Transcriptional profiling at the DLK1/MEG3 domain explains clinical overlap between imprinting disorders

Temple or Silver-Russell syndrome? We reveal the molecular reasons for the clinical resemblance between these two syndromes.


INTRODUCTION
Genomic imprinting is a physiological process defined as the mono allelic expression of a gene according to its parental origin, under the control of a differentially methylated region (DMR), known as the imprinting control region (ICR) (1). More than 150 human genes have been shown to be imprinted. Imprinting disorders (IDs), caused by disturbances of imprinted genes, are a group of congeni tal diseases affecting growth, development, and metabolism in hu mans, leading to diseases with overlapping features, regardless of the genomic region affected (2). Some of this overlap may be ex plained by the coregulation of imprinted genes, which belong to an imprinted gene network (IGN) involved in the control of cellular proliferation and differentiation (3). Recent studies in mammals have shown how the disturbance of one imprinted gene can affect other maternally expressed genes (MEGs) or paternally expressed genes (PEGs) (4)(5)(6). The overlap between SilverRussell syndrome (SRS) (7) and Temple syndrome (TS14) (8) is a particularly demonstrative example of clinical overlap between IDs. Both these syndromes in clude fetal and postnatal growth retardation, early feeding difficul ties, early puberty, and an increase in the risk of metabolic disorders (Fig. 1A) (8,9). Moreover, TS14 patients also have a number of clin ical features in common with another ID, PraderWilli syndrome (PWS), a differential diagnosis for TS14 (Fig. 1A) (8). Most SRS pa tients carry molecular changes in the 11p15.5 region (Fig. 1B), the most prevalent (~50%) of which is hypomethylation of the DMR H19/IGF2:IGDMR (hereinafter referred to as ICR1), decreasing paternal IGF2 (a potent fetal growth factor) expression and increas ing the maternal expression of H19 (10), encoding a lncRNA. TS14 patients present molecular abnormalities at the paternally methyl ated imprinted locus on chromosome 14q32.2 (Fig. 1C). The most frequent of these abnormalities is maternal uniparental disomy or upd (14)mat. Paternal deletions of the imprinted DLK1/MEG3 domain and DLK1/MEG3:IGDMR (hereinafter referred to as IG DMR) hypomethylation are less frequent (8). Last, PWS patients present disturbances of imprinted genes at the SNRPN locus and its ICR (ICSNRPN) in the 15q11q13 region (Fig. 1D), resulting in a loss of the expression of SNRPN/SNURF and IPW, two PEGs map ping to the minimal deletion interval critical for PWS (6). A recent clinical study published in 2017 on a large cohort of 32 TS14 pa tients with 14q32.2 genetic and epigeneticrelated defects revealed both PWS and SRSlike phenotypes in 50% of patients (11). More over, we have recently shown that chromosome 14q32.2 imprinting defects are an alternative molecular diagnosis of SRS (12). Evidence is accumulating that these methylation defects in patients with SRS and TS14 are not isolated events, with some patients having multi locus imprinting disturbances (MLIDs) affecting additional im printed regions (12,13).
Human chromosome 14q32.2 encompasses an imprinted region containing three PEGs (DLK1, RTL1, and DIO3) and a number of MEGs. All the MEGs encode noncoding RNAs (ncRNAs) (MEG3/ GTL2, MEG8, MEG9, and RTL1AS) and several large clusters of microRNAs (miRNAs) and small nucleolar RNAs (snoRNAs) (14). The monoallelic parentspecific expression of these genes is controlled by the germlinederived primary intergenic IGDMR and the postfertilization derived secondary MEG3DMR, both of which are methylated on the paternal allele and unmethylated on the maternal allele (15). PEGs from the 14q32.2 region play a crucial role in cell differentia tion and tissue development, whereas the function of the MEGs remains unclear (16). It has been shown that the hypermethylation of the IGDMR results in the reactivation of the normally silenced maternal allele of PEGs and a loss of expression of MEGs (17). The effect of ICR1 hypomethylation on the expression of IGF2/H19 do main genes has been determined for SRS patients (10), but the effect

of 12
of IGDMR hypomethylation has yet to be determined in TS14 patients.
We studied gene expression following the hypomethylation of IGDMR to characterize the effect of this epigenetic alteration on DLK1/MEG3 domain gene expression. We performed expression profiling for imprinted and nonimprinted genes in human fibro blasts from TS14 (IGDMR hypomethylation) and SRS (ICR1 hy pomethylation) patients (Fig. 2) to identify possible gene expression signatures common to these two IDs, which present a major clinical overlap.

TS14 patients' molecular diagnostics and collected biological materials
We collected serum from seven TS14 patients with IGDMR and MEG3DMR hypomethylation (n = 5) or 14q32.2 paternal deletion (n = 2). We also established fibroblast cell cultures for four TS14 patients with IGDMR hypomethylation, one SRS/TS14 patient with both 11p15.5 ICR1 and IGDMR hypomethylation, five SRS patients with ICR1 hypomethylation, and five controls (cells were provided by Coriell Cell Repositories). Clinical data and methyl ation levels for all patients and controls are listed in tables S1 and S2, respectively. DLK1 is absent from the serum of TS14 patients but present in that of age-matched controls DLK1 is a singlepass transmembrane protein that can be cleaved by extracellular proteases to release a circulating form (18). We as sessed the effect of DLK1/MEG3 domain hypomethylation on DLK1 expression by first measuring the circulating levels of DLK1 in the serum of healthy children (n = 38, 19 boys and 19 girls) between the ages of 0 and 17 years. We found that serum DLK1 levels decreased considerably after birth, but those patients with paternal deletions or hypomethylation of the DLK1/MEG3 domain had barely detect able levels of DLK1, regardless of their sex, age, or the molecular defect at 14q32.2 (Fig. 3A).

MEG3, MEG8, and miRNAs are up-regulated because of their expression from the normally silenced paternal allele
We then investigated the levels of transcription of 14q32.2 MEGs in cultured fibroblasts from TS14 and SRS patients (PEGs were not expressed in fibroblasts). We found that the levels of the lncRNA MEG3 and the ncRNA MEG8 in all TS14 and SRS/TS14 fibroblasts were at least twice those in fibroblasts from controls (Fig. 3, B and C). SRS fibroblasts had normal MEG3 levels, but MEG8 levels were twice those in control fibroblasts (Fig. 3, B and C). In addition to MEG3 and MEG8, we quantified nine miRNAs (encoded by genes distributed throughout the 14q32.2 region), all of which were found  to be upregulated in fibroblasts from TS14 and SRS/TS14 patients relative to SRS patients and controls (Fig. 3D shows data for three of the nine miRNAs). We assessed the stability and passage independence of the overexpression of these genes by quantifying the mRNAs at five different passages for each patient. We found that, in all patients, all the ncRNAs and miRNAs studied were stably overexpressed in all cultures (fig. S1, A and B). Furthermore, singlenucleotide poly morphism (SNP) genotyping (Fig. 3E) provided the first evidence of biallelic expression of MEG3 in fibroblasts from patients with TS14, due to reactivation of the normally silenced paternal allele. No SNP assessment was possible for MEG8 or the miRNAs.

MLIDs at 14q32.2 and 11p15.5 are associated with multi-imprinting relaxation
Fibroblasts from a patient with SRS/TS14 displaying hypomethyl ation at both 14q32.2 and 11p15.5 loci displayed biallelic expression of MEG3 from the DLK1/MEG3 domain. We investigated whether MLID induced multiimprinting relaxation and biallelic expression at the secondary loci affected by sequencing H19 genomic DNA (gDNA) and complementary DNA (cDNA) from SRS/TS14 fibro blasts. We found that H19 was monoallelically expressed in TS14 fibroblasts, whereas it was biallelically expressed in SRS and SRS/ TS14 fibroblasts (Fig. 3F).

11p15.5 and 15q11-q13 PEGs are down-regulated by 14q32.2 hypomethylation
We then studied the expression of a large number of imprinted genes from loci involved in growth control in TS14, SRS, and con trol fibroblasts. For most of the PEGs, expression was barely detect able (PEG3, PEG10, and GNAS-XLa), whereas MEGs were highly variable (H19) or comparable to controls (UBE3A and GRB10) in fibroblast cultures (fig. S1C). We found that the 11p15.5 (IGF2) and 15q11q13 (SNURF and IPW) PEGs were strongly and stably expressed in cultures of fibroblasts from controls and patients. All TS14 fibroblasts displayed low levels of IGF2 expression, as did SRS and SRS/TS14 fibroblasts. Furthermore, both TS14 and SRS fibro blasts displayed low levels of SNURF expression, whereas only TS14 fibroblasts had low levels of IPW expression (Fig. 4). Similar to the ncRNAs and miRNAs, IGF2, SNURF, and IPW were all stably down regulated in all fibroblast cultures, at various passages, for all pa tients (fig. S1D).

MEG3 and MEG8 overexpression and knockdown in control fibroblasts are associated with deregulation of the expression of 11p15.5 and 15q11-q13 PEGs
To study whether the clinical overlap could be explained by the crossregulation between the three imprinted regions, we investi gated the possible role of 14q32.2 ncRNAs MEG3 and MEG8 in the changes in IGF2, SNURF, and IPW expression. For this purpose, we carried out overexpression and knockdown of MEG3 and/or MEG8 in a control primary fibroblast line.
Expression constructs were available only for MEG3. For MEG8, we inserted the fulllength MEG8 cDNA obtained from fibroblasts into an expression vector. MEG8 cDNA amplification revealed the presence of two different transcripts: a 497base pair (bp) transcript corresponding to NR_024149.2 and a previously unreported 639 bp transcript (GenBank KX237564; fig. S2). This new variant has an additional exon, located between exons 1 and 2 of the NR_024149.2 transcript, flanked by donor and acceptor sites for splicing.
We found that the concomitant overexpression of MEG3 and MEG8 in control fibroblast cultures was associated with a moderate but significant downregulation of IGF2 expression (Fig. 5A). As for the 15q11q13 PEGs, we found that overexpressing MEG8 (regard less of the variant used for transfection) in control fibroblasts was associated with a downregulation of SNURF expression, whereas MEG3 overexpression was associated with IPW downregulation (Fig. 5A). To investigate whether the downregulation of MEG3 and MEG8 could also lead to deregulation of theses PEGs, we knocked down both ncRNA expression with siRNA in the same cell line. We found that IGF2 expression increased following MEG3 and/or MEG8 knockdown, whereas IPW and SNURF were upregulated by the knockdown of MEG3 and MEG8, respectively (Fig. 5B). These changes in the expression of IGF2, SNURF, and IPW were observed in all transfected cells, despite the low transfection efficiency due to the use of primary fibroblast cultures as the host cells for transfec tion assays.

Principal components analysis of overall gene expression in TS14 and SRS fibroblasts
We investigated the dynamic changes in global gene expression lev els and searched for possible gene expression signatures common to these two IDs by performing RNA sequencing (RNAseq)-based expression profiling on TS14, SRS, and control fibroblasts [Gene Expression Omnibus (GEO) accession number GSE109408]. We first analyzed and compared the overall gene expression patterns of TS14 and SRS fibroblasts. For this purpose, we used principal com ponents analysis (PCA), a statistical technique that summarizes large datasets, reducing the number of dimensions while illustrating relationships between samples based on the covariance of the data considered (19). PC1, PC2, and PC3 accounted for ~50% of the to tal variance of the original data (Fig. 6A). When all samples (TS14, SRS, and controls) were plotted in a twodimensional space defined by PC1 and PC2 (accounting for ~35% of all variation), we were able to distinguish two distinct groups, the first clustering all the TS14 and SRS patients and the second encompassing all the con trols (Fig. 6B). However, when the samples were plotted in a three dimensional space defined by PC1, PC2, and PC3, we were able to separate the TS14 + SRS group into two separate groups, one corre sponding to TS14 and the other to SRS (Fig. 6C). This analysis pro vides the first evidence for overlapping but distinguishable gene expression profiles in TS14 and SRS patients.

RNA-seq reveals that imprinting defects lead to the genome-wide deregulation of gene expression
We evaluated differential gene expression patterns in TS14 and SRS fibroblasts by grouping our samples based on their imprinting defects: 14q32.2 hypomethylation (four TS14 patients), 11p15.5 hy pomethylation (five SRS patients), and controls (four donors). For this analysis, we selected genes with at least a twofold difference in expression between patients and controls and a q value below 0.05. We increased stringency by considering only genes with a mean expression of ≥1 FPKM (fragments per kilobase million) in at least one group. Using these criteria, we identified a total of 11,005, 11,117, and 11,043 expressed genes in control, SRS, and TS14 fi broblasts, respectively. Out of them, 552 genes were differentially expressed genes (DEGs), with expression levels differing between control fibro blasts and fibroblasts obtained from patients with TS14 or SRS (Fig. 6, D and E). Similar numbers of genes were upregulated (271 genes) and downregulated (281 genes). We found that 61 of the up regulated genes were upregulated in both TS14 and SRS fi broblasts, whereas 62 were upregulated only in SRS fibroblasts, and 148 were up regulated only in TS14 fibroblasts. By contrast, 49 of the down regulated genes were downregulated in both TS14 and SRS fibroblasts, whereas 53 were downregulated only in SRS fi broblasts, and 179 were downregulated only in TS14 fibroblasts (Fig. 6F). Together with the quantitative polymerase chain reaction (qPCR) findings in patients and transfected normal fibroblasts, these results show that imprinting defects induce the deregulation of both imprinted and nonimprinted genes. Moreover, 50% of the DEGs in SRS fibroblasts was also deregulated in TS14 fibroblasts (Fig. 6F). The list of all the DEGs in SRS and TS14 fibroblasts is available in data file S1.

A gene signature common to TS14 and SRS identified by gene set analysis
We then determined whether networks of genes with similar func tional annotations were dysregulated in TS14 and SRS patients relative to controls. We analyzed the genes displaying differential expression between these groups using gene ontology pathways from gene set enrichment analysis (GSEA) to annotate genes on the basis of the functional biological processes in which their products are involved. GSEA with a false discovery rate (q value) threshold of q < 0.05, to correct for multiple testing, identified one major path way as downregulated and two as upregulated in both TS14 and SRS (Fig. 6G:  ). We found that the level of gene expression was lower (87 genes for TS14 and 41 genes for SRS) in both TS14 and SRS fibroblasts for genes encod ing products known to enhance cell or tissue proliferation, growth, and development, whereas genes from the immune system involved in inflammatory, cytokine, and reactive oxygen species reactions were overexpressed. In all cases, TS14 fibroblasts had three to four times more DEGs from these pathways than SRS fibroblasts. Once again, 50% of the DEGs in SRS was also found among the DEGs in TS14 dysregulated pathways.

Allele-specific expression of imprinted genes from TS14 and SRS/TS14 fibroblasts
We previously performed wholeexome sequencing (WES) on DNA extracted from the fibroblasts of two TS14 patients (TS141 and TS142) (12) and the SRS/TS14 patient. We then used a method com bining RNAseq and WES of these patients to accurately assess allelespecific expression (ASE; fig. S3). On average, 3825 genes were found to be expressed biallelically versus 103 expressed monoalleli cally (3941, 3638, and 3895 versus 105, 122, and 82 from TS141, TS142, and SRS/TS14 patients, respectively). We first focused on imprinted human genes from the list of genes available from the geneimprint database (www.geneimprint.com). We compared their ASE to the bulk/singlecell fibroblast ASE recently reported for five healthy individuals (table S3) (20). Consistent with their known im printed status, we found that four imprinted genes had monoallelic expression in both patient and control fibroblasts (OSBPL5, NDN, ZDBF2, and PEG10). The H19 gene was found to be monoallelically expressed in patient TS141, whereas the PRIM2 gene, predicted to be imprinted on the basis of previous studies, also displayed mono allelic expression in both TS14 patients. However, we identified 16 other imprinted genes with biallelic expression in our patients' fi broblasts (table S3). Thirteen of these genes were also biallelically expressed in bulk fibroblast cultures from controls, suggesting that these genes may not be imprinted in fibroblasts. For the remaining three genes, CPA4 was biallelically expressed in all three patients, whereas ZNF597 and H19 (consistent with our Sanger sequencing results for cDNA) were biallelically expressed only in the SRS/TS14 patient.
Fibroblasts from the SRS/TS14 patient displaying hypomethyl ation at both 14q32.2 and 11p15.5 loci displayed biallelic expression of MEG3 and H19 from the DLK1/MEG3 and IGF2/H19 domains, respectively. We investigated whether the biallelic expression of CPA4 and ZNF597 also resulted from MLID by assessing the methylation levels of the two imprinted regions regulating these genes (the PEG1/ MEST and ZNF597DMRs, respectively). The PEG1/MEST locus was hypomethylated in all three patients, whereas ZNF597 was hy pomethylated only in the SRS/TS14 patient, consistent with the bi allelic expression profiles of these patients (table S4). These results confirm our previous finding for MLID and multiimprinting re laxation together with biallelic expression at the affected secondary loci. However, as both CPA4 and ZNF597 are expressed from the (G) Schematic represent ation of gene ontology pathways from the GSEA displaying significant up-and down-regulation in patients relative to controls fibroblasts after correction for multiple testing (q < 0.05). GSEA reveals a dysregulated pathway signature common to TS14 and SRS. methylated maternal allele, their biallelic expression upon hypo methylation of their respective domains is discordant with their parentoforigin expression. Such discordances have been reported for other imprinted genes but never before for these two genes (20). Last, we looked at the remaining genes of unknown imprinting sta tus displaying monoallelic expression (137 genes; table S4) in our patients. We identified seven putative imprinted genes clustering at four different loci across the genome (fig. S4). These variants were not validated by another targeted sequencing technique.

DISCUSSION
IDs often affect growth in humans, leading to diseases with overlap ping features, regardless of the genomic region affected (2). We characterized the effect of a loss of methylation at the imprinted DLK1/MEG3 locus on gene expression at the 14q32.2 locus in TS14 patients, and we investigated whether the overlap in clinical pheno type between TS14 and SRS and PWS might be due to similar pat terns of gene deregulation between these syndromes.
DLK1 is an important PEG in the 14q32.2 domain. This gene encodes a transmembrane protein that generates both membrane bound and serumsoluble isoforms through alternative splicing (16). We found that TS14 patients with paternal deletions or hypomethyl ation of the DLK1/MEG3 domain had no circulating DLK1, where as serum DLK1 concentration is negatively correlated with age after birth in healthy children [consistent with the rapid decrease in serum Dlk1 concentration observed in newborn mice (21)]. Thus, unlike circulating IGF2, which is produced biallelically in the liver after birth, under the control of a specific nonimprinted promoter (22), circulating DLK1 probably continues to be generated by monoallelic expression from the imprinted gene under normal con ditions after birth. An analysis of the MEGs from the 14q32.2 do main showed that MEG3, MEG8, and miRNAs were upregulated in TS14 fibroblasts. We demonstrated that MEG3 overexpression resulted from the reactivation of the normally silenced paternal allele. We hypothesized that the overexpression of MEG8 and miRNA was also due to expression from both parental alleles in TS14 pa tients. This hypothesis is consistent with the finding of previous studies that DLK1/MEG3 hypermethylation leads to a reactivation of the expression of RTL1 and DLK1 from the maternal allele (17). These findings highlight the parentoforigin-specific effects of DNA methylation at ICRs on gene expression in the DLK1/MEG3 domain.
We investigated the possible overlap in gene expression patterns between TS14 and SRS and between TS14 and PWS. We found that imprinted genes within the primary affected domains for SRS (IGF2 at 11p15.5) and PWS (SNURF and IPW at 15q11q13) were down regulated in all TS14 fibroblasts and in SRS/TS14 fibroblasts, despite the normal levels of methylation at their imprinted control regions, that are generally altered in SRS and PWS patients (6,23). We hypothe sized that this deregulation of the IGN might affect gene expression throughout the genome. An unsupervised PCA on the RNA seq data from control and TS14 and SRS patient fibroblast cultures resulted in the clustering of samples into two groups (controls on the one hand and SRS and TS14 patients on the other hand) when the data were projected onto a twodimensional space. However, the SRS and TS14 patients could be distinguished when the data were projected onto a threedimensional space. Gene ontology annotations showed that the genes displaying expression alterations belonged to common pathways. Genes encoding products that promote cell and tissue growth were found to be downregulated in fibroblasts from both TS14 and SRS patients. These results confirm our hy pothesis of an overlapping but distinguishable, genomewide dereg ulation of the pattern of expression in TS14 and SRS patients, affecting not only other imprinted genes from the IGN but also a number of nonimprinted genes. We showed that hypomethylation at the DLK1/ MEG3 domain leads to the downregulation of other imprinted PEGs, such as IGF2, SNURF, and IPW, mostly genes known to enhance growth (24), and of 87 other nonimprinted genes implicated in pro moting growth. Moreover, TS14 fibroblasts had about twice as many deregulated (1.6× up and 2.2× downregulated) genes as SRS fibro blasts, with about 50% of the deregulated genes (61 up and 49 down regulated) common to both syndromes, regardless of the imprinting defect. This finding may partly explain the phenotypic overlap and also some of the differences between SRS and TS14, such as the much early onset of puberty and greater early weight gain in TS14 than in SRS patients.
Last, we hypothesized that these changes in expression might reflect, directly or indirectly, the involvement of 14q32.2 ncRNAs MEG3 and MEG8 in this process. ncRNAs have been shown to reg ulate gene expression both in cis and in trans (25,26), through asso ciation with chromatin modifiers (27). This raises the possibility of 14q32.2 ncRNAs regulating the expression of these genes from the 11p15.5 and 15q11q13 regions. We found that overexpressing or knocking down the expression of MEG3 and MEG8 in normal fi broblast cultures was associated with deregulation of PEGs from the 11p15.5 and 15q11q13 regions. We found in our experimental models of normal fibroblasts that (i) overexpressing MEG3 was as sociated with decreased level of IPW transcripts, (ii) overexpressing MEG8 was associated with lower levels of SNURF, and (iii) concom itant overexpression of MEG3 and MEG8 was associated with lower levels of IGF2 transcripts. These results are in accordance with ob servations in TS14 and SRS fibroblasts, where (i) IPW levels were lower only in TS14 fibroblasts where MEG3 was overexpressed; (ii) SNURF levels were lower in TS14 and SRS patients, both having increased levels of MEG8; and (iii) IGF2 levels were lower in TS14 patients overexpressing MEG3 and MEG8 upon DLK1/MEG3 hy pomethylation. These results suggest that MEG3 and MEG8 can regulate, in trans, the expression of other imprinted genes. We hy pothesize a synergic role of MEG3 and MEG8 to control directly or indirectly the expression of IGF2. Overexpressing one of the two is not enough to increase IGF2 expression, since the other component is missing, while knocking down only one is enough to disrupt the complex and lead to decreased IGF2 expression. Similar patterns of regulation within the IGN have been described before (3,5), involv ing IPW and MEG3 in particular (6), for which IPW was found to repress MEG3 expression. Together with our results, this strongly suggests that a system of reciprocal control operates between ncRNAs at imprinted loci and that this system might contribute to the clini cal overlap between IDs. However, additional functional studies of the role of MEG3 and MEG8 are required to decipher the mecha nisms by which these ncRNAs could regulate gene expression, par ticularly for imprinted genes. These mechanisms are probably unrelated to gene imprinting status because all TS14 patients have normal methylation levels at the ICRs corresponding to the deregu lated imprinted genes (ICR1 for IGF2 and SNRPNIC for SNURF and IPW). Singlecell RNAseq has been shown to be more power ful than bulk RNAseq for studies of the expression of imprinted genes (20) and therefore would allow the generation of a more ex tensive map of the deregulated IGN upon hypomethylation of the DLK1/MEG3 and/or IGF2/H19 domains.
In conclusion, we show that 14q32.2 hypomethylation affects the expression not only of genes within this locus but also of other imprinted and nonimprinted genes, some of which are involved in controlling tissue growth. We also observed changes in gene expres sion common to TS14 and SRS patients, in particular, a diminished IGF2 expression, and to TS14 and PWS patients, with diminished SNURF and IPW expressions (fig. S5). This transcriptomic overlap may account for the clinical overlap between the two syndromes.

Study approval
Patients were followed up at Armand Trousseau Children's Hospital or were referred by other clinical centers for molecular analysis for sus pected TS14 or SRS. A geneticist and/or a pediatric endocrinologist examined each patient, and a detailed clinical form was completed. All patients or their parents and the controls included in this study gave written informed consent for participation, in accordance with French national ethics rules for patients recruited in France (Assis tance Publique-Hôpitaux de Paris authorization no. 682) and the rules of Institutional Review Board I00000204 of the Mount Sinai School of Medicine for patients recruited in the United States.

Quantification of serum DLK1 levels
Serum DLK1 levels from healthy children and TS14 patients were quantified in 96well plates, with the Quantikine Immunoassay Con trol for Human Pref1/DLK1/FA1 (R&D Systems, France), based on sandwich ELISA (enzymelinked immunosorbent assay), according to the manufacturer's instructions.

Skin-derived fibroblast cultures
Skinderived fibroblasts from patients and controls were cultured to confluence in RPMI 1640 (Gibco, Cergy Pontoise, France) supplemented with 10% fetal calf serum and ampicillin (50 U/ml)/ streptomycin (50 g/ml) at 37°C (number of passages, <10). Cells were then treated with trypsin and centrifuged. The cell pellet ob tained was washed twice with 1× phosphatebuffered saline and used for DNA and micro/largescale RNA extractions.
Nucleic acid extraction and quantification DNA was extracted from blood leukocytes and from skinderived fi broblast cultures, by an inhouse protocol, after cell lysis by a salting out procedure, as previously described (28,29). Total RNA and miRNA were extracted from cultured fibroblasts with the NucleoSpin miRNA Kit for the isolation of small and large RNAs (MACHEREY NAGEL, France). Both DNA and RNA were quantified with a NanoDrop ND1000 spectrophotometer (Invitrogen, France).

Reverse transcription and real-time PCR quantification of micro/mRNAs
We synthesized cDNA from micro/mRNAs isolated from fibro blasts and used it for quantitative PCR with the miScript PCR Sys tem (Qiagen, France). Expression in the controls was arbitrarily set to 1, and fold changes (FCs) between two groups were calculated as FC = 2 −Ct . Details from the reverse transcription and realtime quantification analysis are provided in Supplementary Methods. Bisulfite treatment of DNA and methylation analysis gDNA (1 g) was treated with sodium bisulfite, with the EZ DNA Methylation Kit (Zymo Research, Orange, CA), according to the manu facturer's instructions. The methylation status of the studied loci was assessed by allelespecific methylated multiplex realtime quantitative PCR (ASMM RTQPCR), as previously described (30)(31)(32)(33)(34)(35)(36). Details for the treatment and technique are provided in Supplementary Methods. gDNA and cDNA sequencing Two SNPs from the MEG3 and H19 gDNA and cDNA were sequenced by standard Sanger sequencing methods, with the ABI PRISM BigDye Terminator v3.0 Cycle Sequencing Kit and the ABI 3100 Genetic Analyzer. The sequencing products were then analyzed with SeqScape v2.6 (Life Technologies, Courtaboeuf, France).
Cell cultures and transfection assays GM05757 (control human skin-derived fibroblast) cells were cul tured in 24well plates for overexpression and small interfering RNA (siRNA) transfections. Details of culture and transfection condi tions and constructs used are detailed in Supplementary Methods.

WES, RNA-seq, and bioinformatics analysis
Library preparations, sequencing, and data analysis were carried out by IntegraGen SA (Evry, France). The sequencing methods and bio informatics analysis are described in detail in Supplementary Methods.

RT-qPCR statistical analysis
For RTqPCR, we compared data for pairs of groups in MannWhitney tests (expression in fibroblasts from patients versus that in control fibroblasts, with n = 5 each) and unpaired t tests (overexpression and siRNA assays with n > 20 per condition). We considered P val ues below 0.05 to indicate statistical significance. The analyses were performed with GraphPad Prism version 6.00 (GraphPad Software, La Jolla, CA, USA).

SUPPLEMENTARY MATERIALS
Supplementary material for this article is available at http://advances.sciencemag.org/cgi/ content/full/5/2/eaau9425/DC1 Supplementary Methods Fig. S1. Expression of 14q32.2 MEGs and 11p15.5 and 15q11-q13 PEGs in five different passages of cultured fibroblasts from TS14 patients. Fig. S2. Schematic representation of the subcloned transcripts of MEG8 (MEG8a and MEG8b). Fig. S3. Distribution of minor allele frequency according to combined exome and RNA-seq data for fibroblasts from the TS14-1, TS14-2, and SRS/TS14 patients. Fig. S4. Schematic representation of the four putative imprinted loci identified on the basis of allele-specific gene expression and DMRs. Fig. S5. Schematic representation of the molecular findings and the hypothesized mechanism from this study. Table S1. Methylation levels for all patients and controls, as determined by ASMM RT-qPCR. Table S2. Clinical features for TS14 patients with imprinting defects at the DLK1/MEG3 domain described here. Table S3. The allelic status of imprinted genes, as determined by RNA and exome sequencing for TS14 and control fibroblasts. Table S4. List of all genes with monoallelic expression, as determined by RNA and exome sequencing in TS14 patients. Table S5. List and sequences of the primers used in this study. Data file S1. List of all gene FPKMs for each patient and control. Data file S2. Supervised study of DEGs for each group of patients compared to controls. References (37)(38)(39)(40)