Biallelic non-productive enhancer-promoter interactions precede imprinted expression of Kcnk9 during mouse neural commitment

Summary It is only partially understood how constitutive allelic methylation at imprinting control regions (ICRs) interacts with other regulation levels to drive timely parental allele-specific expression along large imprinted domains. The Peg13-Kcnk9 domain is an imprinted domain with important brain functions. To gain insights into its regulation during neural commitment, we performed an integrative analysis of its allele-specific epigenetic, transcriptomic, and cis-spatial organization using a mouse stem cell-based corticogenesis model that recapitulates the control of imprinted gene expression during neurodevelopment. We found that, despite an allelic higher-order chromatin structure associated with the paternally CTCF-bound Peg13 ICR, enhancer-Kcnk9 promoter contacts occurred on both alleles, although they were productive only on the maternal allele. This observation challenges the canonical model in which CTCF binding isolates the enhancer and its target gene on either side and suggests a more nuanced role for allelic CTCF binding at some ICRs.


Introduction
The functional specialization of each cell and tissue type, which is crucial in multicellular organisms, is based on their capacity to respond to developmental and environmental cues by generating specific gene expression profiles.The general principles governing this process have been identified.Key regulatory DNA sequences, sequence-specific transcription factors, epigenetic modifications, and the spatial organization of the genome interact to regulate gene expression levels, 1 raising the question of how their coordinated action is orchestrated.In mammals, this question is particularly important for imprinted genes that are expressed in a parent-of-origin-specific manner.
Genomic imprinting is a key developmental process whereby some mammalian genes are expressed by only one allele, depending on their parental origin.Most of the 200 imprinted genes identified to date are involved in crucial biological processes, such as cell proliferation, fetal and placental growth, energy homeostasis, and metabolic adaptation. 2 Genomic imprinting also plays a central role in brain function and behavior, and many imprinted genes are expressed only in neural lineages. 3Consequently, misregulation of imprinted genes is causally implicated in severe neurobehavioral disorders such as Prader-Willi and Angelman syndromes.
In humans and mice, most imprinted genes are organized in evolutionarily conserved genomic clusters that contain two or more maternally and paternally expressed genes in large regions (up to several megabases in size).In each cluster, allele-specific expression is primarily regulated by DNA methylation at discrete cis-acting regulatory elements known as imprinting control regions (ICRs).Each ICR overlaps with a differentially methylated region (DMR) that harbors allelic DNA methylation inherited from the male or female gamete and subsequently maintained throughout development (i.e., germline DMR).The resulting constitutive allelic DNA methylation at ICRs is critical for orchestrating the allele-specific expression along the imprinted domain by influencing a combination of regulatory mechanisms, some of which are tissue specific, leading to the complex and specific spatiotemporal expression pattern of imprinted genes. 2 Specifically, histone modifications, cis-spatial organization, and tissue-specific regulatory regions have all been documented to contribute, along with DNA methylation, to this longrange, ICR-mediated, tissue-specific regulation of imprinted domains. 4,51][12] In addition, at the maternally methylated ICRs, all of which are also promoters, timely developmental loss or gain of the repressive H3K27me3 mark on the paternal allele contributes to the appropriate tissue-specific paternal expression. 13Allelic methylation at ICRs may also influence long-range chromatin interactions between enhancer and promoters along imprinted domains.Such interactions between regulatory elements and their target genes are facilitated by sub-chromosomal structures called topological associated domains (TADs). 14A study on the H19-Igf2 and Dlk1-Gtl2 domains, which are controlled by a paternally methylated ICR, showed that binding of the methyl-sensitive and boundary protein CTCF to the unmethylated allele of the ICR induces an allele-specific sub-TAD organization that might facilitate the establishment and maintenance of the imprinted transcriptional program. 157][18] A recent study showed that allele-specific CTCF binding at a post-implantation DMR (secondary DMR) structures the Grb10-Ddc locus to direct proper enhancer-promoter interactions in the developing heart.This further illustrates the interplay between DNA methylation and CTCF binding to control instructive allelic chromatin configurations at imprinted loci. 19The observation that CTCF binds to the ICRs of various imprinted loci 20 suggests that the allelic chromatin structure may be a commonly used strategy whereby ICRs direct mono-allelic expression along large genomic imprinted domains.However, for most imprinted clusters, this hypothesis has not been formally evaluated.
Altogether, these data highlight that deciphering how constitutive allelic DNA methylation at ICRs can direct tissue-and stage-specific allele-specific expression along imprinted domains requires the simultaneous analysis of the dynamics of multiple layers of regulation during cell identity acquisition.Here, to gain insights into the regulation of the Peg13-Kcnk9 domain during neural commitment, we precisely monitored the allele-specific epigenetic, transcriptomic, and cis-spatial organization using a mouse stem cellbased corticogenesis model that recapitulates the in vivo epigenetic control of imprinted gene expression. 21The Peg13-Kcnk9 domain is evolutionarily conserved with important functions in the brain.It contains five genes, two of which are imprinted in both humans and mice: the paternally expressed non-coding RNA PEG13 and the potassium channel gene KCNK9, which is maternally expressed specifically in the brain.2][23][24][25][26][27] Mutations in these genes are associated with neurodevelopmental and neurological disorders, [28][29][30][31] including the Birk-Barel intellectual disability syndrome, which is caused by maternally inherited KCNK9 mutations. 32,33ittle is known about the mechanisms that control expression along this domain.3][24] A study using human brain tissues suggests that this DMR controls KCNK9 and PEG13 imprinted expression through a CTCF-mediated enhancer-blocking activity. 24However, this model has not been experimentally validated and it is not known whether it explains the imprinted gene expression kinetics along the domain during neural identity acquisition.The integrative analysis of multiple levels of regulation described in this study provides a comprehensive view of the molecular events that take place during the establishment of maternal Kcnk9 expression in neural commitment.Our main observation challenges the canonical model of CTCF-mediated enhancer-blocking activity and suggests a more nuanced role for allelic CTCF binding at the ICR of this locus.

Material and methods
Details of key reagents and resources are given in Table S1.

Cell culture and embryonic stem cell differentiation
The hybrid male embryonic stem cell (ESC) lines were previously derived from blastocysts obtained from crosses between C57BL/J (B) and JF1 (J) mice 34 and were maintained in gelatin-coated dishes with ESGRO complete plus medium (Millipore, SF001-500P) containing LIF (leukemia inhibitory factor), BMP4 (bone morphogenetic protein 4), and a GSK3-b (glycogen synthase kinase 3b) inhibitor.In vitro corticogenesis was performed as previously described, 35 except that ESCs were plated on Matrigel-coated dishes (human ESC-qualified matrix, Corning), and that the defined default medium was supplemented with B27 (without vitamin A, Gibco) to improve cell survival and with 1 mM dorsomorphin homolog 1 (purified by C.C.H.) to promote neurogenesis. 36Using this protocol, neural precursor cells (NPCs) are the main cell population after 12 days (D12) of in vitro corticogenesis. 35

DNA methylation analysis
DNA extraction and bisulfite sequencing DNA was extracted as previously described. 38Bisulfite conversion was performed with the EZ DNA Methylation-Gold Kit from Zymo (ref.D5006), according to the manufacturer's instructions.PCR amplification, cloning, and sequencing were performed as previously described. 38Details of the primers used are in Table S2.DNA methylation data mining ESC and NPC whole-genome bisulfite sequencing (WGBS) data were obtained from GEO DataSets under the accession numbers GEO: GSM748786 and GEO: GSM748788, respectively.The reads were first processed using TrimGalore and then mapped to the mm39 mouse genome using Bismark.Duplicate reads were removed with the script deduplicate_bismark.CpG methylation levels were computed from the selected alignments using bis-mark_methylation_extractor (-no_header -cutoff 4 -bedgraph) and coverage2cytosine scripts.The output was converted with the bedGraphToBigWig tools to be loaded on the University of California, Santa Cruz (UCSC) Genome Browser.CpG methylation levels of ESCs, NPCs, and frontal cortex in the mm10 genome were obtained from the tracks Stadler 2011 and Lister 2013 of the DNA methylation Hub on UCSC.

Genome production for next-generation sequencing data alignment of hybrid samples
The sequences of the JF1 strain was obtained from the DDBJ database under the accession numbers DDBJ: DRP000326 and DDBJ: DRP000984.Paired-end reads were filtered using the CutAdapt tool to exclude poor quality reads (-minimum-length 101 -pairfilter any -q 20).The remaining reads were mapped to the mm39 genome with bowtie2.Alignments were filtered for poor quality with samtools view (-q 20).Duplicate alignments were excluded using samtools fixmate and markdup (-r).To identify JF1 polymorphisms, the filtered alignments were analyzed using the freebayes tool (-m 20 -q 30 -C 10 -F 0.75) and the output was normalized using bcftools norm.Then, variants were decomposed using the vcflib vcfallelicprimitives (-kg) tool.The resulting vcf file was then processed with the mm39 genome using a custom R script to generate the genomes used for the alignments (library: Biostrings, GenomicRanges).An mm39 genome masked by N at JF1 single-nucleotide polymorphism (SNP) positions was generated.The JF1 genome was reconstructed by converting JF1 SNPs, insertions, and deletions in the mm39 genome.A diploid hybrid genome that consisted, for each chromosome, of the C57BL/6 and JF1 sequences was generated.The hybrid genome has the advantage of taking the JF1 indels into account when determining allelic alignments, but it results in different genomic coordinates for the same element on the C57BL/6 and JF1 genomes.To work with only one reference, a custom R script was written to convert the coordinates of the JF1 alignments into the reference mm39 genome (GenomicRanges).

RNA extraction
RNA was isolated from frozen cell pellets using TRIzol Reagent (Life Technologies, 15596018), according to the manufacturer's recommendations.

RT-qPCR
After treatment with RNase-free DNase I (Life Technologies, 180868-015), first-strand cDNA was generated by reverse transcription with Superscript-IV (Life Technologies, 18090050) using random primers and 500 ng of RNA.Then, cDNA was amplified by real-time PCR with the SYBR Green mixture (Roche) using a LightCycler R 480II (Roche) apparatus.The relative expression level was quantified with the 2-delta Ct method that gives the fold change variation in gene expression normalized to the geometrical mean of the expression of the housekeeping genes Gapdh, Tbp, and Gus.The primer sequences are in Table S2.
For allelic analysis, for each locus of interest, the parental allele origin of expression was assigned following direct sequencing of the cognate RT-PCR product that encompassed a strain-specific SNP (SNP details in Table S2).

Microfluidic-based quantitative analysis
This analysis was performed using a commercial panel of total RNA (mouse total RNA master panel; Ozyme 636644) obtained from pooled samples isolated from several hundred mouse embryos and adults.Following reverse transcription, as described above, first-strand cDNA was pre-amplified for 14 cycles with the pool of primers used for the RT-qPCR analysis and the Taq-Man PreAmplification Master Mix (Life Technologies, 4488593).RT-qPCR was then performed and validated on Fluidigm 96.96 Dynamic Arrays using the Biomark HD system (Fluidigm) according to the manufacturer's instructions.The relative gene expression was quantified using the 2-delta Ct method, which gives the fold changes in gene expression normalized to the geometrical mean of the expression of the housekeeping genes Arbp, Gapdh, and Tbp.For each condition, the presented data were obtained from two independent experiments, each analyzed in duplicate.

RNA sequencing
Paired-end RNA sequencing (RNA-seq) data were generated using ESCs and NPCs in duplicate for the B6xJF1 and JF1xB6 genetic backgrounds.RNA-seq libraries were prepared with the Illumina TruSeq Stranded mRNA Kit or the NEBNext Ultra II mRNA-Seq Kit and sequenced on an HiSeq4000 or NovaSeq6000 apparatus by IntegraGen according to the manufacturer's protocol.To determine the global and allelic expression, RNA-seq reads were mapped on the mm39 masked genome and hybrid genome, respectively, using TopHat2 and a gene annotation file adapted for these genomes based on the UCSC refGene track (-r 350 -mate-std-dev 250 -library-type frfirststrand).Alignments were filtered with samtools for mapping quality and reads mapped in proper pairs (view -f 2 -q 20).This step, on the hybrid alignments, allows obtaining allele-specific mapping.The strand-specific coverages of the RNA-seq data were generated using the C57BL/6 and JF1 specific alignments and the global alignments with bamCoverage (-normalizeUsing RPKM -filterRNAstrand forward/reverse) and visualized on the UCSC genome browser.Replicates were overlaid for allelic and strand-specific coverage using the track collection builder tool for genome exploration.

Gene expression data mining
Expression data of cortex from E13.5 B6xJF1 and JF1xB6 embryos were obtained from the GEO: GSE58523 dataset.The RNA-seq treatment was based on the pipeline described above adapted for single-end RNA-seq.

Chromatin immunoprecipitation
ChIP-qPCR Chromatin immunoprecipitation (ChIP) of native chromatin was performed as described by Brind'Amour et al. 39 using 500,000 cells Human Genetics and Genomics Advances 5, 100271, April 11, 2024 3 per immunoprecipitation.Results presented in this article were obtained from at least three ChIP assays performed using independent chromatin preparations, as indicated in the figure legends.Details of the antisera used can be found in Table S3.Quantitative and allelic analyses were performed as described previously in Maupetit-Me ´houas et al. 13 Details of the SNPs and primers used can be found in Table S2.

ChIP sequencing
ChIP sequencing (ChIP-seq) experiments were performed using native chromatin from ESCs and NPCs (for each cell type: n ¼ 1 in the B/J and n ¼ 1 in the J/B background, respectively), as previously described (Le Boiteux et al. 40 ).Details of the used antisera are in Table S3.Background precipitation levels were determined by performing mock precipitations with a nonspecific immunoglobulin (Ig) G antiserum (Sigma-Aldrich C2288), and experiments were validated by qPCR on diagnostic regions before sequencing.Library preparation (TruSeq ChIP Sample Preparation) and sequencing on a HiSeq 2500 instrument (Illumina) were performed by MGX (Montpellier GenomiX), according to the manufacturer's recommendations (mean of 40 million single reads per sample).To determine the global and allelic alignments, ChIP-seq reads were mapped using Bowtie2 to the mm39 masked genome and hybrid genome, respectively.Alignments filtering was done with samtools (view -q 20), peaks were called using MACS1.4.2 (-nomodel -shiftsize 73 -pvalue 1eÀ5), and the coverage was computed with bamCoverage (-normalizeUsing RPKM -extendReads 200 -ignoreDuplicates -binSize 20).The UCSC track collection builder tool was used to overlay allelic coverages for genome exploration.Cut&Run Cut&Run (C&R) was performed using the CUTANA CUT&RUN Kit (Epicypher) and non-fixed nuclei from ESCs and NPCs (for each cell type: n ¼ 1 in the B/J and n ¼ 1 in the J/B background, respectively), according to the manufacturer's instructions.The antisera used are listed in Table S3.Briefly, nuclei were isolated from fresh ESCs or NPCs and stored in nuclear extraction buffer at À80 C.After thawing, 500,000 nuclei per reaction were aliquoted and incubated with pre-activated concanavalin A-coated beads at room temperature for 10 min, followed by overnight incubation with 0.5 mg of antibody in buffer containing 0.01% digitonin at 4 C.Then, nuclei bound to concanavalin A-coated beads were permeabilized with a buffer containing 0.01% digitonin and incubated with the pAG-MNase fusion protein at room temperature for 10 min.After washing, chromatin-bound pAG-MNase cleavage was induced by addition of calcium chloride to a final concentration of 2 mM.After incubation at 4 C for 2 h, the reaction was stopped by addition of stop buffer (containing fragmented genomic Escherichia coli DNA as spike-in).Following fragmented DNA purification, Illumina sequencing libraries were prepared from $5 ng of purified DNA using the CUTANA CUT&RUN Library Prep Kit (EpiCypher 14-1001 and 14-1002) according to the manufacturer's recommendations.Purified multiplex libraries were diluted to 9 nM concentration (calculated with the Qubit dsDNA HS Assay Kit) and sequenced on a NovaSeq 6000 instrument (Illumina) by IntegraGen SA.Paired-end reads were mapped to the mm39 masked genome and hybrid genome using Bowtie2.Alignment filtering was done with samtools (view -f 2 -q 20), and the coverage was obtained with bamCoverage (global coverage: -scaleFactor ''spike-in DNA'' -normalizeUsing RPKM -binSize 25; allelic coverage: -binSize 25).The UCSC track collection builder tool was used to overlay allelic coverages for genome exploration.Peaks were called with MACS2 using the Cut&Run control sample (IgG) (callpeak -f BAMPE -keep-dup all).
ChIP-seq and ATAC-seq data mining Allelic and global mm9 alignments for acetylation of lysine 27 on histone H3 (H3K27ac) in mouse frontal cortex were obtained from the GEO: GSM751461, GSM751462 datasets.These alignments were converted to coverages using an R script (rtracklayer, GenomicRanges) and were visualized on UCSC.
Assay for transposase-accessible chromatin using sequencing (ATAC-seq) data for ESCs and ESC-derived NPCs were obtained from the GEO: GSE155215 DataSet.Paired-end reads were treated with trim_galore (-paired) and then were aligned to the mm39 genome using bowtie2 (-very-sensitive -X 1000).Only properly paired alignments were conserved with samtools (view -f 2) and alignments to mitochondrial sequences and random chromosomes were excluded.PCR duplicates were removed using picard-tools (MarkDuplicates -REMOVE_DUPLICATES true).The coverage was assessed using bamCoverage (-normalizeUsing RPKM -binSize 20) and was visualized on the UCSC genome browser.Peaks were called using macs2 (callpeak -f BAMPE -broad -broad-cutoff 0.05 -keepdup all).

Circular chromosome conformation capture followed by sequencing
Circular chromosome conformation capture followed by sequencing (4C-seq) experiments were done using ESCs and NPCs from the BxJ genetic background for all viewpoints (for each cell type, Peg13 DMR n ¼ 2; Kcnk9 promoter n ¼ 1; putative enhancer [PE] n ¼ 1) and in ESCs from the JxB genetic background for the Peg13 DMR viewpoint (n ¼ 1).The primers used for each viewpoint can be found in Table S1.4C template preparation was carried out as previously described 41 with some modifications.Briefly, 1 3 10 7 cell suspensions were cross-linked with formaldehyde (final concentration 2%) for 10 min.After cell lysis and permeabilization with SDS and Triton X-100, samples were digested with 600 U of DpnII at 37 C in 13 NEBuffer DpnII (4 h with 200 U, overnight with 200 U, and 4 h with 200 U).The restriction enzyme was inactivated with SDS and Triton X-100.The first ligation was performed in a large volume, 7.2 mL of 13 ligase buffer, and with 50 U of T4 DNA ligase, at 18 C overnight.Cross-linking was reversed with 600 mg of proteinase K at 65 C overnight.After phenol/chloroform purification, DNA was digested with 50 U of NlaIII at 37 C overnight.After phenol/chloroform purification, a second ligation was performed in 14 mL of 13 ligase buffer and with 100 U of T4 DNA ligase at 18 C overnight.The 4C template was concentrated by ethanol precipitation and then purified using the DNA Clean & Concentrator Zymo-25 Kit.To produce a 4C-seq library, 3.2 mg of 4C template was amplified in 16 PCR cycles with viewpoint-specific sequencing primers and 56 U of Expand long template polymerase.PCR reactions were pooled and purified using the High Pure PCR Product Purification Kit.The 4C-seq libraries of the different viewpoints were combined before sequencing on an HiSeq 4000 or NovaSeq 6000 instrument (Illumina) by IntegraGen.Due to the primer design, paired-end reads were used to determine the viewpoint allele and the interacting sequence.To do this, only the expected sequence, corresponding to the viewpoint of the informative reads, was mapped to the mm39 hybrid genome using bowtie2 (-trim5 10 -trim3 ''viewpoint specific'' -local -very-sensitive-local).Only alignments mapped to the viewpoint coordinates and with a minimal quality were conserved to determine the allelic origin of the viewpoint in the reads (samtools view -q 10 [ viewpoint coordinates]).To determine the sequence in interaction, the expected sequence with the DpnII site was mapped to the mm39 masked genome using bowtie2 (-trim5 ''viewpoint specific'' -trim3 ''viewpoint specific'' -local -very-sensitive-local).These alignments were filtered for mapping quality (samtools view -q 10) and were split according to the allelic origin of the viewpoint.To construct the allelic interactome, alignments were processed with the FourCSeq Bioconductor package to count the reads mapped exactly to the end of a DpnII fragment and to generate a smoothed rpm normalized coverage.The UCSC track collection builder tool was used to overlay allelic interactome coverages for genome exploration.Based on these read counts, the 4C-ker package 42 was used to identify paternal and maternal interactions (nearBaitAnalysis; k ¼ 8) in replicate experiments.The differential analysis of maternal and paternal interactions was performed using the function ''differentialAnalysis'' of the 4C-ker package.This function was adapted to handle paired samples and differences were considered significant when the adjusted p value was <0.05.

Statistical analyses
Statistical analyses were performed using GraphPad.The statistical test used for each comparison and the number of independent experiments are indicated in the figure legends.

Results
Kcnk9 gains maternal expression upon neural commitment Using a microfluidic RT-qPCR approach, we observed, in agreement with allelome studies, 25,27 that Peg13 was expressed in a wide range of adult mouse tissues and at different development stages, but particularly in brain tissues.Conversely, Kcnk9 expression was restricted to brain tissues (Figure S1A).Expression analyses in brain tissues from newborn F1 hybrids obtained by crossing C57BL/6J and M. musculus molossinus (JF1) mice confirmed that Peg13 was paternally expressed and Kcnk9 maternally expressed (Figure 1A).To investigate the mechanisms responsible for this brain-specific imprinted expression, we adapted a stem cell-based corticogenesis model that we have previously shown to recapitulate the in vivo epigenetic control of imprinted gene expression 21 to mouse ESC lines we derived from reciprocal crosses of C57BL/6 (B6) and JF1 mice (hereafter, B/J and J/B).Reciprocal crosses allow investigating the parental allele origin using informative SNPs.
We focused on the first 12 days of in vitro corticogenesis.During this period, ESCs predominantly differentiated into NPCs, as indicated by the marked downregulation of the pluripotency marker Pou5f1 and the upregulation of the neural precursor markers Nestin and Pax6 (Figure S1B).RNA-seq and RT-qPCR approaches, performed using B/Jand J/B-derived ESCs, showed that, in ESCs, imprinted expression was restricted to Peg13, which showed weak paternal expression.Upon differentiation to NPCs, paternal Peg13 expression and maternal Kcnk9 expression increased (Figures 1B, 1C, S1C, and S1D).During this time window, the other three genes of the domain, Trapc9, Ago2, and Chrac1, were biallelically expressed (Figure S1E).A similar expression pattern was observed in primary neural stem cells (neurospheres) from newborn mice. 44oreover, re-analysis of RNA-seq data from dorsal telencephalon samples at E13.5 21 demonstrated that, in our corticogenesis model, the imprinted expression pattern at the Peg13 domain in NPCs, restricted to Kcnk9 and Peg13, recapitulated the pattern observed in embryonic brain in vivo (Figures 1B and S1D).Therefore, our ESC-based corticogenesis model to generate NPCs provides a relevant framework to uncover the mechanisms acting at the Peg13 domain, and particularly those involved in the imprinted expression of Peg13 and Knck9, in neural stem cells and during early brain development.
Maternal DNA methylation at the Peg13 DMR is required for Kcnk9 maternal expression The Peg13 DMR is the putative ICR proposed to control the imprinted expression of the entire locus.To more formally evaluate the role of allelic DNA methylation in Kcnk9 expression regulation, we assessed expression of Peg13 and Kcnk9 in brain tissue of E9.5 Dnmt3l À/þ embryos, derived from Dnmt3l À/À females in which DNA methylation imprints at ICRs are not established during oogenesis. 37,45In wild-type embryos, we confirmed the paternal and maternal expression of Peg13 and Kcnk9, respectively (Figure 2).In mutant embryos, the lack of maternal DNA methylation at the Peg13 DMR (Figure S2) resulted in increased and biallelic expression of Peg13, while Kcnk9 expression was lost (Figure 2).This supports the hypothesis that the Peg13 DMR is the ICR of the locus and indicates that its maternal DNA methylation is required for the maternal expression of Kcnk9.
Changes in imprinted expression upon neural commitment are not associated with changes in epigenetic signatures at the Peg13 DMR We then performed an integrative analysis based on allelic ChIP-seq, C&R (both performed with samples from the two reciprocal crosses), and ChIP-qPCR coupled with mining of data obtained using non-allelic ATAC-seq 46 and WGBS 47 to determine whether epigenetic signature changes at the Knck9 and Peg13 promoters could explain their expression change upon ESC differentiation into NPCs.
In ESCs, the Peg13 promoter showed the characteristic feature of an ICR 48 : DNA methylation, the repressive histone mark H3K9me3 and the zinc finger protein ZFP57 associated with its maternal allele, and the permissive histone marks di-methylation and tri-methylation of lysine 4 of histone H3 (H3K4me2 and H3K4me3) and H3K27ac associated with its paternal allele (Figures 3,  S3A, and S3B).This allelic signature was maintained in NPCs, where the permissive histone marks were more widely distributed along the gene on the paternal allele and also in neonatal mouse brain, without major changes despite Peg13 upregulation (Figures 1, 3, and S3B).
The Kcnk9 promoter is in a CpG island that remained unmethylated in both ESCs and NPCs and also in embryonic and adult brain tissues.This indicated that Kcnk9 expression is not controlled by methylation dynamics at its promoter (Figures 3A and S3C).Unlike Peg13, we did not observe any allelic signature at the Kcnk9 promoter in ESCs.A broad biallelic H3K27me3 deposition marked the gene body.This repressive mark was associated with the permissive H3K4me2 mark on both alleles of the promoter, forming a bivalent signature that might poise gene expression. 49In NPCs, H3K27me3 was lost from the gene body, while the promoter retained the bivalent signature, albeit with lower H3K27me3 levels (Figures 3A, 3B, and S3B).ChIP-qPCR, performed using B/J material, also showed that the slight increase of acetylation of lysine 9 on histone H3 (H3K9ac) and H3K27ac, marks associated with active transcription, occurred preferentially on the maternal allele of Kcnk9 (Figure 3B).This trend was further enhanced in the neonatal brain samples, where H3K9ac, H3K27ac, and H3K4me3 marked the maternal allele of the Kcnk9 promoter, while the biallelic bivalent signature H3K4me2/H3K27me3 was maintained (Figure 3B).
These observations suggest that upregulation of paternal Peg13 expression and gain of maternal Kcnk9 expression upon neural commitment are not driven by changes in the Peg13 DMR/putative ICR epigenetic signature.Specifically, for Kcnk9, maternal expression was induced despite the presence of H3K27me3 at the promoter and was accompanied by biallelic loss of H3K27me3 in the gene body and gain of permissive/activating marks, mainly acetylation, on the maternal promoter reflecting transcriptional activity.
Biallelic interactions between the Kcnk9 promoter and its putative regulatory region in ESCs precede the maternally biased interaction in NPCs Besides epigenetic modifications, the higher-order chromatin structure through chromatin looping is another layer of regulation that controls gene expression along imprinted clusters and facilitates enhancer-promoter interactions within TADs.The DNA-binding protein CTCF is a key determinant in the formation of these loops and is frequently found at their base.
Allelic C&R analyses showed that, in both ESCs and NPCs, CTCF bound tightly to the Peg13 DMR/putative ICR in a paternal-specific manner, whereas it bound to both alleles of the Kcnk9 promoter and the 3 0 edge of its unique intron (Figures 4 and S4).To determine whether these regions form chromatin loops and to identify putative distant regulatory regions, we performed allelic 4C-seq using the Peg13 DMR and the Kcnk9 promoter as viewpoints.While the Peg13 domain is all contained within a larger TAD (as defined in cortex by Dixon et al. 14 ), the signal obtained for the Peg13 DMR in ESCs was largely restricted to the imprinted domain, from the downstream Kcnk9 to the upstream Ago2 gene (Figure S4).Strikingly, in the two reciprocal crosses, contacts were exclusively mediated by the paternal unmethylated Peg13 DMR, highlighting that paternal CTCF binding promoted higher-order chromatin structure differences between the parental alleles (Figures 4 and S4 and the next section).Although we observed paternal-specific contacts along the entire imprinted domain, we detected significantly stronger paternal signals at Kcnk9, centered on the CTCF-bound promoter and the 3 0 edge of the intron, and at a biallelic CTCF-bound region in the 5 0 part of Trappc9 (Figures 4 and S4).This second signal peaked in a Trappc9 intron previously identified as a putative regulatory region that controls the tissue-specific expression of the domain. 44These paternal-specific contacts were mainly maintained also in NPCs where the interaction with the Trappc9 intronic putative regulatory region was strengthened (Figures 4 and S4).
The same analysis using the Kcnk9 promoter as viewpoint confirmed that this promoter interacted with the Peg13 DMR only on the paternal allele in ESCs and also in NPCs, although more weakly.In addition, the Kcnk9 promoter interacted with the intronic putative regulatory region in Trappc9, from both alleles in ESCs and with a bias from the maternal allele in NPCs (Figure 4).
These results identified a putative regulatory region (the PE), that interacts with the paternal Peg13 promoter in Human Genetics and Genomics Advances 5, 100271, April 11, 2024 7 ESCs and NPCs, and preferentially, but not exclusively, with the maternal Kcnk9 promoter in NPCs.Therefore, it is a candidate for regulating the imprinted expression of both genes during neural commitment.However, contrary to expectation, contacts with the Kcnk9 promoter were already established in ESCs and from both alleles.This suggests that non-productive biallelic contacts between the Kcnk9 promoter and the putative regulatory region in ESCs precede the maternally biased productive contacts in NPCs.
Contacts between the PE and Peg13 DMR structure the higher-order chromatin conformation in the Peg13 domain To investigate the extent to which the intronic PE influences the chromatin conformation along the Peg13 domain, we performed allelic 4C-seq using this region as a viewpoint and visualized these data together with allelic 4C-seq data for the Peg13 DMR and Kcnk9 promoter.We also re-analyzed high-resolution but non-allelic Hi-C data from ESCs and NPCs in vivo. 50Hi-C data revealed that the Peg13 imprinted domain resides in two sub-TADs that are conserved in ESCs and NPCs (Figure 5).The centromeric sub-TAD was anchored to CTCF-bound regions in the 5 0 part of Trappc9 and in Kcnk9, presumably in the intron (the 5-kb resolution of the Hi-C data did not allow precisely mapping the boundary regions), thus isolating Kcnk9 and Peg13 from Chrac1 and Ago2, which are in the telomeric sub-TAD.The Trappc9 promoter was at the boundary between sub-TADs (Figures 5 and S5).Furthermore, in line with the 4C-seq data, paternal CTCF binding at the Peg13 DMR subdivided the telomeric sub-TAD into two sub-domains, presumably on the paternal allele only, in a structure maintained in ESCs and NPCs (Figure 5).Notably, the CTCF binding sites identified by the Jaspar database 51 in the Peg13 DMR were all in the opposite orientation.This suggests that, unlike the majority of loops, which are anchored to pairs of convergent CTCF sites, 52 the loop between the Peg13 DMR and PE was anchored to a pair of CTCF sites in the same orientation (Figure S5).
Interestingly, in ESCs, this higher-order chromatin structure could not completely isolate sub-TADs or domains from each other.Indeed, the 4C-seq signal obtained for the PE region was mainly, but not entirely, restricted to the centromeric sub-TAD.It was also, albeit to a lesser extent, observed in the telomeric region, including at the Ago2 promoter (Figure 5).In addition, PE strongly contacted the Kcnk9 promoter from both alleles, despite the Peg13 DMR-associated sub-domains on the paternal allele (Figures 5 and S5).
In NPCs, PE contacts were restricted to the centromeric sub-TAD.This coincided with a stronger interaction at the sub-TAD boundary (arrow b in Figure S5) that may enhance its insulating capacity (Figures 5 and S5).Along this centromeric sub-TAD, the pattern observed in ESCs remained largely stable also in NPCs, with a conserved, albeit weaker, contact from the paternal allele with the Peg13 DMR.In addition, the nature of the strong contact with the Kcnk9 promoter, also observed in the Hi-C data (arrow  a in Figure S5), changed from biallelic in ESCs to maternally biased, but still biallelic, in NPCs.This change occurred along the entire sub-TAD where we observed preferential maternal and paternal interactions with the regions located on either side of the Peg13 DMR, respectively (Figure S5).
These data mirror and support those obtained from the 4C-seq data analysis using the Peg13 DMR and Kcnk9 promoter as viewpoints (Figures 4 and 5).They highlighted that the Peg13 DMR organizes the centromeric sub-TAD into two paternal sub-domains that isolate the Kcnk9 promoter from the PE on the paternal allele.However, this structure might be circumvented in ESCs where PE strongly contacted Kcnk9 (both alleles), indicating that these contacts precede Kcnk9 imprinted expression.In NPCs, and consistent with the gain of maternal expression of Kcnk9, contacts between PE and the promoter occurred preferentially, although not exclusively, on the maternal allele (Figures 5 and S5).

PE shows features of a biallelically active enhancer in ESCs, NPCs, and neonatal brain
The PE intronic region is one of the enhancers annotated using mouse transgenic experiments 53 with activity in the mouse brain (dataset ID: mm1679 in Vista Enhancer Browser).This finding and our chromatin structure data suggest that this intronic region could be an enhancer for both Peg13 and Kcnk9 but that the contacts already established in ESCs are not sufficient to induce maternal Kcnk9 expression and increased paternal Peg13 expression.
Therefore, we performed an integrative analysis to determine whether the changes in the molecular signature at this region could account for the change in Kcnk9 and Peg13 expression between ESCs and NPCs.Analysis of non-allelic ATAC-seq and WGBS datasets indicated that this region was in an open chromatin configuration, with a strong ATAC-seq signal and DNA methylation depletion in both cell types and brain tissues (Figures 6A  and S6).Allelic ChIP-seq, C&R (both performed using samples derived from the two reciprocal crosses), and ChIP-qPCR demonstrated that, in both ESCs and NPCs, as well as in neonatal brain tissue, several permissive/activating marks, including H3K27ac (a signature of active enhancers), were enriched on both alleles, whereas the repressive H3K27me3 mark was absent (Figures 6A and  S6B).Besides the histone signature, active enhancers also produce non-coding RNAs called enhancer RNAs (eR-NAs). 54Refined analysis of allelic RNA-seq data identified a biallelically expressed RNA that originated from this re-gion in NPCs and embryonic brain tissues (Figures 6C  and S6C).Time-course analysis by RT-qPCR confirmed that this RNA was biallelically expressed and that its expression slightly increased as ESCs differentiated into NPCs and was maintained in neonatal brains (Figure 6D).
Altogether, these observations suggest that the CTCFbound region in the Trappc9 intron is a bona fide biallelic enhancer that is pre-loaded in an active, but not productive, configuration on the Kcnk9 promoter in ESCs.Increased activity during ESC differentiation into NPCs and then in neonatal brain is associated with an increase in biallelic eRNA production.
Notably, we also observed that eRNA expression was maternally biased in adult brain (Figure S6D) and by data mining 55 that H3K27ac was enriched on the maternal allele of the PE region in the frontal cortex of adult mice (Figure S6E).This suggests that the enhancer activity can switch from a biallelic to a maternal bias in adult brain.

Discussion
In this study, we wanted to understand the regulation of the Peg13-Kcnk9 domain during neural commitment.Our mouse stem cell-based corticogenesis model combined with integrative analyses of multiple layers of regulation allowed obtaining a comprehensive view of the molecular events that take place during the establishment of Kcnk9 maternal expression.We found that, despite the allelic higher-order chromatin structure associated with CTCF, enhancer-Kcnk9 promoter contacts occurred on both alleles, but they were productive only on the maternal allele.This observation challenges the canonical model in which CTCF binding acts as a chromatin boundary and suggests a more refined role for allelic CTCF binding at this DMR and the resulting allelic chromatin loops at this locus.
The molecular patterns detected at the Peg13 locus using our stem cell-based corticogenesis model were in agreement with previous in vivo observations.This consistency in imprinted gene expression patterns, chromatin signature/conformation, and eRNA production provides additional evidence that our in vitro corticogenesis model recapitulates the complex regulations that occur in vivo during early brain development. 21Our observations are also consistent with the results of a study on human brain tissue, 24 thus suggesting that the mechanisms of the Peg13-Kcnk9 domain regulation are evolutionarily conserved.Therefore, our study provides the basis to investigate the etiology of neurodevelopmental and neurological disorders associated with this locus.Particularly, in addition to the documented missense Kcnk9 mutations, 32,33 the enhancer appears to be another target region for mutation and/or epigenetic alteration screening in patients with suspected Birk-Barel syndrome.
The observation that, in ESCs and NPCs, Kcnk9 and the Peg13 DMR are located in a different sub-TAD compared with Trappc9, Chrac1, and Ago2 provides a framework to explain the absence of imprinting in the last three genes in these cell types and more globally in the developing brain.More studies are required to determine whether and how this higher-order chromatin structure is reorganized later during brain development to allow the Peg13 DMR to direct the mechanism by which Trappc9, Chrac1, and Ago2 switch from biallelic to preferential maternal expression in postnatal brain.However, the recent suggestion that enhancerpromoter interactions may be ''memorized'' to influence promoter activity later in development [56][57][58] questions whether the inter-sub-TAD interactions observed between the enhancer and the Ago2 promoter in ESCs may contribute to instruct imprinted expression at later developmental stages.
Consistent with its germline DMR status, 23 our data suggest that the Peg13 promoter is the ICR of the locus.Indeed, it exhibits the characteristic allelic molecular feature of an ICR and its maternal methylation is required to control the maternal expression of Kcnk9, located approximately 250 kb away.Moreover, it is the only region of the domain that recruits CTCF in a parental allele-specific manner.It is reasonable to assume that the resulting higher-order chromatin structure differences between parental alleles provide a framework in which this ICR can impose the imprinted transcriptional program along the domain, as observed at the H19-Igf2 and Dlk1-Gtl2 imprinted loci. 15However, the underlying mechanism does not follow the canonical model in which parental-specific chromatin loops mediated by CTCF restrict enhancer-promoter interactions to the expressing allele only. 15,17As previously documented for a minority of enhancer-promoter pairs, 50,58 Kcnk9 promoterenhancer interactions are pre-established in ESCs that do not express Kcnk9 yet.The absence of Kcnk9 expression at this stage, despite the enhancer active signature, is intriguing.This may be explained by a cell context-dependent dual function of this regulatory element.As observed for other human and mouse regulatory elements, 59,60 it can recruit repressor or activator factors in ESCs and neural cells, respectively.More surprisingly, the enhancer-promoter contacts occur from both alleles, although they are only productive from the maternal allele after differentiation.
These observations suggest an interplay between the pre-existing chromatin structure, the allelic CTCF binding at the Peg13 DMR, and the transcriptional machinery to shape imprinted expression during neural differentiation.In this model (Figure 7), the sub-TAD anchored to the 5 0 part of Trappc9 and to the Kcnk9 intron provides a higher-order chromatin structure where CTCF-anchored chromatin loops (not informative at this stage) are formed in ESCs.Upon differentiation and recruitment of activator transcription factors to the enhancer, this structural organization guides productive contacts.On the paternal allele, the pre-existing interactions of the Peg13 DMR with the enhancer and Kcnk9 promoter allow the gain of Peg13 expression and keep Kcnk9 silent.Specifically, we propose that rather than isolating the enhancer from the promoter, the CTCF-mediated loops induce a threeway Kcnk9 promoter-Peg13 promoter-enhancer contact, where promoter competition for transcription factors and/or a physical barrier formed by the Peg13 DMR between the Kcnk9 promoter and the enhancer keep Kcnk9 silent.It has been proposed that this kind of multi-way interaction between enhancers and promoters, facilitated The sub-TAD anchored to CTCF-bound regions in the Kcnk9 intron and the 5 0 part of Trappc9 provides a higher-order chromatin structure in which CTCF-anchored chromatin loops lead to a three-way Kcnk9 promoter-Peg13 promoter/DMR-enhancer (E) interaction on the paternal allele (Pat.).Due to Peg13 DMR methylation (black circles), interaction occurs only between the Kcnk9 promoter and the enhancer on the maternal allele (Mat.).This scaffold is present but not yet informative in ESCs.During ESC differentiation into NPCs, it directs productive contacts and recruitment of ad hoc activator transcription factors (TF) at the enhancer.On the paternal allele, the pre-existing three-way interaction provides a structure in which the Peg13 DMR acts as a physical barrier between the Kcnk9 promoter and the enhancer, allowing the gain of Peg13 expression while keeping Kcnk9 silent.On the maternal allele, the pre-existing interaction between the enhancer and Kcnk9 promoter induces its maternal expression.The transcription machinery will in turn affect the chromatin structure by strengthening this interaction on the maternal allele as expression increases.
by the ordered chromatin structure, regulates the temporal expression along the a-globin locus. 61,624][65] The absence of any specific interaction on the paternal allele other than with the Peg13 DMR and the absence of a strong repressive signature, such as DNA methylation, on the Kcnk9 promoter rule out the action of a silencer and suggest that this is the main mechanism of Kcnk9 silencing maintenance.The allelic Peg13 DMR-associated sub-domains and enhancer-Kcnk9 promoter interactions we detected on the paternal allele fit with this three-way contact model.On the maternal allele, the pre-existing interaction between the enhancer and the Kcnk9 promoter induces its maternal expression.Moreover, the associated recruitment of RNAPolII, which promotes enhancer-promoter interactions, 66 will influence the chromatin structure by strengthening enhancer-Kcnk9 promoter and enhancer-Peg13 DMR interactions on the maternal and paternal alrespectively.These interactions will become stronger as expression increases.This model, which remains to be validated, explains the enhancer allelic specificity despite biallelic interactions with Kcnk9 and provides an alternative to the canonical isolation model.This model is supported also by the findings of a recent study that overlap and complement our results.Specifically, CRISPRinduced ectopic activation of the TrappC9 intronic enhancer identified here induced ectopic maternal expression of Kcnk9 in ESCs.Moreover, their HiC capture data showed that the interaction between this enhancer region and the Kcnk9 promoter is biallelic in ESCs and maternally biased but biallelic in in vitro-induced neurons and brain tissue. 67ur data are also in line with those obtained in a recent allelic chromatin conformation analysis in human cells showing that the CTCF-mediated insulator model described at the IGF2-H19 locus is not applicable to all imprinted loci. 68Indeed, our observation supports the hypothesis that, although CTCF binds to many ICRs on their unmethylated allele, 69 its function may not be universal at imprinted loci where it may act through different mechanisms.
Data and code availability d The data generated in this study have been deposited at the GEO data repository under the accession number GEO: GSE244147 and are publicly available as of the date of publication.Original data are available from the lead contact (philippe.arnaud@uca.f)on request.
d This paper does not report original codes.d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Figure 1 .
Figure 1.Kcnk9 and Peg13 expression dynamics during neural commitment (A) RT-qPCR analysis of Kcnk9 and Peg13 expression levels in the brain of newborn (NBB; n ¼ 4) and adult (n ¼ 2) B/J mice.The parental origin of expression is shown below.(B) Genome browser view at the Kcnk9 and Peg13 loci to show the allelic-oriented RNA-seq signal in B/J ESCs, NPCs, and embryonic dorsal telencephalon (DT); re-analyzed data from Bouschet et al. 21For each condition, the quantitative and merged parental allelic RNA-seq signals are at the top and bottom, respectively.Maternal and paternal expression levels are shown in red and blue, respectively.(C) RT-qPCR analysis of Kcnk9 and Peg13 expression levels in B/J ESCs (n ¼ 4) and at day 4 (D4; n ¼ 2), D6 (n ¼ 2), D8 (n ¼ 2) of in vitro corticogenesis and in NPCs (D12; n ¼ 4).The parental origin of expression is shown below.Statistical significance was determined with the unpaired t test (p values in the figure).In (A) and (C), the results are presented as percentage of expression relative to the geometric mean of the expression of the three housekeeping genes Gapdh, Gus, and Tbp.Data are the mean 5 SEM.The parental origin of expression was determined by direct sequencing of sample-specific PCR products based on strain-specific SNPs in the regions analyzed; the SNP presence is visible in the traces obtained using hybrid genomic DNA (gDNA) shown in (A).Representative data are shown.

Figure 2 .
Figure 2. Kcnk9 expression is lost in E9.5 Dnmt3l À/þ embryos RT-qPCR analysis of Kcnk9 and Peg13 expression levels in wild-type (n ¼ 2) and Dnmt3l À/þ (n ¼ 3) E9.5 embryos (head).Statistical significance was determined with the unpaired t test (p values in the figure).Data are the mean 5 SEM.The parental origin of expression is shown below.WT, wild type.

Figure 3 .
Figure 3. Epigenetic signatures at Kcnk9 and Peg13 in ESCs, NPCs, and neonatal brain (A) Genome Browser view at the Kcnk9 and Peg13 loci to show CpG island (CGI) positions and ATAC-seq data, methylation (WGBS), ZPF57, H3K4me2, H3K4me3, and H3K27me3 enrichment in ESCs and NPCs.For ZFP57 and the histone marks, data shown were obtained from B/J material, and the quantitative and the merged parental allelic signals are shown in the upper and lower panels, respectively.Maternal and paternal enrichments are shown in red and blue, respectively.(B) Chromatin analysis by native ChIP-qPCR to analyze the deposition of the indicated histone marks at the Peg13 and Kcnk9 promoters.The precipitation level was normalized to that obtained at the Rpl30 promoter (for H327ac, H3K4me2, and H3K4me3), the HoxA3 promoter (for H3K27me3), and IAP (for H3K9me3).For each condition, values are the mean of independent ChIP experiments (n), each performed in duplicate using B/J: ESCs (n ¼ 3), B/J NPCs (n ¼ 3), and B/J neonatal brain (NNB) (n ¼ 5).Data are the mean 5 SEM.The allelic distribution of each histone mark was determined by direct sequencing of the sample-specific PCR products containing a strain-specific SNP in the analyzed region; representative data are shown.

Figure 4 .
Figure 4.The Peg13 DMR and Kcnk9 promoter interact with the same PE in ESCs and NPCs Genome Browser view of the Kcnk9-Trappc9 genomic region to show in B/J ESCs (upper panel) and NPCs (lower panel) allelic 4C-seq data, from the Peg13 DMR and Kcnk9 promoter viewpoints, and CTCF C&R signals.The 4C-seq data are shown by merging the allelic signals; contacts mediated by the paternal and maternal alleles are shown in blue and red, respectively.The maternal/paternal interaction ratio is shown.CTCF-bound regions are highlighted in gray.Binding is biallelic with the exception of the paternally bound Peg13 DMR.

Figure 5 .
Figure 5.The Peg13 DMR and PE interactomes structure the higher-order chromatin conformation at the Peg13 domain Genome Browser view for the TAD at the Peg13 domain in ESCs (top panel) and NPCs (bottom panel) to show re-analyzed Hi-C data, allelic 4C-seq from the Peg13 DMR, Kcnk9 promoter, and the PE viewpoints and CTCF C&R signals.The 4C-seq data, obtained from B/J material, are shown by merging the allelic signals; contacts mediated by the paternal and maternal alleles are in blue and red, respectively.The sub-TADs that divide the imprinted domain are indicated by a dotted line.

Figure 6 .
Figure 6.PE molecular signature dynamics during neural commitment (A) Genome Browser view at the intronic PE region to show ATAC-seq, methylation (WGBS), ZPF57, H3K4me2, H3K4me3, and H3K27me3 enrichment data in ESCs and NPCs.For ZFP57 and the histone marks, data shown were obtained from B/J material; the quantitative and the merged parental allelic signals are shown in the upper and lower panels, respectively.Maternal and paternal enrichments are in red and blue, respectively.(B) Chromatin analysis following native ChIP-qPCR to analyze the deposition of the indicated histone marks.The precipitation level was normalized to that obtained at the Rpl30 promoter (for H3K27ac, H3K4me2, and H3K4me3), the HoxA3 promoter (for H3K27me3), and IAP (for H3K9me3).For each condition, values are the mean of at least three independent ChIP experiments (n), each performed in duplicate in B/J ESCs (n ¼ 4), NPCs (n ¼ 3), and NNB (n ¼ 4).The allelic distribution of each histone mark was determined by direct sequencing of the sample-specific PCR products containing a strain-specific SNP in the analyzed region; representative data are shown.(C) Genome Browser view at the PE region to show the allelic-oriented RNA-seq signal in B/J ESCs, NPCs, and embryonic DT; re-analyzed data from Bouschet et al.For each condition, the quantitative and the merged parental allelic RNA-seq signals are shown in the top and bottom panels, respectively.Maternal and paternal expression levels are in red and blue, respectively.(D) RT-qPCR analyses to assess PE-associated eRNA expression in B/J ESCs (n ¼ 3), at the day 4 (D4; n ¼ 2), D6 (n ¼ 2), D8 (n ¼ 2), and NPC (D12; n ¼ 3) stages of in vitro corticogenesis, and in B/J NNBs (n ¼ 4).Statistical significance was determined with the unpaired t test (p values in the figure).The parental origin of expression is shown in the lower panel.(B and D) Data are the mean 5 SEM.

Figure 7 .
Figure 7. Working modelThe sub-TAD anchored to CTCF-bound regions in the Kcnk9 intron and the 5 0 part of Trappc9 provides a higher-order chromatin structure in which CTCF-anchored chromatin loops lead to a three-way Kcnk9 promoter-Peg13 promoter/DMR-enhancer (E) interaction on the paternal allele (Pat.).Due to Peg13 DMR methylation (black circles), interaction occurs only between the Kcnk9 promoter and the enhancer on the maternal allele (Mat.).This scaffold is present but not yet informative in ESCs.During ESC differentiation into NPCs, it directs productive contacts and recruitment of ad hoc activator transcription factors (TF) at the enhancer.On the paternal allele, the pre-existing three-way interaction provides a structure in which the Peg13 DMR acts as a physical barrier between the Kcnk9 promoter and the enhancer, allowing the gain of Peg13 expression while keeping Kcnk9 silent.On the maternal allele, the pre-existing interaction between the enhancer and Kcnk9 promoter induces its maternal expression.The transcription machinery will in turn affect the chromatin structure by strengthening this interaction on the maternal allele as expression increases.