Localization of RNAs in the nucleus: cis- and trans- regulation

ABSTRACT

The subcellular localization of RNAs correlates with their function and how they are regulated. Most protein-coding mRNAs are exported into the cytoplasm for protein synthesis, while some mRNA species, long noncoding RNAs, and some regulatory element-associated unstable transcripts tend to be retained in the nucleus, where they function as a regulatory unit and/or are regulated by nuclear surveillance pathways. While the mechanisms regulating mRNA export and localization have been well summarized, the mechanisms governing nuclear retention of RNAs, especially of noncoding RNAs, are seldomly reviewed. In this review, we summarize recent advances in the mechanistic study of RNA nuclear retention, especially for noncoding RNAs, from the angle of cis-acting elements embedded in RNA transcripts and their interaction with trans-acting factors. We also try to illustrate the general principles of RNA nuclear retention and we discuss potential areas for future investigation.

Introduction

Pervasive transcription of mammalian genomes produces hundreds of thousands of protein-coding mRNAs and long noncoding RNA (lncRNA) transcripts [1,2]. Also, divergent transcription of active regulatory elements like promoters and enhancers often generates sense and antisense unstable transcripts like upstream antisense RNAs (uaRNAs), and enhancer RNAs (eRNAs) [3–5]. Besides functional differences, the overall subcellular localization of protein-coding mRNAs and these noncoding RNAs (ncRNAs) also differs greatly. Most mRNAs are located in the cytoplasm, while the majority of the ncRNAs tend to be retained in the nucleus [1,5,6]. The mechanisms governing the localization of mRNAs, especially for their processing and export, have been well studied and summarized (see reviews in [7,8]). However, studies of the mechanisms underlying nuclear retention of RNAs, especially of ncRNAs, were relatively scarce until recent years [9–13].

There are several excellent reviews that summarize the mechanism of nuclear retention of lncRNAs and mRNAs [14–16]. In this review, we focus on RNA nuclear retention mainly from the angle of cis-acting elements within the RNA and the trans-acting factors that interact with these elements. We also try to connect nuclear localization of RNA with its function and other processes related to RNA synthesis and processing. In particular, we emphasize one type of RNA nuclear retention – RNA chromatin association – over other types of nuclear retention in a handful of papers that the authors have probed RNA–chromatin interaction. It was believed that chromatin association of RNAs, especially of lncRNAs, might correlate with their function, in regulating gene expression and chromatin structure [17,18]. Non-chromatin-associated RNAs are mainly localized in the nucleoplasm and nuclear bodies such as nuclear speckles [19], which appear to be associated with the processing or degradation of the RNA. To be noted, recent advances also suggest that nuclear speckles can dynamically associate with chromatin, in a manner dependent on and correlated with transcription [20,21], and it was reported that the marker RNAs for nuclear bodies, like MALAT1 or NEAT1 (mainly located in nuclear speckles or paraspeckles, respectively, see below), bind to the chromatin of thousands of active genes [22,23]. These studies indicate that RNA-chromatin association and RNA-nuclear body association might be two highly correlated processes, and the boundary between these two forms of RNA nuclear retention is relatively fuzzy.

We begin with an introduction to how nuclear localization of RNA correlates with its function and regulation, and we illustrate the importance of proper RNA localization for the function of both protein-coding mRNAs and noncoding RNAs. Subsequently, we review mechanisms by which RNAs are retained in the nucleus or on chromatin, from the angles of both cis-acting elements and related trans-acting factors. At last, we discuss the importance of integrating RNA localization with other cellular processes, such as transcription and RNA processing, to better understand the pattern and physiological significance of nuclear retention of noncoding RNA. We also discuss future challenges for studying RNA localization and the function of noncoding RNAs.

Nuclear localization of RNA correlates with its function and regulation

The subcellular localization of RNAs often correlates with their function and regulation. In general, for most protein-coding mRNAs, the mature transcripts are mainly located in the cytoplasm, where they can be occupied by ribosomes for protein synthesis [8]. However, for some mRNAs, their localization is more elaborately regulated in order to modulate the abundance and localization of protein products. For example, it estimated that as many as three-quarters of mammalian multi-exonic genes are transcribed to generate intron-retaining isoforms, and these intron-retaining transcripts are usually retained in the nucleus where they act widely to reduce the expression of host genes [24,25]. Notably, this intron-retention-mediated form of RNA nuclear retention also acts as a means for cells to quickly respond to stimuli such as heat shock, DNA damage, neuronal activation, etc [26–28]. In the mouse neocortex, a subset of well-polyadenylated transcripts retain certain introns and stably accumulate in the nucleus. In response to stimulation, these retained introns undergo splicing and the mRNAs are exported to the cytoplasm for immediate protein synthesis [28] (Fig. 1A). When cells encounter stress conditions such as heat shock, widespread intron retention occurs, and these intron-retaining transcripts are kept in the nucleus to prevent them from engaging with ribosomes. Interestingly, the transcripts from a subset of functionally defined genes can escape this nuclear retention mechanism for normal protein synthesis [27]. Thus, nuclear retention can function as a fast-acting post-transcriptional regulatory mechanism for cells that encounter stress or stimuli.

Figure 1.

Nuclear localization of an RNA correlates with its function and regulation. (A) in the mouse neocortex, a subset of well-polyadenylated transcripts retain selected introns and are retained in the nucleus. These transcripts undergo further posttranscriptional splicing upon neuronal stimulation, and the fully spliced transcript is then exported into the cytoplasm for further protein synthesis. (B) the diagram shows the nuclear location of XIST (on the inactive X chromosome, which forms a condensed structure called the barr body, close to the nuclear membrane), MALAT1 (in nuclear speckles, located in interchromatin regions of the nucleoplasm), and NEAT1 (in paraspeckles, which frequently associate with nuclear speckles). (C) RNA nuclear retention correlates with the distinct functions of the lncRNA FAST in mESCs and hESCs. left: in mESCs, the high abundance of PPIE suppresses the splicing of lncRNA mFast, and these unspliced transcripts are retained in the nucleus; right: the abundance of PPIE in hESCs is low, so hFAST can be fully processed and exported into the cytoplasm, where it interacts with β-TrCP to promote pluripotency maintenance. (D) RNA decay modulates the localization and function of uaRNAs and eRNAs. left: under normal conditions, uaRNAs and eRNAs are removed by the RNA-decay pathway (e.g. by exosomes) during or immediately after their synthesis, before they have a chance to export to move to the cytoplasm. right: after the RNA-decay pathway is blocked, these transcripts accumulate and more RNAs are transferred into the cytoplasm. In the nucleus, the RNA form more R-loop structures and increase genome instability, transcription rate, and chromatin openness; in the cytoplasm, they will occupy more ribosomes and impair the overall translation rate

Unlike protein-coding mRNAs, most lncRNAs are located in the nucleus, and many of them are associated with chromatin [1,29]. The nuclear retention and chromatin association of lncRNAs are often correlated with their function in regulating gene transcription and RNA processing, and also with how they are regulated and processed [17,30–33]. One of the most well-studied lncRNAs is XIST, which mediates the inactivation of one of the X chromosomes in mammalian females and is transcribed from the inactive X chromosome (Xi) [34]. Accordingly, the majority of the XIST RNA ‘coat’ is on the inactivated X chromosome (Fig. 1B) and forms a ‘cloud’-like pattern as viewed by RNA fluorescent in situ hybridization (FISH) [35]. Impairing the chromatin association of XIST abolishes its function in regulating X chromosome inactivation [36,37]. Another example is provided by the lncRNA MALAT1, which was reported to function in regulating gene expression and RNA processing [38,39]. Consistent with this, MALAT1 RNAs mainly locate in nuclear speckles, which are enriched with RNA processing factors (Fig. 1B).

Compared to protein-coding mRNAs, the expression of lncRNAs is more likely to be cell type and tissue-specific [40], implying a functional relevance of lncRNAs in development and cell fate determination [17]. Indeed, an increasing number of studies have revealed that many lncRNAs play roles in development and pathogenesis [41]. Intriguingly, some lncRNAs fulfil their function in these processes by altering their nuclear retention status. For example, TINCR is a lncRNA required for proper epidermal differentiation. The function of TINCR is correlated with changes of its subcellular localization. In undifferentiated epidermal cells, TINCR is mainly located in the nucleus. During differentiation, TINCR levels are elevated and TINCR RNAs are translocated into the cytoplasm, where they interact with and stabilize a range of mRNAs of differentiation-related genes, thus contributing to proper epidermal differentiation [42]. A recent study in human and mouse embryonic stem cells (hESCs) suggests that the distinct subcellular localization pattern of the lncRNA FAST in the two species contributes to the non-conserved function of FAST in ESC pluripotency maintenance [13]. The FAST gene is positionally conserved and specifically expressed in ESCs. Intriguingly, human FAST (hFAST) RNA is mainly localized in the cytoplasm, where it binds to β-TrCP and is required for the pluripotency maintenance of hESCs. Mouse Fast (mFast) RNA, on the other hand, is mainly located in the nucleus, and is not required for the pluripotency maintenance of mESCs. The author found that the nuclear retention of mFast is regulated by the splicing factor PPIE, which is highly expressed in mESCs but not hESCs. Highly expressed PPIE binds to the nascent mFast transcripts and suppresses their splicing, thus promoting their nuclear retention [13] (Fig. 1C).

For some short-lived lncRNAs and regulatory element-associated unstable transcripts like uaRNAs and eRNAs, their subcellular localization correlates with how they are regulated. Experimentally, these noncoding RNAs are usually located in the nucleus and enriched in the chromatin fraction [5,31]. Their nuclear retention at least partially correlates with their short half-lives – they are degraded co-transcriptionally, or immediately after synthesis, mainly through nuclear surveillance pathways [31,43,44], before they have a chance to be exported to the cytoplasm. Although they are short-lived, some of them were proved to play roles in regulating the expression of nearby genes through their transcripts or transcription [5,45,46]. This RNA degradation-mediated nuclear retention may be important for two reasons: The first is that the degradation of these RNAs keeps them at a suitable level for genomic stability and to modulate the proper activity of the regulatory elements which they are associated with. Impeding the degradation of these RNAs by knocking out RNA exosome components alters the activity of regulatory elements like enhancers and promoters, and increases R-loop structures and genomic instability [47,48]. The second reason is that nuclear decay of these ncRNAs prevents their export to the cytoplasm, where they may swap the protein synthesis machinery. Artificial stabilization of these RNAs by blocking the RNA degradation pathway promotes their cytoplasmic localization, which will overwhelm ribosomes and later result in global translational repression [49] (Fig. 1D). Thus, the proper localization of these short-lived ncRNAs correlates with their regulation and prevents them from interfering with other cellular processes.

Mechanisms for nuclear localization and chromatin association of RNA

Both cytoplasmic protein-coding mRNAs and nuclear ncRNAs utilize RNA polymerase II (RNA Pol II) and a similar set of RNA processing factors for their synthesis, packaging, processing, and turnover [1,29]. They may diverge at some point along their maturation pathway before reaching their final destination. Intriguingly, when a nuclear-retained transcript is fused with a cytoplasm located mRNA fragment, the chimeric transcript is mainly retained in the nucleus [50], which suggest that the ‘code’ responsible for RNA localization might be embedded in the RNA sequence. Indeed, studies show that some intrinsic mRNA elements, like the constitutive transport element (CTE) and the RNA transport element (RTE), serve as the ‘postage’ for RNA export and cytoplasmic localization of some viral RNAs and a small subset of mRNAs [51,52]. In addition, it has been reported that an expression and nuclear retention element (ENE) promotes the nuclear retention and stability of PAN transcripts derived from Kaposi’s sarcoma-associated herpesvirus [53]. Thus, identifying key sequences that participate in regulating RNA subcellular localization may serve as an entry point for investigating the mechanism of RNA nuclear retention.

Traditional strategies to identify cis-elements responsible for RNA subcellular localization are mainly based on two principles: 1) truncate the candidate RNA into small fragments, fuse them with a cytoplasm located reporter RNA, and analyse the localization of each fragment to identify the ones that exhibit nuclear localization patterns or promote the nuclear retention of the fused reporter RNA; and 2) serially delete different parts of a candidate RNA to identify key regions that impair the proper localization of the candidate RNA. By utilizing these methods, it has been possible to identify several fragments and cis-elements that are required for nuclear retention of some lncRNAs [38,54–57]. For instance, in the lncRNA MALAT1, two regions called E and M, around 600–1000 nt in length, were identified as being responsible for locating the lncRNA to nuclear speckles [57]. However, these are low-throughput methods, and the identified RNA fragments are usually are long. The low-resolution of these techniques makes it hard to identify the key sequences responsible for the nuclear retention of host RNAs.

New approaches were needed to solve these problems. High-throughput techniques, such as massively parallel reporter assays (MPRAs) or self-transcribing active regulatory region sequencing (STARR-seq), were developed to identify cis-elements that contribute to transcriptional regulation [58,59]. Recently, three groups combined these strategies with subcellular fractionation of RNA, to identify potential cis-elements regulating the nuclear retention of some nucleus-localized RNAs [9–11]. The basic ideas are similar: The first step is to construct MPRA libraries by fusing a cytoplasm-localized reporter gene with tens of thousands of insert sequences derived from synthetic DNAs or randomly fragmented endogenous sequences. These inserts cover all the mature transcript sequences of candidate nuclear-localized RNAs. Next, the reporter library is transfected into cells, followed by subcellular fractionation and isolation of RNAs from the different fractions. Finally, the relative enrichment of reporter-insert-fusion RNAs in each fraction is measured by deep sequencing. It is then possible to identify inserts containing potential cis-elements that promote nuclear retention or chromatin association of RNA. Using this approach, the three groups identified several cis-elements [9–11]. Here we summarize these studies and integrate their discoveries into previously known mechanisms. We will not specifically discuss the mechanism of nuclear surveillance pathway-mediated nuclear retention of unstable nuclear transcripts mentioned above, as it has been well summarized and discussed by previous reviews [44,60,61]. The cis-element-based mechanisms that regulate nuclear retention of RNA can be roughly divided into seven groups (Table 1). These will be described in the following sections.

Table 1.

A list of cis-elements and their interacting trans-factors that participate in regulating RNA nuclear retention

Categories	Cis-elements	Trans-factors	Function	Representative RNAs	References
U1 recognition and splicing	U1-recognition motif (potential 5’ss)	U1 snRNP	nuclear retention, chromatin association, RNA decay	MALAT1, MEG3, Tsix, uaRNAs, eRNA, etc.	[11]
	5’ss	U1 snRNP	nuclear retention	ftz, mvp2, sd4, gag-pol	[67–70]
	NRE	U1 snRNP	nuclear retention	MEG3	[66]
	MALAT1 regions E and M	SON, RNPS1, U1 snRNP	nuclear retention, chromatin association, localization in nuclear speckles	MALAT1	[11,38,57]
SINE-derived elements and C-rich motifs	SIRLON	HNRNPK	nuclear retention	JPX, PVT, etc.	[9]
	HNRNPK binding site (C-rich)	HNRNPK	nuclear retention	MLXIPL, etc.	[9]
	AGCCC	HNRNPK?	nuclear retention	BORG	[54]
	15nt C-rich motif	HNRNPK?	nuclear retention	MALAT1, lincSFPQ, etc.	[10]
IRAlu and paraspeckles	IRAlu	paraspeckles (NONO)	nuclear retention, localization in paraspeckles	CTN-RNA, Linc-p21, etc.	[88,98,99]
Other genomic repeat elements	L1PA16, L2b, MIRb, and MIRc	unknown	nuclear retention	RP11-5407, LINC00173, RP4-806M20.4, etc.	[12]
	CTG expansion	CELF1, MBNL1	nuclear retention, localization in nuclear speckles	DMPK, MBNL1	[113,116,120]
	CAG expansion	MBNL1, U2AF65	nuclear retention, localization in nuclear speckles	HTT	[119,122]
	GGGGCC expansion	TDP43, HNRNPH, MBNL1	nuclear retention, localization in nuclear speckles	noncoding region of C9orf72	[117,118,122]
Tandem repetitive sequences	XIST A-repeat	SPEN	nuclear retention, chromatin association	XIST	[11,126,127]
	XIST LBR binding site	LBR	Lamin localization	XIST	[124]
	XIST B-repeat	HNRNPK, polycomb proteins	nuclear retention, chromatin association, X-chromosome painting	XIST	[90,91,134]
	XIST E-repeat	CIZ1, PTBP1, MATR3, TDP43, CELF	nuclear retention, chromatin association	XIST	[36,37]
	Multiple broad regions in XIST	HNRNPU	nuclear retention, chromatin association	XIST	[37,140]
	RRD	HNRNPU	nuclear retention, chromatin association	FIRRE	[141,142]
R-loop and RNA-DNA Triplex	R-loop	RAD51, Cas9	nuclear retention, chromatin association, telomere targeting	TERRA	[151,152,155]
	RNA/DNA triplex	unknown	nuclear retention, chromatin association	MEG3, Khps1, etc.	[156–160]
SnoRNA-end	snoRNA-end	snoRNA-binding proteins?	nuclear retention, localization in nucleoli and Cajal bodies	sno-lncRNAs	[163,165,166]

3.1. U1 recognition and splicing

The U1-recognition site, which refers to the binding site of U1 snRNP, is also known as the potential 5ʹ splice site (5’ss), and is well known for its function in pre-mRNA splicing [62]. RNA splicing is performed by the spliceosome that assembles on each intron and predominantly comprises small nuclear ribonucleoproteins (snRNPs), including U1, U2, U4, U5, and U6 snRNPs in equal stoichiometry [62,63]. The binding of U1 snRNP to 5’ss is the first step of spliceosome assembly, and it further promotes the assembly of the pre-catalytic spliceosome together with U2 snRNP [64]. Notably, the abundance of U1 snRNP far exceeds that of other splicing-related snRNPs [65], and a transcriptome-wide survey of its RNA binding targets indicates that it associates with a large number of sites located in noncoding regions beyond 5’ss [22]. The high abundance of U1 snRNP and its prevalent binding on RNA transcripts imply non-canonical functions of U1 snRNP beyond splicing.

Recently, it has been reported that U1 snRNP plays a role in promoting the chromatin retention of lncRNAs as well as short-lived noncoding transcripts such as uaRNAs and eRNAs [11]. In this study, Yin et al developed a method named RNA elements for subcellular localization by sequencing (REL-seq), which is strategically similar to the MPRA and STARR-seq-based approaches mentioned above. Different types of reporter were integrated into the genome through the piggyBac system, then a screen was carried out to find cis-elements that promoted the chromatin retention of reporter RNAs. By utilizing this method, they identified 26 chromatin-enriched fragments from 9 representative nucleus-retained transcripts in mouse. For one chromatin enriched 162-nt RNA fragment, which was identified from a retained intron of the gene NXF1, the authors performed random mutagenesis and used the mutated fragments as a new library for REL-seq. In this way, they identified that the U1-recognition motif embedded in the fragment strongly promotes the chromatin retention of reporter RNAs [11].

Notably, they also found that the occurrence of the U1-recognition site and the binding of U1 snRNP is higher in lncRNA transcripts than in protein-coding mRNAs. Impairing the function of U1 snRNP reduces the chromatin association of nearly half of the lncRNAs expressed in mESCs [11]. Consistent with this study, another group also found that knockdown of the U1 snRNP-related protein components impairs the nuclear retention of several lncRNAs, for example MEG3 [66]. Besides, several other studies have revealed that the 5’ss is involved in the nuclear retention of a handful of RNA transcripts encoded by the gene ftz and the viral genes mvp2, sd4, and gag-pol [67–70]. This suggests that U1 recognition is universally employed by many noncoding RNA transcripts, and intron-retaining mRNA transcripts to regulate their chromatin retention.

A systematic comparison of RNA processing between mRNAs and lncRNAs suggests that the overall splicing efficiency of lncRNAs is lower than for mRNAs [6,71] even though lncRNA transcripts harbour a higher density of predicted U1-recognition sites and U1 binding than mRNAs [11,22]. This raises an intriguing question about U1-mediated chromatin retention of lncRNAs: why do these U1-tethered nuclear RNAs not undergo splicing? At least two factors have been reported that may contribute to the inefficient splicing of lncRNAs. The first is a lack of efficient 3ʹ splice sites (3’ss). A genome-wide survey of the distribution of predicted potential 3ʹ ss indicates that compared to mRNA genes, lncRNA genes harbour a substantially lower density of putative 3’ss [11], and the overall strength of 3’ss in lncRNAs seems lower than that in mRNAs [71,72]. The uneven distribution of 5’ss and 3’ss, and the weak internal 3’ss signal may partially account for the inefficient splicing of lncRNA transcripts. The second factor which may affect the splicing efficiency of lncRNA transcript is the binding and abundance of splicing regulators. Consistent with the observation of weak internal 3’ss in lncRNAs, the binding density of the splicing activator U2AF65 on lncRNAs is less than that on mRNAs [71]. In addition, lncRNA exons harbour fewer exonic splicing enhancers (ESEs), which lead to the binding of SR proteins at a lower level in lncRNAs than that in protein-coding mRNAs [72]. Thus, both cis- and trans- elements work in concert to contribute to the inefficient splicing of lncRNAs, which may partially contribute to the constant binding of U1 snRNP. Another intriguing question is how U1 snRNP promote the chromatin retention of lncRNA? U1 snRNP can physically interact with transcriptionally engaged RNA polymerase II (RNA Pol II) [11,73], and the chromatin association of both U1 snRNP and its interacting lncRNAs depends on an active transcription status [11]. It is possible that U1 snRNP tethers and mobilizes its associated lncRNA on chromatin through its interaction with transcriptionally engaged Pol II [11] (Fig. 2A).

Figure 2.

Representative mechanisms regulating RNA nuclear retention and chromatin association. (A) U1 snRNP regulates the chromatin retention of lncRNAs. U1 snRNP binds to U1-recognition motifs embedded in lncRNA transcripts, and further tethers the lncRNA to cis and/or trans chromatin through interaction with transcriptionally engaged Pol II. (B) C-rich motifs interact with HNRNPK to promote RNA nuclear retention. left: HNRNPK interacts with long RNAs harbouring C-rich motifs or SIRLOIN (SINE-derived nuclear RNA LOcalizatIoN, which harbours two C-rich motifs) and promotes their nuclear retention; right: HNRNPK interacts with the super-enhancer (SE)-associated lncRNA CCAT-5 L and with MYC promoter RNA, and promotes chromatin looping between the SE and the MYC promoter through its oligomerization. (C) tandem repetitive sequences interact with multiple regulators to promote the chromatin tethering and spreading of XIST. A-, B-, and E-repeats provide a multivalent platform for recruiting multiple RNA binding proteins and chromatin regulators to promote the chromatin retention of XIST RNA. The oligomerization and phase-separation-potential properties of B-repeat interactors (HNRNPK, polycomb proteins, etc.) and E-repeat interactors (PTBP1, CELF1, MATR3, TDP43, etc.) may promote the spreading of XIST RNA along the X chromosome through a feed-forward mechanism. (D) R-loops (left) and RNA-DNA triplexes (right) regulate the chromatin tethering of RNAs

Another intriguing question related to this mechanism is what happens to Pol II when U1 snRNP cannot be efficiently released? Studies of intron-retaining transcripts in yeast suggest that these unspliced transcripts will inhibit polyadenylation and stall Pol II on chromatin, and promote the degradation of the stalled pre-mRNAs [74]. It is proposed that mammalian cells may use different mechanisms as most of their genes harbour multiple introns [16]. However, as most lncRNAs contain fewer introns than protein-coding mRNAs, and it was reported that cotranscriptional RNA decay is frequently observed for lncRNA [31], it is possible that the constantly binding of U1 snRNP may affect the RNA termination process and promote the decay of some lncRNAs. Consistent with this, when the U1-recognition site is inserted just upstream of an authentic polyadenylation site (PAS), it will inhibit the PAS and trigger RNA decay, possibly through nuclear RNA surveillance-related pathways [11,44,75]. Impairing the function of U1 snRNP indeed increases the RNA abundance and cytoplasmic localization of some lncRNAs, uaRNAs, and eRNAs [11,76]. It has also been reported that U1 snRNP represses cryptic PASs in introns and protects pre-mRNAs from premature cleavage and polyadenylation; furthermore, it was reported that the U1-PAS axis enforces transcriptional directionality [63,77,78]. It is not fully clear how U1 snRNP coordinates the fate of its RNA targets amid all these regulatory events, which occur on chromatin and in a transcription-dependent manner. As U1 snRNP physically interacts with transcriptionally engaged RNA polymerase II [11], U1 snRNP may work in concert with transcriptional machinery and transcription status to ensure proper expression, localization, and turnover of its RNA targets.

3.2. SINE-derived elements and C-rich motifs

Short interspersed elements (SINEs) are a group of retrotransposon elements that widely exist in the mammalian genome. In humans, the most abundant SINEs are Alu elements, which have more than 1 million copies and contribute more than 10% of the genome [79]. Alu elements are unevenly distributed in the genome with a strong preference towards gene-rich regions [80,81]. Most Alu elements are located in the intronic regions of the human genome, where some of them play roles in modulating RNA processing events such as splicing, biogenesis of circular RNA, etc [82–85]. A subset of Alu elements can be exonized by splicing-mediated insertion [85,86]. For most protein-coding genes, this exonization is suppressed [87] and only hundreds of mRNAs harbour Alu elements in their 3ʹUTR [88]. In contrast, Alu elements are commonly present in mature lncRNA transcripts, and these Alu-containing lncRNAs usually exhibit a higher expression level but less tissue specificity [89].

A recent study by Lubelsky and Ulitsky indicated that Alu elements can contribute to the nuclear localization of their long host RNAs in human cells [9]. In their study, the authors constructed an MPRA library containing 5511 sequences derived from 50 nucleus-localized transcripts, including 37 human lncRNAs and 13 3ʹUTRs of mouse mRNAs. Through the screen, they identified three nucleus-enriched regions that overlapped with Alu repeat sequences. They further analysed the sequences and determined that a 42 nt sequence which they named as SIRLOIN (SINE-derived nuclear RNA LOcalizatIoN) can both promote the nuclear retention of reporter RNAs and attenuate their expression [9]. The SIRLOIN elements contain two C-rich consensus sequence motifs (RCCTCCC, R = A/G), and mutation of these motifs abolished the effect of SIRLOIN on both localization and expression level [9] (Fig. 2B, left panel). In another study, Shukla et al. used a similar strategy to identify a C-rich motif, and a similar C-rich pentamer motif (AGCCC) was previously reported as necessary and sufficient for nuclear localization of cytoplasm-localized RNA [10,54]. Moreover, it was reported that XIST harbours multiple C-rich motifs which promote the chromatin tethering of XIST RNA [90,91] (see below). However, the C-rich motif of SIRLON alone failed to promote RNA nuclear retention [9]. Furthermore, the three groups who performed cis-acting element screens did not identify the C-rich motif of XIST, despite the fact that their libraries all contained XIST sequences [9–11]. These observations suggest that C-rich motifs promote nuclear retention of RNA in a context-dependent manner. It is possible that other sequences, possibly related to RNA structure or the binding of other factors, and chromatin environment, may also play a role in this process.

Mechanistic study of SIRLOIN and XIST suggests that these C-rich motifs promote RNA nuclear retention through interaction with an RNA binding protein (RBP), HNRNPK, which is mainly localized in the nucleus. HNRNPK also mediates the nuclear accumulation of its bound RNA outside Alu elements [9]. Moreover, HNRNPK was also reported to interact with the super-enhancer (SE)-associated lncRNA CCAT1-5 L and with MYC promoter RNAs. This interaction further promotes chromatin looping between the SE and the MYC promoter, possibly together with the chromatin organizer CTCF [92], to enhance the Pol II occupancy and transcription of MYC [93] (Fig. 2B, right panel). These studies imply that HNRNPK might serve as a general trans-factor in promoting the nuclear retention of its associated RNAs. An intriguing question is how does HNRNPK promote the nuclear retention of its interacting RNA? A genome-wide survey of HNRNPK binding regions revealed that it mainly binds to gene bodies and open chromatin regions, which indicates that it associates with chromatin, directly or indirectly [94,95]. It has also been shown that the nuclear localization of HNRNPK is dynamically regulated: virus infection or ERK kinase (MEK1) mediated phosphorylation can promote its cytoplasmic translocation [96,97]. However, it is still unknown whether HNRNPK is accompanied by its interacting RNAs when it moves between compartments, and it is unclear how RNA binding affects the function and regulation of HNRNPK. These are some interesting questions that await to be answered in the future.

3.3. Inverted repeated Alu and paraspeckle-related components

Alu elements also promote the nuclear retention of RNAs in another scenario. Specific pairs of Alu elements – inverted repeated Alu (IRAlu) elements, were reported to promote the nuclear retention of their host mRNAs and the lncRNA hLincRNA-p21 [88,98,99]. This regulation is mainly mediated by the interaction between IRAlu and the RNA binding protein p54^nrb/NONO and depends on the secondary structure formed by the IRAlu elements [88,98,99]. The IRAlu RNAs are mainly located in the nucleus and some of them associate with speckle-like, membraneless organelles called ‘paraspeckles’. The paraspeckles frequently associate with nuclear speckles and consist of more than 40 different RBPs, including NONO mentioned above, and an architectural RNA NEAT1 [100–103] (Fig. 1B). NEAT1 and some of the paraspeckle component proteins, such as NONO, PSP, PSF, etc., are essential for the formation of the paraspeckle structure. The morphology and numbers of paraspeckles are regulated by the abundance of NEAT1 RNA. Serial deletion analysis of NEAT1 suggests that its middle domain is required for proper paraspeckle assembly, mainly by recruiting NONO dimers and initiating the assembly of paraspeckles through a phase separation mechanism [55]. The phase-separation process involves interactions between multivalent macromolecules, including RNAs and proteins, which generate micron-sized liquid droplets in aqueous solution [104]. However, studies of CTN-RNA, which is an mRNA isoform of mouse cationic amino acid transporter 2 (mCAT2) gene with a longer 3ʹUTR harbouring an IRAlu element [98], suggest that neither disrupting paraspeckle structure nor deleting NEAT1 RNA obviously impairs the nuclear retention of IRAlu-containing RNAs, but mainly alters their intranuclear distribution [105]. As CTN-RNA interacts with some of the paraspeckle-related protein components independent of NEAT1 RNA or paraspeckle structure [105], it is possible that paraspeckle-related proteins, possibly together with some other proteins, play major roles in promoting the nuclear retention of IRAlu-containing RNAs. NEAT1 RNA, which mainly provides a multivalent platform to facilitate the formation of paraspeckle structure, modulates the interaction between paraspeckle-related proteins and IRAlu-containing RNAs, and influences the intranuclear compartmentalization of IRAlu-containing RNAs.

The nuclear retention of IRAlu elements is dynamically regulated, and this regulation is mainly achieved through three counteracting mechanisms: 1) The IRAlu element can be removed through alternative polyadenylation. Under normal conditions, CTN-RNA is the major isoform and is mainly located in the nucleus, while under stress conditions, CTN-RNA is post-transcriptionally cleaved through alternative polyadenylation to generate an mCAT2 isoform with a shorter 3ʹUTR lacking the IRAlu element. Thus, the mCAT2 RNA will be exported into the cytoplasm for protein synthesis [98]. 2) For a subset of IRAlu-containing mRNAs, the dsRNA-binding protein STAU1 counteracts the binding of NONO to IRAlu, and promotes the export and translation of IRAlu-containing RNAs [106]. 3) The arginine methyltransferase CARM1 dynamically regulates NONO and IRAlu interaction and the abundance of paraspeckles in response to cellular stresses. Under normal physiological conditions, CARM1 on the one hand methylates NONO and reduces its binding to IRAlu-containing transcripts; on the other hand, CARM1 inhibits the formation of paraspeckles by suppressing the expression of NEAT1. These two effects synergistically reduce the nuclear retention effect mediated by IRAlus [107]. In contrast, under certain cellular stresses, the action of CARM1 is attenuated, which increases the level of unmethylated NONO and NEAT1 expression, thus resulting in enhanced nuclear retention of mRNAs containing IRAlus. Besides, under mitochondrial stress conditions, paraspeckles also participate in regulating the nuclear retention of some mitochondrial mRNAs, a process that depends on the presence of IRAlus or AG- or U-rich motifs in these mRNAs. This regulation mainly occurs via increasing NEAT1 transcription and altering the morphology and numbers of paraspeckles [108]. Thus, NEAT1 and paraspeckle-mediated RNA nuclear retention are dynamically regulated as a part of the cell’s responses when stress conditions are encountered.

3.4. Other genomic repeat elements

Besides SINE repeats, many other transposable elements (TEs) were also reported to exist in mature transcripts, especially in the transcripts of lncRNA genes. A survey of 9241 human lncRNAs suggests that about 83% of lncRNAs contain a TE, and TEs comprise 42% of lncRNA sequences [89]. It is hypothesized that some of these inserted TEs might form functional domains that mediate interaction with proteins or nucleic acids, thus contributing to the function and regulation of their host lncRNAs [109–111]. Based on this hypothesis, Carlevaro-Fita et al thoroughly studied the relationship between TEs and lncRNA subcellular localization in an in-silico approach [12]. They uncovered a significant correlation between four repeat elements (L1PA16, L2b, MIRb, and MIRc) and nuclear retention of RNA. The authors validated three lncRNAs harbouring repeat elements L2b, MIRb, and MIRc and confirmed that the nuclear localization of these lncRNAs depends on the presence of wild-type TE [12]. Intriguingly, the authors also uncovered a dose-dependent effect of TEs on nuclear retention of RNAs. These TEs may cooperate with other nuclear retention elements, synergistically promoting the nuclear retention of their host RNAs.

Microsatellite, also called simple sequence repeats (SSRs) or short tandem repeats (STRs), are another type of genomic repeat that comprises about 3% of the human genome [79]. In microsatellites, a short DNA sequence motif, with lengths ranging from one to six nucleotides, is repeated from 5 to more than 50 times [112]. Abnormal expansion of these repeats is responsible for more than 20 neurological and neuromuscular disorders like myotonic dystrophy (DM), amyotrophic lateral sclerosis (ALS), and frontotemporal dementia (FTD) etc [113–115]. For at least some of these diseases, their pathogenesis is partially correlated with nuclear retention of disease gene transcripts, mediated by the presence of the expanded repeat in the RNA [113,116–119] (Table 1). For example, myotonic dystrophy type 1 (DM1) is caused by a CTG trinucleotide expansion in the 3ʹUTR of the Myotonin-protein kinase (DMPK) gene. The RNA transcripts harbouring the expanded CTG repeat are mainly retained in the nucleus and aggregate into nuclear foci. The mutant RNAs contribute to disease pathogenesis by reducing the DMPK protein level and by associating with multiple RBPs involved in RNA processing, thus reducing the availability of these RBPs [113,116]. Interestingly, some RNAs with expanded repeats can form a gel-like structure in vitro and in vivo. The expanded RNAs mainly associate with nuclear speckles, which may contribute to the retention of RNA in the nucleus via a phase-separation mechanism [120].

3.5. Tandem repetitive sequences

The XIST RNA transcript comprises six interspersed repeats: A, B, C, D, E, and F [121]. These repeats are required for the proper function of XIST and X-chromosome inactivation (XCI) [36,37,56,90,122–125]. Cis-fragment deletions and reporter assay analyses suggest that some of these repeats contribute to the nuclear localization and proper chromatin association of XIST [10,11,36,37,90,91,122,126,127]. For instance, deletion of the conserved A-repeat, which consists of 7–8 copies of GC-rich sequences, severely impairs XCI [56]. This deletion does not severely affect the overall X chromosome painting pattern of XIST RNAs detected by RNA FISH [56,128]. However, detailed analysis of RNA–chromatin interactions suggests that loss of the A-repeat impairs the XIST chromatin association, especially at some active gene-dense regions on the X chromosome [126]. Moreover, expressing a 950-nt 5ʹ fragment of XIST containing the A-repeat alone, or fusing the A-repeat with a cytoplasm-localized reporter RNA, causes the majority of RNA transcripts to localize in the nucleus. This indicates that the A-repeat also plays a role in promoting the nuclear retention of XIST RNA [11,127]. The A-repeat mainly interacts with a transcriptional co-repressor protein, SPEN (also known as SHARP), which further recruits the HDAC3 complex to mediate gene silencing and RNA Pol II exclusion [129,130]. Meanwhile, SPEN may also promote the chromatin association of XIST, as depletion of SPEN impairs the chromatin retention of A-repeat-fused GFP reporter RNAs [11]. Also, an LBR (Lamin B receptor) binding region, which located at the downstream of A-repeat and overlap with F-repeat, is required for XCI and targeting of XIST to active genes on X chromosome, mainly through tethering the XIST-coated inactive X chromosome to nuclear lamins [124].

Unlike some other abundant nuclear lncRNAs, which mainly associate with the peripheral chromatin where they are generated (e.g. Haunt [33], Evx1as [32]), or trans-associate with many genomic loci along with nuclear speckle-like structures (e.g. MALAT1 [22,23], NEAT1 [23]), most XIST RNA transcripts ‘spread’ along and ‘coat’ the X chromosome to silence more than 80% of X-linked genes [131]. Intriguingly, RNAs expressed from an ectopic XIST locus on an autosome constantly associate with and mediate the inactivation of the surrounding chromosomal region [132]. This implies that the XIST transcript itself, rather than the X chromosome, or the chromatin environment, plays a major role in determining the specificity of its chromatin association. An interesting question is how is this unique chromatin association pattern achieved? Several studies have systematically investigated the protein interactome of XIST RNA, and revealed that multiple RBPs, transcription regulators, chromatin organizers, etc. interact with XIST RNA directly or indirectly [129,130,133] (Fig. 2C, upper panel). As none of these trans-interacting proteins exclusively interact with XIST, the unique binding pattern of XIST must be regulated by a cooperative regulatory mechanism between these trans-factors and the cis-elements of XIST RNA. Indeed, recently, several studies have indicated that the XIST B and E repeats and their interacting proteins are required for the proper association and spreading of XIST RNA on the X chromosome [36,37,90,91].

Upon depletion of the B-repeat, the ‘cloud-like’ morphology of XIST RNA around the inactivated X chromosome becomes more diffuse, and the spreading of XIST RNA on the X chromosome is severely impaired [90]. The B-repeat consists of multiple C-rich motifs, which are directly bound by HNRNPK, and further recruit the chromatin regulator complexes PRC2 and PRC1 [90,91,134]. Both of these repressive polycomb complexes and the interaction between HNRNPK and the B-repeat are required for the spreading of XIST [90,91]. Intriguingly, it was reported that the polycomb complexes and their marked repressive chromatin domains can be propagated to nearby regions from the nucleation centre where they were initially targeted, partially through modulating nucleosome compaction via a phase separation mechanism [104,135–137]. It has also been reported that HNRNPK can be oligomerized [93,138], and can participate in the formation of paraspeckles [101]. It is possible that the multivalent-interactions of B-repeat-containing RNA, HNRNPK, and the polycomb complexes, create a feed-forward regulatory loop to promote the spreading of XIST RNAs on the X chromosome (Fig. 2C).

Besides the B-repeat, the E-repeat is also required for proper X chromosome association of XIST RNA. Deletion of the E-repeat causes XIST to delocalize from the X chromosome and disperse into the nucleoplasm [36,37]. Mechanistically, the E-repeat acts in two ways. On the one hand, it directly interacts with the nuclear matrix protein CIZ1 (CDKN1-interacting zinc finger protein) to ensure a stable association of XIST RNA with the X chromosome [36,37]. On the other hand, the E-repeat provides a multivalent platform for assembly of the RBPs PTBP1, MATR3, TDP-43, and CELF. These proteins further promote the anchoring of XIST to Xi territory through a phase-separation mechanism [139]. As both the B-repeat and E-repeat are required for proper spreading of XIST RNA along the X chromosome, these two phase-separation-related mechanisms may work in concert with each other to ensure that Xi is coated with XIST RNAs (Fig. 2C, bottom panel).

Depleting another nuclear matrix protein HNRNPU also impaired proper XIST chromatin tethering and abolished XCI, in a mechanism independent of CIZ1 [37,140]. It has been reported that HNRNPU participates in regulating the nuclear retention or chromatin association of other RNAs. For instance, another X chromosome-linked lncRNA, FIRRE, which interacts with HNRNPU through multiple intrinsically repeating RNA domains (RRD) embedded in its transcripts. The interaction between HNRNPU and RRDs promotes nuclear retention and trans-targeting of FIRRE to other chromosomes [141,142]. Moreover, HNRNPU can pervasively interact with chromatin-associated RNAs, and further modulates chromatin structure through its oligomerization in a transcription-dependent manner [143]. The mechanism of how HNRNPU promotes the chromatin tethering of these RNAs is still elusive. However, it prevalently binds to thousands of RNAs derived from both autosomes and inactivated X chromosomes [143,144], and, in contrast to other XCI regulators, it does not exhibit an obviously enriched binding signal on the inactivated X chromosome [37]. These observations imply that HNRNPU may be a general regulator of RNA chromatin tethering, or may act indirectly through modulating the overall nuclear organization. This is consistent with the fact that HNRNPU function was initially reported to function as a nuclear scaffold attachment factor [145].

3.6.  R-loop and RNA-DNA triplex

The above-mentioned mechanisms for retaining RNA in nuclei and on chromatin mainly occur indirectly, through interactions between the RNAs and other factors. It has been reported that some RNAs can associate with their chromatin target directly, mainly through forming R-loop or RNA-DNA triplex structures. Both of these structures are formed by three strands of nucleic acid, including two complementary DNA strands and one single-stranded RNA. Within an R-loop the RNA hybridizes with one of the DNA strands, leaving the other one single-stranded [146]. In the RNA-DNA triplex, on the other hand, the complementary DNA strands hybridize, and the RNA inserts into the major groove of the duplex structure with sequence specificity [147] (Fig. 2D).

The most well-known example of R-loop-mediated RNA-chromatin association and targeting is the CRISPR/Cas system, which is a natural part of the bacterial adaptive immune response against virus infection [148]. The system is now widely used for genome editing and gene expression regulation (reviewed in [149,150]). The most attractive part of this system is that the specificity of the targeting site is controlled by a short RNA sequence, the guide RNA (gRNA), which mediates targeting of the CRISPR/Cas complex through RNA-DNA base-pairing, and finally forms an R-loop structure with the target DNA [151,152]. Intriguingly, this system was engineered to fuse lncRNA transcripts with the gRNA, in order to target lncRNAs to specific chromatin sites [153]. In eukaryotic cells, the R-loop mechanism has been adopted by a few genes for chromatin association of their transcripts. For example, a recent study suggests that TERRA, a lncRNA transcribed from chromosome ends [154], is recruited to chromosome ends through an R-loop-dependent mechanism [155]. However, as the formation of R-loops may trigger genome instability [47,146], cases of RNA targeting via this mechanism are relatively scarce.

Some lncRNAs can also directly associate with their chromatin targets through forming RNA-DNA triplex structures [147,156–160]. This structure usually depends on specific sequence features such as GA-rich homopurine sequences [147]. The RNA-DNA triplex mechanism can mediate the chromatin targeting of RNAs both in cis and in trans. For instance, the human gene encoding dihydrofolate reductase (DHFR) contains two promoters. In quiescent cells, the major promoter of DHFR is repressed, in a manner dependent on a non-coding transcript transcribed from the upstream minor promoter. The noncoding transcripts specifically target the major promoter, which is GC-rich and contains several G-track sequences, by forming a stable triplex structure with it [160]. The targeted non-coding transcripts further repress transcription initiation from the major promoter through interaction with transcription factor TFIIB to dissociate the preinitiation complex from the major promoter [160]. In another example, lncRNA Khps1, which is transcribed in an antisense orientation from the proto-oncogene SPHK1, anchors the promoter region of SPHK1 by forming an RNA-DNA triplex structure with a homopurine stretch upstream of the transcription start site of SPHK1 [157]. For the lncRNA MEG3, on the other hand, the RNA-DNA triplex guides its transcripts to bind to its chromatin targets in trans. Analysis of the targeted chromatin binding sites of MEG3 suggests that they are GA-rich sequences, and the triplex structures are mainly formed by the target sequences and the triplex-forming oligonucleotides (TFOs) located at the 5ʹ of MEG3 RNA [156]. In addition, motif analysis of the lncRNAs HOTAIR and roX2 also found that their targeting sites contain GA-enriched motifs [158], which implies that RNA-DNA triplex structures might also participate in their chromatin association. Nevertheless, GA-rich homopurine sequences are prevalently distributed in the genome, which suggests that other mechanisms are also required to cooperatively determine the specificity of the targeting, especially for the trans-targeting of lncRNAs.

3.7. SnoRNA end

Besides the principal mechanisms that may play general roles in regulating RNA nuclear retention mentioned above, additional mechanisms have also been reported that are related to the nuclear retention of specific sets of RNAs or a specific RNA. For instance, small nucleolar RNAs (snoRNAs) are a family of conserved, nucleus-localized noncoding RNAs that function in RNA modification and rRNA processing [161]. The majority of snoRNA are generated from the introns of snoRNA host genes and protein-coding genes [162]. Intriguingly, the processing of some intronic snoRNAs will generate a class of sno-lncRNAs, whose ends correspond to the positions of intronic snoRNAs. These sno-lncRNAs mainly localize in the nucleus and may associate with specific nuclear bodies such as Cajal bodies and the nucleolus [163]. Notably, fusing snoRNAs to the ends of cytoplasm-localized transcripts will promote the nuclear retention of the fusion-RNAs [164]. Mechanistically, the nuclear retention of sno-lncRNAs depends on their snoRNA ends, which likely recruit interacting partners involved in the function and processing of snoRNAs [163,165,166].

Conclusion and perspectives

In summary, the mechanisms regulating the nuclear retention of noncoding RNAs can be mainly divided into three main categories: 1) association of RNA transcript, via specific nucleotide sequence motifs, with nucleus-localized factors like U1 snRNP, HNRNPK etc., to promote nuclear retention or chromatin association of the RNA (Fig.s 2A-Fig.s 2C); 2) RNA–DNA interactions, which mainly occur through the formation of R-loop or RNA-DNA triplex structures (Fig. 2D); and 3) RNA decay mediated by nuclear surveillance pathways, which degrade some noncoding transcripts before they have a chance to be exported to the cytoplasm (Fig. 1D). Notably, these mechanisms often work together to synergistically promote nuclear localization of RNA. For instance, U1 snRNP can function to promote both chromatin tethering and decay for some of its RNA targets [11]. SPEN-mediated nuclear retention of A-repeat-fused reporter RNA also decreases the expression level of the reporter [11]. It has been proposed that decay is a default fate for nuclear RNA [44]. However, for some stable lncRNAs like MALAT1, NEAT1, etc., other elements, such as those mediating triple-helix structures, may stabilize their transcripts [167,168].

The nuclear retention of an RNA is often regulated by multiple regulators. The best example is XIST. XIST harbours multiple exonic tandem repeat sequences, including the A-, B-, and E- repeats as well as some sequences located downstream of the E-repeat in the last exon, which promote XIST-chromatin association through interaction with different factors [11,36,37,90,91]. These regulators function synergistically to promote proper chromatin tethering and spreading of XIST RNA. Another important observation is that most of the regulators participating in RNA nuclear retention harbour low-complexity sequences (LCS), and some of them can form droplet- or puncta-like structures in vitro and in vivo [55,135,169–171]. Furthermore, many nucleus-retained RNAs are associated with membraneless nuclear organelles like nuclear speckles and paraspeckles [38,55,102]. These findings suggest that a phase-separation mechanism may participate in the regulation of RNA nuclear retention. The detailed mechanism of how phase separation participates in regulating nuclear retention, especially in chromatin tethering, needs to be investigated in the future.

It should be noted that this review focuses on known cis-element and trans-factors, and therefore summarizes only a small portion of the mechanisms, that regulate of RNA nuclear retention. Other factors, such as transcriptional regulation, chromatin environment, RNA modifications, and cellular conditions etc., are also reported or presumed to be involved in this process [172]. Future studies are required to integrate the information derived from all these dimensions, in order to systematically investigate the mechanism and regulation of RNA nuclear retention.

Funding Statement

This work was supported by the National Natural Science Foundation of China [31900439]; Zhejiang University Education Foundation [419000-11111/043].

Disclosure statement

No potential conflict of interest was reported by the authors.