Exploring chromatin structural roles of non-coding RNAs at imprinted domains

Diverse genetic and epigenetic systems of mono-allelic expression have evolved in mammals, together controlling thousands of genes [1]. These mono-allelic gene expression mechanisms provide unique identities to cells, such as in hematopoietic cells or olfactory neurons, or critically modulate the dosage of gene expression, such as in X-chromosome inactivation in females [1]. The epigenetic phenomenon of genomic imprinting is exceptional in that this kind of mono-allelic expression depends entirely on the parental origin of the gene [2]. Some imprinted genes are expressed from the maternally inherited copy only, others only from the paternal copy. About 150 genes are known to be imprinted in humans and mice [3,4] and their correct expression levels are important for fetal growth, development, homeostasis and behavior [5,6].

Imprinting is controlled by oocyte- and sperm-derived DNA methylation marks put onto specialized CpG islands called ‘imprinting control regions’ (ICRs). After fertilization, these epigenetic ‘imprints’ are maintained in the somatic lineages and bring about imprinted expression through diverse mechanisms [2,7,8].

Recent studies have shown that oocyte-acquired histone methylation, particularly histone H3 lysine-27 tri-methylation (H3K27me3), can give rise to parentally biased gene expression as well [9,10]. This non-canonical imprinting is limited to the pre-implantation embryo, and is maintained at only a handful of genes in the extra-embryonic lineages [11–16].

Virtually all the ‘classical’ imprinted genes that are controlled by DNA methylation imprints are clustered in large domains. Most of these imprinted chromosomal domains express one or more long non-coding RNAs (lncRNAs), defined as being more than 200 nucleotides in length [17,18]. Accumulating evidence indicates that these lncRNAs contribute to bringing about imprinted gene expression at close by and distant protein-coding genes. Here, we discuss how imprinted non-coding RNAs control gene expression in cis, with a particular emphasis on their putative roles in chromatin structure. We also discuss emerging insights into trans-regulatory functions.

It is often not well appreciated that numerous non-coding RNAs are imprinted in mammals. For instance, about seven percent of all microRNAs (miRNAs) are imprinted in humans, more than hundred in total. These are mostly transcribed by large host transcription units, each expressing multiple miRNAs [19,20]. One example is the DLK1-DIO3 imprinted domain on human chromosome 14 (mouse chromosome 12), which expresses 53 miRNAs from a 220 kb polycistronic transcription unit, on the maternal chromosome only. Several of these miRNAs control the levels and/or the translation of mRNAs transcribed by other imprinted genes [21–23]. This highlights the considerable interconnectivity between imprinted loci that has arisen during evolution [24,25]. Another large cluster of imprinted miRNAs maps to human chromosome 19. Interestingly, this ‘C19MC’ cluster is primate-specific and expressed in the placenta predominantly [19,26].

Members of one class of small nucleolar RNAs (snoRNAs) are imprinted as well. These so-called C/D-box snoRNAs are thought to guide 2′-O-methylation on specific RNAs, but their precise roles have remained unclear despite recent functional studies [27–29]. The snoRNA DNA sequences are embedded within large transcription units, similarly as the imprinted miRNAs, each expressing multiple C/D-box snoRNAs [19]. One such a host locus is the imprinted DLK1-DIO3 domain, which besides many miRNAs, expresses 38 C/D-box snoRNAs from its maternally expressed ncRNA polycistron. The best-studied cluster of imprinted snoRNAs resides within the SNRPN-UBE3A imprinted domain, which expresses 81 C/D-box snoRNAs from a large polycistronic gene expressed on the paternal chromosome only [19,30].

With respect to chromatin regulation, the most relevant non-coding RNAs are the lncRNAs [2]. In fact, imprinted lncRNAs were amongst the first discovered long non-coding RNAs and have provided many broadly relevant insights [17]. Most imprinted chromosomal domains express at least one lncRNA and these are RNA Polymerase-II transcribed. The very first example was H19 at the imprinted Igf2-H19 domain. This maternally expressed lncRNA was originally described as one of the most highly expressed RNAs during embryonic development, exerting growth-regulating functions [31]. More recent, mechanistic studies revealed that it produces a miRNA (miR-675) that influences muscle development and exerts growth-repressive effects in the placenta [32–34].

Most imprinted lncRNAs originate from their domain's ICR, which acts as a promoter on the unmethylated parental copy. Some are spliced, others not, and several imprinted lncRNAs are retained in the nucleus. These nuclear lncRNAs show different degrees of cis-accumulation onto their locus and exert long-range repressive effects, at some loci across several megabases of chromatin [3,17,35].

In general, lncRNA expression can influence the transcription of protein-coding genes in many different ways [36]. Despite tremendous efforts, however, it has remained complicated to conclude whether observed effects are due to an lncRNA itself or to its transcription [37].

Extensive research during the last years has evoked different models of how lncRNA transcription could interfere with the expression of close by other genes [36,38–40]. As concerns imprinted lncRNAs one transcription-linked mechanism is interference with an overlapping gene transcribed in the opposite orientation (Figure 1A). A well-studied example of this is the imprinted Snrpn-Ube3a domain, where a paternally expressed lncRNA crosses almost one megabase of chromatin, including a distally positioned protein-coding gene called Ube3a. Transcriptional stalling caused by the collision of RNA pol-II complexes coming from opposite directions may explain the lack of Ube3a expression on the paternal chromosome. lncRNA ablation, or expression of truncated forms of the lncRNA that do not overlap Ube3A, cause aberrant activation of this gene on the paternal chromosome [41,42]. Similarly, topoisomerase inhibitors that prevent unwinding of the DNA during transcription —and thereby prevent transcriptional elongation of the lncRNA — reactivate the paternal Ube3A gene [43]. Concordantly, antisense oligonucleotides against the long transcript crossing the domain result in the activation of the paternal UBE3A gene, and such an approach is currently used in different clinical trials to treat Angelman Syndrome, a neuro-behavioural syndrome caused by loss of UBE3A expression [44,45].

A similar model has emerged from detailed studies on the imprinted IGF2-receptor (Igf2r) locus on mouse chromosome 17, which expresses a 117 kb non-spliced lncRNA called Airn that is transcribed oppositely to the Igf2r gene and overlaps its promoter [2]. The allelic lncRNA transcription across the paternal Igf2r promoter blocks RNA polymerase II recruitment, initially in the absence of repressive chromatin marks. Although in differentiating cells there is acquisition of DNA methylation and histone H3 lysine-9 trimethylation (H3K9me3), which provide an additional layer of repression, continued Airn expression is required to keep the paternal Igf2r promoter repressed [46–49].

At other imprinted domains, lncRNA transcription through promoters induces chromatin repression at promoters early in development [17]. At the Gnas locus on mouse chromosome 2, for instance, a lncRNA transcription unit called ‘Nesp-antisense’ (Nespas) overlaps an oppositely transcribed protein-coding gene called Nesp [50] Diverse targeting studies in the mouse, including Nespas truncations, led to activation of the normally silent paternal Nesp allele. This highlights the importance of transcriptional overlap in the promoter repression (Figure 1B), which involves both histone and DNA methylation [51,52]. In a similar manner, at the imprinted Zdbf2 locus on mouse chromosome 1, transient transcription during preimplantation development of a lncRNA (called Liz) brings about DNA methylation, close to the Zdbf2 gene [53,54]. As part of the mechanism, RNA polymerase-II could bring the KMT SETD2 to the chromatin, which induces histone H3 lysine-36 tri-methylation (H3K36me3) across the transcribed region. This histone modification is recognized by the DNA methyltransferase DNMT3B (through its PWWP domain) subsequently, which induces de novo DNA methylation [55–57].

LncRNA-controlled genes at several imprinted domains are located hundreds to thousands of kilobases away from the lncRNA gene [3,58]. These ‘long-distance’ effects have given rise to models in which the lncRNA itself brings about gene repression (Figure 2). Developmental studies have shown that such long-distance repression occurs in a tissue-specific manner at several of the domains [35]. At the Igf2r domain, in the extraembryonic lineages, the paternally expressed Airn mediates the allelic repression of several non-overlapping genes positioned up to several megabases away [3,58]. Upon trunctions of this lncRNA, this long-distance repressive effect no longer occurs [46,58,59].

At Dlk1-Dio3, similarly, the allelic expression of an lncRNA called Meg3 is required to repress a distant protein-coding gene involved in Notch signaling, called Dlk1, in different somatic tissues [60–62]. Knock-out and overexpression studies have suggested that Meg3 expression controls genes on other chromosomes as well, including TGF-Β and p53 pathway genes in human cancer cells [63–67]. A similar trans effect has been reported for the lncRNA IPW generated from the SNHG14 gene at the SNRPN-UBE3A domain (chromosome 15q11–13). This lncRNA dampens in trans the promoter of the MEG3 non-coding polycistron at the DLK1-DIO3 domain, a process that seems to involve repressive H3K9me3 [68]. This provides yet another example of the intricate regulatory links that exist between imprinted loci [25,69–71].

Another locus showing long-range repressive effects of an lncRNA is the Kcnq1 domain on mouse chromosome 7. The integrity of a 91 kb lncRNA called Kcnq1ot1, particularly a 900 bp region at its 5′ end, is important for the allelic repression of no fewer than eight genes at the proximal and distal parts of this multi-megabase domain [72–76]. Several of the target genes show placental-specific imprinting, indicating that lineage-specific factors likely contribute to the long-range repressive effects of this essential lncRNA [3,58,77]. Combined, the above examples illustrate that several imprinted lncRNAs repress protein-coding genes in cis, and that some control genes on other chromosomes as well.

Genome-wide reporter-based studies have revealed that many non-imprinted lncRNA genes exert a positive effect on the expression of other genes in their neighborhood [78,79]. These ‘enhancer-like’ effects of lncRNA gene promoters contrast with the observed effects of imprinted lncRNAs, which mostly repress neighboring genes, through nucleation and spreading of repressive histone modifications across large regions [17].

For the imprinted lncRNAs Kcnq1ot1, Airn and Meg3 evidence has been obtained for a direct role in chromatin repression. All three are retained in the nucleus and show a certain degree of cis-accumulation onto their imprinted domains. This focal accumulation is still detected hours after chemical inhibition of RNA polymerase-II, concordant with the reported intermediate stabilities of these nuclear lncRNAs [58,75,80,81].

Another similarity between Kcnq1ot1, Airn and Meg3 is their reported interaction with components of chromatin regulatory complexes (Table 1). In preimplantation embryonic cells and in the placenta, the paternally expressed, 91 kb Kcnq1ot1 (Kcnq1 domain) co-localizes and interacts with components of the Polycomb repressive complexes 1 (PRC1) and -2 (PRC2) [74,75,82,83]. It also interacts with EHMT2 (also called G9A) [74], a lysine methyltransferase (KMT) that methylates lysine-9 on histone H3. Concordantly, there is allelic acquisition of EZH2 (PRC2)-mediated H3K27me3, RING1B (PRC1)-mediated H2AK119u1 and EHMT2-mediated H3K9me2 across the paternally repressed genes in trophoblast cells and in the placenta [58,75,77]. In trophoblast stem cells (TSCs) that expressed a truncated form of Kcnq1ot1, H3K27me3 levels were strongly reduced across the entire Kcnq1 imprinted domain [58,74]. Ehmt2 knock-out in mice gave biallelic expression of several of the placental-specific Kcnq1ot1 targets in the placenta [84], and, similarly, the essential PRC2 component EED contributes to the process as well [85].

A recent study suggests that the Kcnq1ot1 lncRNA interacts with the nuclear matrix protein hnRNPK. This RNA-interacting protein is essential for the PRC2-mediated H3K27me3 across the imprinted Kcnq1 domain in TSCs [58]. One emerging model (Figure 2) is that hnRNPK enhances the recruitment and spreading of PRC1 complexes, a process that initiates at CpG islands that were already bound by PRC complexes beforehand [58,86].

A similar picture has emerged for the paternally expressed, 117 kb lncRNA Airn (Igf2r domain), which in the extraembryonic lineages represses multiple genes across several megabases. In murine TSCs and in placenta, a truncated form of this lncRNA no longer gave gene repression in cis [3,58,59]. Recent gene targeting studies in mice show that the long-range repressive effects of Airn are not mediated by regulatory sequence elements within the Airn lncRNA gene, excluding transcriptional interference mechanisms at the distant non-overlapping genes controlled by Airn. Rather, these repressive effects correlate with the broad spreading of PRC2-mediated H3K27me3 and PRC1-mediated H2A-lysine-119 mono-ubiquitination (H2AK119u1) on the paternal chromosome predominantly [58,87]. Airn levels are crucial for the allelic recruitment of RING1B (PRC1) and EZH2 (PRC2). Enhancing lncRNA Airn copy numbers per cell, by CRISPR-VP16 mediated transcriptional activation, gave enhanced recruitment of PRC complexes onto the paternal chromosome [58]. Airn had been shown earlier to facilitate EHMT2 recruitment, which correlates with paternal allele-specific H3K9me2/3 enrichment [59]. Also Airn lncRNA seems to interact with hnRNPK and this could enhance recruitment of PRC complexes to the chromatin [88]. In agreement with this hypothesis, the allelic enrichment and spreading of H3K27me3 across the large Igf2r domain requires continued expression of the hnRNPK protein in TSCs [58].

Meg3 lncRNA seems to have a similar mode of action, in somatic tissues. Its expression represses in cis a developmental gene called Dlk1, located on the proximal side of the imprinted domain [61]. Different studies have reported Meg3 association with PRC2 components (EZH2 and JARID2) and RNA precipitation assays on cross-linked chromatin suggest binding to hnRNPK as well [58,82,83]. In the absence of Meg3 lncRNA, there is no longer acquisition of allelic Dlk1 repression, and this is observed following depletion of EZH2 (PRC2 complex) as well [61]. Similarly as for Airn and Kcnq1ot1 [58], the combined data suggest that Meg3 lncRNA enhances in an allelic manner the histone modifying activities and possibly also the spreading of PRC complexes, through still poorly understood mechanisms (Figure 2).

The above examples evoke similarities with X-chromosome inactivation in females, which is a cis repressive mechanism controlled by an lncRNA (called Xist) that involves PRC1, PRC2, EHMT2 and hnRNPK, and other proteins not yet been explored in genomic imprinting [1,88,89]. However, care needs to be taken before drawing firm conclusions. The methodologies used to explore Xist, for instance, have been more focused on the lncRNA itself, with functional identification of chromatin-binding RNA motifs. Complementary technologies have also confirmed a direct interaction between Xist and hnRNPK, which has not yet been shown for Airn, Kcnq1ot1 or Meg3 [90–93].

Because of the parental allele-specific DNA methylation imprints, at several imprinted domains there is allelic association of chromatin structural proteins. At several ICRs, and also at secondary DMRs at which the allelic methylation is acquired during development, there is binding of CCCTC-binding factor (CTCF) to the unmethylated allele only (the protein does not bind methylated DNA) [94–97]. This allelic CTCF binding and the CTCF-associated cohesin complexes contribute to imprinted gene expression [98]. Particularly, CTCF mediates long-range chromatin loops with distant other regions on the CTCF bound parental chromosome. Recent studies have explored these structural interactions by using allelic ‘chromosome conformation capture’ (3C) and 3D DNA FISH-based approaches [98].

At the Igf2-H19 locus, CTCF binding to the unmethylated copy of the ICR brings this region in close proximity to distal regions on the maternal chromosome. Both in mice and humans, this insulates the Igf2 gene from its distally located enhancers, thus leading to the imprinted Igf2 expression from the paternal chromosome mostly [96,99–102].

At the Dlk1-Dio3 domain, CTCF binds the promoter–CpG island of the Meg3 gene, on its unmethylated maternal copy only [96,103]. Also here, allelic CTCF recruitment brings about specific long-distance structural interactions on the maternal chromosome predominantly. Particularly, the Dlk1 gene shows close proximity to the lncRNA focus on the maternal chromosome, and this proximity effect contributes to its imprinted expression from the paternal chromosome predominantly [96]. Interestingly, 3D distance measurements between FISH probes show that the imprinted domain is more loosely compacted on the maternal chromosome (compared with the paternal chromosome), which may facilitate the observed CTCF-mediated looping patterns [96,104].

A similar picture has emerged for the Kcnq1 imprinted domain. Here, CTCF binds the unmethylated paternal copy of the ICR, which also comprises the promoter that drives Kcnq1ot1 expression on this parental chromosome [105]. The allelic CTCF binding mediates specific long-range interactions on the paternal chromosome, detected by 3C-based technology, that correlate with the allelic expression of several genes within the domain [75,102,106]. Another locus that shows both allelic CTCF binding and allelic lncRNA expression is the imprinted Zdbf2 domain [107].

Could the allelic lncRNA expression and the allelic binding of CTCF be mechanistically linked? Possibly, transcription factor binding and lncRNA promoter activity keep CTCF binding sites unmethylated, thus ensuring the continued allelic association of this chromatin structural protein (which does not bind methylated DNA [97]). Continued promoter activity at Meg3 protects indeed against the acquisition of de novo DNA methylation in early embryonic cells [108,109]. Point mutations within transcription-factor binding sites at the ICRs of the human IGF2-H19 and KCNQ1 domains have provided evidence for such a scenario as well [110–114]. Conversely, CTCF itself may protect the unmethylated allele against de novo DNA methylation [110,111,115,116], thus ensuring continued transcription of the lncRNA from the unmethylated parental allele only.

Since Meg3, Kcnq1ot1 and Airn show a relatively stable focal accumulation onto their locus [58,61,75,80], this could locally influence CTCF-linked higher-order chromatin structure. CTCF comprises indeed a putative RNA binding domain (RBD) that is functionally important [117,118]. Recent studies suggest that binding of locally transcribed RNAs to the RBD is important for CTCF's association to many of its recognition sites in the genome. This impacts the 3D organization of the genome through the formation of specific chromatin loops [117,118]. It remains to be explored in mice on an F1 background between two phylogenetically distant strains whether there are direct allelic interactions between CTCF and imprinted lncRNAs and to what extent these may influence chromatin loop formation.

How and when lncRNA–protein compartments are formed at imprinted loci, and what controls their developmental regulation, remains unclear. Structural RNA features could be important. Several recent studies explored in detail the structure of MEG3 in vitro and in cells [63,119], and interacting RNA loops within the lncRNA were shown to be essential for the trans effects of MEG3 on the p53 pathway in cancer cells [63]. Whilst the RNA sequences of imprinted lncRNAs are generally not well conserved, specific secondary and tertiary structures may be comparable between different mammalian species, and may be important as docking sites for RNA–protein interactions.

Specific RNA sequence elements could be important as well; for instance in the association of lncRNAs to specific target genes in trans. In one interesting study on human cancer cells, expression of MEG3 modulated the expression of TGF-B pathway genes, and this was linked to the formation of RNA–DNA triplex structures across several of these target genes [65]. Although further studies are required, such a process could provide specificity to the trans roles of lncRNAs.

The non-imprinted lncRNAs MALAT1 and NEAT1 are linked to the formation of membrane-less nuclear bodies called speckles and paraspeckles in specific cell types and under particular conditions [120–122]. Furthermore, emerging evidence on the heterochromatin-linked satellite RNAs and other non-imprinted RNAs suggest that RNA–protein aggregates can potentially form through liquid-liquid phase separation (LLPS) mechanisms (reviewed in [123], an aspect that has not yet been explored in the context of imprinted domains.

Sub-nuclear localization could impact the process as well, given that at the Dlk1-Dio3, Kcnq1 and other imprinted domains, the lncRNA-expressing parental chromosome displays a more central localization in the nucleus than the opposite parental chromosome [80,124]. The available data so far evoke a model in which focal accumulation of lncRNA and associated chromatin-regulatory complexes creates an aggregate-like organization that brings specific loci in close proximity through protruding chromatin-loop formation and mediates gene repression (Figure 3). At some imprinted domains, interestingly, lncRNA/protein compartments seem to exclude RNA polymerase-II [75], which could be an important aspect of the imprinting process as well. LncRNA-mediated gene repression at imprinted domains is a rather complicated business, and we are only at the beginning of understanding its intricacies.

• Imprinted gene domains have provided strong paradigms for exploring the regulation and roles of lncRNAs in mammals. Ongoing research efforts unravel cis-regulatory chromatin mechanisms and explore how these compare to emerging trans roles of imprinted lncRNAs.

• Besides transcriptional interference mechanisms mediated by the expression of lncRNA genes, it is now well accepted that several imprinted lncRNAs themselves control gene repression. These cis-repressive actions of lncRNAs likely impact chromatin architecture, involve lncRNA–protein interactions, and specific RNA secondary and tertiary structures could be essential as well. In principle, reported trans effects involve the lncRNAs themselves as well [63,65,66,68]. One possibility is that trans targets would be transiently positioned in close proximity to lncRNA–protein aggregates, and several recent studies have started to explore this intriguing possibility [102,125].

• Novel CRISPR technologies may help to distinguish between the effects of lncRNA transcription and those of the imprinted lncRNA transcripts per se [126,127]. Future research should also unravel which sequence motifs and secondary structures within lncRNAs are important for chromatin repression and architecture, and how these control association of specific lncRNA-interacting proteins. Finally, it is timely to determine the importance of lncRNAs and chromatin architecture in human imprinting disorders (IDs) [4,69]. Initial studies have reported altered chromatin structural interactions within the KCNQ1 and IGF2-H19 domains in the growth disorders Beckwith-Wiedemann Syndrome (BWS) and Silver-Russell Syndrome (SRS) [102,128,129].

The authors declare that there are no competing interests associated with the manuscript.

R.F., D.L. and Y.I. wrote the manuscript.

This review is dedicated to the memory of Denise Barlow, who pioneered research on functional long non-coding RNAs. We thank the reviewers for their excellent comments and suggestions and acknowledge grant funding from the Agence National de Recherche through project ANR-18-CE12-0022-02 (‘IMP-REGULOME’) and the Laboratory of Excellence EpiGenMed programme (ANR-10-LABX-12-01). Y.I. acknowledges Fellowship funding from the TOYOBO Biotechnology Foundation, Japan (2019-01) and the Japan Society for the Promotion of Science (JSPS-202160732).

3C

chromosome conformation capture

CTCF

CCCTC-binding factor

DMR

differentially methylated domain

FISH

fluorescence in situ hybridization

ICR

imprinting control region

LLPS

liquid-liquid phase separation

MBD

Methyl CpG Binding Domain

PRC

Polycomb Repressive Complex

TAD

topologically associating domain