or 3′ region(s) of the miRNA are also reported ^{8, 9}. The imperfect base pairing with mRNAs allows a single miRNA to regulate hundreds of targets at both mRNA and protein levels ^10–12. Many miRNAs and their complementary target mRNAs are conserved across species and comprise a complex gene regulatory network ⁶. Yet, genetic studies sometimes reveal that only a few among many miRNA targets are in fact responsible for obvious phenotypes (e.g., ¹³), obscuring the definition of bona fide miRNA targets ¹⁴.

RNA elements

miRNAs are encoded either in their own genes or in introns (or sometimes exons) of regular protein-coding genes, and are transcribed by RNAPII as long pri-miRNAs ^15–17. Pri-miRNAs form one or more hairpin structures, each of which typically contains a stem region of ~33 base pairs (bp) and a terminal loop of ~10 nt on average ¹⁸ (Fig 2A i). The hairpin structure acts as the fundamental RNA element of miRNAs and is first recognized by the nuclear RNase III enzyme Drosha and its cofactor DGCR8 in mammals (Pasha in insects and worms) ^{19, 20, 20–23} (Fig 2A ii). Drosha and DGCR8, also known as Microprocessor, cleaves the stem of the hairpin at ~11 bp away from the junction between the stem and the flanking single-stranded regions, liberating the ~22-bp short hairpin structure with a 2-nt 3′ overhang, called pre-miRNA ¹⁸. pre-miRNAs are then exported by Exportin 5 from the nucleus to the cytoplasm ^24–27, where they are recognized and further cleaved by another RNase III enzyme Dicer and its partner protein TRBP or PACT in mammals ^28–31 (Fig 2A ii). Flies have a specialized Dicer, Dicer-1, and its partner protein Loqs-PB for processing of pre-miRNAs ^32–35. The resultant ~22-nt double-stranded RNAs with 2-nt 3′ overhangs at both ends are termed miRNA/miRNA* duplexes, which are ready for RISC assembly. However, these processing steps by the RNase III enzymes can be skipped in some cases, diversifying the miRNA biogenesis pathways. For example, pre-miRNA-like hairpins can be generated by splicing and debranching of introns independently of Drosha processing ^{36, 37}. In contrast, the hairpin of pre-miR-451 is too short to be processed by Dicer but instead directly assembles into RISC ^{38, 39}. Chemically synthesized miRNA/miRNA* duplexes can properly form RISC independently of Drosha or Dicer, as long as they are appropriate in size, structure, and chemistry.

Figure 2. Elements and machinery for representative small RNA classes.

(A) miRNAs are transcribed as long pri-miRNAs containing hairpin structures as an RNA element (i, red). The hairpin structure is cleaved by the nuclear RNase III enzyme Drosha at the stem (ii, dark blue) and by the cytoplasmic RNaseIII enzyme Dicer at the loop (ii, light blue), generating a ~22-nt miRNA/miRNA* duplex (ii, asterisk). The duplex is loaded into AGO proteins (iii, orange) and the passenger strand (miRNA*) is ejected from AGO, producing mature RISC containing the guide strand (miRNA) as an RNP machinery. RISC binds mRNAs containing the seed-matched sites and recruits GW182/TNRC6 (light green) and CCR4–NOT complex (dark green) to induce translation repression, deadenylation, and mRNA decay (iv). (B) In the siRNA pathway, long double-stranded RNAs act as an RNA element (i, red), which is recognized and cleaved by Dicer (ii, light blue), producing ~21-nt double-stranded siRNAs (ii, asterisk). Upon loading of the siRNAs duplex into catalytic AGO (iii, yellow), the passenger strand is expelled and mature RISC containing the guide strand is formed as an RNP machinery. RISC mediates RNAi by cleaving perfectly complementary target RNAs (iv). (C) Primary piRNAs are transcribed from piRNA clusters as long, single-stranded piRNA precursor transcripts that harbor unknown or no apparent RNA elements (i). piRNA precursor transcripts are processed by nucleases including Zucchini (ii, turquois), generating piRNA precursors of intermediate size (ii, asterisk). The cleaved RNA containing uridine at the 5′ end is preferentially loaded into a PIWI protein (iii, yellow-green) followed by 3′ end trimming by a Trimmer (blue). The resultant piRNA-containing RISC acts as an RNP machinery and silences transposon RNAs by post-transcriptional cleavage and transcriptional silencing (iv). (D) During secondary piRNA biogenesis, RNA sequences that are complementary to primary piRNAs act as RNA elements (i, red) and are cleaved by primary piRNA-containing PIWI (e.g., Aub; ii, green). The cleaved RNA fragments (ii, asterisk), which tend to have adenine at the 10^th base, are loaded into another PIWI protein (e.g., Ago3; blue) and mature into secondary piRNAs by 3′ end trimming (iii). Cleavage of the original piRNA precursor transcript by the secondary piRNA drives the ping-pong cycle of the piRNA amplification loop (iv). (E–G) crRNAs are transcribed as long CRISPR transcripts (pre-CRISPR) containing repetitive sequences as RNA elements (i, red). The repeat is cleaved by specific Cas6 nucleases (types I and III) or tracrRNA and RNaseIII (type II) (ii, turquois and purple, respectively). Cleaved crRNAs containing a protospacer and a part of the repeat (ii, asterisk) may be trimmed by unknown nucleases and form a complex with effector Cas proteins (iii) that act as an RNP machinery with sequence-specific nuclease activity (iv).

RNP machinery

After dicing, miRNA/miRNA* duplexes are loaded into AGO proteins as double-strand RNAs ^{40, 41} (Fig 2A iii). In mammals, miRNA/miRNA* duplexes are incorporated into all Ago1–4 proteins ^{42, 43}, while they are mainly but not exclusively sorted into Ago1 in flies ^44–46. These AGO proteins strictly recognize the 5′ monophosphate on the miRNA strand and prefer uridine as the 5′-end nucleotide ^47–50. Loading of miRNA/miRNA* duplexes into AGO proteins does not occur autonomously but requires ATP and the Hsc70/Hsp90 chaperon complex ^51–53, which likely support a productive conformation of AGO for duplex loading ⁵⁴. Subsequently, the miRNA* strand is expelled from AGO in a manner independent of ATP or the chaperone complex ^{40, 41, 51, 53}, producing mature RISC composed of AGO and the single-stranded miRNA. The seed region of the miRNA embedded in AGO is arranged in an ordered quasi-helical geometry suitable for target RNA binding, explaining why this region plays the central role in target recognition ^55–58. AGO proteins (Ago1–4 in mammals and Ago1 [but not Ago2] in flies) directly interact with the P-body protein TNRC6 in mammals (GW182 in flies), which recruits the CCR4-NOT and PAN2-PAN3 deadenylase complexes ^59–63 (Fig 2A iv). This way, miRNAs promote shortening of the poly(A) tail and decay of their target mRNAs. In addition, miRNAs mediate the repression of translation initiation independently of deadenylation ^64–72, but the underlying molecular mechanism remains unclear.

Small interfering RNAs

Functions

Small interfering RNAs (siRNAs) are another well-characterized class of small ncRNAs. Like miRNAs, siRNAs are eventually matured into ~21-nt-long single-stranded RNAs with 5′ monophosphate ⁷³. Unlike miRNAs, however, siRNAs usually pair with target mRNAs with perfect complementarity and direct their endonucleolytic cleavage and destruction, a phenomenon called RNAi ⁷⁴. In insects and plants, siRNAs are naturally made from genetic invaders such as viruses, transgenes, and transposable elements, and RNAi is critical for protection against those invaders ⁷⁵. The antiviral role of mammalian RNAi had long been elusive, until evidence for its existence was recently reported ^{76, 77}. Endogenous siRNAs are also produced from double-stranded RNA regions of structured loci, bidirectional transcripts, and even mRNAs pairing with their antisense transcripts (e.g., pseudogenes) and may play a role in the regulation of mRNA expression ⁷⁸. From a practical point of view, induction of RNAi by exogenous, synthetic siRNA duplexes can effectively silence virtually any gene in many organisms, and chemically synthesized siRNAs are used as convenient experimental tools and potential therapeutic agents.

RNA elements

siRNAs are naturally generated as ~21-nt double-stranded RNAs with 2-nt 3′ overhangs from long double-stranded RNAs ⁷⁹, which can thus be defined as the fundamental RNA elements for siRNA biogenesis (Fig 2B i). In flies, a specialized Dicer, Dicer-2, and its partner proteins R2D2 or Loqs-PD generate siRNA duplexes from long double-stranded RNAs ^{32, 80–85}, while mammals use the single Dicer and its partner proteins TRBP or PACT for both miRNA and siRNA production ^{31, 86–88} (Fig 2B ii). Mammalian Dicer is more efficient in processing pre-miRNA hairpins than long double-stranded RNAs ⁸⁹. However, mouse oocytes express a specific Dicer isoform lacking the N-terminal helicase domain, which is highly active in siRNA production ⁹⁰. On the other hand, in adult C. elegans soma, Dicer is proteolytically truncated and the resultant small Dicer isoform enhances exogenous RNAi ⁹¹. After dicing, siRNA duplexes are assembled into RISC. Chemically synthesized siRNA duplexes can also form RISC by bypassing the dicing step ⁹².

RNP machinery

Like miRNAs, siRNAs are also loaded into AGO subfamily proteins as double strands (Fig 2B iii) in a manner dependent on the guide strand 5′ monophosphate and the action of the Hsc70/Hsp90 chaperone complex ^{51–53, 93, 94}. While siRNA duplexes are incorporated into all Ago1–4 proteins in mammals ^{42, 43}, they are almost exclusively sorted into Ago2 in flies ^44–46. Fly Ago2 requires Dicer-2—the same protein required for siRNA production—and R2D2 for loading of siRNA duplexes ^{80, 81, 95}, in addition to the chaperone complex. In contrast, Dicer and TRBP are not essential for duplex loading in mammals ^{96, 97}, although they may tune the loading process ⁹⁸. When the siRNA duplex is loaded into mammalian or fly Ago2, the passenger strand is cleaved at the phosphodiester bond between nucleotides 9 and 10 (across from nucleotides 10 and 11 of the siRNA guide strand) by the endonucleolytic “slicer” activity of Ago2, as if it were the first target RNA for Ago2 ^99–102. The cleaved passenger strand is discarded from Ago2 independently of ATP and the chaperone machinery, producing the mature RISC containing the single-stranded guide strand ^{41, 51}. The Translin–Trax endonuclease complex, also known as C3PO, is reported to accelerate this slicer-dependent RISC maturation process ^{97, 103}. In the absence of passenger strand cleavage (e.g., when bound to human Ago1, 3 or 4, which lack slicer activity), the two strands of the siRNA duplex can still be separated, albeit slowly ^{41, 99, 104}. In flies, Ago2-loaded siRNA guide strands are 2′-O-methylated at their 3′ ends by the methyltransferase Hen1 ¹⁰⁵, which protects them from degradation upon target binding ¹⁰⁶. Mature RISC containing slicer-competent Ago2 mediates RNAi by cleaving perfectly complementary target RNAs ^{42–44, 107} (Fig 2B iv). siRNAs with central mismatches to the target site or siRNAs bound to slicer-deficient AGO proteins do not cleave but can still silence their target mRNAs ^{108, 109}, most likely through miRNA-like mechanisms such as translational repression and deadenylation. Conversely, natural miRNAs loaded into Ago2 can cleave perfectly complementary target RNAs ^{43, 110, 111}. In this regard, operating RNP machineries for siRNAs and miRNAs share many common features, compared to the considerable differences in the upstream small RNA biogenesis pathways.

PIWI-interacting RNAs

Functions

Mobile genetic elements such as transposons are potential threats to genome integrity in living organisms. In animals, small RNAs called piRNAs play a pivotal role in protecting the germline genome from transposons ^118–120. piRNAs are abundantly expressed in gonads and are longer than AGO-interacting small RNAs, ranging from 20 to 35 nt in length. Unlike miRNAs, the sequences of piRNAs are highly heterogeneous and not well conserved across species. Intriguingly, however, many piRNAs are complementary to genomic regions in which transposon-related repetitive sequences are encoded ^121–124. piRNAs arise from a subset of these genomic loci, called piRNA clusters, where multiple copies of transposon remnants reside in sense and antisense orientation. Based on genetic, biochemical, and bioinformatic studies, it is proposed that piRNAs derived from piRNA clusters (primary piRNAs) act as guide molecules to silence complementary transposon RNAs either by post-transcriptional cleavage or by transcriptional silencing ^118–120. Target cleavage also triggers the production of secondary piRNAs, thereby boosting the amount of piRNAs, via the so-called ping-pong amplification loop ^{125, 126}. Loss or reduction of piRNAs in nematode, fly, zebrafish, and mouse results in up-regulation of transposon RNAs in the germline ^{125, 127–130}. As a result, transposon insertions, DNA damage and apoptosis are increased and gametogenesis is compromised ^{127, 129, 131, 132}.

Box 1: Secondary siRNAs —

siRNAs are not only made from exogenous or endogenous double-stranded RNAs; they can be amplified through the action of RNA-dependent RNA polymerases (RdRPs). In plants, worms, and fungi (but not in insects and mammals), target binding of siRNAs or target cleavage triggers the recruitment of RdRPs, often assisted by their co-factors. RdRPs use the target RNAs as templates to synthesize long double-stranded RNAs, which are then processed into a new pool of siRNAs (“secondary siRNAs”) by Dicer proteins. In worms, some RdRPs can directly synthesize ~22-nt secondary siRNAs with 5´ triphosphate in a Dicer-independent manner ^112–114. Production of secondary siRNAs can be trigged by small RNA species other than siRNAs. An extreme case is known as tasiRNAs in plants, the production of which is directed by miRNA-mediated cleavage of long ncRNAs ¹¹⁵. In C. elegans, 21U RNAs whose characteristics are similar to insect and mammalian piRNAs trigger RdRP-dependent production of secondary siRNAs called 22G RNAs ^{116, 117}. These examples highlight how distinct pathways of ncRNAs intersect and interact with each other, illustrating their remarkable complexity.

RNA elements

Primary piRNAs are transcribed from piRNA clusters by RNAPII as single-stranded long RNAs (piRNA precursor transcripts) ^{125, 133}. piRNA precursor transcripts are then processed into intermediate-sized RNAs by a Dicer-independent mechanism ^{122, 127}. Because piRNA precursor transcripts lack consensus structures and sequence elements except for the complementarity to transposons, how piRNA precursor transcripts are distinguished from other cellular mRNAs has not been revealed (Fig 2C i). Recently, the mitochondrial protein Zuc, also known as MitoPLD in mammals, has been identified as an RNA-processing protein required for piRNA biogenesis ^134–138 (Fig 2C ii). Yet, Zuc cleaves single-stranded RNA with no apparent sequence preference in vitro ^{137, 138}, leaving the fundamental RNA elements in piRNA precursor transcripts unclear. In vivo, Zuc might recognize as-yet-uncharacterized elements embedded in piRNA precursor transcripts with the help of co-factors such as the DEAD box helicase Armitage, Yb, and Gasz ^{134, 135, 139, 140}. Alternatively, piRNA precursor transcripts might be determined by their genomic origin rather than an RNA element embedded in the piRNA precursor transcript itself. In fly ovaries, a subset of bidirectional piRNA clusters is bound by Rhino, a variant of heterochromatin protein HP1, and requires Rhino for efficient piRNA precursor transcription ¹⁴¹. A nuclear DEAD box helicase UAP56 co-localizes with Rhino and binds piRNA precursor transcripts that are derived from the genomic piRNA clusters ¹⁴². Interestingly, UAP56 localizes in close proximity to the perinuclear cytoplasmic compartment called nuage, where piRNA biogenesis and amplification are thought to take place ^{142, 143}. A tempting model is that piRNA precursor transcripts are marked co-transcriptionally by RNA-binding proteins such as UAP56 and are selectively transferred to the piRNA biogenesis machinery. However, UAP56 is not a piRNA-specific factor but is also involved in pre-mRNA splicing and mRNA export ¹⁴⁴. Transcription factors that direct piRNA precursor expression also appear to be rather general and variable among species ^{133, 145, 146}. Indeed, some piRNAs are generated even from protein-coding mRNAs ^{147, 148}. Therefore, with the current knowledge, it is difficult to define a universal RNA element that discriminates piRNA precursors from other transcripts.

RNP machinery

As their name suggests, piRNAs form the operating RNP machinery with PIWI proteins, an animal gonad-specific subfamily of Ago proteins ¹⁴⁹. Unlike siRNAs and miRNAs, piRNAs are thought to be loaded into PIWI as single-stranded RNAs (Fig 2C iii). Based on in vitro experimental evidence, putative precursor piRNAs generated by Zuc processing supposedly have a random nucleotide at their 5′ end ^{137, 138}. In contrast, mature piRNAs naturally tend to have uridine at the 5′ end according to deep sequencing data ^{125, 126}. This bias is most likely produced by preferential loading of precursor piRNAs containing 5′ U into cognate PIWI proteins ¹⁵⁰ (Fig 2C iii). 5′ U may be a sequence constraint during the formation of the RNP machinery. After being loaded into PIWI proteins, precursor piRNAs are trimmed from the 3′ end by an exonuclease called Trimmer whose identity remains unknown (Fig 2C iii) ¹⁵⁰. This reaction, which is modulated by the mitochondrial protein PAPI in insects or Tdrkh in mice ^{151, 152}, is coupled with methylation by Hen1, generating 22- to 30-nt mature piRNAs with a 2′-O-methyl group at their 3′ end ^{105, 150, 153, 154}. The resultant piRNA-containing RISC acts as a sequence-specific RNA-binding module to induce post-transcriptional and transcriptional silencing of transposon RNAs (Fig 2C iv).

The post-transcriptional silencing of transposon RNAs is mediated by PIWI’s slicer activity, which is a conserved feature of Ago family proteins ^{126, 149, 155}. In addition, the slicer activity is important for amplifying piRNAs via the “ping-pong” cycle ^{125, 126}. For example in Drosophila melanogaster oocytes, primary antisense piRNAs that are loaded into Aubergine (Aub), one of the fly PIWI proteins, cleave complementary sense transposon RNAs (Fig 2D i and ii). The cleaved fragment with a 5′ monophosphate is incorporated by another PIWI protein, Ago3, and matured as a secondary piRNA presumably by 3′ end trimming (Fig 2D-iii). The Ago3-loaded sense piRNA then cleaves the original antisense piRNA precursor transcript (Fig 2D iv) ^{125, 126}. This cyclic process is thought to take place in nuage with help of the conserved germline-specific RNA helicase Vasa and Tudor domain-containing proteins ^{156, 157}. As a consequence, the amount of piRNAs that are complementary to active transposon RNAs is amplified ^{125, 126, 130}. Therefore, piRNA-RISC can also act as an RNA-processing complex in the secondary piRNA biogenesis pathway.

Loss of piRNAs or PIWI proteins not only stabilizes transposon RNAs but also reduces repressive epigenetic marks and activates transcription at transposon loci ^158–163. In addition, some PIWI proteins translocate into the nucleus ^{162, 164}, indicating a function of piRNA-RISC in the transcriptional silencing process, although the exact action of nuclear piRNA-RISC needs to be clarified. So far, it has been shown that Drosophila PIWI functionally and genetically interacts with several nuclear factors including heterochromatin protein HP1a ^{160, 165, 166}, zinc finger protein Gtsf1/Asterix ^167–169, and Maelstrom ¹⁶¹. Recent genome-wide screenings for novel piRNA pathway components in Drosophila identified additional candidate genes with nuclear or cytoplasmic functions ^{139, 140, 169}. Future studies on these factors will fill in the gaps and provide a more complete picture of the piRNA machinery that maintains the integrity of the germline genome.

CRISPR-derived RNAs

Functions

Analogous to piRNAs in animals, bacteria and archaea have a unique genome defense mechanism, and small RNAs lie at its core. CRISPR, together with Cas proteins, forms a widespread DNA segment found in about half of all bacteria and approximately 90% of archaea ¹⁷⁰. CRISPR are characterized by the presence of 20– 50-nt repetitive sequences separated by spacer sequences ^{171, 172}. Each spacer is unique and diverse in sequence, but shares high homology with plasmids and viral genomes that the host cells encounter in the environment ^173–175. Upon infection of viruses and phages, a short fragment of the invading DNA called protospacer is excised either upstream or downstream of a DNA motif called PAM ^{176, 177}. The protospacer is then integrated into the 5′ end of the CRISPR following a repeat sequence, adding a new repeat–spacer pair as a memory of infection ^178–180. The CRISPR locus is transcribed from the 5′ leader sequence and processed into small RNAs called CRISPR-derived RNAs (crRNAs), each of which contains a single spacer sequence that acts as a guide to counteract the foreign DNA ^{172, 181}. Hence, the CRISPR-Cas system is an RNA-based adaptive immune system against foreign DNA elements in prokaryotes ^{179, 182}.

Due to its relatively simple structure and the ease of modification, the CRISPR-Cas system has recently been developed into an engineered nuclease that targets specific sequences in the genome ^183–185. Details of this novel genome editing technique are described elsewhere ¹⁸⁶.

RNA elements

Similar to the biogenesis of piRNAs, multiple crRNAs are processed from long CRISPR transcripts (pre-CRISPR) containing the characteristic repetitive sequences as fundamental RNA elements (Fig 2E i, F i and G i). The biogenesis pathways of pre-CRISPR RNAs have some variations and can be divided into three types. In the type I system, each CRISPR repeat is palindromic and forms an approximately 20-nt stem-loop structure when transcribed as a single-stranded RNA (Fig 2E i) ¹⁷⁰. Endonucleases Cas6e/Cas6f, in complex with Cascade, recognize this stem-loop structure and cleave at the 3′ end of the stem (Fig 2E ii) ^187–189. This reaction generates a mature crRNA containing a single protospacer flanked by a ~8-nt upstream sequence and a ~17- to 20-nt downstream stem-loop, both of which derived from CRISPR repeats. Similarly, in the type III system, Cas6 recognizes the single-stranded region within the repeat and cleaves 8 nt upstream from the repeat–spacer junction (Fig 2G ii) ^{190, 191}. This is followed by trimming of the 3′ end, generating 39- to 45-nt mature crRNA. In contrast, the type II system requires another small RNA termed tracrRNA for crRNA recognition and maturation (Fig 2F ii) ¹⁹². tracrRNA is encoded upstream of the CRISPR locus in antisense orientation and contains a ~25-nt sequence complementary to the CRISPR repeat. RNase III cleaves the duplexed tracrRNA and pre-CRISPR at both strands in the presence of Cas9 ¹⁹², a core protein in this system. A part of the repeat sequence left at the 5′ end of the protospacer is trimmed, generating mature crRNA containing ~40-nt protospacer sequence followed by ~20-nt sequence derived from the adjacent repeat.

RNP machinery

Similar to RISC formed by small RNAs and Ago proteins, mature crRNAs form an operating RNP machinery with Cas proteins acting as sequence-specific nucleases (Fig 2E–G iii and iv). The effector Cas proteins are diverse among the three systems described above. In the type I system, Cascade remains associated with the mature crRNA and is involved in the recognition of the homologous DNA through base pairing of the 5′ end of the protospacer (Fig 2E iii) ¹⁹³. Cascade–crRNA complexes induce R-loop formation at the homologous DNA sequence ¹⁹⁴, which recruits the single-stranded DNA nuclease Cas3 that cleaves the looped, non-complementary DNA strand (Fig 2E iv) ^{195, 196}. crRNAs in the type II system remain incorporated in Cas9, which has HNH and RuvC nuclease domains (Fig 2F iii) ^{170, 183}. These domains are responsible for the cleavage of the complementary and non-complementary target DNA strands, respectively, 3 nt upstream from the 3′ end of the protospacer (Fig 2F iv) ¹⁸³. Cleavage by Cas9 also requires tracrRNA, indicating that Cas9–crRNA–tarcrRNA complexes act as the minimal operating RNP machinery in the type II system ^{183, 197}. crRNA and tracrRNA work as a form of chimeric guide RNA (gRNA), allowing the use of the two-component Cas9-gRNA nuclease for genome engineering ^183–185. Type III systems can be divided into two subclasses, type III-A and type III-B, each of which loads crRNA to different Cas complexes (Fig 2G iii). In the type III-A system, crRNA is incorporated in the Csm complex that targets homologous DNA with help of Cas10 ^{182, 198}. In contrast, in the type III-B system, crRNA is incorporated in the Cmr complex ¹⁹⁹. Interestingly, crRNA-Cmr targets RNA rather than DNA, thus representing the diversity and flexibility of the CRISPR-Cas system (Fig 2G iv) ^{199, 200}.

Box 2: Judgment of coding or non-coding by ribosome profiling —

Ribosome profiling is a recently developed method that utilizes digested RNA and sequencing of the portion that is bound by 80S ribosomes to give a profile of ribosome occupancy along transcripts ²⁰¹. This method was applied to experimentally analyze the protein-coding potential of lncRNAs and revealed that a large number of mammalian lncRNAs are bound by ribosomes ²⁰², raising the possibility that these transcripts encode small proteins. Indeed, several functional small proteins are produced from transcripts that were originally categorized as lncRNAs ^203–206. Conclusions from the ribosome profiling data were revised, however, when a metric termed the ribosome release score (RRS) was considered ²⁰⁷. Ribosomes that are engaged in translation are released from transcripts when they reach a stop codon. As a result, when ribosome profiling is applied to protein-coding transcripts, a sharp drop in ribosome occupancy is seen at the start of the 3′ untranslated regions (UTRs). In contrast, translational termination should not occur for ncRNAs. The RRS was therefore developed to distinguish between coding and non-coding transcripts. Indeed, the RRS categorizes numerous lncRNAs together with well-established lncRNAs, which are distinguished from protein-coding mRNAs, indicating that lncRNAs generally are not translated into proteins.

lncRNAs in epigenetic regulation

Functions

In 2013, the GENCODE ver. 7 catalog of human lncRNAs annotated 9,640 manually curated lncRNA loci ²⁰⁸. Among them, multiple lncRNAs interact with various epigenetic regulatory factors (Fig 3A). Xist RNA, involved in X-chromosome dosage compensation in mammals, is expressed from one of the two X-chromosomes in female cells, which results in epigenetic silencing of this chromosome. Xist directly interacts with the Polycomb repressive complex 2 (PRC2) that leads to proper localization of the PRC2 to the inactive X-chromosome to trimethylate histone H3 at lysine 27 (H3K27me3) ^{209, 210}. Analogously, human HOTAIR lncRNA physically associates with PRC2 and modulates PRC2 and H3K27me3 localization at hundreds of genomic sites ^{211, 212}. Furthermore, RNA immunoprecipitation (RIP) with the antibody against EZH2, a subunit of PRC2, captured 9,788 RNA species including lncRNAs in mouse embryonic stem cells ²¹³. Air and Kcnq1ot1 lncRNAs interact with histone H3 lysine 9 methylase G9a to mediate epigenetic silencing of Igf2r and Kcnq1 loci, respectively ^{214, 215}. In contrast, other lncRNAs bind activating epigenetic regulatory proteins. HOTTIP and Mistral act as enhancer-like lncRNAs by interacting with the mixed lineage leukemia (MLL) complex to activate target genes in the HOXA locus ^{216, 217}. Other enhancer-like lncRNAs, ncRNA-a7 and a5, interact with the Mediator complex for their enhancer-like functions ²¹⁸. Thus, lncRNAs possibly participate in various aspects of epigenetic regulation through interacting with various classes of epigenetic regulators including writers, erasers, and readers of histone as well as DNA marks (Fig 3A) ^{219, 220}.

Figure 3. Categorization of lncRNA actions.

(A) lncRNAs recruit histone modification enzyme complexes such as PRC2 to specific chromosomal loci. The structural domain(s) within the lncRNA directly interacts with a subunit of the complex (e.g., EZH2). It remains to be elucidated how the lncRNAs recognize specific chromosomal loci. (B) lncRNAs act as transcriptional regulators. As an example, the lncRNA that acts as a transcriptional co-activator required for assembly of the transcription initiation complex is shown. An RNA modification shown by a red circle as well as distinct RNA structures are critical elements for the assembly. (C) lncRNAs act as structural cores of subcellular bodies (e.g., paraspeckles). The subcellular body is formed upon the assembly of multiple lncRNA–protein subcomplexes. (D) lncRNAs directly interact with specific mRNAs by base pairing. The base-paired duplex provides a binding platform for regulatory proteins (e.g., STAU1) that modulate mRNA stability and/or translation.

RNA elements and RNP machinery

As shown in Table 1, several functional elements or regions within lncRNAs required for epigenetic regulation have been identified. Xist lncRNA has been shown to include several distinct functional elements ^221–224. PRC2 directly interacts with unique AUCG tetraloop structures present at the A-repeat region at the 5′ terminus of Xist ^{221, 222}. Since this interaction is essential for Xist lncRNA function in X-chromosome inactivation, the A-repeat can be considered as the RNA element that forms the functional machinery responsible for histone modification. EZH2 and SUZ12 in PRC2 interact with the highly structured 5′ domain of HOTAIR, suggesting that this RNA structure acts as the RNA element that constitutes the functional machinery for H3K27 methylation ²¹². Intriguingly, HOTAIR was shown to contain a distinct domain required for interaction with the CoREST–LSD1 complex that is responsible for demethylation of histone H3K4 ²¹². The ability to tether two distinct complexes enables RNA-mediated assembly of PRC2 and LSD1 and coordinates targeting of PRC2 and LSD1 to chromatin for coupled histone H3K27 methylation and K4 demethylation. This suggests that lncRNAs may serve as “scaffolds” by providing binding surfaces to assemble selective histone-modifying enzymes, thereby specifying the pattern of histone modifications on target genes ²¹². No common elements are known among lncRNAs that interact with epigenetic modifying complexes.

Table 1.

Elements and machinery of lncRNAs

RNA species	RNA elements	Element functions	RNA-binding proteins	RNP Machinery functions	Reference
Xist	Repeat A (tetraloop)	Recruitment of histone methylase	EZH2 (PRC2)	X-chromosome inactivation	222
	Repeat A (tetraloop)	lncRNA processing	ASF/SF2		223
	Repeat F	Bridge between lncRNA and DNA	YY1		224
HOTAIR	5′ domain (100- to 300-nt stem-loops)	Recruitment of histone methylase	PRC2	Epigenetic silencing	212
	3′ domain (1500- to 2146-nt stem-loops)	Recruitment of histone demethylase	LSD1/CoREST/REST		212
HOTTIP	Exons 1–2	Recruitment of histone methylase	WDR5	Epigenetic activation	217
Mistral	ABM (stem-loop)	Recruitment of histone methylase	MLL1	Epigenetic activation	216
ecCEBPA	R2, R5, R6 (stem-loop (with CpG))	Binding of DNA methylase	DNMT1	Epigenetic activation	219
SRA	STR7, 10, 11 (stem-loop)	Transcriptional coactivation	?	Transcriptional activation	228
	U206 in STR5 (pseudouridine)	Transcriptional coactivation	Pus1 p, Pus3p (isomerize U206)		229
GAS5	GRE-1, GRE-2 (stem-loop)	Binding of nuclear receptor	Glucocorticoid receptor	Transcriptional suppression	226
ncRNAccnd1	-454S, -341a (GGUG)	Inhibition of histone acetyltransferase	TLS/FUS	Transcriptional suppression	227
NEAT1 (NEAT1_2)	ENE-like (triple helix)	lncRNA stabilization	?	Nuclear body architecture	232, 243, 244
	tRNA-like structure	lncRNA 3′ end processing	RNase P (RNA processing)
PAN	ENE (triple helix)	lncRNA stabilization	?	Viral gene expression	245, 246
MALAT1	ENE-like (triple helix)	lncRNA stabilization	?	Epigenetic activation, splicing regulation	243, 244
	tRNA-like structure	lncRNA 3′ end processing	RNase P (RNA processing)		242
1/2-sbsRNA	Alu	mRNA binding	STAU1 (with target mRNA)	Control of mRNA decay or translation	250
TINCR	TINCR box (complementary sequence)	mRNA binding	STAU1	Control of mRNA decay or translation	252

lncRNAs in transcriptional regulation

Functions

Steroid receptor RNA activator (SRA) is a lncRNA that acts as a co-activator of nuclear receptor (NR)-dependent transcription ²²⁵. SRA acts as a “scaffold” for other regulatory proteins for NR-dependent transcription (Fig 3B). In contrast, GAS5 lncRNA is a negative regulator of NR-dependent transcription ²²⁶. GAS5 lncRNA binds to the DNA-binding domain of the glucocorticoid receptor (GR) and acts as a “decoy” glucocorticoid response element (GRE). As a unique example, the promoter-associated ncRNA_CCND1 serves as a molecular “ligand” for a specific RNA-binding protein, namely TLS, causing an allosteric effect to release it from an inactive conformation ²²⁷. This in turn permits gene-specific interaction between TLS and the histone acetyltransferase (HAT) CBP/p300 resulting in the inhibition of HAT function and the repression of transcription.

RNA elements and RNP machinery

SRA contains multiple stem-loop structures, some of which are required for its co-activator function ²²⁸. Those structures function as RNA elements to form the RNP machinery and may provide the binding platform for the protein factors for NR-dependent transcriptional regulation (Fig 3B). SRA was reported to interact with two RNA pseudouridine synthases, Pus1p and Pus3p, which likely pseudouridylate the SRA transcript ²²⁹. Significantly, the pseudouridylation at U206 within a stem-loop of SRA is required for its co-activator function (Fig 3B). The significance of specific RNA structures is illustrated by the decoy function of the GAS5 lncRNA that captures the GR, thus competing with DNA GR elements for binding to the GR ²²⁶. TLS association and HAT inhibition by the ncRNA_CCND1 are putatively mediated by the GGUG sequences which correspond to the TLS binding sequence identified by in vitro selection procedures ²²⁷.

Architectural lncRNAs

Functions

Recently, a number of relatively abundant lncRNAs have been found to localize to specific subnuclear structures. The nuclear paraspeckle assembly transcript 1 (NEAT1) lncRNA was found to localize specifically to paraspeckles where it forms an essential structural component (Fig 3C) ^230–232. The paraspeckle structure is formed with more than 40 paraspeckle protein components (PSPs), which mostly exhibited features of typical RNA-binding proteins and some of which directly interact with NEAT1. NEAT1 serves as “scaffold” to bring >40 PSPs into the paraspeckle as a huge RNP machinery (~0.36 μm in diameter) (Fig 3C) ^{233, 234}. The paraspeckles reportedly act to regulate gene expression by two different mechanisms through sequestration of mRNA and protein factors ^{235, 236}. Therefore, it also acts as a functional “decoy.” Other RNA-dependent nuclear bodies have been reported. Nuclear stress bodies are formed in response to heat shock. Their formation is initiated through the synthesis of lncRNA derived from pericentric tandem repeats of satellite III (Sat III) sequences ²³⁷. The Sat III lncRNA sequesters several splicing-related factors such as SF2/ASF and hnRNP HAP, suggesting that it affects global gene expression. Formation of a distinct subnuclear structure termed the detention center (DC) is involved in structural and functional adaptations of the nucleolus to heat shock or physiological acidosis. This process depends on the expression of the IGS lncRNA. The formation of DC redistributes nucleolar factors and arrests ribosomal biogenesis. Stress termination causes a decrease in IGS lncRNA levels and a return to the active nucleolar conformation, suggesting that IGS lncRNA regulates the structure and function of the nucleolus ^{238, 239}. Thus, at least several lncRNAs function in mammalian cells as architectural RNAs in subnuclear structures where specific regulatory proteins are sequestered.

RNA elements and RNP machinery

The functional elements for architectural lncRNAs are not well understood. Proper synthesis and accumulation of the NEAT1_2 lncRNA isoform is essential for paraspeckle construction ^{231, 233, 240, 241}. Despite it being transcribed by RNAPII, NEAT1_2 lacks a canonical poly(A) tail. The non-canonical 3′ terminus is generated by RNase P-mediated cleavage of the tRNA-like structure adjacent to the 3′ terminus of NEAT1_2 ^{232, 242}. The non-polyadenylated 3′ terminus forms a characteristic triple helix structure that is critical for NEAT1_2 stabilization ^{243, 244}. Interestingly, a similar structure was first discovered as the expression and nuclear retention element (ENE) at 3′ termini of the Kaposi’s sarcoma-associated herpesvirus (KSHV)-encoded polyadenylated nuclear (PAN) lncRNA ^{245, 246}. The related ENE-like elements were identified in the transcripts from six evolutionarily diverse viruses as well as in the cellular Malat-1 lncRNA ^{243, 244, 247}. Thus, the ENE-like element is considered the first family of functional lncRNA elements promoting stabilization. Extensive RNAi analysis revealed seven RNA-binding proteins that are essential for paraspeckle formation ²³³: Among these, one has a role in alternative 3′ end processing required for NEAT1_2 synthesis, three promote NEAT1_2 stabilization, and the remaining three function in paraspeckle assembly, suggesting that at least three distinct elements required for paraspeckle formation may reside in the NEAT1_2 lncRNA. SatIII and IGS lncRNAs likely possess functional elements required for the formation of distinct nuclear bodies, although they remain uncharacterized. Interestingly, these lncRNAs contain repeats of short sequence stretches that may be the elements for efficient sequestration of multiple proteins on a single lncRNA molecule. This is reminiscent of the sequestration of specific RNA-binding proteins into nuclear foci by a nuclear-retained mRNA with aberrant triplet-nucleotide expansion, which causes neuromuscular diseases such as myotonic dystrophy 1 ²⁴⁸. The sequestration of specific RNA-binding proteins into the nuclear foci is achieved by direct interactions with the expanded repeats.

lncRNAs that base pair with mRNAs

Functions

Besides their nuclear functions, lncRNAs can contribute to distinct regulatory events in the cytoplasm. Recent studies reported cytoplasmic lncRNAs that control gene expression through base pairing with mRNAs (Fig 3D). STAU1 mediates mRNA decay (SMD) by binding to dsRNA structures in the 3′ UTR of translationally active mRNAs ²⁴⁹. It was found that STAU1-binding sites can be formed by base pairing between the target mRNA and a cytoplasmic polyadenylated lncRNA termed 1/2-sbsRNA. 1/2-sbsRNA thereby promotes the degradation of target mRNAs ²⁵⁰. Alternatively, a recent report arguing for a general role for STAU1 in modulating translation elongation through structured mRNA coding regions ²⁵¹ raised another possibility that the lncRNAs regulate translation rather than stability of the base-paired mRNAs. The lncRNA TINCR controls human epidermal differentiation through interacting with a range of differentiation-related mRNAs ²⁵², and STAU1-deficient tissue recapitulated the impaired differentiation observed upon TINCR depletion ²⁵². Unlike 1/2-sbsRNA, however, TINCR directly binds to STAU1 without base paring with other RNAs. UPF1 and UPF2, both of which are required for SMD, did not affect differentiation, suggesting that STAU1 and TINCR function is independent of the SMD pathway. Indeed, the TINCR–STAU1 complex seems to mediate the stabilization of differentiation mRNAs, suggesting a new UPF1/2-independent role for STAU1 in differentiation ²³³.

Box 3:Circular RNAs —

Circular RNAs (circRNAs) are produced by unique head-to-tail splicing between a donor site at the 3´ end of a downstream exon and an acceptor site at the 5´ end of an upstream exon. Thousands of circRNAs are made from regular protein-coding genes as well as from non-coding loci, but their functions have long been unclear. Strikingly, two recent reports revealed that circRNAs can function as natural miRNA inhibitors or “sponges,” with dozens of conserved miRNA-binding sites ^{253, 254}. Because of the lack of their free ends, circRNAs are generally stable and abundantly expressed and therefore well suited to counteract miRNAs that are also abundant in cells. More generally, circRNAs with particular cis-acting elements, for example, specific motifs for RNA-binding proteins, could potentially have broader roles in cells.

Sidebar A:In need of answers —

1. Exactly how do miRNAs silence their targets? How are the contributions and kinetics of different modes of miRNA-mediated gene silencing (i.e., deadenylation and translational repression) determined?

2. How can we determine best bona fide miRNA targets and the physiological outcome of their regulation?

3. How can we best design and deliver siRNAs for efficient knockdown, especially in vivo?

4. How are piRNA precursors discriminated from other transcripts? How are they processed into mature piRNAs? Exactly how do piRNAs silence their targets? What are the roles of subcellular platforms such as nuage and mitochondria for the piRNA pathway?

5. How are crRNAs matured?

6. Are there other small RNA classes?

7. What are the functions of circRNAs besides being miRNA sponges?

8. How do different ncRNA pathways interact, cross-talk, and compete with each other?

9. How do lncRNAs specifically interact with RNA-binding proteins?

10. How can we systematically determine RNA elements, RNP machineries, and functions of diverse lncRNAs? How can we categorize them?

RNA elements and RNP machinery

Imperfect base pairing between an Alu element in the 3′ UTR of SMD-target mRNAs and the Alu element in 1/2-sbsRNA provides a binding platform for STAU1 ²⁵⁰. TINCR–mRNA interaction occurs through a 25-nt “TINCR box” motif that is strongly enriched in the target mRNAs and required for TINCR binding. However, STAU1 binding to TINCR does not require target mRNAs ²⁵², indicating that the RNP machineries of two cytoplasmic lncRNAs are built through distinct mechanisms. In contrast to the common base pairing interactions between small ncRNAs (e.g., miRNA and snoRNA) and their target RNAs, few lncRNAs including antisense transcripts have been demonstrated to directly interact with other RNAs through base pairing. The above-described lncRNAs acting as “guides” for STAU1-mediated control of target mRNAs are among those few examples.

Concluding remarks: toward taxonomy of ncRNAs

As described above, small RNA pathways are now relatively well understood, even though their actual RNA elements and RNP machineries are quite diverse. This is largely due to the remarkable commonality of their biogenesis pathways—multi-step processing of precursor transcripts into mature small RNAs—and their mode of action—guiding the effector RNP machinery to their target mRNAs via base paring. On the other hand, such fundamental commonality is still vague among lncRNAs at this point. Thus, future characterization and categorization will be essential for a systematic understanding of the vast ncRNA world. This is an extremely challenging task akin to (or perhaps more complicated than) categorization of protein families and subfamilies, but further accumulation of knowledge will hopefully allow us to establish a “taxonomy” of ncRNAs by defining their RNA elements, RNP machinery, and molecular functions (Fig 4). To this end, recent technologies for genome-wide analyses of RNA structures ^255–257, RNA modifications ^258–263, and RNA–protein interactions ^264–266 will be without doubt powerful tools for identifying RNA elements and RNP machineries. By integrating all those high-throughput data and experimental validations, we may eventually be able to input the information of an unknown ncRNA as a query and predict its function within minutes, just like we now routinely do for proteins.

Figure 4. Taxonomy of ncRNAs.

In a conventional view, all transcripts but protein-coding mRNAs are defined as ncRNAs in the lump (top). Understanding of fundamental RNA elements and operating RNP machineries may allow for the systematic classification or “taxonomy” of ncRNAs based on the relationships between RNA elements, RNP machinery, and molecular and physiological functions (bottom).

Glossary

Ago

Argonaute

Air

antisense Igf2r RNA

Aub

Aubergine

C3PO

component 3 promoter of RISC

Cas

CRISPR-associated genes

Cascade

CRISPR-associated complex for antiviral defense

CBP

CREB-binding protein

CCR4

carbon catabolite repression 4

circRNA

circular RNA

Cmr

Cas module receptor activity-modifying proteins

CoREST

REST corepressor 1

CRISPR

clustered regularly interspaced short palindromic repeats

crRNA

CRISPR RNA

Csm

CRISPR-Cas subtype Mtube

DGCR8

DiGeorge syndrome critical region gene 8

EZH2

enhancer of zeste homolog 2

GAS5

growth arrest-specific 5

Gasz

germline gene containing ankyrin repeat, SAM, and basic leucine zipper domain

GR

glucocorticoid receptor

Gtsf1

gametocyte-specific factor 1

GW182

glycine-tryptophan protein of 182 KDa

H3K27me3

trimethylate histone H3 at lysine 27

HAP

hnRNP A1-interacting protein

HAT

histone acetyltransferase

Hen1

Hua enhancer1

hnRNP

heterogeneous nuclear ribonucleoprotein

HOTAIR

HOX antisense intergenic RNA

Hottip

HOXA transcript at the distal tip

HOXA

homeobox A

HP1

heterochromatin protein 1

Hsc70

heat shock cognate protein 70

Hsp90

heat shock protein 90

Igf2r

insulin-like growth factor 2 receptor

IGS lncRNA

ribosomal intergenic spacer long non-coding RNA

Kcnq1

potassium voltage-gated channel, KQT-like subfamily, member 1

Kcnq1ot1

KCNQ1 opposite strand/antisense transcript 1

lncRNA

long non-coding RNA

Loqs-PB

PB isoform of loquacious

Loqs-PD

PD isoform of loquacious

LSD1

lysine-specific demethylase 1A

Malat-1

metastasis-associated lung adenocarcinoma transcript 1

miRNA

microRNA

MitoPLD

mitochondrial phospholipase D

ncRNA

non-coding RNA

NEAT1

nuclear paraspeckle assembly transcript 1

NOT

negative on TATA

NR

nuclear receptor

P-body

processing body

PACT

protein activator of the interferon-induced protein kinase

PAM

protospacer adjacent motif

PAN2/3

poly(A)-specific ribonuclease subunit 2/3

PAPI

partner of PIWIs

piRNA

PIWI-interacting RNA

PIWI

P-element-induced wimpy testis

PRC2

polycomb repressive complex 2

pre-miRNA

precursor miRNA

pri-miRNA

primary miRNA

PSPs

paraspeckle protein components

R2D2

two dsRNA-binding domains, associated with DCR-2

RdRP

RNA-dependent RNA polymerase

RIP

RNA immunoprecipitation

RISC

RNA-induced silencing complex

RNAi

RNA interference

RNAPII

RNA polymerase II

RNP

ribonucleoprotein

rRNA

ribosomal RNA

RRS

ribosome release score

SatIII

satellite III

SF2/ASF

pre-mRNA-splicing factor 2/alternative splicing factor 1

siRNA

small interfering RNA

SMD

Staufen-mediated mRNA decay

snoRNA

small nucleolar RNA

snRNA

small nuclear RNA

SRA

steroid receptor RNA activator

STAU1

Staufen 1

SUZ12

suppressor of zeste 12 homolog

tasiRNA

trans-acting siRNA

Tdrkh

Tudor and KH domain-containing protein

TINCR

terminal differentiation-induced ncRNA

TLS

translocated in liposarcoma

TNRC6

trinucleotide repeat containing 6

tracrRNA

trans-acting CRISPR RNA

TRBP

trans-activation response RNA-binding protein

tRNA

transfer RNA

UAP56

U2AF65-associated protein

UPF1

up-frameshift 1

UPF2

up-frameshift 2

XIST

X-inactive specific transcript

Yb

female sterile (1) Yb

1/2-sbsRNA

half-STAU1-binding site RNA

Conflict of interest

The authors declare that they have no conflict of interest.