Long noncoding RNA functionality in imprinted domain regulation

Genomic imprinting is a parent-of-origin dependent phenomenon that restricts transcription to predominantly one parental allele. Since the discovery of the first long noncoding RNA (lncRNA), which notably was an imprinted lncRNA, a body of knowledge has demonstrated pivotal roles for imprinted lncRNAs in regulating parental-specific expression of neighboring imprinted genes. In this Review, we will discuss the multiple functionalities attributed to lncRNAs and how they regulate imprinted gene expression. We also raise unresolved questions about imprinted lncRNA function, which may lead to new avenues of investigation. This Review is dedicated to the memory of Denise Barlow, a giant in the field of genomic imprinting and functional lncRNAs. With her passion for understanding the inner workings of science, her indominable spirit and her consummate curiosity, Denise blazed a path of scientific investigation that made many seminal contributions to genomic imprinting and the wider field of epigenetic regulation, in addition to inspiring future generations of scientists.


Introduction
The H19 RNA was the first long noncoding RNA (lncRNA) within the mammalian genome to be discovered, followed shortly by the X-inactive-specific transcript (Xist) lncRNA [1,2]. Following pioneering work by Denise Barlow's as well as other laboratories, imprinted lncRNAs were found to play pivotal roles in regulating parental-specific expression of neighboring imprinted genes (see references marked in bold for Denise Barlow's contributions to the field). Importantly, identification of imprinted lncRNAs genes and their role in epigenetic regulation has ignited the field of research into lncRNAs. Saliently, thousands of lncRNAs have been found in mammals, with robust research efforts aimed at understanding their functionality and mechanism of action [3][4][5][6][7][8], including in X-inactivation (see recent review on Xist lncRNA function [9]). In this Review, we will describe the state of knowledge about imprinted lncRNAs, their function, and, in the spirit of Denise Barlow, address as well as pose the many fascinating and outstanding questions remaining to be answered.

Genomic imprinting
Genomic imprinting is an epigenetic mechanism whereby gene regulation depends on the sex of the transmitting parent [10]. Imprinted genes, which are governed by this mechanism, exhibit transcriptional silencing when inherited by one parent and active transcription when transmitted by the other parent. Often, imprinted genes reside together in clusters or imprinted domains (2 to 20 genes). Since expressed and repressed alleles of imprinted genes reside in the same nucleus, they must be governed by cis-regulatory mechanisms. One of the seminal discoveries in genomic imprinting was the identification of genetic elements called gametic differentially methylated regions (gDMRs) that harbor the "imprinting mark" acquired in oocytes and sperm [10]. These marks are subsequently inherited by embryos and offspring, directing parental-specific allelic expression. Deletions of various gDMRs, experimentally or naturally, have validated the gDMR as a master switch, which coordinately regulates the repressed or active state of multiple imprinted genes within an imprinted domain [11][12][13][14][15][16], thus, earning the name of imprinting center, imprinting control element, or imprinting control region (ICR). Saliently, within an imprinted domain, ICRs possess the "imprinting mark", rather than genes per se [10]. As a result, Denise Barlow refined terminology in the field, classifying allelic expression as "maternally or paternally expressed or repressed" and not as "maternally or paternally imprinted." Importantly, while less than 1% of genes in the mammalian genome are regulated by genomic imprinting [17,18], deletion or loss of imprinting at these genes can lead to lethality or cause a broad spectrum of effects, including growth, developmental, metabolic, and neurological phenotypes [See reviews 19,20]. As such, genomic imprinting has become a central paradigm for understanding how epigenetic mechanisms control expression and repression of genes, gene clusters, and gene pathways involved in development, health, and disease.

Imprinted lncRNAs
Another seminal discovery in genomic imprinting was the identification of lncRNAs within most imprinted domains. As their name implies, lncRNAs are not known to code for proteins.
In comparison to small noncoding RNAs, lncRNAs are distinguished by their long length (greater than 200 bp). Imprinted lncRNAs range in length from approximately 1.9 kb to approximately 1,000 kb, leading Barlow and colleagues to classify the very long imprinted lncRNAs as macro noncoding RNAs (ncRNAs) due to their extraordinary length [21]. Imprinted lncRNAs also differ from mRNAs in their capacity to be spliced. Long ncRNAs are spliced transcripts with a low intron to exon ratio that are present in the cytoplasm and/or are present predominantly as unspliced transcripts that are retained in the nucleus [4]. As a result, imprinted lncRNAs have short half-lives compared to mRNAs [22][23][24][25][26][27][28]. Another important aspect of imprinted lncRNAs is their intimate relationship to ICRs, with promoters either embedded within or closely positioned to the ICR. Generally, ICR-embedded promoters are maternally methylated, reflecting oocyte-specific acquisition of DNA methylation following transcription elongation and histone H3 lysine 36 trimethylation (H3K36me3) disposition [29-31], while lncRNA promoters near ICRs are paternally methylated, acquiring DNA methylation at intergenic sequences during spermatogenesis (Fig 1). Importantly, imprinted lncRNA expression is dependent on the ICR being in an unmethylated state. Another particular fascinating feature of nuclear lncRNAs is their accumulation as a large volume or "cloud" at the parental domain from which they are expressed. Here, imprinted lncRNAs generally function through cis-acting, long-range effects, silencing alleles of multiple neighboring genes, often bidirectionally, at megabase distances [32]. However, it is still uncertain whether transcription per se is a key feature of lncRNA function, with accumulation reflecting the lncRNAs  [10]. Imprinted protein-coding genes, which are expressed from one parental allele while the other copy is silent, reside in clusters or imprinted domains. Maternally expressed genes are represented by a red arrow and text, while paternally expressed genes are represented by a blue arrow and text. Nonimprinted genes, where both alleles unusual length and processing, or whether lncRNA accumulation is indicative of a physical role for lncRNAs in directing the silencing of neighboring imprinted genes [4,33]. Whichever the case, the mechanisms of repressive function are not yet completely understood. Below, we examine proposed mechanism for lncRNA function, focusing primarily on mouse models of the better-studied imprinted domains. For consistency, we have designated these domains according to the associated lncRNA gene: Airn (antisense of Igfr2; Fig 1A), Kcnq1ot1 (Kcnq1 opposite transcript 1; Fig 1B), Nespas (Nesp antisense; Fig 1C), Snhg14 (Small nucleolar RNA host gene 14, also known as Snrpnlt, Lncat, IC-Snurf-Snrpn, Snrpn-Ube3a, U-Ube3aast, Ube3aast and Ube3a-as; Fig 1D), H19 (Fig 1E), and Gtl2 (also known as Meg3; Fig 1F).

Imprinted lncRNA gene transcription runs interference
How lncRNAs regulate parent-specific expression of multiple neighboring imprinted genes within imprinted domains is of major research interest.  1A). This sets up a situation of convergent transcription of the paternal Airn lncRNA in an antisense direction across the paternal Igf2r promoter. Premature terminations of the Airn lncRNA before but not after the Igf2r promoter result in paternal Igf2r reactivation, suggesting that transcription through the promoter is the primary mechanism of paternal Igf2r silencing [36,39]. Furthermore, in embryonic stem cells (ESCs), inducible Airn transcription is both necessary and sufficient to silence paternal Igf2r expression [40]. Collectively, these findings established a clear role for transcriptional interference as a mechanism for silencing overlapping genes. Several models have been proposed for this suppressive mechanism [35]. The first model involves promoter competition, where the stronger promoter, such as Airn, enables better recruitment of the transcription initiation complex than the weaker promoter, such as Igf2r (Fig 2A). Second is the promoter occlusion model, where one transcript, such as the Airn lncRNA, is more robustly and continuously expressed, causing the elongation complex to block the overlapping Igf2r promoter from recruiting the transcription initiation machinery (Fig 2B). The third model proposes dislodgement, where one transcript is faster at engaging the elongation complex, such as the Airn lncRNA, leading to dislocation of the transcription complex at the slower Igf2r promoter (Fig 2C). In addition to transcriptional interference between the paternal Airn and Igf2r alleles, a transcription interference mechanism has been proposed where transcription of the paternal Airn lncRNA would interfere with activators or enhancers within the Airn gene body, thereby preventing initiation and/or up-regulation of paternal, nonoverlapping upstream Slc22a2 and Slc22a3 alleles [32,41]. However, upon deletion of the entire Airn gene, no regulatory elements have been identified that lead to upregulation of paternal Slc22a2 and Slc22a3 alleles [41]. This indicates that a transcription are expressed, also localize to imprinted domains (black text). At least one lncRNA (wavy line) is present with a domain (dashed wavy line represents potential extension of a lncRNA transcript). Regulation of imprinting across an imprinted domain is through a cis-acting mechanism that is controlled by an ICR (large circles) and its associated lncRNA gene. Note: Some imprinted genes in these domain exhibit tissue-specific imprinted expression. The DNA methylated state of ICRs is depicted by large black circles, while an unmethylated ICR is denoted by white circles. Smaller circles represent methylation at secondary, somatic DMRs. Small ncRNAs housed within lncRNAs are depicted as diamonds for endogenous small interfering RNAs, arrowheads for microRNAs, and rhomboids for snoRNAs.
https://doi.org/10.1371/journal.pgen.1008930.g001 Another form of transcriptional interference is transcriptional collision, which by definition is limited to convergently transcribed genes. The collision model proposes the crash of two converging elongation complexes, leading to premature termination of one complex ( Fig  2D) [35]. This mechanism has been proposed as a silencing mechanism within the Snhg14 imprinted domain [42] (Fig 1D). The Snhg14 lncRNA likely originates from exons upstream of the Snurf-Snrpn gene and extends approximately 1,000 kb through to the Ube3a gene [43,44]. In mouse neurons, the Snhg14 lncRNA gene is paternally expressed, while the Ube3a gene is maternally expressed. Deletions of the Snhg14 promoter or truncation of the paternal Snhg14 lncRNA between Snord115 and Ube3a result in reactivation of the paternally silenced Ube3a allele [42,45,46]. Since both parental Ube3a promoters engage the transcription initiation complex, transcriptional interference via promoter competition, promoter occlusion, and dislodgement are eliminated as mechanisms. Furthermore, the paternally expressed Snhg14 lncRNA extends as far as the intronic region of Ube3a between exons 4 and 5, coinciding with the position of paternal Ube3a termination [42]. Thus, transcriptional collision of RNA polymerases on the paternal allele could account for paternal Ube3a transcription stalling, incomplete elongation, and subsequent degradation of the paternal Ube3a transcript [42]. In support of a collision model, treatment with topoisomerase inhibitors, which block unwinding of DNA during transcription, leads to reactivation of paternal Ube3a, where it is proposed that stalling of elongation complexes prevents Snhg14 lncRNA transcription from extending through to Ube3a [47].
Another imprinted domain, Kcnq1ot1, may also function though a transcription interference mechanism ( Fig 1B). Deletion or conditional deletion of the Kcnq1ot1 ICR (also known as KvDMR1); deletion of the Kcnq1ot1 promoter; and premature truncations of the paternal Kcnq1ot1 lncRNA to less than 3 kb all result in loss of paternal Kcnq1ot1 lncRNA expression and reactivation of paternally silent, protein-coding genes across the domain [48,49]. To differentiate between Kcnq1ot1 ncRNA transcription interference and Kcnq1ot1 lncRNA silencing function, posttranscriptional depletion of the Kcnq1ot1 lncRNA via RNA interference was carried out in embryonic and extraembryonic stem cells. No effect was observed on paternal allelic silencing of protein-coding genes in the domain [50]. As RNA interference acts posttranscriptionally, this strengthens a function through transcriptional interference. However, data are conflicting regarding the length of the Kcnq1ot1 lncRNA gene, extending from 83 [37], 91 [51], 121 [32,52], and 471 kb in length [50], suggesting that the Kcnq1ot1 lncRNA is at least 83 kb with a variable 3' end ( Fig 1B) and that transcriptional length of the Kcnq1ot1 lncRNA could reflect differences between cell types and developmental stages. As the Kcnq1ot1 lncRNA originates from the ICR within intron 11 of the Kcnq1 gene [53], a Kcnq1ot1 lncRNA transcript of 90 to 121 kb would not overlap the convergent Kcnq1 promoter, limiting transcriptional interference to a collision model that prematurely terminates the paternal Kcnq1 transcript but not other downstream paternally silenced genes ( Fig 2D). On the other hand, longer transcription of the Kcnq1ot1 lncRNA gene up to 471 kb would extend through five involves the crash of two convergent elongation complexes between the lncRNA transcript and a second transcript, leading to premature termination of the latter. (E-H) Long RNAs themselves also have direct functions in gene regulation. (E) Long ncRNAs serve as a host for small ncRNAs which are excised from the lncRNA transcript. (F) Scaffold function refers to lncRNAs acting as the framework for RNA-binding proteins and modifier complexes interactions with DNA and chromatin to carry out their function. (G) lncRNAs can function as guides by aligning with specific DNA sequences though lncRNA and DNA pairing, attracting chromatin modifiers and directing chromatin modification to target sequences. (H) Decoy function involves lncRNAs acting as a sink or sponge to sequester proteins away from a target site.
Nespas is another imprinted domain that may operate though transcriptional interference ( Fig 1C). The Nespas lncRNA gene is approximately 30 kb in length [28] (Fig 1C). The Nespas promoter is embedded within the Nespas ICR (also known as Nespas-Gnasxl gDMR), and when unmethylated on the paternal allele, Nespas is expressed [12]. Nesp and Nespas are convergently transcribed and overlapping sense-antisense genes. Paternal truncation of Nespas to 12.6 kb, after the Nesp gene and the Nesp somatic DMR, elicits no change in allelic expression or methylation of the paternal Nesp allele [28]. However, two additional paternal truncations, prior to the Nesp promoter at less than 100 bp, and just short of Nesp exon 2 at 10 kb, produce a loss of secondary methylation at the paternal Nesp DMR and a gain in H3K4me3 [28, 54,55]. This indicates that Nespas lncRNA transcription across the Nesp promoter and DMR is required for a gain of de novo paternal methylation at the Nesp somatic DMR, likely via H3K36me3 deposition [54]. However, only the very premature Nespas truncation at 100 bp reactivated the silent paternal Nesp allele, supporting a mechanism of Nespas lncRNA collision of Nesp transcription between the promoter and exon 2 of the Nesp gene ( Fig 2D). Both Nespas truncations also reduced expression of the paternal Gnasxl allele, likely through loss of Nesp DMR methylation [28, 54,55]. Finally, only the very premature truncation leads to paternal silencing of Exon1A and paternal reactivation of the Gnas promoters, connecting paternal Nesp silencing with paternal-specific regulation of these genes.
Going forward, comprehensive studies are required to determine the exact mechanisms of specific transcriptional interference models at imprinted domains. How is transcriptional interference specifically mediated at the DNA sequence level? R-loops regulating the elongation complex have been proposed as one mechanism [47]. Do other structures such a G-quadraplexes play a role? Adding a twist (pun intended), how do these models function in the context of chromatin structure, since current models are mostly based on chromatin as a linear array?

Imprinted lncRNAs as hosts
One function imprinted lncRNAs serve is as precursor transcripts for small regulatory RNAs ( Fig 2E). Intriguingly, imprinted domains present a higher frequency of small ncRNAs compared to rest of the genome [56]. For example, within an intron of the mouse-and rat-specific, paternally expressed Sfmbt2 gene alone, there is a large cluster of approximately 65 to 72 microRNAs (interspersed with B1 retrotransposons and microsatellite repeats) [57,58]. Imprinted lncRNAs themselves also host multiple small ncRNAs, and in some cases, they hold all the small ncRNAs within an imprinted domain [56]. Within the Airn lncRNA resides the retro-Rangap1 (Au76) pseudogene, which harbors hairpin-loop structures that are processed into endogenous small interfering RNAs (485 sequences [59]; Fig 1A). Within the Snhg14 lncRNA lie two large C/D type small nucleolar RNA (snoRNA) clusters, Snord115 (approximately 71 copies) and Snord116 (approximately 136 copies) plus 3 additional snoRNAs ( Fig  1D) [60]. Situated within the Nespas and H19 lncRNAs are 2 and 1 microRNAs, respectively [61,62] (Fig 1C and 1E). The approximately 220 kb-continuous, polycistronic Gtl2lt lncRNA (Meg3-Rian-Mirg), which includes the lncRNA-encoding sequences Rtl1as (microRNAs), Rian (C/D snoRNAs), and Mirg (microRNAs), contains one of the largest collections of small ncRNA clusters (approximately 40-50 microRNAs) within the genome (Fig 1F), followed by the Snhg14 lncRNA [63][64][65][66]. Strikingly, an imprinted gene on human chromosome 19 is composed entirely of 46 microRNAs interspersed with Alu sequences, which are encoded by a single (or few) paternally expressed transcript [67]. While these small RNAs may have roles in regulating expression of multiple transcript targets [68], their action is through trans-acting function [69]. For example, the maternally expressed, Rtl1as ncRNA gene within the Glt2 domain, which harbors two maternally expressed microRNAs (miR-127 and miR-136), regulates the dosage of the paternally expressed Rtl1 gene [63,[69][70][71], while miR-329 within the maternally expressed miR-379/miR-544 cluster (plus 6 other predicted miRs) regulates the paternally expressed Dlk1 gene [65,72]. To date, no role has been identified for cis-regulated function in silencing alleles of neighboring imprinted genes [56]. That being said, it has been proposed that the repetitive nature of imprinted small ncRNA clusters may play a role in gene silencing [56]. Future studies will need to address whether the repetitive and 3D structural nature of small ncRNAs embedded within imprinted lncRNAs contribute to allelic silencing of adjacent imprinted genes.

Imprinted lncRNAs as scaffolds
Transcriptional interference can account for suppressive function by lncRNA transcription for overlapping downstream genes. However, how lncRNAs silence upstream genes is still an outstanding question. A common feature of many imprinted lncRNAs is that transcripts localize as foci or "clouds" [4,73]. Given the short half-lives of imprinted lncRNAs [25-27], this would require a mechanism that is spatially driven. Thus, an alternative mechanism proposed for lncRNA function is as a scaffold for recruitment of chromatin modifiers to target promoters for allelic silencing [32,74]. In this capacity, lncRNAs act as a backbone onto which proteins are loaded to carry out their function [75,76] (Fig 2F). Within the paternal Airn domain, the Airn lncRNA accumulates and forms a RNA cloud on the paternal allele, spreading from the site of transcription to cover the upstream region containing the paternal silent Slc22a2 and Slc22a3 genes [77]. In trophoblast stem cells (TSCs), the paternal Airn lncRNA cloud associates with Polycomb repressive complex 1 (PRC1) and repressive modifications [H3K27me3, H4K20me1, H2AK119 monoubiquitination (u1)] [27]. In ESCs, PRC2 interacts with the Airn lncRNA [78]. Thus, it has been suggested that the Airn lncRNA recruits PRC2 and PRC1 to paternal alleles of imprinted genes within the domain, forming a repressive compartment [27,41]. In TSCs and visceral yolk sac endoderm, broad segments of parental-specific H3K27me3 and H2AK119u1 (20 kb sliding windows) stretch across a region that harbors the paternal Slc22a3 and Slc22a2 alleles [22,41]. Intriguingly, H3K27me3 levels at these broad segments appear to be dependent on Airn lncRNA levels, with overexpression increasing and reduced expression decreasing H3K27me3 levels in cis [22]. This points to a direct correlation between Airn lncRNA levels, H3K27me3 enrichment, and, potentially, PRC2 occupancy. In the placenta, the paternal Airn lncRNA may also acts as a scaffold with histone H3K9 methyltransferase, EHMT2 (also known as G9a) at the upstream paternal Slc22a3 promoter region, thereby contributing to paternal Slc22a3 silencing [77]. Premature truncation of the Airn lncRNA results in a loss of Airn lncRNA association with the Slc22a3 promoter, EHMT2 is not recruited, and the paternal Slc22a3 allele is reactivated [77]. Interestingly, the Airn lncRNA is not observed to recruit EHMT2 to the Igf2r promoter, reinforcing distinct mechanisms of action for the Airn lncRNA in paternal allelic silencing of imprinted genes within the domain [36,77]. Finally, HNRNPK (heterogeneous nuclear ribonucleoprotein K), an RNA-binding protein, may facilitate Airn lncRNA scaffold function. HNRNPK interacts with the Airn lncRNA, with its deletion reducing H3K27me3 across the paternal Slc22a3 and Slc22a2 alleles in TSCs [22]. As a model, HNRNPK may bind the Airn lncRNA and then interact with PRC2 through protein to protein interactions. DNA and/or histone binding domains in PRC2 proteins would mediate H3K27me3 modification across the paternal Slc22a3 and Slc22a2 alleles.
The paternal Kcnq1ot1 lncRNA is also accredited with long-range repressive function of nine imprinted genes within the domain via scaffold functions (Fig 1B) [15,48,49,79]. As part of a lncRNA repressive function, the Kcnq1ot1 lncRNA accumulates on or "coats" the paternal domain [27,50,51,80,81]. During early development, the paternal Kcnq1ot1 imprinted domain harbors a contracted volume marked by PRC2 and PRC1 enrichment and repressive histone modifications (H2AK119u1, H3K9me3, and/or H4H20me1) [27]. In early embryos, stem cells and placentas, this is accompanied by repressive modifications (H3K27me3, H3K9me2/3, and/or H2AK119u1) at paternal alleles of upstream and downstream imprinted gene promoters [23,27,82-85] as well as broad segments of paternal-specific H3K27me3 and H2AK119u1 [22,41]. The Kcnq1ot1 lncRNA also interacts with PRC2 and EHMT2 in placentas [52,86]. These findings have led to the claim that the paternal Kcnq1ot1 lncRNA acts as a scaffold to recruit chromatin modifiers to target upstream and downstream gene promoters in cis, inducing their silencing by PRC2 and PRC1. An additional scaffold function for the Kcnq1ot1 lncRNA has been postulated in postimplantation embryos, where the Kcnq1ot1 lncRNA interacts with DNMT1 at the paternal Cdkn1c and Slc22a18 loci, catalyzing DNA methylation at the paternal secondary Cdkn1c/Slc22a18 DMRs [85]. Finally, the Kcnq1ot1 lncRNA is bound by HNRNPK in TSCs and its deletion reduces H3K27me3 across the paternal Kcnq1ot1 At the Nespas imprinted domain, it is still unclear whether the Nespas lncRNA could play a role in domain regulation outside of transcription interference mechanisms (Fig 1C). However, two observation support Nespas scaffold function. Firstly, PRC2 has been found to interact with the Nespas lncRNA [78]. Secondly, premature termination of the Nespas lncRNA to less than 100 bp fails to repress the paternal Nesp allele, while truncation of the Nespas lncRNA at 10 kb retains paternal Nesp silencing [28, 54,55]. These findings suggest that any scaffold function enabling repressive function of the Nespas lncRNA is contained within the first 10 kb of the Nespas lncRNA. Indeed, a major binding site for the PRC2 complex lies within 10 kb of the Nespas transcription start site [78]. Together, these data suggest a silencing role for the Nespas lncRNA through a scaffold function with PRC2.
Another imprinted domain for which a lncRNA may function as a scaffold is at the Gtl2 imprinted domain (Fig 1F). The Gtl2 lncRNA accumulates at the site of transcription, the maternal Glt2 ICR (called IG-DMR) and the maternally silent Dlk1 allele [24,87]. In ESCs, Dlk1 is very lowly expressed from both the maternal and paternal alleles [24,87], with both showing EZH2 and H2K27me3 enrichment. The Gtl2 lncRNA directly interacts with PRC2 and JARID2 [78,88]. Upon differentiation into embryoid bodies, cortical neurons or neuronal progenitor cells, Dlk1 attains paternal-specific expression while the maternal allele remains repressed [24,87]. Deletions of the Gtl2 promoter or gene body, as well as Gtl2 depletion fail to repress the maternal Dlk1 allele with reduced EZH2 and H3K27me3 enrichment [24,87]. Finally, Ezh2 deletion also leads to lack of maternal Dlk1 repression, albeit without a change in Gtl2 lncRNA expression [87], suggesting that only scaffold functions of the Gtl2 lncRNA is perturbed by Ezh2 deletion.
Many fascinating questions emerge from the above data. With respect to lncRNA coating of an imprinted domain, is the entire domain coated or only specific regions? If varying lncRNA levels lead to different levels of histone modifications, what are the mechanisms controlling lncRNA levels? Does lncRNAs scaffold function lead to lncRNA tethering to chromatin? Alternatively, is tethering an integral component of scaffold function and does it enable spreading of repression across the imprinted domain? Does the lncRNA form multiple scaffolds for numerous independent chromatin modifiers or does the scaffold exist as one large multimeric, macromolecular complex, executing multiple silencing mechanisms simultaneously?

Imprinted lncRNAs as guides
As a guide, lncRNAs target specific DNA sequences though lncRNA and DNA pairing, attracting chromatin modifiers that directly modify target sequences (Fig 2G) [75,76,89]. As stated above, imprinted lncRNAs interact with neighboring imprinted genes in cis; the Airn lncRNA with the upstream paternal Slc22a3 promoter; the Kcnq1ot1 lncRNA with paternal promoters of at least six imprinted genes in the domain [52,85]; and the Gtl2 lncRNA with the maternal Dlk1 allele (Fig 1A, 1B and 1F) [24,87]. One mechanism for molecular guide function is through the lncRNA making contact with specific target genes in triplex lncRNA-DNA-DNA formation. In silico predictions in the human have revealed numerous triplex-forming DNAbinding sites within imprinted lncRNAs and target sites at promoters of imprinted genes, including the AIRN lncRNA to the IGF2R, SLC22A2, and SLC22A3 promoters; the H19 lncRNA to multiple IGF2 promoters; and the KCNQ1OT1 lncRNA to eight imprinted gene promoters in the domain [90]. In the mouse, multiple DNA-binding sites within the Nespas lncRNA have predicted triplex formation at matched target sites at the Nesp, Gnasxl, and Gnas promoters [90]. For the human GTL2 imprinted domain, the secondary structure of in vitroand ex vivo-produced GTL2 lncRNA (approximately 1,600 nt) has been mapped for guide function [91]. The highly structured, GTL2 lncRNA possesses at least five RNA structural elements including triplex-forming DNA-binding sites in structural motif I and two PRC2 binding sites for EZH2 and SUZ12 in structural motif III [91]. 3D modeling of the DNA-RNA triplex in motif I and PRC2 and lncRNA in motif III place them in close proximity, suggesting that they act in concert to execute H3K27me3 modification of chromatin. However, as best we can tell, imprinted genes within the Gtl2 imprinted domain have not been assessed for triplex target sites.
The function of lncRNAs serving as guides leaves open many outstanding questions. Are there lncRNA specific sequences or 2D and/or 3D structural requirements for generating triplex formation and for modifier complex interactions? Four PRC2 proteins (EZH2, EED, SUZ12, and AEBP2) have known RNA-binding capacities [92][93][94][95] with high specificity to Gquadruplexes [96]. To anchor the RNA to chromatin and then target PRC2 to specific target regions, is RNA-DNA triplex formation required in conjunction with PRC2 binding to RNA? Do other chromatin modifying complexes also bind lncRNAs in a similar fashion? As multiple binding sites within a lncRNA align with different target sites within an imprinted domain, does this necessitate multiple copies of the imprinted lncRNA to target imprinted gene promoters or does the 3D structure of the lncRNA and of chromatin enable a single lncRNA to align with and silence multiple promoters simultaneously? In either case, does loss or interference with one target sequence simultaneously impact the alignment and silencing of other target sites?

Imprinted lncRNAs as decoys
Another function attributed to lncRNAs is that of a molecular decoy. Decoy function is described by lncRNAs acting as a sink or sponge to sequester proteins away from a target site (Fig 2H). This could include transcription factors, signaling proteins, splicing factors, and chromatin modifying proteins or complexes [75,76]. At the Gtl2 imprinted domain, the Gtl2 lncRNA may function as a decoy (Fig 1F). In ES cells, blastocysts and neuronal cells, the maternal Gtl2 ICR harbors an enhancer signature and produces bidirectional enhancer transcripts as well as binds AFF3 (AF4/FMR2 family member 3). AFF3 is a core component of the superelongation complex that facilitates an active chromatin state and promotes transcription of the polycistronic Gtl2lt transcript [24,97]. The Gtl2 lncRNA, in turn, may act as a decoy for the Gtl2 ICR, maintaining its active chromatin state [97,98]. This is supported by PRC2 and JARID2 interactions with the Gtl2 lncRNA (which contrasts with weak binding by EZH2 and JARID2 and the lack of SUZ12 and H3K27me3 enrichment at the maternal Gtl2 ICR) [88,99]. Upon maternal Gtl2 promoter deletion, Gtl2 lncRNA expression is abolished, leading to increased EZH2 binding at the Gtl2 ICR [87]. Furthermore, following Ezh2, Eed, and Jarid2 deletion, the maternal Gtl2 ICR is occupied by DNMT3A and DNMT3L and becomes de novo methylated [97,98]. Together, the data suggest that the Gtl2 lncRNA interacts with JARID2 and PRC2 to mitigate EZH2 methyltransferase activity and DNMT3A/3L recruitment at the maternal Gtl2 ICR.
There are many outstanding questions about lncRNA decoy function. As a molecular decoy, lncRNAs may maintain the ICR in an active state and at the same time, the active ICR directs lncRNA expression. What controls this apparent feedback loop? How susceptible is the active ICR to fluctuations in lncRNA levels? Is decoy function mediated via RNA-binding proteins? What is the relationship between weak and strong chromatin modifier interactions with chromatin and lncRNAs? If lncRNAs act as decoys to titrate away positive and negative regulatory proteins, what mechanisms are in place to balance decoy versus other lncRNA function such as scaffold function?

Imprinted lncRNAs as transregulators
In addition to imprinted lncRNAs operating in cis to regulate expression of imprinted genes within their respective domains, imprinted lncRNAs also regulate other genes or domains in trans [100]. Loss and gain of H19 lncRNA function alter expression of imprinted genes in the Kcnq1ot1, Airn, Nespas, and Gtl2 domains [101], while loss and overexpression of the human IPW lncRNA (from the SNHG14 lncRNA) lead to changes in GTL2 lncRNA, microRNA, and snoRNAs in the GTL2 domain [102]. One factor that may contribute to lncRNAs trans function is that imprinted domains can colocalize to the same 3D nuclear space [103,104], generating imprinted gene networks [105], where expression of an imprinted gene in one domain influences expression of imprinted genes in other domains. While upstream or downstream regulatory pathways could indirectly account for these linked expression patterns [105], imprinted lncRNAs are likely to be directly associated with interactions between imprinted domains. With respect to nonimprinted genes, both Gtl2 and H19 lncRNAs are involved in the regulation of the signaling transforming growth factor β (TGF-β) pathway [106,107], the cell fate Wnt/β-catenin pathway [108,109], and the tumor suppressor p53 response pathway [110,111]. Overall, the trans-actions of these lncRNAs point to either lncRNA scaffold or guide function. As a scaffold function, the H19 lncRNA interacts with the epigenetic modifier methyl-CpG-binding domain protein 1 (MBD1) and targets MBD1 to imprinted genes in trans, modulating repression through H3K9me3 [112]. In a similar fashion, IPW interacts with EHMT2, conferring H3K9me3 to the Gtl2 ICR [102]. The Gtl2 lncRNA interacts with JARID2 and PRC2, directing H3K27me3 to gene targets outside of the Gtl2 domain [88]. As lncRNA guide function, bioinformatic predictions of H19 lncRNA-DNA triplex binding sites in humans and mice identified promoter target sites at multiple imprinted genes, including Igf2r, Gnas, Dlk1, Peg1, and Cdkn1c [113], while the human GTL2 lncRNA has DNA triplex binding sites that map to multiple genome-wide target sites in both human and mouse cell lines [107,114]. While the mechanisms of H19 lncRNA regulation of gene targets in trans remain to be fully elucidated, the Gtl2 lncRNA-DNA triplex structures may act together with its PRC2-binding motifs to modulate gene targets in trans [107].
As stated above, many outstanding questions regarding lncRNA scaffold and guide function remain to be resolved. More specifically for lncRNA trans function, what mechanisms operate for imprinted lncRNAs to find their targets? Is lncRNA trans function at imprinting gene networks and other potential chromosomal interactions left to random associations or are there active mechanisms to target imprinted lncRNAs to single gene promoters or to chromosomal networks?

Emerging functions of posttranscriptional lncRNAs
Imprinted lncRNAs as higher-order structures lncRNAs are able to fold into secondary and tertiary higher-order 3D structures, confining them to the nucleus and influencing their regulatory function [115]. While the dynamics of RNA folding is an important component of human GTL2 lncRNA function [111], the recently identified tertiary structure of the GTL2 lncRNA has shed new light on the importance of higher-order tertiary structures [116]. Here, highly conserved motifs within the GTL2 lncRNA interact to generate pseudoknot structures, with specific loops imparting a stable 3D structure ("kissing loops") [116]. Surprisingly, point mutations to the conserved motifs disrupt formation of the pseudoknot and compromise GTL2 lncRNA function in trans, which is independent of the ability of the GTL2 lncRNA to recruit proteins, such as PRC2. To the best of our knowledge, imprinted genes within the Gtl2 imprinted domain have not been assessed for the effects of altered Gtl2 secondary and tertiary structure on their expression. Specific conserved motifs within Kcnq1ot1 lncRNA also contribute to folding and secondary stem-loop structure of the lncRNA [80]. Mutations in this motif region perturbs the structure of Kcnq1ot1 lncRNA and leads to a relaxation of paternal allelic silencing of imprinted genes within the domain [80]. Whether the Kcnq1ot1 lncRNA forms tertiary structures to carry out its function remains to be determined.
The mechanism of how lncRNA higher-order structure contributes to its ability to regulate expression remains unknown. It has been postulated that spatial organization of lncRNA pilots their function by directing protein positioning and localizing lncRNAs to target sites [80,116]. Thus, is the tertiary-folded lncRNA involved in contributing to scaffold or guide function? If structural motifs with lncRNAs must be in close proximity to function, what would be the consequence of inserting a wedge between these structural motifs? Do all imprinted lncRNAs attain a higher-order, 3D structure?

Imprinted lncRNAs as architectural components
Even with the above lncRNA functions, the question still remains as to how lncRNAs navigate upstream of the lncRNA gene to ultimately carry out long-range silencing. The answers to this question usually involve explanations of chromatin interaction generating active and repressed chromatin loops. Emerging evidence suggests that lncRNAs may play a role by facilitating chromatin interactions. The classic model of chromatin looping at imprinted domains is the H19 imprinted domain. CCCTC-binding factor (CTCF) and the cohesin complex generate insulator activity and chromatin looping between the maternal H19 ICR and downstream enhancer, which leads to upstream maternal Igf2 silencing (Fig 1E) [117][118][119]. A recent analysis of topological-associated domains (TADs) at the H19 and Gtl2 domains revealed that while both parental domains possess the same CTCF-associated TAD organization, at the subTAD level, a maternal-specific CTCF site within the unmethylated H19 ICR and Gtl2 promoter bisected the TAD into two smaller subTADs [120]. This physically isolates the active maternal H19 and Gtl2 alleles from the silent maternal Igf2, and Rtl1 and Dio3 alleles, respectively. However, the maternal subTAD that encloses the Gtl2 ICR and promoter, also includes the upstream, silent Dlk1 allele. Thus, the simple scenario of active alleles in one chromosomal loop/subTAD and repressed alleles in another loop/subTAD is unsubstantiated at this domain, requiring additional mechanisms for allelic silencing for the upstream Dlk1 allele. Since the Gtl2 lncRNA prevents activation of the maternal Dlk1 allele, it was proposed that the Gtl2 lncRNA recruits CTCF to the maternal Gtl2 promoter DMR though CTCF's ability to bind RNA [121,122], focusing Gtl2 lncRNA repressive function within the subTAD to the upstream Dlk1 allele [120].
At the paternal Kcnq1ot1 domain, the Kcnq1ot1 lncRNA may function as an architectural component in chromatin looping. On the paternal allele, the Kcnq1ot1 lncRNA interacts with and coordinates chromatin looping between the paternal Kcnq1ot1 ICR and the Kcnq1 promoter. Loss of Kcnq1ot1 lncRNA through artificial methylation of the paternal Kcnq1ot1 ICR, Kcnq1ot1 promoter deletion, premature termination of the Kcnq1ot1 lncRNA, or posttranscriptional depletion of Kcnq1ot1 lncRNA disrupts chromatin interactions at the paternal Kcnq1ot1 ICR and Kcnq1 promoter in mouse embryonic fibroblasts (MEFs) and neonatal heart [123,124]. In MEFs, this results in paternal allelic reactivation of the downstream Kcnq1 and upstream Cdkn1c genes [123]. These data point to a role for the Kcnq1ot1 lncRNA in directing chromatin interactions that may generate an active loop for the Kcnq1ot1 lncRNA gene while exclusion of paternal upstream and downstream alleles may lead to their repression. This role may be developmentally dependent, as posttranscriptional depletion of Kcnq1ot1 lncRNA did not impact paternal allelic silencing of imprinted genes in ESCs and TSCs [50], and CTCF had little to no interaction with the Kcnq1ot1 lncRNAs in TSCs [22].
At the Airn domain, chromatin-chromatin interactions are observed at the maternal Airn domain between the Airn gene body and the Slc22a3 promoter in visceral yolk sac endoderm [41]. At the paternal domain, the Airn lncRNA disrupts these chromatin interactions, as demonstrated by premature truncation of the Airn lncRNA. Since cohesin complex and CTCF binding had little enrichment at the paternal domain, and CTCF had little to no interaction with the Airn lncRNA in TSCs [22], it is suggested that the paternal domain lacks paternal-specific chromatin loops, and that the Airn lncRNA lacks a role as an architectural component.
Many unresolved questions remain for lncRNAs as architectural components. Does CTCF bind (a subset) of imprinted lncRNAs? If so, do lncRNAs play a role in recruiting or biasing CTCF binding to lncRNA promoters and ICRs? Is there also a role for lncRNAs in inducing chromatin looping? What other lncRNA-protein or lncRNA-chromatin interactions could be involved in chromatin looping? Do imprinted lncRNA have chromatin-templating roles on which chromatin-chromatin interactions are mediated?

Discussion
Since the discovery of the first imprinted lncRNA, significant progress has been made in understanding the functional roles of imprinted lncRNAs in imprinted domain regulation. Compelling evidence described above supports multiple functions for how lncRNAs regulate long-range repressive function. Perhaps the most remarkable feature is that these regulatory mechanisms are not mutually exclusive, with a single lncRNA and its transcription regulating different genes within an imprinted domain. For example, the paternal Airn lncRNA gene and its transcript act through transcriptional interference to silence the paternal Igf2r allele, as a Note that some lncRNA functions may be part of the same mechanism, for example, scaffold and guide function, incorporating higher-order lncRNA structure as well.
https://doi.org/10.1371/journal.pgen.1008930.g003 host for small interfering RNAs, in long-range silencing of upstream paternal alleles through a lncRNA scaffold, and possibly as a guide to direct triplex formation (Fig 3A). By comparison, the maternal Gtl2 lncRNA functions as a scaffold, a decoy, and a host for microRNAs and snoRNAs. It may also act through higher-order structure and as an architectural component upstream in cis, and as a guide in trans (Fig 3B). However, it is also possible that there is less multiplicity of lncRNA functions than the above findings indicate. For example, multiple mechanisms, such as lncRNA scaffold and guide function, may be different sides of the same coin, just investigated from different experimental paradigms. Alternatively, more recent, sophisticated, and detailed analyses may be a better reflection of lncRNA functionalities than earlier studies, requiring future experiments to retest past interpretations.
Moving forward, more in-depth, comprehensive studies are required to understand the properties of imprinted lncRNAs as well as to delve deeper into the actual mechanisms pertaining to imprinted lncRNA function and their mode of action. Case in point, to establish the plausibility of posttranscriptional lncRNA functions, it is essential to determine how imprinted lncRNAs are specifically localized to their cis and trans targets as well as the stoichiometric relationships between lncRNAs, protein partners, and their target sites [125,126]. With regard to emerging imprinted lncRNAs functions, it remains to be seen how broadly applicable they are across imprinted domains. At the very least, these emerging functions emphasize the complexity of imprinted lncRNA regulation and stress the need for continued research into potentially novel lncRNA functions. Furthermore, important questions remain regarding lncRNA regulation of imprinted domains in the context of developmental stage-specific and cell-type specific regulation. Saliently, imprinted lncRNAs and their long-range regulation of imprinted domains will continue to be a rich model for bettering our understanding of the mechanisms of lncRNA functionality and increasing our knowledge of lncRNA-associated pathophysiologies in human disease, which ultimately, may lead to medical advances in RNA therapeutics that target long noncoding RNAs.