Genomic Imprinting in Mammals

3.1. Imprinted Genes Control Embryonic and Neonatal Growth

What is the function of genomic imprinting in mammals? One way to answer this question is to determine the function of known imprinted genes in vivo. This can be performed by mutating the gene sequence to impair its function using the “homologous recombination” technique. The function of many of the known imprinted genes has been determined in this fashion (for original references, see http://www.mousebook.org/catalog.php?catalog=imprinting). The most significantly represented function among imprinted genes includes genes that affect growth of the embryo, placenta, and neonate. In this category are paternally expressed imprinted genes that function as growth promoters (i.e., Igf2, Peg1, Peg3, Rasgrf1, Dlk1) and show growth retardation in embryos deficient for the gene. There are also maternally expressed imprinted genes that function as growth repressors (i.e., Igf2r, Gnas, Cdkn1c, H19, Grb10), as shown by a growth enhancement in embryos deficient for the gene. Another significant category includes genes with behavioral or neurological defects (e.g., Nesp, Ube3a, Kcnq1). These results are, at one level, disappointing because they do not identify one function for all imprinted genes. Nevertheless, these results show that the majority of imprinted genes function as embryonic or neonatal growth regulators. More interestingly, the ability to regulate growth appears to be neatly divided with maternally expressed growth-regulating genes acting to repress growth of the offspring, whereas paternally expressed genes in this category act to increase growth. Moreover, numerous tested imprinted genes are active in neurological processes, some of which affect neonatal growth rate by altering maternal behavior.

3.2. The Function of Genomic Imprinting in Mammals

Can analyses of gene function help us understand why genes are imprinted in mammals? Assessment of genomic imprinting in different types of mammals has been informative. Placental mammals such as mice and humans, and marsupials such as opossum and wallaby, have genomic imprinting. Egg-laying mammals, such as platypus and echidna, appear to lack imprinted genes, although extensive studies have not yet been performed (Renfree et al. 2009). Placental mammals and marsupials are distinguished from egg-laying mammals by a reproductive strategy that allows the embryo to directly influence the amount of maternal resources used for its own growth. In contrast, embryos that develop within eggs are unable to directly influence maternal resources. Most invertebrates and vertebrates use an egg-laying reproductive strategy. Notably, they can also undergo parthenogenesis—a form of reproduction in which the female gamete develops into a new diploid individual without fertilization by a male gamete (note that parthenogenetic embryos arise from the duplication of the same maternal genome, whereas the gynogenetic embryos described in Fig. 2 arise from two different maternal genomes). The ability of organisms to undergo parthenogenesis most likely indicates a complete absence of genomic imprinting as it shows the paternal genome is dispensable. In mammals, however, a direct consequence of imprinted gene expression controlling fetal growth is that parthenogenesis is not possible. Both parents are necessary to produce viable offspring making mammals completely reliant on sexual reproduction to reproduce (Fig. 4). Parthenogenesis has thus not yet been observed in mammals despite claims to the contrary, although manipulating expression of the Igf2 and Dlk1 imprinted clusters has generated some rare mice with a diploid maternal genome (Kawahara et al. 2007).

Why should genomic imprinting have evolved only in some mammals, but not in vertebrates in general? Three features of genomic imprinting—the growth regulatory function of many imprinted genes, the restriction of imprinted genes to placental and marsupial mammals, and last, the necessity of the paternal genome for fetal development, provide evidence that can fit two equally attractive hypotheses.

The first hypothesis proposes that genomic imprinting evolved in response to a “parental conflict” situation (Moore and Haig 1991). This arises from the opposing interests of the maternal and paternal genome: Embryonic growth is dependent on one parent, but influenced by an embryo whose genome comes from two parents. Paternally expressed imprinted genes are proposed to increase embryonic growth, thereby maximizing the fitness of an individual offspring bearing a particular paternal genome. Maternally expressed imprinted genes are proposed to suppress fetal growth. This would allow a more equal distribution of maternal resources to all offspring and increase transmission of the maternal genome to multiple offspring, which may have different paternal genomes.

The second hypothesis is named “trophoblast defense” (Varmuza and Mann 1994). This proposes that the maternal genome is at risk from the consequences of being anatomically equipped for internal reproduction should spontaneous oocyte activation lead to full embryonic development. Because males lack the necessary anatomical equipment for internal reproduction, they do not share the same risks should spontaneous activation of spermatozoa occur. Imprinting is thus proposed to either silence genes on the maternal chromosome that promote placental development or to activate genes that limit this process. The genes necessary for placental invasion of the maternal uterine vasculature would consequently only be expressed from a paternal genome after fertilization has occurred.

Which, if any, of these hypotheses explains the evolution of genomic imprinting in mammals? Both hypotheses indicate a role for imprinted genes in regulating the development and function of the placenta, however, neither the parental conflict nor the trophoblast defense models can provide a full explanation for all the data (Wilkins and Haig 2003). It is interesting to note that imprinted genes have also been identified in the plant endosperm, a tissue that has been compared to the placenta by virtue that it transfers nutrient resources from the parent plant to the embryo (Grossniklaus and Paro 2014). This finding strengthens arguments that genomic imprinting evolved as a means to regulate nutrient transfer between the parent and offspring, but it does not tell us why.

Fuller or alternative explanations of the function of genomic imprinting in mammals could come from two sources. The first would be to examine the function of “imprinting” across a complete gene cluster in contrast to examining the phenotype of mice lacking a single imprinted gene product. This would require an ability to reverse an imprint and generate biparental gene expression across the whole imprinted cluster. The second approach is to learn exactly how genes are imprinted. It is possible that not all genes in a cluster are deliberate targets of the imprinting mechanism and that some may just be “innocent bystanders” of the process, and their function would not be informative about the role of genomic imprinting. The existence of innocent bystander genes affected by the imprinting mechanism may satisfactorily explain the curious abundance of imprinted genes with no obvious biological function in development.

3.3. Imprinted Genes Are Clustered and Controlled by Imprint Control Elements (ICEs)

To date, about 150 imprinted genes have been mapped to 17 mouse chromosomes including the X chromosome. More than 80% of the identified imprinted genes are clustered into 16 genomic regions that contain two or more genes (Wan and Bartolomei 2008). The discovery of clusters of imprinted genes was a strong indication that a common DNA element may regulate imprinted expression of multiple genes in cis. To date, seven of the 16 imprinted clusters have been well characterized, and these are listed in Table 1 by the name of the principle imprinted mRNA gene in the cluster or after a disease association (e.g., the Pws cluster for Prader–Willi syndrome; Beaudet and Zoghbi 2014). These seven clusters contain three to 12 (or more) imprinted genes and are spread over 80–3700 kb of DNA.

A common feature of these seven clusters is the presence of a DNA sequence carrying a gametic methylation imprint that is known as a gametic DMR (differentially DNA-methylated region). A gametic DNA methylation imprint is defined as a methylation imprint established in one gamete and maintained only on one parental chromosome in diploid cells of the embryo. In five clusters (Igf2r, Kcnq1, Gnas, Grb10, and Pws), the gametic DMR has a maternal methylation imprint acquired in oogenesis, whereas in two clusters (Igf2 and Dlk1), it has a paternal methylation imprint acquired during spermatogenesis. In these examples, the gametic DMR controls imprinted expression of the whole or part of the cluster and is therefore designated as the imprint control element, or ICE, for the cluster (Barlow 2011).

Table 1 shows that each imprinted gene cluster contains multiple mRNAs and, with the exception of Grb10, at least one lncRNA. Two trends emerge. First, the imprinted protein-coding genes in each cluster are expressed, for the most part, from the same parental chromosome, whereas the lncRNA is expressed from the opposite parental chromosome (as illustrated in Fig. 5 for a maternal gametic DMR). Second, the ICE deletion causes loss of imprinted expression only when deleted from the parental allele expressing the lncRNA. Table 1 shows that in three clusters (Igf2r, Kcnq1, and Gnas) the lncRNA promoter sits in an intron of one of the imprinted mRNAs, whereas in the remaining clusters the lncRNA promoter is separated, but lies close to the imprinted mRNA genes. This close intermingling of active and silent genes in an imprinted cluster indicates that the silencing and activating mechanisms affecting imprinted genes do not spread and may be restricted to the affected gene. In particular, the fact that the promoter of a silent lncRNA can reside in the intron of an actively transcribed gene indicates that silencing mechanisms may not even spread throughout the length of a gene, but may just be restricted to regulatory elements.

What is the role of the gametic DMR? Despite the fact the gametic DMRs can be maternally or paternally methylated, experiments that deleted these elements have produced broadly similar results albeit with a few interesting exceptions (Fig. 6). For three clusters (Igf2r, Kcnq1, Dlk1), experimental deletion of the methylated gametic DMR produced no effect. In contrast, deletion of the unmethylated gametic DMR eliminated parental-specific expression causing a loss of lncRNA expression in cis and biallelic mRNA expression (Lin et al. 1995; Zwart et al. 2001; Fitzpatrick et al. 2002). Two clusters (Gnas and Pws) appear to contain more than one gametic DMR and show a more complex behavior, yet they still share some similarities with the pattern presented in Figure 6 (Williamson et al. 2006). The Igf2 cluster, however, behaves differently: deletion of both the methylated and unmethylated gametic DMR causes changes in mRNA and lncRNA expression in cis (Thorvaldsen et al. 1998).

The results from the above gametic DMR deletion experiments do not at first glance indicate a common function for gametic DMRs. However, an understanding of their exact function depends on knowing the position of the DMR with respect to the imprinted genes in each cluster. In the three clusters with the simplest pattern (Igf2r, Kcnq1, and Dlk1), the gametic DMR either contains or controls expression of the lncRNA, thus deletion of this element will clearly lead to loss of lncRNA expression. The gametic DMR in the Igf2 cluster, however, does not directly promote H19 transcription, but changes the interaction between Igf2 and H19 and their shared enhancers, and in this way regulates their expression. Despite these differences, in general, the unmethylated gametic DMR is implicated in all six clusters as a positive regulator of lncRNA expression, and the presence of the DNA methylation imprint is associated with repression of the lncRNA. The conclusion from the data obtained from deletion of gametic DMRs clearly identifies these regions as an ICE, whose activity is regulated by DNA methylation.

3.4. Imprinted Gene Clusters Contain at Least One lncRNA

The majority of imprinted clusters contain an lncRNA, which is currently defined as a noncoding transcript >200 nucleotides (Guttman and Rinn 2012). lncRNAs, with the exception of those involved in RNA processing and translation, were previously thought to be a rarity in the mammalian genome. Now, because of the availability of the mouse and human genome sequence, transcriptome analyses have resulted in the identification of all RNA transcripts in a given cell population. This work has shown that a large part of the mammalian transcriptome is composed of lncRNAs. There are several types of mammalian noncoding RNAs (ncRNAs) that possess gene regulatory functions, including “short” ncRNAs that participate in silencing pathways (Martienssen and Moazed 2014), “longer” processed lncRNAs such as Xist, which are involved in X-chromosome activation or inactivation (Brockdorff and Turner 2014; Kuroda and Lucchesi 2014), and lncRNAs that are associated with cis or trans activation or silencing of protein-coding genes (Allis et al. 2014; essay by Rinn 2014).

What types of lncRNAs are associated with imprinted gene clusters? The analysis of the lncRNAs associated with the well-characterized imprinted clusters shown in Table 1 is still incomplete, highlighting some similarities, but also some differences. Three imprinted lncRNAs are unusually long mature RNAs: Airn is 108 kb (Lyle et al. 2000), Kcnq1ot1 is ∼100 kb (Pauler et al. 2012), and Ube3aas may be >1000 kb (Landers et al. 2004). The H19 lncRNA, in contrast, is only 2.3 kb (Brannan et al. 1990). The Gtl2 lncRNA contains multiple alternatively spliced transcripts; however, downstream intergenic transcription has also been noted, suggesting longer transcription units are likely (Tierling et al. 2005). Nespas lncRNA is larger than can be resolved on RNA blots and the full size exceeds 27 kb (Robson et al. 2012). These latter lncRNAs appear to be intron poor with a low intron–exon ratio or are unspliced as mature transcripts (Seidl et al. 2006; Pandey et al. 2008). One further feature is that three imprinted lncRNAs (H19, Ube3aas, and the Gtl2 downstream transcripts) act as host transcripts for snoRNAs (small nucleolar RNAs that direct modifications to ribosomal RNA, snRNAs, and possibly mRNAs thereby acting as posttranscription regulators) and miRNAs (micro RNAs). The snoRNAs are not directed toward the imprinted mRNA genes in the cluster and it is presently unclear if they play a role in the imprinting mechanism itself (Seitz et al. 2004). Similarly, the miRNAs in the H19 and Gtl2 lncRNAs are involved in posttranscriptional repression of mRNA genes, but do not play a direct role in regulating imprinted expression of the cluster (Davis et al. 2005; Keniry et al. 2012).

Two features of imprinted lncRNAs indicate they may play a role in the silencing of the imprinted mRNA (i.e., protein-coding) genes in the cluster. The first is that the lncRNA generally shows reciprocal parental-specific expression compared to the imprinted mRNA genes (Table 1). Second, the DMR that carries the gametic methylation imprint, which controls imprinted expression of the whole cluster, overlaps with the lncRNA promoter in multiple instances (Airn region2, KvDMR1, Snrpn-CGI, and Nespas-DMR). This finding could indicate that imprints evolved to regulate the lncRNA in each imprinted cluster. This interpretation is supported by experiments that deleted the unmethylated sequence carrying the gametic DMR causing a loss of lncRNA expression concomitant with a gain of expression of imprinted mRNA genes (Fig. 6) as tested at the Igf2r, Kcnq1, Gnas, Pws, and Dlk1 clusters (Wutz et al. 1997; Bielinska et al. 2000; Fitzpatrick et al. 2002; Lin et al. 2003; Williamson et al. 2006).

Experiments that directly test the role of the lncRNA itself have now been performed for a number of loci (Airn, Nespas, Kcnq1ot1, and H19 lncRNAs). These lncRNAs were analyzed by genetic manipulation of the endogenous locus. The first three loci were assayed by inserting a polyadenylation signal to truncate the lncRNA. Truncation of the 108-kb Airn lncRNA to 3 kb showed that the lncRNA itself is necessary to silence all three mRNA genes in the Igf2r cluster, indicating a clear regulatory role for this lncRNA (Sleutels et al. 2002). Truncation of the ∼100-kb Kcnq1ot1 lncRNA to 1.5 kb also showed that this lncRNA was directly needed to silence all 10 mRNA genes in the larger Kcnq1 cluster (Mancini-DiNardo et al. 2006). And last, truncation of the ∼27-kb Nespas lncRNA showed it was necessary to silence the overlapped Nesp gene in the Gnas imprinted cluster (Williamson et al. 2011). In contrast, precise deletion of the H19 ncRNA had no effect on imprinting in the Igf2 cluster in endoderm tissues, although some loss of imprinting was seen in mesoderm tissue (Schmidt et al. 1999). Thus, three maternally imprinted clusters (Igf2r, Kcnq1, and Gnas) share a common lncRNA-dependent silencing mechanism, whereas the single paternally imprinted cluster (Igf2) so far examined uses a different, insulator-dependent model (see Section 3.6).

3.5. The Role of DNA Methylation in Genomic Imprinting

The identification of the first three endogenous imprinted genes in 1991 enabled investigators to study how the cell’s epigenetic machinery marked an imprinted gene with its parental identity. The first and most easily testable candidate was DNA methylation, a modification in mammals that covalently adds a methyl group to the cytosine residue in CpG dinucleotides. DNA methylation is acquired through the action of de novo methyltransferases and maintained in situ each time the cell divides by the action of maintenance methyltransferases (see Li and Zhang 2014). Hence, this modification fulfills the criteria outlined in Fig. 3 for a parental identity mark or “imprint” because (1) it can be established in either the sperm or oocyte by de novo methyltransferases that act only in one gamete, (2) it can be stably propagated at each embryonic cell division by a maintenance methyltransferase, and (3) it can be erased in the germline to reset the imprint in the next generation, either by passive demethylation (DNA replication followed by the failure to undergo maintenance methylation) or through the action of a demethylating activity (possibly through conversion of 5-methylcytosine to 5-hydroxymethylcytosine by the ten-eleven translocation family of enzymes or through excision of 5-methylcytosine by the DNA repair machinery [Tan and Shi 2012; see Fig. 6 of Li and Zhang 2014]).

DNA methylation could potentially perform two different functions in genomic imprinting. It could act as the imprinting mark by being acquired de novo only by the chromosomes in one gamete. It could also serve to silence one of the parental alleles because DNA methylation is associated with gene repression (Li and Zhang 2014). To determine which function it has, it is first necessary to show that DNA methylation is present only on one parental chromosome (i.e., that it is a DMR). Second, it is necessary to identify which imprinted gene in the cluster and which regulatory sequences are marked by DNA methylation. The location of methylation marks on a promoter, or on distant positive or negative regulatory elements will have different consequences for gene expression. Finally, it is necessary to identify when the DMR forms during development. If it forms during gametogenesis and is continuously maintained in place in somatic cells (known as a gametic DMR), it may serve as the imprinting mark. If, however, it is placed on the gene after the embryo has become diploid when both parental chromosomes are in the same cell (known as a somatic DMR), it is unlikely to serve as the identity mark, but may serve to maintain parental-specific silencing.

Parental allele-specific DNA methylation has been found at most imprinted clusters that have been examined. For example, the Igf2 cluster has a gametic DMR located 2 kb upstream of the H19 lncRNA promoter that is methylated only in the paternal gamete and is maintained thereafter in all somatic tissues (Bartolomei et al. 1993). A similar gametic DMR was identified covering the promoter of the Airn lncRNA, present only on the silent maternal gene copy, and acquired in the female gamete (Stoger et al. 1993). Surprisingly, gametic DMRs were not identified at the promoters of the principal imprinted protein-coding genes in these clusters (respectively, Igf2 and Igf2r). Instead, the silenced Igf2 promoter is free of DNA methylation, whereas the silenced Igf2r promoter lies within a somatic DMR that is placed after fertilization (Sasaki et al. 1992; Stoger et al. 1993). Similar findings of gametic DMRs methylated on the chromosome carrying the silent copy of the imprinted lncRNA (as illustrated in Fig. 6) have been made for other well-studied imprinted gene clusters, including Pws, Kcnq1, Gnas, Dlk1, and Grb10 (Shemer et al. 1997; Liu et al. 2000; Takada et al. 2002; Yatsuki et al. 2002; Shiura et al. 2009).

Somatic DMRs are relatively rare but have been reported for some imprinted clusters, which suggests that this type of epigenetic modification plays a limited role in maintaining imprinted gene expression (Stoger et al. 1993; Moore et al. 1997; Yatsuki et al. 2002; John and Lefebvre, 2011). Deletions of gametic DMRs in mice result in complete loss of imprinting for multiple genes, thereby proving that this class of DMRs also serves as a major ICE for the whole cluster (Fig. 6) (Wutz et al. 1997; Thorvaldsen et al. 1998; Bielinska et al. 2000; Fitzpatrick et al. 2002; Lin et al. 2003; Williamson et al. 2006). In contrast, deletion of the somatic DMRs affects expression of the adjacent imprinted gene, but imprinted expression is maintained by other genes in the cluster (Constancia et al. 2000; Sleutels et al. 2003).

Genome-wide deficiency in DNA methylation caused by mutations in the Dnmt gene family underscores the essential role of DNA methylation in regulating imprinted gene expression. Mutations in the de novo DNA methyltransferase Dnmt3a, the DNA methyltransferase stimulatory factor Dmnt3L, or the Dnmt1 maintenance DNA methyltransferase generate DNA methylation deficient embryos that show alterations in imprinted gene expression. The type of perturbations shown for four imprinted clusters (Igf2, Igf2r, Kcnq1, and Dlk1) indicates that DNA methylation is generally acting to suppress the action of the gametic DMR. Thus, in the absence of DNA methylation, the gametic DMR cannot function appropriately (i.e., cannot silence the lncRNA). As a consequence, the lncRNA is aberrantly expressed and several imprinted protein-coding genes, including Igf2, Igf2r, Kcnq1, and Dlk1, become repressed on both parental chromosomes. This indicates that these mRNA genes require epigenetic modification of a cis-regulatory element to be expressed. Notably, the H19 lncRNA that is normally only expressed on the chromosome carrying the unmethylated gametic DMR becomes expressed on both parental chromosomes. Some exceptions to this general pattern have been reported for genes that show imprinted expression only in the placenta (Lewis et al. 2004).

Are other types of epigenetic modification used as gametic imprints? Given the sheer abundance of epigenetic mechanisms acting to modify genetic information in the mammalian genome, DNA methylation is unlikely to be the only imprinting mechanism. Histone modifications that affect chromatin activity states are also likely candidates for parental imprints because they could fulfill many of the prerequisites shown in Fig. 3. In one example, the Polycomb group protein known as EED (part of the PRC2 complex that catalyzes methylation of H3K27, i.e., histone H3 at lysine 27) has been shown to affect a few paternally repressed genes in the placenta. The effects of Eed mutation on genomic imprinting, however, are relatively minor compared to that of DNA methylation (Mager et al. 2003). In another example, the EHMT2 histone methyltransferase acting specifically on H3K9 is required to repress a few imprinted genes, but also only in the placenta (Nagano et al. 2008). Thus, evidence to date suggests that histone modifications and modifying enzymes play a minor role in genomic imprinting.

Although much is known about the identity and epigenetic modifications of gametic DMRs, much less is known about how these sequences are chosen for methylation in the gametes. To date, many more maternally than paternally methylated gametic DMRs have been identified (Bartolomei and Ferguson-Smith 2011). The maternally methylated DMRs are methylated during oocyte growth and the paternally methylated DMRs are methylated prenatally in prospermatogonia (Fig. 7) (Lucifero et al. 2002). For maternal gametic DMRs, a sequence comparison of known gametic DMRs reveals no striking sequence conservation although some contain a series of direct repeats that may adopt a secondary structure that attracts DNA methylation (Neumann et al. 1995). The tandem direct repeats in the Igf2r cluster gametic DMR have, for instance, been shown to be essential for oocyte-specific DNA methylation (Koerner et al. 2012). Those in the Kcnq1 cluster gametic DMR, however, are not essential (Mancini-DiNardo et al. 2006). Another feature of maternal DMRs is that they are markedly CpG rich compared to the remainder of the genome. One idea for how these regions are recognized comes from the structural analysis of the complexed carboxy-terminal domains of DNMT3A and DNMT3L, which was obtained by X-ray crystallography (Jia et al. 2007). A tetrameric complex consisting of these two enzymes preferentially methylates a pair of CpGs that are 8–10 base pairs apart (Cheng 2014). Such spacing is found in maternally methylated, but not in paternally methylated imprinted loci. This CpG spacing, however, is widespread in the genome, questioning the specificity of such a mechanism or indicating that additional features are required (Ferguson-Smith and Greally 2007). Additional specificity has been suggested by the demonstration that DNMT3L interacts with the amino terminus of histone H3 if the H3K4 residue is unmethylated, and promotes local DNA methylation (for more detail, see Ooi et al. 2007; Cheng 2014). Another factor contributing to the specificity of de novo DNA methylation at DMRs in the oocyte is transcription across differentially methylated regions (Chotalia et al. 2009). Importantly, only protein-coding transcripts traversing the germline ICEs are thought to be involved in DNA methylation establishment. Although it is, as yet, unclear how this transcription may be attracting the DNA methylation machinery, it has been suggested that transcription across ICEs is required to establish or maintain open chromatin domains that are permissive for the establishment of DNA methylation. To investigate and define the mechanism further, it will be necessary to describe the temporal relationship between transcription and de novo DNA methylation in greater detail. Nevertheless, CpG spacing, posttranslational histone modifications, and transcription in oocytes could provide a starting point for the acquisition of maternal-specific DNA methylation imprints.

There is far less information regarding how paternal-specific DNA methylation imprints are established in the male germline. Nevertheless, early experiments suggest there could be some similarities with the female germline. It has recently been shown that high transcriptional readthrough, predominantly from one strand, is detected at two paternal gametic DMRs in primordial germ cells, H19-DMD and IG-DMR, at the time of imprint establishment (Henckel et al. 2011). It also appears that maternal gametic DMRs, which are protected from DNA methylation, are enriched for H3 lysine 4 trimethylation (H3K4me3) in male primordial germ cells.

One of the most mysterious questions in genomic imprinting is how the DNA methylation marks at imprinted genes escape the genome-wide reprogramming that occurs after fertilization, including the DNA demethylation that occurs in the preimplantation embryo and the subsequent wave of de novo DNA methylation (Fig. 7) (see Morgan et al. 2005; Fig. 3 in Li and Zhang 2014). It is likely that a combination of cis-acting sequences and trans-acting factors are mediating the protection. One maternal factor, PGC7/STELLA, appears to have a general role in maintaining DNA methylation in the early mouse embryo through interactions with H3K9me2 (Nakamura et al. 2012). However, a factor that may be more specific for imprinted genes is ZFP57. Studies have shown that ZFP57 mutations identified in transient neonatal diabetes patients are associated with defects in DNA methylation at multiple imprinted loci (Mackay et al. 2008). Additionally, Zfp57 null mice show embryonic lethality and loss of imprinting at many (but not all) loci (Li et al. 2008). More recently, it has been shown that ZFP57 binds to cofactor KAP1, which can then recruit other epigenetic regulators (Quenneville et al. 2011). Thus, sequence- and DNA methylation-dependent binding of ZFP57 could act as an anchor to specify allelic binding of KAP1, which would subsequently recruit other major repressive epigenetic regulators such as SETDB1, HP1, DNMT1, DNMT3A, and DNMT3B to the heterochromatic, silenced allele at imprinted loci. It is possible that other yet-to-be-identified proteins also maintain DNA methylation at imprinted loci in the early embryo.

3.6. Two Types of cis-Acting Silencing Identified in Imprinted Gene Clusters

Currently, two major classes of cis-acting silencing mechanisms are hypothesized to govern imprinting at various clusters: the insulator model applicable to the Igf2 cluster and the lncRNA-mediated silencing model applicable to the Igf2r and Kcnq1 clusters. Although not yet completely defined, most of the clusters in Table 1 incorporate aspects of one of the two models. The breakthrough that led to the definition of the insulator model at the Igf2 locus was the deletion of the gametic DMR (H19-DMD) that is located 2 kb upstream of the start of H19 transcription and 80 kb downstream of Igf2 (Fig. 8) (Thorvaldsen et al. 1998). When deleted, H19 and Igf2 showed a loss of imprinting regardless of whether the deletion was inherited maternally or paternally, identifying this DMR as an ICE. It was subsequently shown that this ICE bound CTCF, a protein shown to mediate insulator activity at the beta-globin locus, and that the ICE itself functioned as an insulator (Bell and Felsenfeld 2000; Hark et al. 2000). In this context, an insulator is defined as an element that blocks enhancer and promoter interactions when placed between them. Thus, the model for imprinted gene expression at this locus is as follows: on the maternal allele, CTCF binds to the ICE and blocks the access of Igf2 and Ins2 to enhancers shared with the H19 lncRNA that are located downstream of the three genes. This thereby allows H19 exclusive access to the enhancers (Fig. 8). On the paternal allele, the ICE acquires DNA methylation in the male germline, preventing CTCF from binding to it. Thus, on the paternal chromosome, Igf2 and Ins2 interact with the enhancers and are expressed from this chromosome. The presence of DNA methylation on the paternal ICE leads to secondary methylation of the H19 promoter by an unknown mechanism and it becomes silenced on the paternal chromosome. Although the insulator model is widely accepted, it is unclear how the insulator acts at this locus. One of the most widely held views is that CTCF interacts with DNA molecules in cis to insulate genes through the formation of chromatin loops (for more detail, see Dekker and Misteli 2014). Moreover, it has been shown that cohesin interacts with CTCF to form these loops (Nativio et al. 2009). The involvement of CTCF in the insulator model has led to the identification of CTCF binding sites at other imprinted genes such as Rasgrf1, Grb10, and Kcnq1ot1, indicating that the insulator model may operate in other imprinted clusters.

The lncRNA class of imprinting model may, however, be more common. The breakthrough that led to the identification of functional ncRNAs in imprinted clusters was an experiment that truncated the 108-kb Airn lncRNA to 3 kb (Sleutels et al. 2002). This shortened lncRNA retained imprinted expression and the Airn promoter retained imprinted DNA methylation—yet silencing of all three mRNA genes in the Igf2r cluster was lost (Fig. 8). lncRNA-mediated silencing has also now been shown to operate at the Kcnq1 cluster (Mancini-DiNardo et al. 2006), although in a tissue-specific manner, suggesting that another mechanism such as one that uses insulators may also be involved at this cluster (Shin et al. 2008), and in the Gnas imprinted cluster (Williamson et al. 2011). At this time it is not precisely known how lncRNAs silence genes but many models are possible. Two possibilities arise from the sense–antisense overlap between an mRNA and the lncRNA that occurs in each cluster. The first possibility is that double-stranded RNA can form between the mRNA and lncRNA and induce RNA-interference (RNAi) (described in Martienssen and Moazed 2014). Absence of the RNAi machinery, however, does not affect imprinted expression in the Kcnq1 cluster (Redrup et al. 2009). Thus, a second possibility is that this sense–antisense overlap causes a form of transcriptional interference of a promoter or an enhancer, which affects transcription from the mRNA promoter (Pauler et al. 2012). In this case, the first event could be silencing of the overlapped promoter or enhancer followed by accumulation of repressive chromatin that can spread and induce transcriptional gene silencing throughout the cluster. Evidence for this model comes from a series of recombinant endogenous chromosomes generated at the Igf2r/Airn locus in ES (embryonic stem) cells (Latos et al. 2012). The onset of allele-specific expression at this locus in the embryo can be recapitulated by ES cell differentiation, in which Igf2r is initially biallelically expressed, but the initiation of Airn expression results in Igf2r imprinting (Latos et al. 2009). To test whether Airn transcription or the lncRNA itself was required for Igf2r silencing, Airn was shortened to different lengths, with the result that silencing only required Airn transcription overlap of the Igf2r promoter, which interferes with RNA polymerase II recruitment (Latos et al. 2012). This model suggests that Airn acts predominantly through its transcription rather than as an lncRNA.

It is, however, also possible that imprinted lncRNAs act by coating the local chromosomal region and directly recruit repressive chromatin proteins to the imprinted cluster, in a manner similar to that described for the action of the Xist lncRNA in X-chromosome inactivation (Brockdorff and Turner 2014). Evidence for a function of the lncRNA in recruitment of histone posttranslational modification machinery comes from experiments in placental tissues. RNA fluorescence in situ hybridization experiments showed that Airn and Kcnq1ot1 form RNA clouds at their site of transcription (Nagano et al. 2008; Pandey et al. 2008; Terranova et al. 2008; Redrup et al. 2009). Terranova and colleagues show that these long ncRNAs are associated with a repressive histone compartment and Polycomb group proteins (Terranova et al. 2008). This nuclear compartment is also devoid of RNA polymerase II and exists in a three-dimensionally contracted state. Other studies on the Airn lncRNA go further in suggesting that the lncRNAs actively recruit repressive histone modifications (Nagano et al. 2008), but only in the placenta. In this latter case, Airn was shown to actively recruit the EHMT2 H3K9 methyltransferase. This resulted in the paternal-specific silencing of the Slc22a3 gene but not the Igf2r gene. These experiments indicate that lncRNA-mediated silencing of imprinted genes may depend on different downstream mechanisms.

Importantly, other mechanisms of imprinted gene regulation are likely. For example, Wood and colleagues described a new imprinted locus (H13) in which alternative polyadenylation sites are used in an allele-specific manner (Wood et al. 2008). The H13 gene contains a maternally methylated internal CpG island that acquires DNA methylation in oocytes (it has not been tested for ICE activity yet). Hypermethylation of this CpG island ensures synthesis of the full length and functional H13 gene transcript from the maternal chromosome. Experiments showed that the unmethylated CpG island on the paternal allele allowed transcription from the promoter for the Mcts2 retrogene. Mcts2 expression, in turn, correlates with the premature polyadenylation of H13 and, hence, expression of truncated H13 transcripts. This locus raises the possibility that other less widely used mechanisms of genomic imprinting will be identified once the full catalog of imprinted genes is elucidated.