Genomic Imprinting Mechanisms in Mammals

2.1. H19/Igf2 imprinting

The best-characterized cluster that follows a strict insulator model for imprinted expression is the cluster containing maternally expressed H19 and paternally expressed insulin-like growth factor 2 (Igf2) [11,12]. This cluster resides at 11p15.5 in humans and is found in conserved synteny on distal chromosome 7 in mice. While most studies on H19 and Igf2 have been performed in the mouse, many attributes of these genes including their expression profile and regulatory mechanisms are similar in humans. Remarkably Reik and colleagues have recently shown that the cluster and imprinting are conserved in marsupials, making this imprinting mechanism the most ancient identified to date [13]. In both mouse and human, H19 and Igf2 are widely expressed during embryonic development and postnatally downregulated in most tissues. H19 encodes a fully processed 2.3 kb non-coding RNA and was initially implicated as a tumor suppressor [14,15]. However, it has also been shown to have oncogenic properties [16–20]. Igf2 encodes a protein that plays a major role in promoting embryonic and placental growth and development [21]. Both H19 and Igf2 have recently been shown to encode microRNAs of unknown function [22,23].

As with all imprinted clusters, imprinted expression of H19 and IGF2 is regulated by an ICR [designated imprinting center 1 (IC1) in humans and ICR or differentially methylated domain (DMD) in mouse] located between the two genes [24,25]. This region is approximately 5 kb and 2 kb long in humans and mice, respectively (Figure 1). It acts by regulating interactions between the H19 and IGF2 promoters and their shared enhancers, which lie downstream of H19 (Figure 1). Deletion of the ICR/DMD results in loss of imprinting (LOI) at H19 and Igf2 [8]. Proper imprinting of H19 and Igf2 requires that the ICR/DMD is methylated on the paternal allele and unmethylated on the maternal allele. Mutations in CpGs at the ICR/DMD result in hypomethylation and subsequent biallelic expression of H19 and downregulation of Igf2 [26].

An important breakthrough in determining the imprinting mechanism at this locus came from the characterization of conserved sequences at the mouse and human ICRs [27,28]. When unmethylated, these sequences bind to the insulator protein CCCTC-binding factor (CTCF) [29–33](Figure 1A). CTCF was initially shown to mediate insulator or enhancer blocking activity at similar sequences in the β-globin locus [34]. The current model of imprinting regulation at the H19/Igf2 locus is that binding of CTCF to the unmethylated maternal ICR/DMD protects it from de novo methylation and prevents downstream enhancers from activating Igf2, leaving them available to activate transcription at H19 (Figure 1A). CTCF is unable to bind the methylated paternal ICR/DMD resulting in expression of Igf2 while H19 is silenced. Targeted mutation of CTCF sites demonstrated that these sites are necessary for imprint maintenance but not establishment [35–38].

Although it is now well established that the ICR/DMD acts as a CTCF-dependent insulator/enhancer blocker, the mechanism of insulation remains incomplete. Chromosome conformation capture (3C) experiments in mice, which assay for physical interactions between chromosomal regions, have suggested that chromosomal looping is involved in the imprinting mechanism, although the precise nature and function of the looping is debated [39–43]. While there is some consensus that the shared enhancers physically interact with the Igf2 promoters on the paternal chromosome, interactions on the maternal chromosome remain unclear. Karukuti et al report that maternal-specific silencing of Igf2 results when the ICR/DMD interacts with a matrix attachment region and a differentially methylated region at the Igf2 locus to generate a tight loop around the Igf2 gene, thereby physically impeding Igf2 expression [40]. In contrast, Yoon and colleagues demonstrate that the ICR/DMD forms a transcriptionally unproductive association with enhancers and the inactive Igf2 promoters on the maternal chromosome, which leads to silencing of Igf2 [43]. It is unclear how this “decoy” type of model would allow for the interaction between the enhancers and the H19 promoter, which are necessary for maternal H19 expression. More recently, we have shown that on the maternal allele, the enhancers make contacts throughout the H19 coding unit and promoter up to, but not beyond, the ICR/DMD [39]. When the ICR/DMD or CTCF binding sites are deleted on the maternal allele, the enhancers interact throughout the locus, suggesting that the enhancers track along the chromosome until they find a suitable promoter sequence and then the insulator blocks further tracking. There are several reasons why these 3C experiments produce differing results and models. First, the assays were designed to test different interactions (enhancers versus ICR/DMD interactions). Second, different restriction enzymes were used to digest the DNA, which may not have separated regulatory sequences adequately. Third, cells and tissues that were analyzed varied. Finally, the assays were not performed across the entire H19/Igf2 domain in all of the studies and could not be considered highly quantitative. Thus, it is difficult to discern the correctness of one model versus another. Nevertheless, the examination of multiple ICR/DMD mutant alleles demonstrates that long-range allele-specific interactions at the H19/Igf2 locus are dependent upon the ICR/DMD.

Regardless of the precise mechanism of insulator activity, it is clear that distal elements interact. Consistent with the notion of these interactions, recent studies have demonstrated that cohesins, which physically connect sister chromatids, colocalize with CTCF [44,45]. In particular, two cohesin proteins, RAD21 and SMC1, associate with CTCF sites at the H19/Igf2 DMD in an allele-specific manner that is comparable to CTCF binding [46], however it remains unknown whether cohesin binding is required for CTCF association or vice versa. Further studies to determine the timing of binding of cohesin or additional factors at the ICR/DMD as it relates to imprint establishment in the germline and maintenance throughout development will provide important clues in elucidating the molecular mechanism of H19/Igf2 imprinting.

2.2. H19/Igf2 and disease

LOI and/or aberrant expression of H19 and IGF2 is associated with somatic overgrowth and embryonal tumors and has been linked to more than 20 cancer types in human including Wilms’ tumor and colorectal, lung, breast and prostate cancers [47]. In addition, the overgrowth disorder Beckwith-Weidemann Syndrome (BWS) and the dwarfism syndrome Silver-Russell Syndrome (SRS) are strongly associated with defects in H19/IGF2 imprinting [48,49]. Although the precise physiological role of H19 and IGF2 in the development of these diseases is unclear, we are starting to appreciate how epigenetic perturbations of the locus can lead to LOI. Analysis of BWS patients revealed that hypermethylation of the IC1, leading to LOI at H19 and IGF2, can result from either maternally inherited deletions in the IC1 or post-zygotically acquired sporadic epimutations that are IC1 sequence-independent and present mosaically [50]. Likewise, analysis of SRS patients revealed that hypomethylation of IC1 can arise independently of changes in the IC1 sequence [51].

In a recent human study, 1.4 kb to 1.8 kb microdeletions at the H19/IGF2 ICR were shown to result in H19/IGF2 LOI and were associated with BWS [52,53] (Figure 1B). However, in a separate study, a 2.2 kb IC1 deletion was insufficient to cause BWS [54,55] (Figure 1B). This discrepancy was proposed to result from a difference in spacing/pattern of remaining CTCF sites. The shorter mutations deleted fewer sites resulting in abnormally long clusters of CTCF sites (and CpGs), all of which became aberrantly methylated on the maternal allele. On the other hand, the larger deletion resulted in a cluster of CTCF sites similar to one of the two clusters present on the normal allele. Therefore, the bipartite spacing/pattern of CTCF sites may play a role in determining the epigenetic status of the ICR. Alternatively, the two types of deletions may result in different perturbations of additional regulatory elements such as transcripts that have been described at the ICR/DMD in mice [56] or nucleosome positioning sites (NPSs) present in the intervening sequence that have been proposed to regulate CTCF site availability [57]. Needless to say, further scrutiny of the H19/Igf2 ICR/DMD is necessary to definitively characterize its regulatory function in genomic imprinting.

3.1 Overview

The majority of imprinted loci appear to use a second, ncRNA mechanism of regulation of imprinting in clusters (Figure 2). The first, and perhaps the best described, cluster in this class is the Igf2r cluster, which resides on mouse chromosome 17A [58] (Figure 2A). Igf2r and two neighboring genes, Slc22a2 and Slc22a3 (solute carrier 22a2 and 22a3), are expressed maternally. This region also harbors one paternally-expressed transcript, Air (antisense Igf2r RNA) that overlaps Igf2r and whose expression is critical to the silencing of the maternally-expressed genes in cis [59]. Furthermore, similar to other imprinted domains, Air expression is regulated by an ICR with parental-specific epigenetic modifications [60]. However, in contrast to other imprinted loci, the genomic organization and the imprinting pattern of the Igf2r cluster are only partially conserved in the human syntenic region, on chromosome 6q26–27.

Other loci that express long ncRNAs that are likely to be involved in the imprinting process include the Gnas locus [61], the Dlk1/Gtl2 locus [62] and the Snrpn locus [2]. The reader is referred to the excellent reviews for more details of the above loci. Here, we will focus on the Kcnq1 locus. Although the Kcnq1 locus is immediately adjacent to the H19/Igf2 locus in both mouse and human, it is independently regulated. Importantly, much of what we know about the Kcnq1 locus is conserved in human [63]. In addition, over half of the documented BWS cases are associated with defects in imprinted gene expression in the KCNQ1 domain. The majority of these BWS cases exhibit a loss of maternal-specific DNA methylation of the ICR, designated KvDMR1, and subsequent loss of expression of the maternally expressed genes. Also, mutations in the maternally expressed cell-cycle regulator cyclin dependent kinase inhibitor, CDKN1C, constitute about 10% of BWS cases [64].

3.2 ncRNA and the Kcnq1 locus

The Kcnq1 locus contains one paternally-expressed gene encoding a long (>60 kb) ncRNA, Kcnq1ot1, and at least 8 maternally-expressed protein-coding genes, including Cdkn1c, Mash2, Phlda2 [63] (Figure 2B). The locus (and the ncRNA, in particular) is governed by the maternally-methylated ICR, KvDMR1, which is located within an intron of the Kcnq1 gene [65–67]. The promoter for the Kcnq1ot1 gene resides within KvDMR1. Hypomethylation of the promoter on the paternal allele is associated with Kcnq1ot1 expression and repression of the adjacent protein-coding imprinted genes, whereas hypermethylation of KvDMR1 on the maternal allele is associated with repression of the ncRNA and activation of the adjacent imprinted genes. Deletion of KvDMR1 on the paternal allele results in a failure to express Kcnq1ot1 and in biallelic expression of the genes that are normally only expressed on the maternal allele [6,68], suggesting that transcription of the ncRNA Kcnq1ot1 is essential to silence the 8 protein coding genes in cis. Furthermore, insertion of a transcriptional stop signal downstream of the promoter on the paternal allele results in the activation of the normally silenced genes on that allele [68]. Thus, similar to what has been reported for the Igf2r/Air locus, transcription of Kcnq1ot1 or the transcript itself is required for bidirectional repression of genes in cis.

More recently, a new Kcnq1ot truncation allele reported by Higgins and colleagues confirmed that absence of the full length Kcnq1ot transcript resulted in loss of imprinting of the linked genes, with one notable exception. In the embryo proper, Cdkn1c was still imprinted in a subset of tissues [69]. This study demonstrated that Cdkn1c imprinting can be regulated by a mechanism independent of the Kcnqt1ot1 ncRNA. Given that deletion of KvDMR1 results in loss of imprinting of Cdkn1c in all tissues [6], the ncRNA-independent mechanism may rely on element(s) within KvDMR1. As such, two CTCF binding sites within KvDMR1 have been identified that are occupied in vivo only on the unmethylated paternal allele [70]. Thus, it is possible that for a gene as important to the growth of the embryo as Cdkn1c, redundant mechanisms are in place to assure its appropriate imprinting and consequently, its dosage.

Although a role for long ncRNAs appears certain at autosomal imprinted loci, as well as in X-chromosome inactivation (see below), it is still unclear how these RNAs silence overlapping and nonoverlapping genes that are located in cis. Additionally, while some of these silenced genes are several hundred kilobase pairs away, in some loci, such as Igf2r, genes more proximal escape the silencing. Multiple models have been proposed to address this complex regulation [2,71]. First, an RNAi based model is suggested in which the long antisense ncRNA forms a double-stranded RNA intermediate complementary to the silenced gene, which then triggers silencing by one of the strategies documented in other systems, including RNA degradation, translational repression or heterochromatin formation [72]. The major problem with this model, however, is that it cannot explain silencing of nonoverlapping genes with no sequence similarity, nor can it account for silencing in cis. A second possible mechanism invokes an Xist type model of silencing, in which the Xist RNA coats the inactive X-chromosome (see below). The silencing in this case would be much more limited (that is, not coating an entire chromosome), and although coating of the imprinted silenced domains by ncRNA has not yet been reported, such a mechanism cannot be discounted. Finally, models invoking silencing by transcription through the locus have been proposed. In this case, it is the transcription rather than the product of transcription that is important for silencing. Here, transcription could interfere with activators or activate repressors. This particular class of models requires the identity of cis-acting regulatory elements, which remain unknown for most imprinted domains.