Blocked transcription through KvDMR1 results in absence of methylation and gene silencing resembling Beckwith-Wiedemann syndrome

Genomic imprinting plays an essential role in mammalian development, regulating the preferential expression of certain genes in a parent of origin-dependent manner. Imprinted genes are expressed exclusively or preferentially from only one of the two parental alleles, tend to occur in clusters (imprinted domains), and are frequently involved in growth regulation and development, with paternally expressed genes often coding for growth promoters and maternally expressed genes often coding for growth suppressors (Barlow and Bartolomei, 2014; Bartolomei and Ferguson-Smith, 2011; Li and Sasaki, 2011; Tycko and Morison, 2002). Monoallelic expression of genes present in an imprinted domain is controlled by cis-acting short DNA sequences known as imprinting control regions (ICRs) or imprinting centers (ICs) that can function over long genomic distances (Spahn and Barlow, 2003; Verona et al., 2003). As first demonstrated for the H19/Igf2 (Thorvaldsen et al., 1998), the Igf2r/Air (Sleutels et al., 2002) and KvDMR1 imprinted domains (Fitzpatrick et al., 2002), targeted deletion of these sequences in the mouse results in disruption of imprinting. To date, two mechanisms have been described by which an ICR functions to silence genes in somatic cells, through chromatin insulation and via long noncoding RNAs (lncRNAs) (Barlow and Bartolomei, 2014; Pauler and Barlow, 2006; Wan and Bartolomei, 2008). The crucial feature that determines whether an ICR is active is DNA methylation status. ICRs contain CpG island (CGI)-like elements that are methylated (or not) in developing oocytes and sperm; this germline-specific methylation is manifested as differentially methylated regions (DMRs) in somatic cells, which are maintained as a memory of parental origin and mediate monoallelic silencing of imprinted genes. The large majority of characterized imprinted domains contain ICR associated DMRs that become methylated specifically in the female germline. To date, only four imprinted domains (H19/Igf2, Rasgrf1, Dlk1-Gtl2 and Zdbf2) have been shown to contain DMRs methylated in the male germline (Barlow and Bartolomei, 2014; Bartolomei and Ferguson-Smith, 2011; MacDonald and Mann, 2014).

Disruption of imprinted expression in humans leads to several developmental disorders and may contribute to cancer (Lee and Bartolomei, 2013; Soellner et al., 2016; Tomizawa and Sasaki, 2012; Wilkins and Úbeda, 2011). One of the earliest examples of an imprinting disorder is Beckwith-Wiedemann syndrome (BWS), a generalized overgrowth and cancer predisposition condition (Weksberg et al., 2010). Multiple genetic and epigenetic mechanisms give rise to BWS but all have the net effect of disrupting expression of imprinted genes in human chromosome 11p15.5, with IGF2 and CDKN1C thought to play major roles in the overgrowth and cancer susceptibility (Choufani et al., 2013; Demars and Gicquel, 2012; Soejima and Higashimoto, 2013). The 11p15.5 imprinted domain in human, and its orthologous counterpart in mouse distal chromosome 7, comprises two independently regulated subdomains, with the H19/Igf2 ICR (also known as IC1) regulating imprinted expression of the H19 and IGF2 genes, and the KvDMR1 ICR (IC2) controlling a cluster of maternally expressed genes, including CDKN1C (cyclin dependent kinase inhibitor 1C), a negative regulator of cell proliferation (Fitzpatrick et al., 2002). The KvDMR1 ICR includes the promoter for the lncRNA KCNQ1OT1 and is normally methylated on the maternally inherited chromosome, thereby repressing expression of the lncRNA from this allele; on the paternally-inherited allele, KvDMR1 is unmethylated and promotes expression of KCNQ1OT1 (Lee et al., 1999; Smilinich et al., 1999) (Fig. 1). In mouse models, expression of Kcnq1ot1 from the paternal allele has been shown to be instrumental in subdomain-wide silencing of maternal-specific genes on the paternal chromosome (Mancini-Dinardo et al., 2006; Shin et al., 2008). More than half of individuals with BWS have an epimutation (absence of DNA methylation at the maternal allele) at KvDMR1 (Bliek et al., 2001; DeBaun et al., 2002; Engel et al., 2000; Weksberg et al., 2001), which results in biallelic expression of KCNQ1OT1 (Lee et al., 1999; Smilinich et al., 1999) and associated silencing of CDKN1C and other maternally expressed genes in this imprinted domain (Diaz-Meyer et al., 2003). The cause for the absence or loss of methylation at KvDMR1 is not clear but presumably results from improper establishment and/or maintenance during oogenesis or failure of maintenance after fertilization. KvDMR1 may be particularly susceptible to epigenetic disruption, as evidenced by frequent disruption of methylation following assisted-reproductive technologies (Amor and Halliday, 2008; Weksberg et al., 2007; White et al., 2015).

Establishment of methylation imprints in the female mouse germline occurs postnatally in growing oocytes arrested in meiosis I (Hiura et al., 2006; Lucifero et al., 2004). Oocyte-specific methylation is carried out by cooperative activities of DNA methyltransferase 3A (DNMT3A) and its nonenzymatic co-factor DNA methyltransferase 3-like (DNMT3L) (Bourc'his et al., 2001; Hata et al., 2002; Kaneda et al., 2004). Recently, seemingly disparate pieces of evidence have suggested a role for transcription in the establishment of the methylation mark at germline DMRs (reviewed by Veselovska et al., 2015). Most known maternally methylated DMRs are located within transcription units and, in day 10 oocytes (the time when de novo methylation is first detected), methylated intragenic CGIs are more likely to map to active transcription units compared with unmethylated intragenic CGIs. Interestingly, trimethylation of H3K36, a histone modification associated with transcriptional elongation, can recruit DNMT3A (Dhayalan et al., 2010; Zhang et al., 2010). Furthermore, promoter CGIs found to be methylated at this stage frequently overlapped with an adjacent transcription unit (Smallwood et al., 2011). Direct demonstration of a link between transcription and DNA methylation at ICRs was first provided by Kelsey and colleagues for the Gnas locus, where it was shown that disruption of the Nesp transcript, which initiates further upstream in this imprinting domain, precludes normal establishment of methylation of the female germline DMR (Chotalia et al., 2009). More recently, using deep sequencing and de novo assembly of the mouse oocyte transcriptome, the same group demonstrated that transcription was a major determinant of oocyte methylome; moreover, deletion of an oocyte-specific promoter of the Zac1 (Plagl1) gene results in the lack of DNA methylation at the Zac1 ICR (also called Zac1 igDMR) (Veselovska et al., 2015). Importantly, these authors provided evidence that oocyte-specific transcription exists through other maternally methylated ICRs, including KvDMR1 (Chotalia et al., 2009; Veselovska et al., 2015). In the case of KvDMR1, this transcript is co-linear with the Kcnq1 primary transcript and arises through activation of several oocyte-specific transcription start sites (TSS) close to the somatic TSS of Kcnq1 (Chotalia et al., 2009). Further support for transcription-mediated methylation during oogenesis comes from studies of mouse transgenes demonstrating that Snrpn DMR methylation in oocytes depended on activity of an upstream promoter (Smith et al., 2011). Whether transcription through an ICR is a common mechanism that also functions at the KvDMR1 ICR remains to be shown. If it is a common mechanism, then the etiology of some individuals with BWS may involve defective transcription through the KvDMR1 locus, leading to the absence of methylation.

We tested this hypothesis by showing that disruption of Kcnq1 transcription before it reaches KvDMR1 prevents normal establishment and/or maintenance of methylation at KvDMR1. As a consequence, Kcnq1ot1 lncRNA was expressed from the maternal allele, resulting in silencing of maternally expressed genes across the domain. Thus, we provide further evidence that transcription across DMRs is essential for establishment of methylation at maternally methylated gametic DMRs. Moreover, this mouse model partially recapitulates the most common form of BWS, as it lacks maternal methylation at KvDMR1 and has biallelic silencing of Cdkn1c, a crucial growth regulatory gene. This is the first report suggesting impairment of transcription through KvDMR1 as a possible cause of BWS.

Previously, we described a mouse line with a polyadenylation [poly(A)] truncation cassette inserted downstream of KvDMR1 to terminate elongation of Kcnq1ot1 (pA or YJ69, Fig. 1) (Shin et al., 2008). As a control, we inserted the same truncation cassette in the ‘reverse’ orientation (Ap or YJ11). Following paternal transmission of the YJ69 allele, Kcnq1ot1 transcription was blocked and imprinted expression of genes under the control of KvDMR1 was lost, demonstrating the requirement for Kcnq1ot1 transcription in the imprinted silencing of this domain. By contrast, paternal inheritance of the YJ11 had no observable effect on imprinted expression (Shin et al., 2008) (also see Fig. S1). Reports demonstrating the requirement of transcription through the ICRs at the Gnas, Snrpn and Zac1 loci for proper methylation during oogenesis (Chotalia et al., 2009; Smith et al., 2011; Veselovska et al., 2015) prompted us to use the YJ11 mutant line to determine whether transcription of Kcnq1 [which overlaps and transcribes antisense to Kcnq1ot1 (Fig. 1)] through KvDMR1, is required for the establishment of maternal methylation at this ICR. The YJ69 allele should not truncate RNAs transcribed in this direction and therefore serves as a control (Fig. S1). The effect of Kcnq1 truncation was first tested by crossing female mice heterozygous for the YJ11 allele with wild-type males; no pups carrying the Kcnq1 truncation (YJ11) were observed in over 10 litters from this cross, suggesting lethality in utero. Embryos from additional crosses of the same type were analyzed at embryonic day (E) 9.5 and ∼50% of the conceptuses carried the YJ11 allele. Based on previous literature, blocking transcription through the maternal KvDMR1 ICR during oogenesis is predicted to prevent its methylation, therefore allowing maternal expression of Kcnq1ot1, and consequent repression of maternally expressed genes and impaired development, which likely explains the observed lethality. To test this possibility, female mice heterozygous for the YJ11 allele were crossed with males carrying a deletion of KvDMR1 that results in expression of the maternal-specific genes from the paternal chromosome (Fitzpatrick et al., 2002). Significantly, embryonic lethality observed following maternal transmission of the YJ11 allele was rescued (Table S1) supporting this contention.

To directly show the effect of the YJ11 allele on Kcnq1 transcription, real-time RT-PCR was performed on RNA isolated from kidney tissue from 3- to 4-week-old mice and from oocytes collected from 15 days post-partum (dpp) females. As YJ11/+ mice cannot be obtained due to embryonic lethality, these studies were carried out using YJ11/KvDMR1Δmice (i.e. offspring of YJ11^−/+ females and KvDMR1Δ^+/− males). To circumvent possible confounding effects due to insertion of foreign DNA (i.e. the pA termination cassette) into the locus, kidney and oocyte RNA isolated from similarly aged YJ69/KvDMR1Δ mice was used as a control (Fig. S1). cDNA was synthesized using Kcnq1 transcript-specific primers located upstream (Kcnq1URT8) or downstream (Int-ls-RT3) of the poly(A) cassette (Ap or pA) with respect to the direction of Kcnq1 transcription (Fig. S2A), and used as template in PCR assays using primers immediately upstream of the two Kcnq1-specific primers; because the primers corresponding to the downstream assay are located within the deleted region in the KvDMR1Δ allele, this assay only detected transcripts generated from the maternally derived YJ11 and YJ69 alleles (Fig. S2A). As neither the YJ11 nor YJ69 truncation cassettes are predicted to affect transcription upstream of the insertion site, upstream Ct values were used as a normalizing factor for the calculation of ΔCt values. Duplicate experiments analyzing kidney RNA from three YJ69/KvDMR1Δ and three YJ11/KvDMR1Δ animals showed a significant difference in the mean ΔCt (−0.16 for YJ69 and −1.08 for YJ11, P=0.003) corresponding to a 47% decrease in Kcnq1 transcript levels downstream of the YJ11 cassette compared with the YJ69 termination cassette (Fig. S2B). A similar reduction of Kcnq1 primary transcript level was observed in oocytes collected from 15-day-old YJ11/KvDMR1Δ females (Fig. S2C). These results suggested that maternal inheritance of the YJ11 allele only partially impedes Kcnq1 elongation towards KvDMR1. However, upon further investigation, this surprisingly ‘inefficient’ termination of Kcnq1 transcription was found to likely be an artifact due to primer-independent reverse transcription (see Discussion).

To circumvent the apparent problem with RT-PCR at this locus, an RNA in situ hybridization approach was adopted using RNAscope probes, which are specific to regions of the primary Kcnq1 transcript upstream and downstream of the pA and Ap insertion sites (Fig. 2A,B); thus, the downstream probe is designed to hybridize only with RNA transcripts that have elongated through the pA or Ap termination cassettes. The two probes were co-hybridized to sections of ovaries removed from four 15 dpp YJ69/KvDMR1Δ and four YJ11/KvDMR1Δ females. At this developmental stage, chromosomes have been replicated, homologous chromosomes paired and oocytes arrested in the diplotene stage of meiotic prophase 1 (reviewed by Hu et al., 2012; Sánchez and Smitz, 2012). When sectioned oocytes fromYJ69/KvDMR1Δ ovaries were hybridized, signals were detected for both upstream (turquoise) and downstream (magenta) RNAscope probes, indicating elongation of Kcnq1 transcript through the poly(A) cassette (Fig. 2C). By contrast, only upstream (turquoise) signals were detected following hybridization of both probes to sectioned oocytes from YJ11/KvDMR1Δ ovaries. These results were consistent, and statistically significant, across all scorable oocyte sections fromYJ69/KvDMR1Δ (n=9) and YJ11/KvDMR1Δ (n=6) ovaries, with no downstream (magenta) signals observed in YJ11/KvDMR1Δ oocytes (Fig. 2D,E). Therefore, in contrast to RT-PCR analysis of oocytes, which suggested only partial blockage of Kcnq1 transcription by the YJ11 allele, these results suggest that blockage is virtually complete. Consistent with the notion that the YJ11 truncation cassette effectively blocks elongation of the Kcnq1 primary transcript was the finding that it resulted in 40% decrease in the steady state level of spliced Kcnq1 mRNA in kidney from 3- to 4-week-old mice (Fig. S3).

Methylation analysis at KvDMR1 was first carried out on genomic DNA extracted from embryonic tissues of two wild-type (+/SD7) and two mutant (YJ11/SD7) E9.5 conceptuses. KvDMR1 contains two annotated CpG islands (CGI) separated by 250 bp (Fig. S4). First, region C, which is located near the end of CGI 2, was analyzed by COBRA. As expected for an imprinted differentially methylated DMR, approximately half of the PCR products generated using bisulfite-treated DNA from wild-type littermates DNA was digested by BstUI, indicating that roughly 50% of CpG dinucleotides in this restriction site were methylated; by contrast, only a small proportion of PCR product generated using DNA in mutant embryos was cleaved, suggesting little or no methylation at this site (Fig. S4). As a control, COBRA assays were performed on bisulfite-treated DNA from the YJ69 line. Equivalent degrees of cleavage in wild-type and mutant DNA, at three different restriction sites (two BstUI and one TaqI), suggested no significant changes in methylation at KvDMR1 following maternal transmission of the YJ69 allele (Fig. S4). More comprehensive methylation analysis of KvDMR1was carried out using bisulfite sequencing; in total, 10 CpG sites in CGI 1 and 16 CpG sites in CGI 2 were analyzed. Following maternal inheritance of the Kcnq1 truncation (YJ11), methylation was essentially absent at KvDMR1 in mutant embryos, whereas wild-type embryos had ∼50% methylation, as expected (Fig. 3, middle panel).

In addition to failure in methylation establishment during oogenesis, the absence of maternal methylation at KvDMR1 in YJ11 mutant embryos could have resulted from impaired maintenance of DNA methylation at this locus following the wave of demethylation after fertilization. To determine whether the YJ11 allele was methylated prior to fertilization, oocytes, collected from two 15 dpp compound heterozygotes (YJ11/KvDMRΔ) and two ‘wild type’ (+/KvDMRΔ) littermates were analyzed by bisulfite sequencing. As the PCR primers used were located within the region deleted in the KvDMRΔ (Fitzpatrick et al., 2002), PCR products originating from oocyte DNA isolated from these mice originate only from the maternal allele. Oocyte DNA from wild-type littermates showed predominantly methylated alleles (37/52 sequenced molecules) as expected (Fig. 3, lower panel). The unmethylated alleles in the (+/KvDMRΔ) samples likely reflect heterogeneity in the developmental stage of the oocyte population in 15 ddp ovaries (Hiura et al., 2006). By contrast, and consistent with results obtained from E9.5 day embryos, methylation at KvDMR1 was almost completely absent in oocytes following Kcnq1 truncation (YJ11). These results suggest that, in the absence of transcription through KvDMR1, methylation was either not established or was lost later during oocyte maturation; furthermore, the aberrantly unmethylated YJ11 allele remained unmethylated during global post-fertilization methylation events. These data strongly suggest that, similar to the Gnas, Snrpn and Zac1 loci (Chotalia et al., 2009; Smith et al., 2011; Veselovska et al., 2015), transcription through the KvDMR1 is a prerequisite to establish and/or maintain the methylation mark at the maternal allele, and confirm this transcription-associated mechanism at a third endogenous locus.

In the KvDMR1 domain, imprinted protein-coding genes are expressed exclusively or predominately from maternal alleles, whereas the Kcnq1ot1 lncRNA is repressed on the maternal methylated KvDMR1 allele and expressed from the unmethylated paternal allele. To determine whether truncation of the Kcnq1 primary transcript and consequent absence of methylation at KvDMR1 affected gene expression across the imprinted domain, analysis was carried out on E9.5 conceptuses from crosses between females heterozygous for the YJ11 or YJ69 (control) allele and SD7 males (congenic for distal chromosome 7 of M. spretus in a C57BL/6J background); SD7 males were used to provide polymorphisms for allele-specific expression analysis. The imprinted status of the Kcnq1ot1 lncRNA was first tested using primers whose amplicon spans an MwoI RFLP between 129SvJae and SD7 mice (Lewis et al., 2006). Both wild-type placenta and embryo proper expressed only the MwoI-cleaved (SD7) paternal allele, while biallelic expression of Kcnq1ot1 occurred in placental and embryonic tissues with the Kcnq1 truncation (Fig. 4A). This ‘loss of imprinting’ was reflected as a two- to threefold increase in Kcnq1ot1 expression in placental and embryonic tissues with a maternally inherited YJ11 allele, whereas the steady-state level of Kcnq1ot1 lncRNA was not significantly changed in the YJ69 control line (Fig. 4B). Thus, following the maternal Kcnq1 truncation, Kcnq1ot1 becomes biallelically expressed.

In wild-type mice, Kcnq1ot1 is expressed only from the paternal allele and silences (in cis) at least eight maternal-specific genes on the paternal chromosome (Fitzpatrick et al., 2002; Mancini-Dinardo et al., 2006; Shin et al., 2008). Thus, maternal expression of Kcnq1ot1 is also anticipated to silence the same set of protein coding genes on the maternal chromosome. Allele-specific primers were used in qRT-PCR assays to analyze expression of the ubiquitously imprinted genes Cdkn1c and Phlda2, as well as the Ascl2 gene, which is expressed primarily in the placenta. In wild-type controls, expression in wild-type placenta and embryo proper was almost exclusively from the maternal allele; however, following maternal transmission of the YJ11 allele, the maternal allele of each of these genes was partially or completely silenced (Fig. S5, YJ11 X SD7 panels). In E9.5 embryos, maternal transmission of the YJ69 allele (control) had little or no effect on gene expression across the imprinted domain, with the exception of Cdkn1c (Fig. S5, YJ69×SD7 panels) where, in contrast to E9.5 placenta (where expression was unchanged), Cdkn1c expression in E9.5 embryos was almost twice as high in the chromosome carrying the YJ69 termination cassette compared with the wild-type chromosome. This result almost reached statistical significance (P=0.07) (Fig. S5, middle panel, far right) and suggests that the insertion may have disrupted a regulatory element for this gene. The expression of the Kcnq1 and Slc22a18 genes was analyzed using primers that measure overall gene expression levels. These two genes also showed a significant reduction in expression in placental and embryonic tissues, suggesting silencing of the normally expressed maternal allele (Fig. S6). Although, maternal inheritance of the YJ69 allele showed no statistically significant changes in the gene expression of Slc22a18, the expression of Kcnq1 in placenta was unexpectedly increased by ∼40% (P<0.001, Fig. S6), again suggesting potential disruption of a regulatory element. In summary, gene expression analysis suggests that the absence of methylation at KvDMR1 was associated with blocked elongation of a ‘Kcnq1’ transcript through KvDMR1 and allows Kcnq1ot1 to be expressed biallelically. Similar to the situation of the paternal allele, expression of Kcnq1ot1 on the maternal allele directly or indirectly represses expression of maternal-specific genes present in this imprinted domain. Although some inconsistencies are observed [e.g. increased expression of Cdkn1c and Kcnq1 in control YJ69 tissues compared with wild type (Figs S5 and S6), see Discussion], the majority of the expression data described above supports the notion that the net effect of the YJ11 Kcnq1 truncation model is that both parental alleles acquire similar epigenetic states. These findings further support the notion that transcription through an ICR is a common requirement for the establishment and/or maintenance of DNA methylation at maternal gDMRs and regulation of imprinted gene expression. Furthermore, this Kcnq1 truncation mouse model recapitulates the ‘loss of methylation’ molecular phenotype frequently observed in individuals with BWS (Bliek et al., 2004; DeBaun et al., 2002; Engel et al., 2000; Smilinich et al., 1999; Weksberg et al., 2001).

Previous studies have demonstrated that transcription is required for establishment of methylation at maternal gDMRs (Veselovska et al., 2015). These same studies also demonstrated the existence of RNA transcripts in the vicinity of other maternally methylated DMRs, including KvDMR1. Interestingly, a large domain of virtually contiguous hypermethylation was observed across the Kcnq1 gene, beginning just after exon 1 and extending across the entire locus (>300 kb) (Veselovska et al., 2015). Given that, in the majority of individuals with BWS, KvDMR1 is devoid of methylation and genes under its control are repressed, one mechanism by which this anomaly could result is through absence of transcription via this regulatory element. RNA in situ hybridization showed that maternal transmission of the YJ11 allele blocked Kcnq1 primary transcript elongation from reaching KvDMR1. Moreover, the lack of these transcripts correlated with the absence or very low level of methylation at KvDMR1 in E9.5 conceptuses and 15 dpp oocytes. This lack of methylation is likely due to defective establishment, rather than to its developmental postponement because a delay in establishment would not result in the absence of methylation in E9.5 conceptuses (Fig. 3). Thus, similar to other ICRs (Chotalia et al., 2009; Smith et al., 2011; Veselovska et al., 2015), transcription through KvDMR1 appears to be essential for its proper methylation. These results are consistent with the finding that a 260 kb BAC transgene, encompassing the KvDMR1 ICR and most of the Kcnq1 gene, but excluding the promoter region, did not undergo proper imprinting following maternal transmission (John et al., 2001).

RT-PCR analysis of Kcnq1 primary transcription downstream of the YJ11 truncation cassette suggested only partial blockage of transcription, an observation inconsistent with our RNA in situ hybridization analysis, which provided evidence for efficient termination, at least below the level of detection of the assay. However, further investigation has shown that the apparent less than complete blockage of the Kcnq1 primary transcript, as determined by RT-PCR, is due to primer-independent reverse transcription; i.e. some PCR product is generated from cDNA reactions that do not contain a primer for first-strand synthesis. It should be noted that no amplification product was observed in –RT controls, indicating that this product was not amplified from contaminating genomic DNA in the RNA. As this region of the Kcnq1 primary transcript is transcribed from the same genomic region as the Kcnq1ot1 lncRNA transcript, it may be Kcnq1ot1 RNA degradation products that are aberrantly priming cDNA synthesis.

Until recently, the prevailing models of germline-specific methylation of gDMRs presumed targeting of de novo DNMTases to specific loci, perhaps by a combination of localized DNA composition and/or sequence and chromatin structure (reviewed by Bartolomei and Ferguson-Smith, 2011). However, genome-wide methylation studies of the mouse germline have prompted re-evaluation of this notion. First, reduced representation bisulfite sequencing (RRBS) showed that CGIs methylated in oocytes were predominately found within active transcription units rather than at their 5′ends; importantly, these included most maternal gDMRs (Smallwood et al., 2011). Moreover, whole-genome bisulfite sequencing coupled with RNA-seq demonstrated that most methylation in oocytes is within gene bodies and that this methylation was highly correlated with expression levels of those genes (Kobayashi et al., 2012). Thus, the developing paradigm proposes that, at least in the female germline, the DNMT3A/DNMT3L methylation machinery is not specifically targeted to gDMR sequences. Instead gDMR methylation takes place as part of larger more-general gene body methylation, and specific gDMRs become methylated by virtue of their intragenic location (Kelsey and Feil, 2013; Smallwood et al., 2011). Although generally not resulting from studies using oocytes, several lines of evidence suggest that transcription through gene bodies and associated intragenic CGI and gDMRs brings about a specific chromatin structure that promotes the binding and action of the DMNT3A/ DMNT3L complex. The model suggests that transcription-coupled recruitment of histone modifying enzymes, such as the H3K4me1/2 demethylase KDM1B and the H3K36 methylase SETD2, yields a chromatin template with unmethylated H3K4 and trimethylated H3K36, histone modifications that are ‘preferred’ by the DNMT3A/DNMT3L complex (reviewed by Kelsey and Feil, 2013; Stewart et al., 2015). The number of transcription units and intragenic CGIs that are methylated throughout their gene body in oocytes is far greater than the number of known imprinted gDMRs (Kobayashi et al., 2012; Smallwood et al., 2011). Consistently, genome-wide ChIP-seq experiments have demonstrated a relative depletion of ‘protective’ H3K4me2 and H3K4me3, as well as enrichment of ‘permissive’ H3K36me3 marks at the majority of the CGI destined to become methylated during oocyte growth and maturation (Stewart et al., 2015). In the present case, it is noteworthy that KvDMR1, like most maternally methylated ICRs, shows a dramatic decrease in DNA methylation in oocytes isolated from KDM1B knockout mice (Stewart et al., 2015). Thus, the emerging model for how allele-specific methylation occurs at only several tens of loci, including gDMRs is one of ‘selective protection from demethylation’ during early embryonic development (Kelsey and Feil, 2013; Smallwood et al., 2011; Stewart et al., 2015; Veselovska et al., 2015), perhaps by proteins such as ZPF57 (Li et al., 2008; Mackay et al., 2008; Quenneville et al., 2011) and NLRP2 (Meyer et al., 2009).

We propose that the reduction or absence of transcriptional elongation prior to reaching KvDMR1 prevents formation of a chromatin state permissive to DNA methylation by de novo DNA methyltransferases (Fig. 5). Further studies are required to confirm the exact chromatin conformation at KvDMR1 following the maternal inheritance of the YJ11 truncation allele as well as the temporal association and interactions of the transcription process and specific factors that affect de novo methylation (Ciccone et al., 2009; Hiura et al., 2006).

Our observations suggest that maternal transmission of the YJ11 allele causes embryonic lethality (later than E9.5) due to the silencing of one or more maternal-specific genes in the domain. The most likely culprit is Ascl2, expression of which is crucial for placental development, and whose deletion results in embryonic death at E10.5 (Guillemot et al., 1994, 1995; Tunster et al., 2016). Surprisingly, maternal transmission of the YJ69 allele (identical sequence to YJ11 but inserted in opposite orientation) was associated with significantly increased expression of Cdkn1c in E9.5 embryos (Fig. S5) and of Kcnq1 in E9.5 placenta (Fig. S6). This could indicate that insertion of the truncation cassette disrupted a currently undefined negative regulatory element for these two genes; in this regard, it is notable that recent evidence supports the existence of multiple enhancer elements in the KvDMR1/Kcnq1ot1 region (Korostowski et al., 2011; Schultz et al., 2015). Disruption of a gene repressor/silencer at this insertion site may also explain the apparently less efficient silencing effect that insertion of the YJ11 cassette has on Cdkn1c and Kcnq1 (Figs S5 and S6). Despite the apparent complexity of this genomic region, it is unlikely that the absence of methylation at KvDMR1 following the insertion of the YJ11 allele is due simply to insertional mutagenesis, as the same effect is not observed with the YJ69 allele (Fig. S4).

Roughly half of individuals with BWS have an epimutation (absence of DNA methylation at the maternal allele) at KvDMR1 (Bliek et al., 2001; DeBaun et al., 2002; Engel et al., 2000; Weksberg et al., 2001), which results in biallelic expression of KCNQ1OT1 (Lee et al., 1999; Smilinich et al., 1999) and associated silencing of CDKN1C and other maternally expressed genes in this imprinted domain (Diaz-Meyer et al., 2003). Mouse models with mutations in Cdkn1c have been described and in some cases exhibit fetal overgrowth that is reminiscent of BWS but are lethal at birth (Tunster et al., 2011; Yan et al., 1997; Zhang et al., 1997). However, these models do not precisely mirror the majority of cases of BWS as most patients do not have mutations in CDKN1C. The present mouse model results in the absence of methylation at KvDMR1, biallelic expression of the Kcnq1ot1 lncRNA and biallelic silencing of genes under the control of this ICR, closely recapitulating the molecular phenotype observed in the majority of individuals with BWS. The lethality observed in the YJ11 mouse is not seen in BWS because, unless tumors develop, individuals with this condition usually survive into adulthood. The embryonic viability of human embryos may be due to the apparent lack of ASCL2 imprinting in human placenta (Miyamoto et al., 2002) as well as somewhat ‘leaky’ imprinting of CDKN1C (Chung et al., 1996; Matsuoka et al., 1996). Although it is widely thought that maternal expression of KCNQ1OT1 in BWS is responsible for silencing of maternal-specific genes in the domain, it remained a formal possibility that these aberrant expression patterns were not directly related. The present model provides a direct mechanistic link between the absence of methylation at KvDMR1, maternal expression of Kcnq1ot1 and maternal silencing of the ICR regulated genes. It is not known whether any individuals with BWS and an absence of KvDMR1 methylation have a defect in oocyte-related KCNQ1 transcription (Fig. S7). Nevertheless, microdeletions specifically affecting an oocyte-specific promoter may comprise a subset of familial BWS cases that do not have mutations in CDKN1C. This study also predicts that individuals with BWS who have a chromosome translocation that breaks between an oocyte-specific promoter and KvDMR1 would also exhibit absence of methylation at KvDMR1. This may be the case for a patient described in a recent report who had both BWS and long QT syndrome (Kaltenbach et al., 2013). Epigenetic silencing of an oocyte-specific promoter is another possible mechanism by which DNA methylation is not established in this domain, thus leading to BWS. Further analysis of individuals with BWS is necessary to determine to what extent various mechanisms that disrupt transcription through KvDMR1 during oogenesis are responsible for its aberrant methylation.

Mouse studies were approved by the Roswell Park Cancer Institute (RPCI) Institutional Animal Care and Use Committee (IACUC). The generation of the YJ11 and YJ69 mutant lines has been described previously (Shin et al., 2008). Female offspring heterozygous for the YJ11 or YJ69 alleles were crossed with either SD7 (congenic for distal chromosome 7 of M. spretus in a C57BL/6J background, kindly provided by Wolf Reik, Babraham Institute, Cambridge, UK) or KvDMR1Δ males, depending on the analysis.

Oocytes were collected from 15 dpp YJ11/KvDMR1Δ, YJ69/KvDMR1Δ or +/KvDMR1Δ control females to allow exclusive PCR amplification of the maternal KvDMR1 allele (YJ11) in the absence of the paternal allele in oocytes (see Results). Oocytes were collected (Chotalia et al., 2009) (protocol kindly provided by Jiahao Huang and Gavin Kelsey, Babraham Institute, Cambridge, UK). Ovaries were dissected in PBS containing 2 mg/ml collagenase (Sigma, C2674), 0.025% trypsin (Sigma, 93615) and incubated for 30 min at 37°C in a Thermomixer R (Eppendorf) rotating at 600 rpm, with intermittent mixing by pipetting. The digested ovaries were spread out onto a 10 cm cell culture plate and an equal volume of M2 medium was added. Individual oocytes were serially transferred through several droplets of M2 medium to dilute out somatic cells.

RNA in situ hybridization was carried out using RNAscope on ovaries collected from 15 dpp YJ69/KvDMR1Δ and YJ11/KvDMR1Δ. Ovaries were fixed in 4% paraformaldehyde, dehydrated and embedded in paraffin wax. Sections (5 μm) were processed for RNA using the RNAscope chromogenic kit according to the manufacturer's instructions (Advanced Cell Diagnostics; ACD). Custom probes designed and synthesized by ACD to detect Kcnq1 primary transcripts were used for hybridization for 5 h at 40°C. Following a multistep signal amplification, microscopic evaluation of the ovaries was performed by scoring the turquoise and magenta spots in oocyte nuclei.

Methylation at KvDMR1 was assessed by combined bisulfite restriction analysis (COBRA) (Xiong and Laird, 1997) and bisulfite sequencing (Clark et al., 1994). Genomic DNA (1 µg) was bisulfite converted using EZ DNA methylation kit (ZYMO Research). Three fragments within KvDMR1 were amplified by PCR using primers listed in Table S2. For COBRA, PCR products were purified and digested with BstU1 or TaqI restriction enzymes. For bisulfite sequencing, PCR products were cloned into pCRII-TOPO cloning vector and 10 clones for each region were sequenced using M13 primers. Sequencing data was quantified by web-based QUMA methylation analysis tool (http://quma.cdb.riken.jp/).

RNA was extracted using Trizol, digested with RNase-free DNase (Ambion) and used as template to synthesize cDNA using the SuperScript III First Strand cDNA synthesis Kit or Thermoscript (for analysis of Kcnq1 primary transcript, Fig. S2B,C) (ThermoFisher). cDNA syntheses were carried out using random primers, except in Fig. S2B,C; in this case, Kcnq1 strand-specific RT primers (Int-ls-RT3 & Kcnq1URT8) were used. The imprinted expression of the Kcnq1ot1 lncRNA was analyzed by RFLP RT-PCR using an MwoI polymorphism between 129SvJae and SD7 (Lewis et al., 2006). Expression levels of Kcnq1, Kcnq1ot1, Slc22a18, Cdkn1c, Phlda2 and Ascl2 were determined using real-time quantitative RT-PCR in a 20 µl reaction using iQSYBER Green Supermix (Bio-Rad) PCR was carried out using a BIO-RAD CFX96 instrument. Expression levels were analyzed using the ΔΔCt method. In some cases, allele-specific primers were used to distinguish between maternal (C57Bl/6J) and paternal (SD7) alleles (Mohammad et al., 2010). Alternatively, where allele-specific primers did not perform well, RT-PCR was carried out using primers that detect total expression of genes. Except for allele-specific primers (see Mohammad et al., 2010), primer information is given in Table S2.

The expression levels in Fig. 4, Figs S2, S3, S5 and S6 were modeled as a function of mutant type (YJ11 and YJ69), tissue type (kidney, placenta and embryo, oocyte), type (WT-mat, WT-pat, YJ11/69-mat and YJ11/69-mat) and replicate using general linear models. The specific variables included in a given model depend on the experiment being evaluated. The mean expression levels were compared between groups of interest using F-tests about the appropriate linear contrasts of model estimates. All model assumptions were verified graphically using quantile-quantile and residual plots, and transformations were applied as appropriate using the Box-Cox method. All analyses were conducted in SAS v9.4 (Cary, NC) at a significance level of 0.05. P-values less than 0.05 therefore denote statistically significant differences.

The authors thank Dr Jong-Yeon Shin for construction of the YJ11 and YJ69 mouse lines, and Debbie Tabaczynski for animal husbandry and dissections.