Mechanisms of long range silencing by imprinted macro non-coding RNAs.

Non-coding (nc) RNA silencing of imprinted genes in extra-embryonic tissues provides a good model for understanding an underexamined aspect of gene regulation by macro or long ncRNAs, that is their action as long-range cis-silencers. Numerous long intergenic ncRNAs (lincRNAs) have been recently discovered that are thought to regulate gene expression, some of which have been associated with disease. The few shown to regulate protein-coding genes are suggested to act by targeting repressive or active chromatin marks. Correlative evidence also indicates that imprinted macro ncRNAs cause long-range cis-silencing in placenta by targeting repressive histone modifications to imprinted promoters. It is timely, however, to consider alternative explanations consistent with the published data, whereby transcription alone could cause gene silencing at a distance.

Current Opinion in Genetics & Development 2012, 22:283–289

This review comes from a themed issue on Molecular and genetic bases of disease

Edited by Beverly Emanuel and Steve Warren

For a complete overview see the and the

0959-437X/$ – see front matter, © 2012 Elsevier Ltd. All rights reserved.

DOI 10.1016/j.gde.2012.02.005

Genomic imprinting is an epigenetic process that controls parent-of-origin expression of an estimated 1–2% of genes in the mammalian genome [1,2]. Although few in number, many imprinted genes play important roles in development and growth, often in a dose-dependent manner [3]. Imprinted genes mostly occur in clusters in the genome controlled by a CpG rich region known as an Imprint Control Element (ICE). This ICE shows differential DNA methylation, which is established in the germ cells of one parent and maintained on this parental chromosome throughout life. The ICE on the other parental allele remains unmethylated. The unmethylated ICE activates a macro non-coding (nc) RNA in cis, while methylation prevents activation on the other allele. Macro ncRNAs are inefficiently processed long ncRNAs whose main product is unspliced [1]. In three of four cases where the function of the imprinted macro ncRNA has been tested, it acts as a cis-silencer to prevent upregulation of flanking imprinted genes in the cluster [4-6,7]. A hallmark of imprinted genes is that they show developmental and tissue-specific regulation of imprinted expression [8]. For example, the Dlk1 gene is paternally expressed and plays a dose-dependent role in regulating growth of the embryo, but switches to biallelic expression in neural stem cells and niche astrocytes where it is required for normal postnatal neurogenesis [9,10]. Imprinted genes can be divided into two groups based on their tissue-specific imprinted expression pattern. Multi-lineage (ML) genes show imprinted expression in both the embryo and extra-embryonic tissues, while extra-embryonic lineage-specific (EXEL) genes show imprinted expression restricted to specific cell lineages in the placenta and visceral yolk sac. EXEL genes are an example of long-range cis-silencing by a macro ncRNA, as they are located in the outer region of an imprinted cluster at a greater distance from the macro ncRNA than ML genes (Figure 1) [11].

Figure 1

Long-range silencing by macro ncRNAs in two imprinted clusters. (a) Top: The imprinted Kcnq1 cluster on mouse chromosome 7 spans 780 kb. In this cluster ten genes are repressed on the paternal allele by the macro ncRNA Kcnq1ot1. Kcnq1ot1 is repressed on the maternal allele by a DNA methylation mark on the imprint control element (ICE) acquired in the oocyte. Of the ten maternally expressed genes, four show multi-lineage (ML) imprinted expression (bold font) and six show extra-embryonic lineage specific (EXEL) imprinted expression (grey font). Genes reported to be non-imprinted in placenta by a recent reevaluation are indicated (*) [45]. Kcnq1ot1 is located entirely within the protein-coding gene Kcnq1 and is necessary for the silencing of all genes in the cluster [5,25]. Bottom: The Kcnq1 gene is shown visualizing all transcripts in the region by Illumina RNA-Sequencing of mouse fetal head 14.5 days post coitum, a tissue in which Kcnq1ot1, but not Kcnq1, is expressed (pileup of sequencing reads, cut-off at 50 reads, data from [28]). Annotated minus strand unspliced ESTs (same strand as Kcnq1ot1) are also shown as short horizontal bars. Arrows indicate the reported end of Kcnq1ot1 at 83 kb (Refseq annotation), 91 kb [27] and 121 kb [22]. (b) Top: The imprinted Igf2r cluster spans 490 kb on mouse chromosome 17. The Airn macro ncRNA represses three genes on the paternal chromosome, while Airn is repressed on the maternal chromosome by oocyte acquired DNA methylation of the ICE. On the paternal chromosome Airn silences in cis the overlapped ML gene Igf2r (short-range silencing), and the non-overlapped EXEL genes Slc22a2 and Slc22a3 whose promoters lie 157 kb and 234 kb upstream respectively (long-range silencing). Bottom: Part of the Igf2r/Airn region is shown visualizing all transcripts in the region by RNA-Sequencing (as above). Airn is covered continuously with sequence reads and also by plus strand (same strand as Airn) unspliced (short horizontal bars) and spliced (long horizontal bars) ESTs over its annotated 118 kb length. The exons of Igf2r are visible as focal peaks cut-off at 50 reads. Signal from the Rangap1 pseudogene Au76 is indicated (#). The key is given below the Figure. For both loci the names of imprinted genes are above the maternal chromosome, and the names of non-imprinted genes are below.

Long ncRNAs are widespread throughout the genome and include a group known as long intergenic ncRNAs or lincRNAs, which are defined by an H3K4me3-H3K36me3 chromatin signature [12,13]. Some lincRNAs are associated with long-range cis-activation of neighbouring genes [14]; for example, HOTTIP and Mistral activate nearby, but not distant, genes in the HOXD and HOXA clusters by recruiting the H3K4me3 methyltransferase MLL1 [15,16]. Other lincRNAs are implicated in gene silencing. Approximately 20% of lincRNAs are associated with polycomb complex 2 (PRC2), which deposits the repressive H3K27me3 modification [17]. The human lincRNA HOTAIR expressed from the HOXC cluster acts in trans by targeting PCR2 to the HOXD cluster and causing gene silencing [18]; however, this function is not conserved in mouse [19]. The function of most lincRNAs remains unknown, but the example of imprinted macro ncRNAs indicates that some may regulate nearby genes by long-range cis-silencing. Another example of long-range cis-silencing by a long ncRNA is X chromosome inactivation, which is regulated by Xist ncRNA [20]. However, X-inactivation results in silencing of a whole chromosome whereas imprinted macro ncRNAs silence a more limited domain of protein-coding genes, making them the more appropriate model to understand how long-range cis-silencing by lincRNAs may work [21].

Imprinted macro ncRNAs as a model of long-range cis-repressors

Two types of cis-silencing can be mediated by macro ncRNAs: short-range silencing occurs when the ncRNA transcript fully or partially overlaps the regulated gene, while long-range silencing refers to regulation of non-overlapped genes. This review concentrates on recent findings on the mechanism of long-range cis-silencing by ncRNAs. A fundamental question is whether macro ncRNA silencing of gene expression requires the ncRNA product or if transcription alone is responsible for silencing. This question arises because features of imprinted macro ncRNAs, including the lack of sequence conservation, a low splicing rate and their unusually large size do not indicate a function for the RNA product [22,23]. The role of long ncRNAs in regulating genes in the surrounding imprinted cluster has been tested in four cases. The H19 ncRNA is fully spliced and thus not a macro ncRNA, and it is also not responsible for cis-silencing in the Igf2 cluster, but instead has been reported to regulate imprinted genes in trans, a function that may relate to its role as a micro RNA host transcript [24]. By contrast, truncation of the Airn and Kcnq1ot1 macro ncRNAs in the Igf2r and Kcnq1 clusters, respectively, by insertion of polyadenylation cassette resulted in re-expression of all the genes in these clusters from their normally silent paternal allele [5,6,25]. A similar strategy was used recently to show that Nespas macro ncRNA in the Gnas cluster silences Nesp, but the impact of the ncRNA truncation on the other promoters in the cluster has not yet been reported [7].

Genes showing ML imprinted expression may or may not be overlapped by the regulating macro ncRNA. However, all genes showing EXEL imprinted expression are not overlapped and lie further away from the ncRNA, making them a better model to understand long-range cis-silencing by ncRNAs. EXEL imprinted expression is restricted to certain cell types in extra-embryonic tissues, meaning studies of EXEL gene regulation can be compromised when using an intact organ like placenta that contains non-EXEL embryonic cell types as well as maternal endothelial and blood tissue. We have recently shown that visceral endoderm, an EXEL cell type, can be efficiently isolated from visceral yolk sac providing a homogenous cell population to study EXEL gene regulation in vivo [11]. This review examines recent findings that provide information on how imprinted macro ncRNAs may cause long-range cis-silencing of EXEL genes, focusing on silencing by Airn and Kcnq1ot1 in the Igf2r and Kcnq1 clusters (Figure 1).

The Kcnq1ot1 macro ncRNA product may be necessary for long-range silencing

The Kcnq1ot1 macro ncRNA is transcribed from the unmethylated paternal ICE located in intron 10 of Kcnq1 and silences four ML genes and six EXEL genes on the paternal allele (Figure 1a). Using quantitative polymerase chain reaction (qPCR) assays, it was recently reported that Kcnq1ot1 is 471 kb long in all examined tissues, and therefore overlaps all downstream genes, including EXEL genes [26]. However, this finding conflicts with previous reports using qPCR and RNase protection assays that mapped Kcnq1ot1 to be 91 kb or 121 kb [22,27]. In addition, our own RNA sequencing data and the distribution of reported ESTs are consistent with the earlier studies, mapping Kcnq1ot1 to be between 83 and 121 kb, meaning that it would only overlap Kcnq1 introns 10–11 (Figure 1a) [28]. The Kcnq1ot1 RNA is reported to have different behaviour in embryonic versus extra-embryonic tissues. The Kcnq1ot1 RNA fluorescence in situ hybridization (FISH) signal is larger in placenta than in embryo, correlating with the greater number of genes silenced in the placenta [22]. Kcnq1ot1 also shows a greater association with chromatin in placenta than embryo, implying an association between the ncRNA product and the chromosome in placenta [27]. In Trophoblast Stem (TS) cells, the precursor of EXEL cell types in the placenta, Kcnq1ot1 colocalises with a contracted chromosome compartment containing the entire Kcnq1 imprinted cluster and the repressive chromatin modifications H3K9me3, H2A119u1 and H3K27me3 [29]. In addition, the PRC1 and PRC2 complexes, responsible for depositing H2A119u1 and H3K27me3, also colocalise with this putative repressive compartment, and the loss of either complex affects the imprinted expression of both the ML gene Cdkn1c and EXEL genes Cd81 and Tssc4 in extra-embryonic tissues only [29]. By contrast, loss of the H3K9methyltransferase EHMT2 affects imprinted expression of EXEL genes only [30]. Although a direct connection has not been shown, these results imply that the Kcnq1ot1 ncRNA product targets repressive chromatin modifying complexes to imprinted genes in extra-embryonic tissues causing silencing. A recent study reported that RNAi knockdown of Kcnq1ot1 in embryonic (ES), trophoblast (TS) and extra-embryonic endoderm (XEN) stem cells had no effect on the maintenance of imprinted expression raising the possibility that the ncRNA product plays no role in silencing [26]. However these results need to accommodate the finding that Kcnq1ot1 is a nuclear localised ncRNA and it is uncertain if RNAi can act in the mammalian nucleus [27,31].

Airn macro ncRNA may silence by transcription alone

The concept that transcription, rather than the macro ncRNA product, may regulate overlapped imprinted genes is emerging for the Igf2r, Gnas, and Copg2 imprinted gene clusters. Transcriptional interference, where one transcriptional process interferes with another without the involvement of a mature RNA, is a well-established cis-silencing mechanism in non-mammalian organisms like bacteria, yeast, and Drosophila, and has been suggested to occur in mammals [32]. In both the Igf2r and the Gnas clusters, the macro ncRNA overlaps the promoter of a protein-coding gene in an antisense orientation. Truncation of the macro ncRNAs Airn and Nespas, so that the Igf2r and Nesp promoters are not overlapped, respectively, leads to a loss of repression of both protein-coding genes, indicating that repression may result from transcriptional interference; however, these data do not exclude a role for the ncRNA product [6,7]. In the Copg2 cluster, alternative polyadenylation of the paternally expressed Mest gene produces a longer form of this gene called MestXL, specifically in the mouse nervous system. MestXL overlaps the 3′ end of Copg2 in antisense orientation correlating with paternal repression of Copg2, and this repression is lost when MestXL is truncated [34]. This result shows that variants of protein-coding genes can also act like macro ncRNAs to regulate other genes, and was interpreted as silencing by transcriptional interference, which would indicate that transcription across the promoter is not required. However, truncation experiments do not exclude a role for the ncRNA product in silencing, as both transcription and the ncRNA product are lost downstream of the truncation site. In the case of Airn, two aspects of its RNA biology, a short half-life and inefficient splicing [23], make it less likely that the mature ncRNA product is involved in silencing Igf2r in the embryo.

Transcriptional interference of overlapped genes by imprinted macro ncRNAs is consistent with what is known from other species, but long-range cis-silencing of non-overlapped genes by transcription interference is conceptually more difficult to imagine. The Airn and Kcnq1ot1 genes are up to several hundred kilobases away from the EXEL genes they regulate (Figure 1a and b), and in both cases correlative evidence suggests that the ncRNA product is causing repression at a distance, as described for Kcnq1ot1 above. In the placenta, the Airn macro ncRNA product is located in close proximity to the silent paternal promoter of the EXEL gene Slc22a3 that also carries a repressive H3K9me3 histone mark [35]. Silencing of Slc22a3 depends on the lysine methyltransferase EHMT2 [35] whose main activity is to catalyse H3K9me2, but which can also catalyse H3K9me3 at some loci [30,36,37]. As Airn also associates with EHMT2 in placenta, it is possible that the Airn ncRNA product is responsible for the recruitment of EHMT2 to the Slc22a3 promoter and therefore for its silencing. The Tagging and Recovery of Associated Proteins (TRAP) method that is dependent on detecting the ncRNA by in situ hybridization was used to detect the close proximity of the Airn ncRNA and the Slc22a3 promoter [35]. Interestingly this technique was initially used to discover a chromosome loop connecting enhancers in the β-globin locus control region with the β-globin promoter [38]. Applying the same concept to the Airn TRAP data implies that the Slc22a3 promoter is close to the Airn transcription unit in three-dimensional space. With this in mind we propose a model consistent with the published data, where the mature ncRNA product is not responsible for silencing genes at a distance, but rather Airn transcription blocks the binding of transcriptional activators that are required to facilitate chromosomal looping and activation of Slc22a2 and Slc22a3 expression.

In this model, early development is defined by a ground state chromatin conformation that allows low-level biallelic expression of protein-coding genes on both parental alleles (Figure 2a and b, top). This ground state is well established for Igf2r in pre-implantation embryos [39,40], and for Slc22a2 and Slc22a3, which are not upregulated until post-implantation [11]. In this ground state Airn is not made, because DNA methylation of the ICE prevents Airn expression on the maternal chromosome [11] and most probably essential transcription factors are not yet expressed to activate the paternal allele [41]. In the post-implantation embryo, following the binding of transcriptional activators, activating loops form on the maternal chromosome between enhancers and the promoters of Slc22a2 and Slc22a3, causing their upregulation (Figure 2a, middle and bottom). On the paternal allele, Airn transcription prevents upregulation of Igf2r by transcription interference, and blocks the binding of transcriptional activators and the formation of activating loops within the Airn gene body, preventing upregulation of Slc22a2 and Slc22a3 (Figure 2b, middle). In a secondary step, EHMT2 is recruited to the Slc22a2 and Slc22a3 promoters and is required to maintain repression of these genes [35]. The repressed genes then attract PRC1 and PRC2 to catalyse the H2A119u1 and H3K27me3 modifications causing chromatin compaction and the formation of a repressive compartment (Figure 2b bottom). This compaction brings the Airn macro ncRNA, the Slc22a2 and Slc22a3 promoters and EHMT2 in close physical proximity that can be detected by sensitive techniques like TRAP and RNA immunoprecipitation. This model is supported by the formation of a repressive compartment on the paternal chromosome containing Airn ncRNA, a contracted Igf2r cluster, PRC1 and PRC2 and the repressive H2A119u1, H3K27me3 and H3K9me3 modifications [29].

Figure 2

Long-range silencing by interference with active loops. A model for transcription mediated long-range cis-silencing by a macro ncRNA using the example of Airn regulated imprinted expression in the Igf2r cluster. In the ground state, all protein-coding genes show a low level of expression from both the maternal and paternal allele (a and b, top). On the maternal allele binding of transcriptional activators leads to the formation of activating loops and upregulation of Slc22a2 and Slc22a3 (a, middle). The activating loops maintain high expression levels (a, bottom). On the paternal allele Airn transcription prevents Igf2r upregulation by transcriptional interference, and blocks transcription activator binding and the formation of activating loops preventing Slc22a2 and Slc22a3 upregulation (b, middle). The silent promoters attract EHMT2 to deposit H3K9me3, and polycomb complexes then deposit H3K27me3 and H2A119u1 locking the silent state and causing chromatin compaction. This compaction establishes a repressive compartment bringing Airn ncRNA in close proximity to the silenced promoters, the repressive histone modifications and their histone modifying complexes (b, bottom). The key is given below the Figure. The gene colour code is the same as for Figure 1.

Concluding remarks

Recent reports have highlighted the importance of long ncRNAs in disease. Overexpression of the lincRNA HOTAIR in breast and colorectal cancers is associated with increased PRC2 activity and an altered H3K27me3 distribution, and correlates with metastasis and a poor prognosis [42,43]. The prostate cancer associated long ncRNA, PCAT-1, is correlated with aggressive prostate cancer, and appears to have a prostate specific role in regulating cell proliferation [44]. The many long ncRNAs that have been recently discovered are likely to play a role in gene regulation and misregulation in disease, demonstrating the need for well-characterised model systems to understand their different mechanisms of action. Understanding the mechanism of imprinted macro ncRNA action may reveal new drug targets and enable improved therapy for diseases where macro ncRNAs play a role.

References and recommended reading

Papers of particular interest, published within the period of review, have been highlighted as:

• of special interest

•• of outstanding interest