Zfp57 inactivation illustrates the role of ICR methylation in imprinted gene expression during neural differentiation of mouse ESCs

ZFP57 is required to maintain the germline-marked differential methylation at imprinting control regions (ICRs) in mouse embryonic stem cells (ESCs). Although DNA methylation has a key role in genomic imprinting, several imprinted genes are controlled by different mechanisms, and a comprehensive study of the relationship between DMR methylation and imprinted gene expression is lacking. To address the latter issue, we differentiated wild-type and Zfp57-/- hybrid mouse ESCs into neural precursor cells (NPCs) and evaluated allelic expression of imprinted genes. In mutant NPCs, we observed a reduction of allelic bias of all the 32 genes that were imprinted in wild-type cells, demonstrating that ZFP57-dependent methylation is required for maintaining or acquiring imprinted gene expression during differentiation. Analysis of expression levels showed that imprinted genes expressed from the non-methylated chromosome were generally up-regulated, and those expressed from the methylated chromosome were down-regulated in mutant cells. However, expression levels of several imprinted genes acquiring biallelic expression were not affected, suggesting the existence of compensatory mechanisms that control their RNA level. Since neural differentiation was partially impaired in Zfp57-mutant cells, this study also indicates that imprinted genes and/or non-imprinted ZFP57-target genes are required for proper neurogenesis in cultured ESCs.

www.nature.com/scientificreports/ results obtained by RT-PCR and IF (Fig. 2c). In particular, while ESC markers were generally down-regulated in both wild-type and mutant day 12-JB1 cells, Nestin but no other neural markers were activated in the Zfp57 -/cells at a level similar to that of their control cells, or to previously reported hybrid NPCs 26 . Nevertheless, activation of the endo-mesoderm markers Foxa2, Gata4 and Gata6 was similar in wild-type and mutant strains.  www.nature.com/scientificreports/ By using the DESeq2 software 29 , we identified 4912 differentially expressed genes in the Zfp57 -/day 12-JB1 cells with respect to the day 12-JB1 wild-type cells. These included both up-regulated and down-regulated genes in similar proportions (Fig. 2d, Supplementary Table 2). 50 of these genes were reported to be imprinted 3,4 , indicating that the differentially expressed genes were enriched for imprinted genes (p-value < 1e−04).
We previously showed that loss of Zfp57 causes profound changes in DNA methylation and histone marks at its target sites in pluripotent ESCs 24 . To determine if the inheritance of these epigenetic modifications may have directly affected gene expression in NPCs, we calculated the distance of deregulated genes from previously mapped ZFP57 binding sites in ESCs 24 . We found that deregulated imprinted genes were more frequently within 1000 kb from ZFP57 targets than deregulated non-imprinted genes (Fig. 2e). While imprinted gene expression was mostly altered by defective ZFP57 binding to ICRs at the imprinted loci themselves, many deregulated nonimprinted genes were located far from ZFP57 sites and were likely indirectly affected.
We then examined the identity and functional classification of the deregulated genes by gene ontology analysis. While up-regulated genes were mostly linked to intracellular functions and metabolic processes, down-regulated genes were prevalently involved in neurogenesis and neuron development pathways (Fig. 2f,g). Down-regulated genes involved in neuronal and developmental functions included 6 imprinted genes and 68 non-imprinted genes that were located within 100 kb from the ZFP57 binding sites (Fig. 2h, Table 1 and Supplementary Table 3). Significantly, 24 of these non-imprinted genes were overlapping ZFP57 binding sites. This suggests that in addition to disrupting the expression of important imprinted genes, loss of ZFP57 binding may directly alter the expression of a group of non-imprinted genes that controls neuronal differentiation of ESCs in culture.
Overall, these results demonstrate that when exposed to the differentiation conditions, Zfp57 -/-ESCs exit from the pluripotency state and up-regulate some differentiation markers, but fail to fully activate neural markers compared to the wild-type cells. Consistent with these findings, down-regulated genes were prevalently involved in neurogenesis and neuron development pathways. However, although expression differences between the Zfp57 -/and wild-type cells were extensive and affected both imprinted and non-imprinted genes, both gene and protein expression analyses demonstrated that the differentiated mutant cells were significantly different from pluripotent ESCs, likely undergoing a partial differentiation towards the neural fate. Therefore, we went on to analyze the consequence of ZFP57 loss on the allele-specificity of imprinted gene expression in these neural precursor-like cells (NPLCs).

Massive deregulation of allele-specific expression of imprinted genes in Zfp57 -/cells after differentiation.
We previously demonstrated that pluripotent ESCs lose ICR methylation upon Zfp57 inactivation 24 . Now, we investigated ICR methylation levels in wild-type and Zfp57 -/day 12 cells, to see if any change had occurred during differentiation of pluripotent ESCs. Methylation of 10 ICRs (7 maternally methylated and 3 paternally methylated loci) was determined by several methods. In particular, by bisulfite treatment, cloning and sequencing we demonstrated that differential methylation of the paternal and maternal alleles was maintained in the wild-type JB1 cells at day 12 but was not acquired in the day 12-Zfp57 -/-JB1 cells at the Plagl1, Peg13 and Igf2r ICRs ( Supplementary Fig. 3a). In addition, maintenance of the expected 50% methylation level in the wild-type day 0-JB1 and E14 cells and low level of methylation in their cognate Zfp57-mutant cells after differentiation was demonstrated by bisulfite conversion followed by pyrosequencing or direct Sanger sequencing at the Snrpn, Peg3, Kcnq1ot1, Dlk1-Meg3, Mest and H19 ICRs ( Supplementary Fig. 3b-d). Thus, consistent with previous reports, imprinted ICR methylation was maintained in the wild-type NPCs 26 . Conversely, the ICR methylation levels remained very low in the Zfp57 -/-NPLCs, demonstrating that the loss of DNA methylation induced by ZFP57 depletion in pluripotent ESCs was maintained during in vitro corticogenesis.
To determine the consequences of Zfp57 loss on imprinted gene expression, we analysed the RNA-Seq data obtained from NPCs in an allele-specific manner. The mouse ESC lines used for neural differentiation derived from inter-sub-species (JF1 × C57BL6/J)F1 hybrid blastocysts. The genetic divergence between the C57BL6/6 and JF1 inbred mouse strains (> 4 SNPs/kb) allowed us to distinguish the parental origin of most imprinted genes.
Only for a few protein-coding genes, including Gatm, H13, Mcts2, Zim2, and for the Snord115-6 cluster and many miRNAs, allele-specific expression could not be measured because of absence of suitable SNPs. After removal of genes with low-counts and genes located on the X chromosome (the JB1 ESCs are male), allele-specific expression  Table 1). The new datasets are indicated with bold characters. The heatmap was created using R package "pheatmap version 1.0.12" (https:// CRAN.R-proje ct. org/ packa ge= pheat map). (d) Scatter plot showing deregulated genes in Zfp57 −/− day 12-JB1 versus wild-type day 12-JB1 cells. Deregulated genes are indicated by red dots if downregulated (non-imprinted genes by light-red; imprinted genes by dark-red) and by blue dots if upregulated (non-imprinted genes by light-blue; imprinted genes by dark-blue). (e) Cumulative distribution of the distances (bp) of deregulated genes (imprinted genes: green; non-imprinted genes: orange) in Zfp57 -/day 12-JB1 cells from the ZFP57 target sites determined in wild-type day 0-JB1 cells 24 . The difference between the two curves is statistically significant according to the two-sample Kolmogorov-Smirnov test (p-value < 1e-10). (f,g) Gene ontology analysis of differentially expressed genes in Zfp57 −/− day 12-JB1 cells (f, upregulated; g, downregulated). Bars represent GO:BP terms sorted by high to low -log10 (adjusted p-value). (h) Gene ontology analysis of down-regulated genes located < 100 Kb from ZFP57-KAP1 binding sites.  www.nature.com/scientificreports/ was determined by calculating the mat/(mat + pat) allelic ratio for each gene in the RNA-seq datasets. The genes with an average allelic ratio within the interval 0-0.33 were arbitrarily considered paternally expressed, and the genes with an average allelic ratio in the interval 0.67-1 were considered maternally expressed. In our previous study on pluripotent ESCs, we identified only a few genes with imprinted expression 24 . We have now re-evaluated the RNAseq dataset of pluripotent ESCs reported in ref. 24 together with the new NPC dataset by using SNPsplit, an algorithm which efficiently splits the reads between genomes with known SNP genotypes (Supplementary Table 4) 30 . By following this approach, 15 imprinted genes demonstrated significant parent-of-origin-dependent allelic expression in the wild-type day 0-JB1 cells and 32 in the day 12-JB1 cells ( Fig. 3 and Supplementary  Table 5). Among the NPC imprinted genes, 9 were predominantly expressed from the maternal allele, and 23 from the paternal allele, consistent with what was previously demonstrated in mouse tissues 4 . The genes that were imprinted in the NPCs included genes that were either not imprinted in ESCs (e.g. Zdbf2, Phactr2, Peg13 and Impact) or whose expression in ESCs was too weak to assess allele-specificity (Supplementary Table 5). With a few exceptions, these results are consistent with those described in the study of Bouschet and collaborators 26 . We next analyzed the allele-specific expression of the imprinted genes in the Zfp57-mutant JB1 cells at day 0 and day 12 of cyclopamine treatment. We found that all the genes (but Nespas in ESCs) that were imprinted in wild-type cells, reduced their allelic expression bias in the Zfp57 -/cells (Fig. 3, Supplementary Fig. 4 and Supplementary Table 5). However, the shift toward equivalent expression of the maternal and paternal alleles was more dramatic in the mutant cells at day 12 than at day 0. Notably, several genes (e.g. Nespas, Peg3, Plagl1, Meg3, Rian and Mirg) maintained a significant allelic bias in the mutant ESCs. The Zdbf2, Peg13 and Impact genes, which were not imprinted in wild-type and mutant ESCs, acquired imprinting in the wild-type NPCs, but not in the mutant NPLCs.

Loss of allelic bias is associated with different expression patterns of imprinted genes in
Zfp57 -/-NPCs. To better understand the consequence of ZFP57 loss on genomic imprinting in neural differentiation, we integrated the results of gene expression levels with the allelic ratios determined in the Zfp57 -/day 12-JB1 cells. An overall complex relationship between imprinting status and gene expression level was observed ( Table 2). For several genes, the lack of biased allelic expression corresponded to global up-regulation of their RNA level, for others loss of imprinting was associated with down-regulation of their RNA levels. For the remaining ones, the loss of allele-specificity was not accompanied by a corresponding change in global expression levels. www.nature.com/scientificreports/ The effect of ZFP57 deficiency on imprinted gene expression is better understood if each affected imprinted domain is described separately (Fig. 4 and Supplementary Fig. 5). On chromosome 1, the imprinted domain including the DBF-type zinc finger-containing protein 2 gene (Zdbf2) is controlled by a maternal germline DMR that is located within the Gpr1 gene 31 . This DMR regulates the paternal-specific expression of long isoforms of Zdbf2 (Liz), which in turn activates the canonical Zdbf2 promoter on the paternal chromosome during embryonic development 31 . ZFP57 binds the Gpr1 DMR on its methylated maternal allele, which loses its methylation in Zfp57 -/-ESCs 24 . Paternal-specific expression of Zdbf2 was demonstrated in the wild-type day 12-cells, while this gene was activated twofold and imprinted expression was not acquired in the Zfp57 -/cells cultured in the same conditions (Supplementary Fig. 5a). Another gene, Adam23, reported to be preferentially expressed from its paternal allele in mouse brain 32 , is located 100 kb from Zdbf2, in the same transcription orientation. We found Adam23 biallelically expressed in both the wild-type and Zfp57 -/cells, but observed a twofold activation of the WT day 12 JB1 Figure 3. Allele-specific expression of imprinted genes in the wild-type and Zfp57 -/day 0 and day 12-JB1 cells. Dot plot representing the allelic expression of imprinted genes in wild-type and Zfp57 -/day 0-(green and red circles, respectively) and wild-type and Zfp57 -/day 12-(green and red triangles, respectively) JB1 cells. The blue-shaded area corresponds to paternal (B6)-specific expression (allelic ratio interval: 0-0.33), the pink-shaded area corresponds to maternal (JF1)-specific expression (allelic ratio interval: 0.67-1), the white area corresponds to biallelically expressed genes (allelic ratio > 0.33 or < 0.67). The allelic ratio is the average of two biological duplicates (for details, see Supplementary  Fig. 5a).
In the Blcap/Nnat domain, a germline DMR overlaps the Nnat promoter that in turn is located in a Blcap intron 33 . ZFP57 interacts with this region in mouse ESCs, confirming its identity as ICR. We found paternalspecific expression of Nnat in the wild-type but imprinting was not acquired by this gene in the Zfp57-mutant day 12-cells, demonstrating a direct role of ZFP57-dependent methylation in the imprinting control of this locus (Supplementary Fig. 5b). However, differently from Zdbf2, the global mRNA expression level of Nnat was unchanged in the Zfp57-mutant cells. Blcap appears slightly maternally expressed in the wild-type day 12-cells, but its allelic ratio does not fall in the fixed intervals of imprinted expression, possibly because of the presence of non-imprinted isoforms 34 . However, the almost twofold down-regulation of this gene in the Zfp57-mutant day www.nature.com/scientificreports/  www.nature.com/scientificreports/ 12-cells confirms the hypothesis that Blcap imprinting is co-regulated with Nnat by transcriptional interference (Supplementary Fig. 5b). Imprinting of the Gnas domain is controlled by the paternally expressed lncRNA Nespas 35 . Nespas downregulates its sense counterpart Nesp, which in turn controls the reciprocal gamete of origin-specific expression of Gnas and GnasXL. The Nespas promoter overlaps a germline DMR that is maternally methylated and bound by ZFP57 in mouse ESCs 24 . We found Nespas expressed from its paternal allele in the wild-type but activated on both parental alleles with an overall twofold increase of its mRNA level in the Zfp57-mutant day 12-cells, consistent with ZFP57-dependent imprinting regulation (Supplementary Fig. 5c). Unfortunately, Nesp expression was too low to determine its allelism in NPCs, and assess the regulatory role of ZFP57-dependent methylation on this gene.
A ZFP57-marked maternally methylated germline DMR overlaps the promoters of Sgce and Peg10, on proximal chromosome 6 36 . Consistent with Zfp57-dependent imprinting control of these two genes, we observed their paternal-specific expression in wild-type and bi-allelic expression in Zfp57-mutant day 12-cells (Fig. 4a). The change in global mRNA levels of these two genes, however, was not consistent: Peg10 was up-regulated twofold, while Sgce displayed similar levels in wild-type and mutant cells. Another imprinted gene, Asb4, lies 700 kb distal from Sgce/Peg10. We found that Asb4 was preferentially expressed from the maternal allele in the wildtype and lost its allelic bias in the mutant day 12-cells, although its global mRNA level was unchanged (Fig. 4a). Since the Sgce/Peg10 DMR is the closest ZFP57 binding site in mouse ESCs 24 , this finding is consistent with the regulation of Asb4 by this ICR. Of note, two further imprinted genes of this cluster that are highly expressed in neural cells, namely Casd1 and Ppp1r9a 37 , were bi-allelically expressed in wild-type but down-regulated in the Zfp57-mutant day 12-cells. This suggests co-regulation of these genes with the rest of the cluster and is consistent with the incomplete neural differentiation of the Zfp57 -/cells.
A further maternally methylated germline DMR is present on chromosome 6 and overlaps the promoter of Peg1/Mest 38 . Consistent with regulation by this ZFP57-bound DMR, we found that Peg1/Mest and its longer isoform MestXl were paternally expressed in the wild-type and bi-allelically expressed in the Zfp57-mutant day 12-cells ( Supplementary Fig. 5d). We also observed an almost twofold up-regulation of Peg1/Mest in mutant cells. Although we did not detect any allelic bias for Copg2, this gene was down-regulated in Zfp57-mutant cells, consistent with the hypothesis that its expression is controlled by the overlapping anti-sense MestXL 38 . A further chromosome 6 gene, Cntn3, reported as imprinted in placenta but not in brain where it is highly expressed 39 , was bi-allelically expressed in the wild-type but strongly down-regulated in the Zfp57-mutant day 12-cells cells. This gene is not located close to an imprinted domain, but a ZFP57 binding site is located 1 Mbp from its transcription binding site 24 . Its repression may be caused by loss of ZFP57 binding at this site, although it could also be a consequence of the incomplete neural differentiation of the Zfp57-null cells.
On proximal chromosome 7, a ZFP57-bound DMR overlaps the promoters of the divergently transcribed Peg3 and Usp29 genes (Fig. 4b) 24 . Consistent with their imprinting being regulated by this DMR, Peg3 and Usp29 were both paternally expressed in the wild-type but bi-allelically expressed in the Zfp57-mutant day 12-cells. However, at the global level, Peg3 was activated twofold while Usp29 expression remained unchanged. Within the same domain, the maternally expressed Zim1 also reduced its allelic bias, but its expression was strongly down-regulated in the mutant day 12-cells, consistent with the hypothesis that Peg3 is a repressor of Zim1 40 .
In the central part of chromosome 7, another ZFP57-bound DMR overlaps the promoter of Snurf/Snrpn. This gene encodes large alternatively spliced transcripts, including the 460 kb lncRNA Snhg14, which were paternally expressed in the wild-type but were bi-allelic in the Zfp57-mutant day 12-cells (Fig. 4c). This cluster includes several other paternally expressed genes (A230057D06Rik, Ndn, Magel2, Mkrn3 and Peg12) that all lost or reduced their allelic bias in Zfp57-mutant cells, indicating that they are coordinately controlled by methylation of the above mentioned DMR. Concerning their global RNA level, only Snrpn was up-regulated almost twofold, while the others were either unchanged or down-regulated.
On distal chromosome 7, two ZFP57-bound DMRs mark two adjacent imprinted domains. The more centromeric domain includes the H19 and Igf2 genes. These both displayed imprinted expression in the wild-type and lost or reduced, respectively, their allelic bias in the Zfp57-mutant day 12-cells (Fig. 4d). However, the paternally expressed H19 was activated and Igf2 repressed, consistent with the model of the paternally methylated insulator corresponding with the ZFP57-bound DMR 1,41 . Interestingly, the Igf2 antisense transcript Igf2os was down-regulated similar to Igf2, indicating that it is controlled through a similar mechanism. The more distal ZFP57-bound DMR overlaps the promoter of the antisense noncoding gene Kcnq1ot1, which was found paternally expressed in the wild-type and bi-allelically expressed in the mutant day 12-cells, confirming its dependence from DMR methylation (Fig. 4e) 1 . The global level of the Kcnq1ot1 RNA was unchanged. Consistent with the model predicting the regulation of this domain through the cis-repressive function of Kcnq1ot1, the closely located Cdkn1c gene was down-regulated and its maternal expression bias reduced. Th was also repressed but its allelic bias was borderline (adj. p-value = 0.0493). Another gene, Ano1, which lies 1 Mbp from the Kcnq1ot1 DMR and was reported to be maternally expressed in placenta 42 , was found down-regulated in the Zfp57-mutant day 12-cells and therefore possibly coordinately regulated with the Kcnq1ot1 domain, although no allelic expression bias was detected in the wild-type NPCs.
A further ZFP57-bound DMR overlaps with the Plagl1 promoter on chromosome 10 24 . Consistent with regulation by this maternally methylated DMR, the Plagl1 gene was found paternally expressed in the wild-type and bi-allelically expressed in the Zfp57-mutant day 12-cells (Fig. 4f). At the global level, the Plagl1 mRNA was down-regulated in the mutant cells. The overlapping noncoding RNA Hymai is not currently mapped by UCSC, but according to the enrichment of the reads corresponding to its location (Plagl1 intron 1), it appears to have the same behavior as Plagl1. The imprinting status of the close-by Phactr2 gene is controversial 26,43 . However, we observed that Phactr2 was preferentially expressed from the maternal allele in the wild-type cells, and its allelic bias was abolished in the mutant cells, suggesting that its imprinting is co-regulated with Plagl1, although its global expression level was similar in the wild-type and mutant cells. www.nature.com/scientificreports/ Chromosome 11 harbours two maternally methylated ZFP57-bound DMRs 24 . The proximal one overlaps a CTCF binding site and an alternative promoter of the Grb10 gene 44 . It has been proposed that DNA methylation interferes with the enhancer-blocking function of this DMR thereby allowing activation of the main Grb10 promoter on the maternal chromosome. Consistent with this hypothesis, we found Grb10 expressed from the maternal allele in the wild-type and from both parental alleles in the Zfp57-mutant day 12-cells ( Supplementary  Fig. 5f). However, loss of imprinting was accompanied by down-regulation of its global mRNA level. The Ddc gene is located downstream of Grb10. Ddc showed a paternal expression bias in the wild-type day 12-cells, which is consistent with previous reports 26 , but its bias was not strong enough to be classified as imprinted under our stringent criteria. However, its allelic expression was inverted and its global expression activated in the Zfp57 -/day 12-cells, indicating that this gene may also be regulated by the Grb10 methylation-sensitive insulator. The distal chromosome 11 DMR overlaps the promoter of the Zrsr1 gene. Zrsr1 was expressed from its paternal allele in the wild-type and up-regulated twofold on both parental alleles in the Zfp57-mutant day 12-cells, consistent with its regulation from its maternally methylated DMR (Supplementary Fig. 5e). We did not find any allelic expression bias of the overlapping Commd1 gene in NPCs.
In the large cluster of imprinted genes on chromosome 12, ZFP57 binds a paternally methylated DMR that is located between the Dlk1 and Meg3 genes. We found Dlk1 paternally expressed and Meg3 maternally expressed in the wild-type but these genes were expressed from both parental alleles in the Zfp57-mutant day 12-cells, consistent with the model predicting imprinting dependent on the intergenic DMR methylation (Fig. 4g). In addition, Dlk1 was down-regulated, consistent with the repressor function of the lncRNA Meg3 45 . A number of other genes (including Mirg and Rian) that are co-transcribed with Meg3 within a polycistronic RNA drastically reduced their allelic bias (Supplementary Fig. 6a). Although the global RNA level of Meg3 was only slightly changed, those of the small nucleolar RNA-containing gene Rian and the micro RNA-containing gene Mirg were up-regulated twofold and fourfold, respectively, in the mutant day 12-cells.
On chromosome 15, a ZFP57-bound DMR overlaps the promoter of the lncRNA gene Peg13 24 . This gene was paternally expressed in the wild-type and biallelically expressed in the Zfp57-mutant day 12-cells, consistent with being regulated by maternal methylation of this DMR, although its global expression was unchanged (Fig. 4h). The allelic bias of Trappc9, which overlaps Peg13, and that of Kcnk9 located at its 3' were not statistically significant in the wild-type but both of these genes were down-regulated in the Zfp57-mutant day 12-cells , consistent with their co-regulation with Peg13.
On chromosome 17, the promoter of the lncRNA Airn overlaps a maternally methylated and ZFP57-bound DMR 24 . Consistent with the model predicting transcriptional interference between Airn and Igf2r 46 , the former gene was activated and the latter repressed in the Zfp57-mutant day 12-cells ( Supplementary Fig. 5 g). Igf2r appears slightly maternally expressed in the wild-type NPCs but its allelic ratio does not fall in the fixed intervals of imprinted expression, while Airn appears paternally expressed, but its counts are too low to perform a statistical test. However, the symmetrical changes of their allelic ratios observed in the Zfp57-mutant day 12-cells is consistent with their reciprocal imprinting being regulated by Zfp57-dependent DMR methylation.
Finally, on chromosome 18, ZFP57 marks the maternally methylated DMR overlapping the promoter of the Impact gene. The imprinting mechanism controlling Impact is poorly investigated. We found that this gene is paternally expressed in the wild-type and that the allelic bias is reduced in the mutant day 12-cells, with a global upregulation of about fourfold. These findings suggest that its imprinting is controlled by the ZFP57-bound DMR ( Supplementary Fig. 5h).
In conclusion, all the imprinted genes detected in wild-type NPCs lost or reduced their allelic expression bias in the Zfp57 -/-NPLCs. However, those expressed from the non-methylated chromosome were generally up-regulated, and those expressed from the methylated chromosome were generally down-regulated. In addition, several genes (such as Sgce, Asb4, Usp29, Snhg14, Ndn, Magel2, Kcnq1ot1, Phactr2, Meg3 and Peg13), whose expression was expected to be increased upon loss of ICR methylation, acquired biallelic expression but overall were not significantly activated.

Discussion
Differential DNA methylation of ICRs is the best characterized mechanism by which imprinted gene expression is maintained in somatic cells. However, the relationship between ICR methylation and gene expression is not well understood for all imprinted genes. Zfp57 -/-ESCs display a specific DNA methylation defect at the 20 ICRs that are bound by ZFP57 in mouse pluripotent cells, and represent an ideal model to study the role of methylation in the control of imprinted gene expression during differentiation 24 . By employing an in vitro system of neural differentiation of hybrid ESCs, this study highlights the importance of ZFP57-dependent methylation in genomic imprinting, but also illustrates that imprinted gene control is more complex than what previously thought.
An unexpected finding of this study was that although the Zfp57 -/cells were able to leave the pluripotency state they only partially differentiated into NPCs if compared to their derivative wild-type cells, under the conditions used in this study. We cannot exclude that the deregulated expression of some imprinted genes is secondary to the differentiation differences observed between the wild-type and mutant strains. Nevertheless, the general loss of allelic expression and the coordinated expression changes of the genes located in the same domain suggest that most of the imprinting loss is directly caused by lack of ZFP57-dependent ICR methylation.
Despite the presence of differential methylation at ICRs, allele-specific expression was well established at only a few genes in the undifferentiated ESCs. Some genes acquired imprinted expression after differentiation, others increased their allelic bias in NPCs. For several other genes, imprinting could be assessed only in NPCs, because their expression in ESCs was too low. Nevertheless, all the imprinted genes of wild-type NPCs, overall belonging to 15 of the 20 known imprinted gene clusters, lacked or reduced their gamete of origin-specific expression in Zfp57-mutant NPCs. This demonstrates that Zfp57-dependent methylation is generally required www.nature.com/scientificreports/ for both acquisition and maintenance of imprinted gene expression in somatic cells. Interestingly, some genes, such as Nespas, Peg3, Plagl1, Meg3, Rian and Mirg, partially retained their allelic bias in the Zfp57-mutant day 0-cells, but completely lost it in the Zfp57-mutant day 12-cells. Since both the Zfp57-mutant day 0 and Zfp57mutant day 12-cells completely lack ICR methylation, this suggests that Zfp57-dependent methylation is the major determinant of imprinted gene expression in differentiated cells, but additional epigenetic mechanisms contribute to imprinting maintenance in pluripotent cells. Lack of Zfp57 leads to a more limited ICR methylation loss in mouse embryos than in cultured cells 17 . This is likely because the ESCs and the derived early differentiating cells undergo many more cell divisions than the equivalent cells in the early embryo, and there can thus be accumulation of small losses each cell cycle, eventually leading to fixation of the unmethylated state. So, an advantage of our ESC system is that it allows to score regions at which ZFP57 contributes only partly to the maintenance of methylation. The consequence of Zfp57-knockout on imprinted gene expression was investigated in mouse embryos in a very recent study 47 . The data of this study are consistent with ours, but are limited by an incomplete loss of ICR methylation of Zfp57-null embryos. Differently from our neural cell-based system, this in vivo study analyzed whole E13.5 embryos, which did therefore not allow detection of allelic expression of several tissue-specific imprinted genes. It is important to note, however, that because the Zfp57 gene was inactivated prior to differentiation, none of these studies addresses the possible role of this gene in differentiated cells. In humans, ZFP57 inactivation results in a less severe phenotype than in mice, possibly because ZFP57 is not expressed in human oocytes and is activated only post-implantation, and another zinc-finger protein gene, ZNF445, appears to play a major role instead 18,25 .
Among the imprinted genes that were deregulated in the Zfp57-mutant NPLCs, some displayed up-regulation of the silenced allele, others down-regulation of the expressed allele, with respect to the wild-type cells. This behavior is consistent with the relationship between DMR methylation and allele-specific expression of each imprinted gene. In general, the gene in cis with DMR methylation acquired the expression of the other parental allele, in mutant cells. Thus, maternally expressed genes were repressed and paternally expressed genes activated in the domains with maternal ICR methylation. Conversely, maternally expressed genes were activated and paternally expressed genes repressed in the domains with paternal ICR methylation.
Intriguingly, several imprinted genes that were expected to be activated twofold-because now transcribed from both the parental alleles-did not significantly change their global expression level in Zfp57-mutant cells ( Table 2). This suggests that compensatory mechanisms exist for many imprinted genes that finely control their RNA level. Most of these genes, interestingly, are expressed at a lower level in wild-type NPCs. Since their overall expression levels were not affected by expressing two alleles versus one, it may also be speculated that these genes are not the ones that have driven the evolution of imprinted gene expression. In the context of possible dosage compensation mechanisms, it is interesting to note that multiple regulatory links exist between different imprinted domains 48 . For instance, the maternally expressed miRNAs (miR-379/410) encoded by the Dlk1-Meg3 imprinting domain antagonize paternally expressed transcripts from other imprinted domains, including Plagl1, Peg3, Igf2 and Mkrn3 49,50 . The unexpected reduction in Plagl1 and Mkrn3 RNA could be explained by up-regulation of these miRNAs in the Zfp57-mutant day 12-cells ( Supplementary Fig. 6a-c). This could also be responsible of the indirect down-regulation of several neuronal non-imprinted miRNA-target genes in the Zfp57 -/-NPLCs ( Supplementary Fig. 3d).
Zfp57-mutant cells only partially differentiated towards neural precursors, suggesting that genes controlled by ZFP57 are necessary for neural differentiation of ESCs under the conditions used in this study. This finding contrasts with the absence of major problems of neurogenesis reported in Zfp57 mutant embryos 17 . However, the Zfp57 -/-ESCs show more methylation defects at the ZFP57-binding sites and are therefore expected to have more gene expression changes and severe phenotype than the mutant embryos 24 . Consistent with this hypothesis, the Dnmt3l knockout that prevents establishment of methylation at all maternally imprinted DMRs is associated with exencephaly and other neural tube defects in the mouse 51 . A possible explanation of the partial differentiation capacity of the Zfp57 -/cells toward the neural fate is the deregulation of imprinted genes that have a role in the control of neural cell proliferation, differentiation, apoptosis and migration, such as Cdkn1c, Dlk1, Igf2, Peg3 and Plagl1 [52][53][54][55][56] . Alternatively, the expression of non-imprinted genes involved in neurogenesis that are located close to the ZFP57 binding sites may be altered.
In conclusion, this study provides an extensive analysis of the relationship between DMR methylation and imprinted gene expression, and suggests that a number of ZFP57-target genes have a role in neural differentiation of mouse ESCs in vitro.

Materials and methods
Cell lines and culture conditions. The wild-type hybrid ESC line JB1, which is (JF1 × C57BL/6) F1, and the JB1-derived Zfp57 -/-ESC line were described previously 24,57 . ESCs were cultured under serum-free conditions on gelatinized tissue culture dishes and maintained in ESGRO Complete Plus Serum-free Clonal Grade 1i Medium (Merck-Millipore) in the presence of the Gsk3 inhibitor CHIR99021 (at 3 μM). The wild-type inbred 129 ESC line E14 and its derived Zfp57 -/-ESC line were cultured under standard feeder-free conditions as previously described 24 . The wild-type and Zfp57 -/-JB1 and E14 ESCs were differentiated toward NPCs following the protocol described by Gaspard and collaborators (Fig. 1a) 58 . In vitro corticogenesis of ESCs was started with plating at the optimal density of 7.500 ES cells/cm 2 and culturing in chemically defined default medium (DDM). From day 2 to 10 of culture, a Sonic hedgehog inhibitor (cyclopamine) was added at 1 μM concentration. From day 10 to 12, the medium was replaced with DDM only. The cells were cultured at 37 °C under an atmosphere of 5% CO 2 . www.nature.com/scientificreports/ intersected using bedtools, which revealed 457 overlapping peaks. The distance of deregulated genes in the Zfp57 -/day 12 cells from the ZFP57/KAP1 peaks was calculated using bedtools (-closest function) and plotted using ggplot2 R package. The genes overlapping with ZFP57/KAP1 peaks were set at distance 0.
Statistical analysis. Statistical significance for the relative expression of the lineage-specific markers and imprinted genes was conducted by unpaired Student's t-test. Enrichment of differentially expressed imprinted genes over all differentially expressed genes was assessed using the Hypergeometric test. The difference between the two empirical cumulative functions of the distances between genes and ZFP57/KAP1 peaks was evaluated using the two-sample Kolmogorov-Smirnov test. The allele-specific expression of genes in wild-type cells was assessed using a Proportion test and multiplicity correction was performed using the "Benjamini-Hochberg" method.