An in vitro ES cell imprinting model shows that imprinted expression of the Igf2r gene arises from an allele-specific expression bias

Genomic imprinting is a cis-acting epigenetic mechanism resulting in the parental-specific expression of ∼100 mammalian genes(Solter, 2006). Imprinted genes mostly occur in clusters, which are regulated by a cis-acting imprint control element (ICE) that is inactivated by a germline DNA methylation imprint on one parental chromosome (Spahn and Barlow, 2003; Thorvaldsen and Bartolomei, 2007). The unmethylated ICE is active and, in the Igf2 imprinted cluster, binds CTCF to form an insulator that blocks maternal expression (Bell and Felsenfeld,2000; Hark et al.,2000). In the Igf2r and Kcnq1 imprinted clusters, the unmethylated ICE contains an active non-coding (nc) RNA promoter that silences multiple genes on the paternal chromosome(Mancini-Dinardo et al., 2006; Sleutels et al., 2002). Thus,as previously noted, genomic imprinting often constitutes the control of cis-regulatory elements by DNA methylation(Mann et al., 2000). Extensive progress has been made in the last decade towards understanding the mechanism,and today genomic imprinting provides one of the best models of mammalian epigenetic gene regulation.

The Igf2r imprinted cluster contains three maternally expressed mRNA genes (Igf2r, Slc22a2 and Slc22a3) that are silenced on the paternal allele by expression of the Airn ncRNA(Sleutels et al., 2002)(formerly named Air, now renamed Airn by the HUGO Nomenclature Committee) (Fig. 1A). The Airn promoter lies in an antisense orientation in Igf2r intron 2. The resultant 108 kb Airn transcript,which is nuclear localised and largely unspliced, overlaps the 5′ part of Igf2r but lies more than 200 kb upstream of Slc22a2 and Slc22a3 (Seidl et al.,2006). The maternal Airn promoter, which lies in a 3.65 kb BamHI-PacI fragment genetically defined as the ICE, is ubiquitously repressed by a DNA methylation imprint acquired in oocytes(Stoger et al., 1993; Zwart et al., 2001). Airn-mediated silencing of Igf2r is seen throughout the post-implantation embryo and adult, with the exception of post-mitotic neurons(Yamasaki et al., 2005), but its silencing effects on Slc22a2 and Slc22a3 appear to be restricted to the trophoblast placenta(Zwart et al., 2001). Paternal-specific silencing of Igf2r, but not of Slc22a2 and Slc22a3, is accompanied by gain of promoter DNA methylation, but,surprisingly, this methylation mark is not necessary for Igf2rsilencing (Li et al., 1993; Seidl et al., 2006).

Genomic imprinting consists of distinct developmental stages: imprint acquisition in gametes, onset of imprinted expression in early embryos,maintenance of imprinted expression in differentiated cells and, finally,imprint erasure in germ cells of early embryos(Barlow and Bartolomei, 2007). Most studies investigating these processes have involved targeted manipulations in an in vivo mouse model - a long-term and laborious procedure. However, some stages in genomic imprinting are potentially amenable to in vitro analysis. Undifferentiated embryonic stem (ES) cells are a cell culture derivative of the pluripotent blastocyst inner cell mass that can provide an in vitro model of early embryonic development(Evans, 2005). In vitro differentiation of female ES cells has been used to study X-chromosome inactivation in mammals (Heard et al.,2004; Wutz, 2007). Changes in Xist ncRNA expression, coating of the inactive X-chromosome by Xist, gain of histone modifications and DNA methylation, are all recapitulated in correct temporal order during ES cell differentiation. Similarities between X-inactivation and imprinted expression indicate that ES cells may also provide a model in which to investigate genomic imprinting (Reik and Lewis,2005). ES cells might be particularly useful to analyze the onset of Igf2r imprinted expression because undifferentiated ES cells express Igf2r biallelically and lack Airn ncRNA expression(Braidotti et al., 2004; Wang et al., 1994). This mimics the in vivo situation, as pre-implantation embryos express Igf2r biallelically and lack Airn in the blastocyst inner cell mass, whereas post-implantation embryos gain imprinted Igf2rexpression between 4.5 and 6.5 days post-coitum (dpc)(Lerchner and Barlow, 1997; Szabo and Mann, 1995; Terranova et al., 2008). Thus,ES cell in vitro differentiation could provide a reliable model in which to examine the developmental onset and maintenance of imprinted Igf2rexpression.

Recent progress in the reprogramming of somatic cells to pluripotent embryonic-like cells has focussed interest on understanding epigenetic gene regulation in ES cells (Jaenisch and Young, 2008). Histone modifications and DNA methylation have been shown to undergo dynamic changes upon ES cell differentiation(Meshorer and Misteli, 2006). For example, ES cells are globally deficient in DNA methylation and appear insensitive to a lack of DNA methyltransferases, but show the reverse behaviour upon differentiation (Meshorer and Misteli, 2006). However, the role played by epigenetic modifications in regulating ES cell pluripotency is not fully understood because, in contrast to differentiated cells, they lack a clear correlation between expression status and the epigenetic modifications typically associated with expressed or silent genes. For example, when silent, key developmental genes are marked both by repressive H3K27me3 histone modifications and by active H3K4me3 modifications (so-called bivalent domains)(Azuara et al., 2006; Lee et al., 2007). It was also shown recently that in undifferentiated ES cells, peaks of H3K4me3 mark all CpG island promoters irrespective of expression state(Guenther et al., 2007). Genomic imprinting is one of the clearest examples of epigenetic gene regulation and could provide a platform to better understand the role played by epigenetic modifications during in vitro reprogramming and differentiation of ES cells.

Here, we follow the dynamics of Igf2r and Airn imprinted expression during mouse ES cell differentiation. Because several reports have shown that imprinted expression can be disturbed in inter- and intraspecies crosses, possibly owing to cis-regulatory polymorphisms(Jiang et al., 1998; Shi et al., 2005), we used inbred ES cells with a polymorphism introduced into our gene of interest. We show that Igf2r is biallelically expressed in undifferentiated ES cells and that activation of Airn ncRNA expression in differentiating ES cells coincides with gain of imprinted Igf2r expression. This is accompanied by characteristic epigenetic changes, including gain of DNA methylation and H3K9me3. Unexpectedly, even though the paternal Igf2rpromoter gains repressive epigenetic modifications it continues to be expressed at similar levels throughout ES cell differentiation, while expression of the maternal Igf2r promoter increases up to 10-fold. Thus, contrary to expectation, we show that the Airn ncRNA induces imprinted Igf2r expression not by silencing the paternal Igf2r promoter, but by creating an expression bias between the two parental alleles.

ES cells were grown under standard conditions. CCE cells are feeder-independent; D3 cells were grown on irradiated 12.5 dpc mouse embryonic fibroblasts. Differentiation was induced by LIF withdrawal, feeder-cell depletion and 0.27 μM retinoic acid. Embryoid body differentiation was induced by hanging-drop culture for 5 days, followed by 9 days on gelatinised dishes.

The S12 targeting vector was constructed by ligation of a 5 kb BstXI-Eco47III fragment(Mm.Build-37/Chr.17:12909688-12914937) into the EcoRV site, and diphtheria toxin fragment A (DTA) ligated into the SmaI site, of pBluescript KS. The C-to-T change that mutates the exon 12 PstI site(bp:12912731) was generated with the QuikChange XL Site-Directed Mutagenesis Kit (Stratagene). The selection cassette Tk-Neo-loxP511 was introduced into the BsrGI site in intron 11 (bp:12913929).

DNase I-treated RNA was assayed in duplicate or triplicate and normalised to cyclophilin A (peptidylprolyl isomerase A). Forward primers: MUTSEF(5′-CTGGCCTTCCCCTCCTGT-3′) detects the mutated allele, whereas WTSEF (5′-TGGCCTTCCCCTCCTGC-3′) detects the wild-type allele. Common reverse primer: GESER2 (5′-GCTATGACCTGTCTGTGTTGGCT-3′). QPCR was performed using the Power SYBR Green PCR Mastermix (Applied Biosystems), with 9 mM MgCl₂ and 100 nM primers. Cycling: 2 minutes at 50°C, 10 minutes at 95°C, then 40 cycles of 15 seconds at 95°C and 1 minute at 64°C.

Standard conditions were used. Probes: EEi:AJ249895:94104-99081,NEi:AJ249895:98070-99081, BEi:AJ249895:97090-99081,MEi:AJ249895:126086-127237, X12:Mm.Build-37/Chr.17:12916453-12917573,MSi:AJ249895:126087-127237,Oct4a:Mm.Build-37/Chr.17:35,643,244-35,644,955.

Standard conditions were used. RPAIII Kit (Ambion) probes:AirF3b/AJ249895:100185-100446, Igf2rex34/NM_010515:4869-5002,Cycl/:Ambion7675. Northern probes: HX/NM_010515:1001-8877,Airp105/AJ249895:115522-116522, Oct4:(X52437:21-354).

Real-time QPCR was performed as described(Seidl et al., 2006) (details available upon request) using the following primers and probes [5′ to 3′; F, forward primer; TM, Taqman probe (Applied Biosystems); R, reverse primer]:

Airn QPCR: F-GACCAGTTCCGCCCGTTT, TM-TACAAGTGATTATTAACTCCACGCCAGCCTCA,R-GCAAGACCACAAAATATTGAAAAGAC;

Igf2rex48 QPCR: F-TCCTACAAGTACTCAAAGGTCAGCAA,TM-CCAAGACTAGGCAAGGACGGGCAAGA, R-GCGGTTGGTGGTGATATGG;

Igf2rex4/5 QPCR: F-GACTACTGCAGCCTGCAAGAAA,TM-ACATATTTAAAGCTGATAAGGAGGTACCATGCTATGCA, R-AATCATGCTTCTGTAACTTGTCATCAA;

Cyclophilin A QPCR: F-AGGGTTCCTCCTTTCACAGAATT, TM-TCGTGGATCTGACGTGCCGCC,R-GTGCCATTATGGCGTGTAAAGT;

Oct4 QPCR: F-ACCTTCAGGAGATATGCAAATCG, TM-AGACCCTGGTGCAGGCCCGG,R-TTCTCAATGCTAGTTCGTTCGCTTT;

Fgf5 QPCR: F-AACTCCATGCAAGTGCCAAAT, TM-TACGGATGACTGTAAGTTCAGGGAGAGATTCCA,R-GGACGCATAGGTATTATAGCTGTTTTC;

Gata4 QPCR: F-CGCTGTGGCGTCGTAATG, TM-AGCCTGTATGTAATGCCTGCGGCCTC,R-GGAACCCCATGGAGCTTCAT;

Slc22a2 QPCR: F-GGAAATCGGTGCCAGTCTC, TM-CTTCAGAGCCTGACGGCAGATGAGGA,R-AAGGGTTCAATTTCATGCCAGT; and

Slc22a3 QPCR: F-GAAATGCACGCTCATCCTTATG, TM-TTGCTTGGTTCACGAGCGCCGT,R-CAGGCGCATGACAAGTCCTT.

ChIP was performed as described (Regha et al., 2007) using the antibodies and primers listed therein.

RNA FISH was performed as described (PROTO6, http://www.epigenome-noe.net). The single-stranded RNA FISH probe (2658 bp PCR product generated using AIFP1F 5′-GCTGGTCCTTACCTTGTGGA-3′ and AIFP1R 5′-GCAAGACCACATCACACACC-3′ from Igf2r intron 1) was transcribed with T7 RNA polymerase and digoxygenin labelled by reverse transcription. The RNA FISH signal was detected by sheep anti-digoxygenin antibody (Roche) and amplified by rabbit anti-sheep-FITC + goat anti-rabbit-FITC (Calbiochem). Fluorescent spots were independently counted twice.

To study imprinted expression at the Igf2r cluster we used two inbred 129/Sv ES cell lines: the feeder-independent but germline-incompetent CCE line (Keller et al., 1993)and the feeder-dependent and germline-competent D3 line(Doetschman et al., 1985). ES cells were differentiated by exposure to retinoic acid (RA) for 5 days or by embryoid body (EB) formation for 14 days. Three marker genes were assayed by QPCR to follow differentiation: Oct4 (Pou5f1) (an undifferentiated stem cell marker), Gata4 (an endoderm marker) and Fgf5 (a gastrulation marker). Fig. 1B shows RA-treated CCE and D3 cells; for the same analysis after EB formation, see Fig. S1A in the supplementary material. Both CCE and D3 cells showed a sharp decline in Oct4 after 2 days of RA differentiation; the decline was more marked in D3. The Oct4 decline was slower during EB formation than RA treatment and remained at 10-35% of the level found in undifferentiated cells after 14 days differentiation,indicating the continued presence of undifferentiated cells (see Fig. S1A in the supplementary material). After RA treatment or EB formation, expression of Gata4 and Fgf5 increased slowly up to 3 days and then sharply up to 5 days with RA differentiation, but Fgf5 then decreased in EB populations (see Fig. S1A in the supplementary material). These expression patterns show that RA produces more-homogenously differentiated cells than does EB formation and identifies 2-3 days of RA treatment as the window when differentiation markers are induced.

We next determined the expression of genes from the imprinted Igf2r cluster. Based on previous data showing that Igf2r is biallelically expressed in undifferentiated ES cells and that AirnncRNA expression is restricted to differentiated ES cells(Braidotti et al., 2004; Wang et al., 1994), we anticipated that Igf2r expression would decrease during ES cell differentiation, as Airn expression silences the paternal Igf2r promoter. Since QPCR only measures relative differences in gene expression, we first used RNase protection to assess non-amplified RNA levels. Fig. 1C shows that low-level Igf2r expression is detected in undifferentiated CCE and D3 ES cells(lane 0). However, instead of the anticipated decrease, we found that Igf2r was strongly upregulated during differentiation, with the most prominent increase between days 3 and 4 of RA treatment(Fig. 1C). The AirnncRNA behaved as anticipated and was absent from undifferentiated cells (lane 0 is overexposed to show the absence of Airn), then sharply increased between days 3 and 4 of RA treatment. Notably, the levels of Igf2rand Airn increase during the same time window as when the differentiation markers change (Fig. 1B), and both also show a sharp increase in expression during days 3-4. The amount of stable Igf2r mRNA is greater than that of Airn ncRNA, in agreement with their differential stability(Seidl et al., 2006).

We then used QPCR to quantify expression of Igf2r, Airn, Slc22a2and Slc22a3. Fig. 1Dand Fig. S1B in the supplementary material show that in CCE and D3 cells, Igf2r and Airn expression increases during ES differentiation with parallel kinetics: a sharp increase is first seen between days 2 and 3 for both genes. Relative to the day-0 undifferentiated state, Igf2r increased on average ∼5-fold (from ∼20 to 100) and Airn increased ∼100-fold (from background to 100) by day 5 of RA treatment (Fig. 1D and see Fig. S1B, left panel, in the supplementary material) or by day 8 of EB differentiation (see Fig. S1C in the supplementary material). Parallel upregulation of a ncRNA with one mRNA is also seen in the Kcnq1 and Igf2 imprinted clusters during ES cell differentiation (see Fig. S2 in the supplementary material). Expression of Slc22a2 and Slc22a3 was undetectable by QPCR until days 2-3 of differentiation and then increased with similar kinetics to Igf2r and Airn(Fig. 1E and see Fig. S1B,D in the supplementary material). Although expression of both Slc22a2 and Slc22a3 is restricted to the trophoblast placenta(Zwart et al., 2001) and thus would be predicted to be absent from differentiated ES cells, the QPCR threshold cycle (Ct value) indicates that Slc22a2 is expressed at low-to-background levels (Ct 34), whereas Slc22a3 is expressed at clearly detectable levels in differentiated ES cells (Ct 27) (see Fig. S1E in the supplementary material).

Imprinted Igf2r expression in differentiated cells could result from transcriptional or post-transcriptional events. RNA FISH using intronic probes detects nascent transcription; however, probes that contain exons will also detect accumulated transcripts. In day-5 RA-treated ES cells, an intronic RNA FISH signal was detected in 17% of cells, and the majority (88%) of positive cells showed single-spot signals(Fig. 1F). Thus, imprinted expression of Igf2r arises from transcriptional differences between the two parental alleles.

Since Igf2r unexpectedly showed increased expression during ES cell differentiation, we used homologous recombination to generate an inbred ES cell line carrying a single nucleotide polymorphism (SNP) to analyze allele-specific Igf2r expression(Fig. 2). The SNP in Igf2r exon 12 mutated a PstI site but maintained the reading frame. To identify the parental origin of the targeted allele, we made use of the maternal-specific DNA methylation imprint on the ICE that results in a diagnostic 21.9 kb band from a paternally targeted allele but not from a maternally targeted allele, after digestion with EcoRV and the methyl-sensitive MluI. One maternally targeted cell line (S12/+) and one paternally targeted cell line (+/S12) are shown in Fig. 2C (note that the maternal allele is written on the left side throughout the text). We then used an allele-specific QPCR assay that distinguishes the exon 12 SNP to analyze a maternally targeted and a paternally targeted ES line for Igf2rexpression. Fig. 3A shows the specificity of the allele-specific assay using plasmids containing cDNA with(S12) or without (WT) the SNP, indicating that the compatible QPCR assay reaches the detection threshold at least seven Ct cycles earlier than the incompatible assay. As shown in Fig. 3B, in both +/S12 and S12/+ cells, the ratio of maternal to paternal Igf2r expression increased from 1:1 to between 4:1 and 10:1 in differentiated cells. The relative increase in maternal Igf2rexpression during ES differentiation varied between different experiments, but always coincided with the onset of Airn upregulation as shown in Fig. 1D.

In allele-specific QPCR assays, the maternal to paternal Igf2rratio did not exceed 10:1 in differentiated cells, even though the calculated specificity for the QPCR SNP assay was at least 50:1. To examine whether this might indicate incomplete silencing of the paternal Igf2r promoter in differentiated cells, we used D3 ES cells with a lacZ-polyA termination signal targeted into exon 1 that truncates the Igf2r mRNA(Wang et al., 1994). For northern blot analysis, we first used an Igf2r-specific probe downstream to exon 1 that only detects the wild-type parental allele(Fig. 3C). The +/lacZcells showed strong upregulation of the wild-type maternal Igf2rallele during differentiation. However, the lacZ/+ cells also showed that wild-type paternal Igf2r expression is maintained at the same level throughout differentiation, and even increases slightly at day 5. Although this could, unexpectedly, indicate that the paternal allele is not silenced during differentiation, we noted that control double-knockout cells(lacZ/lacZ), which should lack all Igf2rexpression, showed traces of Igf2r. This arises from contaminating wild-type feeder cells that persist at early time points despite feeder depletion. We therefore repeated the analysis using IPdel/Thp feeder cells,which completely lack Igf2r expression because the maternal allele carries an Igf2r promoter deletion and the paternal allele carries a 6 Mbp deletion of the whole imprinted cluster(Sleutels et al., 2003). Northern blots (Fig. 3D) showed that Igf2r is absent from lacZ/lacZ ES cells grown on IPdel/Thp feeder cells, and revealed approximately equal expression of the maternal and paternal Igf2r alleles in undifferentiated ES cells. Igf2r expression in these cells was quantified by QPCR, which showed that paternal Igf2r expression in lacZ/+ ES cells is indeed present throughout differentiation and at approximately constant levels until day 4, with a 2-fold increase at day 5 of RA treatment(Fig. 3E). By contrast, the maternal Igf2r allele (+/lacZ) was upregulated beginning at day 2 of RA treatment and increased sharply between days 4 and 5. Together,these results unexpectedly indicate that imprinted Igf2r expression occurs not by silencing the paternal allele, but by creating an expression bias between the two parental alleles.

The above result shows that the paternal Igf2r allele maintains the same expression level during ES cell differentiation. Since the paternal Igf2r promoter in 13.5 dpc embryos displays partial DNA methylation(Stoger et al., 1993), we tested whether methylation is also acquired in differentiated ES cells. Fig. 4 shows that in undifferentiated ES cells, the Igf2r promoter is unmethylated on both parental alleles, in agreement with the above data showing bi-parental Igf2r expression (Fig. 4A). In differentiated cells, a faint band indicative of a methylated Igf2r promoter was observed after day 3 of RA treatment,the period when Igf2r expression starts to be upregulated and Airn is induced. This methylated Igf2r band remained faint in RA-treated cells, but at day 14 of EB differentiation the intensity of the methylated and unmethylated Igf2r bands was similar (quantified in Fig. 4B), indicating complete methylation of the paternal allele (see the NIH3T3 lane, which shows the equal methylated and unmethylated signals). The Airn promoter is continuously modified by a maternal-specific DNA methylation imprint from the oocyte stage onwards, which silences the Airn ncRNA(Seidl et al., 2006). We confirmed that this methylation imprint is present and stably maintained during ES cell differentiation (Fig. 4A). As the Oct4 gene has been shown to gain DNA methylation during ES cell differentiation(Hattori et al., 2004), we tested whether the kinetics of methylation gain were similar to those observed for the paternal Igf2r promoter. Fig. 4C (top panel) shows that differentiating CCE ES cells gained low-level Oct4 methylation by day 2 and that this increased markedly during days 3 to 5, attaining similar levels to those seen in tail DNA. Igf2r methylation has similar kinetics, with a faint methylated band visible by day 3 (dashed line). However, in contrast to Oct4, Igf2r methylation was still partial by day 5 by comparison with tail DNA, which contains a fully methylated paternal allele.

We have previously shown in 13.5 dpc mouse embryonic fibroblasts (MEFs)(which show a maternal to paternal Igf2r expression ratio of 80:1,see Fig. S3 in the supplementary material) that the silent paternal Igf2r promoter and the silent maternal Airn promoter are each modified by a repressive `heterochromatin' peak composed of H3K9me3/H4K20me3/HP1β, but are devoid of repressive H3K27me3 marks(Regha et al., 2007). In MEFs,the expressed maternal Igf2r and paternal Airn promoters lack repressive marks and are modified only by active histone marks (H3K4me2/3 and H3K9Ac). We therefore used chromatin immunoprecipitation (ChIP) of two active (H3K4me2, H3K9Ac) and two repressive (H3K27me3, H3K9me3) histone marks to test whether these modifications arise during ES cell differentiation. Fig. 5B depicts the expression of Igf2r and Airn in undifferentiated and differentiated ES cells, as described above. Fig. 5C-H shows ChIP analysis of the Igf2r promoter region using 12 primer pairs (the arrow indicates the transcription start site and direction). In undifferentiated ES cells (left panel), the Igf2rpromoter is enriched for H3K27me3 on both sides of the transcription start site (Fig. 5D); however, this modification is lost in differentiated ES cells (right panel). H3K9Ac shows the inverse pattern, with enrichment in differentiated ES cells, mainly downstream of the transcription start (Fig. 5E). H3K4me2 is present downstream of the Igf2rtranscription start in undifferentiated ES cells, but moves further upstream in differentiated cells (Fig. 5G). H3K9me3 is mostly absent from Igf2r in undifferentiated ES cells, but is enriched in differentiated ES cells, mainly downstream of the transcription start site(Fig. 5H).

Fig. 5I-N shows the same modifications analysed over the Airn promoter region. Although Airn is only expressed in differentiated ES cells(Fig. 1C), both H3K9Ac and H3K4me2 show the same dynamic changes as seen at the Igf2r promoter. H3K9me3 differs, as it is already present on the Airn promoter region in undifferentiated ES cells and the pattern is largely unchanged in differentiated ES cells. Similarly, H3K27me3 is lost from the Airnpromoter region in differentiated ES cells.

We then examined the same samples by QPCR. Fig. 5O demonstrates that H3K27me3 is present only in undifferentiated ES cells that lack imprinted Igf2r/Airn expression and is absent (i.e. signals are similar to mock ChIP levels, arrows) in differentiated ES cells that have imprinted expression. Thus, differentiated ES cells lack a correlation between H3K27me3 and imprinted Igf2r/Airn expression, as previously shown for 13.5 dpc MEFs (Regha et al., 2007).

We show here that in vitro differentiating ES cells constitute a reliable model in which to study the onset of Igf2r/Airn imprinted expression. Although it is thought that ES cells are vulnerable to epigenetic instability(Pannetier and Feil, 2007), we show using two independent lines that ES cells not only maintain a gametic DNA methylation imprint, but also mimic the onset of imprinted Igf2rexpression previously only observed in vivo. Differentiating ES cells are a model for: (1) the switch from biallelic to imprinted Igf2rexpression that occurs between 4.5 and 6.5 dpc of development(Lerchner and Barlow, 1997; Szabo and Mann, 1995); (2) the activation of the Airn ncRNA, which is silent in the blastocyst inner cell mass and expressed in post-implantation embryos(Terranova et al., 2008; Zwart et al., 2001); (3) the gain of Igf2r promoter DNA methylation seen in post-implantation embryos (Stoger et al., 1993);and (4) the histone modification pattern present in 13.5 dpc MEF cells(Regha et al., 2007). Contrary to expectation, we find that the paternal Igf2r promoter is expressed at similar low levels in undifferentiated and differentiated ES cells, despite Airn ncRNA expression and gain of DNA methylation on the Igf2r promoter region. Together, these results demonstrate that Airn expression does not silence the paternal Igf2rpromoter, but instead creates an expression bias between the two parental alleles.

DNA methylation is a late event relative to imprinted expression:methylation is incomplete at the time when differentiated ES cells show maximum differences between maternal and paternal Igf2r expression. This is consistent with the ongoing reassessment of the biological role of DNA methylation, which indicates that it largely modifies promoters already downregulated by other means (Suzuki and Bird, 2008). The lack of a direct role for promoter methylation in inducing Igf2r imprinted expression is supported by experiments demonstrating that Igf2r is silenced, but Airn is upregulated, in embryos globally deficient in DNA methylation(Li et al., 1993; Seidl et al., 2006). Note that in this and other imprinted gene clusters, DNA methylation can be viewed as repressing a repressor (e.g. the Airn promoter) on one parental chromosome. The differential behaviour of the Igf2r and Airnpromoters in response to global demethylation indicates an inequality between somatic DNA methylation imprints that coincide with the gain of imprinted expression on one parental chromosome, and gametic DNA methylation imprints that modify the ICE on the other parental chromosome. The former appear to be a consequence of imprinted expression, whereas the latter have been shown to directly repress ICE activity and thereby regulate imprinted gene expression(Barlow and Bartolomei, 2007). The molecular basis of the inequality between gametic-derived and somatic-derived DNA methylation imprints is not yet clear, but might reflect the ability of the ICE to recruit accessory proteins, such as the recently described KRAB zinc-finger ZFP57 protein(Li et al., 2008).

Repressive histone modifications show dynamic changes on the Igf2r/Airn promoters during ES cell differentiation, but in a manner that does not correlate with the gain of imprinted expression. H3K27me3 showed strong enrichment in undifferentiated ES cells on both the biallelically expressed Igf2r and the non-expressed Airn promoters. However, this modification was completely lost in differentiated cells that gain imprinted expression of both Igf2r and Airn. This indicates that H3K27me3 is not used to regulate imprinted Igf2r/Airnexpression. This interpretation is supported by our previous studies of parental-specific histone modification in 13.5 dpc MEF cells, which also demonstrated a complete absence of H3K27me3 over the repressed alleles of Igf2r and Airn (Regha et al., 2007). The lack of a role for H3K27me3 in regulating imprinted expression of Igf2r is further supported by examination of 7.5 dpc embryos deficient for Eed, a PRC2 component that catalyzes H3K27me3. Eed mutant embryos lose imprinted expression of 4/14 of the imprinted genes tested, but Igf2r imprinted expression was unaffected in embryonic and extra-embryonic tissue(Mager et al., 2003). Differentiated ES cells model embryonic differentiated cells, and their analysis does not exclude a role for H3K27me3 in regulating imprinted expression of the Slc22a2 and Slc22a3 genes in the trophoblast placenta. Although this has not been directly tested, a recent publication has described a physical association between the AirnncRNA and a nuclear compartment marked by Polycomb proteins in the trophectoderm cells of the late blastocyst that give rise to the trophoblast cells of the placenta (Terranova et al.,2008).

In contrast to H3K27me3, the repressive H3K9me3 modification is only found on Igf2r when it shows imprinted expression in differentiated ES cells. Notably, this H3K9me3 mark is lost in MEFs that carry a 3 kb Airn ncRNA truncation, a shortened form that is unable to silence Igf2r (Regha et al.,2007). Despite the finding that H3K9me3 also modifies the Airn promoter irrespective of expression status (i.e. in undifferentiated and differentiated ES cells), it is possible that this mark plays a direct or accessory role in allele-specific repression in differentiated cells. H3K9me3 has been shown to specifically mark the maternal DNA-methylated Airn promoter in ES cells(Mikkelsen et al., 2007). We have also shown in 13.5 dpc MEF cells that H3K9me3 specifically marks the repressed paternal Igf2r and repressed maternal Airnpromoters and is probably mediated by the ESET histone methyltransferase(Regha et al., 2007). The absence of viable ES cells lacking ESET(Dodge et al., 2004) precludes the significance of this H3K9me3 mark from being directly tested.

Genomic imprinting is a well-studied model of epigenetic gene regulation that has identified a role for DNA methylation in regulating cis-acting insulator elements and macro ncRNA CpG island promoters. We show here that in vitro ES cell differentiation can faithfully mimic key developmental steps as imprinted expression is established (see Fig. 6 for an in vivo and in vitro comparison). Maternal methylation of the ICE, which is the imprint for the Igf2r cluster and acquired in oocytes and lost in 12.5 dpc germ cells (Labosky et al., 1994; Stoger et al., 1993), cannot be studied in an ES cell model system that mimics events in embryonic somatic cells. However, the ICE gametic methylation imprint is stably maintained during early embryonic development and we show that it is similarly maintained during ES cell differentiation (Fig. 4). The switch from non-imprinted to imprinted expression is also modelled by an in vitro ES system. As we show here, undifferentiated ES cells express Igf2r biallelically in a similar manner to pre-implantation embryos (Lerchner and Barlow,1997; Stoger et al.,1993; Szabo and Mann,1995). The onset of imprinted Igf2r expression that takes place between 4.5 and 6.5 dpc in the embryo(Lerchner and Barlow, 1997) is also recapitulated by the gain of imprinted expression as ES cells differentiate (Fig. 3). Similarly, dynamic changes in histone modifications and gain of methylation on the Igf2r promoter that have been shown to be parental specific in MEF cells (Fournier et al.,2002; Regha et al.,2007) are recapitulated in differentiating ES cells(Fig. 5). One drawback to an ES model system is that it cannot be used to study genes that show placental trophoblast-specific imprinted expression, as ES cells are established from the blastocyst inner cell mass, which does not contribute to the trophoblast. However, alternative stem cells, derived from trophectoderm and primitive endoderm lineages, are available and these might prove useful models of placental trophoblast imprinted expression(Rossant, 2007). Thus, we show that ES cell differentiation offers a reliable model system in which to dissect some stages of genomic imprinting. In addition, analysis of the epigenetic regulation of imprinted genes can itself provide valuable information about existing epigenetic mechanisms related to the control of pluripotency in stem cell populations. For example, we have also shown that absence of the Airn and H19 ncRNAs, as well as equal biallelic expression of Igf2r, are stringent markers of undifferentiated pluripotent ES cells.

Imprinted expression is generally interpreted as an epigenetic silencing event (Solter, 2006). However,we show that the paternal Igf2r allele maintains persistent low-level expression despite the gain of DNA methylation and repressive H3K9me3 histone modifications. Although we use an ES cell in vitro differentiation system, we consider that this accurately reflects imprinted expression because post-implantation mouse embryos similarly show persistent paternal expression in some tissues (Lerchner and Barlow,1997). In addition, 13.5 dpc MEF cells with complete DNA methylation on the paternal promoter also show persistent paternal Igf2r expression that is readily detected in northern blots and by non-quantitative PCR (see Fig. S3 in the supplementary material). Several reports have also noted expression of the `silent' allele of an imprinted gene, including Igf2 (Sasaki et al., 1992) and genes in the Kcnq1, Dlk1 and Sgce-Peg10 imprinted clusters (da Rocha et al., 2007; Fitzpatrick et al., 2002; Ono et al.,2003). A recent review has suggested that the majority of imprinted genes might actually show preferential, rather than exclusive,parental-specific expression (Khatib,2007).

Although the Airn ncRNA has traditionally been viewed as inducing paternal-specific silencing of Igf2r, our data allow an alternative epigenetic regulation model (Fig. 6). In model 1, we propose that Airn does not silence the paternal Igf2r promoter but instead acts to prevent its upregulation. We advocate this maternal-specific upregulation model because we observe that the maternal Igf2r allele, which contains the silent methylated Airn promoter, shows dramatic upregulation during ES cells differentiation. On the paternal chromosome, the unmethylated Airnpromoter is upregulated and low-level paternal Igf2r expression is unchanged. The Airn ncRNA would act in this model to prevent the gain of activating epigenetic marks on either the paternal Igf2r promoter or its enhancers. It is interesting, in view of this model, that H3K9me3 has been shown to block active histone marks such as H3K9Ac and H3Ser10P(Rea et al., 2000). This maternal-specific upregulation model is supported by the parallel kinetics of Igf2r and Airn upregulation and our previous analysis of DNase I hypersensitive sites in this region, which concluded that the Igf2r and Airn promoters share the same cis-acting regulatory elements, albeit on different parental chromosomes(Pauler et al., 2005). In model 2, we propose the more classical viewpoint that Airn does silence the paternal Igf2r promoter but a small percentage of cells,in some unknown stochastic way, escape silencing and express high levels of Igf2r biallelically. Analysis of single-cell Igf2rtranscription would distinguish between these two models. However, RNA FISH is currently insufficiently sensitive to detect either the low-level Igf2r expression seen in undifferentiated ES cells (data not shown),or the full biallelic expression seen in differentiated ES cells that lack a functional Airn ncRNA (Stricker et al., 2008). Thus, we cannot directly distinguish between persistent low-level expression of the paternal Igf2r allele in all cells in the population as proposed by model 1 and a low percentage of cells in the population that express Igf2r biallelically as proposed by model 2. Both models, however, accommodate the view that repressive marks may follow the lack of activation and play accessory roles in regulating the paternal Igf2r promoter, as well as our recent proposal that Airn transcription, and not the Airn ncRNA itself, is important for its function (Pauler et al.,2007). The demonstration here that the gain of imprinted expression can be mimicked in vitro in differentiating ES cells provides a valuable tool with which to determine whether the Airn ncRNA acts by blocking the access of activating epigenetic marks, or by recruiting repressive epigenetic marks, to induce imprinted Igf2rexpression.